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/resourcevar.h> 80 #include <sys/rwlock.h> 81 #include <sys/socket.h> 82 #include <sys/socketvar.h> 83 #include <sys/signalvar.h> 84 #include <sys/stat.h> 85 #include <sys/sx.h> 86 #include <sys/sysctl.h> 87 #include <sys/systm.h> 88 #include <sys/taskqueue.h> 89 #include <sys/un.h> 90 #include <sys/unpcb.h> 91 #include <sys/vnode.h> 92 93 #include <net/vnet.h> 94 95 #ifdef DDB 96 #include <ddb/ddb.h> 97 #endif 98 99 #include <security/mac/mac_framework.h> 100 101 #include <vm/uma.h> 102 103 /* 104 * Locking key: 105 * (l) Locked using list lock 106 * (g) Locked using linkage lock 107 */ 108 109 static uma_zone_t unp_zone; 110 static unp_gen_t unp_gencnt; /* (l) */ 111 static u_int unp_count; /* (l) Count of local sockets. */ 112 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 113 static int unp_rights; /* (g) File descriptors in flight. */ 114 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 115 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 116 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 117 118 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 119 120 /* 121 * Garbage collection of cyclic file descriptor/socket references occurs 122 * asynchronously in a taskqueue context in order to avoid recursion and 123 * reentrance in the UNIX domain socket, file descriptor, and socket layer 124 * code. See unp_gc() for a full description. 125 */ 126 static struct task unp_gc_task; 127 128 /* 129 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 130 * stream sockets, although the total for sender and receiver is actually 131 * only PIPSIZ. 132 * 133 * Datagram sockets really use the sendspace as the maximum datagram size, 134 * and don't really want to reserve the sendspace. Their recvspace should be 135 * large enough for at least one max-size datagram plus address. 136 */ 137 #ifndef PIPSIZ 138 #define PIPSIZ 8192 139 #endif 140 static u_long unpst_sendspace = PIPSIZ; 141 static u_long unpst_recvspace = PIPSIZ; 142 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 143 static u_long unpdg_recvspace = 4*1024; 144 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 145 static u_long unpsp_recvspace = PIPSIZ; 146 147 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 148 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 149 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 150 SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0, 151 "SOCK_SEQPACKET"); 152 153 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 154 &unpst_sendspace, 0, "Default stream send space."); 155 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 156 &unpst_recvspace, 0, "Default stream receive space."); 157 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 158 &unpdg_sendspace, 0, "Default datagram send space."); 159 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 160 &unpdg_recvspace, 0, "Default datagram receive space."); 161 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 162 &unpsp_sendspace, 0, "Default seqpacket send space."); 163 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 164 &unpsp_recvspace, 0, "Default seqpacket receive space."); 165 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 166 "File descriptors in flight."); 167 168 /*- 169 * Locking and synchronization: 170 * 171 * Three types of locks exit in the local domain socket implementation: a 172 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 173 * global locks, the list lock protects the socket count, global generation 174 * number, and stream/datagram global lists. The linkage lock protects the 175 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 176 * held exclusively over the acquisition of multiple unpcb locks to prevent 177 * deadlock. 178 * 179 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 180 * allocated in pru_attach() and freed in pru_detach(). The validity of that 181 * pointer is an invariant, so no lock is required to dereference the so_pcb 182 * pointer if a valid socket reference is held by the caller. In practice, 183 * this is always true during operations performed on a socket. Each unpcb 184 * has a back-pointer to its socket, unp_socket, which will be stable under 185 * the same circumstances. 186 * 187 * This pointer may only be safely dereferenced as long as a valid reference 188 * to the unpcb is held. Typically, this reference will be from the socket, 189 * or from another unpcb when the referring unpcb's lock is held (in order 190 * that the reference not be invalidated during use). For example, to follow 191 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 192 * as unp_socket remains valid as long as the reference to unp_conn is valid. 193 * 194 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 195 * atomic reads without the lock may be performed "lockless", but more 196 * complex reads and read-modify-writes require the mutex to be held. No 197 * lock order is defined between unpcb locks -- multiple unpcb locks may be 198 * acquired at the same time only when holding the linkage rwlock 199 * exclusively, which prevents deadlocks. 200 * 201 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 202 * protocols, bind() is a non-atomic operation, and connect() requires 203 * potential sleeping in the protocol, due to potentially waiting on local or 204 * distributed file systems. We try to separate "lookup" operations, which 205 * may sleep, and the IPC operations themselves, which typically can occur 206 * with relative atomicity as locks can be held over the entire operation. 207 * 208 * Another tricky issue is simultaneous multi-threaded or multi-process 209 * access to a single UNIX domain socket. These are handled by the flags 210 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 211 * binding, both of which involve dropping UNIX domain socket locks in order 212 * to perform namei() and other file system operations. 213 */ 214 static struct rwlock unp_link_rwlock; 215 static struct mtx unp_list_lock; 216 217 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 218 "unp_link_rwlock") 219 220 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 221 RA_LOCKED) 222 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 223 RA_UNLOCKED) 224 225 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 226 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 227 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 228 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 229 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 230 RA_WLOCKED) 231 232 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 233 "unp_list_lock", NULL, MTX_DEF) 234 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 235 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 236 237 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 238 "unp_mtx", "unp_mtx", \ 239 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 240 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 241 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 242 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 243 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 244 245 static int uipc_connect2(struct socket *, struct socket *); 246 static int uipc_ctloutput(struct socket *, struct sockopt *); 247 static int unp_connect(struct socket *, struct sockaddr *, 248 struct thread *); 249 static int unp_connect2(struct socket *so, struct socket *so2, int); 250 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 251 static void unp_dispose(struct mbuf *); 252 static void unp_shutdown(struct unpcb *); 253 static void unp_drop(struct unpcb *, int); 254 static void unp_gc(__unused void *, int); 255 static void unp_scan(struct mbuf *, void (*)(struct file *)); 256 static void unp_discard(struct file *); 257 static void unp_freerights(struct file **, int); 258 static void unp_init(void); 259 static int unp_internalize(struct mbuf **, struct thread *); 260 static void unp_internalize_fp(struct file *); 261 static int unp_externalize(struct mbuf *, struct mbuf **); 262 static void unp_externalize_fp(struct file *); 263 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 264 265 /* 266 * Definitions of protocols supported in the LOCAL domain. 267 */ 268 static struct domain localdomain; 269 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 270 static struct pr_usrreqs uipc_usrreqs_seqpacket; 271 static struct protosw localsw[] = { 272 { 273 .pr_type = SOCK_STREAM, 274 .pr_domain = &localdomain, 275 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 276 .pr_ctloutput = &uipc_ctloutput, 277 .pr_usrreqs = &uipc_usrreqs_stream 278 }, 279 { 280 .pr_type = SOCK_DGRAM, 281 .pr_domain = &localdomain, 282 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 283 .pr_usrreqs = &uipc_usrreqs_dgram 284 }, 285 { 286 .pr_type = SOCK_SEQPACKET, 287 .pr_domain = &localdomain, 288 289 /* 290 * XXXRW: For now, PR_ADDR because soreceive will bump into them 291 * due to our use of sbappendaddr. A new sbappend variants is needed 292 * that supports both atomic record writes and control data. 293 */ 294 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 295 PR_RIGHTS, 296 .pr_usrreqs = &uipc_usrreqs_seqpacket, 297 }, 298 }; 299 300 static struct domain localdomain = { 301 .dom_family = AF_LOCAL, 302 .dom_name = "local", 303 .dom_init = unp_init, 304 .dom_externalize = unp_externalize, 305 .dom_dispose = unp_dispose, 306 .dom_protosw = localsw, 307 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 308 }; 309 DOMAIN_SET(local); 310 311 static void 312 uipc_abort(struct socket *so) 313 { 314 struct unpcb *unp, *unp2; 315 316 unp = sotounpcb(so); 317 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 318 319 UNP_LINK_WLOCK(); 320 UNP_PCB_LOCK(unp); 321 unp2 = unp->unp_conn; 322 if (unp2 != NULL) { 323 UNP_PCB_LOCK(unp2); 324 unp_drop(unp2, ECONNABORTED); 325 UNP_PCB_UNLOCK(unp2); 326 } 327 UNP_PCB_UNLOCK(unp); 328 UNP_LINK_WUNLOCK(); 329 } 330 331 static int 332 uipc_accept(struct socket *so, struct sockaddr **nam) 333 { 334 struct unpcb *unp, *unp2; 335 const struct sockaddr *sa; 336 337 /* 338 * Pass back name of connected socket, if it was bound and we are 339 * still connected (our peer may have closed already!). 340 */ 341 unp = sotounpcb(so); 342 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 343 344 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 345 UNP_LINK_RLOCK(); 346 unp2 = unp->unp_conn; 347 if (unp2 != NULL && unp2->unp_addr != NULL) { 348 UNP_PCB_LOCK(unp2); 349 sa = (struct sockaddr *) unp2->unp_addr; 350 bcopy(sa, *nam, sa->sa_len); 351 UNP_PCB_UNLOCK(unp2); 352 } else { 353 sa = &sun_noname; 354 bcopy(sa, *nam, sa->sa_len); 355 } 356 UNP_LINK_RUNLOCK(); 357 return (0); 358 } 359 360 static int 361 uipc_attach(struct socket *so, int proto, struct thread *td) 362 { 363 u_long sendspace, recvspace; 364 struct unpcb *unp; 365 int error; 366 367 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 368 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 369 switch (so->so_type) { 370 case SOCK_STREAM: 371 sendspace = unpst_sendspace; 372 recvspace = unpst_recvspace; 373 break; 374 375 case SOCK_DGRAM: 376 sendspace = unpdg_sendspace; 377 recvspace = unpdg_recvspace; 378 break; 379 380 case SOCK_SEQPACKET: 381 sendspace = unpsp_sendspace; 382 recvspace = unpsp_recvspace; 383 break; 384 385 default: 386 panic("uipc_attach"); 387 } 388 error = soreserve(so, sendspace, recvspace); 389 if (error) 390 return (error); 391 } 392 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 393 if (unp == NULL) 394 return (ENOBUFS); 395 LIST_INIT(&unp->unp_refs); 396 UNP_PCB_LOCK_INIT(unp); 397 unp->unp_socket = so; 398 so->so_pcb = unp; 399 unp->unp_refcount = 1; 400 401 UNP_LIST_LOCK(); 402 unp->unp_gencnt = ++unp_gencnt; 403 unp_count++; 404 switch (so->so_type) { 405 case SOCK_STREAM: 406 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 407 break; 408 409 case SOCK_DGRAM: 410 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 411 break; 412 413 case SOCK_SEQPACKET: 414 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 415 break; 416 417 default: 418 panic("uipc_attach"); 419 } 420 UNP_LIST_UNLOCK(); 421 422 return (0); 423 } 424 425 static int 426 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 427 { 428 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 429 struct vattr vattr; 430 int error, namelen, vfslocked; 431 struct nameidata nd; 432 struct unpcb *unp; 433 struct vnode *vp; 434 struct mount *mp; 435 char *buf; 436 437 unp = sotounpcb(so); 438 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 439 440 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 441 if (namelen <= 0) 442 return (EINVAL); 443 444 /* 445 * We don't allow simultaneous bind() calls on a single UNIX domain 446 * socket, so flag in-progress operations, and return an error if an 447 * operation is already in progress. 448 * 449 * Historically, we have not allowed a socket to be rebound, so this 450 * also returns an error. Not allowing re-binding simplifies the 451 * implementation and avoids a great many possible failure modes. 452 */ 453 UNP_PCB_LOCK(unp); 454 if (unp->unp_vnode != NULL) { 455 UNP_PCB_UNLOCK(unp); 456 return (EINVAL); 457 } 458 if (unp->unp_flags & UNP_BINDING) { 459 UNP_PCB_UNLOCK(unp); 460 return (EALREADY); 461 } 462 unp->unp_flags |= UNP_BINDING; 463 UNP_PCB_UNLOCK(unp); 464 465 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 466 bcopy(soun->sun_path, buf, namelen); 467 buf[namelen] = 0; 468 469 restart: 470 vfslocked = 0; 471 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 472 UIO_SYSSPACE, buf, td); 473 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 474 error = namei(&nd); 475 if (error) 476 goto error; 477 vp = nd.ni_vp; 478 vfslocked = NDHASGIANT(&nd); 479 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 480 NDFREE(&nd, NDF_ONLY_PNBUF); 481 if (nd.ni_dvp == vp) 482 vrele(nd.ni_dvp); 483 else 484 vput(nd.ni_dvp); 485 if (vp != NULL) { 486 vrele(vp); 487 error = EADDRINUSE; 488 goto error; 489 } 490 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 491 if (error) 492 goto error; 493 VFS_UNLOCK_GIANT(vfslocked); 494 goto restart; 495 } 496 VATTR_NULL(&vattr); 497 vattr.va_type = VSOCK; 498 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 499 #ifdef MAC 500 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 501 &vattr); 502 #endif 503 if (error == 0) 504 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 505 NDFREE(&nd, NDF_ONLY_PNBUF); 506 vput(nd.ni_dvp); 507 if (error) { 508 vn_finished_write(mp); 509 goto error; 510 } 511 vp = nd.ni_vp; 512 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 513 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 514 515 UNP_LINK_WLOCK(); 516 UNP_PCB_LOCK(unp); 517 vp->v_socket = unp->unp_socket; 518 unp->unp_vnode = vp; 519 unp->unp_addr = soun; 520 unp->unp_flags &= ~UNP_BINDING; 521 UNP_PCB_UNLOCK(unp); 522 UNP_LINK_WUNLOCK(); 523 VOP_UNLOCK(vp, 0); 524 vn_finished_write(mp); 525 VFS_UNLOCK_GIANT(vfslocked); 526 free(buf, M_TEMP); 527 return (0); 528 529 error: 530 VFS_UNLOCK_GIANT(vfslocked); 531 UNP_PCB_LOCK(unp); 532 unp->unp_flags &= ~UNP_BINDING; 533 UNP_PCB_UNLOCK(unp); 534 free(buf, M_TEMP); 535 return (error); 536 } 537 538 static int 539 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 540 { 541 int error; 542 543 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 544 UNP_LINK_WLOCK(); 545 error = unp_connect(so, nam, td); 546 UNP_LINK_WUNLOCK(); 547 return (error); 548 } 549 550 static void 551 uipc_close(struct socket *so) 552 { 553 struct unpcb *unp, *unp2; 554 555 unp = sotounpcb(so); 556 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 557 558 UNP_LINK_WLOCK(); 559 UNP_PCB_LOCK(unp); 560 unp2 = unp->unp_conn; 561 if (unp2 != NULL) { 562 UNP_PCB_LOCK(unp2); 563 unp_disconnect(unp, unp2); 564 UNP_PCB_UNLOCK(unp2); 565 } 566 UNP_PCB_UNLOCK(unp); 567 UNP_LINK_WUNLOCK(); 568 } 569 570 static int 571 uipc_connect2(struct socket *so1, struct socket *so2) 572 { 573 struct unpcb *unp, *unp2; 574 int error; 575 576 UNP_LINK_WLOCK(); 577 unp = so1->so_pcb; 578 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 579 UNP_PCB_LOCK(unp); 580 unp2 = so2->so_pcb; 581 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 582 UNP_PCB_LOCK(unp2); 583 error = unp_connect2(so1, so2, PRU_CONNECT2); 584 UNP_PCB_UNLOCK(unp2); 585 UNP_PCB_UNLOCK(unp); 586 UNP_LINK_WUNLOCK(); 587 return (error); 588 } 589 590 static void 591 uipc_detach(struct socket *so) 592 { 593 struct unpcb *unp, *unp2; 594 struct sockaddr_un *saved_unp_addr; 595 struct vnode *vp; 596 int freeunp, local_unp_rights; 597 598 unp = sotounpcb(so); 599 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 600 601 UNP_LINK_WLOCK(); 602 UNP_LIST_LOCK(); 603 UNP_PCB_LOCK(unp); 604 LIST_REMOVE(unp, unp_link); 605 unp->unp_gencnt = ++unp_gencnt; 606 --unp_count; 607 UNP_LIST_UNLOCK(); 608 609 /* 610 * XXXRW: Should assert vp->v_socket == so. 611 */ 612 if ((vp = unp->unp_vnode) != NULL) { 613 unp->unp_vnode->v_socket = NULL; 614 unp->unp_vnode = NULL; 615 } 616 unp2 = unp->unp_conn; 617 if (unp2 != NULL) { 618 UNP_PCB_LOCK(unp2); 619 unp_disconnect(unp, unp2); 620 UNP_PCB_UNLOCK(unp2); 621 } 622 623 /* 624 * We hold the linkage lock exclusively, so it's OK to acquire 625 * multiple pcb locks at a time. 626 */ 627 while (!LIST_EMPTY(&unp->unp_refs)) { 628 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 629 630 UNP_PCB_LOCK(ref); 631 unp_drop(ref, ECONNRESET); 632 UNP_PCB_UNLOCK(ref); 633 } 634 local_unp_rights = unp_rights; 635 UNP_LINK_WUNLOCK(); 636 unp->unp_socket->so_pcb = NULL; 637 saved_unp_addr = unp->unp_addr; 638 unp->unp_addr = NULL; 639 unp->unp_refcount--; 640 freeunp = (unp->unp_refcount == 0); 641 if (saved_unp_addr != NULL) 642 free(saved_unp_addr, M_SONAME); 643 if (freeunp) { 644 UNP_PCB_LOCK_DESTROY(unp); 645 uma_zfree(unp_zone, unp); 646 } else 647 UNP_PCB_UNLOCK(unp); 648 if (vp) { 649 int vfslocked; 650 651 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 652 vrele(vp); 653 VFS_UNLOCK_GIANT(vfslocked); 654 } 655 if (local_unp_rights) 656 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 657 } 658 659 static int 660 uipc_disconnect(struct socket *so) 661 { 662 struct unpcb *unp, *unp2; 663 664 unp = sotounpcb(so); 665 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 666 667 UNP_LINK_WLOCK(); 668 UNP_PCB_LOCK(unp); 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 UNP_PCB_UNLOCK(unp); 676 UNP_LINK_WUNLOCK(); 677 return (0); 678 } 679 680 static int 681 uipc_listen(struct socket *so, int backlog, struct thread *td) 682 { 683 struct unpcb *unp; 684 int error; 685 686 unp = sotounpcb(so); 687 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 688 689 UNP_PCB_LOCK(unp); 690 if (unp->unp_vnode == NULL) { 691 UNP_PCB_UNLOCK(unp); 692 return (EINVAL); 693 } 694 695 SOCK_LOCK(so); 696 error = solisten_proto_check(so); 697 if (error == 0) { 698 cru2x(td->td_ucred, &unp->unp_peercred); 699 unp->unp_flags |= UNP_HAVEPCCACHED; 700 solisten_proto(so, backlog); 701 } 702 SOCK_UNLOCK(so); 703 UNP_PCB_UNLOCK(unp); 704 return (error); 705 } 706 707 static int 708 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 709 { 710 struct unpcb *unp, *unp2; 711 const struct sockaddr *sa; 712 713 unp = sotounpcb(so); 714 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 715 716 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 717 UNP_LINK_RLOCK(); 718 /* 719 * XXX: It seems that this test always fails even when connection is 720 * established. So, this else clause is added as workaround to 721 * return PF_LOCAL sockaddr. 722 */ 723 unp2 = unp->unp_conn; 724 if (unp2 != NULL) { 725 UNP_PCB_LOCK(unp2); 726 if (unp2->unp_addr != NULL) 727 sa = (struct sockaddr *) unp2->unp_addr; 728 else 729 sa = &sun_noname; 730 bcopy(sa, *nam, sa->sa_len); 731 UNP_PCB_UNLOCK(unp2); 732 } else { 733 sa = &sun_noname; 734 bcopy(sa, *nam, sa->sa_len); 735 } 736 UNP_LINK_RUNLOCK(); 737 return (0); 738 } 739 740 static int 741 uipc_rcvd(struct socket *so, int flags) 742 { 743 struct unpcb *unp, *unp2; 744 struct socket *so2; 745 u_int mbcnt, sbcc; 746 u_long newhiwat; 747 748 unp = sotounpcb(so); 749 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 750 751 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 752 panic("uipc_rcvd socktype %d", so->so_type); 753 754 /* 755 * Adjust backpressure on sender and wakeup any waiting to write. 756 * 757 * The unp lock is acquired to maintain the validity of the unp_conn 758 * pointer; no lock on unp2 is required as unp2->unp_socket will be 759 * static as long as we don't permit unp2 to disconnect from unp, 760 * which is prevented by the lock on unp. We cache values from 761 * so_rcv to avoid holding the so_rcv lock over the entire 762 * transaction on the remote so_snd. 763 */ 764 SOCKBUF_LOCK(&so->so_rcv); 765 mbcnt = so->so_rcv.sb_mbcnt; 766 sbcc = so->so_rcv.sb_cc; 767 SOCKBUF_UNLOCK(&so->so_rcv); 768 UNP_PCB_LOCK(unp); 769 unp2 = unp->unp_conn; 770 if (unp2 == NULL) { 771 UNP_PCB_UNLOCK(unp); 772 return (0); 773 } 774 so2 = unp2->unp_socket; 775 SOCKBUF_LOCK(&so2->so_snd); 776 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 777 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 778 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 779 newhiwat, RLIM_INFINITY); 780 sowwakeup_locked(so2); 781 unp->unp_mbcnt = mbcnt; 782 unp->unp_cc = sbcc; 783 UNP_PCB_UNLOCK(unp); 784 return (0); 785 } 786 787 static int 788 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 789 struct mbuf *control, struct thread *td) 790 { 791 struct unpcb *unp, *unp2; 792 struct socket *so2; 793 u_int mbcnt_delta, sbcc; 794 u_long newhiwat; 795 int error = 0; 796 797 unp = sotounpcb(so); 798 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 799 800 if (flags & PRUS_OOB) { 801 error = EOPNOTSUPP; 802 goto release; 803 } 804 if (control != NULL && (error = unp_internalize(&control, td))) 805 goto release; 806 if ((nam != NULL) || (flags & PRUS_EOF)) 807 UNP_LINK_WLOCK(); 808 else 809 UNP_LINK_RLOCK(); 810 switch (so->so_type) { 811 case SOCK_DGRAM: 812 { 813 const struct sockaddr *from; 814 815 unp2 = unp->unp_conn; 816 if (nam != NULL) { 817 UNP_LINK_WLOCK_ASSERT(); 818 if (unp2 != NULL) { 819 error = EISCONN; 820 break; 821 } 822 error = unp_connect(so, nam, td); 823 if (error) 824 break; 825 unp2 = unp->unp_conn; 826 } 827 828 /* 829 * Because connect() and send() are non-atomic in a sendto() 830 * with a target address, it's possible that the socket will 831 * have disconnected before the send() can run. In that case 832 * return the slightly counter-intuitive but otherwise 833 * correct error that the socket is not connected. 834 */ 835 if (unp2 == NULL) { 836 error = ENOTCONN; 837 break; 838 } 839 /* Lockless read. */ 840 if (unp2->unp_flags & UNP_WANTCRED) 841 control = unp_addsockcred(td, control); 842 UNP_PCB_LOCK(unp); 843 if (unp->unp_addr != NULL) 844 from = (struct sockaddr *)unp->unp_addr; 845 else 846 from = &sun_noname; 847 so2 = unp2->unp_socket; 848 SOCKBUF_LOCK(&so2->so_rcv); 849 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 850 sorwakeup_locked(so2); 851 m = NULL; 852 control = NULL; 853 } else { 854 SOCKBUF_UNLOCK(&so2->so_rcv); 855 error = ENOBUFS; 856 } 857 if (nam != NULL) { 858 UNP_LINK_WLOCK_ASSERT(); 859 UNP_PCB_LOCK(unp2); 860 unp_disconnect(unp, unp2); 861 UNP_PCB_UNLOCK(unp2); 862 } 863 UNP_PCB_UNLOCK(unp); 864 break; 865 } 866 867 case SOCK_SEQPACKET: 868 case SOCK_STREAM: 869 if ((so->so_state & SS_ISCONNECTED) == 0) { 870 if (nam != NULL) { 871 UNP_LINK_WLOCK_ASSERT(); 872 error = unp_connect(so, nam, td); 873 if (error) 874 break; /* XXX */ 875 } else { 876 error = ENOTCONN; 877 break; 878 } 879 } 880 881 /* Lockless read. */ 882 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 883 error = EPIPE; 884 break; 885 } 886 887 /* 888 * Because connect() and send() are non-atomic in a sendto() 889 * with a target address, it's possible that the socket will 890 * have disconnected before the send() can run. In that case 891 * return the slightly counter-intuitive but otherwise 892 * correct error that the socket is not connected. 893 * 894 * Locking here must be done carefully: the linkage lock 895 * prevents interconnections between unpcbs from changing, so 896 * we can traverse from unp to unp2 without acquiring unp's 897 * lock. Socket buffer locks follow unpcb locks, so we can 898 * acquire both remote and lock socket buffer locks. 899 */ 900 unp2 = unp->unp_conn; 901 if (unp2 == NULL) { 902 error = ENOTCONN; 903 break; 904 } 905 so2 = unp2->unp_socket; 906 UNP_PCB_LOCK(unp2); 907 SOCKBUF_LOCK(&so2->so_rcv); 908 if (unp2->unp_flags & UNP_WANTCRED) { 909 /* 910 * Credentials are passed only once on SOCK_STREAM. 911 */ 912 unp2->unp_flags &= ~UNP_WANTCRED; 913 control = unp_addsockcred(td, control); 914 } 915 /* 916 * Send to paired receive port, and then reduce send buffer 917 * hiwater marks to maintain backpressure. Wake up readers. 918 */ 919 switch (so->so_type) { 920 case SOCK_STREAM: 921 if (control != NULL) { 922 if (sbappendcontrol_locked(&so2->so_rcv, m, 923 control)) 924 control = NULL; 925 } else 926 sbappend_locked(&so2->so_rcv, m); 927 break; 928 929 case SOCK_SEQPACKET: { 930 const struct sockaddr *from; 931 932 from = &sun_noname; 933 if (sbappendaddr_locked(&so2->so_rcv, from, m, 934 control)) 935 control = NULL; 936 break; 937 } 938 } 939 940 /* 941 * XXXRW: While fine for SOCK_STREAM, this conflates maximum 942 * datagram size and back-pressure for SOCK_SEQPACKET, which 943 * can lead to undesired return of EMSGSIZE on send instead 944 * of more desirable blocking. 945 */ 946 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 947 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 948 sbcc = so2->so_rcv.sb_cc; 949 sorwakeup_locked(so2); 950 951 SOCKBUF_LOCK(&so->so_snd); 952 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 953 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 954 newhiwat, RLIM_INFINITY); 955 so->so_snd.sb_mbmax -= mbcnt_delta; 956 SOCKBUF_UNLOCK(&so->so_snd); 957 unp2->unp_cc = sbcc; 958 UNP_PCB_UNLOCK(unp2); 959 m = NULL; 960 break; 961 962 default: 963 panic("uipc_send unknown socktype"); 964 } 965 966 /* 967 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 968 */ 969 if (flags & PRUS_EOF) { 970 UNP_PCB_LOCK(unp); 971 socantsendmore(so); 972 unp_shutdown(unp); 973 UNP_PCB_UNLOCK(unp); 974 } 975 976 if ((nam != NULL) || (flags & PRUS_EOF)) 977 UNP_LINK_WUNLOCK(); 978 else 979 UNP_LINK_RUNLOCK(); 980 981 if (control != NULL && error != 0) 982 unp_dispose(control); 983 984 release: 985 if (control != NULL) 986 m_freem(control); 987 if (m != NULL) 988 m_freem(m); 989 return (error); 990 } 991 992 static int 993 uipc_sense(struct socket *so, struct stat *sb) 994 { 995 struct unpcb *unp, *unp2; 996 struct socket *so2; 997 998 unp = sotounpcb(so); 999 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1000 1001 sb->st_blksize = so->so_snd.sb_hiwat; 1002 UNP_LINK_RLOCK(); 1003 UNP_PCB_LOCK(unp); 1004 unp2 = unp->unp_conn; 1005 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) && 1006 unp2 != NULL) { 1007 so2 = unp2->unp_socket; 1008 sb->st_blksize += so2->so_rcv.sb_cc; 1009 } 1010 sb->st_dev = NODEV; 1011 if (unp->unp_ino == 0) 1012 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 1013 sb->st_ino = unp->unp_ino; 1014 UNP_PCB_UNLOCK(unp); 1015 UNP_LINK_RUNLOCK(); 1016 return (0); 1017 } 1018 1019 static int 1020 uipc_shutdown(struct socket *so) 1021 { 1022 struct unpcb *unp; 1023 1024 unp = sotounpcb(so); 1025 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1026 1027 UNP_LINK_WLOCK(); 1028 UNP_PCB_LOCK(unp); 1029 socantsendmore(so); 1030 unp_shutdown(unp); 1031 UNP_PCB_UNLOCK(unp); 1032 UNP_LINK_WUNLOCK(); 1033 return (0); 1034 } 1035 1036 static int 1037 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1038 { 1039 struct unpcb *unp; 1040 const struct sockaddr *sa; 1041 1042 unp = sotounpcb(so); 1043 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1044 1045 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1046 UNP_PCB_LOCK(unp); 1047 if (unp->unp_addr != NULL) 1048 sa = (struct sockaddr *) unp->unp_addr; 1049 else 1050 sa = &sun_noname; 1051 bcopy(sa, *nam, sa->sa_len); 1052 UNP_PCB_UNLOCK(unp); 1053 return (0); 1054 } 1055 1056 static struct pr_usrreqs uipc_usrreqs_dgram = { 1057 .pru_abort = uipc_abort, 1058 .pru_accept = uipc_accept, 1059 .pru_attach = uipc_attach, 1060 .pru_bind = uipc_bind, 1061 .pru_connect = uipc_connect, 1062 .pru_connect2 = uipc_connect2, 1063 .pru_detach = uipc_detach, 1064 .pru_disconnect = uipc_disconnect, 1065 .pru_listen = uipc_listen, 1066 .pru_peeraddr = uipc_peeraddr, 1067 .pru_rcvd = uipc_rcvd, 1068 .pru_send = uipc_send, 1069 .pru_sense = uipc_sense, 1070 .pru_shutdown = uipc_shutdown, 1071 .pru_sockaddr = uipc_sockaddr, 1072 .pru_soreceive = soreceive_dgram, 1073 .pru_close = uipc_close, 1074 }; 1075 1076 static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1077 .pru_abort = uipc_abort, 1078 .pru_accept = uipc_accept, 1079 .pru_attach = uipc_attach, 1080 .pru_bind = uipc_bind, 1081 .pru_connect = uipc_connect, 1082 .pru_connect2 = uipc_connect2, 1083 .pru_detach = uipc_detach, 1084 .pru_disconnect = uipc_disconnect, 1085 .pru_listen = uipc_listen, 1086 .pru_peeraddr = uipc_peeraddr, 1087 .pru_rcvd = uipc_rcvd, 1088 .pru_send = uipc_send, 1089 .pru_sense = uipc_sense, 1090 .pru_shutdown = uipc_shutdown, 1091 .pru_sockaddr = uipc_sockaddr, 1092 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1093 .pru_close = uipc_close, 1094 }; 1095 1096 static struct pr_usrreqs uipc_usrreqs_stream = { 1097 .pru_abort = uipc_abort, 1098 .pru_accept = uipc_accept, 1099 .pru_attach = uipc_attach, 1100 .pru_bind = uipc_bind, 1101 .pru_connect = uipc_connect, 1102 .pru_connect2 = uipc_connect2, 1103 .pru_detach = uipc_detach, 1104 .pru_disconnect = uipc_disconnect, 1105 .pru_listen = uipc_listen, 1106 .pru_peeraddr = uipc_peeraddr, 1107 .pru_rcvd = uipc_rcvd, 1108 .pru_send = uipc_send, 1109 .pru_sense = uipc_sense, 1110 .pru_shutdown = uipc_shutdown, 1111 .pru_sockaddr = uipc_sockaddr, 1112 .pru_soreceive = soreceive_generic, 1113 .pru_close = uipc_close, 1114 }; 1115 1116 static int 1117 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1118 { 1119 struct unpcb *unp; 1120 struct xucred xu; 1121 int error, optval; 1122 1123 if (sopt->sopt_level != 0) 1124 return (EINVAL); 1125 1126 unp = sotounpcb(so); 1127 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1128 error = 0; 1129 switch (sopt->sopt_dir) { 1130 case SOPT_GET: 1131 switch (sopt->sopt_name) { 1132 case LOCAL_PEERCRED: 1133 UNP_PCB_LOCK(unp); 1134 if (unp->unp_flags & UNP_HAVEPC) 1135 xu = unp->unp_peercred; 1136 else { 1137 if (so->so_type == SOCK_STREAM) 1138 error = ENOTCONN; 1139 else 1140 error = EINVAL; 1141 } 1142 UNP_PCB_UNLOCK(unp); 1143 if (error == 0) 1144 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1145 break; 1146 1147 case LOCAL_CREDS: 1148 /* Unlocked read. */ 1149 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1150 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1151 break; 1152 1153 case LOCAL_CONNWAIT: 1154 /* Unlocked read. */ 1155 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1156 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1157 break; 1158 1159 default: 1160 error = EOPNOTSUPP; 1161 break; 1162 } 1163 break; 1164 1165 case SOPT_SET: 1166 switch (sopt->sopt_name) { 1167 case LOCAL_CREDS: 1168 case LOCAL_CONNWAIT: 1169 error = sooptcopyin(sopt, &optval, sizeof(optval), 1170 sizeof(optval)); 1171 if (error) 1172 break; 1173 1174 #define OPTSET(bit) do { \ 1175 UNP_PCB_LOCK(unp); \ 1176 if (optval) \ 1177 unp->unp_flags |= bit; \ 1178 else \ 1179 unp->unp_flags &= ~bit; \ 1180 UNP_PCB_UNLOCK(unp); \ 1181 } while (0) 1182 1183 switch (sopt->sopt_name) { 1184 case LOCAL_CREDS: 1185 OPTSET(UNP_WANTCRED); 1186 break; 1187 1188 case LOCAL_CONNWAIT: 1189 OPTSET(UNP_CONNWAIT); 1190 break; 1191 1192 default: 1193 break; 1194 } 1195 break; 1196 #undef OPTSET 1197 default: 1198 error = ENOPROTOOPT; 1199 break; 1200 } 1201 break; 1202 1203 default: 1204 error = EOPNOTSUPP; 1205 break; 1206 } 1207 return (error); 1208 } 1209 1210 static int 1211 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1212 { 1213 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1214 struct vnode *vp; 1215 struct socket *so2, *so3; 1216 struct unpcb *unp, *unp2, *unp3; 1217 int error, len, vfslocked; 1218 struct nameidata nd; 1219 char buf[SOCK_MAXADDRLEN]; 1220 struct sockaddr *sa; 1221 1222 UNP_LINK_WLOCK_ASSERT(); 1223 1224 unp = sotounpcb(so); 1225 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1226 1227 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1228 if (len <= 0) 1229 return (EINVAL); 1230 bcopy(soun->sun_path, buf, len); 1231 buf[len] = 0; 1232 1233 UNP_PCB_LOCK(unp); 1234 if (unp->unp_flags & UNP_CONNECTING) { 1235 UNP_PCB_UNLOCK(unp); 1236 return (EALREADY); 1237 } 1238 UNP_LINK_WUNLOCK(); 1239 unp->unp_flags |= UNP_CONNECTING; 1240 UNP_PCB_UNLOCK(unp); 1241 1242 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1243 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, 1244 td); 1245 error = namei(&nd); 1246 if (error) 1247 vp = NULL; 1248 else 1249 vp = nd.ni_vp; 1250 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1251 vfslocked = NDHASGIANT(&nd); 1252 NDFREE(&nd, NDF_ONLY_PNBUF); 1253 if (error) 1254 goto bad; 1255 1256 if (vp->v_type != VSOCK) { 1257 error = ENOTSOCK; 1258 goto bad; 1259 } 1260 #ifdef MAC 1261 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1262 if (error) 1263 goto bad; 1264 #endif 1265 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1266 if (error) 1267 goto bad; 1268 VFS_UNLOCK_GIANT(vfslocked); 1269 1270 unp = sotounpcb(so); 1271 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1272 1273 /* 1274 * Lock linkage lock for two reasons: make sure v_socket is stable, 1275 * and to protect simultaneous locking of multiple pcbs. 1276 */ 1277 UNP_LINK_WLOCK(); 1278 so2 = vp->v_socket; 1279 if (so2 == NULL) { 1280 error = ECONNREFUSED; 1281 goto bad2; 1282 } 1283 if (so->so_type != so2->so_type) { 1284 error = EPROTOTYPE; 1285 goto bad2; 1286 } 1287 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1288 if (so2->so_options & SO_ACCEPTCONN) { 1289 so3 = sonewconn(so2, 0); 1290 } else 1291 so3 = NULL; 1292 if (so3 == NULL) { 1293 error = ECONNREFUSED; 1294 goto bad2; 1295 } 1296 unp = sotounpcb(so); 1297 unp2 = sotounpcb(so2); 1298 unp3 = sotounpcb(so3); 1299 UNP_PCB_LOCK(unp); 1300 UNP_PCB_LOCK(unp2); 1301 UNP_PCB_LOCK(unp3); 1302 if (unp2->unp_addr != NULL) { 1303 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1304 unp3->unp_addr = (struct sockaddr_un *) sa; 1305 sa = NULL; 1306 } 1307 1308 /* 1309 * The connecter's (client's) credentials are copied from its 1310 * process structure at the time of connect() (which is now). 1311 */ 1312 cru2x(td->td_ucred, &unp3->unp_peercred); 1313 unp3->unp_flags |= UNP_HAVEPC; 1314 1315 /* 1316 * The receiver's (server's) credentials are copied from the 1317 * unp_peercred member of socket on which the former called 1318 * listen(); uipc_listen() cached that process's credentials 1319 * at that time so we can use them now. 1320 */ 1321 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1322 ("unp_connect: listener without cached peercred")); 1323 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1324 sizeof(unp->unp_peercred)); 1325 unp->unp_flags |= UNP_HAVEPC; 1326 if (unp2->unp_flags & UNP_WANTCRED) 1327 unp3->unp_flags |= UNP_WANTCRED; 1328 UNP_PCB_UNLOCK(unp3); 1329 UNP_PCB_UNLOCK(unp2); 1330 UNP_PCB_UNLOCK(unp); 1331 #ifdef MAC 1332 mac_socketpeer_set_from_socket(so, so3); 1333 mac_socketpeer_set_from_socket(so3, so); 1334 #endif 1335 1336 so2 = so3; 1337 } 1338 unp = sotounpcb(so); 1339 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1340 unp2 = sotounpcb(so2); 1341 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1342 UNP_PCB_LOCK(unp); 1343 UNP_PCB_LOCK(unp2); 1344 error = unp_connect2(so, so2, PRU_CONNECT); 1345 UNP_PCB_UNLOCK(unp2); 1346 UNP_PCB_UNLOCK(unp); 1347 bad2: 1348 UNP_LINK_WUNLOCK(); 1349 if (vfslocked) 1350 /* 1351 * Giant has been previously acquired. This means filesystem 1352 * isn't MPSAFE. Do it once again. 1353 */ 1354 mtx_lock(&Giant); 1355 bad: 1356 if (vp != NULL) 1357 vput(vp); 1358 VFS_UNLOCK_GIANT(vfslocked); 1359 free(sa, M_SONAME); 1360 UNP_LINK_WLOCK(); 1361 UNP_PCB_LOCK(unp); 1362 unp->unp_flags &= ~UNP_CONNECTING; 1363 UNP_PCB_UNLOCK(unp); 1364 return (error); 1365 } 1366 1367 static int 1368 unp_connect2(struct socket *so, struct socket *so2, int req) 1369 { 1370 struct unpcb *unp; 1371 struct unpcb *unp2; 1372 1373 unp = sotounpcb(so); 1374 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1375 unp2 = sotounpcb(so2); 1376 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1377 1378 UNP_LINK_WLOCK_ASSERT(); 1379 UNP_PCB_LOCK_ASSERT(unp); 1380 UNP_PCB_LOCK_ASSERT(unp2); 1381 1382 if (so2->so_type != so->so_type) 1383 return (EPROTOTYPE); 1384 unp->unp_conn = unp2; 1385 1386 switch (so->so_type) { 1387 case SOCK_DGRAM: 1388 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1389 soisconnected(so); 1390 break; 1391 1392 case SOCK_STREAM: 1393 case SOCK_SEQPACKET: 1394 unp2->unp_conn = unp; 1395 if (req == PRU_CONNECT && 1396 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1397 soisconnecting(so); 1398 else 1399 soisconnected(so); 1400 soisconnected(so2); 1401 break; 1402 1403 default: 1404 panic("unp_connect2"); 1405 } 1406 return (0); 1407 } 1408 1409 static void 1410 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1411 { 1412 struct socket *so; 1413 1414 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1415 1416 UNP_LINK_WLOCK_ASSERT(); 1417 UNP_PCB_LOCK_ASSERT(unp); 1418 UNP_PCB_LOCK_ASSERT(unp2); 1419 1420 unp->unp_conn = NULL; 1421 switch (unp->unp_socket->so_type) { 1422 case SOCK_DGRAM: 1423 LIST_REMOVE(unp, unp_reflink); 1424 so = unp->unp_socket; 1425 SOCK_LOCK(so); 1426 so->so_state &= ~SS_ISCONNECTED; 1427 SOCK_UNLOCK(so); 1428 break; 1429 1430 case SOCK_STREAM: 1431 case SOCK_SEQPACKET: 1432 soisdisconnected(unp->unp_socket); 1433 unp2->unp_conn = NULL; 1434 soisdisconnected(unp2->unp_socket); 1435 break; 1436 } 1437 } 1438 1439 /* 1440 * unp_pcblist() walks the global list of struct unpcb's to generate a 1441 * pointer list, bumping the refcount on each unpcb. It then copies them out 1442 * sequentially, validating the generation number on each to see if it has 1443 * been detached. All of this is necessary because copyout() may sleep on 1444 * disk I/O. 1445 */ 1446 static int 1447 unp_pcblist(SYSCTL_HANDLER_ARGS) 1448 { 1449 int error, i, n; 1450 int freeunp; 1451 struct unpcb *unp, **unp_list; 1452 unp_gen_t gencnt; 1453 struct xunpgen *xug; 1454 struct unp_head *head; 1455 struct xunpcb *xu; 1456 1457 switch ((intptr_t)arg1) { 1458 case SOCK_STREAM: 1459 head = &unp_shead; 1460 break; 1461 1462 case SOCK_DGRAM: 1463 head = &unp_dhead; 1464 break; 1465 1466 case SOCK_SEQPACKET: 1467 head = &unp_sphead; 1468 break; 1469 1470 default: 1471 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1472 } 1473 1474 /* 1475 * The process of preparing the PCB list is too time-consuming and 1476 * resource-intensive to repeat twice on every request. 1477 */ 1478 if (req->oldptr == NULL) { 1479 n = unp_count; 1480 req->oldidx = 2 * (sizeof *xug) 1481 + (n + n/8) * sizeof(struct xunpcb); 1482 return (0); 1483 } 1484 1485 if (req->newptr != NULL) 1486 return (EPERM); 1487 1488 /* 1489 * OK, now we're committed to doing something. 1490 */ 1491 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1492 UNP_LIST_LOCK(); 1493 gencnt = unp_gencnt; 1494 n = unp_count; 1495 UNP_LIST_UNLOCK(); 1496 1497 xug->xug_len = sizeof *xug; 1498 xug->xug_count = n; 1499 xug->xug_gen = gencnt; 1500 xug->xug_sogen = so_gencnt; 1501 error = SYSCTL_OUT(req, xug, sizeof *xug); 1502 if (error) { 1503 free(xug, M_TEMP); 1504 return (error); 1505 } 1506 1507 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1508 1509 UNP_LIST_LOCK(); 1510 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1511 unp = LIST_NEXT(unp, unp_link)) { 1512 UNP_PCB_LOCK(unp); 1513 if (unp->unp_gencnt <= gencnt) { 1514 if (cr_cansee(req->td->td_ucred, 1515 unp->unp_socket->so_cred)) { 1516 UNP_PCB_UNLOCK(unp); 1517 continue; 1518 } 1519 unp_list[i++] = unp; 1520 unp->unp_refcount++; 1521 } 1522 UNP_PCB_UNLOCK(unp); 1523 } 1524 UNP_LIST_UNLOCK(); 1525 n = i; /* In case we lost some during malloc. */ 1526 1527 error = 0; 1528 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1529 for (i = 0; i < n; i++) { 1530 unp = unp_list[i]; 1531 UNP_PCB_LOCK(unp); 1532 unp->unp_refcount--; 1533 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1534 xu->xu_len = sizeof *xu; 1535 xu->xu_unpp = unp; 1536 /* 1537 * XXX - need more locking here to protect against 1538 * connect/disconnect races for SMP. 1539 */ 1540 if (unp->unp_addr != NULL) 1541 bcopy(unp->unp_addr, &xu->xu_addr, 1542 unp->unp_addr->sun_len); 1543 if (unp->unp_conn != NULL && 1544 unp->unp_conn->unp_addr != NULL) 1545 bcopy(unp->unp_conn->unp_addr, 1546 &xu->xu_caddr, 1547 unp->unp_conn->unp_addr->sun_len); 1548 bcopy(unp, &xu->xu_unp, sizeof *unp); 1549 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1550 UNP_PCB_UNLOCK(unp); 1551 error = SYSCTL_OUT(req, xu, sizeof *xu); 1552 } else { 1553 freeunp = (unp->unp_refcount == 0); 1554 UNP_PCB_UNLOCK(unp); 1555 if (freeunp) { 1556 UNP_PCB_LOCK_DESTROY(unp); 1557 uma_zfree(unp_zone, unp); 1558 } 1559 } 1560 } 1561 free(xu, M_TEMP); 1562 if (!error) { 1563 /* 1564 * Give the user an updated idea of our state. If the 1565 * generation differs from what we told her before, she knows 1566 * that something happened while we were processing this 1567 * request, and it might be necessary to retry. 1568 */ 1569 xug->xug_gen = unp_gencnt; 1570 xug->xug_sogen = so_gencnt; 1571 xug->xug_count = unp_count; 1572 error = SYSCTL_OUT(req, xug, sizeof *xug); 1573 } 1574 free(unp_list, M_TEMP); 1575 free(xug, M_TEMP); 1576 return (error); 1577 } 1578 1579 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1580 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1581 "List of active local datagram sockets"); 1582 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1583 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1584 "List of active local stream sockets"); 1585 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD, 1586 (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1587 "List of active local seqpacket sockets"); 1588 1589 static void 1590 unp_shutdown(struct unpcb *unp) 1591 { 1592 struct unpcb *unp2; 1593 struct socket *so; 1594 1595 UNP_LINK_WLOCK_ASSERT(); 1596 UNP_PCB_LOCK_ASSERT(unp); 1597 1598 unp2 = unp->unp_conn; 1599 if ((unp->unp_socket->so_type == SOCK_STREAM || 1600 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1601 so = unp2->unp_socket; 1602 if (so != NULL) 1603 socantrcvmore(so); 1604 } 1605 } 1606 1607 static void 1608 unp_drop(struct unpcb *unp, int errno) 1609 { 1610 struct socket *so = unp->unp_socket; 1611 struct unpcb *unp2; 1612 1613 UNP_LINK_WLOCK_ASSERT(); 1614 UNP_PCB_LOCK_ASSERT(unp); 1615 1616 so->so_error = errno; 1617 unp2 = unp->unp_conn; 1618 if (unp2 == NULL) 1619 return; 1620 UNP_PCB_LOCK(unp2); 1621 unp_disconnect(unp, unp2); 1622 UNP_PCB_UNLOCK(unp2); 1623 } 1624 1625 static void 1626 unp_freerights(struct file **rp, int fdcount) 1627 { 1628 int i; 1629 struct file *fp; 1630 1631 for (i = 0; i < fdcount; i++) { 1632 fp = *rp; 1633 *rp++ = NULL; 1634 unp_discard(fp); 1635 } 1636 } 1637 1638 static int 1639 unp_externalize(struct mbuf *control, struct mbuf **controlp) 1640 { 1641 struct thread *td = curthread; /* XXX */ 1642 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1643 int i; 1644 int *fdp; 1645 struct file **rp; 1646 struct file *fp; 1647 void *data; 1648 socklen_t clen = control->m_len, datalen; 1649 int error, newfds; 1650 int f; 1651 u_int newlen; 1652 1653 UNP_LINK_UNLOCK_ASSERT(); 1654 1655 error = 0; 1656 if (controlp != NULL) /* controlp == NULL => free control messages */ 1657 *controlp = NULL; 1658 while (cm != NULL) { 1659 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1660 error = EINVAL; 1661 break; 1662 } 1663 data = CMSG_DATA(cm); 1664 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1665 if (cm->cmsg_level == SOL_SOCKET 1666 && cm->cmsg_type == SCM_RIGHTS) { 1667 newfds = datalen / sizeof(struct file *); 1668 rp = data; 1669 1670 /* If we're not outputting the descriptors free them. */ 1671 if (error || controlp == NULL) { 1672 unp_freerights(rp, newfds); 1673 goto next; 1674 } 1675 FILEDESC_XLOCK(td->td_proc->p_fd); 1676 /* if the new FD's will not fit free them. */ 1677 if (!fdavail(td, newfds)) { 1678 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1679 error = EMSGSIZE; 1680 unp_freerights(rp, newfds); 1681 goto next; 1682 } 1683 1684 /* 1685 * Now change each pointer to an fd in the global 1686 * table to an integer that is the index to the local 1687 * fd table entry that we set up to point to the 1688 * global one we are transferring. 1689 */ 1690 newlen = newfds * sizeof(int); 1691 *controlp = sbcreatecontrol(NULL, newlen, 1692 SCM_RIGHTS, SOL_SOCKET); 1693 if (*controlp == NULL) { 1694 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1695 error = E2BIG; 1696 unp_freerights(rp, newfds); 1697 goto next; 1698 } 1699 1700 fdp = (int *) 1701 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1702 for (i = 0; i < newfds; i++) { 1703 if (fdalloc(td, 0, &f)) 1704 panic("unp_externalize fdalloc failed"); 1705 fp = *rp++; 1706 td->td_proc->p_fd->fd_ofiles[f] = fp; 1707 unp_externalize_fp(fp); 1708 *fdp++ = f; 1709 } 1710 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1711 } else { 1712 /* We can just copy anything else across. */ 1713 if (error || controlp == NULL) 1714 goto next; 1715 *controlp = sbcreatecontrol(NULL, datalen, 1716 cm->cmsg_type, cm->cmsg_level); 1717 if (*controlp == NULL) { 1718 error = ENOBUFS; 1719 goto next; 1720 } 1721 bcopy(data, 1722 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1723 datalen); 1724 } 1725 controlp = &(*controlp)->m_next; 1726 1727 next: 1728 if (CMSG_SPACE(datalen) < clen) { 1729 clen -= CMSG_SPACE(datalen); 1730 cm = (struct cmsghdr *) 1731 ((caddr_t)cm + CMSG_SPACE(datalen)); 1732 } else { 1733 clen = 0; 1734 cm = NULL; 1735 } 1736 } 1737 1738 m_freem(control); 1739 return (error); 1740 } 1741 1742 static void 1743 unp_zone_change(void *tag) 1744 { 1745 1746 uma_zone_set_max(unp_zone, maxsockets); 1747 } 1748 1749 static void 1750 unp_init(void) 1751 { 1752 1753 #ifdef VIMAGE 1754 if (!IS_DEFAULT_VNET(curvnet)) 1755 return; 1756 #endif 1757 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1758 NULL, NULL, UMA_ALIGN_PTR, 0); 1759 if (unp_zone == NULL) 1760 panic("unp_init"); 1761 uma_zone_set_max(unp_zone, maxsockets); 1762 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1763 NULL, EVENTHANDLER_PRI_ANY); 1764 LIST_INIT(&unp_dhead); 1765 LIST_INIT(&unp_shead); 1766 LIST_INIT(&unp_sphead); 1767 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1768 UNP_LINK_LOCK_INIT(); 1769 UNP_LIST_LOCK_INIT(); 1770 } 1771 1772 static int 1773 unp_internalize(struct mbuf **controlp, struct thread *td) 1774 { 1775 struct mbuf *control = *controlp; 1776 struct proc *p = td->td_proc; 1777 struct filedesc *fdescp = p->p_fd; 1778 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1779 struct cmsgcred *cmcred; 1780 struct file **rp; 1781 struct file *fp; 1782 struct timeval *tv; 1783 int i, fd, *fdp; 1784 void *data; 1785 socklen_t clen = control->m_len, datalen; 1786 int error, oldfds; 1787 u_int newlen; 1788 1789 UNP_LINK_UNLOCK_ASSERT(); 1790 1791 error = 0; 1792 *controlp = NULL; 1793 while (cm != NULL) { 1794 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1795 || cm->cmsg_len > clen) { 1796 error = EINVAL; 1797 goto out; 1798 } 1799 data = CMSG_DATA(cm); 1800 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1801 1802 switch (cm->cmsg_type) { 1803 /* 1804 * Fill in credential information. 1805 */ 1806 case SCM_CREDS: 1807 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1808 SCM_CREDS, SOL_SOCKET); 1809 if (*controlp == NULL) { 1810 error = ENOBUFS; 1811 goto out; 1812 } 1813 cmcred = (struct cmsgcred *) 1814 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1815 cmcred->cmcred_pid = p->p_pid; 1816 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1817 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1818 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1819 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1820 CMGROUP_MAX); 1821 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1822 cmcred->cmcred_groups[i] = 1823 td->td_ucred->cr_groups[i]; 1824 break; 1825 1826 case SCM_RIGHTS: 1827 oldfds = datalen / sizeof (int); 1828 /* 1829 * Check that all the FDs passed in refer to legal 1830 * files. If not, reject the entire operation. 1831 */ 1832 fdp = data; 1833 FILEDESC_SLOCK(fdescp); 1834 for (i = 0; i < oldfds; i++) { 1835 fd = *fdp++; 1836 if ((unsigned)fd >= fdescp->fd_nfiles || 1837 fdescp->fd_ofiles[fd] == NULL) { 1838 FILEDESC_SUNLOCK(fdescp); 1839 error = EBADF; 1840 goto out; 1841 } 1842 fp = fdescp->fd_ofiles[fd]; 1843 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1844 FILEDESC_SUNLOCK(fdescp); 1845 error = EOPNOTSUPP; 1846 goto out; 1847 } 1848 1849 } 1850 1851 /* 1852 * Now replace the integer FDs with pointers to the 1853 * associated global file table entry.. 1854 */ 1855 newlen = oldfds * sizeof(struct file *); 1856 *controlp = sbcreatecontrol(NULL, newlen, 1857 SCM_RIGHTS, SOL_SOCKET); 1858 if (*controlp == NULL) { 1859 FILEDESC_SUNLOCK(fdescp); 1860 error = E2BIG; 1861 goto out; 1862 } 1863 fdp = data; 1864 rp = (struct file **) 1865 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1866 for (i = 0; i < oldfds; i++) { 1867 fp = fdescp->fd_ofiles[*fdp++]; 1868 *rp++ = fp; 1869 unp_internalize_fp(fp); 1870 } 1871 FILEDESC_SUNLOCK(fdescp); 1872 break; 1873 1874 case SCM_TIMESTAMP: 1875 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1876 SCM_TIMESTAMP, SOL_SOCKET); 1877 if (*controlp == NULL) { 1878 error = ENOBUFS; 1879 goto out; 1880 } 1881 tv = (struct timeval *) 1882 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1883 microtime(tv); 1884 break; 1885 1886 default: 1887 error = EINVAL; 1888 goto out; 1889 } 1890 1891 controlp = &(*controlp)->m_next; 1892 if (CMSG_SPACE(datalen) < clen) { 1893 clen -= CMSG_SPACE(datalen); 1894 cm = (struct cmsghdr *) 1895 ((caddr_t)cm + CMSG_SPACE(datalen)); 1896 } else { 1897 clen = 0; 1898 cm = NULL; 1899 } 1900 } 1901 1902 out: 1903 m_freem(control); 1904 return (error); 1905 } 1906 1907 static struct mbuf * 1908 unp_addsockcred(struct thread *td, struct mbuf *control) 1909 { 1910 struct mbuf *m, *n, *n_prev; 1911 struct sockcred *sc; 1912 const struct cmsghdr *cm; 1913 int ngroups; 1914 int i; 1915 1916 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1917 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1918 if (m == NULL) 1919 return (control); 1920 1921 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1922 sc->sc_uid = td->td_ucred->cr_ruid; 1923 sc->sc_euid = td->td_ucred->cr_uid; 1924 sc->sc_gid = td->td_ucred->cr_rgid; 1925 sc->sc_egid = td->td_ucred->cr_gid; 1926 sc->sc_ngroups = ngroups; 1927 for (i = 0; i < sc->sc_ngroups; i++) 1928 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1929 1930 /* 1931 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1932 * created SCM_CREDS control message (struct sockcred) has another 1933 * format. 1934 */ 1935 if (control != NULL) 1936 for (n = control, n_prev = NULL; n != NULL;) { 1937 cm = mtod(n, struct cmsghdr *); 1938 if (cm->cmsg_level == SOL_SOCKET && 1939 cm->cmsg_type == SCM_CREDS) { 1940 if (n_prev == NULL) 1941 control = n->m_next; 1942 else 1943 n_prev->m_next = n->m_next; 1944 n = m_free(n); 1945 } else { 1946 n_prev = n; 1947 n = n->m_next; 1948 } 1949 } 1950 1951 /* Prepend it to the head. */ 1952 m->m_next = control; 1953 return (m); 1954 } 1955 1956 static struct unpcb * 1957 fptounp(struct file *fp) 1958 { 1959 struct socket *so; 1960 1961 if (fp->f_type != DTYPE_SOCKET) 1962 return (NULL); 1963 if ((so = fp->f_data) == NULL) 1964 return (NULL); 1965 if (so->so_proto->pr_domain != &localdomain) 1966 return (NULL); 1967 return sotounpcb(so); 1968 } 1969 1970 static void 1971 unp_discard(struct file *fp) 1972 { 1973 1974 unp_externalize_fp(fp); 1975 (void) closef(fp, (struct thread *)NULL); 1976 } 1977 1978 static void 1979 unp_internalize_fp(struct file *fp) 1980 { 1981 struct unpcb *unp; 1982 1983 UNP_LINK_WLOCK(); 1984 if ((unp = fptounp(fp)) != NULL) { 1985 unp->unp_file = fp; 1986 unp->unp_msgcount++; 1987 } 1988 fhold(fp); 1989 unp_rights++; 1990 UNP_LINK_WUNLOCK(); 1991 } 1992 1993 static void 1994 unp_externalize_fp(struct file *fp) 1995 { 1996 struct unpcb *unp; 1997 1998 UNP_LINK_WLOCK(); 1999 if ((unp = fptounp(fp)) != NULL) 2000 unp->unp_msgcount--; 2001 unp_rights--; 2002 UNP_LINK_WUNLOCK(); 2003 } 2004 2005 /* 2006 * unp_defer indicates whether additional work has been defered for a future 2007 * pass through unp_gc(). It is thread local and does not require explicit 2008 * synchronization. 2009 */ 2010 static int unp_marked; 2011 static int unp_unreachable; 2012 2013 static void 2014 unp_accessable(struct file *fp) 2015 { 2016 struct unpcb *unp; 2017 2018 if ((unp = fptounp(fp)) == NULL) 2019 return; 2020 if (unp->unp_gcflag & UNPGC_REF) 2021 return; 2022 unp->unp_gcflag &= ~UNPGC_DEAD; 2023 unp->unp_gcflag |= UNPGC_REF; 2024 unp_marked++; 2025 } 2026 2027 static void 2028 unp_gc_process(struct unpcb *unp) 2029 { 2030 struct socket *soa; 2031 struct socket *so; 2032 struct file *fp; 2033 2034 /* Already processed. */ 2035 if (unp->unp_gcflag & UNPGC_SCANNED) 2036 return; 2037 fp = unp->unp_file; 2038 2039 /* 2040 * Check for a socket potentially in a cycle. It must be in a 2041 * queue as indicated by msgcount, and this must equal the file 2042 * reference count. Note that when msgcount is 0 the file is NULL. 2043 */ 2044 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 2045 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 2046 unp->unp_gcflag |= UNPGC_DEAD; 2047 unp_unreachable++; 2048 return; 2049 } 2050 2051 /* 2052 * Mark all sockets we reference with RIGHTS. 2053 */ 2054 so = unp->unp_socket; 2055 SOCKBUF_LOCK(&so->so_rcv); 2056 unp_scan(so->so_rcv.sb_mb, unp_accessable); 2057 SOCKBUF_UNLOCK(&so->so_rcv); 2058 2059 /* 2060 * Mark all sockets in our accept queue. 2061 */ 2062 ACCEPT_LOCK(); 2063 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 2064 SOCKBUF_LOCK(&soa->so_rcv); 2065 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 2066 SOCKBUF_UNLOCK(&soa->so_rcv); 2067 } 2068 ACCEPT_UNLOCK(); 2069 unp->unp_gcflag |= UNPGC_SCANNED; 2070 } 2071 2072 static int unp_recycled; 2073 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2074 "Number of unreachable sockets claimed by the garbage collector."); 2075 2076 static int unp_taskcount; 2077 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2078 "Number of times the garbage collector has run."); 2079 2080 static void 2081 unp_gc(__unused void *arg, int pending) 2082 { 2083 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2084 NULL }; 2085 struct unp_head **head; 2086 struct file **unref; 2087 struct unpcb *unp; 2088 int i; 2089 2090 unp_taskcount++; 2091 UNP_LIST_LOCK(); 2092 /* 2093 * First clear all gc flags from previous runs. 2094 */ 2095 for (head = heads; *head != NULL; head++) 2096 LIST_FOREACH(unp, *head, unp_link) 2097 unp->unp_gcflag = 0; 2098 2099 /* 2100 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 2101 * is reachable all of the sockets it references are reachable. 2102 * Stop the scan once we do a complete loop without discovering 2103 * a new reachable socket. 2104 */ 2105 do { 2106 unp_unreachable = 0; 2107 unp_marked = 0; 2108 for (head = heads; *head != NULL; head++) 2109 LIST_FOREACH(unp, *head, unp_link) 2110 unp_gc_process(unp); 2111 } while (unp_marked); 2112 UNP_LIST_UNLOCK(); 2113 if (unp_unreachable == 0) 2114 return; 2115 2116 /* 2117 * Allocate space for a local list of dead unpcbs. 2118 */ 2119 unref = malloc(unp_unreachable * sizeof(struct file *), 2120 M_TEMP, M_WAITOK); 2121 2122 /* 2123 * Iterate looking for sockets which have been specifically marked 2124 * as as unreachable and store them locally. 2125 */ 2126 UNP_LIST_LOCK(); 2127 for (i = 0, head = heads; *head != NULL; head++) 2128 LIST_FOREACH(unp, *head, unp_link) 2129 if (unp->unp_gcflag & UNPGC_DEAD) { 2130 unref[i++] = unp->unp_file; 2131 fhold(unp->unp_file); 2132 KASSERT(unp->unp_file != NULL, 2133 ("unp_gc: Invalid unpcb.")); 2134 KASSERT(i <= unp_unreachable, 2135 ("unp_gc: incorrect unreachable count.")); 2136 } 2137 UNP_LIST_UNLOCK(); 2138 2139 /* 2140 * Now flush all sockets, free'ing rights. This will free the 2141 * struct files associated with these sockets but leave each socket 2142 * with one remaining ref. 2143 */ 2144 for (i = 0; i < unp_unreachable; i++) 2145 sorflush(unref[i]->f_data); 2146 2147 /* 2148 * And finally release the sockets so they can be reclaimed. 2149 */ 2150 for (i = 0; i < unp_unreachable; i++) 2151 fdrop(unref[i], NULL); 2152 unp_recycled += unp_unreachable; 2153 free(unref, M_TEMP); 2154 } 2155 2156 static void 2157 unp_dispose(struct mbuf *m) 2158 { 2159 2160 if (m) 2161 unp_scan(m, unp_discard); 2162 } 2163 2164 static void 2165 unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2166 { 2167 struct mbuf *m; 2168 struct file **rp; 2169 struct cmsghdr *cm; 2170 void *data; 2171 int i; 2172 socklen_t clen, datalen; 2173 int qfds; 2174 2175 while (m0 != NULL) { 2176 for (m = m0; m; m = m->m_next) { 2177 if (m->m_type != MT_CONTROL) 2178 continue; 2179 2180 cm = mtod(m, struct cmsghdr *); 2181 clen = m->m_len; 2182 2183 while (cm != NULL) { 2184 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2185 break; 2186 2187 data = CMSG_DATA(cm); 2188 datalen = (caddr_t)cm + cm->cmsg_len 2189 - (caddr_t)data; 2190 2191 if (cm->cmsg_level == SOL_SOCKET && 2192 cm->cmsg_type == SCM_RIGHTS) { 2193 qfds = datalen / sizeof (struct file *); 2194 rp = data; 2195 for (i = 0; i < qfds; i++) 2196 (*op)(*rp++); 2197 } 2198 2199 if (CMSG_SPACE(datalen) < clen) { 2200 clen -= CMSG_SPACE(datalen); 2201 cm = (struct cmsghdr *) 2202 ((caddr_t)cm + CMSG_SPACE(datalen)); 2203 } else { 2204 clen = 0; 2205 cm = NULL; 2206 } 2207 } 2208 } 2209 m0 = m0->m_act; 2210 } 2211 } 2212 2213 #ifdef DDB 2214 static void 2215 db_print_indent(int indent) 2216 { 2217 int i; 2218 2219 for (i = 0; i < indent; i++) 2220 db_printf(" "); 2221 } 2222 2223 static void 2224 db_print_unpflags(int unp_flags) 2225 { 2226 int comma; 2227 2228 comma = 0; 2229 if (unp_flags & UNP_HAVEPC) { 2230 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2231 comma = 1; 2232 } 2233 if (unp_flags & UNP_HAVEPCCACHED) { 2234 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2235 comma = 1; 2236 } 2237 if (unp_flags & UNP_WANTCRED) { 2238 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2239 comma = 1; 2240 } 2241 if (unp_flags & UNP_CONNWAIT) { 2242 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2243 comma = 1; 2244 } 2245 if (unp_flags & UNP_CONNECTING) { 2246 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2247 comma = 1; 2248 } 2249 if (unp_flags & UNP_BINDING) { 2250 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2251 comma = 1; 2252 } 2253 } 2254 2255 static void 2256 db_print_xucred(int indent, struct xucred *xu) 2257 { 2258 int comma, i; 2259 2260 db_print_indent(indent); 2261 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2262 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2263 db_print_indent(indent); 2264 db_printf("cr_groups: "); 2265 comma = 0; 2266 for (i = 0; i < xu->cr_ngroups; i++) { 2267 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2268 comma = 1; 2269 } 2270 db_printf("\n"); 2271 } 2272 2273 static void 2274 db_print_unprefs(int indent, struct unp_head *uh) 2275 { 2276 struct unpcb *unp; 2277 int counter; 2278 2279 counter = 0; 2280 LIST_FOREACH(unp, uh, unp_reflink) { 2281 if (counter % 4 == 0) 2282 db_print_indent(indent); 2283 db_printf("%p ", unp); 2284 if (counter % 4 == 3) 2285 db_printf("\n"); 2286 counter++; 2287 } 2288 if (counter != 0 && counter % 4 != 0) 2289 db_printf("\n"); 2290 } 2291 2292 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2293 { 2294 struct unpcb *unp; 2295 2296 if (!have_addr) { 2297 db_printf("usage: show unpcb <addr>\n"); 2298 return; 2299 } 2300 unp = (struct unpcb *)addr; 2301 2302 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2303 unp->unp_vnode); 2304 2305 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2306 unp->unp_conn); 2307 2308 db_printf("unp_refs:\n"); 2309 db_print_unprefs(2, &unp->unp_refs); 2310 2311 /* XXXRW: Would be nice to print the full address, if any. */ 2312 db_printf("unp_addr: %p\n", unp->unp_addr); 2313 2314 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2315 unp->unp_cc, unp->unp_mbcnt, 2316 (unsigned long long)unp->unp_gencnt); 2317 2318 db_printf("unp_flags: %x (", unp->unp_flags); 2319 db_print_unpflags(unp->unp_flags); 2320 db_printf(")\n"); 2321 2322 db_printf("unp_peercred:\n"); 2323 db_print_xucred(2, &unp->unp_peercred); 2324 2325 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2326 } 2327 #endif 2328