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