uipc_socket2.c revision 1.138
1/* $NetBSD: uipc_socket2.c,v 1.138 2020/08/26 22:54:30 christos Exp $ */ 2 3/*- 4 * Copyright (c) 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 26 * POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29/* 30 * Copyright (c) 1982, 1986, 1988, 1990, 1993 31 * The Regents of the University of California. All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 3. Neither the name of the University nor the names of its contributors 42 * may be used to endorse or promote products derived from this software 43 * without specific prior written permission. 44 * 45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 55 * SUCH DAMAGE. 56 * 57 * @(#)uipc_socket2.c 8.2 (Berkeley) 2/14/95 58 */ 59 60#include <sys/cdefs.h> 61__KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.138 2020/08/26 22:54:30 christos Exp $"); 62 63#ifdef _KERNEL_OPT 64#include "opt_ddb.h" 65#include "opt_mbuftrace.h" 66#include "opt_sb_max.h" 67#endif 68 69#include <sys/param.h> 70#include <sys/systm.h> 71#include <sys/proc.h> 72#include <sys/file.h> 73#include <sys/buf.h> 74#include <sys/mbuf.h> 75#include <sys/protosw.h> 76#include <sys/domain.h> 77#include <sys/poll.h> 78#include <sys/socket.h> 79#include <sys/socketvar.h> 80#include <sys/signalvar.h> 81#include <sys/kauth.h> 82#include <sys/pool.h> 83#include <sys/uidinfo.h> 84 85#ifdef DDB 86#include <sys/filedesc.h> 87#endif 88 89/* 90 * Primitive routines for operating on sockets and socket buffers. 91 * 92 * Connection life-cycle: 93 * 94 * Normal sequence from the active (originating) side: 95 * 96 * - soisconnecting() is called during processing of connect() call, 97 * - resulting in an eventual call to soisconnected() if/when the 98 * connection is established. 99 * 100 * When the connection is torn down during processing of disconnect(): 101 * 102 * - soisdisconnecting() is called and, 103 * - soisdisconnected() is called when the connection to the peer 104 * is totally severed. 105 * 106 * The semantics of these routines are such that connectionless protocols 107 * can call soisconnected() and soisdisconnected() only, bypassing the 108 * in-progress calls when setting up a ``connection'' takes no time. 109 * 110 * From the passive side, a socket is created with two queues of sockets: 111 * 112 * - so_q0 (0) for partial connections (i.e. connections in progress) 113 * - so_q (1) for connections already made and awaiting user acceptance. 114 * 115 * As a protocol is preparing incoming connections, it creates a socket 116 * structure queued on so_q0 by calling sonewconn(). When the connection 117 * is established, soisconnected() is called, and transfers the 118 * socket structure to so_q, making it available to accept(). 119 * 120 * If a socket is closed with sockets on either so_q0 or so_q, these 121 * sockets are dropped. 122 * 123 * Locking rules and assumptions: 124 * 125 * o socket::so_lock can change on the fly. The low level routines used 126 * to lock sockets are aware of this. When so_lock is acquired, the 127 * routine locking must check to see if so_lock still points to the 128 * lock that was acquired. If so_lock has changed in the meantime, the 129 * now irrelevant lock that was acquired must be dropped and the lock 130 * operation retried. Although not proven here, this is completely safe 131 * on a multiprocessor system, even with relaxed memory ordering, given 132 * the next two rules: 133 * 134 * o In order to mutate so_lock, the lock pointed to by the current value 135 * of so_lock must be held: i.e., the socket must be held locked by the 136 * changing thread. The thread must issue membar_exit() to prevent 137 * memory accesses being reordered, and can set so_lock to the desired 138 * value. If the lock pointed to by the new value of so_lock is not 139 * held by the changing thread, the socket must then be considered 140 * unlocked. 141 * 142 * o If so_lock is mutated, and the previous lock referred to by so_lock 143 * could still be visible to other threads in the system (e.g. via file 144 * descriptor or protocol-internal reference), then the old lock must 145 * remain valid until the socket and/or protocol control block has been 146 * torn down. 147 * 148 * o If a socket has a non-NULL so_head value (i.e. is in the process of 149 * connecting), then locking the socket must also lock the socket pointed 150 * to by so_head: their lock pointers must match. 151 * 152 * o If a socket has connections in progress (so_q, so_q0 not empty) then 153 * locking the socket must also lock the sockets attached to both queues. 154 * Again, their lock pointers must match. 155 * 156 * o Beyond the initial lock assignment in socreate(), assigning locks to 157 * sockets is the responsibility of the individual protocols / protocol 158 * domains. 159 */ 160 161static pool_cache_t socket_cache; 162u_long sb_max = SB_MAX;/* maximum socket buffer size */ 163static u_long sb_max_adj; /* adjusted sb_max */ 164 165void 166soisconnecting(struct socket *so) 167{ 168 169 KASSERT(solocked(so)); 170 171 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 172 so->so_state |= SS_ISCONNECTING; 173} 174 175void 176soisconnected(struct socket *so) 177{ 178 struct socket *head; 179 180 head = so->so_head; 181 182 KASSERT(solocked(so)); 183 KASSERT(head == NULL || solocked2(so, head)); 184 185 so->so_state &= ~(SS_ISCONNECTING | SS_ISDISCONNECTING); 186 so->so_state |= SS_ISCONNECTED; 187 if (head && so->so_onq == &head->so_q0) { 188 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 189 /* 190 * Re-enqueue and wake up any waiters, e.g. 191 * processes blocking on accept(). 192 */ 193 soqremque(so, 0); 194 soqinsque(head, so, 1); 195 sorwakeup(head); 196 cv_broadcast(&head->so_cv); 197 } else { 198 so->so_upcall = 199 head->so_accf->so_accept_filter->accf_callback; 200 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 201 so->so_rcv.sb_flags |= SB_UPCALL; 202 so->so_options &= ~SO_ACCEPTFILTER; 203 (*so->so_upcall)(so, so->so_upcallarg, 204 POLLIN|POLLRDNORM, M_DONTWAIT); 205 } 206 } else { 207 cv_broadcast(&so->so_cv); 208 sorwakeup(so); 209 sowwakeup(so); 210 } 211} 212 213void 214soisdisconnecting(struct socket *so) 215{ 216 217 KASSERT(solocked(so)); 218 219 so->so_state &= ~SS_ISCONNECTING; 220 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 221 cv_broadcast(&so->so_cv); 222 sowwakeup(so); 223 sorwakeup(so); 224} 225 226void 227soisdisconnected(struct socket *so) 228{ 229 230 KASSERT(solocked(so)); 231 232 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 233 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 234 cv_broadcast(&so->so_cv); 235 sowwakeup(so); 236 sorwakeup(so); 237} 238 239void 240soinit2(void) 241{ 242 243 socket_cache = pool_cache_init(sizeof(struct socket), 0, 0, 0, 244 "socket", NULL, IPL_SOFTNET, NULL, NULL, NULL); 245} 246 247/* 248 * sonewconn: accept a new connection. 249 * 250 * When an attempt at a new connection is noted on a socket which accepts 251 * connections, sonewconn(9) is called. If the connection is possible 252 * (subject to space constraints, etc) then we allocate a new structure, 253 * properly linked into the data structure of the original socket. 254 * 255 * => If 'soready' is true, then socket will become ready for accept() i.e. 256 * inserted into the so_q queue, SS_ISCONNECTED set and waiters awoken. 257 * => May be called from soft-interrupt context. 258 * => Listening socket should be locked. 259 * => Returns the new socket locked. 260 */ 261struct socket * 262sonewconn(struct socket *head, bool soready) 263{ 264 struct socket *so; 265 int soqueue, error; 266 267 KASSERT(solocked(head)); 268 269 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) { 270 /* 271 * Listen queue overflow. If there is an accept filter 272 * active, pass through the oldest cxn it's handling. 273 */ 274 if (head->so_accf == NULL) { 275 return NULL; 276 } else { 277 struct socket *so2, *next; 278 279 /* Pass the oldest connection waiting in the 280 accept filter */ 281 for (so2 = TAILQ_FIRST(&head->so_q0); 282 so2 != NULL; so2 = next) { 283 next = TAILQ_NEXT(so2, so_qe); 284 if (so2->so_upcall == NULL) { 285 continue; 286 } 287 so2->so_upcall = NULL; 288 so2->so_upcallarg = NULL; 289 so2->so_options &= ~SO_ACCEPTFILTER; 290 so2->so_rcv.sb_flags &= ~SB_UPCALL; 291 soisconnected(so2); 292 break; 293 } 294 295 /* If nothing was nudged out of the acept filter, bail 296 * out; otherwise proceed allocating the socket. */ 297 if (so2 == NULL) { 298 return NULL; 299 } 300 } 301 } 302 if ((head->so_options & SO_ACCEPTFILTER) != 0) { 303 soready = false; 304 } 305 soqueue = soready ? 1 : 0; 306 307 if ((so = soget(false)) == NULL) { 308 return NULL; 309 } 310 so->so_type = head->so_type; 311 so->so_options = head->so_options & ~SO_ACCEPTCONN; 312 so->so_linger = head->so_linger; 313 so->so_state = head->so_state | SS_NOFDREF; 314 so->so_proto = head->so_proto; 315 so->so_timeo = head->so_timeo; 316 so->so_pgid = head->so_pgid; 317 so->so_send = head->so_send; 318 so->so_receive = head->so_receive; 319 so->so_uidinfo = head->so_uidinfo; 320 so->so_egid = head->so_egid; 321 so->so_cpid = head->so_cpid; 322 323 /* 324 * Share the lock with the listening-socket, it may get unshared 325 * once the connection is complete. 326 */ 327 mutex_obj_hold(head->so_lock); 328 so->so_lock = head->so_lock; 329 330 /* 331 * Reserve the space for socket buffers. 332 */ 333#ifdef MBUFTRACE 334 so->so_mowner = head->so_mowner; 335 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner; 336 so->so_snd.sb_mowner = head->so_snd.sb_mowner; 337#endif 338 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 339 goto out; 340 } 341 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 342 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 343 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 344 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 345 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 346 so->so_snd.sb_flags |= head->so_snd.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 347 348 /* 349 * Finally, perform the protocol attach. Note: a new socket 350 * lock may be assigned at this point (if so, it will be held). 351 */ 352 error = (*so->so_proto->pr_usrreqs->pr_attach)(so, 0); 353 if (error) { 354out: 355 KASSERT(solocked(so)); 356 KASSERT(so->so_accf == NULL); 357 soput(so); 358 359 /* Note: the listening socket shall stay locked. */ 360 KASSERT(solocked(head)); 361 return NULL; 362 } 363 KASSERT(solocked2(head, so)); 364 365 /* 366 * Insert into the queue. If ready, update the connection status 367 * and wake up any waiters, e.g. processes blocking on accept(). 368 */ 369 soqinsque(head, so, soqueue); 370 if (soready) { 371 so->so_state |= SS_ISCONNECTED; 372 sorwakeup(head); 373 cv_broadcast(&head->so_cv); 374 } 375 return so; 376} 377 378struct socket * 379soget(bool waitok) 380{ 381 struct socket *so; 382 383 so = pool_cache_get(socket_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 384 if (__predict_false(so == NULL)) 385 return (NULL); 386 memset(so, 0, sizeof(*so)); 387 TAILQ_INIT(&so->so_q0); 388 TAILQ_INIT(&so->so_q); 389 cv_init(&so->so_cv, "socket"); 390 cv_init(&so->so_rcv.sb_cv, "netio"); 391 cv_init(&so->so_snd.sb_cv, "netio"); 392 selinit(&so->so_rcv.sb_sel); 393 selinit(&so->so_snd.sb_sel); 394 so->so_rcv.sb_so = so; 395 so->so_snd.sb_so = so; 396 return so; 397} 398 399void 400soput(struct socket *so) 401{ 402 403 KASSERT(!cv_has_waiters(&so->so_cv)); 404 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 405 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 406 seldestroy(&so->so_rcv.sb_sel); 407 seldestroy(&so->so_snd.sb_sel); 408 mutex_obj_free(so->so_lock); 409 cv_destroy(&so->so_cv); 410 cv_destroy(&so->so_rcv.sb_cv); 411 cv_destroy(&so->so_snd.sb_cv); 412 pool_cache_put(socket_cache, so); 413} 414 415/* 416 * soqinsque: insert socket of a new connection into the specified 417 * accept queue of the listening socket (head). 418 * 419 * q = 0: queue of partial connections 420 * q = 1: queue of incoming connections 421 */ 422void 423soqinsque(struct socket *head, struct socket *so, int q) 424{ 425 KASSERT(q == 0 || q == 1); 426 KASSERT(solocked2(head, so)); 427 KASSERT(so->so_onq == NULL); 428 KASSERT(so->so_head == NULL); 429 430 so->so_head = head; 431 if (q == 0) { 432 head->so_q0len++; 433 so->so_onq = &head->so_q0; 434 } else { 435 head->so_qlen++; 436 so->so_onq = &head->so_q; 437 } 438 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 439} 440 441/* 442 * soqremque: remove socket from the specified queue. 443 * 444 * => Returns true if socket was removed from the specified queue. 445 * => False if socket was not removed (because it was in other queue). 446 */ 447bool 448soqremque(struct socket *so, int q) 449{ 450 struct socket *head = so->so_head; 451 452 KASSERT(q == 0 || q == 1); 453 KASSERT(solocked(so)); 454 KASSERT(so->so_onq != NULL); 455 KASSERT(head != NULL); 456 457 if (q == 0) { 458 if (so->so_onq != &head->so_q0) 459 return false; 460 head->so_q0len--; 461 } else { 462 if (so->so_onq != &head->so_q) 463 return false; 464 head->so_qlen--; 465 } 466 KASSERT(solocked2(so, head)); 467 TAILQ_REMOVE(so->so_onq, so, so_qe); 468 so->so_onq = NULL; 469 so->so_head = NULL; 470 return true; 471} 472 473/* 474 * socantsendmore: indicates that no more data will be sent on the 475 * socket; it would normally be applied to a socket when the user 476 * informs the system that no more data is to be sent, by the protocol 477 * code (in case pr_shutdown()). 478 */ 479void 480socantsendmore(struct socket *so) 481{ 482 KASSERT(solocked(so)); 483 484 so->so_state |= SS_CANTSENDMORE; 485 sowwakeup(so); 486} 487 488/* 489 * socantrcvmore(): indicates that no more data will be received and 490 * will normally be applied to the socket by a protocol when it detects 491 * that the peer will send no more data. Data queued for reading in 492 * the socket may yet be read. 493 */ 494void 495socantrcvmore(struct socket *so) 496{ 497 KASSERT(solocked(so)); 498 499 so->so_state |= SS_CANTRCVMORE; 500 sorwakeup(so); 501} 502 503/* 504 * soroverflow(): indicates that data was attempted to be sent 505 * but the receiving buffer overflowed. 506 */ 507void 508soroverflow(struct socket *so) 509{ 510 KASSERT(solocked(so)); 511 512 so->so_rcv.sb_overflowed++; 513 if (so->so_options & SO_RERROR) { 514 so->so_rerror = ENOBUFS; 515 sorwakeup(so); 516 } 517} 518 519/* 520 * Wait for data to arrive at/drain from a socket buffer. 521 */ 522int 523sbwait(struct sockbuf *sb) 524{ 525 struct socket *so; 526 kmutex_t *lock; 527 int error; 528 529 so = sb->sb_so; 530 531 KASSERT(solocked(so)); 532 533 sb->sb_flags |= SB_NOTIFY; 534 lock = so->so_lock; 535 if ((sb->sb_flags & SB_NOINTR) != 0) 536 error = cv_timedwait(&sb->sb_cv, lock, sb->sb_timeo); 537 else 538 error = cv_timedwait_sig(&sb->sb_cv, lock, sb->sb_timeo); 539 if (__predict_false(lock != so->so_lock)) 540 solockretry(so, lock); 541 return error; 542} 543 544/* 545 * Wakeup processes waiting on a socket buffer. 546 * Do asynchronous notification via SIGIO 547 * if the socket buffer has the SB_ASYNC flag set. 548 */ 549void 550sowakeup(struct socket *so, struct sockbuf *sb, int code) 551{ 552 int band; 553 554 KASSERT(solocked(so)); 555 KASSERT(sb->sb_so == so); 556 557 if (code == POLL_IN) 558 band = POLLIN|POLLRDNORM; 559 else 560 band = POLLOUT|POLLWRNORM; 561 sb->sb_flags &= ~SB_NOTIFY; 562 selnotify(&sb->sb_sel, band, NOTE_SUBMIT); 563 cv_broadcast(&sb->sb_cv); 564 if (sb->sb_flags & SB_ASYNC) 565 fownsignal(so->so_pgid, SIGIO, code, band, so); 566 if (sb->sb_flags & SB_UPCALL) 567 (*so->so_upcall)(so, so->so_upcallarg, band, M_DONTWAIT); 568} 569 570/* 571 * Reset a socket's lock pointer. Wake all threads waiting on the 572 * socket's condition variables so that they can restart their waits 573 * using the new lock. The existing lock must be held. 574 */ 575void 576solockreset(struct socket *so, kmutex_t *lock) 577{ 578 579 KASSERT(solocked(so)); 580 581 so->so_lock = lock; 582 cv_broadcast(&so->so_snd.sb_cv); 583 cv_broadcast(&so->so_rcv.sb_cv); 584 cv_broadcast(&so->so_cv); 585} 586 587/* 588 * Socket buffer (struct sockbuf) utility routines. 589 * 590 * Each socket contains two socket buffers: one for sending data and 591 * one for receiving data. Each buffer contains a queue of mbufs, 592 * information about the number of mbufs and amount of data in the 593 * queue, and other fields allowing poll() statements and notification 594 * on data availability to be implemented. 595 * 596 * Data stored in a socket buffer is maintained as a list of records. 597 * Each record is a list of mbufs chained together with the m_next 598 * field. Records are chained together with the m_nextpkt field. The upper 599 * level routine soreceive() expects the following conventions to be 600 * observed when placing information in the receive buffer: 601 * 602 * 1. If the protocol requires each message be preceded by the sender's 603 * name, then a record containing that name must be present before 604 * any associated data (mbuf's must be of type MT_SONAME). 605 * 2. If the protocol supports the exchange of ``access rights'' (really 606 * just additional data associated with the message), and there are 607 * ``rights'' to be received, then a record containing this data 608 * should be present (mbuf's must be of type MT_CONTROL). 609 * 3. If a name or rights record exists, then it must be followed by 610 * a data record, perhaps of zero length. 611 * 612 * Before using a new socket structure it is first necessary to reserve 613 * buffer space to the socket, by calling sbreserve(). This should commit 614 * some of the available buffer space in the system buffer pool for the 615 * socket (currently, it does nothing but enforce limits). The space 616 * should be released by calling sbrelease() when the socket is destroyed. 617 */ 618 619int 620sb_max_set(u_long new_sbmax) 621{ 622 int s; 623 624 if (new_sbmax < (16 * 1024)) 625 return (EINVAL); 626 627 s = splsoftnet(); 628 sb_max = new_sbmax; 629 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES); 630 splx(s); 631 632 return (0); 633} 634 635int 636soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 637{ 638 KASSERT(so->so_pcb == NULL || solocked(so)); 639 640 /* 641 * there's at least one application (a configure script of screen) 642 * which expects a fifo is writable even if it has "some" bytes 643 * in its buffer. 644 * so we want to make sure (hiwat - lowat) >= (some bytes). 645 * 646 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above. 647 * we expect it's large enough for such applications. 648 */ 649 u_long lowat = MAX(sock_loan_thresh, MCLBYTES); 650 u_long hiwat = lowat + PIPE_BUF; 651 652 if (sndcc < hiwat) 653 sndcc = hiwat; 654 if (sbreserve(&so->so_snd, sndcc, so) == 0) 655 goto bad; 656 if (sbreserve(&so->so_rcv, rcvcc, so) == 0) 657 goto bad2; 658 if (so->so_rcv.sb_lowat == 0) 659 so->so_rcv.sb_lowat = 1; 660 if (so->so_snd.sb_lowat == 0) 661 so->so_snd.sb_lowat = lowat; 662 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 663 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 664 return (0); 665 bad2: 666 sbrelease(&so->so_snd, so); 667 bad: 668 return (ENOBUFS); 669} 670 671/* 672 * Allot mbufs to a sockbuf. 673 * Attempt to scale mbmax so that mbcnt doesn't become limiting 674 * if buffering efficiency is near the normal case. 675 */ 676int 677sbreserve(struct sockbuf *sb, u_long cc, struct socket *so) 678{ 679 struct lwp *l = curlwp; /* XXX */ 680 rlim_t maxcc; 681 struct uidinfo *uidinfo; 682 683 KASSERT(so->so_pcb == NULL || solocked(so)); 684 KASSERT(sb->sb_so == so); 685 KASSERT(sb_max_adj != 0); 686 687 if (cc == 0 || cc > sb_max_adj) 688 return (0); 689 690 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur; 691 692 uidinfo = so->so_uidinfo; 693 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc)) 694 return 0; 695 sb->sb_mbmax = uimin(cc * 2, sb_max); 696 if (sb->sb_lowat > sb->sb_hiwat) 697 sb->sb_lowat = sb->sb_hiwat; 698 699 return (1); 700} 701 702/* 703 * Free mbufs held by a socket, and reserved mbuf space. We do not assert 704 * that the socket is held locked here: see sorflush(). 705 */ 706void 707sbrelease(struct sockbuf *sb, struct socket *so) 708{ 709 710 KASSERT(sb->sb_so == so); 711 712 sbflush(sb); 713 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); 714 sb->sb_mbmax = 0; 715} 716 717/* 718 * Routines to add and remove 719 * data from an mbuf queue. 720 * 721 * The routines sbappend() or sbappendrecord() are normally called to 722 * append new mbufs to a socket buffer, after checking that adequate 723 * space is available, comparing the function sbspace() with the amount 724 * of data to be added. sbappendrecord() differs from sbappend() in 725 * that data supplied is treated as the beginning of a new record. 726 * To place a sender's address, optional access rights, and data in a 727 * socket receive buffer, sbappendaddr() should be used. To place 728 * access rights and data in a socket receive buffer, sbappendrights() 729 * should be used. In either case, the new data begins a new record. 730 * Note that unlike sbappend() and sbappendrecord(), these routines check 731 * for the caller that there will be enough space to store the data. 732 * Each fails if there is not enough space, or if it cannot find mbufs 733 * to store additional information in. 734 * 735 * Reliable protocols may use the socket send buffer to hold data 736 * awaiting acknowledgement. Data is normally copied from a socket 737 * send buffer in a protocol with m_copym for output to a peer, 738 * and then removing the data from the socket buffer with sbdrop() 739 * or sbdroprecord() when the data is acknowledged by the peer. 740 */ 741 742#ifdef SOCKBUF_DEBUG 743void 744sblastrecordchk(struct sockbuf *sb, const char *where) 745{ 746 struct mbuf *m = sb->sb_mb; 747 748 KASSERT(solocked(sb->sb_so)); 749 750 while (m && m->m_nextpkt) 751 m = m->m_nextpkt; 752 753 if (m != sb->sb_lastrecord) { 754 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 755 sb->sb_mb, sb->sb_lastrecord, m); 756 printf("packet chain:\n"); 757 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 758 printf("\t%p\n", m); 759 panic("sblastrecordchk from %s", where); 760 } 761} 762 763void 764sblastmbufchk(struct sockbuf *sb, const char *where) 765{ 766 struct mbuf *m = sb->sb_mb; 767 struct mbuf *n; 768 769 KASSERT(solocked(sb->sb_so)); 770 771 while (m && m->m_nextpkt) 772 m = m->m_nextpkt; 773 774 while (m && m->m_next) 775 m = m->m_next; 776 777 if (m != sb->sb_mbtail) { 778 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 779 sb->sb_mb, sb->sb_mbtail, m); 780 printf("packet tree:\n"); 781 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 782 printf("\t"); 783 for (n = m; n != NULL; n = n->m_next) 784 printf("%p ", n); 785 printf("\n"); 786 } 787 panic("sblastmbufchk from %s", where); 788 } 789} 790#endif /* SOCKBUF_DEBUG */ 791 792/* 793 * Link a chain of records onto a socket buffer 794 */ 795#define SBLINKRECORDCHAIN(sb, m0, mlast) \ 796do { \ 797 if ((sb)->sb_lastrecord != NULL) \ 798 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 799 else \ 800 (sb)->sb_mb = (m0); \ 801 (sb)->sb_lastrecord = (mlast); \ 802} while (/*CONSTCOND*/0) 803 804 805#define SBLINKRECORD(sb, m0) \ 806 SBLINKRECORDCHAIN(sb, m0, m0) 807 808/* 809 * Append mbuf chain m to the last record in the 810 * socket buffer sb. The additional space associated 811 * the mbuf chain is recorded in sb. Empty mbufs are 812 * discarded and mbufs are compacted where possible. 813 */ 814void 815sbappend(struct sockbuf *sb, struct mbuf *m) 816{ 817 struct mbuf *n; 818 819 KASSERT(solocked(sb->sb_so)); 820 821 if (m == NULL) 822 return; 823 824#ifdef MBUFTRACE 825 m_claimm(m, sb->sb_mowner); 826#endif 827 828 SBLASTRECORDCHK(sb, "sbappend 1"); 829 830 if ((n = sb->sb_lastrecord) != NULL) { 831 /* 832 * XXX Would like to simply use sb_mbtail here, but 833 * XXX I need to verify that I won't miss an EOR that 834 * XXX way. 835 */ 836 do { 837 if (n->m_flags & M_EOR) { 838 sbappendrecord(sb, m); /* XXXXXX!!!! */ 839 return; 840 } 841 } while (n->m_next && (n = n->m_next)); 842 } else { 843 /* 844 * If this is the first record in the socket buffer, it's 845 * also the last record. 846 */ 847 sb->sb_lastrecord = m; 848 } 849 sbcompress(sb, m, n); 850 SBLASTRECORDCHK(sb, "sbappend 2"); 851} 852 853/* 854 * This version of sbappend() should only be used when the caller 855 * absolutely knows that there will never be more than one record 856 * in the socket buffer, that is, a stream protocol (such as TCP). 857 */ 858void 859sbappendstream(struct sockbuf *sb, struct mbuf *m) 860{ 861 862 KASSERT(solocked(sb->sb_so)); 863 KDASSERT(m->m_nextpkt == NULL); 864 KASSERT(sb->sb_mb == sb->sb_lastrecord); 865 866 SBLASTMBUFCHK(sb, __func__); 867 868#ifdef MBUFTRACE 869 m_claimm(m, sb->sb_mowner); 870#endif 871 872 sbcompress(sb, m, sb->sb_mbtail); 873 874 sb->sb_lastrecord = sb->sb_mb; 875 SBLASTRECORDCHK(sb, __func__); 876} 877 878#ifdef SOCKBUF_DEBUG 879void 880sbcheck(struct sockbuf *sb) 881{ 882 struct mbuf *m, *m2; 883 u_long len, mbcnt; 884 885 KASSERT(solocked(sb->sb_so)); 886 887 len = 0; 888 mbcnt = 0; 889 for (m = sb->sb_mb; m; m = m->m_nextpkt) { 890 for (m2 = m; m2 != NULL; m2 = m2->m_next) { 891 len += m2->m_len; 892 mbcnt += MSIZE; 893 if (m2->m_flags & M_EXT) 894 mbcnt += m2->m_ext.ext_size; 895 if (m2->m_nextpkt != NULL) 896 panic("sbcheck nextpkt"); 897 } 898 } 899 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 900 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 901 mbcnt, sb->sb_mbcnt); 902 panic("sbcheck"); 903 } 904} 905#endif 906 907/* 908 * As above, except the mbuf chain 909 * begins a new record. 910 */ 911void 912sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 913{ 914 struct mbuf *m; 915 916 KASSERT(solocked(sb->sb_so)); 917 918 if (m0 == NULL) 919 return; 920 921#ifdef MBUFTRACE 922 m_claimm(m0, sb->sb_mowner); 923#endif 924 /* 925 * Put the first mbuf on the queue. 926 * Note this permits zero length records. 927 */ 928 sballoc(sb, m0); 929 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 930 SBLINKRECORD(sb, m0); 931 m = m0->m_next; 932 m0->m_next = 0; 933 if (m && (m0->m_flags & M_EOR)) { 934 m0->m_flags &= ~M_EOR; 935 m->m_flags |= M_EOR; 936 } 937 sbcompress(sb, m, m0); 938 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 939} 940 941/* 942 * As above except that OOB data 943 * is inserted at the beginning of the sockbuf, 944 * but after any other OOB data. 945 */ 946void 947sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 948{ 949 struct mbuf *m, **mp; 950 951 KASSERT(solocked(sb->sb_so)); 952 953 if (m0 == NULL) 954 return; 955 956 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 957 958 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 959 again: 960 switch (m->m_type) { 961 962 case MT_OOBDATA: 963 continue; /* WANT next train */ 964 965 case MT_CONTROL: 966 if ((m = m->m_next) != NULL) 967 goto again; /* inspect THIS train further */ 968 } 969 break; 970 } 971 /* 972 * Put the first mbuf on the queue. 973 * Note this permits zero length records. 974 */ 975 sballoc(sb, m0); 976 m0->m_nextpkt = *mp; 977 if (*mp == NULL) { 978 /* m0 is actually the new tail */ 979 sb->sb_lastrecord = m0; 980 } 981 *mp = m0; 982 m = m0->m_next; 983 m0->m_next = 0; 984 if (m && (m0->m_flags & M_EOR)) { 985 m0->m_flags &= ~M_EOR; 986 m->m_flags |= M_EOR; 987 } 988 sbcompress(sb, m, m0); 989 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 990} 991 992/* 993 * Append address and data, and optionally, control (ancillary) data 994 * to the receive queue of a socket. If present, 995 * m0 must include a packet header with total length. 996 * Returns 0 if no space in sockbuf or insufficient mbufs. 997 */ 998int 999sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, 1000 struct mbuf *control) 1001{ 1002 struct mbuf *m, *n, *nlast; 1003 int space, len; 1004 1005 KASSERT(solocked(sb->sb_so)); 1006 1007 space = asa->sa_len; 1008 1009 if (m0 != NULL) { 1010 if ((m0->m_flags & M_PKTHDR) == 0) 1011 panic("sbappendaddr"); 1012 space += m0->m_pkthdr.len; 1013#ifdef MBUFTRACE 1014 m_claimm(m0, sb->sb_mowner); 1015#endif 1016 } 1017 for (n = control; n; n = n->m_next) { 1018 space += n->m_len; 1019 MCLAIM(n, sb->sb_mowner); 1020 if (n->m_next == NULL) /* keep pointer to last control buf */ 1021 break; 1022 } 1023 if (space > sbspace(sb)) 1024 return (0); 1025 m = m_get(M_DONTWAIT, MT_SONAME); 1026 if (m == NULL) 1027 return (0); 1028 MCLAIM(m, sb->sb_mowner); 1029 /* 1030 * XXX avoid 'comparison always true' warning which isn't easily 1031 * avoided. 1032 */ 1033 len = asa->sa_len; 1034 if (len > MLEN) { 1035 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 1036 if ((m->m_flags & M_EXT) == 0) { 1037 m_free(m); 1038 return (0); 1039 } 1040 } 1041 m->m_len = asa->sa_len; 1042 memcpy(mtod(m, void *), asa, asa->sa_len); 1043 if (n) 1044 n->m_next = m0; /* concatenate data to control */ 1045 else 1046 control = m0; 1047 m->m_next = control; 1048 1049 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 1050 1051 for (n = m; n->m_next != NULL; n = n->m_next) 1052 sballoc(sb, n); 1053 sballoc(sb, n); 1054 nlast = n; 1055 SBLINKRECORD(sb, m); 1056 1057 sb->sb_mbtail = nlast; 1058 SBLASTMBUFCHK(sb, "sbappendaddr"); 1059 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 1060 1061 return (1); 1062} 1063 1064/* 1065 * Helper for sbappendchainaddr: prepend a struct sockaddr* to 1066 * an mbuf chain. 1067 */ 1068static inline struct mbuf * 1069m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0, 1070 const struct sockaddr *asa) 1071{ 1072 struct mbuf *m; 1073 const int salen = asa->sa_len; 1074 1075 KASSERT(solocked(sb->sb_so)); 1076 1077 /* only the first in each chain need be a pkthdr */ 1078 m = m_gethdr(M_DONTWAIT, MT_SONAME); 1079 if (m == NULL) 1080 return NULL; 1081 MCLAIM(m, sb->sb_mowner); 1082#ifdef notyet 1083 if (salen > MHLEN) { 1084 MEXTMALLOC(m, salen, M_NOWAIT); 1085 if ((m->m_flags & M_EXT) == 0) { 1086 m_free(m); 1087 return NULL; 1088 } 1089 } 1090#else 1091 KASSERT(salen <= MHLEN); 1092#endif 1093 m->m_len = salen; 1094 memcpy(mtod(m, void *), asa, salen); 1095 m->m_next = m0; 1096 m->m_pkthdr.len = salen + m0->m_pkthdr.len; 1097 1098 return m; 1099} 1100 1101int 1102sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa, 1103 struct mbuf *m0, int sbprio) 1104{ 1105 struct mbuf *m, *n, *n0, *nlast; 1106 int error; 1107 1108 KASSERT(solocked(sb->sb_so)); 1109 1110 /* 1111 * XXX sbprio reserved for encoding priority of this* request: 1112 * SB_PRIO_NONE --> honour normal sb limits 1113 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space, 1114 * take whole chain. Intended for large requests 1115 * that should be delivered atomically (all, or none). 1116 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow 1117 * over normal socket limits, for messages indicating 1118 * buffer overflow in earlier normal/lower-priority messages 1119 * SB_PRIO_BESTEFFORT --> ignore limits entirely. 1120 * Intended for kernel-generated messages only. 1121 * Up to generator to avoid total mbuf resource exhaustion. 1122 */ 1123 (void)sbprio; 1124 1125 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1126 panic("sbappendaddrchain"); 1127 1128#ifdef notyet 1129 space = sbspace(sb); 1130 1131 /* 1132 * Enforce SB_PRIO_* limits as described above. 1133 */ 1134#endif 1135 1136 n0 = NULL; 1137 nlast = NULL; 1138 for (m = m0; m; m = m->m_nextpkt) { 1139 struct mbuf *np; 1140 1141#ifdef MBUFTRACE 1142 m_claimm(m, sb->sb_mowner); 1143#endif 1144 1145 /* Prepend sockaddr to this record (m) of input chain m0 */ 1146 n = m_prepend_sockaddr(sb, m, asa); 1147 if (n == NULL) { 1148 error = ENOBUFS; 1149 goto bad; 1150 } 1151 1152 /* Append record (asa+m) to end of new chain n0 */ 1153 if (n0 == NULL) { 1154 n0 = n; 1155 } else { 1156 nlast->m_nextpkt = n; 1157 } 1158 /* Keep track of last record on new chain */ 1159 nlast = n; 1160 1161 for (np = n; np; np = np->m_next) 1162 sballoc(sb, np); 1163 } 1164 1165 SBLASTRECORDCHK(sb, "sbappendaddrchain 1"); 1166 1167 /* Drop the entire chain of (asa+m) records onto the socket */ 1168 SBLINKRECORDCHAIN(sb, n0, nlast); 1169 1170 SBLASTRECORDCHK(sb, "sbappendaddrchain 2"); 1171 1172 for (m = nlast; m->m_next; m = m->m_next) 1173 ; 1174 sb->sb_mbtail = m; 1175 SBLASTMBUFCHK(sb, "sbappendaddrchain"); 1176 1177 return (1); 1178 1179bad: 1180 /* 1181 * On error, free the prepended addreseses. For consistency 1182 * with sbappendaddr(), leave it to our caller to free 1183 * the input record chain passed to us as m0. 1184 */ 1185 while ((n = n0) != NULL) { 1186 struct mbuf *np; 1187 1188 /* Undo the sballoc() of this record */ 1189 for (np = n; np; np = np->m_next) 1190 sbfree(sb, np); 1191 1192 n0 = n->m_nextpkt; /* iterate at next prepended address */ 1193 np = m_free(n); /* free prepended address (not data) */ 1194 } 1195 return error; 1196} 1197 1198 1199int 1200sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 1201{ 1202 struct mbuf *m, *mlast, *n; 1203 int space; 1204 1205 KASSERT(solocked(sb->sb_so)); 1206 1207 space = 0; 1208 if (control == NULL) 1209 panic("sbappendcontrol"); 1210 for (m = control; ; m = m->m_next) { 1211 space += m->m_len; 1212 MCLAIM(m, sb->sb_mowner); 1213 if (m->m_next == NULL) 1214 break; 1215 } 1216 n = m; /* save pointer to last control buffer */ 1217 for (m = m0; m; m = m->m_next) { 1218 MCLAIM(m, sb->sb_mowner); 1219 space += m->m_len; 1220 } 1221 if (space > sbspace(sb)) 1222 return (0); 1223 n->m_next = m0; /* concatenate data to control */ 1224 1225 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 1226 1227 for (m = control; m->m_next != NULL; m = m->m_next) 1228 sballoc(sb, m); 1229 sballoc(sb, m); 1230 mlast = m; 1231 SBLINKRECORD(sb, control); 1232 1233 sb->sb_mbtail = mlast; 1234 SBLASTMBUFCHK(sb, "sbappendcontrol"); 1235 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 1236 1237 return (1); 1238} 1239 1240/* 1241 * Compress mbuf chain m into the socket 1242 * buffer sb following mbuf n. If n 1243 * is null, the buffer is presumed empty. 1244 */ 1245void 1246sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1247{ 1248 int eor; 1249 struct mbuf *o; 1250 1251 KASSERT(solocked(sb->sb_so)); 1252 1253 eor = 0; 1254 while (m) { 1255 eor |= m->m_flags & M_EOR; 1256 if (m->m_len == 0 && 1257 (eor == 0 || 1258 (((o = m->m_next) || (o = n)) && 1259 o->m_type == m->m_type))) { 1260 if (sb->sb_lastrecord == m) 1261 sb->sb_lastrecord = m->m_next; 1262 m = m_free(m); 1263 continue; 1264 } 1265 if (n && (n->m_flags & M_EOR) == 0 && 1266 /* M_TRAILINGSPACE() checks buffer writeability */ 1267 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 1268 m->m_len <= M_TRAILINGSPACE(n) && 1269 n->m_type == m->m_type) { 1270 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *), 1271 (unsigned)m->m_len); 1272 n->m_len += m->m_len; 1273 sb->sb_cc += m->m_len; 1274 m = m_free(m); 1275 continue; 1276 } 1277 if (n) 1278 n->m_next = m; 1279 else 1280 sb->sb_mb = m; 1281 sb->sb_mbtail = m; 1282 sballoc(sb, m); 1283 n = m; 1284 m->m_flags &= ~M_EOR; 1285 m = m->m_next; 1286 n->m_next = 0; 1287 } 1288 if (eor) { 1289 if (n) 1290 n->m_flags |= eor; 1291 else 1292 printf("semi-panic: sbcompress\n"); 1293 } 1294 SBLASTMBUFCHK(sb, __func__); 1295} 1296 1297/* 1298 * Free all mbufs in a sockbuf. 1299 * Check that all resources are reclaimed. 1300 */ 1301void 1302sbflush(struct sockbuf *sb) 1303{ 1304 1305 KASSERT(solocked(sb->sb_so)); 1306 KASSERT((sb->sb_flags & SB_LOCK) == 0); 1307 1308 while (sb->sb_mbcnt) 1309 sbdrop(sb, (int)sb->sb_cc); 1310 1311 KASSERT(sb->sb_cc == 0); 1312 KASSERT(sb->sb_mb == NULL); 1313 KASSERT(sb->sb_mbtail == NULL); 1314 KASSERT(sb->sb_lastrecord == NULL); 1315} 1316 1317/* 1318 * Drop data from (the front of) a sockbuf. 1319 */ 1320void 1321sbdrop(struct sockbuf *sb, int len) 1322{ 1323 struct mbuf *m, *next; 1324 1325 KASSERT(solocked(sb->sb_so)); 1326 1327 next = (m = sb->sb_mb) ? m->m_nextpkt : NULL; 1328 while (len > 0) { 1329 if (m == NULL) { 1330 if (next == NULL) 1331 panic("sbdrop(%p,%d): cc=%lu", 1332 sb, len, sb->sb_cc); 1333 m = next; 1334 next = m->m_nextpkt; 1335 continue; 1336 } 1337 if (m->m_len > len) { 1338 m->m_len -= len; 1339 m->m_data += len; 1340 sb->sb_cc -= len; 1341 break; 1342 } 1343 len -= m->m_len; 1344 sbfree(sb, m); 1345 m = m_free(m); 1346 } 1347 while (m && m->m_len == 0) { 1348 sbfree(sb, m); 1349 m = m_free(m); 1350 } 1351 if (m) { 1352 sb->sb_mb = m; 1353 m->m_nextpkt = next; 1354 } else 1355 sb->sb_mb = next; 1356 /* 1357 * First part is an inline SB_EMPTY_FIXUP(). Second part 1358 * makes sure sb_lastrecord is up-to-date if we dropped 1359 * part of the last record. 1360 */ 1361 m = sb->sb_mb; 1362 if (m == NULL) { 1363 sb->sb_mbtail = NULL; 1364 sb->sb_lastrecord = NULL; 1365 } else if (m->m_nextpkt == NULL) 1366 sb->sb_lastrecord = m; 1367} 1368 1369/* 1370 * Drop a record off the front of a sockbuf 1371 * and move the next record to the front. 1372 */ 1373void 1374sbdroprecord(struct sockbuf *sb) 1375{ 1376 struct mbuf *m, *mn; 1377 1378 KASSERT(solocked(sb->sb_so)); 1379 1380 m = sb->sb_mb; 1381 if (m) { 1382 sb->sb_mb = m->m_nextpkt; 1383 do { 1384 sbfree(sb, m); 1385 mn = m_free(m); 1386 } while ((m = mn) != NULL); 1387 } 1388 SB_EMPTY_FIXUP(sb); 1389} 1390 1391/* 1392 * Create a "control" mbuf containing the specified data 1393 * with the specified type for presentation on a socket buffer. 1394 */ 1395struct mbuf * 1396sbcreatecontrol1(void **p, int size, int type, int level, int flags) 1397{ 1398 struct cmsghdr *cp; 1399 struct mbuf *m; 1400 int space = CMSG_SPACE(size); 1401 1402 if ((flags & M_DONTWAIT) && space > MCLBYTES) { 1403 printf("%s: message too large %d\n", __func__, space); 1404 return NULL; 1405 } 1406 1407 if ((m = m_get(flags, MT_CONTROL)) == NULL) 1408 return NULL; 1409 if (space > MLEN) { 1410 if (space > MCLBYTES) 1411 MEXTMALLOC(m, space, M_WAITOK); 1412 else 1413 MCLGET(m, flags); 1414 if ((m->m_flags & M_EXT) == 0) { 1415 m_free(m); 1416 return NULL; 1417 } 1418 } 1419 cp = mtod(m, struct cmsghdr *); 1420 *p = CMSG_DATA(cp); 1421 m->m_len = space; 1422 cp->cmsg_len = CMSG_LEN(size); 1423 cp->cmsg_level = level; 1424 cp->cmsg_type = type; 1425 1426 memset(cp + 1, 0, CMSG_LEN(0) - sizeof(*cp)); 1427 memset((uint8_t *)*p + size, 0, CMSG_ALIGN(size) - size); 1428 1429 return m; 1430} 1431 1432struct mbuf * 1433sbcreatecontrol(void *p, int size, int type, int level) 1434{ 1435 struct mbuf *m; 1436 void *v; 1437 1438 m = sbcreatecontrol1(&v, size, type, level, M_DONTWAIT); 1439 if (m == NULL) 1440 return NULL; 1441 memcpy(v, p, size); 1442 return m; 1443} 1444 1445void 1446solockretry(struct socket *so, kmutex_t *lock) 1447{ 1448 1449 while (lock != so->so_lock) { 1450 mutex_exit(lock); 1451 lock = so->so_lock; 1452 mutex_enter(lock); 1453 } 1454} 1455 1456bool 1457solocked(const struct socket *so) 1458{ 1459 1460 return mutex_owned(so->so_lock); 1461} 1462 1463bool 1464solocked2(const struct socket *so1, const struct socket *so2) 1465{ 1466 const kmutex_t *lock; 1467 1468 lock = so1->so_lock; 1469 if (lock != so2->so_lock) 1470 return false; 1471 return mutex_owned(lock); 1472} 1473 1474/* 1475 * sosetlock: assign a default lock to a new socket. 1476 */ 1477void 1478sosetlock(struct socket *so) 1479{ 1480 if (so->so_lock == NULL) { 1481 kmutex_t *lock = softnet_lock; 1482 1483 so->so_lock = lock; 1484 mutex_obj_hold(lock); 1485 mutex_enter(lock); 1486 } 1487 KASSERT(solocked(so)); 1488} 1489 1490/* 1491 * Set lock on sockbuf sb; sleep if lock is already held. 1492 * Unless SB_NOINTR is set on sockbuf, sleep is interruptible. 1493 * Returns error without lock if sleep is interrupted. 1494 */ 1495int 1496sblock(struct sockbuf *sb, int wf) 1497{ 1498 struct socket *so; 1499 kmutex_t *lock; 1500 int error; 1501 1502 KASSERT(solocked(sb->sb_so)); 1503 1504 for (;;) { 1505 if (__predict_true((sb->sb_flags & SB_LOCK) == 0)) { 1506 sb->sb_flags |= SB_LOCK; 1507 return 0; 1508 } 1509 if (wf != M_WAITOK) 1510 return EWOULDBLOCK; 1511 so = sb->sb_so; 1512 lock = so->so_lock; 1513 if ((sb->sb_flags & SB_NOINTR) != 0) { 1514 cv_wait(&so->so_cv, lock); 1515 error = 0; 1516 } else 1517 error = cv_wait_sig(&so->so_cv, lock); 1518 if (__predict_false(lock != so->so_lock)) 1519 solockretry(so, lock); 1520 if (error != 0) 1521 return error; 1522 } 1523} 1524 1525void 1526sbunlock(struct sockbuf *sb) 1527{ 1528 struct socket *so; 1529 1530 so = sb->sb_so; 1531 1532 KASSERT(solocked(so)); 1533 KASSERT((sb->sb_flags & SB_LOCK) != 0); 1534 1535 sb->sb_flags &= ~SB_LOCK; 1536 cv_broadcast(&so->so_cv); 1537} 1538 1539int 1540sowait(struct socket *so, bool catch_p, int timo) 1541{ 1542 kmutex_t *lock; 1543 int error; 1544 1545 KASSERT(solocked(so)); 1546 KASSERT(catch_p || timo != 0); 1547 1548 lock = so->so_lock; 1549 if (catch_p) 1550 error = cv_timedwait_sig(&so->so_cv, lock, timo); 1551 else 1552 error = cv_timedwait(&so->so_cv, lock, timo); 1553 if (__predict_false(lock != so->so_lock)) 1554 solockretry(so, lock); 1555 return error; 1556} 1557 1558#ifdef DDB 1559 1560/* 1561 * Currently, sofindproc() is used only from DDB. It could be used from others 1562 * by using db_mutex_enter() 1563 */ 1564 1565static inline int 1566db_mutex_enter(kmutex_t *mtx) 1567{ 1568 extern int db_active; 1569 int rv; 1570 1571 if (!db_active) { 1572 mutex_enter(mtx); 1573 rv = 1; 1574 } else 1575 rv = mutex_tryenter(mtx); 1576 1577 return rv; 1578} 1579 1580int 1581sofindproc(struct socket *so, int all, void (*pr)(const char *, ...)) 1582{ 1583 proc_t *p; 1584 filedesc_t *fdp; 1585 fdtab_t *dt; 1586 fdfile_t *ff; 1587 file_t *fp = NULL; 1588 int found = 0; 1589 int i, t; 1590 1591 if (so == NULL) 1592 return 0; 1593 1594 t = db_mutex_enter(&proc_lock); 1595 if (!t) { 1596 pr("could not acquire proc_lock mutex\n"); 1597 return 0; 1598 } 1599 PROCLIST_FOREACH(p, &allproc) { 1600 if (p->p_stat == SIDL) 1601 continue; 1602 fdp = p->p_fd; 1603 t = db_mutex_enter(&fdp->fd_lock); 1604 if (!t) { 1605 pr("could not acquire fd_lock mutex\n"); 1606 continue; 1607 } 1608 dt = atomic_load_consume(&fdp->fd_dt); 1609 for (i = 0; i < dt->dt_nfiles; i++) { 1610 ff = dt->dt_ff[i]; 1611 if (ff == NULL) 1612 continue; 1613 1614 fp = atomic_load_consume(&ff->ff_file); 1615 if (fp == NULL) 1616 continue; 1617 1618 t = db_mutex_enter(&fp->f_lock); 1619 if (!t) { 1620 pr("could not acquire f_lock mutex\n"); 1621 continue; 1622 } 1623 if ((struct socket *)fp->f_data != so) { 1624 mutex_exit(&fp->f_lock); 1625 continue; 1626 } 1627 found++; 1628 if (pr) 1629 pr("socket %p: owner %s(pid=%d)\n", 1630 so, p->p_comm, p->p_pid); 1631 mutex_exit(&fp->f_lock); 1632 if (all == 0) 1633 break; 1634 } 1635 mutex_exit(&fdp->fd_lock); 1636 if (all == 0 && found != 0) 1637 break; 1638 } 1639 mutex_exit(&proc_lock); 1640 1641 return found; 1642} 1643 1644void 1645socket_print(const char *modif, void (*pr)(const char *, ...)) 1646{ 1647 file_t *fp; 1648 struct socket *so; 1649 struct sockbuf *sb_snd, *sb_rcv; 1650 struct mbuf *m_rec, *m; 1651 bool opt_v = false; 1652 bool opt_m = false; 1653 bool opt_a = false; 1654 bool opt_p = false; 1655 int nrecs, nmbufs; 1656 char ch; 1657 const char *family; 1658 1659 while ( (ch = *(modif++)) != '\0') { 1660 switch (ch) { 1661 case 'v': 1662 opt_v = true; 1663 break; 1664 case 'm': 1665 opt_m = true; 1666 break; 1667 case 'a': 1668 opt_a = true; 1669 break; 1670 case 'p': 1671 opt_p = true; 1672 break; 1673 } 1674 } 1675 if (opt_v == false && pr) 1676 (pr)("Ignore empty sockets. use /v to print all.\n"); 1677 if (opt_p == true && pr) 1678 (pr)("Don't search owner process.\n"); 1679 1680 LIST_FOREACH(fp, &filehead, f_list) { 1681 if (fp->f_type != DTYPE_SOCKET) 1682 continue; 1683 so = (struct socket *)fp->f_data; 1684 if (so == NULL) 1685 continue; 1686 1687 if (so->so_proto->pr_domain->dom_family == AF_INET) 1688 family = "INET"; 1689#ifdef INET6 1690 else if (so->so_proto->pr_domain->dom_family == AF_INET6) 1691 family = "INET6"; 1692#endif 1693 else if (so->so_proto->pr_domain->dom_family == pseudo_AF_KEY) 1694 family = "KEY"; 1695 else if (so->so_proto->pr_domain->dom_family == AF_ROUTE) 1696 family = "ROUTE"; 1697 else 1698 continue; 1699 1700 sb_snd = &so->so_snd; 1701 sb_rcv = &so->so_rcv; 1702 1703 if (opt_v != true && 1704 sb_snd->sb_cc == 0 && sb_rcv->sb_cc == 0) 1705 continue; 1706 1707 pr("---SOCKET %p: type %s\n", so, family); 1708 if (opt_p != true) 1709 sofindproc(so, opt_a == true ? 1 : 0, pr); 1710 pr("Send Buffer Bytes: %d [bytes]\n", sb_snd->sb_cc); 1711 pr("Send Buffer mbufs:\n"); 1712 m_rec = m = sb_snd->sb_mb; 1713 nrecs = 0; 1714 nmbufs = 0; 1715 while (m_rec) { 1716 nrecs++; 1717 if (opt_m == true) 1718 pr(" mbuf chain %p\n", m_rec); 1719 while (m) { 1720 nmbufs++; 1721 m = m->m_next; 1722 } 1723 m_rec = m = m_rec->m_nextpkt; 1724 } 1725 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1726 1727 pr("Recv Buffer Usage: %d [bytes]\n", sb_rcv->sb_cc); 1728 pr("Recv Buffer mbufs:\n"); 1729 m_rec = m = sb_rcv->sb_mb; 1730 nrecs = 0; 1731 nmbufs = 0; 1732 while (m_rec) { 1733 nrecs++; 1734 if (opt_m == true) 1735 pr(" mbuf chain %p\n", m_rec); 1736 while (m) { 1737 nmbufs++; 1738 m = m->m_next; 1739 } 1740 m_rec = m = m_rec->m_nextpkt; 1741 } 1742 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1743 } 1744} 1745#endif /* DDB */ 1746