uipc_usrreq.c revision 232152
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 232152 2012-02-25 10:15:41Z trociny $"); 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_usrreqs = &uipc_usrreqs_dgram 310}, 311{ 312 .pr_type = SOCK_SEQPACKET, 313 .pr_domain = &localdomain, 314 315 /* 316 * XXXRW: For now, PR_ADDR because soreceive will bump into them 317 * due to our use of sbappendaddr. A new sbappend variants is needed 318 * that supports both atomic record writes and control data. 319 */ 320 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 321 PR_RIGHTS, 322 .pr_usrreqs = &uipc_usrreqs_seqpacket, 323}, 324}; 325 326static struct domain localdomain = { 327 .dom_family = AF_LOCAL, 328 .dom_name = "local", 329 .dom_init = unp_init, 330 .dom_externalize = unp_externalize, 331 .dom_dispose = unp_dispose, 332 .dom_protosw = localsw, 333 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 334}; 335DOMAIN_SET(local); 336 337static void 338uipc_abort(struct socket *so) 339{ 340 struct unpcb *unp, *unp2; 341 342 unp = sotounpcb(so); 343 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 344 345 UNP_LINK_WLOCK(); 346 UNP_PCB_LOCK(unp); 347 unp2 = unp->unp_conn; 348 if (unp2 != NULL) { 349 UNP_PCB_LOCK(unp2); 350 unp_drop(unp2, ECONNABORTED); 351 UNP_PCB_UNLOCK(unp2); 352 } 353 UNP_PCB_UNLOCK(unp); 354 UNP_LINK_WUNLOCK(); 355} 356 357static int 358uipc_accept(struct socket *so, struct sockaddr **nam) 359{ 360 struct unpcb *unp, *unp2; 361 const struct sockaddr *sa; 362 363 /* 364 * Pass back name of connected socket, if it was bound and we are 365 * still connected (our peer may have closed already!). 366 */ 367 unp = sotounpcb(so); 368 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 369 370 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 371 UNP_LINK_RLOCK(); 372 unp2 = unp->unp_conn; 373 if (unp2 != NULL && unp2->unp_addr != NULL) { 374 UNP_PCB_LOCK(unp2); 375 sa = (struct sockaddr *) unp2->unp_addr; 376 bcopy(sa, *nam, sa->sa_len); 377 UNP_PCB_UNLOCK(unp2); 378 } else { 379 sa = &sun_noname; 380 bcopy(sa, *nam, sa->sa_len); 381 } 382 UNP_LINK_RUNLOCK(); 383 return (0); 384} 385 386static int 387uipc_attach(struct socket *so, int proto, struct thread *td) 388{ 389 u_long sendspace, recvspace; 390 struct unpcb *unp; 391 int error; 392 393 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 394 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 395 switch (so->so_type) { 396 case SOCK_STREAM: 397 sendspace = unpst_sendspace; 398 recvspace = unpst_recvspace; 399 break; 400 401 case SOCK_DGRAM: 402 sendspace = unpdg_sendspace; 403 recvspace = unpdg_recvspace; 404 break; 405 406 case SOCK_SEQPACKET: 407 sendspace = unpsp_sendspace; 408 recvspace = unpsp_recvspace; 409 break; 410 411 default: 412 panic("uipc_attach"); 413 } 414 error = soreserve(so, sendspace, recvspace); 415 if (error) 416 return (error); 417 } 418 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 419 if (unp == NULL) 420 return (ENOBUFS); 421 LIST_INIT(&unp->unp_refs); 422 UNP_PCB_LOCK_INIT(unp); 423 unp->unp_socket = so; 424 so->so_pcb = unp; 425 unp->unp_refcount = 1; 426 427 UNP_LIST_LOCK(); 428 unp->unp_gencnt = ++unp_gencnt; 429 unp_count++; 430 switch (so->so_type) { 431 case SOCK_STREAM: 432 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 433 break; 434 435 case SOCK_DGRAM: 436 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 437 break; 438 439 case SOCK_SEQPACKET: 440 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 441 break; 442 443 default: 444 panic("uipc_attach"); 445 } 446 UNP_LIST_UNLOCK(); 447 448 return (0); 449} 450 451static int 452uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 453{ 454 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 455 struct vattr vattr; 456 int error, namelen, vfslocked; 457 struct nameidata nd; 458 struct unpcb *unp; 459 struct vnode *vp; 460 struct mount *mp; 461 char *buf; 462 463 unp = sotounpcb(so); 464 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 465 466 if (soun->sun_len > sizeof(struct sockaddr_un)) 467 return (EINVAL); 468 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 469 if (namelen <= 0) 470 return (EINVAL); 471 472 /* 473 * We don't allow simultaneous bind() calls on a single UNIX domain 474 * socket, so flag in-progress operations, and return an error if an 475 * operation is already in progress. 476 * 477 * Historically, we have not allowed a socket to be rebound, so this 478 * also returns an error. Not allowing re-binding simplifies the 479 * implementation and avoids a great many possible failure modes. 480 */ 481 UNP_PCB_LOCK(unp); 482 if (unp->unp_vnode != NULL) { 483 UNP_PCB_UNLOCK(unp); 484 return (EINVAL); 485 } 486 if (unp->unp_flags & UNP_BINDING) { 487 UNP_PCB_UNLOCK(unp); 488 return (EALREADY); 489 } 490 unp->unp_flags |= UNP_BINDING; 491 UNP_PCB_UNLOCK(unp); 492 493 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 494 bcopy(soun->sun_path, buf, namelen); 495 buf[namelen] = 0; 496 497restart: 498 vfslocked = 0; 499 NDINIT(&nd, CREATE, MPSAFE | 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 vfslocked = NDHASGIANT(&nd); 507 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 508 NDFREE(&nd, NDF_ONLY_PNBUF); 509 if (nd.ni_dvp == vp) 510 vrele(nd.ni_dvp); 511 else 512 vput(nd.ni_dvp); 513 if (vp != NULL) { 514 vrele(vp); 515 error = EADDRINUSE; 516 goto error; 517 } 518 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 519 if (error) 520 goto error; 521 VFS_UNLOCK_GIANT(vfslocked); 522 goto restart; 523 } 524 VATTR_NULL(&vattr); 525 vattr.va_type = VSOCK; 526 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 527#ifdef MAC 528 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 529 &vattr); 530#endif 531 if (error == 0) 532 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 533 NDFREE(&nd, NDF_ONLY_PNBUF); 534 vput(nd.ni_dvp); 535 if (error) { 536 vn_finished_write(mp); 537 goto error; 538 } 539 vp = nd.ni_vp; 540 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 541 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 542 543 UNP_LINK_WLOCK(); 544 UNP_PCB_LOCK(unp); 545 vp->v_socket = unp->unp_socket; 546 unp->unp_vnode = vp; 547 unp->unp_addr = soun; 548 unp->unp_flags &= ~UNP_BINDING; 549 UNP_PCB_UNLOCK(unp); 550 UNP_LINK_WUNLOCK(); 551 VOP_UNLOCK(vp, 0); 552 vn_finished_write(mp); 553 VFS_UNLOCK_GIANT(vfslocked); 554 free(buf, M_TEMP); 555 return (0); 556 557error: 558 VFS_UNLOCK_GIANT(vfslocked); 559 UNP_PCB_LOCK(unp); 560 unp->unp_flags &= ~UNP_BINDING; 561 UNP_PCB_UNLOCK(unp); 562 free(buf, M_TEMP); 563 return (error); 564} 565 566static int 567uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 568{ 569 int error; 570 571 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 572 UNP_LINK_WLOCK(); 573 error = unp_connect(so, nam, td); 574 UNP_LINK_WUNLOCK(); 575 return (error); 576} 577 578static void 579uipc_close(struct socket *so) 580{ 581 struct unpcb *unp, *unp2; 582 583 unp = sotounpcb(so); 584 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 585 586 UNP_LINK_WLOCK(); 587 UNP_PCB_LOCK(unp); 588 unp2 = unp->unp_conn; 589 if (unp2 != NULL) { 590 UNP_PCB_LOCK(unp2); 591 unp_disconnect(unp, unp2); 592 UNP_PCB_UNLOCK(unp2); 593 } 594 UNP_PCB_UNLOCK(unp); 595 UNP_LINK_WUNLOCK(); 596} 597 598static int 599uipc_connect2(struct socket *so1, struct socket *so2) 600{ 601 struct unpcb *unp, *unp2; 602 int error; 603 604 UNP_LINK_WLOCK(); 605 unp = so1->so_pcb; 606 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 607 UNP_PCB_LOCK(unp); 608 unp2 = so2->so_pcb; 609 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 610 UNP_PCB_LOCK(unp2); 611 error = unp_connect2(so1, so2, PRU_CONNECT2); 612 UNP_PCB_UNLOCK(unp2); 613 UNP_PCB_UNLOCK(unp); 614 UNP_LINK_WUNLOCK(); 615 return (error); 616} 617 618static void 619uipc_detach(struct socket *so) 620{ 621 struct unpcb *unp, *unp2; 622 struct sockaddr_un *saved_unp_addr; 623 struct vnode *vp; 624 int freeunp, local_unp_rights; 625 626 unp = sotounpcb(so); 627 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 628 629 UNP_LINK_WLOCK(); 630 UNP_LIST_LOCK(); 631 UNP_PCB_LOCK(unp); 632 LIST_REMOVE(unp, unp_link); 633 unp->unp_gencnt = ++unp_gencnt; 634 --unp_count; 635 UNP_LIST_UNLOCK(); 636 637 /* 638 * XXXRW: Should assert vp->v_socket == so. 639 */ 640 if ((vp = unp->unp_vnode) != NULL) { 641 unp->unp_vnode->v_socket = NULL; 642 unp->unp_vnode = NULL; 643 } 644 unp2 = unp->unp_conn; 645 if (unp2 != NULL) { 646 UNP_PCB_LOCK(unp2); 647 unp_disconnect(unp, unp2); 648 UNP_PCB_UNLOCK(unp2); 649 } 650 651 /* 652 * We hold the linkage lock exclusively, so it's OK to acquire 653 * multiple pcb locks at a time. 654 */ 655 while (!LIST_EMPTY(&unp->unp_refs)) { 656 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 657 658 UNP_PCB_LOCK(ref); 659 unp_drop(ref, ECONNRESET); 660 UNP_PCB_UNLOCK(ref); 661 } 662 local_unp_rights = unp_rights; 663 UNP_LINK_WUNLOCK(); 664 unp->unp_socket->so_pcb = NULL; 665 saved_unp_addr = unp->unp_addr; 666 unp->unp_addr = NULL; 667 unp->unp_refcount--; 668 freeunp = (unp->unp_refcount == 0); 669 if (saved_unp_addr != NULL) 670 free(saved_unp_addr, M_SONAME); 671 if (freeunp) { 672 UNP_PCB_LOCK_DESTROY(unp); 673 uma_zfree(unp_zone, unp); 674 } else 675 UNP_PCB_UNLOCK(unp); 676 if (vp) { 677 int vfslocked; 678 679 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 680 vrele(vp); 681 VFS_UNLOCK_GIANT(vfslocked); 682 } 683 if (local_unp_rights) 684 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 685} 686 687static int 688uipc_disconnect(struct socket *so) 689{ 690 struct unpcb *unp, *unp2; 691 692 unp = sotounpcb(so); 693 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 694 695 UNP_LINK_WLOCK(); 696 UNP_PCB_LOCK(unp); 697 unp2 = unp->unp_conn; 698 if (unp2 != NULL) { 699 UNP_PCB_LOCK(unp2); 700 unp_disconnect(unp, unp2); 701 UNP_PCB_UNLOCK(unp2); 702 } 703 UNP_PCB_UNLOCK(unp); 704 UNP_LINK_WUNLOCK(); 705 return (0); 706} 707 708static int 709uipc_listen(struct socket *so, int backlog, struct thread *td) 710{ 711 struct unpcb *unp; 712 int error; 713 714 unp = sotounpcb(so); 715 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 716 717 UNP_PCB_LOCK(unp); 718 if (unp->unp_vnode == NULL) { 719 UNP_PCB_UNLOCK(unp); 720 return (EINVAL); 721 } 722 723 SOCK_LOCK(so); 724 error = solisten_proto_check(so); 725 if (error == 0) { 726 cru2x(td->td_ucred, &unp->unp_peercred); 727 unp->unp_flags |= UNP_HAVEPCCACHED; 728 solisten_proto(so, backlog); 729 } 730 SOCK_UNLOCK(so); 731 UNP_PCB_UNLOCK(unp); 732 return (error); 733} 734 735static int 736uipc_peeraddr(struct socket *so, struct sockaddr **nam) 737{ 738 struct unpcb *unp, *unp2; 739 const struct sockaddr *sa; 740 741 unp = sotounpcb(so); 742 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 743 744 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 745 UNP_LINK_RLOCK(); 746 /* 747 * XXX: It seems that this test always fails even when connection is 748 * established. So, this else clause is added as workaround to 749 * return PF_LOCAL sockaddr. 750 */ 751 unp2 = unp->unp_conn; 752 if (unp2 != NULL) { 753 UNP_PCB_LOCK(unp2); 754 if (unp2->unp_addr != NULL) 755 sa = (struct sockaddr *) unp2->unp_addr; 756 else 757 sa = &sun_noname; 758 bcopy(sa, *nam, sa->sa_len); 759 UNP_PCB_UNLOCK(unp2); 760 } else { 761 sa = &sun_noname; 762 bcopy(sa, *nam, sa->sa_len); 763 } 764 UNP_LINK_RUNLOCK(); 765 return (0); 766} 767 768static int 769uipc_rcvd(struct socket *so, int flags) 770{ 771 struct unpcb *unp, *unp2; 772 struct socket *so2; 773 u_int mbcnt, sbcc; 774 u_long newhiwat; 775 776 unp = sotounpcb(so); 777 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 778 779 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 780 panic("uipc_rcvd socktype %d", so->so_type); 781 782 /* 783 * Adjust backpressure on sender and wakeup any waiting to write. 784 * 785 * The unp lock is acquired to maintain the validity of the unp_conn 786 * pointer; no lock on unp2 is required as unp2->unp_socket will be 787 * static as long as we don't permit unp2 to disconnect from unp, 788 * which is prevented by the lock on unp. We cache values from 789 * so_rcv to avoid holding the so_rcv lock over the entire 790 * transaction on the remote so_snd. 791 */ 792 SOCKBUF_LOCK(&so->so_rcv); 793 mbcnt = so->so_rcv.sb_mbcnt; 794 sbcc = so->so_rcv.sb_cc; 795 SOCKBUF_UNLOCK(&so->so_rcv); 796 UNP_PCB_LOCK(unp); 797 unp2 = unp->unp_conn; 798 if (unp2 == NULL) { 799 UNP_PCB_UNLOCK(unp); 800 return (0); 801 } 802 so2 = unp2->unp_socket; 803 SOCKBUF_LOCK(&so2->so_snd); 804 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 805 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 806 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 807 newhiwat, RLIM_INFINITY); 808 sowwakeup_locked(so2); 809 unp->unp_mbcnt = mbcnt; 810 unp->unp_cc = sbcc; 811 UNP_PCB_UNLOCK(unp); 812 return (0); 813} 814 815static int 816uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 817 struct mbuf *control, struct thread *td) 818{ 819 struct unpcb *unp, *unp2; 820 struct socket *so2; 821 u_int mbcnt_delta, sbcc; 822 u_int newhiwat; 823 int error = 0; 824 825 unp = sotounpcb(so); 826 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 827 828 if (flags & PRUS_OOB) { 829 error = EOPNOTSUPP; 830 goto release; 831 } 832 if (control != NULL && (error = unp_internalize(&control, td))) 833 goto release; 834 if ((nam != NULL) || (flags & PRUS_EOF)) 835 UNP_LINK_WLOCK(); 836 else 837 UNP_LINK_RLOCK(); 838 switch (so->so_type) { 839 case SOCK_DGRAM: 840 { 841 const struct sockaddr *from; 842 843 unp2 = unp->unp_conn; 844 if (nam != NULL) { 845 UNP_LINK_WLOCK_ASSERT(); 846 if (unp2 != NULL) { 847 error = EISCONN; 848 break; 849 } 850 error = unp_connect(so, nam, td); 851 if (error) 852 break; 853 unp2 = unp->unp_conn; 854 } 855 856 /* 857 * Because connect() and send() are non-atomic in a sendto() 858 * with a target address, it's possible that the socket will 859 * have disconnected before the send() can run. In that case 860 * return the slightly counter-intuitive but otherwise 861 * correct error that the socket is not connected. 862 */ 863 if (unp2 == NULL) { 864 error = ENOTCONN; 865 break; 866 } 867 /* Lockless read. */ 868 if (unp2->unp_flags & UNP_WANTCRED) 869 control = unp_addsockcred(td, control); 870 UNP_PCB_LOCK(unp); 871 if (unp->unp_addr != NULL) 872 from = (struct sockaddr *)unp->unp_addr; 873 else 874 from = &sun_noname; 875 so2 = unp2->unp_socket; 876 SOCKBUF_LOCK(&so2->so_rcv); 877 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 878 sorwakeup_locked(so2); 879 m = NULL; 880 control = NULL; 881 } else { 882 SOCKBUF_UNLOCK(&so2->so_rcv); 883 error = ENOBUFS; 884 } 885 if (nam != NULL) { 886 UNP_LINK_WLOCK_ASSERT(); 887 UNP_PCB_LOCK(unp2); 888 unp_disconnect(unp, unp2); 889 UNP_PCB_UNLOCK(unp2); 890 } 891 UNP_PCB_UNLOCK(unp); 892 break; 893 } 894 895 case SOCK_SEQPACKET: 896 case SOCK_STREAM: 897 if ((so->so_state & SS_ISCONNECTED) == 0) { 898 if (nam != NULL) { 899 UNP_LINK_WLOCK_ASSERT(); 900 error = unp_connect(so, nam, td); 901 if (error) 902 break; /* XXX */ 903 } else { 904 error = ENOTCONN; 905 break; 906 } 907 } 908 909 /* Lockless read. */ 910 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 911 error = EPIPE; 912 break; 913 } 914 915 /* 916 * Because connect() and send() are non-atomic in a sendto() 917 * with a target address, it's possible that the socket will 918 * have disconnected before the send() can run. In that case 919 * return the slightly counter-intuitive but otherwise 920 * correct error that the socket is not connected. 921 * 922 * Locking here must be done carefully: the linkage lock 923 * prevents interconnections between unpcbs from changing, so 924 * we can traverse from unp to unp2 without acquiring unp's 925 * lock. Socket buffer locks follow unpcb locks, so we can 926 * acquire both remote and lock socket buffer locks. 927 */ 928 unp2 = unp->unp_conn; 929 if (unp2 == NULL) { 930 error = ENOTCONN; 931 break; 932 } 933 so2 = unp2->unp_socket; 934 UNP_PCB_LOCK(unp2); 935 SOCKBUF_LOCK(&so2->so_rcv); 936 if (unp2->unp_flags & UNP_WANTCRED) { 937 /* 938 * Credentials are passed only once on SOCK_STREAM. 939 */ 940 unp2->unp_flags &= ~UNP_WANTCRED; 941 control = unp_addsockcred(td, control); 942 } 943 /* 944 * Send to paired receive port, and then reduce send buffer 945 * hiwater marks to maintain backpressure. Wake up readers. 946 */ 947 switch (so->so_type) { 948 case SOCK_STREAM: 949 if (control != NULL) { 950 if (sbappendcontrol_locked(&so2->so_rcv, m, 951 control)) 952 control = NULL; 953 } else 954 sbappend_locked(&so2->so_rcv, m); 955 break; 956 957 case SOCK_SEQPACKET: { 958 const struct sockaddr *from; 959 960 from = &sun_noname; 961 if (sbappendaddr_locked(&so2->so_rcv, from, m, 962 control)) 963 control = NULL; 964 break; 965 } 966 } 967 968 /* 969 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 970 * datagram size and back-pressure for SOCK_SEQPACKET, which 971 * can lead to undesired return of EMSGSIZE on send instead 972 * of more desirable blocking. 973 */ 974 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 975 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 976 sbcc = so2->so_rcv.sb_cc; 977 sorwakeup_locked(so2); 978 979 SOCKBUF_LOCK(&so->so_snd); 980 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc)) 981 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 982 else 983 newhiwat = 0; 984 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 985 newhiwat, RLIM_INFINITY); 986 so->so_snd.sb_mbmax -= mbcnt_delta; 987 SOCKBUF_UNLOCK(&so->so_snd); 988 unp2->unp_cc = sbcc; 989 UNP_PCB_UNLOCK(unp2); 990 m = NULL; 991 break; 992 993 default: 994 panic("uipc_send unknown socktype"); 995 } 996 997 /* 998 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 999 */ 1000 if (flags & PRUS_EOF) { 1001 UNP_PCB_LOCK(unp); 1002 socantsendmore(so); 1003 unp_shutdown(unp); 1004 UNP_PCB_UNLOCK(unp); 1005 } 1006 1007 if ((nam != NULL) || (flags & PRUS_EOF)) 1008 UNP_LINK_WUNLOCK(); 1009 else 1010 UNP_LINK_RUNLOCK(); 1011 1012 if (control != NULL && error != 0) 1013 unp_dispose(control); 1014 1015release: 1016 if (control != NULL) 1017 m_freem(control); 1018 if (m != NULL) 1019 m_freem(m); 1020 return (error); 1021} 1022 1023static int 1024uipc_sense(struct socket *so, struct stat *sb) 1025{ 1026 struct unpcb *unp, *unp2; 1027 struct socket *so2; 1028 1029 unp = sotounpcb(so); 1030 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1031 1032 sb->st_blksize = so->so_snd.sb_hiwat; 1033 UNP_LINK_RLOCK(); 1034 UNP_PCB_LOCK(unp); 1035 unp2 = unp->unp_conn; 1036 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1037 unp2 != NULL) { 1038 so2 = unp2->unp_socket; 1039 sb->st_blksize += so2->so_rcv.sb_cc; 1040 } 1041 sb->st_dev = NODEV; 1042 if (unp->unp_ino == 0) 1043 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1044 sb->st_ino = unp->unp_ino; 1045 UNP_PCB_UNLOCK(unp); 1046 UNP_LINK_RUNLOCK(); 1047 return (0); 1048} 1049 1050static int 1051uipc_shutdown(struct socket *so) 1052{ 1053 struct unpcb *unp; 1054 1055 unp = sotounpcb(so); 1056 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1057 1058 UNP_LINK_WLOCK(); 1059 UNP_PCB_LOCK(unp); 1060 socantsendmore(so); 1061 unp_shutdown(unp); 1062 UNP_PCB_UNLOCK(unp); 1063 UNP_LINK_WUNLOCK(); 1064 return (0); 1065} 1066 1067static int 1068uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1069{ 1070 struct unpcb *unp; 1071 const struct sockaddr *sa; 1072 1073 unp = sotounpcb(so); 1074 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1075 1076 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1077 UNP_PCB_LOCK(unp); 1078 if (unp->unp_addr != NULL) 1079 sa = (struct sockaddr *) unp->unp_addr; 1080 else 1081 sa = &sun_noname; 1082 bcopy(sa, *nam, sa->sa_len); 1083 UNP_PCB_UNLOCK(unp); 1084 return (0); 1085} 1086 1087static struct pr_usrreqs uipc_usrreqs_dgram = { 1088 .pru_abort = uipc_abort, 1089 .pru_accept = uipc_accept, 1090 .pru_attach = uipc_attach, 1091 .pru_bind = uipc_bind, 1092 .pru_connect = uipc_connect, 1093 .pru_connect2 = uipc_connect2, 1094 .pru_detach = uipc_detach, 1095 .pru_disconnect = uipc_disconnect, 1096 .pru_listen = uipc_listen, 1097 .pru_peeraddr = uipc_peeraddr, 1098 .pru_rcvd = uipc_rcvd, 1099 .pru_send = uipc_send, 1100 .pru_sense = uipc_sense, 1101 .pru_shutdown = uipc_shutdown, 1102 .pru_sockaddr = uipc_sockaddr, 1103 .pru_soreceive = soreceive_dgram, 1104 .pru_close = uipc_close, 1105}; 1106 1107static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1108 .pru_abort = uipc_abort, 1109 .pru_accept = uipc_accept, 1110 .pru_attach = uipc_attach, 1111 .pru_bind = uipc_bind, 1112 .pru_connect = uipc_connect, 1113 .pru_connect2 = uipc_connect2, 1114 .pru_detach = uipc_detach, 1115 .pru_disconnect = uipc_disconnect, 1116 .pru_listen = uipc_listen, 1117 .pru_peeraddr = uipc_peeraddr, 1118 .pru_rcvd = uipc_rcvd, 1119 .pru_send = uipc_send, 1120 .pru_sense = uipc_sense, 1121 .pru_shutdown = uipc_shutdown, 1122 .pru_sockaddr = uipc_sockaddr, 1123 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1124 .pru_close = uipc_close, 1125}; 1126 1127static struct pr_usrreqs uipc_usrreqs_stream = { 1128 .pru_abort = uipc_abort, 1129 .pru_accept = uipc_accept, 1130 .pru_attach = uipc_attach, 1131 .pru_bind = uipc_bind, 1132 .pru_connect = uipc_connect, 1133 .pru_connect2 = uipc_connect2, 1134 .pru_detach = uipc_detach, 1135 .pru_disconnect = uipc_disconnect, 1136 .pru_listen = uipc_listen, 1137 .pru_peeraddr = uipc_peeraddr, 1138 .pru_rcvd = uipc_rcvd, 1139 .pru_send = uipc_send, 1140 .pru_sense = uipc_sense, 1141 .pru_shutdown = uipc_shutdown, 1142 .pru_sockaddr = uipc_sockaddr, 1143 .pru_soreceive = soreceive_generic, 1144 .pru_close = uipc_close, 1145}; 1146 1147static int 1148uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1149{ 1150 struct unpcb *unp; 1151 struct xucred xu; 1152 int error, optval; 1153 1154 if (sopt->sopt_level != 0) 1155 return (EINVAL); 1156 1157 unp = sotounpcb(so); 1158 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1159 error = 0; 1160 switch (sopt->sopt_dir) { 1161 case SOPT_GET: 1162 switch (sopt->sopt_name) { 1163 case LOCAL_PEERCRED: 1164 UNP_PCB_LOCK(unp); 1165 if (unp->unp_flags & UNP_HAVEPC) 1166 xu = unp->unp_peercred; 1167 else { 1168 if (so->so_type == SOCK_STREAM) 1169 error = ENOTCONN; 1170 else 1171 error = EINVAL; 1172 } 1173 UNP_PCB_UNLOCK(unp); 1174 if (error == 0) 1175 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1176 break; 1177 1178 case LOCAL_CREDS: 1179 /* Unlocked read. */ 1180 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1181 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1182 break; 1183 1184 case LOCAL_CONNWAIT: 1185 /* Unlocked read. */ 1186 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1187 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1188 break; 1189 1190 default: 1191 error = EOPNOTSUPP; 1192 break; 1193 } 1194 break; 1195 1196 case SOPT_SET: 1197 switch (sopt->sopt_name) { 1198 case LOCAL_CREDS: 1199 case LOCAL_CONNWAIT: 1200 error = sooptcopyin(sopt, &optval, sizeof(optval), 1201 sizeof(optval)); 1202 if (error) 1203 break; 1204 1205#define OPTSET(bit) do { \ 1206 UNP_PCB_LOCK(unp); \ 1207 if (optval) \ 1208 unp->unp_flags |= bit; \ 1209 else \ 1210 unp->unp_flags &= ~bit; \ 1211 UNP_PCB_UNLOCK(unp); \ 1212} while (0) 1213 1214 switch (sopt->sopt_name) { 1215 case LOCAL_CREDS: 1216 OPTSET(UNP_WANTCRED); 1217 break; 1218 1219 case LOCAL_CONNWAIT: 1220 OPTSET(UNP_CONNWAIT); 1221 break; 1222 1223 default: 1224 break; 1225 } 1226 break; 1227#undef OPTSET 1228 default: 1229 error = ENOPROTOOPT; 1230 break; 1231 } 1232 break; 1233 1234 default: 1235 error = EOPNOTSUPP; 1236 break; 1237 } 1238 return (error); 1239} 1240 1241static int 1242unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1243{ 1244 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1245 struct vnode *vp; 1246 struct socket *so2, *so3; 1247 struct unpcb *unp, *unp2, *unp3; 1248 int error, len, vfslocked; 1249 struct nameidata nd; 1250 char buf[SOCK_MAXADDRLEN]; 1251 struct sockaddr *sa; 1252 1253 UNP_LINK_WLOCK_ASSERT(); 1254 1255 unp = sotounpcb(so); 1256 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1257 1258 if (nam->sa_len > sizeof(struct sockaddr_un)) 1259 return (EINVAL); 1260 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1261 if (len <= 0) 1262 return (EINVAL); 1263 bcopy(soun->sun_path, buf, len); 1264 buf[len] = 0; 1265 1266 UNP_PCB_LOCK(unp); 1267 if (unp->unp_flags & UNP_CONNECTING) { 1268 UNP_PCB_UNLOCK(unp); 1269 return (EALREADY); 1270 } 1271 UNP_LINK_WUNLOCK(); 1272 unp->unp_flags |= UNP_CONNECTING; 1273 UNP_PCB_UNLOCK(unp); 1274 1275 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1276 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKSHARED | LOCKLEAF, 1277 UIO_SYSSPACE, buf, td); 1278 error = namei(&nd); 1279 if (error) 1280 vp = NULL; 1281 else 1282 vp = nd.ni_vp; 1283 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1284 vfslocked = NDHASGIANT(&nd); 1285 NDFREE(&nd, NDF_ONLY_PNBUF); 1286 if (error) 1287 goto bad; 1288 1289 if (vp->v_type != VSOCK) { 1290 error = ENOTSOCK; 1291 goto bad; 1292 } 1293#ifdef MAC 1294 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1295 if (error) 1296 goto bad; 1297#endif 1298 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1299 if (error) 1300 goto bad; 1301 VFS_UNLOCK_GIANT(vfslocked); 1302 1303 unp = sotounpcb(so); 1304 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1305 1306 /* 1307 * Lock linkage lock for two reasons: make sure v_socket is stable, 1308 * and to protect simultaneous locking of multiple pcbs. 1309 */ 1310 UNP_LINK_WLOCK(); 1311 so2 = vp->v_socket; 1312 if (so2 == NULL) { 1313 error = ECONNREFUSED; 1314 goto bad2; 1315 } 1316 if (so->so_type != so2->so_type) { 1317 error = EPROTOTYPE; 1318 goto bad2; 1319 } 1320 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1321 if (so2->so_options & SO_ACCEPTCONN) { 1322 CURVNET_SET(so2->so_vnet); 1323 so3 = sonewconn(so2, 0); 1324 CURVNET_RESTORE(); 1325 } else 1326 so3 = NULL; 1327 if (so3 == NULL) { 1328 error = ECONNREFUSED; 1329 goto bad2; 1330 } 1331 unp = sotounpcb(so); 1332 unp2 = sotounpcb(so2); 1333 unp3 = sotounpcb(so3); 1334 UNP_PCB_LOCK(unp); 1335 UNP_PCB_LOCK(unp2); 1336 UNP_PCB_LOCK(unp3); 1337 if (unp2->unp_addr != NULL) { 1338 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1339 unp3->unp_addr = (struct sockaddr_un *) sa; 1340 sa = NULL; 1341 } 1342 1343 /* 1344 * The connecter's (client's) credentials are copied from its 1345 * process structure at the time of connect() (which is now). 1346 */ 1347 cru2x(td->td_ucred, &unp3->unp_peercred); 1348 unp3->unp_flags |= UNP_HAVEPC; 1349 1350 /* 1351 * The receiver's (server's) credentials are copied from the 1352 * unp_peercred member of socket on which the former called 1353 * listen(); uipc_listen() cached that process's credentials 1354 * at that time so we can use them now. 1355 */ 1356 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1357 ("unp_connect: listener without cached peercred")); 1358 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1359 sizeof(unp->unp_peercred)); 1360 unp->unp_flags |= UNP_HAVEPC; 1361 if (unp2->unp_flags & UNP_WANTCRED) 1362 unp3->unp_flags |= UNP_WANTCRED; 1363 UNP_PCB_UNLOCK(unp3); 1364 UNP_PCB_UNLOCK(unp2); 1365 UNP_PCB_UNLOCK(unp); 1366#ifdef MAC 1367 mac_socketpeer_set_from_socket(so, so3); 1368 mac_socketpeer_set_from_socket(so3, so); 1369#endif 1370 1371 so2 = so3; 1372 } 1373 unp = sotounpcb(so); 1374 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1375 unp2 = sotounpcb(so2); 1376 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1377 UNP_PCB_LOCK(unp); 1378 UNP_PCB_LOCK(unp2); 1379 error = unp_connect2(so, so2, PRU_CONNECT); 1380 UNP_PCB_UNLOCK(unp2); 1381 UNP_PCB_UNLOCK(unp); 1382bad2: 1383 UNP_LINK_WUNLOCK(); 1384 if (vfslocked) 1385 /* 1386 * Giant has been previously acquired. This means filesystem 1387 * isn't MPSAFE. Do it once again. 1388 */ 1389 mtx_lock(&Giant); 1390bad: 1391 if (vp != NULL) 1392 vput(vp); 1393 VFS_UNLOCK_GIANT(vfslocked); 1394 free(sa, M_SONAME); 1395 UNP_LINK_WLOCK(); 1396 UNP_PCB_LOCK(unp); 1397 unp->unp_flags &= ~UNP_CONNECTING; 1398 UNP_PCB_UNLOCK(unp); 1399 return (error); 1400} 1401 1402static int 1403unp_connect2(struct socket *so, struct socket *so2, int req) 1404{ 1405 struct unpcb *unp; 1406 struct unpcb *unp2; 1407 1408 unp = sotounpcb(so); 1409 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1410 unp2 = sotounpcb(so2); 1411 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1412 1413 UNP_LINK_WLOCK_ASSERT(); 1414 UNP_PCB_LOCK_ASSERT(unp); 1415 UNP_PCB_LOCK_ASSERT(unp2); 1416 1417 if (so2->so_type != so->so_type) 1418 return (EPROTOTYPE); 1419 unp->unp_conn = unp2; 1420 1421 switch (so->so_type) { 1422 case SOCK_DGRAM: 1423 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1424 soisconnected(so); 1425 break; 1426 1427 case SOCK_STREAM: 1428 case SOCK_SEQPACKET: 1429 unp2->unp_conn = unp; 1430 if (req == PRU_CONNECT && 1431 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1432 soisconnecting(so); 1433 else 1434 soisconnected(so); 1435 soisconnected(so2); 1436 break; 1437 1438 default: 1439 panic("unp_connect2"); 1440 } 1441 return (0); 1442} 1443 1444static void 1445unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1446{ 1447 struct socket *so; 1448 1449 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1450 1451 UNP_LINK_WLOCK_ASSERT(); 1452 UNP_PCB_LOCK_ASSERT(unp); 1453 UNP_PCB_LOCK_ASSERT(unp2); 1454 1455 unp->unp_conn = NULL; 1456 switch (unp->unp_socket->so_type) { 1457 case SOCK_DGRAM: 1458 LIST_REMOVE(unp, unp_reflink); 1459 so = unp->unp_socket; 1460 SOCK_LOCK(so); 1461 so->so_state &= ~SS_ISCONNECTED; 1462 SOCK_UNLOCK(so); 1463 break; 1464 1465 case SOCK_STREAM: 1466 case SOCK_SEQPACKET: 1467 soisdisconnected(unp->unp_socket); 1468 unp2->unp_conn = NULL; 1469 soisdisconnected(unp2->unp_socket); 1470 break; 1471 } 1472} 1473 1474/* 1475 * unp_pcblist() walks the global list of struct unpcb's to generate a 1476 * pointer list, bumping the refcount on each unpcb. It then copies them out 1477 * sequentially, validating the generation number on each to see if it has 1478 * been detached. All of this is necessary because copyout() may sleep on 1479 * disk I/O. 1480 */ 1481static int 1482unp_pcblist(SYSCTL_HANDLER_ARGS) 1483{ 1484 int error, i, n; 1485 int freeunp; 1486 struct unpcb *unp, **unp_list; 1487 unp_gen_t gencnt; 1488 struct xunpgen *xug; 1489 struct unp_head *head; 1490 struct xunpcb *xu; 1491 1492 switch ((intptr_t)arg1) { 1493 case SOCK_STREAM: 1494 head = &unp_shead; 1495 break; 1496 1497 case SOCK_DGRAM: 1498 head = &unp_dhead; 1499 break; 1500 1501 case SOCK_SEQPACKET: 1502 head = &unp_sphead; 1503 break; 1504 1505 default: 1506 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1507 } 1508 1509 /* 1510 * The process of preparing the PCB list is too time-consuming and 1511 * resource-intensive to repeat twice on every request. 1512 */ 1513 if (req->oldptr == NULL) { 1514 n = unp_count; 1515 req->oldidx = 2 * (sizeof *xug) 1516 + (n + n/8) * sizeof(struct xunpcb); 1517 return (0); 1518 } 1519 1520 if (req->newptr != NULL) 1521 return (EPERM); 1522 1523 /* 1524 * OK, now we're committed to doing something. 1525 */ 1526 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1527 UNP_LIST_LOCK(); 1528 gencnt = unp_gencnt; 1529 n = unp_count; 1530 UNP_LIST_UNLOCK(); 1531 1532 xug->xug_len = sizeof *xug; 1533 xug->xug_count = n; 1534 xug->xug_gen = gencnt; 1535 xug->xug_sogen = so_gencnt; 1536 error = SYSCTL_OUT(req, xug, sizeof *xug); 1537 if (error) { 1538 free(xug, M_TEMP); 1539 return (error); 1540 } 1541 1542 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1543 1544 UNP_LIST_LOCK(); 1545 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1546 unp = LIST_NEXT(unp, unp_link)) { 1547 UNP_PCB_LOCK(unp); 1548 if (unp->unp_gencnt <= gencnt) { 1549 if (cr_cansee(req->td->td_ucred, 1550 unp->unp_socket->so_cred)) { 1551 UNP_PCB_UNLOCK(unp); 1552 continue; 1553 } 1554 unp_list[i++] = unp; 1555 unp->unp_refcount++; 1556 } 1557 UNP_PCB_UNLOCK(unp); 1558 } 1559 UNP_LIST_UNLOCK(); 1560 n = i; /* In case we lost some during malloc. */ 1561 1562 error = 0; 1563 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1564 for (i = 0; i < n; i++) { 1565 unp = unp_list[i]; 1566 UNP_PCB_LOCK(unp); 1567 unp->unp_refcount--; 1568 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1569 xu->xu_len = sizeof *xu; 1570 xu->xu_unpp = unp; 1571 /* 1572 * XXX - need more locking here to protect against 1573 * connect/disconnect races for SMP. 1574 */ 1575 if (unp->unp_addr != NULL) 1576 bcopy(unp->unp_addr, &xu->xu_addr, 1577 unp->unp_addr->sun_len); 1578 if (unp->unp_conn != NULL && 1579 unp->unp_conn->unp_addr != NULL) 1580 bcopy(unp->unp_conn->unp_addr, 1581 &xu->xu_caddr, 1582 unp->unp_conn->unp_addr->sun_len); 1583 bcopy(unp, &xu->xu_unp, sizeof *unp); 1584 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1585 UNP_PCB_UNLOCK(unp); 1586 error = SYSCTL_OUT(req, xu, sizeof *xu); 1587 } else { 1588 freeunp = (unp->unp_refcount == 0); 1589 UNP_PCB_UNLOCK(unp); 1590 if (freeunp) { 1591 UNP_PCB_LOCK_DESTROY(unp); 1592 uma_zfree(unp_zone, unp); 1593 } 1594 } 1595 } 1596 free(xu, M_TEMP); 1597 if (!error) { 1598 /* 1599 * Give the user an updated idea of our state. If the 1600 * generation differs from what we told her before, she knows 1601 * that something happened while we were processing this 1602 * request, and it might be necessary to retry. 1603 */ 1604 xug->xug_gen = unp_gencnt; 1605 xug->xug_sogen = so_gencnt; 1606 xug->xug_count = unp_count; 1607 error = SYSCTL_OUT(req, xug, sizeof *xug); 1608 } 1609 free(unp_list, M_TEMP); 1610 free(xug, M_TEMP); 1611 return (error); 1612} 1613 1614SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1615 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1616 "List of active local datagram sockets"); 1617SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 1618 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1619 "List of active local stream sockets"); 1620SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1621 CTLTYPE_OPAQUE | CTLFLAG_RD, 1622 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1623 "List of active local seqpacket sockets"); 1624 1625static void 1626unp_shutdown(struct unpcb *unp) 1627{ 1628 struct unpcb *unp2; 1629 struct socket *so; 1630 1631 UNP_LINK_WLOCK_ASSERT(); 1632 UNP_PCB_LOCK_ASSERT(unp); 1633 1634 unp2 = unp->unp_conn; 1635 if ((unp->unp_socket->so_type == SOCK_STREAM || 1636 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1637 so = unp2->unp_socket; 1638 if (so != NULL) 1639 socantrcvmore(so); 1640 } 1641} 1642 1643static void 1644unp_drop(struct unpcb *unp, int errno) 1645{ 1646 struct socket *so = unp->unp_socket; 1647 struct unpcb *unp2; 1648 1649 UNP_LINK_WLOCK_ASSERT(); 1650 UNP_PCB_LOCK_ASSERT(unp); 1651 1652 so->so_error = errno; 1653 unp2 = unp->unp_conn; 1654 if (unp2 == NULL) 1655 return; 1656 UNP_PCB_LOCK(unp2); 1657 unp_disconnect(unp, unp2); 1658 UNP_PCB_UNLOCK(unp2); 1659} 1660 1661static void 1662unp_freerights(struct file **rp, int fdcount) 1663{ 1664 int i; 1665 struct file *fp; 1666 1667 for (i = 0; i < fdcount; i++) { 1668 fp = *rp; 1669 *rp++ = NULL; 1670 unp_discard(fp); 1671 } 1672} 1673 1674static int 1675unp_externalize(struct mbuf *control, struct mbuf **controlp) 1676{ 1677 struct thread *td = curthread; /* XXX */ 1678 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1679 int i; 1680 int *fdp; 1681 struct file **rp; 1682 struct file *fp; 1683 void *data; 1684 socklen_t clen = control->m_len, datalen; 1685 int error, newfds; 1686 int f; 1687 u_int newlen; 1688 1689 UNP_LINK_UNLOCK_ASSERT(); 1690 1691 error = 0; 1692 if (controlp != NULL) /* controlp == NULL => free control messages */ 1693 *controlp = NULL; 1694 while (cm != NULL) { 1695 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1696 error = EINVAL; 1697 break; 1698 } 1699 data = CMSG_DATA(cm); 1700 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1701 if (cm->cmsg_level == SOL_SOCKET 1702 && cm->cmsg_type == SCM_RIGHTS) { 1703 newfds = datalen / sizeof(struct file *); 1704 rp = data; 1705 1706 /* If we're not outputting the descriptors free them. */ 1707 if (error || controlp == NULL) { 1708 unp_freerights(rp, newfds); 1709 goto next; 1710 } 1711 FILEDESC_XLOCK(td->td_proc->p_fd); 1712 /* if the new FD's will not fit free them. */ 1713 if (!fdavail(td, newfds)) { 1714 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1715 error = EMSGSIZE; 1716 unp_freerights(rp, newfds); 1717 goto next; 1718 } 1719 1720 /* 1721 * Now change each pointer to an fd in the global 1722 * table to an integer that is the index to the local 1723 * fd table entry that we set up to point to the 1724 * global one we are transferring. 1725 */ 1726 newlen = newfds * sizeof(int); 1727 *controlp = sbcreatecontrol(NULL, newlen, 1728 SCM_RIGHTS, SOL_SOCKET); 1729 if (*controlp == NULL) { 1730 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1731 error = E2BIG; 1732 unp_freerights(rp, newfds); 1733 goto next; 1734 } 1735 1736 fdp = (int *) 1737 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1738 for (i = 0; i < newfds; i++) { 1739 if (fdalloc(td, 0, &f)) 1740 panic("unp_externalize fdalloc failed"); 1741 fp = *rp++; 1742 td->td_proc->p_fd->fd_ofiles[f] = fp; 1743 unp_externalize_fp(fp); 1744 *fdp++ = f; 1745 } 1746 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1747 } else { 1748 /* We can just copy anything else across. */ 1749 if (error || controlp == NULL) 1750 goto next; 1751 *controlp = sbcreatecontrol(NULL, datalen, 1752 cm->cmsg_type, cm->cmsg_level); 1753 if (*controlp == NULL) { 1754 error = ENOBUFS; 1755 goto next; 1756 } 1757 bcopy(data, 1758 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1759 datalen); 1760 } 1761 controlp = &(*controlp)->m_next; 1762 1763next: 1764 if (CMSG_SPACE(datalen) < clen) { 1765 clen -= CMSG_SPACE(datalen); 1766 cm = (struct cmsghdr *) 1767 ((caddr_t)cm + CMSG_SPACE(datalen)); 1768 } else { 1769 clen = 0; 1770 cm = NULL; 1771 } 1772 } 1773 1774 m_freem(control); 1775 return (error); 1776} 1777 1778static void 1779unp_zone_change(void *tag) 1780{ 1781 1782 uma_zone_set_max(unp_zone, maxsockets); 1783} 1784 1785static void 1786unp_init(void) 1787{ 1788 1789#ifdef VIMAGE 1790 if (!IS_DEFAULT_VNET(curvnet)) 1791 return; 1792#endif 1793 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1794 NULL, NULL, UMA_ALIGN_PTR, 0); 1795 if (unp_zone == NULL) 1796 panic("unp_init"); 1797 uma_zone_set_max(unp_zone, maxsockets); 1798 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1799 NULL, EVENTHANDLER_PRI_ANY); 1800 LIST_INIT(&unp_dhead); 1801 LIST_INIT(&unp_shead); 1802 LIST_INIT(&unp_sphead); 1803 SLIST_INIT(&unp_defers); 1804 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1805 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 1806 UNP_LINK_LOCK_INIT(); 1807 UNP_LIST_LOCK_INIT(); 1808 UNP_DEFERRED_LOCK_INIT(); 1809} 1810 1811static int 1812unp_internalize(struct mbuf **controlp, struct thread *td) 1813{ 1814 struct mbuf *control = *controlp; 1815 struct proc *p = td->td_proc; 1816 struct filedesc *fdescp = p->p_fd; 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 default: 1926 error = EINVAL; 1927 goto out; 1928 } 1929 1930 controlp = &(*controlp)->m_next; 1931 if (CMSG_SPACE(datalen) < clen) { 1932 clen -= CMSG_SPACE(datalen); 1933 cm = (struct cmsghdr *) 1934 ((caddr_t)cm + CMSG_SPACE(datalen)); 1935 } else { 1936 clen = 0; 1937 cm = NULL; 1938 } 1939 } 1940 1941out: 1942 m_freem(control); 1943 return (error); 1944} 1945 1946static struct mbuf * 1947unp_addsockcred(struct thread *td, struct mbuf *control) 1948{ 1949 struct mbuf *m, *n, *n_prev; 1950 struct sockcred *sc; 1951 const struct cmsghdr *cm; 1952 int ngroups; 1953 int i; 1954 1955 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1956 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1957 if (m == NULL) 1958 return (control); 1959 1960 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1961 sc->sc_uid = td->td_ucred->cr_ruid; 1962 sc->sc_euid = td->td_ucred->cr_uid; 1963 sc->sc_gid = td->td_ucred->cr_rgid; 1964 sc->sc_egid = td->td_ucred->cr_gid; 1965 sc->sc_ngroups = ngroups; 1966 for (i = 0; i < sc->sc_ngroups; i++) 1967 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1968 1969 /* 1970 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1971 * created SCM_CREDS control message (struct sockcred) has another 1972 * format. 1973 */ 1974 if (control != NULL) 1975 for (n = control, n_prev = NULL; n != NULL;) { 1976 cm = mtod(n, struct cmsghdr *); 1977 if (cm->cmsg_level == SOL_SOCKET && 1978 cm->cmsg_type == SCM_CREDS) { 1979 if (n_prev == NULL) 1980 control = n->m_next; 1981 else 1982 n_prev->m_next = n->m_next; 1983 n = m_free(n); 1984 } else { 1985 n_prev = n; 1986 n = n->m_next; 1987 } 1988 } 1989 1990 /* Prepend it to the head. */ 1991 m->m_next = control; 1992 return (m); 1993} 1994 1995static struct unpcb * 1996fptounp(struct file *fp) 1997{ 1998 struct socket *so; 1999 2000 if (fp->f_type != DTYPE_SOCKET) 2001 return (NULL); 2002 if ((so = fp->f_data) == NULL) 2003 return (NULL); 2004 if (so->so_proto->pr_domain != &localdomain) 2005 return (NULL); 2006 return sotounpcb(so); 2007} 2008 2009static void 2010unp_discard(struct file *fp) 2011{ 2012 struct unp_defer *dr; 2013 2014 if (unp_externalize_fp(fp)) { 2015 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2016 dr->ud_fp = fp; 2017 UNP_DEFERRED_LOCK(); 2018 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2019 UNP_DEFERRED_UNLOCK(); 2020 atomic_add_int(&unp_defers_count, 1); 2021 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2022 } else 2023 (void) closef(fp, (struct thread *)NULL); 2024} 2025 2026static void 2027unp_process_defers(void *arg __unused, int pending) 2028{ 2029 struct unp_defer *dr; 2030 SLIST_HEAD(, unp_defer) drl; 2031 int count; 2032 2033 SLIST_INIT(&drl); 2034 for (;;) { 2035 UNP_DEFERRED_LOCK(); 2036 if (SLIST_FIRST(&unp_defers) == NULL) { 2037 UNP_DEFERRED_UNLOCK(); 2038 break; 2039 } 2040 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2041 UNP_DEFERRED_UNLOCK(); 2042 count = 0; 2043 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2044 SLIST_REMOVE_HEAD(&drl, ud_link); 2045 closef(dr->ud_fp, NULL); 2046 free(dr, M_TEMP); 2047 count++; 2048 } 2049 atomic_add_int(&unp_defers_count, -count); 2050 } 2051} 2052 2053static void 2054unp_internalize_fp(struct file *fp) 2055{ 2056 struct unpcb *unp; 2057 2058 UNP_LINK_WLOCK(); 2059 if ((unp = fptounp(fp)) != NULL) { 2060 unp->unp_file = fp; 2061 unp->unp_msgcount++; 2062 } 2063 fhold(fp); 2064 unp_rights++; 2065 UNP_LINK_WUNLOCK(); 2066} 2067 2068static int 2069unp_externalize_fp(struct file *fp) 2070{ 2071 struct unpcb *unp; 2072 int ret; 2073 2074 UNP_LINK_WLOCK(); 2075 if ((unp = fptounp(fp)) != NULL) { 2076 unp->unp_msgcount--; 2077 ret = 1; 2078 } else 2079 ret = 0; 2080 unp_rights--; 2081 UNP_LINK_WUNLOCK(); 2082 return (ret); 2083} 2084 2085/* 2086 * unp_defer indicates whether additional work has been defered for a future 2087 * pass through unp_gc(). It is thread local and does not require explicit 2088 * synchronization. 2089 */ 2090static int unp_marked; 2091static int unp_unreachable; 2092 2093static void 2094unp_accessable(struct file *fp) 2095{ 2096 struct unpcb *unp; 2097 2098 if ((unp = fptounp(fp)) == NULL) 2099 return; 2100 if (unp->unp_gcflag & UNPGC_REF) 2101 return; 2102 unp->unp_gcflag &= ~UNPGC_DEAD; 2103 unp->unp_gcflag |= UNPGC_REF; 2104 unp_marked++; 2105} 2106 2107static void 2108unp_gc_process(struct unpcb *unp) 2109{ 2110 struct socket *soa; 2111 struct socket *so; 2112 struct file *fp; 2113 2114 /* Already processed. */ 2115 if (unp->unp_gcflag & UNPGC_SCANNED) 2116 return; 2117 fp = unp->unp_file; 2118 2119 /* 2120 * Check for a socket potentially in a cycle. It must be in a 2121 * queue as indicated by msgcount, and this must equal the file 2122 * reference count. Note that when msgcount is 0 the file is NULL. 2123 */ 2124 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2125 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2126 unp->unp_gcflag |= UNPGC_DEAD; 2127 unp_unreachable++; 2128 return; 2129 } 2130 2131 /* 2132 * Mark all sockets we reference with RIGHTS. 2133 */ 2134 so = unp->unp_socket; 2135 SOCKBUF_LOCK(&so->so_rcv); 2136 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2137 SOCKBUF_UNLOCK(&so->so_rcv); 2138 2139 /* 2140 * Mark all sockets in our accept queue. 2141 */ 2142 ACCEPT_LOCK(); 2143 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2144 SOCKBUF_LOCK(&soa->so_rcv); 2145 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2146 SOCKBUF_UNLOCK(&soa->so_rcv); 2147 } 2148 ACCEPT_UNLOCK(); 2149 unp->unp_gcflag |= UNPGC_SCANNED; 2150} 2151 2152static int unp_recycled; 2153SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2154 "Number of unreachable sockets claimed by the garbage collector."); 2155 2156static int unp_taskcount; 2157SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2158 "Number of times the garbage collector has run."); 2159 2160static void 2161unp_gc(__unused void *arg, int pending) 2162{ 2163 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2164 NULL }; 2165 struct unp_head **head; 2166 struct file *f, **unref; 2167 struct unpcb *unp; 2168 int i, total; 2169 2170 unp_taskcount++; 2171 UNP_LIST_LOCK(); 2172 /* 2173 * First clear all gc flags from previous runs. 2174 */ 2175 for (head = heads; *head != NULL; head++) 2176 LIST_FOREACH(unp, *head, unp_link) 2177 unp->unp_gcflag = 0; 2178 2179 /* 2180 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2181 * is reachable all of the sockets it references are reachable. 2182 * Stop the scan once we do a complete loop without discovering 2183 * a new reachable socket. 2184 */ 2185 do { 2186 unp_unreachable = 0; 2187 unp_marked = 0; 2188 for (head = heads; *head != NULL; head++) 2189 LIST_FOREACH(unp, *head, unp_link) 2190 unp_gc_process(unp); 2191 } while (unp_marked); 2192 UNP_LIST_UNLOCK(); 2193 if (unp_unreachable == 0) 2194 return; 2195 2196 /* 2197 * Allocate space for a local list of dead unpcbs. 2198 */ 2199 unref = malloc(unp_unreachable * sizeof(struct file *), 2200 M_TEMP, M_WAITOK); 2201 2202 /* 2203 * Iterate looking for sockets which have been specifically marked 2204 * as as unreachable and store them locally. 2205 */ 2206 UNP_LINK_RLOCK(); 2207 UNP_LIST_LOCK(); 2208 for (total = 0, head = heads; *head != NULL; head++) 2209 LIST_FOREACH(unp, *head, unp_link) 2210 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) { 2211 f = unp->unp_file; 2212 if (unp->unp_msgcount == 0 || f == NULL || 2213 f->f_count != unp->unp_msgcount) 2214 continue; 2215 unref[total++] = f; 2216 fhold(f); 2217 KASSERT(total <= unp_unreachable, 2218 ("unp_gc: incorrect unreachable count.")); 2219 } 2220 UNP_LIST_UNLOCK(); 2221 UNP_LINK_RUNLOCK(); 2222 2223 /* 2224 * Now flush all sockets, free'ing rights. This will free the 2225 * struct files associated with these sockets but leave each socket 2226 * with one remaining ref. 2227 */ 2228 for (i = 0; i < total; i++) { 2229 struct socket *so; 2230 2231 so = unref[i]->f_data; 2232 CURVNET_SET(so->so_vnet); 2233 sorflush(so); 2234 CURVNET_RESTORE(); 2235 } 2236 2237 /* 2238 * And finally release the sockets so they can be reclaimed. 2239 */ 2240 for (i = 0; i < total; i++) 2241 fdrop(unref[i], NULL); 2242 unp_recycled += total; 2243 free(unref, M_TEMP); 2244} 2245 2246static void 2247unp_dispose(struct mbuf *m) 2248{ 2249 2250 if (m) 2251 unp_scan(m, unp_discard); 2252} 2253 2254static void 2255unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2256{ 2257 struct mbuf *m; 2258 struct file **rp; 2259 struct cmsghdr *cm; 2260 void *data; 2261 int i; 2262 socklen_t clen, datalen; 2263 int qfds; 2264 2265 while (m0 != NULL) { 2266 for (m = m0; m; m = m->m_next) { 2267 if (m->m_type != MT_CONTROL) 2268 continue; 2269 2270 cm = mtod(m, struct cmsghdr *); 2271 clen = m->m_len; 2272 2273 while (cm != NULL) { 2274 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2275 break; 2276 2277 data = CMSG_DATA(cm); 2278 datalen = (caddr_t)cm + cm->cmsg_len 2279 - (caddr_t)data; 2280 2281 if (cm->cmsg_level == SOL_SOCKET && 2282 cm->cmsg_type == SCM_RIGHTS) { 2283 qfds = datalen / sizeof (struct file *); 2284 rp = data; 2285 for (i = 0; i < qfds; i++) 2286 (*op)(*rp++); 2287 } 2288 2289 if (CMSG_SPACE(datalen) < clen) { 2290 clen -= CMSG_SPACE(datalen); 2291 cm = (struct cmsghdr *) 2292 ((caddr_t)cm + CMSG_SPACE(datalen)); 2293 } else { 2294 clen = 0; 2295 cm = NULL; 2296 } 2297 } 2298 } 2299 m0 = m0->m_act; 2300 } 2301} 2302 2303/* 2304 * A helper function called by VFS before socket-type vnode reclamation. 2305 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2306 * use count. 2307 */ 2308void 2309vfs_unp_reclaim(struct vnode *vp) 2310{ 2311 struct socket *so; 2312 struct unpcb *unp; 2313 int active; 2314 2315 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2316 KASSERT(vp->v_type == VSOCK, 2317 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2318 2319 active = 0; 2320 UNP_LINK_WLOCK(); 2321 so = vp->v_socket; 2322 if (so == NULL) 2323 goto done; 2324 unp = sotounpcb(so); 2325 if (unp == NULL) 2326 goto done; 2327 UNP_PCB_LOCK(unp); 2328 if (unp->unp_vnode != NULL) { 2329 KASSERT(unp->unp_vnode == vp, 2330 ("vfs_unp_reclaim: vp != unp->unp_vnode")); 2331 vp->v_socket = NULL; 2332 unp->unp_vnode = NULL; 2333 active = 1; 2334 } 2335 UNP_PCB_UNLOCK(unp); 2336done: 2337 UNP_LINK_WUNLOCK(); 2338 if (active) 2339 vunref(vp); 2340} 2341 2342#ifdef DDB 2343static void 2344db_print_indent(int indent) 2345{ 2346 int i; 2347 2348 for (i = 0; i < indent; i++) 2349 db_printf(" "); 2350} 2351 2352static void 2353db_print_unpflags(int unp_flags) 2354{ 2355 int comma; 2356 2357 comma = 0; 2358 if (unp_flags & UNP_HAVEPC) { 2359 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2360 comma = 1; 2361 } 2362 if (unp_flags & UNP_HAVEPCCACHED) { 2363 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2364 comma = 1; 2365 } 2366 if (unp_flags & UNP_WANTCRED) { 2367 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2368 comma = 1; 2369 } 2370 if (unp_flags & UNP_CONNWAIT) { 2371 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2372 comma = 1; 2373 } 2374 if (unp_flags & UNP_CONNECTING) { 2375 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2376 comma = 1; 2377 } 2378 if (unp_flags & UNP_BINDING) { 2379 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2380 comma = 1; 2381 } 2382} 2383 2384static void 2385db_print_xucred(int indent, struct xucred *xu) 2386{ 2387 int comma, i; 2388 2389 db_print_indent(indent); 2390 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2391 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2392 db_print_indent(indent); 2393 db_printf("cr_groups: "); 2394 comma = 0; 2395 for (i = 0; i < xu->cr_ngroups; i++) { 2396 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2397 comma = 1; 2398 } 2399 db_printf("\n"); 2400} 2401 2402static void 2403db_print_unprefs(int indent, struct unp_head *uh) 2404{ 2405 struct unpcb *unp; 2406 int counter; 2407 2408 counter = 0; 2409 LIST_FOREACH(unp, uh, unp_reflink) { 2410 if (counter % 4 == 0) 2411 db_print_indent(indent); 2412 db_printf("%p ", unp); 2413 if (counter % 4 == 3) 2414 db_printf("\n"); 2415 counter++; 2416 } 2417 if (counter != 0 && counter % 4 != 0) 2418 db_printf("\n"); 2419} 2420 2421DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2422{ 2423 struct unpcb *unp; 2424 2425 if (!have_addr) { 2426 db_printf("usage: show unpcb <addr>\n"); 2427 return; 2428 } 2429 unp = (struct unpcb *)addr; 2430 2431 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2432 unp->unp_vnode); 2433 2434 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2435 unp->unp_conn); 2436 2437 db_printf("unp_refs:\n"); 2438 db_print_unprefs(2, &unp->unp_refs); 2439 2440 /* XXXRW: Would be nice to print the full address, if any. */ 2441 db_printf("unp_addr: %p\n", unp->unp_addr); 2442 2443 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2444 unp->unp_cc, unp->unp_mbcnt, 2445 (unsigned long long)unp->unp_gencnt); 2446 2447 db_printf("unp_flags: %x (", unp->unp_flags); 2448 db_print_unpflags(unp->unp_flags); 2449 db_printf(")\n"); 2450 2451 db_printf("unp_peercred:\n"); 2452 db_print_xucred(2, &unp->unp_peercred); 2453 2454 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2455} 2456#endif 2457