uipc_usrreq.c revision 247667
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: head/sys/kern/uipc_usrreq.c 247667 2013-03-02 21:11:30Z pjd $"); 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> 104 105/* 106 * Locking key: 107 * (l) Locked using list lock 108 * (g) Locked using linkage lock 109 */ 110 111static uma_zone_t unp_zone; 112static unp_gen_t unp_gencnt; /* (l) */ 113static u_int unp_count; /* (l) Count of local sockets. */ 114static ino_t unp_ino; /* Prototype for fake inode numbers. */ 115static int unp_rights; /* (g) File descriptors in flight. */ 116static struct unp_head unp_shead; /* (l) List of stream sockets. */ 117static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 118static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 119 120struct unp_defer { 121 SLIST_ENTRY(unp_defer) ud_link; 122 struct file *ud_fp; 123}; 124static SLIST_HEAD(, unp_defer) unp_defers; 125static int unp_defers_count; 126 127static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 128 129/* 130 * Garbage collection of cyclic file descriptor/socket references occurs 131 * asynchronously in a taskqueue context in order to avoid recursion and 132 * reentrance in the UNIX domain socket, file descriptor, and socket layer 133 * code. See unp_gc() for a full description. 134 */ 135static struct timeout_task unp_gc_task; 136 137/* 138 * The close of unix domain sockets attached as SCM_RIGHTS is 139 * postponed to the taskqueue, to avoid arbitrary recursion depth. 140 * The attached sockets might have another sockets attached. 141 */ 142static struct task unp_defer_task; 143 144/* 145 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 146 * stream sockets, although the total for sender and receiver is actually 147 * only PIPSIZ. 148 * 149 * Datagram sockets really use the sendspace as the maximum datagram size, 150 * and don't really want to reserve the sendspace. Their recvspace should be 151 * large enough for at least one max-size datagram plus address. 152 */ 153#ifndef PIPSIZ 154#define PIPSIZ 8192 155#endif 156static u_long unpst_sendspace = PIPSIZ; 157static u_long unpst_recvspace = PIPSIZ; 158static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 159static u_long unpdg_recvspace = 4*1024; 160static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 161static u_long unpsp_recvspace = PIPSIZ; 162 163static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 164static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, 165 "SOCK_STREAM"); 166static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 167static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, 168 "SOCK_SEQPACKET"); 169 170SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 171 &unpst_sendspace, 0, "Default stream send space."); 172SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 173 &unpst_recvspace, 0, "Default stream receive space."); 174SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 175 &unpdg_sendspace, 0, "Default datagram send space."); 176SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 177 &unpdg_recvspace, 0, "Default datagram receive space."); 178SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 179 &unpsp_sendspace, 0, "Default seqpacket send space."); 180SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 181 &unpsp_recvspace, 0, "Default seqpacket receive space."); 182SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 183 "File descriptors in flight."); 184SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 185 &unp_defers_count, 0, 186 "File descriptors deferred to taskqueue for close."); 187 188/* 189 * Locking and synchronization: 190 * 191 * Three types of locks exit in the local domain socket implementation: a 192 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 193 * global locks, the list lock protects the socket count, global generation 194 * number, and stream/datagram global lists. The linkage lock protects the 195 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 196 * held exclusively over the acquisition of multiple unpcb locks to prevent 197 * deadlock. 198 * 199 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 200 * allocated in pru_attach() and freed in pru_detach(). The validity of that 201 * pointer is an invariant, so no lock is required to dereference the so_pcb 202 * pointer if a valid socket reference is held by the caller. In practice, 203 * this is always true during operations performed on a socket. Each unpcb 204 * has a back-pointer to its socket, unp_socket, which will be stable under 205 * the same circumstances. 206 * 207 * This pointer may only be safely dereferenced as long as a valid reference 208 * to the unpcb is held. Typically, this reference will be from the socket, 209 * or from another unpcb when the referring unpcb's lock is held (in order 210 * that the reference not be invalidated during use). For example, to follow 211 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 212 * as unp_socket remains valid as long as the reference to unp_conn is valid. 213 * 214 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 215 * atomic reads without the lock may be performed "lockless", but more 216 * complex reads and read-modify-writes require the mutex to be held. No 217 * lock order is defined between unpcb locks -- multiple unpcb locks may be 218 * acquired at the same time only when holding the linkage rwlock 219 * exclusively, which prevents deadlocks. 220 * 221 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 222 * protocols, bind() is a non-atomic operation, and connect() requires 223 * potential sleeping in the protocol, due to potentially waiting on local or 224 * distributed file systems. We try to separate "lookup" operations, which 225 * may sleep, and the IPC operations themselves, which typically can occur 226 * with relative atomicity as locks can be held over the entire operation. 227 * 228 * Another tricky issue is simultaneous multi-threaded or multi-process 229 * access to a single UNIX domain socket. These are handled by the flags 230 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 231 * binding, both of which involve dropping UNIX domain socket locks in order 232 * to perform namei() and other file system operations. 233 */ 234static struct rwlock unp_link_rwlock; 235static struct mtx unp_list_lock; 236static struct mtx unp_defers_lock; 237 238#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 239 "unp_link_rwlock") 240 241#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 242 RA_LOCKED) 243#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 244 RA_UNLOCKED) 245 246#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 247#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 248#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 249#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 250#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 251 RA_WLOCKED) 252 253#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 254 "unp_list_lock", NULL, MTX_DEF) 255#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 256#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 257 258#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 259 "unp_defer", NULL, MTX_DEF) 260#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 261#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 262 263#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 264 "unp_mtx", "unp_mtx", \ 265 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 266#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 267#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 268#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 269#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 270 271static int uipc_connect2(struct socket *, struct socket *); 272static int uipc_ctloutput(struct socket *, struct sockopt *); 273static int unp_connect(struct socket *, struct sockaddr *, 274 struct thread *); 275static int unp_connectat(int, struct socket *, struct sockaddr *, 276 struct thread *); 277static int unp_connect2(struct socket *so, struct socket *so2, int); 278static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 279static void unp_dispose(struct mbuf *); 280static void unp_shutdown(struct unpcb *); 281static void unp_drop(struct unpcb *, int); 282static void unp_gc(__unused void *, int); 283static void unp_scan(struct mbuf *, void (*)(struct file *)); 284static void unp_discard(struct file *); 285static void unp_freerights(struct filedescent *, int); 286static void unp_init(void); 287static int unp_internalize(struct mbuf **, struct thread *); 288static void unp_internalize_fp(struct file *); 289static int unp_externalize(struct mbuf *, struct mbuf **); 290static int unp_externalize_fp(struct file *); 291static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 292static void unp_process_defers(void * __unused, int); 293 294/* 295 * Definitions of protocols supported in the LOCAL domain. 296 */ 297static struct domain localdomain; 298static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 299static struct pr_usrreqs uipc_usrreqs_seqpacket; 300static struct protosw localsw[] = { 301{ 302 .pr_type = SOCK_STREAM, 303 .pr_domain = &localdomain, 304 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 305 .pr_ctloutput = &uipc_ctloutput, 306 .pr_usrreqs = &uipc_usrreqs_stream 307}, 308{ 309 .pr_type = SOCK_DGRAM, 310 .pr_domain = &localdomain, 311 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 312 .pr_ctloutput = &uipc_ctloutput, 313 .pr_usrreqs = &uipc_usrreqs_dgram 314}, 315{ 316 .pr_type = SOCK_SEQPACKET, 317 .pr_domain = &localdomain, 318 319 /* 320 * XXXRW: For now, PR_ADDR because soreceive will bump into them 321 * due to our use of sbappendaddr. A new sbappend variants is needed 322 * that supports both atomic record writes and control data. 323 */ 324 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 325 PR_RIGHTS, 326 .pr_usrreqs = &uipc_usrreqs_seqpacket, 327}, 328}; 329 330static struct domain localdomain = { 331 .dom_family = AF_LOCAL, 332 .dom_name = "local", 333 .dom_init = unp_init, 334 .dom_externalize = unp_externalize, 335 .dom_dispose = unp_dispose, 336 .dom_protosw = localsw, 337 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 338}; 339DOMAIN_SET(local); 340 341static void 342uipc_abort(struct socket *so) 343{ 344 struct unpcb *unp, *unp2; 345 346 unp = sotounpcb(so); 347 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 348 349 UNP_LINK_WLOCK(); 350 UNP_PCB_LOCK(unp); 351 unp2 = unp->unp_conn; 352 if (unp2 != NULL) { 353 UNP_PCB_LOCK(unp2); 354 unp_drop(unp2, ECONNABORTED); 355 UNP_PCB_UNLOCK(unp2); 356 } 357 UNP_PCB_UNLOCK(unp); 358 UNP_LINK_WUNLOCK(); 359} 360 361static int 362uipc_accept(struct socket *so, struct sockaddr **nam) 363{ 364 struct unpcb *unp, *unp2; 365 const struct sockaddr *sa; 366 367 /* 368 * Pass back name of connected socket, if it was bound and we are 369 * still connected (our peer may have closed already!). 370 */ 371 unp = sotounpcb(so); 372 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 373 374 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 375 UNP_LINK_RLOCK(); 376 unp2 = unp->unp_conn; 377 if (unp2 != NULL && unp2->unp_addr != NULL) { 378 UNP_PCB_LOCK(unp2); 379 sa = (struct sockaddr *) unp2->unp_addr; 380 bcopy(sa, *nam, sa->sa_len); 381 UNP_PCB_UNLOCK(unp2); 382 } else { 383 sa = &sun_noname; 384 bcopy(sa, *nam, sa->sa_len); 385 } 386 UNP_LINK_RUNLOCK(); 387 return (0); 388} 389 390static int 391uipc_attach(struct socket *so, int proto, struct thread *td) 392{ 393 u_long sendspace, recvspace; 394 struct unpcb *unp; 395 int error; 396 397 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 398 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 399 switch (so->so_type) { 400 case SOCK_STREAM: 401 sendspace = unpst_sendspace; 402 recvspace = unpst_recvspace; 403 break; 404 405 case SOCK_DGRAM: 406 sendspace = unpdg_sendspace; 407 recvspace = unpdg_recvspace; 408 break; 409 410 case SOCK_SEQPACKET: 411 sendspace = unpsp_sendspace; 412 recvspace = unpsp_recvspace; 413 break; 414 415 default: 416 panic("uipc_attach"); 417 } 418 error = soreserve(so, sendspace, recvspace); 419 if (error) 420 return (error); 421 } 422 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 423 if (unp == NULL) 424 return (ENOBUFS); 425 LIST_INIT(&unp->unp_refs); 426 UNP_PCB_LOCK_INIT(unp); 427 unp->unp_socket = so; 428 so->so_pcb = unp; 429 unp->unp_refcount = 1; 430 431 UNP_LIST_LOCK(); 432 unp->unp_gencnt = ++unp_gencnt; 433 unp_count++; 434 switch (so->so_type) { 435 case SOCK_STREAM: 436 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 437 break; 438 439 case SOCK_DGRAM: 440 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 441 break; 442 443 case SOCK_SEQPACKET: 444 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 445 break; 446 447 default: 448 panic("uipc_attach"); 449 } 450 UNP_LIST_UNLOCK(); 451 452 return (0); 453} 454 455static int 456uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 457{ 458 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 459 struct vattr vattr; 460 int error, namelen; 461 struct nameidata nd; 462 struct unpcb *unp; 463 struct vnode *vp; 464 struct mount *mp; 465 char *buf; 466 467 unp = sotounpcb(so); 468 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 469 470 if (soun->sun_len > sizeof(struct sockaddr_un)) 471 return (EINVAL); 472 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 473 if (namelen <= 0) 474 return (EINVAL); 475 476 /* 477 * We don't allow simultaneous bind() calls on a single UNIX domain 478 * socket, so flag in-progress operations, and return an error if an 479 * operation is already in progress. 480 * 481 * Historically, we have not allowed a socket to be rebound, so this 482 * also returns an error. Not allowing re-binding simplifies the 483 * implementation and avoids a great many possible failure modes. 484 */ 485 UNP_PCB_LOCK(unp); 486 if (unp->unp_vnode != NULL) { 487 UNP_PCB_UNLOCK(unp); 488 return (EINVAL); 489 } 490 if (unp->unp_flags & UNP_BINDING) { 491 UNP_PCB_UNLOCK(unp); 492 return (EALREADY); 493 } 494 unp->unp_flags |= UNP_BINDING; 495 UNP_PCB_UNLOCK(unp); 496 497 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 498 bcopy(soun->sun_path, buf, namelen); 499 buf[namelen] = 0; 500 501restart: 502 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, 503 UIO_SYSSPACE, buf, fd, CAP_BINDAT, td); 504/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 505 error = namei(&nd); 506 if (error) 507 goto error; 508 vp = nd.ni_vp; 509 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 510 NDFREE(&nd, NDF_ONLY_PNBUF); 511 if (nd.ni_dvp == vp) 512 vrele(nd.ni_dvp); 513 else 514 vput(nd.ni_dvp); 515 if (vp != NULL) { 516 vrele(vp); 517 error = EADDRINUSE; 518 goto error; 519 } 520 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 521 if (error) 522 goto error; 523 goto restart; 524 } 525 VATTR_NULL(&vattr); 526 vattr.va_type = VSOCK; 527 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 528#ifdef MAC 529 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 530 &vattr); 531#endif 532 if (error == 0) 533 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 534 NDFREE(&nd, NDF_ONLY_PNBUF); 535 vput(nd.ni_dvp); 536 if (error) { 537 vn_finished_write(mp); 538 goto error; 539 } 540 vp = nd.ni_vp; 541 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 542 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 543 544 UNP_LINK_WLOCK(); 545 UNP_PCB_LOCK(unp); 546 VOP_UNP_BIND(vp, unp->unp_socket); 547 unp->unp_vnode = vp; 548 unp->unp_addr = soun; 549 unp->unp_flags &= ~UNP_BINDING; 550 UNP_PCB_UNLOCK(unp); 551 UNP_LINK_WUNLOCK(); 552 VOP_UNLOCK(vp, 0); 553 vn_finished_write(mp); 554 free(buf, M_TEMP); 555 return (0); 556 557error: 558 UNP_PCB_LOCK(unp); 559 unp->unp_flags &= ~UNP_BINDING; 560 UNP_PCB_UNLOCK(unp); 561 free(buf, M_TEMP); 562 return (error); 563} 564 565static int 566uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 567{ 568 569 return (uipc_bindat(AT_FDCWD, so, nam, td)); 570} 571 572static int 573uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 574{ 575 int error; 576 577 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 578 UNP_LINK_WLOCK(); 579 error = unp_connect(so, nam, td); 580 UNP_LINK_WUNLOCK(); 581 return (error); 582} 583 584static int 585uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 586 struct thread *td) 587{ 588 int error; 589 590 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 591 UNP_LINK_WLOCK(); 592 error = unp_connectat(fd, so, nam, td); 593 UNP_LINK_WUNLOCK(); 594 return (error); 595} 596 597static void 598uipc_close(struct socket *so) 599{ 600 struct unpcb *unp, *unp2; 601 602 unp = sotounpcb(so); 603 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 604 605 UNP_LINK_WLOCK(); 606 UNP_PCB_LOCK(unp); 607 unp2 = unp->unp_conn; 608 if (unp2 != NULL) { 609 UNP_PCB_LOCK(unp2); 610 unp_disconnect(unp, unp2); 611 UNP_PCB_UNLOCK(unp2); 612 } 613 UNP_PCB_UNLOCK(unp); 614 UNP_LINK_WUNLOCK(); 615} 616 617static int 618uipc_connect2(struct socket *so1, struct socket *so2) 619{ 620 struct unpcb *unp, *unp2; 621 int error; 622 623 UNP_LINK_WLOCK(); 624 unp = so1->so_pcb; 625 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 626 UNP_PCB_LOCK(unp); 627 unp2 = so2->so_pcb; 628 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 629 UNP_PCB_LOCK(unp2); 630 error = unp_connect2(so1, so2, PRU_CONNECT2); 631 UNP_PCB_UNLOCK(unp2); 632 UNP_PCB_UNLOCK(unp); 633 UNP_LINK_WUNLOCK(); 634 return (error); 635} 636 637static void 638uipc_detach(struct socket *so) 639{ 640 struct unpcb *unp, *unp2; 641 struct sockaddr_un *saved_unp_addr; 642 struct vnode *vp; 643 int freeunp, local_unp_rights; 644 645 unp = sotounpcb(so); 646 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 647 648 UNP_LINK_WLOCK(); 649 UNP_LIST_LOCK(); 650 UNP_PCB_LOCK(unp); 651 LIST_REMOVE(unp, unp_link); 652 unp->unp_gencnt = ++unp_gencnt; 653 --unp_count; 654 UNP_LIST_UNLOCK(); 655 656 /* 657 * XXXRW: Should assert vp->v_socket == so. 658 */ 659 if ((vp = unp->unp_vnode) != NULL) { 660 VOP_UNP_DETACH(vp); 661 unp->unp_vnode = NULL; 662 } 663 unp2 = unp->unp_conn; 664 if (unp2 != NULL) { 665 UNP_PCB_LOCK(unp2); 666 unp_disconnect(unp, unp2); 667 UNP_PCB_UNLOCK(unp2); 668 } 669 670 /* 671 * We hold the linkage lock exclusively, so it's OK to acquire 672 * multiple pcb locks at a time. 673 */ 674 while (!LIST_EMPTY(&unp->unp_refs)) { 675 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 676 677 UNP_PCB_LOCK(ref); 678 unp_drop(ref, ECONNRESET); 679 UNP_PCB_UNLOCK(ref); 680 } 681 local_unp_rights = unp_rights; 682 UNP_LINK_WUNLOCK(); 683 unp->unp_socket->so_pcb = NULL; 684 saved_unp_addr = unp->unp_addr; 685 unp->unp_addr = NULL; 686 unp->unp_refcount--; 687 freeunp = (unp->unp_refcount == 0); 688 if (saved_unp_addr != NULL) 689 free(saved_unp_addr, M_SONAME); 690 if (freeunp) { 691 UNP_PCB_LOCK_DESTROY(unp); 692 uma_zfree(unp_zone, unp); 693 } else 694 UNP_PCB_UNLOCK(unp); 695 if (vp) 696 vrele(vp); 697 if (local_unp_rights) 698 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 699} 700 701static int 702uipc_disconnect(struct socket *so) 703{ 704 struct unpcb *unp, *unp2; 705 706 unp = sotounpcb(so); 707 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 708 709 UNP_LINK_WLOCK(); 710 UNP_PCB_LOCK(unp); 711 unp2 = unp->unp_conn; 712 if (unp2 != NULL) { 713 UNP_PCB_LOCK(unp2); 714 unp_disconnect(unp, unp2); 715 UNP_PCB_UNLOCK(unp2); 716 } 717 UNP_PCB_UNLOCK(unp); 718 UNP_LINK_WUNLOCK(); 719 return (0); 720} 721 722static int 723uipc_listen(struct socket *so, int backlog, struct thread *td) 724{ 725 struct unpcb *unp; 726 int error; 727 728 unp = sotounpcb(so); 729 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 730 731 UNP_PCB_LOCK(unp); 732 if (unp->unp_vnode == NULL) { 733 UNP_PCB_UNLOCK(unp); 734 return (EINVAL); 735 } 736 737 SOCK_LOCK(so); 738 error = solisten_proto_check(so); 739 if (error == 0) { 740 cru2x(td->td_ucred, &unp->unp_peercred); 741 unp->unp_flags |= UNP_HAVEPCCACHED; 742 solisten_proto(so, backlog); 743 } 744 SOCK_UNLOCK(so); 745 UNP_PCB_UNLOCK(unp); 746 return (error); 747} 748 749static int 750uipc_peeraddr(struct socket *so, struct sockaddr **nam) 751{ 752 struct unpcb *unp, *unp2; 753 const struct sockaddr *sa; 754 755 unp = sotounpcb(so); 756 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 757 758 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 759 UNP_LINK_RLOCK(); 760 /* 761 * XXX: It seems that this test always fails even when connection is 762 * established. So, this else clause is added as workaround to 763 * return PF_LOCAL sockaddr. 764 */ 765 unp2 = unp->unp_conn; 766 if (unp2 != NULL) { 767 UNP_PCB_LOCK(unp2); 768 if (unp2->unp_addr != NULL) 769 sa = (struct sockaddr *) unp2->unp_addr; 770 else 771 sa = &sun_noname; 772 bcopy(sa, *nam, sa->sa_len); 773 UNP_PCB_UNLOCK(unp2); 774 } else { 775 sa = &sun_noname; 776 bcopy(sa, *nam, sa->sa_len); 777 } 778 UNP_LINK_RUNLOCK(); 779 return (0); 780} 781 782static int 783uipc_rcvd(struct socket *so, int flags) 784{ 785 struct unpcb *unp, *unp2; 786 struct socket *so2; 787 u_int mbcnt, sbcc; 788 u_long newhiwat; 789 790 unp = sotounpcb(so); 791 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 792 793 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 794 panic("uipc_rcvd socktype %d", so->so_type); 795 796 /* 797 * Adjust backpressure on sender and wakeup any waiting to write. 798 * 799 * The unp lock is acquired to maintain the validity of the unp_conn 800 * pointer; no lock on unp2 is required as unp2->unp_socket will be 801 * static as long as we don't permit unp2 to disconnect from unp, 802 * which is prevented by the lock on unp. We cache values from 803 * so_rcv to avoid holding the so_rcv lock over the entire 804 * transaction on the remote so_snd. 805 */ 806 SOCKBUF_LOCK(&so->so_rcv); 807 mbcnt = so->so_rcv.sb_mbcnt; 808 sbcc = so->so_rcv.sb_cc; 809 SOCKBUF_UNLOCK(&so->so_rcv); 810 UNP_PCB_LOCK(unp); 811 unp2 = unp->unp_conn; 812 if (unp2 == NULL) { 813 UNP_PCB_UNLOCK(unp); 814 return (0); 815 } 816 so2 = unp2->unp_socket; 817 SOCKBUF_LOCK(&so2->so_snd); 818 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 819 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 820 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 821 newhiwat, RLIM_INFINITY); 822 sowwakeup_locked(so2); 823 unp->unp_mbcnt = mbcnt; 824 unp->unp_cc = sbcc; 825 UNP_PCB_UNLOCK(unp); 826 return (0); 827} 828 829static int 830uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 831 struct mbuf *control, struct thread *td) 832{ 833 struct unpcb *unp, *unp2; 834 struct socket *so2; 835 u_int mbcnt_delta, sbcc; 836 u_int newhiwat; 837 int error = 0; 838 839 unp = sotounpcb(so); 840 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 841 842 if (flags & PRUS_OOB) { 843 error = EOPNOTSUPP; 844 goto release; 845 } 846 if (control != NULL && (error = unp_internalize(&control, td))) 847 goto release; 848 if ((nam != NULL) || (flags & PRUS_EOF)) 849 UNP_LINK_WLOCK(); 850 else 851 UNP_LINK_RLOCK(); 852 switch (so->so_type) { 853 case SOCK_DGRAM: 854 { 855 const struct sockaddr *from; 856 857 unp2 = unp->unp_conn; 858 if (nam != NULL) { 859 UNP_LINK_WLOCK_ASSERT(); 860 if (unp2 != NULL) { 861 error = EISCONN; 862 break; 863 } 864 error = unp_connect(so, nam, td); 865 if (error) 866 break; 867 unp2 = unp->unp_conn; 868 } 869 870 /* 871 * Because connect() and send() are non-atomic in a sendto() 872 * with a target address, it's possible that the socket will 873 * have disconnected before the send() can run. In that case 874 * return the slightly counter-intuitive but otherwise 875 * correct error that the socket is not connected. 876 */ 877 if (unp2 == NULL) { 878 error = ENOTCONN; 879 break; 880 } 881 /* Lockless read. */ 882 if (unp2->unp_flags & UNP_WANTCRED) 883 control = unp_addsockcred(td, control); 884 UNP_PCB_LOCK(unp); 885 if (unp->unp_addr != NULL) 886 from = (struct sockaddr *)unp->unp_addr; 887 else 888 from = &sun_noname; 889 so2 = unp2->unp_socket; 890 SOCKBUF_LOCK(&so2->so_rcv); 891 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 892 sorwakeup_locked(so2); 893 m = NULL; 894 control = NULL; 895 } else { 896 SOCKBUF_UNLOCK(&so2->so_rcv); 897 error = ENOBUFS; 898 } 899 if (nam != NULL) { 900 UNP_LINK_WLOCK_ASSERT(); 901 UNP_PCB_LOCK(unp2); 902 unp_disconnect(unp, unp2); 903 UNP_PCB_UNLOCK(unp2); 904 } 905 UNP_PCB_UNLOCK(unp); 906 break; 907 } 908 909 case SOCK_SEQPACKET: 910 case SOCK_STREAM: 911 if ((so->so_state & SS_ISCONNECTED) == 0) { 912 if (nam != NULL) { 913 UNP_LINK_WLOCK_ASSERT(); 914 error = unp_connect(so, nam, td); 915 if (error) 916 break; /* XXX */ 917 } else { 918 error = ENOTCONN; 919 break; 920 } 921 } 922 923 /* Lockless read. */ 924 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 925 error = EPIPE; 926 break; 927 } 928 929 /* 930 * Because connect() and send() are non-atomic in a sendto() 931 * with a target address, it's possible that the socket will 932 * have disconnected before the send() can run. In that case 933 * return the slightly counter-intuitive but otherwise 934 * correct error that the socket is not connected. 935 * 936 * Locking here must be done carefully: the linkage lock 937 * prevents interconnections between unpcbs from changing, so 938 * we can traverse from unp to unp2 without acquiring unp's 939 * lock. Socket buffer locks follow unpcb locks, so we can 940 * acquire both remote and lock socket buffer locks. 941 */ 942 unp2 = unp->unp_conn; 943 if (unp2 == NULL) { 944 error = ENOTCONN; 945 break; 946 } 947 so2 = unp2->unp_socket; 948 UNP_PCB_LOCK(unp2); 949 SOCKBUF_LOCK(&so2->so_rcv); 950 if (unp2->unp_flags & UNP_WANTCRED) { 951 /* 952 * Credentials are passed only once on SOCK_STREAM 953 * and SOCK_SEQPACKET. 954 */ 955 unp2->unp_flags &= ~UNP_WANTCRED; 956 control = unp_addsockcred(td, control); 957 } 958 /* 959 * Send to paired receive port, and then reduce send buffer 960 * hiwater marks to maintain backpressure. Wake up readers. 961 */ 962 switch (so->so_type) { 963 case SOCK_STREAM: 964 if (control != NULL) { 965 if (sbappendcontrol_locked(&so2->so_rcv, m, 966 control)) 967 control = NULL; 968 } else 969 sbappend_locked(&so2->so_rcv, m); 970 break; 971 972 case SOCK_SEQPACKET: { 973 const struct sockaddr *from; 974 975 from = &sun_noname; 976 if (sbappendaddr_locked(&so2->so_rcv, from, m, 977 control)) 978 control = NULL; 979 break; 980 } 981 } 982 983 /* 984 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 985 * datagram size and back-pressure for SOCK_SEQPACKET, which 986 * can lead to undesired return of EMSGSIZE on send instead 987 * of more desirable blocking. 988 */ 989 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 990 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 991 sbcc = so2->so_rcv.sb_cc; 992 sorwakeup_locked(so2); 993 994 SOCKBUF_LOCK(&so->so_snd); 995 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 996 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 997 else 998 newhiwat = 0; 999 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 1000 newhiwat, RLIM_INFINITY); 1001 so->so_snd.sb_mbmax -= mbcnt_delta; 1002 SOCKBUF_UNLOCK(&so->so_snd); 1003 unp2->unp_cc = sbcc; 1004 UNP_PCB_UNLOCK(unp2); 1005 m = NULL; 1006 break; 1007 1008 default: 1009 panic("uipc_send unknown socktype"); 1010 } 1011 1012 /* 1013 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1014 */ 1015 if (flags & PRUS_EOF) { 1016 UNP_PCB_LOCK(unp); 1017 socantsendmore(so); 1018 unp_shutdown(unp); 1019 UNP_PCB_UNLOCK(unp); 1020 } 1021 1022 if ((nam != NULL) || (flags & PRUS_EOF)) 1023 UNP_LINK_WUNLOCK(); 1024 else 1025 UNP_LINK_RUNLOCK(); 1026 1027 if (control != NULL && error != 0) 1028 unp_dispose(control); 1029 1030release: 1031 if (control != NULL) 1032 m_freem(control); 1033 if (m != NULL) 1034 m_freem(m); 1035 return (error); 1036} 1037 1038static int 1039uipc_sense(struct socket *so, struct stat *sb) 1040{ 1041 struct unpcb *unp, *unp2; 1042 struct socket *so2; 1043 1044 unp = sotounpcb(so); 1045 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1046 1047 sb->st_blksize = so->so_snd.sb_hiwat; 1048 UNP_LINK_RLOCK(); 1049 UNP_PCB_LOCK(unp); 1050 unp2 = unp->unp_conn; 1051 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1052 unp2 != NULL) { 1053 so2 = unp2->unp_socket; 1054 sb->st_blksize += so2->so_rcv.sb_cc; 1055 } 1056 sb->st_dev = NODEV; 1057 if (unp->unp_ino == 0) 1058 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1059 sb->st_ino = unp->unp_ino; 1060 UNP_PCB_UNLOCK(unp); 1061 UNP_LINK_RUNLOCK(); 1062 return (0); 1063} 1064 1065static int 1066uipc_shutdown(struct socket *so) 1067{ 1068 struct unpcb *unp; 1069 1070 unp = sotounpcb(so); 1071 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1072 1073 UNP_LINK_WLOCK(); 1074 UNP_PCB_LOCK(unp); 1075 socantsendmore(so); 1076 unp_shutdown(unp); 1077 UNP_PCB_UNLOCK(unp); 1078 UNP_LINK_WUNLOCK(); 1079 return (0); 1080} 1081 1082static int 1083uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1084{ 1085 struct unpcb *unp; 1086 const struct sockaddr *sa; 1087 1088 unp = sotounpcb(so); 1089 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1090 1091 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1092 UNP_PCB_LOCK(unp); 1093 if (unp->unp_addr != NULL) 1094 sa = (struct sockaddr *) unp->unp_addr; 1095 else 1096 sa = &sun_noname; 1097 bcopy(sa, *nam, sa->sa_len); 1098 UNP_PCB_UNLOCK(unp); 1099 return (0); 1100} 1101 1102static struct pr_usrreqs uipc_usrreqs_dgram = { 1103 .pru_abort = uipc_abort, 1104 .pru_accept = uipc_accept, 1105 .pru_attach = uipc_attach, 1106 .pru_bind = uipc_bind, 1107 .pru_bindat = uipc_bindat, 1108 .pru_connect = uipc_connect, 1109 .pru_connectat = uipc_connectat, 1110 .pru_connect2 = uipc_connect2, 1111 .pru_detach = uipc_detach, 1112 .pru_disconnect = uipc_disconnect, 1113 .pru_listen = uipc_listen, 1114 .pru_peeraddr = uipc_peeraddr, 1115 .pru_rcvd = uipc_rcvd, 1116 .pru_send = uipc_send, 1117 .pru_sense = uipc_sense, 1118 .pru_shutdown = uipc_shutdown, 1119 .pru_sockaddr = uipc_sockaddr, 1120 .pru_soreceive = soreceive_dgram, 1121 .pru_close = uipc_close, 1122}; 1123 1124static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1125 .pru_abort = uipc_abort, 1126 .pru_accept = uipc_accept, 1127 .pru_attach = uipc_attach, 1128 .pru_bind = uipc_bind, 1129 .pru_bindat = uipc_bindat, 1130 .pru_connect = uipc_connect, 1131 .pru_connectat = uipc_connectat, 1132 .pru_connect2 = uipc_connect2, 1133 .pru_detach = uipc_detach, 1134 .pru_disconnect = uipc_disconnect, 1135 .pru_listen = uipc_listen, 1136 .pru_peeraddr = uipc_peeraddr, 1137 .pru_rcvd = uipc_rcvd, 1138 .pru_send = uipc_send, 1139 .pru_sense = uipc_sense, 1140 .pru_shutdown = uipc_shutdown, 1141 .pru_sockaddr = uipc_sockaddr, 1142 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1143 .pru_close = uipc_close, 1144}; 1145 1146static struct pr_usrreqs uipc_usrreqs_stream = { 1147 .pru_abort = uipc_abort, 1148 .pru_accept = uipc_accept, 1149 .pru_attach = uipc_attach, 1150 .pru_bind = uipc_bind, 1151 .pru_bindat = uipc_bindat, 1152 .pru_connect = uipc_connect, 1153 .pru_connectat = uipc_connectat, 1154 .pru_connect2 = uipc_connect2, 1155 .pru_detach = uipc_detach, 1156 .pru_disconnect = uipc_disconnect, 1157 .pru_listen = uipc_listen, 1158 .pru_peeraddr = uipc_peeraddr, 1159 .pru_rcvd = uipc_rcvd, 1160 .pru_send = uipc_send, 1161 .pru_sense = uipc_sense, 1162 .pru_shutdown = uipc_shutdown, 1163 .pru_sockaddr = uipc_sockaddr, 1164 .pru_soreceive = soreceive_generic, 1165 .pru_close = uipc_close, 1166}; 1167 1168static int 1169uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1170{ 1171 struct unpcb *unp; 1172 struct xucred xu; 1173 int error, optval; 1174 1175 if (sopt->sopt_level != 0) 1176 return (EINVAL); 1177 1178 unp = sotounpcb(so); 1179 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1180 error = 0; 1181 switch (sopt->sopt_dir) { 1182 case SOPT_GET: 1183 switch (sopt->sopt_name) { 1184 case LOCAL_PEERCRED: 1185 UNP_PCB_LOCK(unp); 1186 if (unp->unp_flags & UNP_HAVEPC) 1187 xu = unp->unp_peercred; 1188 else { 1189 if (so->so_type == SOCK_STREAM) 1190 error = ENOTCONN; 1191 else 1192 error = EINVAL; 1193 } 1194 UNP_PCB_UNLOCK(unp); 1195 if (error == 0) 1196 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1197 break; 1198 1199 case LOCAL_CREDS: 1200 /* Unlocked read. */ 1201 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1202 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1203 break; 1204 1205 case LOCAL_CONNWAIT: 1206 /* Unlocked read. */ 1207 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1208 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1209 break; 1210 1211 default: 1212 error = EOPNOTSUPP; 1213 break; 1214 } 1215 break; 1216 1217 case SOPT_SET: 1218 switch (sopt->sopt_name) { 1219 case LOCAL_CREDS: 1220 case LOCAL_CONNWAIT: 1221 error = sooptcopyin(sopt, &optval, sizeof(optval), 1222 sizeof(optval)); 1223 if (error) 1224 break; 1225 1226#define OPTSET(bit) do { \ 1227 UNP_PCB_LOCK(unp); \ 1228 if (optval) \ 1229 unp->unp_flags |= bit; \ 1230 else \ 1231 unp->unp_flags &= ~bit; \ 1232 UNP_PCB_UNLOCK(unp); \ 1233} while (0) 1234 1235 switch (sopt->sopt_name) { 1236 case LOCAL_CREDS: 1237 OPTSET(UNP_WANTCRED); 1238 break; 1239 1240 case LOCAL_CONNWAIT: 1241 OPTSET(UNP_CONNWAIT); 1242 break; 1243 1244 default: 1245 break; 1246 } 1247 break; 1248#undef OPTSET 1249 default: 1250 error = ENOPROTOOPT; 1251 break; 1252 } 1253 break; 1254 1255 default: 1256 error = EOPNOTSUPP; 1257 break; 1258 } 1259 return (error); 1260} 1261 1262static int 1263unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1264{ 1265 1266 return (unp_connectat(AT_FDCWD, so, nam, td)); 1267} 1268 1269static int 1270unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1271 struct thread *td) 1272{ 1273 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1274 struct vnode *vp; 1275 struct socket *so2, *so3; 1276 struct unpcb *unp, *unp2, *unp3; 1277 int error, len; 1278 struct nameidata nd; 1279 char buf[SOCK_MAXADDRLEN]; 1280 struct sockaddr *sa; 1281 1282 UNP_LINK_WLOCK_ASSERT(); 1283 1284 unp = sotounpcb(so); 1285 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1286 1287 if (nam->sa_len > sizeof(struct sockaddr_un)) 1288 return (EINVAL); 1289 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1290 if (len <= 0) 1291 return (EINVAL); 1292 bcopy(soun->sun_path, buf, len); 1293 buf[len] = 0; 1294 1295 UNP_PCB_LOCK(unp); 1296 if (unp->unp_flags & UNP_CONNECTING) { 1297 UNP_PCB_UNLOCK(unp); 1298 return (EALREADY); 1299 } 1300 UNP_LINK_WUNLOCK(); 1301 unp->unp_flags |= UNP_CONNECTING; 1302 UNP_PCB_UNLOCK(unp); 1303 1304 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1305 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1306 UIO_SYSSPACE, buf, fd, CAP_CONNECTAT, td); 1307 error = namei(&nd); 1308 if (error) 1309 vp = NULL; 1310 else 1311 vp = nd.ni_vp; 1312 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1313 NDFREE(&nd, NDF_ONLY_PNBUF); 1314 if (error) 1315 goto bad; 1316 1317 if (vp->v_type != VSOCK) { 1318 error = ENOTSOCK; 1319 goto bad; 1320 } 1321#ifdef MAC 1322 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1323 if (error) 1324 goto bad; 1325#endif 1326 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1327 if (error) 1328 goto bad; 1329 1330 unp = sotounpcb(so); 1331 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1332 1333 /* 1334 * Lock linkage lock for two reasons: make sure v_socket is stable, 1335 * and to protect simultaneous locking of multiple pcbs. 1336 */ 1337 UNP_LINK_WLOCK(); 1338 VOP_UNP_CONNECT(vp, &so2); 1339 if (so2 == NULL) { 1340 error = ECONNREFUSED; 1341 goto bad2; 1342 } 1343 if (so->so_type != so2->so_type) { 1344 error = EPROTOTYPE; 1345 goto bad2; 1346 } 1347 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1348 if (so2->so_options & SO_ACCEPTCONN) { 1349 CURVNET_SET(so2->so_vnet); 1350 so3 = sonewconn(so2, 0); 1351 CURVNET_RESTORE(); 1352 } else 1353 so3 = NULL; 1354 if (so3 == NULL) { 1355 error = ECONNREFUSED; 1356 goto bad2; 1357 } 1358 unp = sotounpcb(so); 1359 unp2 = sotounpcb(so2); 1360 unp3 = sotounpcb(so3); 1361 UNP_PCB_LOCK(unp); 1362 UNP_PCB_LOCK(unp2); 1363 UNP_PCB_LOCK(unp3); 1364 if (unp2->unp_addr != NULL) { 1365 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1366 unp3->unp_addr = (struct sockaddr_un *) sa; 1367 sa = NULL; 1368 } 1369 1370 /* 1371 * The connecter's (client's) credentials are copied from its 1372 * process structure at the time of connect() (which is now). 1373 */ 1374 cru2x(td->td_ucred, &unp3->unp_peercred); 1375 unp3->unp_flags |= UNP_HAVEPC; 1376 1377 /* 1378 * The receiver's (server's) credentials are copied from the 1379 * unp_peercred member of socket on which the former called 1380 * listen(); uipc_listen() cached that process's credentials 1381 * at that time so we can use them now. 1382 */ 1383 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1384 ("unp_connect: listener without cached peercred")); 1385 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1386 sizeof(unp->unp_peercred)); 1387 unp->unp_flags |= UNP_HAVEPC; 1388 if (unp2->unp_flags & UNP_WANTCRED) 1389 unp3->unp_flags |= UNP_WANTCRED; 1390 UNP_PCB_UNLOCK(unp3); 1391 UNP_PCB_UNLOCK(unp2); 1392 UNP_PCB_UNLOCK(unp); 1393#ifdef MAC 1394 mac_socketpeer_set_from_socket(so, so3); 1395 mac_socketpeer_set_from_socket(so3, so); 1396#endif 1397 1398 so2 = so3; 1399 } 1400 unp = sotounpcb(so); 1401 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1402 unp2 = sotounpcb(so2); 1403 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1404 UNP_PCB_LOCK(unp); 1405 UNP_PCB_LOCK(unp2); 1406 error = unp_connect2(so, so2, PRU_CONNECT); 1407 UNP_PCB_UNLOCK(unp2); 1408 UNP_PCB_UNLOCK(unp); 1409bad2: 1410 UNP_LINK_WUNLOCK(); 1411bad: 1412 if (vp != NULL) 1413 vput(vp); 1414 free(sa, M_SONAME); 1415 UNP_LINK_WLOCK(); 1416 UNP_PCB_LOCK(unp); 1417 unp->unp_flags &= ~UNP_CONNECTING; 1418 UNP_PCB_UNLOCK(unp); 1419 return (error); 1420} 1421 1422static int 1423unp_connect2(struct socket *so, struct socket *so2, int req) 1424{ 1425 struct unpcb *unp; 1426 struct unpcb *unp2; 1427 1428 unp = sotounpcb(so); 1429 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1430 unp2 = sotounpcb(so2); 1431 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1432 1433 UNP_LINK_WLOCK_ASSERT(); 1434 UNP_PCB_LOCK_ASSERT(unp); 1435 UNP_PCB_LOCK_ASSERT(unp2); 1436 1437 if (so2->so_type != so->so_type) 1438 return (EPROTOTYPE); 1439 unp->unp_conn = unp2; 1440 1441 switch (so->so_type) { 1442 case SOCK_DGRAM: 1443 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1444 soisconnected(so); 1445 break; 1446 1447 case SOCK_STREAM: 1448 case SOCK_SEQPACKET: 1449 unp2->unp_conn = unp; 1450 if (req == PRU_CONNECT && 1451 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1452 soisconnecting(so); 1453 else 1454 soisconnected(so); 1455 soisconnected(so2); 1456 break; 1457 1458 default: 1459 panic("unp_connect2"); 1460 } 1461 return (0); 1462} 1463 1464static void 1465unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1466{ 1467 struct socket *so; 1468 1469 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1470 1471 UNP_LINK_WLOCK_ASSERT(); 1472 UNP_PCB_LOCK_ASSERT(unp); 1473 UNP_PCB_LOCK_ASSERT(unp2); 1474 1475 unp->unp_conn = NULL; 1476 switch (unp->unp_socket->so_type) { 1477 case SOCK_DGRAM: 1478 LIST_REMOVE(unp, unp_reflink); 1479 so = unp->unp_socket; 1480 SOCK_LOCK(so); 1481 so->so_state &= ~SS_ISCONNECTED; 1482 SOCK_UNLOCK(so); 1483 break; 1484 1485 case SOCK_STREAM: 1486 case SOCK_SEQPACKET: 1487 soisdisconnected(unp->unp_socket); 1488 unp2->unp_conn = NULL; 1489 soisdisconnected(unp2->unp_socket); 1490 break; 1491 } 1492} 1493 1494/* 1495 * unp_pcblist() walks the global list of struct unpcb's to generate a 1496 * pointer list, bumping the refcount on each unpcb. It then copies them out 1497 * sequentially, validating the generation number on each to see if it has 1498 * been detached. All of this is necessary because copyout() may sleep on 1499 * disk I/O. 1500 */ 1501static int 1502unp_pcblist(SYSCTL_HANDLER_ARGS) 1503{ 1504 int error, i, n; 1505 int freeunp; 1506 struct unpcb *unp, **unp_list; 1507 unp_gen_t gencnt; 1508 struct xunpgen *xug; 1509 struct unp_head *head; 1510 struct xunpcb *xu; 1511 1512 switch ((intptr_t)arg1) { 1513 case SOCK_STREAM: 1514 head = &unp_shead; 1515 break; 1516 1517 case SOCK_DGRAM: 1518 head = &unp_dhead; 1519 break; 1520 1521 case SOCK_SEQPACKET: 1522 head = &unp_sphead; 1523 break; 1524 1525 default: 1526 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1527 } 1528 1529 /* 1530 * The process of preparing the PCB list is too time-consuming and 1531 * resource-intensive to repeat twice on every request. 1532 */ 1533 if (req->oldptr == NULL) { 1534 n = unp_count; 1535 req->oldidx = 2 * (sizeof *xug) 1536 + (n + n/8) * sizeof(struct xunpcb); 1537 return (0); 1538 } 1539 1540 if (req->newptr != NULL) 1541 return (EPERM); 1542 1543 /* 1544 * OK, now we're committed to doing something. 1545 */ 1546 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1547 UNP_LIST_LOCK(); 1548 gencnt = unp_gencnt; 1549 n = unp_count; 1550 UNP_LIST_UNLOCK(); 1551 1552 xug->xug_len = sizeof *xug; 1553 xug->xug_count = n; 1554 xug->xug_gen = gencnt; 1555 xug->xug_sogen = so_gencnt; 1556 error = SYSCTL_OUT(req, xug, sizeof *xug); 1557 if (error) { 1558 free(xug, M_TEMP); 1559 return (error); 1560 } 1561 1562 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1563 1564 UNP_LIST_LOCK(); 1565 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1566 unp = LIST_NEXT(unp, unp_link)) { 1567 UNP_PCB_LOCK(unp); 1568 if (unp->unp_gencnt <= gencnt) { 1569 if (cr_cansee(req->td->td_ucred, 1570 unp->unp_socket->so_cred)) { 1571 UNP_PCB_UNLOCK(unp); 1572 continue; 1573 } 1574 unp_list[i++] = unp; 1575 unp->unp_refcount++; 1576 } 1577 UNP_PCB_UNLOCK(unp); 1578 } 1579 UNP_LIST_UNLOCK(); 1580 n = i; /* In case we lost some during malloc. */ 1581 1582 error = 0; 1583 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1584 for (i = 0; i < n; i++) { 1585 unp = unp_list[i]; 1586 UNP_PCB_LOCK(unp); 1587 unp->unp_refcount--; 1588 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1589 xu->xu_len = sizeof *xu; 1590 xu->xu_unpp = unp; 1591 /* 1592 * XXX - need more locking here to protect against 1593 * connect/disconnect races for SMP. 1594 */ 1595 if (unp->unp_addr != NULL) 1596 bcopy(unp->unp_addr, &xu->xu_addr, 1597 unp->unp_addr->sun_len); 1598 if (unp->unp_conn != NULL && 1599 unp->unp_conn->unp_addr != NULL) 1600 bcopy(unp->unp_conn->unp_addr, 1601 &xu->xu_caddr, 1602 unp->unp_conn->unp_addr->sun_len); 1603 bcopy(unp, &xu->xu_unp, sizeof *unp); 1604 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1605 UNP_PCB_UNLOCK(unp); 1606 error = SYSCTL_OUT(req, xu, sizeof *xu); 1607 } else { 1608 freeunp = (unp->unp_refcount == 0); 1609 UNP_PCB_UNLOCK(unp); 1610 if (freeunp) { 1611 UNP_PCB_LOCK_DESTROY(unp); 1612 uma_zfree(unp_zone, unp); 1613 } 1614 } 1615 } 1616 free(xu, M_TEMP); 1617 if (!error) { 1618 /* 1619 * Give the user an updated idea of our state. If the 1620 * generation differs from what we told her before, she knows 1621 * that something happened while we were processing this 1622 * request, and it might be necessary to retry. 1623 */ 1624 xug->xug_gen = unp_gencnt; 1625 xug->xug_sogen = so_gencnt; 1626 xug->xug_count = unp_count; 1627 error = SYSCTL_OUT(req, xug, sizeof *xug); 1628 } 1629 free(unp_list, M_TEMP); 1630 free(xug, M_TEMP); 1631 return (error); 1632} 1633 1634SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1635 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1636 "List of active local datagram sockets"); 1637SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1638 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1639 "List of active local stream sockets"); 1640SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1641 CTLTYPE_OPAQUE | CTLFLAG_RD, 1642 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1643 "List of active local seqpacket sockets"); 1644 1645static void 1646unp_shutdown(struct unpcb *unp) 1647{ 1648 struct unpcb *unp2; 1649 struct socket *so; 1650 1651 UNP_LINK_WLOCK_ASSERT(); 1652 UNP_PCB_LOCK_ASSERT(unp); 1653 1654 unp2 = unp->unp_conn; 1655 if ((unp->unp_socket->so_type == SOCK_STREAM || 1656 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1657 so = unp2->unp_socket; 1658 if (so != NULL) 1659 socantrcvmore(so); 1660 } 1661} 1662 1663static void 1664unp_drop(struct unpcb *unp, int errno) 1665{ 1666 struct socket *so = unp->unp_socket; 1667 struct unpcb *unp2; 1668 1669 UNP_LINK_WLOCK_ASSERT(); 1670 UNP_PCB_LOCK_ASSERT(unp); 1671 1672 so->so_error = errno; 1673 unp2 = unp->unp_conn; 1674 if (unp2 == NULL) 1675 return; 1676 UNP_PCB_LOCK(unp2); 1677 unp_disconnect(unp, unp2); 1678 UNP_PCB_UNLOCK(unp2); 1679} 1680 1681static void 1682unp_freerights(struct filedescent *fde, int fdcount) 1683{ 1684 struct file *fp; 1685 int i; 1686 1687 for (i = 0; i < fdcount; i++, fde++) { 1688 fp = fde->fde_file; 1689 bzero(fde, sizeof(*fde)); 1690 unp_discard(fp); 1691 } 1692} 1693 1694static int 1695unp_externalize(struct mbuf *control, struct mbuf **controlp) 1696{ 1697 struct thread *td = curthread; /* XXX */ 1698 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1699 int i; 1700 int *fdp; 1701 struct filedesc *fdesc = td->td_proc->p_fd; 1702 struct filedescent *fde, *fdep; 1703 void *data; 1704 socklen_t clen = control->m_len, datalen; 1705 int error, newfds; 1706 int f; 1707 u_int newlen; 1708 1709 UNP_LINK_UNLOCK_ASSERT(); 1710 1711 error = 0; 1712 if (controlp != NULL) /* controlp == NULL => free control messages */ 1713 *controlp = NULL; 1714 while (cm != NULL) { 1715 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1716 error = EINVAL; 1717 break; 1718 } 1719 data = CMSG_DATA(cm); 1720 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1721 if (cm->cmsg_level == SOL_SOCKET 1722 && cm->cmsg_type == SCM_RIGHTS) { 1723 newfds = datalen / sizeof(*fdep); 1724 fdep = data; 1725 1726 /* If we're not outputting the descriptors free them. */ 1727 if (error || controlp == NULL) { 1728 unp_freerights(fdep, newfds); 1729 goto next; 1730 } 1731 FILEDESC_XLOCK(fdesc); 1732 /* if the new FD's will not fit free them. */ 1733 if (!fdavail(td, newfds)) { 1734 FILEDESC_XUNLOCK(fdesc); 1735 error = EMSGSIZE; 1736 unp_freerights(fdep, newfds); 1737 goto next; 1738 } 1739 1740 /* 1741 * Now change each pointer to an fd in the global 1742 * table to an integer that is the index to the local 1743 * fd table entry that we set up to point to the 1744 * global one we are transferring. 1745 */ 1746 newlen = newfds * sizeof(int); 1747 *controlp = sbcreatecontrol(NULL, newlen, 1748 SCM_RIGHTS, SOL_SOCKET); 1749 if (*controlp == NULL) { 1750 FILEDESC_XUNLOCK(fdesc); 1751 error = E2BIG; 1752 unp_freerights(fdep, newfds); 1753 goto next; 1754 } 1755 1756 fdp = (int *) 1757 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1758 for (i = 0; i < newfds; i++, fdep++, fdp++) { 1759 if (fdalloc(td, 0, &f)) 1760 panic("unp_externalize fdalloc failed"); 1761 fde = &fdesc->fd_ofiles[f]; 1762 fde->fde_file = fdep->fde_file; 1763 filecaps_copy(&fdep->fde_caps, &fde->fde_caps); 1764 unp_externalize_fp(fde->fde_file); 1765 *fdp = f; 1766 } 1767 FILEDESC_XUNLOCK(fdesc); 1768 } else { 1769 /* We can just copy anything else across. */ 1770 if (error || controlp == NULL) 1771 goto next; 1772 *controlp = sbcreatecontrol(NULL, datalen, 1773 cm->cmsg_type, cm->cmsg_level); 1774 if (*controlp == NULL) { 1775 error = ENOBUFS; 1776 goto next; 1777 } 1778 bcopy(data, 1779 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1780 datalen); 1781 } 1782 controlp = &(*controlp)->m_next; 1783 1784next: 1785 if (CMSG_SPACE(datalen) < clen) { 1786 clen -= CMSG_SPACE(datalen); 1787 cm = (struct cmsghdr *) 1788 ((caddr_t)cm + CMSG_SPACE(datalen)); 1789 } else { 1790 clen = 0; 1791 cm = NULL; 1792 } 1793 } 1794 1795 m_freem(control); 1796 return (error); 1797} 1798 1799static void 1800unp_zone_change(void *tag) 1801{ 1802 1803 uma_zone_set_max(unp_zone, maxsockets); 1804} 1805 1806static void 1807unp_init(void) 1808{ 1809 1810#ifdef VIMAGE 1811 if (!IS_DEFAULT_VNET(curvnet)) 1812 return; 1813#endif 1814 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1815 NULL, NULL, UMA_ALIGN_PTR, 0); 1816 if (unp_zone == NULL) 1817 panic("unp_init"); 1818 uma_zone_set_max(unp_zone, maxsockets); 1819 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 1820 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1821 NULL, EVENTHANDLER_PRI_ANY); 1822 LIST_INIT(&unp_dhead); 1823 LIST_INIT(&unp_shead); 1824 LIST_INIT(&unp_sphead); 1825 SLIST_INIT(&unp_defers); 1826 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 1827 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1828 UNP_LINK_LOCK_INIT(); 1829 UNP_LIST_LOCK_INIT(); 1830 UNP_DEFERRED_LOCK_INIT(); 1831} 1832 1833static int 1834unp_internalize(struct mbuf **controlp, struct thread *td) 1835{ 1836 struct mbuf *control = *controlp; 1837 struct proc *p = td->td_proc; 1838 struct filedesc *fdesc = p->p_fd; 1839 struct bintime *bt; 1840 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1841 struct cmsgcred *cmcred; 1842 struct filedescent *fde, *fdep; 1843 struct file *fp; 1844 struct timeval *tv; 1845 int i, fd, *fdp; 1846 void *data; 1847 socklen_t clen = control->m_len, datalen; 1848 int error, oldfds; 1849 u_int newlen; 1850 1851 UNP_LINK_UNLOCK_ASSERT(); 1852 1853 error = 0; 1854 *controlp = NULL; 1855 while (cm != NULL) { 1856 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1857 || cm->cmsg_len > clen) { 1858 error = EINVAL; 1859 goto out; 1860 } 1861 data = CMSG_DATA(cm); 1862 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1863 1864 switch (cm->cmsg_type) { 1865 /* 1866 * Fill in credential information. 1867 */ 1868 case SCM_CREDS: 1869 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1870 SCM_CREDS, SOL_SOCKET); 1871 if (*controlp == NULL) { 1872 error = ENOBUFS; 1873 goto out; 1874 } 1875 cmcred = (struct cmsgcred *) 1876 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1877 cmcred->cmcred_pid = p->p_pid; 1878 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1879 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1880 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1881 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1882 CMGROUP_MAX); 1883 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1884 cmcred->cmcred_groups[i] = 1885 td->td_ucred->cr_groups[i]; 1886 break; 1887 1888 case SCM_RIGHTS: 1889 oldfds = datalen / sizeof (int); 1890 /* 1891 * Check that all the FDs passed in refer to legal 1892 * files. If not, reject the entire operation. 1893 */ 1894 fdp = data; 1895 FILEDESC_SLOCK(fdesc); 1896 for (i = 0; i < oldfds; i++) { 1897 fd = *fdp++; 1898 if (fget_locked(fdesc, fd) == NULL) { 1899 FILEDESC_SUNLOCK(fdesc); 1900 error = EBADF; 1901 goto out; 1902 } 1903 fp = fdesc->fd_ofiles[fd].fde_file; 1904 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1905 FILEDESC_SUNLOCK(fdesc); 1906 error = EOPNOTSUPP; 1907 goto out; 1908 } 1909 1910 } 1911 1912 /* 1913 * Now replace the integer FDs with pointers to the 1914 * file structure and capability rights. 1915 */ 1916 newlen = oldfds * sizeof(*fdep); 1917 *controlp = sbcreatecontrol(NULL, newlen, 1918 SCM_RIGHTS, SOL_SOCKET); 1919 if (*controlp == NULL) { 1920 FILEDESC_SUNLOCK(fdesc); 1921 error = E2BIG; 1922 goto out; 1923 } 1924 fdp = data; 1925 fdep = (struct filedescent *) 1926 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1927 for (i = 0; i < oldfds; i++, fdep++, fdp++) { 1928 fde = &fdesc->fd_ofiles[*fdp]; 1929 fdep->fde_file = fde->fde_file; 1930 filecaps_copy(&fde->fde_caps, &fdep->fde_caps); 1931 unp_internalize_fp(fdep->fde_file); 1932 } 1933 FILEDESC_SUNLOCK(fdesc); 1934 break; 1935 1936 case SCM_TIMESTAMP: 1937 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1938 SCM_TIMESTAMP, SOL_SOCKET); 1939 if (*controlp == NULL) { 1940 error = ENOBUFS; 1941 goto out; 1942 } 1943 tv = (struct timeval *) 1944 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1945 microtime(tv); 1946 break; 1947 1948 case SCM_BINTIME: 1949 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 1950 SCM_BINTIME, SOL_SOCKET); 1951 if (*controlp == NULL) { 1952 error = ENOBUFS; 1953 goto out; 1954 } 1955 bt = (struct bintime *) 1956 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1957 bintime(bt); 1958 break; 1959 1960 default: 1961 error = EINVAL; 1962 goto out; 1963 } 1964 1965 controlp = &(*controlp)->m_next; 1966 if (CMSG_SPACE(datalen) < clen) { 1967 clen -= CMSG_SPACE(datalen); 1968 cm = (struct cmsghdr *) 1969 ((caddr_t)cm + CMSG_SPACE(datalen)); 1970 } else { 1971 clen = 0; 1972 cm = NULL; 1973 } 1974 } 1975 1976out: 1977 m_freem(control); 1978 return (error); 1979} 1980 1981static struct mbuf * 1982unp_addsockcred(struct thread *td, struct mbuf *control) 1983{ 1984 struct mbuf *m, *n, *n_prev; 1985 struct sockcred *sc; 1986 const struct cmsghdr *cm; 1987 int ngroups; 1988 int i; 1989 1990 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1991 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1992 if (m == NULL) 1993 return (control); 1994 1995 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1996 sc->sc_uid = td->td_ucred->cr_ruid; 1997 sc->sc_euid = td->td_ucred->cr_uid; 1998 sc->sc_gid = td->td_ucred->cr_rgid; 1999 sc->sc_egid = td->td_ucred->cr_gid; 2000 sc->sc_ngroups = ngroups; 2001 for (i = 0; i < sc->sc_ngroups; i++) 2002 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2003 2004 /* 2005 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2006 * created SCM_CREDS control message (struct sockcred) has another 2007 * format. 2008 */ 2009 if (control != NULL) 2010 for (n = control, n_prev = NULL; n != NULL;) { 2011 cm = mtod(n, struct cmsghdr *); 2012 if (cm->cmsg_level == SOL_SOCKET && 2013 cm->cmsg_type == SCM_CREDS) { 2014 if (n_prev == NULL) 2015 control = n->m_next; 2016 else 2017 n_prev->m_next = n->m_next; 2018 n = m_free(n); 2019 } else { 2020 n_prev = n; 2021 n = n->m_next; 2022 } 2023 } 2024 2025 /* Prepend it to the head. */ 2026 m->m_next = control; 2027 return (m); 2028} 2029 2030static struct unpcb * 2031fptounp(struct file *fp) 2032{ 2033 struct socket *so; 2034 2035 if (fp->f_type != DTYPE_SOCKET) 2036 return (NULL); 2037 if ((so = fp->f_data) == NULL) 2038 return (NULL); 2039 if (so->so_proto->pr_domain != &localdomain) 2040 return (NULL); 2041 return sotounpcb(so); 2042} 2043 2044static void 2045unp_discard(struct file *fp) 2046{ 2047 struct unp_defer *dr; 2048 2049 if (unp_externalize_fp(fp)) { 2050 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2051 dr->ud_fp = fp; 2052 UNP_DEFERRED_LOCK(); 2053 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2054 UNP_DEFERRED_UNLOCK(); 2055 atomic_add_int(&unp_defers_count, 1); 2056 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2057 } else 2058 (void) closef(fp, (struct thread *)NULL); 2059} 2060 2061static void 2062unp_process_defers(void *arg __unused, int pending) 2063{ 2064 struct unp_defer *dr; 2065 SLIST_HEAD(, unp_defer) drl; 2066 int count; 2067 2068 SLIST_INIT(&drl); 2069 for (;;) { 2070 UNP_DEFERRED_LOCK(); 2071 if (SLIST_FIRST(&unp_defers) == NULL) { 2072 UNP_DEFERRED_UNLOCK(); 2073 break; 2074 } 2075 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2076 UNP_DEFERRED_UNLOCK(); 2077 count = 0; 2078 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2079 SLIST_REMOVE_HEAD(&drl, ud_link); 2080 closef(dr->ud_fp, NULL); 2081 free(dr, M_TEMP); 2082 count++; 2083 } 2084 atomic_add_int(&unp_defers_count, -count); 2085 } 2086} 2087 2088static void 2089unp_internalize_fp(struct file *fp) 2090{ 2091 struct unpcb *unp; 2092 2093 UNP_LINK_WLOCK(); 2094 if ((unp = fptounp(fp)) != NULL) { 2095 unp->unp_file = fp; 2096 unp->unp_msgcount++; 2097 } 2098 fhold(fp); 2099 unp_rights++; 2100 UNP_LINK_WUNLOCK(); 2101} 2102 2103static int 2104unp_externalize_fp(struct file *fp) 2105{ 2106 struct unpcb *unp; 2107 int ret; 2108 2109 UNP_LINK_WLOCK(); 2110 if ((unp = fptounp(fp)) != NULL) { 2111 unp->unp_msgcount--; 2112 ret = 1; 2113 } else 2114 ret = 0; 2115 unp_rights--; 2116 UNP_LINK_WUNLOCK(); 2117 return (ret); 2118} 2119 2120/* 2121 * unp_defer indicates whether additional work has been defered for a future 2122 * pass through unp_gc(). It is thread local and does not require explicit 2123 * synchronization. 2124 */ 2125static int unp_marked; 2126static int unp_unreachable; 2127 2128static void 2129unp_accessable(struct file *fp) 2130{ 2131 struct unpcb *unp; 2132 2133 if ((unp = fptounp(fp)) == NULL) 2134 return; 2135 if (unp->unp_gcflag & UNPGC_REF) 2136 return; 2137 unp->unp_gcflag &= ~UNPGC_DEAD; 2138 unp->unp_gcflag |= UNPGC_REF; 2139 unp_marked++; 2140} 2141 2142static void 2143unp_gc_process(struct unpcb *unp) 2144{ 2145 struct socket *soa; 2146 struct socket *so; 2147 struct file *fp; 2148 2149 /* Already processed. */ 2150 if (unp->unp_gcflag & UNPGC_SCANNED) 2151 return; 2152 fp = unp->unp_file; 2153 2154 /* 2155 * Check for a socket potentially in a cycle. It must be in a 2156 * queue as indicated by msgcount, and this must equal the file 2157 * reference count. Note that when msgcount is 0 the file is NULL. 2158 */ 2159 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2160 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2161 unp->unp_gcflag |= UNPGC_DEAD; 2162 unp_unreachable++; 2163 return; 2164 } 2165 2166 /* 2167 * Mark all sockets we reference with RIGHTS. 2168 */ 2169 so = unp->unp_socket; 2170 SOCKBUF_LOCK(&so->so_rcv); 2171 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2172 SOCKBUF_UNLOCK(&so->so_rcv); 2173 2174 /* 2175 * Mark all sockets in our accept queue. 2176 */ 2177 ACCEPT_LOCK(); 2178 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2179 SOCKBUF_LOCK(&soa->so_rcv); 2180 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2181 SOCKBUF_UNLOCK(&soa->so_rcv); 2182 } 2183 ACCEPT_UNLOCK(); 2184 unp->unp_gcflag |= UNPGC_SCANNED; 2185} 2186 2187static int unp_recycled; 2188SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2189 "Number of unreachable sockets claimed by the garbage collector."); 2190 2191static int unp_taskcount; 2192SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2193 "Number of times the garbage collector has run."); 2194 2195static void 2196unp_gc(__unused void *arg, int pending) 2197{ 2198 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2199 NULL }; 2200 struct unp_head **head; 2201 struct file *f, **unref; 2202 struct unpcb *unp; 2203 int i, total; 2204 2205 unp_taskcount++; 2206 UNP_LIST_LOCK(); 2207 /* 2208 * First clear all gc flags from previous runs. 2209 */ 2210 for (head = heads; *head != NULL; head++) 2211 LIST_FOREACH(unp, *head, unp_link) 2212 unp->unp_gcflag = 0; 2213 2214 /* 2215 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2216 * is reachable all of the sockets it references are reachable. 2217 * Stop the scan once we do a complete loop without discovering 2218 * a new reachable socket. 2219 */ 2220 do { 2221 unp_unreachable = 0; 2222 unp_marked = 0; 2223 for (head = heads; *head != NULL; head++) 2224 LIST_FOREACH(unp, *head, unp_link) 2225 unp_gc_process(unp); 2226 } while (unp_marked); 2227 UNP_LIST_UNLOCK(); 2228 if (unp_unreachable == 0) 2229 return; 2230 2231 /* 2232 * Allocate space for a local list of dead unpcbs. 2233 */ 2234 unref = malloc(unp_unreachable * sizeof(struct file *), 2235 M_TEMP, M_WAITOK); 2236 2237 /* 2238 * Iterate looking for sockets which have been specifically marked 2239 * as as unreachable and store them locally. 2240 */ 2241 UNP_LINK_RLOCK(); 2242 UNP_LIST_LOCK(); 2243 for (total = 0, head = heads; *head != NULL; head++) 2244 LIST_FOREACH(unp, *head, unp_link) 2245 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2246 f = unp->unp_file; 2247 if (unp->unp_msgcount == 0 || f == NULL || 2248 f->f_count != unp->unp_msgcount) 2249 continue; 2250 unref[total++] = f; 2251 fhold(f); 2252 KASSERT(total <= unp_unreachable, 2253 ("unp_gc: incorrect unreachable count.")); 2254 } 2255 UNP_LIST_UNLOCK(); 2256 UNP_LINK_RUNLOCK(); 2257 2258 /* 2259 * Now flush all sockets, free'ing rights. This will free the 2260 * struct files associated with these sockets but leave each socket 2261 * with one remaining ref. 2262 */ 2263 for (i = 0; i < total; i++) { 2264 struct socket *so; 2265 2266 so = unref[i]->f_data; 2267 CURVNET_SET(so->so_vnet); 2268 sorflush(so); 2269 CURVNET_RESTORE(); 2270 } 2271 2272 /* 2273 * And finally release the sockets so they can be reclaimed. 2274 */ 2275 for (i = 0; i < total; i++) 2276 fdrop(unref[i], NULL); 2277 unp_recycled += total; 2278 free(unref, M_TEMP); 2279} 2280 2281static void 2282unp_dispose(struct mbuf *m) 2283{ 2284 2285 if (m) 2286 unp_scan(m, unp_discard); 2287} 2288 2289static void 2290unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2291{ 2292 struct mbuf *m; 2293 struct filedescent *fdep; 2294 struct cmsghdr *cm; 2295 void *data; 2296 int i; 2297 socklen_t clen, datalen; 2298 int qfds; 2299 2300 while (m0 != NULL) { 2301 for (m = m0; m; m = m->m_next) { 2302 if (m->m_type != MT_CONTROL) 2303 continue; 2304 2305 cm = mtod(m, struct cmsghdr *); 2306 clen = m->m_len; 2307 2308 while (cm != NULL) { 2309 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2310 break; 2311 2312 data = CMSG_DATA(cm); 2313 datalen = (caddr_t)cm + cm->cmsg_len 2314 - (caddr_t)data; 2315 2316 if (cm->cmsg_level == SOL_SOCKET && 2317 cm->cmsg_type == SCM_RIGHTS) { 2318 qfds = datalen / sizeof(*fdep); 2319 fdep = data; 2320 for (i = 0; i < qfds; i++, fdep++) 2321 (*op)(fdep->fde_file); 2322 } 2323 2324 if (CMSG_SPACE(datalen) < clen) { 2325 clen -= CMSG_SPACE(datalen); 2326 cm = (struct cmsghdr *) 2327 ((caddr_t)cm + CMSG_SPACE(datalen)); 2328 } else { 2329 clen = 0; 2330 cm = NULL; 2331 } 2332 } 2333 } 2334 m0 = m0->m_act; 2335 } 2336} 2337 2338/* 2339 * A helper function called by VFS before socket-type vnode reclamation. 2340 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2341 * use count. 2342 */ 2343void 2344vfs_unp_reclaim(struct vnode *vp) 2345{ 2346 struct socket *so; 2347 struct unpcb *unp; 2348 int active; 2349 2350 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2351 KASSERT(vp->v_type == VSOCK, 2352 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2353 2354 active = 0; 2355 UNP_LINK_WLOCK(); 2356 VOP_UNP_CONNECT(vp, &so); 2357 if (so == NULL) 2358 goto done; 2359 unp = sotounpcb(so); 2360 if (unp == NULL) 2361 goto done; 2362 UNP_PCB_LOCK(unp); 2363 if (unp->unp_vnode == vp) { 2364 VOP_UNP_DETACH(vp); 2365 unp->unp_vnode = NULL; 2366 active = 1; 2367 } 2368 UNP_PCB_UNLOCK(unp); 2369done: 2370 UNP_LINK_WUNLOCK(); 2371 if (active) 2372 vunref(vp); 2373} 2374 2375#ifdef DDB 2376static void 2377db_print_indent(int indent) 2378{ 2379 int i; 2380 2381 for (i = 0; i < indent; i++) 2382 db_printf(" "); 2383} 2384 2385static void 2386db_print_unpflags(int unp_flags) 2387{ 2388 int comma; 2389 2390 comma = 0; 2391 if (unp_flags & UNP_HAVEPC) { 2392 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2393 comma = 1; 2394 } 2395 if (unp_flags & UNP_HAVEPCCACHED) { 2396 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2397 comma = 1; 2398 } 2399 if (unp_flags & UNP_WANTCRED) { 2400 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2401 comma = 1; 2402 } 2403 if (unp_flags & UNP_CONNWAIT) { 2404 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2405 comma = 1; 2406 } 2407 if (unp_flags & UNP_CONNECTING) { 2408 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2409 comma = 1; 2410 } 2411 if (unp_flags & UNP_BINDING) { 2412 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2413 comma = 1; 2414 } 2415} 2416 2417static void 2418db_print_xucred(int indent, struct xucred *xu) 2419{ 2420 int comma, i; 2421 2422 db_print_indent(indent); 2423 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2424 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2425 db_print_indent(indent); 2426 db_printf("cr_groups: "); 2427 comma = 0; 2428 for (i = 0; i < xu->cr_ngroups; i++) { 2429 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2430 comma = 1; 2431 } 2432 db_printf("\n"); 2433} 2434 2435static void 2436db_print_unprefs(int indent, struct unp_head *uh) 2437{ 2438 struct unpcb *unp; 2439 int counter; 2440 2441 counter = 0; 2442 LIST_FOREACH(unp, uh, unp_reflink) { 2443 if (counter % 4 == 0) 2444 db_print_indent(indent); 2445 db_printf("%p ", unp); 2446 if (counter % 4 == 3) 2447 db_printf("\n"); 2448 counter++; 2449 } 2450 if (counter != 0 && counter % 4 != 0) 2451 db_printf("\n"); 2452} 2453 2454DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2455{ 2456 struct unpcb *unp; 2457 2458 if (!have_addr) { 2459 db_printf("usage: show unpcb <addr>\n"); 2460 return; 2461 } 2462 unp = (struct unpcb *)addr; 2463 2464 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2465 unp->unp_vnode); 2466 2467 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 2468 unp->unp_conn); 2469 2470 db_printf("unp_refs:\n"); 2471 db_print_unprefs(2, &unp->unp_refs); 2472 2473 /* XXXRW: Would be nice to print the full address, if any. */ 2474 db_printf("unp_addr: %p\n", unp->unp_addr); 2475 2476 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2477 unp->unp_cc, unp->unp_mbcnt, 2478 (unsigned long long)unp->unp_gencnt); 2479 2480 db_printf("unp_flags: %x (", unp->unp_flags); 2481 db_print_unpflags(unp->unp_flags); 2482 db_printf(")\n"); 2483 2484 db_printf("unp_peercred:\n"); 2485 db_print_xucred(2, &unp->unp_peercred); 2486 2487 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2488} 2489#endif 2490