uipc_socket2.c revision 1.137
1/* $NetBSD: uipc_socket2.c,v 1.137 2020/05/23 23:42:43 ad 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 1.137 2020/05/23 23:42:43 ad 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_cpid = head->so_cpid; 321 322 /* 323 * Share the lock with the listening-socket, it may get unshared 324 * once the connection is complete. 325 */ 326 mutex_obj_hold(head->so_lock); 327 so->so_lock = head->so_lock; 328 329 /* 330 * Reserve the space for socket buffers. 331 */ 332#ifdef MBUFTRACE 333 so->so_mowner = head->so_mowner; 334 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner; 335 so->so_snd.sb_mowner = head->so_snd.sb_mowner; 336#endif 337 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 338 goto out; 339 } 340 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 341 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 342 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 343 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 344 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 345 so->so_snd.sb_flags |= head->so_snd.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 346 347 /* 348 * Finally, perform the protocol attach. Note: a new socket 349 * lock may be assigned at this point (if so, it will be held). 350 */ 351 error = (*so->so_proto->pr_usrreqs->pr_attach)(so, 0); 352 if (error) { 353out: 354 KASSERT(solocked(so)); 355 KASSERT(so->so_accf == NULL); 356 soput(so); 357 358 /* Note: the listening socket shall stay locked. */ 359 KASSERT(solocked(head)); 360 return NULL; 361 } 362 KASSERT(solocked2(head, so)); 363 364 /* 365 * Insert into the queue. If ready, update the connection status 366 * and wake up any waiters, e.g. processes blocking on accept(). 367 */ 368 soqinsque(head, so, soqueue); 369 if (soready) { 370 so->so_state |= SS_ISCONNECTED; 371 sorwakeup(head); 372 cv_broadcast(&head->so_cv); 373 } 374 return so; 375} 376 377struct socket * 378soget(bool waitok) 379{ 380 struct socket *so; 381 382 so = pool_cache_get(socket_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 383 if (__predict_false(so == NULL)) 384 return (NULL); 385 memset(so, 0, sizeof(*so)); 386 TAILQ_INIT(&so->so_q0); 387 TAILQ_INIT(&so->so_q); 388 cv_init(&so->so_cv, "socket"); 389 cv_init(&so->so_rcv.sb_cv, "netio"); 390 cv_init(&so->so_snd.sb_cv, "netio"); 391 selinit(&so->so_rcv.sb_sel); 392 selinit(&so->so_snd.sb_sel); 393 so->so_rcv.sb_so = so; 394 so->so_snd.sb_so = so; 395 return so; 396} 397 398void 399soput(struct socket *so) 400{ 401 402 KASSERT(!cv_has_waiters(&so->so_cv)); 403 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 404 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 405 seldestroy(&so->so_rcv.sb_sel); 406 seldestroy(&so->so_snd.sb_sel); 407 mutex_obj_free(so->so_lock); 408 cv_destroy(&so->so_cv); 409 cv_destroy(&so->so_rcv.sb_cv); 410 cv_destroy(&so->so_snd.sb_cv); 411 pool_cache_put(socket_cache, so); 412} 413 414/* 415 * soqinsque: insert socket of a new connection into the specified 416 * accept queue of the listening socket (head). 417 * 418 * q = 0: queue of partial connections 419 * q = 1: queue of incoming connections 420 */ 421void 422soqinsque(struct socket *head, struct socket *so, int q) 423{ 424 KASSERT(q == 0 || q == 1); 425 KASSERT(solocked2(head, so)); 426 KASSERT(so->so_onq == NULL); 427 KASSERT(so->so_head == NULL); 428 429 so->so_head = head; 430 if (q == 0) { 431 head->so_q0len++; 432 so->so_onq = &head->so_q0; 433 } else { 434 head->so_qlen++; 435 so->so_onq = &head->so_q; 436 } 437 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 438} 439 440/* 441 * soqremque: remove socket from the specified queue. 442 * 443 * => Returns true if socket was removed from the specified queue. 444 * => False if socket was not removed (because it was in other queue). 445 */ 446bool 447soqremque(struct socket *so, int q) 448{ 449 struct socket *head = so->so_head; 450 451 KASSERT(q == 0 || q == 1); 452 KASSERT(solocked(so)); 453 KASSERT(so->so_onq != NULL); 454 KASSERT(head != NULL); 455 456 if (q == 0) { 457 if (so->so_onq != &head->so_q0) 458 return false; 459 head->so_q0len--; 460 } else { 461 if (so->so_onq != &head->so_q) 462 return false; 463 head->so_qlen--; 464 } 465 KASSERT(solocked2(so, head)); 466 TAILQ_REMOVE(so->so_onq, so, so_qe); 467 so->so_onq = NULL; 468 so->so_head = NULL; 469 return true; 470} 471 472/* 473 * socantsendmore: indicates that no more data will be sent on the 474 * socket; it would normally be applied to a socket when the user 475 * informs the system that no more data is to be sent, by the protocol 476 * code (in case pr_shutdown()). 477 */ 478void 479socantsendmore(struct socket *so) 480{ 481 KASSERT(solocked(so)); 482 483 so->so_state |= SS_CANTSENDMORE; 484 sowwakeup(so); 485} 486 487/* 488 * socantrcvmore(): indicates that no more data will be received and 489 * will normally be applied to the socket by a protocol when it detects 490 * that the peer will send no more data. Data queued for reading in 491 * the socket may yet be read. 492 */ 493void 494socantrcvmore(struct socket *so) 495{ 496 KASSERT(solocked(so)); 497 498 so->so_state |= SS_CANTRCVMORE; 499 sorwakeup(so); 500} 501 502/* 503 * soroverflow(): indicates that data was attempted to be sent 504 * but the receiving buffer overflowed. 505 */ 506void 507soroverflow(struct socket *so) 508{ 509 KASSERT(solocked(so)); 510 511 so->so_rcv.sb_overflowed++; 512 if (so->so_options & SO_RERROR) { 513 so->so_rerror = ENOBUFS; 514 sorwakeup(so); 515 } 516} 517 518/* 519 * Wait for data to arrive at/drain from a socket buffer. 520 */ 521int 522sbwait(struct sockbuf *sb) 523{ 524 struct socket *so; 525 kmutex_t *lock; 526 int error; 527 528 so = sb->sb_so; 529 530 KASSERT(solocked(so)); 531 532 sb->sb_flags |= SB_NOTIFY; 533 lock = so->so_lock; 534 if ((sb->sb_flags & SB_NOINTR) != 0) 535 error = cv_timedwait(&sb->sb_cv, lock, sb->sb_timeo); 536 else 537 error = cv_timedwait_sig(&sb->sb_cv, lock, sb->sb_timeo); 538 if (__predict_false(lock != so->so_lock)) 539 solockretry(so, lock); 540 return error; 541} 542 543/* 544 * Wakeup processes waiting on a socket buffer. 545 * Do asynchronous notification via SIGIO 546 * if the socket buffer has the SB_ASYNC flag set. 547 */ 548void 549sowakeup(struct socket *so, struct sockbuf *sb, int code) 550{ 551 int band; 552 553 KASSERT(solocked(so)); 554 KASSERT(sb->sb_so == so); 555 556 if (code == POLL_IN) 557 band = POLLIN|POLLRDNORM; 558 else 559 band = POLLOUT|POLLWRNORM; 560 sb->sb_flags &= ~SB_NOTIFY; 561 selnotify(&sb->sb_sel, band, NOTE_SUBMIT); 562 cv_broadcast(&sb->sb_cv); 563 if (sb->sb_flags & SB_ASYNC) 564 fownsignal(so->so_pgid, SIGIO, code, band, so); 565 if (sb->sb_flags & SB_UPCALL) 566 (*so->so_upcall)(so, so->so_upcallarg, band, M_DONTWAIT); 567} 568 569/* 570 * Reset a socket's lock pointer. Wake all threads waiting on the 571 * socket's condition variables so that they can restart their waits 572 * using the new lock. The existing lock must be held. 573 */ 574void 575solockreset(struct socket *so, kmutex_t *lock) 576{ 577 578 KASSERT(solocked(so)); 579 580 so->so_lock = lock; 581 cv_broadcast(&so->so_snd.sb_cv); 582 cv_broadcast(&so->so_rcv.sb_cv); 583 cv_broadcast(&so->so_cv); 584} 585 586/* 587 * Socket buffer (struct sockbuf) utility routines. 588 * 589 * Each socket contains two socket buffers: one for sending data and 590 * one for receiving data. Each buffer contains a queue of mbufs, 591 * information about the number of mbufs and amount of data in the 592 * queue, and other fields allowing poll() statements and notification 593 * on data availability to be implemented. 594 * 595 * Data stored in a socket buffer is maintained as a list of records. 596 * Each record is a list of mbufs chained together with the m_next 597 * field. Records are chained together with the m_nextpkt field. The upper 598 * level routine soreceive() expects the following conventions to be 599 * observed when placing information in the receive buffer: 600 * 601 * 1. If the protocol requires each message be preceded by the sender's 602 * name, then a record containing that name must be present before 603 * any associated data (mbuf's must be of type MT_SONAME). 604 * 2. If the protocol supports the exchange of ``access rights'' (really 605 * just additional data associated with the message), and there are 606 * ``rights'' to be received, then a record containing this data 607 * should be present (mbuf's must be of type MT_CONTROL). 608 * 3. If a name or rights record exists, then it must be followed by 609 * a data record, perhaps of zero length. 610 * 611 * Before using a new socket structure it is first necessary to reserve 612 * buffer space to the socket, by calling sbreserve(). This should commit 613 * some of the available buffer space in the system buffer pool for the 614 * socket (currently, it does nothing but enforce limits). The space 615 * should be released by calling sbrelease() when the socket is destroyed. 616 */ 617 618int 619sb_max_set(u_long new_sbmax) 620{ 621 int s; 622 623 if (new_sbmax < (16 * 1024)) 624 return (EINVAL); 625 626 s = splsoftnet(); 627 sb_max = new_sbmax; 628 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES); 629 splx(s); 630 631 return (0); 632} 633 634int 635soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 636{ 637 KASSERT(so->so_pcb == NULL || solocked(so)); 638 639 /* 640 * there's at least one application (a configure script of screen) 641 * which expects a fifo is writable even if it has "some" bytes 642 * in its buffer. 643 * so we want to make sure (hiwat - lowat) >= (some bytes). 644 * 645 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above. 646 * we expect it's large enough for such applications. 647 */ 648 u_long lowat = MAX(sock_loan_thresh, MCLBYTES); 649 u_long hiwat = lowat + PIPE_BUF; 650 651 if (sndcc < hiwat) 652 sndcc = hiwat; 653 if (sbreserve(&so->so_snd, sndcc, so) == 0) 654 goto bad; 655 if (sbreserve(&so->so_rcv, rcvcc, so) == 0) 656 goto bad2; 657 if (so->so_rcv.sb_lowat == 0) 658 so->so_rcv.sb_lowat = 1; 659 if (so->so_snd.sb_lowat == 0) 660 so->so_snd.sb_lowat = lowat; 661 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 662 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 663 return (0); 664 bad2: 665 sbrelease(&so->so_snd, so); 666 bad: 667 return (ENOBUFS); 668} 669 670/* 671 * Allot mbufs to a sockbuf. 672 * Attempt to scale mbmax so that mbcnt doesn't become limiting 673 * if buffering efficiency is near the normal case. 674 */ 675int 676sbreserve(struct sockbuf *sb, u_long cc, struct socket *so) 677{ 678 struct lwp *l = curlwp; /* XXX */ 679 rlim_t maxcc; 680 struct uidinfo *uidinfo; 681 682 KASSERT(so->so_pcb == NULL || solocked(so)); 683 KASSERT(sb->sb_so == so); 684 KASSERT(sb_max_adj != 0); 685 686 if (cc == 0 || cc > sb_max_adj) 687 return (0); 688 689 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur; 690 691 uidinfo = so->so_uidinfo; 692 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc)) 693 return 0; 694 sb->sb_mbmax = uimin(cc * 2, sb_max); 695 if (sb->sb_lowat > sb->sb_hiwat) 696 sb->sb_lowat = sb->sb_hiwat; 697 698 return (1); 699} 700 701/* 702 * Free mbufs held by a socket, and reserved mbuf space. We do not assert 703 * that the socket is held locked here: see sorflush(). 704 */ 705void 706sbrelease(struct sockbuf *sb, struct socket *so) 707{ 708 709 KASSERT(sb->sb_so == so); 710 711 sbflush(sb); 712 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); 713 sb->sb_mbmax = 0; 714} 715 716/* 717 * Routines to add and remove 718 * data from an mbuf queue. 719 * 720 * The routines sbappend() or sbappendrecord() are normally called to 721 * append new mbufs to a socket buffer, after checking that adequate 722 * space is available, comparing the function sbspace() with the amount 723 * of data to be added. sbappendrecord() differs from sbappend() in 724 * that data supplied is treated as the beginning of a new record. 725 * To place a sender's address, optional access rights, and data in a 726 * socket receive buffer, sbappendaddr() should be used. To place 727 * access rights and data in a socket receive buffer, sbappendrights() 728 * should be used. In either case, the new data begins a new record. 729 * Note that unlike sbappend() and sbappendrecord(), these routines check 730 * for the caller that there will be enough space to store the data. 731 * Each fails if there is not enough space, or if it cannot find mbufs 732 * to store additional information in. 733 * 734 * Reliable protocols may use the socket send buffer to hold data 735 * awaiting acknowledgement. Data is normally copied from a socket 736 * send buffer in a protocol with m_copym for output to a peer, 737 * and then removing the data from the socket buffer with sbdrop() 738 * or sbdroprecord() when the data is acknowledged by the peer. 739 */ 740 741#ifdef SOCKBUF_DEBUG 742void 743sblastrecordchk(struct sockbuf *sb, const char *where) 744{ 745 struct mbuf *m = sb->sb_mb; 746 747 KASSERT(solocked(sb->sb_so)); 748 749 while (m && m->m_nextpkt) 750 m = m->m_nextpkt; 751 752 if (m != sb->sb_lastrecord) { 753 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 754 sb->sb_mb, sb->sb_lastrecord, m); 755 printf("packet chain:\n"); 756 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 757 printf("\t%p\n", m); 758 panic("sblastrecordchk from %s", where); 759 } 760} 761 762void 763sblastmbufchk(struct sockbuf *sb, const char *where) 764{ 765 struct mbuf *m = sb->sb_mb; 766 struct mbuf *n; 767 768 KASSERT(solocked(sb->sb_so)); 769 770 while (m && m->m_nextpkt) 771 m = m->m_nextpkt; 772 773 while (m && m->m_next) 774 m = m->m_next; 775 776 if (m != sb->sb_mbtail) { 777 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 778 sb->sb_mb, sb->sb_mbtail, m); 779 printf("packet tree:\n"); 780 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 781 printf("\t"); 782 for (n = m; n != NULL; n = n->m_next) 783 printf("%p ", n); 784 printf("\n"); 785 } 786 panic("sblastmbufchk from %s", where); 787 } 788} 789#endif /* SOCKBUF_DEBUG */ 790 791/* 792 * Link a chain of records onto a socket buffer 793 */ 794#define SBLINKRECORDCHAIN(sb, m0, mlast) \ 795do { \ 796 if ((sb)->sb_lastrecord != NULL) \ 797 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 798 else \ 799 (sb)->sb_mb = (m0); \ 800 (sb)->sb_lastrecord = (mlast); \ 801} while (/*CONSTCOND*/0) 802 803 804#define SBLINKRECORD(sb, m0) \ 805 SBLINKRECORDCHAIN(sb, m0, m0) 806 807/* 808 * Append mbuf chain m to the last record in the 809 * socket buffer sb. The additional space associated 810 * the mbuf chain is recorded in sb. Empty mbufs are 811 * discarded and mbufs are compacted where possible. 812 */ 813void 814sbappend(struct sockbuf *sb, struct mbuf *m) 815{ 816 struct mbuf *n; 817 818 KASSERT(solocked(sb->sb_so)); 819 820 if (m == NULL) 821 return; 822 823#ifdef MBUFTRACE 824 m_claimm(m, sb->sb_mowner); 825#endif 826 827 SBLASTRECORDCHK(sb, "sbappend 1"); 828 829 if ((n = sb->sb_lastrecord) != NULL) { 830 /* 831 * XXX Would like to simply use sb_mbtail here, but 832 * XXX I need to verify that I won't miss an EOR that 833 * XXX way. 834 */ 835 do { 836 if (n->m_flags & M_EOR) { 837 sbappendrecord(sb, m); /* XXXXXX!!!! */ 838 return; 839 } 840 } while (n->m_next && (n = n->m_next)); 841 } else { 842 /* 843 * If this is the first record in the socket buffer, it's 844 * also the last record. 845 */ 846 sb->sb_lastrecord = m; 847 } 848 sbcompress(sb, m, n); 849 SBLASTRECORDCHK(sb, "sbappend 2"); 850} 851 852/* 853 * This version of sbappend() should only be used when the caller 854 * absolutely knows that there will never be more than one record 855 * in the socket buffer, that is, a stream protocol (such as TCP). 856 */ 857void 858sbappendstream(struct sockbuf *sb, struct mbuf *m) 859{ 860 861 KASSERT(solocked(sb->sb_so)); 862 KDASSERT(m->m_nextpkt == NULL); 863 KASSERT(sb->sb_mb == sb->sb_lastrecord); 864 865 SBLASTMBUFCHK(sb, __func__); 866 867#ifdef MBUFTRACE 868 m_claimm(m, sb->sb_mowner); 869#endif 870 871 sbcompress(sb, m, sb->sb_mbtail); 872 873 sb->sb_lastrecord = sb->sb_mb; 874 SBLASTRECORDCHK(sb, __func__); 875} 876 877#ifdef SOCKBUF_DEBUG 878void 879sbcheck(struct sockbuf *sb) 880{ 881 struct mbuf *m, *m2; 882 u_long len, mbcnt; 883 884 KASSERT(solocked(sb->sb_so)); 885 886 len = 0; 887 mbcnt = 0; 888 for (m = sb->sb_mb; m; m = m->m_nextpkt) { 889 for (m2 = m; m2 != NULL; m2 = m2->m_next) { 890 len += m2->m_len; 891 mbcnt += MSIZE; 892 if (m2->m_flags & M_EXT) 893 mbcnt += m2->m_ext.ext_size; 894 if (m2->m_nextpkt != NULL) 895 panic("sbcheck nextpkt"); 896 } 897 } 898 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 899 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 900 mbcnt, sb->sb_mbcnt); 901 panic("sbcheck"); 902 } 903} 904#endif 905 906/* 907 * As above, except the mbuf chain 908 * begins a new record. 909 */ 910void 911sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 912{ 913 struct mbuf *m; 914 915 KASSERT(solocked(sb->sb_so)); 916 917 if (m0 == NULL) 918 return; 919 920#ifdef MBUFTRACE 921 m_claimm(m0, sb->sb_mowner); 922#endif 923 /* 924 * Put the first mbuf on the queue. 925 * Note this permits zero length records. 926 */ 927 sballoc(sb, m0); 928 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 929 SBLINKRECORD(sb, m0); 930 m = m0->m_next; 931 m0->m_next = 0; 932 if (m && (m0->m_flags & M_EOR)) { 933 m0->m_flags &= ~M_EOR; 934 m->m_flags |= M_EOR; 935 } 936 sbcompress(sb, m, m0); 937 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 938} 939 940/* 941 * As above except that OOB data 942 * is inserted at the beginning of the sockbuf, 943 * but after any other OOB data. 944 */ 945void 946sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 947{ 948 struct mbuf *m, **mp; 949 950 KASSERT(solocked(sb->sb_so)); 951 952 if (m0 == NULL) 953 return; 954 955 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 956 957 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 958 again: 959 switch (m->m_type) { 960 961 case MT_OOBDATA: 962 continue; /* WANT next train */ 963 964 case MT_CONTROL: 965 if ((m = m->m_next) != NULL) 966 goto again; /* inspect THIS train further */ 967 } 968 break; 969 } 970 /* 971 * Put the first mbuf on the queue. 972 * Note this permits zero length records. 973 */ 974 sballoc(sb, m0); 975 m0->m_nextpkt = *mp; 976 if (*mp == NULL) { 977 /* m0 is actually the new tail */ 978 sb->sb_lastrecord = m0; 979 } 980 *mp = m0; 981 m = m0->m_next; 982 m0->m_next = 0; 983 if (m && (m0->m_flags & M_EOR)) { 984 m0->m_flags &= ~M_EOR; 985 m->m_flags |= M_EOR; 986 } 987 sbcompress(sb, m, m0); 988 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 989} 990 991/* 992 * Append address and data, and optionally, control (ancillary) data 993 * to the receive queue of a socket. If present, 994 * m0 must include a packet header with total length. 995 * Returns 0 if no space in sockbuf or insufficient mbufs. 996 */ 997int 998sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, 999 struct mbuf *control) 1000{ 1001 struct mbuf *m, *n, *nlast; 1002 int space, len; 1003 1004 KASSERT(solocked(sb->sb_so)); 1005 1006 space = asa->sa_len; 1007 1008 if (m0 != NULL) { 1009 if ((m0->m_flags & M_PKTHDR) == 0) 1010 panic("sbappendaddr"); 1011 space += m0->m_pkthdr.len; 1012#ifdef MBUFTRACE 1013 m_claimm(m0, sb->sb_mowner); 1014#endif 1015 } 1016 for (n = control; n; n = n->m_next) { 1017 space += n->m_len; 1018 MCLAIM(n, sb->sb_mowner); 1019 if (n->m_next == NULL) /* keep pointer to last control buf */ 1020 break; 1021 } 1022 if (space > sbspace(sb)) 1023 return (0); 1024 m = m_get(M_DONTWAIT, MT_SONAME); 1025 if (m == NULL) 1026 return (0); 1027 MCLAIM(m, sb->sb_mowner); 1028 /* 1029 * XXX avoid 'comparison always true' warning which isn't easily 1030 * avoided. 1031 */ 1032 len = asa->sa_len; 1033 if (len > MLEN) { 1034 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 1035 if ((m->m_flags & M_EXT) == 0) { 1036 m_free(m); 1037 return (0); 1038 } 1039 } 1040 m->m_len = asa->sa_len; 1041 memcpy(mtod(m, void *), asa, asa->sa_len); 1042 if (n) 1043 n->m_next = m0; /* concatenate data to control */ 1044 else 1045 control = m0; 1046 m->m_next = control; 1047 1048 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 1049 1050 for (n = m; n->m_next != NULL; n = n->m_next) 1051 sballoc(sb, n); 1052 sballoc(sb, n); 1053 nlast = n; 1054 SBLINKRECORD(sb, m); 1055 1056 sb->sb_mbtail = nlast; 1057 SBLASTMBUFCHK(sb, "sbappendaddr"); 1058 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 1059 1060 return (1); 1061} 1062 1063/* 1064 * Helper for sbappendchainaddr: prepend a struct sockaddr* to 1065 * an mbuf chain. 1066 */ 1067static inline struct mbuf * 1068m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0, 1069 const struct sockaddr *asa) 1070{ 1071 struct mbuf *m; 1072 const int salen = asa->sa_len; 1073 1074 KASSERT(solocked(sb->sb_so)); 1075 1076 /* only the first in each chain need be a pkthdr */ 1077 m = m_gethdr(M_DONTWAIT, MT_SONAME); 1078 if (m == NULL) 1079 return NULL; 1080 MCLAIM(m, sb->sb_mowner); 1081#ifdef notyet 1082 if (salen > MHLEN) { 1083 MEXTMALLOC(m, salen, M_NOWAIT); 1084 if ((m->m_flags & M_EXT) == 0) { 1085 m_free(m); 1086 return NULL; 1087 } 1088 } 1089#else 1090 KASSERT(salen <= MHLEN); 1091#endif 1092 m->m_len = salen; 1093 memcpy(mtod(m, void *), asa, salen); 1094 m->m_next = m0; 1095 m->m_pkthdr.len = salen + m0->m_pkthdr.len; 1096 1097 return m; 1098} 1099 1100int 1101sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa, 1102 struct mbuf *m0, int sbprio) 1103{ 1104 struct mbuf *m, *n, *n0, *nlast; 1105 int error; 1106 1107 KASSERT(solocked(sb->sb_so)); 1108 1109 /* 1110 * XXX sbprio reserved for encoding priority of this* request: 1111 * SB_PRIO_NONE --> honour normal sb limits 1112 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space, 1113 * take whole chain. Intended for large requests 1114 * that should be delivered atomically (all, or none). 1115 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow 1116 * over normal socket limits, for messages indicating 1117 * buffer overflow in earlier normal/lower-priority messages 1118 * SB_PRIO_BESTEFFORT --> ignore limits entirely. 1119 * Intended for kernel-generated messages only. 1120 * Up to generator to avoid total mbuf resource exhaustion. 1121 */ 1122 (void)sbprio; 1123 1124 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1125 panic("sbappendaddrchain"); 1126 1127#ifdef notyet 1128 space = sbspace(sb); 1129 1130 /* 1131 * Enforce SB_PRIO_* limits as described above. 1132 */ 1133#endif 1134 1135 n0 = NULL; 1136 nlast = NULL; 1137 for (m = m0; m; m = m->m_nextpkt) { 1138 struct mbuf *np; 1139 1140#ifdef MBUFTRACE 1141 m_claimm(m, sb->sb_mowner); 1142#endif 1143 1144 /* Prepend sockaddr to this record (m) of input chain m0 */ 1145 n = m_prepend_sockaddr(sb, m, asa); 1146 if (n == NULL) { 1147 error = ENOBUFS; 1148 goto bad; 1149 } 1150 1151 /* Append record (asa+m) to end of new chain n0 */ 1152 if (n0 == NULL) { 1153 n0 = n; 1154 } else { 1155 nlast->m_nextpkt = n; 1156 } 1157 /* Keep track of last record on new chain */ 1158 nlast = n; 1159 1160 for (np = n; np; np = np->m_next) 1161 sballoc(sb, np); 1162 } 1163 1164 SBLASTRECORDCHK(sb, "sbappendaddrchain 1"); 1165 1166 /* Drop the entire chain of (asa+m) records onto the socket */ 1167 SBLINKRECORDCHAIN(sb, n0, nlast); 1168 1169 SBLASTRECORDCHK(sb, "sbappendaddrchain 2"); 1170 1171 for (m = nlast; m->m_next; m = m->m_next) 1172 ; 1173 sb->sb_mbtail = m; 1174 SBLASTMBUFCHK(sb, "sbappendaddrchain"); 1175 1176 return (1); 1177 1178bad: 1179 /* 1180 * On error, free the prepended addreseses. For consistency 1181 * with sbappendaddr(), leave it to our caller to free 1182 * the input record chain passed to us as m0. 1183 */ 1184 while ((n = n0) != NULL) { 1185 struct mbuf *np; 1186 1187 /* Undo the sballoc() of this record */ 1188 for (np = n; np; np = np->m_next) 1189 sbfree(sb, np); 1190 1191 n0 = n->m_nextpkt; /* iterate at next prepended address */ 1192 np = m_free(n); /* free prepended address (not data) */ 1193 } 1194 return error; 1195} 1196 1197 1198int 1199sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 1200{ 1201 struct mbuf *m, *mlast, *n; 1202 int space; 1203 1204 KASSERT(solocked(sb->sb_so)); 1205 1206 space = 0; 1207 if (control == NULL) 1208 panic("sbappendcontrol"); 1209 for (m = control; ; m = m->m_next) { 1210 space += m->m_len; 1211 MCLAIM(m, sb->sb_mowner); 1212 if (m->m_next == NULL) 1213 break; 1214 } 1215 n = m; /* save pointer to last control buffer */ 1216 for (m = m0; m; m = m->m_next) { 1217 MCLAIM(m, sb->sb_mowner); 1218 space += m->m_len; 1219 } 1220 if (space > sbspace(sb)) 1221 return (0); 1222 n->m_next = m0; /* concatenate data to control */ 1223 1224 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 1225 1226 for (m = control; m->m_next != NULL; m = m->m_next) 1227 sballoc(sb, m); 1228 sballoc(sb, m); 1229 mlast = m; 1230 SBLINKRECORD(sb, control); 1231 1232 sb->sb_mbtail = mlast; 1233 SBLASTMBUFCHK(sb, "sbappendcontrol"); 1234 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 1235 1236 return (1); 1237} 1238 1239/* 1240 * Compress mbuf chain m into the socket 1241 * buffer sb following mbuf n. If n 1242 * is null, the buffer is presumed empty. 1243 */ 1244void 1245sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1246{ 1247 int eor; 1248 struct mbuf *o; 1249 1250 KASSERT(solocked(sb->sb_so)); 1251 1252 eor = 0; 1253 while (m) { 1254 eor |= m->m_flags & M_EOR; 1255 if (m->m_len == 0 && 1256 (eor == 0 || 1257 (((o = m->m_next) || (o = n)) && 1258 o->m_type == m->m_type))) { 1259 if (sb->sb_lastrecord == m) 1260 sb->sb_lastrecord = m->m_next; 1261 m = m_free(m); 1262 continue; 1263 } 1264 if (n && (n->m_flags & M_EOR) == 0 && 1265 /* M_TRAILINGSPACE() checks buffer writeability */ 1266 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 1267 m->m_len <= M_TRAILINGSPACE(n) && 1268 n->m_type == m->m_type) { 1269 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *), 1270 (unsigned)m->m_len); 1271 n->m_len += m->m_len; 1272 sb->sb_cc += m->m_len; 1273 m = m_free(m); 1274 continue; 1275 } 1276 if (n) 1277 n->m_next = m; 1278 else 1279 sb->sb_mb = m; 1280 sb->sb_mbtail = m; 1281 sballoc(sb, m); 1282 n = m; 1283 m->m_flags &= ~M_EOR; 1284 m = m->m_next; 1285 n->m_next = 0; 1286 } 1287 if (eor) { 1288 if (n) 1289 n->m_flags |= eor; 1290 else 1291 printf("semi-panic: sbcompress\n"); 1292 } 1293 SBLASTMBUFCHK(sb, __func__); 1294} 1295 1296/* 1297 * Free all mbufs in a sockbuf. 1298 * Check that all resources are reclaimed. 1299 */ 1300void 1301sbflush(struct sockbuf *sb) 1302{ 1303 1304 KASSERT(solocked(sb->sb_so)); 1305 KASSERT((sb->sb_flags & SB_LOCK) == 0); 1306 1307 while (sb->sb_mbcnt) 1308 sbdrop(sb, (int)sb->sb_cc); 1309 1310 KASSERT(sb->sb_cc == 0); 1311 KASSERT(sb->sb_mb == NULL); 1312 KASSERT(sb->sb_mbtail == NULL); 1313 KASSERT(sb->sb_lastrecord == NULL); 1314} 1315 1316/* 1317 * Drop data from (the front of) a sockbuf. 1318 */ 1319void 1320sbdrop(struct sockbuf *sb, int len) 1321{ 1322 struct mbuf *m, *next; 1323 1324 KASSERT(solocked(sb->sb_so)); 1325 1326 next = (m = sb->sb_mb) ? m->m_nextpkt : NULL; 1327 while (len > 0) { 1328 if (m == NULL) { 1329 if (next == NULL) 1330 panic("sbdrop(%p,%d): cc=%lu", 1331 sb, len, sb->sb_cc); 1332 m = next; 1333 next = m->m_nextpkt; 1334 continue; 1335 } 1336 if (m->m_len > len) { 1337 m->m_len -= len; 1338 m->m_data += len; 1339 sb->sb_cc -= len; 1340 break; 1341 } 1342 len -= m->m_len; 1343 sbfree(sb, m); 1344 m = m_free(m); 1345 } 1346 while (m && m->m_len == 0) { 1347 sbfree(sb, m); 1348 m = m_free(m); 1349 } 1350 if (m) { 1351 sb->sb_mb = m; 1352 m->m_nextpkt = next; 1353 } else 1354 sb->sb_mb = next; 1355 /* 1356 * First part is an inline SB_EMPTY_FIXUP(). Second part 1357 * makes sure sb_lastrecord is up-to-date if we dropped 1358 * part of the last record. 1359 */ 1360 m = sb->sb_mb; 1361 if (m == NULL) { 1362 sb->sb_mbtail = NULL; 1363 sb->sb_lastrecord = NULL; 1364 } else if (m->m_nextpkt == NULL) 1365 sb->sb_lastrecord = m; 1366} 1367 1368/* 1369 * Drop a record off the front of a sockbuf 1370 * and move the next record to the front. 1371 */ 1372void 1373sbdroprecord(struct sockbuf *sb) 1374{ 1375 struct mbuf *m, *mn; 1376 1377 KASSERT(solocked(sb->sb_so)); 1378 1379 m = sb->sb_mb; 1380 if (m) { 1381 sb->sb_mb = m->m_nextpkt; 1382 do { 1383 sbfree(sb, m); 1384 mn = m_free(m); 1385 } while ((m = mn) != NULL); 1386 } 1387 SB_EMPTY_FIXUP(sb); 1388} 1389 1390/* 1391 * Create a "control" mbuf containing the specified data 1392 * with the specified type for presentation on a socket buffer. 1393 */ 1394struct mbuf * 1395sbcreatecontrol1(void **p, int size, int type, int level, int flags) 1396{ 1397 struct cmsghdr *cp; 1398 struct mbuf *m; 1399 int space = CMSG_SPACE(size); 1400 1401 if ((flags & M_DONTWAIT) && space > MCLBYTES) { 1402 printf("%s: message too large %d\n", __func__, space); 1403 return NULL; 1404 } 1405 1406 if ((m = m_get(flags, MT_CONTROL)) == NULL) 1407 return NULL; 1408 if (space > MLEN) { 1409 if (space > MCLBYTES) 1410 MEXTMALLOC(m, space, M_WAITOK); 1411 else 1412 MCLGET(m, flags); 1413 if ((m->m_flags & M_EXT) == 0) { 1414 m_free(m); 1415 return NULL; 1416 } 1417 } 1418 cp = mtod(m, struct cmsghdr *); 1419 *p = CMSG_DATA(cp); 1420 m->m_len = space; 1421 cp->cmsg_len = CMSG_LEN(size); 1422 cp->cmsg_level = level; 1423 cp->cmsg_type = type; 1424 1425 memset(cp + 1, 0, CMSG_LEN(0) - sizeof(*cp)); 1426 memset((uint8_t *)*p + size, 0, CMSG_ALIGN(size) - size); 1427 1428 return m; 1429} 1430 1431struct mbuf * 1432sbcreatecontrol(void *p, int size, int type, int level) 1433{ 1434 struct mbuf *m; 1435 void *v; 1436 1437 m = sbcreatecontrol1(&v, size, type, level, M_DONTWAIT); 1438 if (m == NULL) 1439 return NULL; 1440 memcpy(v, p, size); 1441 return m; 1442} 1443 1444void 1445solockretry(struct socket *so, kmutex_t *lock) 1446{ 1447 1448 while (lock != so->so_lock) { 1449 mutex_exit(lock); 1450 lock = so->so_lock; 1451 mutex_enter(lock); 1452 } 1453} 1454 1455bool 1456solocked(const struct socket *so) 1457{ 1458 1459 return mutex_owned(so->so_lock); 1460} 1461 1462bool 1463solocked2(const struct socket *so1, const struct socket *so2) 1464{ 1465 const kmutex_t *lock; 1466 1467 lock = so1->so_lock; 1468 if (lock != so2->so_lock) 1469 return false; 1470 return mutex_owned(lock); 1471} 1472 1473/* 1474 * sosetlock: assign a default lock to a new socket. 1475 */ 1476void 1477sosetlock(struct socket *so) 1478{ 1479 if (so->so_lock == NULL) { 1480 kmutex_t *lock = softnet_lock; 1481 1482 so->so_lock = lock; 1483 mutex_obj_hold(lock); 1484 mutex_enter(lock); 1485 } 1486 KASSERT(solocked(so)); 1487} 1488 1489/* 1490 * Set lock on sockbuf sb; sleep if lock is already held. 1491 * Unless SB_NOINTR is set on sockbuf, sleep is interruptible. 1492 * Returns error without lock if sleep is interrupted. 1493 */ 1494int 1495sblock(struct sockbuf *sb, int wf) 1496{ 1497 struct socket *so; 1498 kmutex_t *lock; 1499 int error; 1500 1501 KASSERT(solocked(sb->sb_so)); 1502 1503 for (;;) { 1504 if (__predict_true((sb->sb_flags & SB_LOCK) == 0)) { 1505 sb->sb_flags |= SB_LOCK; 1506 return 0; 1507 } 1508 if (wf != M_WAITOK) 1509 return EWOULDBLOCK; 1510 so = sb->sb_so; 1511 lock = so->so_lock; 1512 if ((sb->sb_flags & SB_NOINTR) != 0) { 1513 cv_wait(&so->so_cv, lock); 1514 error = 0; 1515 } else 1516 error = cv_wait_sig(&so->so_cv, lock); 1517 if (__predict_false(lock != so->so_lock)) 1518 solockretry(so, lock); 1519 if (error != 0) 1520 return error; 1521 } 1522} 1523 1524void 1525sbunlock(struct sockbuf *sb) 1526{ 1527 struct socket *so; 1528 1529 so = sb->sb_so; 1530 1531 KASSERT(solocked(so)); 1532 KASSERT((sb->sb_flags & SB_LOCK) != 0); 1533 1534 sb->sb_flags &= ~SB_LOCK; 1535 cv_broadcast(&so->so_cv); 1536} 1537 1538int 1539sowait(struct socket *so, bool catch_p, int timo) 1540{ 1541 kmutex_t *lock; 1542 int error; 1543 1544 KASSERT(solocked(so)); 1545 KASSERT(catch_p || timo != 0); 1546 1547 lock = so->so_lock; 1548 if (catch_p) 1549 error = cv_timedwait_sig(&so->so_cv, lock, timo); 1550 else 1551 error = cv_timedwait(&so->so_cv, lock, timo); 1552 if (__predict_false(lock != so->so_lock)) 1553 solockretry(so, lock); 1554 return error; 1555} 1556 1557#ifdef DDB 1558 1559/* 1560 * Currently, sofindproc() is used only from DDB. It could be used from others 1561 * by using db_mutex_enter() 1562 */ 1563 1564static inline int 1565db_mutex_enter(kmutex_t *mtx) 1566{ 1567 extern int db_active; 1568 int rv; 1569 1570 if (!db_active) { 1571 mutex_enter(mtx); 1572 rv = 1; 1573 } else 1574 rv = mutex_tryenter(mtx); 1575 1576 return rv; 1577} 1578 1579int 1580sofindproc(struct socket *so, int all, void (*pr)(const char *, ...)) 1581{ 1582 proc_t *p; 1583 filedesc_t *fdp; 1584 fdtab_t *dt; 1585 fdfile_t *ff; 1586 file_t *fp = NULL; 1587 int found = 0; 1588 int i, t; 1589 1590 if (so == NULL) 1591 return 0; 1592 1593 t = db_mutex_enter(&proc_lock); 1594 if (!t) { 1595 pr("could not acquire proc_lock mutex\n"); 1596 return 0; 1597 } 1598 PROCLIST_FOREACH(p, &allproc) { 1599 if (p->p_stat == SIDL) 1600 continue; 1601 fdp = p->p_fd; 1602 t = db_mutex_enter(&fdp->fd_lock); 1603 if (!t) { 1604 pr("could not acquire fd_lock mutex\n"); 1605 continue; 1606 } 1607 dt = atomic_load_consume(&fdp->fd_dt); 1608 for (i = 0; i < dt->dt_nfiles; i++) { 1609 ff = dt->dt_ff[i]; 1610 if (ff == NULL) 1611 continue; 1612 1613 fp = atomic_load_consume(&ff->ff_file); 1614 if (fp == NULL) 1615 continue; 1616 1617 t = db_mutex_enter(&fp->f_lock); 1618 if (!t) { 1619 pr("could not acquire f_lock mutex\n"); 1620 continue; 1621 } 1622 if ((struct socket *)fp->f_data != so) { 1623 mutex_exit(&fp->f_lock); 1624 continue; 1625 } 1626 found++; 1627 if (pr) 1628 pr("socket %p: owner %s(pid=%d)\n", 1629 so, p->p_comm, p->p_pid); 1630 mutex_exit(&fp->f_lock); 1631 if (all == 0) 1632 break; 1633 } 1634 mutex_exit(&fdp->fd_lock); 1635 if (all == 0 && found != 0) 1636 break; 1637 } 1638 mutex_exit(&proc_lock); 1639 1640 return found; 1641} 1642 1643void 1644socket_print(const char *modif, void (*pr)(const char *, ...)) 1645{ 1646 file_t *fp; 1647 struct socket *so; 1648 struct sockbuf *sb_snd, *sb_rcv; 1649 struct mbuf *m_rec, *m; 1650 bool opt_v = false; 1651 bool opt_m = false; 1652 bool opt_a = false; 1653 bool opt_p = false; 1654 int nrecs, nmbufs; 1655 char ch; 1656 const char *family; 1657 1658 while ( (ch = *(modif++)) != '\0') { 1659 switch (ch) { 1660 case 'v': 1661 opt_v = true; 1662 break; 1663 case 'm': 1664 opt_m = true; 1665 break; 1666 case 'a': 1667 opt_a = true; 1668 break; 1669 case 'p': 1670 opt_p = true; 1671 break; 1672 } 1673 } 1674 if (opt_v == false && pr) 1675 (pr)("Ignore empty sockets. use /v to print all.\n"); 1676 if (opt_p == true && pr) 1677 (pr)("Don't search owner process.\n"); 1678 1679 LIST_FOREACH(fp, &filehead, f_list) { 1680 if (fp->f_type != DTYPE_SOCKET) 1681 continue; 1682 so = (struct socket *)fp->f_data; 1683 if (so == NULL) 1684 continue; 1685 1686 if (so->so_proto->pr_domain->dom_family == AF_INET) 1687 family = "INET"; 1688#ifdef INET6 1689 else if (so->so_proto->pr_domain->dom_family == AF_INET6) 1690 family = "INET6"; 1691#endif 1692 else if (so->so_proto->pr_domain->dom_family == pseudo_AF_KEY) 1693 family = "KEY"; 1694 else if (so->so_proto->pr_domain->dom_family == AF_ROUTE) 1695 family = "ROUTE"; 1696 else 1697 continue; 1698 1699 sb_snd = &so->so_snd; 1700 sb_rcv = &so->so_rcv; 1701 1702 if (opt_v != true && 1703 sb_snd->sb_cc == 0 && sb_rcv->sb_cc == 0) 1704 continue; 1705 1706 pr("---SOCKET %p: type %s\n", so, family); 1707 if (opt_p != true) 1708 sofindproc(so, opt_a == true ? 1 : 0, pr); 1709 pr("Send Buffer Bytes: %d [bytes]\n", sb_snd->sb_cc); 1710 pr("Send Buffer mbufs:\n"); 1711 m_rec = m = sb_snd->sb_mb; 1712 nrecs = 0; 1713 nmbufs = 0; 1714 while (m_rec) { 1715 nrecs++; 1716 if (opt_m == true) 1717 pr(" mbuf chain %p\n", m_rec); 1718 while (m) { 1719 nmbufs++; 1720 m = m->m_next; 1721 } 1722 m_rec = m = m_rec->m_nextpkt; 1723 } 1724 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1725 1726 pr("Recv Buffer Usage: %d [bytes]\n", sb_rcv->sb_cc); 1727 pr("Recv Buffer mbufs:\n"); 1728 m_rec = m = sb_rcv->sb_mb; 1729 nrecs = 0; 1730 nmbufs = 0; 1731 while (m_rec) { 1732 nrecs++; 1733 if (opt_m == true) 1734 pr(" mbuf chain %p\n", m_rec); 1735 while (m) { 1736 nmbufs++; 1737 m = m->m_next; 1738 } 1739 m_rec = m = m_rec->m_nextpkt; 1740 } 1741 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1742 } 1743} 1744#endif /* DDB */ 1745