uipc_usrreq.c revision 260598
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/9/sys/kern/uipc_usrreq.c 260598 2014-01-13 18:57:30Z jhb $"); 61 62#include "opt_ddb.h" 63 64#include <sys/param.h> 65#include <sys/domain.h> 66#include <sys/fcntl.h> 67#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 68#include <sys/eventhandler.h> 69#include <sys/file.h> 70#include <sys/filedesc.h> 71#include <sys/kernel.h> 72#include <sys/lock.h> 73#include <sys/mbuf.h> 74#include <sys/mount.h> 75#include <sys/mutex.h> 76#include <sys/namei.h> 77#include <sys/proc.h> 78#include <sys/protosw.h> 79#include <sys/queue.h> 80#include <sys/resourcevar.h> 81#include <sys/rwlock.h> 82#include <sys/socket.h> 83#include <sys/socketvar.h> 84#include <sys/signalvar.h> 85#include <sys/stat.h> 86#include <sys/sx.h> 87#include <sys/sysctl.h> 88#include <sys/systm.h> 89#include <sys/taskqueue.h> 90#include <sys/un.h> 91#include <sys/unpcb.h> 92#include <sys/vnode.h> 93 94#include <net/vnet.h> 95 96#ifdef DDB 97#include <ddb/ddb.h> 98#endif 99 100#include <security/mac/mac_framework.h> 101 102#include <vm/uma.h> 103 104/* 105 * Locking key: 106 * (l) Locked using list lock 107 * (g) Locked using linkage lock 108 */ 109 110static uma_zone_t unp_zone; 111static unp_gen_t unp_gencnt; /* (l) */ 112static u_int unp_count; /* (l) Count of local sockets. */ 113static ino_t unp_ino; /* Prototype for fake inode numbers. */ 114static int unp_rights; /* (g) File descriptors in flight. */ 115static struct unp_head unp_shead; /* (l) List of stream sockets. */ 116static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 117static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 118 119struct unp_defer { 120 SLIST_ENTRY(unp_defer) ud_link; 121 struct file *ud_fp; 122}; 123static SLIST_HEAD(, unp_defer) unp_defers; 124static int unp_defers_count; 125 126static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 127 128/* 129 * Garbage collection of cyclic file descriptor/socket references occurs 130 * asynchronously in a taskqueue context in order to avoid recursion and 131 * reentrance in the UNIX domain socket, file descriptor, and socket layer 132 * code. See unp_gc() for a full description. 133 */ 134static struct timeout_task unp_gc_task; 135 136/* 137 * The close of unix domain sockets attached as SCM_RIGHTS is 138 * postponed to the taskqueue, to avoid arbitrary recursion depth. 139 * The attached sockets might have another sockets attached. 140 */ 141static struct task unp_defer_task; 142 143/* 144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 145 * stream sockets, although the total for sender and receiver is actually 146 * only PIPSIZ. 147 * 148 * Datagram sockets really use the sendspace as the maximum datagram size, 149 * and don't really want to reserve the sendspace. Their recvspace should be 150 * large enough for at least one max-size datagram plus address. 151 */ 152#ifndef PIPSIZ 153#define PIPSIZ 8192 154#endif 155static u_long unpst_sendspace = PIPSIZ; 156static u_long unpst_recvspace = PIPSIZ; 157static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 158static u_long unpdg_recvspace = 4*1024; 159static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 160static u_long unpsp_recvspace = PIPSIZ; 161 162static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 163static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, 164 "SOCK_STREAM"); 165static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 166static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, 167 "SOCK_SEQPACKET"); 168 169SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 170 &unpst_sendspace, 0, "Default stream send space."); 171SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 172 &unpst_recvspace, 0, "Default stream receive space."); 173SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 174 &unpdg_sendspace, 0, "Default datagram send space."); 175SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 176 &unpdg_recvspace, 0, "Default datagram receive space."); 177SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 178 &unpsp_sendspace, 0, "Default seqpacket send space."); 179SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 180 &unpsp_recvspace, 0, "Default seqpacket receive space."); 181SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 182 "File descriptors in flight."); 183SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 184 &unp_defers_count, 0, 185 "File descriptors deferred to taskqueue for close."); 186 187/* 188 * Locking and synchronization: 189 * 190 * Three types of locks exit in the local domain socket implementation: a 191 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 192 * global locks, the list lock protects the socket count, global generation 193 * number, and stream/datagram global lists. The linkage lock protects the 194 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 195 * held exclusively over the acquisition of multiple unpcb locks to prevent 196 * deadlock. 197 * 198 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 199 * allocated in pru_attach() and freed in pru_detach(). The validity of that 200 * pointer is an invariant, so no lock is required to dereference the so_pcb 201 * pointer if a valid socket reference is held by the caller. In practice, 202 * this is always true during operations performed on a socket. Each unpcb 203 * has a back-pointer to its socket, unp_socket, which will be stable under 204 * the same circumstances. 205 * 206 * This pointer may only be safely dereferenced as long as a valid reference 207 * to the unpcb is held. Typically, this reference will be from the socket, 208 * or from another unpcb when the referring unpcb's lock is held (in order 209 * that the reference not be invalidated during use). For example, to follow 210 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 211 * as unp_socket remains valid as long as the reference to unp_conn is valid. 212 * 213 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 214 * atomic reads without the lock may be performed "lockless", but more 215 * complex reads and read-modify-writes require the mutex to be held. No 216 * lock order is defined between unpcb locks -- multiple unpcb locks may be 217 * acquired at the same time only when holding the linkage rwlock 218 * exclusively, which prevents deadlocks. 219 * 220 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 221 * protocols, bind() is a non-atomic operation, and connect() requires 222 * potential sleeping in the protocol, due to potentially waiting on local or 223 * distributed file systems. We try to separate "lookup" operations, which 224 * may sleep, and the IPC operations themselves, which typically can occur 225 * with relative atomicity as locks can be held over the entire operation. 226 * 227 * Another tricky issue is simultaneous multi-threaded or multi-process 228 * access to a single UNIX domain socket. These are handled by the flags 229 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 230 * binding, both of which involve dropping UNIX domain socket locks in order 231 * to perform namei() and other file system operations. 232 */ 233static struct rwlock unp_link_rwlock; 234static struct mtx unp_list_lock; 235static struct mtx unp_defers_lock; 236 237#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 238 "unp_link_rwlock") 239 240#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 241 RA_LOCKED) 242#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 243 RA_UNLOCKED) 244 245#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 246#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 247#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 248#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 249#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 250 RA_WLOCKED) 251 252#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 253 "unp_list_lock", NULL, MTX_DEF) 254#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 255#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 256 257#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 258 "unp_defer", NULL, MTX_DEF) 259#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 260#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 261 262#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 263 "unp_mtx", "unp_mtx", \ 264 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 265#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 266#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 267#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 268#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 269 270static int uipc_connect2(struct socket *, struct socket *); 271static int uipc_ctloutput(struct socket *, struct sockopt *); 272static int unp_connect(struct socket *, struct sockaddr *, 273 struct thread *); 274static int unp_connect2(struct socket *so, struct socket *so2, int); 275static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 276static void unp_dispose(struct mbuf *); 277static void unp_shutdown(struct unpcb *); 278static void unp_drop(struct unpcb *, int); 279static void unp_gc(__unused void *, int); 280static void unp_scan(struct mbuf *, void (*)(struct file *)); 281static void unp_discard(struct file *); 282static void unp_freerights(struct file **, int); 283static void unp_init(void); 284static int unp_internalize(struct mbuf **, struct thread *); 285static void unp_internalize_fp(struct file *); 286static int unp_externalize(struct mbuf *, struct mbuf **); 287static int unp_externalize_fp(struct file *); 288static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 289static void unp_process_defers(void * __unused, int); 290 291/* 292 * Definitions of protocols supported in the LOCAL domain. 293 */ 294static struct domain localdomain; 295static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 296static struct pr_usrreqs uipc_usrreqs_seqpacket; 297static struct protosw localsw[] = { 298{ 299 .pr_type = SOCK_STREAM, 300 .pr_domain = &localdomain, 301 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 302 .pr_ctloutput = &uipc_ctloutput, 303 .pr_usrreqs = &uipc_usrreqs_stream 304}, 305{ 306 .pr_type = SOCK_DGRAM, 307 .pr_domain = &localdomain, 308 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 309 .pr_ctloutput = &uipc_ctloutput, 310 .pr_usrreqs = &uipc_usrreqs_dgram 311}, 312{ 313 .pr_type = SOCK_SEQPACKET, 314 .pr_domain = &localdomain, 315 316 /* 317 * XXXRW: For now, PR_ADDR because soreceive will bump into them 318 * due to our use of sbappendaddr. A new sbappend variants is needed 319 * that supports both atomic record writes and control data. 320 */ 321 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 322 PR_RIGHTS, 323 .pr_ctloutput = &uipc_ctloutput, 324 .pr_usrreqs = &uipc_usrreqs_seqpacket, 325}, 326}; 327 328static struct domain localdomain = { 329 .dom_family = AF_LOCAL, 330 .dom_name = "local", 331 .dom_init = unp_init, 332 .dom_externalize = unp_externalize, 333 .dom_dispose = unp_dispose, 334 .dom_protosw = localsw, 335 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 336}; 337DOMAIN_SET(local); 338 339static void 340uipc_abort(struct socket *so) 341{ 342 struct unpcb *unp, *unp2; 343 344 unp = sotounpcb(so); 345 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 346 347 UNP_LINK_WLOCK(); 348 UNP_PCB_LOCK(unp); 349 unp2 = unp->unp_conn; 350 if (unp2 != NULL) { 351 UNP_PCB_LOCK(unp2); 352 unp_drop(unp2, ECONNABORTED); 353 UNP_PCB_UNLOCK(unp2); 354 } 355 UNP_PCB_UNLOCK(unp); 356 UNP_LINK_WUNLOCK(); 357} 358 359static int 360uipc_accept(struct socket *so, struct sockaddr **nam) 361{ 362 struct unpcb *unp, *unp2; 363 const struct sockaddr *sa; 364 365 /* 366 * Pass back name of connected socket, if it was bound and we are 367 * still connected (our peer may have closed already!). 368 */ 369 unp = sotounpcb(so); 370 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 371 372 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 373 UNP_LINK_RLOCK(); 374 unp2 = unp->unp_conn; 375 if (unp2 != NULL && unp2->unp_addr != NULL) { 376 UNP_PCB_LOCK(unp2); 377 sa = (struct sockaddr *) unp2->unp_addr; 378 bcopy(sa, *nam, sa->sa_len); 379 UNP_PCB_UNLOCK(unp2); 380 } else { 381 sa = &sun_noname; 382 bcopy(sa, *nam, sa->sa_len); 383 } 384 UNP_LINK_RUNLOCK(); 385 return (0); 386} 387 388static int 389uipc_attach(struct socket *so, int proto, struct thread *td) 390{ 391 u_long sendspace, recvspace; 392 struct unpcb *unp; 393 int error; 394 395 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 396 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 397 switch (so->so_type) { 398 case SOCK_STREAM: 399 sendspace = unpst_sendspace; 400 recvspace = unpst_recvspace; 401 break; 402 403 case SOCK_DGRAM: 404 sendspace = unpdg_sendspace; 405 recvspace = unpdg_recvspace; 406 break; 407 408 case SOCK_SEQPACKET: 409 sendspace = unpsp_sendspace; 410 recvspace = unpsp_recvspace; 411 break; 412 413 default: 414 panic("uipc_attach"); 415 } 416 error = soreserve(so, sendspace, recvspace); 417 if (error) 418 return (error); 419 } 420 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 421 if (unp == NULL) 422 return (ENOBUFS); 423 LIST_INIT(&unp->unp_refs); 424 UNP_PCB_LOCK_INIT(unp); 425 unp->unp_socket = so; 426 so->so_pcb = unp; 427 unp->unp_refcount = 1; 428 429 UNP_LIST_LOCK(); 430 unp->unp_gencnt = ++unp_gencnt; 431 unp_count++; 432 switch (so->so_type) { 433 case SOCK_STREAM: 434 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 435 break; 436 437 case SOCK_DGRAM: 438 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 439 break; 440 441 case SOCK_SEQPACKET: 442 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 443 break; 444 445 default: 446 panic("uipc_attach"); 447 } 448 UNP_LIST_UNLOCK(); 449 450 return (0); 451} 452 453static int 454uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 455{ 456 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 457 struct vattr vattr; 458 int error, namelen, vfslocked; 459 struct nameidata nd; 460 struct unpcb *unp; 461 struct vnode *vp; 462 struct mount *mp; 463 char *buf; 464 465 unp = sotounpcb(so); 466 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 467 468 if (soun->sun_len > sizeof(struct sockaddr_un)) 469 return (EINVAL); 470 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 471 if (namelen <= 0) 472 return (EINVAL); 473 474 /* 475 * We don't allow simultaneous bind() calls on a single UNIX domain 476 * socket, so flag in-progress operations, and return an error if an 477 * operation is already in progress. 478 * 479 * Historically, we have not allowed a socket to be rebound, so this 480 * also returns an error. Not allowing re-binding simplifies the 481 * implementation and avoids a great many possible failure modes. 482 */ 483 UNP_PCB_LOCK(unp); 484 if (unp->unp_vnode != NULL) { 485 UNP_PCB_UNLOCK(unp); 486 return (EINVAL); 487 } 488 if (unp->unp_flags & UNP_BINDING) { 489 UNP_PCB_UNLOCK(unp); 490 return (EALREADY); 491 } 492 unp->unp_flags |= UNP_BINDING; 493 UNP_PCB_UNLOCK(unp); 494 495 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 496 bcopy(soun->sun_path, buf, namelen); 497 buf[namelen] = 0; 498 499restart: 500 vfslocked = 0; 501 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 502 UIO_SYSSPACE, buf, td); 503/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 504 error = namei(&nd); 505 if (error) 506 goto error; 507 vp = nd.ni_vp; 508 vfslocked = NDHASGIANT(&nd); 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 VFS_UNLOCK_GIANT(vfslocked); 524 goto restart; 525 } 526 VATTR_NULL(&vattr); 527 vattr.va_type = VSOCK; 528 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 529#ifdef MAC 530 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 531 &vattr); 532#endif 533 if (error == 0) 534 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 535 NDFREE(&nd, NDF_ONLY_PNBUF); 536 vput(nd.ni_dvp); 537 if (error) { 538 vn_finished_write(mp); 539 goto error; 540 } 541 vp = nd.ni_vp; 542 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 543 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 544 545 UNP_LINK_WLOCK(); 546 UNP_PCB_LOCK(unp); 547 VOP_UNP_BIND(vp, unp->unp_socket); 548 unp->unp_vnode = vp; 549 unp->unp_addr = soun; 550 unp->unp_flags &= ~UNP_BINDING; 551 UNP_PCB_UNLOCK(unp); 552 UNP_LINK_WUNLOCK(); 553 VOP_UNLOCK(vp, 0); 554 vn_finished_write(mp); 555 VFS_UNLOCK_GIANT(vfslocked); 556 free(buf, M_TEMP); 557 return (0); 558 559error: 560 VFS_UNLOCK_GIANT(vfslocked); 561 UNP_PCB_LOCK(unp); 562 unp->unp_flags &= ~UNP_BINDING; 563 UNP_PCB_UNLOCK(unp); 564 free(buf, M_TEMP); 565 return (error); 566} 567 568static int 569uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 570{ 571 int error; 572 573 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 574 UNP_LINK_WLOCK(); 575 error = unp_connect(so, nam, td); 576 UNP_LINK_WUNLOCK(); 577 return (error); 578} 579 580static void 581uipc_close(struct socket *so) 582{ 583 struct unpcb *unp, *unp2; 584 585 unp = sotounpcb(so); 586 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 587 588 UNP_LINK_WLOCK(); 589 UNP_PCB_LOCK(unp); 590 unp2 = unp->unp_conn; 591 if (unp2 != NULL) { 592 UNP_PCB_LOCK(unp2); 593 unp_disconnect(unp, unp2); 594 UNP_PCB_UNLOCK(unp2); 595 } 596 UNP_PCB_UNLOCK(unp); 597 UNP_LINK_WUNLOCK(); 598} 599 600static int 601uipc_connect2(struct socket *so1, struct socket *so2) 602{ 603 struct unpcb *unp, *unp2; 604 int error; 605 606 UNP_LINK_WLOCK(); 607 unp = so1->so_pcb; 608 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 609 UNP_PCB_LOCK(unp); 610 unp2 = so2->so_pcb; 611 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 612 UNP_PCB_LOCK(unp2); 613 error = unp_connect2(so1, so2, PRU_CONNECT2); 614 UNP_PCB_UNLOCK(unp2); 615 UNP_PCB_UNLOCK(unp); 616 UNP_LINK_WUNLOCK(); 617 return (error); 618} 619 620static void 621uipc_detach(struct socket *so) 622{ 623 struct unpcb *unp, *unp2; 624 struct sockaddr_un *saved_unp_addr; 625 struct vnode *vp; 626 int freeunp, local_unp_rights; 627 628 unp = sotounpcb(so); 629 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 630 631 UNP_LINK_WLOCK(); 632 UNP_LIST_LOCK(); 633 UNP_PCB_LOCK(unp); 634 LIST_REMOVE(unp, unp_link); 635 unp->unp_gencnt = ++unp_gencnt; 636 --unp_count; 637 UNP_LIST_UNLOCK(); 638 639 /* 640 * XXXRW: Should assert vp->v_socket == so. 641 */ 642 if ((vp = unp->unp_vnode) != NULL) { 643 VOP_UNP_DETACH(vp); 644 unp->unp_vnode = NULL; 645 } 646 unp2 = unp->unp_conn; 647 if (unp2 != NULL) { 648 UNP_PCB_LOCK(unp2); 649 unp_disconnect(unp, unp2); 650 UNP_PCB_UNLOCK(unp2); 651 } 652 653 /* 654 * We hold the linkage lock exclusively, so it's OK to acquire 655 * multiple pcb locks at a time. 656 */ 657 while (!LIST_EMPTY(&unp->unp_refs)) { 658 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 659 660 UNP_PCB_LOCK(ref); 661 unp_drop(ref, ECONNRESET); 662 UNP_PCB_UNLOCK(ref); 663 } 664 local_unp_rights = unp_rights; 665 UNP_LINK_WUNLOCK(); 666 unp->unp_socket->so_pcb = NULL; 667 saved_unp_addr = unp->unp_addr; 668 unp->unp_addr = NULL; 669 unp->unp_refcount--; 670 freeunp = (unp->unp_refcount == 0); 671 if (saved_unp_addr != NULL) 672 free(saved_unp_addr, M_SONAME); 673 if (freeunp) { 674 UNP_PCB_LOCK_DESTROY(unp); 675 uma_zfree(unp_zone, unp); 676 } else 677 UNP_PCB_UNLOCK(unp); 678 if (vp) { 679 int vfslocked; 680 681 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 682 vrele(vp); 683 VFS_UNLOCK_GIANT(vfslocked); 684 } 685 if (local_unp_rights) 686 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 687} 688 689static int 690uipc_disconnect(struct socket *so) 691{ 692 struct unpcb *unp, *unp2; 693 694 unp = sotounpcb(so); 695 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 696 697 UNP_LINK_WLOCK(); 698 UNP_PCB_LOCK(unp); 699 unp2 = unp->unp_conn; 700 if (unp2 != NULL) { 701 UNP_PCB_LOCK(unp2); 702 unp_disconnect(unp, unp2); 703 UNP_PCB_UNLOCK(unp2); 704 } 705 UNP_PCB_UNLOCK(unp); 706 UNP_LINK_WUNLOCK(); 707 return (0); 708} 709 710static int 711uipc_listen(struct socket *so, int backlog, struct thread *td) 712{ 713 struct unpcb *unp; 714 int error; 715 716 unp = sotounpcb(so); 717 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 718 719 UNP_PCB_LOCK(unp); 720 if (unp->unp_vnode == NULL) { 721 UNP_PCB_UNLOCK(unp); 722 return (EINVAL); 723 } 724 725 SOCK_LOCK(so); 726 error = solisten_proto_check(so); 727 if (error == 0) { 728 cru2x(td->td_ucred, &unp->unp_peercred); 729 unp->unp_flags |= UNP_HAVEPCCACHED; 730 solisten_proto(so, backlog); 731 } 732 SOCK_UNLOCK(so); 733 UNP_PCB_UNLOCK(unp); 734 return (error); 735} 736 737static int 738uipc_peeraddr(struct socket *so, struct sockaddr **nam) 739{ 740 struct unpcb *unp, *unp2; 741 const struct sockaddr *sa; 742 743 unp = sotounpcb(so); 744 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 745 746 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 747 UNP_LINK_RLOCK(); 748 /* 749 * XXX: It seems that this test always fails even when connection is 750 * established. So, this else clause is added as workaround to 751 * return PF_LOCAL sockaddr. 752 */ 753 unp2 = unp->unp_conn; 754 if (unp2 != NULL) { 755 UNP_PCB_LOCK(unp2); 756 if (unp2->unp_addr != NULL) 757 sa = (struct sockaddr *) unp2->unp_addr; 758 else 759 sa = &sun_noname; 760 bcopy(sa, *nam, sa->sa_len); 761 UNP_PCB_UNLOCK(unp2); 762 } else { 763 sa = &sun_noname; 764 bcopy(sa, *nam, sa->sa_len); 765 } 766 UNP_LINK_RUNLOCK(); 767 return (0); 768} 769 770static int 771uipc_rcvd(struct socket *so, int flags) 772{ 773 struct unpcb *unp, *unp2; 774 struct socket *so2; 775 u_int mbcnt, sbcc; 776 u_long newhiwat; 777 778 unp = sotounpcb(so); 779 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 780 781 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 782 panic("uipc_rcvd socktype %d", so->so_type); 783 784 /* 785 * Adjust backpressure on sender and wakeup any waiting to write. 786 * 787 * The unp lock is acquired to maintain the validity of the unp_conn 788 * pointer; no lock on unp2 is required as unp2->unp_socket will be 789 * static as long as we don't permit unp2 to disconnect from unp, 790 * which is prevented by the lock on unp. We cache values from 791 * so_rcv to avoid holding the so_rcv lock over the entire 792 * transaction on the remote so_snd. 793 */ 794 SOCKBUF_LOCK(&so->so_rcv); 795 mbcnt = so->so_rcv.sb_mbcnt; 796 sbcc = so->so_rcv.sb_cc; 797 SOCKBUF_UNLOCK(&so->so_rcv); 798 UNP_PCB_LOCK(unp); 799 unp2 = unp->unp_conn; 800 if (unp2 == NULL) { 801 UNP_PCB_UNLOCK(unp); 802 return (0); 803 } 804 so2 = unp2->unp_socket; 805 SOCKBUF_LOCK(&so2->so_snd); 806 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 807 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 808 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 809 newhiwat, RLIM_INFINITY); 810 sowwakeup_locked(so2); 811 unp->unp_mbcnt = mbcnt; 812 unp->unp_cc = sbcc; 813 UNP_PCB_UNLOCK(unp); 814 return (0); 815} 816 817static int 818uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 819 struct mbuf *control, struct thread *td) 820{ 821 struct unpcb *unp, *unp2; 822 struct socket *so2; 823 u_int mbcnt_delta, sbcc; 824 u_int newhiwat; 825 int error = 0; 826 827 unp = sotounpcb(so); 828 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 829 830 if (flags & PRUS_OOB) { 831 error = EOPNOTSUPP; 832 goto release; 833 } 834 if (control != NULL && (error = unp_internalize(&control, td))) 835 goto release; 836 if ((nam != NULL) || (flags & PRUS_EOF)) 837 UNP_LINK_WLOCK(); 838 else 839 UNP_LINK_RLOCK(); 840 switch (so->so_type) { 841 case SOCK_DGRAM: 842 { 843 const struct sockaddr *from; 844 845 unp2 = unp->unp_conn; 846 if (nam != NULL) { 847 UNP_LINK_WLOCK_ASSERT(); 848 if (unp2 != NULL) { 849 error = EISCONN; 850 break; 851 } 852 error = unp_connect(so, nam, td); 853 if (error) 854 break; 855 unp2 = unp->unp_conn; 856 } 857 858 /* 859 * Because connect() and send() are non-atomic in a sendto() 860 * with a target address, it's possible that the socket will 861 * have disconnected before the send() can run. In that case 862 * return the slightly counter-intuitive but otherwise 863 * correct error that the socket is not connected. 864 */ 865 if (unp2 == NULL) { 866 error = ENOTCONN; 867 break; 868 } 869 /* Lockless read. */ 870 if (unp2->unp_flags & UNP_WANTCRED) 871 control = unp_addsockcred(td, control); 872 UNP_PCB_LOCK(unp); 873 if (unp->unp_addr != NULL) 874 from = (struct sockaddr *)unp->unp_addr; 875 else 876 from = &sun_noname; 877 so2 = unp2->unp_socket; 878 SOCKBUF_LOCK(&so2->so_rcv); 879 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 880 sorwakeup_locked(so2); 881 m = NULL; 882 control = NULL; 883 } else { 884 SOCKBUF_UNLOCK(&so2->so_rcv); 885 error = ENOBUFS; 886 } 887 if (nam != NULL) { 888 UNP_LINK_WLOCK_ASSERT(); 889 UNP_PCB_LOCK(unp2); 890 unp_disconnect(unp, unp2); 891 UNP_PCB_UNLOCK(unp2); 892 } 893 UNP_PCB_UNLOCK(unp); 894 break; 895 } 896 897 case SOCK_SEQPACKET: 898 case SOCK_STREAM: 899 if ((so->so_state & SS_ISCONNECTED) == 0) { 900 if (nam != NULL) { 901 UNP_LINK_WLOCK_ASSERT(); 902 error = unp_connect(so, nam, td); 903 if (error) 904 break; /* XXX */ 905 } else { 906 error = ENOTCONN; 907 break; 908 } 909 } 910 911 /* Lockless read. */ 912 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 913 error = EPIPE; 914 break; 915 } 916 917 /* 918 * Because connect() and send() are non-atomic in a sendto() 919 * with a target address, it's possible that the socket will 920 * have disconnected before the send() can run. In that case 921 * return the slightly counter-intuitive but otherwise 922 * correct error that the socket is not connected. 923 * 924 * Locking here must be done carefully: the linkage lock 925 * prevents interconnections between unpcbs from changing, so 926 * we can traverse from unp to unp2 without acquiring unp's 927 * lock. Socket buffer locks follow unpcb locks, so we can 928 * acquire both remote and lock socket buffer locks. 929 */ 930 unp2 = unp->unp_conn; 931 if (unp2 == NULL) { 932 error = ENOTCONN; 933 break; 934 } 935 so2 = unp2->unp_socket; 936 UNP_PCB_LOCK(unp2); 937 SOCKBUF_LOCK(&so2->so_rcv); 938 if (unp2->unp_flags & UNP_WANTCRED) { 939 /* 940 * Credentials are passed only once on SOCK_STREAM 941 * and SOCK_SEQPACKET. 942 */ 943 unp2->unp_flags &= ~UNP_WANTCRED; 944 control = unp_addsockcred(td, control); 945 } 946 /* 947 * Send to paired receive port, and then reduce send buffer 948 * hiwater marks to maintain backpressure. Wake up readers. 949 */ 950 switch (so->so_type) { 951 case SOCK_STREAM: 952 if (control != NULL) { 953 if (sbappendcontrol_locked(&so2->so_rcv, m, 954 control)) 955 control = NULL; 956 } else 957 sbappend_locked(&so2->so_rcv, m); 958 break; 959 960 case SOCK_SEQPACKET: { 961 const struct sockaddr *from; 962 963 from = &sun_noname; 964 if (sbappendaddr_locked(&so2->so_rcv, from, m, 965 control)) 966 control = NULL; 967 break; 968 } 969 } 970 971 /* 972 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 973 * datagram size and back-pressure for SOCK_SEQPACKET, which 974 * can lead to undesired return of EMSGSIZE on send instead 975 * of more desirable blocking. 976 */ 977 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 978 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 979 sbcc = so2->so_rcv.sb_cc; 980 sorwakeup_locked(so2); 981 982 SOCKBUF_LOCK(&so->so_snd); 983 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 984 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 985 else 986 newhiwat = 0; 987 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 988 newhiwat, RLIM_INFINITY); 989 so->so_snd.sb_mbmax -= mbcnt_delta; 990 SOCKBUF_UNLOCK(&so->so_snd); 991 unp2->unp_cc = sbcc; 992 UNP_PCB_UNLOCK(unp2); 993 m = NULL; 994 break; 995 996 default: 997 panic("uipc_send unknown socktype"); 998 } 999 1000 /* 1001 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1002 */ 1003 if (flags & PRUS_EOF) { 1004 UNP_PCB_LOCK(unp); 1005 socantsendmore(so); 1006 unp_shutdown(unp); 1007 UNP_PCB_UNLOCK(unp); 1008 } 1009 1010 if ((nam != NULL) || (flags & PRUS_EOF)) 1011 UNP_LINK_WUNLOCK(); 1012 else 1013 UNP_LINK_RUNLOCK(); 1014 1015 if (control != NULL && error != 0) 1016 unp_dispose(control); 1017 1018release: 1019 if (control != NULL) 1020 m_freem(control); 1021 if (m != NULL) 1022 m_freem(m); 1023 return (error); 1024} 1025 1026static int 1027uipc_sense(struct socket *so, struct stat *sb) 1028{ 1029 struct unpcb *unp, *unp2; 1030 struct socket *so2; 1031 1032 unp = sotounpcb(so); 1033 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1034 1035 sb->st_blksize = so->so_snd.sb_hiwat; 1036 UNP_LINK_RLOCK(); 1037 UNP_PCB_LOCK(unp); 1038 unp2 = unp->unp_conn; 1039 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1040 unp2 != NULL) { 1041 so2 = unp2->unp_socket; 1042 sb->st_blksize += so2->so_rcv.sb_cc; 1043 } 1044 sb->st_dev = NODEV; 1045 if (unp->unp_ino == 0) 1046 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1047 sb->st_ino = unp->unp_ino; 1048 UNP_PCB_UNLOCK(unp); 1049 UNP_LINK_RUNLOCK(); 1050 return (0); 1051} 1052 1053static int 1054uipc_shutdown(struct socket *so) 1055{ 1056 struct unpcb *unp; 1057 1058 unp = sotounpcb(so); 1059 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1060 1061 UNP_LINK_WLOCK(); 1062 UNP_PCB_LOCK(unp); 1063 socantsendmore(so); 1064 unp_shutdown(unp); 1065 UNP_PCB_UNLOCK(unp); 1066 UNP_LINK_WUNLOCK(); 1067 return (0); 1068} 1069 1070static int 1071uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1072{ 1073 struct unpcb *unp; 1074 const struct sockaddr *sa; 1075 1076 unp = sotounpcb(so); 1077 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1078 1079 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1080 UNP_PCB_LOCK(unp); 1081 if (unp->unp_addr != NULL) 1082 sa = (struct sockaddr *) unp->unp_addr; 1083 else 1084 sa = &sun_noname; 1085 bcopy(sa, *nam, sa->sa_len); 1086 UNP_PCB_UNLOCK(unp); 1087 return (0); 1088} 1089 1090static struct pr_usrreqs uipc_usrreqs_dgram = { 1091 .pru_abort = uipc_abort, 1092 .pru_accept = uipc_accept, 1093 .pru_attach = uipc_attach, 1094 .pru_bind = uipc_bind, 1095 .pru_connect = uipc_connect, 1096 .pru_connect2 = uipc_connect2, 1097 .pru_detach = uipc_detach, 1098 .pru_disconnect = uipc_disconnect, 1099 .pru_listen = uipc_listen, 1100 .pru_peeraddr = uipc_peeraddr, 1101 .pru_rcvd = uipc_rcvd, 1102 .pru_send = uipc_send, 1103 .pru_sense = uipc_sense, 1104 .pru_shutdown = uipc_shutdown, 1105 .pru_sockaddr = uipc_sockaddr, 1106 .pru_soreceive = soreceive_dgram, 1107 .pru_close = uipc_close, 1108}; 1109 1110static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1111 .pru_abort = uipc_abort, 1112 .pru_accept = uipc_accept, 1113 .pru_attach = uipc_attach, 1114 .pru_bind = uipc_bind, 1115 .pru_connect = uipc_connect, 1116 .pru_connect2 = uipc_connect2, 1117 .pru_detach = uipc_detach, 1118 .pru_disconnect = uipc_disconnect, 1119 .pru_listen = uipc_listen, 1120 .pru_peeraddr = uipc_peeraddr, 1121 .pru_rcvd = uipc_rcvd, 1122 .pru_send = uipc_send, 1123 .pru_sense = uipc_sense, 1124 .pru_shutdown = uipc_shutdown, 1125 .pru_sockaddr = uipc_sockaddr, 1126 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1127 .pru_close = uipc_close, 1128}; 1129 1130static struct pr_usrreqs uipc_usrreqs_stream = { 1131 .pru_abort = uipc_abort, 1132 .pru_accept = uipc_accept, 1133 .pru_attach = uipc_attach, 1134 .pru_bind = uipc_bind, 1135 .pru_connect = uipc_connect, 1136 .pru_connect2 = uipc_connect2, 1137 .pru_detach = uipc_detach, 1138 .pru_disconnect = uipc_disconnect, 1139 .pru_listen = uipc_listen, 1140 .pru_peeraddr = uipc_peeraddr, 1141 .pru_rcvd = uipc_rcvd, 1142 .pru_send = uipc_send, 1143 .pru_sense = uipc_sense, 1144 .pru_shutdown = uipc_shutdown, 1145 .pru_sockaddr = uipc_sockaddr, 1146 .pru_soreceive = soreceive_generic, 1147 .pru_close = uipc_close, 1148}; 1149 1150static int 1151uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1152{ 1153 struct unpcb *unp; 1154 struct xucred xu; 1155 int error, optval; 1156 1157 if (sopt->sopt_level != 0) 1158 return (EINVAL); 1159 1160 unp = sotounpcb(so); 1161 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1162 error = 0; 1163 switch (sopt->sopt_dir) { 1164 case SOPT_GET: 1165 switch (sopt->sopt_name) { 1166 case LOCAL_PEERCRED: 1167 UNP_PCB_LOCK(unp); 1168 if (unp->unp_flags & UNP_HAVEPC) 1169 xu = unp->unp_peercred; 1170 else { 1171 if (so->so_type == SOCK_STREAM) 1172 error = ENOTCONN; 1173 else 1174 error = EINVAL; 1175 } 1176 UNP_PCB_UNLOCK(unp); 1177 if (error == 0) 1178 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1179 break; 1180 1181 case LOCAL_CREDS: 1182 /* Unlocked read. */ 1183 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1184 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1185 break; 1186 1187 case LOCAL_CONNWAIT: 1188 /* Unlocked read. */ 1189 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1190 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1191 break; 1192 1193 default: 1194 error = EOPNOTSUPP; 1195 break; 1196 } 1197 break; 1198 1199 case SOPT_SET: 1200 switch (sopt->sopt_name) { 1201 case LOCAL_CREDS: 1202 case LOCAL_CONNWAIT: 1203 error = sooptcopyin(sopt, &optval, sizeof(optval), 1204 sizeof(optval)); 1205 if (error) 1206 break; 1207 1208#define OPTSET(bit) do { \ 1209 UNP_PCB_LOCK(unp); \ 1210 if (optval) \ 1211 unp->unp_flags |= bit; \ 1212 else \ 1213 unp->unp_flags &= ~bit; \ 1214 UNP_PCB_UNLOCK(unp); \ 1215} while (0) 1216 1217 switch (sopt->sopt_name) { 1218 case LOCAL_CREDS: 1219 OPTSET(UNP_WANTCRED); 1220 break; 1221 1222 case LOCAL_CONNWAIT: 1223 OPTSET(UNP_CONNWAIT); 1224 break; 1225 1226 default: 1227 break; 1228 } 1229 break; 1230#undef OPTSET 1231 default: 1232 error = ENOPROTOOPT; 1233 break; 1234 } 1235 break; 1236 1237 default: 1238 error = EOPNOTSUPP; 1239 break; 1240 } 1241 return (error); 1242} 1243 1244static int 1245unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1246{ 1247 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1248 struct vnode *vp; 1249 struct socket *so2, *so3; 1250 struct unpcb *unp, *unp2, *unp3; 1251 int error, len, vfslocked; 1252 struct nameidata nd; 1253 char buf[SOCK_MAXADDRLEN]; 1254 struct sockaddr *sa; 1255 1256 UNP_LINK_WLOCK_ASSERT(); 1257 1258 unp = sotounpcb(so); 1259 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1260 1261 if (nam->sa_len > sizeof(struct sockaddr_un)) 1262 return (EINVAL); 1263 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1264 if (len <= 0) 1265 return (EINVAL); 1266 bcopy(soun->sun_path, buf, len); 1267 buf[len] = 0; 1268 1269 UNP_PCB_LOCK(unp); 1270 if (unp->unp_flags & UNP_CONNECTING) { 1271 UNP_PCB_UNLOCK(unp); 1272 return (EALREADY); 1273 } 1274 UNP_LINK_WUNLOCK(); 1275 unp->unp_flags |= UNP_CONNECTING; 1276 UNP_PCB_UNLOCK(unp); 1277 1278 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1279 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKSHARED | LOCKLEAF, 1280 UIO_SYSSPACE, buf, td); 1281 error = namei(&nd); 1282 if (error) 1283 vp = NULL; 1284 else 1285 vp = nd.ni_vp; 1286 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1287 vfslocked = NDHASGIANT(&nd); 1288 NDFREE(&nd, NDF_ONLY_PNBUF); 1289 if (error) 1290 goto bad; 1291 1292 if (vp->v_type != VSOCK) { 1293 error = ENOTSOCK; 1294 goto bad; 1295 } 1296#ifdef MAC 1297 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1298 if (error) 1299 goto bad; 1300#endif 1301 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1302 if (error) 1303 goto bad; 1304 VFS_UNLOCK_GIANT(vfslocked); 1305 1306 unp = sotounpcb(so); 1307 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1308 1309 /* 1310 * Lock linkage lock for two reasons: make sure v_socket is stable, 1311 * and to protect simultaneous locking of multiple pcbs. 1312 */ 1313 UNP_LINK_WLOCK(); 1314 VOP_UNP_CONNECT(vp, &so2); 1315 if (so2 == NULL) { 1316 error = ECONNREFUSED; 1317 goto bad2; 1318 } 1319 if (so->so_type != so2->so_type) { 1320 error = EPROTOTYPE; 1321 goto bad2; 1322 } 1323 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1324 if (so2->so_options & SO_ACCEPTCONN) { 1325 CURVNET_SET(so2->so_vnet); 1326 so3 = sonewconn(so2, 0); 1327 CURVNET_RESTORE(); 1328 } else 1329 so3 = NULL; 1330 if (so3 == NULL) { 1331 error = ECONNREFUSED; 1332 goto bad2; 1333 } 1334 unp = sotounpcb(so); 1335 unp2 = sotounpcb(so2); 1336 unp3 = sotounpcb(so3); 1337 UNP_PCB_LOCK(unp); 1338 UNP_PCB_LOCK(unp2); 1339 UNP_PCB_LOCK(unp3); 1340 if (unp2->unp_addr != NULL) { 1341 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1342 unp3->unp_addr = (struct sockaddr_un *) sa; 1343 sa = NULL; 1344 } 1345 1346 /* 1347 * The connector's (client's) credentials are copied from its 1348 * process structure at the time of connect() (which is now). 1349 */ 1350 cru2x(td->td_ucred, &unp3->unp_peercred); 1351 unp3->unp_flags |= UNP_HAVEPC; 1352 1353 /* 1354 * The receiver's (server's) credentials are copied from the 1355 * unp_peercred member of socket on which the former called 1356 * listen(); uipc_listen() cached that process's credentials 1357 * at that time so we can use them now. 1358 */ 1359 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1360 ("unp_connect: listener without cached peercred")); 1361 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1362 sizeof(unp->unp_peercred)); 1363 unp->unp_flags |= UNP_HAVEPC; 1364 if (unp2->unp_flags & UNP_WANTCRED) 1365 unp3->unp_flags |= UNP_WANTCRED; 1366 UNP_PCB_UNLOCK(unp3); 1367 UNP_PCB_UNLOCK(unp2); 1368 UNP_PCB_UNLOCK(unp); 1369#ifdef MAC 1370 mac_socketpeer_set_from_socket(so, so3); 1371 mac_socketpeer_set_from_socket(so3, so); 1372#endif 1373 1374 so2 = so3; 1375 } 1376 unp = sotounpcb(so); 1377 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1378 unp2 = sotounpcb(so2); 1379 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1380 UNP_PCB_LOCK(unp); 1381 UNP_PCB_LOCK(unp2); 1382 error = unp_connect2(so, so2, PRU_CONNECT); 1383 UNP_PCB_UNLOCK(unp2); 1384 UNP_PCB_UNLOCK(unp); 1385bad2: 1386 UNP_LINK_WUNLOCK(); 1387 if (vfslocked) 1388 /* 1389 * Giant has been previously acquired. This means filesystem 1390 * isn't MPSAFE. Do it once again. 1391 */ 1392 mtx_lock(&Giant); 1393bad: 1394 if (vp != NULL) 1395 vput(vp); 1396 VFS_UNLOCK_GIANT(vfslocked); 1397 free(sa, M_SONAME); 1398 UNP_LINK_WLOCK(); 1399 UNP_PCB_LOCK(unp); 1400 unp->unp_flags &= ~UNP_CONNECTING; 1401 UNP_PCB_UNLOCK(unp); 1402 return (error); 1403} 1404 1405static int 1406unp_connect2(struct socket *so, struct socket *so2, int req) 1407{ 1408 struct unpcb *unp; 1409 struct unpcb *unp2; 1410 1411 unp = sotounpcb(so); 1412 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1413 unp2 = sotounpcb(so2); 1414 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1415 1416 UNP_LINK_WLOCK_ASSERT(); 1417 UNP_PCB_LOCK_ASSERT(unp); 1418 UNP_PCB_LOCK_ASSERT(unp2); 1419 1420 if (so2->so_type != so->so_type) 1421 return (EPROTOTYPE); 1422 unp->unp_conn = unp2; 1423 1424 switch (so->so_type) { 1425 case SOCK_DGRAM: 1426 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1427 soisconnected(so); 1428 break; 1429 1430 case SOCK_STREAM: 1431 case SOCK_SEQPACKET: 1432 unp2->unp_conn = unp; 1433 if (req == PRU_CONNECT && 1434 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1435 soisconnecting(so); 1436 else 1437 soisconnected(so); 1438 soisconnected(so2); 1439 break; 1440 1441 default: 1442 panic("unp_connect2"); 1443 } 1444 return (0); 1445} 1446 1447static void 1448unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1449{ 1450 struct socket *so; 1451 1452 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1453 1454 UNP_LINK_WLOCK_ASSERT(); 1455 UNP_PCB_LOCK_ASSERT(unp); 1456 UNP_PCB_LOCK_ASSERT(unp2); 1457 1458 unp->unp_conn = NULL; 1459 switch (unp->unp_socket->so_type) { 1460 case SOCK_DGRAM: 1461 LIST_REMOVE(unp, unp_reflink); 1462 so = unp->unp_socket; 1463 SOCK_LOCK(so); 1464 so->so_state &= ~SS_ISCONNECTED; 1465 SOCK_UNLOCK(so); 1466 break; 1467 1468 case SOCK_STREAM: 1469 case SOCK_SEQPACKET: 1470 soisdisconnected(unp->unp_socket); 1471 unp2->unp_conn = NULL; 1472 soisdisconnected(unp2->unp_socket); 1473 break; 1474 } 1475} 1476 1477/* 1478 * unp_pcblist() walks the global list of struct unpcb's to generate a 1479 * pointer list, bumping the refcount on each unpcb. It then copies them out 1480 * sequentially, validating the generation number on each to see if it has 1481 * been detached. All of this is necessary because copyout() may sleep on 1482 * disk I/O. 1483 */ 1484static int 1485unp_pcblist(SYSCTL_HANDLER_ARGS) 1486{ 1487 int error, i, n; 1488 int freeunp; 1489 struct unpcb *unp, **unp_list; 1490 unp_gen_t gencnt; 1491 struct xunpgen *xug; 1492 struct unp_head *head; 1493 struct xunpcb *xu; 1494 1495 switch ((intptr_t)arg1) { 1496 case SOCK_STREAM: 1497 head = &unp_shead; 1498 break; 1499 1500 case SOCK_DGRAM: 1501 head = &unp_dhead; 1502 break; 1503 1504 case SOCK_SEQPACKET: 1505 head = &unp_sphead; 1506 break; 1507 1508 default: 1509 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1510 } 1511 1512 /* 1513 * The process of preparing the PCB list is too time-consuming and 1514 * resource-intensive to repeat twice on every request. 1515 */ 1516 if (req->oldptr == NULL) { 1517 n = unp_count; 1518 req->oldidx = 2 * (sizeof *xug) 1519 + (n + n/8) * sizeof(struct xunpcb); 1520 return (0); 1521 } 1522 1523 if (req->newptr != NULL) 1524 return (EPERM); 1525 1526 /* 1527 * OK, now we're committed to doing something. 1528 */ 1529 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1530 UNP_LIST_LOCK(); 1531 gencnt = unp_gencnt; 1532 n = unp_count; 1533 UNP_LIST_UNLOCK(); 1534 1535 xug->xug_len = sizeof *xug; 1536 xug->xug_count = n; 1537 xug->xug_gen = gencnt; 1538 xug->xug_sogen = so_gencnt; 1539 error = SYSCTL_OUT(req, xug, sizeof *xug); 1540 if (error) { 1541 free(xug, M_TEMP); 1542 return (error); 1543 } 1544 1545 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1546 1547 UNP_LIST_LOCK(); 1548 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1549 unp = LIST_NEXT(unp, unp_link)) { 1550 UNP_PCB_LOCK(unp); 1551 if (unp->unp_gencnt <= gencnt) { 1552 if (cr_cansee(req->td->td_ucred, 1553 unp->unp_socket->so_cred)) { 1554 UNP_PCB_UNLOCK(unp); 1555 continue; 1556 } 1557 unp_list[i++] = unp; 1558 unp->unp_refcount++; 1559 } 1560 UNP_PCB_UNLOCK(unp); 1561 } 1562 UNP_LIST_UNLOCK(); 1563 n = i; /* In case we lost some during malloc. */ 1564 1565 error = 0; 1566 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1567 for (i = 0; i < n; i++) { 1568 unp = unp_list[i]; 1569 UNP_PCB_LOCK(unp); 1570 unp->unp_refcount--; 1571 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1572 xu->xu_len = sizeof *xu; 1573 xu->xu_unpp = unp; 1574 /* 1575 * XXX - need more locking here to protect against 1576 * connect/disconnect races for SMP. 1577 */ 1578 if (unp->unp_addr != NULL) 1579 bcopy(unp->unp_addr, &xu->xu_addr, 1580 unp->unp_addr->sun_len); 1581 if (unp->unp_conn != NULL && 1582 unp->unp_conn->unp_addr != NULL) 1583 bcopy(unp->unp_conn->unp_addr, 1584 &xu->xu_caddr, 1585 unp->unp_conn->unp_addr->sun_len); 1586 bcopy(unp, &xu->xu_unp, sizeof *unp); 1587 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1588 UNP_PCB_UNLOCK(unp); 1589 error = SYSCTL_OUT(req, xu, sizeof *xu); 1590 } else { 1591 freeunp = (unp->unp_refcount == 0); 1592 UNP_PCB_UNLOCK(unp); 1593 if (freeunp) { 1594 UNP_PCB_LOCK_DESTROY(unp); 1595 uma_zfree(unp_zone, unp); 1596 } 1597 } 1598 } 1599 free(xu, M_TEMP); 1600 if (!error) { 1601 /* 1602 * Give the user an updated idea of our state. If the 1603 * generation differs from what we told her before, she knows 1604 * that something happened while we were processing this 1605 * request, and it might be necessary to retry. 1606 */ 1607 xug->xug_gen = unp_gencnt; 1608 xug->xug_sogen = so_gencnt; 1609 xug->xug_count = unp_count; 1610 error = SYSCTL_OUT(req, xug, sizeof *xug); 1611 } 1612 free(unp_list, M_TEMP); 1613 free(xug, M_TEMP); 1614 return (error); 1615} 1616 1617SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1618 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1619 "List of active local datagram sockets"); 1620SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1621 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1622 "List of active local stream sockets"); 1623SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1624 CTLTYPE_OPAQUE | CTLFLAG_RD, 1625 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1626 "List of active local seqpacket sockets"); 1627 1628static void 1629unp_shutdown(struct unpcb *unp) 1630{ 1631 struct unpcb *unp2; 1632 struct socket *so; 1633 1634 UNP_LINK_WLOCK_ASSERT(); 1635 UNP_PCB_LOCK_ASSERT(unp); 1636 1637 unp2 = unp->unp_conn; 1638 if ((unp->unp_socket->so_type == SOCK_STREAM || 1639 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1640 so = unp2->unp_socket; 1641 if (so != NULL) 1642 socantrcvmore(so); 1643 } 1644} 1645 1646static void 1647unp_drop(struct unpcb *unp, int errno) 1648{ 1649 struct socket *so = unp->unp_socket; 1650 struct unpcb *unp2; 1651 1652 UNP_LINK_WLOCK_ASSERT(); 1653 UNP_PCB_LOCK_ASSERT(unp); 1654 1655 so->so_error = errno; 1656 unp2 = unp->unp_conn; 1657 if (unp2 == NULL) 1658 return; 1659 UNP_PCB_LOCK(unp2); 1660 unp_disconnect(unp, unp2); 1661 UNP_PCB_UNLOCK(unp2); 1662} 1663 1664static void 1665unp_freerights(struct file **rp, int fdcount) 1666{ 1667 int i; 1668 struct file *fp; 1669 1670 for (i = 0; i < fdcount; i++) { 1671 fp = *rp; 1672 *rp++ = NULL; 1673 unp_discard(fp); 1674 } 1675} 1676 1677static int 1678unp_externalize(struct mbuf *control, struct mbuf **controlp) 1679{ 1680 struct thread *td = curthread; /* XXX */ 1681 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1682 int i; 1683 int *fdp; 1684 struct file **rp; 1685 struct file *fp; 1686 void *data; 1687 socklen_t clen = control->m_len, datalen; 1688 int error, newfds; 1689 u_int newlen; 1690 1691 UNP_LINK_UNLOCK_ASSERT(); 1692 1693 error = 0; 1694 if (controlp != NULL) /* controlp == NULL => free control messages */ 1695 *controlp = NULL; 1696 while (cm != NULL) { 1697 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1698 error = EINVAL; 1699 break; 1700 } 1701 data = CMSG_DATA(cm); 1702 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1703 if (cm->cmsg_level == SOL_SOCKET 1704 && cm->cmsg_type == SCM_RIGHTS) { 1705 newfds = datalen / sizeof(struct file *); 1706 rp = data; 1707 1708 /* If we're not outputting the descriptors free them. */ 1709 if (error || controlp == NULL) { 1710 unp_freerights(rp, newfds); 1711 goto next; 1712 } 1713 FILEDESC_XLOCK(td->td_proc->p_fd); 1714 /* 1715 * Now change each pointer to an fd in the global 1716 * table to an integer that is the index to the local 1717 * fd table entry that we set up to point to the 1718 * global one we are transferring. 1719 */ 1720 newlen = newfds * sizeof(int); 1721 *controlp = sbcreatecontrol(NULL, newlen, 1722 SCM_RIGHTS, SOL_SOCKET); 1723 if (*controlp == NULL) { 1724 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1725 error = E2BIG; 1726 unp_freerights(rp, newfds); 1727 goto next; 1728 } 1729 1730 fdp = (int *) 1731 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1732 if (fdallocn(td, 0, fdp, newfds) != 0) { 1733 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1734 error = EMSGSIZE; 1735 unp_freerights(rp, newfds); 1736 m_freem(*controlp); 1737 *controlp = NULL; 1738 goto next; 1739 } 1740 for (i = 0; i < newfds; i++) { 1741 fp = *rp++; 1742 td->td_proc->p_fd->fd_ofiles[fdp[i]] = fp; 1743 unp_externalize_fp(fp); 1744 } 1745 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1746 } else { 1747 /* We can just copy anything else across. */ 1748 if (error || controlp == NULL) 1749 goto next; 1750 *controlp = sbcreatecontrol(NULL, datalen, 1751 cm->cmsg_type, cm->cmsg_level); 1752 if (*controlp == NULL) { 1753 error = ENOBUFS; 1754 goto next; 1755 } 1756 bcopy(data, 1757 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1758 datalen); 1759 } 1760 controlp = &(*controlp)->m_next; 1761 1762next: 1763 if (CMSG_SPACE(datalen) < clen) { 1764 clen -= CMSG_SPACE(datalen); 1765 cm = (struct cmsghdr *) 1766 ((caddr_t)cm + CMSG_SPACE(datalen)); 1767 } else { 1768 clen = 0; 1769 cm = NULL; 1770 } 1771 } 1772 1773 m_freem(control); 1774 return (error); 1775} 1776 1777static void 1778unp_zone_change(void *tag) 1779{ 1780 1781 uma_zone_set_max(unp_zone, maxsockets); 1782} 1783 1784static void 1785unp_init(void) 1786{ 1787 1788#ifdef VIMAGE 1789 if (!IS_DEFAULT_VNET(curvnet)) 1790 return; 1791#endif 1792 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1793 NULL, NULL, UMA_ALIGN_PTR, 0); 1794 if (unp_zone == NULL) 1795 panic("unp_init"); 1796 uma_zone_set_max(unp_zone, maxsockets); 1797 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1798 NULL, EVENTHANDLER_PRI_ANY); 1799 LIST_INIT(&unp_dhead); 1800 LIST_INIT(&unp_shead); 1801 LIST_INIT(&unp_sphead); 1802 SLIST_INIT(&unp_defers); 1803 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 1804 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1805 UNP_LINK_LOCK_INIT(); 1806 UNP_LIST_LOCK_INIT(); 1807 UNP_DEFERRED_LOCK_INIT(); 1808} 1809 1810static int 1811unp_internalize(struct mbuf **controlp, struct thread *td) 1812{ 1813 struct mbuf *control = *controlp; 1814 struct proc *p = td->td_proc; 1815 struct filedesc *fdescp = p->p_fd; 1816 struct bintime *bt; 1817 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1818 struct cmsgcred *cmcred; 1819 struct file **rp; 1820 struct file *fp; 1821 struct timeval *tv; 1822 int i, fd, *fdp; 1823 void *data; 1824 socklen_t clen = control->m_len, datalen; 1825 int error, oldfds; 1826 u_int newlen; 1827 1828 UNP_LINK_UNLOCK_ASSERT(); 1829 1830 error = 0; 1831 *controlp = NULL; 1832 while (cm != NULL) { 1833 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1834 || cm->cmsg_len > clen) { 1835 error = EINVAL; 1836 goto out; 1837 } 1838 data = CMSG_DATA(cm); 1839 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1840 1841 switch (cm->cmsg_type) { 1842 /* 1843 * Fill in credential information. 1844 */ 1845 case SCM_CREDS: 1846 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1847 SCM_CREDS, SOL_SOCKET); 1848 if (*controlp == NULL) { 1849 error = ENOBUFS; 1850 goto out; 1851 } 1852 cmcred = (struct cmsgcred *) 1853 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1854 cmcred->cmcred_pid = p->p_pid; 1855 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1856 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1857 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1858 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1859 CMGROUP_MAX); 1860 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1861 cmcred->cmcred_groups[i] = 1862 td->td_ucred->cr_groups[i]; 1863 break; 1864 1865 case SCM_RIGHTS: 1866 oldfds = datalen / sizeof (int); 1867 /* 1868 * Check that all the FDs passed in refer to legal 1869 * files. If not, reject the entire operation. 1870 */ 1871 fdp = data; 1872 FILEDESC_SLOCK(fdescp); 1873 for (i = 0; i < oldfds; i++) { 1874 fd = *fdp++; 1875 if ((unsigned)fd >= fdescp->fd_nfiles || 1876 fdescp->fd_ofiles[fd] == NULL) { 1877 FILEDESC_SUNLOCK(fdescp); 1878 error = EBADF; 1879 goto out; 1880 } 1881 fp = fdescp->fd_ofiles[fd]; 1882 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1883 FILEDESC_SUNLOCK(fdescp); 1884 error = EOPNOTSUPP; 1885 goto out; 1886 } 1887 1888 } 1889 1890 /* 1891 * Now replace the integer FDs with pointers to the 1892 * associated global file table entry.. 1893 */ 1894 newlen = oldfds * sizeof(struct file *); 1895 *controlp = sbcreatecontrol(NULL, newlen, 1896 SCM_RIGHTS, SOL_SOCKET); 1897 if (*controlp == NULL) { 1898 FILEDESC_SUNLOCK(fdescp); 1899 error = E2BIG; 1900 goto out; 1901 } 1902 fdp = data; 1903 rp = (struct file **) 1904 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1905 for (i = 0; i < oldfds; i++) { 1906 fp = fdescp->fd_ofiles[*fdp++]; 1907 *rp++ = fp; 1908 unp_internalize_fp(fp); 1909 } 1910 FILEDESC_SUNLOCK(fdescp); 1911 break; 1912 1913 case SCM_TIMESTAMP: 1914 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1915 SCM_TIMESTAMP, SOL_SOCKET); 1916 if (*controlp == NULL) { 1917 error = ENOBUFS; 1918 goto out; 1919 } 1920 tv = (struct timeval *) 1921 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1922 microtime(tv); 1923 break; 1924 1925 case SCM_BINTIME: 1926 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 1927 SCM_BINTIME, SOL_SOCKET); 1928 if (*controlp == NULL) { 1929 error = ENOBUFS; 1930 goto out; 1931 } 1932 bt = (struct bintime *) 1933 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1934 bintime(bt); 1935 break; 1936 1937 default: 1938 error = EINVAL; 1939 goto out; 1940 } 1941 1942 controlp = &(*controlp)->m_next; 1943 if (CMSG_SPACE(datalen) < clen) { 1944 clen -= CMSG_SPACE(datalen); 1945 cm = (struct cmsghdr *) 1946 ((caddr_t)cm + CMSG_SPACE(datalen)); 1947 } else { 1948 clen = 0; 1949 cm = NULL; 1950 } 1951 } 1952 1953out: 1954 m_freem(control); 1955 return (error); 1956} 1957 1958static struct mbuf * 1959unp_addsockcred(struct thread *td, struct mbuf *control) 1960{ 1961 struct mbuf *m, *n, *n_prev; 1962 struct sockcred *sc; 1963 const struct cmsghdr *cm; 1964 int ngroups; 1965 int i; 1966 1967 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1968 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1969 if (m == NULL) 1970 return (control); 1971 1972 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1973 sc->sc_uid = td->td_ucred->cr_ruid; 1974 sc->sc_euid = td->td_ucred->cr_uid; 1975 sc->sc_gid = td->td_ucred->cr_rgid; 1976 sc->sc_egid = td->td_ucred->cr_gid; 1977 sc->sc_ngroups = ngroups; 1978 for (i = 0; i < sc->sc_ngroups; i++) 1979 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1980 1981 /* 1982 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1983 * created SCM_CREDS control message (struct sockcred) has another 1984 * format. 1985 */ 1986 if (control != NULL) 1987 for (n = control, n_prev = NULL; n != NULL;) { 1988 cm = mtod(n, struct cmsghdr *); 1989 if (cm->cmsg_level == SOL_SOCKET && 1990 cm->cmsg_type == SCM_CREDS) { 1991 if (n_prev == NULL) 1992 control = n->m_next; 1993 else 1994 n_prev->m_next = n->m_next; 1995 n = m_free(n); 1996 } else { 1997 n_prev = n; 1998 n = n->m_next; 1999 } 2000 } 2001 2002 /* Prepend it to the head. */ 2003 m->m_next = control; 2004 return (m); 2005} 2006 2007static struct unpcb * 2008fptounp(struct file *fp) 2009{ 2010 struct socket *so; 2011 2012 if (fp->f_type != DTYPE_SOCKET) 2013 return (NULL); 2014 if ((so = fp->f_data) == NULL) 2015 return (NULL); 2016 if (so->so_proto->pr_domain != &localdomain) 2017 return (NULL); 2018 return sotounpcb(so); 2019} 2020 2021static void 2022unp_discard(struct file *fp) 2023{ 2024 struct unp_defer *dr; 2025 2026 if (unp_externalize_fp(fp)) { 2027 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2028 dr->ud_fp = fp; 2029 UNP_DEFERRED_LOCK(); 2030 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2031 UNP_DEFERRED_UNLOCK(); 2032 atomic_add_int(&unp_defers_count, 1); 2033 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2034 } else 2035 (void) closef(fp, (struct thread *)NULL); 2036} 2037 2038static void 2039unp_process_defers(void *arg __unused, int pending) 2040{ 2041 struct unp_defer *dr; 2042 SLIST_HEAD(, unp_defer) drl; 2043 int count; 2044 2045 SLIST_INIT(&drl); 2046 for (;;) { 2047 UNP_DEFERRED_LOCK(); 2048 if (SLIST_FIRST(&unp_defers) == NULL) { 2049 UNP_DEFERRED_UNLOCK(); 2050 break; 2051 } 2052 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2053 UNP_DEFERRED_UNLOCK(); 2054 count = 0; 2055 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2056 SLIST_REMOVE_HEAD(&drl, ud_link); 2057 closef(dr->ud_fp, NULL); 2058 free(dr, M_TEMP); 2059 count++; 2060 } 2061 atomic_add_int(&unp_defers_count, -count); 2062 } 2063} 2064 2065static void 2066unp_internalize_fp(struct file *fp) 2067{ 2068 struct unpcb *unp; 2069 2070 UNP_LINK_WLOCK(); 2071 if ((unp = fptounp(fp)) != NULL) { 2072 unp->unp_file = fp; 2073 unp->unp_msgcount++; 2074 } 2075 fhold(fp); 2076 unp_rights++; 2077 UNP_LINK_WUNLOCK(); 2078} 2079 2080static int 2081unp_externalize_fp(struct file *fp) 2082{ 2083 struct unpcb *unp; 2084 int ret; 2085 2086 UNP_LINK_WLOCK(); 2087 if ((unp = fptounp(fp)) != NULL) { 2088 unp->unp_msgcount--; 2089 ret = 1; 2090 } else 2091 ret = 0; 2092 unp_rights--; 2093 UNP_LINK_WUNLOCK(); 2094 return (ret); 2095} 2096 2097/* 2098 * unp_defer indicates whether additional work has been defered for a future 2099 * pass through unp_gc(). It is thread local and does not require explicit 2100 * synchronization. 2101 */ 2102static int unp_marked; 2103static int unp_unreachable; 2104 2105static void 2106unp_accessable(struct file *fp) 2107{ 2108 struct unpcb *unp; 2109 2110 if ((unp = fptounp(fp)) == NULL) 2111 return; 2112 if (unp->unp_gcflag & UNPGC_REF) 2113 return; 2114 unp->unp_gcflag &= ~UNPGC_DEAD; 2115 unp->unp_gcflag |= UNPGC_REF; 2116 unp_marked++; 2117} 2118 2119static void 2120unp_gc_process(struct unpcb *unp) 2121{ 2122 struct socket *soa; 2123 struct socket *so; 2124 struct file *fp; 2125 2126 /* Already processed. */ 2127 if (unp->unp_gcflag & UNPGC_SCANNED) 2128 return; 2129 fp = unp->unp_file; 2130 2131 /* 2132 * Check for a socket potentially in a cycle. It must be in a 2133 * queue as indicated by msgcount, and this must equal the file 2134 * reference count. Note that when msgcount is 0 the file is NULL. 2135 */ 2136 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2137 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2138 unp->unp_gcflag |= UNPGC_DEAD; 2139 unp_unreachable++; 2140 return; 2141 } 2142 2143 /* 2144 * Mark all sockets we reference with RIGHTS. 2145 */ 2146 so = unp->unp_socket; 2147 SOCKBUF_LOCK(&so->so_rcv); 2148 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2149 SOCKBUF_UNLOCK(&so->so_rcv); 2150 2151 /* 2152 * Mark all sockets in our accept queue. 2153 */ 2154 ACCEPT_LOCK(); 2155 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2156 SOCKBUF_LOCK(&soa->so_rcv); 2157 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2158 SOCKBUF_UNLOCK(&soa->so_rcv); 2159 } 2160 ACCEPT_UNLOCK(); 2161 unp->unp_gcflag |= UNPGC_SCANNED; 2162} 2163 2164static int unp_recycled; 2165SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2166 "Number of unreachable sockets claimed by the garbage collector."); 2167 2168static int unp_taskcount; 2169SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2170 "Number of times the garbage collector has run."); 2171 2172static void 2173unp_gc(__unused void *arg, int pending) 2174{ 2175 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2176 NULL }; 2177 struct unp_head **head; 2178 struct file *f, **unref; 2179 struct unpcb *unp; 2180 int i, total; 2181 2182 unp_taskcount++; 2183 UNP_LIST_LOCK(); 2184 /* 2185 * First clear all gc flags from previous runs. 2186 */ 2187 for (head = heads; *head != NULL; head++) 2188 LIST_FOREACH(unp, *head, unp_link) 2189 unp->unp_gcflag = 0; 2190 2191 /* 2192 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2193 * is reachable all of the sockets it references are reachable. 2194 * Stop the scan once we do a complete loop without discovering 2195 * a new reachable socket. 2196 */ 2197 do { 2198 unp_unreachable = 0; 2199 unp_marked = 0; 2200 for (head = heads; *head != NULL; head++) 2201 LIST_FOREACH(unp, *head, unp_link) 2202 unp_gc_process(unp); 2203 } while (unp_marked); 2204 UNP_LIST_UNLOCK(); 2205 if (unp_unreachable == 0) 2206 return; 2207 2208 /* 2209 * Allocate space for a local list of dead unpcbs. 2210 */ 2211 unref = malloc(unp_unreachable * sizeof(struct file *), 2212 M_TEMP, M_WAITOK); 2213 2214 /* 2215 * Iterate looking for sockets which have been specifically marked 2216 * as as unreachable and store them locally. 2217 */ 2218 UNP_LINK_RLOCK(); 2219 UNP_LIST_LOCK(); 2220 for (total = 0, head = heads; *head != NULL; head++) 2221 LIST_FOREACH(unp, *head, unp_link) 2222 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2223 f = unp->unp_file; 2224 if (unp->unp_msgcount == 0 || f == NULL || 2225 f->f_count != unp->unp_msgcount) 2226 continue; 2227 unref[total++] = f; 2228 fhold(f); 2229 KASSERT(total <= unp_unreachable, 2230 ("unp_gc: incorrect unreachable count.")); 2231 } 2232 UNP_LIST_UNLOCK(); 2233 UNP_LINK_RUNLOCK(); 2234 2235 /* 2236 * Now flush all sockets, free'ing rights. This will free the 2237 * struct files associated with these sockets but leave each socket 2238 * with one remaining ref. 2239 */ 2240 for (i = 0; i < total; i++) { 2241 struct socket *so; 2242 2243 so = unref[i]->f_data; 2244 CURVNET_SET(so->so_vnet); 2245 sorflush(so); 2246 CURVNET_RESTORE(); 2247 } 2248 2249 /* 2250 * And finally release the sockets so they can be reclaimed. 2251 */ 2252 for (i = 0; i < total; i++) 2253 fdrop(unref[i], NULL); 2254 unp_recycled += total; 2255 free(unref, M_TEMP); 2256} 2257 2258static void 2259unp_dispose(struct mbuf *m) 2260{ 2261 2262 if (m) 2263 unp_scan(m, unp_discard); 2264} 2265 2266static void 2267unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2268{ 2269 struct mbuf *m; 2270 struct file **rp; 2271 struct cmsghdr *cm; 2272 void *data; 2273 int i; 2274 socklen_t clen, datalen; 2275 int qfds; 2276 2277 while (m0 != NULL) { 2278 for (m = m0; m; m = m->m_next) { 2279 if (m->m_type != MT_CONTROL) 2280 continue; 2281 2282 cm = mtod(m, struct cmsghdr *); 2283 clen = m->m_len; 2284 2285 while (cm != NULL) { 2286 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2287 break; 2288 2289 data = CMSG_DATA(cm); 2290 datalen = (caddr_t)cm + cm->cmsg_len 2291 - (caddr_t)data; 2292 2293 if (cm->cmsg_level == SOL_SOCKET && 2294 cm->cmsg_type == SCM_RIGHTS) { 2295 qfds = datalen / sizeof (struct file *); 2296 rp = data; 2297 for (i = 0; i < qfds; i++) 2298 (*op)(*rp++); 2299 } 2300 2301 if (CMSG_SPACE(datalen) < clen) { 2302 clen -= CMSG_SPACE(datalen); 2303 cm = (struct cmsghdr *) 2304 ((caddr_t)cm + CMSG_SPACE(datalen)); 2305 } else { 2306 clen = 0; 2307 cm = NULL; 2308 } 2309 } 2310 } 2311 m0 = m0->m_act; 2312 } 2313} 2314 2315/* 2316 * A helper function called by VFS before socket-type vnode reclamation. 2317 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2318 * use count. 2319 */ 2320void 2321vfs_unp_reclaim(struct vnode *vp) 2322{ 2323 struct socket *so; 2324 struct unpcb *unp; 2325 int active; 2326 2327 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2328 KASSERT(vp->v_type == VSOCK, 2329 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2330 2331 active = 0; 2332 UNP_LINK_WLOCK(); 2333 VOP_UNP_CONNECT(vp, &so); 2334 if (so == NULL) 2335 goto done; 2336 unp = sotounpcb(so); 2337 if (unp == NULL) 2338 goto done; 2339 UNP_PCB_LOCK(unp); 2340 if (unp->unp_vnode == vp) { 2341 VOP_UNP_DETACH(vp); 2342 unp->unp_vnode = NULL; 2343 active = 1; 2344 } 2345 UNP_PCB_UNLOCK(unp); 2346done: 2347 UNP_LINK_WUNLOCK(); 2348 if (active) 2349 vunref(vp); 2350} 2351 2352#ifdef DDB 2353static void 2354db_print_indent(int indent) 2355{ 2356 int i; 2357 2358 for (i = 0; i < indent; i++) 2359 db_printf(" "); 2360} 2361 2362static void 2363db_print_unpflags(int unp_flags) 2364{ 2365 int comma; 2366 2367 comma = 0; 2368 if (unp_flags & UNP_HAVEPC) { 2369 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2370 comma = 1; 2371 } 2372 if (unp_flags & UNP_HAVEPCCACHED) { 2373 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2374 comma = 1; 2375 } 2376 if (unp_flags & UNP_WANTCRED) { 2377 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2378 comma = 1; 2379 } 2380 if (unp_flags & UNP_CONNWAIT) { 2381 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2382 comma = 1; 2383 } 2384 if (unp_flags & UNP_CONNECTING) { 2385 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2386 comma = 1; 2387 } 2388 if (unp_flags & UNP_BINDING) { 2389 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2390 comma = 1; 2391 } 2392} 2393 2394static void 2395db_print_xucred(int indent, struct xucred *xu) 2396{ 2397 int comma, i; 2398 2399 db_print_indent(indent); 2400 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2401 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2402 db_print_indent(indent); 2403 db_printf("cr_groups: "); 2404 comma = 0; 2405 for (i = 0; i < xu->cr_ngroups; i++) { 2406 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2407 comma = 1; 2408 } 2409 db_printf("\n"); 2410} 2411 2412static void 2413db_print_unprefs(int indent, struct unp_head *uh) 2414{ 2415 struct unpcb *unp; 2416 int counter; 2417 2418 counter = 0; 2419 LIST_FOREACH(unp, uh, unp_reflink) { 2420 if (counter % 4 == 0) 2421 db_print_indent(indent); 2422 db_printf("%p ", unp); 2423 if (counter % 4 == 3) 2424 db_printf("\n"); 2425 counter++; 2426 } 2427 if (counter != 0 && counter % 4 != 0) 2428 db_printf("\n"); 2429} 2430 2431DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2432{ 2433 struct unpcb *unp; 2434 2435 if (!have_addr) { 2436 db_printf("usage: show unpcb <addr>\n"); 2437 return; 2438 } 2439 unp = (struct unpcb *)addr; 2440 2441 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2442 unp->unp_vnode); 2443 2444 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2445 unp->unp_conn); 2446 2447 db_printf("unp_refs:\n"); 2448 db_print_unprefs(2, &unp->unp_refs); 2449 2450 /* XXXRW: Would be nice to print the full address, if any. */ 2451 db_printf("unp_addr: %p\n", unp->unp_addr); 2452 2453 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2454 unp->unp_cc, unp->unp_mbcnt, 2455 (unsigned long long)unp->unp_gencnt); 2456 2457 db_printf("unp_flags: %x (", unp->unp_flags); 2458 db_print_unpflags(unp->unp_flags); 2459 db_printf(")\n"); 2460 2461 db_printf("unp_peercred:\n"); 2462 db_print_xucred(2, &unp->unp_peercred); 2463 2464 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2465} 2466#endif 2467