| 1 | /* |
| 2 | * Copyright (c) 1982, 1986, 1988, 1990, 1993 |
| 3 | * The Regents of the University of California. All rights reserved. |
| 4 | * |
| 5 | * Redistribution and use in source and binary forms, with or without |
| 6 | * modification, are permitted provided that the following conditions |
| 7 | * are met: |
| 8 | * 1. Redistributions of source code must retain the above copyright |
| 9 | * notice, this list of conditions and the following disclaimer. |
| 10 | * 2. Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in the |
| 12 | * documentation and/or other materials provided with the distribution. |
| 13 | * 3. All advertising materials mentioning features or use of this software |
| 14 | * must display the following acknowledgement: |
| 15 | * This product includes software developed by the University of |
| 16 | * California, Berkeley and its contributors. |
| 17 | * 4. Neither the name of the University nor the names of its contributors |
| 18 | * may be used to endorse or promote products derived from this software |
| 19 | * without specific prior written permission. |
| 20 | * |
| 21 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 22 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 23 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 24 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 25 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 26 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 27 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 28 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 29 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 30 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 31 | * SUCH DAMAGE. |
| 32 | * |
| 33 | * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 |
| 34 | * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $ |
| 35 | * $DragonFly: src/sys/kern/uipc_socket2.c,v 1.15 2005/01/26 23:09:57 hsu Exp $ |
| 36 | */ |
| 37 | |
| 38 | #include "opt_param.h" |
| 39 | #include <sys/param.h> |
| 40 | #include <sys/systm.h> |
| 41 | #include <sys/domain.h> |
| 42 | #include <sys/file.h> /* for maxfiles */ |
| 43 | #include <sys/kernel.h> |
| 44 | #include <sys/proc.h> |
| 45 | #include <sys/malloc.h> |
| 46 | #include <sys/mbuf.h> |
| 47 | #include <sys/protosw.h> |
| 48 | #include <sys/resourcevar.h> |
| 49 | #include <sys/stat.h> |
| 50 | #include <sys/socket.h> |
| 51 | #include <sys/socketvar.h> |
| 52 | #include <sys/signalvar.h> |
| 53 | #include <sys/sysctl.h> |
| 54 | #include <sys/aio.h> /* for aio_swake proto */ |
| 55 | #include <sys/event.h> |
| 56 | |
| 57 | #include <sys/thread2.h> |
| 58 | #include <sys/msgport2.h> |
| 59 | |
| 60 | int maxsockets; |
| 61 | |
| 62 | /* |
| 63 | * Primitive routines for operating on sockets and socket buffers |
| 64 | */ |
| 65 | |
| 66 | u_long sb_max = SB_MAX; |
| 67 | u_long sb_max_adj = |
| 68 | SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ |
| 69 | |
| 70 | static u_long sb_efficiency = 8; /* parameter for sbreserve() */ |
| 71 | |
| 72 | /* |
| 73 | * Procedures to manipulate state flags of socket |
| 74 | * and do appropriate wakeups. Normal sequence from the |
| 75 | * active (originating) side is that soisconnecting() is |
| 76 | * called during processing of connect() call, |
| 77 | * resulting in an eventual call to soisconnected() if/when the |
| 78 | * connection is established. When the connection is torn down |
| 79 | * soisdisconnecting() is called during processing of disconnect() call, |
| 80 | * and soisdisconnected() is called when the connection to the peer |
| 81 | * is totally severed. The semantics of these routines are such that |
| 82 | * connectionless protocols can call soisconnected() and soisdisconnected() |
| 83 | * only, bypassing the in-progress calls when setting up a ``connection'' |
| 84 | * takes no time. |
| 85 | * |
| 86 | * From the passive side, a socket is created with |
| 87 | * two queues of sockets: so_incomp for connections in progress |
| 88 | * and so_comp for connections already made and awaiting user acceptance. |
| 89 | * As a protocol is preparing incoming connections, it creates a socket |
| 90 | * structure queued on so_incomp by calling sonewconn(). When the connection |
| 91 | * is established, soisconnected() is called, and transfers the |
| 92 | * socket structure to so_comp, making it available to accept(). |
| 93 | * |
| 94 | * If a socket is closed with sockets on either |
| 95 | * so_incomp or so_comp, these sockets are dropped. |
| 96 | * |
| 97 | * If higher level protocols are implemented in |
| 98 | * the kernel, the wakeups done here will sometimes |
| 99 | * cause software-interrupt process scheduling. |
| 100 | */ |
| 101 | |
| 102 | void |
| 103 | soisconnecting(so) |
| 104 | struct socket *so; |
| 105 | { |
| 106 | |
| 107 | so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); |
| 108 | so->so_state |= SS_ISCONNECTING; |
| 109 | } |
| 110 | |
| 111 | void |
| 112 | soisconnected(so) |
| 113 | struct socket *so; |
| 114 | { |
| 115 | struct socket *head = so->so_head; |
| 116 | |
| 117 | so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); |
| 118 | so->so_state |= SS_ISCONNECTED; |
| 119 | if (head && (so->so_state & SS_INCOMP)) { |
| 120 | if ((so->so_options & SO_ACCEPTFILTER) != 0) { |
| 121 | so->so_upcall = head->so_accf->so_accept_filter->accf_callback; |
| 122 | so->so_upcallarg = head->so_accf->so_accept_filter_arg; |
| 123 | so->so_rcv.sb_flags |= SB_UPCALL; |
| 124 | so->so_options &= ~SO_ACCEPTFILTER; |
| 125 | so->so_upcall(so, so->so_upcallarg, 0); |
| 126 | return; |
| 127 | } |
| 128 | TAILQ_REMOVE(&head->so_incomp, so, so_list); |
| 129 | head->so_incqlen--; |
| 130 | so->so_state &= ~SS_INCOMP; |
| 131 | TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); |
| 132 | head->so_qlen++; |
| 133 | so->so_state |= SS_COMP; |
| 134 | sorwakeup(head); |
| 135 | wakeup_one(&head->so_timeo); |
| 136 | } else { |
| 137 | wakeup(&so->so_timeo); |
| 138 | sorwakeup(so); |
| 139 | sowwakeup(so); |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | void |
| 144 | soisdisconnecting(so) |
| 145 | struct socket *so; |
| 146 | { |
| 147 | |
| 148 | so->so_state &= ~SS_ISCONNECTING; |
| 149 | so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); |
| 150 | wakeup((caddr_t)&so->so_timeo); |
| 151 | sowwakeup(so); |
| 152 | sorwakeup(so); |
| 153 | } |
| 154 | |
| 155 | void |
| 156 | soisdisconnected(so) |
| 157 | struct socket *so; |
| 158 | { |
| 159 | |
| 160 | so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); |
| 161 | so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); |
| 162 | wakeup((caddr_t)&so->so_timeo); |
| 163 | sbdrop(&so->so_snd, so->so_snd.sb_cc); |
| 164 | sowwakeup(so); |
| 165 | sorwakeup(so); |
| 166 | } |
| 167 | |
| 168 | /* |
| 169 | * When an attempt at a new connection is noted on a socket |
| 170 | * which accepts connections, sonewconn is called. If the |
| 171 | * connection is possible (subject to space constraints, etc.) |
| 172 | * then we allocate a new structure, propoerly linked into the |
| 173 | * data structure of the original socket, and return this. |
| 174 | * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. |
| 175 | */ |
| 176 | struct socket * |
| 177 | sonewconn(struct socket *head, int connstatus) |
| 178 | { |
| 179 | struct socket *so; |
| 180 | struct pru_attach_info ai; |
| 181 | |
| 182 | if (head->so_qlen > 3 * head->so_qlimit / 2) |
| 183 | return ((struct socket *)0); |
| 184 | so = soalloc(0); |
| 185 | if (so == NULL) |
| 186 | return ((struct socket *)0); |
| 187 | if ((head->so_options & SO_ACCEPTFILTER) != 0) |
| 188 | connstatus = 0; |
| 189 | so->so_head = head; |
| 190 | so->so_type = head->so_type; |
| 191 | so->so_options = head->so_options &~ SO_ACCEPTCONN; |
| 192 | so->so_linger = head->so_linger; |
| 193 | so->so_state = head->so_state | SS_NOFDREF; |
| 194 | so->so_proto = head->so_proto; |
| 195 | so->so_timeo = head->so_timeo; |
| 196 | so->so_cred = crhold(head->so_cred); |
| 197 | ai.sb_rlimit = NULL; |
| 198 | ai.p_ucred = NULL; |
| 199 | ai.fd_rdir = NULL; /* jail code cruft XXX JH */ |
| 200 | if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat, NULL) || |
| 201 | /* Directly call function since we're already at protocol level. */ |
| 202 | (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) { |
| 203 | sodealloc(so); |
| 204 | return ((struct socket *)0); |
| 205 | } |
| 206 | |
| 207 | if (connstatus) { |
| 208 | TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); |
| 209 | so->so_state |= SS_COMP; |
| 210 | head->so_qlen++; |
| 211 | } else { |
| 212 | if (head->so_incqlen > head->so_qlimit) { |
| 213 | struct socket *sp; |
| 214 | sp = TAILQ_FIRST(&head->so_incomp); |
| 215 | (void) soabort(sp); |
| 216 | } |
| 217 | TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); |
| 218 | so->so_state |= SS_INCOMP; |
| 219 | head->so_incqlen++; |
| 220 | } |
| 221 | if (connstatus) { |
| 222 | sorwakeup(head); |
| 223 | wakeup((caddr_t)&head->so_timeo); |
| 224 | so->so_state |= connstatus; |
| 225 | } |
| 226 | return (so); |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * Socantsendmore indicates that no more data will be sent on the |
| 231 | * socket; it would normally be applied to a socket when the user |
| 232 | * informs the system that no more data is to be sent, by the protocol |
| 233 | * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data |
| 234 | * will be received, and will normally be applied to the socket by a |
| 235 | * protocol when it detects that the peer will send no more data. |
| 236 | * Data queued for reading in the socket may yet be read. |
| 237 | */ |
| 238 | |
| 239 | void |
| 240 | socantsendmore(so) |
| 241 | struct socket *so; |
| 242 | { |
| 243 | |
| 244 | so->so_state |= SS_CANTSENDMORE; |
| 245 | sowwakeup(so); |
| 246 | } |
| 247 | |
| 248 | void |
| 249 | socantrcvmore(so) |
| 250 | struct socket *so; |
| 251 | { |
| 252 | |
| 253 | so->so_state |= SS_CANTRCVMORE; |
| 254 | sorwakeup(so); |
| 255 | } |
| 256 | |
| 257 | /* |
| 258 | * Wait for data to arrive at/drain from a socket buffer. |
| 259 | */ |
| 260 | int |
| 261 | sbwait(sb) |
| 262 | struct sockbuf *sb; |
| 263 | { |
| 264 | |
| 265 | sb->sb_flags |= SB_WAIT; |
| 266 | return (tsleep((caddr_t)&sb->sb_cc, |
| 267 | ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), |
| 268 | "sbwait", |
| 269 | sb->sb_timeo)); |
| 270 | } |
| 271 | |
| 272 | /* |
| 273 | * Lock a sockbuf already known to be locked; |
| 274 | * return any error returned from sleep (EINTR). |
| 275 | */ |
| 276 | int |
| 277 | sb_lock(sb) |
| 278 | struct sockbuf *sb; |
| 279 | { |
| 280 | int error; |
| 281 | |
| 282 | while (sb->sb_flags & SB_LOCK) { |
| 283 | sb->sb_flags |= SB_WANT; |
| 284 | error = tsleep((caddr_t)&sb->sb_flags, |
| 285 | ((sb->sb_flags & SB_NOINTR) ? 0 : PCATCH), |
| 286 | "sblock", 0); |
| 287 | if (error) |
| 288 | return (error); |
| 289 | } |
| 290 | sb->sb_flags |= SB_LOCK; |
| 291 | return (0); |
| 292 | } |
| 293 | |
| 294 | /* |
| 295 | * Wakeup processes waiting on a socket buffer. Do asynchronous notification |
| 296 | * via SIGIO if the socket has the SS_ASYNC flag set. |
| 297 | */ |
| 298 | void |
| 299 | sowakeup(so, sb) |
| 300 | struct socket *so; |
| 301 | struct sockbuf *sb; |
| 302 | { |
| 303 | struct selinfo *selinfo = &sb->sb_sel; |
| 304 | |
| 305 | selwakeup(selinfo); |
| 306 | sb->sb_flags &= ~SB_SEL; |
| 307 | if (sb->sb_flags & SB_WAIT) { |
| 308 | sb->sb_flags &= ~SB_WAIT; |
| 309 | wakeup((caddr_t)&sb->sb_cc); |
| 310 | } |
| 311 | if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) |
| 312 | pgsigio(so->so_sigio, SIGIO, 0); |
| 313 | if (sb->sb_flags & SB_UPCALL) |
| 314 | (*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT); |
| 315 | if (sb->sb_flags & SB_AIO) |
| 316 | aio_swake(so, sb); |
| 317 | KNOTE(&selinfo->si_note, 0); |
| 318 | if (sb->sb_flags & SB_MEVENT) { |
| 319 | struct netmsg_so_notify *msg, *nmsg; |
| 320 | |
| 321 | TAILQ_FOREACH_MUTABLE(msg, &selinfo->si_mlist, nm_list, nmsg) { |
| 322 | if (msg->nm_predicate((struct netmsg *)msg)) { |
| 323 | TAILQ_REMOVE(&selinfo->si_mlist, msg, nm_list); |
| 324 | lwkt_replymsg(&msg->nm_lmsg, |
| 325 | msg->nm_lmsg.ms_error); |
| 326 | } |
| 327 | } |
| 328 | if (TAILQ_EMPTY(&sb->sb_sel.si_mlist)) |
| 329 | sb->sb_flags &= ~SB_MEVENT; |
| 330 | } |
| 331 | } |
| 332 | |
| 333 | /* |
| 334 | * Socket buffer (struct sockbuf) utility routines. |
| 335 | * |
| 336 | * Each socket contains two socket buffers: one for sending data and |
| 337 | * one for receiving data. Each buffer contains a queue of mbufs, |
| 338 | * information about the number of mbufs and amount of data in the |
| 339 | * queue, and other fields allowing select() statements and notification |
| 340 | * on data availability to be implemented. |
| 341 | * |
| 342 | * Data stored in a socket buffer is maintained as a list of records. |
| 343 | * Each record is a list of mbufs chained together with the m_next |
| 344 | * field. Records are chained together with the m_nextpkt field. The upper |
| 345 | * level routine soreceive() expects the following conventions to be |
| 346 | * observed when placing information in the receive buffer: |
| 347 | * |
| 348 | * 1. If the protocol requires each message be preceded by the sender's |
| 349 | * name, then a record containing that name must be present before |
| 350 | * any associated data (mbuf's must be of type MT_SONAME). |
| 351 | * 2. If the protocol supports the exchange of ``access rights'' (really |
| 352 | * just additional data associated with the message), and there are |
| 353 | * ``rights'' to be received, then a record containing this data |
| 354 | * should be present (mbuf's must be of type MT_RIGHTS). |
| 355 | * 3. If a name or rights record exists, then it must be followed by |
| 356 | * a data record, perhaps of zero length. |
| 357 | * |
| 358 | * Before using a new socket structure it is first necessary to reserve |
| 359 | * buffer space to the socket, by calling sbreserve(). This should commit |
| 360 | * some of the available buffer space in the system buffer pool for the |
| 361 | * socket (currently, it does nothing but enforce limits). The space |
| 362 | * should be released by calling sbrelease() when the socket is destroyed. |
| 363 | */ |
| 364 | |
| 365 | int |
| 366 | soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) |
| 367 | { |
| 368 | if (sbreserve(&so->so_snd, sndcc, so, rl) == 0) |
| 369 | goto bad; |
| 370 | if (sbreserve(&so->so_rcv, rcvcc, so, rl) == 0) |
| 371 | goto bad2; |
| 372 | if (so->so_rcv.sb_lowat == 0) |
| 373 | so->so_rcv.sb_lowat = 1; |
| 374 | if (so->so_snd.sb_lowat == 0) |
| 375 | so->so_snd.sb_lowat = MCLBYTES; |
| 376 | if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) |
| 377 | so->so_snd.sb_lowat = so->so_snd.sb_hiwat; |
| 378 | return (0); |
| 379 | bad2: |
| 380 | sbrelease(&so->so_snd, so); |
| 381 | bad: |
| 382 | return (ENOBUFS); |
| 383 | } |
| 384 | |
| 385 | static int |
| 386 | sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) |
| 387 | { |
| 388 | int error = 0; |
| 389 | u_long old_sb_max = sb_max; |
| 390 | |
| 391 | error = SYSCTL_OUT(req, arg1, sizeof(int)); |
| 392 | if (error || !req->newptr) |
| 393 | return (error); |
| 394 | error = SYSCTL_IN(req, arg1, sizeof(int)); |
| 395 | if (error) |
| 396 | return (error); |
| 397 | if (sb_max < MSIZE + MCLBYTES) { |
| 398 | sb_max = old_sb_max; |
| 399 | return (EINVAL); |
| 400 | } |
| 401 | sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); |
| 402 | return (0); |
| 403 | } |
| 404 | |
| 405 | /* |
| 406 | * Allot mbufs to a sockbuf. |
| 407 | * Attempt to scale mbmax so that mbcnt doesn't become limiting |
| 408 | * if buffering efficiency is near the normal case. |
| 409 | */ |
| 410 | int |
| 411 | sbreserve(struct sockbuf *sb, u_long cc, struct socket *so, struct rlimit *rl) |
| 412 | { |
| 413 | |
| 414 | /* |
| 415 | * rl will only be NULL when we're in an interrupt (eg, in tcp_input) |
| 416 | * or when called from netgraph (ie, ngd_attach) |
| 417 | */ |
| 418 | if (cc > sb_max_adj) |
| 419 | return (0); |
| 420 | if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc, |
| 421 | rl ? rl->rlim_cur : RLIM_INFINITY)) { |
| 422 | return (0); |
| 423 | } |
| 424 | sb->sb_mbmax = min(cc * sb_efficiency, sb_max); |
| 425 | if (sb->sb_lowat > sb->sb_hiwat) |
| 426 | sb->sb_lowat = sb->sb_hiwat; |
| 427 | return (1); |
| 428 | } |
| 429 | |
| 430 | /* |
| 431 | * Free mbufs held by a socket, and reserved mbuf space. |
| 432 | */ |
| 433 | void |
| 434 | sbrelease(sb, so) |
| 435 | struct sockbuf *sb; |
| 436 | struct socket *so; |
| 437 | { |
| 438 | |
| 439 | sbflush(sb); |
| 440 | (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, |
| 441 | RLIM_INFINITY); |
| 442 | sb->sb_mbmax = 0; |
| 443 | } |
| 444 | |
| 445 | /* |
| 446 | * Routines to add and remove |
| 447 | * data from an mbuf queue. |
| 448 | * |
| 449 | * The routines sbappend() or sbappendrecord() are normally called to |
| 450 | * append new mbufs to a socket buffer, after checking that adequate |
| 451 | * space is available, comparing the function sbspace() with the amount |
| 452 | * of data to be added. sbappendrecord() differs from sbappend() in |
| 453 | * that data supplied is treated as the beginning of a new record. |
| 454 | * To place a sender's address, optional access rights, and data in a |
| 455 | * socket receive buffer, sbappendaddr() should be used. To place |
| 456 | * access rights and data in a socket receive buffer, sbappendrights() |
| 457 | * should be used. In either case, the new data begins a new record. |
| 458 | * Note that unlike sbappend() and sbappendrecord(), these routines check |
| 459 | * for the caller that there will be enough space to store the data. |
| 460 | * Each fails if there is not enough space, or if it cannot find mbufs |
| 461 | * to store additional information in. |
| 462 | * |
| 463 | * Reliable protocols may use the socket send buffer to hold data |
| 464 | * awaiting acknowledgement. Data is normally copied from a socket |
| 465 | * send buffer in a protocol with m_copy for output to a peer, |
| 466 | * and then removing the data from the socket buffer with sbdrop() |
| 467 | * or sbdroprecord() when the data is acknowledged by the peer. |
| 468 | */ |
| 469 | |
| 470 | /* |
| 471 | * Append mbuf chain m to the last record in the |
| 472 | * socket buffer sb. The additional space associated |
| 473 | * the mbuf chain is recorded in sb. Empty mbufs are |
| 474 | * discarded and mbufs are compacted where possible. |
| 475 | */ |
| 476 | void |
| 477 | sbappend(sb, m) |
| 478 | struct sockbuf *sb; |
| 479 | struct mbuf *m; |
| 480 | { |
| 481 | struct mbuf *n; |
| 482 | |
| 483 | if (m == 0) |
| 484 | return; |
| 485 | n = sb->sb_mb; |
| 486 | if (n) { |
| 487 | while (n->m_nextpkt) |
| 488 | n = n->m_nextpkt; |
| 489 | do { |
| 490 | if (n->m_flags & M_EOR) { |
| 491 | sbappendrecord(sb, m); /* XXXXXX!!!! */ |
| 492 | return; |
| 493 | } |
| 494 | } while (n->m_next && (n = n->m_next)); |
| 495 | } |
| 496 | sbcompress(sb, m, n); |
| 497 | } |
| 498 | |
| 499 | /* |
| 500 | * sbappendstream() is an optimized form of sbappend() for protocols |
| 501 | * such as TCP that only have one record in the socket buffer, are |
| 502 | * not PR_ATOMIC, nor allow MT_CONTROL data. |
| 503 | */ |
| 504 | void |
| 505 | sbappendstream(struct sockbuf *sb, struct mbuf *m) |
| 506 | { |
| 507 | KKASSERT(m->m_nextpkt == NULL); |
| 508 | sbcompress(sb, m, sb->sb_lastmbuf); |
| 509 | } |
| 510 | |
| 511 | #ifdef SOCKBUF_DEBUG |
| 512 | void |
| 513 | sbcheck(sb) |
| 514 | struct sockbuf *sb; |
| 515 | { |
| 516 | struct mbuf *m; |
| 517 | struct mbuf *n = 0; |
| 518 | u_long len = 0, mbcnt = 0; |
| 519 | |
| 520 | for (m = sb->sb_mb; m; m = n) { |
| 521 | n = m->m_nextpkt; |
| 522 | for (; m; m = m->m_next) { |
| 523 | len += m->m_len; |
| 524 | mbcnt += MSIZE; |
| 525 | if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */ |
| 526 | mbcnt += m->m_ext.ext_size; |
| 527 | } |
| 528 | } |
| 529 | if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { |
| 530 | printf("cc %ld != %ld || mbcnt %ld != %ld\n", len, sb->sb_cc, |
| 531 | mbcnt, sb->sb_mbcnt); |
| 532 | panic("sbcheck"); |
| 533 | } |
| 534 | } |
| 535 | #endif |
| 536 | |
| 537 | /* |
| 538 | * As above, except the mbuf chain |
| 539 | * begins a new record. |
| 540 | */ |
| 541 | void |
| 542 | sbappendrecord(sb, m0) |
| 543 | struct sockbuf *sb; |
| 544 | struct mbuf *m0; |
| 545 | { |
| 546 | struct mbuf *m; |
| 547 | |
| 548 | if (m0 == 0) |
| 549 | return; |
| 550 | m = sb->sb_mb; |
| 551 | if (m) |
| 552 | while (m->m_nextpkt) |
| 553 | m = m->m_nextpkt; |
| 554 | /* |
| 555 | * Put the first mbuf on the queue. |
| 556 | * Note this permits zero length records. |
| 557 | */ |
| 558 | sballoc(sb, m0); |
| 559 | if (m) |
| 560 | m->m_nextpkt = m0; |
| 561 | else |
| 562 | sb->sb_mb = m0; |
| 563 | m = m0->m_next; |
| 564 | m0->m_next = 0; |
| 565 | if (m && (m0->m_flags & M_EOR)) { |
| 566 | m0->m_flags &= ~M_EOR; |
| 567 | m->m_flags |= M_EOR; |
| 568 | } |
| 569 | sbcompress(sb, m, m0); |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * As above except that OOB data |
| 574 | * is inserted at the beginning of the sockbuf, |
| 575 | * but after any other OOB data. |
| 576 | */ |
| 577 | void |
| 578 | sbinsertoob(sb, m0) |
| 579 | struct sockbuf *sb; |
| 580 | struct mbuf *m0; |
| 581 | { |
| 582 | struct mbuf *m; |
| 583 | struct mbuf **mp; |
| 584 | |
| 585 | if (m0 == 0) |
| 586 | return; |
| 587 | for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { |
| 588 | m = *mp; |
| 589 | again: |
| 590 | switch (m->m_type) { |
| 591 | |
| 592 | case MT_OOBDATA: |
| 593 | continue; /* WANT next train */ |
| 594 | |
| 595 | case MT_CONTROL: |
| 596 | m = m->m_next; |
| 597 | if (m) |
| 598 | goto again; /* inspect THIS train further */ |
| 599 | } |
| 600 | break; |
| 601 | } |
| 602 | /* |
| 603 | * Put the first mbuf on the queue. |
| 604 | * Note this permits zero length records. |
| 605 | */ |
| 606 | sballoc(sb, m0); |
| 607 | m0->m_nextpkt = *mp; |
| 608 | *mp = m0; |
| 609 | m = m0->m_next; |
| 610 | m0->m_next = 0; |
| 611 | if (m && (m0->m_flags & M_EOR)) { |
| 612 | m0->m_flags &= ~M_EOR; |
| 613 | m->m_flags |= M_EOR; |
| 614 | } |
| 615 | sbcompress(sb, m, m0); |
| 616 | } |
| 617 | |
| 618 | /* |
| 619 | * Append address and data, and optionally, control (ancillary) data |
| 620 | * to the receive queue of a socket. If present, |
| 621 | * m0 must include a packet header with total length. |
| 622 | * Returns 0 if no space in sockbuf or insufficient mbufs. |
| 623 | */ |
| 624 | int |
| 625 | sbappendaddr(sb, asa, m0, control) |
| 626 | struct sockbuf *sb; |
| 627 | const struct sockaddr *asa; |
| 628 | struct mbuf *m0, *control; |
| 629 | { |
| 630 | struct mbuf *m, *n; |
| 631 | int space = asa->sa_len; |
| 632 | |
| 633 | if (m0 && (m0->m_flags & M_PKTHDR) == 0) |
| 634 | panic("sbappendaddr"); |
| 635 | |
| 636 | if (m0) |
| 637 | space += m0->m_pkthdr.len; |
| 638 | for (n = control; n; n = n->m_next) { |
| 639 | space += n->m_len; |
| 640 | if (n->m_next == 0) /* keep pointer to last control buf */ |
| 641 | break; |
| 642 | } |
| 643 | if (space > sbspace(sb)) |
| 644 | return (0); |
| 645 | if (asa->sa_len > MLEN) |
| 646 | return (0); |
| 647 | MGET(m, MB_DONTWAIT, MT_SONAME); |
| 648 | if (m == 0) |
| 649 | return (0); |
| 650 | m->m_len = asa->sa_len; |
| 651 | bcopy(asa, mtod(m, caddr_t), asa->sa_len); |
| 652 | if (n) |
| 653 | n->m_next = m0; /* concatenate data to control */ |
| 654 | else |
| 655 | control = m0; |
| 656 | m->m_next = control; |
| 657 | for (n = m; n; n = n->m_next) |
| 658 | sballoc(sb, n); |
| 659 | n = sb->sb_mb; |
| 660 | if (n) { |
| 661 | while (n->m_nextpkt) |
| 662 | n = n->m_nextpkt; |
| 663 | n->m_nextpkt = m; |
| 664 | } else |
| 665 | sb->sb_mb = m; |
| 666 | return (1); |
| 667 | } |
| 668 | |
| 669 | int |
| 670 | sbappendcontrol(sb, m0, control) |
| 671 | struct sockbuf *sb; |
| 672 | struct mbuf *control, *m0; |
| 673 | { |
| 674 | struct mbuf *m, *n; |
| 675 | int space = 0; |
| 676 | |
| 677 | if (control == 0) |
| 678 | panic("sbappendcontrol"); |
| 679 | for (m = control; ; m = m->m_next) { |
| 680 | space += m->m_len; |
| 681 | if (m->m_next == 0) |
| 682 | break; |
| 683 | } |
| 684 | n = m; /* save pointer to last control buffer */ |
| 685 | for (m = m0; m; m = m->m_next) |
| 686 | space += m->m_len; |
| 687 | if (space > sbspace(sb)) |
| 688 | return (0); |
| 689 | n->m_next = m0; /* concatenate data to control */ |
| 690 | for (m = control; m; m = m->m_next) |
| 691 | sballoc(sb, m); |
| 692 | n = sb->sb_mb; |
| 693 | if (n) { |
| 694 | while (n->m_nextpkt) |
| 695 | n = n->m_nextpkt; |
| 696 | n->m_nextpkt = control; |
| 697 | } else |
| 698 | sb->sb_mb = control; |
| 699 | return (1); |
| 700 | } |
| 701 | |
| 702 | /* |
| 703 | * Compress mbuf chain m into the socket |
| 704 | * buffer sb following mbuf n. If n |
| 705 | * is null, the buffer is presumed empty. |
| 706 | */ |
| 707 | void |
| 708 | sbcompress(sb, m, n) |
| 709 | struct sockbuf *sb; |
| 710 | struct mbuf *m, *n; |
| 711 | { |
| 712 | int eor = 0; |
| 713 | struct mbuf *o; |
| 714 | |
| 715 | while (m) { |
| 716 | eor |= m->m_flags & M_EOR; |
| 717 | if (m->m_len == 0 && |
| 718 | (eor == 0 || |
| 719 | (((o = m->m_next) || (o = n)) && |
| 720 | o->m_type == m->m_type))) { |
| 721 | m = m_free(m); |
| 722 | continue; |
| 723 | } |
| 724 | if (n && (n->m_flags & M_EOR) == 0 && |
| 725 | M_WRITABLE(n) && |
| 726 | m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */ |
| 727 | m->m_len <= M_TRAILINGSPACE(n) && |
| 728 | n->m_type == m->m_type) { |
| 729 | bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len, |
| 730 | (unsigned)m->m_len); |
| 731 | n->m_len += m->m_len; |
| 732 | sb->sb_cc += m->m_len; |
| 733 | m = m_free(m); |
| 734 | continue; |
| 735 | } |
| 736 | if (n) |
| 737 | n->m_next = m; |
| 738 | else |
| 739 | sb->sb_mb = m; |
| 740 | sb->sb_lastmbuf = m; |
| 741 | sballoc(sb, m); |
| 742 | n = m; |
| 743 | m->m_flags &= ~M_EOR; |
| 744 | m = m->m_next; |
| 745 | n->m_next = 0; |
| 746 | } |
| 747 | if (eor) { |
| 748 | if (n) |
| 749 | n->m_flags |= eor; |
| 750 | else |
| 751 | printf("semi-panic: sbcompress"); |
| 752 | } |
| 753 | } |
| 754 | |
| 755 | /* |
| 756 | * Free all mbufs in a sockbuf. |
| 757 | * Check that all resources are reclaimed. |
| 758 | */ |
| 759 | void |
| 760 | sbflush(sb) |
| 761 | struct sockbuf *sb; |
| 762 | { |
| 763 | |
| 764 | if (sb->sb_flags & SB_LOCK) |
| 765 | panic("sbflush: locked"); |
| 766 | while (sb->sb_mbcnt) { |
| 767 | /* |
| 768 | * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty: |
| 769 | * we would loop forever. Panic instead. |
| 770 | */ |
| 771 | if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len)) |
| 772 | break; |
| 773 | sbdrop(sb, (int)sb->sb_cc); |
| 774 | } |
| 775 | KASSERT(!(sb->sb_cc || sb->sb_mb || sb->sb_mbcnt || sb->sb_lastmbuf), |
| 776 | ("sbflush: cc %ld || mb %p || mbcnt %ld || mbtail %p", |
| 777 | sb->sb_cc, sb->sb_mb, sb->sb_mbcnt, sb->sb_lastmbuf)); |
| 778 | } |
| 779 | |
| 780 | /* |
| 781 | * Drop data from (the front of) a sockbuf. |
| 782 | */ |
| 783 | void |
| 784 | sbdrop(sb, len) |
| 785 | struct sockbuf *sb; |
| 786 | int len; |
| 787 | { |
| 788 | struct mbuf *m; |
| 789 | struct mbuf *next; |
| 790 | |
| 791 | next = (m = sb->sb_mb) ? m->m_nextpkt : 0; |
| 792 | while (len > 0) { |
| 793 | if (m == 0) { |
| 794 | if (next == 0) |
| 795 | panic("sbdrop"); |
| 796 | m = next; |
| 797 | next = m->m_nextpkt; |
| 798 | continue; |
| 799 | } |
| 800 | if (m->m_len > len) { |
| 801 | m->m_len -= len; |
| 802 | m->m_data += len; |
| 803 | sb->sb_cc -= len; |
| 804 | break; |
| 805 | } |
| 806 | len -= m->m_len; |
| 807 | sbfree(sb, m); |
| 808 | m = m_free(m); |
| 809 | } |
| 810 | while (m && m->m_len == 0) { |
| 811 | sbfree(sb, m); |
| 812 | m = m_free(m); |
| 813 | } |
| 814 | if (m) { |
| 815 | sb->sb_mb = m; |
| 816 | m->m_nextpkt = next; |
| 817 | } else { |
| 818 | sb->sb_mb = next; |
| 819 | sb->sb_lastmbuf = NULL; |
| 820 | } |
| 821 | } |
| 822 | |
| 823 | /* |
| 824 | * Drop a record off the front of a sockbuf |
| 825 | * and move the next record to the front. |
| 826 | */ |
| 827 | void |
| 828 | sbdroprecord(sb) |
| 829 | struct sockbuf *sb; |
| 830 | { |
| 831 | struct mbuf *m; |
| 832 | |
| 833 | m = sb->sb_mb; |
| 834 | if (m) { |
| 835 | sb->sb_mb = m->m_nextpkt; |
| 836 | do { |
| 837 | sbfree(sb, m); |
| 838 | m = m_free(m); |
| 839 | } while (m); |
| 840 | } |
| 841 | } |
| 842 | |
| 843 | /* |
| 844 | * Create a "control" mbuf containing the specified data |
| 845 | * with the specified type for presentation on a socket buffer. |
| 846 | */ |
| 847 | struct mbuf * |
| 848 | sbcreatecontrol(p, size, type, level) |
| 849 | caddr_t p; |
| 850 | int size; |
| 851 | int type, level; |
| 852 | { |
| 853 | struct cmsghdr *cp; |
| 854 | struct mbuf *m; |
| 855 | |
| 856 | if (CMSG_SPACE((u_int)size) > MCLBYTES) |
| 857 | return ((struct mbuf *) NULL); |
| 858 | if ((m = m_get(MB_DONTWAIT, MT_CONTROL)) == NULL) |
| 859 | return ((struct mbuf *) NULL); |
| 860 | if (CMSG_SPACE((u_int)size) > MLEN) { |
| 861 | MCLGET(m, MB_DONTWAIT); |
| 862 | if ((m->m_flags & M_EXT) == 0) { |
| 863 | m_free(m); |
| 864 | return ((struct mbuf *) NULL); |
| 865 | } |
| 866 | } |
| 867 | cp = mtod(m, struct cmsghdr *); |
| 868 | m->m_len = 0; |
| 869 | KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m), |
| 870 | ("sbcreatecontrol: short mbuf")); |
| 871 | if (p != NULL) |
| 872 | (void)memcpy(CMSG_DATA(cp), p, size); |
| 873 | m->m_len = CMSG_SPACE(size); |
| 874 | cp->cmsg_len = CMSG_LEN(size); |
| 875 | cp->cmsg_level = level; |
| 876 | cp->cmsg_type = type; |
| 877 | return (m); |
| 878 | } |
| 879 | |
| 880 | /* |
| 881 | * Some routines that return EOPNOTSUPP for entry points that are not |
| 882 | * supported by a protocol. Fill in as needed. |
| 883 | */ |
| 884 | int |
| 885 | pru_accept_notsupp(struct socket *so, struct sockaddr **nam) |
| 886 | { |
| 887 | return EOPNOTSUPP; |
| 888 | } |
| 889 | |
| 890 | int |
| 891 | pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) |
| 892 | { |
| 893 | return EOPNOTSUPP; |
| 894 | } |
| 895 | |
| 896 | int |
| 897 | pru_connect2_notsupp(struct socket *so1, struct socket *so2) |
| 898 | { |
| 899 | return EOPNOTSUPP; |
| 900 | } |
| 901 | |
| 902 | int |
| 903 | pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, |
| 904 | struct ifnet *ifp, struct thread *td) |
| 905 | { |
| 906 | return EOPNOTSUPP; |
| 907 | } |
| 908 | |
| 909 | int |
| 910 | pru_listen_notsupp(struct socket *so, struct thread *td) |
| 911 | { |
| 912 | return EOPNOTSUPP; |
| 913 | } |
| 914 | |
| 915 | int |
| 916 | pru_rcvd_notsupp(struct socket *so, int flags) |
| 917 | { |
| 918 | return EOPNOTSUPP; |
| 919 | } |
| 920 | |
| 921 | int |
| 922 | pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) |
| 923 | { |
| 924 | return EOPNOTSUPP; |
| 925 | } |
| 926 | |
| 927 | /* |
| 928 | * This isn't really a ``null'' operation, but it's the default one |
| 929 | * and doesn't do anything destructive. |
| 930 | */ |
| 931 | int |
| 932 | pru_sense_null(struct socket *so, struct stat *sb) |
| 933 | { |
| 934 | sb->st_blksize = so->so_snd.sb_hiwat; |
| 935 | return 0; |
| 936 | } |
| 937 | |
| 938 | /* |
| 939 | * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers |
| 940 | * of this routine assume that it always succeeds, so we have to use a |
| 941 | * blockable allocation even though we might be called from a critical thread. |
| 942 | */ |
| 943 | struct sockaddr * |
| 944 | dup_sockaddr(const struct sockaddr *sa) |
| 945 | { |
| 946 | struct sockaddr *sa2; |
| 947 | |
| 948 | sa2 = malloc(sa->sa_len, M_SONAME, M_INTWAIT); |
| 949 | bcopy(sa, sa2, sa->sa_len); |
| 950 | return (sa2); |
| 951 | } |
| 952 | |
| 953 | /* |
| 954 | * Create an external-format (``xsocket'') structure using the information |
| 955 | * in the kernel-format socket structure pointed to by so. This is done |
| 956 | * to reduce the spew of irrelevant information over this interface, |
| 957 | * to isolate user code from changes in the kernel structure, and |
| 958 | * potentially to provide information-hiding if we decide that |
| 959 | * some of this information should be hidden from users. |
| 960 | */ |
| 961 | void |
| 962 | sotoxsocket(struct socket *so, struct xsocket *xso) |
| 963 | { |
| 964 | xso->xso_len = sizeof *xso; |
| 965 | xso->xso_so = so; |
| 966 | xso->so_type = so->so_type; |
| 967 | xso->so_options = so->so_options; |
| 968 | xso->so_linger = so->so_linger; |
| 969 | xso->so_state = so->so_state; |
| 970 | xso->so_pcb = so->so_pcb; |
| 971 | xso->xso_protocol = so->so_proto->pr_protocol; |
| 972 | xso->xso_family = so->so_proto->pr_domain->dom_family; |
| 973 | xso->so_qlen = so->so_qlen; |
| 974 | xso->so_incqlen = so->so_incqlen; |
| 975 | xso->so_qlimit = so->so_qlimit; |
| 976 | xso->so_timeo = so->so_timeo; |
| 977 | xso->so_error = so->so_error; |
| 978 | xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; |
| 979 | xso->so_oobmark = so->so_oobmark; |
| 980 | sbtoxsockbuf(&so->so_snd, &xso->so_snd); |
| 981 | sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); |
| 982 | xso->so_uid = so->so_cred->cr_uid; |
| 983 | } |
| 984 | |
| 985 | /* |
| 986 | * This does the same for sockbufs. Note that the xsockbuf structure, |
| 987 | * since it is always embedded in a socket, does not include a self |
| 988 | * pointer nor a length. We make this entry point public in case |
| 989 | * some other mechanism needs it. |
| 990 | */ |
| 991 | void |
| 992 | sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) |
| 993 | { |
| 994 | xsb->sb_cc = sb->sb_cc; |
| 995 | xsb->sb_hiwat = sb->sb_hiwat; |
| 996 | xsb->sb_mbcnt = sb->sb_mbcnt; |
| 997 | xsb->sb_mbmax = sb->sb_mbmax; |
| 998 | xsb->sb_lowat = sb->sb_lowat; |
| 999 | xsb->sb_flags = sb->sb_flags; |
| 1000 | xsb->sb_timeo = sb->sb_timeo; |
| 1001 | } |
| 1002 | |
| 1003 | /* |
| 1004 | * Here is the definition of some of the basic objects in the kern.ipc |
| 1005 | * branch of the MIB. |
| 1006 | */ |
| 1007 | SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); |
| 1008 | |
| 1009 | /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ |
| 1010 | static int dummy; |
| 1011 | SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); |
| 1012 | SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, |
| 1013 | &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); |
| 1014 | SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, |
| 1015 | &maxsockets, 0, "Maximum number of sockets avaliable"); |
| 1016 | SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, |
| 1017 | &sb_efficiency, 0, ""); |
| 1018 | |
| 1019 | /* |
| 1020 | * Initialise maxsockets |
| 1021 | */ |
| 1022 | static void init_maxsockets(void *ignored) |
| 1023 | { |
| 1024 | TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); |
| 1025 | maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); |
| 1026 | } |
| 1027 | SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); |