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 * Copyright (c) 2022 Gleb Smirnoff <glebius@FreeBSD.org> 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /* 36 * UNIX Domain (Local) Sockets 37 * 38 * This is an implementation of UNIX (local) domain sockets. Each socket has 39 * an associated struct unpcb (UNIX protocol control block). Stream sockets 40 * may be connected to 0 or 1 other socket. Datagram sockets may be 41 * connected to 0, 1, or many other sockets. Sockets may be created and 42 * connected in pairs (socketpair(2)), or bound/connected to using the file 43 * system name space. For most purposes, only the receive socket buffer is 44 * used, as sending on one socket delivers directly to the receive socket 45 * buffer of a second socket. 46 * 47 * The implementation is substantially complicated by the fact that 48 * "ancillary data", such as file descriptors or credentials, may be passed 49 * across UNIX domain sockets. The potential for passing UNIX domain sockets 50 * over other UNIX domain sockets requires the implementation of a simple 51 * garbage collector to find and tear down cycles of disconnected sockets. 52 * 53 * TODO: 54 * RDM 55 * rethink name space problems 56 * need a proper out-of-band 57 */ 58 59 #include <sys/cdefs.h> 60 #include "opt_ddb.h" 61 62 #include <sys/param.h> 63 #include <sys/capsicum.h> 64 #include <sys/domain.h> 65 #include <sys/eventhandler.h> 66 #include <sys/fcntl.h> 67 #include <sys/file.h> 68 #include <sys/filedesc.h> 69 #include <sys/kernel.h> 70 #include <sys/lock.h> 71 #include <sys/malloc.h> 72 #include <sys/mbuf.h> 73 #include <sys/mount.h> 74 #include <sys/mutex.h> 75 #include <sys/namei.h> 76 #include <sys/proc.h> 77 #include <sys/protosw.h> 78 #include <sys/queue.h> 79 #include <sys/resourcevar.h> 80 #include <sys/rwlock.h> 81 #include <sys/socket.h> 82 #include <sys/socketvar.h> 83 #include <sys/signalvar.h> 84 #include <sys/stat.h> 85 #include <sys/sx.h> 86 #include <sys/sysctl.h> 87 #include <sys/systm.h> 88 #include <sys/taskqueue.h> 89 #include <sys/un.h> 90 #include <sys/unpcb.h> 91 #include <sys/vnode.h> 92 93 #include <net/vnet.h> 94 95 #ifdef DDB 96 #include <ddb/ddb.h> 97 #endif 98 99 #include <security/mac/mac_framework.h> 100 101 #include <vm/uma.h> 102 103 MALLOC_DECLARE(M_FILECAPS); 104 105 static struct domain localdomain; 106 107 static uma_zone_t unp_zone; 108 static unp_gen_t unp_gencnt; /* (l) */ 109 static u_int unp_count; /* (l) Count of local sockets. */ 110 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 111 static int unp_rights; /* (g) File descriptors in flight. */ 112 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 113 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 114 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 115 116 struct unp_defer { 117 SLIST_ENTRY(unp_defer) ud_link; 118 struct file *ud_fp; 119 }; 120 static SLIST_HEAD(, unp_defer) unp_defers; 121 static int unp_defers_count; 122 123 static const struct sockaddr sun_noname = { 124 .sa_len = sizeof(sun_noname), 125 .sa_family = AF_LOCAL, 126 }; 127 128 /* 129 * Garbage collection of cyclic file descriptor/socket references occurs 130 * asynchronously in a taskqueue context in order to avoid recursion and 131 * reentrance in the UNIX domain socket, file descriptor, and socket layer 132 * code. See unp_gc() for a full description. 133 */ 134 static struct timeout_task unp_gc_task; 135 136 /* 137 * The close of unix domain sockets attached as SCM_RIGHTS is 138 * postponed to the taskqueue, to avoid arbitrary recursion depth. 139 * The attached sockets might have another sockets attached. 140 */ 141 static struct task unp_defer_task; 142 143 /* 144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 145 * stream sockets, although the total for sender and receiver is actually 146 * only PIPSIZ. 147 * 148 * Datagram sockets really use the sendspace as the maximum datagram size, 149 * and don't really want to reserve the sendspace. Their recvspace should be 150 * large enough for at least one max-size datagram plus address. 151 */ 152 #ifndef PIPSIZ 153 #define PIPSIZ 8192 154 #endif 155 static u_long unpst_sendspace = PIPSIZ; 156 static u_long unpst_recvspace = PIPSIZ; 157 static u_long unpdg_maxdgram = 8*1024; /* support 8KB syslog msgs */ 158 static u_long unpdg_recvspace = 16*1024; 159 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 160 static u_long unpsp_recvspace = PIPSIZ; 161 162 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 163 "Local domain"); 164 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 165 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 166 "SOCK_STREAM"); 167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 169 "SOCK_DGRAM"); 170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 172 "SOCK_SEQPACKET"); 173 174 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 175 &unpst_sendspace, 0, "Default stream send space."); 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 177 &unpst_recvspace, 0, "Default stream receive space."); 178 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 179 &unpdg_maxdgram, 0, "Maximum datagram size."); 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 181 &unpdg_recvspace, 0, "Default datagram receive space."); 182 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 183 &unpsp_sendspace, 0, "Default seqpacket send space."); 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 185 &unpsp_recvspace, 0, "Default seqpacket receive space."); 186 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 187 "File descriptors in flight."); 188 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 189 &unp_defers_count, 0, 190 "File descriptors deferred to taskqueue for close."); 191 192 /* 193 * Locking and synchronization: 194 * 195 * Several types of locks exist in the local domain socket implementation: 196 * - a global linkage lock 197 * - a global connection list lock 198 * - the mtxpool lock 199 * - per-unpcb mutexes 200 * 201 * The linkage lock protects the global socket lists, the generation number 202 * counter and garbage collector state. 203 * 204 * The connection list lock protects the list of referring sockets in a datagram 205 * socket PCB. This lock is also overloaded to protect a global list of 206 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 207 * messages. To avoid recursion, such references are released by a dedicated 208 * thread. 209 * 210 * The mtxpool lock protects the vnode from being modified while referenced. 211 * Lock ordering rules require that it be acquired before any PCB locks. 212 * 213 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 214 * unpcb. This includes the unp_conn field, which either links two connected 215 * PCBs together (for connected socket types) or points at the destination 216 * socket (for connectionless socket types). The operations of creating or 217 * destroying a connection therefore involve locking multiple PCBs. To avoid 218 * lock order reversals, in some cases this involves dropping a PCB lock and 219 * using a reference counter to maintain liveness. 220 * 221 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 222 * allocated in pr_attach() and freed in pr_detach(). The validity of that 223 * pointer is an invariant, so no lock is required to dereference the so_pcb 224 * pointer if a valid socket reference is held by the caller. In practice, 225 * this is always true during operations performed on a socket. Each unpcb 226 * has a back-pointer to its socket, unp_socket, which will be stable under 227 * the same circumstances. 228 * 229 * This pointer may only be safely dereferenced as long as a valid reference 230 * to the unpcb is held. Typically, this reference will be from the socket, 231 * or from another unpcb when the referring unpcb's lock is held (in order 232 * that the reference not be invalidated during use). For example, to follow 233 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 234 * that detach is not run clearing unp_socket. 235 * 236 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 237 * protocols, bind() is a non-atomic operation, and connect() requires 238 * potential sleeping in the protocol, due to potentially waiting on local or 239 * distributed file systems. We try to separate "lookup" operations, which 240 * may sleep, and the IPC operations themselves, which typically can occur 241 * with relative atomicity as locks can be held over the entire operation. 242 * 243 * Another tricky issue is simultaneous multi-threaded or multi-process 244 * access to a single UNIX domain socket. These are handled by the flags 245 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 246 * binding, both of which involve dropping UNIX domain socket locks in order 247 * to perform namei() and other file system operations. 248 */ 249 static struct rwlock unp_link_rwlock; 250 static struct mtx unp_defers_lock; 251 252 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 253 "unp_link_rwlock") 254 255 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 256 RA_LOCKED) 257 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 258 RA_UNLOCKED) 259 260 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 261 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 262 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 263 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 264 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 265 RA_WLOCKED) 266 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 267 268 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 269 "unp_defer", NULL, MTX_DEF) 270 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 271 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 272 273 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 274 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 275 276 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 277 "unp", "unp", \ 278 MTX_DUPOK|MTX_DEF) 279 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 280 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 281 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 282 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 283 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 284 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 285 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 286 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 287 288 static int uipc_connect2(struct socket *, struct socket *); 289 static int uipc_ctloutput(struct socket *, struct sockopt *); 290 static int unp_connect(struct socket *, struct sockaddr *, 291 struct thread *); 292 static int unp_connectat(int, struct socket *, struct sockaddr *, 293 struct thread *, bool); 294 static void unp_connect2(struct socket *so, struct socket *so2); 295 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 296 static void unp_dispose(struct socket *so); 297 static void unp_shutdown(struct unpcb *); 298 static void unp_drop(struct unpcb *); 299 static void unp_gc(__unused void *, int); 300 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 301 static void unp_discard(struct file *); 302 static void unp_freerights(struct filedescent **, int); 303 static int unp_internalize(struct mbuf **, struct thread *, 304 struct mbuf **, u_int *, u_int *); 305 static void unp_internalize_fp(struct file *); 306 static int unp_externalize(struct mbuf *, struct mbuf **, int); 307 static int unp_externalize_fp(struct file *); 308 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 309 int, struct mbuf **, u_int *, u_int *); 310 static void unp_process_defers(void * __unused, int); 311 312 static void 313 unp_pcb_hold(struct unpcb *unp) 314 { 315 u_int old __unused; 316 317 old = refcount_acquire(&unp->unp_refcount); 318 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 319 } 320 321 static __result_use_check bool 322 unp_pcb_rele(struct unpcb *unp) 323 { 324 bool ret; 325 326 UNP_PCB_LOCK_ASSERT(unp); 327 328 if ((ret = refcount_release(&unp->unp_refcount))) { 329 UNP_PCB_UNLOCK(unp); 330 UNP_PCB_LOCK_DESTROY(unp); 331 uma_zfree(unp_zone, unp); 332 } 333 return (ret); 334 } 335 336 static void 337 unp_pcb_rele_notlast(struct unpcb *unp) 338 { 339 bool ret __unused; 340 341 ret = refcount_release(&unp->unp_refcount); 342 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 343 } 344 345 static void 346 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 347 { 348 UNP_PCB_UNLOCK_ASSERT(unp); 349 UNP_PCB_UNLOCK_ASSERT(unp2); 350 351 if (unp == unp2) { 352 UNP_PCB_LOCK(unp); 353 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 354 UNP_PCB_LOCK(unp); 355 UNP_PCB_LOCK(unp2); 356 } else { 357 UNP_PCB_LOCK(unp2); 358 UNP_PCB_LOCK(unp); 359 } 360 } 361 362 static void 363 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 364 { 365 UNP_PCB_UNLOCK(unp); 366 if (unp != unp2) 367 UNP_PCB_UNLOCK(unp2); 368 } 369 370 /* 371 * Try to lock the connected peer of an already locked socket. In some cases 372 * this requires that we unlock the current socket. The pairbusy counter is 373 * used to block concurrent connection attempts while the lock is dropped. The 374 * caller must be careful to revalidate PCB state. 375 */ 376 static struct unpcb * 377 unp_pcb_lock_peer(struct unpcb *unp) 378 { 379 struct unpcb *unp2; 380 381 UNP_PCB_LOCK_ASSERT(unp); 382 unp2 = unp->unp_conn; 383 if (unp2 == NULL) 384 return (NULL); 385 if (__predict_false(unp == unp2)) 386 return (unp); 387 388 UNP_PCB_UNLOCK_ASSERT(unp2); 389 390 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 391 return (unp2); 392 if ((uintptr_t)unp2 > (uintptr_t)unp) { 393 UNP_PCB_LOCK(unp2); 394 return (unp2); 395 } 396 unp->unp_pairbusy++; 397 unp_pcb_hold(unp2); 398 UNP_PCB_UNLOCK(unp); 399 400 UNP_PCB_LOCK(unp2); 401 UNP_PCB_LOCK(unp); 402 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 403 ("%s: socket %p was reconnected", __func__, unp)); 404 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 405 unp->unp_flags &= ~UNP_WAITING; 406 wakeup(unp); 407 } 408 if (unp_pcb_rele(unp2)) { 409 /* unp2 is unlocked. */ 410 return (NULL); 411 } 412 if (unp->unp_conn == NULL) { 413 UNP_PCB_UNLOCK(unp2); 414 return (NULL); 415 } 416 return (unp2); 417 } 418 419 static void 420 uipc_abort(struct socket *so) 421 { 422 struct unpcb *unp, *unp2; 423 424 unp = sotounpcb(so); 425 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 426 UNP_PCB_UNLOCK_ASSERT(unp); 427 428 UNP_PCB_LOCK(unp); 429 unp2 = unp->unp_conn; 430 if (unp2 != NULL) { 431 unp_pcb_hold(unp2); 432 UNP_PCB_UNLOCK(unp); 433 unp_drop(unp2); 434 } else 435 UNP_PCB_UNLOCK(unp); 436 } 437 438 static int 439 uipc_attach(struct socket *so, int proto, struct thread *td) 440 { 441 u_long sendspace, recvspace; 442 struct unpcb *unp; 443 int error; 444 bool locked; 445 446 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 447 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 448 switch (so->so_type) { 449 case SOCK_STREAM: 450 sendspace = unpst_sendspace; 451 recvspace = unpst_recvspace; 452 break; 453 454 case SOCK_DGRAM: 455 STAILQ_INIT(&so->so_rcv.uxdg_mb); 456 STAILQ_INIT(&so->so_snd.uxdg_mb); 457 TAILQ_INIT(&so->so_rcv.uxdg_conns); 458 /* 459 * Since send buffer is either bypassed or is a part 460 * of one-to-many receive buffer, we assign both space 461 * limits to unpdg_recvspace. 462 */ 463 sendspace = recvspace = unpdg_recvspace; 464 break; 465 466 case SOCK_SEQPACKET: 467 sendspace = unpsp_sendspace; 468 recvspace = unpsp_recvspace; 469 break; 470 471 default: 472 panic("uipc_attach"); 473 } 474 error = soreserve(so, sendspace, recvspace); 475 if (error) 476 return (error); 477 } 478 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 479 if (unp == NULL) 480 return (ENOBUFS); 481 LIST_INIT(&unp->unp_refs); 482 UNP_PCB_LOCK_INIT(unp); 483 unp->unp_socket = so; 484 so->so_pcb = unp; 485 refcount_init(&unp->unp_refcount, 1); 486 487 if ((locked = UNP_LINK_WOWNED()) == false) 488 UNP_LINK_WLOCK(); 489 490 unp->unp_gencnt = ++unp_gencnt; 491 unp->unp_ino = ++unp_ino; 492 unp_count++; 493 switch (so->so_type) { 494 case SOCK_STREAM: 495 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 496 break; 497 498 case SOCK_DGRAM: 499 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 500 break; 501 502 case SOCK_SEQPACKET: 503 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 504 break; 505 506 default: 507 panic("uipc_attach"); 508 } 509 510 if (locked == false) 511 UNP_LINK_WUNLOCK(); 512 513 return (0); 514 } 515 516 static int 517 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 518 { 519 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 520 struct vattr vattr; 521 int error, namelen; 522 struct nameidata nd; 523 struct unpcb *unp; 524 struct vnode *vp; 525 struct mount *mp; 526 cap_rights_t rights; 527 char *buf; 528 529 if (nam->sa_family != AF_UNIX) 530 return (EAFNOSUPPORT); 531 532 unp = sotounpcb(so); 533 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 534 535 if (soun->sun_len > sizeof(struct sockaddr_un)) 536 return (EINVAL); 537 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 538 if (namelen <= 0) 539 return (EINVAL); 540 541 /* 542 * We don't allow simultaneous bind() calls on a single UNIX domain 543 * socket, so flag in-progress operations, and return an error if an 544 * operation is already in progress. 545 * 546 * Historically, we have not allowed a socket to be rebound, so this 547 * also returns an error. Not allowing re-binding simplifies the 548 * implementation and avoids a great many possible failure modes. 549 */ 550 UNP_PCB_LOCK(unp); 551 if (unp->unp_vnode != NULL) { 552 UNP_PCB_UNLOCK(unp); 553 return (EINVAL); 554 } 555 if (unp->unp_flags & UNP_BINDING) { 556 UNP_PCB_UNLOCK(unp); 557 return (EALREADY); 558 } 559 unp->unp_flags |= UNP_BINDING; 560 UNP_PCB_UNLOCK(unp); 561 562 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 563 bcopy(soun->sun_path, buf, namelen); 564 buf[namelen] = 0; 565 566 restart: 567 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, 568 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 569 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 570 error = namei(&nd); 571 if (error) 572 goto error; 573 vp = nd.ni_vp; 574 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 575 NDFREE_PNBUF(&nd); 576 if (nd.ni_dvp == vp) 577 vrele(nd.ni_dvp); 578 else 579 vput(nd.ni_dvp); 580 if (vp != NULL) { 581 vrele(vp); 582 error = EADDRINUSE; 583 goto error; 584 } 585 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); 586 if (error) 587 goto error; 588 goto restart; 589 } 590 VATTR_NULL(&vattr); 591 vattr.va_type = VSOCK; 592 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 593 #ifdef MAC 594 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 595 &vattr); 596 #endif 597 if (error == 0) 598 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 599 NDFREE_PNBUF(&nd); 600 if (error) { 601 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 602 vn_finished_write(mp); 603 if (error == ERELOOKUP) 604 goto restart; 605 goto error; 606 } 607 vp = nd.ni_vp; 608 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 609 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 610 611 UNP_PCB_LOCK(unp); 612 VOP_UNP_BIND(vp, unp); 613 unp->unp_vnode = vp; 614 unp->unp_addr = soun; 615 unp->unp_flags &= ~UNP_BINDING; 616 UNP_PCB_UNLOCK(unp); 617 vref(vp); 618 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 619 vn_finished_write(mp); 620 free(buf, M_TEMP); 621 return (0); 622 623 error: 624 UNP_PCB_LOCK(unp); 625 unp->unp_flags &= ~UNP_BINDING; 626 UNP_PCB_UNLOCK(unp); 627 free(buf, M_TEMP); 628 return (error); 629 } 630 631 static int 632 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 633 { 634 635 return (uipc_bindat(AT_FDCWD, so, nam, td)); 636 } 637 638 static int 639 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 640 { 641 int error; 642 643 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 644 error = unp_connect(so, nam, td); 645 return (error); 646 } 647 648 static int 649 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 650 struct thread *td) 651 { 652 int error; 653 654 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 655 error = unp_connectat(fd, so, nam, td, false); 656 return (error); 657 } 658 659 static void 660 uipc_close(struct socket *so) 661 { 662 struct unpcb *unp, *unp2; 663 struct vnode *vp = NULL; 664 struct mtx *vplock; 665 666 unp = sotounpcb(so); 667 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 668 669 vplock = NULL; 670 if ((vp = unp->unp_vnode) != NULL) { 671 vplock = mtx_pool_find(mtxpool_sleep, vp); 672 mtx_lock(vplock); 673 } 674 UNP_PCB_LOCK(unp); 675 if (vp && unp->unp_vnode == NULL) { 676 mtx_unlock(vplock); 677 vp = NULL; 678 } 679 if (vp != NULL) { 680 VOP_UNP_DETACH(vp); 681 unp->unp_vnode = NULL; 682 } 683 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 684 unp_disconnect(unp, unp2); 685 else 686 UNP_PCB_UNLOCK(unp); 687 if (vp) { 688 mtx_unlock(vplock); 689 vrele(vp); 690 } 691 } 692 693 static int 694 uipc_connect2(struct socket *so1, struct socket *so2) 695 { 696 struct unpcb *unp, *unp2; 697 698 if (so1->so_type != so2->so_type) 699 return (EPROTOTYPE); 700 701 unp = so1->so_pcb; 702 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 703 unp2 = so2->so_pcb; 704 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 705 unp_pcb_lock_pair(unp, unp2); 706 unp_connect2(so1, so2); 707 unp_pcb_unlock_pair(unp, unp2); 708 709 return (0); 710 } 711 712 static void 713 uipc_detach(struct socket *so) 714 { 715 struct unpcb *unp, *unp2; 716 struct mtx *vplock; 717 struct vnode *vp; 718 int local_unp_rights; 719 720 unp = sotounpcb(so); 721 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 722 723 vp = NULL; 724 vplock = NULL; 725 726 if (!SOLISTENING(so)) 727 unp_dispose(so); 728 729 UNP_LINK_WLOCK(); 730 LIST_REMOVE(unp, unp_link); 731 if (unp->unp_gcflag & UNPGC_DEAD) 732 LIST_REMOVE(unp, unp_dead); 733 unp->unp_gencnt = ++unp_gencnt; 734 --unp_count; 735 UNP_LINK_WUNLOCK(); 736 737 UNP_PCB_UNLOCK_ASSERT(unp); 738 restart: 739 if ((vp = unp->unp_vnode) != NULL) { 740 vplock = mtx_pool_find(mtxpool_sleep, vp); 741 mtx_lock(vplock); 742 } 743 UNP_PCB_LOCK(unp); 744 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 745 if (vplock) 746 mtx_unlock(vplock); 747 UNP_PCB_UNLOCK(unp); 748 goto restart; 749 } 750 if ((vp = unp->unp_vnode) != NULL) { 751 VOP_UNP_DETACH(vp); 752 unp->unp_vnode = NULL; 753 } 754 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 755 unp_disconnect(unp, unp2); 756 else 757 UNP_PCB_UNLOCK(unp); 758 759 UNP_REF_LIST_LOCK(); 760 while (!LIST_EMPTY(&unp->unp_refs)) { 761 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 762 763 unp_pcb_hold(ref); 764 UNP_REF_LIST_UNLOCK(); 765 766 MPASS(ref != unp); 767 UNP_PCB_UNLOCK_ASSERT(ref); 768 unp_drop(ref); 769 UNP_REF_LIST_LOCK(); 770 } 771 UNP_REF_LIST_UNLOCK(); 772 773 UNP_PCB_LOCK(unp); 774 local_unp_rights = unp_rights; 775 unp->unp_socket->so_pcb = NULL; 776 unp->unp_socket = NULL; 777 free(unp->unp_addr, M_SONAME); 778 unp->unp_addr = NULL; 779 if (!unp_pcb_rele(unp)) 780 UNP_PCB_UNLOCK(unp); 781 if (vp) { 782 mtx_unlock(vplock); 783 vrele(vp); 784 } 785 if (local_unp_rights) 786 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 787 788 switch (so->so_type) { 789 case SOCK_DGRAM: 790 /* 791 * Everything should have been unlinked/freed by unp_dispose() 792 * and/or unp_disconnect(). 793 */ 794 MPASS(so->so_rcv.uxdg_peeked == NULL); 795 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 796 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 797 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 798 } 799 } 800 801 static int 802 uipc_disconnect(struct socket *so) 803 { 804 struct unpcb *unp, *unp2; 805 806 unp = sotounpcb(so); 807 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 808 809 UNP_PCB_LOCK(unp); 810 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 811 unp_disconnect(unp, unp2); 812 else 813 UNP_PCB_UNLOCK(unp); 814 return (0); 815 } 816 817 static int 818 uipc_listen(struct socket *so, int backlog, struct thread *td) 819 { 820 struct unpcb *unp; 821 int error; 822 823 MPASS(so->so_type != SOCK_DGRAM); 824 825 /* 826 * Synchronize with concurrent connection attempts. 827 */ 828 error = 0; 829 unp = sotounpcb(so); 830 UNP_PCB_LOCK(unp); 831 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 832 error = EINVAL; 833 else if (unp->unp_vnode == NULL) 834 error = EDESTADDRREQ; 835 if (error != 0) { 836 UNP_PCB_UNLOCK(unp); 837 return (error); 838 } 839 840 SOCK_LOCK(so); 841 error = solisten_proto_check(so); 842 if (error == 0) { 843 cru2xt(td, &unp->unp_peercred); 844 solisten_proto(so, backlog); 845 } 846 SOCK_UNLOCK(so); 847 UNP_PCB_UNLOCK(unp); 848 return (error); 849 } 850 851 static int 852 uipc_peeraddr(struct socket *so, struct sockaddr *ret) 853 { 854 struct unpcb *unp, *unp2; 855 const struct sockaddr *sa; 856 857 unp = sotounpcb(so); 858 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 859 860 UNP_PCB_LOCK(unp); 861 unp2 = unp_pcb_lock_peer(unp); 862 if (unp2 != NULL) { 863 if (unp2->unp_addr != NULL) 864 sa = (struct sockaddr *)unp2->unp_addr; 865 else 866 sa = &sun_noname; 867 bcopy(sa, ret, sa->sa_len); 868 unp_pcb_unlock_pair(unp, unp2); 869 } else { 870 UNP_PCB_UNLOCK(unp); 871 sa = &sun_noname; 872 bcopy(sa, ret, sa->sa_len); 873 } 874 return (0); 875 } 876 877 static int 878 uipc_rcvd(struct socket *so, int flags) 879 { 880 struct unpcb *unp, *unp2; 881 struct socket *so2; 882 u_int mbcnt, sbcc; 883 884 unp = sotounpcb(so); 885 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 886 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 887 ("%s: socktype %d", __func__, so->so_type)); 888 889 /* 890 * Adjust backpressure on sender and wakeup any waiting to write. 891 * 892 * The unp lock is acquired to maintain the validity of the unp_conn 893 * pointer; no lock on unp2 is required as unp2->unp_socket will be 894 * static as long as we don't permit unp2 to disconnect from unp, 895 * which is prevented by the lock on unp. We cache values from 896 * so_rcv to avoid holding the so_rcv lock over the entire 897 * transaction on the remote so_snd. 898 */ 899 SOCKBUF_LOCK(&so->so_rcv); 900 mbcnt = so->so_rcv.sb_mbcnt; 901 sbcc = sbavail(&so->so_rcv); 902 SOCKBUF_UNLOCK(&so->so_rcv); 903 /* 904 * There is a benign race condition at this point. If we're planning to 905 * clear SB_STOP, but uipc_send is called on the connected socket at 906 * this instant, it might add data to the sockbuf and set SB_STOP. Then 907 * we would erroneously clear SB_STOP below, even though the sockbuf is 908 * full. The race is benign because the only ill effect is to allow the 909 * sockbuf to exceed its size limit, and the size limits are not 910 * strictly guaranteed anyway. 911 */ 912 UNP_PCB_LOCK(unp); 913 unp2 = unp->unp_conn; 914 if (unp2 == NULL) { 915 UNP_PCB_UNLOCK(unp); 916 return (0); 917 } 918 so2 = unp2->unp_socket; 919 SOCKBUF_LOCK(&so2->so_snd); 920 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 921 so2->so_snd.sb_flags &= ~SB_STOP; 922 sowwakeup_locked(so2); 923 UNP_PCB_UNLOCK(unp); 924 return (0); 925 } 926 927 static int 928 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 929 struct mbuf *control, struct thread *td) 930 { 931 struct unpcb *unp, *unp2; 932 struct socket *so2; 933 u_int mbcnt, sbcc; 934 int error; 935 936 unp = sotounpcb(so); 937 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 938 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 939 ("%s: socktype %d", __func__, so->so_type)); 940 941 error = 0; 942 if (flags & PRUS_OOB) { 943 error = EOPNOTSUPP; 944 goto release; 945 } 946 if (control != NULL && 947 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 948 goto release; 949 950 unp2 = NULL; 951 if ((so->so_state & SS_ISCONNECTED) == 0) { 952 if (nam != NULL) { 953 if ((error = unp_connect(so, nam, td)) != 0) 954 goto out; 955 } else { 956 error = ENOTCONN; 957 goto out; 958 } 959 } 960 961 UNP_PCB_LOCK(unp); 962 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 963 UNP_PCB_UNLOCK(unp); 964 error = ENOTCONN; 965 goto out; 966 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 967 unp_pcb_unlock_pair(unp, unp2); 968 error = EPIPE; 969 goto out; 970 } 971 UNP_PCB_UNLOCK(unp); 972 if ((so2 = unp2->unp_socket) == NULL) { 973 UNP_PCB_UNLOCK(unp2); 974 error = ENOTCONN; 975 goto out; 976 } 977 SOCKBUF_LOCK(&so2->so_rcv); 978 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 979 /* 980 * Credentials are passed only once on SOCK_STREAM and 981 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 982 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 983 */ 984 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 985 NULL, NULL); 986 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 987 } 988 989 /* 990 * Send to paired receive port and wake up readers. Don't 991 * check for space available in the receive buffer if we're 992 * attaching ancillary data; Unix domain sockets only check 993 * for space in the sending sockbuf, and that check is 994 * performed one level up the stack. At that level we cannot 995 * precisely account for the amount of buffer space used 996 * (e.g., because control messages are not yet internalized). 997 */ 998 switch (so->so_type) { 999 case SOCK_STREAM: 1000 if (control != NULL) { 1001 sbappendcontrol_locked(&so2->so_rcv, 1002 m->m_len > 0 ? m : NULL, control, flags); 1003 control = NULL; 1004 } else 1005 sbappend_locked(&so2->so_rcv, m, flags); 1006 break; 1007 1008 case SOCK_SEQPACKET: 1009 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1010 &sun_noname, m, control)) 1011 control = NULL; 1012 break; 1013 } 1014 1015 mbcnt = so2->so_rcv.sb_mbcnt; 1016 sbcc = sbavail(&so2->so_rcv); 1017 if (sbcc) 1018 sorwakeup_locked(so2); 1019 else 1020 SOCKBUF_UNLOCK(&so2->so_rcv); 1021 1022 /* 1023 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1024 * it would be possible for uipc_rcvd to be called at this 1025 * point, drain the receiving sockbuf, clear SB_STOP, and then 1026 * we would set SB_STOP below. That could lead to an empty 1027 * sockbuf having SB_STOP set 1028 */ 1029 SOCKBUF_LOCK(&so->so_snd); 1030 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1031 so->so_snd.sb_flags |= SB_STOP; 1032 SOCKBUF_UNLOCK(&so->so_snd); 1033 UNP_PCB_UNLOCK(unp2); 1034 m = NULL; 1035 out: 1036 /* 1037 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1038 */ 1039 if (flags & PRUS_EOF) { 1040 UNP_PCB_LOCK(unp); 1041 socantsendmore(so); 1042 unp_shutdown(unp); 1043 UNP_PCB_UNLOCK(unp); 1044 } 1045 if (control != NULL && error != 0) 1046 unp_scan(control, unp_freerights); 1047 1048 release: 1049 if (control != NULL) 1050 m_freem(control); 1051 /* 1052 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1053 * for freeing memory. 1054 */ 1055 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1056 m_freem(m); 1057 return (error); 1058 } 1059 1060 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1061 static inline bool 1062 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1063 { 1064 u_int bleft, mleft; 1065 1066 /* 1067 * Negative space may happen if send(2) is followed by 1068 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1069 */ 1070 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1071 sb->sb_mbmax < sb->uxdg_mbcnt)) 1072 return (false); 1073 1074 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1075 return (false); 1076 1077 bleft = sb->sb_hiwat - sb->uxdg_cc; 1078 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1079 1080 return (bleft >= cc && mleft >= mbcnt); 1081 } 1082 1083 /* 1084 * PF_UNIX/SOCK_DGRAM send 1085 * 1086 * Allocate a record consisting of 3 mbufs in the sequence of 1087 * from -> control -> data and append it to the socket buffer. 1088 * 1089 * The first mbuf carries sender's name and is a pkthdr that stores 1090 * overall length of datagram, its memory consumption and control length. 1091 */ 1092 #define ctllen PH_loc.thirtytwo[1] 1093 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1094 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1095 static int 1096 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1097 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1098 { 1099 struct unpcb *unp, *unp2; 1100 const struct sockaddr *from; 1101 struct socket *so2; 1102 struct sockbuf *sb; 1103 struct mbuf *f, *clast; 1104 u_int cc, ctl, mbcnt; 1105 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1106 int error; 1107 1108 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1109 1110 error = 0; 1111 f = NULL; 1112 ctl = 0; 1113 1114 if (__predict_false(flags & MSG_OOB)) { 1115 error = EOPNOTSUPP; 1116 goto out; 1117 } 1118 if (m == NULL) { 1119 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1120 error = EMSGSIZE; 1121 goto out; 1122 } 1123 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1124 if (__predict_false(m == NULL)) { 1125 error = EFAULT; 1126 goto out; 1127 } 1128 f = m_gethdr(M_WAITOK, MT_SONAME); 1129 cc = m->m_pkthdr.len; 1130 mbcnt = MSIZE + m->m_pkthdr.memlen; 1131 if (c != NULL && 1132 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1133 goto out; 1134 } else { 1135 /* pr_sosend() with mbuf usually is a kernel thread. */ 1136 1137 M_ASSERTPKTHDR(m); 1138 if (__predict_false(c != NULL)) 1139 panic("%s: control from a kernel thread", __func__); 1140 1141 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1142 error = EMSGSIZE; 1143 goto out; 1144 } 1145 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1146 error = ENOBUFS; 1147 goto out; 1148 } 1149 /* Condition the foreign mbuf to our standards. */ 1150 m_clrprotoflags(m); 1151 m_tag_delete_chain(m, NULL); 1152 m->m_pkthdr.rcvif = NULL; 1153 m->m_pkthdr.flowid = 0; 1154 m->m_pkthdr.csum_flags = 0; 1155 m->m_pkthdr.fibnum = 0; 1156 m->m_pkthdr.rsstype = 0; 1157 1158 cc = m->m_pkthdr.len; 1159 mbcnt = MSIZE; 1160 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1161 mbcnt += MSIZE; 1162 if (mb->m_flags & M_EXT) 1163 mbcnt += mb->m_ext.ext_size; 1164 } 1165 } 1166 1167 unp = sotounpcb(so); 1168 MPASS(unp); 1169 1170 /* 1171 * XXXGL: would be cool to fully remove so_snd out of the equation 1172 * and avoid this lock, which is not only extraneous, but also being 1173 * released, thus still leaving possibility for a race. We can easily 1174 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1175 * is more difficult to invent something to handle so_error. 1176 */ 1177 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1178 if (error) 1179 goto out2; 1180 SOCK_SENDBUF_LOCK(so); 1181 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1182 SOCK_SENDBUF_UNLOCK(so); 1183 error = EPIPE; 1184 goto out3; 1185 } 1186 if (so->so_error != 0) { 1187 error = so->so_error; 1188 so->so_error = 0; 1189 SOCK_SENDBUF_UNLOCK(so); 1190 goto out3; 1191 } 1192 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1193 SOCK_SENDBUF_UNLOCK(so); 1194 error = EDESTADDRREQ; 1195 goto out3; 1196 } 1197 SOCK_SENDBUF_UNLOCK(so); 1198 1199 if (addr != NULL) { 1200 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1201 goto out3; 1202 UNP_PCB_LOCK_ASSERT(unp); 1203 unp2 = unp->unp_conn; 1204 UNP_PCB_LOCK_ASSERT(unp2); 1205 } else { 1206 UNP_PCB_LOCK(unp); 1207 unp2 = unp_pcb_lock_peer(unp); 1208 if (unp2 == NULL) { 1209 UNP_PCB_UNLOCK(unp); 1210 error = ENOTCONN; 1211 goto out3; 1212 } 1213 } 1214 1215 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1216 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1217 &mbcnt); 1218 if (unp->unp_addr != NULL) 1219 from = (struct sockaddr *)unp->unp_addr; 1220 else 1221 from = &sun_noname; 1222 f->m_len = from->sa_len; 1223 MPASS(from->sa_len <= MLEN); 1224 bcopy(from, mtod(f, void *), from->sa_len); 1225 ctl += f->m_len; 1226 1227 /* 1228 * Concatenate mbufs: from -> control -> data. 1229 * Save overall cc and mbcnt in "from" mbuf. 1230 */ 1231 if (c != NULL) { 1232 #ifdef INVARIANTS 1233 struct mbuf *mc; 1234 1235 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1236 MPASS(mc == clast); 1237 #endif 1238 f->m_next = c; 1239 clast->m_next = m; 1240 c = NULL; 1241 } else 1242 f->m_next = m; 1243 m = NULL; 1244 #ifdef INVARIANTS 1245 dcc = dctl = dmbcnt = 0; 1246 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1247 if (mb->m_type == MT_DATA) 1248 dcc += mb->m_len; 1249 else 1250 dctl += mb->m_len; 1251 dmbcnt += MSIZE; 1252 if (mb->m_flags & M_EXT) 1253 dmbcnt += mb->m_ext.ext_size; 1254 } 1255 MPASS(dcc == cc); 1256 MPASS(dctl == ctl); 1257 MPASS(dmbcnt == mbcnt); 1258 #endif 1259 f->m_pkthdr.len = cc + ctl; 1260 f->m_pkthdr.memlen = mbcnt; 1261 f->m_pkthdr.ctllen = ctl; 1262 1263 /* 1264 * Destination socket buffer selection. 1265 * 1266 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1267 * destination address is supplied, create a temporary connection for 1268 * the run time of the function (see call to unp_connectat() above and 1269 * to unp_disconnect() below). We distinguish them by condition of 1270 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1271 * that condition, since, again, through the run time of this code we 1272 * are always connected. For such "unconnected" sends, the destination 1273 * buffer would be the receive buffer of destination socket so2. 1274 * 1275 * For connected sends, data lands on the send buffer of the sender's 1276 * socket "so". Then, if we just added the very first datagram 1277 * on this send buffer, we need to add the send buffer on to the 1278 * receiving socket's buffer list. We put ourselves on top of the 1279 * list. Such logic gives infrequent senders priority over frequent 1280 * senders. 1281 * 1282 * Note on byte count management. As long as event methods kevent(2), 1283 * select(2) are not protocol specific (yet), we need to maintain 1284 * meaningful values on the receive buffer. So, the receive buffer 1285 * would accumulate counters from all connected buffers potentially 1286 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1287 */ 1288 so2 = unp2->unp_socket; 1289 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1290 SOCK_RECVBUF_LOCK(so2); 1291 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1292 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1293 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1294 uxdg_clist); 1295 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1296 sb->uxdg_cc += cc + ctl; 1297 sb->uxdg_ctl += ctl; 1298 sb->uxdg_mbcnt += mbcnt; 1299 so2->so_rcv.sb_acc += cc + ctl; 1300 so2->so_rcv.sb_ccc += cc + ctl; 1301 so2->so_rcv.sb_ctl += ctl; 1302 so2->so_rcv.sb_mbcnt += mbcnt; 1303 sorwakeup_locked(so2); 1304 f = NULL; 1305 } else { 1306 soroverflow_locked(so2); 1307 error = ENOBUFS; 1308 if (f->m_next->m_type == MT_CONTROL) { 1309 c = f->m_next; 1310 f->m_next = NULL; 1311 } 1312 } 1313 1314 if (addr != NULL) 1315 unp_disconnect(unp, unp2); 1316 else 1317 unp_pcb_unlock_pair(unp, unp2); 1318 1319 td->td_ru.ru_msgsnd++; 1320 1321 out3: 1322 SOCK_IO_SEND_UNLOCK(so); 1323 out2: 1324 if (c) 1325 unp_scan(c, unp_freerights); 1326 out: 1327 if (f) 1328 m_freem(f); 1329 if (c) 1330 m_freem(c); 1331 if (m) 1332 m_freem(m); 1333 1334 return (error); 1335 } 1336 1337 /* 1338 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1339 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1340 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1341 */ 1342 static int 1343 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1344 struct uio *uio, struct mbuf **controlp, int *flagsp) 1345 { 1346 ssize_t len = 0; 1347 int error; 1348 1349 so->so_rcv.uxdg_peeked = m; 1350 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1351 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1352 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1353 SOCK_RECVBUF_UNLOCK(so); 1354 1355 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1356 if (psa != NULL) 1357 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1358 1359 m = m->m_next; 1360 KASSERT(m, ("%s: no data or control after soname", __func__)); 1361 1362 /* 1363 * With MSG_PEEK the control isn't executed, just copied. 1364 */ 1365 while (m != NULL && m->m_type == MT_CONTROL) { 1366 if (controlp != NULL) { 1367 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1368 controlp = &(*controlp)->m_next; 1369 } 1370 m = m->m_next; 1371 } 1372 KASSERT(m == NULL || m->m_type == MT_DATA, 1373 ("%s: not MT_DATA mbuf %p", __func__, m)); 1374 while (m != NULL && uio->uio_resid > 0) { 1375 len = uio->uio_resid; 1376 if (len > m->m_len) 1377 len = m->m_len; 1378 error = uiomove(mtod(m, char *), (int)len, uio); 1379 if (error) { 1380 SOCK_IO_RECV_UNLOCK(so); 1381 return (error); 1382 } 1383 if (len == m->m_len) 1384 m = m->m_next; 1385 } 1386 SOCK_IO_RECV_UNLOCK(so); 1387 1388 if (flagsp != NULL) { 1389 if (m != NULL) { 1390 if (*flagsp & MSG_TRUNC) { 1391 /* Report real length of the packet */ 1392 uio->uio_resid -= m_length(m, NULL) - len; 1393 } 1394 *flagsp |= MSG_TRUNC; 1395 } else 1396 *flagsp &= ~MSG_TRUNC; 1397 } 1398 1399 return (0); 1400 } 1401 1402 /* 1403 * PF_UNIX/SOCK_DGRAM receive 1404 */ 1405 static int 1406 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1407 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1408 { 1409 struct sockbuf *sb = NULL; 1410 struct mbuf *m; 1411 int flags, error; 1412 ssize_t len = 0; 1413 bool nonblock; 1414 1415 MPASS(mp0 == NULL); 1416 1417 if (psa != NULL) 1418 *psa = NULL; 1419 if (controlp != NULL) 1420 *controlp = NULL; 1421 1422 flags = flagsp != NULL ? *flagsp : 0; 1423 nonblock = (so->so_state & SS_NBIO) || 1424 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1425 1426 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1427 if (__predict_false(error)) 1428 return (error); 1429 1430 /* 1431 * Loop blocking while waiting for a datagram. Prioritize connected 1432 * peers over unconnected sends. Set sb to selected socket buffer 1433 * containing an mbuf on exit from the wait loop. A datagram that 1434 * had already been peeked at has top priority. 1435 */ 1436 SOCK_RECVBUF_LOCK(so); 1437 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1438 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1439 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1440 if (so->so_error) { 1441 error = so->so_error; 1442 if (!(flags & MSG_PEEK)) 1443 so->so_error = 0; 1444 SOCK_RECVBUF_UNLOCK(so); 1445 SOCK_IO_RECV_UNLOCK(so); 1446 return (error); 1447 } 1448 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1449 uio->uio_resid == 0) { 1450 SOCK_RECVBUF_UNLOCK(so); 1451 SOCK_IO_RECV_UNLOCK(so); 1452 return (0); 1453 } 1454 if (nonblock) { 1455 SOCK_RECVBUF_UNLOCK(so); 1456 SOCK_IO_RECV_UNLOCK(so); 1457 return (EWOULDBLOCK); 1458 } 1459 error = sbwait(so, SO_RCV); 1460 if (error) { 1461 SOCK_RECVBUF_UNLOCK(so); 1462 SOCK_IO_RECV_UNLOCK(so); 1463 return (error); 1464 } 1465 } 1466 1467 if (sb == NULL) 1468 sb = &so->so_rcv; 1469 else if (m == NULL) 1470 m = STAILQ_FIRST(&sb->uxdg_mb); 1471 else 1472 MPASS(m == so->so_rcv.uxdg_peeked); 1473 1474 MPASS(sb->uxdg_cc > 0); 1475 M_ASSERTPKTHDR(m); 1476 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1477 1478 if (uio->uio_td) 1479 uio->uio_td->td_ru.ru_msgrcv++; 1480 1481 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1482 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1483 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1484 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1485 } else 1486 so->so_rcv.uxdg_peeked = NULL; 1487 1488 sb->uxdg_cc -= m->m_pkthdr.len; 1489 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1490 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1491 1492 if (__predict_false(flags & MSG_PEEK)) 1493 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1494 1495 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1496 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1497 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1498 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1499 SOCK_RECVBUF_UNLOCK(so); 1500 1501 if (psa != NULL) 1502 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1503 m = m_free(m); 1504 KASSERT(m, ("%s: no data or control after soname", __func__)); 1505 1506 /* 1507 * Packet to copyout() is now in 'm' and it is disconnected from the 1508 * queue. 1509 * 1510 * Process one or more MT_CONTROL mbufs present before any data mbufs 1511 * in the first mbuf chain on the socket buffer. We call into the 1512 * unp_externalize() to perform externalization (or freeing if 1513 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1514 * without MT_DATA mbufs. 1515 */ 1516 while (m != NULL && m->m_type == MT_CONTROL) { 1517 struct mbuf *cm; 1518 1519 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1520 * it frees whole chain, so we must disconnect the mbuf. 1521 */ 1522 cm = m; m = m->m_next; cm->m_next = NULL; 1523 error = unp_externalize(cm, controlp, flags); 1524 if (error != 0) { 1525 SOCK_IO_RECV_UNLOCK(so); 1526 unp_scan(m, unp_freerights); 1527 m_freem(m); 1528 return (error); 1529 } 1530 if (controlp != NULL) { 1531 while (*controlp != NULL) 1532 controlp = &(*controlp)->m_next; 1533 } 1534 } 1535 KASSERT(m == NULL || m->m_type == MT_DATA, 1536 ("%s: not MT_DATA mbuf %p", __func__, m)); 1537 while (m != NULL && uio->uio_resid > 0) { 1538 len = uio->uio_resid; 1539 if (len > m->m_len) 1540 len = m->m_len; 1541 error = uiomove(mtod(m, char *), (int)len, uio); 1542 if (error) { 1543 SOCK_IO_RECV_UNLOCK(so); 1544 m_freem(m); 1545 return (error); 1546 } 1547 if (len == m->m_len) 1548 m = m_free(m); 1549 else { 1550 m->m_data += len; 1551 m->m_len -= len; 1552 } 1553 } 1554 SOCK_IO_RECV_UNLOCK(so); 1555 1556 if (m != NULL) { 1557 if (flagsp != NULL) { 1558 if (flags & MSG_TRUNC) { 1559 /* Report real length of the packet */ 1560 uio->uio_resid -= m_length(m, NULL); 1561 } 1562 *flagsp |= MSG_TRUNC; 1563 } 1564 m_freem(m); 1565 } else if (flagsp != NULL) 1566 *flagsp &= ~MSG_TRUNC; 1567 1568 return (0); 1569 } 1570 1571 static bool 1572 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1573 { 1574 struct mbuf *mb, *n; 1575 struct sockbuf *sb; 1576 1577 SOCK_LOCK(so); 1578 if (SOLISTENING(so)) { 1579 SOCK_UNLOCK(so); 1580 return (false); 1581 } 1582 mb = NULL; 1583 sb = &so->so_rcv; 1584 SOCKBUF_LOCK(sb); 1585 if (sb->sb_fnrdy != NULL) { 1586 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1587 if (mb == m) { 1588 *errorp = sbready(sb, m, count); 1589 break; 1590 } 1591 mb = mb->m_next; 1592 if (mb == NULL) { 1593 mb = n; 1594 if (mb != NULL) 1595 n = mb->m_nextpkt; 1596 } 1597 } 1598 } 1599 SOCKBUF_UNLOCK(sb); 1600 SOCK_UNLOCK(so); 1601 return (mb != NULL); 1602 } 1603 1604 static int 1605 uipc_ready(struct socket *so, struct mbuf *m, int count) 1606 { 1607 struct unpcb *unp, *unp2; 1608 struct socket *so2; 1609 int error, i; 1610 1611 unp = sotounpcb(so); 1612 1613 KASSERT(so->so_type == SOCK_STREAM, 1614 ("%s: unexpected socket type for %p", __func__, so)); 1615 1616 UNP_PCB_LOCK(unp); 1617 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1618 UNP_PCB_UNLOCK(unp); 1619 so2 = unp2->unp_socket; 1620 SOCKBUF_LOCK(&so2->so_rcv); 1621 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1622 sorwakeup_locked(so2); 1623 else 1624 SOCKBUF_UNLOCK(&so2->so_rcv); 1625 UNP_PCB_UNLOCK(unp2); 1626 return (error); 1627 } 1628 UNP_PCB_UNLOCK(unp); 1629 1630 /* 1631 * The receiving socket has been disconnected, but may still be valid. 1632 * In this case, the now-ready mbufs are still present in its socket 1633 * buffer, so perform an exhaustive search before giving up and freeing 1634 * the mbufs. 1635 */ 1636 UNP_LINK_RLOCK(); 1637 LIST_FOREACH(unp, &unp_shead, unp_link) { 1638 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1639 break; 1640 } 1641 UNP_LINK_RUNLOCK(); 1642 1643 if (unp == NULL) { 1644 for (i = 0; i < count; i++) 1645 m = m_free(m); 1646 error = ECONNRESET; 1647 } 1648 return (error); 1649 } 1650 1651 static int 1652 uipc_sense(struct socket *so, struct stat *sb) 1653 { 1654 struct unpcb *unp; 1655 1656 unp = sotounpcb(so); 1657 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1658 1659 sb->st_blksize = so->so_snd.sb_hiwat; 1660 sb->st_dev = NODEV; 1661 sb->st_ino = unp->unp_ino; 1662 return (0); 1663 } 1664 1665 static int 1666 uipc_shutdown(struct socket *so, enum shutdown_how how) 1667 { 1668 struct unpcb *unp = sotounpcb(so); 1669 int error; 1670 1671 SOCK_LOCK(so); 1672 if (SOLISTENING(so)) { 1673 if (how != SHUT_WR) { 1674 so->so_error = ECONNABORTED; 1675 solisten_wakeup(so); /* unlocks so */ 1676 } else 1677 SOCK_UNLOCK(so); 1678 return (ENOTCONN); 1679 } else if ((so->so_state & 1680 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 1681 /* 1682 * POSIX mandates us to just return ENOTCONN when shutdown(2) is 1683 * invoked on a datagram sockets, however historically we would 1684 * actually tear socket down. This is known to be leveraged by 1685 * some applications to unblock process waiting in recv(2) by 1686 * other process that it shares that socket with. Try to meet 1687 * both backward-compatibility and POSIX requirements by forcing 1688 * ENOTCONN but still flushing buffers and performing wakeup(9). 1689 * 1690 * XXXGL: it remains unknown what applications expect this 1691 * behavior and is this isolated to unix/dgram or inet/dgram or 1692 * both. See: D10351, D3039. 1693 */ 1694 error = ENOTCONN; 1695 if (so->so_type != SOCK_DGRAM) { 1696 SOCK_UNLOCK(so); 1697 return (error); 1698 } 1699 } else 1700 error = 0; 1701 SOCK_UNLOCK(so); 1702 1703 switch (how) { 1704 case SHUT_RD: 1705 socantrcvmore(so); 1706 unp_dispose(so); 1707 break; 1708 case SHUT_RDWR: 1709 socantrcvmore(so); 1710 unp_dispose(so); 1711 /* FALLTHROUGH */ 1712 case SHUT_WR: 1713 UNP_PCB_LOCK(unp); 1714 socantsendmore(so); 1715 unp_shutdown(unp); 1716 UNP_PCB_UNLOCK(unp); 1717 } 1718 wakeup(&so->so_timeo); 1719 1720 return (error); 1721 } 1722 1723 static int 1724 uipc_sockaddr(struct socket *so, struct sockaddr *ret) 1725 { 1726 struct unpcb *unp; 1727 const struct sockaddr *sa; 1728 1729 unp = sotounpcb(so); 1730 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1731 1732 UNP_PCB_LOCK(unp); 1733 if (unp->unp_addr != NULL) 1734 sa = (struct sockaddr *) unp->unp_addr; 1735 else 1736 sa = &sun_noname; 1737 bcopy(sa, ret, sa->sa_len); 1738 UNP_PCB_UNLOCK(unp); 1739 return (0); 1740 } 1741 1742 static int 1743 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1744 { 1745 struct unpcb *unp; 1746 struct xucred xu; 1747 int error, optval; 1748 1749 if (sopt->sopt_level != SOL_LOCAL) 1750 return (EINVAL); 1751 1752 unp = sotounpcb(so); 1753 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1754 error = 0; 1755 switch (sopt->sopt_dir) { 1756 case SOPT_GET: 1757 switch (sopt->sopt_name) { 1758 case LOCAL_PEERCRED: 1759 UNP_PCB_LOCK(unp); 1760 if (unp->unp_flags & UNP_HAVEPC) 1761 xu = unp->unp_peercred; 1762 else { 1763 if (so->so_type == SOCK_STREAM) 1764 error = ENOTCONN; 1765 else 1766 error = EINVAL; 1767 } 1768 UNP_PCB_UNLOCK(unp); 1769 if (error == 0) 1770 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1771 break; 1772 1773 case LOCAL_CREDS: 1774 /* Unlocked read. */ 1775 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1776 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1777 break; 1778 1779 case LOCAL_CREDS_PERSISTENT: 1780 /* Unlocked read. */ 1781 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1782 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1783 break; 1784 1785 default: 1786 error = EOPNOTSUPP; 1787 break; 1788 } 1789 break; 1790 1791 case SOPT_SET: 1792 switch (sopt->sopt_name) { 1793 case LOCAL_CREDS: 1794 case LOCAL_CREDS_PERSISTENT: 1795 error = sooptcopyin(sopt, &optval, sizeof(optval), 1796 sizeof(optval)); 1797 if (error) 1798 break; 1799 1800 #define OPTSET(bit, exclusive) do { \ 1801 UNP_PCB_LOCK(unp); \ 1802 if (optval) { \ 1803 if ((unp->unp_flags & (exclusive)) != 0) { \ 1804 UNP_PCB_UNLOCK(unp); \ 1805 error = EINVAL; \ 1806 break; \ 1807 } \ 1808 unp->unp_flags |= (bit); \ 1809 } else \ 1810 unp->unp_flags &= ~(bit); \ 1811 UNP_PCB_UNLOCK(unp); \ 1812 } while (0) 1813 1814 switch (sopt->sopt_name) { 1815 case LOCAL_CREDS: 1816 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1817 break; 1818 1819 case LOCAL_CREDS_PERSISTENT: 1820 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1821 break; 1822 1823 default: 1824 break; 1825 } 1826 break; 1827 #undef OPTSET 1828 default: 1829 error = ENOPROTOOPT; 1830 break; 1831 } 1832 break; 1833 1834 default: 1835 error = EOPNOTSUPP; 1836 break; 1837 } 1838 return (error); 1839 } 1840 1841 static int 1842 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1843 { 1844 1845 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1846 } 1847 1848 static int 1849 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1850 struct thread *td, bool return_locked) 1851 { 1852 struct mtx *vplock; 1853 struct sockaddr_un *soun; 1854 struct vnode *vp; 1855 struct socket *so2; 1856 struct unpcb *unp, *unp2, *unp3; 1857 struct nameidata nd; 1858 char buf[SOCK_MAXADDRLEN]; 1859 struct sockaddr *sa; 1860 cap_rights_t rights; 1861 int error, len; 1862 bool connreq; 1863 1864 if (nam->sa_family != AF_UNIX) 1865 return (EAFNOSUPPORT); 1866 if (nam->sa_len > sizeof(struct sockaddr_un)) 1867 return (EINVAL); 1868 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1869 if (len <= 0) 1870 return (EINVAL); 1871 soun = (struct sockaddr_un *)nam; 1872 bcopy(soun->sun_path, buf, len); 1873 buf[len] = 0; 1874 1875 error = 0; 1876 unp = sotounpcb(so); 1877 UNP_PCB_LOCK(unp); 1878 for (;;) { 1879 /* 1880 * Wait for connection state to stabilize. If a connection 1881 * already exists, give up. For datagram sockets, which permit 1882 * multiple consecutive connect(2) calls, upper layers are 1883 * responsible for disconnecting in advance of a subsequent 1884 * connect(2), but this is not synchronized with PCB connection 1885 * state. 1886 * 1887 * Also make sure that no threads are currently attempting to 1888 * lock the peer socket, to ensure that unp_conn cannot 1889 * transition between two valid sockets while locks are dropped. 1890 */ 1891 if (SOLISTENING(so)) 1892 error = EOPNOTSUPP; 1893 else if (unp->unp_conn != NULL) 1894 error = EISCONN; 1895 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1896 error = EALREADY; 1897 } 1898 if (error != 0) { 1899 UNP_PCB_UNLOCK(unp); 1900 return (error); 1901 } 1902 if (unp->unp_pairbusy > 0) { 1903 unp->unp_flags |= UNP_WAITING; 1904 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1905 continue; 1906 } 1907 break; 1908 } 1909 unp->unp_flags |= UNP_CONNECTING; 1910 UNP_PCB_UNLOCK(unp); 1911 1912 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1913 if (connreq) 1914 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1915 else 1916 sa = NULL; 1917 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1918 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1919 error = namei(&nd); 1920 if (error) 1921 vp = NULL; 1922 else 1923 vp = nd.ni_vp; 1924 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1925 if (error) 1926 goto bad; 1927 NDFREE_PNBUF(&nd); 1928 1929 if (vp->v_type != VSOCK) { 1930 error = ENOTSOCK; 1931 goto bad; 1932 } 1933 #ifdef MAC 1934 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1935 if (error) 1936 goto bad; 1937 #endif 1938 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1939 if (error) 1940 goto bad; 1941 1942 unp = sotounpcb(so); 1943 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1944 1945 vplock = mtx_pool_find(mtxpool_sleep, vp); 1946 mtx_lock(vplock); 1947 VOP_UNP_CONNECT(vp, &unp2); 1948 if (unp2 == NULL) { 1949 error = ECONNREFUSED; 1950 goto bad2; 1951 } 1952 so2 = unp2->unp_socket; 1953 if (so->so_type != so2->so_type) { 1954 error = EPROTOTYPE; 1955 goto bad2; 1956 } 1957 if (connreq) { 1958 if (SOLISTENING(so2)) { 1959 CURVNET_SET(so2->so_vnet); 1960 so2 = sonewconn(so2, 0); 1961 CURVNET_RESTORE(); 1962 } else 1963 so2 = NULL; 1964 if (so2 == NULL) { 1965 error = ECONNREFUSED; 1966 goto bad2; 1967 } 1968 unp3 = sotounpcb(so2); 1969 unp_pcb_lock_pair(unp2, unp3); 1970 if (unp2->unp_addr != NULL) { 1971 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1972 unp3->unp_addr = (struct sockaddr_un *) sa; 1973 sa = NULL; 1974 } 1975 1976 unp_copy_peercred(td, unp3, unp, unp2); 1977 1978 UNP_PCB_UNLOCK(unp2); 1979 unp2 = unp3; 1980 1981 /* 1982 * It is safe to block on the PCB lock here since unp2 is 1983 * nascent and cannot be connected to any other sockets. 1984 */ 1985 UNP_PCB_LOCK(unp); 1986 #ifdef MAC 1987 mac_socketpeer_set_from_socket(so, so2); 1988 mac_socketpeer_set_from_socket(so2, so); 1989 #endif 1990 } else { 1991 unp_pcb_lock_pair(unp, unp2); 1992 } 1993 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1994 sotounpcb(so2) == unp2, 1995 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1996 unp_connect2(so, so2); 1997 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1998 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1999 unp->unp_flags &= ~UNP_CONNECTING; 2000 if (!return_locked) 2001 unp_pcb_unlock_pair(unp, unp2); 2002 bad2: 2003 mtx_unlock(vplock); 2004 bad: 2005 if (vp != NULL) { 2006 /* 2007 * If we are returning locked (called via uipc_sosend_dgram()), 2008 * we need to be sure that vput() won't sleep. This is 2009 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2010 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2011 */ 2012 MPASS(!(return_locked && connreq)); 2013 vput(vp); 2014 } 2015 free(sa, M_SONAME); 2016 if (__predict_false(error)) { 2017 UNP_PCB_LOCK(unp); 2018 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2019 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2020 unp->unp_flags &= ~UNP_CONNECTING; 2021 UNP_PCB_UNLOCK(unp); 2022 } 2023 return (error); 2024 } 2025 2026 /* 2027 * Set socket peer credentials at connection time. 2028 * 2029 * The client's PCB credentials are copied from its process structure. The 2030 * server's PCB credentials are copied from the socket on which it called 2031 * listen(2). uipc_listen cached that process's credentials at the time. 2032 */ 2033 void 2034 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2035 struct unpcb *server_unp, struct unpcb *listen_unp) 2036 { 2037 cru2xt(td, &client_unp->unp_peercred); 2038 client_unp->unp_flags |= UNP_HAVEPC; 2039 2040 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2041 sizeof(server_unp->unp_peercred)); 2042 server_unp->unp_flags |= UNP_HAVEPC; 2043 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2044 } 2045 2046 static void 2047 unp_connect2(struct socket *so, struct socket *so2) 2048 { 2049 struct unpcb *unp; 2050 struct unpcb *unp2; 2051 2052 MPASS(so2->so_type == so->so_type); 2053 unp = sotounpcb(so); 2054 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2055 unp2 = sotounpcb(so2); 2056 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2057 2058 UNP_PCB_LOCK_ASSERT(unp); 2059 UNP_PCB_LOCK_ASSERT(unp2); 2060 KASSERT(unp->unp_conn == NULL, 2061 ("%s: socket %p is already connected", __func__, unp)); 2062 2063 unp->unp_conn = unp2; 2064 unp_pcb_hold(unp2); 2065 unp_pcb_hold(unp); 2066 switch (so->so_type) { 2067 case SOCK_DGRAM: 2068 UNP_REF_LIST_LOCK(); 2069 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2070 UNP_REF_LIST_UNLOCK(); 2071 soisconnected(so); 2072 break; 2073 2074 case SOCK_STREAM: 2075 case SOCK_SEQPACKET: 2076 KASSERT(unp2->unp_conn == NULL, 2077 ("%s: socket %p is already connected", __func__, unp2)); 2078 unp2->unp_conn = unp; 2079 soisconnected(so); 2080 soisconnected(so2); 2081 break; 2082 2083 default: 2084 panic("unp_connect2"); 2085 } 2086 } 2087 2088 static void 2089 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2090 { 2091 struct socket *so, *so2; 2092 struct mbuf *m = NULL; 2093 #ifdef INVARIANTS 2094 struct unpcb *unptmp; 2095 #endif 2096 2097 UNP_PCB_LOCK_ASSERT(unp); 2098 UNP_PCB_LOCK_ASSERT(unp2); 2099 KASSERT(unp->unp_conn == unp2, 2100 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2101 2102 unp->unp_conn = NULL; 2103 so = unp->unp_socket; 2104 so2 = unp2->unp_socket; 2105 switch (unp->unp_socket->so_type) { 2106 case SOCK_DGRAM: 2107 /* 2108 * Remove our send socket buffer from the peer's receive buffer. 2109 * Move the data to the receive buffer only if it is empty. 2110 * This is a protection against a scenario where a peer 2111 * connects, floods and disconnects, effectively blocking 2112 * sendto() from unconnected sockets. 2113 */ 2114 SOCK_RECVBUF_LOCK(so2); 2115 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2116 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2117 uxdg_clist); 2118 if (__predict_true((so2->so_rcv.sb_state & 2119 SBS_CANTRCVMORE) == 0) && 2120 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2121 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2122 &so->so_snd.uxdg_mb); 2123 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2124 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2125 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2126 } else { 2127 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2128 STAILQ_INIT(&so->so_snd.uxdg_mb); 2129 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2130 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2131 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2132 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2133 } 2134 /* Note: so may reconnect. */ 2135 so->so_snd.uxdg_cc = 0; 2136 so->so_snd.uxdg_ctl = 0; 2137 so->so_snd.uxdg_mbcnt = 0; 2138 } 2139 SOCK_RECVBUF_UNLOCK(so2); 2140 UNP_REF_LIST_LOCK(); 2141 #ifdef INVARIANTS 2142 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2143 if (unptmp == unp) 2144 break; 2145 } 2146 KASSERT(unptmp != NULL, 2147 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2148 #endif 2149 LIST_REMOVE(unp, unp_reflink); 2150 UNP_REF_LIST_UNLOCK(); 2151 if (so) { 2152 SOCK_LOCK(so); 2153 so->so_state &= ~SS_ISCONNECTED; 2154 SOCK_UNLOCK(so); 2155 } 2156 break; 2157 2158 case SOCK_STREAM: 2159 case SOCK_SEQPACKET: 2160 if (so) 2161 soisdisconnected(so); 2162 MPASS(unp2->unp_conn == unp); 2163 unp2->unp_conn = NULL; 2164 if (so2) 2165 soisdisconnected(so2); 2166 break; 2167 } 2168 2169 if (unp == unp2) { 2170 unp_pcb_rele_notlast(unp); 2171 if (!unp_pcb_rele(unp)) 2172 UNP_PCB_UNLOCK(unp); 2173 } else { 2174 if (!unp_pcb_rele(unp)) 2175 UNP_PCB_UNLOCK(unp); 2176 if (!unp_pcb_rele(unp2)) 2177 UNP_PCB_UNLOCK(unp2); 2178 } 2179 2180 if (m != NULL) { 2181 unp_scan(m, unp_freerights); 2182 m_freem(m); 2183 } 2184 } 2185 2186 /* 2187 * unp_pcblist() walks the global list of struct unpcb's to generate a 2188 * pointer list, bumping the refcount on each unpcb. It then copies them out 2189 * sequentially, validating the generation number on each to see if it has 2190 * been detached. All of this is necessary because copyout() may sleep on 2191 * disk I/O. 2192 */ 2193 static int 2194 unp_pcblist(SYSCTL_HANDLER_ARGS) 2195 { 2196 struct unpcb *unp, **unp_list; 2197 unp_gen_t gencnt; 2198 struct xunpgen *xug; 2199 struct unp_head *head; 2200 struct xunpcb *xu; 2201 u_int i; 2202 int error, n; 2203 2204 switch ((intptr_t)arg1) { 2205 case SOCK_STREAM: 2206 head = &unp_shead; 2207 break; 2208 2209 case SOCK_DGRAM: 2210 head = &unp_dhead; 2211 break; 2212 2213 case SOCK_SEQPACKET: 2214 head = &unp_sphead; 2215 break; 2216 2217 default: 2218 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2219 } 2220 2221 /* 2222 * The process of preparing the PCB list is too time-consuming and 2223 * resource-intensive to repeat twice on every request. 2224 */ 2225 if (req->oldptr == NULL) { 2226 n = unp_count; 2227 req->oldidx = 2 * (sizeof *xug) 2228 + (n + n/8) * sizeof(struct xunpcb); 2229 return (0); 2230 } 2231 2232 if (req->newptr != NULL) 2233 return (EPERM); 2234 2235 /* 2236 * OK, now we're committed to doing something. 2237 */ 2238 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2239 UNP_LINK_RLOCK(); 2240 gencnt = unp_gencnt; 2241 n = unp_count; 2242 UNP_LINK_RUNLOCK(); 2243 2244 xug->xug_len = sizeof *xug; 2245 xug->xug_count = n; 2246 xug->xug_gen = gencnt; 2247 xug->xug_sogen = so_gencnt; 2248 error = SYSCTL_OUT(req, xug, sizeof *xug); 2249 if (error) { 2250 free(xug, M_TEMP); 2251 return (error); 2252 } 2253 2254 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2255 2256 UNP_LINK_RLOCK(); 2257 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2258 unp = LIST_NEXT(unp, unp_link)) { 2259 UNP_PCB_LOCK(unp); 2260 if (unp->unp_gencnt <= gencnt) { 2261 if (cr_cansee(req->td->td_ucred, 2262 unp->unp_socket->so_cred)) { 2263 UNP_PCB_UNLOCK(unp); 2264 continue; 2265 } 2266 unp_list[i++] = unp; 2267 unp_pcb_hold(unp); 2268 } 2269 UNP_PCB_UNLOCK(unp); 2270 } 2271 UNP_LINK_RUNLOCK(); 2272 n = i; /* In case we lost some during malloc. */ 2273 2274 error = 0; 2275 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2276 for (i = 0; i < n; i++) { 2277 unp = unp_list[i]; 2278 UNP_PCB_LOCK(unp); 2279 if (unp_pcb_rele(unp)) 2280 continue; 2281 2282 if (unp->unp_gencnt <= gencnt) { 2283 xu->xu_len = sizeof *xu; 2284 xu->xu_unpp = (uintptr_t)unp; 2285 /* 2286 * XXX - need more locking here to protect against 2287 * connect/disconnect races for SMP. 2288 */ 2289 if (unp->unp_addr != NULL) 2290 bcopy(unp->unp_addr, &xu->xu_addr, 2291 unp->unp_addr->sun_len); 2292 else 2293 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2294 if (unp->unp_conn != NULL && 2295 unp->unp_conn->unp_addr != NULL) 2296 bcopy(unp->unp_conn->unp_addr, 2297 &xu->xu_caddr, 2298 unp->unp_conn->unp_addr->sun_len); 2299 else 2300 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2301 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2302 xu->unp_conn = (uintptr_t)unp->unp_conn; 2303 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2304 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2305 xu->unp_gencnt = unp->unp_gencnt; 2306 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2307 UNP_PCB_UNLOCK(unp); 2308 error = SYSCTL_OUT(req, xu, sizeof *xu); 2309 } else { 2310 UNP_PCB_UNLOCK(unp); 2311 } 2312 } 2313 free(xu, M_TEMP); 2314 if (!error) { 2315 /* 2316 * Give the user an updated idea of our state. If the 2317 * generation differs from what we told her before, she knows 2318 * that something happened while we were processing this 2319 * request, and it might be necessary to retry. 2320 */ 2321 xug->xug_gen = unp_gencnt; 2322 xug->xug_sogen = so_gencnt; 2323 xug->xug_count = unp_count; 2324 error = SYSCTL_OUT(req, xug, sizeof *xug); 2325 } 2326 free(unp_list, M_TEMP); 2327 free(xug, M_TEMP); 2328 return (error); 2329 } 2330 2331 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2332 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2333 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2334 "List of active local datagram sockets"); 2335 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2336 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2337 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2338 "List of active local stream sockets"); 2339 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2340 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2341 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2342 "List of active local seqpacket sockets"); 2343 2344 static void 2345 unp_shutdown(struct unpcb *unp) 2346 { 2347 struct unpcb *unp2; 2348 struct socket *so; 2349 2350 UNP_PCB_LOCK_ASSERT(unp); 2351 2352 unp2 = unp->unp_conn; 2353 if ((unp->unp_socket->so_type == SOCK_STREAM || 2354 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2355 so = unp2->unp_socket; 2356 if (so != NULL) 2357 socantrcvmore(so); 2358 } 2359 } 2360 2361 static void 2362 unp_drop(struct unpcb *unp) 2363 { 2364 struct socket *so; 2365 struct unpcb *unp2; 2366 2367 /* 2368 * Regardless of whether the socket's peer dropped the connection 2369 * with this socket by aborting or disconnecting, POSIX requires 2370 * that ECONNRESET is returned. 2371 */ 2372 2373 UNP_PCB_LOCK(unp); 2374 so = unp->unp_socket; 2375 if (so) 2376 so->so_error = ECONNRESET; 2377 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2378 /* Last reference dropped in unp_disconnect(). */ 2379 unp_pcb_rele_notlast(unp); 2380 unp_disconnect(unp, unp2); 2381 } else if (!unp_pcb_rele(unp)) { 2382 UNP_PCB_UNLOCK(unp); 2383 } 2384 } 2385 2386 static void 2387 unp_freerights(struct filedescent **fdep, int fdcount) 2388 { 2389 struct file *fp; 2390 int i; 2391 2392 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2393 2394 for (i = 0; i < fdcount; i++) { 2395 fp = fdep[i]->fde_file; 2396 filecaps_free(&fdep[i]->fde_caps); 2397 unp_discard(fp); 2398 } 2399 free(fdep[0], M_FILECAPS); 2400 } 2401 2402 static int 2403 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2404 { 2405 struct thread *td = curthread; /* XXX */ 2406 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2407 int i; 2408 int *fdp; 2409 struct filedesc *fdesc = td->td_proc->p_fd; 2410 struct filedescent **fdep; 2411 void *data; 2412 socklen_t clen = control->m_len, datalen; 2413 int error, newfds; 2414 u_int newlen; 2415 2416 UNP_LINK_UNLOCK_ASSERT(); 2417 2418 error = 0; 2419 if (controlp != NULL) /* controlp == NULL => free control messages */ 2420 *controlp = NULL; 2421 while (cm != NULL) { 2422 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2423 2424 data = CMSG_DATA(cm); 2425 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2426 if (cm->cmsg_level == SOL_SOCKET 2427 && cm->cmsg_type == SCM_RIGHTS) { 2428 newfds = datalen / sizeof(*fdep); 2429 if (newfds == 0) 2430 goto next; 2431 fdep = data; 2432 2433 /* If we're not outputting the descriptors free them. */ 2434 if (error || controlp == NULL) { 2435 unp_freerights(fdep, newfds); 2436 goto next; 2437 } 2438 FILEDESC_XLOCK(fdesc); 2439 2440 /* 2441 * Now change each pointer to an fd in the global 2442 * table to an integer that is the index to the local 2443 * fd table entry that we set up to point to the 2444 * global one we are transferring. 2445 */ 2446 newlen = newfds * sizeof(int); 2447 *controlp = sbcreatecontrol(NULL, newlen, 2448 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2449 2450 fdp = (int *) 2451 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2452 if ((error = fdallocn(td, 0, fdp, newfds))) { 2453 FILEDESC_XUNLOCK(fdesc); 2454 unp_freerights(fdep, newfds); 2455 m_freem(*controlp); 2456 *controlp = NULL; 2457 goto next; 2458 } 2459 for (i = 0; i < newfds; i++, fdp++) { 2460 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2461 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2462 &fdep[i]->fde_caps); 2463 unp_externalize_fp(fdep[i]->fde_file); 2464 } 2465 2466 /* 2467 * The new type indicates that the mbuf data refers to 2468 * kernel resources that may need to be released before 2469 * the mbuf is freed. 2470 */ 2471 m_chtype(*controlp, MT_EXTCONTROL); 2472 FILEDESC_XUNLOCK(fdesc); 2473 free(fdep[0], M_FILECAPS); 2474 } else { 2475 /* We can just copy anything else across. */ 2476 if (error || controlp == NULL) 2477 goto next; 2478 *controlp = sbcreatecontrol(NULL, datalen, 2479 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2480 bcopy(data, 2481 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2482 datalen); 2483 } 2484 controlp = &(*controlp)->m_next; 2485 2486 next: 2487 if (CMSG_SPACE(datalen) < clen) { 2488 clen -= CMSG_SPACE(datalen); 2489 cm = (struct cmsghdr *) 2490 ((caddr_t)cm + CMSG_SPACE(datalen)); 2491 } else { 2492 clen = 0; 2493 cm = NULL; 2494 } 2495 } 2496 2497 m_freem(control); 2498 return (error); 2499 } 2500 2501 static void 2502 unp_zone_change(void *tag) 2503 { 2504 2505 uma_zone_set_max(unp_zone, maxsockets); 2506 } 2507 2508 #ifdef INVARIANTS 2509 static void 2510 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2511 { 2512 struct unpcb *unp; 2513 2514 unp = mem; 2515 2516 KASSERT(LIST_EMPTY(&unp->unp_refs), 2517 ("%s: unpcb %p has lingering refs", __func__, unp)); 2518 KASSERT(unp->unp_socket == NULL, 2519 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2520 KASSERT(unp->unp_vnode == NULL, 2521 ("%s: unpcb %p has vnode references", __func__, unp)); 2522 KASSERT(unp->unp_conn == NULL, 2523 ("%s: unpcb %p is still connected", __func__, unp)); 2524 KASSERT(unp->unp_addr == NULL, 2525 ("%s: unpcb %p has leaked addr", __func__, unp)); 2526 } 2527 #endif 2528 2529 static void 2530 unp_init(void *arg __unused) 2531 { 2532 uma_dtor dtor; 2533 2534 #ifdef INVARIANTS 2535 dtor = unp_zdtor; 2536 #else 2537 dtor = NULL; 2538 #endif 2539 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2540 NULL, NULL, UMA_ALIGN_CACHE, 0); 2541 uma_zone_set_max(unp_zone, maxsockets); 2542 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2543 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2544 NULL, EVENTHANDLER_PRI_ANY); 2545 LIST_INIT(&unp_dhead); 2546 LIST_INIT(&unp_shead); 2547 LIST_INIT(&unp_sphead); 2548 SLIST_INIT(&unp_defers); 2549 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2550 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2551 UNP_LINK_LOCK_INIT(); 2552 UNP_DEFERRED_LOCK_INIT(); 2553 } 2554 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2555 2556 static void 2557 unp_internalize_cleanup_rights(struct mbuf *control) 2558 { 2559 struct cmsghdr *cp; 2560 struct mbuf *m; 2561 void *data; 2562 socklen_t datalen; 2563 2564 for (m = control; m != NULL; m = m->m_next) { 2565 cp = mtod(m, struct cmsghdr *); 2566 if (cp->cmsg_level != SOL_SOCKET || 2567 cp->cmsg_type != SCM_RIGHTS) 2568 continue; 2569 data = CMSG_DATA(cp); 2570 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2571 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2572 } 2573 } 2574 2575 static int 2576 unp_internalize(struct mbuf **controlp, struct thread *td, 2577 struct mbuf **clast, u_int *space, u_int *mbcnt) 2578 { 2579 struct mbuf *control, **initial_controlp; 2580 struct proc *p; 2581 struct filedesc *fdesc; 2582 struct bintime *bt; 2583 struct cmsghdr *cm; 2584 struct cmsgcred *cmcred; 2585 struct filedescent *fde, **fdep, *fdev; 2586 struct file *fp; 2587 struct timeval *tv; 2588 struct timespec *ts; 2589 void *data; 2590 socklen_t clen, datalen; 2591 int i, j, error, *fdp, oldfds; 2592 u_int newlen; 2593 2594 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2595 UNP_LINK_UNLOCK_ASSERT(); 2596 2597 p = td->td_proc; 2598 fdesc = p->p_fd; 2599 error = 0; 2600 control = *controlp; 2601 *controlp = NULL; 2602 initial_controlp = controlp; 2603 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2604 data = CMSG_DATA(cm); 2605 2606 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2607 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2608 (char *)cm + cm->cmsg_len >= (char *)data; 2609 2610 clen -= min(CMSG_SPACE(datalen), clen), 2611 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2612 data = CMSG_DATA(cm)) { 2613 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2614 switch (cm->cmsg_type) { 2615 case SCM_CREDS: 2616 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2617 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2618 cmcred = (struct cmsgcred *) 2619 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2620 cmcred->cmcred_pid = p->p_pid; 2621 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2622 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2623 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2624 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2625 CMGROUP_MAX); 2626 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2627 cmcred->cmcred_groups[i] = 2628 td->td_ucred->cr_groups[i]; 2629 break; 2630 2631 case SCM_RIGHTS: 2632 oldfds = datalen / sizeof (int); 2633 if (oldfds == 0) 2634 continue; 2635 /* On some machines sizeof pointer is bigger than 2636 * sizeof int, so we need to check if data fits into 2637 * single mbuf. We could allocate several mbufs, and 2638 * unp_externalize() should even properly handle that. 2639 * But it is not worth to complicate the code for an 2640 * insane scenario of passing over 200 file descriptors 2641 * at once. 2642 */ 2643 newlen = oldfds * sizeof(fdep[0]); 2644 if (CMSG_SPACE(newlen) > MCLBYTES) { 2645 error = EMSGSIZE; 2646 goto out; 2647 } 2648 /* 2649 * Check that all the FDs passed in refer to legal 2650 * files. If not, reject the entire operation. 2651 */ 2652 fdp = data; 2653 FILEDESC_SLOCK(fdesc); 2654 for (i = 0; i < oldfds; i++, fdp++) { 2655 fp = fget_noref(fdesc, *fdp); 2656 if (fp == NULL) { 2657 FILEDESC_SUNLOCK(fdesc); 2658 error = EBADF; 2659 goto out; 2660 } 2661 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2662 FILEDESC_SUNLOCK(fdesc); 2663 error = EOPNOTSUPP; 2664 goto out; 2665 } 2666 } 2667 2668 /* 2669 * Now replace the integer FDs with pointers to the 2670 * file structure and capability rights. 2671 */ 2672 *controlp = sbcreatecontrol(NULL, newlen, 2673 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2674 fdp = data; 2675 for (i = 0; i < oldfds; i++, fdp++) { 2676 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2677 fdp = data; 2678 for (j = 0; j < i; j++, fdp++) { 2679 fdrop(fdesc->fd_ofiles[*fdp]. 2680 fde_file, td); 2681 } 2682 FILEDESC_SUNLOCK(fdesc); 2683 error = EBADF; 2684 goto out; 2685 } 2686 } 2687 fdp = data; 2688 fdep = (struct filedescent **) 2689 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2690 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2691 M_WAITOK); 2692 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2693 fde = &fdesc->fd_ofiles[*fdp]; 2694 fdep[i] = fdev; 2695 fdep[i]->fde_file = fde->fde_file; 2696 filecaps_copy(&fde->fde_caps, 2697 &fdep[i]->fde_caps, true); 2698 unp_internalize_fp(fdep[i]->fde_file); 2699 } 2700 FILEDESC_SUNLOCK(fdesc); 2701 break; 2702 2703 case SCM_TIMESTAMP: 2704 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2705 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2706 tv = (struct timeval *) 2707 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2708 microtime(tv); 2709 break; 2710 2711 case SCM_BINTIME: 2712 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2713 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2714 bt = (struct bintime *) 2715 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2716 bintime(bt); 2717 break; 2718 2719 case SCM_REALTIME: 2720 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2721 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2722 ts = (struct timespec *) 2723 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2724 nanotime(ts); 2725 break; 2726 2727 case SCM_MONOTONIC: 2728 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2729 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2730 ts = (struct timespec *) 2731 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2732 nanouptime(ts); 2733 break; 2734 2735 default: 2736 error = EINVAL; 2737 goto out; 2738 } 2739 2740 if (space != NULL) { 2741 *space += (*controlp)->m_len; 2742 *mbcnt += MSIZE; 2743 if ((*controlp)->m_flags & M_EXT) 2744 *mbcnt += (*controlp)->m_ext.ext_size; 2745 *clast = *controlp; 2746 } 2747 controlp = &(*controlp)->m_next; 2748 } 2749 if (clen > 0) 2750 error = EINVAL; 2751 2752 out: 2753 if (error != 0 && initial_controlp != NULL) 2754 unp_internalize_cleanup_rights(*initial_controlp); 2755 m_freem(control); 2756 return (error); 2757 } 2758 2759 static struct mbuf * 2760 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2761 struct mbuf **clast, u_int *space, u_int *mbcnt) 2762 { 2763 struct mbuf *m, *n, *n_prev; 2764 const struct cmsghdr *cm; 2765 int ngroups, i, cmsgtype; 2766 size_t ctrlsz; 2767 2768 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2769 if (mode & UNP_WANTCRED_ALWAYS) { 2770 ctrlsz = SOCKCRED2SIZE(ngroups); 2771 cmsgtype = SCM_CREDS2; 2772 } else { 2773 ctrlsz = SOCKCREDSIZE(ngroups); 2774 cmsgtype = SCM_CREDS; 2775 } 2776 2777 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2778 if (m == NULL) 2779 return (control); 2780 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2781 2782 if (mode & UNP_WANTCRED_ALWAYS) { 2783 struct sockcred2 *sc; 2784 2785 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2786 sc->sc_version = 0; 2787 sc->sc_pid = td->td_proc->p_pid; 2788 sc->sc_uid = td->td_ucred->cr_ruid; 2789 sc->sc_euid = td->td_ucred->cr_uid; 2790 sc->sc_gid = td->td_ucred->cr_rgid; 2791 sc->sc_egid = td->td_ucred->cr_gid; 2792 sc->sc_ngroups = ngroups; 2793 for (i = 0; i < sc->sc_ngroups; i++) 2794 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2795 } else { 2796 struct sockcred *sc; 2797 2798 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2799 sc->sc_uid = td->td_ucred->cr_ruid; 2800 sc->sc_euid = td->td_ucred->cr_uid; 2801 sc->sc_gid = td->td_ucred->cr_rgid; 2802 sc->sc_egid = td->td_ucred->cr_gid; 2803 sc->sc_ngroups = ngroups; 2804 for (i = 0; i < sc->sc_ngroups; i++) 2805 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2806 } 2807 2808 /* 2809 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2810 * created SCM_CREDS control message (struct sockcred) has another 2811 * format. 2812 */ 2813 if (control != NULL && cmsgtype == SCM_CREDS) 2814 for (n = control, n_prev = NULL; n != NULL;) { 2815 cm = mtod(n, struct cmsghdr *); 2816 if (cm->cmsg_level == SOL_SOCKET && 2817 cm->cmsg_type == SCM_CREDS) { 2818 if (n_prev == NULL) 2819 control = n->m_next; 2820 else 2821 n_prev->m_next = n->m_next; 2822 if (space != NULL) { 2823 MPASS(*space >= n->m_len); 2824 *space -= n->m_len; 2825 MPASS(*mbcnt >= MSIZE); 2826 *mbcnt -= MSIZE; 2827 if (n->m_flags & M_EXT) { 2828 MPASS(*mbcnt >= 2829 n->m_ext.ext_size); 2830 *mbcnt -= n->m_ext.ext_size; 2831 } 2832 MPASS(clast); 2833 if (*clast == n) { 2834 MPASS(n->m_next == NULL); 2835 if (n_prev == NULL) 2836 *clast = m; 2837 else 2838 *clast = n_prev; 2839 } 2840 } 2841 n = m_free(n); 2842 } else { 2843 n_prev = n; 2844 n = n->m_next; 2845 } 2846 } 2847 2848 /* Prepend it to the head. */ 2849 m->m_next = control; 2850 if (space != NULL) { 2851 *space += m->m_len; 2852 *mbcnt += MSIZE; 2853 if (control == NULL) 2854 *clast = m; 2855 } 2856 return (m); 2857 } 2858 2859 static struct unpcb * 2860 fptounp(struct file *fp) 2861 { 2862 struct socket *so; 2863 2864 if (fp->f_type != DTYPE_SOCKET) 2865 return (NULL); 2866 if ((so = fp->f_data) == NULL) 2867 return (NULL); 2868 if (so->so_proto->pr_domain != &localdomain) 2869 return (NULL); 2870 return sotounpcb(so); 2871 } 2872 2873 static void 2874 unp_discard(struct file *fp) 2875 { 2876 struct unp_defer *dr; 2877 2878 if (unp_externalize_fp(fp)) { 2879 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2880 dr->ud_fp = fp; 2881 UNP_DEFERRED_LOCK(); 2882 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2883 UNP_DEFERRED_UNLOCK(); 2884 atomic_add_int(&unp_defers_count, 1); 2885 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2886 } else 2887 closef_nothread(fp); 2888 } 2889 2890 static void 2891 unp_process_defers(void *arg __unused, int pending) 2892 { 2893 struct unp_defer *dr; 2894 SLIST_HEAD(, unp_defer) drl; 2895 int count; 2896 2897 SLIST_INIT(&drl); 2898 for (;;) { 2899 UNP_DEFERRED_LOCK(); 2900 if (SLIST_FIRST(&unp_defers) == NULL) { 2901 UNP_DEFERRED_UNLOCK(); 2902 break; 2903 } 2904 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2905 UNP_DEFERRED_UNLOCK(); 2906 count = 0; 2907 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2908 SLIST_REMOVE_HEAD(&drl, ud_link); 2909 closef_nothread(dr->ud_fp); 2910 free(dr, M_TEMP); 2911 count++; 2912 } 2913 atomic_add_int(&unp_defers_count, -count); 2914 } 2915 } 2916 2917 static void 2918 unp_internalize_fp(struct file *fp) 2919 { 2920 struct unpcb *unp; 2921 2922 UNP_LINK_WLOCK(); 2923 if ((unp = fptounp(fp)) != NULL) { 2924 unp->unp_file = fp; 2925 unp->unp_msgcount++; 2926 } 2927 unp_rights++; 2928 UNP_LINK_WUNLOCK(); 2929 } 2930 2931 static int 2932 unp_externalize_fp(struct file *fp) 2933 { 2934 struct unpcb *unp; 2935 int ret; 2936 2937 UNP_LINK_WLOCK(); 2938 if ((unp = fptounp(fp)) != NULL) { 2939 unp->unp_msgcount--; 2940 ret = 1; 2941 } else 2942 ret = 0; 2943 unp_rights--; 2944 UNP_LINK_WUNLOCK(); 2945 return (ret); 2946 } 2947 2948 /* 2949 * unp_defer indicates whether additional work has been defered for a future 2950 * pass through unp_gc(). It is thread local and does not require explicit 2951 * synchronization. 2952 */ 2953 static int unp_marked; 2954 2955 static void 2956 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2957 { 2958 struct unpcb *unp; 2959 struct file *fp; 2960 int i; 2961 2962 /* 2963 * This function can only be called from the gc task. 2964 */ 2965 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2966 ("%s: not on gc callout", __func__)); 2967 UNP_LINK_LOCK_ASSERT(); 2968 2969 for (i = 0; i < fdcount; i++) { 2970 fp = fdep[i]->fde_file; 2971 if ((unp = fptounp(fp)) == NULL) 2972 continue; 2973 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2974 continue; 2975 unp->unp_gcrefs--; 2976 } 2977 } 2978 2979 static void 2980 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2981 { 2982 struct unpcb *unp; 2983 struct file *fp; 2984 int i; 2985 2986 /* 2987 * This function can only be called from the gc task. 2988 */ 2989 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2990 ("%s: not on gc callout", __func__)); 2991 UNP_LINK_LOCK_ASSERT(); 2992 2993 for (i = 0; i < fdcount; i++) { 2994 fp = fdep[i]->fde_file; 2995 if ((unp = fptounp(fp)) == NULL) 2996 continue; 2997 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2998 continue; 2999 unp->unp_gcrefs++; 3000 unp_marked++; 3001 } 3002 } 3003 3004 static void 3005 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3006 { 3007 struct sockbuf *sb; 3008 3009 SOCK_LOCK_ASSERT(so); 3010 3011 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3012 return; 3013 3014 SOCK_RECVBUF_LOCK(so); 3015 switch (so->so_type) { 3016 case SOCK_DGRAM: 3017 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3018 unp_scan(so->so_rcv.uxdg_peeked, op); 3019 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3020 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3021 break; 3022 case SOCK_STREAM: 3023 case SOCK_SEQPACKET: 3024 unp_scan(so->so_rcv.sb_mb, op); 3025 break; 3026 } 3027 SOCK_RECVBUF_UNLOCK(so); 3028 } 3029 3030 static void 3031 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3032 { 3033 struct socket *so, *soa; 3034 3035 so = unp->unp_socket; 3036 SOCK_LOCK(so); 3037 if (SOLISTENING(so)) { 3038 /* 3039 * Mark all sockets in our accept queue. 3040 */ 3041 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3042 unp_scan_socket(soa, op); 3043 } else { 3044 /* 3045 * Mark all sockets we reference with RIGHTS. 3046 */ 3047 unp_scan_socket(so, op); 3048 } 3049 SOCK_UNLOCK(so); 3050 } 3051 3052 static int unp_recycled; 3053 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3054 "Number of unreachable sockets claimed by the garbage collector."); 3055 3056 static int unp_taskcount; 3057 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3058 "Number of times the garbage collector has run."); 3059 3060 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3061 "Number of active local sockets."); 3062 3063 static void 3064 unp_gc(__unused void *arg, int pending) 3065 { 3066 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3067 NULL }; 3068 struct unp_head **head; 3069 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3070 struct file *f, **unref; 3071 struct unpcb *unp, *unptmp; 3072 int i, total, unp_unreachable; 3073 3074 LIST_INIT(&unp_deadhead); 3075 unp_taskcount++; 3076 UNP_LINK_RLOCK(); 3077 /* 3078 * First determine which sockets may be in cycles. 3079 */ 3080 unp_unreachable = 0; 3081 3082 for (head = heads; *head != NULL; head++) 3083 LIST_FOREACH(unp, *head, unp_link) { 3084 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3085 ("%s: unp %p has unexpected gc flags 0x%x", 3086 __func__, unp, (unsigned int)unp->unp_gcflag)); 3087 3088 f = unp->unp_file; 3089 3090 /* 3091 * Check for an unreachable socket potentially in a 3092 * cycle. It must be in a queue as indicated by 3093 * msgcount, and this must equal the file reference 3094 * count. Note that when msgcount is 0 the file is 3095 * NULL. 3096 */ 3097 if (f != NULL && unp->unp_msgcount != 0 && 3098 refcount_load(&f->f_count) == unp->unp_msgcount) { 3099 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3100 unp->unp_gcflag |= UNPGC_DEAD; 3101 unp->unp_gcrefs = unp->unp_msgcount; 3102 unp_unreachable++; 3103 } 3104 } 3105 3106 /* 3107 * Scan all sockets previously marked as potentially being in a cycle 3108 * and remove the references each socket holds on any UNPGC_DEAD 3109 * sockets in its queue. After this step, all remaining references on 3110 * sockets marked UNPGC_DEAD should not be part of any cycle. 3111 */ 3112 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3113 unp_gc_scan(unp, unp_remove_dead_ref); 3114 3115 /* 3116 * If a socket still has a non-negative refcount, it cannot be in a 3117 * cycle. In this case increment refcount of all children iteratively. 3118 * Stop the scan once we do a complete loop without discovering 3119 * a new reachable socket. 3120 */ 3121 do { 3122 unp_marked = 0; 3123 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3124 if (unp->unp_gcrefs > 0) { 3125 unp->unp_gcflag &= ~UNPGC_DEAD; 3126 LIST_REMOVE(unp, unp_dead); 3127 KASSERT(unp_unreachable > 0, 3128 ("%s: unp_unreachable underflow.", 3129 __func__)); 3130 unp_unreachable--; 3131 unp_gc_scan(unp, unp_restore_undead_ref); 3132 } 3133 } while (unp_marked); 3134 3135 UNP_LINK_RUNLOCK(); 3136 3137 if (unp_unreachable == 0) 3138 return; 3139 3140 /* 3141 * Allocate space for a local array of dead unpcbs. 3142 * TODO: can this path be simplified by instead using the local 3143 * dead list at unp_deadhead, after taking out references 3144 * on the file object and/or unpcb and dropping the link lock? 3145 */ 3146 unref = malloc(unp_unreachable * sizeof(struct file *), 3147 M_TEMP, M_WAITOK); 3148 3149 /* 3150 * Iterate looking for sockets which have been specifically marked 3151 * as unreachable and store them locally. 3152 */ 3153 UNP_LINK_RLOCK(); 3154 total = 0; 3155 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3156 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3157 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3158 unp->unp_gcflag &= ~UNPGC_DEAD; 3159 f = unp->unp_file; 3160 if (unp->unp_msgcount == 0 || f == NULL || 3161 refcount_load(&f->f_count) != unp->unp_msgcount || 3162 !fhold(f)) 3163 continue; 3164 unref[total++] = f; 3165 KASSERT(total <= unp_unreachable, 3166 ("%s: incorrect unreachable count.", __func__)); 3167 } 3168 UNP_LINK_RUNLOCK(); 3169 3170 /* 3171 * Now flush all sockets, free'ing rights. This will free the 3172 * struct files associated with these sockets but leave each socket 3173 * with one remaining ref. 3174 */ 3175 for (i = 0; i < total; i++) { 3176 struct socket *so; 3177 3178 so = unref[i]->f_data; 3179 CURVNET_SET(so->so_vnet); 3180 socantrcvmore(so); 3181 unp_dispose(so); 3182 CURVNET_RESTORE(); 3183 } 3184 3185 /* 3186 * And finally release the sockets so they can be reclaimed. 3187 */ 3188 for (i = 0; i < total; i++) 3189 fdrop(unref[i], NULL); 3190 unp_recycled += total; 3191 free(unref, M_TEMP); 3192 } 3193 3194 /* 3195 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3196 */ 3197 static void 3198 unp_dispose(struct socket *so) 3199 { 3200 struct sockbuf *sb; 3201 struct unpcb *unp; 3202 struct mbuf *m; 3203 int error __diagused; 3204 3205 MPASS(!SOLISTENING(so)); 3206 3207 unp = sotounpcb(so); 3208 UNP_LINK_WLOCK(); 3209 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3210 UNP_LINK_WUNLOCK(); 3211 3212 /* 3213 * Grab our special mbufs before calling sbrelease(). 3214 */ 3215 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 3216 MPASS(!error); 3217 SOCK_RECVBUF_LOCK(so); 3218 switch (so->so_type) { 3219 case SOCK_DGRAM: 3220 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3221 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3222 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3223 /* Note: socket of sb may reconnect. */ 3224 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3225 } 3226 sb = &so->so_rcv; 3227 if (sb->uxdg_peeked != NULL) { 3228 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3229 m_stailqpkt); 3230 sb->uxdg_peeked = NULL; 3231 } 3232 m = STAILQ_FIRST(&sb->uxdg_mb); 3233 STAILQ_INIT(&sb->uxdg_mb); 3234 /* XXX: our shortened sbrelease() */ 3235 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3236 RLIM_INFINITY); 3237 /* 3238 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3239 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3240 * data to the socket. 3241 * We came here either through shutdown(2) or from the final 3242 * sofree(). The sofree() case is simple as it guarantees 3243 * that no more sends will happen, however we can race with 3244 * unp_disconnect() from our peer. The shutdown(2) case is 3245 * more exotic. It would call into unp_dispose() only if 3246 * socket is SS_ISCONNECTED. This is possible if we did 3247 * connect(2) on this socket and we also had it bound with 3248 * bind(2) and receive connections from other sockets. 3249 * Because uipc_shutdown() violates POSIX (see comment 3250 * there) we will end up here shutting down our receive side. 3251 * Of course this will have affect not only on the peer we 3252 * connect(2)ed to, but also on all of the peers who had 3253 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3254 */ 3255 sb->sb_state |= SBS_CANTRCVMORE; 3256 break; 3257 case SOCK_STREAM: 3258 case SOCK_SEQPACKET: 3259 sb = &so->so_rcv; 3260 m = sbcut_locked(sb, sb->sb_ccc); 3261 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3262 ("%s: ccc %u mb %p mbcnt %u", __func__, 3263 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3264 sbrelease_locked(so, SO_RCV); 3265 break; 3266 } 3267 SOCK_RECVBUF_UNLOCK(so); 3268 SOCK_IO_RECV_UNLOCK(so); 3269 3270 if (m != NULL) { 3271 unp_scan(m, unp_freerights); 3272 m_freem(m); 3273 } 3274 } 3275 3276 static void 3277 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3278 { 3279 struct mbuf *m; 3280 struct cmsghdr *cm; 3281 void *data; 3282 socklen_t clen, datalen; 3283 3284 while (m0 != NULL) { 3285 for (m = m0; m; m = m->m_next) { 3286 if (m->m_type != MT_CONTROL) 3287 continue; 3288 3289 cm = mtod(m, struct cmsghdr *); 3290 clen = m->m_len; 3291 3292 while (cm != NULL) { 3293 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3294 break; 3295 3296 data = CMSG_DATA(cm); 3297 datalen = (caddr_t)cm + cm->cmsg_len 3298 - (caddr_t)data; 3299 3300 if (cm->cmsg_level == SOL_SOCKET && 3301 cm->cmsg_type == SCM_RIGHTS) { 3302 (*op)(data, datalen / 3303 sizeof(struct filedescent *)); 3304 } 3305 3306 if (CMSG_SPACE(datalen) < clen) { 3307 clen -= CMSG_SPACE(datalen); 3308 cm = (struct cmsghdr *) 3309 ((caddr_t)cm + CMSG_SPACE(datalen)); 3310 } else { 3311 clen = 0; 3312 cm = NULL; 3313 } 3314 } 3315 } 3316 m0 = m0->m_nextpkt; 3317 } 3318 } 3319 3320 /* 3321 * Definitions of protocols supported in the LOCAL domain. 3322 */ 3323 static struct protosw streamproto = { 3324 .pr_type = SOCK_STREAM, 3325 .pr_flags = PR_CONNREQUIRED | PR_WANTRCVD | PR_CAPATTACH, 3326 .pr_ctloutput = &uipc_ctloutput, 3327 .pr_abort = uipc_abort, 3328 .pr_accept = uipc_peeraddr, 3329 .pr_attach = uipc_attach, 3330 .pr_bind = uipc_bind, 3331 .pr_bindat = uipc_bindat, 3332 .pr_connect = uipc_connect, 3333 .pr_connectat = uipc_connectat, 3334 .pr_connect2 = uipc_connect2, 3335 .pr_detach = uipc_detach, 3336 .pr_disconnect = uipc_disconnect, 3337 .pr_listen = uipc_listen, 3338 .pr_peeraddr = uipc_peeraddr, 3339 .pr_rcvd = uipc_rcvd, 3340 .pr_send = uipc_send, 3341 .pr_ready = uipc_ready, 3342 .pr_sense = uipc_sense, 3343 .pr_shutdown = uipc_shutdown, 3344 .pr_sockaddr = uipc_sockaddr, 3345 .pr_soreceive = soreceive_generic, 3346 .pr_close = uipc_close, 3347 }; 3348 3349 static struct protosw dgramproto = { 3350 .pr_type = SOCK_DGRAM, 3351 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH | PR_SOCKBUF, 3352 .pr_ctloutput = &uipc_ctloutput, 3353 .pr_abort = uipc_abort, 3354 .pr_accept = uipc_peeraddr, 3355 .pr_attach = uipc_attach, 3356 .pr_bind = uipc_bind, 3357 .pr_bindat = uipc_bindat, 3358 .pr_connect = uipc_connect, 3359 .pr_connectat = uipc_connectat, 3360 .pr_connect2 = uipc_connect2, 3361 .pr_detach = uipc_detach, 3362 .pr_disconnect = uipc_disconnect, 3363 .pr_peeraddr = uipc_peeraddr, 3364 .pr_sosend = uipc_sosend_dgram, 3365 .pr_sense = uipc_sense, 3366 .pr_shutdown = uipc_shutdown, 3367 .pr_sockaddr = uipc_sockaddr, 3368 .pr_soreceive = uipc_soreceive_dgram, 3369 .pr_close = uipc_close, 3370 }; 3371 3372 static struct protosw seqpacketproto = { 3373 .pr_type = SOCK_SEQPACKET, 3374 /* 3375 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3376 * due to our use of sbappendaddr. A new sbappend variants is needed 3377 * that supports both atomic record writes and control data. 3378 */ 3379 .pr_flags = PR_ADDR | PR_ATOMIC | PR_CONNREQUIRED | 3380 PR_WANTRCVD | PR_CAPATTACH, 3381 .pr_ctloutput = &uipc_ctloutput, 3382 .pr_abort = uipc_abort, 3383 .pr_accept = uipc_peeraddr, 3384 .pr_attach = uipc_attach, 3385 .pr_bind = uipc_bind, 3386 .pr_bindat = uipc_bindat, 3387 .pr_connect = uipc_connect, 3388 .pr_connectat = uipc_connectat, 3389 .pr_connect2 = uipc_connect2, 3390 .pr_detach = uipc_detach, 3391 .pr_disconnect = uipc_disconnect, 3392 .pr_listen = uipc_listen, 3393 .pr_peeraddr = uipc_peeraddr, 3394 .pr_rcvd = uipc_rcvd, 3395 .pr_send = uipc_send, 3396 .pr_sense = uipc_sense, 3397 .pr_shutdown = uipc_shutdown, 3398 .pr_sockaddr = uipc_sockaddr, 3399 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3400 .pr_close = uipc_close, 3401 }; 3402 3403 static struct domain localdomain = { 3404 .dom_family = AF_LOCAL, 3405 .dom_name = "local", 3406 .dom_externalize = unp_externalize, 3407 .dom_nprotosw = 3, 3408 .dom_protosw = { 3409 &streamproto, 3410 &dgramproto, 3411 &seqpacketproto, 3412 } 3413 }; 3414 DOMAIN_SET(local); 3415 3416 /* 3417 * A helper function called by VFS before socket-type vnode reclamation. 3418 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3419 * use count. 3420 */ 3421 void 3422 vfs_unp_reclaim(struct vnode *vp) 3423 { 3424 struct unpcb *unp; 3425 int active; 3426 struct mtx *vplock; 3427 3428 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3429 KASSERT(vp->v_type == VSOCK, 3430 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3431 3432 active = 0; 3433 vplock = mtx_pool_find(mtxpool_sleep, vp); 3434 mtx_lock(vplock); 3435 VOP_UNP_CONNECT(vp, &unp); 3436 if (unp == NULL) 3437 goto done; 3438 UNP_PCB_LOCK(unp); 3439 if (unp->unp_vnode == vp) { 3440 VOP_UNP_DETACH(vp); 3441 unp->unp_vnode = NULL; 3442 active = 1; 3443 } 3444 UNP_PCB_UNLOCK(unp); 3445 done: 3446 mtx_unlock(vplock); 3447 if (active) 3448 vunref(vp); 3449 } 3450 3451 #ifdef DDB 3452 static void 3453 db_print_indent(int indent) 3454 { 3455 int i; 3456 3457 for (i = 0; i < indent; i++) 3458 db_printf(" "); 3459 } 3460 3461 static void 3462 db_print_unpflags(int unp_flags) 3463 { 3464 int comma; 3465 3466 comma = 0; 3467 if (unp_flags & UNP_HAVEPC) { 3468 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3469 comma = 1; 3470 } 3471 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3472 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3473 comma = 1; 3474 } 3475 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3476 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3477 comma = 1; 3478 } 3479 if (unp_flags & UNP_CONNECTING) { 3480 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3481 comma = 1; 3482 } 3483 if (unp_flags & UNP_BINDING) { 3484 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3485 comma = 1; 3486 } 3487 } 3488 3489 static void 3490 db_print_xucred(int indent, struct xucred *xu) 3491 { 3492 int comma, i; 3493 3494 db_print_indent(indent); 3495 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3496 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3497 db_print_indent(indent); 3498 db_printf("cr_groups: "); 3499 comma = 0; 3500 for (i = 0; i < xu->cr_ngroups; i++) { 3501 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3502 comma = 1; 3503 } 3504 db_printf("\n"); 3505 } 3506 3507 static void 3508 db_print_unprefs(int indent, struct unp_head *uh) 3509 { 3510 struct unpcb *unp; 3511 int counter; 3512 3513 counter = 0; 3514 LIST_FOREACH(unp, uh, unp_reflink) { 3515 if (counter % 4 == 0) 3516 db_print_indent(indent); 3517 db_printf("%p ", unp); 3518 if (counter % 4 == 3) 3519 db_printf("\n"); 3520 counter++; 3521 } 3522 if (counter != 0 && counter % 4 != 0) 3523 db_printf("\n"); 3524 } 3525 3526 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3527 { 3528 struct unpcb *unp; 3529 3530 if (!have_addr) { 3531 db_printf("usage: show unpcb <addr>\n"); 3532 return; 3533 } 3534 unp = (struct unpcb *)addr; 3535 3536 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3537 unp->unp_vnode); 3538 3539 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3540 unp->unp_conn); 3541 3542 db_printf("unp_refs:\n"); 3543 db_print_unprefs(2, &unp->unp_refs); 3544 3545 /* XXXRW: Would be nice to print the full address, if any. */ 3546 db_printf("unp_addr: %p\n", unp->unp_addr); 3547 3548 db_printf("unp_gencnt: %llu\n", 3549 (unsigned long long)unp->unp_gencnt); 3550 3551 db_printf("unp_flags: %x (", unp->unp_flags); 3552 db_print_unpflags(unp->unp_flags); 3553 db_printf(")\n"); 3554 3555 db_printf("unp_peercred:\n"); 3556 db_print_xucred(2, &unp->unp_peercred); 3557 3558 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3559 } 3560 #endif 3561