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 * SEQPACKET, RDM 54 * rethink name space problems 55 * need a proper out-of-band 56 */ 57 58 #include <sys/cdefs.h> 59 __FBSDID("$FreeBSD$"); 60 61 #include "opt_ddb.h" 62 #include "opt_mac.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/jail.h> 72 #include <sys/kernel.h> 73 #include <sys/lock.h> 74 #include <sys/mbuf.h> 75 #include <sys/mount.h> 76 #include <sys/mutex.h> 77 #include <sys/namei.h> 78 #include <sys/proc.h> 79 #include <sys/protosw.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 #include <sys/vimage.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. */ 116 117 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 118 119 /* 120 * Garbage collection of cyclic file descriptor/socket references occurs 121 * asynchronously in a taskqueue context in order to avoid recursion and 122 * reentrance in the UNIX domain socket, file descriptor, and socket layer 123 * code. See unp_gc() for a full description. 124 */ 125 static struct task unp_gc_task; 126 127 /* 128 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 129 * stream sockets, although the total for sender and receiver is actually 130 * only PIPSIZ. 131 * 132 * Datagram sockets really use the sendspace as the maximum datagram size, 133 * and don't really want to reserve the sendspace. Their recvspace should be 134 * large enough for at least one max-size datagram plus address. 135 */ 136 #ifndef PIPSIZ 137 #define PIPSIZ 8192 138 #endif 139 static u_long unpst_sendspace = PIPSIZ; 140 static u_long unpst_recvspace = PIPSIZ; 141 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 142 static u_long unpdg_recvspace = 4*1024; 143 144 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 145 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 146 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 147 148 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 149 &unpst_sendspace, 0, "Default stream send space."); 150 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 151 &unpst_recvspace, 0, "Default stream receive space."); 152 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 153 &unpdg_sendspace, 0, "Default datagram send space."); 154 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 155 &unpdg_recvspace, 0, "Default datagram receive space."); 156 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 157 "File descriptors in flight."); 158 159 /*- 160 * Locking and synchronization: 161 * 162 * Three types of locks exit in the local domain socket implementation: a 163 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the 164 * global locks, the list lock protects the socket count, global generation 165 * number, and stream/datagram global lists. The linkage lock protects the 166 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be 167 * held exclusively over the acquisition of multiple unpcb locks to prevent 168 * deadlock. 169 * 170 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 171 * allocated in pru_attach() and freed in pru_detach(). The validity of that 172 * pointer is an invariant, so no lock is required to dereference the so_pcb 173 * pointer if a valid socket reference is held by the caller. In practice, 174 * this is always true during operations performed on a socket. Each unpcb 175 * has a back-pointer to its socket, unp_socket, which will be stable under 176 * the same circumstances. 177 * 178 * This pointer may only be safely dereferenced as long as a valid reference 179 * to the unpcb is held. Typically, this reference will be from the socket, 180 * or from another unpcb when the referring unpcb's lock is held (in order 181 * that the reference not be invalidated during use). For example, to follow 182 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 183 * as unp_socket remains valid as long as the reference to unp_conn is valid. 184 * 185 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 186 * atomic reads without the lock may be performed "lockless", but more 187 * complex reads and read-modify-writes require the mutex to be held. No 188 * lock order is defined between unpcb locks -- multiple unpcb locks may be 189 * acquired at the same time only when holding the linkage rwlock 190 * exclusively, which prevents deadlocks. 191 * 192 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 193 * protocols, bind() is a non-atomic operation, and connect() requires 194 * potential sleeping in the protocol, due to potentially waiting on local or 195 * distributed file systems. We try to separate "lookup" operations, which 196 * may sleep, and the IPC operations themselves, which typically can occur 197 * with relative atomicity as locks can be held over the entire operation. 198 * 199 * Another tricky issue is simultaneous multi-threaded or multi-process 200 * access to a single UNIX domain socket. These are handled by the flags 201 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 202 * binding, both of which involve dropping UNIX domain socket locks in order 203 * to perform namei() and other file system operations. 204 */ 205 static struct rwlock unp_link_rwlock; 206 static struct mtx unp_list_lock; 207 208 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 209 "unp_link_rwlock") 210 211 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 212 RA_LOCKED) 213 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 214 RA_UNLOCKED) 215 216 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 217 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 218 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 219 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 220 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 221 RA_WLOCKED) 222 223 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \ 224 "unp_list_lock", NULL, MTX_DEF) 225 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock) 226 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock) 227 228 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 229 "unp_mtx", "unp_mtx", \ 230 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 231 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 232 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 233 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 234 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 235 236 static int uipc_connect2(struct socket *, struct socket *); 237 static int uipc_ctloutput(struct socket *, struct sockopt *); 238 static int unp_connect(struct socket *, struct sockaddr *, 239 struct thread *); 240 static int unp_connect2(struct socket *so, struct socket *so2, int); 241 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 242 static void unp_dispose(struct mbuf *); 243 static void unp_shutdown(struct unpcb *); 244 static void unp_drop(struct unpcb *, int); 245 static void unp_gc(__unused void *, int); 246 static void unp_scan(struct mbuf *, void (*)(struct file *)); 247 static void unp_discard(struct file *); 248 static void unp_freerights(struct file **, int); 249 static void unp_init(void); 250 static int unp_internalize(struct mbuf **, struct thread *); 251 static void unp_internalize_fp(struct file *); 252 static int unp_externalize(struct mbuf *, struct mbuf **); 253 static void unp_externalize_fp(struct file *); 254 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 255 256 /* 257 * Definitions of protocols supported in the LOCAL domain. 258 */ 259 static struct domain localdomain; 260 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 261 static struct protosw localsw[] = { 262 { 263 .pr_type = SOCK_STREAM, 264 .pr_domain = &localdomain, 265 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 266 .pr_ctloutput = &uipc_ctloutput, 267 .pr_usrreqs = &uipc_usrreqs_stream 268 }, 269 { 270 .pr_type = SOCK_DGRAM, 271 .pr_domain = &localdomain, 272 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 273 .pr_usrreqs = &uipc_usrreqs_dgram 274 }, 275 }; 276 277 static struct domain localdomain = { 278 .dom_family = AF_LOCAL, 279 .dom_name = "local", 280 .dom_init = unp_init, 281 .dom_externalize = unp_externalize, 282 .dom_dispose = unp_dispose, 283 .dom_protosw = localsw, 284 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 285 }; 286 DOMAIN_SET(local); 287 288 static void 289 uipc_abort(struct socket *so) 290 { 291 struct unpcb *unp, *unp2; 292 293 unp = sotounpcb(so); 294 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 295 296 UNP_LINK_WLOCK(); 297 UNP_PCB_LOCK(unp); 298 unp2 = unp->unp_conn; 299 if (unp2 != NULL) { 300 UNP_PCB_LOCK(unp2); 301 unp_drop(unp2, ECONNABORTED); 302 UNP_PCB_UNLOCK(unp2); 303 } 304 UNP_PCB_UNLOCK(unp); 305 UNP_LINK_WUNLOCK(); 306 } 307 308 static int 309 uipc_accept(struct socket *so, struct sockaddr **nam) 310 { 311 struct unpcb *unp, *unp2; 312 const struct sockaddr *sa; 313 314 /* 315 * Pass back name of connected socket, if it was bound and we are 316 * still connected (our peer may have closed already!). 317 */ 318 unp = sotounpcb(so); 319 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 320 321 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 322 UNP_LINK_RLOCK(); 323 unp2 = unp->unp_conn; 324 if (unp2 != NULL && unp2->unp_addr != NULL) { 325 UNP_PCB_LOCK(unp2); 326 sa = (struct sockaddr *) unp2->unp_addr; 327 bcopy(sa, *nam, sa->sa_len); 328 UNP_PCB_UNLOCK(unp2); 329 } else { 330 sa = &sun_noname; 331 bcopy(sa, *nam, sa->sa_len); 332 } 333 UNP_LINK_RUNLOCK(); 334 return (0); 335 } 336 337 static int 338 uipc_attach(struct socket *so, int proto, struct thread *td) 339 { 340 u_long sendspace, recvspace; 341 struct unpcb *unp; 342 int error; 343 344 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 345 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 346 switch (so->so_type) { 347 case SOCK_STREAM: 348 sendspace = unpst_sendspace; 349 recvspace = unpst_recvspace; 350 break; 351 352 case SOCK_DGRAM: 353 sendspace = unpdg_sendspace; 354 recvspace = unpdg_recvspace; 355 break; 356 357 default: 358 panic("uipc_attach"); 359 } 360 error = soreserve(so, sendspace, recvspace); 361 if (error) 362 return (error); 363 } 364 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 365 if (unp == NULL) 366 return (ENOBUFS); 367 LIST_INIT(&unp->unp_refs); 368 UNP_PCB_LOCK_INIT(unp); 369 unp->unp_socket = so; 370 so->so_pcb = unp; 371 unp->unp_refcount = 1; 372 373 UNP_LIST_LOCK(); 374 unp->unp_gencnt = ++unp_gencnt; 375 unp_count++; 376 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead, 377 unp, unp_link); 378 UNP_LIST_UNLOCK(); 379 380 return (0); 381 } 382 383 static int 384 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 385 { 386 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 387 struct vattr vattr; 388 int error, namelen, vfslocked; 389 struct nameidata nd; 390 struct unpcb *unp; 391 struct vnode *vp; 392 struct mount *mp; 393 char *buf; 394 395 unp = sotounpcb(so); 396 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 397 398 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 399 if (namelen <= 0) 400 return (EINVAL); 401 402 /* 403 * We don't allow simultaneous bind() calls on a single UNIX domain 404 * socket, so flag in-progress operations, and return an error if an 405 * operation is already in progress. 406 * 407 * Historically, we have not allowed a socket to be rebound, so this 408 * also returns an error. Not allowing re-binding simplifies the 409 * implementation and avoids a great many possible failure modes. 410 */ 411 UNP_PCB_LOCK(unp); 412 if (unp->unp_vnode != NULL) { 413 UNP_PCB_UNLOCK(unp); 414 return (EINVAL); 415 } 416 if (unp->unp_flags & UNP_BINDING) { 417 UNP_PCB_UNLOCK(unp); 418 return (EALREADY); 419 } 420 unp->unp_flags |= UNP_BINDING; 421 UNP_PCB_UNLOCK(unp); 422 423 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 424 bcopy(soun->sun_path, buf, namelen); 425 buf[namelen] = 0; 426 427 restart: 428 vfslocked = 0; 429 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 430 UIO_SYSSPACE, buf, td); 431 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 432 error = namei(&nd); 433 if (error) 434 goto error; 435 vp = nd.ni_vp; 436 vfslocked = NDHASGIANT(&nd); 437 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 438 NDFREE(&nd, NDF_ONLY_PNBUF); 439 if (nd.ni_dvp == vp) 440 vrele(nd.ni_dvp); 441 else 442 vput(nd.ni_dvp); 443 if (vp != NULL) { 444 vrele(vp); 445 error = EADDRINUSE; 446 goto error; 447 } 448 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 449 if (error) 450 goto error; 451 VFS_UNLOCK_GIANT(vfslocked); 452 goto restart; 453 } 454 VATTR_NULL(&vattr); 455 vattr.va_type = VSOCK; 456 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 457 #ifdef MAC 458 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 459 &vattr); 460 #endif 461 if (error == 0) 462 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 463 NDFREE(&nd, NDF_ONLY_PNBUF); 464 vput(nd.ni_dvp); 465 if (error) { 466 vn_finished_write(mp); 467 goto error; 468 } 469 vp = nd.ni_vp; 470 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 471 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 472 473 UNP_LINK_WLOCK(); 474 UNP_PCB_LOCK(unp); 475 vp->v_socket = unp->unp_socket; 476 unp->unp_vnode = vp; 477 unp->unp_addr = soun; 478 unp->unp_flags &= ~UNP_BINDING; 479 UNP_PCB_UNLOCK(unp); 480 UNP_LINK_WUNLOCK(); 481 VOP_UNLOCK(vp, 0); 482 vn_finished_write(mp); 483 VFS_UNLOCK_GIANT(vfslocked); 484 free(buf, M_TEMP); 485 return (0); 486 487 error: 488 VFS_UNLOCK_GIANT(vfslocked); 489 UNP_PCB_LOCK(unp); 490 unp->unp_flags &= ~UNP_BINDING; 491 UNP_PCB_UNLOCK(unp); 492 free(buf, M_TEMP); 493 return (error); 494 } 495 496 static int 497 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 498 { 499 int error; 500 501 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 502 UNP_LINK_WLOCK(); 503 error = unp_connect(so, nam, td); 504 UNP_LINK_WUNLOCK(); 505 return (error); 506 } 507 508 static void 509 uipc_close(struct socket *so) 510 { 511 struct unpcb *unp, *unp2; 512 513 unp = sotounpcb(so); 514 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 515 516 UNP_LINK_WLOCK(); 517 UNP_PCB_LOCK(unp); 518 unp2 = unp->unp_conn; 519 if (unp2 != NULL) { 520 UNP_PCB_LOCK(unp2); 521 unp_disconnect(unp, unp2); 522 UNP_PCB_UNLOCK(unp2); 523 } 524 UNP_PCB_UNLOCK(unp); 525 UNP_LINK_WUNLOCK(); 526 } 527 528 static int 529 uipc_connect2(struct socket *so1, struct socket *so2) 530 { 531 struct unpcb *unp, *unp2; 532 int error; 533 534 UNP_LINK_WLOCK(); 535 unp = so1->so_pcb; 536 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 537 UNP_PCB_LOCK(unp); 538 unp2 = so2->so_pcb; 539 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 540 UNP_PCB_LOCK(unp2); 541 error = unp_connect2(so1, so2, PRU_CONNECT2); 542 UNP_PCB_UNLOCK(unp2); 543 UNP_PCB_UNLOCK(unp); 544 UNP_LINK_WUNLOCK(); 545 return (error); 546 } 547 548 static void 549 uipc_detach(struct socket *so) 550 { 551 struct unpcb *unp, *unp2; 552 struct sockaddr_un *saved_unp_addr; 553 struct vnode *vp; 554 int freeunp, local_unp_rights; 555 556 unp = sotounpcb(so); 557 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 558 559 UNP_LINK_WLOCK(); 560 UNP_LIST_LOCK(); 561 UNP_PCB_LOCK(unp); 562 LIST_REMOVE(unp, unp_link); 563 unp->unp_gencnt = ++unp_gencnt; 564 --unp_count; 565 UNP_LIST_UNLOCK(); 566 567 /* 568 * XXXRW: Should assert vp->v_socket == so. 569 */ 570 if ((vp = unp->unp_vnode) != NULL) { 571 unp->unp_vnode->v_socket = NULL; 572 unp->unp_vnode = NULL; 573 } 574 unp2 = unp->unp_conn; 575 if (unp2 != NULL) { 576 UNP_PCB_LOCK(unp2); 577 unp_disconnect(unp, unp2); 578 UNP_PCB_UNLOCK(unp2); 579 } 580 581 /* 582 * We hold the linkage lock exclusively, so it's OK to acquire 583 * multiple pcb locks at a time. 584 */ 585 while (!LIST_EMPTY(&unp->unp_refs)) { 586 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 587 588 UNP_PCB_LOCK(ref); 589 unp_drop(ref, ECONNRESET); 590 UNP_PCB_UNLOCK(ref); 591 } 592 local_unp_rights = unp_rights; 593 UNP_LINK_WUNLOCK(); 594 unp->unp_socket->so_pcb = NULL; 595 saved_unp_addr = unp->unp_addr; 596 unp->unp_addr = NULL; 597 unp->unp_refcount--; 598 freeunp = (unp->unp_refcount == 0); 599 if (saved_unp_addr != NULL) 600 free(saved_unp_addr, M_SONAME); 601 if (freeunp) { 602 UNP_PCB_LOCK_DESTROY(unp); 603 uma_zfree(unp_zone, unp); 604 } else 605 UNP_PCB_UNLOCK(unp); 606 if (vp) { 607 int vfslocked; 608 609 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 610 vrele(vp); 611 VFS_UNLOCK_GIANT(vfslocked); 612 } 613 if (local_unp_rights) 614 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 615 } 616 617 static int 618 uipc_disconnect(struct socket *so) 619 { 620 struct unpcb *unp, *unp2; 621 622 unp = sotounpcb(so); 623 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 624 625 UNP_LINK_WLOCK(); 626 UNP_PCB_LOCK(unp); 627 unp2 = unp->unp_conn; 628 if (unp2 != NULL) { 629 UNP_PCB_LOCK(unp2); 630 unp_disconnect(unp, unp2); 631 UNP_PCB_UNLOCK(unp2); 632 } 633 UNP_PCB_UNLOCK(unp); 634 UNP_LINK_WUNLOCK(); 635 return (0); 636 } 637 638 static int 639 uipc_listen(struct socket *so, int backlog, struct thread *td) 640 { 641 struct unpcb *unp; 642 int error; 643 644 unp = sotounpcb(so); 645 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 646 647 UNP_PCB_LOCK(unp); 648 if (unp->unp_vnode == NULL) { 649 UNP_PCB_UNLOCK(unp); 650 return (EINVAL); 651 } 652 653 SOCK_LOCK(so); 654 error = solisten_proto_check(so); 655 if (error == 0) { 656 cru2x(td->td_ucred, &unp->unp_peercred); 657 unp->unp_flags |= UNP_HAVEPCCACHED; 658 solisten_proto(so, backlog); 659 } 660 SOCK_UNLOCK(so); 661 UNP_PCB_UNLOCK(unp); 662 return (error); 663 } 664 665 static int 666 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 667 { 668 struct unpcb *unp, *unp2; 669 const struct sockaddr *sa; 670 671 unp = sotounpcb(so); 672 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 673 674 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 675 UNP_PCB_LOCK(unp); 676 /* 677 * XXX: It seems that this test always fails even when connection is 678 * established. So, this else clause is added as workaround to 679 * return PF_LOCAL sockaddr. 680 */ 681 unp2 = unp->unp_conn; 682 if (unp2 != NULL) { 683 UNP_PCB_LOCK(unp2); 684 if (unp2->unp_addr != NULL) 685 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 686 else 687 sa = &sun_noname; 688 bcopy(sa, *nam, sa->sa_len); 689 UNP_PCB_UNLOCK(unp2); 690 } else { 691 sa = &sun_noname; 692 bcopy(sa, *nam, sa->sa_len); 693 } 694 UNP_PCB_UNLOCK(unp); 695 return (0); 696 } 697 698 static int 699 uipc_rcvd(struct socket *so, int flags) 700 { 701 struct unpcb *unp, *unp2; 702 struct socket *so2; 703 u_int mbcnt, sbcc; 704 u_long newhiwat; 705 706 unp = sotounpcb(so); 707 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 708 709 if (so->so_type == SOCK_DGRAM) 710 panic("uipc_rcvd DGRAM?"); 711 712 if (so->so_type != SOCK_STREAM) 713 panic("uipc_rcvd unknown socktype"); 714 715 /* 716 * Adjust backpressure on sender and wakeup any waiting to write. 717 * 718 * The unp lock is acquired to maintain the validity of the unp_conn 719 * pointer; no lock on unp2 is required as unp2->unp_socket will be 720 * static as long as we don't permit unp2 to disconnect from unp, 721 * which is prevented by the lock on unp. We cache values from 722 * so_rcv to avoid holding the so_rcv lock over the entire 723 * transaction on the remote so_snd. 724 */ 725 SOCKBUF_LOCK(&so->so_rcv); 726 mbcnt = so->so_rcv.sb_mbcnt; 727 sbcc = so->so_rcv.sb_cc; 728 SOCKBUF_UNLOCK(&so->so_rcv); 729 UNP_PCB_LOCK(unp); 730 unp2 = unp->unp_conn; 731 if (unp2 == NULL) { 732 UNP_PCB_UNLOCK(unp); 733 return (0); 734 } 735 so2 = unp2->unp_socket; 736 SOCKBUF_LOCK(&so2->so_snd); 737 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 738 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 739 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 740 newhiwat, RLIM_INFINITY); 741 sowwakeup_locked(so2); 742 unp->unp_mbcnt = mbcnt; 743 unp->unp_cc = sbcc; 744 UNP_PCB_UNLOCK(unp); 745 return (0); 746 } 747 748 static int 749 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 750 struct mbuf *control, struct thread *td) 751 { 752 struct unpcb *unp, *unp2; 753 struct socket *so2; 754 u_int mbcnt_delta, sbcc; 755 u_long newhiwat; 756 int error = 0; 757 758 unp = sotounpcb(so); 759 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 760 761 if (flags & PRUS_OOB) { 762 error = EOPNOTSUPP; 763 goto release; 764 } 765 if (control != NULL && (error = unp_internalize(&control, td))) 766 goto release; 767 if ((nam != NULL) || (flags & PRUS_EOF)) 768 UNP_LINK_WLOCK(); 769 else 770 UNP_LINK_RLOCK(); 771 switch (so->so_type) { 772 case SOCK_DGRAM: 773 { 774 const struct sockaddr *from; 775 776 unp2 = unp->unp_conn; 777 if (nam != NULL) { 778 UNP_LINK_WLOCK_ASSERT(); 779 if (unp2 != NULL) { 780 error = EISCONN; 781 break; 782 } 783 error = unp_connect(so, nam, td); 784 if (error) 785 break; 786 unp2 = unp->unp_conn; 787 } 788 789 /* 790 * Because connect() and send() are non-atomic in a sendto() 791 * with a target address, it's possible that the socket will 792 * have disconnected before the send() can run. In that case 793 * return the slightly counter-intuitive but otherwise 794 * correct error that the socket is not connected. 795 */ 796 if (unp2 == NULL) { 797 error = ENOTCONN; 798 break; 799 } 800 /* Lockless read. */ 801 if (unp2->unp_flags & UNP_WANTCRED) 802 control = unp_addsockcred(td, control); 803 UNP_PCB_LOCK(unp); 804 if (unp->unp_addr != NULL) 805 from = (struct sockaddr *)unp->unp_addr; 806 else 807 from = &sun_noname; 808 so2 = unp2->unp_socket; 809 SOCKBUF_LOCK(&so2->so_rcv); 810 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 811 sorwakeup_locked(so2); 812 m = NULL; 813 control = NULL; 814 } else { 815 SOCKBUF_UNLOCK(&so2->so_rcv); 816 error = ENOBUFS; 817 } 818 if (nam != NULL) { 819 UNP_LINK_WLOCK_ASSERT(); 820 UNP_PCB_LOCK(unp2); 821 unp_disconnect(unp, unp2); 822 UNP_PCB_UNLOCK(unp2); 823 } 824 UNP_PCB_UNLOCK(unp); 825 break; 826 } 827 828 case SOCK_STREAM: 829 if ((so->so_state & SS_ISCONNECTED) == 0) { 830 if (nam != NULL) { 831 UNP_LINK_WLOCK_ASSERT(); 832 error = unp_connect(so, nam, td); 833 if (error) 834 break; /* XXX */ 835 } else { 836 error = ENOTCONN; 837 break; 838 } 839 } 840 841 /* Lockless read. */ 842 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 843 error = EPIPE; 844 break; 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 * Locking here must be done carefully: the inkage lock 855 * prevents interconnections between unpcbs from changing, so 856 * we can traverse from unp to unp2 without acquiring unp's 857 * lock. Socket buffer locks follow unpcb locks, so we can 858 * acquire both remote and lock socket buffer locks. 859 */ 860 unp2 = unp->unp_conn; 861 if (unp2 == NULL) { 862 error = ENOTCONN; 863 break; 864 } 865 so2 = unp2->unp_socket; 866 UNP_PCB_LOCK(unp2); 867 SOCKBUF_LOCK(&so2->so_rcv); 868 if (unp2->unp_flags & UNP_WANTCRED) { 869 /* 870 * Credentials are passed only once on SOCK_STREAM. 871 */ 872 unp2->unp_flags &= ~UNP_WANTCRED; 873 control = unp_addsockcred(td, control); 874 } 875 /* 876 * Send to paired receive port, and then reduce send buffer 877 * hiwater marks to maintain backpressure. Wake up readers. 878 */ 879 if (control != NULL) { 880 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 881 control = NULL; 882 } else 883 sbappend_locked(&so2->so_rcv, m); 884 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 885 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 886 sbcc = so2->so_rcv.sb_cc; 887 sorwakeup_locked(so2); 888 889 SOCKBUF_LOCK(&so->so_snd); 890 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 891 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 892 newhiwat, RLIM_INFINITY); 893 so->so_snd.sb_mbmax -= mbcnt_delta; 894 SOCKBUF_UNLOCK(&so->so_snd); 895 unp2->unp_cc = sbcc; 896 UNP_PCB_UNLOCK(unp2); 897 m = NULL; 898 break; 899 900 default: 901 panic("uipc_send unknown socktype"); 902 } 903 904 /* 905 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 906 */ 907 if (flags & PRUS_EOF) { 908 UNP_PCB_LOCK(unp); 909 socantsendmore(so); 910 unp_shutdown(unp); 911 UNP_PCB_UNLOCK(unp); 912 } 913 914 if ((nam != NULL) || (flags & PRUS_EOF)) 915 UNP_LINK_WUNLOCK(); 916 else 917 UNP_LINK_RUNLOCK(); 918 919 if (control != NULL && error != 0) 920 unp_dispose(control); 921 922 release: 923 if (control != NULL) 924 m_freem(control); 925 if (m != NULL) 926 m_freem(m); 927 return (error); 928 } 929 930 static int 931 uipc_sense(struct socket *so, struct stat *sb) 932 { 933 struct unpcb *unp, *unp2; 934 struct socket *so2; 935 936 unp = sotounpcb(so); 937 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 938 939 sb->st_blksize = so->so_snd.sb_hiwat; 940 UNP_LINK_RLOCK(); 941 UNP_PCB_LOCK(unp); 942 unp2 = unp->unp_conn; 943 if (so->so_type == SOCK_STREAM && unp2 != NULL) { 944 so2 = unp2->unp_socket; 945 sb->st_blksize += so2->so_rcv.sb_cc; 946 } 947 sb->st_dev = NODEV; 948 if (unp->unp_ino == 0) 949 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 950 sb->st_ino = unp->unp_ino; 951 UNP_PCB_UNLOCK(unp); 952 UNP_LINK_RUNLOCK(); 953 return (0); 954 } 955 956 static int 957 uipc_shutdown(struct socket *so) 958 { 959 struct unpcb *unp; 960 961 unp = sotounpcb(so); 962 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 963 964 UNP_LINK_WLOCK(); 965 UNP_PCB_LOCK(unp); 966 socantsendmore(so); 967 unp_shutdown(unp); 968 UNP_PCB_UNLOCK(unp); 969 UNP_LINK_WUNLOCK(); 970 return (0); 971 } 972 973 static int 974 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 975 { 976 struct unpcb *unp; 977 const struct sockaddr *sa; 978 979 unp = sotounpcb(so); 980 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 981 982 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 983 UNP_PCB_LOCK(unp); 984 if (unp->unp_addr != NULL) 985 sa = (struct sockaddr *) unp->unp_addr; 986 else 987 sa = &sun_noname; 988 bcopy(sa, *nam, sa->sa_len); 989 UNP_PCB_UNLOCK(unp); 990 return (0); 991 } 992 993 static struct pr_usrreqs uipc_usrreqs_dgram = { 994 .pru_abort = uipc_abort, 995 .pru_accept = uipc_accept, 996 .pru_attach = uipc_attach, 997 .pru_bind = uipc_bind, 998 .pru_connect = uipc_connect, 999 .pru_connect2 = uipc_connect2, 1000 .pru_detach = uipc_detach, 1001 .pru_disconnect = uipc_disconnect, 1002 .pru_listen = uipc_listen, 1003 .pru_peeraddr = uipc_peeraddr, 1004 .pru_rcvd = uipc_rcvd, 1005 .pru_send = uipc_send, 1006 .pru_sense = uipc_sense, 1007 .pru_shutdown = uipc_shutdown, 1008 .pru_sockaddr = uipc_sockaddr, 1009 .pru_soreceive = soreceive_dgram, 1010 .pru_close = uipc_close, 1011 }; 1012 1013 static struct pr_usrreqs uipc_usrreqs_stream = { 1014 .pru_abort = uipc_abort, 1015 .pru_accept = uipc_accept, 1016 .pru_attach = uipc_attach, 1017 .pru_bind = uipc_bind, 1018 .pru_connect = uipc_connect, 1019 .pru_connect2 = uipc_connect2, 1020 .pru_detach = uipc_detach, 1021 .pru_disconnect = uipc_disconnect, 1022 .pru_listen = uipc_listen, 1023 .pru_peeraddr = uipc_peeraddr, 1024 .pru_rcvd = uipc_rcvd, 1025 .pru_send = uipc_send, 1026 .pru_sense = uipc_sense, 1027 .pru_shutdown = uipc_shutdown, 1028 .pru_sockaddr = uipc_sockaddr, 1029 .pru_soreceive = soreceive_generic, 1030 .pru_close = uipc_close, 1031 }; 1032 1033 static int 1034 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1035 { 1036 struct unpcb *unp; 1037 struct xucred xu; 1038 int error, optval; 1039 1040 if (sopt->sopt_level != 0) 1041 return (EINVAL); 1042 1043 unp = sotounpcb(so); 1044 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1045 error = 0; 1046 switch (sopt->sopt_dir) { 1047 case SOPT_GET: 1048 switch (sopt->sopt_name) { 1049 case LOCAL_PEERCRED: 1050 UNP_PCB_LOCK(unp); 1051 if (unp->unp_flags & UNP_HAVEPC) 1052 xu = unp->unp_peercred; 1053 else { 1054 if (so->so_type == SOCK_STREAM) 1055 error = ENOTCONN; 1056 else 1057 error = EINVAL; 1058 } 1059 UNP_PCB_UNLOCK(unp); 1060 if (error == 0) 1061 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1062 break; 1063 1064 case LOCAL_CREDS: 1065 /* Unlocked read. */ 1066 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1067 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1068 break; 1069 1070 case LOCAL_CONNWAIT: 1071 /* Unlocked read. */ 1072 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1073 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1074 break; 1075 1076 default: 1077 error = EOPNOTSUPP; 1078 break; 1079 } 1080 break; 1081 1082 case SOPT_SET: 1083 switch (sopt->sopt_name) { 1084 case LOCAL_CREDS: 1085 case LOCAL_CONNWAIT: 1086 error = sooptcopyin(sopt, &optval, sizeof(optval), 1087 sizeof(optval)); 1088 if (error) 1089 break; 1090 1091 #define OPTSET(bit) do { \ 1092 UNP_PCB_LOCK(unp); \ 1093 if (optval) \ 1094 unp->unp_flags |= bit; \ 1095 else \ 1096 unp->unp_flags &= ~bit; \ 1097 UNP_PCB_UNLOCK(unp); \ 1098 } while (0) 1099 1100 switch (sopt->sopt_name) { 1101 case LOCAL_CREDS: 1102 OPTSET(UNP_WANTCRED); 1103 break; 1104 1105 case LOCAL_CONNWAIT: 1106 OPTSET(UNP_CONNWAIT); 1107 break; 1108 1109 default: 1110 break; 1111 } 1112 break; 1113 #undef OPTSET 1114 default: 1115 error = ENOPROTOOPT; 1116 break; 1117 } 1118 break; 1119 1120 default: 1121 error = EOPNOTSUPP; 1122 break; 1123 } 1124 return (error); 1125 } 1126 1127 static int 1128 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1129 { 1130 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1131 struct vnode *vp; 1132 struct socket *so2, *so3; 1133 struct unpcb *unp, *unp2, *unp3; 1134 int error, len, vfslocked; 1135 struct nameidata nd; 1136 char buf[SOCK_MAXADDRLEN]; 1137 struct sockaddr *sa; 1138 1139 UNP_LINK_WLOCK_ASSERT(); 1140 1141 unp = sotounpcb(so); 1142 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1143 1144 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1145 if (len <= 0) 1146 return (EINVAL); 1147 bcopy(soun->sun_path, buf, len); 1148 buf[len] = 0; 1149 1150 UNP_PCB_LOCK(unp); 1151 if (unp->unp_flags & UNP_CONNECTING) { 1152 UNP_PCB_UNLOCK(unp); 1153 return (EALREADY); 1154 } 1155 UNP_LINK_WUNLOCK(); 1156 unp->unp_flags |= UNP_CONNECTING; 1157 UNP_PCB_UNLOCK(unp); 1158 1159 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1160 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, 1161 td); 1162 error = namei(&nd); 1163 if (error) 1164 vp = NULL; 1165 else 1166 vp = nd.ni_vp; 1167 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1168 vfslocked = NDHASGIANT(&nd); 1169 NDFREE(&nd, NDF_ONLY_PNBUF); 1170 if (error) 1171 goto bad; 1172 1173 if (vp->v_type != VSOCK) { 1174 error = ENOTSOCK; 1175 goto bad; 1176 } 1177 #ifdef MAC 1178 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1179 if (error) 1180 goto bad; 1181 #endif 1182 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1183 if (error) 1184 goto bad; 1185 VFS_UNLOCK_GIANT(vfslocked); 1186 1187 unp = sotounpcb(so); 1188 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1189 1190 /* 1191 * Lock linkage lock for two reasons: make sure v_socket is stable, 1192 * and to protect simultaneous locking of multiple pcbs. 1193 */ 1194 UNP_LINK_WLOCK(); 1195 so2 = vp->v_socket; 1196 if (so2 == NULL) { 1197 error = ECONNREFUSED; 1198 goto bad2; 1199 } 1200 if (so->so_type != so2->so_type) { 1201 error = EPROTOTYPE; 1202 goto bad2; 1203 } 1204 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1205 if (so2->so_options & SO_ACCEPTCONN) { 1206 so3 = sonewconn(so2, 0); 1207 } else 1208 so3 = NULL; 1209 if (so3 == NULL) { 1210 error = ECONNREFUSED; 1211 goto bad2; 1212 } 1213 unp = sotounpcb(so); 1214 unp2 = sotounpcb(so2); 1215 unp3 = sotounpcb(so3); 1216 UNP_PCB_LOCK(unp); 1217 UNP_PCB_LOCK(unp2); 1218 UNP_PCB_LOCK(unp3); 1219 if (unp2->unp_addr != NULL) { 1220 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1221 unp3->unp_addr = (struct sockaddr_un *) sa; 1222 sa = NULL; 1223 } 1224 1225 /* 1226 * The connecter's (client's) credentials are copied from its 1227 * process structure at the time of connect() (which is now). 1228 */ 1229 cru2x(td->td_ucred, &unp3->unp_peercred); 1230 unp3->unp_flags |= UNP_HAVEPC; 1231 1232 /* 1233 * The receiver's (server's) credentials are copied from the 1234 * unp_peercred member of socket on which the former called 1235 * listen(); uipc_listen() cached that process's credentials 1236 * at that time so we can use them now. 1237 */ 1238 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1239 ("unp_connect: listener without cached peercred")); 1240 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1241 sizeof(unp->unp_peercred)); 1242 unp->unp_flags |= UNP_HAVEPC; 1243 if (unp2->unp_flags & UNP_WANTCRED) 1244 unp3->unp_flags |= UNP_WANTCRED; 1245 UNP_PCB_UNLOCK(unp3); 1246 UNP_PCB_UNLOCK(unp2); 1247 UNP_PCB_UNLOCK(unp); 1248 #ifdef MAC 1249 SOCK_LOCK(so); 1250 mac_socketpeer_set_from_socket(so, so3); 1251 mac_socketpeer_set_from_socket(so3, so); 1252 SOCK_UNLOCK(so); 1253 #endif 1254 1255 so2 = so3; 1256 } 1257 unp = sotounpcb(so); 1258 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1259 unp2 = sotounpcb(so2); 1260 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1261 UNP_PCB_LOCK(unp); 1262 UNP_PCB_LOCK(unp2); 1263 error = unp_connect2(so, so2, PRU_CONNECT); 1264 UNP_PCB_UNLOCK(unp2); 1265 UNP_PCB_UNLOCK(unp); 1266 bad2: 1267 UNP_LINK_WUNLOCK(); 1268 if (vfslocked) 1269 /* 1270 * Giant has been previously acquired. This means filesystem 1271 * isn't MPSAFE. Do it once again. 1272 */ 1273 mtx_lock(&Giant); 1274 bad: 1275 if (vp != NULL) 1276 vput(vp); 1277 VFS_UNLOCK_GIANT(vfslocked); 1278 free(sa, M_SONAME); 1279 UNP_LINK_WLOCK(); 1280 UNP_PCB_LOCK(unp); 1281 unp->unp_flags &= ~UNP_CONNECTING; 1282 UNP_PCB_UNLOCK(unp); 1283 return (error); 1284 } 1285 1286 static int 1287 unp_connect2(struct socket *so, struct socket *so2, int req) 1288 { 1289 struct unpcb *unp; 1290 struct unpcb *unp2; 1291 1292 unp = sotounpcb(so); 1293 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1294 unp2 = sotounpcb(so2); 1295 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1296 1297 UNP_LINK_WLOCK_ASSERT(); 1298 UNP_PCB_LOCK_ASSERT(unp); 1299 UNP_PCB_LOCK_ASSERT(unp2); 1300 1301 if (so2->so_type != so->so_type) 1302 return (EPROTOTYPE); 1303 unp->unp_conn = unp2; 1304 1305 switch (so->so_type) { 1306 case SOCK_DGRAM: 1307 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1308 soisconnected(so); 1309 break; 1310 1311 case SOCK_STREAM: 1312 unp2->unp_conn = unp; 1313 if (req == PRU_CONNECT && 1314 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1315 soisconnecting(so); 1316 else 1317 soisconnected(so); 1318 soisconnected(so2); 1319 break; 1320 1321 default: 1322 panic("unp_connect2"); 1323 } 1324 return (0); 1325 } 1326 1327 static void 1328 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1329 { 1330 struct socket *so; 1331 1332 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1333 1334 UNP_LINK_WLOCK_ASSERT(); 1335 UNP_PCB_LOCK_ASSERT(unp); 1336 UNP_PCB_LOCK_ASSERT(unp2); 1337 1338 unp->unp_conn = NULL; 1339 switch (unp->unp_socket->so_type) { 1340 case SOCK_DGRAM: 1341 LIST_REMOVE(unp, unp_reflink); 1342 so = unp->unp_socket; 1343 SOCK_LOCK(so); 1344 so->so_state &= ~SS_ISCONNECTED; 1345 SOCK_UNLOCK(so); 1346 break; 1347 1348 case SOCK_STREAM: 1349 soisdisconnected(unp->unp_socket); 1350 unp2->unp_conn = NULL; 1351 soisdisconnected(unp2->unp_socket); 1352 break; 1353 } 1354 } 1355 1356 /* 1357 * unp_pcblist() walks the global list of struct unpcb's to generate a 1358 * pointer list, bumping the refcount on each unpcb. It then copies them out 1359 * sequentially, validating the generation number on each to see if it has 1360 * been detached. All of this is necessary because copyout() may sleep on 1361 * disk I/O. 1362 */ 1363 static int 1364 unp_pcblist(SYSCTL_HANDLER_ARGS) 1365 { 1366 int error, i, n; 1367 int freeunp; 1368 struct unpcb *unp, **unp_list; 1369 unp_gen_t gencnt; 1370 struct xunpgen *xug; 1371 struct unp_head *head; 1372 struct xunpcb *xu; 1373 1374 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1375 1376 /* 1377 * The process of preparing the PCB list is too time-consuming and 1378 * resource-intensive to repeat twice on every request. 1379 */ 1380 if (req->oldptr == NULL) { 1381 n = unp_count; 1382 req->oldidx = 2 * (sizeof *xug) 1383 + (n + n/8) * sizeof(struct xunpcb); 1384 return (0); 1385 } 1386 1387 if (req->newptr != NULL) 1388 return (EPERM); 1389 1390 /* 1391 * OK, now we're committed to doing something. 1392 */ 1393 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1394 UNP_LIST_LOCK(); 1395 gencnt = unp_gencnt; 1396 n = unp_count; 1397 UNP_LIST_UNLOCK(); 1398 1399 xug->xug_len = sizeof *xug; 1400 xug->xug_count = n; 1401 xug->xug_gen = gencnt; 1402 xug->xug_sogen = so_gencnt; 1403 error = SYSCTL_OUT(req, xug, sizeof *xug); 1404 if (error) { 1405 free(xug, M_TEMP); 1406 return (error); 1407 } 1408 1409 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1410 1411 UNP_LIST_LOCK(); 1412 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1413 unp = LIST_NEXT(unp, unp_link)) { 1414 UNP_PCB_LOCK(unp); 1415 if (unp->unp_gencnt <= gencnt) { 1416 if (cr_cansee(req->td->td_ucred, 1417 unp->unp_socket->so_cred)) { 1418 UNP_PCB_UNLOCK(unp); 1419 continue; 1420 } 1421 unp_list[i++] = unp; 1422 unp->unp_refcount++; 1423 } 1424 UNP_PCB_UNLOCK(unp); 1425 } 1426 UNP_LIST_UNLOCK(); 1427 n = i; /* In case we lost some during malloc. */ 1428 1429 error = 0; 1430 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1431 for (i = 0; i < n; i++) { 1432 unp = unp_list[i]; 1433 UNP_PCB_LOCK(unp); 1434 unp->unp_refcount--; 1435 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1436 xu->xu_len = sizeof *xu; 1437 xu->xu_unpp = unp; 1438 /* 1439 * XXX - need more locking here to protect against 1440 * connect/disconnect races for SMP. 1441 */ 1442 if (unp->unp_addr != NULL) 1443 bcopy(unp->unp_addr, &xu->xu_addr, 1444 unp->unp_addr->sun_len); 1445 if (unp->unp_conn != NULL && 1446 unp->unp_conn->unp_addr != NULL) 1447 bcopy(unp->unp_conn->unp_addr, 1448 &xu->xu_caddr, 1449 unp->unp_conn->unp_addr->sun_len); 1450 bcopy(unp, &xu->xu_unp, sizeof *unp); 1451 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1452 UNP_PCB_UNLOCK(unp); 1453 error = SYSCTL_OUT(req, xu, sizeof *xu); 1454 } else { 1455 freeunp = (unp->unp_refcount == 0); 1456 UNP_PCB_UNLOCK(unp); 1457 if (freeunp) { 1458 UNP_PCB_LOCK_DESTROY(unp); 1459 uma_zfree(unp_zone, unp); 1460 } 1461 } 1462 } 1463 free(xu, M_TEMP); 1464 if (!error) { 1465 /* 1466 * Give the user an updated idea of our state. If the 1467 * generation differs from what we told her before, she knows 1468 * that something happened while we were processing this 1469 * request, and it might be necessary to retry. 1470 */ 1471 xug->xug_gen = unp_gencnt; 1472 xug->xug_sogen = so_gencnt; 1473 xug->xug_count = unp_count; 1474 error = SYSCTL_OUT(req, xug, sizeof *xug); 1475 } 1476 free(unp_list, M_TEMP); 1477 free(xug, M_TEMP); 1478 return (error); 1479 } 1480 1481 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1482 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1483 "List of active local datagram sockets"); 1484 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1485 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1486 "List of active local stream sockets"); 1487 1488 static void 1489 unp_shutdown(struct unpcb *unp) 1490 { 1491 struct unpcb *unp2; 1492 struct socket *so; 1493 1494 UNP_LINK_WLOCK_ASSERT(); 1495 UNP_PCB_LOCK_ASSERT(unp); 1496 1497 unp2 = unp->unp_conn; 1498 if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) { 1499 so = unp2->unp_socket; 1500 if (so != NULL) 1501 socantrcvmore(so); 1502 } 1503 } 1504 1505 static void 1506 unp_drop(struct unpcb *unp, int errno) 1507 { 1508 struct socket *so = unp->unp_socket; 1509 struct unpcb *unp2; 1510 1511 UNP_LINK_WLOCK_ASSERT(); 1512 UNP_PCB_LOCK_ASSERT(unp); 1513 1514 so->so_error = errno; 1515 unp2 = unp->unp_conn; 1516 if (unp2 == NULL) 1517 return; 1518 UNP_PCB_LOCK(unp2); 1519 unp_disconnect(unp, unp2); 1520 UNP_PCB_UNLOCK(unp2); 1521 } 1522 1523 static void 1524 unp_freerights(struct file **rp, int fdcount) 1525 { 1526 int i; 1527 struct file *fp; 1528 1529 for (i = 0; i < fdcount; i++) { 1530 fp = *rp; 1531 *rp++ = NULL; 1532 unp_discard(fp); 1533 } 1534 } 1535 1536 static int 1537 unp_externalize(struct mbuf *control, struct mbuf **controlp) 1538 { 1539 struct thread *td = curthread; /* XXX */ 1540 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1541 int i; 1542 int *fdp; 1543 struct file **rp; 1544 struct file *fp; 1545 void *data; 1546 socklen_t clen = control->m_len, datalen; 1547 int error, newfds; 1548 int f; 1549 u_int newlen; 1550 1551 UNP_LINK_UNLOCK_ASSERT(); 1552 1553 error = 0; 1554 if (controlp != NULL) /* controlp == NULL => free control messages */ 1555 *controlp = NULL; 1556 while (cm != NULL) { 1557 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1558 error = EINVAL; 1559 break; 1560 } 1561 data = CMSG_DATA(cm); 1562 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1563 if (cm->cmsg_level == SOL_SOCKET 1564 && cm->cmsg_type == SCM_RIGHTS) { 1565 newfds = datalen / sizeof(struct file *); 1566 rp = data; 1567 1568 /* If we're not outputting the descriptors free them. */ 1569 if (error || controlp == NULL) { 1570 unp_freerights(rp, newfds); 1571 goto next; 1572 } 1573 FILEDESC_XLOCK(td->td_proc->p_fd); 1574 /* if the new FD's will not fit free them. */ 1575 if (!fdavail(td, newfds)) { 1576 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1577 error = EMSGSIZE; 1578 unp_freerights(rp, newfds); 1579 goto next; 1580 } 1581 1582 /* 1583 * Now change each pointer to an fd in the global 1584 * table to an integer that is the index to the local 1585 * fd table entry that we set up to point to the 1586 * global one we are transferring. 1587 */ 1588 newlen = newfds * sizeof(int); 1589 *controlp = sbcreatecontrol(NULL, newlen, 1590 SCM_RIGHTS, SOL_SOCKET); 1591 if (*controlp == NULL) { 1592 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1593 error = E2BIG; 1594 unp_freerights(rp, newfds); 1595 goto next; 1596 } 1597 1598 fdp = (int *) 1599 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1600 for (i = 0; i < newfds; i++) { 1601 if (fdalloc(td, 0, &f)) 1602 panic("unp_externalize fdalloc failed"); 1603 fp = *rp++; 1604 td->td_proc->p_fd->fd_ofiles[f] = fp; 1605 unp_externalize_fp(fp); 1606 *fdp++ = f; 1607 } 1608 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1609 } else { 1610 /* We can just copy anything else across. */ 1611 if (error || controlp == NULL) 1612 goto next; 1613 *controlp = sbcreatecontrol(NULL, datalen, 1614 cm->cmsg_type, cm->cmsg_level); 1615 if (*controlp == NULL) { 1616 error = ENOBUFS; 1617 goto next; 1618 } 1619 bcopy(data, 1620 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1621 datalen); 1622 } 1623 controlp = &(*controlp)->m_next; 1624 1625 next: 1626 if (CMSG_SPACE(datalen) < clen) { 1627 clen -= CMSG_SPACE(datalen); 1628 cm = (struct cmsghdr *) 1629 ((caddr_t)cm + CMSG_SPACE(datalen)); 1630 } else { 1631 clen = 0; 1632 cm = NULL; 1633 } 1634 } 1635 1636 m_freem(control); 1637 return (error); 1638 } 1639 1640 static void 1641 unp_zone_change(void *tag) 1642 { 1643 1644 uma_zone_set_max(unp_zone, maxsockets); 1645 } 1646 1647 static void 1648 unp_init(void) 1649 { 1650 1651 #ifdef VIMAGE 1652 if (!IS_DEFAULT_VNET(curvnet)) 1653 return; 1654 #endif 1655 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1656 NULL, NULL, UMA_ALIGN_PTR, 0); 1657 if (unp_zone == NULL) 1658 panic("unp_init"); 1659 uma_zone_set_max(unp_zone, maxsockets); 1660 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1661 NULL, EVENTHANDLER_PRI_ANY); 1662 LIST_INIT(&unp_dhead); 1663 LIST_INIT(&unp_shead); 1664 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1665 UNP_LINK_LOCK_INIT(); 1666 UNP_LIST_LOCK_INIT(); 1667 } 1668 1669 static int 1670 unp_internalize(struct mbuf **controlp, struct thread *td) 1671 { 1672 struct mbuf *control = *controlp; 1673 struct proc *p = td->td_proc; 1674 struct filedesc *fdescp = p->p_fd; 1675 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1676 struct cmsgcred *cmcred; 1677 struct file **rp; 1678 struct file *fp; 1679 struct timeval *tv; 1680 int i, fd, *fdp; 1681 void *data; 1682 socklen_t clen = control->m_len, datalen; 1683 int error, oldfds; 1684 u_int newlen; 1685 1686 UNP_LINK_UNLOCK_ASSERT(); 1687 1688 error = 0; 1689 *controlp = NULL; 1690 while (cm != NULL) { 1691 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1692 || cm->cmsg_len > clen) { 1693 error = EINVAL; 1694 goto out; 1695 } 1696 data = CMSG_DATA(cm); 1697 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1698 1699 switch (cm->cmsg_type) { 1700 /* 1701 * Fill in credential information. 1702 */ 1703 case SCM_CREDS: 1704 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1705 SCM_CREDS, SOL_SOCKET); 1706 if (*controlp == NULL) { 1707 error = ENOBUFS; 1708 goto out; 1709 } 1710 cmcred = (struct cmsgcred *) 1711 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1712 cmcred->cmcred_pid = p->p_pid; 1713 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1714 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1715 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1716 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1717 CMGROUP_MAX); 1718 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1719 cmcred->cmcred_groups[i] = 1720 td->td_ucred->cr_groups[i]; 1721 break; 1722 1723 case SCM_RIGHTS: 1724 oldfds = datalen / sizeof (int); 1725 /* 1726 * Check that all the FDs passed in refer to legal 1727 * files. If not, reject the entire operation. 1728 */ 1729 fdp = data; 1730 FILEDESC_SLOCK(fdescp); 1731 for (i = 0; i < oldfds; i++) { 1732 fd = *fdp++; 1733 if ((unsigned)fd >= fdescp->fd_nfiles || 1734 fdescp->fd_ofiles[fd] == NULL) { 1735 FILEDESC_SUNLOCK(fdescp); 1736 error = EBADF; 1737 goto out; 1738 } 1739 fp = fdescp->fd_ofiles[fd]; 1740 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1741 FILEDESC_SUNLOCK(fdescp); 1742 error = EOPNOTSUPP; 1743 goto out; 1744 } 1745 1746 } 1747 1748 /* 1749 * Now replace the integer FDs with pointers to the 1750 * associated global file table entry.. 1751 */ 1752 newlen = oldfds * sizeof(struct file *); 1753 *controlp = sbcreatecontrol(NULL, newlen, 1754 SCM_RIGHTS, SOL_SOCKET); 1755 if (*controlp == NULL) { 1756 FILEDESC_SUNLOCK(fdescp); 1757 error = E2BIG; 1758 goto out; 1759 } 1760 fdp = data; 1761 rp = (struct file **) 1762 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1763 for (i = 0; i < oldfds; i++) { 1764 fp = fdescp->fd_ofiles[*fdp++]; 1765 *rp++ = fp; 1766 unp_internalize_fp(fp); 1767 } 1768 FILEDESC_SUNLOCK(fdescp); 1769 break; 1770 1771 case SCM_TIMESTAMP: 1772 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1773 SCM_TIMESTAMP, SOL_SOCKET); 1774 if (*controlp == NULL) { 1775 error = ENOBUFS; 1776 goto out; 1777 } 1778 tv = (struct timeval *) 1779 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1780 microtime(tv); 1781 break; 1782 1783 default: 1784 error = EINVAL; 1785 goto out; 1786 } 1787 1788 controlp = &(*controlp)->m_next; 1789 if (CMSG_SPACE(datalen) < clen) { 1790 clen -= CMSG_SPACE(datalen); 1791 cm = (struct cmsghdr *) 1792 ((caddr_t)cm + CMSG_SPACE(datalen)); 1793 } else { 1794 clen = 0; 1795 cm = NULL; 1796 } 1797 } 1798 1799 out: 1800 m_freem(control); 1801 return (error); 1802 } 1803 1804 static struct mbuf * 1805 unp_addsockcred(struct thread *td, struct mbuf *control) 1806 { 1807 struct mbuf *m, *n, *n_prev; 1808 struct sockcred *sc; 1809 const struct cmsghdr *cm; 1810 int ngroups; 1811 int i; 1812 1813 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1814 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1815 if (m == NULL) 1816 return (control); 1817 1818 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1819 sc->sc_uid = td->td_ucred->cr_ruid; 1820 sc->sc_euid = td->td_ucred->cr_uid; 1821 sc->sc_gid = td->td_ucred->cr_rgid; 1822 sc->sc_egid = td->td_ucred->cr_gid; 1823 sc->sc_ngroups = ngroups; 1824 for (i = 0; i < sc->sc_ngroups; i++) 1825 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1826 1827 /* 1828 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1829 * created SCM_CREDS control message (struct sockcred) has another 1830 * format. 1831 */ 1832 if (control != NULL) 1833 for (n = control, n_prev = NULL; n != NULL;) { 1834 cm = mtod(n, struct cmsghdr *); 1835 if (cm->cmsg_level == SOL_SOCKET && 1836 cm->cmsg_type == SCM_CREDS) { 1837 if (n_prev == NULL) 1838 control = n->m_next; 1839 else 1840 n_prev->m_next = n->m_next; 1841 n = m_free(n); 1842 } else { 1843 n_prev = n; 1844 n = n->m_next; 1845 } 1846 } 1847 1848 /* Prepend it to the head. */ 1849 m->m_next = control; 1850 return (m); 1851 } 1852 1853 static struct unpcb * 1854 fptounp(struct file *fp) 1855 { 1856 struct socket *so; 1857 1858 if (fp->f_type != DTYPE_SOCKET) 1859 return (NULL); 1860 if ((so = fp->f_data) == NULL) 1861 return (NULL); 1862 if (so->so_proto->pr_domain != &localdomain) 1863 return (NULL); 1864 return sotounpcb(so); 1865 } 1866 1867 static void 1868 unp_discard(struct file *fp) 1869 { 1870 1871 unp_externalize_fp(fp); 1872 (void) closef(fp, (struct thread *)NULL); 1873 } 1874 1875 static void 1876 unp_internalize_fp(struct file *fp) 1877 { 1878 struct unpcb *unp; 1879 1880 UNP_LINK_WLOCK(); 1881 if ((unp = fptounp(fp)) != NULL) { 1882 unp->unp_file = fp; 1883 unp->unp_msgcount++; 1884 } 1885 fhold(fp); 1886 unp_rights++; 1887 UNP_LINK_WUNLOCK(); 1888 } 1889 1890 static void 1891 unp_externalize_fp(struct file *fp) 1892 { 1893 struct unpcb *unp; 1894 1895 UNP_LINK_WLOCK(); 1896 if ((unp = fptounp(fp)) != NULL) 1897 unp->unp_msgcount--; 1898 unp_rights--; 1899 UNP_LINK_WUNLOCK(); 1900 } 1901 1902 /* 1903 * unp_defer indicates whether additional work has been defered for a future 1904 * pass through unp_gc(). It is thread local and does not require explicit 1905 * synchronization. 1906 */ 1907 static int unp_marked; 1908 static int unp_unreachable; 1909 1910 static void 1911 unp_accessable(struct file *fp) 1912 { 1913 struct unpcb *unp; 1914 1915 if ((unp = fptounp(fp)) == NULL) 1916 return; 1917 if (unp->unp_gcflag & UNPGC_REF) 1918 return; 1919 unp->unp_gcflag &= ~UNPGC_DEAD; 1920 unp->unp_gcflag |= UNPGC_REF; 1921 unp_marked++; 1922 } 1923 1924 static void 1925 unp_gc_process(struct unpcb *unp) 1926 { 1927 struct socket *soa; 1928 struct socket *so; 1929 struct file *fp; 1930 1931 /* Already processed. */ 1932 if (unp->unp_gcflag & UNPGC_SCANNED) 1933 return; 1934 fp = unp->unp_file; 1935 1936 /* 1937 * Check for a socket potentially in a cycle. It must be in a 1938 * queue as indicated by msgcount, and this must equal the file 1939 * reference count. Note that when msgcount is 0 the file is NULL. 1940 */ 1941 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 1942 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 1943 unp->unp_gcflag |= UNPGC_DEAD; 1944 unp_unreachable++; 1945 return; 1946 } 1947 1948 /* 1949 * Mark all sockets we reference with RIGHTS. 1950 */ 1951 so = unp->unp_socket; 1952 SOCKBUF_LOCK(&so->so_rcv); 1953 unp_scan(so->so_rcv.sb_mb, unp_accessable); 1954 SOCKBUF_UNLOCK(&so->so_rcv); 1955 1956 /* 1957 * Mark all sockets in our accept queue. 1958 */ 1959 ACCEPT_LOCK(); 1960 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 1961 SOCKBUF_LOCK(&soa->so_rcv); 1962 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 1963 SOCKBUF_UNLOCK(&soa->so_rcv); 1964 } 1965 ACCEPT_UNLOCK(); 1966 unp->unp_gcflag |= UNPGC_SCANNED; 1967 } 1968 1969 static int unp_recycled; 1970 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 1971 "Number of unreachable sockets claimed by the garbage collector."); 1972 1973 static int unp_taskcount; 1974 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 1975 "Number of times the garbage collector has run."); 1976 1977 static void 1978 unp_gc(__unused void *arg, int pending) 1979 { 1980 struct unp_head *heads[] = { &unp_dhead, &unp_shead, NULL }; 1981 struct unp_head **head; 1982 struct file **unref; 1983 struct unpcb *unp; 1984 int i; 1985 1986 unp_taskcount++; 1987 UNP_LIST_LOCK(); 1988 /* 1989 * First clear all gc flags from previous runs. 1990 */ 1991 for (head = heads; *head != NULL; head++) 1992 LIST_FOREACH(unp, *head, unp_link) 1993 unp->unp_gcflag = 0; 1994 1995 /* 1996 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 1997 * is reachable all of the sockets it references are reachable. 1998 * Stop the scan once we do a complete loop without discovering 1999 * a new reachable socket. 2000 */ 2001 do { 2002 unp_unreachable = 0; 2003 unp_marked = 0; 2004 for (head = heads; *head != NULL; head++) 2005 LIST_FOREACH(unp, *head, unp_link) 2006 unp_gc_process(unp); 2007 } while (unp_marked); 2008 UNP_LIST_UNLOCK(); 2009 if (unp_unreachable == 0) 2010 return; 2011 2012 /* 2013 * Allocate space for a local list of dead unpcbs. 2014 */ 2015 unref = malloc(unp_unreachable * sizeof(struct file *), 2016 M_TEMP, M_WAITOK); 2017 2018 /* 2019 * Iterate looking for sockets which have been specifically marked 2020 * as as unreachable and store them locally. 2021 */ 2022 UNP_LIST_LOCK(); 2023 for (i = 0, head = heads; *head != NULL; head++) 2024 LIST_FOREACH(unp, *head, unp_link) 2025 if (unp->unp_gcflag & UNPGC_DEAD) { 2026 unref[i++] = unp->unp_file; 2027 fhold(unp->unp_file); 2028 KASSERT(unp->unp_file != NULL, 2029 ("unp_gc: Invalid unpcb.")); 2030 KASSERT(i <= unp_unreachable, 2031 ("unp_gc: incorrect unreachable count.")); 2032 } 2033 UNP_LIST_UNLOCK(); 2034 2035 /* 2036 * Now flush all sockets, free'ing rights. This will free the 2037 * struct files associated with these sockets but leave each socket 2038 * with one remaining ref. 2039 */ 2040 for (i = 0; i < unp_unreachable; i++) 2041 sorflush(unref[i]->f_data); 2042 2043 /* 2044 * And finally release the sockets so they can be reclaimed. 2045 */ 2046 for (i = 0; i < unp_unreachable; i++) 2047 fdrop(unref[i], NULL); 2048 unp_recycled += unp_unreachable; 2049 free(unref, M_TEMP); 2050 } 2051 2052 static void 2053 unp_dispose(struct mbuf *m) 2054 { 2055 2056 if (m) 2057 unp_scan(m, unp_discard); 2058 } 2059 2060 static void 2061 unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2062 { 2063 struct mbuf *m; 2064 struct file **rp; 2065 struct cmsghdr *cm; 2066 void *data; 2067 int i; 2068 socklen_t clen, datalen; 2069 int qfds; 2070 2071 while (m0 != NULL) { 2072 for (m = m0; m; m = m->m_next) { 2073 if (m->m_type != MT_CONTROL) 2074 continue; 2075 2076 cm = mtod(m, struct cmsghdr *); 2077 clen = m->m_len; 2078 2079 while (cm != NULL) { 2080 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2081 break; 2082 2083 data = CMSG_DATA(cm); 2084 datalen = (caddr_t)cm + cm->cmsg_len 2085 - (caddr_t)data; 2086 2087 if (cm->cmsg_level == SOL_SOCKET && 2088 cm->cmsg_type == SCM_RIGHTS) { 2089 qfds = datalen / sizeof (struct file *); 2090 rp = data; 2091 for (i = 0; i < qfds; i++) 2092 (*op)(*rp++); 2093 } 2094 2095 if (CMSG_SPACE(datalen) < clen) { 2096 clen -= CMSG_SPACE(datalen); 2097 cm = (struct cmsghdr *) 2098 ((caddr_t)cm + CMSG_SPACE(datalen)); 2099 } else { 2100 clen = 0; 2101 cm = NULL; 2102 } 2103 } 2104 } 2105 m0 = m0->m_act; 2106 } 2107 } 2108 2109 #ifdef DDB 2110 static void 2111 db_print_indent(int indent) 2112 { 2113 int i; 2114 2115 for (i = 0; i < indent; i++) 2116 db_printf(" "); 2117 } 2118 2119 static void 2120 db_print_unpflags(int unp_flags) 2121 { 2122 int comma; 2123 2124 comma = 0; 2125 if (unp_flags & UNP_HAVEPC) { 2126 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2127 comma = 1; 2128 } 2129 if (unp_flags & UNP_HAVEPCCACHED) { 2130 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2131 comma = 1; 2132 } 2133 if (unp_flags & UNP_WANTCRED) { 2134 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2135 comma = 1; 2136 } 2137 if (unp_flags & UNP_CONNWAIT) { 2138 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2139 comma = 1; 2140 } 2141 if (unp_flags & UNP_CONNECTING) { 2142 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2143 comma = 1; 2144 } 2145 if (unp_flags & UNP_BINDING) { 2146 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2147 comma = 1; 2148 } 2149 } 2150 2151 static void 2152 db_print_xucred(int indent, struct xucred *xu) 2153 { 2154 int comma, i; 2155 2156 db_print_indent(indent); 2157 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2158 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2159 db_print_indent(indent); 2160 db_printf("cr_groups: "); 2161 comma = 0; 2162 for (i = 0; i < xu->cr_ngroups; i++) { 2163 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2164 comma = 1; 2165 } 2166 db_printf("\n"); 2167 } 2168 2169 static void 2170 db_print_unprefs(int indent, struct unp_head *uh) 2171 { 2172 struct unpcb *unp; 2173 int counter; 2174 2175 counter = 0; 2176 LIST_FOREACH(unp, uh, unp_reflink) { 2177 if (counter % 4 == 0) 2178 db_print_indent(indent); 2179 db_printf("%p ", unp); 2180 if (counter % 4 == 3) 2181 db_printf("\n"); 2182 counter++; 2183 } 2184 if (counter != 0 && counter % 4 != 0) 2185 db_printf("\n"); 2186 } 2187 2188 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2189 { 2190 struct unpcb *unp; 2191 2192 if (!have_addr) { 2193 db_printf("usage: show unpcb <addr>\n"); 2194 return; 2195 } 2196 unp = (struct unpcb *)addr; 2197 2198 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2199 unp->unp_vnode); 2200 2201 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2202 unp->unp_conn); 2203 2204 db_printf("unp_refs:\n"); 2205 db_print_unprefs(2, &unp->unp_refs); 2206 2207 /* XXXRW: Would be nice to print the full address, if any. */ 2208 db_printf("unp_addr: %p\n", unp->unp_addr); 2209 2210 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2211 unp->unp_cc, unp->unp_mbcnt, 2212 (unsigned long long)unp->unp_gencnt); 2213 2214 db_printf("unp_flags: %x (", unp->unp_flags); 2215 db_print_unpflags(unp->unp_flags); 2216 db_printf(")\n"); 2217 2218 db_printf("unp_peercred:\n"); 2219 db_print_xucred(2, &unp->unp_peercred); 2220 2221 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2222 } 2223 #endif 2224