1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004-2008 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 94 #ifdef DDB 95 #include <ddb/ddb.h> 96 #endif 97 98 #include <security/mac/mac_framework.h> 99 100 #include <vm/uma.h> 101 102 static uma_zone_t unp_zone; 103 static unp_gen_t unp_gencnt; 104 static u_int unp_count; /* Count of local sockets. */ 105 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 106 static int unp_rights; /* File descriptors in flight. */ 107 static struct unp_head unp_shead; /* List of local stream sockets. */ 108 static struct unp_head unp_dhead; /* List of local datagram sockets. */ 109 110 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 111 112 /* 113 * Garbage collection of cyclic file descriptor/socket references occurs 114 * asynchronously in a taskqueue context in order to avoid recursion and 115 * reentrance in the UNIX domain socket, file descriptor, and socket layer 116 * code. See unp_gc() for a full description. 117 */ 118 static struct task unp_gc_task; 119 120 /* 121 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 122 * stream sockets, although the total for sender and receiver is actually 123 * only PIPSIZ. 124 * 125 * Datagram sockets really use the sendspace as the maximum datagram size, 126 * and don't really want to reserve the sendspace. Their recvspace should be 127 * large enough for at least one max-size datagram plus address. 128 */ 129 #ifndef PIPSIZ 130 #define PIPSIZ 8192 131 #endif 132 static u_long unpst_sendspace = PIPSIZ; 133 static u_long unpst_recvspace = PIPSIZ; 134 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 135 static u_long unpdg_recvspace = 4*1024; 136 137 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 138 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 139 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 140 141 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 142 &unpst_sendspace, 0, "Default stream send space."); 143 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 144 &unpst_recvspace, 0, "Default stream receive space."); 145 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 146 &unpdg_sendspace, 0, "Default datagram send space."); 147 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 148 &unpdg_recvspace, 0, "Default datagram receive space."); 149 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 150 "File descriptors in flight."); 151 152 /*- 153 * Locking and synchronization: 154 * 155 * The global UNIX domain socket rwlock (unp_global_rwlock) protects all 156 * global variables, including the linked lists tracking the set of allocated 157 * UNIX domain sockets. The global rwlock also serves to prevent deadlock 158 * when more than one PCB lock is acquired at a time (i.e., during 159 * connect()). Finally, the global rwlock protects uncounted references from 160 * vnodes to sockets bound to those vnodes: to safely dereference the 161 * v_socket pointer, the global rwlock must be held while a full reference is 162 * acquired. 163 * 164 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 165 * allocated in pru_attach() and freed in pru_detach(). The validity of that 166 * pointer is an invariant, so no lock is required to dereference the so_pcb 167 * pointer if a valid socket reference is held by the caller. In practice, 168 * this is always true during operations performed on a socket. Each unpcb 169 * has a back-pointer to its socket, unp_socket, which will be stable under 170 * the same circumstances. 171 * 172 * This pointer may only be safely dereferenced as long as a valid reference 173 * to the unpcb is held. Typically, this reference will be from the socket, 174 * or from another unpcb when the referring unpcb's lock is held (in order 175 * that the reference not be invalidated during use). For example, to follow 176 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 177 * as unp_socket remains valid as long as the reference to unp_conn is valid. 178 * 179 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 180 * atomic reads without the lock may be performed "lockless", but more 181 * complex reads and read-modify-writes require the mutex to be held. No 182 * lock order is defined between unpcb locks -- multiple unpcb locks may be 183 * acquired at the same time only when holding the global UNIX domain socket 184 * rwlock exclusively, which prevents deadlocks. 185 * 186 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 187 * protocols, bind() is a non-atomic operation, and connect() requires 188 * potential sleeping in the protocol, due to potentially waiting on local or 189 * distributed file systems. We try to separate "lookup" operations, which 190 * may sleep, and the IPC operations themselves, which typically can occur 191 * with relative atomicity as locks can be held over the entire operation. 192 * 193 * Another tricky issue is simultaneous multi-threaded or multi-process 194 * access to a single UNIX domain socket. These are handled by the flags 195 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 196 * binding, both of which involve dropping UNIX domain socket locks in order 197 * to perform namei() and other file system operations. 198 */ 199 static struct rwlock unp_global_rwlock; 200 201 #define UNP_GLOBAL_LOCK_INIT() rw_init(&unp_global_rwlock, \ 202 "unp_global_rwlock") 203 204 #define UNP_GLOBAL_LOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 205 RA_LOCKED) 206 #define UNP_GLOBAL_UNLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 207 RA_UNLOCKED) 208 209 #define UNP_GLOBAL_WLOCK() rw_wlock(&unp_global_rwlock) 210 #define UNP_GLOBAL_WUNLOCK() rw_wunlock(&unp_global_rwlock) 211 #define UNP_GLOBAL_WLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 212 RA_WLOCKED) 213 #define UNP_GLOBAL_WOWNED() rw_wowned(&unp_global_rwlock) 214 215 #define UNP_GLOBAL_RLOCK() rw_rlock(&unp_global_rwlock) 216 #define UNP_GLOBAL_RUNLOCK() rw_runlock(&unp_global_rwlock) 217 #define UNP_GLOBAL_RLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 218 RA_RLOCKED) 219 220 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 221 "unp_mtx", "unp_mtx", \ 222 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 223 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 224 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 225 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 226 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 227 228 static int uipc_connect2(struct socket *, struct socket *); 229 static int uipc_ctloutput(struct socket *, struct sockopt *); 230 static int unp_connect(struct socket *, struct sockaddr *, 231 struct thread *); 232 static int unp_connect2(struct socket *so, struct socket *so2, int); 233 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 234 static void unp_dispose(struct mbuf *); 235 static void unp_shutdown(struct unpcb *); 236 static void unp_drop(struct unpcb *, int); 237 static void unp_gc(__unused void *, int); 238 static void unp_scan(struct mbuf *, void (*)(struct file *)); 239 static void unp_discard(struct file *); 240 static void unp_freerights(struct file **, int); 241 static void unp_init(void); 242 static int unp_internalize(struct mbuf **, struct thread *); 243 static void unp_internalize_fp(struct file *); 244 static int unp_externalize(struct mbuf *, struct mbuf **); 245 static void unp_externalize_fp(struct file *); 246 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 247 248 /* 249 * Definitions of protocols supported in the LOCAL domain. 250 */ 251 static struct domain localdomain; 252 static struct pr_usrreqs uipc_usrreqs; 253 static struct protosw localsw[] = { 254 { 255 .pr_type = SOCK_STREAM, 256 .pr_domain = &localdomain, 257 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 258 .pr_ctloutput = &uipc_ctloutput, 259 .pr_usrreqs = &uipc_usrreqs 260 }, 261 { 262 .pr_type = SOCK_DGRAM, 263 .pr_domain = &localdomain, 264 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 265 .pr_usrreqs = &uipc_usrreqs 266 }, 267 }; 268 269 static struct domain localdomain = { 270 .dom_family = AF_LOCAL, 271 .dom_name = "local", 272 .dom_init = unp_init, 273 .dom_externalize = unp_externalize, 274 .dom_dispose = unp_dispose, 275 .dom_protosw = localsw, 276 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 277 }; 278 DOMAIN_SET(local); 279 280 static void 281 uipc_abort(struct socket *so) 282 { 283 struct unpcb *unp, *unp2; 284 285 unp = sotounpcb(so); 286 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 287 288 UNP_GLOBAL_WLOCK(); 289 UNP_PCB_LOCK(unp); 290 unp2 = unp->unp_conn; 291 if (unp2 != NULL) { 292 UNP_PCB_LOCK(unp2); 293 unp_drop(unp2, ECONNABORTED); 294 UNP_PCB_UNLOCK(unp2); 295 } 296 UNP_PCB_UNLOCK(unp); 297 UNP_GLOBAL_WUNLOCK(); 298 } 299 300 static int 301 uipc_accept(struct socket *so, struct sockaddr **nam) 302 { 303 struct unpcb *unp, *unp2; 304 const struct sockaddr *sa; 305 306 /* 307 * Pass back name of connected socket, if it was bound and we are 308 * still connected (our peer may have closed already!). 309 */ 310 unp = sotounpcb(so); 311 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 312 313 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 314 UNP_GLOBAL_RLOCK(); 315 unp2 = unp->unp_conn; 316 if (unp2 != NULL && unp2->unp_addr != NULL) { 317 UNP_PCB_LOCK(unp2); 318 sa = (struct sockaddr *) unp2->unp_addr; 319 bcopy(sa, *nam, sa->sa_len); 320 UNP_PCB_UNLOCK(unp2); 321 } else { 322 sa = &sun_noname; 323 bcopy(sa, *nam, sa->sa_len); 324 } 325 UNP_GLOBAL_RUNLOCK(); 326 return (0); 327 } 328 329 static int 330 uipc_attach(struct socket *so, int proto, struct thread *td) 331 { 332 u_long sendspace, recvspace; 333 struct unpcb *unp; 334 int error, locked; 335 336 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 337 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 338 switch (so->so_type) { 339 case SOCK_STREAM: 340 sendspace = unpst_sendspace; 341 recvspace = unpst_recvspace; 342 break; 343 344 case SOCK_DGRAM: 345 sendspace = unpdg_sendspace; 346 recvspace = unpdg_recvspace; 347 break; 348 349 default: 350 panic("uipc_attach"); 351 } 352 error = soreserve(so, sendspace, recvspace); 353 if (error) 354 return (error); 355 } 356 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 357 if (unp == NULL) 358 return (ENOBUFS); 359 LIST_INIT(&unp->unp_refs); 360 UNP_PCB_LOCK_INIT(unp); 361 unp->unp_socket = so; 362 so->so_pcb = unp; 363 unp->unp_refcount = 1; 364 365 /* 366 * uipc_attach() may be called indirectly from within the UNIX domain 367 * socket code via sonewconn() in unp_connect(). Since rwlocks can 368 * not be recursed, we do the closest thing. 369 */ 370 locked = 0; 371 if (!UNP_GLOBAL_WOWNED()) { 372 UNP_GLOBAL_WLOCK(); 373 locked = 1; 374 } 375 unp->unp_gencnt = ++unp_gencnt; 376 unp_count++; 377 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead, 378 unp, unp_link); 379 if (locked) 380 UNP_GLOBAL_WUNLOCK(); 381 382 return (0); 383 } 384 385 static int 386 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 387 { 388 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 389 struct vattr vattr; 390 int error, namelen, vfslocked; 391 struct nameidata nd; 392 struct unpcb *unp; 393 struct vnode *vp; 394 struct mount *mp; 395 char *buf; 396 397 unp = sotounpcb(so); 398 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 399 400 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 401 if (namelen <= 0) 402 return (EINVAL); 403 404 /* 405 * We don't allow simultaneous bind() calls on a single UNIX domain 406 * socket, so flag in-progress operations, and return an error if an 407 * operation is already in progress. 408 * 409 * Historically, we have not allowed a socket to be rebound, so this 410 * also returns an error. Not allowing re-binding simplifies the 411 * implementation and avoids a great many possible failure modes. 412 */ 413 UNP_PCB_LOCK(unp); 414 if (unp->unp_vnode != NULL) { 415 UNP_PCB_UNLOCK(unp); 416 return (EINVAL); 417 } 418 if (unp->unp_flags & UNP_BINDING) { 419 UNP_PCB_UNLOCK(unp); 420 return (EALREADY); 421 } 422 unp->unp_flags |= UNP_BINDING; 423 UNP_PCB_UNLOCK(unp); 424 425 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 426 bcopy(soun->sun_path, buf, namelen); 427 buf[namelen] = 0; 428 429 restart: 430 vfslocked = 0; 431 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 432 UIO_SYSSPACE, buf, td); 433 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 434 error = namei(&nd); 435 if (error) 436 goto error; 437 vp = nd.ni_vp; 438 vfslocked = NDHASGIANT(&nd); 439 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 440 NDFREE(&nd, NDF_ONLY_PNBUF); 441 if (nd.ni_dvp == vp) 442 vrele(nd.ni_dvp); 443 else 444 vput(nd.ni_dvp); 445 if (vp != NULL) { 446 vrele(vp); 447 error = EADDRINUSE; 448 goto error; 449 } 450 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 451 if (error) 452 goto error; 453 VFS_UNLOCK_GIANT(vfslocked); 454 goto restart; 455 } 456 VATTR_NULL(&vattr); 457 vattr.va_type = VSOCK; 458 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 459 #ifdef MAC 460 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 461 &vattr); 462 #endif 463 if (error == 0) { 464 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 465 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 466 } 467 NDFREE(&nd, NDF_ONLY_PNBUF); 468 vput(nd.ni_dvp); 469 if (error) { 470 vn_finished_write(mp); 471 goto error; 472 } 473 vp = nd.ni_vp; 474 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 475 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 476 477 UNP_GLOBAL_WLOCK(); 478 UNP_PCB_LOCK(unp); 479 vp->v_socket = unp->unp_socket; 480 unp->unp_vnode = vp; 481 unp->unp_addr = soun; 482 unp->unp_flags &= ~UNP_BINDING; 483 UNP_PCB_UNLOCK(unp); 484 UNP_GLOBAL_WUNLOCK(); 485 VOP_UNLOCK(vp, 0); 486 vn_finished_write(mp); 487 VFS_UNLOCK_GIANT(vfslocked); 488 free(buf, M_TEMP); 489 return (0); 490 491 error: 492 VFS_UNLOCK_GIANT(vfslocked); 493 UNP_PCB_LOCK(unp); 494 unp->unp_flags &= ~UNP_BINDING; 495 UNP_PCB_UNLOCK(unp); 496 free(buf, M_TEMP); 497 return (error); 498 } 499 500 static int 501 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 502 { 503 int error; 504 505 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 506 UNP_GLOBAL_WLOCK(); 507 error = unp_connect(so, nam, td); 508 UNP_GLOBAL_WUNLOCK(); 509 return (error); 510 } 511 512 static void 513 uipc_close(struct socket *so) 514 { 515 struct unpcb *unp, *unp2; 516 517 unp = sotounpcb(so); 518 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 519 520 UNP_GLOBAL_WLOCK(); 521 UNP_PCB_LOCK(unp); 522 unp2 = unp->unp_conn; 523 if (unp2 != NULL) { 524 UNP_PCB_LOCK(unp2); 525 unp_disconnect(unp, unp2); 526 UNP_PCB_UNLOCK(unp2); 527 } 528 UNP_PCB_UNLOCK(unp); 529 UNP_GLOBAL_WUNLOCK(); 530 } 531 532 /* 533 * uipc_connect2() is not static as it is invoked directly by fifofs. 534 */ 535 static int 536 uipc_connect2(struct socket *so1, struct socket *so2) 537 { 538 struct unpcb *unp, *unp2; 539 int error; 540 541 UNP_GLOBAL_WLOCK(); 542 unp = so1->so_pcb; 543 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 544 UNP_PCB_LOCK(unp); 545 unp2 = so2->so_pcb; 546 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 547 UNP_PCB_LOCK(unp2); 548 error = unp_connect2(so1, so2, PRU_CONNECT2); 549 UNP_PCB_UNLOCK(unp2); 550 UNP_PCB_UNLOCK(unp); 551 UNP_GLOBAL_WUNLOCK(); 552 return (error); 553 } 554 555 static void 556 uipc_detach(struct socket *so) 557 { 558 struct unpcb *unp, *unp2; 559 struct sockaddr_un *saved_unp_addr; 560 struct vnode *vp; 561 int freeunp, local_unp_rights; 562 563 unp = sotounpcb(so); 564 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 565 566 UNP_GLOBAL_WLOCK(); 567 UNP_PCB_LOCK(unp); 568 569 LIST_REMOVE(unp, unp_link); 570 unp->unp_gencnt = ++unp_gencnt; 571 --unp_count; 572 573 /* 574 * XXXRW: Should assert vp->v_socket == so. 575 */ 576 if ((vp = unp->unp_vnode) != NULL) { 577 unp->unp_vnode->v_socket = NULL; 578 unp->unp_vnode = NULL; 579 } 580 unp2 = unp->unp_conn; 581 if (unp2 != NULL) { 582 UNP_PCB_LOCK(unp2); 583 unp_disconnect(unp, unp2); 584 UNP_PCB_UNLOCK(unp2); 585 } 586 587 /* 588 * We hold the global lock exclusively, so it's OK to acquire 589 * multiple pcb locks at a time. 590 */ 591 while (!LIST_EMPTY(&unp->unp_refs)) { 592 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 593 594 UNP_PCB_LOCK(ref); 595 unp_drop(ref, ECONNRESET); 596 UNP_PCB_UNLOCK(ref); 597 } 598 local_unp_rights = unp_rights; 599 UNP_GLOBAL_WUNLOCK(); 600 unp->unp_socket->so_pcb = NULL; 601 saved_unp_addr = unp->unp_addr; 602 unp->unp_addr = NULL; 603 unp->unp_refcount--; 604 freeunp = (unp->unp_refcount == 0); 605 if (saved_unp_addr != NULL) 606 FREE(saved_unp_addr, M_SONAME); 607 if (freeunp) { 608 UNP_PCB_LOCK_DESTROY(unp); 609 uma_zfree(unp_zone, unp); 610 } else 611 UNP_PCB_UNLOCK(unp); 612 if (vp) { 613 int vfslocked; 614 615 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 616 vrele(vp); 617 VFS_UNLOCK_GIANT(vfslocked); 618 } 619 if (local_unp_rights) 620 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 621 } 622 623 static int 624 uipc_disconnect(struct socket *so) 625 { 626 struct unpcb *unp, *unp2; 627 628 unp = sotounpcb(so); 629 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 630 631 UNP_GLOBAL_WLOCK(); 632 UNP_PCB_LOCK(unp); 633 unp2 = unp->unp_conn; 634 if (unp2 != NULL) { 635 UNP_PCB_LOCK(unp2); 636 unp_disconnect(unp, unp2); 637 UNP_PCB_UNLOCK(unp2); 638 } 639 UNP_PCB_UNLOCK(unp); 640 UNP_GLOBAL_WUNLOCK(); 641 return (0); 642 } 643 644 static int 645 uipc_listen(struct socket *so, int backlog, struct thread *td) 646 { 647 struct unpcb *unp; 648 int error; 649 650 unp = sotounpcb(so); 651 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 652 653 UNP_PCB_LOCK(unp); 654 if (unp->unp_vnode == NULL) { 655 UNP_PCB_UNLOCK(unp); 656 return (EINVAL); 657 } 658 659 SOCK_LOCK(so); 660 error = solisten_proto_check(so); 661 if (error == 0) { 662 cru2x(td->td_ucred, &unp->unp_peercred); 663 unp->unp_flags |= UNP_HAVEPCCACHED; 664 solisten_proto(so, backlog); 665 } 666 SOCK_UNLOCK(so); 667 UNP_PCB_UNLOCK(unp); 668 return (error); 669 } 670 671 static int 672 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 673 { 674 struct unpcb *unp, *unp2; 675 const struct sockaddr *sa; 676 677 unp = sotounpcb(so); 678 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 679 680 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 681 UNP_PCB_LOCK(unp); 682 /* 683 * XXX: It seems that this test always fails even when connection is 684 * established. So, this else clause is added as workaround to 685 * return PF_LOCAL sockaddr. 686 */ 687 unp2 = unp->unp_conn; 688 if (unp2 != NULL) { 689 UNP_PCB_LOCK(unp2); 690 if (unp2->unp_addr != NULL) 691 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 692 else 693 sa = &sun_noname; 694 bcopy(sa, *nam, sa->sa_len); 695 UNP_PCB_UNLOCK(unp2); 696 } else { 697 sa = &sun_noname; 698 bcopy(sa, *nam, sa->sa_len); 699 } 700 UNP_PCB_UNLOCK(unp); 701 return (0); 702 } 703 704 static int 705 uipc_rcvd(struct socket *so, int flags) 706 { 707 struct unpcb *unp, *unp2; 708 struct socket *so2; 709 u_int mbcnt, sbcc; 710 u_long newhiwat; 711 712 unp = sotounpcb(so); 713 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 714 715 if (so->so_type == SOCK_DGRAM) 716 panic("uipc_rcvd DGRAM?"); 717 718 if (so->so_type != SOCK_STREAM) 719 panic("uipc_rcvd unknown socktype"); 720 721 /* 722 * Adjust backpressure on sender and wakeup any waiting to write. 723 * 724 * The unp lock is acquired to maintain the validity of the unp_conn 725 * pointer; no lock on unp2 is required as unp2->unp_socket will be 726 * static as long as we don't permit unp2 to disconnect from unp, 727 * which is prevented by the lock on unp. We cache values from 728 * so_rcv to avoid holding the so_rcv lock over the entire 729 * transaction on the remote so_snd. 730 */ 731 SOCKBUF_LOCK(&so->so_rcv); 732 mbcnt = so->so_rcv.sb_mbcnt; 733 sbcc = so->so_rcv.sb_cc; 734 SOCKBUF_UNLOCK(&so->so_rcv); 735 UNP_PCB_LOCK(unp); 736 unp2 = unp->unp_conn; 737 if (unp2 == NULL) { 738 UNP_PCB_UNLOCK(unp); 739 return (0); 740 } 741 so2 = unp2->unp_socket; 742 SOCKBUF_LOCK(&so2->so_snd); 743 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 744 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 745 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 746 newhiwat, RLIM_INFINITY); 747 sowwakeup_locked(so2); 748 unp->unp_mbcnt = mbcnt; 749 unp->unp_cc = sbcc; 750 UNP_PCB_UNLOCK(unp); 751 return (0); 752 } 753 754 static int 755 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 756 struct mbuf *control, struct thread *td) 757 { 758 struct unpcb *unp, *unp2; 759 struct socket *so2; 760 u_int mbcnt, sbcc; 761 u_long newhiwat; 762 int error = 0; 763 764 unp = sotounpcb(so); 765 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 766 767 if (flags & PRUS_OOB) { 768 error = EOPNOTSUPP; 769 goto release; 770 } 771 if (control != NULL && (error = unp_internalize(&control, td))) 772 goto release; 773 if ((nam != NULL) || (flags & PRUS_EOF)) 774 UNP_GLOBAL_WLOCK(); 775 else 776 UNP_GLOBAL_RLOCK(); 777 switch (so->so_type) { 778 case SOCK_DGRAM: 779 { 780 const struct sockaddr *from; 781 782 unp2 = unp->unp_conn; 783 if (nam != NULL) { 784 UNP_GLOBAL_WLOCK_ASSERT(); 785 if (unp2 != NULL) { 786 error = EISCONN; 787 break; 788 } 789 error = unp_connect(so, nam, td); 790 if (error) 791 break; 792 unp2 = unp->unp_conn; 793 } 794 795 /* 796 * Because connect() and send() are non-atomic in a sendto() 797 * with a target address, it's possible that the socket will 798 * have disconnected before the send() can run. In that case 799 * return the slightly counter-intuitive but otherwise 800 * correct error that the socket is not connected. 801 */ 802 if (unp2 == NULL) { 803 error = ENOTCONN; 804 break; 805 } 806 /* Lockless read. */ 807 if (unp2->unp_flags & UNP_WANTCRED) 808 control = unp_addsockcred(td, control); 809 UNP_PCB_LOCK(unp); 810 if (unp->unp_addr != NULL) 811 from = (struct sockaddr *)unp->unp_addr; 812 else 813 from = &sun_noname; 814 so2 = unp2->unp_socket; 815 SOCKBUF_LOCK(&so2->so_rcv); 816 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 817 sorwakeup_locked(so2); 818 m = NULL; 819 control = NULL; 820 } else { 821 SOCKBUF_UNLOCK(&so2->so_rcv); 822 error = ENOBUFS; 823 } 824 if (nam != NULL) { 825 UNP_GLOBAL_WLOCK_ASSERT(); 826 UNP_PCB_LOCK(unp2); 827 unp_disconnect(unp, unp2); 828 UNP_PCB_UNLOCK(unp2); 829 } 830 UNP_PCB_UNLOCK(unp); 831 break; 832 } 833 834 case SOCK_STREAM: 835 if ((so->so_state & SS_ISCONNECTED) == 0) { 836 if (nam != NULL) { 837 UNP_GLOBAL_WLOCK_ASSERT(); 838 error = unp_connect(so, nam, td); 839 if (error) 840 break; /* XXX */ 841 } else { 842 error = ENOTCONN; 843 break; 844 } 845 } 846 847 /* Lockless read. */ 848 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 849 error = EPIPE; 850 break; 851 } 852 853 /* 854 * Because connect() and send() are non-atomic in a sendto() 855 * with a target address, it's possible that the socket will 856 * have disconnected before the send() can run. In that case 857 * return the slightly counter-intuitive but otherwise 858 * correct error that the socket is not connected. 859 * 860 * Locking here must be done carefully: the global lock 861 * prevents interconnections between unpcbs from changing, so 862 * we can traverse from unp to unp2 without acquiring unp's 863 * lock. Socket buffer locks follow unpcb locks, so we can 864 * acquire both remote and lock socket buffer locks. 865 */ 866 unp2 = unp->unp_conn; 867 if (unp2 == NULL) { 868 error = ENOTCONN; 869 break; 870 } 871 so2 = unp2->unp_socket; 872 UNP_PCB_LOCK(unp2); 873 SOCKBUF_LOCK(&so2->so_rcv); 874 if (unp2->unp_flags & UNP_WANTCRED) { 875 /* 876 * Credentials are passed only once on SOCK_STREAM. 877 */ 878 unp2->unp_flags &= ~UNP_WANTCRED; 879 control = unp_addsockcred(td, control); 880 } 881 /* 882 * Send to paired receive port, and then reduce send buffer 883 * hiwater marks to maintain backpressure. Wake up readers. 884 */ 885 if (control != NULL) { 886 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 887 control = NULL; 888 } else 889 sbappend_locked(&so2->so_rcv, m); 890 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 891 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 892 sbcc = so2->so_rcv.sb_cc; 893 sorwakeup_locked(so2); 894 895 SOCKBUF_LOCK(&so->so_snd); 896 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 897 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 898 newhiwat, RLIM_INFINITY); 899 so->so_snd.sb_mbmax -= mbcnt; 900 SOCKBUF_UNLOCK(&so->so_snd); 901 unp2->unp_cc = sbcc; 902 UNP_PCB_UNLOCK(unp2); 903 m = NULL; 904 break; 905 906 default: 907 panic("uipc_send unknown socktype"); 908 } 909 910 /* 911 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 912 */ 913 if (flags & PRUS_EOF) { 914 UNP_PCB_LOCK(unp); 915 socantsendmore(so); 916 unp_shutdown(unp); 917 UNP_PCB_UNLOCK(unp); 918 } 919 920 if ((nam != NULL) || (flags & PRUS_EOF)) 921 UNP_GLOBAL_WUNLOCK(); 922 else 923 UNP_GLOBAL_RUNLOCK(); 924 925 if (control != NULL && error != 0) 926 unp_dispose(control); 927 928 release: 929 if (control != NULL) 930 m_freem(control); 931 if (m != NULL) 932 m_freem(m); 933 return (error); 934 } 935 936 static int 937 uipc_sense(struct socket *so, struct stat *sb) 938 { 939 struct unpcb *unp, *unp2; 940 struct socket *so2; 941 942 unp = sotounpcb(so); 943 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 944 945 sb->st_blksize = so->so_snd.sb_hiwat; 946 UNP_GLOBAL_RLOCK(); 947 UNP_PCB_LOCK(unp); 948 unp2 = unp->unp_conn; 949 if (so->so_type == SOCK_STREAM && unp2 != NULL) { 950 so2 = unp2->unp_socket; 951 sb->st_blksize += so2->so_rcv.sb_cc; 952 } 953 sb->st_dev = NODEV; 954 if (unp->unp_ino == 0) 955 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 956 sb->st_ino = unp->unp_ino; 957 UNP_PCB_UNLOCK(unp); 958 UNP_GLOBAL_RUNLOCK(); 959 return (0); 960 } 961 962 static int 963 uipc_shutdown(struct socket *so) 964 { 965 struct unpcb *unp; 966 967 unp = sotounpcb(so); 968 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 969 970 UNP_GLOBAL_WLOCK(); 971 UNP_PCB_LOCK(unp); 972 socantsendmore(so); 973 unp_shutdown(unp); 974 UNP_PCB_UNLOCK(unp); 975 UNP_GLOBAL_WUNLOCK(); 976 return (0); 977 } 978 979 static int 980 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 981 { 982 struct unpcb *unp; 983 const struct sockaddr *sa; 984 985 unp = sotounpcb(so); 986 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 987 988 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 989 UNP_PCB_LOCK(unp); 990 if (unp->unp_addr != NULL) 991 sa = (struct sockaddr *) unp->unp_addr; 992 else 993 sa = &sun_noname; 994 bcopy(sa, *nam, sa->sa_len); 995 UNP_PCB_UNLOCK(unp); 996 return (0); 997 } 998 999 static struct pr_usrreqs uipc_usrreqs = { 1000 .pru_abort = uipc_abort, 1001 .pru_accept = uipc_accept, 1002 .pru_attach = uipc_attach, 1003 .pru_bind = uipc_bind, 1004 .pru_connect = uipc_connect, 1005 .pru_connect2 = uipc_connect2, 1006 .pru_detach = uipc_detach, 1007 .pru_disconnect = uipc_disconnect, 1008 .pru_listen = uipc_listen, 1009 .pru_peeraddr = uipc_peeraddr, 1010 .pru_rcvd = uipc_rcvd, 1011 .pru_send = uipc_send, 1012 .pru_sense = uipc_sense, 1013 .pru_shutdown = uipc_shutdown, 1014 .pru_sockaddr = uipc_sockaddr, 1015 .pru_close = uipc_close, 1016 }; 1017 1018 static int 1019 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1020 { 1021 struct unpcb *unp; 1022 struct xucred xu; 1023 int error, optval; 1024 1025 if (sopt->sopt_level != 0) 1026 return (EINVAL); 1027 1028 unp = sotounpcb(so); 1029 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1030 error = 0; 1031 switch (sopt->sopt_dir) { 1032 case SOPT_GET: 1033 switch (sopt->sopt_name) { 1034 case LOCAL_PEERCRED: 1035 UNP_PCB_LOCK(unp); 1036 if (unp->unp_flags & UNP_HAVEPC) 1037 xu = unp->unp_peercred; 1038 else { 1039 if (so->so_type == SOCK_STREAM) 1040 error = ENOTCONN; 1041 else 1042 error = EINVAL; 1043 } 1044 UNP_PCB_UNLOCK(unp); 1045 if (error == 0) 1046 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1047 break; 1048 1049 case LOCAL_CREDS: 1050 /* Unlocked read. */ 1051 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1052 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1053 break; 1054 1055 case LOCAL_CONNWAIT: 1056 /* Unlocked read. */ 1057 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1058 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1059 break; 1060 1061 default: 1062 error = EOPNOTSUPP; 1063 break; 1064 } 1065 break; 1066 1067 case SOPT_SET: 1068 switch (sopt->sopt_name) { 1069 case LOCAL_CREDS: 1070 case LOCAL_CONNWAIT: 1071 error = sooptcopyin(sopt, &optval, sizeof(optval), 1072 sizeof(optval)); 1073 if (error) 1074 break; 1075 1076 #define OPTSET(bit) do { \ 1077 UNP_PCB_LOCK(unp); \ 1078 if (optval) \ 1079 unp->unp_flags |= bit; \ 1080 else \ 1081 unp->unp_flags &= ~bit; \ 1082 UNP_PCB_UNLOCK(unp); \ 1083 } while (0) 1084 1085 switch (sopt->sopt_name) { 1086 case LOCAL_CREDS: 1087 OPTSET(UNP_WANTCRED); 1088 break; 1089 1090 case LOCAL_CONNWAIT: 1091 OPTSET(UNP_CONNWAIT); 1092 break; 1093 1094 default: 1095 break; 1096 } 1097 break; 1098 #undef OPTSET 1099 default: 1100 error = ENOPROTOOPT; 1101 break; 1102 } 1103 break; 1104 1105 default: 1106 error = EOPNOTSUPP; 1107 break; 1108 } 1109 return (error); 1110 } 1111 1112 static int 1113 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1114 { 1115 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1116 struct vnode *vp; 1117 struct socket *so2, *so3; 1118 struct unpcb *unp, *unp2, *unp3; 1119 int error, len, vfslocked; 1120 struct nameidata nd; 1121 char buf[SOCK_MAXADDRLEN]; 1122 struct sockaddr *sa; 1123 1124 UNP_GLOBAL_WLOCK_ASSERT(); 1125 1126 unp = sotounpcb(so); 1127 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1128 1129 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1130 if (len <= 0) 1131 return (EINVAL); 1132 bcopy(soun->sun_path, buf, len); 1133 buf[len] = 0; 1134 1135 UNP_PCB_LOCK(unp); 1136 if (unp->unp_flags & UNP_CONNECTING) { 1137 UNP_PCB_UNLOCK(unp); 1138 return (EALREADY); 1139 } 1140 UNP_GLOBAL_WUNLOCK(); 1141 unp->unp_flags |= UNP_CONNECTING; 1142 UNP_PCB_UNLOCK(unp); 1143 1144 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1145 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, 1146 td); 1147 error = namei(&nd); 1148 if (error) 1149 vp = NULL; 1150 else 1151 vp = nd.ni_vp; 1152 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1153 vfslocked = NDHASGIANT(&nd); 1154 NDFREE(&nd, NDF_ONLY_PNBUF); 1155 if (error) 1156 goto bad; 1157 1158 if (vp->v_type != VSOCK) { 1159 error = ENOTSOCK; 1160 goto bad; 1161 } 1162 #ifdef MAC 1163 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1164 if (error) 1165 goto bad; 1166 #endif 1167 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1168 if (error) 1169 goto bad; 1170 VFS_UNLOCK_GIANT(vfslocked); 1171 1172 unp = sotounpcb(so); 1173 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1174 1175 /* 1176 * Lock global lock for two reasons: make sure v_socket is stable, 1177 * and to protect simultaneous locking of multiple pcbs. 1178 */ 1179 UNP_GLOBAL_WLOCK(); 1180 so2 = vp->v_socket; 1181 if (so2 == NULL) { 1182 error = ECONNREFUSED; 1183 goto bad2; 1184 } 1185 if (so->so_type != so2->so_type) { 1186 error = EPROTOTYPE; 1187 goto bad2; 1188 } 1189 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1190 if (so2->so_options & SO_ACCEPTCONN) { 1191 /* 1192 * We can't drop the global lock here or 'so2' may 1193 * become invalid. As a result, we need to handle 1194 * possibly lock recursion in uipc_attach. 1195 */ 1196 so3 = sonewconn(so2, 0); 1197 } else 1198 so3 = NULL; 1199 if (so3 == NULL) { 1200 error = ECONNREFUSED; 1201 goto bad2; 1202 } 1203 unp = sotounpcb(so); 1204 unp2 = sotounpcb(so2); 1205 unp3 = sotounpcb(so3); 1206 UNP_PCB_LOCK(unp); 1207 UNP_PCB_LOCK(unp2); 1208 UNP_PCB_LOCK(unp3); 1209 if (unp2->unp_addr != NULL) { 1210 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1211 unp3->unp_addr = (struct sockaddr_un *) sa; 1212 sa = NULL; 1213 } 1214 /* 1215 * unp_peercred management: 1216 * 1217 * The connecter's (client's) credentials are copied from its 1218 * process structure at the time of connect() (which is now). 1219 */ 1220 cru2x(td->td_ucred, &unp3->unp_peercred); 1221 unp3->unp_flags |= UNP_HAVEPC; 1222 /* 1223 * The receiver's (server's) credentials are copied from the 1224 * unp_peercred member of socket on which the former called 1225 * listen(); uipc_listen() cached that process's credentials 1226 * at that time so we can use them now. 1227 */ 1228 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1229 ("unp_connect: listener without cached peercred")); 1230 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1231 sizeof(unp->unp_peercred)); 1232 unp->unp_flags |= UNP_HAVEPC; 1233 if (unp2->unp_flags & UNP_WANTCRED) 1234 unp3->unp_flags |= UNP_WANTCRED; 1235 UNP_PCB_UNLOCK(unp3); 1236 UNP_PCB_UNLOCK(unp2); 1237 UNP_PCB_UNLOCK(unp); 1238 #ifdef MAC 1239 SOCK_LOCK(so); 1240 mac_socketpeer_set_from_socket(so, so3); 1241 mac_socketpeer_set_from_socket(so3, so); 1242 SOCK_UNLOCK(so); 1243 #endif 1244 1245 so2 = so3; 1246 } 1247 unp = sotounpcb(so); 1248 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1249 unp2 = sotounpcb(so2); 1250 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1251 UNP_PCB_LOCK(unp); 1252 UNP_PCB_LOCK(unp2); 1253 error = unp_connect2(so, so2, PRU_CONNECT); 1254 UNP_PCB_UNLOCK(unp2); 1255 UNP_PCB_UNLOCK(unp); 1256 bad2: 1257 UNP_GLOBAL_WUNLOCK(); 1258 if (vfslocked) 1259 /* 1260 * Giant has been previously acquired. This means filesystem 1261 * isn't MPSAFE. Do it once again. 1262 */ 1263 mtx_lock(&Giant); 1264 bad: 1265 if (vp != NULL) 1266 vput(vp); 1267 VFS_UNLOCK_GIANT(vfslocked); 1268 free(sa, M_SONAME); 1269 UNP_GLOBAL_WLOCK(); 1270 UNP_PCB_LOCK(unp); 1271 unp->unp_flags &= ~UNP_CONNECTING; 1272 UNP_PCB_UNLOCK(unp); 1273 return (error); 1274 } 1275 1276 static int 1277 unp_connect2(struct socket *so, struct socket *so2, int req) 1278 { 1279 struct unpcb *unp; 1280 struct unpcb *unp2; 1281 1282 unp = sotounpcb(so); 1283 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1284 unp2 = sotounpcb(so2); 1285 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1286 1287 UNP_GLOBAL_WLOCK_ASSERT(); 1288 UNP_PCB_LOCK_ASSERT(unp); 1289 UNP_PCB_LOCK_ASSERT(unp2); 1290 1291 if (so2->so_type != so->so_type) 1292 return (EPROTOTYPE); 1293 unp->unp_conn = unp2; 1294 1295 switch (so->so_type) { 1296 case SOCK_DGRAM: 1297 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1298 soisconnected(so); 1299 break; 1300 1301 case SOCK_STREAM: 1302 unp2->unp_conn = unp; 1303 if (req == PRU_CONNECT && 1304 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1305 soisconnecting(so); 1306 else 1307 soisconnected(so); 1308 soisconnected(so2); 1309 break; 1310 1311 default: 1312 panic("unp_connect2"); 1313 } 1314 return (0); 1315 } 1316 1317 static void 1318 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1319 { 1320 struct socket *so; 1321 1322 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1323 1324 UNP_GLOBAL_WLOCK_ASSERT(); 1325 UNP_PCB_LOCK_ASSERT(unp); 1326 UNP_PCB_LOCK_ASSERT(unp2); 1327 1328 unp->unp_conn = NULL; 1329 switch (unp->unp_socket->so_type) { 1330 case SOCK_DGRAM: 1331 LIST_REMOVE(unp, unp_reflink); 1332 so = unp->unp_socket; 1333 SOCK_LOCK(so); 1334 so->so_state &= ~SS_ISCONNECTED; 1335 SOCK_UNLOCK(so); 1336 break; 1337 1338 case SOCK_STREAM: 1339 soisdisconnected(unp->unp_socket); 1340 unp2->unp_conn = NULL; 1341 soisdisconnected(unp2->unp_socket); 1342 break; 1343 } 1344 } 1345 1346 /* 1347 * unp_pcblist() walks the global list of struct unpcb's to generate a 1348 * pointer list, bumping the refcount on each unpcb. It then copies them out 1349 * sequentially, validating the generation number on each to see if it has 1350 * been detached. All of this is necessary because copyout() may sleep on 1351 * disk I/O. 1352 */ 1353 static int 1354 unp_pcblist(SYSCTL_HANDLER_ARGS) 1355 { 1356 int error, i, n; 1357 int freeunp; 1358 struct unpcb *unp, **unp_list; 1359 unp_gen_t gencnt; 1360 struct xunpgen *xug; 1361 struct unp_head *head; 1362 struct xunpcb *xu; 1363 1364 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1365 1366 /* 1367 * The process of preparing the PCB list is too time-consuming and 1368 * resource-intensive to repeat twice on every request. 1369 */ 1370 if (req->oldptr == NULL) { 1371 n = unp_count; 1372 req->oldidx = 2 * (sizeof *xug) 1373 + (n + n/8) * sizeof(struct xunpcb); 1374 return (0); 1375 } 1376 1377 if (req->newptr != NULL) 1378 return (EPERM); 1379 1380 /* 1381 * OK, now we're committed to doing something. 1382 */ 1383 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1384 UNP_GLOBAL_RLOCK(); 1385 gencnt = unp_gencnt; 1386 n = unp_count; 1387 UNP_GLOBAL_RUNLOCK(); 1388 1389 xug->xug_len = sizeof *xug; 1390 xug->xug_count = n; 1391 xug->xug_gen = gencnt; 1392 xug->xug_sogen = so_gencnt; 1393 error = SYSCTL_OUT(req, xug, sizeof *xug); 1394 if (error) { 1395 free(xug, M_TEMP); 1396 return (error); 1397 } 1398 1399 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1400 1401 UNP_GLOBAL_RLOCK(); 1402 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1403 unp = LIST_NEXT(unp, unp_link)) { 1404 UNP_PCB_LOCK(unp); 1405 if (unp->unp_gencnt <= gencnt) { 1406 if (cr_cansee(req->td->td_ucred, 1407 unp->unp_socket->so_cred)) { 1408 UNP_PCB_UNLOCK(unp); 1409 continue; 1410 } 1411 unp_list[i++] = unp; 1412 unp->unp_refcount++; 1413 } 1414 UNP_PCB_UNLOCK(unp); 1415 } 1416 UNP_GLOBAL_RUNLOCK(); 1417 n = i; /* In case we lost some during malloc. */ 1418 1419 error = 0; 1420 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1421 for (i = 0; i < n; i++) { 1422 unp = unp_list[i]; 1423 UNP_PCB_LOCK(unp); 1424 unp->unp_refcount--; 1425 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1426 xu->xu_len = sizeof *xu; 1427 xu->xu_unpp = unp; 1428 /* 1429 * XXX - need more locking here to protect against 1430 * connect/disconnect races for SMP. 1431 */ 1432 if (unp->unp_addr != NULL) 1433 bcopy(unp->unp_addr, &xu->xu_addr, 1434 unp->unp_addr->sun_len); 1435 if (unp->unp_conn != NULL && 1436 unp->unp_conn->unp_addr != NULL) 1437 bcopy(unp->unp_conn->unp_addr, 1438 &xu->xu_caddr, 1439 unp->unp_conn->unp_addr->sun_len); 1440 bcopy(unp, &xu->xu_unp, sizeof *unp); 1441 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1442 UNP_PCB_UNLOCK(unp); 1443 error = SYSCTL_OUT(req, xu, sizeof *xu); 1444 } else { 1445 freeunp = (unp->unp_refcount == 0); 1446 UNP_PCB_UNLOCK(unp); 1447 if (freeunp) { 1448 UNP_PCB_LOCK_DESTROY(unp); 1449 uma_zfree(unp_zone, unp); 1450 } 1451 } 1452 } 1453 free(xu, M_TEMP); 1454 if (!error) { 1455 /* 1456 * Give the user an updated idea of our state. If the 1457 * generation differs from what we told her before, she knows 1458 * that something happened while we were processing this 1459 * request, and it might be necessary to retry. 1460 */ 1461 xug->xug_gen = unp_gencnt; 1462 xug->xug_sogen = so_gencnt; 1463 xug->xug_count = unp_count; 1464 error = SYSCTL_OUT(req, xug, sizeof *xug); 1465 } 1466 free(unp_list, M_TEMP); 1467 free(xug, M_TEMP); 1468 return (error); 1469 } 1470 1471 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1472 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1473 "List of active local datagram sockets"); 1474 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1475 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1476 "List of active local stream sockets"); 1477 1478 static void 1479 unp_shutdown(struct unpcb *unp) 1480 { 1481 struct unpcb *unp2; 1482 struct socket *so; 1483 1484 UNP_GLOBAL_WLOCK_ASSERT(); 1485 UNP_PCB_LOCK_ASSERT(unp); 1486 1487 unp2 = unp->unp_conn; 1488 if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) { 1489 so = unp2->unp_socket; 1490 if (so != NULL) 1491 socantrcvmore(so); 1492 } 1493 } 1494 1495 static void 1496 unp_drop(struct unpcb *unp, int errno) 1497 { 1498 struct socket *so = unp->unp_socket; 1499 struct unpcb *unp2; 1500 1501 UNP_GLOBAL_WLOCK_ASSERT(); 1502 UNP_PCB_LOCK_ASSERT(unp); 1503 1504 so->so_error = errno; 1505 unp2 = unp->unp_conn; 1506 if (unp2 == NULL) 1507 return; 1508 UNP_PCB_LOCK(unp2); 1509 unp_disconnect(unp, unp2); 1510 UNP_PCB_UNLOCK(unp2); 1511 } 1512 1513 static void 1514 unp_freerights(struct file **rp, int fdcount) 1515 { 1516 int i; 1517 struct file *fp; 1518 1519 for (i = 0; i < fdcount; i++) { 1520 /* 1521 * Zero the pointer before calling unp_discard since it may 1522 * end up in unp_gc().. 1523 * 1524 * XXXRW: This is less true than it used to be. 1525 */ 1526 fp = *rp; 1527 *rp++ = NULL; 1528 unp_discard(fp); 1529 } 1530 } 1531 1532 static int 1533 unp_externalize(struct mbuf *control, struct mbuf **controlp) 1534 { 1535 struct thread *td = curthread; /* XXX */ 1536 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1537 int i; 1538 int *fdp; 1539 struct file **rp; 1540 struct file *fp; 1541 void *data; 1542 socklen_t clen = control->m_len, datalen; 1543 int error, newfds; 1544 int f; 1545 u_int newlen; 1546 1547 UNP_GLOBAL_UNLOCK_ASSERT(); 1548 1549 error = 0; 1550 if (controlp != NULL) /* controlp == NULL => free control messages */ 1551 *controlp = NULL; 1552 while (cm != NULL) { 1553 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1554 error = EINVAL; 1555 break; 1556 } 1557 data = CMSG_DATA(cm); 1558 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1559 if (cm->cmsg_level == SOL_SOCKET 1560 && cm->cmsg_type == SCM_RIGHTS) { 1561 newfds = datalen / sizeof(struct file *); 1562 rp = data; 1563 1564 /* If we're not outputting the descriptors free them. */ 1565 if (error || controlp == NULL) { 1566 unp_freerights(rp, newfds); 1567 goto next; 1568 } 1569 FILEDESC_XLOCK(td->td_proc->p_fd); 1570 /* if the new FD's will not fit free them. */ 1571 if (!fdavail(td, newfds)) { 1572 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1573 error = EMSGSIZE; 1574 unp_freerights(rp, newfds); 1575 goto next; 1576 } 1577 1578 /* 1579 * Now change each pointer to an fd in the global 1580 * table to an integer that is the index to the local 1581 * fd table entry that we set up to point to the 1582 * global one we are transferring. 1583 */ 1584 newlen = newfds * sizeof(int); 1585 *controlp = sbcreatecontrol(NULL, newlen, 1586 SCM_RIGHTS, SOL_SOCKET); 1587 if (*controlp == NULL) { 1588 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1589 error = E2BIG; 1590 unp_freerights(rp, newfds); 1591 goto next; 1592 } 1593 1594 fdp = (int *) 1595 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1596 for (i = 0; i < newfds; i++) { 1597 if (fdalloc(td, 0, &f)) 1598 panic("unp_externalize fdalloc failed"); 1599 fp = *rp++; 1600 td->td_proc->p_fd->fd_ofiles[f] = fp; 1601 unp_externalize_fp(fp); 1602 *fdp++ = f; 1603 } 1604 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1605 } else { 1606 /* We can just copy anything else across. */ 1607 if (error || controlp == NULL) 1608 goto next; 1609 *controlp = sbcreatecontrol(NULL, datalen, 1610 cm->cmsg_type, cm->cmsg_level); 1611 if (*controlp == NULL) { 1612 error = ENOBUFS; 1613 goto next; 1614 } 1615 bcopy(data, 1616 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1617 datalen); 1618 } 1619 controlp = &(*controlp)->m_next; 1620 1621 next: 1622 if (CMSG_SPACE(datalen) < clen) { 1623 clen -= CMSG_SPACE(datalen); 1624 cm = (struct cmsghdr *) 1625 ((caddr_t)cm + CMSG_SPACE(datalen)); 1626 } else { 1627 clen = 0; 1628 cm = NULL; 1629 } 1630 } 1631 1632 m_freem(control); 1633 return (error); 1634 } 1635 1636 static void 1637 unp_zone_change(void *tag) 1638 { 1639 1640 uma_zone_set_max(unp_zone, maxsockets); 1641 } 1642 1643 static void 1644 unp_init(void) 1645 { 1646 1647 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1648 NULL, NULL, UMA_ALIGN_PTR, 0); 1649 if (unp_zone == NULL) 1650 panic("unp_init"); 1651 uma_zone_set_max(unp_zone, maxsockets); 1652 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1653 NULL, EVENTHANDLER_PRI_ANY); 1654 LIST_INIT(&unp_dhead); 1655 LIST_INIT(&unp_shead); 1656 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1657 UNP_GLOBAL_LOCK_INIT(); 1658 } 1659 1660 static int 1661 unp_internalize(struct mbuf **controlp, struct thread *td) 1662 { 1663 struct mbuf *control = *controlp; 1664 struct proc *p = td->td_proc; 1665 struct filedesc *fdescp = p->p_fd; 1666 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1667 struct cmsgcred *cmcred; 1668 struct file **rp; 1669 struct file *fp; 1670 struct timeval *tv; 1671 int i, fd, *fdp; 1672 void *data; 1673 socklen_t clen = control->m_len, datalen; 1674 int error, oldfds; 1675 u_int newlen; 1676 1677 UNP_GLOBAL_UNLOCK_ASSERT(); 1678 1679 error = 0; 1680 *controlp = NULL; 1681 while (cm != NULL) { 1682 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1683 || cm->cmsg_len > clen) { 1684 error = EINVAL; 1685 goto out; 1686 } 1687 data = CMSG_DATA(cm); 1688 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1689 1690 switch (cm->cmsg_type) { 1691 /* 1692 * Fill in credential information. 1693 */ 1694 case SCM_CREDS: 1695 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1696 SCM_CREDS, SOL_SOCKET); 1697 if (*controlp == NULL) { 1698 error = ENOBUFS; 1699 goto out; 1700 } 1701 cmcred = (struct cmsgcred *) 1702 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1703 cmcred->cmcred_pid = p->p_pid; 1704 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1705 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1706 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1707 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1708 CMGROUP_MAX); 1709 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1710 cmcred->cmcred_groups[i] = 1711 td->td_ucred->cr_groups[i]; 1712 break; 1713 1714 case SCM_RIGHTS: 1715 oldfds = datalen / sizeof (int); 1716 /* 1717 * Check that all the FDs passed in refer to legal 1718 * files. If not, reject the entire operation. 1719 */ 1720 fdp = data; 1721 FILEDESC_SLOCK(fdescp); 1722 for (i = 0; i < oldfds; i++) { 1723 fd = *fdp++; 1724 if ((unsigned)fd >= fdescp->fd_nfiles || 1725 fdescp->fd_ofiles[fd] == NULL) { 1726 FILEDESC_SUNLOCK(fdescp); 1727 error = EBADF; 1728 goto out; 1729 } 1730 fp = fdescp->fd_ofiles[fd]; 1731 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1732 FILEDESC_SUNLOCK(fdescp); 1733 error = EOPNOTSUPP; 1734 goto out; 1735 } 1736 1737 } 1738 1739 /* 1740 * Now replace the integer FDs with pointers to the 1741 * associated global file table entry.. 1742 */ 1743 newlen = oldfds * sizeof(struct file *); 1744 *controlp = sbcreatecontrol(NULL, newlen, 1745 SCM_RIGHTS, SOL_SOCKET); 1746 if (*controlp == NULL) { 1747 FILEDESC_SUNLOCK(fdescp); 1748 error = E2BIG; 1749 goto out; 1750 } 1751 fdp = data; 1752 rp = (struct file **) 1753 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1754 for (i = 0; i < oldfds; i++) { 1755 fp = fdescp->fd_ofiles[*fdp++]; 1756 *rp++ = fp; 1757 unp_internalize_fp(fp); 1758 } 1759 FILEDESC_SUNLOCK(fdescp); 1760 break; 1761 1762 case SCM_TIMESTAMP: 1763 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1764 SCM_TIMESTAMP, SOL_SOCKET); 1765 if (*controlp == NULL) { 1766 error = ENOBUFS; 1767 goto out; 1768 } 1769 tv = (struct timeval *) 1770 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1771 microtime(tv); 1772 break; 1773 1774 default: 1775 error = EINVAL; 1776 goto out; 1777 } 1778 1779 controlp = &(*controlp)->m_next; 1780 if (CMSG_SPACE(datalen) < clen) { 1781 clen -= CMSG_SPACE(datalen); 1782 cm = (struct cmsghdr *) 1783 ((caddr_t)cm + CMSG_SPACE(datalen)); 1784 } else { 1785 clen = 0; 1786 cm = NULL; 1787 } 1788 } 1789 1790 out: 1791 m_freem(control); 1792 return (error); 1793 } 1794 1795 static struct mbuf * 1796 unp_addsockcred(struct thread *td, struct mbuf *control) 1797 { 1798 struct mbuf *m, *n, *n_prev; 1799 struct sockcred *sc; 1800 const struct cmsghdr *cm; 1801 int ngroups; 1802 int i; 1803 1804 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1805 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1806 if (m == NULL) 1807 return (control); 1808 1809 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1810 sc->sc_uid = td->td_ucred->cr_ruid; 1811 sc->sc_euid = td->td_ucred->cr_uid; 1812 sc->sc_gid = td->td_ucred->cr_rgid; 1813 sc->sc_egid = td->td_ucred->cr_gid; 1814 sc->sc_ngroups = ngroups; 1815 for (i = 0; i < sc->sc_ngroups; i++) 1816 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1817 1818 /* 1819 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1820 * created SCM_CREDS control message (struct sockcred) has another 1821 * format. 1822 */ 1823 if (control != NULL) 1824 for (n = control, n_prev = NULL; n != NULL;) { 1825 cm = mtod(n, struct cmsghdr *); 1826 if (cm->cmsg_level == SOL_SOCKET && 1827 cm->cmsg_type == SCM_CREDS) { 1828 if (n_prev == NULL) 1829 control = n->m_next; 1830 else 1831 n_prev->m_next = n->m_next; 1832 n = m_free(n); 1833 } else { 1834 n_prev = n; 1835 n = n->m_next; 1836 } 1837 } 1838 1839 /* Prepend it to the head. */ 1840 m->m_next = control; 1841 return (m); 1842 } 1843 1844 static struct unpcb * 1845 fptounp(struct file *fp) 1846 { 1847 struct socket *so; 1848 1849 if (fp->f_type != DTYPE_SOCKET) 1850 return (NULL); 1851 if ((so = fp->f_data) == NULL) 1852 return (NULL); 1853 if (so->so_proto->pr_domain != &localdomain) 1854 return (NULL); 1855 return sotounpcb(so); 1856 } 1857 1858 static void 1859 unp_discard(struct file *fp) 1860 { 1861 1862 unp_externalize_fp(fp); 1863 (void) closef(fp, (struct thread *)NULL); 1864 } 1865 1866 static void 1867 unp_internalize_fp(struct file *fp) 1868 { 1869 struct unpcb *unp; 1870 1871 UNP_GLOBAL_WLOCK(); 1872 if ((unp = fptounp(fp)) != NULL) { 1873 unp->unp_file = fp; 1874 unp->unp_msgcount++; 1875 } 1876 fhold(fp); 1877 unp_rights++; 1878 UNP_GLOBAL_WUNLOCK(); 1879 } 1880 1881 static void 1882 unp_externalize_fp(struct file *fp) 1883 { 1884 struct unpcb *unp; 1885 1886 UNP_GLOBAL_WLOCK(); 1887 if ((unp = fptounp(fp)) != NULL) 1888 unp->unp_msgcount--; 1889 unp_rights--; 1890 UNP_GLOBAL_WUNLOCK(); 1891 } 1892 1893 /* 1894 * unp_defer indicates whether additional work has been defered for a future 1895 * pass through unp_gc(). It is thread local and does not require explicit 1896 * synchronization. 1897 */ 1898 static int unp_marked; 1899 static int unp_unreachable; 1900 1901 static void 1902 unp_accessable(struct file *fp) 1903 { 1904 struct unpcb *unp; 1905 1906 if ((unp = fptounp(fp)) == NULL) 1907 return; 1908 if (unp->unp_gcflag & UNPGC_REF) 1909 return; 1910 unp->unp_gcflag &= ~UNPGC_DEAD; 1911 unp->unp_gcflag |= UNPGC_REF; 1912 unp_marked++; 1913 } 1914 1915 static void 1916 unp_gc_process(struct unpcb *unp) 1917 { 1918 struct socket *soa; 1919 struct socket *so; 1920 struct file *fp; 1921 1922 /* Already processed. */ 1923 if (unp->unp_gcflag & UNPGC_SCANNED) 1924 return; 1925 fp = unp->unp_file; 1926 1927 /* 1928 * Check for a socket potentially in a cycle. It must be in a 1929 * queue as indicated by msgcount, and this must equal the file 1930 * reference count. Note that when msgcount is 0 the file is NULL. 1931 */ 1932 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp && 1933 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) { 1934 unp->unp_gcflag |= UNPGC_DEAD; 1935 unp_unreachable++; 1936 return; 1937 } 1938 1939 /* 1940 * Mark all sockets we reference with RIGHTS. 1941 */ 1942 so = unp->unp_socket; 1943 SOCKBUF_LOCK(&so->so_rcv); 1944 unp_scan(so->so_rcv.sb_mb, unp_accessable); 1945 SOCKBUF_UNLOCK(&so->so_rcv); 1946 1947 /* 1948 * Mark all sockets in our accept queue. 1949 */ 1950 ACCEPT_LOCK(); 1951 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 1952 SOCKBUF_LOCK(&soa->so_rcv); 1953 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 1954 SOCKBUF_UNLOCK(&soa->so_rcv); 1955 } 1956 ACCEPT_UNLOCK(); 1957 unp->unp_gcflag |= UNPGC_SCANNED; 1958 } 1959 1960 static int unp_recycled; 1961 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 1962 "Number of unreachable sockets claimed by the garbage collector."); 1963 1964 static int unp_taskcount; 1965 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 1966 "Number of times the garbage collector has run."); 1967 1968 static void 1969 unp_gc(__unused void *arg, int pending) 1970 { 1971 struct unp_head *heads[] = { &unp_dhead, &unp_shead, NULL }; 1972 struct unp_head **head; 1973 struct file **unref; 1974 struct unpcb *unp; 1975 int i; 1976 1977 unp_taskcount++; 1978 UNP_GLOBAL_RLOCK(); 1979 /* 1980 * First clear all gc flags from previous runs. 1981 */ 1982 for (head = heads; *head != NULL; head++) 1983 LIST_FOREACH(unp, *head, unp_link) 1984 unp->unp_gcflag = 0; 1985 1986 /* 1987 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 1988 * is reachable all of the sockets it references are reachable. 1989 * Stop the scan once we do a complete loop without discovering 1990 * a new reachable socket. 1991 */ 1992 do { 1993 unp_unreachable = 0; 1994 unp_marked = 0; 1995 for (head = heads; *head != NULL; head++) 1996 LIST_FOREACH(unp, *head, unp_link) 1997 unp_gc_process(unp); 1998 } while (unp_marked); 1999 UNP_GLOBAL_RUNLOCK(); 2000 if (unp_unreachable == 0) 2001 return; 2002 2003 /* 2004 * Allocate space for a local list of dead unpcbs. 2005 */ 2006 unref = malloc(unp_unreachable * sizeof(struct file *), 2007 M_TEMP, M_WAITOK); 2008 2009 /* 2010 * Iterate looking for sockets which have been specifically marked 2011 * as as unreachable and store them locally. 2012 */ 2013 UNP_GLOBAL_RLOCK(); 2014 for (i = 0, head = heads; *head != NULL; head++) 2015 LIST_FOREACH(unp, *head, unp_link) 2016 if (unp->unp_gcflag & UNPGC_DEAD) { 2017 unref[i++] = unp->unp_file; 2018 fhold(unp->unp_file); 2019 KASSERT(unp->unp_file != NULL, 2020 ("unp_gc: Invalid unpcb.")); 2021 KASSERT(i <= unp_unreachable, 2022 ("unp_gc: incorrect unreachable count.")); 2023 } 2024 UNP_GLOBAL_RUNLOCK(); 2025 2026 /* 2027 * Now flush all sockets, free'ing rights. This will free the 2028 * struct files associated with these sockets but leave each socket 2029 * with one remaining ref. 2030 */ 2031 for (i = 0; i < unp_unreachable; i++) 2032 sorflush(unref[i]->f_data); 2033 2034 /* 2035 * And finally release the sockets so they can be reclaimed. 2036 */ 2037 for (i = 0; i < unp_unreachable; i++) 2038 fdrop(unref[i], NULL); 2039 unp_recycled += unp_unreachable; 2040 free(unref, M_TEMP); 2041 } 2042 2043 static void 2044 unp_dispose(struct mbuf *m) 2045 { 2046 2047 if (m) 2048 unp_scan(m, unp_discard); 2049 } 2050 2051 static void 2052 unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2053 { 2054 struct mbuf *m; 2055 struct file **rp; 2056 struct cmsghdr *cm; 2057 void *data; 2058 int i; 2059 socklen_t clen, datalen; 2060 int qfds; 2061 2062 while (m0 != NULL) { 2063 for (m = m0; m; m = m->m_next) { 2064 if (m->m_type != MT_CONTROL) 2065 continue; 2066 2067 cm = mtod(m, struct cmsghdr *); 2068 clen = m->m_len; 2069 2070 while (cm != NULL) { 2071 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2072 break; 2073 2074 data = CMSG_DATA(cm); 2075 datalen = (caddr_t)cm + cm->cmsg_len 2076 - (caddr_t)data; 2077 2078 if (cm->cmsg_level == SOL_SOCKET && 2079 cm->cmsg_type == SCM_RIGHTS) { 2080 qfds = datalen / sizeof (struct file *); 2081 rp = data; 2082 for (i = 0; i < qfds; i++) 2083 (*op)(*rp++); 2084 } 2085 2086 if (CMSG_SPACE(datalen) < clen) { 2087 clen -= CMSG_SPACE(datalen); 2088 cm = (struct cmsghdr *) 2089 ((caddr_t)cm + CMSG_SPACE(datalen)); 2090 } else { 2091 clen = 0; 2092 cm = NULL; 2093 } 2094 } 2095 } 2096 m0 = m0->m_act; 2097 } 2098 } 2099 2100 #ifdef DDB 2101 static void 2102 db_print_indent(int indent) 2103 { 2104 int i; 2105 2106 for (i = 0; i < indent; i++) 2107 db_printf(" "); 2108 } 2109 2110 static void 2111 db_print_unpflags(int unp_flags) 2112 { 2113 int comma; 2114 2115 comma = 0; 2116 if (unp_flags & UNP_HAVEPC) { 2117 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2118 comma = 1; 2119 } 2120 if (unp_flags & UNP_HAVEPCCACHED) { 2121 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2122 comma = 1; 2123 } 2124 if (unp_flags & UNP_WANTCRED) { 2125 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2126 comma = 1; 2127 } 2128 if (unp_flags & UNP_CONNWAIT) { 2129 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2130 comma = 1; 2131 } 2132 if (unp_flags & UNP_CONNECTING) { 2133 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2134 comma = 1; 2135 } 2136 if (unp_flags & UNP_BINDING) { 2137 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2138 comma = 1; 2139 } 2140 } 2141 2142 static void 2143 db_print_xucred(int indent, struct xucred *xu) 2144 { 2145 int comma, i; 2146 2147 db_print_indent(indent); 2148 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2149 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2150 db_print_indent(indent); 2151 db_printf("cr_groups: "); 2152 comma = 0; 2153 for (i = 0; i < xu->cr_ngroups; i++) { 2154 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2155 comma = 1; 2156 } 2157 db_printf("\n"); 2158 } 2159 2160 static void 2161 db_print_unprefs(int indent, struct unp_head *uh) 2162 { 2163 struct unpcb *unp; 2164 int counter; 2165 2166 counter = 0; 2167 LIST_FOREACH(unp, uh, unp_reflink) { 2168 if (counter % 4 == 0) 2169 db_print_indent(indent); 2170 db_printf("%p ", unp); 2171 if (counter % 4 == 3) 2172 db_printf("\n"); 2173 counter++; 2174 } 2175 if (counter != 0 && counter % 4 != 0) 2176 db_printf("\n"); 2177 } 2178 2179 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2180 { 2181 struct unpcb *unp; 2182 2183 if (!have_addr) { 2184 db_printf("usage: show unpcb <addr>\n"); 2185 return; 2186 } 2187 unp = (struct unpcb *)addr; 2188 2189 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2190 unp->unp_vnode); 2191 2192 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2193 unp->unp_conn); 2194 2195 db_printf("unp_refs:\n"); 2196 db_print_unprefs(2, &unp->unp_refs); 2197 2198 /* XXXRW: Would be nice to print the full address, if any. */ 2199 db_printf("unp_addr: %p\n", unp->unp_addr); 2200 2201 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2202 unp->unp_cc, unp->unp_mbcnt, 2203 (unsigned long long)unp->unp_gencnt); 2204 2205 db_printf("unp_flags: %x (", unp->unp_flags); 2206 db_print_unpflags(unp->unp_flags); 2207 db_printf(")\n"); 2208 2209 db_printf("unp_peercred:\n"); 2210 db_print_xucred(2, &unp->unp_peercred); 2211 2212 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2213 } 2214 #endif 2215