1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All Rights Reserved. 6 * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved. 7 * Copyright (c) 2018 Matthew Macy 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 34 */ 35 36 /* 37 * UNIX Domain (Local) Sockets 38 * 39 * This is an implementation of UNIX (local) domain sockets. Each socket has 40 * an associated struct unpcb (UNIX protocol control block). Stream sockets 41 * may be connected to 0 or 1 other socket. Datagram sockets may be 42 * connected to 0, 1, or many other sockets. Sockets may be created and 43 * connected in pairs (socketpair(2)), or bound/connected to using the file 44 * system name space. For most purposes, only the receive socket buffer is 45 * used, as sending on one socket delivers directly to the receive socket 46 * buffer of a second socket. 47 * 48 * The implementation is substantially complicated by the fact that 49 * "ancillary data", such as file descriptors or credentials, may be passed 50 * across UNIX domain sockets. The potential for passing UNIX domain sockets 51 * over other UNIX domain sockets requires the implementation of a simple 52 * garbage collector to find and tear down cycles of disconnected sockets. 53 * 54 * TODO: 55 * RDM 56 * rethink name space problems 57 * need a proper out-of-band 58 */ 59 60 #include <sys/cdefs.h> 61 __FBSDID("$FreeBSD$"); 62 63 #include "opt_ddb.h" 64 65 #include <sys/param.h> 66 #include <sys/capsicum.h> 67 #include <sys/domain.h> 68 #include <sys/fcntl.h> 69 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 70 #include <sys/eventhandler.h> 71 #include <sys/file.h> 72 #include <sys/filedesc.h> 73 #include <sys/kernel.h> 74 #include <sys/lock.h> 75 #include <sys/mbuf.h> 76 #include <sys/mount.h> 77 #include <sys/mutex.h> 78 #include <sys/namei.h> 79 #include <sys/proc.h> 80 #include <sys/protosw.h> 81 #include <sys/queue.h> 82 #include <sys/resourcevar.h> 83 #include <sys/rwlock.h> 84 #include <sys/socket.h> 85 #include <sys/socketvar.h> 86 #include <sys/signalvar.h> 87 #include <sys/stat.h> 88 #include <sys/sx.h> 89 #include <sys/sysctl.h> 90 #include <sys/systm.h> 91 #include <sys/taskqueue.h> 92 #include <sys/un.h> 93 #include <sys/unpcb.h> 94 #include <sys/vnode.h> 95 96 #include <net/vnet.h> 97 98 #ifdef DDB 99 #include <ddb/ddb.h> 100 #endif 101 102 #include <security/mac/mac_framework.h> 103 104 #include <vm/uma.h> 105 106 MALLOC_DECLARE(M_FILECAPS); 107 108 /* 109 * Locking key: 110 * (l) Locked using list lock 111 * (g) Locked using linkage lock 112 */ 113 114 static uma_zone_t unp_zone; 115 static unp_gen_t unp_gencnt; /* (l) */ 116 static u_int unp_count; /* (l) Count of local sockets. */ 117 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 118 static int unp_rights; /* (g) File descriptors in flight. */ 119 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 120 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 121 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 122 123 struct unp_defer { 124 SLIST_ENTRY(unp_defer) ud_link; 125 struct file *ud_fp; 126 }; 127 static SLIST_HEAD(, unp_defer) unp_defers; 128 static int unp_defers_count; 129 130 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 131 132 /* 133 * Garbage collection of cyclic file descriptor/socket references occurs 134 * asynchronously in a taskqueue context in order to avoid recursion and 135 * reentrance in the UNIX domain socket, file descriptor, and socket layer 136 * code. See unp_gc() for a full description. 137 */ 138 static struct timeout_task unp_gc_task; 139 140 /* 141 * The close of unix domain sockets attached as SCM_RIGHTS is 142 * postponed to the taskqueue, to avoid arbitrary recursion depth. 143 * The attached sockets might have another sockets attached. 144 */ 145 static struct task unp_defer_task; 146 147 /* 148 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 149 * stream sockets, although the total for sender and receiver is actually 150 * only PIPSIZ. 151 * 152 * Datagram sockets really use the sendspace as the maximum datagram size, 153 * and don't really want to reserve the sendspace. Their recvspace should be 154 * large enough for at least one max-size datagram plus address. 155 */ 156 #ifndef PIPSIZ 157 #define PIPSIZ 8192 158 #endif 159 static u_long unpst_sendspace = PIPSIZ; 160 static u_long unpst_recvspace = PIPSIZ; 161 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 162 static u_long unpdg_recvspace = 4*1024; 163 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 164 static u_long unpsp_recvspace = PIPSIZ; 165 166 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 167 "Local domain"); 168 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 170 "SOCK_STREAM"); 171 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 173 "SOCK_DGRAM"); 174 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 175 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 176 "SOCK_SEQPACKET"); 177 178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 179 &unpst_sendspace, 0, "Default stream send space."); 180 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 181 &unpst_recvspace, 0, "Default stream receive space."); 182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 183 &unpdg_sendspace, 0, "Default datagram send space."); 184 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 185 &unpdg_recvspace, 0, "Default datagram receive space."); 186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 187 &unpsp_sendspace, 0, "Default seqpacket send space."); 188 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 189 &unpsp_recvspace, 0, "Default seqpacket receive space."); 190 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 191 "File descriptors in flight."); 192 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 193 &unp_defers_count, 0, 194 "File descriptors deferred to taskqueue for close."); 195 196 /* 197 * Locking and synchronization: 198 * 199 * Three types of locks exist in the local domain socket implementation: a 200 * a global linkage rwlock, the mtxpool lock, and per-unpcb mutexes. 201 * The linkage lock protects the socket count, global generation number, 202 * and stream/datagram global lists. 203 * 204 * The mtxpool lock protects the vnode from being modified while referenced. 205 * Lock ordering requires that it be acquired before any unpcb locks. 206 * 207 * The unpcb lock (unp_mtx) protects all fields in the unpcb. Of particular 208 * note is that this includes the unp_conn field. So long as the unpcb lock 209 * is held the reference to the unpcb pointed to by unp_conn is valid. If we 210 * require that the unpcb pointed to by unp_conn remain live in cases where 211 * we need to drop the unp_mtx as when we need to acquire the lock for a 212 * second unpcb the caller must first acquire an additional reference on the 213 * second unpcb and then revalidate any state (typically check that unp_conn 214 * is non-NULL) upon requiring the initial unpcb lock. The lock ordering 215 * between unpcbs is the conventional ascending address order. Two helper 216 * routines exist for this: 217 * 218 * - unp_pcb_lock2(unp, unp2) - which just acquires the two locks in the 219 * safe ordering. 220 * 221 * - unp_pcb_owned_lock2(unp, unp2, freed) - the lock for unp is held 222 * when called. If unp is unlocked and unp2 is subsequently freed 223 * freed will be set to 1. 224 * 225 * The helper routines for references are: 226 * 227 * - unp_pcb_hold(unp): Can be called any time we currently hold a valid 228 * reference to unp. 229 * 230 * - unp_pcb_rele(unp): The caller must hold the unp lock. If we are 231 * releasing the last reference, detach must have been called thus 232 * unp->unp_socket be NULL. 233 * 234 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 235 * allocated in pru_attach() and freed in pru_detach(). The validity of that 236 * pointer is an invariant, so no lock is required to dereference the so_pcb 237 * pointer if a valid socket reference is held by the caller. In practice, 238 * this is always true during operations performed on a socket. Each unpcb 239 * has a back-pointer to its socket, unp_socket, which will be stable under 240 * the same circumstances. 241 * 242 * This pointer may only be safely dereferenced as long as a valid reference 243 * to the unpcb is held. Typically, this reference will be from the socket, 244 * or from another unpcb when the referring unpcb's lock is held (in order 245 * that the reference not be invalidated during use). For example, to follow 246 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 247 * that detach is not run clearing unp_socket. 248 * 249 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 250 * protocols, bind() is a non-atomic operation, and connect() requires 251 * potential sleeping in the protocol, due to potentially waiting on local or 252 * distributed file systems. We try to separate "lookup" operations, which 253 * may sleep, and the IPC operations themselves, which typically can occur 254 * with relative atomicity as locks can be held over the entire operation. 255 * 256 * Another tricky issue is simultaneous multi-threaded or multi-process 257 * access to a single UNIX domain socket. These are handled by the flags 258 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 259 * binding, both of which involve dropping UNIX domain socket locks in order 260 * to perform namei() and other file system operations. 261 */ 262 static struct rwlock unp_link_rwlock; 263 static struct mtx unp_defers_lock; 264 265 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 266 "unp_link_rwlock") 267 268 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 269 RA_LOCKED) 270 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 271 RA_UNLOCKED) 272 273 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 274 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 275 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 276 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 277 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 278 RA_WLOCKED) 279 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 280 281 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 282 "unp_defer", NULL, MTX_DEF) 283 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 284 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 285 286 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 287 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 288 289 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 290 "unp", "unp", \ 291 MTX_DUPOK|MTX_DEF) 292 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 293 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 294 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 295 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 296 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 297 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 298 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 299 300 static int uipc_connect2(struct socket *, struct socket *); 301 static int uipc_ctloutput(struct socket *, struct sockopt *); 302 static int unp_connect(struct socket *, struct sockaddr *, 303 struct thread *); 304 static int unp_connectat(int, struct socket *, struct sockaddr *, 305 struct thread *); 306 static int unp_connect2(struct socket *so, struct socket *so2, int); 307 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 308 static void unp_dispose(struct socket *so); 309 static void unp_dispose_mbuf(struct mbuf *); 310 static void unp_shutdown(struct unpcb *); 311 static void unp_drop(struct unpcb *); 312 static void unp_gc(__unused void *, int); 313 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 314 static void unp_discard(struct file *); 315 static void unp_freerights(struct filedescent **, int); 316 static void unp_init(void); 317 static int unp_internalize(struct mbuf **, struct thread *); 318 static void unp_internalize_fp(struct file *); 319 static int unp_externalize(struct mbuf *, struct mbuf **, int); 320 static int unp_externalize_fp(struct file *); 321 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 322 static void unp_process_defers(void * __unused, int); 323 324 static void 325 unp_pcb_hold(struct unpcb *unp) 326 { 327 MPASS(unp->unp_refcount); 328 refcount_acquire(&unp->unp_refcount); 329 } 330 331 static int 332 unp_pcb_rele(struct unpcb *unp) 333 { 334 int freed; 335 336 UNP_PCB_LOCK_ASSERT(unp); 337 MPASS(unp->unp_refcount); 338 if ((freed = refcount_release(&unp->unp_refcount))) { 339 /* we got here with having detached? */ 340 MPASS(unp->unp_socket == NULL); 341 UNP_PCB_UNLOCK(unp); 342 UNP_PCB_LOCK_DESTROY(unp); 343 uma_zfree(unp_zone, unp); 344 } 345 return (freed); 346 } 347 348 static void 349 unp_pcb_lock2(struct unpcb *unp, struct unpcb *unp2) 350 { 351 MPASS(unp != unp2); 352 UNP_PCB_UNLOCK_ASSERT(unp); 353 UNP_PCB_UNLOCK_ASSERT(unp2); 354 if ((uintptr_t)unp2 > (uintptr_t)unp) { 355 UNP_PCB_LOCK(unp); 356 UNP_PCB_LOCK(unp2); 357 } else { 358 UNP_PCB_LOCK(unp2); 359 UNP_PCB_LOCK(unp); 360 } 361 } 362 363 static __noinline void 364 unp_pcb_owned_lock2_slowpath(struct unpcb *unp, struct unpcb **unp2p, 365 int *freed) 366 { 367 struct unpcb *unp2; 368 369 unp2 = *unp2p; 370 unp_pcb_hold(unp2); 371 UNP_PCB_UNLOCK(unp); 372 UNP_PCB_LOCK(unp2); 373 UNP_PCB_LOCK(unp); 374 *freed = unp_pcb_rele(unp2); 375 if (*freed) 376 *unp2p = NULL; 377 } 378 379 #define unp_pcb_owned_lock2(unp, unp2, freed) do { \ 380 freed = 0; \ 381 UNP_PCB_LOCK_ASSERT(unp); \ 382 UNP_PCB_UNLOCK_ASSERT(unp2); \ 383 MPASS((unp) != (unp2)); \ 384 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) \ 385 break; \ 386 else if ((uintptr_t)(unp2) > (uintptr_t)(unp)) \ 387 UNP_PCB_LOCK(unp2); \ 388 else \ 389 unp_pcb_owned_lock2_slowpath((unp), &(unp2), &freed); \ 390 } while (0) 391 392 /* 393 * Definitions of protocols supported in the LOCAL domain. 394 */ 395 static struct domain localdomain; 396 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 397 static struct pr_usrreqs uipc_usrreqs_seqpacket; 398 static struct protosw localsw[] = { 399 { 400 .pr_type = SOCK_STREAM, 401 .pr_domain = &localdomain, 402 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 403 .pr_ctloutput = &uipc_ctloutput, 404 .pr_usrreqs = &uipc_usrreqs_stream 405 }, 406 { 407 .pr_type = SOCK_DGRAM, 408 .pr_domain = &localdomain, 409 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 410 .pr_ctloutput = &uipc_ctloutput, 411 .pr_usrreqs = &uipc_usrreqs_dgram 412 }, 413 { 414 .pr_type = SOCK_SEQPACKET, 415 .pr_domain = &localdomain, 416 417 /* 418 * XXXRW: For now, PR_ADDR because soreceive will bump into them 419 * due to our use of sbappendaddr. A new sbappend variants is needed 420 * that supports both atomic record writes and control data. 421 */ 422 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| 423 PR_RIGHTS, 424 .pr_ctloutput = &uipc_ctloutput, 425 .pr_usrreqs = &uipc_usrreqs_seqpacket, 426 }, 427 }; 428 429 static struct domain localdomain = { 430 .dom_family = AF_LOCAL, 431 .dom_name = "local", 432 .dom_init = unp_init, 433 .dom_externalize = unp_externalize, 434 .dom_dispose = unp_dispose, 435 .dom_protosw = localsw, 436 .dom_protoswNPROTOSW = &localsw[nitems(localsw)] 437 }; 438 DOMAIN_SET(local); 439 440 static void 441 uipc_abort(struct socket *so) 442 { 443 struct unpcb *unp, *unp2; 444 445 unp = sotounpcb(so); 446 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 447 UNP_PCB_UNLOCK_ASSERT(unp); 448 449 UNP_PCB_LOCK(unp); 450 unp2 = unp->unp_conn; 451 if (unp2 != NULL) { 452 unp_pcb_hold(unp2); 453 UNP_PCB_UNLOCK(unp); 454 unp_drop(unp2); 455 } else 456 UNP_PCB_UNLOCK(unp); 457 } 458 459 static int 460 uipc_accept(struct socket *so, struct sockaddr **nam) 461 { 462 struct unpcb *unp, *unp2; 463 const struct sockaddr *sa; 464 465 /* 466 * Pass back name of connected socket, if it was bound and we are 467 * still connected (our peer may have closed already!). 468 */ 469 unp = sotounpcb(so); 470 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 471 472 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 473 UNP_LINK_RLOCK(); 474 unp2 = unp->unp_conn; 475 if (unp2 != NULL && unp2->unp_addr != NULL) { 476 UNP_PCB_LOCK(unp2); 477 sa = (struct sockaddr *) unp2->unp_addr; 478 bcopy(sa, *nam, sa->sa_len); 479 UNP_PCB_UNLOCK(unp2); 480 } else { 481 sa = &sun_noname; 482 bcopy(sa, *nam, sa->sa_len); 483 } 484 UNP_LINK_RUNLOCK(); 485 return (0); 486 } 487 488 static int 489 uipc_attach(struct socket *so, int proto, struct thread *td) 490 { 491 u_long sendspace, recvspace; 492 struct unpcb *unp; 493 int error; 494 bool locked; 495 496 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 497 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 498 switch (so->so_type) { 499 case SOCK_STREAM: 500 sendspace = unpst_sendspace; 501 recvspace = unpst_recvspace; 502 break; 503 504 case SOCK_DGRAM: 505 sendspace = unpdg_sendspace; 506 recvspace = unpdg_recvspace; 507 break; 508 509 case SOCK_SEQPACKET: 510 sendspace = unpsp_sendspace; 511 recvspace = unpsp_recvspace; 512 break; 513 514 default: 515 panic("uipc_attach"); 516 } 517 error = soreserve(so, sendspace, recvspace); 518 if (error) 519 return (error); 520 } 521 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 522 if (unp == NULL) 523 return (ENOBUFS); 524 LIST_INIT(&unp->unp_refs); 525 UNP_PCB_LOCK_INIT(unp); 526 unp->unp_socket = so; 527 so->so_pcb = unp; 528 unp->unp_refcount = 1; 529 530 if ((locked = UNP_LINK_WOWNED()) == false) 531 UNP_LINK_WLOCK(); 532 533 unp->unp_gencnt = ++unp_gencnt; 534 unp->unp_ino = ++unp_ino; 535 unp_count++; 536 switch (so->so_type) { 537 case SOCK_STREAM: 538 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 539 break; 540 541 case SOCK_DGRAM: 542 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 543 break; 544 545 case SOCK_SEQPACKET: 546 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 547 break; 548 549 default: 550 panic("uipc_attach"); 551 } 552 553 if (locked == false) 554 UNP_LINK_WUNLOCK(); 555 556 return (0); 557 } 558 559 static int 560 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 561 { 562 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 563 struct vattr vattr; 564 int error, namelen; 565 struct nameidata nd; 566 struct unpcb *unp; 567 struct vnode *vp; 568 struct mount *mp; 569 cap_rights_t rights; 570 char *buf; 571 572 if (nam->sa_family != AF_UNIX) 573 return (EAFNOSUPPORT); 574 575 unp = sotounpcb(so); 576 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 577 578 if (soun->sun_len > sizeof(struct sockaddr_un)) 579 return (EINVAL); 580 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 581 if (namelen <= 0) 582 return (EINVAL); 583 584 /* 585 * We don't allow simultaneous bind() calls on a single UNIX domain 586 * socket, so flag in-progress operations, and return an error if an 587 * operation is already in progress. 588 * 589 * Historically, we have not allowed a socket to be rebound, so this 590 * also returns an error. Not allowing re-binding simplifies the 591 * implementation and avoids a great many possible failure modes. 592 */ 593 UNP_PCB_LOCK(unp); 594 if (unp->unp_vnode != NULL) { 595 UNP_PCB_UNLOCK(unp); 596 return (EINVAL); 597 } 598 if (unp->unp_flags & UNP_BINDING) { 599 UNP_PCB_UNLOCK(unp); 600 return (EALREADY); 601 } 602 unp->unp_flags |= UNP_BINDING; 603 UNP_PCB_UNLOCK(unp); 604 605 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 606 bcopy(soun->sun_path, buf, namelen); 607 buf[namelen] = 0; 608 609 restart: 610 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, 611 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td); 612 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 613 error = namei(&nd); 614 if (error) 615 goto error; 616 vp = nd.ni_vp; 617 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 618 NDFREE(&nd, NDF_ONLY_PNBUF); 619 if (nd.ni_dvp == vp) 620 vrele(nd.ni_dvp); 621 else 622 vput(nd.ni_dvp); 623 if (vp != NULL) { 624 vrele(vp); 625 error = EADDRINUSE; 626 goto error; 627 } 628 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 629 if (error) 630 goto error; 631 goto restart; 632 } 633 VATTR_NULL(&vattr); 634 vattr.va_type = VSOCK; 635 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 636 #ifdef MAC 637 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 638 &vattr); 639 #endif 640 if (error == 0) 641 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 642 NDFREE(&nd, NDF_ONLY_PNBUF); 643 vput(nd.ni_dvp); 644 if (error) { 645 vn_finished_write(mp); 646 goto error; 647 } 648 vp = nd.ni_vp; 649 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 650 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 651 652 UNP_PCB_LOCK(unp); 653 VOP_UNP_BIND(vp, unp); 654 unp->unp_vnode = vp; 655 unp->unp_addr = soun; 656 unp->unp_flags &= ~UNP_BINDING; 657 UNP_PCB_UNLOCK(unp); 658 VOP_UNLOCK(vp); 659 vn_finished_write(mp); 660 free(buf, M_TEMP); 661 return (0); 662 663 error: 664 UNP_PCB_LOCK(unp); 665 unp->unp_flags &= ~UNP_BINDING; 666 UNP_PCB_UNLOCK(unp); 667 free(buf, M_TEMP); 668 return (error); 669 } 670 671 static int 672 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 673 { 674 675 return (uipc_bindat(AT_FDCWD, so, nam, td)); 676 } 677 678 static int 679 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 680 { 681 int error; 682 683 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 684 error = unp_connect(so, nam, td); 685 return (error); 686 } 687 688 static int 689 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 690 struct thread *td) 691 { 692 int error; 693 694 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 695 error = unp_connectat(fd, so, nam, td); 696 return (error); 697 } 698 699 static void 700 uipc_close(struct socket *so) 701 { 702 struct unpcb *unp, *unp2; 703 struct vnode *vp = NULL; 704 struct mtx *vplock; 705 int freed; 706 unp = sotounpcb(so); 707 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 708 709 vplock = NULL; 710 if ((vp = unp->unp_vnode) != NULL) { 711 vplock = mtx_pool_find(mtxpool_sleep, vp); 712 mtx_lock(vplock); 713 } 714 UNP_PCB_LOCK(unp); 715 if (vp && unp->unp_vnode == NULL) { 716 mtx_unlock(vplock); 717 vp = NULL; 718 } 719 if (vp != NULL) { 720 VOP_UNP_DETACH(vp); 721 unp->unp_vnode = NULL; 722 } 723 unp2 = unp->unp_conn; 724 unp_pcb_hold(unp); 725 if (__predict_false(unp == unp2)) { 726 unp_disconnect(unp, unp2); 727 } else if (unp2 != NULL) { 728 unp_pcb_hold(unp2); 729 unp_pcb_owned_lock2(unp, unp2, freed); 730 unp_disconnect(unp, unp2); 731 if (unp_pcb_rele(unp2) == 0) 732 UNP_PCB_UNLOCK(unp2); 733 } 734 if (unp_pcb_rele(unp) == 0) 735 UNP_PCB_UNLOCK(unp); 736 if (vp) { 737 mtx_unlock(vplock); 738 vrele(vp); 739 } 740 } 741 742 static int 743 uipc_connect2(struct socket *so1, struct socket *so2) 744 { 745 struct unpcb *unp, *unp2; 746 int error; 747 748 unp = so1->so_pcb; 749 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 750 unp2 = so2->so_pcb; 751 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 752 if (unp != unp2) 753 unp_pcb_lock2(unp, unp2); 754 else 755 UNP_PCB_LOCK(unp); 756 error = unp_connect2(so1, so2, PRU_CONNECT2); 757 if (unp != unp2) 758 UNP_PCB_UNLOCK(unp2); 759 UNP_PCB_UNLOCK(unp); 760 return (error); 761 } 762 763 static void 764 uipc_detach(struct socket *so) 765 { 766 struct unpcb *unp, *unp2; 767 struct mtx *vplock; 768 struct vnode *vp; 769 int freeunp, local_unp_rights; 770 771 unp = sotounpcb(so); 772 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 773 774 vp = NULL; 775 vplock = NULL; 776 local_unp_rights = 0; 777 778 SOCK_LOCK(so); 779 if (!SOLISTENING(so)) { 780 /* 781 * Once the socket is removed from the global lists, 782 * uipc_ready() will not be able to locate its socket buffer, so 783 * clear the buffer now. At this point internalized rights have 784 * already been disposed of. 785 */ 786 sbrelease(&so->so_rcv, so); 787 } 788 SOCK_UNLOCK(so); 789 790 UNP_LINK_WLOCK(); 791 LIST_REMOVE(unp, unp_link); 792 if (unp->unp_gcflag & UNPGC_DEAD) 793 LIST_REMOVE(unp, unp_dead); 794 unp->unp_gencnt = ++unp_gencnt; 795 --unp_count; 796 UNP_LINK_WUNLOCK(); 797 798 UNP_PCB_UNLOCK_ASSERT(unp); 799 restart: 800 if ((vp = unp->unp_vnode) != NULL) { 801 vplock = mtx_pool_find(mtxpool_sleep, vp); 802 mtx_lock(vplock); 803 } 804 UNP_PCB_LOCK(unp); 805 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 806 if (vplock) 807 mtx_unlock(vplock); 808 UNP_PCB_UNLOCK(unp); 809 goto restart; 810 } 811 if ((vp = unp->unp_vnode) != NULL) { 812 VOP_UNP_DETACH(vp); 813 unp->unp_vnode = NULL; 814 } 815 if (__predict_false(unp == unp->unp_conn)) { 816 unp_disconnect(unp, unp); 817 unp2 = NULL; 818 } else { 819 if ((unp2 = unp->unp_conn) != NULL) { 820 unp_pcb_owned_lock2(unp, unp2, freeunp); 821 if (freeunp) 822 unp2 = NULL; 823 } 824 unp_pcb_hold(unp); 825 if (unp2 != NULL) { 826 unp_pcb_hold(unp2); 827 unp_disconnect(unp, unp2); 828 if (unp_pcb_rele(unp2) == 0) 829 UNP_PCB_UNLOCK(unp2); 830 } 831 } 832 UNP_PCB_UNLOCK(unp); 833 UNP_REF_LIST_LOCK(); 834 while (!LIST_EMPTY(&unp->unp_refs)) { 835 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 836 837 unp_pcb_hold(ref); 838 UNP_REF_LIST_UNLOCK(); 839 840 MPASS(ref != unp); 841 UNP_PCB_UNLOCK_ASSERT(ref); 842 unp_drop(ref); 843 UNP_REF_LIST_LOCK(); 844 } 845 846 UNP_REF_LIST_UNLOCK(); 847 UNP_PCB_LOCK(unp); 848 freeunp = unp_pcb_rele(unp); 849 MPASS(freeunp == 0); 850 local_unp_rights = unp_rights; 851 unp->unp_socket->so_pcb = NULL; 852 unp->unp_socket = NULL; 853 free(unp->unp_addr, M_SONAME); 854 unp->unp_addr = NULL; 855 if (!unp_pcb_rele(unp)) 856 UNP_PCB_UNLOCK(unp); 857 if (vp) { 858 mtx_unlock(vplock); 859 vrele(vp); 860 } 861 if (local_unp_rights) 862 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 863 } 864 865 static int 866 uipc_disconnect(struct socket *so) 867 { 868 struct unpcb *unp, *unp2; 869 int freed; 870 871 unp = sotounpcb(so); 872 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 873 874 UNP_PCB_LOCK(unp); 875 if ((unp2 = unp->unp_conn) == NULL) { 876 UNP_PCB_UNLOCK(unp); 877 return (0); 878 } 879 if (__predict_true(unp != unp2)) { 880 unp_pcb_owned_lock2(unp, unp2, freed); 881 if (__predict_false(freed)) { 882 UNP_PCB_UNLOCK(unp); 883 return (0); 884 } 885 unp_pcb_hold(unp2); 886 } 887 unp_pcb_hold(unp); 888 unp_disconnect(unp, unp2); 889 if (unp_pcb_rele(unp) == 0) 890 UNP_PCB_UNLOCK(unp); 891 if ((unp != unp2) && unp_pcb_rele(unp2) == 0) 892 UNP_PCB_UNLOCK(unp2); 893 return (0); 894 } 895 896 static int 897 uipc_listen(struct socket *so, int backlog, struct thread *td) 898 { 899 struct unpcb *unp; 900 int error; 901 902 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) 903 return (EOPNOTSUPP); 904 905 unp = sotounpcb(so); 906 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 907 908 UNP_PCB_LOCK(unp); 909 if (unp->unp_vnode == NULL) { 910 /* Already connected or not bound to an address. */ 911 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ; 912 UNP_PCB_UNLOCK(unp); 913 return (error); 914 } 915 916 SOCK_LOCK(so); 917 error = solisten_proto_check(so); 918 if (error == 0) { 919 cru2xt(td, &unp->unp_peercred); 920 solisten_proto(so, backlog); 921 } 922 SOCK_UNLOCK(so); 923 UNP_PCB_UNLOCK(unp); 924 return (error); 925 } 926 927 static int 928 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 929 { 930 struct unpcb *unp, *unp2; 931 const struct sockaddr *sa; 932 933 unp = sotounpcb(so); 934 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 935 936 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 937 UNP_LINK_RLOCK(); 938 /* 939 * XXX: It seems that this test always fails even when connection is 940 * established. So, this else clause is added as workaround to 941 * return PF_LOCAL sockaddr. 942 */ 943 unp2 = unp->unp_conn; 944 if (unp2 != NULL) { 945 UNP_PCB_LOCK(unp2); 946 if (unp2->unp_addr != NULL) 947 sa = (struct sockaddr *) unp2->unp_addr; 948 else 949 sa = &sun_noname; 950 bcopy(sa, *nam, sa->sa_len); 951 UNP_PCB_UNLOCK(unp2); 952 } else { 953 sa = &sun_noname; 954 bcopy(sa, *nam, sa->sa_len); 955 } 956 UNP_LINK_RUNLOCK(); 957 return (0); 958 } 959 960 static int 961 uipc_rcvd(struct socket *so, int flags) 962 { 963 struct unpcb *unp, *unp2; 964 struct socket *so2; 965 u_int mbcnt, sbcc; 966 967 unp = sotounpcb(so); 968 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 969 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 970 ("%s: socktype %d", __func__, so->so_type)); 971 972 /* 973 * Adjust backpressure on sender and wakeup any waiting to write. 974 * 975 * The unp lock is acquired to maintain the validity of the unp_conn 976 * pointer; no lock on unp2 is required as unp2->unp_socket will be 977 * static as long as we don't permit unp2 to disconnect from unp, 978 * which is prevented by the lock on unp. We cache values from 979 * so_rcv to avoid holding the so_rcv lock over the entire 980 * transaction on the remote so_snd. 981 */ 982 SOCKBUF_LOCK(&so->so_rcv); 983 mbcnt = so->so_rcv.sb_mbcnt; 984 sbcc = sbavail(&so->so_rcv); 985 SOCKBUF_UNLOCK(&so->so_rcv); 986 /* 987 * There is a benign race condition at this point. If we're planning to 988 * clear SB_STOP, but uipc_send is called on the connected socket at 989 * this instant, it might add data to the sockbuf and set SB_STOP. Then 990 * we would erroneously clear SB_STOP below, even though the sockbuf is 991 * full. The race is benign because the only ill effect is to allow the 992 * sockbuf to exceed its size limit, and the size limits are not 993 * strictly guaranteed anyway. 994 */ 995 UNP_PCB_LOCK(unp); 996 unp2 = unp->unp_conn; 997 if (unp2 == NULL) { 998 UNP_PCB_UNLOCK(unp); 999 return (0); 1000 } 1001 so2 = unp2->unp_socket; 1002 SOCKBUF_LOCK(&so2->so_snd); 1003 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 1004 so2->so_snd.sb_flags &= ~SB_STOP; 1005 sowwakeup_locked(so2); 1006 UNP_PCB_UNLOCK(unp); 1007 return (0); 1008 } 1009 1010 static int 1011 connect_internal(struct socket *so, struct sockaddr *nam, struct thread *td) 1012 { 1013 int error; 1014 struct unpcb *unp; 1015 1016 unp = so->so_pcb; 1017 if (unp->unp_conn != NULL) 1018 return (EISCONN); 1019 error = unp_connect(so, nam, td); 1020 if (error) 1021 return (error); 1022 UNP_PCB_LOCK(unp); 1023 if (unp->unp_conn == NULL) { 1024 UNP_PCB_UNLOCK(unp); 1025 if (error == 0) 1026 error = ENOTCONN; 1027 } 1028 return (error); 1029 } 1030 1031 static int 1032 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 1033 struct mbuf *control, struct thread *td) 1034 { 1035 struct unpcb *unp, *unp2; 1036 struct socket *so2; 1037 u_int mbcnt, sbcc; 1038 int freed, error; 1039 1040 unp = sotounpcb(so); 1041 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 1042 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM || 1043 so->so_type == SOCK_SEQPACKET, 1044 ("%s: socktype %d", __func__, so->so_type)); 1045 1046 freed = error = 0; 1047 if (flags & PRUS_OOB) { 1048 error = EOPNOTSUPP; 1049 goto release; 1050 } 1051 if (control != NULL && (error = unp_internalize(&control, td))) 1052 goto release; 1053 1054 unp2 = NULL; 1055 switch (so->so_type) { 1056 case SOCK_DGRAM: 1057 { 1058 const struct sockaddr *from; 1059 1060 if (nam != NULL) { 1061 /* 1062 * We return with UNP_PCB_LOCK_HELD so we know that 1063 * the reference is live if the pointer is valid. 1064 */ 1065 if ((error = connect_internal(so, nam, td))) 1066 break; 1067 MPASS(unp->unp_conn != NULL); 1068 unp2 = unp->unp_conn; 1069 } else { 1070 UNP_PCB_LOCK(unp); 1071 1072 /* 1073 * Because connect() and send() are non-atomic in a sendto() 1074 * with a target address, it's possible that the socket will 1075 * have disconnected before the send() can run. In that case 1076 * return the slightly counter-intuitive but otherwise 1077 * correct error that the socket is not connected. 1078 */ 1079 if ((unp2 = unp->unp_conn) == NULL) { 1080 UNP_PCB_UNLOCK(unp); 1081 error = ENOTCONN; 1082 break; 1083 } 1084 } 1085 if (__predict_false(unp == unp2)) { 1086 if (unp->unp_socket == NULL) { 1087 error = ENOTCONN; 1088 break; 1089 } 1090 goto connect_self; 1091 } 1092 unp_pcb_owned_lock2(unp, unp2, freed); 1093 if (__predict_false(freed)) { 1094 UNP_PCB_UNLOCK(unp); 1095 error = ENOTCONN; 1096 break; 1097 } 1098 /* 1099 * The socket referencing unp2 may have been closed 1100 * or unp may have been disconnected if the unp lock 1101 * was dropped to acquire unp2. 1102 */ 1103 if (__predict_false(unp->unp_conn == NULL) || 1104 unp2->unp_socket == NULL) { 1105 UNP_PCB_UNLOCK(unp); 1106 if (unp_pcb_rele(unp2) == 0) 1107 UNP_PCB_UNLOCK(unp2); 1108 error = ENOTCONN; 1109 break; 1110 } 1111 connect_self: 1112 if (unp2->unp_flags & UNP_WANTCRED) 1113 control = unp_addsockcred(td, control); 1114 if (unp->unp_addr != NULL) 1115 from = (struct sockaddr *)unp->unp_addr; 1116 else 1117 from = &sun_noname; 1118 so2 = unp2->unp_socket; 1119 SOCKBUF_LOCK(&so2->so_rcv); 1120 if (sbappendaddr_locked(&so2->so_rcv, from, m, 1121 control)) { 1122 sorwakeup_locked(so2); 1123 m = NULL; 1124 control = NULL; 1125 } else { 1126 SOCKBUF_UNLOCK(&so2->so_rcv); 1127 error = ENOBUFS; 1128 } 1129 if (nam != NULL) 1130 unp_disconnect(unp, unp2); 1131 if (__predict_true(unp != unp2)) 1132 UNP_PCB_UNLOCK(unp2); 1133 UNP_PCB_UNLOCK(unp); 1134 break; 1135 } 1136 1137 case SOCK_SEQPACKET: 1138 case SOCK_STREAM: 1139 if ((so->so_state & SS_ISCONNECTED) == 0) { 1140 if (nam != NULL) { 1141 error = connect_internal(so, nam, td); 1142 if (error != 0) 1143 break; 1144 } else { 1145 error = ENOTCONN; 1146 break; 1147 } 1148 } else { 1149 UNP_PCB_LOCK(unp); 1150 } 1151 1152 if ((unp2 = unp->unp_conn) == NULL) { 1153 UNP_PCB_UNLOCK(unp); 1154 error = ENOTCONN; 1155 break; 1156 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1157 UNP_PCB_UNLOCK(unp); 1158 error = EPIPE; 1159 break; 1160 } else if ((unp2 = unp->unp_conn) == NULL) { 1161 UNP_PCB_UNLOCK(unp); 1162 error = ENOTCONN; 1163 break; 1164 } 1165 unp_pcb_owned_lock2(unp, unp2, freed); 1166 UNP_PCB_UNLOCK(unp); 1167 if (__predict_false(freed)) { 1168 error = ENOTCONN; 1169 break; 1170 } 1171 if ((so2 = unp2->unp_socket) == NULL) { 1172 UNP_PCB_UNLOCK(unp2); 1173 error = ENOTCONN; 1174 break; 1175 } 1176 SOCKBUF_LOCK(&so2->so_rcv); 1177 if (unp2->unp_flags & UNP_WANTCRED) { 1178 /* 1179 * Credentials are passed only once on SOCK_STREAM 1180 * and SOCK_SEQPACKET. 1181 */ 1182 unp2->unp_flags &= ~UNP_WANTCRED; 1183 control = unp_addsockcred(td, control); 1184 } 1185 1186 /* 1187 * Send to paired receive port and wake up readers. Don't 1188 * check for space available in the receive buffer if we're 1189 * attaching ancillary data; Unix domain sockets only check 1190 * for space in the sending sockbuf, and that check is 1191 * performed one level up the stack. At that level we cannot 1192 * precisely account for the amount of buffer space used 1193 * (e.g., because control messages are not yet internalized). 1194 */ 1195 switch (so->so_type) { 1196 case SOCK_STREAM: 1197 if (control != NULL) { 1198 sbappendcontrol_locked(&so2->so_rcv, m, 1199 control, flags); 1200 control = NULL; 1201 } else 1202 sbappend_locked(&so2->so_rcv, m, flags); 1203 break; 1204 1205 case SOCK_SEQPACKET: 1206 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1207 &sun_noname, m, control)) 1208 control = NULL; 1209 break; 1210 } 1211 1212 mbcnt = so2->so_rcv.sb_mbcnt; 1213 sbcc = sbavail(&so2->so_rcv); 1214 if (sbcc) 1215 sorwakeup_locked(so2); 1216 else 1217 SOCKBUF_UNLOCK(&so2->so_rcv); 1218 1219 /* 1220 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1221 * it would be possible for uipc_rcvd to be called at this 1222 * point, drain the receiving sockbuf, clear SB_STOP, and then 1223 * we would set SB_STOP below. That could lead to an empty 1224 * sockbuf having SB_STOP set 1225 */ 1226 SOCKBUF_LOCK(&so->so_snd); 1227 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1228 so->so_snd.sb_flags |= SB_STOP; 1229 SOCKBUF_UNLOCK(&so->so_snd); 1230 UNP_PCB_UNLOCK(unp2); 1231 m = NULL; 1232 break; 1233 } 1234 1235 /* 1236 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1237 */ 1238 if (flags & PRUS_EOF) { 1239 UNP_PCB_LOCK(unp); 1240 socantsendmore(so); 1241 unp_shutdown(unp); 1242 UNP_PCB_UNLOCK(unp); 1243 } 1244 if (control != NULL && error != 0) 1245 unp_dispose_mbuf(control); 1246 1247 release: 1248 if (control != NULL) 1249 m_freem(control); 1250 /* 1251 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1252 * for freeing memory. 1253 */ 1254 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1255 m_freem(m); 1256 return (error); 1257 } 1258 1259 static bool 1260 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1261 { 1262 struct mbuf *mb, *n; 1263 struct sockbuf *sb; 1264 1265 SOCK_LOCK(so); 1266 if (SOLISTENING(so)) { 1267 SOCK_UNLOCK(so); 1268 return (false); 1269 } 1270 mb = NULL; 1271 sb = &so->so_rcv; 1272 SOCKBUF_LOCK(sb); 1273 if (sb->sb_fnrdy != NULL) { 1274 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1275 if (mb == m) { 1276 *errorp = sbready(sb, m, count); 1277 break; 1278 } 1279 mb = mb->m_next; 1280 if (mb == NULL) { 1281 mb = n; 1282 if (mb != NULL) 1283 n = mb->m_nextpkt; 1284 } 1285 } 1286 } 1287 SOCKBUF_UNLOCK(sb); 1288 SOCK_UNLOCK(so); 1289 return (mb != NULL); 1290 } 1291 1292 static int 1293 uipc_ready(struct socket *so, struct mbuf *m, int count) 1294 { 1295 struct unpcb *unp, *unp2; 1296 struct socket *so2; 1297 int error, i; 1298 1299 unp = sotounpcb(so); 1300 1301 KASSERT(so->so_type == SOCK_STREAM, 1302 ("%s: unexpected socket type for %p", __func__, so)); 1303 1304 UNP_PCB_LOCK(unp); 1305 if ((unp2 = unp->unp_conn) == NULL) { 1306 UNP_PCB_UNLOCK(unp); 1307 goto search; 1308 } 1309 if (unp != unp2) { 1310 if (UNP_PCB_TRYLOCK(unp2) == 0) { 1311 unp_pcb_hold(unp2); 1312 UNP_PCB_UNLOCK(unp); 1313 UNP_PCB_LOCK(unp2); 1314 if (unp_pcb_rele(unp2)) 1315 goto search; 1316 } else 1317 UNP_PCB_UNLOCK(unp); 1318 } 1319 so2 = unp2->unp_socket; 1320 SOCKBUF_LOCK(&so2->so_rcv); 1321 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1322 sorwakeup_locked(so2); 1323 else 1324 SOCKBUF_UNLOCK(&so2->so_rcv); 1325 UNP_PCB_UNLOCK(unp2); 1326 return (error); 1327 1328 search: 1329 /* 1330 * The receiving socket has been disconnected, but may still be valid. 1331 * In this case, the now-ready mbufs are still present in its socket 1332 * buffer, so perform an exhaustive search before giving up and freeing 1333 * the mbufs. 1334 */ 1335 UNP_LINK_RLOCK(); 1336 LIST_FOREACH(unp, &unp_shead, unp_link) { 1337 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1338 break; 1339 } 1340 UNP_LINK_RUNLOCK(); 1341 1342 if (unp == NULL) { 1343 for (i = 0; i < count; i++) 1344 m = m_free(m); 1345 error = ECONNRESET; 1346 } 1347 return (error); 1348 } 1349 1350 static int 1351 uipc_sense(struct socket *so, struct stat *sb) 1352 { 1353 struct unpcb *unp; 1354 1355 unp = sotounpcb(so); 1356 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1357 1358 sb->st_blksize = so->so_snd.sb_hiwat; 1359 sb->st_dev = NODEV; 1360 sb->st_ino = unp->unp_ino; 1361 return (0); 1362 } 1363 1364 static int 1365 uipc_shutdown(struct socket *so) 1366 { 1367 struct unpcb *unp; 1368 1369 unp = sotounpcb(so); 1370 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1371 1372 UNP_PCB_LOCK(unp); 1373 socantsendmore(so); 1374 unp_shutdown(unp); 1375 UNP_PCB_UNLOCK(unp); 1376 return (0); 1377 } 1378 1379 static int 1380 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1381 { 1382 struct unpcb *unp; 1383 const struct sockaddr *sa; 1384 1385 unp = sotounpcb(so); 1386 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1387 1388 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1389 UNP_PCB_LOCK(unp); 1390 if (unp->unp_addr != NULL) 1391 sa = (struct sockaddr *) unp->unp_addr; 1392 else 1393 sa = &sun_noname; 1394 bcopy(sa, *nam, sa->sa_len); 1395 UNP_PCB_UNLOCK(unp); 1396 return (0); 1397 } 1398 1399 static struct pr_usrreqs uipc_usrreqs_dgram = { 1400 .pru_abort = uipc_abort, 1401 .pru_accept = uipc_accept, 1402 .pru_attach = uipc_attach, 1403 .pru_bind = uipc_bind, 1404 .pru_bindat = uipc_bindat, 1405 .pru_connect = uipc_connect, 1406 .pru_connectat = uipc_connectat, 1407 .pru_connect2 = uipc_connect2, 1408 .pru_detach = uipc_detach, 1409 .pru_disconnect = uipc_disconnect, 1410 .pru_listen = uipc_listen, 1411 .pru_peeraddr = uipc_peeraddr, 1412 .pru_rcvd = uipc_rcvd, 1413 .pru_send = uipc_send, 1414 .pru_sense = uipc_sense, 1415 .pru_shutdown = uipc_shutdown, 1416 .pru_sockaddr = uipc_sockaddr, 1417 .pru_soreceive = soreceive_dgram, 1418 .pru_close = uipc_close, 1419 }; 1420 1421 static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1422 .pru_abort = uipc_abort, 1423 .pru_accept = uipc_accept, 1424 .pru_attach = uipc_attach, 1425 .pru_bind = uipc_bind, 1426 .pru_bindat = uipc_bindat, 1427 .pru_connect = uipc_connect, 1428 .pru_connectat = uipc_connectat, 1429 .pru_connect2 = uipc_connect2, 1430 .pru_detach = uipc_detach, 1431 .pru_disconnect = uipc_disconnect, 1432 .pru_listen = uipc_listen, 1433 .pru_peeraddr = uipc_peeraddr, 1434 .pru_rcvd = uipc_rcvd, 1435 .pru_send = uipc_send, 1436 .pru_sense = uipc_sense, 1437 .pru_shutdown = uipc_shutdown, 1438 .pru_sockaddr = uipc_sockaddr, 1439 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1440 .pru_close = uipc_close, 1441 }; 1442 1443 static struct pr_usrreqs uipc_usrreqs_stream = { 1444 .pru_abort = uipc_abort, 1445 .pru_accept = uipc_accept, 1446 .pru_attach = uipc_attach, 1447 .pru_bind = uipc_bind, 1448 .pru_bindat = uipc_bindat, 1449 .pru_connect = uipc_connect, 1450 .pru_connectat = uipc_connectat, 1451 .pru_connect2 = uipc_connect2, 1452 .pru_detach = uipc_detach, 1453 .pru_disconnect = uipc_disconnect, 1454 .pru_listen = uipc_listen, 1455 .pru_peeraddr = uipc_peeraddr, 1456 .pru_rcvd = uipc_rcvd, 1457 .pru_send = uipc_send, 1458 .pru_ready = uipc_ready, 1459 .pru_sense = uipc_sense, 1460 .pru_shutdown = uipc_shutdown, 1461 .pru_sockaddr = uipc_sockaddr, 1462 .pru_soreceive = soreceive_generic, 1463 .pru_close = uipc_close, 1464 }; 1465 1466 static int 1467 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1468 { 1469 struct unpcb *unp; 1470 struct xucred xu; 1471 int error, optval; 1472 1473 if (sopt->sopt_level != SOL_LOCAL) 1474 return (EINVAL); 1475 1476 unp = sotounpcb(so); 1477 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1478 error = 0; 1479 switch (sopt->sopt_dir) { 1480 case SOPT_GET: 1481 switch (sopt->sopt_name) { 1482 case LOCAL_PEERCRED: 1483 UNP_PCB_LOCK(unp); 1484 if (unp->unp_flags & UNP_HAVEPC) 1485 xu = unp->unp_peercred; 1486 else { 1487 if (so->so_type == SOCK_STREAM) 1488 error = ENOTCONN; 1489 else 1490 error = EINVAL; 1491 } 1492 UNP_PCB_UNLOCK(unp); 1493 if (error == 0) 1494 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1495 break; 1496 1497 case LOCAL_CREDS: 1498 /* Unlocked read. */ 1499 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1500 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1501 break; 1502 1503 case LOCAL_CONNWAIT: 1504 /* Unlocked read. */ 1505 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1506 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1507 break; 1508 1509 default: 1510 error = EOPNOTSUPP; 1511 break; 1512 } 1513 break; 1514 1515 case SOPT_SET: 1516 switch (sopt->sopt_name) { 1517 case LOCAL_CREDS: 1518 case LOCAL_CONNWAIT: 1519 error = sooptcopyin(sopt, &optval, sizeof(optval), 1520 sizeof(optval)); 1521 if (error) 1522 break; 1523 1524 #define OPTSET(bit) do { \ 1525 UNP_PCB_LOCK(unp); \ 1526 if (optval) \ 1527 unp->unp_flags |= bit; \ 1528 else \ 1529 unp->unp_flags &= ~bit; \ 1530 UNP_PCB_UNLOCK(unp); \ 1531 } while (0) 1532 1533 switch (sopt->sopt_name) { 1534 case LOCAL_CREDS: 1535 OPTSET(UNP_WANTCRED); 1536 break; 1537 1538 case LOCAL_CONNWAIT: 1539 OPTSET(UNP_CONNWAIT); 1540 break; 1541 1542 default: 1543 break; 1544 } 1545 break; 1546 #undef OPTSET 1547 default: 1548 error = ENOPROTOOPT; 1549 break; 1550 } 1551 break; 1552 1553 default: 1554 error = EOPNOTSUPP; 1555 break; 1556 } 1557 return (error); 1558 } 1559 1560 static int 1561 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1562 { 1563 1564 return (unp_connectat(AT_FDCWD, so, nam, td)); 1565 } 1566 1567 static int 1568 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1569 struct thread *td) 1570 { 1571 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1572 struct vnode *vp; 1573 struct socket *so2; 1574 struct unpcb *unp, *unp2, *unp3; 1575 struct nameidata nd; 1576 char buf[SOCK_MAXADDRLEN]; 1577 struct sockaddr *sa; 1578 cap_rights_t rights; 1579 int error, len, freed; 1580 struct mtx *vplock; 1581 1582 if (nam->sa_family != AF_UNIX) 1583 return (EAFNOSUPPORT); 1584 if (nam->sa_len > sizeof(struct sockaddr_un)) 1585 return (EINVAL); 1586 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1587 if (len <= 0) 1588 return (EINVAL); 1589 bcopy(soun->sun_path, buf, len); 1590 buf[len] = 0; 1591 1592 unp = sotounpcb(so); 1593 UNP_PCB_LOCK(unp); 1594 if (unp->unp_flags & UNP_CONNECTING) { 1595 UNP_PCB_UNLOCK(unp); 1596 return (EALREADY); 1597 } 1598 unp->unp_flags |= UNP_CONNECTING; 1599 UNP_PCB_UNLOCK(unp); 1600 1601 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1602 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1603 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td); 1604 error = namei(&nd); 1605 if (error) 1606 vp = NULL; 1607 else 1608 vp = nd.ni_vp; 1609 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1610 NDFREE(&nd, NDF_ONLY_PNBUF); 1611 if (error) 1612 goto bad; 1613 1614 if (vp->v_type != VSOCK) { 1615 error = ENOTSOCK; 1616 goto bad; 1617 } 1618 #ifdef MAC 1619 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1620 if (error) 1621 goto bad; 1622 #endif 1623 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1624 if (error) 1625 goto bad; 1626 1627 unp = sotounpcb(so); 1628 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1629 1630 vplock = mtx_pool_find(mtxpool_sleep, vp); 1631 mtx_lock(vplock); 1632 VOP_UNP_CONNECT(vp, &unp2); 1633 if (unp2 == NULL) { 1634 error = ECONNREFUSED; 1635 goto bad2; 1636 } 1637 so2 = unp2->unp_socket; 1638 if (so->so_type != so2->so_type) { 1639 error = EPROTOTYPE; 1640 goto bad2; 1641 } 1642 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1643 if (so2->so_options & SO_ACCEPTCONN) { 1644 CURVNET_SET(so2->so_vnet); 1645 so2 = sonewconn(so2, 0); 1646 CURVNET_RESTORE(); 1647 } else 1648 so2 = NULL; 1649 if (so2 == NULL) { 1650 error = ECONNREFUSED; 1651 goto bad2; 1652 } 1653 unp3 = sotounpcb(so2); 1654 unp_pcb_lock2(unp2, unp3); 1655 if (unp2->unp_addr != NULL) { 1656 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1657 unp3->unp_addr = (struct sockaddr_un *) sa; 1658 sa = NULL; 1659 } 1660 1661 unp_copy_peercred(td, unp3, unp, unp2); 1662 1663 UNP_PCB_UNLOCK(unp2); 1664 unp2 = unp3; 1665 unp_pcb_owned_lock2(unp2, unp, freed); 1666 if (__predict_false(freed)) { 1667 UNP_PCB_UNLOCK(unp2); 1668 error = ECONNREFUSED; 1669 goto bad2; 1670 } 1671 #ifdef MAC 1672 mac_socketpeer_set_from_socket(so, so2); 1673 mac_socketpeer_set_from_socket(so2, so); 1674 #endif 1675 } else { 1676 if (unp == unp2) 1677 UNP_PCB_LOCK(unp); 1678 else 1679 unp_pcb_lock2(unp, unp2); 1680 } 1681 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1682 sotounpcb(so2) == unp2, 1683 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1684 error = unp_connect2(so, so2, PRU_CONNECT); 1685 if (unp != unp2) 1686 UNP_PCB_UNLOCK(unp2); 1687 UNP_PCB_UNLOCK(unp); 1688 bad2: 1689 mtx_unlock(vplock); 1690 bad: 1691 if (vp != NULL) { 1692 vput(vp); 1693 } 1694 free(sa, M_SONAME); 1695 UNP_PCB_LOCK(unp); 1696 unp->unp_flags &= ~UNP_CONNECTING; 1697 UNP_PCB_UNLOCK(unp); 1698 return (error); 1699 } 1700 1701 /* 1702 * Set socket peer credentials at connection time. 1703 * 1704 * The client's PCB credentials are copied from its process structure. The 1705 * server's PCB credentials are copied from the socket on which it called 1706 * listen(2). uipc_listen cached that process's credentials at the time. 1707 */ 1708 void 1709 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 1710 struct unpcb *server_unp, struct unpcb *listen_unp) 1711 { 1712 cru2xt(td, &client_unp->unp_peercred); 1713 client_unp->unp_flags |= UNP_HAVEPC; 1714 1715 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 1716 sizeof(server_unp->unp_peercred)); 1717 server_unp->unp_flags |= UNP_HAVEPC; 1718 if (listen_unp->unp_flags & UNP_WANTCRED) 1719 client_unp->unp_flags |= UNP_WANTCRED; 1720 } 1721 1722 static int 1723 unp_connect2(struct socket *so, struct socket *so2, int req) 1724 { 1725 struct unpcb *unp; 1726 struct unpcb *unp2; 1727 1728 unp = sotounpcb(so); 1729 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1730 unp2 = sotounpcb(so2); 1731 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1732 1733 UNP_PCB_LOCK_ASSERT(unp); 1734 UNP_PCB_LOCK_ASSERT(unp2); 1735 1736 if (so2->so_type != so->so_type) 1737 return (EPROTOTYPE); 1738 unp->unp_conn = unp2; 1739 unp_pcb_hold(unp2); 1740 unp_pcb_hold(unp); 1741 switch (so->so_type) { 1742 case SOCK_DGRAM: 1743 UNP_REF_LIST_LOCK(); 1744 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1745 UNP_REF_LIST_UNLOCK(); 1746 soisconnected(so); 1747 break; 1748 1749 case SOCK_STREAM: 1750 case SOCK_SEQPACKET: 1751 unp2->unp_conn = unp; 1752 if (req == PRU_CONNECT && 1753 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1754 soisconnecting(so); 1755 else 1756 soisconnected(so); 1757 soisconnected(so2); 1758 break; 1759 1760 default: 1761 panic("unp_connect2"); 1762 } 1763 return (0); 1764 } 1765 1766 static void 1767 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1768 { 1769 struct socket *so, *so2; 1770 int freed __unused; 1771 1772 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1773 1774 UNP_PCB_LOCK_ASSERT(unp); 1775 UNP_PCB_LOCK_ASSERT(unp2); 1776 1777 if (unp->unp_conn == NULL && unp2->unp_conn == NULL) 1778 return; 1779 1780 MPASS(unp->unp_conn == unp2); 1781 unp->unp_conn = NULL; 1782 so = unp->unp_socket; 1783 so2 = unp2->unp_socket; 1784 switch (unp->unp_socket->so_type) { 1785 case SOCK_DGRAM: 1786 UNP_REF_LIST_LOCK(); 1787 LIST_REMOVE(unp, unp_reflink); 1788 UNP_REF_LIST_UNLOCK(); 1789 if (so) { 1790 SOCK_LOCK(so); 1791 so->so_state &= ~SS_ISCONNECTED; 1792 SOCK_UNLOCK(so); 1793 } 1794 break; 1795 1796 case SOCK_STREAM: 1797 case SOCK_SEQPACKET: 1798 if (so) 1799 soisdisconnected(so); 1800 MPASS(unp2->unp_conn == unp); 1801 unp2->unp_conn = NULL; 1802 if (so2) 1803 soisdisconnected(so2); 1804 break; 1805 } 1806 freed = unp_pcb_rele(unp); 1807 MPASS(freed == 0); 1808 freed = unp_pcb_rele(unp2); 1809 MPASS(freed == 0); 1810 } 1811 1812 /* 1813 * unp_pcblist() walks the global list of struct unpcb's to generate a 1814 * pointer list, bumping the refcount on each unpcb. It then copies them out 1815 * sequentially, validating the generation number on each to see if it has 1816 * been detached. All of this is necessary because copyout() may sleep on 1817 * disk I/O. 1818 */ 1819 static int 1820 unp_pcblist(SYSCTL_HANDLER_ARGS) 1821 { 1822 struct unpcb *unp, **unp_list; 1823 unp_gen_t gencnt; 1824 struct xunpgen *xug; 1825 struct unp_head *head; 1826 struct xunpcb *xu; 1827 u_int i; 1828 int error, freeunp, n; 1829 1830 switch ((intptr_t)arg1) { 1831 case SOCK_STREAM: 1832 head = &unp_shead; 1833 break; 1834 1835 case SOCK_DGRAM: 1836 head = &unp_dhead; 1837 break; 1838 1839 case SOCK_SEQPACKET: 1840 head = &unp_sphead; 1841 break; 1842 1843 default: 1844 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 1845 } 1846 1847 /* 1848 * The process of preparing the PCB list is too time-consuming and 1849 * resource-intensive to repeat twice on every request. 1850 */ 1851 if (req->oldptr == NULL) { 1852 n = unp_count; 1853 req->oldidx = 2 * (sizeof *xug) 1854 + (n + n/8) * sizeof(struct xunpcb); 1855 return (0); 1856 } 1857 1858 if (req->newptr != NULL) 1859 return (EPERM); 1860 1861 /* 1862 * OK, now we're committed to doing something. 1863 */ 1864 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 1865 UNP_LINK_RLOCK(); 1866 gencnt = unp_gencnt; 1867 n = unp_count; 1868 UNP_LINK_RUNLOCK(); 1869 1870 xug->xug_len = sizeof *xug; 1871 xug->xug_count = n; 1872 xug->xug_gen = gencnt; 1873 xug->xug_sogen = so_gencnt; 1874 error = SYSCTL_OUT(req, xug, sizeof *xug); 1875 if (error) { 1876 free(xug, M_TEMP); 1877 return (error); 1878 } 1879 1880 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1881 1882 UNP_LINK_RLOCK(); 1883 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1884 unp = LIST_NEXT(unp, unp_link)) { 1885 UNP_PCB_LOCK(unp); 1886 if (unp->unp_gencnt <= gencnt) { 1887 if (cr_cansee(req->td->td_ucred, 1888 unp->unp_socket->so_cred)) { 1889 UNP_PCB_UNLOCK(unp); 1890 continue; 1891 } 1892 unp_list[i++] = unp; 1893 unp_pcb_hold(unp); 1894 } 1895 UNP_PCB_UNLOCK(unp); 1896 } 1897 UNP_LINK_RUNLOCK(); 1898 n = i; /* In case we lost some during malloc. */ 1899 1900 error = 0; 1901 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1902 for (i = 0; i < n; i++) { 1903 unp = unp_list[i]; 1904 UNP_PCB_LOCK(unp); 1905 freeunp = unp_pcb_rele(unp); 1906 1907 if (freeunp == 0 && unp->unp_gencnt <= gencnt) { 1908 xu->xu_len = sizeof *xu; 1909 xu->xu_unpp = (uintptr_t)unp; 1910 /* 1911 * XXX - need more locking here to protect against 1912 * connect/disconnect races for SMP. 1913 */ 1914 if (unp->unp_addr != NULL) 1915 bcopy(unp->unp_addr, &xu->xu_addr, 1916 unp->unp_addr->sun_len); 1917 else 1918 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 1919 if (unp->unp_conn != NULL && 1920 unp->unp_conn->unp_addr != NULL) 1921 bcopy(unp->unp_conn->unp_addr, 1922 &xu->xu_caddr, 1923 unp->unp_conn->unp_addr->sun_len); 1924 else 1925 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 1926 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 1927 xu->unp_conn = (uintptr_t)unp->unp_conn; 1928 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 1929 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 1930 xu->unp_gencnt = unp->unp_gencnt; 1931 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1932 UNP_PCB_UNLOCK(unp); 1933 error = SYSCTL_OUT(req, xu, sizeof *xu); 1934 } else if (freeunp == 0) 1935 UNP_PCB_UNLOCK(unp); 1936 } 1937 free(xu, M_TEMP); 1938 if (!error) { 1939 /* 1940 * Give the user an updated idea of our state. If the 1941 * generation differs from what we told her before, she knows 1942 * that something happened while we were processing this 1943 * request, and it might be necessary to retry. 1944 */ 1945 xug->xug_gen = unp_gencnt; 1946 xug->xug_sogen = so_gencnt; 1947 xug->xug_count = unp_count; 1948 error = SYSCTL_OUT(req, xug, sizeof *xug); 1949 } 1950 free(unp_list, M_TEMP); 1951 free(xug, M_TEMP); 1952 return (error); 1953 } 1954 1955 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 1956 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1957 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1958 "List of active local datagram sockets"); 1959 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 1960 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1961 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1962 "List of active local stream sockets"); 1963 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 1964 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1965 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 1966 "List of active local seqpacket sockets"); 1967 1968 static void 1969 unp_shutdown(struct unpcb *unp) 1970 { 1971 struct unpcb *unp2; 1972 struct socket *so; 1973 1974 UNP_PCB_LOCK_ASSERT(unp); 1975 1976 unp2 = unp->unp_conn; 1977 if ((unp->unp_socket->so_type == SOCK_STREAM || 1978 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 1979 so = unp2->unp_socket; 1980 if (so != NULL) 1981 socantrcvmore(so); 1982 } 1983 } 1984 1985 static void 1986 unp_drop(struct unpcb *unp) 1987 { 1988 struct socket *so = unp->unp_socket; 1989 struct unpcb *unp2; 1990 int freed; 1991 1992 /* 1993 * Regardless of whether the socket's peer dropped the connection 1994 * with this socket by aborting or disconnecting, POSIX requires 1995 * that ECONNRESET is returned. 1996 */ 1997 /* acquire a reference so that unp isn't freed from underneath us */ 1998 1999 UNP_PCB_LOCK(unp); 2000 if (so) 2001 so->so_error = ECONNRESET; 2002 unp2 = unp->unp_conn; 2003 if (unp2 == unp) { 2004 unp_disconnect(unp, unp2); 2005 } else if (unp2 != NULL) { 2006 unp_pcb_hold(unp2); 2007 unp_pcb_owned_lock2(unp, unp2, freed); 2008 unp_disconnect(unp, unp2); 2009 if (unp_pcb_rele(unp2) == 0) 2010 UNP_PCB_UNLOCK(unp2); 2011 } 2012 if (unp_pcb_rele(unp) == 0) 2013 UNP_PCB_UNLOCK(unp); 2014 } 2015 2016 static void 2017 unp_freerights(struct filedescent **fdep, int fdcount) 2018 { 2019 struct file *fp; 2020 int i; 2021 2022 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2023 2024 for (i = 0; i < fdcount; i++) { 2025 fp = fdep[i]->fde_file; 2026 filecaps_free(&fdep[i]->fde_caps); 2027 unp_discard(fp); 2028 } 2029 free(fdep[0], M_FILECAPS); 2030 } 2031 2032 static int 2033 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2034 { 2035 struct thread *td = curthread; /* XXX */ 2036 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2037 int i; 2038 int *fdp; 2039 struct filedesc *fdesc = td->td_proc->p_fd; 2040 struct filedescent **fdep; 2041 void *data; 2042 socklen_t clen = control->m_len, datalen; 2043 int error, newfds; 2044 u_int newlen; 2045 2046 UNP_LINK_UNLOCK_ASSERT(); 2047 2048 error = 0; 2049 if (controlp != NULL) /* controlp == NULL => free control messages */ 2050 *controlp = NULL; 2051 while (cm != NULL) { 2052 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 2053 error = EINVAL; 2054 break; 2055 } 2056 data = CMSG_DATA(cm); 2057 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2058 if (cm->cmsg_level == SOL_SOCKET 2059 && cm->cmsg_type == SCM_RIGHTS) { 2060 newfds = datalen / sizeof(*fdep); 2061 if (newfds == 0) 2062 goto next; 2063 fdep = data; 2064 2065 /* If we're not outputting the descriptors free them. */ 2066 if (error || controlp == NULL) { 2067 unp_freerights(fdep, newfds); 2068 goto next; 2069 } 2070 FILEDESC_XLOCK(fdesc); 2071 2072 /* 2073 * Now change each pointer to an fd in the global 2074 * table to an integer that is the index to the local 2075 * fd table entry that we set up to point to the 2076 * global one we are transferring. 2077 */ 2078 newlen = newfds * sizeof(int); 2079 *controlp = sbcreatecontrol(NULL, newlen, 2080 SCM_RIGHTS, SOL_SOCKET); 2081 if (*controlp == NULL) { 2082 FILEDESC_XUNLOCK(fdesc); 2083 error = E2BIG; 2084 unp_freerights(fdep, newfds); 2085 goto next; 2086 } 2087 2088 fdp = (int *) 2089 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2090 if (fdallocn(td, 0, fdp, newfds) != 0) { 2091 FILEDESC_XUNLOCK(fdesc); 2092 error = EMSGSIZE; 2093 unp_freerights(fdep, newfds); 2094 m_freem(*controlp); 2095 *controlp = NULL; 2096 goto next; 2097 } 2098 for (i = 0; i < newfds; i++, fdp++) { 2099 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2100 (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0, 2101 &fdep[i]->fde_caps); 2102 unp_externalize_fp(fdep[i]->fde_file); 2103 } 2104 2105 /* 2106 * The new type indicates that the mbuf data refers to 2107 * kernel resources that may need to be released before 2108 * the mbuf is freed. 2109 */ 2110 m_chtype(*controlp, MT_EXTCONTROL); 2111 FILEDESC_XUNLOCK(fdesc); 2112 free(fdep[0], M_FILECAPS); 2113 } else { 2114 /* We can just copy anything else across. */ 2115 if (error || controlp == NULL) 2116 goto next; 2117 *controlp = sbcreatecontrol(NULL, datalen, 2118 cm->cmsg_type, cm->cmsg_level); 2119 if (*controlp == NULL) { 2120 error = ENOBUFS; 2121 goto next; 2122 } 2123 bcopy(data, 2124 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2125 datalen); 2126 } 2127 controlp = &(*controlp)->m_next; 2128 2129 next: 2130 if (CMSG_SPACE(datalen) < clen) { 2131 clen -= CMSG_SPACE(datalen); 2132 cm = (struct cmsghdr *) 2133 ((caddr_t)cm + CMSG_SPACE(datalen)); 2134 } else { 2135 clen = 0; 2136 cm = NULL; 2137 } 2138 } 2139 2140 m_freem(control); 2141 return (error); 2142 } 2143 2144 static void 2145 unp_zone_change(void *tag) 2146 { 2147 2148 uma_zone_set_max(unp_zone, maxsockets); 2149 } 2150 2151 static void 2152 unp_init(void) 2153 { 2154 2155 #ifdef VIMAGE 2156 if (!IS_DEFAULT_VNET(curvnet)) 2157 return; 2158 #endif 2159 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 2160 NULL, NULL, UMA_ALIGN_CACHE, 0); 2161 if (unp_zone == NULL) 2162 panic("unp_init"); 2163 uma_zone_set_max(unp_zone, maxsockets); 2164 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2165 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2166 NULL, EVENTHANDLER_PRI_ANY); 2167 LIST_INIT(&unp_dhead); 2168 LIST_INIT(&unp_shead); 2169 LIST_INIT(&unp_sphead); 2170 SLIST_INIT(&unp_defers); 2171 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2172 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2173 UNP_LINK_LOCK_INIT(); 2174 UNP_DEFERRED_LOCK_INIT(); 2175 } 2176 2177 static void 2178 unp_internalize_cleanup_rights(struct mbuf *control) 2179 { 2180 struct cmsghdr *cp; 2181 struct mbuf *m; 2182 void *data; 2183 socklen_t datalen; 2184 2185 for (m = control; m != NULL; m = m->m_next) { 2186 cp = mtod(m, struct cmsghdr *); 2187 if (cp->cmsg_level != SOL_SOCKET || 2188 cp->cmsg_type != SCM_RIGHTS) 2189 continue; 2190 data = CMSG_DATA(cp); 2191 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2192 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2193 } 2194 } 2195 2196 static int 2197 unp_internalize(struct mbuf **controlp, struct thread *td) 2198 { 2199 struct mbuf *control, **initial_controlp; 2200 struct proc *p; 2201 struct filedesc *fdesc; 2202 struct bintime *bt; 2203 struct cmsghdr *cm; 2204 struct cmsgcred *cmcred; 2205 struct filedescent *fde, **fdep, *fdev; 2206 struct file *fp; 2207 struct timeval *tv; 2208 struct timespec *ts; 2209 void *data; 2210 socklen_t clen, datalen; 2211 int i, j, error, *fdp, oldfds; 2212 u_int newlen; 2213 2214 UNP_LINK_UNLOCK_ASSERT(); 2215 2216 p = td->td_proc; 2217 fdesc = p->p_fd; 2218 error = 0; 2219 control = *controlp; 2220 clen = control->m_len; 2221 *controlp = NULL; 2222 initial_controlp = controlp; 2223 for (cm = mtod(control, struct cmsghdr *); cm != NULL;) { 2224 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 2225 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) { 2226 error = EINVAL; 2227 goto out; 2228 } 2229 data = CMSG_DATA(cm); 2230 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2231 2232 switch (cm->cmsg_type) { 2233 /* 2234 * Fill in credential information. 2235 */ 2236 case SCM_CREDS: 2237 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2238 SCM_CREDS, SOL_SOCKET); 2239 if (*controlp == NULL) { 2240 error = ENOBUFS; 2241 goto out; 2242 } 2243 cmcred = (struct cmsgcred *) 2244 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2245 cmcred->cmcred_pid = p->p_pid; 2246 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2247 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2248 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2249 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2250 CMGROUP_MAX); 2251 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2252 cmcred->cmcred_groups[i] = 2253 td->td_ucred->cr_groups[i]; 2254 break; 2255 2256 case SCM_RIGHTS: 2257 oldfds = datalen / sizeof (int); 2258 if (oldfds == 0) 2259 break; 2260 /* 2261 * Check that all the FDs passed in refer to legal 2262 * files. If not, reject the entire operation. 2263 */ 2264 fdp = data; 2265 FILEDESC_SLOCK(fdesc); 2266 for (i = 0; i < oldfds; i++, fdp++) { 2267 fp = fget_locked(fdesc, *fdp); 2268 if (fp == NULL) { 2269 FILEDESC_SUNLOCK(fdesc); 2270 error = EBADF; 2271 goto out; 2272 } 2273 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2274 FILEDESC_SUNLOCK(fdesc); 2275 error = EOPNOTSUPP; 2276 goto out; 2277 } 2278 } 2279 2280 /* 2281 * Now replace the integer FDs with pointers to the 2282 * file structure and capability rights. 2283 */ 2284 newlen = oldfds * sizeof(fdep[0]); 2285 *controlp = sbcreatecontrol(NULL, newlen, 2286 SCM_RIGHTS, SOL_SOCKET); 2287 if (*controlp == NULL) { 2288 FILEDESC_SUNLOCK(fdesc); 2289 error = E2BIG; 2290 goto out; 2291 } 2292 fdp = data; 2293 for (i = 0; i < oldfds; i++, fdp++) { 2294 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2295 fdp = data; 2296 for (j = 0; j < i; j++, fdp++) { 2297 fdrop(fdesc->fd_ofiles[*fdp]. 2298 fde_file, td); 2299 } 2300 FILEDESC_SUNLOCK(fdesc); 2301 error = EBADF; 2302 goto out; 2303 } 2304 } 2305 fdp = data; 2306 fdep = (struct filedescent **) 2307 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2308 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2309 M_WAITOK); 2310 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2311 fde = &fdesc->fd_ofiles[*fdp]; 2312 fdep[i] = fdev; 2313 fdep[i]->fde_file = fde->fde_file; 2314 filecaps_copy(&fde->fde_caps, 2315 &fdep[i]->fde_caps, true); 2316 unp_internalize_fp(fdep[i]->fde_file); 2317 } 2318 FILEDESC_SUNLOCK(fdesc); 2319 break; 2320 2321 case SCM_TIMESTAMP: 2322 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2323 SCM_TIMESTAMP, SOL_SOCKET); 2324 if (*controlp == NULL) { 2325 error = ENOBUFS; 2326 goto out; 2327 } 2328 tv = (struct timeval *) 2329 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2330 microtime(tv); 2331 break; 2332 2333 case SCM_BINTIME: 2334 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2335 SCM_BINTIME, SOL_SOCKET); 2336 if (*controlp == NULL) { 2337 error = ENOBUFS; 2338 goto out; 2339 } 2340 bt = (struct bintime *) 2341 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2342 bintime(bt); 2343 break; 2344 2345 case SCM_REALTIME: 2346 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2347 SCM_REALTIME, SOL_SOCKET); 2348 if (*controlp == NULL) { 2349 error = ENOBUFS; 2350 goto out; 2351 } 2352 ts = (struct timespec *) 2353 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2354 nanotime(ts); 2355 break; 2356 2357 case SCM_MONOTONIC: 2358 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2359 SCM_MONOTONIC, SOL_SOCKET); 2360 if (*controlp == NULL) { 2361 error = ENOBUFS; 2362 goto out; 2363 } 2364 ts = (struct timespec *) 2365 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2366 nanouptime(ts); 2367 break; 2368 2369 default: 2370 error = EINVAL; 2371 goto out; 2372 } 2373 2374 if (*controlp != NULL) 2375 controlp = &(*controlp)->m_next; 2376 if (CMSG_SPACE(datalen) < clen) { 2377 clen -= CMSG_SPACE(datalen); 2378 cm = (struct cmsghdr *) 2379 ((caddr_t)cm + CMSG_SPACE(datalen)); 2380 } else { 2381 clen = 0; 2382 cm = NULL; 2383 } 2384 } 2385 2386 out: 2387 if (error != 0 && initial_controlp != NULL) 2388 unp_internalize_cleanup_rights(*initial_controlp); 2389 m_freem(control); 2390 return (error); 2391 } 2392 2393 static struct mbuf * 2394 unp_addsockcred(struct thread *td, struct mbuf *control) 2395 { 2396 struct mbuf *m, *n, *n_prev; 2397 struct sockcred *sc; 2398 const struct cmsghdr *cm; 2399 int ngroups; 2400 int i; 2401 2402 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2403 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 2404 if (m == NULL) 2405 return (control); 2406 2407 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 2408 sc->sc_uid = td->td_ucred->cr_ruid; 2409 sc->sc_euid = td->td_ucred->cr_uid; 2410 sc->sc_gid = td->td_ucred->cr_rgid; 2411 sc->sc_egid = td->td_ucred->cr_gid; 2412 sc->sc_ngroups = ngroups; 2413 for (i = 0; i < sc->sc_ngroups; i++) 2414 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2415 2416 /* 2417 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2418 * created SCM_CREDS control message (struct sockcred) has another 2419 * format. 2420 */ 2421 if (control != NULL) 2422 for (n = control, n_prev = NULL; n != NULL;) { 2423 cm = mtod(n, struct cmsghdr *); 2424 if (cm->cmsg_level == SOL_SOCKET && 2425 cm->cmsg_type == SCM_CREDS) { 2426 if (n_prev == NULL) 2427 control = n->m_next; 2428 else 2429 n_prev->m_next = n->m_next; 2430 n = m_free(n); 2431 } else { 2432 n_prev = n; 2433 n = n->m_next; 2434 } 2435 } 2436 2437 /* Prepend it to the head. */ 2438 m->m_next = control; 2439 return (m); 2440 } 2441 2442 static struct unpcb * 2443 fptounp(struct file *fp) 2444 { 2445 struct socket *so; 2446 2447 if (fp->f_type != DTYPE_SOCKET) 2448 return (NULL); 2449 if ((so = fp->f_data) == NULL) 2450 return (NULL); 2451 if (so->so_proto->pr_domain != &localdomain) 2452 return (NULL); 2453 return sotounpcb(so); 2454 } 2455 2456 static void 2457 unp_discard(struct file *fp) 2458 { 2459 struct unp_defer *dr; 2460 2461 if (unp_externalize_fp(fp)) { 2462 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2463 dr->ud_fp = fp; 2464 UNP_DEFERRED_LOCK(); 2465 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2466 UNP_DEFERRED_UNLOCK(); 2467 atomic_add_int(&unp_defers_count, 1); 2468 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2469 } else 2470 (void) closef(fp, (struct thread *)NULL); 2471 } 2472 2473 static void 2474 unp_process_defers(void *arg __unused, int pending) 2475 { 2476 struct unp_defer *dr; 2477 SLIST_HEAD(, unp_defer) drl; 2478 int count; 2479 2480 SLIST_INIT(&drl); 2481 for (;;) { 2482 UNP_DEFERRED_LOCK(); 2483 if (SLIST_FIRST(&unp_defers) == NULL) { 2484 UNP_DEFERRED_UNLOCK(); 2485 break; 2486 } 2487 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2488 UNP_DEFERRED_UNLOCK(); 2489 count = 0; 2490 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2491 SLIST_REMOVE_HEAD(&drl, ud_link); 2492 closef(dr->ud_fp, NULL); 2493 free(dr, M_TEMP); 2494 count++; 2495 } 2496 atomic_add_int(&unp_defers_count, -count); 2497 } 2498 } 2499 2500 static void 2501 unp_internalize_fp(struct file *fp) 2502 { 2503 struct unpcb *unp; 2504 2505 UNP_LINK_WLOCK(); 2506 if ((unp = fptounp(fp)) != NULL) { 2507 unp->unp_file = fp; 2508 unp->unp_msgcount++; 2509 } 2510 unp_rights++; 2511 UNP_LINK_WUNLOCK(); 2512 } 2513 2514 static int 2515 unp_externalize_fp(struct file *fp) 2516 { 2517 struct unpcb *unp; 2518 int ret; 2519 2520 UNP_LINK_WLOCK(); 2521 if ((unp = fptounp(fp)) != NULL) { 2522 unp->unp_msgcount--; 2523 ret = 1; 2524 } else 2525 ret = 0; 2526 unp_rights--; 2527 UNP_LINK_WUNLOCK(); 2528 return (ret); 2529 } 2530 2531 /* 2532 * unp_defer indicates whether additional work has been defered for a future 2533 * pass through unp_gc(). It is thread local and does not require explicit 2534 * synchronization. 2535 */ 2536 static int unp_marked; 2537 2538 static void 2539 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2540 { 2541 struct unpcb *unp; 2542 struct file *fp; 2543 int i; 2544 2545 /* 2546 * This function can only be called from the gc task. 2547 */ 2548 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2549 ("%s: not on gc callout", __func__)); 2550 UNP_LINK_LOCK_ASSERT(); 2551 2552 for (i = 0; i < fdcount; i++) { 2553 fp = fdep[i]->fde_file; 2554 if ((unp = fptounp(fp)) == NULL) 2555 continue; 2556 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2557 continue; 2558 unp->unp_gcrefs--; 2559 } 2560 } 2561 2562 static void 2563 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2564 { 2565 struct unpcb *unp; 2566 struct file *fp; 2567 int i; 2568 2569 /* 2570 * This function can only be called from the gc task. 2571 */ 2572 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2573 ("%s: not on gc callout", __func__)); 2574 UNP_LINK_LOCK_ASSERT(); 2575 2576 for (i = 0; i < fdcount; i++) { 2577 fp = fdep[i]->fde_file; 2578 if ((unp = fptounp(fp)) == NULL) 2579 continue; 2580 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2581 continue; 2582 unp->unp_gcrefs++; 2583 unp_marked++; 2584 } 2585 } 2586 2587 static void 2588 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 2589 { 2590 struct socket *so, *soa; 2591 2592 so = unp->unp_socket; 2593 SOCK_LOCK(so); 2594 if (SOLISTENING(so)) { 2595 /* 2596 * Mark all sockets in our accept queue. 2597 */ 2598 TAILQ_FOREACH(soa, &so->sol_comp, so_list) { 2599 if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 2600 continue; 2601 SOCKBUF_LOCK(&soa->so_rcv); 2602 unp_scan(soa->so_rcv.sb_mb, op); 2603 SOCKBUF_UNLOCK(&soa->so_rcv); 2604 } 2605 } else { 2606 /* 2607 * Mark all sockets we reference with RIGHTS. 2608 */ 2609 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) { 2610 SOCKBUF_LOCK(&so->so_rcv); 2611 unp_scan(so->so_rcv.sb_mb, op); 2612 SOCKBUF_UNLOCK(&so->so_rcv); 2613 } 2614 } 2615 SOCK_UNLOCK(so); 2616 } 2617 2618 static int unp_recycled; 2619 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 2620 "Number of unreachable sockets claimed by the garbage collector."); 2621 2622 static int unp_taskcount; 2623 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 2624 "Number of times the garbage collector has run."); 2625 2626 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 2627 "Number of active local sockets."); 2628 2629 static void 2630 unp_gc(__unused void *arg, int pending) 2631 { 2632 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 2633 NULL }; 2634 struct unp_head **head; 2635 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 2636 struct file *f, **unref; 2637 struct unpcb *unp, *unptmp; 2638 int i, total, unp_unreachable; 2639 2640 LIST_INIT(&unp_deadhead); 2641 unp_taskcount++; 2642 UNP_LINK_RLOCK(); 2643 /* 2644 * First determine which sockets may be in cycles. 2645 */ 2646 unp_unreachable = 0; 2647 2648 for (head = heads; *head != NULL; head++) 2649 LIST_FOREACH(unp, *head, unp_link) { 2650 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 2651 ("%s: unp %p has unexpected gc flags 0x%x", 2652 __func__, unp, (unsigned int)unp->unp_gcflag)); 2653 2654 f = unp->unp_file; 2655 2656 /* 2657 * Check for an unreachable socket potentially in a 2658 * cycle. It must be in a queue as indicated by 2659 * msgcount, and this must equal the file reference 2660 * count. Note that when msgcount is 0 the file is 2661 * NULL. 2662 */ 2663 if (f != NULL && unp->unp_msgcount != 0 && 2664 f->f_count == unp->unp_msgcount) { 2665 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 2666 unp->unp_gcflag |= UNPGC_DEAD; 2667 unp->unp_gcrefs = unp->unp_msgcount; 2668 unp_unreachable++; 2669 } 2670 } 2671 2672 /* 2673 * Scan all sockets previously marked as potentially being in a cycle 2674 * and remove the references each socket holds on any UNPGC_DEAD 2675 * sockets in its queue. After this step, all remaining references on 2676 * sockets marked UNPGC_DEAD should not be part of any cycle. 2677 */ 2678 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 2679 unp_gc_scan(unp, unp_remove_dead_ref); 2680 2681 /* 2682 * If a socket still has a non-negative refcount, it cannot be in a 2683 * cycle. In this case increment refcount of all children iteratively. 2684 * Stop the scan once we do a complete loop without discovering 2685 * a new reachable socket. 2686 */ 2687 do { 2688 unp_marked = 0; 2689 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 2690 if (unp->unp_gcrefs > 0) { 2691 unp->unp_gcflag &= ~UNPGC_DEAD; 2692 LIST_REMOVE(unp, unp_dead); 2693 KASSERT(unp_unreachable > 0, 2694 ("%s: unp_unreachable underflow.", 2695 __func__)); 2696 unp_unreachable--; 2697 unp_gc_scan(unp, unp_restore_undead_ref); 2698 } 2699 } while (unp_marked); 2700 2701 UNP_LINK_RUNLOCK(); 2702 2703 if (unp_unreachable == 0) 2704 return; 2705 2706 /* 2707 * Allocate space for a local array of dead unpcbs. 2708 * TODO: can this path be simplified by instead using the local 2709 * dead list at unp_deadhead, after taking out references 2710 * on the file object and/or unpcb and dropping the link lock? 2711 */ 2712 unref = malloc(unp_unreachable * sizeof(struct file *), 2713 M_TEMP, M_WAITOK); 2714 2715 /* 2716 * Iterate looking for sockets which have been specifically marked 2717 * as unreachable and store them locally. 2718 */ 2719 UNP_LINK_RLOCK(); 2720 total = 0; 2721 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 2722 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 2723 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 2724 unp->unp_gcflag &= ~UNPGC_DEAD; 2725 f = unp->unp_file; 2726 if (unp->unp_msgcount == 0 || f == NULL || 2727 f->f_count != unp->unp_msgcount || 2728 !fhold(f)) 2729 continue; 2730 unref[total++] = f; 2731 KASSERT(total <= unp_unreachable, 2732 ("%s: incorrect unreachable count.", __func__)); 2733 } 2734 UNP_LINK_RUNLOCK(); 2735 2736 /* 2737 * Now flush all sockets, free'ing rights. This will free the 2738 * struct files associated with these sockets but leave each socket 2739 * with one remaining ref. 2740 */ 2741 for (i = 0; i < total; i++) { 2742 struct socket *so; 2743 2744 so = unref[i]->f_data; 2745 CURVNET_SET(so->so_vnet); 2746 sorflush(so); 2747 CURVNET_RESTORE(); 2748 } 2749 2750 /* 2751 * And finally release the sockets so they can be reclaimed. 2752 */ 2753 for (i = 0; i < total; i++) 2754 fdrop(unref[i], NULL); 2755 unp_recycled += total; 2756 free(unref, M_TEMP); 2757 } 2758 2759 static void 2760 unp_dispose_mbuf(struct mbuf *m) 2761 { 2762 2763 if (m) 2764 unp_scan(m, unp_freerights); 2765 } 2766 2767 /* 2768 * Synchronize against unp_gc, which can trip over data as we are freeing it. 2769 */ 2770 static void 2771 unp_dispose(struct socket *so) 2772 { 2773 struct unpcb *unp; 2774 2775 unp = sotounpcb(so); 2776 UNP_LINK_WLOCK(); 2777 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 2778 UNP_LINK_WUNLOCK(); 2779 if (!SOLISTENING(so)) 2780 unp_dispose_mbuf(so->so_rcv.sb_mb); 2781 } 2782 2783 static void 2784 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 2785 { 2786 struct mbuf *m; 2787 struct cmsghdr *cm; 2788 void *data; 2789 socklen_t clen, datalen; 2790 2791 while (m0 != NULL) { 2792 for (m = m0; m; m = m->m_next) { 2793 if (m->m_type != MT_CONTROL) 2794 continue; 2795 2796 cm = mtod(m, struct cmsghdr *); 2797 clen = m->m_len; 2798 2799 while (cm != NULL) { 2800 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2801 break; 2802 2803 data = CMSG_DATA(cm); 2804 datalen = (caddr_t)cm + cm->cmsg_len 2805 - (caddr_t)data; 2806 2807 if (cm->cmsg_level == SOL_SOCKET && 2808 cm->cmsg_type == SCM_RIGHTS) { 2809 (*op)(data, datalen / 2810 sizeof(struct filedescent *)); 2811 } 2812 2813 if (CMSG_SPACE(datalen) < clen) { 2814 clen -= CMSG_SPACE(datalen); 2815 cm = (struct cmsghdr *) 2816 ((caddr_t)cm + CMSG_SPACE(datalen)); 2817 } else { 2818 clen = 0; 2819 cm = NULL; 2820 } 2821 } 2822 } 2823 m0 = m0->m_nextpkt; 2824 } 2825 } 2826 2827 /* 2828 * A helper function called by VFS before socket-type vnode reclamation. 2829 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 2830 * use count. 2831 */ 2832 void 2833 vfs_unp_reclaim(struct vnode *vp) 2834 { 2835 struct unpcb *unp; 2836 int active; 2837 struct mtx *vplock; 2838 2839 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 2840 KASSERT(vp->v_type == VSOCK, 2841 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 2842 2843 active = 0; 2844 vplock = mtx_pool_find(mtxpool_sleep, vp); 2845 mtx_lock(vplock); 2846 VOP_UNP_CONNECT(vp, &unp); 2847 if (unp == NULL) 2848 goto done; 2849 UNP_PCB_LOCK(unp); 2850 if (unp->unp_vnode == vp) { 2851 VOP_UNP_DETACH(vp); 2852 unp->unp_vnode = NULL; 2853 active = 1; 2854 } 2855 UNP_PCB_UNLOCK(unp); 2856 done: 2857 mtx_unlock(vplock); 2858 if (active) 2859 vunref(vp); 2860 } 2861 2862 #ifdef DDB 2863 static void 2864 db_print_indent(int indent) 2865 { 2866 int i; 2867 2868 for (i = 0; i < indent; i++) 2869 db_printf(" "); 2870 } 2871 2872 static void 2873 db_print_unpflags(int unp_flags) 2874 { 2875 int comma; 2876 2877 comma = 0; 2878 if (unp_flags & UNP_HAVEPC) { 2879 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2880 comma = 1; 2881 } 2882 if (unp_flags & UNP_WANTCRED) { 2883 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2884 comma = 1; 2885 } 2886 if (unp_flags & UNP_CONNWAIT) { 2887 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2888 comma = 1; 2889 } 2890 if (unp_flags & UNP_CONNECTING) { 2891 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2892 comma = 1; 2893 } 2894 if (unp_flags & UNP_BINDING) { 2895 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2896 comma = 1; 2897 } 2898 } 2899 2900 static void 2901 db_print_xucred(int indent, struct xucred *xu) 2902 { 2903 int comma, i; 2904 2905 db_print_indent(indent); 2906 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 2907 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 2908 db_print_indent(indent); 2909 db_printf("cr_groups: "); 2910 comma = 0; 2911 for (i = 0; i < xu->cr_ngroups; i++) { 2912 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2913 comma = 1; 2914 } 2915 db_printf("\n"); 2916 } 2917 2918 static void 2919 db_print_unprefs(int indent, struct unp_head *uh) 2920 { 2921 struct unpcb *unp; 2922 int counter; 2923 2924 counter = 0; 2925 LIST_FOREACH(unp, uh, unp_reflink) { 2926 if (counter % 4 == 0) 2927 db_print_indent(indent); 2928 db_printf("%p ", unp); 2929 if (counter % 4 == 3) 2930 db_printf("\n"); 2931 counter++; 2932 } 2933 if (counter != 0 && counter % 4 != 0) 2934 db_printf("\n"); 2935 } 2936 2937 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2938 { 2939 struct unpcb *unp; 2940 2941 if (!have_addr) { 2942 db_printf("usage: show unpcb <addr>\n"); 2943 return; 2944 } 2945 unp = (struct unpcb *)addr; 2946 2947 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2948 unp->unp_vnode); 2949 2950 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 2951 unp->unp_conn); 2952 2953 db_printf("unp_refs:\n"); 2954 db_print_unprefs(2, &unp->unp_refs); 2955 2956 /* XXXRW: Would be nice to print the full address, if any. */ 2957 db_printf("unp_addr: %p\n", unp->unp_addr); 2958 2959 db_printf("unp_gencnt: %llu\n", 2960 (unsigned long long)unp->unp_gencnt); 2961 2962 db_printf("unp_flags: %x (", unp->unp_flags); 2963 db_print_unpflags(unp->unp_flags); 2964 db_printf(")\n"); 2965 2966 db_printf("unp_peercred:\n"); 2967 db_print_xucred(2, &unp->unp_peercred); 2968 2969 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2970 } 2971 #endif 2972