uipc_usrreq.c revision 241896
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 241896 2012-10-22 17:50:54Z kib $"); 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 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_usrreqs = &uipc_usrreqs_seqpacket, 324}, 325}; 326 327static struct domain localdomain = { 328 .dom_family = AF_LOCAL, 329 .dom_name = "local", 330 .dom_init = unp_init, 331 .dom_externalize = unp_externalize, 332 .dom_dispose = unp_dispose, 333 .dom_protosw = localsw, 334 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 335}; 336DOMAIN_SET(local); 337 338static void 339uipc_abort(struct socket *so) 340{ 341 struct unpcb *unp, *unp2; 342 343 unp = sotounpcb(so); 344 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 345 346 UNP_LINK_WLOCK(); 347 UNP_PCB_LOCK(unp); 348 unp2 = unp->unp_conn; 349 if (unp2 != NULL) { 350 UNP_PCB_LOCK(unp2); 351 unp_drop(unp2, ECONNABORTED); 352 UNP_PCB_UNLOCK(unp2); 353 } 354 UNP_PCB_UNLOCK(unp); 355 UNP_LINK_WUNLOCK(); 356} 357 358static int 359uipc_accept(struct socket *so, struct sockaddr **nam) 360{ 361 struct unpcb *unp, *unp2; 362 const struct sockaddr *sa; 363 364 /* 365 * Pass back name of connected socket, if it was bound and we are 366 * still connected (our peer may have closed already!). 367 */ 368 unp = sotounpcb(so); 369 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 370 371 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 372 UNP_LINK_RLOCK(); 373 unp2 = unp->unp_conn; 374 if (unp2 != NULL && unp2->unp_addr != NULL) { 375 UNP_PCB_LOCK(unp2); 376 sa = (struct sockaddr *) unp2->unp_addr; 377 bcopy(sa, *nam, sa->sa_len); 378 UNP_PCB_UNLOCK(unp2); 379 } else { 380 sa = &sun_noname; 381 bcopy(sa, *nam, sa->sa_len); 382 } 383 UNP_LINK_RUNLOCK(); 384 return (0); 385} 386 387static int 388uipc_attach(struct socket *so, int proto, struct thread *td) 389{ 390 u_long sendspace, recvspace; 391 struct unpcb *unp; 392 int error; 393 394 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 395 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 396 switch (so->so_type) { 397 case SOCK_STREAM: 398 sendspace = unpst_sendspace; 399 recvspace = unpst_recvspace; 400 break; 401 402 case SOCK_DGRAM: 403 sendspace = unpdg_sendspace; 404 recvspace = unpdg_recvspace; 405 break; 406 407 case SOCK_SEQPACKET: 408 sendspace = unpsp_sendspace; 409 recvspace = unpsp_recvspace; 410 break; 411 412 default: 413 panic("uipc_attach"); 414 } 415 error = soreserve(so, sendspace, recvspace); 416 if (error) 417 return (error); 418 } 419 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 420 if (unp == NULL) 421 return (ENOBUFS); 422 LIST_INIT(&unp->unp_refs); 423 UNP_PCB_LOCK_INIT(unp); 424 unp->unp_socket = so; 425 so->so_pcb = unp; 426 unp->unp_refcount = 1; 427 428 UNP_LIST_LOCK(); 429 unp->unp_gencnt = ++unp_gencnt; 430 unp_count++; 431 switch (so->so_type) { 432 case SOCK_STREAM: 433 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 434 break; 435 436 case SOCK_DGRAM: 437 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 438 break; 439 440 case SOCK_SEQPACKET: 441 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 442 break; 443 444 default: 445 panic("uipc_attach"); 446 } 447 UNP_LIST_UNLOCK(); 448 449 return (0); 450} 451 452static int 453uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 454{ 455 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 456 struct vattr vattr; 457 int error, namelen; 458 struct nameidata nd; 459 struct unpcb *unp; 460 struct vnode *vp; 461 struct mount *mp; 462 char *buf; 463 464 unp = sotounpcb(so); 465 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 466 467 if (soun->sun_len > sizeof(struct sockaddr_un)) 468 return (EINVAL); 469 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 470 if (namelen <= 0) 471 return (EINVAL); 472 473 /* 474 * We don't allow simultaneous bind() calls on a single UNIX domain 475 * socket, so flag in-progress operations, and return an error if an 476 * operation is already in progress. 477 * 478 * Historically, we have not allowed a socket to be rebound, so this 479 * also returns an error. Not allowing re-binding simplifies the 480 * implementation and avoids a great many possible failure modes. 481 */ 482 UNP_PCB_LOCK(unp); 483 if (unp->unp_vnode != NULL) { 484 UNP_PCB_UNLOCK(unp); 485 return (EINVAL); 486 } 487 if (unp->unp_flags & UNP_BINDING) { 488 UNP_PCB_UNLOCK(unp); 489 return (EALREADY); 490 } 491 unp->unp_flags |= UNP_BINDING; 492 UNP_PCB_UNLOCK(unp); 493 494 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 495 bcopy(soun->sun_path, buf, namelen); 496 buf[namelen] = 0; 497 498restart: 499 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, 500 UIO_SYSSPACE, buf, td); 501/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 502 error = namei(&nd); 503 if (error) 504 goto error; 505 vp = nd.ni_vp; 506 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 507 NDFREE(&nd, NDF_ONLY_PNBUF); 508 if (nd.ni_dvp == vp) 509 vrele(nd.ni_dvp); 510 else 511 vput(nd.ni_dvp); 512 if (vp != NULL) { 513 vrele(vp); 514 error = EADDRINUSE; 515 goto error; 516 } 517 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 518 if (error) 519 goto error; 520 goto restart; 521 } 522 VATTR_NULL(&vattr); 523 vattr.va_type = VSOCK; 524 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 525#ifdef MAC 526 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 527 &vattr); 528#endif 529 if (error == 0) 530 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 531 NDFREE(&nd, NDF_ONLY_PNBUF); 532 vput(nd.ni_dvp); 533 if (error) { 534 vn_finished_write(mp); 535 goto error; 536 } 537 vp = nd.ni_vp; 538 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 539 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 540 541 UNP_LINK_WLOCK(); 542 UNP_PCB_LOCK(unp); 543 VOP_UNP_BIND(vp, unp->unp_socket); 544 unp->unp_vnode = vp; 545 unp->unp_addr = soun; 546 unp->unp_flags &= ~UNP_BINDING; 547 UNP_PCB_UNLOCK(unp); 548 UNP_LINK_WUNLOCK(); 549 VOP_UNLOCK(vp, 0); 550 vn_finished_write(mp); 551 free(buf, M_TEMP); 552 return (0); 553 554error: 555 UNP_PCB_LOCK(unp); 556 unp->unp_flags &= ~UNP_BINDING; 557 UNP_PCB_UNLOCK(unp); 558 free(buf, M_TEMP); 559 return (error); 560} 561 562static int 563uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 564{ 565 int error; 566 567 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 568 UNP_LINK_WLOCK(); 569 error = unp_connect(so, nam, td); 570 UNP_LINK_WUNLOCK(); 571 return (error); 572} 573 574static void 575uipc_close(struct socket *so) 576{ 577 struct unpcb *unp, *unp2; 578 579 unp = sotounpcb(so); 580 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 581 582 UNP_LINK_WLOCK(); 583 UNP_PCB_LOCK(unp); 584 unp2 = unp->unp_conn; 585 if (unp2 != NULL) { 586 UNP_PCB_LOCK(unp2); 587 unp_disconnect(unp, unp2); 588 UNP_PCB_UNLOCK(unp2); 589 } 590 UNP_PCB_UNLOCK(unp); 591 UNP_LINK_WUNLOCK(); 592} 593 594static int 595uipc_connect2(struct socket *so1, struct socket *so2) 596{ 597 struct unpcb *unp, *unp2; 598 int error; 599 600 UNP_LINK_WLOCK(); 601 unp = so1->so_pcb; 602 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 603 UNP_PCB_LOCK(unp); 604 unp2 = so2->so_pcb; 605 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 606 UNP_PCB_LOCK(unp2); 607 error = unp_connect2(so1, so2, PRU_CONNECT2); 608 UNP_PCB_UNLOCK(unp2); 609 UNP_PCB_UNLOCK(unp); 610 UNP_LINK_WUNLOCK(); 611 return (error); 612} 613 614static void 615uipc_detach(struct socket *so) 616{ 617 struct unpcb *unp, *unp2; 618 struct sockaddr_un *saved_unp_addr; 619 struct vnode *vp; 620 int freeunp, local_unp_rights; 621 622 unp = sotounpcb(so); 623 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 624 625 UNP_LINK_WLOCK(); 626 UNP_LIST_LOCK(); 627 UNP_PCB_LOCK(unp); 628 LIST_REMOVE(unp, unp_link); 629 unp->unp_gencnt = ++unp_gencnt; 630 --unp_count; 631 UNP_LIST_UNLOCK(); 632 633 /* 634 * XXXRW: Should assert vp->v_socket == so. 635 */ 636 if ((vp = unp->unp_vnode) != NULL) { 637 VOP_UNP_DETACH(vp); 638 unp->unp_vnode = NULL; 639 } 640 unp2 = unp->unp_conn; 641 if (unp2 != NULL) { 642 UNP_PCB_LOCK(unp2); 643 unp_disconnect(unp, unp2); 644 UNP_PCB_UNLOCK(unp2); 645 } 646 647 /* 648 * We hold the linkage lock exclusively, so it's OK to acquire 649 * multiple pcb locks at a time. 650 */ 651 while (!LIST_EMPTY(&unp->unp_refs)) { 652 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 653 654 UNP_PCB_LOCK(ref); 655 unp_drop(ref, ECONNRESET); 656 UNP_PCB_UNLOCK(ref); 657 } 658 local_unp_rights = unp_rights; 659 UNP_LINK_WUNLOCK(); 660 unp->unp_socket->so_pcb = NULL; 661 saved_unp_addr = unp->unp_addr; 662 unp->unp_addr = NULL; 663 unp->unp_refcount--; 664 freeunp = (unp->unp_refcount == 0); 665 if (saved_unp_addr != NULL) 666 free(saved_unp_addr, M_SONAME); 667 if (freeunp) { 668 UNP_PCB_LOCK_DESTROY(unp); 669 uma_zfree(unp_zone, unp); 670 } else 671 UNP_PCB_UNLOCK(unp); 672 if (vp) 673 vrele(vp); 674 if (local_unp_rights) 675 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 676} 677 678static int 679uipc_disconnect(struct socket *so) 680{ 681 struct unpcb *unp, *unp2; 682 683 unp = sotounpcb(so); 684 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 685 686 UNP_LINK_WLOCK(); 687 UNP_PCB_LOCK(unp); 688 unp2 = unp->unp_conn; 689 if (unp2 != NULL) { 690 UNP_PCB_LOCK(unp2); 691 unp_disconnect(unp, unp2); 692 UNP_PCB_UNLOCK(unp2); 693 } 694 UNP_PCB_UNLOCK(unp); 695 UNP_LINK_WUNLOCK(); 696 return (0); 697} 698 699static int 700uipc_listen(struct socket *so, int backlog, struct thread *td) 701{ 702 struct unpcb *unp; 703 int error; 704 705 unp = sotounpcb(so); 706 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 707 708 UNP_PCB_LOCK(unp); 709 if (unp->unp_vnode == NULL) { 710 UNP_PCB_UNLOCK(unp); 711 return (EINVAL); 712 } 713 714 SOCK_LOCK(so); 715 error = solisten_proto_check(so); 716 if (error == 0) { 717 cru2x(td->td_ucred, &unp->unp_peercred); 718 unp->unp_flags |= UNP_HAVEPCCACHED; 719 solisten_proto(so, backlog); 720 } 721 SOCK_UNLOCK(so); 722 UNP_PCB_UNLOCK(unp); 723 return (error); 724} 725 726static int 727uipc_peeraddr(struct socket *so, struct sockaddr **nam) 728{ 729 struct unpcb *unp, *unp2; 730 const struct sockaddr *sa; 731 732 unp = sotounpcb(so); 733 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 734 735 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 736 UNP_LINK_RLOCK(); 737 /* 738 * XXX: It seems that this test always fails even when connection is 739 * established. So, this else clause is added as workaround to 740 * return PF_LOCAL sockaddr. 741 */ 742 unp2 = unp->unp_conn; 743 if (unp2 != NULL) { 744 UNP_PCB_LOCK(unp2); 745 if (unp2->unp_addr != NULL) 746 sa = (struct sockaddr *) unp2->unp_addr; 747 else 748 sa = &sun_noname; 749 bcopy(sa, *nam, sa->sa_len); 750 UNP_PCB_UNLOCK(unp2); 751 } else { 752 sa = &sun_noname; 753 bcopy(sa, *nam, sa->sa_len); 754 } 755 UNP_LINK_RUNLOCK(); 756 return (0); 757} 758 759static int 760uipc_rcvd(struct socket *so, int flags) 761{ 762 struct unpcb *unp, *unp2; 763 struct socket *so2; 764 u_int mbcnt, sbcc; 765 u_long newhiwat; 766 767 unp = sotounpcb(so); 768 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 769 770 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 771 panic("uipc_rcvd socktype %d", so->so_type); 772 773 /* 774 * Adjust backpressure on sender and wakeup any waiting to write. 775 * 776 * The unp lock is acquired to maintain the validity of the unp_conn 777 * pointer; no lock on unp2 is required as unp2->unp_socket will be 778 * static as long as we don't permit unp2 to disconnect from unp, 779 * which is prevented by the lock on unp. We cache values from 780 * so_rcv to avoid holding the so_rcv lock over the entire 781 * transaction on the remote so_snd. 782 */ 783 SOCKBUF_LOCK(&so->so_rcv); 784 mbcnt = so->so_rcv.sb_mbcnt; 785 sbcc = so->so_rcv.sb_cc; 786 SOCKBUF_UNLOCK(&so->so_rcv); 787 UNP_PCB_LOCK(unp); 788 unp2 = unp->unp_conn; 789 if (unp2 == NULL) { 790 UNP_PCB_UNLOCK(unp); 791 return (0); 792 } 793 so2 = unp2->unp_socket; 794 SOCKBUF_LOCK(&so2->so_snd); 795 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 796 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 797 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 798 newhiwat, RLIM_INFINITY); 799 sowwakeup_locked(so2); 800 unp->unp_mbcnt = mbcnt; 801 unp->unp_cc = sbcc; 802 UNP_PCB_UNLOCK(unp); 803 return (0); 804} 805 806static int 807uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 808 struct mbuf *control, struct thread *td) 809{ 810 struct unpcb *unp, *unp2; 811 struct socket *so2; 812 u_int mbcnt_delta, sbcc; 813 u_int newhiwat; 814 int error = 0; 815 816 unp = sotounpcb(so); 817 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 818 819 if (flags & PRUS_OOB) { 820 error = EOPNOTSUPP; 821 goto release; 822 } 823 if (control != NULL && (error = unp_internalize(&control, td))) 824 goto release; 825 if ((nam != NULL) || (flags & PRUS_EOF)) 826 UNP_LINK_WLOCK(); 827 else 828 UNP_LINK_RLOCK(); 829 switch (so->so_type) { 830 case SOCK_DGRAM: 831 { 832 const struct sockaddr *from; 833 834 unp2 = unp->unp_conn; 835 if (nam != NULL) { 836 UNP_LINK_WLOCK_ASSERT(); 837 if (unp2 != NULL) { 838 error = EISCONN; 839 break; 840 } 841 error = unp_connect(so, nam, td); 842 if (error) 843 break; 844 unp2 = unp->unp_conn; 845 } 846 847 /* 848 * Because connect() and send() are non-atomic in a sendto() 849 * with a target address, it's possible that the socket will 850 * have disconnected before the send() can run. In that case 851 * return the slightly counter-intuitive but otherwise 852 * correct error that the socket is not connected. 853 */ 854 if (unp2 == NULL) { 855 error = ENOTCONN; 856 break; 857 } 858 /* Lockless read. */ 859 if (unp2->unp_flags & UNP_WANTCRED) 860 control = unp_addsockcred(td, control); 861 UNP_PCB_LOCK(unp); 862 if (unp->unp_addr != NULL) 863 from = (struct sockaddr *)unp->unp_addr; 864 else 865 from = &sun_noname; 866 so2 = unp2->unp_socket; 867 SOCKBUF_LOCK(&so2->so_rcv); 868 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 869 sorwakeup_locked(so2); 870 m = NULL; 871 control = NULL; 872 } else { 873 SOCKBUF_UNLOCK(&so2->so_rcv); 874 error = ENOBUFS; 875 } 876 if (nam != NULL) { 877 UNP_LINK_WLOCK_ASSERT(); 878 UNP_PCB_LOCK(unp2); 879 unp_disconnect(unp, unp2); 880 UNP_PCB_UNLOCK(unp2); 881 } 882 UNP_PCB_UNLOCK(unp); 883 break; 884 } 885 886 case SOCK_SEQPACKET: 887 case SOCK_STREAM: 888 if ((so->so_state & SS_ISCONNECTED) == 0) { 889 if (nam != NULL) { 890 UNP_LINK_WLOCK_ASSERT(); 891 error = unp_connect(so, nam, td); 892 if (error) 893 break; /* XXX */ 894 } else { 895 error = ENOTCONN; 896 break; 897 } 898 } 899 900 /* Lockless read. */ 901 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 902 error = EPIPE; 903 break; 904 } 905 906 /* 907 * Because connect() and send() are non-atomic in a sendto() 908 * with a target address, it's possible that the socket will 909 * have disconnected before the send() can run. In that case 910 * return the slightly counter-intuitive but otherwise 911 * correct error that the socket is not connected. 912 * 913 * Locking here must be done carefully: the linkage lock 914 * prevents interconnections between unpcbs from changing, so 915 * we can traverse from unp to unp2 without acquiring unp's 916 * lock. Socket buffer locks follow unpcb locks, so we can 917 * acquire both remote and lock socket buffer locks. 918 */ 919 unp2 = unp->unp_conn; 920 if (unp2 == NULL) { 921 error = ENOTCONN; 922 break; 923 } 924 so2 = unp2->unp_socket; 925 UNP_PCB_LOCK(unp2); 926 SOCKBUF_LOCK(&so2->so_rcv); 927 if (unp2->unp_flags & UNP_WANTCRED) { 928 /* 929 * Credentials are passed only once on SOCK_STREAM. 930 */ 931 unp2->unp_flags &= ~UNP_WANTCRED; 932 control = unp_addsockcred(td, control); 933 } 934 /* 935 * Send to paired receive port, and then reduce send buffer 936 * hiwater marks to maintain backpressure. Wake up readers. 937 */ 938 switch (so->so_type) { 939 case SOCK_STREAM: 940 if (control != NULL) { 941 if (sbappendcontrol_locked(&so2->so_rcv, m, 942 control)) 943 control = NULL; 944 } else 945 sbappend_locked(&so2->so_rcv, m); 946 break; 947 948 case SOCK_SEQPACKET: { 949 const struct sockaddr *from; 950 951 from = &sun_noname; 952 if (sbappendaddr_locked(&so2->so_rcv, from, m, 953 control)) 954 control = NULL; 955 break; 956 } 957 } 958 959 /* 960 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 961 * datagram size and back-pressure for SOCK_SEQPACKET, which 962 * can lead to undesired return of EMSGSIZE on send instead 963 * of more desirable blocking. 964 */ 965 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 966 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 967 sbcc = so2->so_rcv.sb_cc; 968 sorwakeup_locked(so2); 969 970 SOCKBUF_LOCK(&so->so_snd); 971 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 972 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 973 else 974 newhiwat = 0; 975 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 976 newhiwat, RLIM_INFINITY); 977 so->so_snd.sb_mbmax -= mbcnt_delta; 978 SOCKBUF_UNLOCK(&so->so_snd); 979 unp2->unp_cc = sbcc; 980 UNP_PCB_UNLOCK(unp2); 981 m = NULL; 982 break; 983 984 default: 985 panic("uipc_send unknown socktype"); 986 } 987 988 /* 989 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 990 */ 991 if (flags & PRUS_EOF) { 992 UNP_PCB_LOCK(unp); 993 socantsendmore(so); 994 unp_shutdown(unp); 995 UNP_PCB_UNLOCK(unp); 996 } 997 998 if ((nam != NULL) || (flags & PRUS_EOF)) 999 UNP_LINK_WUNLOCK(); 1000 else 1001 UNP_LINK_RUNLOCK(); 1002 1003 if (control != NULL && error != 0) 1004 unp_dispose(control); 1005 1006release: 1007 if (control != NULL) 1008 m_freem(control); 1009 if (m != NULL) 1010 m_freem(m); 1011 return (error); 1012} 1013 1014static int 1015uipc_sense(struct socket *so, struct stat *sb) 1016{ 1017 struct unpcb *unp, *unp2; 1018 struct socket *so2; 1019 1020 unp = sotounpcb(so); 1021 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1022 1023 sb->st_blksize = so->so_snd.sb_hiwat; 1024 UNP_LINK_RLOCK(); 1025 UNP_PCB_LOCK(unp); 1026 unp2 = unp->unp_conn; 1027 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1028 unp2 != NULL) { 1029 so2 = unp2->unp_socket; 1030 sb->st_blksize += so2->so_rcv.sb_cc; 1031 } 1032 sb->st_dev = NODEV; 1033 if (unp->unp_ino == 0) 1034 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1035 sb->st_ino = unp->unp_ino; 1036 UNP_PCB_UNLOCK(unp); 1037 UNP_LINK_RUNLOCK(); 1038 return (0); 1039} 1040 1041static int 1042uipc_shutdown(struct socket *so) 1043{ 1044 struct unpcb *unp; 1045 1046 unp = sotounpcb(so); 1047 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1048 1049 UNP_LINK_WLOCK(); 1050 UNP_PCB_LOCK(unp); 1051 socantsendmore(so); 1052 unp_shutdown(unp); 1053 UNP_PCB_UNLOCK(unp); 1054 UNP_LINK_WUNLOCK(); 1055 return (0); 1056} 1057 1058static int 1059uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1060{ 1061 struct unpcb *unp; 1062 const struct sockaddr *sa; 1063 1064 unp = sotounpcb(so); 1065 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1066 1067 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1068 UNP_PCB_LOCK(unp); 1069 if (unp->unp_addr != NULL) 1070 sa = (struct sockaddr *) unp->unp_addr; 1071 else 1072 sa = &sun_noname; 1073 bcopy(sa, *nam, sa->sa_len); 1074 UNP_PCB_UNLOCK(unp); 1075 return (0); 1076} 1077 1078static struct pr_usrreqs uipc_usrreqs_dgram = { 1079 .pru_abort = uipc_abort, 1080 .pru_accept = uipc_accept, 1081 .pru_attach = uipc_attach, 1082 .pru_bind = uipc_bind, 1083 .pru_connect = uipc_connect, 1084 .pru_connect2 = uipc_connect2, 1085 .pru_detach = uipc_detach, 1086 .pru_disconnect = uipc_disconnect, 1087 .pru_listen = uipc_listen, 1088 .pru_peeraddr = uipc_peeraddr, 1089 .pru_rcvd = uipc_rcvd, 1090 .pru_send = uipc_send, 1091 .pru_sense = uipc_sense, 1092 .pru_shutdown = uipc_shutdown, 1093 .pru_sockaddr = uipc_sockaddr, 1094 .pru_soreceive = soreceive_dgram, 1095 .pru_close = uipc_close, 1096}; 1097 1098static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1099 .pru_abort = uipc_abort, 1100 .pru_accept = uipc_accept, 1101 .pru_attach = uipc_attach, 1102 .pru_bind = uipc_bind, 1103 .pru_connect = uipc_connect, 1104 .pru_connect2 = uipc_connect2, 1105 .pru_detach = uipc_detach, 1106 .pru_disconnect = uipc_disconnect, 1107 .pru_listen = uipc_listen, 1108 .pru_peeraddr = uipc_peeraddr, 1109 .pru_rcvd = uipc_rcvd, 1110 .pru_send = uipc_send, 1111 .pru_sense = uipc_sense, 1112 .pru_shutdown = uipc_shutdown, 1113 .pru_sockaddr = uipc_sockaddr, 1114 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1115 .pru_close = uipc_close, 1116}; 1117 1118static struct pr_usrreqs uipc_usrreqs_stream = { 1119 .pru_abort = uipc_abort, 1120 .pru_accept = uipc_accept, 1121 .pru_attach = uipc_attach, 1122 .pru_bind = uipc_bind, 1123 .pru_connect = uipc_connect, 1124 .pru_connect2 = uipc_connect2, 1125 .pru_detach = uipc_detach, 1126 .pru_disconnect = uipc_disconnect, 1127 .pru_listen = uipc_listen, 1128 .pru_peeraddr = uipc_peeraddr, 1129 .pru_rcvd = uipc_rcvd, 1130 .pru_send = uipc_send, 1131 .pru_sense = uipc_sense, 1132 .pru_shutdown = uipc_shutdown, 1133 .pru_sockaddr = uipc_sockaddr, 1134 .pru_soreceive = soreceive_generic, 1135 .pru_close = uipc_close, 1136}; 1137 1138static int 1139uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1140{ 1141 struct unpcb *unp; 1142 struct xucred xu; 1143 int error, optval; 1144 1145 if (sopt->sopt_level != 0) 1146 return (EINVAL); 1147 1148 unp = sotounpcb(so); 1149 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1150 error = 0; 1151 switch (sopt->sopt_dir) { 1152 case SOPT_GET: 1153 switch (sopt->sopt_name) { 1154 case LOCAL_PEERCRED: 1155 UNP_PCB_LOCK(unp); 1156 if (unp->unp_flags & UNP_HAVEPC) 1157 xu = unp->unp_peercred; 1158 else { 1159 if (so->so_type == SOCK_STREAM) 1160 error = ENOTCONN; 1161 else 1162 error = EINVAL; 1163 } 1164 UNP_PCB_UNLOCK(unp); 1165 if (error == 0) 1166 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1167 break; 1168 1169 case LOCAL_CREDS: 1170 /* Unlocked read. */ 1171 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1172 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1173 break; 1174 1175 case LOCAL_CONNWAIT: 1176 /* Unlocked read. */ 1177 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1178 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1179 break; 1180 1181 default: 1182 error = EOPNOTSUPP; 1183 break; 1184 } 1185 break; 1186 1187 case SOPT_SET: 1188 switch (sopt->sopt_name) { 1189 case LOCAL_CREDS: 1190 case LOCAL_CONNWAIT: 1191 error = sooptcopyin(sopt, &optval, sizeof(optval), 1192 sizeof(optval)); 1193 if (error) 1194 break; 1195 1196#define OPTSET(bit) do { \ 1197 UNP_PCB_LOCK(unp); \ 1198 if (optval) \ 1199 unp->unp_flags |= bit; \ 1200 else \ 1201 unp->unp_flags &= ~bit; \ 1202 UNP_PCB_UNLOCK(unp); \ 1203} while (0) 1204 1205 switch (sopt->sopt_name) { 1206 case LOCAL_CREDS: 1207 OPTSET(UNP_WANTCRED); 1208 break; 1209 1210 case LOCAL_CONNWAIT: 1211 OPTSET(UNP_CONNWAIT); 1212 break; 1213 1214 default: 1215 break; 1216 } 1217 break; 1218#undef OPTSET 1219 default: 1220 error = ENOPROTOOPT; 1221 break; 1222 } 1223 break; 1224 1225 default: 1226 error = EOPNOTSUPP; 1227 break; 1228 } 1229 return (error); 1230} 1231 1232static int 1233unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1234{ 1235 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1236 struct vnode *vp; 1237 struct socket *so2, *so3; 1238 struct unpcb *unp, *unp2, *unp3; 1239 int error, len; 1240 struct nameidata nd; 1241 char buf[SOCK_MAXADDRLEN]; 1242 struct sockaddr *sa; 1243 1244 UNP_LINK_WLOCK_ASSERT(); 1245 1246 unp = sotounpcb(so); 1247 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1248 1249 if (nam->sa_len > sizeof(struct sockaddr_un)) 1250 return (EINVAL); 1251 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1252 if (len <= 0) 1253 return (EINVAL); 1254 bcopy(soun->sun_path, buf, len); 1255 buf[len] = 0; 1256 1257 UNP_PCB_LOCK(unp); 1258 if (unp->unp_flags & UNP_CONNECTING) { 1259 UNP_PCB_UNLOCK(unp); 1260 return (EALREADY); 1261 } 1262 UNP_LINK_WUNLOCK(); 1263 unp->unp_flags |= UNP_CONNECTING; 1264 UNP_PCB_UNLOCK(unp); 1265 1266 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1267 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1268 UIO_SYSSPACE, buf, td); 1269 error = namei(&nd); 1270 if (error) 1271 vp = NULL; 1272 else 1273 vp = nd.ni_vp; 1274 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1275 NDFREE(&nd, NDF_ONLY_PNBUF); 1276 if (error) 1277 goto bad; 1278 1279 if (vp->v_type != VSOCK) { 1280 error = ENOTSOCK; 1281 goto bad; 1282 } 1283#ifdef MAC 1284 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1285 if (error) 1286 goto bad; 1287#endif 1288 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1289 if (error) 1290 goto bad; 1291 1292 unp = sotounpcb(so); 1293 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1294 1295 /* 1296 * Lock linkage lock for two reasons: make sure v_socket is stable, 1297 * and to protect simultaneous locking of multiple pcbs. 1298 */ 1299 UNP_LINK_WLOCK(); 1300 VOP_UNP_CONNECT(vp, &so2); 1301 if (so2 == NULL) { 1302 error = ECONNREFUSED; 1303 goto bad2; 1304 } 1305 if (so->so_type != so2->so_type) { 1306 error = EPROTOTYPE; 1307 goto bad2; 1308 } 1309 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1310 if (so2->so_options & SO_ACCEPTCONN) { 1311 CURVNET_SET(so2->so_vnet); 1312 so3 = sonewconn(so2, 0); 1313 CURVNET_RESTORE(); 1314 } else 1315 so3 = NULL; 1316 if (so3 == NULL) { 1317 error = ECONNREFUSED; 1318 goto bad2; 1319 } 1320 unp = sotounpcb(so); 1321 unp2 = sotounpcb(so2); 1322 unp3 = sotounpcb(so3); 1323 UNP_PCB_LOCK(unp); 1324 UNP_PCB_LOCK(unp2); 1325 UNP_PCB_LOCK(unp3); 1326 if (unp2->unp_addr != NULL) { 1327 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1328 unp3->unp_addr = (struct sockaddr_un *) sa; 1329 sa = NULL; 1330 } 1331 1332 /* 1333 * The connecter's (client's) credentials are copied from its 1334 * process structure at the time of connect() (which is now). 1335 */ 1336 cru2x(td->td_ucred, &unp3->unp_peercred); 1337 unp3->unp_flags |= UNP_HAVEPC; 1338 1339 /* 1340 * The receiver's (server's) credentials are copied from the 1341 * unp_peercred member of socket on which the former called 1342 * listen(); uipc_listen() cached that process's credentials 1343 * at that time so we can use them now. 1344 */ 1345 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1346 ("unp_connect: listener without cached peercred")); 1347 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1348 sizeof(unp->unp_peercred)); 1349 unp->unp_flags |= UNP_HAVEPC; 1350 if (unp2->unp_flags & UNP_WANTCRED) 1351 unp3->unp_flags |= UNP_WANTCRED; 1352 UNP_PCB_UNLOCK(unp3); 1353 UNP_PCB_UNLOCK(unp2); 1354 UNP_PCB_UNLOCK(unp); 1355#ifdef MAC 1356 mac_socketpeer_set_from_socket(so, so3); 1357 mac_socketpeer_set_from_socket(so3, so); 1358#endif 1359 1360 so2 = so3; 1361 } 1362 unp = sotounpcb(so); 1363 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1364 unp2 = sotounpcb(so2); 1365 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1366 UNP_PCB_LOCK(unp); 1367 UNP_PCB_LOCK(unp2); 1368 error = unp_connect2(so, so2, PRU_CONNECT); 1369 UNP_PCB_UNLOCK(unp2); 1370 UNP_PCB_UNLOCK(unp); 1371bad2: 1372 UNP_LINK_WUNLOCK(); 1373bad: 1374 if (vp != NULL) 1375 vput(vp); 1376 free(sa, M_SONAME); 1377 UNP_LINK_WLOCK(); 1378 UNP_PCB_LOCK(unp); 1379 unp->unp_flags &= ~UNP_CONNECTING; 1380 UNP_PCB_UNLOCK(unp); 1381 return (error); 1382} 1383 1384static int 1385unp_connect2(struct socket *so, struct socket *so2, int req) 1386{ 1387 struct unpcb *unp; 1388 struct unpcb *unp2; 1389 1390 unp = sotounpcb(so); 1391 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1392 unp2 = sotounpcb(so2); 1393 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1394 1395 UNP_LINK_WLOCK_ASSERT(); 1396 UNP_PCB_LOCK_ASSERT(unp); 1397 UNP_PCB_LOCK_ASSERT(unp2); 1398 1399 if (so2->so_type != so->so_type) 1400 return (EPROTOTYPE); 1401 unp->unp_conn = unp2; 1402 1403 switch (so->so_type) { 1404 case SOCK_DGRAM: 1405 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1406 soisconnected(so); 1407 break; 1408 1409 case SOCK_STREAM: 1410 case SOCK_SEQPACKET: 1411 unp2->unp_conn = unp; 1412 if (req == PRU_CONNECT && 1413 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1414 soisconnecting(so); 1415 else 1416 soisconnected(so); 1417 soisconnected(so2); 1418 break; 1419 1420 default: 1421 panic("unp_connect2"); 1422 } 1423 return (0); 1424} 1425 1426static void 1427unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1428{ 1429 struct socket *so; 1430 1431 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1432 1433 UNP_LINK_WLOCK_ASSERT(); 1434 UNP_PCB_LOCK_ASSERT(unp); 1435 UNP_PCB_LOCK_ASSERT(unp2); 1436 1437 unp->unp_conn = NULL; 1438 switch (unp->unp_socket->so_type) { 1439 case SOCK_DGRAM: 1440 LIST_REMOVE(unp, unp_reflink); 1441 so = unp->unp_socket; 1442 SOCK_LOCK(so); 1443 so->so_state &= ~SS_ISCONNECTED; 1444 SOCK_UNLOCK(so); 1445 break; 1446 1447 case SOCK_STREAM: 1448 case SOCK_SEQPACKET: 1449 soisdisconnected(unp->unp_socket); 1450 unp2->unp_conn = NULL; 1451 soisdisconnected(unp2->unp_socket); 1452 break; 1453 } 1454} 1455 1456/* 1457 * unp_pcblist() walks the global list of struct unpcb's to generate a 1458 * pointer list, bumping the refcount on each unpcb. It then copies them out 1459 * sequentially, validating the generation number on each to see if it has 1460 * been detached. All of this is necessary because copyout() may sleep on 1461 * disk I/O. 1462 */ 1463static int 1464unp_pcblist(SYSCTL_HANDLER_ARGS) 1465{ 1466 int error, i, n; 1467 int freeunp; 1468 struct unpcb *unp, **unp_list; 1469 unp_gen_t gencnt; 1470 struct xunpgen *xug; 1471 struct unp_head *head; 1472 struct xunpcb *xu; 1473 1474 switch ((intptr_t)arg1) { 1475 case SOCK_STREAM: 1476 head = &unp_shead; 1477 break; 1478 1479 case SOCK_DGRAM: 1480 head = &unp_dhead; 1481 break; 1482 1483 case SOCK_SEQPACKET: 1484 head = &unp_sphead; 1485 break; 1486 1487 default: 1488 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1489 } 1490 1491 /* 1492 * The process of preparing the PCB list is too time-consuming and 1493 * resource-intensive to repeat twice on every request. 1494 */ 1495 if (req->oldptr == NULL) { 1496 n = unp_count; 1497 req->oldidx = 2 * (sizeof *xug) 1498 + (n + n/8) * sizeof(struct xunpcb); 1499 return (0); 1500 } 1501 1502 if (req->newptr != NULL) 1503 return (EPERM); 1504 1505 /* 1506 * OK, now we're committed to doing something. 1507 */ 1508 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1509 UNP_LIST_LOCK(); 1510 gencnt = unp_gencnt; 1511 n = unp_count; 1512 UNP_LIST_UNLOCK(); 1513 1514 xug->xug_len = sizeof *xug; 1515 xug->xug_count = n; 1516 xug->xug_gen = gencnt; 1517 xug->xug_sogen = so_gencnt; 1518 error = SYSCTL_OUT(req, xug, sizeof *xug); 1519 if (error) { 1520 free(xug, M_TEMP); 1521 return (error); 1522 } 1523 1524 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1525 1526 UNP_LIST_LOCK(); 1527 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1528 unp = LIST_NEXT(unp, unp_link)) { 1529 UNP_PCB_LOCK(unp); 1530 if (unp->unp_gencnt <= gencnt) { 1531 if (cr_cansee(req->td->td_ucred, 1532 unp->unp_socket->so_cred)) { 1533 UNP_PCB_UNLOCK(unp); 1534 continue; 1535 } 1536 unp_list[i++] = unp; 1537 unp->unp_refcount++; 1538 } 1539 UNP_PCB_UNLOCK(unp); 1540 } 1541 UNP_LIST_UNLOCK(); 1542 n = i; /* In case we lost some during malloc. */ 1543 1544 error = 0; 1545 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1546 for (i = 0; i < n; i++) { 1547 unp = unp_list[i]; 1548 UNP_PCB_LOCK(unp); 1549 unp->unp_refcount--; 1550 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1551 xu->xu_len = sizeof *xu; 1552 xu->xu_unpp = unp; 1553 /* 1554 * XXX - need more locking here to protect against 1555 * connect/disconnect races for SMP. 1556 */ 1557 if (unp->unp_addr != NULL) 1558 bcopy(unp->unp_addr, &xu->xu_addr, 1559 unp->unp_addr->sun_len); 1560 if (unp->unp_conn != NULL && 1561 unp->unp_conn->unp_addr != NULL) 1562 bcopy(unp->unp_conn->unp_addr, 1563 &xu->xu_caddr, 1564 unp->unp_conn->unp_addr->sun_len); 1565 bcopy(unp, &xu->xu_unp, sizeof *unp); 1566 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1567 UNP_PCB_UNLOCK(unp); 1568 error = SYSCTL_OUT(req, xu, sizeof *xu); 1569 } else { 1570 freeunp = (unp->unp_refcount == 0); 1571 UNP_PCB_UNLOCK(unp); 1572 if (freeunp) { 1573 UNP_PCB_LOCK_DESTROY(unp); 1574 uma_zfree(unp_zone, unp); 1575 } 1576 } 1577 } 1578 free(xu, M_TEMP); 1579 if (!error) { 1580 /* 1581 * Give the user an updated idea of our state. If the 1582 * generation differs from what we told her before, she knows 1583 * that something happened while we were processing this 1584 * request, and it might be necessary to retry. 1585 */ 1586 xug->xug_gen = unp_gencnt; 1587 xug->xug_sogen = so_gencnt; 1588 xug->xug_count = unp_count; 1589 error = SYSCTL_OUT(req, xug, sizeof *xug); 1590 } 1591 free(unp_list, M_TEMP); 1592 free(xug, M_TEMP); 1593 return (error); 1594} 1595 1596SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1597 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1598 "List of active local datagram sockets"); 1599SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1600 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1601 "List of active local stream sockets"); 1602SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1603 CTLTYPE_OPAQUE | CTLFLAG_RD, 1604 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1605 "List of active local seqpacket sockets"); 1606 1607static void 1608unp_shutdown(struct unpcb *unp) 1609{ 1610 struct unpcb *unp2; 1611 struct socket *so; 1612 1613 UNP_LINK_WLOCK_ASSERT(); 1614 UNP_PCB_LOCK_ASSERT(unp); 1615 1616 unp2 = unp->unp_conn; 1617 if ((unp->unp_socket->so_type == SOCK_STREAM || 1618 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1619 so = unp2->unp_socket; 1620 if (so != NULL) 1621 socantrcvmore(so); 1622 } 1623} 1624 1625static void 1626unp_drop(struct unpcb *unp, int errno) 1627{ 1628 struct socket *so = unp->unp_socket; 1629 struct unpcb *unp2; 1630 1631 UNP_LINK_WLOCK_ASSERT(); 1632 UNP_PCB_LOCK_ASSERT(unp); 1633 1634 so->so_error = errno; 1635 unp2 = unp->unp_conn; 1636 if (unp2 == NULL) 1637 return; 1638 UNP_PCB_LOCK(unp2); 1639 unp_disconnect(unp, unp2); 1640 UNP_PCB_UNLOCK(unp2); 1641} 1642 1643static void 1644unp_freerights(struct file **rp, int fdcount) 1645{ 1646 int i; 1647 struct file *fp; 1648 1649 for (i = 0; i < fdcount; i++) { 1650 fp = *rp; 1651 *rp++ = NULL; 1652 unp_discard(fp); 1653 } 1654} 1655 1656static int 1657unp_externalize(struct mbuf *control, struct mbuf **controlp) 1658{ 1659 struct thread *td = curthread; /* XXX */ 1660 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1661 int i; 1662 int *fdp; 1663 struct file **rp; 1664 struct file *fp; 1665 void *data; 1666 socklen_t clen = control->m_len, datalen; 1667 int error, newfds; 1668 int f; 1669 u_int newlen; 1670 1671 UNP_LINK_UNLOCK_ASSERT(); 1672 1673 error = 0; 1674 if (controlp != NULL) /* controlp == NULL => free control messages */ 1675 *controlp = NULL; 1676 while (cm != NULL) { 1677 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1678 error = EINVAL; 1679 break; 1680 } 1681 data = CMSG_DATA(cm); 1682 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1683 if (cm->cmsg_level == SOL_SOCKET 1684 && cm->cmsg_type == SCM_RIGHTS) { 1685 newfds = datalen / sizeof(struct file *); 1686 rp = data; 1687 1688 /* If we're not outputting the descriptors free them. */ 1689 if (error || controlp == NULL) { 1690 unp_freerights(rp, newfds); 1691 goto next; 1692 } 1693 FILEDESC_XLOCK(td->td_proc->p_fd); 1694 /* if the new FD's will not fit free them. */ 1695 if (!fdavail(td, newfds)) { 1696 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1697 error = EMSGSIZE; 1698 unp_freerights(rp, newfds); 1699 goto next; 1700 } 1701 1702 /* 1703 * Now change each pointer to an fd in the global 1704 * table to an integer that is the index to the local 1705 * fd table entry that we set up to point to the 1706 * global one we are transferring. 1707 */ 1708 newlen = newfds * sizeof(int); 1709 *controlp = sbcreatecontrol(NULL, newlen, 1710 SCM_RIGHTS, SOL_SOCKET); 1711 if (*controlp == NULL) { 1712 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1713 error = E2BIG; 1714 unp_freerights(rp, newfds); 1715 goto next; 1716 } 1717 1718 fdp = (int *) 1719 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1720 for (i = 0; i < newfds; i++) { 1721 if (fdalloc(td, 0, &f)) 1722 panic("unp_externalize fdalloc failed"); 1723 fp = *rp++; 1724 td->td_proc->p_fd->fd_ofiles[f] = fp; 1725 unp_externalize_fp(fp); 1726 *fdp++ = f; 1727 } 1728 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1729 } else { 1730 /* We can just copy anything else across. */ 1731 if (error || controlp == NULL) 1732 goto next; 1733 *controlp = sbcreatecontrol(NULL, datalen, 1734 cm->cmsg_type, cm->cmsg_level); 1735 if (*controlp == NULL) { 1736 error = ENOBUFS; 1737 goto next; 1738 } 1739 bcopy(data, 1740 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1741 datalen); 1742 } 1743 controlp = &(*controlp)->m_next; 1744 1745next: 1746 if (CMSG_SPACE(datalen) < clen) { 1747 clen -= CMSG_SPACE(datalen); 1748 cm = (struct cmsghdr *) 1749 ((caddr_t)cm + CMSG_SPACE(datalen)); 1750 } else { 1751 clen = 0; 1752 cm = NULL; 1753 } 1754 } 1755 1756 m_freem(control); 1757 return (error); 1758} 1759 1760static void 1761unp_zone_change(void *tag) 1762{ 1763 1764 uma_zone_set_max(unp_zone, maxsockets); 1765} 1766 1767static void 1768unp_init(void) 1769{ 1770 1771#ifdef VIMAGE 1772 if (!IS_DEFAULT_VNET(curvnet)) 1773 return; 1774#endif 1775 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1776 NULL, NULL, UMA_ALIGN_PTR, 0); 1777 if (unp_zone == NULL) 1778 panic("unp_init"); 1779 uma_zone_set_max(unp_zone, maxsockets); 1780 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1781 NULL, EVENTHANDLER_PRI_ANY); 1782 LIST_INIT(&unp_dhead); 1783 LIST_INIT(&unp_shead); 1784 LIST_INIT(&unp_sphead); 1785 SLIST_INIT(&unp_defers); 1786 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1787 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1788 UNP_LINK_LOCK_INIT(); 1789 UNP_LIST_LOCK_INIT(); 1790 UNP_DEFERRED_LOCK_INIT(); 1791} 1792 1793static int 1794unp_internalize(struct mbuf **controlp, struct thread *td) 1795{ 1796 struct mbuf *control = *controlp; 1797 struct proc *p = td->td_proc; 1798 struct filedesc *fdescp = p->p_fd; 1799 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1800 struct cmsgcred *cmcred; 1801 struct file **rp; 1802 struct file *fp; 1803 struct timeval *tv; 1804 int i, fd, *fdp; 1805 void *data; 1806 socklen_t clen = control->m_len, datalen; 1807 int error, oldfds; 1808 u_int newlen; 1809 1810 UNP_LINK_UNLOCK_ASSERT(); 1811 1812 error = 0; 1813 *controlp = NULL; 1814 while (cm != NULL) { 1815 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1816 || cm->cmsg_len > clen) { 1817 error = EINVAL; 1818 goto out; 1819 } 1820 data = CMSG_DATA(cm); 1821 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1822 1823 switch (cm->cmsg_type) { 1824 /* 1825 * Fill in credential information. 1826 */ 1827 case SCM_CREDS: 1828 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1829 SCM_CREDS, SOL_SOCKET); 1830 if (*controlp == NULL) { 1831 error = ENOBUFS; 1832 goto out; 1833 } 1834 cmcred = (struct cmsgcred *) 1835 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1836 cmcred->cmcred_pid = p->p_pid; 1837 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1838 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1839 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1840 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1841 CMGROUP_MAX); 1842 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1843 cmcred->cmcred_groups[i] = 1844 td->td_ucred->cr_groups[i]; 1845 break; 1846 1847 case SCM_RIGHTS: 1848 oldfds = datalen / sizeof (int); 1849 /* 1850 * Check that all the FDs passed in refer to legal 1851 * files. If not, reject the entire operation. 1852 */ 1853 fdp = data; 1854 FILEDESC_SLOCK(fdescp); 1855 for (i = 0; i < oldfds; i++) { 1856 fd = *fdp++; 1857 if (fd < 0 || fd >= fdescp->fd_nfiles || 1858 fdescp->fd_ofiles[fd] == NULL) { 1859 FILEDESC_SUNLOCK(fdescp); 1860 error = EBADF; 1861 goto out; 1862 } 1863 fp = fdescp->fd_ofiles[fd]; 1864 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1865 FILEDESC_SUNLOCK(fdescp); 1866 error = EOPNOTSUPP; 1867 goto out; 1868 } 1869 1870 } 1871 1872 /* 1873 * Now replace the integer FDs with pointers to the 1874 * associated global file table entry.. 1875 */ 1876 newlen = oldfds * sizeof(struct file *); 1877 *controlp = sbcreatecontrol(NULL, newlen, 1878 SCM_RIGHTS, SOL_SOCKET); 1879 if (*controlp == NULL) { 1880 FILEDESC_SUNLOCK(fdescp); 1881 error = E2BIG; 1882 goto out; 1883 } 1884 fdp = data; 1885 rp = (struct file **) 1886 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1887 for (i = 0; i < oldfds; i++) { 1888 fp = fdescp->fd_ofiles[*fdp++]; 1889 *rp++ = fp; 1890 unp_internalize_fp(fp); 1891 } 1892 FILEDESC_SUNLOCK(fdescp); 1893 break; 1894 1895 case SCM_TIMESTAMP: 1896 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1897 SCM_TIMESTAMP, SOL_SOCKET); 1898 if (*controlp == NULL) { 1899 error = ENOBUFS; 1900 goto out; 1901 } 1902 tv = (struct timeval *) 1903 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1904 microtime(tv); 1905 break; 1906 1907 default: 1908 error = EINVAL; 1909 goto out; 1910 } 1911 1912 controlp = &(*controlp)->m_next; 1913 if (CMSG_SPACE(datalen) < clen) { 1914 clen -= CMSG_SPACE(datalen); 1915 cm = (struct cmsghdr *) 1916 ((caddr_t)cm + CMSG_SPACE(datalen)); 1917 } else { 1918 clen = 0; 1919 cm = NULL; 1920 } 1921 } 1922 1923out: 1924 m_freem(control); 1925 return (error); 1926} 1927 1928static struct mbuf * 1929unp_addsockcred(struct thread *td, struct mbuf *control) 1930{ 1931 struct mbuf *m, *n, *n_prev; 1932 struct sockcred *sc; 1933 const struct cmsghdr *cm; 1934 int ngroups; 1935 int i; 1936 1937 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1938 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1939 if (m == NULL) 1940 return (control); 1941 1942 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1943 sc->sc_uid = td->td_ucred->cr_ruid; 1944 sc->sc_euid = td->td_ucred->cr_uid; 1945 sc->sc_gid = td->td_ucred->cr_rgid; 1946 sc->sc_egid = td->td_ucred->cr_gid; 1947 sc->sc_ngroups = ngroups; 1948 for (i = 0; i < sc->sc_ngroups; i++) 1949 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1950 1951 /* 1952 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1953 * created SCM_CREDS control message (struct sockcred) has another 1954 * format. 1955 */ 1956 if (control != NULL) 1957 for (n = control, n_prev = NULL; n != NULL;) { 1958 cm = mtod(n, struct cmsghdr *); 1959 if (cm->cmsg_level == SOL_SOCKET && 1960 cm->cmsg_type == SCM_CREDS) { 1961 if (n_prev == NULL) 1962 control = n->m_next; 1963 else 1964 n_prev->m_next = n->m_next; 1965 n = m_free(n); 1966 } else { 1967 n_prev = n; 1968 n = n->m_next; 1969 } 1970 } 1971 1972 /* Prepend it to the head. */ 1973 m->m_next = control; 1974 return (m); 1975} 1976 1977static struct unpcb * 1978fptounp(struct file *fp) 1979{ 1980 struct socket *so; 1981 1982 if (fp->f_type != DTYPE_SOCKET) 1983 return (NULL); 1984 if ((so = fp->f_data) == NULL) 1985 return (NULL); 1986 if (so->so_proto->pr_domain != &localdomain) 1987 return (NULL); 1988 return sotounpcb(so); 1989} 1990 1991static void 1992unp_discard(struct file *fp) 1993{ 1994 struct unp_defer *dr; 1995 1996 if (unp_externalize_fp(fp)) { 1997 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 1998 dr->ud_fp = fp; 1999 UNP_DEFERRED_LOCK(); 2000 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2001 UNP_DEFERRED_UNLOCK(); 2002 atomic_add_int(&unp_defers_count, 1); 2003 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2004 } else 2005 (void) closef(fp, (struct thread *)NULL); 2006} 2007 2008static void 2009unp_process_defers(void *arg __unused, int pending) 2010{ 2011 struct unp_defer *dr; 2012 SLIST_HEAD(, unp_defer) drl; 2013 int count; 2014 2015 SLIST_INIT(&drl); 2016 for (;;) { 2017 UNP_DEFERRED_LOCK(); 2018 if (SLIST_FIRST(&unp_defers) == NULL) { 2019 UNP_DEFERRED_UNLOCK(); 2020 break; 2021 } 2022 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2023 UNP_DEFERRED_UNLOCK(); 2024 count = 0; 2025 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2026 SLIST_REMOVE_HEAD(&drl, ud_link); 2027 closef(dr->ud_fp, NULL); 2028 free(dr, M_TEMP); 2029 count++; 2030 } 2031 atomic_add_int(&unp_defers_count, -count); 2032 } 2033} 2034 2035static void 2036unp_internalize_fp(struct file *fp) 2037{ 2038 struct unpcb *unp; 2039 2040 UNP_LINK_WLOCK(); 2041 if ((unp = fptounp(fp)) != NULL) { 2042 unp->unp_file = fp; 2043 unp->unp_msgcount++; 2044 } 2045 fhold(fp); 2046 unp_rights++; 2047 UNP_LINK_WUNLOCK(); 2048} 2049 2050static int 2051unp_externalize_fp(struct file *fp) 2052{ 2053 struct unpcb *unp; 2054 int ret; 2055 2056 UNP_LINK_WLOCK(); 2057 if ((unp = fptounp(fp)) != NULL) { 2058 unp->unp_msgcount--; 2059 ret = 1; 2060 } else 2061 ret = 0; 2062 unp_rights--; 2063 UNP_LINK_WUNLOCK(); 2064 return (ret); 2065} 2066 2067/* 2068 * unp_defer indicates whether additional work has been defered for a future 2069 * pass through unp_gc(). It is thread local and does not require explicit 2070 * synchronization. 2071 */ 2072static int unp_marked; 2073static int unp_unreachable; 2074 2075static void 2076unp_accessable(struct file *fp) 2077{ 2078 struct unpcb *unp; 2079 2080 if ((unp = fptounp(fp)) == NULL) 2081 return; 2082 if (unp->unp_gcflag & UNPGC_REF) 2083 return; 2084 unp->unp_gcflag &= ~UNPGC_DEAD; 2085 unp->unp_gcflag |= UNPGC_REF; 2086 unp_marked++; 2087} 2088 2089static void 2090unp_gc_process(struct unpcb *unp) 2091{ 2092 struct socket *soa; 2093 struct socket *so; 2094 struct file *fp; 2095 2096 /* Already processed. */ 2097 if (unp->unp_gcflag & UNPGC_SCANNED) 2098 return; 2099 fp = unp->unp_file; 2100 2101 /* 2102 * Check for a socket potentially in a cycle. It must be in a 2103 * queue as indicated by msgcount, and this must equal the file 2104 * reference count. Note that when msgcount is 0 the file is NULL. 2105 */ 2106 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2107 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2108 unp->unp_gcflag |= UNPGC_DEAD; 2109 unp_unreachable++; 2110 return; 2111 } 2112 2113 /* 2114 * Mark all sockets we reference with RIGHTS. 2115 */ 2116 so = unp->unp_socket; 2117 SOCKBUF_LOCK(&so->so_rcv); 2118 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2119 SOCKBUF_UNLOCK(&so->so_rcv); 2120 2121 /* 2122 * Mark all sockets in our accept queue. 2123 */ 2124 ACCEPT_LOCK(); 2125 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2126 SOCKBUF_LOCK(&soa->so_rcv); 2127 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2128 SOCKBUF_UNLOCK(&soa->so_rcv); 2129 } 2130 ACCEPT_UNLOCK(); 2131 unp->unp_gcflag |= UNPGC_SCANNED; 2132} 2133 2134static int unp_recycled; 2135SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2136 "Number of unreachable sockets claimed by the garbage collector."); 2137 2138static int unp_taskcount; 2139SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2140 "Number of times the garbage collector has run."); 2141 2142static void 2143unp_gc(__unused void *arg, int pending) 2144{ 2145 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2146 NULL }; 2147 struct unp_head **head; 2148 struct file *f, **unref; 2149 struct unpcb *unp; 2150 int i, total; 2151 2152 unp_taskcount++; 2153 UNP_LIST_LOCK(); 2154 /* 2155 * First clear all gc flags from previous runs. 2156 */ 2157 for (head = heads; *head != NULL; head++) 2158 LIST_FOREACH(unp, *head, unp_link) 2159 unp->unp_gcflag = 0; 2160 2161 /* 2162 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2163 * is reachable all of the sockets it references are reachable. 2164 * Stop the scan once we do a complete loop without discovering 2165 * a new reachable socket. 2166 */ 2167 do { 2168 unp_unreachable = 0; 2169 unp_marked = 0; 2170 for (head = heads; *head != NULL; head++) 2171 LIST_FOREACH(unp, *head, unp_link) 2172 unp_gc_process(unp); 2173 } while (unp_marked); 2174 UNP_LIST_UNLOCK(); 2175 if (unp_unreachable == 0) 2176 return; 2177 2178 /* 2179 * Allocate space for a local list of dead unpcbs. 2180 */ 2181 unref = malloc(unp_unreachable * sizeof(struct file *), 2182 M_TEMP, M_WAITOK); 2183 2184 /* 2185 * Iterate looking for sockets which have been specifically marked 2186 * as as unreachable and store them locally. 2187 */ 2188 UNP_LINK_RLOCK(); 2189 UNP_LIST_LOCK(); 2190 for (total = 0, head = heads; *head != NULL; head++) 2191 LIST_FOREACH(unp, *head, unp_link) 2192 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2193 f = unp->unp_file; 2194 if (unp->unp_msgcount == 0 || f == NULL || 2195 f->f_count != unp->unp_msgcount) 2196 continue; 2197 unref[total++] = f; 2198 fhold(f); 2199 KASSERT(total <= unp_unreachable, 2200 ("unp_gc: incorrect unreachable count.")); 2201 } 2202 UNP_LIST_UNLOCK(); 2203 UNP_LINK_RUNLOCK(); 2204 2205 /* 2206 * Now flush all sockets, free'ing rights. This will free the 2207 * struct files associated with these sockets but leave each socket 2208 * with one remaining ref. 2209 */ 2210 for (i = 0; i < total; i++) { 2211 struct socket *so; 2212 2213 so = unref[i]->f_data; 2214 CURVNET_SET(so->so_vnet); 2215 sorflush(so); 2216 CURVNET_RESTORE(); 2217 } 2218 2219 /* 2220 * And finally release the sockets so they can be reclaimed. 2221 */ 2222 for (i = 0; i < total; i++) 2223 fdrop(unref[i], NULL); 2224 unp_recycled += total; 2225 free(unref, M_TEMP); 2226} 2227 2228static void 2229unp_dispose(struct mbuf *m) 2230{ 2231 2232 if (m) 2233 unp_scan(m, unp_discard); 2234} 2235 2236static void 2237unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2238{ 2239 struct mbuf *m; 2240 struct file **rp; 2241 struct cmsghdr *cm; 2242 void *data; 2243 int i; 2244 socklen_t clen, datalen; 2245 int qfds; 2246 2247 while (m0 != NULL) { 2248 for (m = m0; m; m = m->m_next) { 2249 if (m->m_type != MT_CONTROL) 2250 continue; 2251 2252 cm = mtod(m, struct cmsghdr *); 2253 clen = m->m_len; 2254 2255 while (cm != NULL) { 2256 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2257 break; 2258 2259 data = CMSG_DATA(cm); 2260 datalen = (caddr_t)cm + cm->cmsg_len 2261 - (caddr_t)data; 2262 2263 if (cm->cmsg_level == SOL_SOCKET && 2264 cm->cmsg_type == SCM_RIGHTS) { 2265 qfds = datalen / sizeof (struct file *); 2266 rp = data; 2267 for (i = 0; i < qfds; i++) 2268 (*op)(*rp++); 2269 } 2270 2271 if (CMSG_SPACE(datalen) < clen) { 2272 clen -= CMSG_SPACE(datalen); 2273 cm = (struct cmsghdr *) 2274 ((caddr_t)cm + CMSG_SPACE(datalen)); 2275 } else { 2276 clen = 0; 2277 cm = NULL; 2278 } 2279 } 2280 } 2281 m0 = m0->m_act; 2282 } 2283} 2284 2285/* 2286 * A helper function called by VFS before socket-type vnode reclamation. 2287 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2288 * use count. 2289 */ 2290void 2291vfs_unp_reclaim(struct vnode *vp) 2292{ 2293 struct socket *so; 2294 struct unpcb *unp; 2295 int active; 2296 2297 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2298 KASSERT(vp->v_type == VSOCK, 2299 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2300 2301 active = 0; 2302 UNP_LINK_WLOCK(); 2303 VOP_UNP_CONNECT(vp, &so); 2304 if (so == NULL) 2305 goto done; 2306 unp = sotounpcb(so); 2307 if (unp == NULL) 2308 goto done; 2309 UNP_PCB_LOCK(unp); 2310 if (unp->unp_vnode == vp) { 2311 VOP_UNP_DETACH(vp); 2312 unp->unp_vnode = NULL; 2313 active = 1; 2314 } 2315 UNP_PCB_UNLOCK(unp); 2316done: 2317 UNP_LINK_WUNLOCK(); 2318 if (active) 2319 vunref(vp); 2320} 2321 2322#ifdef DDB 2323static void 2324db_print_indent(int indent) 2325{ 2326 int i; 2327 2328 for (i = 0; i < indent; i++) 2329 db_printf(" "); 2330} 2331 2332static void 2333db_print_unpflags(int unp_flags) 2334{ 2335 int comma; 2336 2337 comma = 0; 2338 if (unp_flags & UNP_HAVEPC) { 2339 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2340 comma = 1; 2341 } 2342 if (unp_flags & UNP_HAVEPCCACHED) { 2343 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2344 comma = 1; 2345 } 2346 if (unp_flags & UNP_WANTCRED) { 2347 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2348 comma = 1; 2349 } 2350 if (unp_flags & UNP_CONNWAIT) { 2351 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2352 comma = 1; 2353 } 2354 if (unp_flags & UNP_CONNECTING) { 2355 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2356 comma = 1; 2357 } 2358 if (unp_flags & UNP_BINDING) { 2359 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2360 comma = 1; 2361 } 2362} 2363 2364static void 2365db_print_xucred(int indent, struct xucred *xu) 2366{ 2367 int comma, i; 2368 2369 db_print_indent(indent); 2370 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2371 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2372 db_print_indent(indent); 2373 db_printf("cr_groups: "); 2374 comma = 0; 2375 for (i = 0; i < xu->cr_ngroups; i++) { 2376 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2377 comma = 1; 2378 } 2379 db_printf("\n"); 2380} 2381 2382static void 2383db_print_unprefs(int indent, struct unp_head *uh) 2384{ 2385 struct unpcb *unp; 2386 int counter; 2387 2388 counter = 0; 2389 LIST_FOREACH(unp, uh, unp_reflink) { 2390 if (counter % 4 == 0) 2391 db_print_indent(indent); 2392 db_printf("%p ", unp); 2393 if (counter % 4 == 3) 2394 db_printf("\n"); 2395 counter++; 2396 } 2397 if (counter != 0 && counter % 4 != 0) 2398 db_printf("\n"); 2399} 2400 2401DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2402{ 2403 struct unpcb *unp; 2404 2405 if (!have_addr) { 2406 db_printf("usage: show unpcb <addr>\n"); 2407 return; 2408 } 2409 unp = (struct unpcb *)addr; 2410 2411 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2412 unp->unp_vnode); 2413 2414 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 2415 unp->unp_conn); 2416 2417 db_printf("unp_refs:\n"); 2418 db_print_unprefs(2, &unp->unp_refs); 2419 2420 /* XXXRW: Would be nice to print the full address, if any. */ 2421 db_printf("unp_addr: %p\n", unp->unp_addr); 2422 2423 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2424 unp->unp_cc, unp->unp_mbcnt, 2425 (unsigned long long)unp->unp_gencnt); 2426 2427 db_printf("unp_flags: %x (", unp->unp_flags); 2428 db_print_unpflags(unp->unp_flags); 2429 db_printf(")\n"); 2430 2431 db_printf("unp_peercred:\n"); 2432 db_print_xucred(2, &unp->unp_peercred); 2433 2434 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2435} 2436#endif 2437