uipc_usrreq.c revision 263820
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004-2009 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. 45 * 46 * The implementation is substantially complicated by the fact that 47 * "ancillary data", such as file descriptors or credentials, may be passed 48 * across UNIX domain sockets. The potential for passing UNIX domain sockets 49 * over other UNIX domain sockets requires the implementation of a simple 50 * garbage collector to find and tear down cycles of disconnected sockets. 51 * 52 * TODO: 53 * RDM 54 * distinguish datagram size limits from flow control limits in SEQPACKET 55 * rethink name space problems 56 * need a proper out-of-band 57 */ 58 59#include <sys/cdefs.h> 60__FBSDID("$FreeBSD: stable/10/sys/kern/uipc_usrreq.c 263820 2014-03-27 16:47:35Z asomers $"); 61 62#include "opt_ddb.h" 63 64#include <sys/param.h> 65#include <sys/capability.h> 66#include <sys/domain.h> 67#include <sys/fcntl.h> 68#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 69#include <sys/eventhandler.h> 70#include <sys/file.h> 71#include <sys/filedesc.h> 72#include <sys/kernel.h> 73#include <sys/lock.h> 74#include <sys/mbuf.h> 75#include <sys/mount.h> 76#include <sys/mutex.h> 77#include <sys/namei.h> 78#include <sys/proc.h> 79#include <sys/protosw.h> 80#include <sys/queue.h> 81#include <sys/resourcevar.h> 82#include <sys/rwlock.h> 83#include <sys/socket.h> 84#include <sys/socketvar.h> 85#include <sys/signalvar.h> 86#include <sys/stat.h> 87#include <sys/sx.h> 88#include <sys/sysctl.h> 89#include <sys/systm.h> 90#include <sys/taskqueue.h> 91#include <sys/un.h> 92#include <sys/unpcb.h> 93#include <sys/vnode.h> 94 95#include <net/vnet.h> 96 97#ifdef DDB 98#include <ddb/ddb.h> 99#endif 100 101#include <security/mac/mac_framework.h> 102 103#include <vm/uma.h> 104 105MALLOC_DECLARE(M_FILECAPS); 106 107/* 108 * Locking key: 109 * (l) Locked using list lock 110 * (g) Locked using linkage lock 111 */ 112 113static uma_zone_t unp_zone; 114static unp_gen_t unp_gencnt; /* (l) */ 115static u_int unp_count; /* (l) Count of local sockets. */ 116static ino_t unp_ino; /* Prototype for fake inode numbers. */ 117static int unp_rights; /* (g) File descriptors in flight. */ 118static struct unp_head unp_shead; /* (l) List of stream sockets. */ 119static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 120static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 121 122struct unp_defer { 123 SLIST_ENTRY(unp_defer) ud_link; 124 struct file *ud_fp; 125}; 126static SLIST_HEAD(, unp_defer) unp_defers; 127static int unp_defers_count; 128 129static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 130 131/* 132 * Garbage collection of cyclic file descriptor/socket references occurs 133 * asynchronously in a taskqueue context in order to avoid recursion and 134 * reentrance in the UNIX domain socket, file descriptor, and socket layer 135 * code. See unp_gc() for a full description. 136 */ 137static struct timeout_task unp_gc_task; 138 139/* 140 * The close of unix domain sockets attached as SCM_RIGHTS is 141 * postponed to the taskqueue, to avoid arbitrary recursion depth. 142 * The attached sockets might have another sockets attached. 143 */ 144static struct task unp_defer_task; 145 146/* 147 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 148 * stream sockets, although the total for sender and receiver is actually 149 * only PIPSIZ. 150 * 151 * Datagram sockets really use the sendspace as the maximum datagram size, 152 * and don't really want to reserve the sendspace. Their recvspace should be 153 * large enough for at least one max-size datagram plus address. 154 */ 155#ifndef PIPSIZ 156#define PIPSIZ 8192 157#endif 158static u_long unpst_sendspace = PIPSIZ; 159static u_long unpst_recvspace = PIPSIZ; 160static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 161static u_long unpdg_recvspace = 4*1024; 162static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 163static u_long unpsp_recvspace = PIPSIZ; 164 165static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 166static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, 167 "SOCK_STREAM"); 168static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 169static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, 170 "SOCK_SEQPACKET"); 171 172SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 173 &unpst_sendspace, 0, "Default stream send space."); 174SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 175 &unpst_recvspace, 0, "Default stream receive space."); 176SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 177 &unpdg_sendspace, 0, "Default datagram send space."); 178SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 179 &unpdg_recvspace, 0, "Default datagram receive space."); 180SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 181 &unpsp_sendspace, 0, "Default seqpacket send space."); 182SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 183 &unpsp_recvspace, 0, "Default seqpacket receive space."); 184SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 185 "File descriptors in flight."); 186SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 187 &unp_defers_count, 0, 188 "File descriptors deferred to taskqueue for close."); 189 190/* 191 * Locking and synchronization: 192 * 193 * Three types of locks exit in the local domain socket implementation: a 194 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 195 * global locks, the list lock protects the socket count, global generation 196 * number, and stream/datagram global lists. The linkage lock protects the 197 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 198 * held exclusively over the acquisition of multiple unpcb locks to prevent 199 * deadlock. 200 * 201 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 202 * allocated in pru_attach() and freed in pru_detach(). The validity of that 203 * pointer is an invariant, so no lock is required to dereference the so_pcb 204 * pointer if a valid socket reference is held by the caller. In practice, 205 * this is always true during operations performed on a socket. Each unpcb 206 * has a back-pointer to its socket, unp_socket, which will be stable under 207 * the same circumstances. 208 * 209 * This pointer may only be safely dereferenced as long as a valid reference 210 * to the unpcb is held. Typically, this reference will be from the socket, 211 * or from another unpcb when the referring unpcb's lock is held (in order 212 * that the reference not be invalidated during use). For example, to follow 213 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 214 * as unp_socket remains valid as long as the reference to unp_conn is valid. 215 * 216 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 217 * atomic reads without the lock may be performed "lockless", but more 218 * complex reads and read-modify-writes require the mutex to be held. No 219 * lock order is defined between unpcb locks -- multiple unpcb locks may be 220 * acquired at the same time only when holding the linkage rwlock 221 * exclusively, which prevents deadlocks. 222 * 223 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 224 * protocols, bind() is a non-atomic operation, and connect() requires 225 * potential sleeping in the protocol, due to potentially waiting on local or 226 * distributed file systems. We try to separate "lookup" operations, which 227 * may sleep, and the IPC operations themselves, which typically can occur 228 * with relative atomicity as locks can be held over the entire operation. 229 * 230 * Another tricky issue is simultaneous multi-threaded or multi-process 231 * access to a single UNIX domain socket. These are handled by the flags 232 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 233 * binding, both of which involve dropping UNIX domain socket locks in order 234 * to perform namei() and other file system operations. 235 */ 236static struct rwlock unp_link_rwlock; 237static struct mtx unp_list_lock; 238static struct mtx unp_defers_lock; 239 240#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 241 "unp_link_rwlock") 242 243#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 244 RA_LOCKED) 245#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 246 RA_UNLOCKED) 247 248#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 249#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 250#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 251#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 252#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 253 RA_WLOCKED) 254 255#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 256 "unp_list_lock", NULL, MTX_DEF) 257#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 258#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 259 260#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 261 "unp_defer", NULL, MTX_DEF) 262#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 263#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 264 265#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 266 "unp_mtx", "unp_mtx", \ 267 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 268#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 269#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 270#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 271#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 272 273static int uipc_connect2(struct socket *, struct socket *); 274static int uipc_ctloutput(struct socket *, struct sockopt *); 275static int unp_connect(struct socket *, struct sockaddr *, 276 struct thread *); 277static int unp_connectat(int, struct socket *, struct sockaddr *, 278 struct thread *); 279static int unp_connect2(struct socket *so, struct socket *so2, int); 280static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 281static void unp_dispose(struct mbuf *); 282static void unp_shutdown(struct unpcb *); 283static void unp_drop(struct unpcb *, int); 284static void unp_gc(__unused void *, int); 285static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 286static void unp_discard(struct file *); 287static void unp_freerights(struct filedescent **, int); 288static void unp_init(void); 289static int unp_internalize(struct mbuf **, struct thread *); 290static void unp_internalize_fp(struct file *); 291static int unp_externalize(struct mbuf *, struct mbuf **, int); 292static int unp_externalize_fp(struct file *); 293static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 294static void unp_process_defers(void * __unused, int); 295 296/* 297 * Definitions of protocols supported in the LOCAL domain. 298 */ 299static struct domain localdomain; 300static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 301static struct pr_usrreqs uipc_usrreqs_seqpacket; 302static struct protosw localsw[] = { 303{ 304 .pr_type = SOCK_STREAM, 305 .pr_domain = &localdomain, 306 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 307 .pr_ctloutput = &uipc_ctloutput, 308 .pr_usrreqs = &uipc_usrreqs_stream 309}, 310{ 311 .pr_type = SOCK_DGRAM, 312 .pr_domain = &localdomain, 313 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 314 .pr_ctloutput = &uipc_ctloutput, 315 .pr_usrreqs = &uipc_usrreqs_dgram 316}, 317{ 318 .pr_type = SOCK_SEQPACKET, 319 .pr_domain = &localdomain, 320 321 /* 322 * XXXRW: For now, PR_ADDR because soreceive will bump into them 323 * due to our use of sbappendaddr. A new sbappend variants is needed 324 * that supports both atomic record writes and control data. 325 */ 326 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 327 PR_RIGHTS, 328 .pr_ctloutput = &uipc_ctloutput, 329 .pr_usrreqs = &uipc_usrreqs_seqpacket, 330}, 331}; 332 333static struct domain localdomain = { 334 .dom_family = AF_LOCAL, 335 .dom_name = "local", 336 .dom_init = unp_init, 337 .dom_externalize = unp_externalize, 338 .dom_dispose = unp_dispose, 339 .dom_protosw = localsw, 340 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 341}; 342DOMAIN_SET(local); 343 344static void 345uipc_abort(struct socket *so) 346{ 347 struct unpcb *unp, *unp2; 348 349 unp = sotounpcb(so); 350 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 351 352 UNP_LINK_WLOCK(); 353 UNP_PCB_LOCK(unp); 354 unp2 = unp->unp_conn; 355 if (unp2 != NULL) { 356 UNP_PCB_LOCK(unp2); 357 unp_drop(unp2, ECONNABORTED); 358 UNP_PCB_UNLOCK(unp2); 359 } 360 UNP_PCB_UNLOCK(unp); 361 UNP_LINK_WUNLOCK(); 362} 363 364static int 365uipc_accept(struct socket *so, struct sockaddr **nam) 366{ 367 struct unpcb *unp, *unp2; 368 const struct sockaddr *sa; 369 370 /* 371 * Pass back name of connected socket, if it was bound and we are 372 * still connected (our peer may have closed already!). 373 */ 374 unp = sotounpcb(so); 375 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 376 377 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 378 UNP_LINK_RLOCK(); 379 unp2 = unp->unp_conn; 380 if (unp2 != NULL && unp2->unp_addr != NULL) { 381 UNP_PCB_LOCK(unp2); 382 sa = (struct sockaddr *) unp2->unp_addr; 383 bcopy(sa, *nam, sa->sa_len); 384 UNP_PCB_UNLOCK(unp2); 385 } else { 386 sa = &sun_noname; 387 bcopy(sa, *nam, sa->sa_len); 388 } 389 UNP_LINK_RUNLOCK(); 390 return (0); 391} 392 393static int 394uipc_attach(struct socket *so, int proto, struct thread *td) 395{ 396 u_long sendspace, recvspace; 397 struct unpcb *unp; 398 int error; 399 400 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 401 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 402 switch (so->so_type) { 403 case SOCK_STREAM: 404 sendspace = unpst_sendspace; 405 recvspace = unpst_recvspace; 406 break; 407 408 case SOCK_DGRAM: 409 sendspace = unpdg_sendspace; 410 recvspace = unpdg_recvspace; 411 break; 412 413 case SOCK_SEQPACKET: 414 sendspace = unpsp_sendspace; 415 recvspace = unpsp_recvspace; 416 break; 417 418 default: 419 panic("uipc_attach"); 420 } 421 error = soreserve(so, sendspace, recvspace); 422 if (error) 423 return (error); 424 } 425 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 426 if (unp == NULL) 427 return (ENOBUFS); 428 LIST_INIT(&unp->unp_refs); 429 UNP_PCB_LOCK_INIT(unp); 430 unp->unp_socket = so; 431 so->so_pcb = unp; 432 unp->unp_refcount = 1; 433 434 UNP_LIST_LOCK(); 435 unp->unp_gencnt = ++unp_gencnt; 436 unp_count++; 437 switch (so->so_type) { 438 case SOCK_STREAM: 439 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 440 break; 441 442 case SOCK_DGRAM: 443 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 444 break; 445 446 case SOCK_SEQPACKET: 447 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 448 break; 449 450 default: 451 panic("uipc_attach"); 452 } 453 UNP_LIST_UNLOCK(); 454 455 return (0); 456} 457 458static int 459uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 460{ 461 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 462 struct vattr vattr; 463 int error, namelen; 464 struct nameidata nd; 465 struct unpcb *unp; 466 struct vnode *vp; 467 struct mount *mp; 468 cap_rights_t rights; 469 char *buf; 470 471 unp = sotounpcb(so); 472 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 473 474 if (soun->sun_len > sizeof(struct sockaddr_un)) 475 return (EINVAL); 476 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 477 if (namelen <= 0) 478 return (EINVAL); 479 480 /* 481 * We don't allow simultaneous bind() calls on a single UNIX domain 482 * socket, so flag in-progress operations, and return an error if an 483 * operation is already in progress. 484 * 485 * Historically, we have not allowed a socket to be rebound, so this 486 * also returns an error. Not allowing re-binding simplifies the 487 * implementation and avoids a great many possible failure modes. 488 */ 489 UNP_PCB_LOCK(unp); 490 if (unp->unp_vnode != NULL) { 491 UNP_PCB_UNLOCK(unp); 492 return (EINVAL); 493 } 494 if (unp->unp_flags & UNP_BINDING) { 495 UNP_PCB_UNLOCK(unp); 496 return (EALREADY); 497 } 498 unp->unp_flags |= UNP_BINDING; 499 UNP_PCB_UNLOCK(unp); 500 501 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 502 bcopy(soun->sun_path, buf, namelen); 503 buf[namelen] = 0; 504 505restart: 506 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, 507 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td); 508/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 509 error = namei(&nd); 510 if (error) 511 goto error; 512 vp = nd.ni_vp; 513 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 514 NDFREE(&nd, NDF_ONLY_PNBUF); 515 if (nd.ni_dvp == vp) 516 vrele(nd.ni_dvp); 517 else 518 vput(nd.ni_dvp); 519 if (vp != NULL) { 520 vrele(vp); 521 error = EADDRINUSE; 522 goto error; 523 } 524 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 525 if (error) 526 goto error; 527 goto restart; 528 } 529 VATTR_NULL(&vattr); 530 vattr.va_type = VSOCK; 531 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 532#ifdef MAC 533 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 534 &vattr); 535#endif 536 if (error == 0) 537 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 538 NDFREE(&nd, NDF_ONLY_PNBUF); 539 vput(nd.ni_dvp); 540 if (error) { 541 vn_finished_write(mp); 542 goto error; 543 } 544 vp = nd.ni_vp; 545 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 546 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 547 548 UNP_LINK_WLOCK(); 549 UNP_PCB_LOCK(unp); 550 VOP_UNP_BIND(vp, unp->unp_socket); 551 unp->unp_vnode = vp; 552 unp->unp_addr = soun; 553 unp->unp_flags &= ~UNP_BINDING; 554 UNP_PCB_UNLOCK(unp); 555 UNP_LINK_WUNLOCK(); 556 VOP_UNLOCK(vp, 0); 557 vn_finished_write(mp); 558 free(buf, M_TEMP); 559 return (0); 560 561error: 562 UNP_PCB_LOCK(unp); 563 unp->unp_flags &= ~UNP_BINDING; 564 UNP_PCB_UNLOCK(unp); 565 free(buf, M_TEMP); 566 return (error); 567} 568 569static int 570uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 571{ 572 573 return (uipc_bindat(AT_FDCWD, so, nam, td)); 574} 575 576static int 577uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 578{ 579 int error; 580 581 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 582 UNP_LINK_WLOCK(); 583 error = unp_connect(so, nam, td); 584 UNP_LINK_WUNLOCK(); 585 return (error); 586} 587 588static int 589uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 590 struct thread *td) 591{ 592 int error; 593 594 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 595 UNP_LINK_WLOCK(); 596 error = unp_connectat(fd, so, nam, td); 597 UNP_LINK_WUNLOCK(); 598 return (error); 599} 600 601static void 602uipc_close(struct socket *so) 603{ 604 struct unpcb *unp, *unp2; 605 606 unp = sotounpcb(so); 607 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 608 609 UNP_LINK_WLOCK(); 610 UNP_PCB_LOCK(unp); 611 unp2 = unp->unp_conn; 612 if (unp2 != NULL) { 613 UNP_PCB_LOCK(unp2); 614 unp_disconnect(unp, unp2); 615 UNP_PCB_UNLOCK(unp2); 616 } 617 UNP_PCB_UNLOCK(unp); 618 UNP_LINK_WUNLOCK(); 619} 620 621static int 622uipc_connect2(struct socket *so1, struct socket *so2) 623{ 624 struct unpcb *unp, *unp2; 625 int error; 626 627 UNP_LINK_WLOCK(); 628 unp = so1->so_pcb; 629 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 630 UNP_PCB_LOCK(unp); 631 unp2 = so2->so_pcb; 632 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 633 UNP_PCB_LOCK(unp2); 634 error = unp_connect2(so1, so2, PRU_CONNECT2); 635 UNP_PCB_UNLOCK(unp2); 636 UNP_PCB_UNLOCK(unp); 637 UNP_LINK_WUNLOCK(); 638 return (error); 639} 640 641static void 642uipc_detach(struct socket *so) 643{ 644 struct unpcb *unp, *unp2; 645 struct sockaddr_un *saved_unp_addr; 646 struct vnode *vp; 647 int freeunp, local_unp_rights; 648 649 unp = sotounpcb(so); 650 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 651 652 UNP_LINK_WLOCK(); 653 UNP_LIST_LOCK(); 654 UNP_PCB_LOCK(unp); 655 LIST_REMOVE(unp, unp_link); 656 unp->unp_gencnt = ++unp_gencnt; 657 --unp_count; 658 UNP_LIST_UNLOCK(); 659 660 /* 661 * XXXRW: Should assert vp->v_socket == so. 662 */ 663 if ((vp = unp->unp_vnode) != NULL) { 664 VOP_UNP_DETACH(vp); 665 unp->unp_vnode = NULL; 666 } 667 unp2 = unp->unp_conn; 668 if (unp2 != NULL) { 669 UNP_PCB_LOCK(unp2); 670 unp_disconnect(unp, unp2); 671 UNP_PCB_UNLOCK(unp2); 672 } 673 674 /* 675 * We hold the linkage lock exclusively, so it's OK to acquire 676 * multiple pcb locks at a time. 677 */ 678 while (!LIST_EMPTY(&unp->unp_refs)) { 679 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 680 681 UNP_PCB_LOCK(ref); 682 unp_drop(ref, ECONNRESET); 683 UNP_PCB_UNLOCK(ref); 684 } 685 local_unp_rights = unp_rights; 686 UNP_LINK_WUNLOCK(); 687 unp->unp_socket->so_pcb = NULL; 688 saved_unp_addr = unp->unp_addr; 689 unp->unp_addr = NULL; 690 unp->unp_refcount--; 691 freeunp = (unp->unp_refcount == 0); 692 if (saved_unp_addr != NULL) 693 free(saved_unp_addr, M_SONAME); 694 if (freeunp) { 695 UNP_PCB_LOCK_DESTROY(unp); 696 uma_zfree(unp_zone, unp); 697 } else 698 UNP_PCB_UNLOCK(unp); 699 if (vp) 700 vrele(vp); 701 if (local_unp_rights) 702 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 703} 704 705static int 706uipc_disconnect(struct socket *so) 707{ 708 struct unpcb *unp, *unp2; 709 710 unp = sotounpcb(so); 711 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 712 713 UNP_LINK_WLOCK(); 714 UNP_PCB_LOCK(unp); 715 unp2 = unp->unp_conn; 716 if (unp2 != NULL) { 717 UNP_PCB_LOCK(unp2); 718 unp_disconnect(unp, unp2); 719 UNP_PCB_UNLOCK(unp2); 720 } 721 UNP_PCB_UNLOCK(unp); 722 UNP_LINK_WUNLOCK(); 723 return (0); 724} 725 726static int 727uipc_listen(struct socket *so, int backlog, struct thread *td) 728{ 729 struct unpcb *unp; 730 int error; 731 732 unp = sotounpcb(so); 733 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 734 735 UNP_PCB_LOCK(unp); 736 if (unp->unp_vnode == NULL) { 737 UNP_PCB_UNLOCK(unp); 738 return (EINVAL); 739 } 740 741 SOCK_LOCK(so); 742 error = solisten_proto_check(so); 743 if (error == 0) { 744 cru2x(td->td_ucred, &unp->unp_peercred); 745 unp->unp_flags |= UNP_HAVEPCCACHED; 746 solisten_proto(so, backlog); 747 } 748 SOCK_UNLOCK(so); 749 UNP_PCB_UNLOCK(unp); 750 return (error); 751} 752 753static int 754uipc_peeraddr(struct socket *so, struct sockaddr **nam) 755{ 756 struct unpcb *unp, *unp2; 757 const struct sockaddr *sa; 758 759 unp = sotounpcb(so); 760 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 761 762 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 763 UNP_LINK_RLOCK(); 764 /* 765 * XXX: It seems that this test always fails even when connection is 766 * established. So, this else clause is added as workaround to 767 * return PF_LOCAL sockaddr. 768 */ 769 unp2 = unp->unp_conn; 770 if (unp2 != NULL) { 771 UNP_PCB_LOCK(unp2); 772 if (unp2->unp_addr != NULL) 773 sa = (struct sockaddr *) unp2->unp_addr; 774 else 775 sa = &sun_noname; 776 bcopy(sa, *nam, sa->sa_len); 777 UNP_PCB_UNLOCK(unp2); 778 } else { 779 sa = &sun_noname; 780 bcopy(sa, *nam, sa->sa_len); 781 } 782 UNP_LINK_RUNLOCK(); 783 return (0); 784} 785 786static int 787uipc_rcvd(struct socket *so, int flags) 788{ 789 struct unpcb *unp, *unp2; 790 struct socket *so2; 791 u_int mbcnt, sbcc; 792 u_long newhiwat; 793 794 unp = sotounpcb(so); 795 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 796 797 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 798 panic("uipc_rcvd socktype %d", so->so_type); 799 800 /* 801 * Adjust backpressure on sender and wakeup any waiting to write. 802 * 803 * The unp lock is acquired to maintain the validity of the unp_conn 804 * pointer; no lock on unp2 is required as unp2->unp_socket will be 805 * static as long as we don't permit unp2 to disconnect from unp, 806 * which is prevented by the lock on unp. We cache values from 807 * so_rcv to avoid holding the so_rcv lock over the entire 808 * transaction on the remote so_snd. 809 */ 810 SOCKBUF_LOCK(&so->so_rcv); 811 mbcnt = so->so_rcv.sb_mbcnt; 812 sbcc = so->so_rcv.sb_cc; 813 SOCKBUF_UNLOCK(&so->so_rcv); 814 UNP_PCB_LOCK(unp); 815 unp2 = unp->unp_conn; 816 if (unp2 == NULL) { 817 UNP_PCB_UNLOCK(unp); 818 return (0); 819 } 820 so2 = unp2->unp_socket; 821 SOCKBUF_LOCK(&so2->so_snd); 822 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 823 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 824 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 825 newhiwat, RLIM_INFINITY); 826 sowwakeup_locked(so2); 827 unp->unp_mbcnt = mbcnt; 828 unp->unp_cc = sbcc; 829 UNP_PCB_UNLOCK(unp); 830 return (0); 831} 832 833static int 834uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 835 struct mbuf *control, struct thread *td) 836{ 837 struct unpcb *unp, *unp2; 838 struct socket *so2; 839 u_int mbcnt_delta, sbcc; 840 u_int newhiwat; 841 int error = 0; 842 843 unp = sotounpcb(so); 844 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 845 846 if (flags & PRUS_OOB) { 847 error = EOPNOTSUPP; 848 goto release; 849 } 850 if (control != NULL && (error = unp_internalize(&control, td))) 851 goto release; 852 if ((nam != NULL) || (flags & PRUS_EOF)) 853 UNP_LINK_WLOCK(); 854 else 855 UNP_LINK_RLOCK(); 856 switch (so->so_type) { 857 case SOCK_DGRAM: 858 { 859 const struct sockaddr *from; 860 861 unp2 = unp->unp_conn; 862 if (nam != NULL) { 863 UNP_LINK_WLOCK_ASSERT(); 864 if (unp2 != NULL) { 865 error = EISCONN; 866 break; 867 } 868 error = unp_connect(so, nam, td); 869 if (error) 870 break; 871 unp2 = unp->unp_conn; 872 } 873 874 /* 875 * Because connect() and send() are non-atomic in a sendto() 876 * with a target address, it's possible that the socket will 877 * have disconnected before the send() can run. In that case 878 * return the slightly counter-intuitive but otherwise 879 * correct error that the socket is not connected. 880 */ 881 if (unp2 == NULL) { 882 error = ENOTCONN; 883 break; 884 } 885 /* Lockless read. */ 886 if (unp2->unp_flags & UNP_WANTCRED) 887 control = unp_addsockcred(td, control); 888 UNP_PCB_LOCK(unp); 889 if (unp->unp_addr != NULL) 890 from = (struct sockaddr *)unp->unp_addr; 891 else 892 from = &sun_noname; 893 so2 = unp2->unp_socket; 894 SOCKBUF_LOCK(&so2->so_rcv); 895 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, from, m, 896 control)) { 897 sorwakeup_locked(so2); 898 m = NULL; 899 control = NULL; 900 } else { 901 SOCKBUF_UNLOCK(&so2->so_rcv); 902 error = ENOBUFS; 903 } 904 if (nam != NULL) { 905 UNP_LINK_WLOCK_ASSERT(); 906 UNP_PCB_LOCK(unp2); 907 unp_disconnect(unp, unp2); 908 UNP_PCB_UNLOCK(unp2); 909 } 910 UNP_PCB_UNLOCK(unp); 911 break; 912 } 913 914 case SOCK_SEQPACKET: 915 case SOCK_STREAM: 916 if ((so->so_state & SS_ISCONNECTED) == 0) { 917 if (nam != NULL) { 918 UNP_LINK_WLOCK_ASSERT(); 919 error = unp_connect(so, nam, td); 920 if (error) 921 break; /* XXX */ 922 } else { 923 error = ENOTCONN; 924 break; 925 } 926 } 927 928 /* Lockless read. */ 929 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 930 error = EPIPE; 931 break; 932 } 933 934 /* 935 * Because connect() and send() are non-atomic in a sendto() 936 * with a target address, it's possible that the socket will 937 * have disconnected before the send() can run. In that case 938 * return the slightly counter-intuitive but otherwise 939 * correct error that the socket is not connected. 940 * 941 * Locking here must be done carefully: the linkage lock 942 * prevents interconnections between unpcbs from changing, so 943 * we can traverse from unp to unp2 without acquiring unp's 944 * lock. Socket buffer locks follow unpcb locks, so we can 945 * acquire both remote and lock socket buffer locks. 946 */ 947 unp2 = unp->unp_conn; 948 if (unp2 == NULL) { 949 error = ENOTCONN; 950 break; 951 } 952 so2 = unp2->unp_socket; 953 UNP_PCB_LOCK(unp2); 954 SOCKBUF_LOCK(&so2->so_rcv); 955 if (unp2->unp_flags & UNP_WANTCRED) { 956 /* 957 * Credentials are passed only once on SOCK_STREAM 958 * and SOCK_SEQPACKET. 959 */ 960 unp2->unp_flags &= ~UNP_WANTCRED; 961 control = unp_addsockcred(td, control); 962 } 963 /* 964 * Send to paired receive port, and then reduce send buffer 965 * hiwater marks to maintain backpressure. Wake up readers. 966 */ 967 switch (so->so_type) { 968 case SOCK_STREAM: 969 if (control != NULL) { 970 if (sbappendcontrol_locked(&so2->so_rcv, m, 971 control)) 972 control = NULL; 973 } else 974 sbappend_locked(&so2->so_rcv, m); 975 break; 976 977 case SOCK_SEQPACKET: { 978 const struct sockaddr *from; 979 980 from = &sun_noname; 981 /* 982 * Don't check for space available in so2->so_rcv. 983 * Unix domain sockets only check for space in the 984 * sending sockbuf, and that check is performed one 985 * level up the stack. 986 */ 987 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 988 from, m, control)) 989 control = NULL; 990 break; 991 } 992 } 993 994 /* 995 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 996 * datagram size and back-pressure for SOCK_SEQPACKET, which 997 * can lead to undesired return of EMSGSIZE on send instead 998 * of more desirable blocking. 999 */ 1000 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 1001 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 1002 sbcc = so2->so_rcv.sb_cc; 1003 sorwakeup_locked(so2); 1004 1005 SOCKBUF_LOCK(&so->so_snd); 1006 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 1007 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 1008 else 1009 newhiwat = 0; 1010 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 1011 newhiwat, RLIM_INFINITY); 1012 so->so_snd.sb_mbmax -= mbcnt_delta; 1013 SOCKBUF_UNLOCK(&so->so_snd); 1014 unp2->unp_cc = sbcc; 1015 UNP_PCB_UNLOCK(unp2); 1016 m = NULL; 1017 break; 1018 1019 default: 1020 panic("uipc_send unknown socktype"); 1021 } 1022 1023 /* 1024 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1025 */ 1026 if (flags & PRUS_EOF) { 1027 UNP_PCB_LOCK(unp); 1028 socantsendmore(so); 1029 unp_shutdown(unp); 1030 UNP_PCB_UNLOCK(unp); 1031 } 1032 1033 if ((nam != NULL) || (flags & PRUS_EOF)) 1034 UNP_LINK_WUNLOCK(); 1035 else 1036 UNP_LINK_RUNLOCK(); 1037 1038 if (control != NULL && error != 0) 1039 unp_dispose(control); 1040 1041release: 1042 if (control != NULL) 1043 m_freem(control); 1044 if (m != NULL) 1045 m_freem(m); 1046 return (error); 1047} 1048 1049static int 1050uipc_sense(struct socket *so, struct stat *sb) 1051{ 1052 struct unpcb *unp, *unp2; 1053 struct socket *so2; 1054 1055 unp = sotounpcb(so); 1056 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1057 1058 sb->st_blksize = so->so_snd.sb_hiwat; 1059 UNP_LINK_RLOCK(); 1060 UNP_PCB_LOCK(unp); 1061 unp2 = unp->unp_conn; 1062 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1063 unp2 != NULL) { 1064 so2 = unp2->unp_socket; 1065 sb->st_blksize += so2->so_rcv.sb_cc; 1066 } 1067 sb->st_dev = NODEV; 1068 if (unp->unp_ino == 0) 1069 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1070 sb->st_ino = unp->unp_ino; 1071 UNP_PCB_UNLOCK(unp); 1072 UNP_LINK_RUNLOCK(); 1073 return (0); 1074} 1075 1076static int 1077uipc_shutdown(struct socket *so) 1078{ 1079 struct unpcb *unp; 1080 1081 unp = sotounpcb(so); 1082 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1083 1084 UNP_LINK_WLOCK(); 1085 UNP_PCB_LOCK(unp); 1086 socantsendmore(so); 1087 unp_shutdown(unp); 1088 UNP_PCB_UNLOCK(unp); 1089 UNP_LINK_WUNLOCK(); 1090 return (0); 1091} 1092 1093static int 1094uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1095{ 1096 struct unpcb *unp; 1097 const struct sockaddr *sa; 1098 1099 unp = sotounpcb(so); 1100 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1101 1102 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1103 UNP_PCB_LOCK(unp); 1104 if (unp->unp_addr != NULL) 1105 sa = (struct sockaddr *) unp->unp_addr; 1106 else 1107 sa = &sun_noname; 1108 bcopy(sa, *nam, sa->sa_len); 1109 UNP_PCB_UNLOCK(unp); 1110 return (0); 1111} 1112 1113static struct pr_usrreqs uipc_usrreqs_dgram = { 1114 .pru_abort = uipc_abort, 1115 .pru_accept = uipc_accept, 1116 .pru_attach = uipc_attach, 1117 .pru_bind = uipc_bind, 1118 .pru_bindat = uipc_bindat, 1119 .pru_connect = uipc_connect, 1120 .pru_connectat = uipc_connectat, 1121 .pru_connect2 = uipc_connect2, 1122 .pru_detach = uipc_detach, 1123 .pru_disconnect = uipc_disconnect, 1124 .pru_listen = uipc_listen, 1125 .pru_peeraddr = uipc_peeraddr, 1126 .pru_rcvd = uipc_rcvd, 1127 .pru_send = uipc_send, 1128 .pru_sense = uipc_sense, 1129 .pru_shutdown = uipc_shutdown, 1130 .pru_sockaddr = uipc_sockaddr, 1131 .pru_soreceive = soreceive_dgram, 1132 .pru_close = uipc_close, 1133}; 1134 1135static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1136 .pru_abort = uipc_abort, 1137 .pru_accept = uipc_accept, 1138 .pru_attach = uipc_attach, 1139 .pru_bind = uipc_bind, 1140 .pru_bindat = uipc_bindat, 1141 .pru_connect = uipc_connect, 1142 .pru_connectat = uipc_connectat, 1143 .pru_connect2 = uipc_connect2, 1144 .pru_detach = uipc_detach, 1145 .pru_disconnect = uipc_disconnect, 1146 .pru_listen = uipc_listen, 1147 .pru_peeraddr = uipc_peeraddr, 1148 .pru_rcvd = uipc_rcvd, 1149 .pru_send = uipc_send, 1150 .pru_sense = uipc_sense, 1151 .pru_shutdown = uipc_shutdown, 1152 .pru_sockaddr = uipc_sockaddr, 1153 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1154 .pru_close = uipc_close, 1155}; 1156 1157static struct pr_usrreqs uipc_usrreqs_stream = { 1158 .pru_abort = uipc_abort, 1159 .pru_accept = uipc_accept, 1160 .pru_attach = uipc_attach, 1161 .pru_bind = uipc_bind, 1162 .pru_bindat = uipc_bindat, 1163 .pru_connect = uipc_connect, 1164 .pru_connectat = uipc_connectat, 1165 .pru_connect2 = uipc_connect2, 1166 .pru_detach = uipc_detach, 1167 .pru_disconnect = uipc_disconnect, 1168 .pru_listen = uipc_listen, 1169 .pru_peeraddr = uipc_peeraddr, 1170 .pru_rcvd = uipc_rcvd, 1171 .pru_send = uipc_send, 1172 .pru_sense = uipc_sense, 1173 .pru_shutdown = uipc_shutdown, 1174 .pru_sockaddr = uipc_sockaddr, 1175 .pru_soreceive = soreceive_generic, 1176 .pru_close = uipc_close, 1177}; 1178 1179static int 1180uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1181{ 1182 struct unpcb *unp; 1183 struct xucred xu; 1184 int error, optval; 1185 1186 if (sopt->sopt_level != 0) 1187 return (EINVAL); 1188 1189 unp = sotounpcb(so); 1190 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1191 error = 0; 1192 switch (sopt->sopt_dir) { 1193 case SOPT_GET: 1194 switch (sopt->sopt_name) { 1195 case LOCAL_PEERCRED: 1196 UNP_PCB_LOCK(unp); 1197 if (unp->unp_flags & UNP_HAVEPC) 1198 xu = unp->unp_peercred; 1199 else { 1200 if (so->so_type == SOCK_STREAM) 1201 error = ENOTCONN; 1202 else 1203 error = EINVAL; 1204 } 1205 UNP_PCB_UNLOCK(unp); 1206 if (error == 0) 1207 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1208 break; 1209 1210 case LOCAL_CREDS: 1211 /* Unlocked read. */ 1212 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1213 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1214 break; 1215 1216 case LOCAL_CONNWAIT: 1217 /* Unlocked read. */ 1218 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1219 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1220 break; 1221 1222 default: 1223 error = EOPNOTSUPP; 1224 break; 1225 } 1226 break; 1227 1228 case SOPT_SET: 1229 switch (sopt->sopt_name) { 1230 case LOCAL_CREDS: 1231 case LOCAL_CONNWAIT: 1232 error = sooptcopyin(sopt, &optval, sizeof(optval), 1233 sizeof(optval)); 1234 if (error) 1235 break; 1236 1237#define OPTSET(bit) do { \ 1238 UNP_PCB_LOCK(unp); \ 1239 if (optval) \ 1240 unp->unp_flags |= bit; \ 1241 else \ 1242 unp->unp_flags &= ~bit; \ 1243 UNP_PCB_UNLOCK(unp); \ 1244} while (0) 1245 1246 switch (sopt->sopt_name) { 1247 case LOCAL_CREDS: 1248 OPTSET(UNP_WANTCRED); 1249 break; 1250 1251 case LOCAL_CONNWAIT: 1252 OPTSET(UNP_CONNWAIT); 1253 break; 1254 1255 default: 1256 break; 1257 } 1258 break; 1259#undef OPTSET 1260 default: 1261 error = ENOPROTOOPT; 1262 break; 1263 } 1264 break; 1265 1266 default: 1267 error = EOPNOTSUPP; 1268 break; 1269 } 1270 return (error); 1271} 1272 1273static int 1274unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1275{ 1276 1277 return (unp_connectat(AT_FDCWD, so, nam, td)); 1278} 1279 1280static int 1281unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1282 struct thread *td) 1283{ 1284 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1285 struct vnode *vp; 1286 struct socket *so2, *so3; 1287 struct unpcb *unp, *unp2, *unp3; 1288 struct nameidata nd; 1289 char buf[SOCK_MAXADDRLEN]; 1290 struct sockaddr *sa; 1291 cap_rights_t rights; 1292 int error, len; 1293 1294 UNP_LINK_WLOCK_ASSERT(); 1295 1296 unp = sotounpcb(so); 1297 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1298 1299 if (nam->sa_len > sizeof(struct sockaddr_un)) 1300 return (EINVAL); 1301 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1302 if (len <= 0) 1303 return (EINVAL); 1304 bcopy(soun->sun_path, buf, len); 1305 buf[len] = 0; 1306 1307 UNP_PCB_LOCK(unp); 1308 if (unp->unp_flags & UNP_CONNECTING) { 1309 UNP_PCB_UNLOCK(unp); 1310 return (EALREADY); 1311 } 1312 UNP_LINK_WUNLOCK(); 1313 unp->unp_flags |= UNP_CONNECTING; 1314 UNP_PCB_UNLOCK(unp); 1315 1316 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1317 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1318 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td); 1319 error = namei(&nd); 1320 if (error) 1321 vp = NULL; 1322 else 1323 vp = nd.ni_vp; 1324 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1325 NDFREE(&nd, NDF_ONLY_PNBUF); 1326 if (error) 1327 goto bad; 1328 1329 if (vp->v_type != VSOCK) { 1330 error = ENOTSOCK; 1331 goto bad; 1332 } 1333#ifdef MAC 1334 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1335 if (error) 1336 goto bad; 1337#endif 1338 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1339 if (error) 1340 goto bad; 1341 1342 unp = sotounpcb(so); 1343 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1344 1345 /* 1346 * Lock linkage lock for two reasons: make sure v_socket is stable, 1347 * and to protect simultaneous locking of multiple pcbs. 1348 */ 1349 UNP_LINK_WLOCK(); 1350 VOP_UNP_CONNECT(vp, &so2); 1351 if (so2 == NULL) { 1352 error = ECONNREFUSED; 1353 goto bad2; 1354 } 1355 if (so->so_type != so2->so_type) { 1356 error = EPROTOTYPE; 1357 goto bad2; 1358 } 1359 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1360 if (so2->so_options & SO_ACCEPTCONN) { 1361 CURVNET_SET(so2->so_vnet); 1362 so3 = sonewconn(so2, 0); 1363 CURVNET_RESTORE(); 1364 } else 1365 so3 = NULL; 1366 if (so3 == NULL) { 1367 error = ECONNREFUSED; 1368 goto bad2; 1369 } 1370 unp = sotounpcb(so); 1371 unp2 = sotounpcb(so2); 1372 unp3 = sotounpcb(so3); 1373 UNP_PCB_LOCK(unp); 1374 UNP_PCB_LOCK(unp2); 1375 UNP_PCB_LOCK(unp3); 1376 if (unp2->unp_addr != NULL) { 1377 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1378 unp3->unp_addr = (struct sockaddr_un *) sa; 1379 sa = NULL; 1380 } 1381 1382 /* 1383 * The connector's (client's) credentials are copied from its 1384 * process structure at the time of connect() (which is now). 1385 */ 1386 cru2x(td->td_ucred, &unp3->unp_peercred); 1387 unp3->unp_flags |= UNP_HAVEPC; 1388 1389 /* 1390 * The receiver's (server's) credentials are copied from the 1391 * unp_peercred member of socket on which the former called 1392 * listen(); uipc_listen() cached that process's credentials 1393 * at that time so we can use them now. 1394 */ 1395 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1396 ("unp_connect: listener without cached peercred")); 1397 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1398 sizeof(unp->unp_peercred)); 1399 unp->unp_flags |= UNP_HAVEPC; 1400 if (unp2->unp_flags & UNP_WANTCRED) 1401 unp3->unp_flags |= UNP_WANTCRED; 1402 UNP_PCB_UNLOCK(unp3); 1403 UNP_PCB_UNLOCK(unp2); 1404 UNP_PCB_UNLOCK(unp); 1405#ifdef MAC 1406 mac_socketpeer_set_from_socket(so, so3); 1407 mac_socketpeer_set_from_socket(so3, so); 1408#endif 1409 1410 so2 = so3; 1411 } 1412 unp = sotounpcb(so); 1413 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1414 unp2 = sotounpcb(so2); 1415 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1416 UNP_PCB_LOCK(unp); 1417 UNP_PCB_LOCK(unp2); 1418 error = unp_connect2(so, so2, PRU_CONNECT); 1419 UNP_PCB_UNLOCK(unp2); 1420 UNP_PCB_UNLOCK(unp); 1421bad2: 1422 UNP_LINK_WUNLOCK(); 1423bad: 1424 if (vp != NULL) 1425 vput(vp); 1426 free(sa, M_SONAME); 1427 UNP_LINK_WLOCK(); 1428 UNP_PCB_LOCK(unp); 1429 unp->unp_flags &= ~UNP_CONNECTING; 1430 UNP_PCB_UNLOCK(unp); 1431 return (error); 1432} 1433 1434static int 1435unp_connect2(struct socket *so, struct socket *so2, int req) 1436{ 1437 struct unpcb *unp; 1438 struct unpcb *unp2; 1439 1440 unp = sotounpcb(so); 1441 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1442 unp2 = sotounpcb(so2); 1443 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1444 1445 UNP_LINK_WLOCK_ASSERT(); 1446 UNP_PCB_LOCK_ASSERT(unp); 1447 UNP_PCB_LOCK_ASSERT(unp2); 1448 1449 if (so2->so_type != so->so_type) 1450 return (EPROTOTYPE); 1451 unp->unp_conn = unp2; 1452 1453 switch (so->so_type) { 1454 case SOCK_DGRAM: 1455 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1456 soisconnected(so); 1457 break; 1458 1459 case SOCK_STREAM: 1460 case SOCK_SEQPACKET: 1461 unp2->unp_conn = unp; 1462 if (req == PRU_CONNECT && 1463 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1464 soisconnecting(so); 1465 else 1466 soisconnected(so); 1467 soisconnected(so2); 1468 break; 1469 1470 default: 1471 panic("unp_connect2"); 1472 } 1473 return (0); 1474} 1475 1476static void 1477unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1478{ 1479 struct socket *so; 1480 1481 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1482 1483 UNP_LINK_WLOCK_ASSERT(); 1484 UNP_PCB_LOCK_ASSERT(unp); 1485 UNP_PCB_LOCK_ASSERT(unp2); 1486 1487 unp->unp_conn = NULL; 1488 switch (unp->unp_socket->so_type) { 1489 case SOCK_DGRAM: 1490 LIST_REMOVE(unp, unp_reflink); 1491 so = unp->unp_socket; 1492 SOCK_LOCK(so); 1493 so->so_state &= ~SS_ISCONNECTED; 1494 SOCK_UNLOCK(so); 1495 break; 1496 1497 case SOCK_STREAM: 1498 case SOCK_SEQPACKET: 1499 soisdisconnected(unp->unp_socket); 1500 unp2->unp_conn = NULL; 1501 soisdisconnected(unp2->unp_socket); 1502 break; 1503 } 1504} 1505 1506/* 1507 * unp_pcblist() walks the global list of struct unpcb's to generate a 1508 * pointer list, bumping the refcount on each unpcb. It then copies them out 1509 * sequentially, validating the generation number on each to see if it has 1510 * been detached. All of this is necessary because copyout() may sleep on 1511 * disk I/O. 1512 */ 1513static int 1514unp_pcblist(SYSCTL_HANDLER_ARGS) 1515{ 1516 int error, i, n; 1517 int freeunp; 1518 struct unpcb *unp, **unp_list; 1519 unp_gen_t gencnt; 1520 struct xunpgen *xug; 1521 struct unp_head *head; 1522 struct xunpcb *xu; 1523 1524 switch ((intptr_t)arg1) { 1525 case SOCK_STREAM: 1526 head = &unp_shead; 1527 break; 1528 1529 case SOCK_DGRAM: 1530 head = &unp_dhead; 1531 break; 1532 1533 case SOCK_SEQPACKET: 1534 head = &unp_sphead; 1535 break; 1536 1537 default: 1538 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1539 } 1540 1541 /* 1542 * The process of preparing the PCB list is too time-consuming and 1543 * resource-intensive to repeat twice on every request. 1544 */ 1545 if (req->oldptr == NULL) { 1546 n = unp_count; 1547 req->oldidx = 2 * (sizeof *xug) 1548 + (n + n/8) * sizeof(struct xunpcb); 1549 return (0); 1550 } 1551 1552 if (req->newptr != NULL) 1553 return (EPERM); 1554 1555 /* 1556 * OK, now we're committed to doing something. 1557 */ 1558 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1559 UNP_LIST_LOCK(); 1560 gencnt = unp_gencnt; 1561 n = unp_count; 1562 UNP_LIST_UNLOCK(); 1563 1564 xug->xug_len = sizeof *xug; 1565 xug->xug_count = n; 1566 xug->xug_gen = gencnt; 1567 xug->xug_sogen = so_gencnt; 1568 error = SYSCTL_OUT(req, xug, sizeof *xug); 1569 if (error) { 1570 free(xug, M_TEMP); 1571 return (error); 1572 } 1573 1574 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1575 1576 UNP_LIST_LOCK(); 1577 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1578 unp = LIST_NEXT(unp, unp_link)) { 1579 UNP_PCB_LOCK(unp); 1580 if (unp->unp_gencnt <= gencnt) { 1581 if (cr_cansee(req->td->td_ucred, 1582 unp->unp_socket->so_cred)) { 1583 UNP_PCB_UNLOCK(unp); 1584 continue; 1585 } 1586 unp_list[i++] = unp; 1587 unp->unp_refcount++; 1588 } 1589 UNP_PCB_UNLOCK(unp); 1590 } 1591 UNP_LIST_UNLOCK(); 1592 n = i; /* In case we lost some during malloc. */ 1593 1594 error = 0; 1595 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1596 for (i = 0; i < n; i++) { 1597 unp = unp_list[i]; 1598 UNP_PCB_LOCK(unp); 1599 unp->unp_refcount--; 1600 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1601 xu->xu_len = sizeof *xu; 1602 xu->xu_unpp = unp; 1603 /* 1604 * XXX - need more locking here to protect against 1605 * connect/disconnect races for SMP. 1606 */ 1607 if (unp->unp_addr != NULL) 1608 bcopy(unp->unp_addr, &xu->xu_addr, 1609 unp->unp_addr->sun_len); 1610 if (unp->unp_conn != NULL && 1611 unp->unp_conn->unp_addr != NULL) 1612 bcopy(unp->unp_conn->unp_addr, 1613 &xu->xu_caddr, 1614 unp->unp_conn->unp_addr->sun_len); 1615 bcopy(unp, &xu->xu_unp, sizeof *unp); 1616 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1617 UNP_PCB_UNLOCK(unp); 1618 error = SYSCTL_OUT(req, xu, sizeof *xu); 1619 } else { 1620 freeunp = (unp->unp_refcount == 0); 1621 UNP_PCB_UNLOCK(unp); 1622 if (freeunp) { 1623 UNP_PCB_LOCK_DESTROY(unp); 1624 uma_zfree(unp_zone, unp); 1625 } 1626 } 1627 } 1628 free(xu, M_TEMP); 1629 if (!error) { 1630 /* 1631 * Give the user an updated idea of our state. If the 1632 * generation differs from what we told her before, she knows 1633 * that something happened while we were processing this 1634 * request, and it might be necessary to retry. 1635 */ 1636 xug->xug_gen = unp_gencnt; 1637 xug->xug_sogen = so_gencnt; 1638 xug->xug_count = unp_count; 1639 error = SYSCTL_OUT(req, xug, sizeof *xug); 1640 } 1641 free(unp_list, M_TEMP); 1642 free(xug, M_TEMP); 1643 return (error); 1644} 1645 1646SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1647 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1648 "List of active local datagram sockets"); 1649SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1650 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1651 "List of active local stream sockets"); 1652SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1653 CTLTYPE_OPAQUE | CTLFLAG_RD, 1654 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1655 "List of active local seqpacket sockets"); 1656 1657static void 1658unp_shutdown(struct unpcb *unp) 1659{ 1660 struct unpcb *unp2; 1661 struct socket *so; 1662 1663 UNP_LINK_WLOCK_ASSERT(); 1664 UNP_PCB_LOCK_ASSERT(unp); 1665 1666 unp2 = unp->unp_conn; 1667 if ((unp->unp_socket->so_type == SOCK_STREAM || 1668 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1669 so = unp2->unp_socket; 1670 if (so != NULL) 1671 socantrcvmore(so); 1672 } 1673} 1674 1675static void 1676unp_drop(struct unpcb *unp, int errno) 1677{ 1678 struct socket *so = unp->unp_socket; 1679 struct unpcb *unp2; 1680 1681 UNP_LINK_WLOCK_ASSERT(); 1682 UNP_PCB_LOCK_ASSERT(unp); 1683 1684 so->so_error = errno; 1685 unp2 = unp->unp_conn; 1686 if (unp2 == NULL) 1687 return; 1688 UNP_PCB_LOCK(unp2); 1689 unp_disconnect(unp, unp2); 1690 UNP_PCB_UNLOCK(unp2); 1691} 1692 1693static void 1694unp_freerights(struct filedescent **fdep, int fdcount) 1695{ 1696 struct file *fp; 1697 int i; 1698 1699 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 1700 1701 for (i = 0; i < fdcount; i++) { 1702 fp = fdep[i]->fde_file; 1703 filecaps_free(&fdep[i]->fde_caps); 1704 unp_discard(fp); 1705 } 1706 free(fdep[0], M_FILECAPS); 1707} 1708 1709static int 1710unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 1711{ 1712 struct thread *td = curthread; /* XXX */ 1713 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1714 int i; 1715 int *fdp; 1716 struct filedesc *fdesc = td->td_proc->p_fd; 1717 struct filedescent *fde, **fdep; 1718 void *data; 1719 socklen_t clen = control->m_len, datalen; 1720 int error, newfds; 1721 u_int newlen; 1722 1723 UNP_LINK_UNLOCK_ASSERT(); 1724 1725 error = 0; 1726 if (controlp != NULL) /* controlp == NULL => free control messages */ 1727 *controlp = NULL; 1728 while (cm != NULL) { 1729 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1730 error = EINVAL; 1731 break; 1732 } 1733 data = CMSG_DATA(cm); 1734 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1735 if (cm->cmsg_level == SOL_SOCKET 1736 && cm->cmsg_type == SCM_RIGHTS) { 1737 newfds = datalen / sizeof(*fdep); 1738 if (newfds == 0) 1739 goto next; 1740 fdep = data; 1741 1742 /* If we're not outputting the descriptors free them. */ 1743 if (error || controlp == NULL) { 1744 unp_freerights(fdep, newfds); 1745 goto next; 1746 } 1747 FILEDESC_XLOCK(fdesc); 1748 1749 /* 1750 * Now change each pointer to an fd in the global 1751 * table to an integer that is the index to the local 1752 * fd table entry that we set up to point to the 1753 * global one we are transferring. 1754 */ 1755 newlen = newfds * sizeof(int); 1756 *controlp = sbcreatecontrol(NULL, newlen, 1757 SCM_RIGHTS, SOL_SOCKET); 1758 if (*controlp == NULL) { 1759 FILEDESC_XUNLOCK(fdesc); 1760 error = E2BIG; 1761 unp_freerights(fdep, newfds); 1762 goto next; 1763 } 1764 1765 fdp = (int *) 1766 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1767 if (fdallocn(td, 0, fdp, newfds) != 0) { 1768 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1769 error = EMSGSIZE; 1770 unp_freerights(fdep, newfds); 1771 m_freem(*controlp); 1772 *controlp = NULL; 1773 goto next; 1774 } 1775 for (i = 0; i < newfds; i++, fdp++) { 1776 fde = &fdesc->fd_ofiles[*fdp]; 1777 fde->fde_file = fdep[i]->fde_file; 1778 filecaps_move(&fdep[i]->fde_caps, 1779 &fde->fde_caps); 1780 if ((flags & MSG_CMSG_CLOEXEC) != 0) 1781 fde->fde_flags |= UF_EXCLOSE; 1782 unp_externalize_fp(fde->fde_file); 1783 } 1784 FILEDESC_XUNLOCK(fdesc); 1785 free(fdep[0], M_FILECAPS); 1786 } else { 1787 /* We can just copy anything else across. */ 1788 if (error || controlp == NULL) 1789 goto next; 1790 *controlp = sbcreatecontrol(NULL, datalen, 1791 cm->cmsg_type, cm->cmsg_level); 1792 if (*controlp == NULL) { 1793 error = ENOBUFS; 1794 goto next; 1795 } 1796 bcopy(data, 1797 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1798 datalen); 1799 } 1800 controlp = &(*controlp)->m_next; 1801 1802next: 1803 if (CMSG_SPACE(datalen) < clen) { 1804 clen -= CMSG_SPACE(datalen); 1805 cm = (struct cmsghdr *) 1806 ((caddr_t)cm + CMSG_SPACE(datalen)); 1807 } else { 1808 clen = 0; 1809 cm = NULL; 1810 } 1811 } 1812 1813 m_freem(control); 1814 return (error); 1815} 1816 1817static void 1818unp_zone_change(void *tag) 1819{ 1820 1821 uma_zone_set_max(unp_zone, maxsockets); 1822} 1823 1824static void 1825unp_init(void) 1826{ 1827 1828#ifdef VIMAGE 1829 if (!IS_DEFAULT_VNET(curvnet)) 1830 return; 1831#endif 1832 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1833 NULL, NULL, UMA_ALIGN_PTR, 0); 1834 if (unp_zone == NULL) 1835 panic("unp_init"); 1836 uma_zone_set_max(unp_zone, maxsockets); 1837 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 1838 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1839 NULL, EVENTHANDLER_PRI_ANY); 1840 LIST_INIT(&unp_dhead); 1841 LIST_INIT(&unp_shead); 1842 LIST_INIT(&unp_sphead); 1843 SLIST_INIT(&unp_defers); 1844 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 1845 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1846 UNP_LINK_LOCK_INIT(); 1847 UNP_LIST_LOCK_INIT(); 1848 UNP_DEFERRED_LOCK_INIT(); 1849} 1850 1851static int 1852unp_internalize(struct mbuf **controlp, struct thread *td) 1853{ 1854 struct mbuf *control = *controlp; 1855 struct proc *p = td->td_proc; 1856 struct filedesc *fdesc = p->p_fd; 1857 struct bintime *bt; 1858 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1859 struct cmsgcred *cmcred; 1860 struct filedescent *fde, **fdep, *fdev; 1861 struct file *fp; 1862 struct timeval *tv; 1863 int i, fd, *fdp; 1864 void *data; 1865 socklen_t clen = control->m_len, datalen; 1866 int error, oldfds; 1867 u_int newlen; 1868 1869 UNP_LINK_UNLOCK_ASSERT(); 1870 1871 error = 0; 1872 *controlp = NULL; 1873 while (cm != NULL) { 1874 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1875 || cm->cmsg_len > clen) { 1876 error = EINVAL; 1877 goto out; 1878 } 1879 data = CMSG_DATA(cm); 1880 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1881 1882 switch (cm->cmsg_type) { 1883 /* 1884 * Fill in credential information. 1885 */ 1886 case SCM_CREDS: 1887 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1888 SCM_CREDS, SOL_SOCKET); 1889 if (*controlp == NULL) { 1890 error = ENOBUFS; 1891 goto out; 1892 } 1893 cmcred = (struct cmsgcred *) 1894 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1895 cmcred->cmcred_pid = p->p_pid; 1896 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1897 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1898 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1899 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1900 CMGROUP_MAX); 1901 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1902 cmcred->cmcred_groups[i] = 1903 td->td_ucred->cr_groups[i]; 1904 break; 1905 1906 case SCM_RIGHTS: 1907 oldfds = datalen / sizeof (int); 1908 if (oldfds == 0) 1909 break; 1910 /* 1911 * Check that all the FDs passed in refer to legal 1912 * files. If not, reject the entire operation. 1913 */ 1914 fdp = data; 1915 FILEDESC_SLOCK(fdesc); 1916 for (i = 0; i < oldfds; i++) { 1917 fd = *fdp++; 1918 if (fget_locked(fdesc, fd) == NULL) { 1919 FILEDESC_SUNLOCK(fdesc); 1920 error = EBADF; 1921 goto out; 1922 } 1923 fp = fdesc->fd_ofiles[fd].fde_file; 1924 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1925 FILEDESC_SUNLOCK(fdesc); 1926 error = EOPNOTSUPP; 1927 goto out; 1928 } 1929 1930 } 1931 1932 /* 1933 * Now replace the integer FDs with pointers to the 1934 * file structure and capability rights. 1935 */ 1936 newlen = oldfds * sizeof(fdep[0]); 1937 *controlp = sbcreatecontrol(NULL, newlen, 1938 SCM_RIGHTS, SOL_SOCKET); 1939 if (*controlp == NULL) { 1940 FILEDESC_SUNLOCK(fdesc); 1941 error = E2BIG; 1942 goto out; 1943 } 1944 fdp = data; 1945 fdep = (struct filedescent **) 1946 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1947 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 1948 M_WAITOK); 1949 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 1950 fde = &fdesc->fd_ofiles[*fdp]; 1951 fdep[i] = fdev; 1952 fdep[i]->fde_file = fde->fde_file; 1953 filecaps_copy(&fde->fde_caps, 1954 &fdep[i]->fde_caps); 1955 unp_internalize_fp(fdep[i]->fde_file); 1956 } 1957 FILEDESC_SUNLOCK(fdesc); 1958 break; 1959 1960 case SCM_TIMESTAMP: 1961 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1962 SCM_TIMESTAMP, SOL_SOCKET); 1963 if (*controlp == NULL) { 1964 error = ENOBUFS; 1965 goto out; 1966 } 1967 tv = (struct timeval *) 1968 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1969 microtime(tv); 1970 break; 1971 1972 case SCM_BINTIME: 1973 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 1974 SCM_BINTIME, SOL_SOCKET); 1975 if (*controlp == NULL) { 1976 error = ENOBUFS; 1977 goto out; 1978 } 1979 bt = (struct bintime *) 1980 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1981 bintime(bt); 1982 break; 1983 1984 default: 1985 error = EINVAL; 1986 goto out; 1987 } 1988 1989 controlp = &(*controlp)->m_next; 1990 if (CMSG_SPACE(datalen) < clen) { 1991 clen -= CMSG_SPACE(datalen); 1992 cm = (struct cmsghdr *) 1993 ((caddr_t)cm + CMSG_SPACE(datalen)); 1994 } else { 1995 clen = 0; 1996 cm = NULL; 1997 } 1998 } 1999 2000out: 2001 m_freem(control); 2002 return (error); 2003} 2004 2005static struct mbuf * 2006unp_addsockcred(struct thread *td, struct mbuf *control) 2007{ 2008 struct mbuf *m, *n, *n_prev; 2009 struct sockcred *sc; 2010 const struct cmsghdr *cm; 2011 int ngroups; 2012 int i; 2013 2014 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2015 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 2016 if (m == NULL) 2017 return (control); 2018 2019 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 2020 sc->sc_uid = td->td_ucred->cr_ruid; 2021 sc->sc_euid = td->td_ucred->cr_uid; 2022 sc->sc_gid = td->td_ucred->cr_rgid; 2023 sc->sc_egid = td->td_ucred->cr_gid; 2024 sc->sc_ngroups = ngroups; 2025 for (i = 0; i < sc->sc_ngroups; i++) 2026 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2027 2028 /* 2029 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2030 * created SCM_CREDS control message (struct sockcred) has another 2031 * format. 2032 */ 2033 if (control != NULL) 2034 for (n = control, n_prev = NULL; n != NULL;) { 2035 cm = mtod(n, struct cmsghdr *); 2036 if (cm->cmsg_level == SOL_SOCKET && 2037 cm->cmsg_type == SCM_CREDS) { 2038 if (n_prev == NULL) 2039 control = n->m_next; 2040 else 2041 n_prev->m_next = n->m_next; 2042 n = m_free(n); 2043 } else { 2044 n_prev = n; 2045 n = n->m_next; 2046 } 2047 } 2048 2049 /* Prepend it to the head. */ 2050 m->m_next = control; 2051 return (m); 2052} 2053 2054static struct unpcb * 2055fptounp(struct file *fp) 2056{ 2057 struct socket *so; 2058 2059 if (fp->f_type != DTYPE_SOCKET) 2060 return (NULL); 2061 if ((so = fp->f_data) == NULL) 2062 return (NULL); 2063 if (so->so_proto->pr_domain != &localdomain) 2064 return (NULL); 2065 return sotounpcb(so); 2066} 2067 2068static void 2069unp_discard(struct file *fp) 2070{ 2071 struct unp_defer *dr; 2072 2073 if (unp_externalize_fp(fp)) { 2074 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2075 dr->ud_fp = fp; 2076 UNP_DEFERRED_LOCK(); 2077 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2078 UNP_DEFERRED_UNLOCK(); 2079 atomic_add_int(&unp_defers_count, 1); 2080 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2081 } else 2082 (void) closef(fp, (struct thread *)NULL); 2083} 2084 2085static void 2086unp_process_defers(void *arg __unused, int pending) 2087{ 2088 struct unp_defer *dr; 2089 SLIST_HEAD(, unp_defer) drl; 2090 int count; 2091 2092 SLIST_INIT(&drl); 2093 for (;;) { 2094 UNP_DEFERRED_LOCK(); 2095 if (SLIST_FIRST(&unp_defers) == NULL) { 2096 UNP_DEFERRED_UNLOCK(); 2097 break; 2098 } 2099 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2100 UNP_DEFERRED_UNLOCK(); 2101 count = 0; 2102 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2103 SLIST_REMOVE_HEAD(&drl, ud_link); 2104 closef(dr->ud_fp, NULL); 2105 free(dr, M_TEMP); 2106 count++; 2107 } 2108 atomic_add_int(&unp_defers_count, -count); 2109 } 2110} 2111 2112static void 2113unp_internalize_fp(struct file *fp) 2114{ 2115 struct unpcb *unp; 2116 2117 UNP_LINK_WLOCK(); 2118 if ((unp = fptounp(fp)) != NULL) { 2119 unp->unp_file = fp; 2120 unp->unp_msgcount++; 2121 } 2122 fhold(fp); 2123 unp_rights++; 2124 UNP_LINK_WUNLOCK(); 2125} 2126 2127static int 2128unp_externalize_fp(struct file *fp) 2129{ 2130 struct unpcb *unp; 2131 int ret; 2132 2133 UNP_LINK_WLOCK(); 2134 if ((unp = fptounp(fp)) != NULL) { 2135 unp->unp_msgcount--; 2136 ret = 1; 2137 } else 2138 ret = 0; 2139 unp_rights--; 2140 UNP_LINK_WUNLOCK(); 2141 return (ret); 2142} 2143 2144/* 2145 * unp_defer indicates whether additional work has been defered for a future 2146 * pass through unp_gc(). It is thread local and does not require explicit 2147 * synchronization. 2148 */ 2149static int unp_marked; 2150static int unp_unreachable; 2151 2152static void 2153unp_accessable(struct filedescent **fdep, int fdcount) 2154{ 2155 struct unpcb *unp; 2156 struct file *fp; 2157 int i; 2158 2159 for (i = 0; i < fdcount; i++) { 2160 fp = fdep[i]->fde_file; 2161 if ((unp = fptounp(fp)) == NULL) 2162 continue; 2163 if (unp->unp_gcflag & UNPGC_REF) 2164 continue; 2165 unp->unp_gcflag &= ~UNPGC_DEAD; 2166 unp->unp_gcflag |= UNPGC_REF; 2167 unp_marked++; 2168 } 2169} 2170 2171static void 2172unp_gc_process(struct unpcb *unp) 2173{ 2174 struct socket *soa; 2175 struct socket *so; 2176 struct file *fp; 2177 2178 /* Already processed. */ 2179 if (unp->unp_gcflag & UNPGC_SCANNED) 2180 return; 2181 fp = unp->unp_file; 2182 2183 /* 2184 * Check for a socket potentially in a cycle. It must be in a 2185 * queue as indicated by msgcount, and this must equal the file 2186 * reference count. Note that when msgcount is 0 the file is NULL. 2187 */ 2188 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2189 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2190 unp->unp_gcflag |= UNPGC_DEAD; 2191 unp_unreachable++; 2192 return; 2193 } 2194 2195 /* 2196 * Mark all sockets we reference with RIGHTS. 2197 */ 2198 so = unp->unp_socket; 2199 SOCKBUF_LOCK(&so->so_rcv); 2200 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2201 SOCKBUF_UNLOCK(&so->so_rcv); 2202 2203 /* 2204 * Mark all sockets in our accept queue. 2205 */ 2206 ACCEPT_LOCK(); 2207 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2208 SOCKBUF_LOCK(&soa->so_rcv); 2209 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2210 SOCKBUF_UNLOCK(&soa->so_rcv); 2211 } 2212 ACCEPT_UNLOCK(); 2213 unp->unp_gcflag |= UNPGC_SCANNED; 2214} 2215 2216static int unp_recycled; 2217SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2218 "Number of unreachable sockets claimed by the garbage collector."); 2219 2220static int unp_taskcount; 2221SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2222 "Number of times the garbage collector has run."); 2223 2224static void 2225unp_gc(__unused void *arg, int pending) 2226{ 2227 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2228 NULL }; 2229 struct unp_head **head; 2230 struct file *f, **unref; 2231 struct unpcb *unp; 2232 int i, total; 2233 2234 unp_taskcount++; 2235 UNP_LIST_LOCK(); 2236 /* 2237 * First clear all gc flags from previous runs. 2238 */ 2239 for (head = heads; *head != NULL; head++) 2240 LIST_FOREACH(unp, *head, unp_link) 2241 unp->unp_gcflag = 0; 2242 2243 /* 2244 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2245 * is reachable all of the sockets it references are reachable. 2246 * Stop the scan once we do a complete loop without discovering 2247 * a new reachable socket. 2248 */ 2249 do { 2250 unp_unreachable = 0; 2251 unp_marked = 0; 2252 for (head = heads; *head != NULL; head++) 2253 LIST_FOREACH(unp, *head, unp_link) 2254 unp_gc_process(unp); 2255 } while (unp_marked); 2256 UNP_LIST_UNLOCK(); 2257 if (unp_unreachable == 0) 2258 return; 2259 2260 /* 2261 * Allocate space for a local list of dead unpcbs. 2262 */ 2263 unref = malloc(unp_unreachable * sizeof(struct file *), 2264 M_TEMP, M_WAITOK); 2265 2266 /* 2267 * Iterate looking for sockets which have been specifically marked 2268 * as as unreachable and store them locally. 2269 */ 2270 UNP_LINK_RLOCK(); 2271 UNP_LIST_LOCK(); 2272 for (total = 0, head = heads; *head != NULL; head++) 2273 LIST_FOREACH(unp, *head, unp_link) 2274 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2275 f = unp->unp_file; 2276 if (unp->unp_msgcount == 0 || f == NULL || 2277 f->f_count != unp->unp_msgcount) 2278 continue; 2279 unref[total++] = f; 2280 fhold(f); 2281 KASSERT(total <= unp_unreachable, 2282 ("unp_gc: incorrect unreachable count.")); 2283 } 2284 UNP_LIST_UNLOCK(); 2285 UNP_LINK_RUNLOCK(); 2286 2287 /* 2288 * Now flush all sockets, free'ing rights. This will free the 2289 * struct files associated with these sockets but leave each socket 2290 * with one remaining ref. 2291 */ 2292 for (i = 0; i < total; i++) { 2293 struct socket *so; 2294 2295 so = unref[i]->f_data; 2296 CURVNET_SET(so->so_vnet); 2297 sorflush(so); 2298 CURVNET_RESTORE(); 2299 } 2300 2301 /* 2302 * And finally release the sockets so they can be reclaimed. 2303 */ 2304 for (i = 0; i < total; i++) 2305 fdrop(unref[i], NULL); 2306 unp_recycled += total; 2307 free(unref, M_TEMP); 2308} 2309 2310static void 2311unp_dispose(struct mbuf *m) 2312{ 2313 2314 if (m) 2315 unp_scan(m, unp_freerights); 2316} 2317 2318static void 2319unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 2320{ 2321 struct mbuf *m; 2322 struct cmsghdr *cm; 2323 void *data; 2324 socklen_t clen, datalen; 2325 2326 while (m0 != NULL) { 2327 for (m = m0; m; m = m->m_next) { 2328 if (m->m_type != MT_CONTROL) 2329 continue; 2330 2331 cm = mtod(m, struct cmsghdr *); 2332 clen = m->m_len; 2333 2334 while (cm != NULL) { 2335 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2336 break; 2337 2338 data = CMSG_DATA(cm); 2339 datalen = (caddr_t)cm + cm->cmsg_len 2340 - (caddr_t)data; 2341 2342 if (cm->cmsg_level == SOL_SOCKET && 2343 cm->cmsg_type == SCM_RIGHTS) { 2344 (*op)(data, datalen / 2345 sizeof(struct filedescent *)); 2346 } 2347 2348 if (CMSG_SPACE(datalen) < clen) { 2349 clen -= CMSG_SPACE(datalen); 2350 cm = (struct cmsghdr *) 2351 ((caddr_t)cm + CMSG_SPACE(datalen)); 2352 } else { 2353 clen = 0; 2354 cm = NULL; 2355 } 2356 } 2357 } 2358 m0 = m0->m_act; 2359 } 2360} 2361 2362/* 2363 * A helper function called by VFS before socket-type vnode reclamation. 2364 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2365 * use count. 2366 */ 2367void 2368vfs_unp_reclaim(struct vnode *vp) 2369{ 2370 struct socket *so; 2371 struct unpcb *unp; 2372 int active; 2373 2374 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2375 KASSERT(vp->v_type == VSOCK, 2376 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2377 2378 active = 0; 2379 UNP_LINK_WLOCK(); 2380 VOP_UNP_CONNECT(vp, &so); 2381 if (so == NULL) 2382 goto done; 2383 unp = sotounpcb(so); 2384 if (unp == NULL) 2385 goto done; 2386 UNP_PCB_LOCK(unp); 2387 if (unp->unp_vnode == vp) { 2388 VOP_UNP_DETACH(vp); 2389 unp->unp_vnode = NULL; 2390 active = 1; 2391 } 2392 UNP_PCB_UNLOCK(unp); 2393done: 2394 UNP_LINK_WUNLOCK(); 2395 if (active) 2396 vunref(vp); 2397} 2398 2399#ifdef DDB 2400static void 2401db_print_indent(int indent) 2402{ 2403 int i; 2404 2405 for (i = 0; i < indent; i++) 2406 db_printf(" "); 2407} 2408 2409static void 2410db_print_unpflags(int unp_flags) 2411{ 2412 int comma; 2413 2414 comma = 0; 2415 if (unp_flags & UNP_HAVEPC) { 2416 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2417 comma = 1; 2418 } 2419 if (unp_flags & UNP_HAVEPCCACHED) { 2420 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2421 comma = 1; 2422 } 2423 if (unp_flags & UNP_WANTCRED) { 2424 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2425 comma = 1; 2426 } 2427 if (unp_flags & UNP_CONNWAIT) { 2428 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2429 comma = 1; 2430 } 2431 if (unp_flags & UNP_CONNECTING) { 2432 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2433 comma = 1; 2434 } 2435 if (unp_flags & UNP_BINDING) { 2436 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2437 comma = 1; 2438 } 2439} 2440 2441static void 2442db_print_xucred(int indent, struct xucred *xu) 2443{ 2444 int comma, i; 2445 2446 db_print_indent(indent); 2447 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2448 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2449 db_print_indent(indent); 2450 db_printf("cr_groups: "); 2451 comma = 0; 2452 for (i = 0; i < xu->cr_ngroups; i++) { 2453 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2454 comma = 1; 2455 } 2456 db_printf("\n"); 2457} 2458 2459static void 2460db_print_unprefs(int indent, struct unp_head *uh) 2461{ 2462 struct unpcb *unp; 2463 int counter; 2464 2465 counter = 0; 2466 LIST_FOREACH(unp, uh, unp_reflink) { 2467 if (counter % 4 == 0) 2468 db_print_indent(indent); 2469 db_printf("%p ", unp); 2470 if (counter % 4 == 3) 2471 db_printf("\n"); 2472 counter++; 2473 } 2474 if (counter != 0 && counter % 4 != 0) 2475 db_printf("\n"); 2476} 2477 2478DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2479{ 2480 struct unpcb *unp; 2481 2482 if (!have_addr) { 2483 db_printf("usage: show unpcb <addr>\n"); 2484 return; 2485 } 2486 unp = (struct unpcb *)addr; 2487 2488 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2489 unp->unp_vnode); 2490 2491 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 2492 unp->unp_conn); 2493 2494 db_printf("unp_refs:\n"); 2495 db_print_unprefs(2, &unp->unp_refs); 2496 2497 /* XXXRW: Would be nice to print the full address, if any. */ 2498 db_printf("unp_addr: %p\n", unp->unp_addr); 2499 2500 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2501 unp->unp_cc, unp->unp_mbcnt, 2502 (unsigned long long)unp->unp_gencnt); 2503 2504 db_printf("unp_flags: %x (", unp->unp_flags); 2505 db_print_unpflags(unp->unp_flags); 2506 db_printf(")\n"); 2507 2508 db_printf("unp_peercred:\n"); 2509 db_print_xucred(2, &unp->unp_peercred); 2510 2511 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2512} 2513#endif 2514