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