1 /* $NetBSD: uipc_socket2.c,v 1.141 2022/04/09 23:52:23 riastradh 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 1.141 2022/04/09 23:52:23 riastradh Exp $"); 62 63 #ifdef _KERNEL_OPT 64 #include "opt_ddb.h" 65 #include "opt_inet.h" 66 #include "opt_mbuftrace.h" 67 #include "opt_sb_max.h" 68 #endif 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/proc.h> 73 #include <sys/file.h> 74 #include <sys/buf.h> 75 #include <sys/mbuf.h> 76 #include <sys/protosw.h> 77 #include <sys/domain.h> 78 #include <sys/poll.h> 79 #include <sys/socket.h> 80 #include <sys/socketvar.h> 81 #include <sys/signalvar.h> 82 #include <sys/kauth.h> 83 #include <sys/pool.h> 84 #include <sys/uidinfo.h> 85 86 #ifdef DDB 87 #include <sys/filedesc.h> 88 #endif 89 90 /* 91 * Primitive routines for operating on sockets and socket buffers. 92 * 93 * Connection life-cycle: 94 * 95 * Normal sequence from the active (originating) side: 96 * 97 * - soisconnecting() is called during processing of connect() call, 98 * - resulting in an eventual call to soisconnected() if/when the 99 * connection is established. 100 * 101 * When the connection is torn down during processing of disconnect(): 102 * 103 * - soisdisconnecting() is called and, 104 * - soisdisconnected() is called when the connection to the peer 105 * is totally severed. 106 * 107 * The semantics of these routines are such that connectionless protocols 108 * can call soisconnected() and soisdisconnected() only, bypassing the 109 * in-progress calls when setting up a ``connection'' takes no time. 110 * 111 * From the passive side, a socket is created with two queues of sockets: 112 * 113 * - so_q0 (0) for partial connections (i.e. connections in progress) 114 * - so_q (1) for connections already made and awaiting user acceptance. 115 * 116 * As a protocol is preparing incoming connections, it creates a socket 117 * structure queued on so_q0 by calling sonewconn(). When the connection 118 * is established, soisconnected() is called, and transfers the 119 * socket structure to so_q, making it available to accept(). 120 * 121 * If a socket is closed with sockets on either so_q0 or so_q, these 122 * sockets are dropped. 123 * 124 * Locking rules and assumptions: 125 * 126 * o socket::so_lock can change on the fly. The low level routines used 127 * to lock sockets are aware of this. When so_lock is acquired, the 128 * routine locking must check to see if so_lock still points to the 129 * lock that was acquired. If so_lock has changed in the meantime, the 130 * now irrelevant lock that was acquired must be dropped and the lock 131 * operation retried. Although not proven here, this is completely safe 132 * on a multiprocessor system, even with relaxed memory ordering, given 133 * the next two rules: 134 * 135 * o In order to mutate so_lock, the lock pointed to by the current value 136 * of so_lock must be held: i.e., the socket must be held locked by the 137 * changing thread. The thread must issue membar_release() to prevent 138 * memory accesses being reordered, and can set so_lock to the desired 139 * value. If the lock pointed to by the new value of so_lock is not 140 * held by the changing thread, the socket must then be considered 141 * unlocked. 142 * 143 * o If so_lock is mutated, and the previous lock referred to by so_lock 144 * could still be visible to other threads in the system (e.g. via file 145 * descriptor or protocol-internal reference), then the old lock must 146 * remain valid until the socket and/or protocol control block has been 147 * torn down. 148 * 149 * o If a socket has a non-NULL so_head value (i.e. is in the process of 150 * connecting), then locking the socket must also lock the socket pointed 151 * to by so_head: their lock pointers must match. 152 * 153 * o If a socket has connections in progress (so_q, so_q0 not empty) then 154 * locking the socket must also lock the sockets attached to both queues. 155 * Again, their lock pointers must match. 156 * 157 * o Beyond the initial lock assignment in socreate(), assigning locks to 158 * sockets is the responsibility of the individual protocols / protocol 159 * domains. 160 */ 161 162 static pool_cache_t socket_cache; 163 u_long sb_max = SB_MAX;/* maximum socket buffer size */ 164 static u_long sb_max_adj; /* adjusted sb_max */ 165 166 void 167 soisconnecting(struct socket *so) 168 { 169 170 KASSERT(solocked(so)); 171 172 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 173 so->so_state |= SS_ISCONNECTING; 174 } 175 176 void 177 soisconnected(struct socket *so) 178 { 179 struct socket *head; 180 181 head = so->so_head; 182 183 KASSERT(solocked(so)); 184 KASSERT(head == NULL || solocked2(so, head)); 185 186 so->so_state &= ~(SS_ISCONNECTING | SS_ISDISCONNECTING); 187 so->so_state |= SS_ISCONNECTED; 188 if (head && so->so_onq == &head->so_q0) { 189 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 190 /* 191 * Re-enqueue and wake up any waiters, e.g. 192 * processes blocking on accept(). 193 */ 194 soqremque(so, 0); 195 soqinsque(head, so, 1); 196 sorwakeup(head); 197 cv_broadcast(&head->so_cv); 198 } else { 199 so->so_upcall = 200 head->so_accf->so_accept_filter->accf_callback; 201 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 202 so->so_rcv.sb_flags |= SB_UPCALL; 203 so->so_options &= ~SO_ACCEPTFILTER; 204 (*so->so_upcall)(so, so->so_upcallarg, 205 POLLIN|POLLRDNORM, M_DONTWAIT); 206 } 207 } else { 208 cv_broadcast(&so->so_cv); 209 sorwakeup(so); 210 sowwakeup(so); 211 } 212 } 213 214 void 215 soisdisconnecting(struct socket *so) 216 { 217 218 KASSERT(solocked(so)); 219 220 so->so_state &= ~SS_ISCONNECTING; 221 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 222 cv_broadcast(&so->so_cv); 223 sowwakeup(so); 224 sorwakeup(so); 225 } 226 227 void 228 soisdisconnected(struct socket *so) 229 { 230 231 KASSERT(solocked(so)); 232 233 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 234 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 235 cv_broadcast(&so->so_cv); 236 sowwakeup(so); 237 sorwakeup(so); 238 } 239 240 void 241 soinit2(void) 242 { 243 244 socket_cache = pool_cache_init(sizeof(struct socket), 0, 0, 0, 245 "socket", NULL, IPL_SOFTNET, NULL, NULL, NULL); 246 } 247 248 /* 249 * sonewconn: accept a new connection. 250 * 251 * When an attempt at a new connection is noted on a socket which accepts 252 * connections, sonewconn(9) is called. If the connection is possible 253 * (subject to space constraints, etc) then we allocate a new structure, 254 * properly linked into the data structure of the original socket. 255 * 256 * => If 'soready' is true, then socket will become ready for accept() i.e. 257 * inserted into the so_q queue, SS_ISCONNECTED set and waiters awoken. 258 * => May be called from soft-interrupt context. 259 * => Listening socket should be locked. 260 * => Returns the new socket locked. 261 */ 262 struct socket * 263 sonewconn(struct socket *head, bool soready) 264 { 265 struct socket *so; 266 int soqueue, error; 267 268 KASSERT(solocked(head)); 269 270 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) { 271 /* 272 * Listen queue overflow. If there is an accept filter 273 * active, pass through the oldest cxn it's handling. 274 */ 275 if (head->so_accf == NULL) { 276 return NULL; 277 } else { 278 struct socket *so2, *next; 279 280 /* Pass the oldest connection waiting in the 281 accept filter */ 282 for (so2 = TAILQ_FIRST(&head->so_q0); 283 so2 != NULL; so2 = next) { 284 next = TAILQ_NEXT(so2, so_qe); 285 if (so2->so_upcall == NULL) { 286 continue; 287 } 288 so2->so_upcall = NULL; 289 so2->so_upcallarg = NULL; 290 so2->so_options &= ~SO_ACCEPTFILTER; 291 so2->so_rcv.sb_flags &= ~SB_UPCALL; 292 soisconnected(so2); 293 break; 294 } 295 296 /* If nothing was nudged out of the acept filter, bail 297 * out; otherwise proceed allocating the socket. */ 298 if (so2 == NULL) { 299 return NULL; 300 } 301 } 302 } 303 if ((head->so_options & SO_ACCEPTFILTER) != 0) { 304 soready = false; 305 } 306 soqueue = soready ? 1 : 0; 307 308 if ((so = soget(false)) == NULL) { 309 return NULL; 310 } 311 so->so_type = head->so_type; 312 so->so_options = head->so_options & ~SO_ACCEPTCONN; 313 so->so_linger = head->so_linger; 314 so->so_state = head->so_state | SS_NOFDREF; 315 so->so_proto = head->so_proto; 316 so->so_timeo = head->so_timeo; 317 so->so_pgid = head->so_pgid; 318 so->so_send = head->so_send; 319 so->so_receive = head->so_receive; 320 so->so_uidinfo = head->so_uidinfo; 321 so->so_egid = head->so_egid; 322 so->so_cpid = head->so_cpid; 323 324 /* 325 * Share the lock with the listening-socket, it may get unshared 326 * once the connection is complete. 327 * 328 * so_lock is stable while we hold the socket locked, so no 329 * need for atomic_load_* here. 330 */ 331 mutex_obj_hold(head->so_lock); 332 so->so_lock = head->so_lock; 333 334 /* 335 * Reserve the space for socket buffers. 336 */ 337 #ifdef MBUFTRACE 338 so->so_mowner = head->so_mowner; 339 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner; 340 so->so_snd.sb_mowner = head->so_snd.sb_mowner; 341 #endif 342 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 343 goto out; 344 } 345 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 346 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 347 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 348 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 349 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 350 so->so_snd.sb_flags |= head->so_snd.sb_flags & (SB_AUTOSIZE | SB_ASYNC); 351 352 /* 353 * Finally, perform the protocol attach. Note: a new socket 354 * lock may be assigned at this point (if so, it will be held). 355 */ 356 error = (*so->so_proto->pr_usrreqs->pr_attach)(so, 0); 357 if (error) { 358 out: 359 KASSERT(solocked(so)); 360 KASSERT(so->so_accf == NULL); 361 soput(so); 362 363 /* Note: the listening socket shall stay locked. */ 364 KASSERT(solocked(head)); 365 return NULL; 366 } 367 KASSERT(solocked2(head, so)); 368 369 /* 370 * Insert into the queue. If ready, update the connection status 371 * and wake up any waiters, e.g. processes blocking on accept(). 372 */ 373 soqinsque(head, so, soqueue); 374 if (soready) { 375 so->so_state |= SS_ISCONNECTED; 376 sorwakeup(head); 377 cv_broadcast(&head->so_cv); 378 } 379 return so; 380 } 381 382 struct socket * 383 soget(bool waitok) 384 { 385 struct socket *so; 386 387 so = pool_cache_get(socket_cache, (waitok ? PR_WAITOK : PR_NOWAIT)); 388 if (__predict_false(so == NULL)) 389 return (NULL); 390 memset(so, 0, sizeof(*so)); 391 TAILQ_INIT(&so->so_q0); 392 TAILQ_INIT(&so->so_q); 393 cv_init(&so->so_cv, "socket"); 394 cv_init(&so->so_rcv.sb_cv, "netio"); 395 cv_init(&so->so_snd.sb_cv, "netio"); 396 selinit(&so->so_rcv.sb_sel); 397 selinit(&so->so_snd.sb_sel); 398 so->so_rcv.sb_so = so; 399 so->so_snd.sb_so = so; 400 return so; 401 } 402 403 void 404 soput(struct socket *so) 405 { 406 407 KASSERT(!cv_has_waiters(&so->so_cv)); 408 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv)); 409 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv)); 410 seldestroy(&so->so_rcv.sb_sel); 411 seldestroy(&so->so_snd.sb_sel); 412 mutex_obj_free(so->so_lock); 413 cv_destroy(&so->so_cv); 414 cv_destroy(&so->so_rcv.sb_cv); 415 cv_destroy(&so->so_snd.sb_cv); 416 pool_cache_put(socket_cache, so); 417 } 418 419 /* 420 * soqinsque: insert socket of a new connection into the specified 421 * accept queue of the listening socket (head). 422 * 423 * q = 0: queue of partial connections 424 * q = 1: queue of incoming connections 425 */ 426 void 427 soqinsque(struct socket *head, struct socket *so, int q) 428 { 429 KASSERT(q == 0 || q == 1); 430 KASSERT(solocked2(head, so)); 431 KASSERT(so->so_onq == NULL); 432 KASSERT(so->so_head == NULL); 433 434 so->so_head = head; 435 if (q == 0) { 436 head->so_q0len++; 437 so->so_onq = &head->so_q0; 438 } else { 439 head->so_qlen++; 440 so->so_onq = &head->so_q; 441 } 442 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe); 443 } 444 445 /* 446 * soqremque: remove socket from the specified queue. 447 * 448 * => Returns true if socket was removed from the specified queue. 449 * => False if socket was not removed (because it was in other queue). 450 */ 451 bool 452 soqremque(struct socket *so, int q) 453 { 454 struct socket *head = so->so_head; 455 456 KASSERT(q == 0 || q == 1); 457 KASSERT(solocked(so)); 458 KASSERT(so->so_onq != NULL); 459 KASSERT(head != NULL); 460 461 if (q == 0) { 462 if (so->so_onq != &head->so_q0) 463 return false; 464 head->so_q0len--; 465 } else { 466 if (so->so_onq != &head->so_q) 467 return false; 468 head->so_qlen--; 469 } 470 KASSERT(solocked2(so, head)); 471 TAILQ_REMOVE(so->so_onq, so, so_qe); 472 so->so_onq = NULL; 473 so->so_head = NULL; 474 return true; 475 } 476 477 /* 478 * socantsendmore: indicates that no more data will be sent on the 479 * socket; it would normally be applied to a socket when the user 480 * informs the system that no more data is to be sent, by the protocol 481 * code (in case pr_shutdown()). 482 */ 483 void 484 socantsendmore(struct socket *so) 485 { 486 KASSERT(solocked(so)); 487 488 so->so_state |= SS_CANTSENDMORE; 489 sowwakeup(so); 490 } 491 492 /* 493 * socantrcvmore(): indicates that no more data will be received and 494 * will normally be applied to the socket by a protocol when it detects 495 * that the peer will send no more data. Data queued for reading in 496 * the socket may yet be read. 497 */ 498 void 499 socantrcvmore(struct socket *so) 500 { 501 KASSERT(solocked(so)); 502 503 so->so_state |= SS_CANTRCVMORE; 504 sorwakeup(so); 505 } 506 507 /* 508 * soroverflow(): indicates that data was attempted to be sent 509 * but the receiving buffer overflowed. 510 */ 511 void 512 soroverflow(struct socket *so) 513 { 514 KASSERT(solocked(so)); 515 516 so->so_rcv.sb_overflowed++; 517 if (so->so_options & SO_RERROR) { 518 so->so_rerror = ENOBUFS; 519 sorwakeup(so); 520 } 521 } 522 523 /* 524 * Wait for data to arrive at/drain from a socket buffer. 525 */ 526 int 527 sbwait(struct sockbuf *sb) 528 { 529 struct socket *so; 530 kmutex_t *lock; 531 int error; 532 533 so = sb->sb_so; 534 535 KASSERT(solocked(so)); 536 537 sb->sb_flags |= SB_NOTIFY; 538 lock = so->so_lock; 539 if ((sb->sb_flags & SB_NOINTR) != 0) 540 error = cv_timedwait(&sb->sb_cv, lock, sb->sb_timeo); 541 else 542 error = cv_timedwait_sig(&sb->sb_cv, lock, sb->sb_timeo); 543 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock))) 544 solockretry(so, lock); 545 return error; 546 } 547 548 /* 549 * Wakeup processes waiting on a socket buffer. 550 * Do asynchronous notification via SIGIO 551 * if the socket buffer has the SB_ASYNC flag set. 552 */ 553 void 554 sowakeup(struct socket *so, struct sockbuf *sb, int code) 555 { 556 int band; 557 558 KASSERT(solocked(so)); 559 KASSERT(sb->sb_so == so); 560 561 switch (code) { 562 case POLL_IN: 563 band = POLLIN|POLLRDNORM; 564 break; 565 566 case POLL_OUT: 567 band = POLLOUT|POLLWRNORM; 568 break; 569 570 case POLL_HUP: 571 band = POLLHUP; 572 break; 573 574 default: 575 band = 0; 576 #ifdef DIAGNOSTIC 577 printf("bad siginfo code %d in socket notification.\n", code); 578 #endif 579 break; 580 } 581 582 sb->sb_flags &= ~SB_NOTIFY; 583 selnotify(&sb->sb_sel, band, NOTE_SUBMIT); 584 cv_broadcast(&sb->sb_cv); 585 if (sb->sb_flags & SB_ASYNC) 586 fownsignal(so->so_pgid, SIGIO, code, band, so); 587 if (sb->sb_flags & SB_UPCALL) 588 (*so->so_upcall)(so, so->so_upcallarg, band, M_DONTWAIT); 589 } 590 591 /* 592 * Reset a socket's lock pointer. Wake all threads waiting on the 593 * socket's condition variables so that they can restart their waits 594 * using the new lock. The existing lock must be held. 595 * 596 * Caller must have issued membar_release before this. 597 */ 598 void 599 solockreset(struct socket *so, kmutex_t *lock) 600 { 601 602 KASSERT(solocked(so)); 603 604 so->so_lock = lock; 605 cv_broadcast(&so->so_snd.sb_cv); 606 cv_broadcast(&so->so_rcv.sb_cv); 607 cv_broadcast(&so->so_cv); 608 } 609 610 /* 611 * Socket buffer (struct sockbuf) utility routines. 612 * 613 * Each socket contains two socket buffers: one for sending data and 614 * one for receiving data. Each buffer contains a queue of mbufs, 615 * information about the number of mbufs and amount of data in the 616 * queue, and other fields allowing poll() statements and notification 617 * on data availability to be implemented. 618 * 619 * Data stored in a socket buffer is maintained as a list of records. 620 * Each record is a list of mbufs chained together with the m_next 621 * field. Records are chained together with the m_nextpkt field. The upper 622 * level routine soreceive() expects the following conventions to be 623 * observed when placing information in the receive buffer: 624 * 625 * 1. If the protocol requires each message be preceded by the sender's 626 * name, then a record containing that name must be present before 627 * any associated data (mbuf's must be of type MT_SONAME). 628 * 2. If the protocol supports the exchange of ``access rights'' (really 629 * just additional data associated with the message), and there are 630 * ``rights'' to be received, then a record containing this data 631 * should be present (mbuf's must be of type MT_CONTROL). 632 * 3. If a name or rights record exists, then it must be followed by 633 * a data record, perhaps of zero length. 634 * 635 * Before using a new socket structure it is first necessary to reserve 636 * buffer space to the socket, by calling sbreserve(). This should commit 637 * some of the available buffer space in the system buffer pool for the 638 * socket (currently, it does nothing but enforce limits). The space 639 * should be released by calling sbrelease() when the socket is destroyed. 640 */ 641 642 int 643 sb_max_set(u_long new_sbmax) 644 { 645 int s; 646 647 if (new_sbmax < (16 * 1024)) 648 return (EINVAL); 649 650 s = splsoftnet(); 651 sb_max = new_sbmax; 652 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES); 653 splx(s); 654 655 return (0); 656 } 657 658 int 659 soreserve(struct socket *so, u_long sndcc, u_long rcvcc) 660 { 661 KASSERT(so->so_pcb == NULL || solocked(so)); 662 663 /* 664 * there's at least one application (a configure script of screen) 665 * which expects a fifo is writable even if it has "some" bytes 666 * in its buffer. 667 * so we want to make sure (hiwat - lowat) >= (some bytes). 668 * 669 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above. 670 * we expect it's large enough for such applications. 671 */ 672 u_long lowat = MAX(sock_loan_thresh, MCLBYTES); 673 u_long hiwat = lowat + PIPE_BUF; 674 675 if (sndcc < hiwat) 676 sndcc = hiwat; 677 if (sbreserve(&so->so_snd, sndcc, so) == 0) 678 goto bad; 679 if (sbreserve(&so->so_rcv, rcvcc, so) == 0) 680 goto bad2; 681 if (so->so_rcv.sb_lowat == 0) 682 so->so_rcv.sb_lowat = 1; 683 if (so->so_snd.sb_lowat == 0) 684 so->so_snd.sb_lowat = lowat; 685 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat) 686 so->so_snd.sb_lowat = so->so_snd.sb_hiwat; 687 return (0); 688 bad2: 689 sbrelease(&so->so_snd, so); 690 bad: 691 return (ENOBUFS); 692 } 693 694 /* 695 * Allot mbufs to a sockbuf. 696 * Attempt to scale mbmax so that mbcnt doesn't become limiting 697 * if buffering efficiency is near the normal case. 698 */ 699 int 700 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so) 701 { 702 struct lwp *l = curlwp; /* XXX */ 703 rlim_t maxcc; 704 struct uidinfo *uidinfo; 705 706 KASSERT(so->so_pcb == NULL || solocked(so)); 707 KASSERT(sb->sb_so == so); 708 KASSERT(sb_max_adj != 0); 709 710 if (cc == 0 || cc > sb_max_adj) 711 return (0); 712 713 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur; 714 715 uidinfo = so->so_uidinfo; 716 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc)) 717 return 0; 718 sb->sb_mbmax = uimin(cc * 2, sb_max); 719 if (sb->sb_lowat > sb->sb_hiwat) 720 sb->sb_lowat = sb->sb_hiwat; 721 722 return (1); 723 } 724 725 /* 726 * Free mbufs held by a socket, and reserved mbuf space. We do not assert 727 * that the socket is held locked here: see sorflush(). 728 */ 729 void 730 sbrelease(struct sockbuf *sb, struct socket *so) 731 { 732 733 KASSERT(sb->sb_so == so); 734 735 sbflush(sb); 736 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY); 737 sb->sb_mbmax = 0; 738 } 739 740 /* 741 * Routines to add and remove 742 * data from an mbuf queue. 743 * 744 * The routines sbappend() or sbappendrecord() are normally called to 745 * append new mbufs to a socket buffer, after checking that adequate 746 * space is available, comparing the function sbspace() with the amount 747 * of data to be added. sbappendrecord() differs from sbappend() in 748 * that data supplied is treated as the beginning of a new record. 749 * To place a sender's address, optional access rights, and data in a 750 * socket receive buffer, sbappendaddr() should be used. To place 751 * access rights and data in a socket receive buffer, sbappendrights() 752 * should be used. In either case, the new data begins a new record. 753 * Note that unlike sbappend() and sbappendrecord(), these routines check 754 * for the caller that there will be enough space to store the data. 755 * Each fails if there is not enough space, or if it cannot find mbufs 756 * to store additional information in. 757 * 758 * Reliable protocols may use the socket send buffer to hold data 759 * awaiting acknowledgement. Data is normally copied from a socket 760 * send buffer in a protocol with m_copym for output to a peer, 761 * and then removing the data from the socket buffer with sbdrop() 762 * or sbdroprecord() when the data is acknowledged by the peer. 763 */ 764 765 #ifdef SOCKBUF_DEBUG 766 void 767 sblastrecordchk(struct sockbuf *sb, const char *where) 768 { 769 struct mbuf *m = sb->sb_mb; 770 771 KASSERT(solocked(sb->sb_so)); 772 773 while (m && m->m_nextpkt) 774 m = m->m_nextpkt; 775 776 if (m != sb->sb_lastrecord) { 777 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n", 778 sb->sb_mb, sb->sb_lastrecord, m); 779 printf("packet chain:\n"); 780 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) 781 printf("\t%p\n", m); 782 panic("sblastrecordchk from %s", where); 783 } 784 } 785 786 void 787 sblastmbufchk(struct sockbuf *sb, const char *where) 788 { 789 struct mbuf *m = sb->sb_mb; 790 struct mbuf *n; 791 792 KASSERT(solocked(sb->sb_so)); 793 794 while (m && m->m_nextpkt) 795 m = m->m_nextpkt; 796 797 while (m && m->m_next) 798 m = m->m_next; 799 800 if (m != sb->sb_mbtail) { 801 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n", 802 sb->sb_mb, sb->sb_mbtail, m); 803 printf("packet tree:\n"); 804 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) { 805 printf("\t"); 806 for (n = m; n != NULL; n = n->m_next) 807 printf("%p ", n); 808 printf("\n"); 809 } 810 panic("sblastmbufchk from %s", where); 811 } 812 } 813 #endif /* SOCKBUF_DEBUG */ 814 815 /* 816 * Link a chain of records onto a socket buffer 817 */ 818 #define SBLINKRECORDCHAIN(sb, m0, mlast) \ 819 do { \ 820 if ((sb)->sb_lastrecord != NULL) \ 821 (sb)->sb_lastrecord->m_nextpkt = (m0); \ 822 else \ 823 (sb)->sb_mb = (m0); \ 824 (sb)->sb_lastrecord = (mlast); \ 825 } while (/*CONSTCOND*/0) 826 827 828 #define SBLINKRECORD(sb, m0) \ 829 SBLINKRECORDCHAIN(sb, m0, m0) 830 831 /* 832 * Append mbuf chain m to the last record in the 833 * socket buffer sb. The additional space associated 834 * the mbuf chain is recorded in sb. Empty mbufs are 835 * discarded and mbufs are compacted where possible. 836 */ 837 void 838 sbappend(struct sockbuf *sb, struct mbuf *m) 839 { 840 struct mbuf *n; 841 842 KASSERT(solocked(sb->sb_so)); 843 844 if (m == NULL) 845 return; 846 847 #ifdef MBUFTRACE 848 m_claimm(m, sb->sb_mowner); 849 #endif 850 851 SBLASTRECORDCHK(sb, "sbappend 1"); 852 853 if ((n = sb->sb_lastrecord) != NULL) { 854 /* 855 * XXX Would like to simply use sb_mbtail here, but 856 * XXX I need to verify that I won't miss an EOR that 857 * XXX way. 858 */ 859 do { 860 if (n->m_flags & M_EOR) { 861 sbappendrecord(sb, m); /* XXXXXX!!!! */ 862 return; 863 } 864 } while (n->m_next && (n = n->m_next)); 865 } else { 866 /* 867 * If this is the first record in the socket buffer, it's 868 * also the last record. 869 */ 870 sb->sb_lastrecord = m; 871 } 872 sbcompress(sb, m, n); 873 SBLASTRECORDCHK(sb, "sbappend 2"); 874 } 875 876 /* 877 * This version of sbappend() should only be used when the caller 878 * absolutely knows that there will never be more than one record 879 * in the socket buffer, that is, a stream protocol (such as TCP). 880 */ 881 void 882 sbappendstream(struct sockbuf *sb, struct mbuf *m) 883 { 884 885 KASSERT(solocked(sb->sb_so)); 886 KDASSERT(m->m_nextpkt == NULL); 887 KASSERT(sb->sb_mb == sb->sb_lastrecord); 888 889 SBLASTMBUFCHK(sb, __func__); 890 891 #ifdef MBUFTRACE 892 m_claimm(m, sb->sb_mowner); 893 #endif 894 895 sbcompress(sb, m, sb->sb_mbtail); 896 897 sb->sb_lastrecord = sb->sb_mb; 898 SBLASTRECORDCHK(sb, __func__); 899 } 900 901 #ifdef SOCKBUF_DEBUG 902 void 903 sbcheck(struct sockbuf *sb) 904 { 905 struct mbuf *m, *m2; 906 u_long len, mbcnt; 907 908 KASSERT(solocked(sb->sb_so)); 909 910 len = 0; 911 mbcnt = 0; 912 for (m = sb->sb_mb; m; m = m->m_nextpkt) { 913 for (m2 = m; m2 != NULL; m2 = m2->m_next) { 914 len += m2->m_len; 915 mbcnt += MSIZE; 916 if (m2->m_flags & M_EXT) 917 mbcnt += m2->m_ext.ext_size; 918 if (m2->m_nextpkt != NULL) 919 panic("sbcheck nextpkt"); 920 } 921 } 922 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) { 923 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc, 924 mbcnt, sb->sb_mbcnt); 925 panic("sbcheck"); 926 } 927 } 928 #endif 929 930 /* 931 * As above, except the mbuf chain 932 * begins a new record. 933 */ 934 void 935 sbappendrecord(struct sockbuf *sb, struct mbuf *m0) 936 { 937 struct mbuf *m; 938 939 KASSERT(solocked(sb->sb_so)); 940 941 if (m0 == NULL) 942 return; 943 944 #ifdef MBUFTRACE 945 m_claimm(m0, sb->sb_mowner); 946 #endif 947 /* 948 * Put the first mbuf on the queue. 949 * Note this permits zero length records. 950 */ 951 sballoc(sb, m0); 952 SBLASTRECORDCHK(sb, "sbappendrecord 1"); 953 SBLINKRECORD(sb, m0); 954 m = m0->m_next; 955 m0->m_next = 0; 956 if (m && (m0->m_flags & M_EOR)) { 957 m0->m_flags &= ~M_EOR; 958 m->m_flags |= M_EOR; 959 } 960 sbcompress(sb, m, m0); 961 SBLASTRECORDCHK(sb, "sbappendrecord 2"); 962 } 963 964 /* 965 * As above except that OOB data 966 * is inserted at the beginning of the sockbuf, 967 * but after any other OOB data. 968 */ 969 void 970 sbinsertoob(struct sockbuf *sb, struct mbuf *m0) 971 { 972 struct mbuf *m, **mp; 973 974 KASSERT(solocked(sb->sb_so)); 975 976 if (m0 == NULL) 977 return; 978 979 SBLASTRECORDCHK(sb, "sbinsertoob 1"); 980 981 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) { 982 again: 983 switch (m->m_type) { 984 985 case MT_OOBDATA: 986 continue; /* WANT next train */ 987 988 case MT_CONTROL: 989 if ((m = m->m_next) != NULL) 990 goto again; /* inspect THIS train further */ 991 } 992 break; 993 } 994 /* 995 * Put the first mbuf on the queue. 996 * Note this permits zero length records. 997 */ 998 sballoc(sb, m0); 999 m0->m_nextpkt = *mp; 1000 if (*mp == NULL) { 1001 /* m0 is actually the new tail */ 1002 sb->sb_lastrecord = m0; 1003 } 1004 *mp = m0; 1005 m = m0->m_next; 1006 m0->m_next = 0; 1007 if (m && (m0->m_flags & M_EOR)) { 1008 m0->m_flags &= ~M_EOR; 1009 m->m_flags |= M_EOR; 1010 } 1011 sbcompress(sb, m, m0); 1012 SBLASTRECORDCHK(sb, "sbinsertoob 2"); 1013 } 1014 1015 /* 1016 * Append address and data, and optionally, control (ancillary) data 1017 * to the receive queue of a socket. If present, 1018 * m0 must include a packet header with total length. 1019 * Returns 0 if no space in sockbuf or insufficient mbufs. 1020 */ 1021 int 1022 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0, 1023 struct mbuf *control) 1024 { 1025 struct mbuf *m, *n, *nlast; 1026 int space, len; 1027 1028 KASSERT(solocked(sb->sb_so)); 1029 1030 space = asa->sa_len; 1031 1032 if (m0 != NULL) { 1033 if ((m0->m_flags & M_PKTHDR) == 0) 1034 panic("sbappendaddr"); 1035 space += m0->m_pkthdr.len; 1036 #ifdef MBUFTRACE 1037 m_claimm(m0, sb->sb_mowner); 1038 #endif 1039 } 1040 for (n = control; n; n = n->m_next) { 1041 space += n->m_len; 1042 MCLAIM(n, sb->sb_mowner); 1043 if (n->m_next == NULL) /* keep pointer to last control buf */ 1044 break; 1045 } 1046 if (space > sbspace(sb)) 1047 return (0); 1048 m = m_get(M_DONTWAIT, MT_SONAME); 1049 if (m == NULL) 1050 return (0); 1051 MCLAIM(m, sb->sb_mowner); 1052 /* 1053 * XXX avoid 'comparison always true' warning which isn't easily 1054 * avoided. 1055 */ 1056 len = asa->sa_len; 1057 if (len > MLEN) { 1058 MEXTMALLOC(m, asa->sa_len, M_NOWAIT); 1059 if ((m->m_flags & M_EXT) == 0) { 1060 m_free(m); 1061 return (0); 1062 } 1063 } 1064 m->m_len = asa->sa_len; 1065 memcpy(mtod(m, void *), asa, asa->sa_len); 1066 if (n) 1067 n->m_next = m0; /* concatenate data to control */ 1068 else 1069 control = m0; 1070 m->m_next = control; 1071 1072 SBLASTRECORDCHK(sb, "sbappendaddr 1"); 1073 1074 for (n = m; n->m_next != NULL; n = n->m_next) 1075 sballoc(sb, n); 1076 sballoc(sb, n); 1077 nlast = n; 1078 SBLINKRECORD(sb, m); 1079 1080 sb->sb_mbtail = nlast; 1081 SBLASTMBUFCHK(sb, "sbappendaddr"); 1082 SBLASTRECORDCHK(sb, "sbappendaddr 2"); 1083 1084 return (1); 1085 } 1086 1087 /* 1088 * Helper for sbappendchainaddr: prepend a struct sockaddr* to 1089 * an mbuf chain. 1090 */ 1091 static inline struct mbuf * 1092 m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0, 1093 const struct sockaddr *asa) 1094 { 1095 struct mbuf *m; 1096 const int salen = asa->sa_len; 1097 1098 KASSERT(solocked(sb->sb_so)); 1099 1100 /* only the first in each chain need be a pkthdr */ 1101 m = m_gethdr(M_DONTWAIT, MT_SONAME); 1102 if (m == NULL) 1103 return NULL; 1104 MCLAIM(m, sb->sb_mowner); 1105 #ifdef notyet 1106 if (salen > MHLEN) { 1107 MEXTMALLOC(m, salen, M_NOWAIT); 1108 if ((m->m_flags & M_EXT) == 0) { 1109 m_free(m); 1110 return NULL; 1111 } 1112 } 1113 #else 1114 KASSERT(salen <= MHLEN); 1115 #endif 1116 m->m_len = salen; 1117 memcpy(mtod(m, void *), asa, salen); 1118 m->m_next = m0; 1119 m->m_pkthdr.len = salen + m0->m_pkthdr.len; 1120 1121 return m; 1122 } 1123 1124 int 1125 sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa, 1126 struct mbuf *m0, int sbprio) 1127 { 1128 struct mbuf *m, *n, *n0, *nlast; 1129 int error; 1130 1131 KASSERT(solocked(sb->sb_so)); 1132 1133 /* 1134 * XXX sbprio reserved for encoding priority of this* request: 1135 * SB_PRIO_NONE --> honour normal sb limits 1136 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space, 1137 * take whole chain. Intended for large requests 1138 * that should be delivered atomically (all, or none). 1139 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow 1140 * over normal socket limits, for messages indicating 1141 * buffer overflow in earlier normal/lower-priority messages 1142 * SB_PRIO_BESTEFFORT --> ignore limits entirely. 1143 * Intended for kernel-generated messages only. 1144 * Up to generator to avoid total mbuf resource exhaustion. 1145 */ 1146 (void)sbprio; 1147 1148 if (m0 && (m0->m_flags & M_PKTHDR) == 0) 1149 panic("sbappendaddrchain"); 1150 1151 #ifdef notyet 1152 space = sbspace(sb); 1153 1154 /* 1155 * Enforce SB_PRIO_* limits as described above. 1156 */ 1157 #endif 1158 1159 n0 = NULL; 1160 nlast = NULL; 1161 for (m = m0; m; m = m->m_nextpkt) { 1162 struct mbuf *np; 1163 1164 #ifdef MBUFTRACE 1165 m_claimm(m, sb->sb_mowner); 1166 #endif 1167 1168 /* Prepend sockaddr to this record (m) of input chain m0 */ 1169 n = m_prepend_sockaddr(sb, m, asa); 1170 if (n == NULL) { 1171 error = ENOBUFS; 1172 goto bad; 1173 } 1174 1175 /* Append record (asa+m) to end of new chain n0 */ 1176 if (n0 == NULL) { 1177 n0 = n; 1178 } else { 1179 nlast->m_nextpkt = n; 1180 } 1181 /* Keep track of last record on new chain */ 1182 nlast = n; 1183 1184 for (np = n; np; np = np->m_next) 1185 sballoc(sb, np); 1186 } 1187 1188 SBLASTRECORDCHK(sb, "sbappendaddrchain 1"); 1189 1190 /* Drop the entire chain of (asa+m) records onto the socket */ 1191 SBLINKRECORDCHAIN(sb, n0, nlast); 1192 1193 SBLASTRECORDCHK(sb, "sbappendaddrchain 2"); 1194 1195 for (m = nlast; m->m_next; m = m->m_next) 1196 ; 1197 sb->sb_mbtail = m; 1198 SBLASTMBUFCHK(sb, "sbappendaddrchain"); 1199 1200 return (1); 1201 1202 bad: 1203 /* 1204 * On error, free the prepended addreseses. For consistency 1205 * with sbappendaddr(), leave it to our caller to free 1206 * the input record chain passed to us as m0. 1207 */ 1208 while ((n = n0) != NULL) { 1209 struct mbuf *np; 1210 1211 /* Undo the sballoc() of this record */ 1212 for (np = n; np; np = np->m_next) 1213 sbfree(sb, np); 1214 1215 n0 = n->m_nextpkt; /* iterate at next prepended address */ 1216 np = m_free(n); /* free prepended address (not data) */ 1217 } 1218 return error; 1219 } 1220 1221 1222 int 1223 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control) 1224 { 1225 struct mbuf *m, *mlast, *n; 1226 int space; 1227 1228 KASSERT(solocked(sb->sb_so)); 1229 1230 space = 0; 1231 if (control == NULL) 1232 panic("sbappendcontrol"); 1233 for (m = control; ; m = m->m_next) { 1234 space += m->m_len; 1235 MCLAIM(m, sb->sb_mowner); 1236 if (m->m_next == NULL) 1237 break; 1238 } 1239 n = m; /* save pointer to last control buffer */ 1240 for (m = m0; m; m = m->m_next) { 1241 MCLAIM(m, sb->sb_mowner); 1242 space += m->m_len; 1243 } 1244 if (space > sbspace(sb)) 1245 return (0); 1246 n->m_next = m0; /* concatenate data to control */ 1247 1248 SBLASTRECORDCHK(sb, "sbappendcontrol 1"); 1249 1250 for (m = control; m->m_next != NULL; m = m->m_next) 1251 sballoc(sb, m); 1252 sballoc(sb, m); 1253 mlast = m; 1254 SBLINKRECORD(sb, control); 1255 1256 sb->sb_mbtail = mlast; 1257 SBLASTMBUFCHK(sb, "sbappendcontrol"); 1258 SBLASTRECORDCHK(sb, "sbappendcontrol 2"); 1259 1260 return (1); 1261 } 1262 1263 /* 1264 * Compress mbuf chain m into the socket 1265 * buffer sb following mbuf n. If n 1266 * is null, the buffer is presumed empty. 1267 */ 1268 void 1269 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n) 1270 { 1271 int eor; 1272 struct mbuf *o; 1273 1274 KASSERT(solocked(sb->sb_so)); 1275 1276 eor = 0; 1277 while (m) { 1278 eor |= m->m_flags & M_EOR; 1279 if (m->m_len == 0 && 1280 (eor == 0 || 1281 (((o = m->m_next) || (o = n)) && 1282 o->m_type == m->m_type))) { 1283 if (sb->sb_lastrecord == m) 1284 sb->sb_lastrecord = m->m_next; 1285 m = m_free(m); 1286 continue; 1287 } 1288 if (n && (n->m_flags & M_EOR) == 0 && 1289 /* M_TRAILINGSPACE() checks buffer writeability */ 1290 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */ 1291 m->m_len <= M_TRAILINGSPACE(n) && 1292 n->m_type == m->m_type) { 1293 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *), 1294 (unsigned)m->m_len); 1295 n->m_len += m->m_len; 1296 sb->sb_cc += m->m_len; 1297 m = m_free(m); 1298 continue; 1299 } 1300 if (n) 1301 n->m_next = m; 1302 else 1303 sb->sb_mb = m; 1304 sb->sb_mbtail = m; 1305 sballoc(sb, m); 1306 n = m; 1307 m->m_flags &= ~M_EOR; 1308 m = m->m_next; 1309 n->m_next = 0; 1310 } 1311 if (eor) { 1312 if (n) 1313 n->m_flags |= eor; 1314 else 1315 printf("semi-panic: sbcompress\n"); 1316 } 1317 SBLASTMBUFCHK(sb, __func__); 1318 } 1319 1320 /* 1321 * Free all mbufs in a sockbuf. 1322 * Check that all resources are reclaimed. 1323 */ 1324 void 1325 sbflush(struct sockbuf *sb) 1326 { 1327 1328 KASSERT(solocked(sb->sb_so)); 1329 KASSERT((sb->sb_flags & SB_LOCK) == 0); 1330 1331 while (sb->sb_mbcnt) 1332 sbdrop(sb, (int)sb->sb_cc); 1333 1334 KASSERT(sb->sb_cc == 0); 1335 KASSERT(sb->sb_mb == NULL); 1336 KASSERT(sb->sb_mbtail == NULL); 1337 KASSERT(sb->sb_lastrecord == NULL); 1338 } 1339 1340 /* 1341 * Drop data from (the front of) a sockbuf. 1342 */ 1343 void 1344 sbdrop(struct sockbuf *sb, int len) 1345 { 1346 struct mbuf *m, *next; 1347 1348 KASSERT(solocked(sb->sb_so)); 1349 1350 next = (m = sb->sb_mb) ? m->m_nextpkt : NULL; 1351 while (len > 0) { 1352 if (m == NULL) { 1353 if (next == NULL) 1354 panic("sbdrop(%p,%d): cc=%lu", 1355 sb, len, sb->sb_cc); 1356 m = next; 1357 next = m->m_nextpkt; 1358 continue; 1359 } 1360 if (m->m_len > len) { 1361 m->m_len -= len; 1362 m->m_data += len; 1363 sb->sb_cc -= len; 1364 break; 1365 } 1366 len -= m->m_len; 1367 sbfree(sb, m); 1368 m = m_free(m); 1369 } 1370 while (m && m->m_len == 0) { 1371 sbfree(sb, m); 1372 m = m_free(m); 1373 } 1374 if (m) { 1375 sb->sb_mb = m; 1376 m->m_nextpkt = next; 1377 } else 1378 sb->sb_mb = next; 1379 /* 1380 * First part is an inline SB_EMPTY_FIXUP(). Second part 1381 * makes sure sb_lastrecord is up-to-date if we dropped 1382 * part of the last record. 1383 */ 1384 m = sb->sb_mb; 1385 if (m == NULL) { 1386 sb->sb_mbtail = NULL; 1387 sb->sb_lastrecord = NULL; 1388 } else if (m->m_nextpkt == NULL) 1389 sb->sb_lastrecord = m; 1390 } 1391 1392 /* 1393 * Drop a record off the front of a sockbuf 1394 * and move the next record to the front. 1395 */ 1396 void 1397 sbdroprecord(struct sockbuf *sb) 1398 { 1399 struct mbuf *m, *mn; 1400 1401 KASSERT(solocked(sb->sb_so)); 1402 1403 m = sb->sb_mb; 1404 if (m) { 1405 sb->sb_mb = m->m_nextpkt; 1406 do { 1407 sbfree(sb, m); 1408 mn = m_free(m); 1409 } while ((m = mn) != NULL); 1410 } 1411 SB_EMPTY_FIXUP(sb); 1412 } 1413 1414 /* 1415 * Create a "control" mbuf containing the specified data 1416 * with the specified type for presentation on a socket buffer. 1417 */ 1418 struct mbuf * 1419 sbcreatecontrol1(void **p, int size, int type, int level, int flags) 1420 { 1421 struct cmsghdr *cp; 1422 struct mbuf *m; 1423 int space = CMSG_SPACE(size); 1424 1425 if ((flags & M_DONTWAIT) && space > MCLBYTES) { 1426 printf("%s: message too large %d\n", __func__, space); 1427 return NULL; 1428 } 1429 1430 if ((m = m_get(flags, MT_CONTROL)) == NULL) 1431 return NULL; 1432 if (space > MLEN) { 1433 if (space > MCLBYTES) 1434 MEXTMALLOC(m, space, M_WAITOK); 1435 else 1436 MCLGET(m, flags); 1437 if ((m->m_flags & M_EXT) == 0) { 1438 m_free(m); 1439 return NULL; 1440 } 1441 } 1442 cp = mtod(m, struct cmsghdr *); 1443 *p = CMSG_DATA(cp); 1444 m->m_len = space; 1445 cp->cmsg_len = CMSG_LEN(size); 1446 cp->cmsg_level = level; 1447 cp->cmsg_type = type; 1448 1449 memset(cp + 1, 0, CMSG_LEN(0) - sizeof(*cp)); 1450 memset((uint8_t *)*p + size, 0, CMSG_ALIGN(size) - size); 1451 1452 return m; 1453 } 1454 1455 struct mbuf * 1456 sbcreatecontrol(void *p, int size, int type, int level) 1457 { 1458 struct mbuf *m; 1459 void *v; 1460 1461 m = sbcreatecontrol1(&v, size, type, level, M_DONTWAIT); 1462 if (m == NULL) 1463 return NULL; 1464 memcpy(v, p, size); 1465 return m; 1466 } 1467 1468 void 1469 solockretry(struct socket *so, kmutex_t *lock) 1470 { 1471 1472 while (lock != atomic_load_relaxed(&so->so_lock)) { 1473 mutex_exit(lock); 1474 lock = atomic_load_consume(&so->so_lock); 1475 mutex_enter(lock); 1476 } 1477 } 1478 1479 bool 1480 solocked(const struct socket *so) 1481 { 1482 1483 /* 1484 * Used only for diagnostic assertions, so so_lock should be 1485 * stable at this point, hence on need for atomic_load_*. 1486 */ 1487 return mutex_owned(so->so_lock); 1488 } 1489 1490 bool 1491 solocked2(const struct socket *so1, const struct socket *so2) 1492 { 1493 const kmutex_t *lock; 1494 1495 /* 1496 * Used only for diagnostic assertions, so so_lock should be 1497 * stable at this point, hence on need for atomic_load_*. 1498 */ 1499 lock = so1->so_lock; 1500 if (lock != so2->so_lock) 1501 return false; 1502 return mutex_owned(lock); 1503 } 1504 1505 /* 1506 * sosetlock: assign a default lock to a new socket. 1507 */ 1508 void 1509 sosetlock(struct socket *so) 1510 { 1511 if (so->so_lock == NULL) { 1512 kmutex_t *lock = softnet_lock; 1513 1514 so->so_lock = lock; 1515 mutex_obj_hold(lock); 1516 mutex_enter(lock); 1517 } 1518 KASSERT(solocked(so)); 1519 } 1520 1521 /* 1522 * Set lock on sockbuf sb; sleep if lock is already held. 1523 * Unless SB_NOINTR is set on sockbuf, sleep is interruptible. 1524 * Returns error without lock if sleep is interrupted. 1525 */ 1526 int 1527 sblock(struct sockbuf *sb, int wf) 1528 { 1529 struct socket *so; 1530 kmutex_t *lock; 1531 int error; 1532 1533 KASSERT(solocked(sb->sb_so)); 1534 1535 for (;;) { 1536 if (__predict_true((sb->sb_flags & SB_LOCK) == 0)) { 1537 sb->sb_flags |= SB_LOCK; 1538 return 0; 1539 } 1540 if (wf != M_WAITOK) 1541 return EWOULDBLOCK; 1542 so = sb->sb_so; 1543 lock = so->so_lock; 1544 if ((sb->sb_flags & SB_NOINTR) != 0) { 1545 cv_wait(&so->so_cv, lock); 1546 error = 0; 1547 } else 1548 error = cv_wait_sig(&so->so_cv, lock); 1549 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock))) 1550 solockretry(so, lock); 1551 if (error != 0) 1552 return error; 1553 } 1554 } 1555 1556 void 1557 sbunlock(struct sockbuf *sb) 1558 { 1559 struct socket *so; 1560 1561 so = sb->sb_so; 1562 1563 KASSERT(solocked(so)); 1564 KASSERT((sb->sb_flags & SB_LOCK) != 0); 1565 1566 sb->sb_flags &= ~SB_LOCK; 1567 cv_broadcast(&so->so_cv); 1568 } 1569 1570 int 1571 sowait(struct socket *so, bool catch_p, int timo) 1572 { 1573 kmutex_t *lock; 1574 int error; 1575 1576 KASSERT(solocked(so)); 1577 KASSERT(catch_p || timo != 0); 1578 1579 lock = so->so_lock; 1580 if (catch_p) 1581 error = cv_timedwait_sig(&so->so_cv, lock, timo); 1582 else 1583 error = cv_timedwait(&so->so_cv, lock, timo); 1584 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock))) 1585 solockretry(so, lock); 1586 return error; 1587 } 1588 1589 #ifdef DDB 1590 1591 /* 1592 * Currently, sofindproc() is used only from DDB. It could be used from others 1593 * by using db_mutex_enter() 1594 */ 1595 1596 static inline int 1597 db_mutex_enter(kmutex_t *mtx) 1598 { 1599 extern int db_active; 1600 int rv; 1601 1602 if (!db_active) { 1603 mutex_enter(mtx); 1604 rv = 1; 1605 } else 1606 rv = mutex_tryenter(mtx); 1607 1608 return rv; 1609 } 1610 1611 int 1612 sofindproc(struct socket *so, int all, void (*pr)(const char *, ...)) 1613 { 1614 proc_t *p; 1615 filedesc_t *fdp; 1616 fdtab_t *dt; 1617 fdfile_t *ff; 1618 file_t *fp = NULL; 1619 int found = 0; 1620 int i, t; 1621 1622 if (so == NULL) 1623 return 0; 1624 1625 t = db_mutex_enter(&proc_lock); 1626 if (!t) { 1627 pr("could not acquire proc_lock mutex\n"); 1628 return 0; 1629 } 1630 PROCLIST_FOREACH(p, &allproc) { 1631 if (p->p_stat == SIDL) 1632 continue; 1633 fdp = p->p_fd; 1634 t = db_mutex_enter(&fdp->fd_lock); 1635 if (!t) { 1636 pr("could not acquire fd_lock mutex\n"); 1637 continue; 1638 } 1639 dt = atomic_load_consume(&fdp->fd_dt); 1640 for (i = 0; i < dt->dt_nfiles; i++) { 1641 ff = dt->dt_ff[i]; 1642 if (ff == NULL) 1643 continue; 1644 1645 fp = atomic_load_consume(&ff->ff_file); 1646 if (fp == NULL) 1647 continue; 1648 1649 t = db_mutex_enter(&fp->f_lock); 1650 if (!t) { 1651 pr("could not acquire f_lock mutex\n"); 1652 continue; 1653 } 1654 if ((struct socket *)fp->f_data != so) { 1655 mutex_exit(&fp->f_lock); 1656 continue; 1657 } 1658 found++; 1659 if (pr) 1660 pr("socket %p: owner %s(pid=%d)\n", 1661 so, p->p_comm, p->p_pid); 1662 mutex_exit(&fp->f_lock); 1663 if (all == 0) 1664 break; 1665 } 1666 mutex_exit(&fdp->fd_lock); 1667 if (all == 0 && found != 0) 1668 break; 1669 } 1670 mutex_exit(&proc_lock); 1671 1672 return found; 1673 } 1674 1675 void 1676 socket_print(const char *modif, void (*pr)(const char *, ...)) 1677 { 1678 file_t *fp; 1679 struct socket *so; 1680 struct sockbuf *sb_snd, *sb_rcv; 1681 struct mbuf *m_rec, *m; 1682 bool opt_v = false; 1683 bool opt_m = false; 1684 bool opt_a = false; 1685 bool opt_p = false; 1686 int nrecs, nmbufs; 1687 char ch; 1688 const char *family; 1689 1690 while ( (ch = *(modif++)) != '\0') { 1691 switch (ch) { 1692 case 'v': 1693 opt_v = true; 1694 break; 1695 case 'm': 1696 opt_m = true; 1697 break; 1698 case 'a': 1699 opt_a = true; 1700 break; 1701 case 'p': 1702 opt_p = true; 1703 break; 1704 } 1705 } 1706 if (opt_v == false && pr) 1707 (pr)("Ignore empty sockets. use /v to print all.\n"); 1708 if (opt_p == true && pr) 1709 (pr)("Don't search owner process.\n"); 1710 1711 LIST_FOREACH(fp, &filehead, f_list) { 1712 if (fp->f_type != DTYPE_SOCKET) 1713 continue; 1714 so = (struct socket *)fp->f_data; 1715 if (so == NULL) 1716 continue; 1717 1718 if (so->so_proto->pr_domain->dom_family == AF_INET) 1719 family = "INET"; 1720 #ifdef INET6 1721 else if (so->so_proto->pr_domain->dom_family == AF_INET6) 1722 family = "INET6"; 1723 #endif 1724 else if (so->so_proto->pr_domain->dom_family == pseudo_AF_KEY) 1725 family = "KEY"; 1726 else if (so->so_proto->pr_domain->dom_family == AF_ROUTE) 1727 family = "ROUTE"; 1728 else 1729 continue; 1730 1731 sb_snd = &so->so_snd; 1732 sb_rcv = &so->so_rcv; 1733 1734 if (opt_v != true && 1735 sb_snd->sb_cc == 0 && sb_rcv->sb_cc == 0) 1736 continue; 1737 1738 pr("---SOCKET %p: type %s\n", so, family); 1739 if (opt_p != true) 1740 sofindproc(so, opt_a == true ? 1 : 0, pr); 1741 pr("Send Buffer Bytes: %d [bytes]\n", sb_snd->sb_cc); 1742 pr("Send Buffer mbufs:\n"); 1743 m_rec = m = sb_snd->sb_mb; 1744 nrecs = 0; 1745 nmbufs = 0; 1746 while (m_rec) { 1747 nrecs++; 1748 if (opt_m == true) 1749 pr(" mbuf chain %p\n", m_rec); 1750 while (m) { 1751 nmbufs++; 1752 m = m->m_next; 1753 } 1754 m_rec = m = m_rec->m_nextpkt; 1755 } 1756 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1757 1758 pr("Recv Buffer Usage: %d [bytes]\n", sb_rcv->sb_cc); 1759 pr("Recv Buffer mbufs:\n"); 1760 m_rec = m = sb_rcv->sb_mb; 1761 nrecs = 0; 1762 nmbufs = 0; 1763 while (m_rec) { 1764 nrecs++; 1765 if (opt_m == true) 1766 pr(" mbuf chain %p\n", m_rec); 1767 while (m) { 1768 nmbufs++; 1769 m = m->m_next; 1770 } 1771 m_rec = m = m_rec->m_nextpkt; 1772 } 1773 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs); 1774 } 1775 } 1776 #endif /* DDB */ 1777