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; 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 (m != NULL && flagsp != NULL) 1463 *flagsp |= MSG_TRUNC; 1464 1465 return (0); 1466 } 1467 1468 /* 1469 * PF_UNIX/SOCK_DGRAM receive 1470 */ 1471 static int 1472 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1473 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1474 { 1475 struct sockbuf *sb = NULL; 1476 struct mbuf *m; 1477 int flags, error; 1478 ssize_t len; 1479 bool nonblock; 1480 1481 MPASS(mp0 == NULL); 1482 1483 if (psa != NULL) 1484 *psa = NULL; 1485 if (controlp != NULL) 1486 *controlp = NULL; 1487 1488 flags = flagsp != NULL ? *flagsp : 0; 1489 nonblock = (so->so_state & SS_NBIO) || 1490 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1491 1492 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1493 if (__predict_false(error)) 1494 return (error); 1495 1496 /* 1497 * Loop blocking while waiting for a datagram. Prioritize connected 1498 * peers over unconnected sends. Set sb to selected socket buffer 1499 * containing an mbuf on exit from the wait loop. A datagram that 1500 * had already been peeked at has top priority. 1501 */ 1502 SOCK_RECVBUF_LOCK(so); 1503 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1504 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1505 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1506 if (so->so_error) { 1507 error = so->so_error; 1508 so->so_error = 0; 1509 SOCK_RECVBUF_UNLOCK(so); 1510 SOCK_IO_RECV_UNLOCK(so); 1511 return (error); 1512 } 1513 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1514 uio->uio_resid == 0) { 1515 SOCK_RECVBUF_UNLOCK(so); 1516 SOCK_IO_RECV_UNLOCK(so); 1517 return (0); 1518 } 1519 if (nonblock) { 1520 SOCK_RECVBUF_UNLOCK(so); 1521 SOCK_IO_RECV_UNLOCK(so); 1522 return (EWOULDBLOCK); 1523 } 1524 error = sbwait(so, SO_RCV); 1525 if (error) { 1526 SOCK_RECVBUF_UNLOCK(so); 1527 SOCK_IO_RECV_UNLOCK(so); 1528 return (error); 1529 } 1530 } 1531 1532 if (sb == NULL) 1533 sb = &so->so_rcv; 1534 else if (m == NULL) 1535 m = STAILQ_FIRST(&sb->uxdg_mb); 1536 else 1537 MPASS(m == so->so_rcv.uxdg_peeked); 1538 1539 MPASS(sb->uxdg_cc > 0); 1540 M_ASSERTPKTHDR(m); 1541 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1542 1543 if (uio->uio_td) 1544 uio->uio_td->td_ru.ru_msgrcv++; 1545 1546 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1547 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1548 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1549 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1550 } else 1551 so->so_rcv.uxdg_peeked = NULL; 1552 1553 sb->uxdg_cc -= m->m_pkthdr.len; 1554 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1555 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1556 1557 if (__predict_false(flags & MSG_PEEK)) 1558 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1559 1560 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1561 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1562 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1563 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1564 SOCK_RECVBUF_UNLOCK(so); 1565 1566 if (psa != NULL) 1567 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1568 m = m_free(m); 1569 KASSERT(m, ("%s: no data or control after soname", __func__)); 1570 1571 /* 1572 * Packet to copyout() is now in 'm' and it is disconnected from the 1573 * queue. 1574 * 1575 * Process one or more MT_CONTROL mbufs present before any data mbufs 1576 * in the first mbuf chain on the socket buffer. We call into the 1577 * unp_externalize() to perform externalization (or freeing if 1578 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1579 * without MT_DATA mbufs. 1580 */ 1581 while (m != NULL && m->m_type == MT_CONTROL) { 1582 struct mbuf *cm; 1583 1584 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1585 * it frees whole chain, so we must disconnect the mbuf. 1586 */ 1587 cm = m; m = m->m_next; cm->m_next = NULL; 1588 error = unp_externalize(cm, controlp, flags); 1589 if (error != 0) { 1590 SOCK_IO_RECV_UNLOCK(so); 1591 unp_scan(m, unp_freerights); 1592 m_freem(m); 1593 return (error); 1594 } 1595 if (controlp != NULL) { 1596 while (*controlp != NULL) 1597 controlp = &(*controlp)->m_next; 1598 } 1599 } 1600 KASSERT(m == NULL || m->m_type == MT_DATA, 1601 ("%s: not MT_DATA mbuf %p", __func__, m)); 1602 while (m != NULL && uio->uio_resid > 0) { 1603 len = uio->uio_resid; 1604 if (len > m->m_len) 1605 len = m->m_len; 1606 error = uiomove(mtod(m, char *), (int)len, uio); 1607 if (error) { 1608 SOCK_IO_RECV_UNLOCK(so); 1609 m_freem(m); 1610 return (error); 1611 } 1612 if (len == m->m_len) 1613 m = m_free(m); 1614 else { 1615 m->m_data += len; 1616 m->m_len -= len; 1617 } 1618 } 1619 SOCK_IO_RECV_UNLOCK(so); 1620 1621 if (m != NULL) { 1622 flags |= MSG_TRUNC; 1623 m_freem(m); 1624 } 1625 if (flagsp != NULL) 1626 *flagsp |= flags; 1627 1628 return (0); 1629 } 1630 1631 static bool 1632 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1633 { 1634 struct mbuf *mb, *n; 1635 struct sockbuf *sb; 1636 1637 SOCK_LOCK(so); 1638 if (SOLISTENING(so)) { 1639 SOCK_UNLOCK(so); 1640 return (false); 1641 } 1642 mb = NULL; 1643 sb = &so->so_rcv; 1644 SOCKBUF_LOCK(sb); 1645 if (sb->sb_fnrdy != NULL) { 1646 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1647 if (mb == m) { 1648 *errorp = sbready(sb, m, count); 1649 break; 1650 } 1651 mb = mb->m_next; 1652 if (mb == NULL) { 1653 mb = n; 1654 if (mb != NULL) 1655 n = mb->m_nextpkt; 1656 } 1657 } 1658 } 1659 SOCKBUF_UNLOCK(sb); 1660 SOCK_UNLOCK(so); 1661 return (mb != NULL); 1662 } 1663 1664 static int 1665 uipc_ready(struct socket *so, struct mbuf *m, int count) 1666 { 1667 struct unpcb *unp, *unp2; 1668 struct socket *so2; 1669 int error, i; 1670 1671 unp = sotounpcb(so); 1672 1673 KASSERT(so->so_type == SOCK_STREAM, 1674 ("%s: unexpected socket type for %p", __func__, so)); 1675 1676 UNP_PCB_LOCK(unp); 1677 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1678 UNP_PCB_UNLOCK(unp); 1679 so2 = unp2->unp_socket; 1680 SOCKBUF_LOCK(&so2->so_rcv); 1681 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1682 sorwakeup_locked(so2); 1683 else 1684 SOCKBUF_UNLOCK(&so2->so_rcv); 1685 UNP_PCB_UNLOCK(unp2); 1686 return (error); 1687 } 1688 UNP_PCB_UNLOCK(unp); 1689 1690 /* 1691 * The receiving socket has been disconnected, but may still be valid. 1692 * In this case, the now-ready mbufs are still present in its socket 1693 * buffer, so perform an exhaustive search before giving up and freeing 1694 * the mbufs. 1695 */ 1696 UNP_LINK_RLOCK(); 1697 LIST_FOREACH(unp, &unp_shead, unp_link) { 1698 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1699 break; 1700 } 1701 UNP_LINK_RUNLOCK(); 1702 1703 if (unp == NULL) { 1704 for (i = 0; i < count; i++) 1705 m = m_free(m); 1706 error = ECONNRESET; 1707 } 1708 return (error); 1709 } 1710 1711 static int 1712 uipc_sense(struct socket *so, struct stat *sb) 1713 { 1714 struct unpcb *unp; 1715 1716 unp = sotounpcb(so); 1717 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1718 1719 sb->st_blksize = so->so_snd.sb_hiwat; 1720 sb->st_dev = NODEV; 1721 sb->st_ino = unp->unp_ino; 1722 return (0); 1723 } 1724 1725 static int 1726 uipc_shutdown(struct socket *so) 1727 { 1728 struct unpcb *unp; 1729 1730 unp = sotounpcb(so); 1731 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1732 1733 UNP_PCB_LOCK(unp); 1734 socantsendmore(so); 1735 unp_shutdown(unp); 1736 UNP_PCB_UNLOCK(unp); 1737 return (0); 1738 } 1739 1740 static int 1741 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1742 { 1743 struct unpcb *unp; 1744 const struct sockaddr *sa; 1745 1746 unp = sotounpcb(so); 1747 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1748 1749 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1750 UNP_PCB_LOCK(unp); 1751 if (unp->unp_addr != NULL) 1752 sa = (struct sockaddr *) unp->unp_addr; 1753 else 1754 sa = &sun_noname; 1755 bcopy(sa, *nam, sa->sa_len); 1756 UNP_PCB_UNLOCK(unp); 1757 return (0); 1758 } 1759 1760 static struct pr_usrreqs uipc_usrreqs_dgram = { 1761 .pru_abort = uipc_abort, 1762 .pru_accept = uipc_accept, 1763 .pru_attach = uipc_attach, 1764 .pru_bind = uipc_bind, 1765 .pru_bindat = uipc_bindat, 1766 .pru_connect = uipc_connect, 1767 .pru_connectat = uipc_connectat, 1768 .pru_connect2 = uipc_connect2, 1769 .pru_detach = uipc_detach, 1770 .pru_disconnect = uipc_disconnect, 1771 .pru_peeraddr = uipc_peeraddr, 1772 .pru_sosend = uipc_sosend_dgram, 1773 .pru_sense = uipc_sense, 1774 .pru_shutdown = uipc_shutdown, 1775 .pru_sockaddr = uipc_sockaddr, 1776 .pru_soreceive = uipc_soreceive_dgram, 1777 .pru_close = uipc_close, 1778 }; 1779 1780 static struct pr_usrreqs uipc_usrreqs_seqpacket = { 1781 .pru_abort = uipc_abort, 1782 .pru_accept = uipc_accept, 1783 .pru_attach = uipc_attach, 1784 .pru_bind = uipc_bind, 1785 .pru_bindat = uipc_bindat, 1786 .pru_connect = uipc_connect, 1787 .pru_connectat = uipc_connectat, 1788 .pru_connect2 = uipc_connect2, 1789 .pru_detach = uipc_detach, 1790 .pru_disconnect = uipc_disconnect, 1791 .pru_listen = uipc_listen, 1792 .pru_peeraddr = uipc_peeraddr, 1793 .pru_rcvd = uipc_rcvd, 1794 .pru_send = uipc_send, 1795 .pru_sense = uipc_sense, 1796 .pru_shutdown = uipc_shutdown, 1797 .pru_sockaddr = uipc_sockaddr, 1798 .pru_soreceive = soreceive_generic, /* XXX: or...? */ 1799 .pru_close = uipc_close, 1800 }; 1801 1802 static struct pr_usrreqs uipc_usrreqs_stream = { 1803 .pru_abort = uipc_abort, 1804 .pru_accept = uipc_accept, 1805 .pru_attach = uipc_attach, 1806 .pru_bind = uipc_bind, 1807 .pru_bindat = uipc_bindat, 1808 .pru_connect = uipc_connect, 1809 .pru_connectat = uipc_connectat, 1810 .pru_connect2 = uipc_connect2, 1811 .pru_detach = uipc_detach, 1812 .pru_disconnect = uipc_disconnect, 1813 .pru_listen = uipc_listen, 1814 .pru_peeraddr = uipc_peeraddr, 1815 .pru_rcvd = uipc_rcvd, 1816 .pru_send = uipc_send, 1817 .pru_ready = uipc_ready, 1818 .pru_sense = uipc_sense, 1819 .pru_shutdown = uipc_shutdown, 1820 .pru_sockaddr = uipc_sockaddr, 1821 .pru_soreceive = soreceive_generic, 1822 .pru_close = uipc_close, 1823 }; 1824 1825 static int 1826 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1827 { 1828 struct unpcb *unp; 1829 struct xucred xu; 1830 int error, optval; 1831 1832 if (sopt->sopt_level != SOL_LOCAL) 1833 return (EINVAL); 1834 1835 unp = sotounpcb(so); 1836 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1837 error = 0; 1838 switch (sopt->sopt_dir) { 1839 case SOPT_GET: 1840 switch (sopt->sopt_name) { 1841 case LOCAL_PEERCRED: 1842 UNP_PCB_LOCK(unp); 1843 if (unp->unp_flags & UNP_HAVEPC) 1844 xu = unp->unp_peercred; 1845 else { 1846 if (so->so_type == SOCK_STREAM) 1847 error = ENOTCONN; 1848 else 1849 error = EINVAL; 1850 } 1851 UNP_PCB_UNLOCK(unp); 1852 if (error == 0) 1853 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1854 break; 1855 1856 case LOCAL_CREDS: 1857 /* Unlocked read. */ 1858 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1859 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1860 break; 1861 1862 case LOCAL_CREDS_PERSISTENT: 1863 /* Unlocked read. */ 1864 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1865 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1866 break; 1867 1868 case LOCAL_CONNWAIT: 1869 /* Unlocked read. */ 1870 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1871 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1872 break; 1873 1874 default: 1875 error = EOPNOTSUPP; 1876 break; 1877 } 1878 break; 1879 1880 case SOPT_SET: 1881 switch (sopt->sopt_name) { 1882 case LOCAL_CREDS: 1883 case LOCAL_CREDS_PERSISTENT: 1884 case LOCAL_CONNWAIT: 1885 error = sooptcopyin(sopt, &optval, sizeof(optval), 1886 sizeof(optval)); 1887 if (error) 1888 break; 1889 1890 #define OPTSET(bit, exclusive) do { \ 1891 UNP_PCB_LOCK(unp); \ 1892 if (optval) { \ 1893 if ((unp->unp_flags & (exclusive)) != 0) { \ 1894 UNP_PCB_UNLOCK(unp); \ 1895 error = EINVAL; \ 1896 break; \ 1897 } \ 1898 unp->unp_flags |= (bit); \ 1899 } else \ 1900 unp->unp_flags &= ~(bit); \ 1901 UNP_PCB_UNLOCK(unp); \ 1902 } while (0) 1903 1904 switch (sopt->sopt_name) { 1905 case LOCAL_CREDS: 1906 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1907 break; 1908 1909 case LOCAL_CREDS_PERSISTENT: 1910 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1911 break; 1912 1913 case LOCAL_CONNWAIT: 1914 OPTSET(UNP_CONNWAIT, 0); 1915 break; 1916 1917 default: 1918 break; 1919 } 1920 break; 1921 #undef OPTSET 1922 default: 1923 error = ENOPROTOOPT; 1924 break; 1925 } 1926 break; 1927 1928 default: 1929 error = EOPNOTSUPP; 1930 break; 1931 } 1932 return (error); 1933 } 1934 1935 static int 1936 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1937 { 1938 1939 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1940 } 1941 1942 static int 1943 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1944 struct thread *td, bool return_locked) 1945 { 1946 struct mtx *vplock; 1947 struct sockaddr_un *soun; 1948 struct vnode *vp; 1949 struct socket *so2; 1950 struct unpcb *unp, *unp2, *unp3; 1951 struct nameidata nd; 1952 char buf[SOCK_MAXADDRLEN]; 1953 struct sockaddr *sa; 1954 cap_rights_t rights; 1955 int error, len; 1956 bool connreq; 1957 1958 if (nam->sa_family != AF_UNIX) 1959 return (EAFNOSUPPORT); 1960 if (nam->sa_len > sizeof(struct sockaddr_un)) 1961 return (EINVAL); 1962 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1963 if (len <= 0) 1964 return (EINVAL); 1965 soun = (struct sockaddr_un *)nam; 1966 bcopy(soun->sun_path, buf, len); 1967 buf[len] = 0; 1968 1969 error = 0; 1970 unp = sotounpcb(so); 1971 UNP_PCB_LOCK(unp); 1972 for (;;) { 1973 /* 1974 * Wait for connection state to stabilize. If a connection 1975 * already exists, give up. For datagram sockets, which permit 1976 * multiple consecutive connect(2) calls, upper layers are 1977 * responsible for disconnecting in advance of a subsequent 1978 * connect(2), but this is not synchronized with PCB connection 1979 * state. 1980 * 1981 * Also make sure that no threads are currently attempting to 1982 * lock the peer socket, to ensure that unp_conn cannot 1983 * transition between two valid sockets while locks are dropped. 1984 */ 1985 if (SOLISTENING(so)) 1986 error = EOPNOTSUPP; 1987 else if (unp->unp_conn != NULL) 1988 error = EISCONN; 1989 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1990 error = EALREADY; 1991 } 1992 if (error != 0) { 1993 UNP_PCB_UNLOCK(unp); 1994 return (error); 1995 } 1996 if (unp->unp_pairbusy > 0) { 1997 unp->unp_flags |= UNP_WAITING; 1998 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1999 continue; 2000 } 2001 break; 2002 } 2003 unp->unp_flags |= UNP_CONNECTING; 2004 UNP_PCB_UNLOCK(unp); 2005 2006 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 2007 if (connreq) 2008 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 2009 else 2010 sa = NULL; 2011 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 2012 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 2013 error = namei(&nd); 2014 if (error) 2015 vp = NULL; 2016 else 2017 vp = nd.ni_vp; 2018 ASSERT_VOP_LOCKED(vp, "unp_connect"); 2019 NDFREE_NOTHING(&nd); 2020 if (error) 2021 goto bad; 2022 2023 if (vp->v_type != VSOCK) { 2024 error = ENOTSOCK; 2025 goto bad; 2026 } 2027 #ifdef MAC 2028 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 2029 if (error) 2030 goto bad; 2031 #endif 2032 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 2033 if (error) 2034 goto bad; 2035 2036 unp = sotounpcb(so); 2037 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 2038 2039 vplock = mtx_pool_find(mtxpool_sleep, vp); 2040 mtx_lock(vplock); 2041 VOP_UNP_CONNECT(vp, &unp2); 2042 if (unp2 == NULL) { 2043 error = ECONNREFUSED; 2044 goto bad2; 2045 } 2046 so2 = unp2->unp_socket; 2047 if (so->so_type != so2->so_type) { 2048 error = EPROTOTYPE; 2049 goto bad2; 2050 } 2051 if (connreq) { 2052 if (SOLISTENING(so2)) { 2053 CURVNET_SET(so2->so_vnet); 2054 so2 = sonewconn(so2, 0); 2055 CURVNET_RESTORE(); 2056 } else 2057 so2 = NULL; 2058 if (so2 == NULL) { 2059 error = ECONNREFUSED; 2060 goto bad2; 2061 } 2062 unp3 = sotounpcb(so2); 2063 unp_pcb_lock_pair(unp2, unp3); 2064 if (unp2->unp_addr != NULL) { 2065 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 2066 unp3->unp_addr = (struct sockaddr_un *) sa; 2067 sa = NULL; 2068 } 2069 2070 unp_copy_peercred(td, unp3, unp, unp2); 2071 2072 UNP_PCB_UNLOCK(unp2); 2073 unp2 = unp3; 2074 2075 /* 2076 * It is safe to block on the PCB lock here since unp2 is 2077 * nascent and cannot be connected to any other sockets. 2078 */ 2079 UNP_PCB_LOCK(unp); 2080 #ifdef MAC 2081 mac_socketpeer_set_from_socket(so, so2); 2082 mac_socketpeer_set_from_socket(so2, so); 2083 #endif 2084 } else { 2085 unp_pcb_lock_pair(unp, unp2); 2086 } 2087 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 2088 sotounpcb(so2) == unp2, 2089 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 2090 unp_connect2(so, so2, PRU_CONNECT); 2091 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2092 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2093 unp->unp_flags &= ~UNP_CONNECTING; 2094 if (!return_locked) 2095 unp_pcb_unlock_pair(unp, unp2); 2096 bad2: 2097 mtx_unlock(vplock); 2098 bad: 2099 if (vp != NULL) { 2100 /* 2101 * If we are returning locked (called via uipc_sosend_dgram()), 2102 * we need to be sure that vput() won't sleep. This is 2103 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2104 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2105 */ 2106 MPASS(!(return_locked && connreq)); 2107 vput(vp); 2108 } 2109 free(sa, M_SONAME); 2110 if (__predict_false(error)) { 2111 UNP_PCB_LOCK(unp); 2112 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2113 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2114 unp->unp_flags &= ~UNP_CONNECTING; 2115 UNP_PCB_UNLOCK(unp); 2116 } 2117 return (error); 2118 } 2119 2120 /* 2121 * Set socket peer credentials at connection time. 2122 * 2123 * The client's PCB credentials are copied from its process structure. The 2124 * server's PCB credentials are copied from the socket on which it called 2125 * listen(2). uipc_listen cached that process's credentials at the time. 2126 */ 2127 void 2128 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2129 struct unpcb *server_unp, struct unpcb *listen_unp) 2130 { 2131 cru2xt(td, &client_unp->unp_peercred); 2132 client_unp->unp_flags |= UNP_HAVEPC; 2133 2134 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2135 sizeof(server_unp->unp_peercred)); 2136 server_unp->unp_flags |= UNP_HAVEPC; 2137 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2138 } 2139 2140 static void 2141 unp_connect2(struct socket *so, struct socket *so2, int req) 2142 { 2143 struct unpcb *unp; 2144 struct unpcb *unp2; 2145 2146 MPASS(so2->so_type == so->so_type); 2147 unp = sotounpcb(so); 2148 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2149 unp2 = sotounpcb(so2); 2150 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2151 2152 UNP_PCB_LOCK_ASSERT(unp); 2153 UNP_PCB_LOCK_ASSERT(unp2); 2154 KASSERT(unp->unp_conn == NULL, 2155 ("%s: socket %p is already connected", __func__, unp)); 2156 2157 unp->unp_conn = unp2; 2158 unp_pcb_hold(unp2); 2159 unp_pcb_hold(unp); 2160 switch (so->so_type) { 2161 case SOCK_DGRAM: 2162 UNP_REF_LIST_LOCK(); 2163 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2164 UNP_REF_LIST_UNLOCK(); 2165 soisconnected(so); 2166 break; 2167 2168 case SOCK_STREAM: 2169 case SOCK_SEQPACKET: 2170 KASSERT(unp2->unp_conn == NULL, 2171 ("%s: socket %p is already connected", __func__, unp2)); 2172 unp2->unp_conn = unp; 2173 if (req == PRU_CONNECT && 2174 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2175 soisconnecting(so); 2176 else 2177 soisconnected(so); 2178 soisconnected(so2); 2179 break; 2180 2181 default: 2182 panic("unp_connect2"); 2183 } 2184 } 2185 2186 static void 2187 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2188 { 2189 struct socket *so, *so2; 2190 struct mbuf *m = NULL; 2191 #ifdef INVARIANTS 2192 struct unpcb *unptmp; 2193 #endif 2194 2195 UNP_PCB_LOCK_ASSERT(unp); 2196 UNP_PCB_LOCK_ASSERT(unp2); 2197 KASSERT(unp->unp_conn == unp2, 2198 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2199 2200 unp->unp_conn = NULL; 2201 so = unp->unp_socket; 2202 so2 = unp2->unp_socket; 2203 switch (unp->unp_socket->so_type) { 2204 case SOCK_DGRAM: 2205 /* 2206 * Remove our send socket buffer from the peer's receive buffer. 2207 * Move the data to the receive buffer only if it is empty. 2208 * This is a protection against a scenario where a peer 2209 * connects, floods and disconnects, effectively blocking 2210 * sendto() from unconnected sockets. 2211 */ 2212 SOCK_RECVBUF_LOCK(so2); 2213 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2214 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2215 uxdg_clist); 2216 if (__predict_true((so2->so_rcv.sb_state & 2217 SBS_CANTRCVMORE) == 0) && 2218 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2219 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2220 &so->so_snd.uxdg_mb); 2221 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2222 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2223 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2224 } else { 2225 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2226 STAILQ_INIT(&so->so_snd.uxdg_mb); 2227 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2228 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2229 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2230 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2231 } 2232 /* Note: so may reconnect. */ 2233 so->so_snd.uxdg_cc = 0; 2234 so->so_snd.uxdg_ctl = 0; 2235 so->so_snd.uxdg_mbcnt = 0; 2236 } 2237 SOCK_RECVBUF_UNLOCK(so2); 2238 UNP_REF_LIST_LOCK(); 2239 #ifdef INVARIANTS 2240 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2241 if (unptmp == unp) 2242 break; 2243 } 2244 KASSERT(unptmp != NULL, 2245 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2246 #endif 2247 LIST_REMOVE(unp, unp_reflink); 2248 UNP_REF_LIST_UNLOCK(); 2249 if (so) { 2250 SOCK_LOCK(so); 2251 so->so_state &= ~SS_ISCONNECTED; 2252 SOCK_UNLOCK(so); 2253 } 2254 break; 2255 2256 case SOCK_STREAM: 2257 case SOCK_SEQPACKET: 2258 if (so) 2259 soisdisconnected(so); 2260 MPASS(unp2->unp_conn == unp); 2261 unp2->unp_conn = NULL; 2262 if (so2) 2263 soisdisconnected(so2); 2264 break; 2265 } 2266 2267 if (unp == unp2) { 2268 unp_pcb_rele_notlast(unp); 2269 if (!unp_pcb_rele(unp)) 2270 UNP_PCB_UNLOCK(unp); 2271 } else { 2272 if (!unp_pcb_rele(unp)) 2273 UNP_PCB_UNLOCK(unp); 2274 if (!unp_pcb_rele(unp2)) 2275 UNP_PCB_UNLOCK(unp2); 2276 } 2277 2278 if (m != NULL) { 2279 unp_scan(m, unp_freerights); 2280 m_freem(m); 2281 } 2282 } 2283 2284 /* 2285 * unp_pcblist() walks the global list of struct unpcb's to generate a 2286 * pointer list, bumping the refcount on each unpcb. It then copies them out 2287 * sequentially, validating the generation number on each to see if it has 2288 * been detached. All of this is necessary because copyout() may sleep on 2289 * disk I/O. 2290 */ 2291 static int 2292 unp_pcblist(SYSCTL_HANDLER_ARGS) 2293 { 2294 struct unpcb *unp, **unp_list; 2295 unp_gen_t gencnt; 2296 struct xunpgen *xug; 2297 struct unp_head *head; 2298 struct xunpcb *xu; 2299 u_int i; 2300 int error, n; 2301 2302 switch ((intptr_t)arg1) { 2303 case SOCK_STREAM: 2304 head = &unp_shead; 2305 break; 2306 2307 case SOCK_DGRAM: 2308 head = &unp_dhead; 2309 break; 2310 2311 case SOCK_SEQPACKET: 2312 head = &unp_sphead; 2313 break; 2314 2315 default: 2316 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2317 } 2318 2319 /* 2320 * The process of preparing the PCB list is too time-consuming and 2321 * resource-intensive to repeat twice on every request. 2322 */ 2323 if (req->oldptr == NULL) { 2324 n = unp_count; 2325 req->oldidx = 2 * (sizeof *xug) 2326 + (n + n/8) * sizeof(struct xunpcb); 2327 return (0); 2328 } 2329 2330 if (req->newptr != NULL) 2331 return (EPERM); 2332 2333 /* 2334 * OK, now we're committed to doing something. 2335 */ 2336 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2337 UNP_LINK_RLOCK(); 2338 gencnt = unp_gencnt; 2339 n = unp_count; 2340 UNP_LINK_RUNLOCK(); 2341 2342 xug->xug_len = sizeof *xug; 2343 xug->xug_count = n; 2344 xug->xug_gen = gencnt; 2345 xug->xug_sogen = so_gencnt; 2346 error = SYSCTL_OUT(req, xug, sizeof *xug); 2347 if (error) { 2348 free(xug, M_TEMP); 2349 return (error); 2350 } 2351 2352 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2353 2354 UNP_LINK_RLOCK(); 2355 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2356 unp = LIST_NEXT(unp, unp_link)) { 2357 UNP_PCB_LOCK(unp); 2358 if (unp->unp_gencnt <= gencnt) { 2359 if (cr_cansee(req->td->td_ucred, 2360 unp->unp_socket->so_cred)) { 2361 UNP_PCB_UNLOCK(unp); 2362 continue; 2363 } 2364 unp_list[i++] = unp; 2365 unp_pcb_hold(unp); 2366 } 2367 UNP_PCB_UNLOCK(unp); 2368 } 2369 UNP_LINK_RUNLOCK(); 2370 n = i; /* In case we lost some during malloc. */ 2371 2372 error = 0; 2373 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2374 for (i = 0; i < n; i++) { 2375 unp = unp_list[i]; 2376 UNP_PCB_LOCK(unp); 2377 if (unp_pcb_rele(unp)) 2378 continue; 2379 2380 if (unp->unp_gencnt <= gencnt) { 2381 xu->xu_len = sizeof *xu; 2382 xu->xu_unpp = (uintptr_t)unp; 2383 /* 2384 * XXX - need more locking here to protect against 2385 * connect/disconnect races for SMP. 2386 */ 2387 if (unp->unp_addr != NULL) 2388 bcopy(unp->unp_addr, &xu->xu_addr, 2389 unp->unp_addr->sun_len); 2390 else 2391 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2392 if (unp->unp_conn != NULL && 2393 unp->unp_conn->unp_addr != NULL) 2394 bcopy(unp->unp_conn->unp_addr, 2395 &xu->xu_caddr, 2396 unp->unp_conn->unp_addr->sun_len); 2397 else 2398 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2399 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2400 xu->unp_conn = (uintptr_t)unp->unp_conn; 2401 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2402 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2403 xu->unp_gencnt = unp->unp_gencnt; 2404 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2405 UNP_PCB_UNLOCK(unp); 2406 error = SYSCTL_OUT(req, xu, sizeof *xu); 2407 } else { 2408 UNP_PCB_UNLOCK(unp); 2409 } 2410 } 2411 free(xu, M_TEMP); 2412 if (!error) { 2413 /* 2414 * Give the user an updated idea of our state. If the 2415 * generation differs from what we told her before, she knows 2416 * that something happened while we were processing this 2417 * request, and it might be necessary to retry. 2418 */ 2419 xug->xug_gen = unp_gencnt; 2420 xug->xug_sogen = so_gencnt; 2421 xug->xug_count = unp_count; 2422 error = SYSCTL_OUT(req, xug, sizeof *xug); 2423 } 2424 free(unp_list, M_TEMP); 2425 free(xug, M_TEMP); 2426 return (error); 2427 } 2428 2429 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2430 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2431 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2432 "List of active local datagram sockets"); 2433 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2434 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2435 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2436 "List of active local stream sockets"); 2437 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2438 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2439 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2440 "List of active local seqpacket sockets"); 2441 2442 static void 2443 unp_shutdown(struct unpcb *unp) 2444 { 2445 struct unpcb *unp2; 2446 struct socket *so; 2447 2448 UNP_PCB_LOCK_ASSERT(unp); 2449 2450 unp2 = unp->unp_conn; 2451 if ((unp->unp_socket->so_type == SOCK_STREAM || 2452 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2453 so = unp2->unp_socket; 2454 if (so != NULL) 2455 socantrcvmore(so); 2456 } 2457 } 2458 2459 static void 2460 unp_drop(struct unpcb *unp) 2461 { 2462 struct socket *so; 2463 struct unpcb *unp2; 2464 2465 /* 2466 * Regardless of whether the socket's peer dropped the connection 2467 * with this socket by aborting or disconnecting, POSIX requires 2468 * that ECONNRESET is returned. 2469 */ 2470 2471 UNP_PCB_LOCK(unp); 2472 so = unp->unp_socket; 2473 if (so) 2474 so->so_error = ECONNRESET; 2475 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2476 /* Last reference dropped in unp_disconnect(). */ 2477 unp_pcb_rele_notlast(unp); 2478 unp_disconnect(unp, unp2); 2479 } else if (!unp_pcb_rele(unp)) { 2480 UNP_PCB_UNLOCK(unp); 2481 } 2482 } 2483 2484 static void 2485 unp_freerights(struct filedescent **fdep, int fdcount) 2486 { 2487 struct file *fp; 2488 int i; 2489 2490 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2491 2492 for (i = 0; i < fdcount; i++) { 2493 fp = fdep[i]->fde_file; 2494 filecaps_free(&fdep[i]->fde_caps); 2495 unp_discard(fp); 2496 } 2497 free(fdep[0], M_FILECAPS); 2498 } 2499 2500 static int 2501 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2502 { 2503 struct thread *td = curthread; /* XXX */ 2504 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2505 int i; 2506 int *fdp; 2507 struct filedesc *fdesc = td->td_proc->p_fd; 2508 struct filedescent **fdep; 2509 void *data; 2510 socklen_t clen = control->m_len, datalen; 2511 int error, newfds; 2512 u_int newlen; 2513 2514 UNP_LINK_UNLOCK_ASSERT(); 2515 2516 error = 0; 2517 if (controlp != NULL) /* controlp == NULL => free control messages */ 2518 *controlp = NULL; 2519 while (cm != NULL) { 2520 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2521 2522 data = CMSG_DATA(cm); 2523 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2524 if (cm->cmsg_level == SOL_SOCKET 2525 && cm->cmsg_type == SCM_RIGHTS) { 2526 newfds = datalen / sizeof(*fdep); 2527 if (newfds == 0) 2528 goto next; 2529 fdep = data; 2530 2531 /* If we're not outputting the descriptors free them. */ 2532 if (error || controlp == NULL) { 2533 unp_freerights(fdep, newfds); 2534 goto next; 2535 } 2536 FILEDESC_XLOCK(fdesc); 2537 2538 /* 2539 * Now change each pointer to an fd in the global 2540 * table to an integer that is the index to the local 2541 * fd table entry that we set up to point to the 2542 * global one we are transferring. 2543 */ 2544 newlen = newfds * sizeof(int); 2545 *controlp = sbcreatecontrol(NULL, newlen, 2546 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2547 2548 fdp = (int *) 2549 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2550 if ((error = fdallocn(td, 0, fdp, newfds))) { 2551 FILEDESC_XUNLOCK(fdesc); 2552 unp_freerights(fdep, newfds); 2553 m_freem(*controlp); 2554 *controlp = NULL; 2555 goto next; 2556 } 2557 for (i = 0; i < newfds; i++, fdp++) { 2558 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2559 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2560 &fdep[i]->fde_caps); 2561 unp_externalize_fp(fdep[i]->fde_file); 2562 } 2563 2564 /* 2565 * The new type indicates that the mbuf data refers to 2566 * kernel resources that may need to be released before 2567 * the mbuf is freed. 2568 */ 2569 m_chtype(*controlp, MT_EXTCONTROL); 2570 FILEDESC_XUNLOCK(fdesc); 2571 free(fdep[0], M_FILECAPS); 2572 } else { 2573 /* We can just copy anything else across. */ 2574 if (error || controlp == NULL) 2575 goto next; 2576 *controlp = sbcreatecontrol(NULL, datalen, 2577 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2578 bcopy(data, 2579 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2580 datalen); 2581 } 2582 controlp = &(*controlp)->m_next; 2583 2584 next: 2585 if (CMSG_SPACE(datalen) < clen) { 2586 clen -= CMSG_SPACE(datalen); 2587 cm = (struct cmsghdr *) 2588 ((caddr_t)cm + CMSG_SPACE(datalen)); 2589 } else { 2590 clen = 0; 2591 cm = NULL; 2592 } 2593 } 2594 2595 m_freem(control); 2596 return (error); 2597 } 2598 2599 static void 2600 unp_zone_change(void *tag) 2601 { 2602 2603 uma_zone_set_max(unp_zone, maxsockets); 2604 } 2605 2606 #ifdef INVARIANTS 2607 static void 2608 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2609 { 2610 struct unpcb *unp; 2611 2612 unp = mem; 2613 2614 KASSERT(LIST_EMPTY(&unp->unp_refs), 2615 ("%s: unpcb %p has lingering refs", __func__, unp)); 2616 KASSERT(unp->unp_socket == NULL, 2617 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2618 KASSERT(unp->unp_vnode == NULL, 2619 ("%s: unpcb %p has vnode references", __func__, unp)); 2620 KASSERT(unp->unp_conn == NULL, 2621 ("%s: unpcb %p is still connected", __func__, unp)); 2622 KASSERT(unp->unp_addr == NULL, 2623 ("%s: unpcb %p has leaked addr", __func__, unp)); 2624 } 2625 #endif 2626 2627 static void 2628 unp_init(void *arg __unused) 2629 { 2630 uma_dtor dtor; 2631 2632 #ifdef INVARIANTS 2633 dtor = unp_zdtor; 2634 #else 2635 dtor = NULL; 2636 #endif 2637 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2638 NULL, NULL, UMA_ALIGN_CACHE, 0); 2639 uma_zone_set_max(unp_zone, maxsockets); 2640 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2641 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2642 NULL, EVENTHANDLER_PRI_ANY); 2643 LIST_INIT(&unp_dhead); 2644 LIST_INIT(&unp_shead); 2645 LIST_INIT(&unp_sphead); 2646 SLIST_INIT(&unp_defers); 2647 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2648 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2649 UNP_LINK_LOCK_INIT(); 2650 UNP_DEFERRED_LOCK_INIT(); 2651 } 2652 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2653 2654 static void 2655 unp_internalize_cleanup_rights(struct mbuf *control) 2656 { 2657 struct cmsghdr *cp; 2658 struct mbuf *m; 2659 void *data; 2660 socklen_t datalen; 2661 2662 for (m = control; m != NULL; m = m->m_next) { 2663 cp = mtod(m, struct cmsghdr *); 2664 if (cp->cmsg_level != SOL_SOCKET || 2665 cp->cmsg_type != SCM_RIGHTS) 2666 continue; 2667 data = CMSG_DATA(cp); 2668 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2669 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2670 } 2671 } 2672 2673 static int 2674 unp_internalize(struct mbuf **controlp, struct thread *td, 2675 struct mbuf **clast, u_int *space, u_int *mbcnt) 2676 { 2677 struct mbuf *control, **initial_controlp; 2678 struct proc *p; 2679 struct filedesc *fdesc; 2680 struct bintime *bt; 2681 struct cmsghdr *cm; 2682 struct cmsgcred *cmcred; 2683 struct filedescent *fde, **fdep, *fdev; 2684 struct file *fp; 2685 struct timeval *tv; 2686 struct timespec *ts; 2687 void *data; 2688 socklen_t clen, datalen; 2689 int i, j, error, *fdp, oldfds; 2690 u_int newlen; 2691 2692 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2693 UNP_LINK_UNLOCK_ASSERT(); 2694 2695 p = td->td_proc; 2696 fdesc = p->p_fd; 2697 error = 0; 2698 control = *controlp; 2699 *controlp = NULL; 2700 initial_controlp = controlp; 2701 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2702 data = CMSG_DATA(cm); 2703 2704 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2705 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2706 (char *)cm + cm->cmsg_len >= (char *)data; 2707 2708 clen -= min(CMSG_SPACE(datalen), clen), 2709 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2710 data = CMSG_DATA(cm)) { 2711 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2712 switch (cm->cmsg_type) { 2713 case SCM_CREDS: 2714 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2715 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2716 cmcred = (struct cmsgcred *) 2717 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2718 cmcred->cmcred_pid = p->p_pid; 2719 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2720 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2721 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2722 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2723 CMGROUP_MAX); 2724 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2725 cmcred->cmcred_groups[i] = 2726 td->td_ucred->cr_groups[i]; 2727 break; 2728 2729 case SCM_RIGHTS: 2730 oldfds = datalen / sizeof (int); 2731 if (oldfds == 0) 2732 continue; 2733 /* On some machines sizeof pointer is bigger than 2734 * sizeof int, so we need to check if data fits into 2735 * single mbuf. We could allocate several mbufs, and 2736 * unp_externalize() should even properly handle that. 2737 * But it is not worth to complicate the code for an 2738 * insane scenario of passing over 200 file descriptors 2739 * at once. 2740 */ 2741 newlen = oldfds * sizeof(fdep[0]); 2742 if (CMSG_SPACE(newlen) > MCLBYTES) { 2743 error = EMSGSIZE; 2744 goto out; 2745 } 2746 /* 2747 * Check that all the FDs passed in refer to legal 2748 * files. If not, reject the entire operation. 2749 */ 2750 fdp = data; 2751 FILEDESC_SLOCK(fdesc); 2752 for (i = 0; i < oldfds; i++, fdp++) { 2753 fp = fget_noref(fdesc, *fdp); 2754 if (fp == NULL) { 2755 FILEDESC_SUNLOCK(fdesc); 2756 error = EBADF; 2757 goto out; 2758 } 2759 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2760 FILEDESC_SUNLOCK(fdesc); 2761 error = EOPNOTSUPP; 2762 goto out; 2763 } 2764 } 2765 2766 /* 2767 * Now replace the integer FDs with pointers to the 2768 * file structure and capability rights. 2769 */ 2770 *controlp = sbcreatecontrol(NULL, newlen, 2771 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2772 fdp = data; 2773 for (i = 0; i < oldfds; i++, fdp++) { 2774 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2775 fdp = data; 2776 for (j = 0; j < i; j++, fdp++) { 2777 fdrop(fdesc->fd_ofiles[*fdp]. 2778 fde_file, td); 2779 } 2780 FILEDESC_SUNLOCK(fdesc); 2781 error = EBADF; 2782 goto out; 2783 } 2784 } 2785 fdp = data; 2786 fdep = (struct filedescent **) 2787 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2788 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2789 M_WAITOK); 2790 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2791 fde = &fdesc->fd_ofiles[*fdp]; 2792 fdep[i] = fdev; 2793 fdep[i]->fde_file = fde->fde_file; 2794 filecaps_copy(&fde->fde_caps, 2795 &fdep[i]->fde_caps, true); 2796 unp_internalize_fp(fdep[i]->fde_file); 2797 } 2798 FILEDESC_SUNLOCK(fdesc); 2799 break; 2800 2801 case SCM_TIMESTAMP: 2802 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2803 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2804 tv = (struct timeval *) 2805 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2806 microtime(tv); 2807 break; 2808 2809 case SCM_BINTIME: 2810 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2811 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2812 bt = (struct bintime *) 2813 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2814 bintime(bt); 2815 break; 2816 2817 case SCM_REALTIME: 2818 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2819 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2820 ts = (struct timespec *) 2821 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2822 nanotime(ts); 2823 break; 2824 2825 case SCM_MONOTONIC: 2826 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2827 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2828 ts = (struct timespec *) 2829 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2830 nanouptime(ts); 2831 break; 2832 2833 default: 2834 error = EINVAL; 2835 goto out; 2836 } 2837 2838 if (space != NULL) { 2839 *space += (*controlp)->m_len; 2840 *mbcnt += MSIZE; 2841 if ((*controlp)->m_flags & M_EXT) 2842 *mbcnt += (*controlp)->m_ext.ext_size; 2843 *clast = *controlp; 2844 } 2845 controlp = &(*controlp)->m_next; 2846 } 2847 if (clen > 0) 2848 error = EINVAL; 2849 2850 out: 2851 if (error != 0 && initial_controlp != NULL) 2852 unp_internalize_cleanup_rights(*initial_controlp); 2853 m_freem(control); 2854 return (error); 2855 } 2856 2857 static struct mbuf * 2858 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2859 struct mbuf **clast, u_int *space, u_int *mbcnt) 2860 { 2861 struct mbuf *m, *n, *n_prev; 2862 const struct cmsghdr *cm; 2863 int ngroups, i, cmsgtype; 2864 size_t ctrlsz; 2865 2866 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2867 if (mode & UNP_WANTCRED_ALWAYS) { 2868 ctrlsz = SOCKCRED2SIZE(ngroups); 2869 cmsgtype = SCM_CREDS2; 2870 } else { 2871 ctrlsz = SOCKCREDSIZE(ngroups); 2872 cmsgtype = SCM_CREDS; 2873 } 2874 2875 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2876 if (m == NULL) 2877 return (control); 2878 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2879 2880 if (mode & UNP_WANTCRED_ALWAYS) { 2881 struct sockcred2 *sc; 2882 2883 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2884 sc->sc_version = 0; 2885 sc->sc_pid = td->td_proc->p_pid; 2886 sc->sc_uid = td->td_ucred->cr_ruid; 2887 sc->sc_euid = td->td_ucred->cr_uid; 2888 sc->sc_gid = td->td_ucred->cr_rgid; 2889 sc->sc_egid = td->td_ucred->cr_gid; 2890 sc->sc_ngroups = ngroups; 2891 for (i = 0; i < sc->sc_ngroups; i++) 2892 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2893 } else { 2894 struct sockcred *sc; 2895 2896 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2897 sc->sc_uid = td->td_ucred->cr_ruid; 2898 sc->sc_euid = td->td_ucred->cr_uid; 2899 sc->sc_gid = td->td_ucred->cr_rgid; 2900 sc->sc_egid = td->td_ucred->cr_gid; 2901 sc->sc_ngroups = ngroups; 2902 for (i = 0; i < sc->sc_ngroups; i++) 2903 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2904 } 2905 2906 /* 2907 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2908 * created SCM_CREDS control message (struct sockcred) has another 2909 * format. 2910 */ 2911 if (control != NULL && cmsgtype == SCM_CREDS) 2912 for (n = control, n_prev = NULL; n != NULL;) { 2913 cm = mtod(n, struct cmsghdr *); 2914 if (cm->cmsg_level == SOL_SOCKET && 2915 cm->cmsg_type == SCM_CREDS) { 2916 if (n_prev == NULL) 2917 control = n->m_next; 2918 else 2919 n_prev->m_next = n->m_next; 2920 if (space != NULL) { 2921 MPASS(*space >= n->m_len); 2922 *space -= n->m_len; 2923 MPASS(*mbcnt >= MSIZE); 2924 *mbcnt -= MSIZE; 2925 if (n->m_flags & M_EXT) { 2926 MPASS(*mbcnt >= 2927 n->m_ext.ext_size); 2928 *mbcnt -= n->m_ext.ext_size; 2929 } 2930 MPASS(clast); 2931 if (*clast == n) { 2932 MPASS(n->m_next == NULL); 2933 if (n_prev == NULL) 2934 *clast = m; 2935 else 2936 *clast = n_prev; 2937 } 2938 } 2939 n = m_free(n); 2940 } else { 2941 n_prev = n; 2942 n = n->m_next; 2943 } 2944 } 2945 2946 /* Prepend it to the head. */ 2947 m->m_next = control; 2948 if (space != NULL) { 2949 *space += m->m_len; 2950 *mbcnt += MSIZE; 2951 if (control == NULL) 2952 *clast = m; 2953 } 2954 return (m); 2955 } 2956 2957 static struct unpcb * 2958 fptounp(struct file *fp) 2959 { 2960 struct socket *so; 2961 2962 if (fp->f_type != DTYPE_SOCKET) 2963 return (NULL); 2964 if ((so = fp->f_data) == NULL) 2965 return (NULL); 2966 if (so->so_proto->pr_domain != &localdomain) 2967 return (NULL); 2968 return sotounpcb(so); 2969 } 2970 2971 static void 2972 unp_discard(struct file *fp) 2973 { 2974 struct unp_defer *dr; 2975 2976 if (unp_externalize_fp(fp)) { 2977 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2978 dr->ud_fp = fp; 2979 UNP_DEFERRED_LOCK(); 2980 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2981 UNP_DEFERRED_UNLOCK(); 2982 atomic_add_int(&unp_defers_count, 1); 2983 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2984 } else 2985 closef_nothread(fp); 2986 } 2987 2988 static void 2989 unp_process_defers(void *arg __unused, int pending) 2990 { 2991 struct unp_defer *dr; 2992 SLIST_HEAD(, unp_defer) drl; 2993 int count; 2994 2995 SLIST_INIT(&drl); 2996 for (;;) { 2997 UNP_DEFERRED_LOCK(); 2998 if (SLIST_FIRST(&unp_defers) == NULL) { 2999 UNP_DEFERRED_UNLOCK(); 3000 break; 3001 } 3002 SLIST_SWAP(&unp_defers, &drl, unp_defer); 3003 UNP_DEFERRED_UNLOCK(); 3004 count = 0; 3005 while ((dr = SLIST_FIRST(&drl)) != NULL) { 3006 SLIST_REMOVE_HEAD(&drl, ud_link); 3007 closef_nothread(dr->ud_fp); 3008 free(dr, M_TEMP); 3009 count++; 3010 } 3011 atomic_add_int(&unp_defers_count, -count); 3012 } 3013 } 3014 3015 static void 3016 unp_internalize_fp(struct file *fp) 3017 { 3018 struct unpcb *unp; 3019 3020 UNP_LINK_WLOCK(); 3021 if ((unp = fptounp(fp)) != NULL) { 3022 unp->unp_file = fp; 3023 unp->unp_msgcount++; 3024 } 3025 unp_rights++; 3026 UNP_LINK_WUNLOCK(); 3027 } 3028 3029 static int 3030 unp_externalize_fp(struct file *fp) 3031 { 3032 struct unpcb *unp; 3033 int ret; 3034 3035 UNP_LINK_WLOCK(); 3036 if ((unp = fptounp(fp)) != NULL) { 3037 unp->unp_msgcount--; 3038 ret = 1; 3039 } else 3040 ret = 0; 3041 unp_rights--; 3042 UNP_LINK_WUNLOCK(); 3043 return (ret); 3044 } 3045 3046 /* 3047 * unp_defer indicates whether additional work has been defered for a future 3048 * pass through unp_gc(). It is thread local and does not require explicit 3049 * synchronization. 3050 */ 3051 static int unp_marked; 3052 3053 static void 3054 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 3055 { 3056 struct unpcb *unp; 3057 struct file *fp; 3058 int i; 3059 3060 /* 3061 * This function can only be called from the gc task. 3062 */ 3063 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 3064 ("%s: not on gc callout", __func__)); 3065 UNP_LINK_LOCK_ASSERT(); 3066 3067 for (i = 0; i < fdcount; i++) { 3068 fp = fdep[i]->fde_file; 3069 if ((unp = fptounp(fp)) == NULL) 3070 continue; 3071 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3072 continue; 3073 unp->unp_gcrefs--; 3074 } 3075 } 3076 3077 static void 3078 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 3079 { 3080 struct unpcb *unp; 3081 struct file *fp; 3082 int i; 3083 3084 /* 3085 * This function can only be called from the gc task. 3086 */ 3087 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 3088 ("%s: not on gc callout", __func__)); 3089 UNP_LINK_LOCK_ASSERT(); 3090 3091 for (i = 0; i < fdcount; i++) { 3092 fp = fdep[i]->fde_file; 3093 if ((unp = fptounp(fp)) == NULL) 3094 continue; 3095 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3096 continue; 3097 unp->unp_gcrefs++; 3098 unp_marked++; 3099 } 3100 } 3101 3102 static void 3103 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3104 { 3105 struct sockbuf *sb; 3106 3107 SOCK_LOCK_ASSERT(so); 3108 3109 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3110 return; 3111 3112 SOCK_RECVBUF_LOCK(so); 3113 switch (so->so_type) { 3114 case SOCK_DGRAM: 3115 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3116 unp_scan(so->so_rcv.uxdg_peeked, op); 3117 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3118 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3119 break; 3120 case SOCK_STREAM: 3121 case SOCK_SEQPACKET: 3122 unp_scan(so->so_rcv.sb_mb, op); 3123 break; 3124 } 3125 SOCK_RECVBUF_UNLOCK(so); 3126 } 3127 3128 static void 3129 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3130 { 3131 struct socket *so, *soa; 3132 3133 so = unp->unp_socket; 3134 SOCK_LOCK(so); 3135 if (SOLISTENING(so)) { 3136 /* 3137 * Mark all sockets in our accept queue. 3138 */ 3139 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3140 unp_scan_socket(soa, op); 3141 } else { 3142 /* 3143 * Mark all sockets we reference with RIGHTS. 3144 */ 3145 unp_scan_socket(so, op); 3146 } 3147 SOCK_UNLOCK(so); 3148 } 3149 3150 static int unp_recycled; 3151 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3152 "Number of unreachable sockets claimed by the garbage collector."); 3153 3154 static int unp_taskcount; 3155 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3156 "Number of times the garbage collector has run."); 3157 3158 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3159 "Number of active local sockets."); 3160 3161 static void 3162 unp_gc(__unused void *arg, int pending) 3163 { 3164 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3165 NULL }; 3166 struct unp_head **head; 3167 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3168 struct file *f, **unref; 3169 struct unpcb *unp, *unptmp; 3170 int i, total, unp_unreachable; 3171 3172 LIST_INIT(&unp_deadhead); 3173 unp_taskcount++; 3174 UNP_LINK_RLOCK(); 3175 /* 3176 * First determine which sockets may be in cycles. 3177 */ 3178 unp_unreachable = 0; 3179 3180 for (head = heads; *head != NULL; head++) 3181 LIST_FOREACH(unp, *head, unp_link) { 3182 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3183 ("%s: unp %p has unexpected gc flags 0x%x", 3184 __func__, unp, (unsigned int)unp->unp_gcflag)); 3185 3186 f = unp->unp_file; 3187 3188 /* 3189 * Check for an unreachable socket potentially in a 3190 * cycle. It must be in a queue as indicated by 3191 * msgcount, and this must equal the file reference 3192 * count. Note that when msgcount is 0 the file is 3193 * NULL. 3194 */ 3195 if (f != NULL && unp->unp_msgcount != 0 && 3196 refcount_load(&f->f_count) == unp->unp_msgcount) { 3197 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3198 unp->unp_gcflag |= UNPGC_DEAD; 3199 unp->unp_gcrefs = unp->unp_msgcount; 3200 unp_unreachable++; 3201 } 3202 } 3203 3204 /* 3205 * Scan all sockets previously marked as potentially being in a cycle 3206 * and remove the references each socket holds on any UNPGC_DEAD 3207 * sockets in its queue. After this step, all remaining references on 3208 * sockets marked UNPGC_DEAD should not be part of any cycle. 3209 */ 3210 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3211 unp_gc_scan(unp, unp_remove_dead_ref); 3212 3213 /* 3214 * If a socket still has a non-negative refcount, it cannot be in a 3215 * cycle. In this case increment refcount of all children iteratively. 3216 * Stop the scan once we do a complete loop without discovering 3217 * a new reachable socket. 3218 */ 3219 do { 3220 unp_marked = 0; 3221 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3222 if (unp->unp_gcrefs > 0) { 3223 unp->unp_gcflag &= ~UNPGC_DEAD; 3224 LIST_REMOVE(unp, unp_dead); 3225 KASSERT(unp_unreachable > 0, 3226 ("%s: unp_unreachable underflow.", 3227 __func__)); 3228 unp_unreachable--; 3229 unp_gc_scan(unp, unp_restore_undead_ref); 3230 } 3231 } while (unp_marked); 3232 3233 UNP_LINK_RUNLOCK(); 3234 3235 if (unp_unreachable == 0) 3236 return; 3237 3238 /* 3239 * Allocate space for a local array of dead unpcbs. 3240 * TODO: can this path be simplified by instead using the local 3241 * dead list at unp_deadhead, after taking out references 3242 * on the file object and/or unpcb and dropping the link lock? 3243 */ 3244 unref = malloc(unp_unreachable * sizeof(struct file *), 3245 M_TEMP, M_WAITOK); 3246 3247 /* 3248 * Iterate looking for sockets which have been specifically marked 3249 * as unreachable and store them locally. 3250 */ 3251 UNP_LINK_RLOCK(); 3252 total = 0; 3253 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3254 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3255 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3256 unp->unp_gcflag &= ~UNPGC_DEAD; 3257 f = unp->unp_file; 3258 if (unp->unp_msgcount == 0 || f == NULL || 3259 refcount_load(&f->f_count) != unp->unp_msgcount || 3260 !fhold(f)) 3261 continue; 3262 unref[total++] = f; 3263 KASSERT(total <= unp_unreachable, 3264 ("%s: incorrect unreachable count.", __func__)); 3265 } 3266 UNP_LINK_RUNLOCK(); 3267 3268 /* 3269 * Now flush all sockets, free'ing rights. This will free the 3270 * struct files associated with these sockets but leave each socket 3271 * with one remaining ref. 3272 */ 3273 for (i = 0; i < total; i++) { 3274 struct socket *so; 3275 3276 so = unref[i]->f_data; 3277 CURVNET_SET(so->so_vnet); 3278 sorflush(so); 3279 CURVNET_RESTORE(); 3280 } 3281 3282 /* 3283 * And finally release the sockets so they can be reclaimed. 3284 */ 3285 for (i = 0; i < total; i++) 3286 fdrop(unref[i], NULL); 3287 unp_recycled += total; 3288 free(unref, M_TEMP); 3289 } 3290 3291 /* 3292 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3293 */ 3294 static void 3295 unp_dispose(struct socket *so) 3296 { 3297 struct sockbuf *sb; 3298 struct unpcb *unp; 3299 struct mbuf *m; 3300 3301 MPASS(!SOLISTENING(so)); 3302 3303 unp = sotounpcb(so); 3304 UNP_LINK_WLOCK(); 3305 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3306 UNP_LINK_WUNLOCK(); 3307 3308 /* 3309 * Grab our special mbufs before calling sbrelease(). 3310 */ 3311 SOCK_RECVBUF_LOCK(so); 3312 switch (so->so_type) { 3313 case SOCK_DGRAM: 3314 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3315 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3316 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3317 /* Note: socket of sb may reconnect. */ 3318 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3319 } 3320 sb = &so->so_rcv; 3321 if (sb->uxdg_peeked != NULL) { 3322 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3323 m_stailqpkt); 3324 sb->uxdg_peeked = NULL; 3325 } 3326 m = STAILQ_FIRST(&sb->uxdg_mb); 3327 STAILQ_INIT(&sb->uxdg_mb); 3328 /* XXX: our shortened sbrelease() */ 3329 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3330 RLIM_INFINITY); 3331 /* 3332 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3333 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3334 * data to the socket. 3335 * We are now in dom_dispose and it could be a call from 3336 * soshutdown() or from the final sofree(). The sofree() case 3337 * is simple as it guarantees that no more sends will happen, 3338 * however we can race with unp_disconnect() from our peer. 3339 * The shutdown(2) case is more exotic. It would call into 3340 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3341 * possible if we did connect(2) on this socket and we also 3342 * had it bound with bind(2) and receive connections from other 3343 * sockets. Because soshutdown() violates POSIX (see comment 3344 * there) we will end up here shutting down our receive side. 3345 * Of course this will have affect not only on the peer we 3346 * connect(2)ed to, but also on all of the peers who had 3347 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3348 */ 3349 sb->sb_state |= SBS_CANTRCVMORE; 3350 break; 3351 case SOCK_STREAM: 3352 case SOCK_SEQPACKET: 3353 sb = &so->so_rcv; 3354 m = sbcut_locked(sb, sb->sb_ccc); 3355 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3356 ("%s: ccc %u mb %p mbcnt %u", __func__, 3357 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3358 sbrelease_locked(so, SO_RCV); 3359 break; 3360 } 3361 SOCK_RECVBUF_UNLOCK(so); 3362 if (SOCK_IO_RECV_OWNED(so)) 3363 SOCK_IO_RECV_UNLOCK(so); 3364 3365 if (m != NULL) { 3366 unp_scan(m, unp_freerights); 3367 m_freem(m); 3368 } 3369 } 3370 3371 static void 3372 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3373 { 3374 struct mbuf *m; 3375 struct cmsghdr *cm; 3376 void *data; 3377 socklen_t clen, datalen; 3378 3379 while (m0 != NULL) { 3380 for (m = m0; m; m = m->m_next) { 3381 if (m->m_type != MT_CONTROL) 3382 continue; 3383 3384 cm = mtod(m, struct cmsghdr *); 3385 clen = m->m_len; 3386 3387 while (cm != NULL) { 3388 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3389 break; 3390 3391 data = CMSG_DATA(cm); 3392 datalen = (caddr_t)cm + cm->cmsg_len 3393 - (caddr_t)data; 3394 3395 if (cm->cmsg_level == SOL_SOCKET && 3396 cm->cmsg_type == SCM_RIGHTS) { 3397 (*op)(data, datalen / 3398 sizeof(struct filedescent *)); 3399 } 3400 3401 if (CMSG_SPACE(datalen) < clen) { 3402 clen -= CMSG_SPACE(datalen); 3403 cm = (struct cmsghdr *) 3404 ((caddr_t)cm + CMSG_SPACE(datalen)); 3405 } else { 3406 clen = 0; 3407 cm = NULL; 3408 } 3409 } 3410 } 3411 m0 = m0->m_nextpkt; 3412 } 3413 } 3414 3415 /* 3416 * A helper function called by VFS before socket-type vnode reclamation. 3417 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3418 * use count. 3419 */ 3420 void 3421 vfs_unp_reclaim(struct vnode *vp) 3422 { 3423 struct unpcb *unp; 3424 int active; 3425 struct mtx *vplock; 3426 3427 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3428 KASSERT(vp->v_type == VSOCK, 3429 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3430 3431 active = 0; 3432 vplock = mtx_pool_find(mtxpool_sleep, vp); 3433 mtx_lock(vplock); 3434 VOP_UNP_CONNECT(vp, &unp); 3435 if (unp == NULL) 3436 goto done; 3437 UNP_PCB_LOCK(unp); 3438 if (unp->unp_vnode == vp) { 3439 VOP_UNP_DETACH(vp); 3440 unp->unp_vnode = NULL; 3441 active = 1; 3442 } 3443 UNP_PCB_UNLOCK(unp); 3444 done: 3445 mtx_unlock(vplock); 3446 if (active) 3447 vunref(vp); 3448 } 3449 3450 #ifdef DDB 3451 static void 3452 db_print_indent(int indent) 3453 { 3454 int i; 3455 3456 for (i = 0; i < indent; i++) 3457 db_printf(" "); 3458 } 3459 3460 static void 3461 db_print_unpflags(int unp_flags) 3462 { 3463 int comma; 3464 3465 comma = 0; 3466 if (unp_flags & UNP_HAVEPC) { 3467 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3468 comma = 1; 3469 } 3470 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3471 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3472 comma = 1; 3473 } 3474 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3475 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3476 comma = 1; 3477 } 3478 if (unp_flags & UNP_CONNWAIT) { 3479 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3480 comma = 1; 3481 } 3482 if (unp_flags & UNP_CONNECTING) { 3483 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3484 comma = 1; 3485 } 3486 if (unp_flags & UNP_BINDING) { 3487 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3488 comma = 1; 3489 } 3490 } 3491 3492 static void 3493 db_print_xucred(int indent, struct xucred *xu) 3494 { 3495 int comma, i; 3496 3497 db_print_indent(indent); 3498 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3499 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3500 db_print_indent(indent); 3501 db_printf("cr_groups: "); 3502 comma = 0; 3503 for (i = 0; i < xu->cr_ngroups; i++) { 3504 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3505 comma = 1; 3506 } 3507 db_printf("\n"); 3508 } 3509 3510 static void 3511 db_print_unprefs(int indent, struct unp_head *uh) 3512 { 3513 struct unpcb *unp; 3514 int counter; 3515 3516 counter = 0; 3517 LIST_FOREACH(unp, uh, unp_reflink) { 3518 if (counter % 4 == 0) 3519 db_print_indent(indent); 3520 db_printf("%p ", unp); 3521 if (counter % 4 == 3) 3522 db_printf("\n"); 3523 counter++; 3524 } 3525 if (counter != 0 && counter % 4 != 0) 3526 db_printf("\n"); 3527 } 3528 3529 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3530 { 3531 struct unpcb *unp; 3532 3533 if (!have_addr) { 3534 db_printf("usage: show unpcb <addr>\n"); 3535 return; 3536 } 3537 unp = (struct unpcb *)addr; 3538 3539 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3540 unp->unp_vnode); 3541 3542 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3543 unp->unp_conn); 3544 3545 db_printf("unp_refs:\n"); 3546 db_print_unprefs(2, &unp->unp_refs); 3547 3548 /* XXXRW: Would be nice to print the full address, if any. */ 3549 db_printf("unp_addr: %p\n", unp->unp_addr); 3550 3551 db_printf("unp_gencnt: %llu\n", 3552 (unsigned long long)unp->unp_gencnt); 3553 3554 db_printf("unp_flags: %x (", unp->unp_flags); 3555 db_print_unpflags(unp->unp_flags); 3556 db_printf(")\n"); 3557 3558 db_printf("unp_peercred:\n"); 3559 db_print_xucred(2, &unp->unp_peercred); 3560 3561 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3562 } 3563 #endif 3564