uipc_usrreq.c revision 160619
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright 2004-2006 Robert N. M. Watson 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 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34/* 35 * UNIX Domain (Local) Sockets 36 * 37 * This is an implementation of UNIX (local) domain sockets. Each socket has 38 * an associated struct unpcb (UNIX protocol control block). Stream sockets 39 * may be connected to 0 or 1 other socket. Datagram sockets may be 40 * connected to 0, 1, or many other sockets. Sockets may be created and 41 * connected in pairs (socketpair(2)), or bound/connected to using the file 42 * system name space. For most purposes, only the receive socket buffer is 43 * used, as sending on one socket delivers directly to the receive socket 44 * buffer of a second socket. The implementation is substantially 45 * complicated by the fact that "ancillary data", such as file descriptors or 46 * credentials, may be passed across UNIX domain sockets. The potential for 47 * passing UNIX domain sockets over other UNIX domain sockets requires the 48 * implementation of a simple garbage collector to find and tear down cycles 49 * of disconnected sockets. 50 */ 51 52#include <sys/cdefs.h> 53__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 160619 2006-07-24 15:20:08Z rwatson $"); 54 55#include "opt_mac.h" 56 57#include <sys/param.h> 58#include <sys/domain.h> 59#include <sys/fcntl.h> 60#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 61#include <sys/eventhandler.h> 62#include <sys/file.h> 63#include <sys/filedesc.h> 64#include <sys/jail.h> 65#include <sys/kernel.h> 66#include <sys/lock.h> 67#include <sys/mac.h> 68#include <sys/mbuf.h> 69#include <sys/mount.h> 70#include <sys/mutex.h> 71#include <sys/namei.h> 72#include <sys/proc.h> 73#include <sys/protosw.h> 74#include <sys/resourcevar.h> 75#include <sys/socket.h> 76#include <sys/socketvar.h> 77#include <sys/signalvar.h> 78#include <sys/stat.h> 79#include <sys/sx.h> 80#include <sys/sysctl.h> 81#include <sys/systm.h> 82#include <sys/taskqueue.h> 83#include <sys/un.h> 84#include <sys/unpcb.h> 85#include <sys/vnode.h> 86 87#include <vm/uma.h> 88 89static uma_zone_t unp_zone; 90static unp_gen_t unp_gencnt; 91static u_int unp_count; 92 93static struct unp_head unp_shead, unp_dhead; 94 95/* 96 * Unix communications domain. 97 * 98 * TODO: 99 * SEQPACKET, RDM 100 * rethink name space problems 101 * need a proper out-of-band 102 * lock pushdown 103 */ 104static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 105static ino_t unp_ino; /* prototype for fake inode numbers */ 106struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 107 108/* 109 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 110 * stream sockets, although the total for sender and receiver is actually 111 * only PIPSIZ. 112 * 113 * Datagram sockets really use the sendspace as the maximum datagram size, 114 * and don't really want to reserve the sendspace. Their recvspace should be 115 * large enough for at least one max-size datagram plus address. 116 */ 117#ifndef PIPSIZ 118#define PIPSIZ 8192 119#endif 120static u_long unpst_sendspace = PIPSIZ; 121static u_long unpst_recvspace = PIPSIZ; 122static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 123static u_long unpdg_recvspace = 4*1024; 124 125static int unp_rights; /* file descriptors in flight */ 126 127SYSCTL_DECL(_net_local_stream); 128SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 129 &unpst_sendspace, 0, ""); 130SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 131 &unpst_recvspace, 0, ""); 132SYSCTL_DECL(_net_local_dgram); 133SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 134 &unpdg_sendspace, 0, ""); 135SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 136 &unpdg_recvspace, 0, ""); 137SYSCTL_DECL(_net_local); 138SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 139 140/* 141 * Currently, UNIX domain sockets are protected by a single subsystem lock, 142 * which covers global data structures and variables, the contents of each 143 * per-socket unpcb structure, and the so_pcb field in sockets attached to 144 * the UNIX domain. This provides for a moderate degree of paralellism, as 145 * receive operations on UNIX domain sockets do not need to acquire the 146 * subsystem lock. Finer grained locking to permit send() without acquiring 147 * a global lock would be a logical next step. 148 * 149 * The UNIX domain socket lock preceds all socket layer locks, including the 150 * socket lock and socket buffer lock, permitting UNIX domain socket code to 151 * call into socket support routines without releasing its locks. 152 * 153 * Some caution is required in areas where the UNIX domain socket code enters 154 * VFS in order to create or find rendezvous points. This results in 155 * dropping of the UNIX domain socket subsystem lock, acquisition of the 156 * Giant lock, and potential sleeping. This increases the chances of races, 157 * and exposes weaknesses in the socket->protocol API by offering poor 158 * failure modes. 159 */ 160static struct mtx unp_mtx; 161#define UNP_LOCK_INIT() \ 162 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF) 163#define UNP_LOCK() mtx_lock(&unp_mtx) 164#define UNP_UNLOCK() mtx_unlock(&unp_mtx) 165#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) 166#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED) 167 168/* 169 * Garbage collection of cyclic file descriptor/socket references occurs 170 * asynchronously in a taskqueue context in order to avoid recursion and 171 * reentrance in the UNIX domain socket, file descriptor, and socket layer 172 * code. See unp_gc() for a full description. 173 */ 174static struct task unp_gc_task; 175 176static int unp_connect(struct socket *,struct sockaddr *, struct thread *); 177static int unp_connect2(struct socket *so, struct socket *so2, int); 178static void unp_disconnect(struct unpcb *); 179static void unp_shutdown(struct unpcb *); 180static void unp_drop(struct unpcb *, int); 181static void unp_gc(__unused void *, int); 182static void unp_scan(struct mbuf *, void (*)(struct file *)); 183static void unp_mark(struct file *); 184static void unp_discard(struct file *); 185static void unp_freerights(struct file **, int); 186static int unp_internalize(struct mbuf **, struct thread *); 187static int unp_listen(struct socket *, struct unpcb *, int, 188 struct thread *); 189 190static void 191uipc_abort(struct socket *so) 192{ 193 struct unpcb *unp; 194 195 unp = sotounpcb(so); 196 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 197 UNP_LOCK(); 198 unp_drop(unp, ECONNABORTED); 199 UNP_UNLOCK(); 200} 201 202static int 203uipc_accept(struct socket *so, struct sockaddr **nam) 204{ 205 struct unpcb *unp; 206 const struct sockaddr *sa; 207 208 /* 209 * Pass back name of connected socket, if it was bound and we are 210 * still connected (our peer may have closed already!). 211 */ 212 unp = sotounpcb(so); 213 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 214 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 215 UNP_LOCK(); 216 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) 217 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 218 else 219 sa = &sun_noname; 220 bcopy(sa, *nam, sa->sa_len); 221 UNP_UNLOCK(); 222 return (0); 223} 224 225static int 226uipc_attach(struct socket *so, int proto, struct thread *td) 227{ 228 struct unpcb *unp; 229 int error; 230 231 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 232 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 233 switch (so->so_type) { 234 235 case SOCK_STREAM: 236 error = soreserve(so, unpst_sendspace, unpst_recvspace); 237 break; 238 239 case SOCK_DGRAM: 240 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 241 break; 242 243 default: 244 panic("unp_attach"); 245 } 246 if (error) 247 return (error); 248 } 249 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO); 250 if (unp == NULL) 251 return (ENOBUFS); 252 LIST_INIT(&unp->unp_refs); 253 unp->unp_socket = so; 254 so->so_pcb = unp; 255 256 UNP_LOCK(); 257 unp->unp_gencnt = ++unp_gencnt; 258 unp_count++; 259 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead 260 : &unp_shead, unp, unp_link); 261 UNP_UNLOCK(); 262 263 return (0); 264} 265 266static int 267uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 268{ 269 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 270 struct vattr vattr; 271 int error, namelen; 272 struct nameidata nd; 273 struct unpcb *unp; 274 struct vnode *vp; 275 struct mount *mp; 276 char *buf; 277 278 unp = sotounpcb(so); 279 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 280 281 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 282 if (namelen <= 0) 283 return (EINVAL); 284 285 /* 286 * We don't allow simultaneous bind() calls on a single UNIX domain 287 * socket, so flag in-progress operations, and return an error if an 288 * operation is already in progress. 289 * 290 * Historically, we have not allowed a socket to be rebound, so this 291 * also returns an error. Not allowing re-binding certainly 292 * simplifies the implementation and avoids a great many possible 293 * failure modes. 294 */ 295 UNP_LOCK(); 296 if (unp->unp_vnode != NULL) { 297 UNP_UNLOCK(); 298 return (EINVAL); 299 } 300 if (unp->unp_flags & UNP_BINDING) { 301 UNP_UNLOCK(); 302 return (EALREADY); 303 } 304 unp->unp_flags |= UNP_BINDING; 305 UNP_UNLOCK(); 306 307 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 308 strlcpy(buf, soun->sun_path, namelen + 1); 309 310 mtx_lock(&Giant); 311restart: 312 mtx_assert(&Giant, MA_OWNED); 313 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, 314 buf, td); 315/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 316 error = namei(&nd); 317 if (error) 318 goto error; 319 vp = nd.ni_vp; 320 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 321 NDFREE(&nd, NDF_ONLY_PNBUF); 322 if (nd.ni_dvp == vp) 323 vrele(nd.ni_dvp); 324 else 325 vput(nd.ni_dvp); 326 if (vp != NULL) { 327 vrele(vp); 328 error = EADDRINUSE; 329 goto error; 330 } 331 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 332 if (error) 333 goto error; 334 goto restart; 335 } 336 VATTR_NULL(&vattr); 337 vattr.va_type = VSOCK; 338 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 339#ifdef MAC 340 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 341 &vattr); 342#endif 343 if (error == 0) { 344 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 345 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 346 } 347 NDFREE(&nd, NDF_ONLY_PNBUF); 348 vput(nd.ni_dvp); 349 if (error) { 350 vn_finished_write(mp); 351 goto error; 352 } 353 vp = nd.ni_vp; 354 ASSERT_VOP_LOCKED(vp, "uipc_bind"); 355 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 356 UNP_LOCK(); 357 vp->v_socket = unp->unp_socket; 358 unp->unp_vnode = vp; 359 unp->unp_addr = soun; 360 unp->unp_flags &= ~UNP_BINDING; 361 UNP_UNLOCK(); 362 VOP_UNLOCK(vp, 0, td); 363 vn_finished_write(mp); 364 mtx_unlock(&Giant); 365 free(buf, M_TEMP); 366 return (0); 367error: 368 UNP_LOCK(); 369 unp->unp_flags &= ~UNP_BINDING; 370 UNP_UNLOCK(); 371 mtx_unlock(&Giant); 372 free(buf, M_TEMP); 373 return (error); 374} 375 376static int 377uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 378{ 379 int error; 380 381 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 382 UNP_LOCK(); 383 error = unp_connect(so, nam, td); 384 UNP_UNLOCK(); 385 return (error); 386} 387 388/* 389 * XXXRW: Should also unbind? 390 */ 391static void 392uipc_close(struct socket *so) 393{ 394 struct unpcb *unp; 395 396 unp = sotounpcb(so); 397 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 398 UNP_LOCK(); 399 unp_disconnect(unp); 400 UNP_UNLOCK(); 401} 402 403int 404uipc_connect2(struct socket *so1, struct socket *so2) 405{ 406 struct unpcb *unp; 407 int error; 408 409 unp = sotounpcb(so1); 410 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 411 UNP_LOCK(); 412 error = unp_connect2(so1, so2, PRU_CONNECT2); 413 UNP_UNLOCK(); 414 return (error); 415} 416 417/* control is EOPNOTSUPP */ 418 419static void 420uipc_detach(struct socket *so) 421{ 422 int local_unp_rights; 423 struct unpcb *unp; 424 struct vnode *vp; 425 426 unp = sotounpcb(so); 427 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 428 UNP_LOCK(); 429 LIST_REMOVE(unp, unp_link); 430 unp->unp_gencnt = ++unp_gencnt; 431 --unp_count; 432 if ((vp = unp->unp_vnode) != NULL) { 433 /* 434 * XXXRW: should v_socket be frobbed only while holding 435 * Giant? 436 */ 437 unp->unp_vnode->v_socket = NULL; 438 unp->unp_vnode = NULL; 439 } 440 if (unp->unp_conn != NULL) 441 unp_disconnect(unp); 442 while (!LIST_EMPTY(&unp->unp_refs)) { 443 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 444 unp_drop(ref, ECONNRESET); 445 } 446 soisdisconnected(unp->unp_socket); 447 unp->unp_socket->so_pcb = NULL; 448 local_unp_rights = unp_rights; 449 UNP_UNLOCK(); 450 if (unp->unp_addr != NULL) 451 FREE(unp->unp_addr, M_SONAME); 452 uma_zfree(unp_zone, unp); 453 if (vp) { 454 int vfslocked; 455 456 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 457 vrele(vp); 458 VFS_UNLOCK_GIANT(vfslocked); 459 } 460 if (local_unp_rights) 461 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 462} 463 464static int 465uipc_disconnect(struct socket *so) 466{ 467 struct unpcb *unp; 468 469 unp = sotounpcb(so); 470 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 471 UNP_LOCK(); 472 unp_disconnect(unp); 473 UNP_UNLOCK(); 474 return (0); 475} 476 477static int 478uipc_listen(struct socket *so, int backlog, struct thread *td) 479{ 480 struct unpcb *unp; 481 int error; 482 483 unp = sotounpcb(so); 484 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 485 UNP_LOCK(); 486 if (unp->unp_vnode == NULL) { 487 UNP_UNLOCK(); 488 return (EINVAL); 489 } 490 error = unp_listen(so, unp, backlog, td); 491 UNP_UNLOCK(); 492 return (error); 493} 494 495static int 496uipc_peeraddr(struct socket *so, struct sockaddr **nam) 497{ 498 struct unpcb *unp; 499 const struct sockaddr *sa; 500 501 unp = sotounpcb(so); 502 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 503 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 504 UNP_LOCK(); 505 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL) 506 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 507 else { 508 /* 509 * XXX: It seems that this test always fails even when 510 * connection is established. So, this else clause is 511 * added as workaround to return PF_LOCAL sockaddr. 512 */ 513 sa = &sun_noname; 514 } 515 bcopy(sa, *nam, sa->sa_len); 516 UNP_UNLOCK(); 517 return (0); 518} 519 520static int 521uipc_rcvd(struct socket *so, int flags) 522{ 523 struct unpcb *unp; 524 struct socket *so2; 525 u_int mbcnt, sbcc; 526 u_long newhiwat; 527 528 unp = sotounpcb(so); 529 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 530 switch (so->so_type) { 531 case SOCK_DGRAM: 532 panic("uipc_rcvd DGRAM?"); 533 /*NOTREACHED*/ 534 535 case SOCK_STREAM: 536 /* 537 * Adjust backpressure on sender and wakeup any waiting to 538 * write. 539 */ 540 SOCKBUF_LOCK(&so->so_rcv); 541 mbcnt = so->so_rcv.sb_mbcnt; 542 sbcc = so->so_rcv.sb_cc; 543 SOCKBUF_UNLOCK(&so->so_rcv); 544 UNP_LOCK(); 545 if (unp->unp_conn == NULL) { 546 UNP_UNLOCK(); 547 break; 548 } 549 so2 = unp->unp_conn->unp_socket; 550 SOCKBUF_LOCK(&so2->so_snd); 551 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 552 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 553 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 554 newhiwat, RLIM_INFINITY); 555 sowwakeup_locked(so2); 556 unp->unp_mbcnt = mbcnt; 557 unp->unp_cc = sbcc; 558 UNP_UNLOCK(); 559 break; 560 561 default: 562 panic("uipc_rcvd unknown socktype"); 563 } 564 return (0); 565} 566 567/* pru_rcvoob is EOPNOTSUPP */ 568 569static int 570uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 571 struct mbuf *control, struct thread *td) 572{ 573 struct unpcb *unp, *unp2; 574 struct socket *so2; 575 u_int mbcnt, sbcc; 576 u_long newhiwat; 577 int error = 0; 578 579 unp = sotounpcb(so); 580 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 581 if (flags & PRUS_OOB) { 582 error = EOPNOTSUPP; 583 goto release; 584 } 585 586 if (control != NULL && (error = unp_internalize(&control, td))) 587 goto release; 588 589 UNP_LOCK(); 590 switch (so->so_type) { 591 case SOCK_DGRAM: 592 { 593 const struct sockaddr *from; 594 595 if (nam != NULL) { 596 if (unp->unp_conn != NULL) { 597 error = EISCONN; 598 break; 599 } 600 error = unp_connect(so, nam, td); 601 if (error) 602 break; 603 } else { 604 if (unp->unp_conn == NULL) { 605 error = ENOTCONN; 606 break; 607 } 608 } 609 unp2 = unp->unp_conn; 610 so2 = unp2->unp_socket; 611 if (unp->unp_addr != NULL) 612 from = (struct sockaddr *)unp->unp_addr; 613 else 614 from = &sun_noname; 615 if (unp2->unp_flags & UNP_WANTCRED) 616 control = unp_addsockcred(td, control); 617 SOCKBUF_LOCK(&so2->so_rcv); 618 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 619 sorwakeup_locked(so2); 620 m = NULL; 621 control = NULL; 622 } else { 623 SOCKBUF_UNLOCK(&so2->so_rcv); 624 error = ENOBUFS; 625 } 626 if (nam != NULL) 627 unp_disconnect(unp); 628 break; 629 } 630 631 case SOCK_STREAM: 632 /* 633 * Connect if not connected yet. 634 * 635 * Note: A better implementation would complain if not equal 636 * to the peer's address. 637 */ 638 if ((so->so_state & SS_ISCONNECTED) == 0) { 639 if (nam != NULL) { 640 error = unp_connect(so, nam, td); 641 if (error) 642 break; /* XXX */ 643 } else { 644 error = ENOTCONN; 645 break; 646 } 647 } 648 649 /* Lockless read. */ 650 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 651 error = EPIPE; 652 break; 653 } 654 unp2 = unp->unp_conn; 655 if (unp2 == NULL) 656 panic("uipc_send connected but no connection?"); 657 so2 = unp2->unp_socket; 658 SOCKBUF_LOCK(&so2->so_rcv); 659 if (unp2->unp_flags & UNP_WANTCRED) { 660 /* 661 * Credentials are passed only once on 662 * SOCK_STREAM. 663 */ 664 unp2->unp_flags &= ~UNP_WANTCRED; 665 control = unp_addsockcred(td, control); 666 } 667 /* 668 * Send to paired receive port, and then reduce send buffer 669 * hiwater marks to maintain backpressure. Wake up readers. 670 */ 671 if (control != NULL) { 672 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 673 control = NULL; 674 } else { 675 sbappend_locked(&so2->so_rcv, m); 676 } 677 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 678 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 679 sbcc = so2->so_rcv.sb_cc; 680 sorwakeup_locked(so2); 681 682 SOCKBUF_LOCK(&so->so_snd); 683 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 684 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 685 newhiwat, RLIM_INFINITY); 686 so->so_snd.sb_mbmax -= mbcnt; 687 SOCKBUF_UNLOCK(&so->so_snd); 688 689 unp2->unp_cc = sbcc; 690 m = NULL; 691 break; 692 693 default: 694 panic("uipc_send unknown socktype"); 695 } 696 697 /* 698 * SEND_EOF is equivalent to a SEND followed by 699 * a SHUTDOWN. 700 */ 701 if (flags & PRUS_EOF) { 702 socantsendmore(so); 703 unp_shutdown(unp); 704 } 705 UNP_UNLOCK(); 706 707 if (control != NULL && error != 0) 708 unp_dispose(control); 709 710release: 711 if (control != NULL) 712 m_freem(control); 713 if (m != NULL) 714 m_freem(m); 715 return (error); 716} 717 718static int 719uipc_sense(struct socket *so, struct stat *sb) 720{ 721 struct unpcb *unp; 722 struct socket *so2; 723 724 unp = sotounpcb(so); 725 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 726 UNP_LOCK(); 727 sb->st_blksize = so->so_snd.sb_hiwat; 728 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) { 729 so2 = unp->unp_conn->unp_socket; 730 sb->st_blksize += so2->so_rcv.sb_cc; 731 } 732 sb->st_dev = NODEV; 733 if (unp->unp_ino == 0) 734 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 735 sb->st_ino = unp->unp_ino; 736 UNP_UNLOCK(); 737 return (0); 738} 739 740static int 741uipc_shutdown(struct socket *so) 742{ 743 struct unpcb *unp; 744 745 unp = sotounpcb(so); 746 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 747 UNP_LOCK(); 748 socantsendmore(so); 749 unp_shutdown(unp); 750 UNP_UNLOCK(); 751 return (0); 752} 753 754static int 755uipc_sockaddr(struct socket *so, struct sockaddr **nam) 756{ 757 struct unpcb *unp; 758 const struct sockaddr *sa; 759 760 unp = sotounpcb(so); 761 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 762 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 763 UNP_LOCK(); 764 if (unp->unp_addr != NULL) 765 sa = (struct sockaddr *) unp->unp_addr; 766 else 767 sa = &sun_noname; 768 bcopy(sa, *nam, sa->sa_len); 769 UNP_UNLOCK(); 770 return (0); 771} 772 773struct pr_usrreqs uipc_usrreqs = { 774 .pru_abort = uipc_abort, 775 .pru_accept = uipc_accept, 776 .pru_attach = uipc_attach, 777 .pru_bind = uipc_bind, 778 .pru_connect = uipc_connect, 779 .pru_connect2 = uipc_connect2, 780 .pru_detach = uipc_detach, 781 .pru_disconnect = uipc_disconnect, 782 .pru_listen = uipc_listen, 783 .pru_peeraddr = uipc_peeraddr, 784 .pru_rcvd = uipc_rcvd, 785 .pru_send = uipc_send, 786 .pru_sense = uipc_sense, 787 .pru_shutdown = uipc_shutdown, 788 .pru_sockaddr = uipc_sockaddr, 789 .pru_sosend = sosend_generic, 790 .pru_soreceive = soreceive_generic, 791 .pru_sopoll = sopoll_generic, 792 .pru_close = uipc_close, 793}; 794 795int 796uipc_ctloutput(struct socket *so, struct sockopt *sopt) 797{ 798 struct unpcb *unp; 799 struct xucred xu; 800 int error, optval; 801 802 if (sopt->sopt_level != 0) 803 return (EINVAL); 804 805 unp = sotounpcb(so); 806 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 807 UNP_LOCK(); 808 error = 0; 809 switch (sopt->sopt_dir) { 810 case SOPT_GET: 811 switch (sopt->sopt_name) { 812 case LOCAL_PEERCRED: 813 if (unp->unp_flags & UNP_HAVEPC) 814 xu = unp->unp_peercred; 815 else { 816 if (so->so_type == SOCK_STREAM) 817 error = ENOTCONN; 818 else 819 error = EINVAL; 820 } 821 if (error == 0) 822 error = sooptcopyout(sopt, &xu, sizeof(xu)); 823 break; 824 case LOCAL_CREDS: 825 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 826 error = sooptcopyout(sopt, &optval, sizeof(optval)); 827 break; 828 case LOCAL_CONNWAIT: 829 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 830 error = sooptcopyout(sopt, &optval, sizeof(optval)); 831 break; 832 default: 833 error = EOPNOTSUPP; 834 break; 835 } 836 break; 837 case SOPT_SET: 838 switch (sopt->sopt_name) { 839 case LOCAL_CREDS: 840 case LOCAL_CONNWAIT: 841 error = sooptcopyin(sopt, &optval, sizeof(optval), 842 sizeof(optval)); 843 if (error) 844 break; 845 846#define OPTSET(bit) \ 847 if (optval) \ 848 unp->unp_flags |= bit; \ 849 else \ 850 unp->unp_flags &= ~bit; 851 852 switch (sopt->sopt_name) { 853 case LOCAL_CREDS: 854 OPTSET(UNP_WANTCRED); 855 break; 856 case LOCAL_CONNWAIT: 857 OPTSET(UNP_CONNWAIT); 858 break; 859 default: 860 break; 861 } 862 break; 863#undef OPTSET 864 default: 865 error = ENOPROTOOPT; 866 break; 867 } 868 break; 869 default: 870 error = EOPNOTSUPP; 871 break; 872 } 873 UNP_UNLOCK(); 874 return (error); 875} 876 877static int 878unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 879{ 880 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 881 struct vnode *vp; 882 struct socket *so2, *so3; 883 struct unpcb *unp, *unp2, *unp3; 884 int error, len; 885 struct nameidata nd; 886 char buf[SOCK_MAXADDRLEN]; 887 struct sockaddr *sa; 888 889 UNP_LOCK_ASSERT(); 890 891 unp = sotounpcb(so); 892 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 893 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 894 if (len <= 0) 895 return (EINVAL); 896 strlcpy(buf, soun->sun_path, len + 1); 897 if (unp->unp_flags & UNP_CONNECTING) { 898 UNP_UNLOCK(); 899 return (EALREADY); 900 } 901 UNP_UNLOCK(); 902 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 903 mtx_lock(&Giant); 904 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); 905 error = namei(&nd); 906 if (error) 907 vp = NULL; 908 else 909 vp = nd.ni_vp; 910 ASSERT_VOP_LOCKED(vp, "unp_connect"); 911 NDFREE(&nd, NDF_ONLY_PNBUF); 912 if (error) 913 goto bad; 914 915 if (vp->v_type != VSOCK) { 916 error = ENOTSOCK; 917 goto bad; 918 } 919 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 920 if (error) 921 goto bad; 922 mtx_unlock(&Giant); 923 UNP_LOCK(); 924 unp = sotounpcb(so); 925 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 926 so2 = vp->v_socket; 927 if (so2 == NULL) { 928 error = ECONNREFUSED; 929 goto bad2; 930 } 931 if (so->so_type != so2->so_type) { 932 error = EPROTOTYPE; 933 goto bad2; 934 } 935 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 936 if (so2->so_options & SO_ACCEPTCONN) { 937 /* 938 * NB: drop locks here so unp_attach is entered w/o 939 * locks; this avoids a recursive lock of the head 940 * and holding sleep locks across a (potentially) 941 * blocking malloc. 942 */ 943 UNP_UNLOCK(); 944 so3 = sonewconn(so2, 0); 945 UNP_LOCK(); 946 } else 947 so3 = NULL; 948 if (so3 == NULL) { 949 error = ECONNREFUSED; 950 goto bad2; 951 } 952 unp = sotounpcb(so); 953 unp2 = sotounpcb(so2); 954 unp3 = sotounpcb(so3); 955 if (unp2->unp_addr != NULL) { 956 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 957 unp3->unp_addr = (struct sockaddr_un *) sa; 958 sa = NULL; 959 } 960 /* 961 * unp_peercred management: 962 * 963 * The connecter's (client's) credentials are copied from its 964 * process structure at the time of connect() (which is now). 965 */ 966 cru2x(td->td_ucred, &unp3->unp_peercred); 967 unp3->unp_flags |= UNP_HAVEPC; 968 /* 969 * The receiver's (server's) credentials are copied from the 970 * unp_peercred member of socket on which the former called 971 * listen(); unp_listen() cached that process's credentials 972 * at that time so we can use them now. 973 */ 974 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 975 ("unp_connect: listener without cached peercred")); 976 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 977 sizeof(unp->unp_peercred)); 978 unp->unp_flags |= UNP_HAVEPC; 979 if (unp2->unp_flags & UNP_WANTCRED) 980 unp3->unp_flags |= UNP_WANTCRED; 981#ifdef MAC 982 SOCK_LOCK(so); 983 mac_set_socket_peer_from_socket(so, so3); 984 mac_set_socket_peer_from_socket(so3, so); 985 SOCK_UNLOCK(so); 986#endif 987 988 so2 = so3; 989 } 990 error = unp_connect2(so, so2, PRU_CONNECT); 991bad2: 992 UNP_UNLOCK(); 993 mtx_lock(&Giant); 994bad: 995 mtx_assert(&Giant, MA_OWNED); 996 if (vp != NULL) 997 vput(vp); 998 mtx_unlock(&Giant); 999 free(sa, M_SONAME); 1000 UNP_LOCK(); 1001 unp->unp_flags &= ~UNP_CONNECTING; 1002 return (error); 1003} 1004 1005static int 1006unp_connect2(struct socket *so, struct socket *so2, int req) 1007{ 1008 struct unpcb *unp = sotounpcb(so); 1009 struct unpcb *unp2; 1010 1011 UNP_LOCK_ASSERT(); 1012 1013 if (so2->so_type != so->so_type) 1014 return (EPROTOTYPE); 1015 unp2 = sotounpcb(so2); 1016 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1017 unp->unp_conn = unp2; 1018 switch (so->so_type) { 1019 1020 case SOCK_DGRAM: 1021 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1022 soisconnected(so); 1023 break; 1024 1025 case SOCK_STREAM: 1026 unp2->unp_conn = unp; 1027 if (req == PRU_CONNECT && 1028 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1029 soisconnecting(so); 1030 else 1031 soisconnected(so); 1032 soisconnected(so2); 1033 break; 1034 1035 default: 1036 panic("unp_connect2"); 1037 } 1038 return (0); 1039} 1040 1041static void 1042unp_disconnect(struct unpcb *unp) 1043{ 1044 struct unpcb *unp2 = unp->unp_conn; 1045 struct socket *so; 1046 1047 UNP_LOCK_ASSERT(); 1048 1049 if (unp2 == NULL) 1050 return; 1051 unp->unp_conn = NULL; 1052 switch (unp->unp_socket->so_type) { 1053 case SOCK_DGRAM: 1054 LIST_REMOVE(unp, unp_reflink); 1055 so = unp->unp_socket; 1056 SOCK_LOCK(so); 1057 so->so_state &= ~SS_ISCONNECTED; 1058 SOCK_UNLOCK(so); 1059 break; 1060 1061 case SOCK_STREAM: 1062 soisdisconnected(unp->unp_socket); 1063 unp2->unp_conn = NULL; 1064 soisdisconnected(unp2->unp_socket); 1065 break; 1066 } 1067} 1068 1069/* 1070 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by 1071 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are 1072 * safe to reference. It first scans the list of struct unpcb's to generate 1073 * a pointer list, then it rescans its list one entry at a time to 1074 * externalize and copyout. It checks the generation number to see if a 1075 * struct unpcb has been reused, and will skip it if so. 1076 */ 1077static int 1078unp_pcblist(SYSCTL_HANDLER_ARGS) 1079{ 1080 int error, i, n; 1081 struct unpcb *unp, **unp_list; 1082 unp_gen_t gencnt; 1083 struct xunpgen *xug; 1084 struct unp_head *head; 1085 struct xunpcb *xu; 1086 1087 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1088 1089 /* 1090 * The process of preparing the PCB list is too time-consuming and 1091 * resource-intensive to repeat twice on every request. 1092 */ 1093 if (req->oldptr == NULL) { 1094 n = unp_count; 1095 req->oldidx = 2 * (sizeof *xug) 1096 + (n + n/8) * sizeof(struct xunpcb); 1097 return (0); 1098 } 1099 1100 if (req->newptr != NULL) 1101 return (EPERM); 1102 1103 /* 1104 * OK, now we're committed to doing something. 1105 */ 1106 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1107 UNP_LOCK(); 1108 gencnt = unp_gencnt; 1109 n = unp_count; 1110 UNP_UNLOCK(); 1111 1112 xug->xug_len = sizeof *xug; 1113 xug->xug_count = n; 1114 xug->xug_gen = gencnt; 1115 xug->xug_sogen = so_gencnt; 1116 error = SYSCTL_OUT(req, xug, sizeof *xug); 1117 if (error) { 1118 free(xug, M_TEMP); 1119 return (error); 1120 } 1121 1122 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1123 1124 UNP_LOCK(); 1125 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1126 unp = LIST_NEXT(unp, unp_link)) { 1127 if (unp->unp_gencnt <= gencnt) { 1128 if (cr_cansee(req->td->td_ucred, 1129 unp->unp_socket->so_cred)) 1130 continue; 1131 unp_list[i++] = unp; 1132 } 1133 } 1134 UNP_UNLOCK(); 1135 n = i; /* In case we lost some during malloc. */ 1136 1137 error = 0; 1138 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1139 for (i = 0; i < n; i++) { 1140 unp = unp_list[i]; 1141 if (unp->unp_gencnt <= gencnt) { 1142 xu->xu_len = sizeof *xu; 1143 xu->xu_unpp = unp; 1144 /* 1145 * XXX - need more locking here to protect against 1146 * connect/disconnect races for SMP. 1147 */ 1148 if (unp->unp_addr != NULL) 1149 bcopy(unp->unp_addr, &xu->xu_addr, 1150 unp->unp_addr->sun_len); 1151 if (unp->unp_conn != NULL && 1152 unp->unp_conn->unp_addr != NULL) 1153 bcopy(unp->unp_conn->unp_addr, 1154 &xu->xu_caddr, 1155 unp->unp_conn->unp_addr->sun_len); 1156 bcopy(unp, &xu->xu_unp, sizeof *unp); 1157 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1158 error = SYSCTL_OUT(req, xu, sizeof *xu); 1159 } 1160 } 1161 free(xu, M_TEMP); 1162 if (!error) { 1163 /* 1164 * Give the user an updated idea of our state. If the 1165 * generation differs from what we told her before, she knows 1166 * that something happened while we were processing this 1167 * request, and it might be necessary to retry. 1168 */ 1169 xug->xug_gen = unp_gencnt; 1170 xug->xug_sogen = so_gencnt; 1171 xug->xug_count = unp_count; 1172 error = SYSCTL_OUT(req, xug, sizeof *xug); 1173 } 1174 free(unp_list, M_TEMP); 1175 free(xug, M_TEMP); 1176 return (error); 1177} 1178 1179SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1180 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1181 "List of active local datagram sockets"); 1182SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1183 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1184 "List of active local stream sockets"); 1185 1186static void 1187unp_shutdown(struct unpcb *unp) 1188{ 1189 struct socket *so; 1190 1191 UNP_LOCK_ASSERT(); 1192 1193 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 1194 (so = unp->unp_conn->unp_socket)) 1195 socantrcvmore(so); 1196} 1197 1198static void 1199unp_drop(struct unpcb *unp, int errno) 1200{ 1201 struct socket *so = unp->unp_socket; 1202 1203 UNP_LOCK_ASSERT(); 1204 1205 so->so_error = errno; 1206 unp_disconnect(unp); 1207} 1208 1209static void 1210unp_freerights(struct file **rp, int fdcount) 1211{ 1212 int i; 1213 struct file *fp; 1214 1215 for (i = 0; i < fdcount; i++) { 1216 fp = *rp; 1217 /* 1218 * Zero the pointer before calling unp_discard since it may 1219 * end up in unp_gc().. 1220 * 1221 * XXXRW: This is less true than it used to be. 1222 */ 1223 *rp++ = 0; 1224 unp_discard(fp); 1225 } 1226} 1227 1228int 1229unp_externalize(struct mbuf *control, struct mbuf **controlp) 1230{ 1231 struct thread *td = curthread; /* XXX */ 1232 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1233 int i; 1234 int *fdp; 1235 struct file **rp; 1236 struct file *fp; 1237 void *data; 1238 socklen_t clen = control->m_len, datalen; 1239 int error, newfds; 1240 int f; 1241 u_int newlen; 1242 1243 UNP_UNLOCK_ASSERT(); 1244 1245 error = 0; 1246 if (controlp != NULL) /* controlp == NULL => free control messages */ 1247 *controlp = NULL; 1248 1249 while (cm != NULL) { 1250 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1251 error = EINVAL; 1252 break; 1253 } 1254 1255 data = CMSG_DATA(cm); 1256 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1257 1258 if (cm->cmsg_level == SOL_SOCKET 1259 && cm->cmsg_type == SCM_RIGHTS) { 1260 newfds = datalen / sizeof(struct file *); 1261 rp = data; 1262 1263 /* If we're not outputting the descriptors free them. */ 1264 if (error || controlp == NULL) { 1265 unp_freerights(rp, newfds); 1266 goto next; 1267 } 1268 FILEDESC_LOCK(td->td_proc->p_fd); 1269 /* if the new FD's will not fit free them. */ 1270 if (!fdavail(td, newfds)) { 1271 FILEDESC_UNLOCK(td->td_proc->p_fd); 1272 error = EMSGSIZE; 1273 unp_freerights(rp, newfds); 1274 goto next; 1275 } 1276 /* 1277 * Now change each pointer to an fd in the global 1278 * table to an integer that is the index to the local 1279 * fd table entry that we set up to point to the 1280 * global one we are transferring. 1281 */ 1282 newlen = newfds * sizeof(int); 1283 *controlp = sbcreatecontrol(NULL, newlen, 1284 SCM_RIGHTS, SOL_SOCKET); 1285 if (*controlp == NULL) { 1286 FILEDESC_UNLOCK(td->td_proc->p_fd); 1287 error = E2BIG; 1288 unp_freerights(rp, newfds); 1289 goto next; 1290 } 1291 1292 fdp = (int *) 1293 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1294 for (i = 0; i < newfds; i++) { 1295 if (fdalloc(td, 0, &f)) 1296 panic("unp_externalize fdalloc failed"); 1297 fp = *rp++; 1298 td->td_proc->p_fd->fd_ofiles[f] = fp; 1299 FILE_LOCK(fp); 1300 fp->f_msgcount--; 1301 FILE_UNLOCK(fp); 1302 unp_rights--; 1303 *fdp++ = f; 1304 } 1305 FILEDESC_UNLOCK(td->td_proc->p_fd); 1306 } else { 1307 /* We can just copy anything else across. */ 1308 if (error || controlp == NULL) 1309 goto next; 1310 *controlp = sbcreatecontrol(NULL, datalen, 1311 cm->cmsg_type, cm->cmsg_level); 1312 if (*controlp == NULL) { 1313 error = ENOBUFS; 1314 goto next; 1315 } 1316 bcopy(data, 1317 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1318 datalen); 1319 } 1320 1321 controlp = &(*controlp)->m_next; 1322 1323next: 1324 if (CMSG_SPACE(datalen) < clen) { 1325 clen -= CMSG_SPACE(datalen); 1326 cm = (struct cmsghdr *) 1327 ((caddr_t)cm + CMSG_SPACE(datalen)); 1328 } else { 1329 clen = 0; 1330 cm = NULL; 1331 } 1332 } 1333 1334 m_freem(control); 1335 1336 return (error); 1337} 1338 1339static void 1340unp_zone_change(void *tag) 1341{ 1342 1343 uma_zone_set_max(unp_zone, maxsockets); 1344} 1345 1346void 1347unp_init(void) 1348{ 1349 1350 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1351 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1352 if (unp_zone == NULL) 1353 panic("unp_init"); 1354 uma_zone_set_max(unp_zone, maxsockets); 1355 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1356 NULL, EVENTHANDLER_PRI_ANY); 1357 LIST_INIT(&unp_dhead); 1358 LIST_INIT(&unp_shead); 1359 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1360 UNP_LOCK_INIT(); 1361} 1362 1363static int 1364unp_internalize(struct mbuf **controlp, struct thread *td) 1365{ 1366 struct mbuf *control = *controlp; 1367 struct proc *p = td->td_proc; 1368 struct filedesc *fdescp = p->p_fd; 1369 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1370 struct cmsgcred *cmcred; 1371 struct file **rp; 1372 struct file *fp; 1373 struct timeval *tv; 1374 int i, fd, *fdp; 1375 void *data; 1376 socklen_t clen = control->m_len, datalen; 1377 int error, oldfds; 1378 u_int newlen; 1379 1380 UNP_UNLOCK_ASSERT(); 1381 1382 error = 0; 1383 *controlp = NULL; 1384 1385 while (cm != NULL) { 1386 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1387 || cm->cmsg_len > clen) { 1388 error = EINVAL; 1389 goto out; 1390 } 1391 1392 data = CMSG_DATA(cm); 1393 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1394 1395 switch (cm->cmsg_type) { 1396 /* 1397 * Fill in credential information. 1398 */ 1399 case SCM_CREDS: 1400 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1401 SCM_CREDS, SOL_SOCKET); 1402 if (*controlp == NULL) { 1403 error = ENOBUFS; 1404 goto out; 1405 } 1406 1407 cmcred = (struct cmsgcred *) 1408 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1409 cmcred->cmcred_pid = p->p_pid; 1410 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1411 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1412 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1413 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1414 CMGROUP_MAX); 1415 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1416 cmcred->cmcred_groups[i] = 1417 td->td_ucred->cr_groups[i]; 1418 break; 1419 1420 case SCM_RIGHTS: 1421 oldfds = datalen / sizeof (int); 1422 /* 1423 * Check that all the FDs passed in refer to legal 1424 * files. If not, reject the entire operation. 1425 */ 1426 fdp = data; 1427 FILEDESC_LOCK(fdescp); 1428 for (i = 0; i < oldfds; i++) { 1429 fd = *fdp++; 1430 if ((unsigned)fd >= fdescp->fd_nfiles || 1431 fdescp->fd_ofiles[fd] == NULL) { 1432 FILEDESC_UNLOCK(fdescp); 1433 error = EBADF; 1434 goto out; 1435 } 1436 fp = fdescp->fd_ofiles[fd]; 1437 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1438 FILEDESC_UNLOCK(fdescp); 1439 error = EOPNOTSUPP; 1440 goto out; 1441 } 1442 1443 } 1444 /* 1445 * Now replace the integer FDs with pointers to the 1446 * associated global file table entry.. 1447 */ 1448 newlen = oldfds * sizeof(struct file *); 1449 *controlp = sbcreatecontrol(NULL, newlen, 1450 SCM_RIGHTS, SOL_SOCKET); 1451 if (*controlp == NULL) { 1452 FILEDESC_UNLOCK(fdescp); 1453 error = E2BIG; 1454 goto out; 1455 } 1456 1457 fdp = data; 1458 rp = (struct file **) 1459 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1460 for (i = 0; i < oldfds; i++) { 1461 fp = fdescp->fd_ofiles[*fdp++]; 1462 *rp++ = fp; 1463 FILE_LOCK(fp); 1464 fp->f_count++; 1465 fp->f_msgcount++; 1466 FILE_UNLOCK(fp); 1467 unp_rights++; 1468 } 1469 FILEDESC_UNLOCK(fdescp); 1470 break; 1471 1472 case SCM_TIMESTAMP: 1473 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1474 SCM_TIMESTAMP, SOL_SOCKET); 1475 if (*controlp == NULL) { 1476 error = ENOBUFS; 1477 goto out; 1478 } 1479 tv = (struct timeval *) 1480 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1481 microtime(tv); 1482 break; 1483 1484 default: 1485 error = EINVAL; 1486 goto out; 1487 } 1488 1489 controlp = &(*controlp)->m_next; 1490 1491 if (CMSG_SPACE(datalen) < clen) { 1492 clen -= CMSG_SPACE(datalen); 1493 cm = (struct cmsghdr *) 1494 ((caddr_t)cm + CMSG_SPACE(datalen)); 1495 } else { 1496 clen = 0; 1497 cm = NULL; 1498 } 1499 } 1500 1501out: 1502 m_freem(control); 1503 1504 return (error); 1505} 1506 1507struct mbuf * 1508unp_addsockcred(struct thread *td, struct mbuf *control) 1509{ 1510 struct mbuf *m, *n, *n_prev; 1511 struct sockcred *sc; 1512 const struct cmsghdr *cm; 1513 int ngroups; 1514 int i; 1515 1516 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1517 1518 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1519 if (m == NULL) 1520 return (control); 1521 1522 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1523 sc->sc_uid = td->td_ucred->cr_ruid; 1524 sc->sc_euid = td->td_ucred->cr_uid; 1525 sc->sc_gid = td->td_ucred->cr_rgid; 1526 sc->sc_egid = td->td_ucred->cr_gid; 1527 sc->sc_ngroups = ngroups; 1528 for (i = 0; i < sc->sc_ngroups; i++) 1529 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1530 1531 /* 1532 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1533 * created SCM_CREDS control message (struct sockcred) has another 1534 * format. 1535 */ 1536 if (control != NULL) 1537 for (n = control, n_prev = NULL; n != NULL;) { 1538 cm = mtod(n, struct cmsghdr *); 1539 if (cm->cmsg_level == SOL_SOCKET && 1540 cm->cmsg_type == SCM_CREDS) { 1541 if (n_prev == NULL) 1542 control = n->m_next; 1543 else 1544 n_prev->m_next = n->m_next; 1545 n = m_free(n); 1546 } else { 1547 n_prev = n; 1548 n = n->m_next; 1549 } 1550 } 1551 1552 /* Prepend it to the head. */ 1553 m->m_next = control; 1554 1555 return (m); 1556} 1557 1558/* 1559 * unp_defer indicates whether additional work has been defered for a future 1560 * pass through unp_gc(). It is thread local and does not require explicit 1561 * synchronization. 1562 */ 1563static int unp_defer; 1564 1565static int unp_taskcount; 1566SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1567 1568static int unp_recycled; 1569SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1570 1571static void 1572unp_gc(__unused void *arg, int pending) 1573{ 1574 struct file *fp, *nextfp; 1575 struct socket *so; 1576 struct file **extra_ref, **fpp; 1577 int nunref, i; 1578 int nfiles_snap; 1579 int nfiles_slack = 20; 1580 1581 unp_taskcount++; 1582 unp_defer = 0; 1583 /* 1584 * Before going through all this, set all FDs to be NOT defered and 1585 * NOT externally accessible. 1586 */ 1587 sx_slock(&filelist_lock); 1588 LIST_FOREACH(fp, &filehead, f_list) 1589 fp->f_gcflag &= ~(FMARK|FDEFER); 1590 do { 1591 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer)); 1592 LIST_FOREACH(fp, &filehead, f_list) { 1593 FILE_LOCK(fp); 1594 /* 1595 * If the file is not open, skip it -- could be a 1596 * file in the process of being opened, or in the 1597 * process of being closed. If the file is 1598 * "closing", it may have been marked for deferred 1599 * consideration. Clear the flag now if so. 1600 */ 1601 if (fp->f_count == 0) { 1602 if (fp->f_gcflag & FDEFER) 1603 unp_defer--; 1604 fp->f_gcflag &= ~(FMARK|FDEFER); 1605 FILE_UNLOCK(fp); 1606 continue; 1607 } 1608 /* 1609 * If we already marked it as 'defer' in a previous 1610 * pass, then try process it this time and un-mark 1611 * it. 1612 */ 1613 if (fp->f_gcflag & FDEFER) { 1614 fp->f_gcflag &= ~FDEFER; 1615 unp_defer--; 1616 } else { 1617 /* 1618 * if it's not defered, then check if it's 1619 * already marked.. if so skip it 1620 */ 1621 if (fp->f_gcflag & FMARK) { 1622 FILE_UNLOCK(fp); 1623 continue; 1624 } 1625 /* 1626 * If all references are from messages in 1627 * transit, then skip it. it's not externally 1628 * accessible. 1629 */ 1630 if (fp->f_count == fp->f_msgcount) { 1631 FILE_UNLOCK(fp); 1632 continue; 1633 } 1634 /* 1635 * If it got this far then it must be 1636 * externally accessible. 1637 */ 1638 fp->f_gcflag |= FMARK; 1639 } 1640 /* 1641 * Either it was defered, or it is externally 1642 * accessible and not already marked so. Now check 1643 * if it is possibly one of OUR sockets. 1644 */ 1645 if (fp->f_type != DTYPE_SOCKET || 1646 (so = fp->f_data) == NULL) { 1647 FILE_UNLOCK(fp); 1648 continue; 1649 } 1650 FILE_UNLOCK(fp); 1651 if (so->so_proto->pr_domain != &localdomain || 1652 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 1653 continue; 1654 /* 1655 * So, Ok, it's one of our sockets and it IS 1656 * externally accessible (or was defered). Now we 1657 * look to see if we hold any file descriptors in its 1658 * message buffers. Follow those links and mark them 1659 * as accessible too. 1660 */ 1661 SOCKBUF_LOCK(&so->so_rcv); 1662 unp_scan(so->so_rcv.sb_mb, unp_mark); 1663 SOCKBUF_UNLOCK(&so->so_rcv); 1664 } 1665 } while (unp_defer); 1666 sx_sunlock(&filelist_lock); 1667 /* 1668 * XXXRW: The following comments need updating for a post-SMPng and 1669 * deferred unp_gc() world, but are still generally accurate. 1670 * 1671 * We grab an extra reference to each of the file table entries that 1672 * are not otherwise accessible and then free the rights that are 1673 * stored in messages on them. 1674 * 1675 * The bug in the orginal code is a little tricky, so I'll describe 1676 * what's wrong with it here. 1677 * 1678 * It is incorrect to simply unp_discard each entry for f_msgcount 1679 * times -- consider the case of sockets A and B that contain 1680 * references to each other. On a last close of some other socket, 1681 * we trigger a gc since the number of outstanding rights (unp_rights) 1682 * is non-zero. If during the sweep phase the gc code unp_discards, 1683 * we end up doing a (full) closef on the descriptor. A closef on A 1684 * results in the following chain. Closef calls soo_close, which 1685 * calls soclose. Soclose calls first (through the switch 1686 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply 1687 * returns because the previous instance had set unp_gcing, and we 1688 * return all the way back to soclose, which marks the socket with 1689 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free 1690 * up the rights that are queued in messages on the socket A, i.e., 1691 * the reference on B. The sorflush calls via the dom_dispose switch 1692 * unp_dispose, which unp_scans with unp_discard. This second 1693 * instance of unp_discard just calls closef on B. 1694 * 1695 * Well, a similar chain occurs on B, resulting in a sorflush on B, 1696 * which results in another closef on A. Unfortunately, A is already 1697 * being closed, and the descriptor has already been marked with 1698 * SS_NOFDREF, and soclose panics at this point. 1699 * 1700 * Here, we first take an extra reference to each inaccessible 1701 * descriptor. Then, we call sorflush ourself, since we know it is a 1702 * Unix domain socket anyhow. After we destroy all the rights 1703 * carried in messages, we do a last closef to get rid of our extra 1704 * reference. This is the last close, and the unp_detach etc will 1705 * shut down the socket. 1706 * 1707 * 91/09/19, bsy@cs.cmu.edu 1708 */ 1709again: 1710 nfiles_snap = openfiles + nfiles_slack; /* some slack */ 1711 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP, 1712 M_WAITOK); 1713 sx_slock(&filelist_lock); 1714 if (nfiles_snap < openfiles) { 1715 sx_sunlock(&filelist_lock); 1716 free(extra_ref, M_TEMP); 1717 nfiles_slack += 20; 1718 goto again; 1719 } 1720 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; 1721 fp != NULL; fp = nextfp) { 1722 nextfp = LIST_NEXT(fp, f_list); 1723 FILE_LOCK(fp); 1724 /* 1725 * If it's not open, skip it 1726 */ 1727 if (fp->f_count == 0) { 1728 FILE_UNLOCK(fp); 1729 continue; 1730 } 1731 /* 1732 * If all refs are from msgs, and it's not marked accessible 1733 * then it must be referenced from some unreachable cycle of 1734 * (shut-down) FDs, so include it in our list of FDs to 1735 * remove. 1736 */ 1737 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) { 1738 *fpp++ = fp; 1739 nunref++; 1740 fp->f_count++; 1741 } 1742 FILE_UNLOCK(fp); 1743 } 1744 sx_sunlock(&filelist_lock); 1745 /* 1746 * For each FD on our hit list, do the following two things: 1747 */ 1748 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1749 struct file *tfp = *fpp; 1750 FILE_LOCK(tfp); 1751 if (tfp->f_type == DTYPE_SOCKET && 1752 tfp->f_data != NULL) { 1753 FILE_UNLOCK(tfp); 1754 sorflush(tfp->f_data); 1755 } else { 1756 FILE_UNLOCK(tfp); 1757 } 1758 } 1759 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) { 1760 closef(*fpp, (struct thread *) NULL); 1761 unp_recycled++; 1762 } 1763 free(extra_ref, M_TEMP); 1764} 1765 1766void 1767unp_dispose(struct mbuf *m) 1768{ 1769 1770 if (m) 1771 unp_scan(m, unp_discard); 1772} 1773 1774static int 1775unp_listen(struct socket *so, struct unpcb *unp, int backlog, 1776 struct thread *td) 1777{ 1778 int error; 1779 1780 UNP_LOCK_ASSERT(); 1781 1782 SOCK_LOCK(so); 1783 error = solisten_proto_check(so); 1784 if (error == 0) { 1785 cru2x(td->td_ucred, &unp->unp_peercred); 1786 unp->unp_flags |= UNP_HAVEPCCACHED; 1787 solisten_proto(so, backlog); 1788 } 1789 SOCK_UNLOCK(so); 1790 return (error); 1791} 1792 1793static void 1794unp_scan(struct mbuf *m0, void (*op)(struct file *)) 1795{ 1796 struct mbuf *m; 1797 struct file **rp; 1798 struct cmsghdr *cm; 1799 void *data; 1800 int i; 1801 socklen_t clen, datalen; 1802 int qfds; 1803 1804 while (m0 != NULL) { 1805 for (m = m0; m; m = m->m_next) { 1806 if (m->m_type != MT_CONTROL) 1807 continue; 1808 1809 cm = mtod(m, struct cmsghdr *); 1810 clen = m->m_len; 1811 1812 while (cm != NULL) { 1813 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 1814 break; 1815 1816 data = CMSG_DATA(cm); 1817 datalen = (caddr_t)cm + cm->cmsg_len 1818 - (caddr_t)data; 1819 1820 if (cm->cmsg_level == SOL_SOCKET && 1821 cm->cmsg_type == SCM_RIGHTS) { 1822 qfds = datalen / sizeof (struct file *); 1823 rp = data; 1824 for (i = 0; i < qfds; i++) 1825 (*op)(*rp++); 1826 } 1827 1828 if (CMSG_SPACE(datalen) < clen) { 1829 clen -= CMSG_SPACE(datalen); 1830 cm = (struct cmsghdr *) 1831 ((caddr_t)cm + CMSG_SPACE(datalen)); 1832 } else { 1833 clen = 0; 1834 cm = NULL; 1835 } 1836 } 1837 } 1838 m0 = m0->m_act; 1839 } 1840} 1841 1842static void 1843unp_mark(struct file *fp) 1844{ 1845 if (fp->f_gcflag & FMARK) 1846 return; 1847 unp_defer++; 1848 fp->f_gcflag |= (FMARK|FDEFER); 1849} 1850 1851static void 1852unp_discard(struct file *fp) 1853{ 1854 UNP_LOCK(); 1855 FILE_LOCK(fp); 1856 fp->f_msgcount--; 1857 unp_rights--; 1858 FILE_UNLOCK(fp); 1859 UNP_UNLOCK(); 1860 (void) closef(fp, (struct thread *)NULL); 1861} 1862