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/eventhandler.h> 69 #include <sys/fcntl.h> 70 #include <sys/file.h> 71 #include <sys/filedesc.h> 72 #include <sys/kernel.h> 73 #include <sys/lock.h> 74 #include <sys/malloc.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 * See unpcb.h for the locking key. 110 */ 111 112 static uma_zone_t unp_zone; 113 static unp_gen_t unp_gencnt; /* (l) */ 114 static u_int unp_count; /* (l) Count of local sockets. */ 115 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 116 static int unp_rights; /* (g) File descriptors in flight. */ 117 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 118 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 119 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 120 121 struct unp_defer { 122 SLIST_ENTRY(unp_defer) ud_link; 123 struct file *ud_fp; 124 }; 125 static SLIST_HEAD(, unp_defer) unp_defers; 126 static int unp_defers_count; 127 128 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 129 130 /* 131 * Garbage collection of cyclic file descriptor/socket references occurs 132 * asynchronously in a taskqueue context in order to avoid recursion and 133 * reentrance in the UNIX domain socket, file descriptor, and socket layer 134 * code. See unp_gc() for a full description. 135 */ 136 static struct timeout_task unp_gc_task; 137 138 /* 139 * The close of unix domain sockets attached as SCM_RIGHTS is 140 * postponed to the taskqueue, to avoid arbitrary recursion depth. 141 * The attached sockets might have another sockets attached. 142 */ 143 static struct task unp_defer_task; 144 145 /* 146 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 147 * stream sockets, although the total for sender and receiver is actually 148 * only PIPSIZ. 149 * 150 * Datagram sockets really use the sendspace as the maximum datagram size, 151 * and don't really want to reserve the sendspace. Their recvspace should be 152 * large enough for at least one max-size datagram plus address. 153 */ 154 #ifndef PIPSIZ 155 #define PIPSIZ 8192 156 #endif 157 static u_long unpst_sendspace = PIPSIZ; 158 static u_long unpst_recvspace = PIPSIZ; 159 static u_long unpdg_maxdgram = 2*1024; 160 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */ 161 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 162 static u_long unpsp_recvspace = PIPSIZ; 163 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 165 "Local domain"); 166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 167 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 168 "SOCK_STREAM"); 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 171 "SOCK_DGRAM"); 172 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 173 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 174 "SOCK_SEQPACKET"); 175 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 177 &unpst_sendspace, 0, "Default stream send space."); 178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 179 &unpst_recvspace, 0, "Default stream receive space."); 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 181 &unpdg_maxdgram, 0, "Maximum datagram size."); 182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 183 &unpdg_recvspace, 0, "Default datagram receive space."); 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 185 &unpsp_sendspace, 0, "Default seqpacket send space."); 186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 187 &unpsp_recvspace, 0, "Default seqpacket receive space."); 188 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 189 "File descriptors in flight."); 190 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 191 &unp_defers_count, 0, 192 "File descriptors deferred to taskqueue for close."); 193 194 /* 195 * Locking and synchronization: 196 * 197 * Several types of locks exist in the local domain socket implementation: 198 * - a global linkage lock 199 * - a global connection list lock 200 * - the mtxpool lock 201 * - per-unpcb mutexes 202 * 203 * The linkage lock protects the global socket lists, the generation number 204 * counter and garbage collector state. 205 * 206 * The connection list lock protects the list of referring sockets in a datagram 207 * socket PCB. This lock is also overloaded to protect a global list of 208 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 209 * messages. To avoid recursion, such references are released by a dedicated 210 * thread. 211 * 212 * The mtxpool lock protects the vnode from being modified while referenced. 213 * Lock ordering rules require that it be acquired before any PCB locks. 214 * 215 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 216 * unpcb. This includes the unp_conn field, which either links two connected 217 * PCBs together (for connected socket types) or points at the destination 218 * socket (for connectionless socket types). The operations of creating or 219 * destroying a connection therefore involve locking multiple PCBs. To avoid 220 * lock order reversals, in some cases this involves dropping a PCB lock and 221 * using a reference counter to maintain liveness. 222 * 223 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 224 * allocated in pru_attach() and freed in pru_detach(). The validity of that 225 * pointer is an invariant, so no lock is required to dereference the so_pcb 226 * pointer if a valid socket reference is held by the caller. In practice, 227 * this is always true during operations performed on a socket. Each unpcb 228 * has a back-pointer to its socket, unp_socket, which will be stable under 229 * the same circumstances. 230 * 231 * This pointer may only be safely dereferenced as long as a valid reference 232 * to the unpcb is held. Typically, this reference will be from the socket, 233 * or from another unpcb when the referring unpcb's lock is held (in order 234 * that the reference not be invalidated during use). For example, to follow 235 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 236 * that detach is not run clearing unp_socket. 237 * 238 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 239 * protocols, bind() is a non-atomic operation, and connect() requires 240 * potential sleeping in the protocol, due to potentially waiting on local or 241 * distributed file systems. We try to separate "lookup" operations, which 242 * may sleep, and the IPC operations themselves, which typically can occur 243 * with relative atomicity as locks can be held over the entire operation. 244 * 245 * Another tricky issue is simultaneous multi-threaded or multi-process 246 * access to a single UNIX domain socket. These are handled by the flags 247 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 248 * binding, both of which involve dropping UNIX domain socket locks in order 249 * to perform namei() and other file system operations. 250 */ 251 static struct rwlock unp_link_rwlock; 252 static struct mtx unp_defers_lock; 253 254 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 255 "unp_link_rwlock") 256 257 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 258 RA_LOCKED) 259 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 260 RA_UNLOCKED) 261 262 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 263 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 264 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 265 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 266 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 267 RA_WLOCKED) 268 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 269 270 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 271 "unp_defer", NULL, MTX_DEF) 272 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 273 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 274 275 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 276 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 277 278 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 279 "unp", "unp", \ 280 MTX_DUPOK|MTX_DEF) 281 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 282 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 283 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 284 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 285 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 286 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 287 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 288 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 289 290 static int uipc_connect2(struct socket *, struct socket *); 291 static int uipc_ctloutput(struct socket *, struct sockopt *); 292 static int unp_connect(struct socket *, struct sockaddr *, 293 struct thread *); 294 static int unp_connectat(int, struct socket *, struct sockaddr *, 295 struct thread *, bool); 296 static void unp_connect2(struct socket *so, struct socket *so2, int); 297 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 298 static void unp_dispose(struct socket *so); 299 static void unp_shutdown(struct unpcb *); 300 static void unp_drop(struct unpcb *); 301 static void unp_gc(__unused void *, int); 302 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 303 static void unp_discard(struct file *); 304 static void unp_freerights(struct filedescent **, int); 305 static int unp_internalize(struct mbuf **, struct thread *, 306 struct mbuf **, u_int *, u_int *); 307 static void unp_internalize_fp(struct file *); 308 static int unp_externalize(struct mbuf *, struct mbuf **, int); 309 static int unp_externalize_fp(struct file *); 310 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 311 int, struct mbuf **, u_int *, u_int *); 312 static void unp_process_defers(void * __unused, int); 313 314 static void 315 unp_pcb_hold(struct unpcb *unp) 316 { 317 u_int old __unused; 318 319 old = refcount_acquire(&unp->unp_refcount); 320 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 321 } 322 323 static __result_use_check bool 324 unp_pcb_rele(struct unpcb *unp) 325 { 326 bool ret; 327 328 UNP_PCB_LOCK_ASSERT(unp); 329 330 if ((ret = refcount_release(&unp->unp_refcount))) { 331 UNP_PCB_UNLOCK(unp); 332 UNP_PCB_LOCK_DESTROY(unp); 333 uma_zfree(unp_zone, unp); 334 } 335 return (ret); 336 } 337 338 static void 339 unp_pcb_rele_notlast(struct unpcb *unp) 340 { 341 bool ret __unused; 342 343 ret = refcount_release(&unp->unp_refcount); 344 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 345 } 346 347 static void 348 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 349 { 350 UNP_PCB_UNLOCK_ASSERT(unp); 351 UNP_PCB_UNLOCK_ASSERT(unp2); 352 353 if (unp == unp2) { 354 UNP_PCB_LOCK(unp); 355 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 356 UNP_PCB_LOCK(unp); 357 UNP_PCB_LOCK(unp2); 358 } else { 359 UNP_PCB_LOCK(unp2); 360 UNP_PCB_LOCK(unp); 361 } 362 } 363 364 static void 365 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 366 { 367 UNP_PCB_UNLOCK(unp); 368 if (unp != unp2) 369 UNP_PCB_UNLOCK(unp2); 370 } 371 372 /* 373 * Try to lock the connected peer of an already locked socket. In some cases 374 * this requires that we unlock the current socket. The pairbusy counter is 375 * used to block concurrent connection attempts while the lock is dropped. The 376 * caller must be careful to revalidate PCB state. 377 */ 378 static struct unpcb * 379 unp_pcb_lock_peer(struct unpcb *unp) 380 { 381 struct unpcb *unp2; 382 383 UNP_PCB_LOCK_ASSERT(unp); 384 unp2 = unp->unp_conn; 385 if (unp2 == NULL) 386 return (NULL); 387 if (__predict_false(unp == unp2)) 388 return (unp); 389 390 UNP_PCB_UNLOCK_ASSERT(unp2); 391 392 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 393 return (unp2); 394 if ((uintptr_t)unp2 > (uintptr_t)unp) { 395 UNP_PCB_LOCK(unp2); 396 return (unp2); 397 } 398 unp->unp_pairbusy++; 399 unp_pcb_hold(unp2); 400 UNP_PCB_UNLOCK(unp); 401 402 UNP_PCB_LOCK(unp2); 403 UNP_PCB_LOCK(unp); 404 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 405 ("%s: socket %p was reconnected", __func__, unp)); 406 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 407 unp->unp_flags &= ~UNP_WAITING; 408 wakeup(unp); 409 } 410 if (unp_pcb_rele(unp2)) { 411 /* unp2 is unlocked. */ 412 return (NULL); 413 } 414 if (unp->unp_conn == NULL) { 415 UNP_PCB_UNLOCK(unp2); 416 return (NULL); 417 } 418 return (unp2); 419 } 420 421 /* 422 * Definitions of protocols supported in the LOCAL domain. 423 */ 424 static struct domain localdomain; 425 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; 426 static struct pr_usrreqs uipc_usrreqs_seqpacket; 427 static struct protosw localsw[] = { 428 { 429 .pr_type = SOCK_STREAM, 430 .pr_domain = &localdomain, 431 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 432 PR_CAPATTACH, 433 .pr_ctloutput = &uipc_ctloutput, 434 .pr_usrreqs = &uipc_usrreqs_stream 435 }, 436 { 437 .pr_type = SOCK_DGRAM, 438 .pr_domain = &localdomain, 439 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 440 PR_SOCKBUF, 441 .pr_ctloutput = &uipc_ctloutput, 442 .pr_usrreqs = &uipc_usrreqs_dgram 443 }, 444 { 445 .pr_type = SOCK_SEQPACKET, 446 .pr_domain = &localdomain, 447 448 /* 449 * XXXRW: For now, PR_ADDR because soreceive will bump into them 450 * due to our use of sbappendaddr. A new sbappend variants is needed 451 * that supports both atomic record writes and control data. 452 */ 453 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 454 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 455 .pr_ctloutput = &uipc_ctloutput, 456 .pr_usrreqs = &uipc_usrreqs_seqpacket, 457 }, 458 }; 459 460 static struct domain localdomain = { 461 .dom_family = AF_LOCAL, 462 .dom_name = "local", 463 .dom_externalize = unp_externalize, 464 .dom_dispose = unp_dispose, 465 .dom_protosw = localsw, 466 .dom_protoswNPROTOSW = &localsw[nitems(localsw)] 467 }; 468 DOMAIN_SET(local); 469 470 static void 471 uipc_abort(struct socket *so) 472 { 473 struct unpcb *unp, *unp2; 474 475 unp = sotounpcb(so); 476 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 477 UNP_PCB_UNLOCK_ASSERT(unp); 478 479 UNP_PCB_LOCK(unp); 480 unp2 = unp->unp_conn; 481 if (unp2 != NULL) { 482 unp_pcb_hold(unp2); 483 UNP_PCB_UNLOCK(unp); 484 unp_drop(unp2); 485 } else 486 UNP_PCB_UNLOCK(unp); 487 } 488 489 static int 490 uipc_accept(struct socket *so, struct sockaddr **nam) 491 { 492 struct unpcb *unp, *unp2; 493 const struct sockaddr *sa; 494 495 /* 496 * Pass back name of connected socket, if it was bound and we are 497 * still connected (our peer may have closed already!). 498 */ 499 unp = sotounpcb(so); 500 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 501 502 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 503 UNP_PCB_LOCK(unp); 504 unp2 = unp_pcb_lock_peer(unp); 505 if (unp2 != NULL && unp2->unp_addr != NULL) 506 sa = (struct sockaddr *)unp2->unp_addr; 507 else 508 sa = &sun_noname; 509 bcopy(sa, *nam, sa->sa_len); 510 if (unp2 != NULL) 511 unp_pcb_unlock_pair(unp, unp2); 512 else 513 UNP_PCB_UNLOCK(unp); 514 return (0); 515 } 516 517 static int 518 uipc_attach(struct socket *so, int proto, struct thread *td) 519 { 520 u_long sendspace, recvspace; 521 struct unpcb *unp; 522 int error; 523 bool locked; 524 525 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 526 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 527 switch (so->so_type) { 528 case SOCK_STREAM: 529 sendspace = unpst_sendspace; 530 recvspace = unpst_recvspace; 531 break; 532 533 case SOCK_DGRAM: 534 STAILQ_INIT(&so->so_rcv.uxdg_mb); 535 STAILQ_INIT(&so->so_snd.uxdg_mb); 536 TAILQ_INIT(&so->so_rcv.uxdg_conns); 537 /* 538 * Since send buffer is either bypassed or is a part 539 * of one-to-many receive buffer, we assign both space 540 * limits to unpdg_recvspace. 541 */ 542 sendspace = recvspace = unpdg_recvspace; 543 break; 544 545 case SOCK_SEQPACKET: 546 sendspace = unpsp_sendspace; 547 recvspace = unpsp_recvspace; 548 break; 549 550 default: 551 panic("uipc_attach"); 552 } 553 error = soreserve(so, sendspace, recvspace); 554 if (error) 555 return (error); 556 } 557 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 558 if (unp == NULL) 559 return (ENOBUFS); 560 LIST_INIT(&unp->unp_refs); 561 UNP_PCB_LOCK_INIT(unp); 562 unp->unp_socket = so; 563 so->so_pcb = unp; 564 refcount_init(&unp->unp_refcount, 1); 565 566 if ((locked = UNP_LINK_WOWNED()) == false) 567 UNP_LINK_WLOCK(); 568 569 unp->unp_gencnt = ++unp_gencnt; 570 unp->unp_ino = ++unp_ino; 571 unp_count++; 572 switch (so->so_type) { 573 case SOCK_STREAM: 574 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 575 break; 576 577 case SOCK_DGRAM: 578 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 579 break; 580 581 case SOCK_SEQPACKET: 582 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 583 break; 584 585 default: 586 panic("uipc_attach"); 587 } 588 589 if (locked == false) 590 UNP_LINK_WUNLOCK(); 591 592 return (0); 593 } 594 595 static int 596 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 597 { 598 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 599 struct vattr vattr; 600 int error, namelen; 601 struct nameidata nd; 602 struct unpcb *unp; 603 struct vnode *vp; 604 struct mount *mp; 605 cap_rights_t rights; 606 char *buf; 607 608 if (nam->sa_family != AF_UNIX) 609 return (EAFNOSUPPORT); 610 611 unp = sotounpcb(so); 612 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 613 614 if (soun->sun_len > sizeof(struct sockaddr_un)) 615 return (EINVAL); 616 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 617 if (namelen <= 0) 618 return (EINVAL); 619 620 /* 621 * We don't allow simultaneous bind() calls on a single UNIX domain 622 * socket, so flag in-progress operations, and return an error if an 623 * operation is already in progress. 624 * 625 * Historically, we have not allowed a socket to be rebound, so this 626 * also returns an error. Not allowing re-binding simplifies the 627 * implementation and avoids a great many possible failure modes. 628 */ 629 UNP_PCB_LOCK(unp); 630 if (unp->unp_vnode != NULL) { 631 UNP_PCB_UNLOCK(unp); 632 return (EINVAL); 633 } 634 if (unp->unp_flags & UNP_BINDING) { 635 UNP_PCB_UNLOCK(unp); 636 return (EALREADY); 637 } 638 unp->unp_flags |= UNP_BINDING; 639 UNP_PCB_UNLOCK(unp); 640 641 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 642 bcopy(soun->sun_path, buf, namelen); 643 buf[namelen] = 0; 644 645 restart: 646 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, 647 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 648 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 649 error = namei(&nd); 650 if (error) 651 goto error; 652 vp = nd.ni_vp; 653 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 654 NDFREE_PNBUF(&nd); 655 if (nd.ni_dvp == vp) 656 vrele(nd.ni_dvp); 657 else 658 vput(nd.ni_dvp); 659 if (vp != NULL) { 660 vrele(vp); 661 error = EADDRINUSE; 662 goto error; 663 } 664 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 665 if (error) 666 goto error; 667 goto restart; 668 } 669 VATTR_NULL(&vattr); 670 vattr.va_type = VSOCK; 671 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 672 #ifdef MAC 673 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 674 &vattr); 675 #endif 676 if (error == 0) 677 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 678 NDFREE_PNBUF(&nd); 679 if (error) { 680 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 681 vn_finished_write(mp); 682 if (error == ERELOOKUP) 683 goto restart; 684 goto error; 685 } 686 vp = nd.ni_vp; 687 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 688 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 689 690 UNP_PCB_LOCK(unp); 691 VOP_UNP_BIND(vp, unp); 692 unp->unp_vnode = vp; 693 unp->unp_addr = soun; 694 unp->unp_flags &= ~UNP_BINDING; 695 UNP_PCB_UNLOCK(unp); 696 vref(vp); 697 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 698 vn_finished_write(mp); 699 free(buf, M_TEMP); 700 return (0); 701 702 error: 703 UNP_PCB_LOCK(unp); 704 unp->unp_flags &= ~UNP_BINDING; 705 UNP_PCB_UNLOCK(unp); 706 free(buf, M_TEMP); 707 return (error); 708 } 709 710 static int 711 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 712 { 713 714 return (uipc_bindat(AT_FDCWD, so, nam, td)); 715 } 716 717 static int 718 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 719 { 720 int error; 721 722 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 723 error = unp_connect(so, nam, td); 724 return (error); 725 } 726 727 static int 728 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 729 struct thread *td) 730 { 731 int error; 732 733 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 734 error = unp_connectat(fd, so, nam, td, false); 735 return (error); 736 } 737 738 static void 739 uipc_close(struct socket *so) 740 { 741 struct unpcb *unp, *unp2; 742 struct vnode *vp = NULL; 743 struct mtx *vplock; 744 745 unp = sotounpcb(so); 746 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 747 748 vplock = NULL; 749 if ((vp = unp->unp_vnode) != NULL) { 750 vplock = mtx_pool_find(mtxpool_sleep, vp); 751 mtx_lock(vplock); 752 } 753 UNP_PCB_LOCK(unp); 754 if (vp && unp->unp_vnode == NULL) { 755 mtx_unlock(vplock); 756 vp = NULL; 757 } 758 if (vp != NULL) { 759 VOP_UNP_DETACH(vp); 760 unp->unp_vnode = NULL; 761 } 762 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 763 unp_disconnect(unp, unp2); 764 else 765 UNP_PCB_UNLOCK(unp); 766 if (vp) { 767 mtx_unlock(vplock); 768 vrele(vp); 769 } 770 } 771 772 static int 773 uipc_connect2(struct socket *so1, struct socket *so2) 774 { 775 struct unpcb *unp, *unp2; 776 777 if (so1->so_type != so2->so_type) 778 return (EPROTOTYPE); 779 780 unp = so1->so_pcb; 781 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 782 unp2 = so2->so_pcb; 783 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 784 unp_pcb_lock_pair(unp, unp2); 785 unp_connect2(so1, so2, PRU_CONNECT2); 786 unp_pcb_unlock_pair(unp, unp2); 787 788 return (0); 789 } 790 791 static void 792 uipc_detach(struct socket *so) 793 { 794 struct unpcb *unp, *unp2; 795 struct mtx *vplock; 796 struct vnode *vp; 797 int local_unp_rights; 798 799 unp = sotounpcb(so); 800 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 801 802 vp = NULL; 803 vplock = NULL; 804 805 UNP_LINK_WLOCK(); 806 LIST_REMOVE(unp, unp_link); 807 if (unp->unp_gcflag & UNPGC_DEAD) 808 LIST_REMOVE(unp, unp_dead); 809 unp->unp_gencnt = ++unp_gencnt; 810 --unp_count; 811 UNP_LINK_WUNLOCK(); 812 813 UNP_PCB_UNLOCK_ASSERT(unp); 814 restart: 815 if ((vp = unp->unp_vnode) != NULL) { 816 vplock = mtx_pool_find(mtxpool_sleep, vp); 817 mtx_lock(vplock); 818 } 819 UNP_PCB_LOCK(unp); 820 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 821 if (vplock) 822 mtx_unlock(vplock); 823 UNP_PCB_UNLOCK(unp); 824 goto restart; 825 } 826 if ((vp = unp->unp_vnode) != NULL) { 827 VOP_UNP_DETACH(vp); 828 unp->unp_vnode = NULL; 829 } 830 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 831 unp_disconnect(unp, unp2); 832 else 833 UNP_PCB_UNLOCK(unp); 834 835 UNP_REF_LIST_LOCK(); 836 while (!LIST_EMPTY(&unp->unp_refs)) { 837 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 838 839 unp_pcb_hold(ref); 840 UNP_REF_LIST_UNLOCK(); 841 842 MPASS(ref != unp); 843 UNP_PCB_UNLOCK_ASSERT(ref); 844 unp_drop(ref); 845 UNP_REF_LIST_LOCK(); 846 } 847 UNP_REF_LIST_UNLOCK(); 848 849 UNP_PCB_LOCK(unp); 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 switch (so->so_type) { 865 case SOCK_DGRAM: 866 /* 867 * Everything should have been unlinked/freed by unp_dispose() 868 * and/or unp_disconnect(). 869 */ 870 MPASS(so->so_rcv.uxdg_peeked == NULL); 871 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 872 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 873 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 874 } 875 } 876 877 static int 878 uipc_disconnect(struct socket *so) 879 { 880 struct unpcb *unp, *unp2; 881 882 unp = sotounpcb(so); 883 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 884 885 UNP_PCB_LOCK(unp); 886 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 887 unp_disconnect(unp, unp2); 888 else 889 UNP_PCB_UNLOCK(unp); 890 return (0); 891 } 892 893 static int 894 uipc_listen(struct socket *so, int backlog, struct thread *td) 895 { 896 struct unpcb *unp; 897 int error; 898 899 MPASS(so->so_type != SOCK_DGRAM); 900 901 /* 902 * Synchronize with concurrent connection attempts. 903 */ 904 error = 0; 905 unp = sotounpcb(so); 906 UNP_PCB_LOCK(unp); 907 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 908 error = EINVAL; 909 else if (unp->unp_vnode == NULL) 910 error = EDESTADDRREQ; 911 if (error != 0) { 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 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 1012 struct mbuf *control, struct thread *td) 1013 { 1014 struct unpcb *unp, *unp2; 1015 struct socket *so2; 1016 u_int mbcnt, sbcc; 1017 int error; 1018 1019 unp = sotounpcb(so); 1020 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 1021 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 1022 ("%s: socktype %d", __func__, so->so_type)); 1023 1024 error = 0; 1025 if (flags & PRUS_OOB) { 1026 error = EOPNOTSUPP; 1027 goto release; 1028 } 1029 if (control != NULL && 1030 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 1031 goto release; 1032 1033 unp2 = NULL; 1034 if ((so->so_state & SS_ISCONNECTED) == 0) { 1035 if (nam != NULL) { 1036 if ((error = unp_connect(so, nam, td)) != 0) 1037 goto out; 1038 } else { 1039 error = ENOTCONN; 1040 goto out; 1041 } 1042 } 1043 1044 UNP_PCB_LOCK(unp); 1045 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 1046 UNP_PCB_UNLOCK(unp); 1047 error = ENOTCONN; 1048 goto out; 1049 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1050 unp_pcb_unlock_pair(unp, unp2); 1051 error = EPIPE; 1052 goto out; 1053 } 1054 UNP_PCB_UNLOCK(unp); 1055 if ((so2 = unp2->unp_socket) == NULL) { 1056 UNP_PCB_UNLOCK(unp2); 1057 error = ENOTCONN; 1058 goto out; 1059 } 1060 SOCKBUF_LOCK(&so2->so_rcv); 1061 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 1062 /* 1063 * Credentials are passed only once on SOCK_STREAM and 1064 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 1065 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 1066 */ 1067 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 1068 NULL, NULL); 1069 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 1070 } 1071 1072 /* 1073 * Send to paired receive port and wake up readers. Don't 1074 * check for space available in the receive buffer if we're 1075 * attaching ancillary data; Unix domain sockets only check 1076 * for space in the sending sockbuf, and that check is 1077 * performed one level up the stack. At that level we cannot 1078 * precisely account for the amount of buffer space used 1079 * (e.g., because control messages are not yet internalized). 1080 */ 1081 switch (so->so_type) { 1082 case SOCK_STREAM: 1083 if (control != NULL) { 1084 sbappendcontrol_locked(&so2->so_rcv, m, 1085 control, flags); 1086 control = NULL; 1087 } else 1088 sbappend_locked(&so2->so_rcv, m, flags); 1089 break; 1090 1091 case SOCK_SEQPACKET: 1092 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1093 &sun_noname, m, control)) 1094 control = NULL; 1095 break; 1096 } 1097 1098 mbcnt = so2->so_rcv.sb_mbcnt; 1099 sbcc = sbavail(&so2->so_rcv); 1100 if (sbcc) 1101 sorwakeup_locked(so2); 1102 else 1103 SOCKBUF_UNLOCK(&so2->so_rcv); 1104 1105 /* 1106 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1107 * it would be possible for uipc_rcvd to be called at this 1108 * point, drain the receiving sockbuf, clear SB_STOP, and then 1109 * we would set SB_STOP below. That could lead to an empty 1110 * sockbuf having SB_STOP set 1111 */ 1112 SOCKBUF_LOCK(&so->so_snd); 1113 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1114 so->so_snd.sb_flags |= SB_STOP; 1115 SOCKBUF_UNLOCK(&so->so_snd); 1116 UNP_PCB_UNLOCK(unp2); 1117 m = NULL; 1118 out: 1119 /* 1120 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. 1121 */ 1122 if (flags & PRUS_EOF) { 1123 UNP_PCB_LOCK(unp); 1124 socantsendmore(so); 1125 unp_shutdown(unp); 1126 UNP_PCB_UNLOCK(unp); 1127 } 1128 if (control != NULL && error != 0) 1129 unp_scan(control, unp_freerights); 1130 1131 release: 1132 if (control != NULL) 1133 m_freem(control); 1134 /* 1135 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1136 * for freeing memory. 1137 */ 1138 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1139 m_freem(m); 1140 return (error); 1141 } 1142 1143 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1144 static inline bool 1145 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1146 { 1147 u_int bleft, mleft; 1148 1149 MPASS(sb->sb_hiwat >= sb->uxdg_cc); 1150 MPASS(sb->sb_mbmax >= sb->uxdg_mbcnt); 1151 1152 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1153 return (false); 1154 1155 bleft = sb->sb_hiwat - sb->uxdg_cc; 1156 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1157 1158 return (bleft >= cc && mleft >= mbcnt); 1159 } 1160 1161 /* 1162 * PF_UNIX/SOCK_DGRAM send 1163 * 1164 * Allocate a record consisting of 3 mbufs in the sequence of 1165 * from -> control -> data and append it to the socket buffer. 1166 * 1167 * The first mbuf carries sender's name and is a pkthdr that stores 1168 * overall length of datagram, its memory consumption and control length. 1169 */ 1170 #define ctllen PH_loc.thirtytwo[1] 1171 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1172 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1173 static int 1174 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1175 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1176 { 1177 struct unpcb *unp, *unp2; 1178 const struct sockaddr *from; 1179 struct socket *so2; 1180 struct sockbuf *sb; 1181 struct mbuf *f, *clast; 1182 u_int cc, ctl, mbcnt; 1183 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1184 int error; 1185 1186 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1187 1188 error = 0; 1189 f = NULL; 1190 ctl = 0; 1191 1192 if (__predict_false(flags & MSG_OOB)) { 1193 error = EOPNOTSUPP; 1194 goto out; 1195 } 1196 if (m == NULL) { 1197 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1198 error = EMSGSIZE; 1199 goto out; 1200 } 1201 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1202 if (__predict_false(m == NULL)) { 1203 error = EFAULT; 1204 goto out; 1205 } 1206 f = m_gethdr(M_WAITOK, MT_SONAME); 1207 cc = m->m_pkthdr.len; 1208 mbcnt = MSIZE + m->m_pkthdr.memlen; 1209 if (c != NULL && 1210 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1211 goto out; 1212 } else { 1213 /* pru_sosend() with mbuf usually is a kernel thread. */ 1214 1215 M_ASSERTPKTHDR(m); 1216 if (__predict_false(c != NULL)) 1217 panic("%s: control from a kernel thread", __func__); 1218 1219 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1220 error = EMSGSIZE; 1221 goto out; 1222 } 1223 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1224 error = ENOBUFS; 1225 goto out; 1226 } 1227 /* Condition the foreign mbuf to our standards. */ 1228 m_clrprotoflags(m); 1229 m_tag_delete_chain(m, NULL); 1230 m->m_pkthdr.rcvif = NULL; 1231 m->m_pkthdr.flowid = 0; 1232 m->m_pkthdr.csum_flags = 0; 1233 m->m_pkthdr.fibnum = 0; 1234 m->m_pkthdr.rsstype = 0; 1235 1236 cc = m->m_pkthdr.len; 1237 mbcnt = MSIZE; 1238 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1239 mbcnt += MSIZE; 1240 if (mb->m_flags & M_EXT) 1241 mbcnt += mb->m_ext.ext_size; 1242 } 1243 } 1244 1245 unp = sotounpcb(so); 1246 MPASS(unp); 1247 1248 /* 1249 * XXXGL: would be cool to fully remove so_snd out of the equation 1250 * and avoid this lock, which is not only extraneous, but also being 1251 * released, thus still leaving possibility for a race. We can easily 1252 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1253 * is more difficult to invent something to handle so_error. 1254 */ 1255 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1256 if (error) 1257 goto out2; 1258 SOCK_SENDBUF_LOCK(so); 1259 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1260 SOCK_SENDBUF_UNLOCK(so); 1261 error = EPIPE; 1262 goto out3; 1263 } 1264 if (so->so_error != 0) { 1265 error = so->so_error; 1266 so->so_error = 0; 1267 SOCK_SENDBUF_UNLOCK(so); 1268 goto out3; 1269 } 1270 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1271 SOCK_SENDBUF_UNLOCK(so); 1272 error = EDESTADDRREQ; 1273 goto out3; 1274 } 1275 SOCK_SENDBUF_UNLOCK(so); 1276 1277 if (addr != NULL) { 1278 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1279 goto out3; 1280 UNP_PCB_LOCK_ASSERT(unp); 1281 unp2 = unp->unp_conn; 1282 UNP_PCB_LOCK_ASSERT(unp2); 1283 } else { 1284 UNP_PCB_LOCK(unp); 1285 unp2 = unp_pcb_lock_peer(unp); 1286 if (unp2 == NULL) { 1287 UNP_PCB_UNLOCK(unp); 1288 error = ENOTCONN; 1289 goto out3; 1290 } 1291 } 1292 1293 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1294 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1295 &mbcnt); 1296 if (unp->unp_addr != NULL) 1297 from = (struct sockaddr *)unp->unp_addr; 1298 else 1299 from = &sun_noname; 1300 f->m_len = from->sa_len; 1301 MPASS(from->sa_len <= MLEN); 1302 bcopy(from, mtod(f, void *), from->sa_len); 1303 ctl += f->m_len; 1304 1305 /* 1306 * Concatenate mbufs: from -> control -> data. 1307 * Save overall cc and mbcnt in "from" mbuf. 1308 */ 1309 if (c != NULL) { 1310 #ifdef INVARIANTS 1311 struct mbuf *mc; 1312 1313 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1314 MPASS(mc == clast); 1315 #endif 1316 f->m_next = c; 1317 clast->m_next = m; 1318 c = NULL; 1319 } else 1320 f->m_next = m; 1321 m = NULL; 1322 #ifdef INVARIANTS 1323 dcc = dctl = dmbcnt = 0; 1324 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1325 if (mb->m_type == MT_DATA) 1326 dcc += mb->m_len; 1327 else 1328 dctl += mb->m_len; 1329 dmbcnt += MSIZE; 1330 if (mb->m_flags & M_EXT) 1331 dmbcnt += mb->m_ext.ext_size; 1332 } 1333 MPASS(dcc == cc); 1334 MPASS(dctl == ctl); 1335 MPASS(dmbcnt == mbcnt); 1336 #endif 1337 f->m_pkthdr.len = cc + ctl; 1338 f->m_pkthdr.memlen = mbcnt; 1339 f->m_pkthdr.ctllen = ctl; 1340 1341 /* 1342 * Destination socket buffer selection. 1343 * 1344 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1345 * destination address is supplied, create a temporary connection for 1346 * the run time of the function (see call to unp_connectat() above and 1347 * to unp_disconnect() below). We distinguish them by condition of 1348 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1349 * that condition, since, again, through the run time of this code we 1350 * are always connected. For such "unconnected" sends, the destination 1351 * buffer would be the receive buffer of destination socket so2. 1352 * 1353 * For connected sends, data lands on the send buffer of the sender's 1354 * socket "so". Then, if we just added the very first datagram 1355 * on this send buffer, we need to add the send buffer on to the 1356 * receiving socket's buffer list. We put ourselves on top of the 1357 * list. Such logic gives infrequent senders priority over frequent 1358 * senders. 1359 * 1360 * Note on byte count management. As long as event methods kevent(2), 1361 * select(2) are not protocol specific (yet), we need to maintain 1362 * meaningful values on the receive buffer. So, the receive buffer 1363 * would accumulate counters from all connected buffers potentially 1364 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1365 */ 1366 so2 = unp2->unp_socket; 1367 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1368 SOCK_RECVBUF_LOCK(so2); 1369 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1370 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1371 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1372 uxdg_clist); 1373 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1374 sb->uxdg_cc += cc + ctl; 1375 sb->uxdg_ctl += ctl; 1376 sb->uxdg_mbcnt += mbcnt; 1377 so2->so_rcv.sb_acc += cc + ctl; 1378 so2->so_rcv.sb_ccc += cc + ctl; 1379 so2->so_rcv.sb_ctl += ctl; 1380 so2->so_rcv.sb_mbcnt += mbcnt; 1381 sorwakeup_locked(so2); 1382 f = NULL; 1383 } else { 1384 soroverflow_locked(so2); 1385 error = (so->so_state & SS_NBIO) ? EAGAIN : ENOBUFS; 1386 } 1387 1388 if (addr != NULL) 1389 unp_disconnect(unp, unp2); 1390 else 1391 unp_pcb_unlock_pair(unp, unp2); 1392 1393 td->td_ru.ru_msgsnd++; 1394 1395 out3: 1396 SOCK_IO_SEND_UNLOCK(so); 1397 out2: 1398 if (c) 1399 unp_scan(c, unp_freerights); 1400 out: 1401 if (f) 1402 m_freem(f); 1403 if (c) 1404 m_freem(c); 1405 if (m) 1406 m_freem(m); 1407 1408 return (error); 1409 } 1410 1411 /* 1412 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1413 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1414 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1415 */ 1416 static int 1417 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1418 struct uio *uio, struct mbuf **controlp, int *flagsp) 1419 { 1420 ssize_t len = 0; 1421 int error; 1422 1423 so->so_rcv.uxdg_peeked = m; 1424 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1425 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1426 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1427 SOCK_RECVBUF_UNLOCK(so); 1428 1429 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1430 if (psa != NULL) 1431 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1432 1433 m = m->m_next; 1434 KASSERT(m, ("%s: no data or control after soname", __func__)); 1435 1436 /* 1437 * With MSG_PEEK the control isn't executed, just copied. 1438 */ 1439 while (m != NULL && m->m_type == MT_CONTROL) { 1440 if (controlp != NULL) { 1441 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1442 controlp = &(*controlp)->m_next; 1443 } 1444 m = m->m_next; 1445 } 1446 KASSERT(m == NULL || m->m_type == MT_DATA, 1447 ("%s: not MT_DATA mbuf %p", __func__, m)); 1448 while (m != NULL && uio->uio_resid > 0) { 1449 len = uio->uio_resid; 1450 if (len > m->m_len) 1451 len = m->m_len; 1452 error = uiomove(mtod(m, char *), (int)len, uio); 1453 if (error) { 1454 SOCK_IO_RECV_UNLOCK(so); 1455 return (error); 1456 } 1457 if (len == m->m_len) 1458 m = m->m_next; 1459 } 1460 SOCK_IO_RECV_UNLOCK(so); 1461 1462 if (flagsp != NULL) { 1463 if (m != NULL) { 1464 if (*flagsp & MSG_TRUNC) { 1465 /* Report real length of the packet */ 1466 uio->uio_resid -= m_length(m, NULL) - len; 1467 } 1468 *flagsp |= MSG_TRUNC; 1469 } else 1470 *flagsp &= ~MSG_TRUNC; 1471 } 1472 1473 return (0); 1474 } 1475 1476 /* 1477 * PF_UNIX/SOCK_DGRAM receive 1478 */ 1479 static int 1480 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1481 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1482 { 1483 struct sockbuf *sb = NULL; 1484 struct mbuf *m; 1485 int flags, error; 1486 ssize_t len = 0; 1487 bool nonblock; 1488 1489 MPASS(mp0 == NULL); 1490 1491 if (psa != NULL) 1492 *psa = NULL; 1493 if (controlp != NULL) 1494 *controlp = NULL; 1495 1496 flags = flagsp != NULL ? *flagsp : 0; 1497 nonblock = (so->so_state & SS_NBIO) || 1498 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1499 1500 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1501 if (__predict_false(error)) 1502 return (error); 1503 1504 /* 1505 * Loop blocking while waiting for a datagram. Prioritize connected 1506 * peers over unconnected sends. Set sb to selected socket buffer 1507 * containing an mbuf on exit from the wait loop. A datagram that 1508 * had already been peeked at has top priority. 1509 */ 1510 SOCK_RECVBUF_LOCK(so); 1511 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1512 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1513 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1514 if (so->so_error) { 1515 error = so->so_error; 1516 so->so_error = 0; 1517 SOCK_RECVBUF_UNLOCK(so); 1518 SOCK_IO_RECV_UNLOCK(so); 1519 return (error); 1520 } 1521 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1522 uio->uio_resid == 0) { 1523 SOCK_RECVBUF_UNLOCK(so); 1524 SOCK_IO_RECV_UNLOCK(so); 1525 return (0); 1526 } 1527 if (nonblock) { 1528 SOCK_RECVBUF_UNLOCK(so); 1529 SOCK_IO_RECV_UNLOCK(so); 1530 return (EWOULDBLOCK); 1531 } 1532 error = sbwait(so, SO_RCV); 1533 if (error) { 1534 SOCK_RECVBUF_UNLOCK(so); 1535 SOCK_IO_RECV_UNLOCK(so); 1536 return (error); 1537 } 1538 } 1539 1540 if (sb == NULL) 1541 sb = &so->so_rcv; 1542 else if (m == NULL) 1543 m = STAILQ_FIRST(&sb->uxdg_mb); 1544 else 1545 MPASS(m == so->so_rcv.uxdg_peeked); 1546 1547 MPASS(sb->uxdg_cc > 0); 1548 M_ASSERTPKTHDR(m); 1549 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1550 1551 if (uio->uio_td) 1552 uio->uio_td->td_ru.ru_msgrcv++; 1553 1554 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1555 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1556 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1557 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1558 } else 1559 so->so_rcv.uxdg_peeked = NULL; 1560 1561 sb->uxdg_cc -= m->m_pkthdr.len; 1562 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1563 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1564 1565 if (__predict_false(flags & MSG_PEEK)) 1566 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1567 1568 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1569 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1570 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1571 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1572 SOCK_RECVBUF_UNLOCK(so); 1573 1574 if (psa != NULL) 1575 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1576 m = m_free(m); 1577 KASSERT(m, ("%s: no data or control after soname", __func__)); 1578 1579 /* 1580 * Packet to copyout() is now in 'm' and it is disconnected from the 1581 * queue. 1582 * 1583 * Process one or more MT_CONTROL mbufs present before any data mbufs 1584 * in the first mbuf chain on the socket buffer. We call into the 1585 * unp_externalize() to perform externalization (or freeing if 1586 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1587 * without MT_DATA mbufs. 1588 */ 1589 while (m != NULL && m->m_type == MT_CONTROL) { 1590 struct mbuf *cm; 1591 1592 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1593 * it frees whole chain, so we must disconnect the mbuf. 1594 */ 1595 cm = m; m = m->m_next; cm->m_next = NULL; 1596 error = unp_externalize(cm, controlp, flags); 1597 if (error != 0) { 1598 SOCK_IO_RECV_UNLOCK(so); 1599 unp_scan(m, unp_freerights); 1600 m_freem(m); 1601 return (error); 1602 } 1603 if (controlp != NULL) { 1604 while (*controlp != NULL) 1605 controlp = &(*controlp)->m_next; 1606 } 1607 } 1608 KASSERT(m == NULL || m->m_type == MT_DATA, 1609 ("%s: not MT_DATA mbuf %p", __func__, m)); 1610 while (m != NULL && uio->uio_resid > 0) { 1611 len = uio->uio_resid; 1612 if (len > m->m_len) 1613 len = m->m_len; 1614 error = uiomove(mtod(m, char *), (int)len, uio); 1615 if (error) { 1616 SOCK_IO_RECV_UNLOCK(so); 1617 m_freem(m); 1618 return (error); 1619 } 1620 if (len == m->m_len) 1621 m = m_free(m); 1622 else { 1623 m->m_data += len; 1624 m->m_len -= len; 1625 } 1626 } 1627 SOCK_IO_RECV_UNLOCK(so); 1628 1629 if (m != NULL) { 1630 if (flagsp != NULL) { 1631 if (flags & MSG_TRUNC) { 1632 /* Report real length of the packet */ 1633 uio->uio_resid -= m_length(m, NULL); 1634 } 1635 *flagsp |= MSG_TRUNC; 1636 } 1637 m_freem(m); 1638 } else if (flagsp != NULL) 1639 *flagsp &= ~MSG_TRUNC; 1640 1641 return (0); 1642 } 1643 1644 static bool 1645 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1646 { 1647 struct mbuf *mb, *n; 1648 struct sockbuf *sb; 1649 1650 SOCK_LOCK(so); 1651 if (SOLISTENING(so)) { 1652 SOCK_UNLOCK(so); 1653 return (false); 1654 } 1655 mb = NULL; 1656 sb = &so->so_rcv; 1657 SOCKBUF_LOCK(sb); 1658 if (sb->sb_fnrdy != NULL) { 1659 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1660 if (mb == m) { 1661 *errorp = sbready(sb, m, count); 1662 break; 1663 } 1664 mb = mb->m_next; 1665 if (mb == NULL) { 1666 mb = n; 1667 if (mb != NULL) 1668 n = mb->m_nextpkt; 1669 } 1670 } 1671 } 1672 SOCKBUF_UNLOCK(sb); 1673 SOCK_UNLOCK(so); 1674 return (mb != NULL); 1675 } 1676 1677 static int 1678 uipc_ready(struct socket *so, struct mbuf *m, int count) 1679 { 1680 struct unpcb *unp, *unp2; 1681 struct socket *so2; 1682 int error, i; 1683 1684 unp = sotounpcb(so); 1685 1686 KASSERT(so->so_type == SOCK_STREAM, 1687 ("%s: unexpected socket type for %p", __func__, so)); 1688 1689 UNP_PCB_LOCK(unp); 1690 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1691 UNP_PCB_UNLOCK(unp); 1692 so2 = unp2->unp_socket; 1693 SOCKBUF_LOCK(&so2->so_rcv); 1694 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1695 sorwakeup_locked(so2); 1696 else 1697 SOCKBUF_UNLOCK(&so2->so_rcv); 1698 UNP_PCB_UNLOCK(unp2); 1699 return (error); 1700 } 1701 UNP_PCB_UNLOCK(unp); 1702 1703 /* 1704 * The receiving socket has been disconnected, but may still be valid. 1705 * In this case, the now-ready mbufs are still present in its socket 1706 * buffer, so perform an exhaustive search before giving up and freeing 1707 * the mbufs. 1708 */ 1709 UNP_LINK_RLOCK(); 1710 LIST_FOREACH(unp, &unp_shead, unp_link) { 1711 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1712 break; 1713 } 1714 UNP_LINK_RUNLOCK(); 1715 1716 if (unp == NULL) { 1717 for (i = 0; i < count; i++) 1718 m = m_free(m); 1719 error = ECONNRESET; 1720 } 1721 return (error); 1722 } 1723 1724 static int 1725 uipc_sense(struct socket *so, struct stat *sb) 1726 { 1727 struct unpcb *unp; 1728 1729 unp = sotounpcb(so); 1730 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1731 1732 sb->st_blksize = so->so_snd.sb_hiwat; 1733 sb->st_dev = NODEV; 1734 sb->st_ino = unp->unp_ino; 1735 return (0); 1736 } 1737 1738 static int 1739 uipc_shutdown(struct socket *so) 1740 { 1741 struct unpcb *unp; 1742 1743 unp = sotounpcb(so); 1744 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1745 1746 UNP_PCB_LOCK(unp); 1747 socantsendmore(so); 1748 unp_shutdown(unp); 1749 UNP_PCB_UNLOCK(unp); 1750 return (0); 1751 } 1752 1753 static int 1754 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1755 { 1756 struct unpcb *unp; 1757 const struct sockaddr *sa; 1758 1759 unp = sotounpcb(so); 1760 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1761 1762 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1763 UNP_PCB_LOCK(unp); 1764 if (unp->unp_addr != NULL) 1765 sa = (struct sockaddr *) unp->unp_addr; 1766 else 1767 sa = &sun_noname; 1768 bcopy(sa, *nam, sa->sa_len); 1769 UNP_PCB_UNLOCK(unp); 1770 return (0); 1771 } 1772 1773 static struct pr_usrreqs uipc_usrreqs_dgram = { 1774 .pru_abort = uipc_abort, 1775 .pru_accept = uipc_accept, 1776 .pru_attach = uipc_attach, 1777 .pru_bind = uipc_bind, 1778 .pru_bindat = uipc_bindat, 1779 .pru_connect = uipc_connect, 1780 .pru_connectat = uipc_connectat, 1781 .pru_connect2 = uipc_connect2, 1782 .pru_detach = uipc_detach, 1783 .pru_disconnect = uipc_disconnect, 1784 .pru_peeraddr = uipc_peeraddr, 1785 .pru_sosend = uipc_sosend_dgram, 1786 .pru_sense = uipc_sense, 1787 .pru_shutdown = uipc_shutdown, 1788 .pru_sockaddr = uipc_sockaddr, 1789 .pru_soreceive = uipc_soreceive_dgram, 1790 .pru_close = uipc_close, 1791 }; 1792 1793 static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1794 .pru_abort = uipc_abort, 1795 .pru_accept = uipc_accept, 1796 .pru_attach = uipc_attach, 1797 .pru_bind = uipc_bind, 1798 .pru_bindat = uipc_bindat, 1799 .pru_connect = uipc_connect, 1800 .pru_connectat = uipc_connectat, 1801 .pru_connect2 = uipc_connect2, 1802 .pru_detach = uipc_detach, 1803 .pru_disconnect = uipc_disconnect, 1804 .pru_listen = uipc_listen, 1805 .pru_peeraddr = uipc_peeraddr, 1806 .pru_rcvd = uipc_rcvd, 1807 .pru_send = uipc_send, 1808 .pru_sense = uipc_sense, 1809 .pru_shutdown = uipc_shutdown, 1810 .pru_sockaddr = uipc_sockaddr, 1811 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1812 .pru_close = uipc_close, 1813 }; 1814 1815 static struct pr_usrreqs uipc_usrreqs_stream = { 1816 .pru_abort = uipc_abort, 1817 .pru_accept = uipc_accept, 1818 .pru_attach = uipc_attach, 1819 .pru_bind = uipc_bind, 1820 .pru_bindat = uipc_bindat, 1821 .pru_connect = uipc_connect, 1822 .pru_connectat = uipc_connectat, 1823 .pru_connect2 = uipc_connect2, 1824 .pru_detach = uipc_detach, 1825 .pru_disconnect = uipc_disconnect, 1826 .pru_listen = uipc_listen, 1827 .pru_peeraddr = uipc_peeraddr, 1828 .pru_rcvd = uipc_rcvd, 1829 .pru_send = uipc_send, 1830 .pru_ready = uipc_ready, 1831 .pru_sense = uipc_sense, 1832 .pru_shutdown = uipc_shutdown, 1833 .pru_sockaddr = uipc_sockaddr, 1834 .pru_soreceive = soreceive_generic, 1835 .pru_close = uipc_close, 1836 }; 1837 1838 static int 1839 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1840 { 1841 struct unpcb *unp; 1842 struct xucred xu; 1843 int error, optval; 1844 1845 if (sopt->sopt_level != SOL_LOCAL) 1846 return (EINVAL); 1847 1848 unp = sotounpcb(so); 1849 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1850 error = 0; 1851 switch (sopt->sopt_dir) { 1852 case SOPT_GET: 1853 switch (sopt->sopt_name) { 1854 case LOCAL_PEERCRED: 1855 UNP_PCB_LOCK(unp); 1856 if (unp->unp_flags & UNP_HAVEPC) 1857 xu = unp->unp_peercred; 1858 else { 1859 if (so->so_type == SOCK_STREAM) 1860 error = ENOTCONN; 1861 else 1862 error = EINVAL; 1863 } 1864 UNP_PCB_UNLOCK(unp); 1865 if (error == 0) 1866 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1867 break; 1868 1869 case LOCAL_CREDS: 1870 /* Unlocked read. */ 1871 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1872 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1873 break; 1874 1875 case LOCAL_CREDS_PERSISTENT: 1876 /* Unlocked read. */ 1877 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1878 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1879 break; 1880 1881 case LOCAL_CONNWAIT: 1882 /* Unlocked read. */ 1883 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1884 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1885 break; 1886 1887 default: 1888 error = EOPNOTSUPP; 1889 break; 1890 } 1891 break; 1892 1893 case SOPT_SET: 1894 switch (sopt->sopt_name) { 1895 case LOCAL_CREDS: 1896 case LOCAL_CREDS_PERSISTENT: 1897 case LOCAL_CONNWAIT: 1898 error = sooptcopyin(sopt, &optval, sizeof(optval), 1899 sizeof(optval)); 1900 if (error) 1901 break; 1902 1903 #define OPTSET(bit, exclusive) do { \ 1904 UNP_PCB_LOCK(unp); \ 1905 if (optval) { \ 1906 if ((unp->unp_flags & (exclusive)) != 0) { \ 1907 UNP_PCB_UNLOCK(unp); \ 1908 error = EINVAL; \ 1909 break; \ 1910 } \ 1911 unp->unp_flags |= (bit); \ 1912 } else \ 1913 unp->unp_flags &= ~(bit); \ 1914 UNP_PCB_UNLOCK(unp); \ 1915 } while (0) 1916 1917 switch (sopt->sopt_name) { 1918 case LOCAL_CREDS: 1919 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1920 break; 1921 1922 case LOCAL_CREDS_PERSISTENT: 1923 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1924 break; 1925 1926 case LOCAL_CONNWAIT: 1927 OPTSET(UNP_CONNWAIT, 0); 1928 break; 1929 1930 default: 1931 break; 1932 } 1933 break; 1934 #undef OPTSET 1935 default: 1936 error = ENOPROTOOPT; 1937 break; 1938 } 1939 break; 1940 1941 default: 1942 error = EOPNOTSUPP; 1943 break; 1944 } 1945 return (error); 1946 } 1947 1948 static int 1949 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1950 { 1951 1952 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1953 } 1954 1955 static int 1956 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1957 struct thread *td, bool return_locked) 1958 { 1959 struct mtx *vplock; 1960 struct sockaddr_un *soun; 1961 struct vnode *vp; 1962 struct socket *so2; 1963 struct unpcb *unp, *unp2, *unp3; 1964 struct nameidata nd; 1965 char buf[SOCK_MAXADDRLEN]; 1966 struct sockaddr *sa; 1967 cap_rights_t rights; 1968 int error, len; 1969 bool connreq; 1970 1971 if (nam->sa_family != AF_UNIX) 1972 return (EAFNOSUPPORT); 1973 if (nam->sa_len > sizeof(struct sockaddr_un)) 1974 return (EINVAL); 1975 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1976 if (len <= 0) 1977 return (EINVAL); 1978 soun = (struct sockaddr_un *)nam; 1979 bcopy(soun->sun_path, buf, len); 1980 buf[len] = 0; 1981 1982 error = 0; 1983 unp = sotounpcb(so); 1984 UNP_PCB_LOCK(unp); 1985 for (;;) { 1986 /* 1987 * Wait for connection state to stabilize. If a connection 1988 * already exists, give up. For datagram sockets, which permit 1989 * multiple consecutive connect(2) calls, upper layers are 1990 * responsible for disconnecting in advance of a subsequent 1991 * connect(2), but this is not synchronized with PCB connection 1992 * state. 1993 * 1994 * Also make sure that no threads are currently attempting to 1995 * lock the peer socket, to ensure that unp_conn cannot 1996 * transition between two valid sockets while locks are dropped. 1997 */ 1998 if (SOLISTENING(so)) 1999 error = EOPNOTSUPP; 2000 else if (unp->unp_conn != NULL) 2001 error = EISCONN; 2002 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 2003 error = EALREADY; 2004 } 2005 if (error != 0) { 2006 UNP_PCB_UNLOCK(unp); 2007 return (error); 2008 } 2009 if (unp->unp_pairbusy > 0) { 2010 unp->unp_flags |= UNP_WAITING; 2011 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 2012 continue; 2013 } 2014 break; 2015 } 2016 unp->unp_flags |= UNP_CONNECTING; 2017 UNP_PCB_UNLOCK(unp); 2018 2019 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 2020 if (connreq) 2021 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 2022 else 2023 sa = NULL; 2024 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 2025 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 2026 error = namei(&nd); 2027 if (error) 2028 vp = NULL; 2029 else 2030 vp = nd.ni_vp; 2031 ASSERT_VOP_LOCKED(vp, "unp_connect"); 2032 NDFREE_NOTHING(&nd); 2033 if (error) 2034 goto bad; 2035 2036 if (vp->v_type != VSOCK) { 2037 error = ENOTSOCK; 2038 goto bad; 2039 } 2040 #ifdef MAC 2041 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 2042 if (error) 2043 goto bad; 2044 #endif 2045 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 2046 if (error) 2047 goto bad; 2048 2049 unp = sotounpcb(so); 2050 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 2051 2052 vplock = mtx_pool_find(mtxpool_sleep, vp); 2053 mtx_lock(vplock); 2054 VOP_UNP_CONNECT(vp, &unp2); 2055 if (unp2 == NULL) { 2056 error = ECONNREFUSED; 2057 goto bad2; 2058 } 2059 so2 = unp2->unp_socket; 2060 if (so->so_type != so2->so_type) { 2061 error = EPROTOTYPE; 2062 goto bad2; 2063 } 2064 if (connreq) { 2065 if (SOLISTENING(so2)) { 2066 CURVNET_SET(so2->so_vnet); 2067 so2 = sonewconn(so2, 0); 2068 CURVNET_RESTORE(); 2069 } else 2070 so2 = NULL; 2071 if (so2 == NULL) { 2072 error = ECONNREFUSED; 2073 goto bad2; 2074 } 2075 unp3 = sotounpcb(so2); 2076 unp_pcb_lock_pair(unp2, unp3); 2077 if (unp2->unp_addr != NULL) { 2078 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 2079 unp3->unp_addr = (struct sockaddr_un *) sa; 2080 sa = NULL; 2081 } 2082 2083 unp_copy_peercred(td, unp3, unp, unp2); 2084 2085 UNP_PCB_UNLOCK(unp2); 2086 unp2 = unp3; 2087 2088 /* 2089 * It is safe to block on the PCB lock here since unp2 is 2090 * nascent and cannot be connected to any other sockets. 2091 */ 2092 UNP_PCB_LOCK(unp); 2093 #ifdef MAC 2094 mac_socketpeer_set_from_socket(so, so2); 2095 mac_socketpeer_set_from_socket(so2, so); 2096 #endif 2097 } else { 2098 unp_pcb_lock_pair(unp, unp2); 2099 } 2100 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 2101 sotounpcb(so2) == unp2, 2102 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 2103 unp_connect2(so, so2, PRU_CONNECT); 2104 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2105 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2106 unp->unp_flags &= ~UNP_CONNECTING; 2107 if (!return_locked) 2108 unp_pcb_unlock_pair(unp, unp2); 2109 bad2: 2110 mtx_unlock(vplock); 2111 bad: 2112 if (vp != NULL) { 2113 /* 2114 * If we are returning locked (called via uipc_sosend_dgram()), 2115 * we need to be sure that vput() won't sleep. This is 2116 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2117 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2118 */ 2119 MPASS(!(return_locked && connreq)); 2120 vput(vp); 2121 } 2122 free(sa, M_SONAME); 2123 if (__predict_false(error)) { 2124 UNP_PCB_LOCK(unp); 2125 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2126 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2127 unp->unp_flags &= ~UNP_CONNECTING; 2128 UNP_PCB_UNLOCK(unp); 2129 } 2130 return (error); 2131 } 2132 2133 /* 2134 * Set socket peer credentials at connection time. 2135 * 2136 * The client's PCB credentials are copied from its process structure. The 2137 * server's PCB credentials are copied from the socket on which it called 2138 * listen(2). uipc_listen cached that process's credentials at the time. 2139 */ 2140 void 2141 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2142 struct unpcb *server_unp, struct unpcb *listen_unp) 2143 { 2144 cru2xt(td, &client_unp->unp_peercred); 2145 client_unp->unp_flags |= UNP_HAVEPC; 2146 2147 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2148 sizeof(server_unp->unp_peercred)); 2149 server_unp->unp_flags |= UNP_HAVEPC; 2150 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2151 } 2152 2153 static void 2154 unp_connect2(struct socket *so, struct socket *so2, int req) 2155 { 2156 struct unpcb *unp; 2157 struct unpcb *unp2; 2158 2159 MPASS(so2->so_type == so->so_type); 2160 unp = sotounpcb(so); 2161 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2162 unp2 = sotounpcb(so2); 2163 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2164 2165 UNP_PCB_LOCK_ASSERT(unp); 2166 UNP_PCB_LOCK_ASSERT(unp2); 2167 KASSERT(unp->unp_conn == NULL, 2168 ("%s: socket %p is already connected", __func__, unp)); 2169 2170 unp->unp_conn = unp2; 2171 unp_pcb_hold(unp2); 2172 unp_pcb_hold(unp); 2173 switch (so->so_type) { 2174 case SOCK_DGRAM: 2175 UNP_REF_LIST_LOCK(); 2176 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2177 UNP_REF_LIST_UNLOCK(); 2178 soisconnected(so); 2179 break; 2180 2181 case SOCK_STREAM: 2182 case SOCK_SEQPACKET: 2183 KASSERT(unp2->unp_conn == NULL, 2184 ("%s: socket %p is already connected", __func__, unp2)); 2185 unp2->unp_conn = unp; 2186 if (req == PRU_CONNECT && 2187 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2188 soisconnecting(so); 2189 else 2190 soisconnected(so); 2191 soisconnected(so2); 2192 break; 2193 2194 default: 2195 panic("unp_connect2"); 2196 } 2197 } 2198 2199 static void 2200 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2201 { 2202 struct socket *so, *so2; 2203 struct mbuf *m = NULL; 2204 #ifdef INVARIANTS 2205 struct unpcb *unptmp; 2206 #endif 2207 2208 UNP_PCB_LOCK_ASSERT(unp); 2209 UNP_PCB_LOCK_ASSERT(unp2); 2210 KASSERT(unp->unp_conn == unp2, 2211 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2212 2213 unp->unp_conn = NULL; 2214 so = unp->unp_socket; 2215 so2 = unp2->unp_socket; 2216 switch (unp->unp_socket->so_type) { 2217 case SOCK_DGRAM: 2218 /* 2219 * Remove our send socket buffer from the peer's receive buffer. 2220 * Move the data to the receive buffer only if it is empty. 2221 * This is a protection against a scenario where a peer 2222 * connects, floods and disconnects, effectively blocking 2223 * sendto() from unconnected sockets. 2224 */ 2225 SOCK_RECVBUF_LOCK(so2); 2226 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2227 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2228 uxdg_clist); 2229 if (__predict_true((so2->so_rcv.sb_state & 2230 SBS_CANTRCVMORE) == 0) && 2231 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2232 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2233 &so->so_snd.uxdg_mb); 2234 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2235 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2236 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2237 } else { 2238 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2239 STAILQ_INIT(&so->so_snd.uxdg_mb); 2240 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2241 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2242 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2243 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2244 } 2245 /* Note: so may reconnect. */ 2246 so->so_snd.uxdg_cc = 0; 2247 so->so_snd.uxdg_ctl = 0; 2248 so->so_snd.uxdg_mbcnt = 0; 2249 } 2250 SOCK_RECVBUF_UNLOCK(so2); 2251 UNP_REF_LIST_LOCK(); 2252 #ifdef INVARIANTS 2253 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2254 if (unptmp == unp) 2255 break; 2256 } 2257 KASSERT(unptmp != NULL, 2258 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2259 #endif 2260 LIST_REMOVE(unp, unp_reflink); 2261 UNP_REF_LIST_UNLOCK(); 2262 if (so) { 2263 SOCK_LOCK(so); 2264 so->so_state &= ~SS_ISCONNECTED; 2265 SOCK_UNLOCK(so); 2266 } 2267 break; 2268 2269 case SOCK_STREAM: 2270 case SOCK_SEQPACKET: 2271 if (so) 2272 soisdisconnected(so); 2273 MPASS(unp2->unp_conn == unp); 2274 unp2->unp_conn = NULL; 2275 if (so2) 2276 soisdisconnected(so2); 2277 break; 2278 } 2279 2280 if (unp == unp2) { 2281 unp_pcb_rele_notlast(unp); 2282 if (!unp_pcb_rele(unp)) 2283 UNP_PCB_UNLOCK(unp); 2284 } else { 2285 if (!unp_pcb_rele(unp)) 2286 UNP_PCB_UNLOCK(unp); 2287 if (!unp_pcb_rele(unp2)) 2288 UNP_PCB_UNLOCK(unp2); 2289 } 2290 2291 if (m != NULL) { 2292 unp_scan(m, unp_freerights); 2293 m_freem(m); 2294 } 2295 } 2296 2297 /* 2298 * unp_pcblist() walks the global list of struct unpcb's to generate a 2299 * pointer list, bumping the refcount on each unpcb. It then copies them out 2300 * sequentially, validating the generation number on each to see if it has 2301 * been detached. All of this is necessary because copyout() may sleep on 2302 * disk I/O. 2303 */ 2304 static int 2305 unp_pcblist(SYSCTL_HANDLER_ARGS) 2306 { 2307 struct unpcb *unp, **unp_list; 2308 unp_gen_t gencnt; 2309 struct xunpgen *xug; 2310 struct unp_head *head; 2311 struct xunpcb *xu; 2312 u_int i; 2313 int error, n; 2314 2315 switch ((intptr_t)arg1) { 2316 case SOCK_STREAM: 2317 head = &unp_shead; 2318 break; 2319 2320 case SOCK_DGRAM: 2321 head = &unp_dhead; 2322 break; 2323 2324 case SOCK_SEQPACKET: 2325 head = &unp_sphead; 2326 break; 2327 2328 default: 2329 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2330 } 2331 2332 /* 2333 * The process of preparing the PCB list is too time-consuming and 2334 * resource-intensive to repeat twice on every request. 2335 */ 2336 if (req->oldptr == NULL) { 2337 n = unp_count; 2338 req->oldidx = 2 * (sizeof *xug) 2339 + (n + n/8) * sizeof(struct xunpcb); 2340 return (0); 2341 } 2342 2343 if (req->newptr != NULL) 2344 return (EPERM); 2345 2346 /* 2347 * OK, now we're committed to doing something. 2348 */ 2349 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2350 UNP_LINK_RLOCK(); 2351 gencnt = unp_gencnt; 2352 n = unp_count; 2353 UNP_LINK_RUNLOCK(); 2354 2355 xug->xug_len = sizeof *xug; 2356 xug->xug_count = n; 2357 xug->xug_gen = gencnt; 2358 xug->xug_sogen = so_gencnt; 2359 error = SYSCTL_OUT(req, xug, sizeof *xug); 2360 if (error) { 2361 free(xug, M_TEMP); 2362 return (error); 2363 } 2364 2365 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2366 2367 UNP_LINK_RLOCK(); 2368 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2369 unp = LIST_NEXT(unp, unp_link)) { 2370 UNP_PCB_LOCK(unp); 2371 if (unp->unp_gencnt <= gencnt) { 2372 if (cr_cansee(req->td->td_ucred, 2373 unp->unp_socket->so_cred)) { 2374 UNP_PCB_UNLOCK(unp); 2375 continue; 2376 } 2377 unp_list[i++] = unp; 2378 unp_pcb_hold(unp); 2379 } 2380 UNP_PCB_UNLOCK(unp); 2381 } 2382 UNP_LINK_RUNLOCK(); 2383 n = i; /* In case we lost some during malloc. */ 2384 2385 error = 0; 2386 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2387 for (i = 0; i < n; i++) { 2388 unp = unp_list[i]; 2389 UNP_PCB_LOCK(unp); 2390 if (unp_pcb_rele(unp)) 2391 continue; 2392 2393 if (unp->unp_gencnt <= gencnt) { 2394 xu->xu_len = sizeof *xu; 2395 xu->xu_unpp = (uintptr_t)unp; 2396 /* 2397 * XXX - need more locking here to protect against 2398 * connect/disconnect races for SMP. 2399 */ 2400 if (unp->unp_addr != NULL) 2401 bcopy(unp->unp_addr, &xu->xu_addr, 2402 unp->unp_addr->sun_len); 2403 else 2404 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2405 if (unp->unp_conn != NULL && 2406 unp->unp_conn->unp_addr != NULL) 2407 bcopy(unp->unp_conn->unp_addr, 2408 &xu->xu_caddr, 2409 unp->unp_conn->unp_addr->sun_len); 2410 else 2411 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2412 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2413 xu->unp_conn = (uintptr_t)unp->unp_conn; 2414 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2415 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2416 xu->unp_gencnt = unp->unp_gencnt; 2417 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2418 UNP_PCB_UNLOCK(unp); 2419 error = SYSCTL_OUT(req, xu, sizeof *xu); 2420 } else { 2421 UNP_PCB_UNLOCK(unp); 2422 } 2423 } 2424 free(xu, M_TEMP); 2425 if (!error) { 2426 /* 2427 * Give the user an updated idea of our state. If the 2428 * generation differs from what we told her before, she knows 2429 * that something happened while we were processing this 2430 * request, and it might be necessary to retry. 2431 */ 2432 xug->xug_gen = unp_gencnt; 2433 xug->xug_sogen = so_gencnt; 2434 xug->xug_count = unp_count; 2435 error = SYSCTL_OUT(req, xug, sizeof *xug); 2436 } 2437 free(unp_list, M_TEMP); 2438 free(xug, M_TEMP); 2439 return (error); 2440 } 2441 2442 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2443 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2444 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2445 "List of active local datagram sockets"); 2446 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2447 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2448 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2449 "List of active local stream sockets"); 2450 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2451 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2452 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2453 "List of active local seqpacket sockets"); 2454 2455 static void 2456 unp_shutdown(struct unpcb *unp) 2457 { 2458 struct unpcb *unp2; 2459 struct socket *so; 2460 2461 UNP_PCB_LOCK_ASSERT(unp); 2462 2463 unp2 = unp->unp_conn; 2464 if ((unp->unp_socket->so_type == SOCK_STREAM || 2465 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2466 so = unp2->unp_socket; 2467 if (so != NULL) 2468 socantrcvmore(so); 2469 } 2470 } 2471 2472 static void 2473 unp_drop(struct unpcb *unp) 2474 { 2475 struct socket *so; 2476 struct unpcb *unp2; 2477 2478 /* 2479 * Regardless of whether the socket's peer dropped the connection 2480 * with this socket by aborting or disconnecting, POSIX requires 2481 * that ECONNRESET is returned. 2482 */ 2483 2484 UNP_PCB_LOCK(unp); 2485 so = unp->unp_socket; 2486 if (so) 2487 so->so_error = ECONNRESET; 2488 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2489 /* Last reference dropped in unp_disconnect(). */ 2490 unp_pcb_rele_notlast(unp); 2491 unp_disconnect(unp, unp2); 2492 } else if (!unp_pcb_rele(unp)) { 2493 UNP_PCB_UNLOCK(unp); 2494 } 2495 } 2496 2497 static void 2498 unp_freerights(struct filedescent **fdep, int fdcount) 2499 { 2500 struct file *fp; 2501 int i; 2502 2503 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2504 2505 for (i = 0; i < fdcount; i++) { 2506 fp = fdep[i]->fde_file; 2507 filecaps_free(&fdep[i]->fde_caps); 2508 unp_discard(fp); 2509 } 2510 free(fdep[0], M_FILECAPS); 2511 } 2512 2513 static int 2514 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2515 { 2516 struct thread *td = curthread; /* XXX */ 2517 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2518 int i; 2519 int *fdp; 2520 struct filedesc *fdesc = td->td_proc->p_fd; 2521 struct filedescent **fdep; 2522 void *data; 2523 socklen_t clen = control->m_len, datalen; 2524 int error, newfds; 2525 u_int newlen; 2526 2527 UNP_LINK_UNLOCK_ASSERT(); 2528 2529 error = 0; 2530 if (controlp != NULL) /* controlp == NULL => free control messages */ 2531 *controlp = NULL; 2532 while (cm != NULL) { 2533 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2534 2535 data = CMSG_DATA(cm); 2536 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2537 if (cm->cmsg_level == SOL_SOCKET 2538 && cm->cmsg_type == SCM_RIGHTS) { 2539 newfds = datalen / sizeof(*fdep); 2540 if (newfds == 0) 2541 goto next; 2542 fdep = data; 2543 2544 /* If we're not outputting the descriptors free them. */ 2545 if (error || controlp == NULL) { 2546 unp_freerights(fdep, newfds); 2547 goto next; 2548 } 2549 FILEDESC_XLOCK(fdesc); 2550 2551 /* 2552 * Now change each pointer to an fd in the global 2553 * table to an integer that is the index to the local 2554 * fd table entry that we set up to point to the 2555 * global one we are transferring. 2556 */ 2557 newlen = newfds * sizeof(int); 2558 *controlp = sbcreatecontrol(NULL, newlen, 2559 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2560 2561 fdp = (int *) 2562 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2563 if ((error = fdallocn(td, 0, fdp, newfds))) { 2564 FILEDESC_XUNLOCK(fdesc); 2565 unp_freerights(fdep, newfds); 2566 m_freem(*controlp); 2567 *controlp = NULL; 2568 goto next; 2569 } 2570 for (i = 0; i < newfds; i++, fdp++) { 2571 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2572 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2573 &fdep[i]->fde_caps); 2574 unp_externalize_fp(fdep[i]->fde_file); 2575 } 2576 2577 /* 2578 * The new type indicates that the mbuf data refers to 2579 * kernel resources that may need to be released before 2580 * the mbuf is freed. 2581 */ 2582 m_chtype(*controlp, MT_EXTCONTROL); 2583 FILEDESC_XUNLOCK(fdesc); 2584 free(fdep[0], M_FILECAPS); 2585 } else { 2586 /* We can just copy anything else across. */ 2587 if (error || controlp == NULL) 2588 goto next; 2589 *controlp = sbcreatecontrol(NULL, datalen, 2590 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2591 bcopy(data, 2592 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2593 datalen); 2594 } 2595 controlp = &(*controlp)->m_next; 2596 2597 next: 2598 if (CMSG_SPACE(datalen) < clen) { 2599 clen -= CMSG_SPACE(datalen); 2600 cm = (struct cmsghdr *) 2601 ((caddr_t)cm + CMSG_SPACE(datalen)); 2602 } else { 2603 clen = 0; 2604 cm = NULL; 2605 } 2606 } 2607 2608 m_freem(control); 2609 return (error); 2610 } 2611 2612 static void 2613 unp_zone_change(void *tag) 2614 { 2615 2616 uma_zone_set_max(unp_zone, maxsockets); 2617 } 2618 2619 #ifdef INVARIANTS 2620 static void 2621 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2622 { 2623 struct unpcb *unp; 2624 2625 unp = mem; 2626 2627 KASSERT(LIST_EMPTY(&unp->unp_refs), 2628 ("%s: unpcb %p has lingering refs", __func__, unp)); 2629 KASSERT(unp->unp_socket == NULL, 2630 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2631 KASSERT(unp->unp_vnode == NULL, 2632 ("%s: unpcb %p has vnode references", __func__, unp)); 2633 KASSERT(unp->unp_conn == NULL, 2634 ("%s: unpcb %p is still connected", __func__, unp)); 2635 KASSERT(unp->unp_addr == NULL, 2636 ("%s: unpcb %p has leaked addr", __func__, unp)); 2637 } 2638 #endif 2639 2640 static void 2641 unp_init(void *arg __unused) 2642 { 2643 uma_dtor dtor; 2644 2645 #ifdef INVARIANTS 2646 dtor = unp_zdtor; 2647 #else 2648 dtor = NULL; 2649 #endif 2650 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2651 NULL, NULL, UMA_ALIGN_CACHE, 0); 2652 uma_zone_set_max(unp_zone, maxsockets); 2653 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2654 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2655 NULL, EVENTHANDLER_PRI_ANY); 2656 LIST_INIT(&unp_dhead); 2657 LIST_INIT(&unp_shead); 2658 LIST_INIT(&unp_sphead); 2659 SLIST_INIT(&unp_defers); 2660 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2661 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2662 UNP_LINK_LOCK_INIT(); 2663 UNP_DEFERRED_LOCK_INIT(); 2664 } 2665 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2666 2667 static void 2668 unp_internalize_cleanup_rights(struct mbuf *control) 2669 { 2670 struct cmsghdr *cp; 2671 struct mbuf *m; 2672 void *data; 2673 socklen_t datalen; 2674 2675 for (m = control; m != NULL; m = m->m_next) { 2676 cp = mtod(m, struct cmsghdr *); 2677 if (cp->cmsg_level != SOL_SOCKET || 2678 cp->cmsg_type != SCM_RIGHTS) 2679 continue; 2680 data = CMSG_DATA(cp); 2681 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2682 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2683 } 2684 } 2685 2686 static int 2687 unp_internalize(struct mbuf **controlp, struct thread *td, 2688 struct mbuf **clast, u_int *space, u_int *mbcnt) 2689 { 2690 struct mbuf *control, **initial_controlp; 2691 struct proc *p; 2692 struct filedesc *fdesc; 2693 struct bintime *bt; 2694 struct cmsghdr *cm; 2695 struct cmsgcred *cmcred; 2696 struct filedescent *fde, **fdep, *fdev; 2697 struct file *fp; 2698 struct timeval *tv; 2699 struct timespec *ts; 2700 void *data; 2701 socklen_t clen, datalen; 2702 int i, j, error, *fdp, oldfds; 2703 u_int newlen; 2704 2705 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2706 UNP_LINK_UNLOCK_ASSERT(); 2707 2708 p = td->td_proc; 2709 fdesc = p->p_fd; 2710 error = 0; 2711 control = *controlp; 2712 *controlp = NULL; 2713 initial_controlp = controlp; 2714 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2715 data = CMSG_DATA(cm); 2716 2717 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2718 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2719 (char *)cm + cm->cmsg_len >= (char *)data; 2720 2721 clen -= min(CMSG_SPACE(datalen), clen), 2722 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2723 data = CMSG_DATA(cm)) { 2724 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2725 switch (cm->cmsg_type) { 2726 case SCM_CREDS: 2727 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2728 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2729 cmcred = (struct cmsgcred *) 2730 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2731 cmcred->cmcred_pid = p->p_pid; 2732 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2733 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2734 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2735 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2736 CMGROUP_MAX); 2737 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2738 cmcred->cmcred_groups[i] = 2739 td->td_ucred->cr_groups[i]; 2740 break; 2741 2742 case SCM_RIGHTS: 2743 oldfds = datalen / sizeof (int); 2744 if (oldfds == 0) 2745 continue; 2746 /* On some machines sizeof pointer is bigger than 2747 * sizeof int, so we need to check if data fits into 2748 * single mbuf. We could allocate several mbufs, and 2749 * unp_externalize() should even properly handle that. 2750 * But it is not worth to complicate the code for an 2751 * insane scenario of passing over 200 file descriptors 2752 * at once. 2753 */ 2754 newlen = oldfds * sizeof(fdep[0]); 2755 if (CMSG_SPACE(newlen) > MCLBYTES) { 2756 error = EMSGSIZE; 2757 goto out; 2758 } 2759 /* 2760 * Check that all the FDs passed in refer to legal 2761 * files. If not, reject the entire operation. 2762 */ 2763 fdp = data; 2764 FILEDESC_SLOCK(fdesc); 2765 for (i = 0; i < oldfds; i++, fdp++) { 2766 fp = fget_noref(fdesc, *fdp); 2767 if (fp == NULL) { 2768 FILEDESC_SUNLOCK(fdesc); 2769 error = EBADF; 2770 goto out; 2771 } 2772 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2773 FILEDESC_SUNLOCK(fdesc); 2774 error = EOPNOTSUPP; 2775 goto out; 2776 } 2777 } 2778 2779 /* 2780 * Now replace the integer FDs with pointers to the 2781 * file structure and capability rights. 2782 */ 2783 *controlp = sbcreatecontrol(NULL, newlen, 2784 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2785 fdp = data; 2786 for (i = 0; i < oldfds; i++, fdp++) { 2787 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2788 fdp = data; 2789 for (j = 0; j < i; j++, fdp++) { 2790 fdrop(fdesc->fd_ofiles[*fdp]. 2791 fde_file, td); 2792 } 2793 FILEDESC_SUNLOCK(fdesc); 2794 error = EBADF; 2795 goto out; 2796 } 2797 } 2798 fdp = data; 2799 fdep = (struct filedescent **) 2800 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2801 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2802 M_WAITOK); 2803 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2804 fde = &fdesc->fd_ofiles[*fdp]; 2805 fdep[i] = fdev; 2806 fdep[i]->fde_file = fde->fde_file; 2807 filecaps_copy(&fde->fde_caps, 2808 &fdep[i]->fde_caps, true); 2809 unp_internalize_fp(fdep[i]->fde_file); 2810 } 2811 FILEDESC_SUNLOCK(fdesc); 2812 break; 2813 2814 case SCM_TIMESTAMP: 2815 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2816 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2817 tv = (struct timeval *) 2818 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2819 microtime(tv); 2820 break; 2821 2822 case SCM_BINTIME: 2823 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2824 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2825 bt = (struct bintime *) 2826 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2827 bintime(bt); 2828 break; 2829 2830 case SCM_REALTIME: 2831 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2832 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2833 ts = (struct timespec *) 2834 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2835 nanotime(ts); 2836 break; 2837 2838 case SCM_MONOTONIC: 2839 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2840 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2841 ts = (struct timespec *) 2842 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2843 nanouptime(ts); 2844 break; 2845 2846 default: 2847 error = EINVAL; 2848 goto out; 2849 } 2850 2851 if (space != NULL) { 2852 *space += (*controlp)->m_len; 2853 *mbcnt += MSIZE; 2854 if ((*controlp)->m_flags & M_EXT) 2855 *mbcnt += (*controlp)->m_ext.ext_size; 2856 *clast = *controlp; 2857 } 2858 controlp = &(*controlp)->m_next; 2859 } 2860 if (clen > 0) 2861 error = EINVAL; 2862 2863 out: 2864 if (error != 0 && initial_controlp != NULL) 2865 unp_internalize_cleanup_rights(*initial_controlp); 2866 m_freem(control); 2867 return (error); 2868 } 2869 2870 static struct mbuf * 2871 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2872 struct mbuf **clast, u_int *space, u_int *mbcnt) 2873 { 2874 struct mbuf *m, *n, *n_prev; 2875 const struct cmsghdr *cm; 2876 int ngroups, i, cmsgtype; 2877 size_t ctrlsz; 2878 2879 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2880 if (mode & UNP_WANTCRED_ALWAYS) { 2881 ctrlsz = SOCKCRED2SIZE(ngroups); 2882 cmsgtype = SCM_CREDS2; 2883 } else { 2884 ctrlsz = SOCKCREDSIZE(ngroups); 2885 cmsgtype = SCM_CREDS; 2886 } 2887 2888 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2889 if (m == NULL) 2890 return (control); 2891 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2892 2893 if (mode & UNP_WANTCRED_ALWAYS) { 2894 struct sockcred2 *sc; 2895 2896 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2897 sc->sc_version = 0; 2898 sc->sc_pid = td->td_proc->p_pid; 2899 sc->sc_uid = td->td_ucred->cr_ruid; 2900 sc->sc_euid = td->td_ucred->cr_uid; 2901 sc->sc_gid = td->td_ucred->cr_rgid; 2902 sc->sc_egid = td->td_ucred->cr_gid; 2903 sc->sc_ngroups = ngroups; 2904 for (i = 0; i < sc->sc_ngroups; i++) 2905 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2906 } else { 2907 struct sockcred *sc; 2908 2909 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2910 sc->sc_uid = td->td_ucred->cr_ruid; 2911 sc->sc_euid = td->td_ucred->cr_uid; 2912 sc->sc_gid = td->td_ucred->cr_rgid; 2913 sc->sc_egid = td->td_ucred->cr_gid; 2914 sc->sc_ngroups = ngroups; 2915 for (i = 0; i < sc->sc_ngroups; i++) 2916 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2917 } 2918 2919 /* 2920 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2921 * created SCM_CREDS control message (struct sockcred) has another 2922 * format. 2923 */ 2924 if (control != NULL && cmsgtype == SCM_CREDS) 2925 for (n = control, n_prev = NULL; n != NULL;) { 2926 cm = mtod(n, struct cmsghdr *); 2927 if (cm->cmsg_level == SOL_SOCKET && 2928 cm->cmsg_type == SCM_CREDS) { 2929 if (n_prev == NULL) 2930 control = n->m_next; 2931 else 2932 n_prev->m_next = n->m_next; 2933 if (space != NULL) { 2934 MPASS(*space >= n->m_len); 2935 *space -= n->m_len; 2936 MPASS(*mbcnt >= MSIZE); 2937 *mbcnt -= MSIZE; 2938 if (n->m_flags & M_EXT) { 2939 MPASS(*mbcnt >= 2940 n->m_ext.ext_size); 2941 *mbcnt -= n->m_ext.ext_size; 2942 } 2943 MPASS(clast); 2944 if (*clast == n) { 2945 MPASS(n->m_next == NULL); 2946 if (n_prev == NULL) 2947 *clast = m; 2948 else 2949 *clast = n_prev; 2950 } 2951 } 2952 n = m_free(n); 2953 } else { 2954 n_prev = n; 2955 n = n->m_next; 2956 } 2957 } 2958 2959 /* Prepend it to the head. */ 2960 m->m_next = control; 2961 if (space != NULL) { 2962 *space += m->m_len; 2963 *mbcnt += MSIZE; 2964 if (control == NULL) 2965 *clast = m; 2966 } 2967 return (m); 2968 } 2969 2970 static struct unpcb * 2971 fptounp(struct file *fp) 2972 { 2973 struct socket *so; 2974 2975 if (fp->f_type != DTYPE_SOCKET) 2976 return (NULL); 2977 if ((so = fp->f_data) == NULL) 2978 return (NULL); 2979 if (so->so_proto->pr_domain != &localdomain) 2980 return (NULL); 2981 return sotounpcb(so); 2982 } 2983 2984 static void 2985 unp_discard(struct file *fp) 2986 { 2987 struct unp_defer *dr; 2988 2989 if (unp_externalize_fp(fp)) { 2990 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2991 dr->ud_fp = fp; 2992 UNP_DEFERRED_LOCK(); 2993 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2994 UNP_DEFERRED_UNLOCK(); 2995 atomic_add_int(&unp_defers_count, 1); 2996 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2997 } else 2998 closef_nothread(fp); 2999 } 3000 3001 static void 3002 unp_process_defers(void *arg __unused, int pending) 3003 { 3004 struct unp_defer *dr; 3005 SLIST_HEAD(, unp_defer) drl; 3006 int count; 3007 3008 SLIST_INIT(&drl); 3009 for (;;) { 3010 UNP_DEFERRED_LOCK(); 3011 if (SLIST_FIRST(&unp_defers) == NULL) { 3012 UNP_DEFERRED_UNLOCK(); 3013 break; 3014 } 3015 SLIST_SWAP(&unp_defers, &drl, unp_defer); 3016 UNP_DEFERRED_UNLOCK(); 3017 count = 0; 3018 while ((dr = SLIST_FIRST(&drl)) != NULL) { 3019 SLIST_REMOVE_HEAD(&drl, ud_link); 3020 closef_nothread(dr->ud_fp); 3021 free(dr, M_TEMP); 3022 count++; 3023 } 3024 atomic_add_int(&unp_defers_count, -count); 3025 } 3026 } 3027 3028 static void 3029 unp_internalize_fp(struct file *fp) 3030 { 3031 struct unpcb *unp; 3032 3033 UNP_LINK_WLOCK(); 3034 if ((unp = fptounp(fp)) != NULL) { 3035 unp->unp_file = fp; 3036 unp->unp_msgcount++; 3037 } 3038 unp_rights++; 3039 UNP_LINK_WUNLOCK(); 3040 } 3041 3042 static int 3043 unp_externalize_fp(struct file *fp) 3044 { 3045 struct unpcb *unp; 3046 int ret; 3047 3048 UNP_LINK_WLOCK(); 3049 if ((unp = fptounp(fp)) != NULL) { 3050 unp->unp_msgcount--; 3051 ret = 1; 3052 } else 3053 ret = 0; 3054 unp_rights--; 3055 UNP_LINK_WUNLOCK(); 3056 return (ret); 3057 } 3058 3059 /* 3060 * unp_defer indicates whether additional work has been defered for a future 3061 * pass through unp_gc(). It is thread local and does not require explicit 3062 * synchronization. 3063 */ 3064 static int unp_marked; 3065 3066 static void 3067 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 3068 { 3069 struct unpcb *unp; 3070 struct file *fp; 3071 int i; 3072 3073 /* 3074 * This function can only be called from the gc task. 3075 */ 3076 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 3077 ("%s: not on gc callout", __func__)); 3078 UNP_LINK_LOCK_ASSERT(); 3079 3080 for (i = 0; i < fdcount; i++) { 3081 fp = fdep[i]->fde_file; 3082 if ((unp = fptounp(fp)) == NULL) 3083 continue; 3084 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3085 continue; 3086 unp->unp_gcrefs--; 3087 } 3088 } 3089 3090 static void 3091 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 3092 { 3093 struct unpcb *unp; 3094 struct file *fp; 3095 int i; 3096 3097 /* 3098 * This function can only be called from the gc task. 3099 */ 3100 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 3101 ("%s: not on gc callout", __func__)); 3102 UNP_LINK_LOCK_ASSERT(); 3103 3104 for (i = 0; i < fdcount; i++) { 3105 fp = fdep[i]->fde_file; 3106 if ((unp = fptounp(fp)) == NULL) 3107 continue; 3108 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3109 continue; 3110 unp->unp_gcrefs++; 3111 unp_marked++; 3112 } 3113 } 3114 3115 static void 3116 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3117 { 3118 struct sockbuf *sb; 3119 3120 SOCK_LOCK_ASSERT(so); 3121 3122 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3123 return; 3124 3125 SOCK_RECVBUF_LOCK(so); 3126 switch (so->so_type) { 3127 case SOCK_DGRAM: 3128 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3129 unp_scan(so->so_rcv.uxdg_peeked, op); 3130 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3131 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3132 break; 3133 case SOCK_STREAM: 3134 case SOCK_SEQPACKET: 3135 unp_scan(so->so_rcv.sb_mb, op); 3136 break; 3137 } 3138 SOCK_RECVBUF_UNLOCK(so); 3139 } 3140 3141 static void 3142 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3143 { 3144 struct socket *so, *soa; 3145 3146 so = unp->unp_socket; 3147 SOCK_LOCK(so); 3148 if (SOLISTENING(so)) { 3149 /* 3150 * Mark all sockets in our accept queue. 3151 */ 3152 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3153 unp_scan_socket(soa, op); 3154 } else { 3155 /* 3156 * Mark all sockets we reference with RIGHTS. 3157 */ 3158 unp_scan_socket(so, op); 3159 } 3160 SOCK_UNLOCK(so); 3161 } 3162 3163 static int unp_recycled; 3164 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3165 "Number of unreachable sockets claimed by the garbage collector."); 3166 3167 static int unp_taskcount; 3168 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3169 "Number of times the garbage collector has run."); 3170 3171 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3172 "Number of active local sockets."); 3173 3174 static void 3175 unp_gc(__unused void *arg, int pending) 3176 { 3177 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3178 NULL }; 3179 struct unp_head **head; 3180 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3181 struct file *f, **unref; 3182 struct unpcb *unp, *unptmp; 3183 int i, total, unp_unreachable; 3184 3185 LIST_INIT(&unp_deadhead); 3186 unp_taskcount++; 3187 UNP_LINK_RLOCK(); 3188 /* 3189 * First determine which sockets may be in cycles. 3190 */ 3191 unp_unreachable = 0; 3192 3193 for (head = heads; *head != NULL; head++) 3194 LIST_FOREACH(unp, *head, unp_link) { 3195 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3196 ("%s: unp %p has unexpected gc flags 0x%x", 3197 __func__, unp, (unsigned int)unp->unp_gcflag)); 3198 3199 f = unp->unp_file; 3200 3201 /* 3202 * Check for an unreachable socket potentially in a 3203 * cycle. It must be in a queue as indicated by 3204 * msgcount, and this must equal the file reference 3205 * count. Note that when msgcount is 0 the file is 3206 * NULL. 3207 */ 3208 if (f != NULL && unp->unp_msgcount != 0 && 3209 refcount_load(&f->f_count) == unp->unp_msgcount) { 3210 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3211 unp->unp_gcflag |= UNPGC_DEAD; 3212 unp->unp_gcrefs = unp->unp_msgcount; 3213 unp_unreachable++; 3214 } 3215 } 3216 3217 /* 3218 * Scan all sockets previously marked as potentially being in a cycle 3219 * and remove the references each socket holds on any UNPGC_DEAD 3220 * sockets in its queue. After this step, all remaining references on 3221 * sockets marked UNPGC_DEAD should not be part of any cycle. 3222 */ 3223 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3224 unp_gc_scan(unp, unp_remove_dead_ref); 3225 3226 /* 3227 * If a socket still has a non-negative refcount, it cannot be in a 3228 * cycle. In this case increment refcount of all children iteratively. 3229 * Stop the scan once we do a complete loop without discovering 3230 * a new reachable socket. 3231 */ 3232 do { 3233 unp_marked = 0; 3234 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3235 if (unp->unp_gcrefs > 0) { 3236 unp->unp_gcflag &= ~UNPGC_DEAD; 3237 LIST_REMOVE(unp, unp_dead); 3238 KASSERT(unp_unreachable > 0, 3239 ("%s: unp_unreachable underflow.", 3240 __func__)); 3241 unp_unreachable--; 3242 unp_gc_scan(unp, unp_restore_undead_ref); 3243 } 3244 } while (unp_marked); 3245 3246 UNP_LINK_RUNLOCK(); 3247 3248 if (unp_unreachable == 0) 3249 return; 3250 3251 /* 3252 * Allocate space for a local array of dead unpcbs. 3253 * TODO: can this path be simplified by instead using the local 3254 * dead list at unp_deadhead, after taking out references 3255 * on the file object and/or unpcb and dropping the link lock? 3256 */ 3257 unref = malloc(unp_unreachable * sizeof(struct file *), 3258 M_TEMP, M_WAITOK); 3259 3260 /* 3261 * Iterate looking for sockets which have been specifically marked 3262 * as unreachable and store them locally. 3263 */ 3264 UNP_LINK_RLOCK(); 3265 total = 0; 3266 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3267 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3268 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3269 unp->unp_gcflag &= ~UNPGC_DEAD; 3270 f = unp->unp_file; 3271 if (unp->unp_msgcount == 0 || f == NULL || 3272 refcount_load(&f->f_count) != unp->unp_msgcount || 3273 !fhold(f)) 3274 continue; 3275 unref[total++] = f; 3276 KASSERT(total <= unp_unreachable, 3277 ("%s: incorrect unreachable count.", __func__)); 3278 } 3279 UNP_LINK_RUNLOCK(); 3280 3281 /* 3282 * Now flush all sockets, free'ing rights. This will free the 3283 * struct files associated with these sockets but leave each socket 3284 * with one remaining ref. 3285 */ 3286 for (i = 0; i < total; i++) { 3287 struct socket *so; 3288 3289 so = unref[i]->f_data; 3290 CURVNET_SET(so->so_vnet); 3291 sorflush(so); 3292 CURVNET_RESTORE(); 3293 } 3294 3295 /* 3296 * And finally release the sockets so they can be reclaimed. 3297 */ 3298 for (i = 0; i < total; i++) 3299 fdrop(unref[i], NULL); 3300 unp_recycled += total; 3301 free(unref, M_TEMP); 3302 } 3303 3304 /* 3305 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3306 */ 3307 static void 3308 unp_dispose(struct socket *so) 3309 { 3310 struct sockbuf *sb; 3311 struct unpcb *unp; 3312 struct mbuf *m; 3313 3314 MPASS(!SOLISTENING(so)); 3315 3316 unp = sotounpcb(so); 3317 UNP_LINK_WLOCK(); 3318 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3319 UNP_LINK_WUNLOCK(); 3320 3321 /* 3322 * Grab our special mbufs before calling sbrelease(). 3323 */ 3324 SOCK_RECVBUF_LOCK(so); 3325 switch (so->so_type) { 3326 case SOCK_DGRAM: 3327 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3328 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3329 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3330 /* Note: socket of sb may reconnect. */ 3331 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3332 } 3333 sb = &so->so_rcv; 3334 if (sb->uxdg_peeked != NULL) { 3335 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3336 m_stailqpkt); 3337 sb->uxdg_peeked = NULL; 3338 } 3339 m = STAILQ_FIRST(&sb->uxdg_mb); 3340 STAILQ_INIT(&sb->uxdg_mb); 3341 /* XXX: our shortened sbrelease() */ 3342 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3343 RLIM_INFINITY); 3344 /* 3345 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3346 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3347 * data to the socket. 3348 * We are now in dom_dispose and it could be a call from 3349 * soshutdown() or from the final sofree(). The sofree() case 3350 * is simple as it guarantees that no more sends will happen, 3351 * however we can race with unp_disconnect() from our peer. 3352 * The shutdown(2) case is more exotic. It would call into 3353 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3354 * possible if we did connect(2) on this socket and we also 3355 * had it bound with bind(2) and receive connections from other 3356 * sockets. Because soshutdown() violates POSIX (see comment 3357 * there) we will end up here shutting down our receive side. 3358 * Of course this will have affect not only on the peer we 3359 * connect(2)ed to, but also on all of the peers who had 3360 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3361 */ 3362 sb->sb_state |= SBS_CANTRCVMORE; 3363 break; 3364 case SOCK_STREAM: 3365 case SOCK_SEQPACKET: 3366 sb = &so->so_rcv; 3367 m = sbcut_locked(sb, sb->sb_ccc); 3368 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3369 ("%s: ccc %u mb %p mbcnt %u", __func__, 3370 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3371 sbrelease_locked(so, SO_RCV); 3372 break; 3373 } 3374 SOCK_RECVBUF_UNLOCK(so); 3375 if (SOCK_IO_RECV_OWNED(so)) 3376 SOCK_IO_RECV_UNLOCK(so); 3377 3378 if (m != NULL) { 3379 unp_scan(m, unp_freerights); 3380 m_freem(m); 3381 } 3382 } 3383 3384 static void 3385 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3386 { 3387 struct mbuf *m; 3388 struct cmsghdr *cm; 3389 void *data; 3390 socklen_t clen, datalen; 3391 3392 while (m0 != NULL) { 3393 for (m = m0; m; m = m->m_next) { 3394 if (m->m_type != MT_CONTROL) 3395 continue; 3396 3397 cm = mtod(m, struct cmsghdr *); 3398 clen = m->m_len; 3399 3400 while (cm != NULL) { 3401 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3402 break; 3403 3404 data = CMSG_DATA(cm); 3405 datalen = (caddr_t)cm + cm->cmsg_len 3406 - (caddr_t)data; 3407 3408 if (cm->cmsg_level == SOL_SOCKET && 3409 cm->cmsg_type == SCM_RIGHTS) { 3410 (*op)(data, datalen / 3411 sizeof(struct filedescent *)); 3412 } 3413 3414 if (CMSG_SPACE(datalen) < clen) { 3415 clen -= CMSG_SPACE(datalen); 3416 cm = (struct cmsghdr *) 3417 ((caddr_t)cm + CMSG_SPACE(datalen)); 3418 } else { 3419 clen = 0; 3420 cm = NULL; 3421 } 3422 } 3423 } 3424 m0 = m0->m_nextpkt; 3425 } 3426 } 3427 3428 /* 3429 * A helper function called by VFS before socket-type vnode reclamation. 3430 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3431 * use count. 3432 */ 3433 void 3434 vfs_unp_reclaim(struct vnode *vp) 3435 { 3436 struct unpcb *unp; 3437 int active; 3438 struct mtx *vplock; 3439 3440 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3441 KASSERT(vp->v_type == VSOCK, 3442 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3443 3444 active = 0; 3445 vplock = mtx_pool_find(mtxpool_sleep, vp); 3446 mtx_lock(vplock); 3447 VOP_UNP_CONNECT(vp, &unp); 3448 if (unp == NULL) 3449 goto done; 3450 UNP_PCB_LOCK(unp); 3451 if (unp->unp_vnode == vp) { 3452 VOP_UNP_DETACH(vp); 3453 unp->unp_vnode = NULL; 3454 active = 1; 3455 } 3456 UNP_PCB_UNLOCK(unp); 3457 done: 3458 mtx_unlock(vplock); 3459 if (active) 3460 vunref(vp); 3461 } 3462 3463 #ifdef DDB 3464 static void 3465 db_print_indent(int indent) 3466 { 3467 int i; 3468 3469 for (i = 0; i < indent; i++) 3470 db_printf(" "); 3471 } 3472 3473 static void 3474 db_print_unpflags(int unp_flags) 3475 { 3476 int comma; 3477 3478 comma = 0; 3479 if (unp_flags & UNP_HAVEPC) { 3480 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3481 comma = 1; 3482 } 3483 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3484 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3485 comma = 1; 3486 } 3487 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3488 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3489 comma = 1; 3490 } 3491 if (unp_flags & UNP_CONNWAIT) { 3492 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3493 comma = 1; 3494 } 3495 if (unp_flags & UNP_CONNECTING) { 3496 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3497 comma = 1; 3498 } 3499 if (unp_flags & UNP_BINDING) { 3500 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3501 comma = 1; 3502 } 3503 } 3504 3505 static void 3506 db_print_xucred(int indent, struct xucred *xu) 3507 { 3508 int comma, i; 3509 3510 db_print_indent(indent); 3511 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3512 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3513 db_print_indent(indent); 3514 db_printf("cr_groups: "); 3515 comma = 0; 3516 for (i = 0; i < xu->cr_ngroups; i++) { 3517 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3518 comma = 1; 3519 } 3520 db_printf("\n"); 3521 } 3522 3523 static void 3524 db_print_unprefs(int indent, struct unp_head *uh) 3525 { 3526 struct unpcb *unp; 3527 int counter; 3528 3529 counter = 0; 3530 LIST_FOREACH(unp, uh, unp_reflink) { 3531 if (counter % 4 == 0) 3532 db_print_indent(indent); 3533 db_printf("%p ", unp); 3534 if (counter % 4 == 3) 3535 db_printf("\n"); 3536 counter++; 3537 } 3538 if (counter != 0 && counter % 4 != 0) 3539 db_printf("\n"); 3540 } 3541 3542 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3543 { 3544 struct unpcb *unp; 3545 3546 if (!have_addr) { 3547 db_printf("usage: show unpcb <addr>\n"); 3548 return; 3549 } 3550 unp = (struct unpcb *)addr; 3551 3552 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3553 unp->unp_vnode); 3554 3555 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3556 unp->unp_conn); 3557 3558 db_printf("unp_refs:\n"); 3559 db_print_unprefs(2, &unp->unp_refs); 3560 3561 /* XXXRW: Would be nice to print the full address, if any. */ 3562 db_printf("unp_addr: %p\n", unp->unp_addr); 3563 3564 db_printf("unp_gencnt: %llu\n", 3565 (unsigned long long)unp->unp_gencnt); 3566 3567 db_printf("unp_flags: %x (", unp->unp_flags); 3568 db_print_unpflags(unp->unp_flags); 3569 db_printf(")\n"); 3570 3571 db_printf("unp_peercred:\n"); 3572 db_print_xucred(2, &unp->unp_peercred); 3573 3574 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3575 } 3576 #endif 3577