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 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how; 295 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how); 296 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 297 static void unp_dispose(struct socket *so); 298 static void unp_shutdown(struct unpcb *); 299 static void unp_drop(struct unpcb *); 300 static void unp_gc(__unused void *, int); 301 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 302 static void unp_discard(struct file *); 303 static void unp_freerights(struct filedescent **, int); 304 static int unp_internalize(struct mbuf **, struct thread *, 305 struct mbuf **, u_int *, u_int *); 306 static void unp_internalize_fp(struct file *); 307 static int unp_externalize(struct mbuf *, struct mbuf **, int); 308 static int unp_externalize_fp(struct file *); 309 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 310 int, struct mbuf **, u_int *, u_int *); 311 static void unp_process_defers(void * __unused, int); 312 313 static void 314 unp_pcb_hold(struct unpcb *unp) 315 { 316 u_int old __unused; 317 318 old = refcount_acquire(&unp->unp_refcount); 319 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 320 } 321 322 static __result_use_check bool 323 unp_pcb_rele(struct unpcb *unp) 324 { 325 bool ret; 326 327 UNP_PCB_LOCK_ASSERT(unp); 328 329 if ((ret = refcount_release(&unp->unp_refcount))) { 330 UNP_PCB_UNLOCK(unp); 331 UNP_PCB_LOCK_DESTROY(unp); 332 uma_zfree(unp_zone, unp); 333 } 334 return (ret); 335 } 336 337 static void 338 unp_pcb_rele_notlast(struct unpcb *unp) 339 { 340 bool ret __unused; 341 342 ret = refcount_release(&unp->unp_refcount); 343 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 344 } 345 346 static void 347 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 348 { 349 UNP_PCB_UNLOCK_ASSERT(unp); 350 UNP_PCB_UNLOCK_ASSERT(unp2); 351 352 if (unp == unp2) { 353 UNP_PCB_LOCK(unp); 354 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 355 UNP_PCB_LOCK(unp); 356 UNP_PCB_LOCK(unp2); 357 } else { 358 UNP_PCB_LOCK(unp2); 359 UNP_PCB_LOCK(unp); 360 } 361 } 362 363 static void 364 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 365 { 366 UNP_PCB_UNLOCK(unp); 367 if (unp != unp2) 368 UNP_PCB_UNLOCK(unp2); 369 } 370 371 /* 372 * Try to lock the connected peer of an already locked socket. In some cases 373 * this requires that we unlock the current socket. The pairbusy counter is 374 * used to block concurrent connection attempts while the lock is dropped. The 375 * caller must be careful to revalidate PCB state. 376 */ 377 static struct unpcb * 378 unp_pcb_lock_peer(struct unpcb *unp) 379 { 380 struct unpcb *unp2; 381 382 UNP_PCB_LOCK_ASSERT(unp); 383 unp2 = unp->unp_conn; 384 if (unp2 == NULL) 385 return (NULL); 386 if (__predict_false(unp == unp2)) 387 return (unp); 388 389 UNP_PCB_UNLOCK_ASSERT(unp2); 390 391 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 392 return (unp2); 393 if ((uintptr_t)unp2 > (uintptr_t)unp) { 394 UNP_PCB_LOCK(unp2); 395 return (unp2); 396 } 397 unp->unp_pairbusy++; 398 unp_pcb_hold(unp2); 399 UNP_PCB_UNLOCK(unp); 400 401 UNP_PCB_LOCK(unp2); 402 UNP_PCB_LOCK(unp); 403 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 404 ("%s: socket %p was reconnected", __func__, unp)); 405 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 406 unp->unp_flags &= ~UNP_WAITING; 407 wakeup(unp); 408 } 409 if (unp_pcb_rele(unp2)) { 410 /* unp2 is unlocked. */ 411 return (NULL); 412 } 413 if (unp->unp_conn == NULL) { 414 UNP_PCB_UNLOCK(unp2); 415 return (NULL); 416 } 417 return (unp2); 418 } 419 420 static void 421 uipc_abort(struct socket *so) 422 { 423 struct unpcb *unp, *unp2; 424 425 unp = sotounpcb(so); 426 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 427 UNP_PCB_UNLOCK_ASSERT(unp); 428 429 UNP_PCB_LOCK(unp); 430 unp2 = unp->unp_conn; 431 if (unp2 != NULL) { 432 unp_pcb_hold(unp2); 433 UNP_PCB_UNLOCK(unp); 434 unp_drop(unp2); 435 } else 436 UNP_PCB_UNLOCK(unp); 437 } 438 439 static int 440 uipc_attach(struct socket *so, int proto, struct thread *td) 441 { 442 u_long sendspace, recvspace; 443 struct unpcb *unp; 444 int error; 445 bool locked; 446 447 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 448 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 449 switch (so->so_type) { 450 case SOCK_STREAM: 451 sendspace = unpst_sendspace; 452 recvspace = unpst_recvspace; 453 break; 454 455 case SOCK_DGRAM: 456 STAILQ_INIT(&so->so_rcv.uxdg_mb); 457 STAILQ_INIT(&so->so_snd.uxdg_mb); 458 TAILQ_INIT(&so->so_rcv.uxdg_conns); 459 /* 460 * Since send buffer is either bypassed or is a part 461 * of one-to-many receive buffer, we assign both space 462 * limits to unpdg_recvspace. 463 */ 464 sendspace = recvspace = unpdg_recvspace; 465 break; 466 467 case SOCK_SEQPACKET: 468 sendspace = unpsp_sendspace; 469 recvspace = unpsp_recvspace; 470 break; 471 472 default: 473 panic("uipc_attach"); 474 } 475 error = soreserve(so, sendspace, recvspace); 476 if (error) 477 return (error); 478 } 479 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 480 if (unp == NULL) 481 return (ENOBUFS); 482 LIST_INIT(&unp->unp_refs); 483 UNP_PCB_LOCK_INIT(unp); 484 unp->unp_socket = so; 485 so->so_pcb = unp; 486 refcount_init(&unp->unp_refcount, 1); 487 488 if ((locked = UNP_LINK_WOWNED()) == false) 489 UNP_LINK_WLOCK(); 490 491 unp->unp_gencnt = ++unp_gencnt; 492 unp->unp_ino = ++unp_ino; 493 unp_count++; 494 switch (so->so_type) { 495 case SOCK_STREAM: 496 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 497 break; 498 499 case SOCK_DGRAM: 500 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 501 break; 502 503 case SOCK_SEQPACKET: 504 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 505 break; 506 507 default: 508 panic("uipc_attach"); 509 } 510 511 if (locked == false) 512 UNP_LINK_WUNLOCK(); 513 514 return (0); 515 } 516 517 static int 518 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 519 { 520 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 521 struct vattr vattr; 522 int error, namelen; 523 struct nameidata nd; 524 struct unpcb *unp; 525 struct vnode *vp; 526 struct mount *mp; 527 cap_rights_t rights; 528 char *buf; 529 530 if (nam->sa_family != AF_UNIX) 531 return (EAFNOSUPPORT); 532 533 unp = sotounpcb(so); 534 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 535 536 if (soun->sun_len > sizeof(struct sockaddr_un)) 537 return (EINVAL); 538 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 539 if (namelen <= 0) 540 return (EINVAL); 541 542 /* 543 * We don't allow simultaneous bind() calls on a single UNIX domain 544 * socket, so flag in-progress operations, and return an error if an 545 * operation is already in progress. 546 * 547 * Historically, we have not allowed a socket to be rebound, so this 548 * also returns an error. Not allowing re-binding simplifies the 549 * implementation and avoids a great many possible failure modes. 550 */ 551 UNP_PCB_LOCK(unp); 552 if (unp->unp_vnode != NULL) { 553 UNP_PCB_UNLOCK(unp); 554 return (EINVAL); 555 } 556 if (unp->unp_flags & UNP_BINDING) { 557 UNP_PCB_UNLOCK(unp); 558 return (EALREADY); 559 } 560 unp->unp_flags |= UNP_BINDING; 561 UNP_PCB_UNLOCK(unp); 562 563 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 564 bcopy(soun->sun_path, buf, namelen); 565 buf[namelen] = 0; 566 567 restart: 568 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, 569 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 570 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 571 error = namei(&nd); 572 if (error) 573 goto error; 574 vp = nd.ni_vp; 575 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 576 NDFREE_PNBUF(&nd); 577 if (nd.ni_dvp == vp) 578 vrele(nd.ni_dvp); 579 else 580 vput(nd.ni_dvp); 581 if (vp != NULL) { 582 vrele(vp); 583 error = EADDRINUSE; 584 goto error; 585 } 586 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); 587 if (error) 588 goto error; 589 goto restart; 590 } 591 VATTR_NULL(&vattr); 592 vattr.va_type = VSOCK; 593 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 594 #ifdef MAC 595 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 596 &vattr); 597 #endif 598 if (error == 0) 599 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 600 NDFREE_PNBUF(&nd); 601 if (error) { 602 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 603 vn_finished_write(mp); 604 if (error == ERELOOKUP) 605 goto restart; 606 goto error; 607 } 608 vp = nd.ni_vp; 609 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 610 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 611 612 UNP_PCB_LOCK(unp); 613 VOP_UNP_BIND(vp, unp); 614 unp->unp_vnode = vp; 615 unp->unp_addr = soun; 616 unp->unp_flags &= ~UNP_BINDING; 617 UNP_PCB_UNLOCK(unp); 618 vref(vp); 619 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 620 vn_finished_write(mp); 621 free(buf, M_TEMP); 622 return (0); 623 624 error: 625 UNP_PCB_LOCK(unp); 626 unp->unp_flags &= ~UNP_BINDING; 627 UNP_PCB_UNLOCK(unp); 628 free(buf, M_TEMP); 629 return (error); 630 } 631 632 static int 633 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 634 { 635 636 return (uipc_bindat(AT_FDCWD, so, nam, td)); 637 } 638 639 static int 640 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 641 { 642 int error; 643 644 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 645 error = unp_connect(so, nam, td); 646 return (error); 647 } 648 649 static int 650 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 651 struct thread *td) 652 { 653 int error; 654 655 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 656 error = unp_connectat(fd, so, nam, td, false); 657 return (error); 658 } 659 660 static void 661 uipc_close(struct socket *so) 662 { 663 struct unpcb *unp, *unp2; 664 struct vnode *vp = NULL; 665 struct mtx *vplock; 666 667 unp = sotounpcb(so); 668 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 669 670 vplock = NULL; 671 if ((vp = unp->unp_vnode) != NULL) { 672 vplock = mtx_pool_find(mtxpool_sleep, vp); 673 mtx_lock(vplock); 674 } 675 UNP_PCB_LOCK(unp); 676 if (vp && unp->unp_vnode == NULL) { 677 mtx_unlock(vplock); 678 vp = NULL; 679 } 680 if (vp != NULL) { 681 VOP_UNP_DETACH(vp); 682 unp->unp_vnode = NULL; 683 } 684 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 685 unp_disconnect(unp, unp2); 686 else 687 UNP_PCB_UNLOCK(unp); 688 if (vp) { 689 mtx_unlock(vplock); 690 vrele(vp); 691 } 692 } 693 694 static int 695 uipc_connect2(struct socket *so1, struct socket *so2) 696 { 697 struct unpcb *unp, *unp2; 698 699 if (so1->so_type != so2->so_type) 700 return (EPROTOTYPE); 701 702 unp = so1->so_pcb; 703 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 704 unp2 = so2->so_pcb; 705 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 706 unp_pcb_lock_pair(unp, unp2); 707 unp_connect2(so1, so2, PRU_CONNECT2); 708 unp_pcb_unlock_pair(unp, unp2); 709 710 return (0); 711 } 712 713 static void 714 uipc_detach(struct socket *so) 715 { 716 struct unpcb *unp, *unp2; 717 struct mtx *vplock; 718 struct vnode *vp; 719 int local_unp_rights; 720 721 unp = sotounpcb(so); 722 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 723 724 vp = NULL; 725 vplock = NULL; 726 727 UNP_LINK_WLOCK(); 728 LIST_REMOVE(unp, unp_link); 729 if (unp->unp_gcflag & UNPGC_DEAD) 730 LIST_REMOVE(unp, unp_dead); 731 unp->unp_gencnt = ++unp_gencnt; 732 --unp_count; 733 UNP_LINK_WUNLOCK(); 734 735 UNP_PCB_UNLOCK_ASSERT(unp); 736 restart: 737 if ((vp = unp->unp_vnode) != NULL) { 738 vplock = mtx_pool_find(mtxpool_sleep, vp); 739 mtx_lock(vplock); 740 } 741 UNP_PCB_LOCK(unp); 742 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 743 if (vplock) 744 mtx_unlock(vplock); 745 UNP_PCB_UNLOCK(unp); 746 goto restart; 747 } 748 if ((vp = unp->unp_vnode) != NULL) { 749 VOP_UNP_DETACH(vp); 750 unp->unp_vnode = NULL; 751 } 752 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 753 unp_disconnect(unp, unp2); 754 else 755 UNP_PCB_UNLOCK(unp); 756 757 UNP_REF_LIST_LOCK(); 758 while (!LIST_EMPTY(&unp->unp_refs)) { 759 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 760 761 unp_pcb_hold(ref); 762 UNP_REF_LIST_UNLOCK(); 763 764 MPASS(ref != unp); 765 UNP_PCB_UNLOCK_ASSERT(ref); 766 unp_drop(ref); 767 UNP_REF_LIST_LOCK(); 768 } 769 UNP_REF_LIST_UNLOCK(); 770 771 UNP_PCB_LOCK(unp); 772 local_unp_rights = unp_rights; 773 unp->unp_socket->so_pcb = NULL; 774 unp->unp_socket = NULL; 775 free(unp->unp_addr, M_SONAME); 776 unp->unp_addr = NULL; 777 if (!unp_pcb_rele(unp)) 778 UNP_PCB_UNLOCK(unp); 779 if (vp) { 780 mtx_unlock(vplock); 781 vrele(vp); 782 } 783 if (local_unp_rights) 784 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 785 786 switch (so->so_type) { 787 case SOCK_DGRAM: 788 /* 789 * Everything should have been unlinked/freed by unp_dispose() 790 * and/or unp_disconnect(). 791 */ 792 MPASS(so->so_rcv.uxdg_peeked == NULL); 793 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 794 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 795 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 796 } 797 } 798 799 static int 800 uipc_disconnect(struct socket *so) 801 { 802 struct unpcb *unp, *unp2; 803 804 unp = sotounpcb(so); 805 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 806 807 UNP_PCB_LOCK(unp); 808 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 809 unp_disconnect(unp, unp2); 810 else 811 UNP_PCB_UNLOCK(unp); 812 return (0); 813 } 814 815 static int 816 uipc_listen(struct socket *so, int backlog, struct thread *td) 817 { 818 struct unpcb *unp; 819 int error; 820 821 MPASS(so->so_type != SOCK_DGRAM); 822 823 /* 824 * Synchronize with concurrent connection attempts. 825 */ 826 error = 0; 827 unp = sotounpcb(so); 828 UNP_PCB_LOCK(unp); 829 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 830 error = EINVAL; 831 else if (unp->unp_vnode == NULL) 832 error = EDESTADDRREQ; 833 if (error != 0) { 834 UNP_PCB_UNLOCK(unp); 835 return (error); 836 } 837 838 SOCK_LOCK(so); 839 error = solisten_proto_check(so); 840 if (error == 0) { 841 cru2xt(td, &unp->unp_peercred); 842 solisten_proto(so, backlog); 843 } 844 SOCK_UNLOCK(so); 845 UNP_PCB_UNLOCK(unp); 846 return (error); 847 } 848 849 static int 850 uipc_peeraddr(struct socket *so, struct sockaddr *ret) 851 { 852 struct unpcb *unp, *unp2; 853 const struct sockaddr *sa; 854 855 unp = sotounpcb(so); 856 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 857 858 UNP_PCB_LOCK(unp); 859 unp2 = unp_pcb_lock_peer(unp); 860 if (unp2 != NULL) { 861 if (unp2->unp_addr != NULL) 862 sa = (struct sockaddr *)unp2->unp_addr; 863 else 864 sa = &sun_noname; 865 bcopy(sa, ret, sa->sa_len); 866 unp_pcb_unlock_pair(unp, unp2); 867 } else { 868 UNP_PCB_UNLOCK(unp); 869 sa = &sun_noname; 870 bcopy(sa, ret, sa->sa_len); 871 } 872 return (0); 873 } 874 875 static int 876 uipc_rcvd(struct socket *so, int flags) 877 { 878 struct unpcb *unp, *unp2; 879 struct socket *so2; 880 u_int mbcnt, sbcc; 881 882 unp = sotounpcb(so); 883 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 884 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 885 ("%s: socktype %d", __func__, so->so_type)); 886 887 /* 888 * Adjust backpressure on sender and wakeup any waiting to write. 889 * 890 * The unp lock is acquired to maintain the validity of the unp_conn 891 * pointer; no lock on unp2 is required as unp2->unp_socket will be 892 * static as long as we don't permit unp2 to disconnect from unp, 893 * which is prevented by the lock on unp. We cache values from 894 * so_rcv to avoid holding the so_rcv lock over the entire 895 * transaction on the remote so_snd. 896 */ 897 SOCKBUF_LOCK(&so->so_rcv); 898 mbcnt = so->so_rcv.sb_mbcnt; 899 sbcc = sbavail(&so->so_rcv); 900 SOCKBUF_UNLOCK(&so->so_rcv); 901 /* 902 * There is a benign race condition at this point. If we're planning to 903 * clear SB_STOP, but uipc_send is called on the connected socket at 904 * this instant, it might add data to the sockbuf and set SB_STOP. Then 905 * we would erroneously clear SB_STOP below, even though the sockbuf is 906 * full. The race is benign because the only ill effect is to allow the 907 * sockbuf to exceed its size limit, and the size limits are not 908 * strictly guaranteed anyway. 909 */ 910 UNP_PCB_LOCK(unp); 911 unp2 = unp->unp_conn; 912 if (unp2 == NULL) { 913 UNP_PCB_UNLOCK(unp); 914 return (0); 915 } 916 so2 = unp2->unp_socket; 917 SOCKBUF_LOCK(&so2->so_snd); 918 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 919 so2->so_snd.sb_flags &= ~SB_STOP; 920 sowwakeup_locked(so2); 921 UNP_PCB_UNLOCK(unp); 922 return (0); 923 } 924 925 static int 926 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 927 struct mbuf *control, struct thread *td) 928 { 929 struct unpcb *unp, *unp2; 930 struct socket *so2; 931 u_int mbcnt, sbcc; 932 int error; 933 934 unp = sotounpcb(so); 935 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 936 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 937 ("%s: socktype %d", __func__, so->so_type)); 938 939 error = 0; 940 if (flags & PRUS_OOB) { 941 error = EOPNOTSUPP; 942 goto release; 943 } 944 if (control != NULL && 945 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 946 goto release; 947 948 unp2 = NULL; 949 if ((so->so_state & SS_ISCONNECTED) == 0) { 950 if (nam != NULL) { 951 if ((error = unp_connect(so, nam, td)) != 0) 952 goto out; 953 } else { 954 error = ENOTCONN; 955 goto out; 956 } 957 } 958 959 UNP_PCB_LOCK(unp); 960 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 961 UNP_PCB_UNLOCK(unp); 962 error = ENOTCONN; 963 goto out; 964 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 965 unp_pcb_unlock_pair(unp, unp2); 966 error = EPIPE; 967 goto out; 968 } 969 UNP_PCB_UNLOCK(unp); 970 if ((so2 = unp2->unp_socket) == NULL) { 971 UNP_PCB_UNLOCK(unp2); 972 error = ENOTCONN; 973 goto out; 974 } 975 SOCKBUF_LOCK(&so2->so_rcv); 976 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 977 /* 978 * Credentials are passed only once on SOCK_STREAM and 979 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 980 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 981 */ 982 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 983 NULL, NULL); 984 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 985 } 986 987 /* 988 * Send to paired receive port and wake up readers. Don't 989 * check for space available in the receive buffer if we're 990 * attaching ancillary data; Unix domain sockets only check 991 * for space in the sending sockbuf, and that check is 992 * performed one level up the stack. At that level we cannot 993 * precisely account for the amount of buffer space used 994 * (e.g., because control messages are not yet internalized). 995 */ 996 switch (so->so_type) { 997 case SOCK_STREAM: 998 if (control != NULL) { 999 sbappendcontrol_locked(&so2->so_rcv, m, 1000 control, flags); 1001 control = NULL; 1002 } else 1003 sbappend_locked(&so2->so_rcv, m, flags); 1004 break; 1005 1006 case SOCK_SEQPACKET: 1007 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1008 &sun_noname, m, control)) 1009 control = NULL; 1010 break; 1011 } 1012 1013 mbcnt = so2->so_rcv.sb_mbcnt; 1014 sbcc = sbavail(&so2->so_rcv); 1015 if (sbcc) 1016 sorwakeup_locked(so2); 1017 else 1018 SOCKBUF_UNLOCK(&so2->so_rcv); 1019 1020 /* 1021 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1022 * it would be possible for uipc_rcvd to be called at this 1023 * point, drain the receiving sockbuf, clear SB_STOP, and then 1024 * we would set SB_STOP below. That could lead to an empty 1025 * sockbuf having SB_STOP set 1026 */ 1027 SOCKBUF_LOCK(&so->so_snd); 1028 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1029 so->so_snd.sb_flags |= SB_STOP; 1030 SOCKBUF_UNLOCK(&so->so_snd); 1031 UNP_PCB_UNLOCK(unp2); 1032 m = NULL; 1033 out: 1034 /* 1035 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1036 */ 1037 if (flags & PRUS_EOF) { 1038 UNP_PCB_LOCK(unp); 1039 socantsendmore(so); 1040 unp_shutdown(unp); 1041 UNP_PCB_UNLOCK(unp); 1042 } 1043 if (control != NULL && error != 0) 1044 unp_scan(control, unp_freerights); 1045 1046 release: 1047 if (control != NULL) 1048 m_freem(control); 1049 /* 1050 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1051 * for freeing memory. 1052 */ 1053 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1054 m_freem(m); 1055 return (error); 1056 } 1057 1058 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1059 static inline bool 1060 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1061 { 1062 u_int bleft, mleft; 1063 1064 /* 1065 * Negative space may happen if send(2) is followed by 1066 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1067 */ 1068 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1069 sb->sb_mbmax < sb->uxdg_mbcnt)) 1070 return (false); 1071 1072 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1073 return (false); 1074 1075 bleft = sb->sb_hiwat - sb->uxdg_cc; 1076 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1077 1078 return (bleft >= cc && mleft >= mbcnt); 1079 } 1080 1081 /* 1082 * PF_UNIX/SOCK_DGRAM send 1083 * 1084 * Allocate a record consisting of 3 mbufs in the sequence of 1085 * from -> control -> data and append it to the socket buffer. 1086 * 1087 * The first mbuf carries sender's name and is a pkthdr that stores 1088 * overall length of datagram, its memory consumption and control length. 1089 */ 1090 #define ctllen PH_loc.thirtytwo[1] 1091 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1092 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1093 static int 1094 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1095 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1096 { 1097 struct unpcb *unp, *unp2; 1098 const struct sockaddr *from; 1099 struct socket *so2; 1100 struct sockbuf *sb; 1101 struct mbuf *f, *clast; 1102 u_int cc, ctl, mbcnt; 1103 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1104 int error; 1105 1106 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1107 1108 error = 0; 1109 f = NULL; 1110 ctl = 0; 1111 1112 if (__predict_false(flags & MSG_OOB)) { 1113 error = EOPNOTSUPP; 1114 goto out; 1115 } 1116 if (m == NULL) { 1117 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1118 error = EMSGSIZE; 1119 goto out; 1120 } 1121 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1122 if (__predict_false(m == NULL)) { 1123 error = EFAULT; 1124 goto out; 1125 } 1126 f = m_gethdr(M_WAITOK, MT_SONAME); 1127 cc = m->m_pkthdr.len; 1128 mbcnt = MSIZE + m->m_pkthdr.memlen; 1129 if (c != NULL && 1130 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1131 goto out; 1132 } else { 1133 /* pr_sosend() with mbuf usually is a kernel thread. */ 1134 1135 M_ASSERTPKTHDR(m); 1136 if (__predict_false(c != NULL)) 1137 panic("%s: control from a kernel thread", __func__); 1138 1139 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1140 error = EMSGSIZE; 1141 goto out; 1142 } 1143 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1144 error = ENOBUFS; 1145 goto out; 1146 } 1147 /* Condition the foreign mbuf to our standards. */ 1148 m_clrprotoflags(m); 1149 m_tag_delete_chain(m, NULL); 1150 m->m_pkthdr.rcvif = NULL; 1151 m->m_pkthdr.flowid = 0; 1152 m->m_pkthdr.csum_flags = 0; 1153 m->m_pkthdr.fibnum = 0; 1154 m->m_pkthdr.rsstype = 0; 1155 1156 cc = m->m_pkthdr.len; 1157 mbcnt = MSIZE; 1158 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1159 mbcnt += MSIZE; 1160 if (mb->m_flags & M_EXT) 1161 mbcnt += mb->m_ext.ext_size; 1162 } 1163 } 1164 1165 unp = sotounpcb(so); 1166 MPASS(unp); 1167 1168 /* 1169 * XXXGL: would be cool to fully remove so_snd out of the equation 1170 * and avoid this lock, which is not only extraneous, but also being 1171 * released, thus still leaving possibility for a race. We can easily 1172 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1173 * is more difficult to invent something to handle so_error. 1174 */ 1175 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1176 if (error) 1177 goto out2; 1178 SOCK_SENDBUF_LOCK(so); 1179 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1180 SOCK_SENDBUF_UNLOCK(so); 1181 error = EPIPE; 1182 goto out3; 1183 } 1184 if (so->so_error != 0) { 1185 error = so->so_error; 1186 so->so_error = 0; 1187 SOCK_SENDBUF_UNLOCK(so); 1188 goto out3; 1189 } 1190 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1191 SOCK_SENDBUF_UNLOCK(so); 1192 error = EDESTADDRREQ; 1193 goto out3; 1194 } 1195 SOCK_SENDBUF_UNLOCK(so); 1196 1197 if (addr != NULL) { 1198 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1199 goto out3; 1200 UNP_PCB_LOCK_ASSERT(unp); 1201 unp2 = unp->unp_conn; 1202 UNP_PCB_LOCK_ASSERT(unp2); 1203 } else { 1204 UNP_PCB_LOCK(unp); 1205 unp2 = unp_pcb_lock_peer(unp); 1206 if (unp2 == NULL) { 1207 UNP_PCB_UNLOCK(unp); 1208 error = ENOTCONN; 1209 goto out3; 1210 } 1211 } 1212 1213 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1214 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1215 &mbcnt); 1216 if (unp->unp_addr != NULL) 1217 from = (struct sockaddr *)unp->unp_addr; 1218 else 1219 from = &sun_noname; 1220 f->m_len = from->sa_len; 1221 MPASS(from->sa_len <= MLEN); 1222 bcopy(from, mtod(f, void *), from->sa_len); 1223 ctl += f->m_len; 1224 1225 /* 1226 * Concatenate mbufs: from -> control -> data. 1227 * Save overall cc and mbcnt in "from" mbuf. 1228 */ 1229 if (c != NULL) { 1230 #ifdef INVARIANTS 1231 struct mbuf *mc; 1232 1233 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1234 MPASS(mc == clast); 1235 #endif 1236 f->m_next = c; 1237 clast->m_next = m; 1238 c = NULL; 1239 } else 1240 f->m_next = m; 1241 m = NULL; 1242 #ifdef INVARIANTS 1243 dcc = dctl = dmbcnt = 0; 1244 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1245 if (mb->m_type == MT_DATA) 1246 dcc += mb->m_len; 1247 else 1248 dctl += mb->m_len; 1249 dmbcnt += MSIZE; 1250 if (mb->m_flags & M_EXT) 1251 dmbcnt += mb->m_ext.ext_size; 1252 } 1253 MPASS(dcc == cc); 1254 MPASS(dctl == ctl); 1255 MPASS(dmbcnt == mbcnt); 1256 #endif 1257 f->m_pkthdr.len = cc + ctl; 1258 f->m_pkthdr.memlen = mbcnt; 1259 f->m_pkthdr.ctllen = ctl; 1260 1261 /* 1262 * Destination socket buffer selection. 1263 * 1264 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1265 * destination address is supplied, create a temporary connection for 1266 * the run time of the function (see call to unp_connectat() above and 1267 * to unp_disconnect() below). We distinguish them by condition of 1268 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1269 * that condition, since, again, through the run time of this code we 1270 * are always connected. For such "unconnected" sends, the destination 1271 * buffer would be the receive buffer of destination socket so2. 1272 * 1273 * For connected sends, data lands on the send buffer of the sender's 1274 * socket "so". Then, if we just added the very first datagram 1275 * on this send buffer, we need to add the send buffer on to the 1276 * receiving socket's buffer list. We put ourselves on top of the 1277 * list. Such logic gives infrequent senders priority over frequent 1278 * senders. 1279 * 1280 * Note on byte count management. As long as event methods kevent(2), 1281 * select(2) are not protocol specific (yet), we need to maintain 1282 * meaningful values on the receive buffer. So, the receive buffer 1283 * would accumulate counters from all connected buffers potentially 1284 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1285 */ 1286 so2 = unp2->unp_socket; 1287 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1288 SOCK_RECVBUF_LOCK(so2); 1289 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1290 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1291 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1292 uxdg_clist); 1293 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1294 sb->uxdg_cc += cc + ctl; 1295 sb->uxdg_ctl += ctl; 1296 sb->uxdg_mbcnt += mbcnt; 1297 so2->so_rcv.sb_acc += cc + ctl; 1298 so2->so_rcv.sb_ccc += cc + ctl; 1299 so2->so_rcv.sb_ctl += ctl; 1300 so2->so_rcv.sb_mbcnt += mbcnt; 1301 sorwakeup_locked(so2); 1302 f = NULL; 1303 } else { 1304 soroverflow_locked(so2); 1305 error = ENOBUFS; 1306 if (f->m_next->m_type == MT_CONTROL) { 1307 c = f->m_next; 1308 f->m_next = NULL; 1309 } 1310 } 1311 1312 if (addr != NULL) 1313 unp_disconnect(unp, unp2); 1314 else 1315 unp_pcb_unlock_pair(unp, unp2); 1316 1317 td->td_ru.ru_msgsnd++; 1318 1319 out3: 1320 SOCK_IO_SEND_UNLOCK(so); 1321 out2: 1322 if (c) 1323 unp_scan(c, unp_freerights); 1324 out: 1325 if (f) 1326 m_freem(f); 1327 if (c) 1328 m_freem(c); 1329 if (m) 1330 m_freem(m); 1331 1332 return (error); 1333 } 1334 1335 /* 1336 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1337 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1338 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1339 */ 1340 static int 1341 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1342 struct uio *uio, struct mbuf **controlp, int *flagsp) 1343 { 1344 ssize_t len = 0; 1345 int error; 1346 1347 so->so_rcv.uxdg_peeked = m; 1348 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1349 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1350 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1351 SOCK_RECVBUF_UNLOCK(so); 1352 1353 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1354 if (psa != NULL) 1355 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1356 1357 m = m->m_next; 1358 KASSERT(m, ("%s: no data or control after soname", __func__)); 1359 1360 /* 1361 * With MSG_PEEK the control isn't executed, just copied. 1362 */ 1363 while (m != NULL && m->m_type == MT_CONTROL) { 1364 if (controlp != NULL) { 1365 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1366 controlp = &(*controlp)->m_next; 1367 } 1368 m = m->m_next; 1369 } 1370 KASSERT(m == NULL || m->m_type == MT_DATA, 1371 ("%s: not MT_DATA mbuf %p", __func__, m)); 1372 while (m != NULL && uio->uio_resid > 0) { 1373 len = uio->uio_resid; 1374 if (len > m->m_len) 1375 len = m->m_len; 1376 error = uiomove(mtod(m, char *), (int)len, uio); 1377 if (error) { 1378 SOCK_IO_RECV_UNLOCK(so); 1379 return (error); 1380 } 1381 if (len == m->m_len) 1382 m = m->m_next; 1383 } 1384 SOCK_IO_RECV_UNLOCK(so); 1385 1386 if (flagsp != NULL) { 1387 if (m != NULL) { 1388 if (*flagsp & MSG_TRUNC) { 1389 /* Report real length of the packet */ 1390 uio->uio_resid -= m_length(m, NULL) - len; 1391 } 1392 *flagsp |= MSG_TRUNC; 1393 } else 1394 *flagsp &= ~MSG_TRUNC; 1395 } 1396 1397 return (0); 1398 } 1399 1400 /* 1401 * PF_UNIX/SOCK_DGRAM receive 1402 */ 1403 static int 1404 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1405 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1406 { 1407 struct sockbuf *sb = NULL; 1408 struct mbuf *m; 1409 int flags, error; 1410 ssize_t len = 0; 1411 bool nonblock; 1412 1413 MPASS(mp0 == NULL); 1414 1415 if (psa != NULL) 1416 *psa = NULL; 1417 if (controlp != NULL) 1418 *controlp = NULL; 1419 1420 flags = flagsp != NULL ? *flagsp : 0; 1421 nonblock = (so->so_state & SS_NBIO) || 1422 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1423 1424 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1425 if (__predict_false(error)) 1426 return (error); 1427 1428 /* 1429 * Loop blocking while waiting for a datagram. Prioritize connected 1430 * peers over unconnected sends. Set sb to selected socket buffer 1431 * containing an mbuf on exit from the wait loop. A datagram that 1432 * had already been peeked at has top priority. 1433 */ 1434 SOCK_RECVBUF_LOCK(so); 1435 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1436 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1437 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1438 if (so->so_error) { 1439 error = so->so_error; 1440 so->so_error = 0; 1441 SOCK_RECVBUF_UNLOCK(so); 1442 SOCK_IO_RECV_UNLOCK(so); 1443 return (error); 1444 } 1445 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1446 uio->uio_resid == 0) { 1447 SOCK_RECVBUF_UNLOCK(so); 1448 SOCK_IO_RECV_UNLOCK(so); 1449 return (0); 1450 } 1451 if (nonblock) { 1452 SOCK_RECVBUF_UNLOCK(so); 1453 SOCK_IO_RECV_UNLOCK(so); 1454 return (EWOULDBLOCK); 1455 } 1456 error = sbwait(so, SO_RCV); 1457 if (error) { 1458 SOCK_RECVBUF_UNLOCK(so); 1459 SOCK_IO_RECV_UNLOCK(so); 1460 return (error); 1461 } 1462 } 1463 1464 if (sb == NULL) 1465 sb = &so->so_rcv; 1466 else if (m == NULL) 1467 m = STAILQ_FIRST(&sb->uxdg_mb); 1468 else 1469 MPASS(m == so->so_rcv.uxdg_peeked); 1470 1471 MPASS(sb->uxdg_cc > 0); 1472 M_ASSERTPKTHDR(m); 1473 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1474 1475 if (uio->uio_td) 1476 uio->uio_td->td_ru.ru_msgrcv++; 1477 1478 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1479 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1480 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1481 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1482 } else 1483 so->so_rcv.uxdg_peeked = NULL; 1484 1485 sb->uxdg_cc -= m->m_pkthdr.len; 1486 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1487 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1488 1489 if (__predict_false(flags & MSG_PEEK)) 1490 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1491 1492 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1493 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1494 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1495 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1496 SOCK_RECVBUF_UNLOCK(so); 1497 1498 if (psa != NULL) 1499 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1500 m = m_free(m); 1501 KASSERT(m, ("%s: no data or control after soname", __func__)); 1502 1503 /* 1504 * Packet to copyout() is now in 'm' and it is disconnected from the 1505 * queue. 1506 * 1507 * Process one or more MT_CONTROL mbufs present before any data mbufs 1508 * in the first mbuf chain on the socket buffer. We call into the 1509 * unp_externalize() to perform externalization (or freeing if 1510 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1511 * without MT_DATA mbufs. 1512 */ 1513 while (m != NULL && m->m_type == MT_CONTROL) { 1514 struct mbuf *cm; 1515 1516 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1517 * it frees whole chain, so we must disconnect the mbuf. 1518 */ 1519 cm = m; m = m->m_next; cm->m_next = NULL; 1520 error = unp_externalize(cm, controlp, flags); 1521 if (error != 0) { 1522 SOCK_IO_RECV_UNLOCK(so); 1523 unp_scan(m, unp_freerights); 1524 m_freem(m); 1525 return (error); 1526 } 1527 if (controlp != NULL) { 1528 while (*controlp != NULL) 1529 controlp = &(*controlp)->m_next; 1530 } 1531 } 1532 KASSERT(m == NULL || m->m_type == MT_DATA, 1533 ("%s: not MT_DATA mbuf %p", __func__, m)); 1534 while (m != NULL && uio->uio_resid > 0) { 1535 len = uio->uio_resid; 1536 if (len > m->m_len) 1537 len = m->m_len; 1538 error = uiomove(mtod(m, char *), (int)len, uio); 1539 if (error) { 1540 SOCK_IO_RECV_UNLOCK(so); 1541 m_freem(m); 1542 return (error); 1543 } 1544 if (len == m->m_len) 1545 m = m_free(m); 1546 else { 1547 m->m_data += len; 1548 m->m_len -= len; 1549 } 1550 } 1551 SOCK_IO_RECV_UNLOCK(so); 1552 1553 if (m != NULL) { 1554 if (flagsp != NULL) { 1555 if (flags & MSG_TRUNC) { 1556 /* Report real length of the packet */ 1557 uio->uio_resid -= m_length(m, NULL); 1558 } 1559 *flagsp |= MSG_TRUNC; 1560 } 1561 m_freem(m); 1562 } else if (flagsp != NULL) 1563 *flagsp &= ~MSG_TRUNC; 1564 1565 return (0); 1566 } 1567 1568 static bool 1569 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1570 { 1571 struct mbuf *mb, *n; 1572 struct sockbuf *sb; 1573 1574 SOCK_LOCK(so); 1575 if (SOLISTENING(so)) { 1576 SOCK_UNLOCK(so); 1577 return (false); 1578 } 1579 mb = NULL; 1580 sb = &so->so_rcv; 1581 SOCKBUF_LOCK(sb); 1582 if (sb->sb_fnrdy != NULL) { 1583 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1584 if (mb == m) { 1585 *errorp = sbready(sb, m, count); 1586 break; 1587 } 1588 mb = mb->m_next; 1589 if (mb == NULL) { 1590 mb = n; 1591 if (mb != NULL) 1592 n = mb->m_nextpkt; 1593 } 1594 } 1595 } 1596 SOCKBUF_UNLOCK(sb); 1597 SOCK_UNLOCK(so); 1598 return (mb != NULL); 1599 } 1600 1601 static int 1602 uipc_ready(struct socket *so, struct mbuf *m, int count) 1603 { 1604 struct unpcb *unp, *unp2; 1605 struct socket *so2; 1606 int error, i; 1607 1608 unp = sotounpcb(so); 1609 1610 KASSERT(so->so_type == SOCK_STREAM, 1611 ("%s: unexpected socket type for %p", __func__, so)); 1612 1613 UNP_PCB_LOCK(unp); 1614 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1615 UNP_PCB_UNLOCK(unp); 1616 so2 = unp2->unp_socket; 1617 SOCKBUF_LOCK(&so2->so_rcv); 1618 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1619 sorwakeup_locked(so2); 1620 else 1621 SOCKBUF_UNLOCK(&so2->so_rcv); 1622 UNP_PCB_UNLOCK(unp2); 1623 return (error); 1624 } 1625 UNP_PCB_UNLOCK(unp); 1626 1627 /* 1628 * The receiving socket has been disconnected, but may still be valid. 1629 * In this case, the now-ready mbufs are still present in its socket 1630 * buffer, so perform an exhaustive search before giving up and freeing 1631 * the mbufs. 1632 */ 1633 UNP_LINK_RLOCK(); 1634 LIST_FOREACH(unp, &unp_shead, unp_link) { 1635 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1636 break; 1637 } 1638 UNP_LINK_RUNLOCK(); 1639 1640 if (unp == NULL) { 1641 for (i = 0; i < count; i++) 1642 m = m_free(m); 1643 error = ECONNRESET; 1644 } 1645 return (error); 1646 } 1647 1648 static int 1649 uipc_sense(struct socket *so, struct stat *sb) 1650 { 1651 struct unpcb *unp; 1652 1653 unp = sotounpcb(so); 1654 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1655 1656 sb->st_blksize = so->so_snd.sb_hiwat; 1657 sb->st_dev = NODEV; 1658 sb->st_ino = unp->unp_ino; 1659 return (0); 1660 } 1661 1662 static int 1663 uipc_shutdown(struct socket *so) 1664 { 1665 struct unpcb *unp; 1666 1667 unp = sotounpcb(so); 1668 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1669 1670 UNP_PCB_LOCK(unp); 1671 socantsendmore(so); 1672 unp_shutdown(unp); 1673 UNP_PCB_UNLOCK(unp); 1674 return (0); 1675 } 1676 1677 static int 1678 uipc_sockaddr(struct socket *so, struct sockaddr *ret) 1679 { 1680 struct unpcb *unp; 1681 const struct sockaddr *sa; 1682 1683 unp = sotounpcb(so); 1684 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1685 1686 UNP_PCB_LOCK(unp); 1687 if (unp->unp_addr != NULL) 1688 sa = (struct sockaddr *) unp->unp_addr; 1689 else 1690 sa = &sun_noname; 1691 bcopy(sa, ret, sa->sa_len); 1692 UNP_PCB_UNLOCK(unp); 1693 return (0); 1694 } 1695 1696 static int 1697 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1698 { 1699 struct unpcb *unp; 1700 struct xucred xu; 1701 int error, optval; 1702 1703 if (sopt->sopt_level != SOL_LOCAL) 1704 return (EINVAL); 1705 1706 unp = sotounpcb(so); 1707 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1708 error = 0; 1709 switch (sopt->sopt_dir) { 1710 case SOPT_GET: 1711 switch (sopt->sopt_name) { 1712 case LOCAL_PEERCRED: 1713 UNP_PCB_LOCK(unp); 1714 if (unp->unp_flags & UNP_HAVEPC) 1715 xu = unp->unp_peercred; 1716 else { 1717 if (so->so_type == SOCK_STREAM) 1718 error = ENOTCONN; 1719 else 1720 error = EINVAL; 1721 } 1722 UNP_PCB_UNLOCK(unp); 1723 if (error == 0) 1724 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1725 break; 1726 1727 case LOCAL_CREDS: 1728 /* Unlocked read. */ 1729 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1730 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1731 break; 1732 1733 case LOCAL_CREDS_PERSISTENT: 1734 /* Unlocked read. */ 1735 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1736 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1737 break; 1738 1739 case LOCAL_CONNWAIT: 1740 /* Unlocked read. */ 1741 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1742 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1743 break; 1744 1745 default: 1746 error = EOPNOTSUPP; 1747 break; 1748 } 1749 break; 1750 1751 case SOPT_SET: 1752 switch (sopt->sopt_name) { 1753 case LOCAL_CREDS: 1754 case LOCAL_CREDS_PERSISTENT: 1755 case LOCAL_CONNWAIT: 1756 error = sooptcopyin(sopt, &optval, sizeof(optval), 1757 sizeof(optval)); 1758 if (error) 1759 break; 1760 1761 #define OPTSET(bit, exclusive) do { \ 1762 UNP_PCB_LOCK(unp); \ 1763 if (optval) { \ 1764 if ((unp->unp_flags & (exclusive)) != 0) { \ 1765 UNP_PCB_UNLOCK(unp); \ 1766 error = EINVAL; \ 1767 break; \ 1768 } \ 1769 unp->unp_flags |= (bit); \ 1770 } else \ 1771 unp->unp_flags &= ~(bit); \ 1772 UNP_PCB_UNLOCK(unp); \ 1773 } while (0) 1774 1775 switch (sopt->sopt_name) { 1776 case LOCAL_CREDS: 1777 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1778 break; 1779 1780 case LOCAL_CREDS_PERSISTENT: 1781 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1782 break; 1783 1784 case LOCAL_CONNWAIT: 1785 OPTSET(UNP_CONNWAIT, 0); 1786 break; 1787 1788 default: 1789 break; 1790 } 1791 break; 1792 #undef OPTSET 1793 default: 1794 error = ENOPROTOOPT; 1795 break; 1796 } 1797 break; 1798 1799 default: 1800 error = EOPNOTSUPP; 1801 break; 1802 } 1803 return (error); 1804 } 1805 1806 static int 1807 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1808 { 1809 1810 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1811 } 1812 1813 static int 1814 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1815 struct thread *td, bool return_locked) 1816 { 1817 struct mtx *vplock; 1818 struct sockaddr_un *soun; 1819 struct vnode *vp; 1820 struct socket *so2; 1821 struct unpcb *unp, *unp2, *unp3; 1822 struct nameidata nd; 1823 char buf[SOCK_MAXADDRLEN]; 1824 struct sockaddr *sa; 1825 cap_rights_t rights; 1826 int error, len; 1827 bool connreq; 1828 1829 if (nam->sa_family != AF_UNIX) 1830 return (EAFNOSUPPORT); 1831 if (nam->sa_len > sizeof(struct sockaddr_un)) 1832 return (EINVAL); 1833 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1834 if (len <= 0) 1835 return (EINVAL); 1836 soun = (struct sockaddr_un *)nam; 1837 bcopy(soun->sun_path, buf, len); 1838 buf[len] = 0; 1839 1840 error = 0; 1841 unp = sotounpcb(so); 1842 UNP_PCB_LOCK(unp); 1843 for (;;) { 1844 /* 1845 * Wait for connection state to stabilize. If a connection 1846 * already exists, give up. For datagram sockets, which permit 1847 * multiple consecutive connect(2) calls, upper layers are 1848 * responsible for disconnecting in advance of a subsequent 1849 * connect(2), but this is not synchronized with PCB connection 1850 * state. 1851 * 1852 * Also make sure that no threads are currently attempting to 1853 * lock the peer socket, to ensure that unp_conn cannot 1854 * transition between two valid sockets while locks are dropped. 1855 */ 1856 if (SOLISTENING(so)) 1857 error = EOPNOTSUPP; 1858 else if (unp->unp_conn != NULL) 1859 error = EISCONN; 1860 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1861 error = EALREADY; 1862 } 1863 if (error != 0) { 1864 UNP_PCB_UNLOCK(unp); 1865 return (error); 1866 } 1867 if (unp->unp_pairbusy > 0) { 1868 unp->unp_flags |= UNP_WAITING; 1869 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1870 continue; 1871 } 1872 break; 1873 } 1874 unp->unp_flags |= UNP_CONNECTING; 1875 UNP_PCB_UNLOCK(unp); 1876 1877 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1878 if (connreq) 1879 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1880 else 1881 sa = NULL; 1882 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1883 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1884 error = namei(&nd); 1885 if (error) 1886 vp = NULL; 1887 else 1888 vp = nd.ni_vp; 1889 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1890 if (error) 1891 goto bad; 1892 NDFREE_PNBUF(&nd); 1893 1894 if (vp->v_type != VSOCK) { 1895 error = ENOTSOCK; 1896 goto bad; 1897 } 1898 #ifdef MAC 1899 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1900 if (error) 1901 goto bad; 1902 #endif 1903 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1904 if (error) 1905 goto bad; 1906 1907 unp = sotounpcb(so); 1908 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1909 1910 vplock = mtx_pool_find(mtxpool_sleep, vp); 1911 mtx_lock(vplock); 1912 VOP_UNP_CONNECT(vp, &unp2); 1913 if (unp2 == NULL) { 1914 error = ECONNREFUSED; 1915 goto bad2; 1916 } 1917 so2 = unp2->unp_socket; 1918 if (so->so_type != so2->so_type) { 1919 error = EPROTOTYPE; 1920 goto bad2; 1921 } 1922 if (connreq) { 1923 if (SOLISTENING(so2)) { 1924 CURVNET_SET(so2->so_vnet); 1925 so2 = sonewconn(so2, 0); 1926 CURVNET_RESTORE(); 1927 } else 1928 so2 = NULL; 1929 if (so2 == NULL) { 1930 error = ECONNREFUSED; 1931 goto bad2; 1932 } 1933 unp3 = sotounpcb(so2); 1934 unp_pcb_lock_pair(unp2, unp3); 1935 if (unp2->unp_addr != NULL) { 1936 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1937 unp3->unp_addr = (struct sockaddr_un *) sa; 1938 sa = NULL; 1939 } 1940 1941 unp_copy_peercred(td, unp3, unp, unp2); 1942 1943 UNP_PCB_UNLOCK(unp2); 1944 unp2 = unp3; 1945 1946 /* 1947 * It is safe to block on the PCB lock here since unp2 is 1948 * nascent and cannot be connected to any other sockets. 1949 */ 1950 UNP_PCB_LOCK(unp); 1951 #ifdef MAC 1952 mac_socketpeer_set_from_socket(so, so2); 1953 mac_socketpeer_set_from_socket(so2, so); 1954 #endif 1955 } else { 1956 unp_pcb_lock_pair(unp, unp2); 1957 } 1958 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1959 sotounpcb(so2) == unp2, 1960 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1961 unp_connect2(so, so2, PRU_CONNECT); 1962 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1963 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1964 unp->unp_flags &= ~UNP_CONNECTING; 1965 if (!return_locked) 1966 unp_pcb_unlock_pair(unp, unp2); 1967 bad2: 1968 mtx_unlock(vplock); 1969 bad: 1970 if (vp != NULL) { 1971 /* 1972 * If we are returning locked (called via uipc_sosend_dgram()), 1973 * we need to be sure that vput() won't sleep. This is 1974 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 1975 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 1976 */ 1977 MPASS(!(return_locked && connreq)); 1978 vput(vp); 1979 } 1980 free(sa, M_SONAME); 1981 if (__predict_false(error)) { 1982 UNP_PCB_LOCK(unp); 1983 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1984 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1985 unp->unp_flags &= ~UNP_CONNECTING; 1986 UNP_PCB_UNLOCK(unp); 1987 } 1988 return (error); 1989 } 1990 1991 /* 1992 * Set socket peer credentials at connection time. 1993 * 1994 * The client's PCB credentials are copied from its process structure. The 1995 * server's PCB credentials are copied from the socket on which it called 1996 * listen(2). uipc_listen cached that process's credentials at the time. 1997 */ 1998 void 1999 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2000 struct unpcb *server_unp, struct unpcb *listen_unp) 2001 { 2002 cru2xt(td, &client_unp->unp_peercred); 2003 client_unp->unp_flags |= UNP_HAVEPC; 2004 2005 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2006 sizeof(server_unp->unp_peercred)); 2007 server_unp->unp_flags |= UNP_HAVEPC; 2008 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2009 } 2010 2011 static void 2012 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2013 { 2014 struct unpcb *unp; 2015 struct unpcb *unp2; 2016 2017 MPASS(so2->so_type == so->so_type); 2018 unp = sotounpcb(so); 2019 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2020 unp2 = sotounpcb(so2); 2021 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2022 2023 UNP_PCB_LOCK_ASSERT(unp); 2024 UNP_PCB_LOCK_ASSERT(unp2); 2025 KASSERT(unp->unp_conn == NULL, 2026 ("%s: socket %p is already connected", __func__, unp)); 2027 2028 unp->unp_conn = unp2; 2029 unp_pcb_hold(unp2); 2030 unp_pcb_hold(unp); 2031 switch (so->so_type) { 2032 case SOCK_DGRAM: 2033 UNP_REF_LIST_LOCK(); 2034 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2035 UNP_REF_LIST_UNLOCK(); 2036 soisconnected(so); 2037 break; 2038 2039 case SOCK_STREAM: 2040 case SOCK_SEQPACKET: 2041 KASSERT(unp2->unp_conn == NULL, 2042 ("%s: socket %p is already connected", __func__, unp2)); 2043 unp2->unp_conn = unp; 2044 if (req == PRU_CONNECT && 2045 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2046 soisconnecting(so); 2047 else 2048 soisconnected(so); 2049 soisconnected(so2); 2050 break; 2051 2052 default: 2053 panic("unp_connect2"); 2054 } 2055 } 2056 2057 static void 2058 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2059 { 2060 struct socket *so, *so2; 2061 struct mbuf *m = NULL; 2062 #ifdef INVARIANTS 2063 struct unpcb *unptmp; 2064 #endif 2065 2066 UNP_PCB_LOCK_ASSERT(unp); 2067 UNP_PCB_LOCK_ASSERT(unp2); 2068 KASSERT(unp->unp_conn == unp2, 2069 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2070 2071 unp->unp_conn = NULL; 2072 so = unp->unp_socket; 2073 so2 = unp2->unp_socket; 2074 switch (unp->unp_socket->so_type) { 2075 case SOCK_DGRAM: 2076 /* 2077 * Remove our send socket buffer from the peer's receive buffer. 2078 * Move the data to the receive buffer only if it is empty. 2079 * This is a protection against a scenario where a peer 2080 * connects, floods and disconnects, effectively blocking 2081 * sendto() from unconnected sockets. 2082 */ 2083 SOCK_RECVBUF_LOCK(so2); 2084 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2085 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2086 uxdg_clist); 2087 if (__predict_true((so2->so_rcv.sb_state & 2088 SBS_CANTRCVMORE) == 0) && 2089 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2090 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2091 &so->so_snd.uxdg_mb); 2092 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2093 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2094 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2095 } else { 2096 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2097 STAILQ_INIT(&so->so_snd.uxdg_mb); 2098 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2099 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2100 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2101 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2102 } 2103 /* Note: so may reconnect. */ 2104 so->so_snd.uxdg_cc = 0; 2105 so->so_snd.uxdg_ctl = 0; 2106 so->so_snd.uxdg_mbcnt = 0; 2107 } 2108 SOCK_RECVBUF_UNLOCK(so2); 2109 UNP_REF_LIST_LOCK(); 2110 #ifdef INVARIANTS 2111 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2112 if (unptmp == unp) 2113 break; 2114 } 2115 KASSERT(unptmp != NULL, 2116 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2117 #endif 2118 LIST_REMOVE(unp, unp_reflink); 2119 UNP_REF_LIST_UNLOCK(); 2120 if (so) { 2121 SOCK_LOCK(so); 2122 so->so_state &= ~SS_ISCONNECTED; 2123 SOCK_UNLOCK(so); 2124 } 2125 break; 2126 2127 case SOCK_STREAM: 2128 case SOCK_SEQPACKET: 2129 if (so) 2130 soisdisconnected(so); 2131 MPASS(unp2->unp_conn == unp); 2132 unp2->unp_conn = NULL; 2133 if (so2) 2134 soisdisconnected(so2); 2135 break; 2136 } 2137 2138 if (unp == unp2) { 2139 unp_pcb_rele_notlast(unp); 2140 if (!unp_pcb_rele(unp)) 2141 UNP_PCB_UNLOCK(unp); 2142 } else { 2143 if (!unp_pcb_rele(unp)) 2144 UNP_PCB_UNLOCK(unp); 2145 if (!unp_pcb_rele(unp2)) 2146 UNP_PCB_UNLOCK(unp2); 2147 } 2148 2149 if (m != NULL) { 2150 unp_scan(m, unp_freerights); 2151 m_freem(m); 2152 } 2153 } 2154 2155 /* 2156 * unp_pcblist() walks the global list of struct unpcb's to generate a 2157 * pointer list, bumping the refcount on each unpcb. It then copies them out 2158 * sequentially, validating the generation number on each to see if it has 2159 * been detached. All of this is necessary because copyout() may sleep on 2160 * disk I/O. 2161 */ 2162 static int 2163 unp_pcblist(SYSCTL_HANDLER_ARGS) 2164 { 2165 struct unpcb *unp, **unp_list; 2166 unp_gen_t gencnt; 2167 struct xunpgen *xug; 2168 struct unp_head *head; 2169 struct xunpcb *xu; 2170 u_int i; 2171 int error, n; 2172 2173 switch ((intptr_t)arg1) { 2174 case SOCK_STREAM: 2175 head = &unp_shead; 2176 break; 2177 2178 case SOCK_DGRAM: 2179 head = &unp_dhead; 2180 break; 2181 2182 case SOCK_SEQPACKET: 2183 head = &unp_sphead; 2184 break; 2185 2186 default: 2187 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2188 } 2189 2190 /* 2191 * The process of preparing the PCB list is too time-consuming and 2192 * resource-intensive to repeat twice on every request. 2193 */ 2194 if (req->oldptr == NULL) { 2195 n = unp_count; 2196 req->oldidx = 2 * (sizeof *xug) 2197 + (n + n/8) * sizeof(struct xunpcb); 2198 return (0); 2199 } 2200 2201 if (req->newptr != NULL) 2202 return (EPERM); 2203 2204 /* 2205 * OK, now we're committed to doing something. 2206 */ 2207 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2208 UNP_LINK_RLOCK(); 2209 gencnt = unp_gencnt; 2210 n = unp_count; 2211 UNP_LINK_RUNLOCK(); 2212 2213 xug->xug_len = sizeof *xug; 2214 xug->xug_count = n; 2215 xug->xug_gen = gencnt; 2216 xug->xug_sogen = so_gencnt; 2217 error = SYSCTL_OUT(req, xug, sizeof *xug); 2218 if (error) { 2219 free(xug, M_TEMP); 2220 return (error); 2221 } 2222 2223 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2224 2225 UNP_LINK_RLOCK(); 2226 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2227 unp = LIST_NEXT(unp, unp_link)) { 2228 UNP_PCB_LOCK(unp); 2229 if (unp->unp_gencnt <= gencnt) { 2230 if (cr_cansee(req->td->td_ucred, 2231 unp->unp_socket->so_cred)) { 2232 UNP_PCB_UNLOCK(unp); 2233 continue; 2234 } 2235 unp_list[i++] = unp; 2236 unp_pcb_hold(unp); 2237 } 2238 UNP_PCB_UNLOCK(unp); 2239 } 2240 UNP_LINK_RUNLOCK(); 2241 n = i; /* In case we lost some during malloc. */ 2242 2243 error = 0; 2244 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2245 for (i = 0; i < n; i++) { 2246 unp = unp_list[i]; 2247 UNP_PCB_LOCK(unp); 2248 if (unp_pcb_rele(unp)) 2249 continue; 2250 2251 if (unp->unp_gencnt <= gencnt) { 2252 xu->xu_len = sizeof *xu; 2253 xu->xu_unpp = (uintptr_t)unp; 2254 /* 2255 * XXX - need more locking here to protect against 2256 * connect/disconnect races for SMP. 2257 */ 2258 if (unp->unp_addr != NULL) 2259 bcopy(unp->unp_addr, &xu->xu_addr, 2260 unp->unp_addr->sun_len); 2261 else 2262 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2263 if (unp->unp_conn != NULL && 2264 unp->unp_conn->unp_addr != NULL) 2265 bcopy(unp->unp_conn->unp_addr, 2266 &xu->xu_caddr, 2267 unp->unp_conn->unp_addr->sun_len); 2268 else 2269 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2270 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2271 xu->unp_conn = (uintptr_t)unp->unp_conn; 2272 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2273 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2274 xu->unp_gencnt = unp->unp_gencnt; 2275 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2276 UNP_PCB_UNLOCK(unp); 2277 error = SYSCTL_OUT(req, xu, sizeof *xu); 2278 } else { 2279 UNP_PCB_UNLOCK(unp); 2280 } 2281 } 2282 free(xu, M_TEMP); 2283 if (!error) { 2284 /* 2285 * Give the user an updated idea of our state. If the 2286 * generation differs from what we told her before, she knows 2287 * that something happened while we were processing this 2288 * request, and it might be necessary to retry. 2289 */ 2290 xug->xug_gen = unp_gencnt; 2291 xug->xug_sogen = so_gencnt; 2292 xug->xug_count = unp_count; 2293 error = SYSCTL_OUT(req, xug, sizeof *xug); 2294 } 2295 free(unp_list, M_TEMP); 2296 free(xug, M_TEMP); 2297 return (error); 2298 } 2299 2300 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2301 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2302 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2303 "List of active local datagram sockets"); 2304 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2305 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2306 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2307 "List of active local stream sockets"); 2308 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2309 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2310 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2311 "List of active local seqpacket sockets"); 2312 2313 static void 2314 unp_shutdown(struct unpcb *unp) 2315 { 2316 struct unpcb *unp2; 2317 struct socket *so; 2318 2319 UNP_PCB_LOCK_ASSERT(unp); 2320 2321 unp2 = unp->unp_conn; 2322 if ((unp->unp_socket->so_type == SOCK_STREAM || 2323 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2324 so = unp2->unp_socket; 2325 if (so != NULL) 2326 socantrcvmore(so); 2327 } 2328 } 2329 2330 static void 2331 unp_drop(struct unpcb *unp) 2332 { 2333 struct socket *so; 2334 struct unpcb *unp2; 2335 2336 /* 2337 * Regardless of whether the socket's peer dropped the connection 2338 * with this socket by aborting or disconnecting, POSIX requires 2339 * that ECONNRESET is returned. 2340 */ 2341 2342 UNP_PCB_LOCK(unp); 2343 so = unp->unp_socket; 2344 if (so) 2345 so->so_error = ECONNRESET; 2346 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2347 /* Last reference dropped in unp_disconnect(). */ 2348 unp_pcb_rele_notlast(unp); 2349 unp_disconnect(unp, unp2); 2350 } else if (!unp_pcb_rele(unp)) { 2351 UNP_PCB_UNLOCK(unp); 2352 } 2353 } 2354 2355 static void 2356 unp_freerights(struct filedescent **fdep, int fdcount) 2357 { 2358 struct file *fp; 2359 int i; 2360 2361 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2362 2363 for (i = 0; i < fdcount; i++) { 2364 fp = fdep[i]->fde_file; 2365 filecaps_free(&fdep[i]->fde_caps); 2366 unp_discard(fp); 2367 } 2368 free(fdep[0], M_FILECAPS); 2369 } 2370 2371 static int 2372 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2373 { 2374 struct thread *td = curthread; /* XXX */ 2375 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2376 int i; 2377 int *fdp; 2378 struct filedesc *fdesc = td->td_proc->p_fd; 2379 struct filedescent **fdep; 2380 void *data; 2381 socklen_t clen = control->m_len, datalen; 2382 int error, newfds; 2383 u_int newlen; 2384 2385 UNP_LINK_UNLOCK_ASSERT(); 2386 2387 error = 0; 2388 if (controlp != NULL) /* controlp == NULL => free control messages */ 2389 *controlp = NULL; 2390 while (cm != NULL) { 2391 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2392 2393 data = CMSG_DATA(cm); 2394 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2395 if (cm->cmsg_level == SOL_SOCKET 2396 && cm->cmsg_type == SCM_RIGHTS) { 2397 newfds = datalen / sizeof(*fdep); 2398 if (newfds == 0) 2399 goto next; 2400 fdep = data; 2401 2402 /* If we're not outputting the descriptors free them. */ 2403 if (error || controlp == NULL) { 2404 unp_freerights(fdep, newfds); 2405 goto next; 2406 } 2407 FILEDESC_XLOCK(fdesc); 2408 2409 /* 2410 * Now change each pointer to an fd in the global 2411 * table to an integer that is the index to the local 2412 * fd table entry that we set up to point to the 2413 * global one we are transferring. 2414 */ 2415 newlen = newfds * sizeof(int); 2416 *controlp = sbcreatecontrol(NULL, newlen, 2417 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2418 2419 fdp = (int *) 2420 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2421 if ((error = fdallocn(td, 0, fdp, newfds))) { 2422 FILEDESC_XUNLOCK(fdesc); 2423 unp_freerights(fdep, newfds); 2424 m_freem(*controlp); 2425 *controlp = NULL; 2426 goto next; 2427 } 2428 for (i = 0; i < newfds; i++, fdp++) { 2429 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2430 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2431 &fdep[i]->fde_caps); 2432 unp_externalize_fp(fdep[i]->fde_file); 2433 } 2434 2435 /* 2436 * The new type indicates that the mbuf data refers to 2437 * kernel resources that may need to be released before 2438 * the mbuf is freed. 2439 */ 2440 m_chtype(*controlp, MT_EXTCONTROL); 2441 FILEDESC_XUNLOCK(fdesc); 2442 free(fdep[0], M_FILECAPS); 2443 } else { 2444 /* We can just copy anything else across. */ 2445 if (error || controlp == NULL) 2446 goto next; 2447 *controlp = sbcreatecontrol(NULL, datalen, 2448 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2449 bcopy(data, 2450 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2451 datalen); 2452 } 2453 controlp = &(*controlp)->m_next; 2454 2455 next: 2456 if (CMSG_SPACE(datalen) < clen) { 2457 clen -= CMSG_SPACE(datalen); 2458 cm = (struct cmsghdr *) 2459 ((caddr_t)cm + CMSG_SPACE(datalen)); 2460 } else { 2461 clen = 0; 2462 cm = NULL; 2463 } 2464 } 2465 2466 m_freem(control); 2467 return (error); 2468 } 2469 2470 static void 2471 unp_zone_change(void *tag) 2472 { 2473 2474 uma_zone_set_max(unp_zone, maxsockets); 2475 } 2476 2477 #ifdef INVARIANTS 2478 static void 2479 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2480 { 2481 struct unpcb *unp; 2482 2483 unp = mem; 2484 2485 KASSERT(LIST_EMPTY(&unp->unp_refs), 2486 ("%s: unpcb %p has lingering refs", __func__, unp)); 2487 KASSERT(unp->unp_socket == NULL, 2488 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2489 KASSERT(unp->unp_vnode == NULL, 2490 ("%s: unpcb %p has vnode references", __func__, unp)); 2491 KASSERT(unp->unp_conn == NULL, 2492 ("%s: unpcb %p is still connected", __func__, unp)); 2493 KASSERT(unp->unp_addr == NULL, 2494 ("%s: unpcb %p has leaked addr", __func__, unp)); 2495 } 2496 #endif 2497 2498 static void 2499 unp_init(void *arg __unused) 2500 { 2501 uma_dtor dtor; 2502 2503 #ifdef INVARIANTS 2504 dtor = unp_zdtor; 2505 #else 2506 dtor = NULL; 2507 #endif 2508 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2509 NULL, NULL, UMA_ALIGN_CACHE, 0); 2510 uma_zone_set_max(unp_zone, maxsockets); 2511 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2512 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2513 NULL, EVENTHANDLER_PRI_ANY); 2514 LIST_INIT(&unp_dhead); 2515 LIST_INIT(&unp_shead); 2516 LIST_INIT(&unp_sphead); 2517 SLIST_INIT(&unp_defers); 2518 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2519 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2520 UNP_LINK_LOCK_INIT(); 2521 UNP_DEFERRED_LOCK_INIT(); 2522 } 2523 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2524 2525 static void 2526 unp_internalize_cleanup_rights(struct mbuf *control) 2527 { 2528 struct cmsghdr *cp; 2529 struct mbuf *m; 2530 void *data; 2531 socklen_t datalen; 2532 2533 for (m = control; m != NULL; m = m->m_next) { 2534 cp = mtod(m, struct cmsghdr *); 2535 if (cp->cmsg_level != SOL_SOCKET || 2536 cp->cmsg_type != SCM_RIGHTS) 2537 continue; 2538 data = CMSG_DATA(cp); 2539 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2540 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2541 } 2542 } 2543 2544 static int 2545 unp_internalize(struct mbuf **controlp, struct thread *td, 2546 struct mbuf **clast, u_int *space, u_int *mbcnt) 2547 { 2548 struct mbuf *control, **initial_controlp; 2549 struct proc *p; 2550 struct filedesc *fdesc; 2551 struct bintime *bt; 2552 struct cmsghdr *cm; 2553 struct cmsgcred *cmcred; 2554 struct filedescent *fde, **fdep, *fdev; 2555 struct file *fp; 2556 struct timeval *tv; 2557 struct timespec *ts; 2558 void *data; 2559 socklen_t clen, datalen; 2560 int i, j, error, *fdp, oldfds; 2561 u_int newlen; 2562 2563 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2564 UNP_LINK_UNLOCK_ASSERT(); 2565 2566 p = td->td_proc; 2567 fdesc = p->p_fd; 2568 error = 0; 2569 control = *controlp; 2570 *controlp = NULL; 2571 initial_controlp = controlp; 2572 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2573 data = CMSG_DATA(cm); 2574 2575 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2576 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2577 (char *)cm + cm->cmsg_len >= (char *)data; 2578 2579 clen -= min(CMSG_SPACE(datalen), clen), 2580 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2581 data = CMSG_DATA(cm)) { 2582 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2583 switch (cm->cmsg_type) { 2584 case SCM_CREDS: 2585 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2586 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2587 cmcred = (struct cmsgcred *) 2588 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2589 cmcred->cmcred_pid = p->p_pid; 2590 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2591 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2592 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2593 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2594 CMGROUP_MAX); 2595 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2596 cmcred->cmcred_groups[i] = 2597 td->td_ucred->cr_groups[i]; 2598 break; 2599 2600 case SCM_RIGHTS: 2601 oldfds = datalen / sizeof (int); 2602 if (oldfds == 0) 2603 continue; 2604 /* On some machines sizeof pointer is bigger than 2605 * sizeof int, so we need to check if data fits into 2606 * single mbuf. We could allocate several mbufs, and 2607 * unp_externalize() should even properly handle that. 2608 * But it is not worth to complicate the code for an 2609 * insane scenario of passing over 200 file descriptors 2610 * at once. 2611 */ 2612 newlen = oldfds * sizeof(fdep[0]); 2613 if (CMSG_SPACE(newlen) > MCLBYTES) { 2614 error = EMSGSIZE; 2615 goto out; 2616 } 2617 /* 2618 * Check that all the FDs passed in refer to legal 2619 * files. If not, reject the entire operation. 2620 */ 2621 fdp = data; 2622 FILEDESC_SLOCK(fdesc); 2623 for (i = 0; i < oldfds; i++, fdp++) { 2624 fp = fget_noref(fdesc, *fdp); 2625 if (fp == NULL) { 2626 FILEDESC_SUNLOCK(fdesc); 2627 error = EBADF; 2628 goto out; 2629 } 2630 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2631 FILEDESC_SUNLOCK(fdesc); 2632 error = EOPNOTSUPP; 2633 goto out; 2634 } 2635 } 2636 2637 /* 2638 * Now replace the integer FDs with pointers to the 2639 * file structure and capability rights. 2640 */ 2641 *controlp = sbcreatecontrol(NULL, newlen, 2642 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2643 fdp = data; 2644 for (i = 0; i < oldfds; i++, fdp++) { 2645 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2646 fdp = data; 2647 for (j = 0; j < i; j++, fdp++) { 2648 fdrop(fdesc->fd_ofiles[*fdp]. 2649 fde_file, td); 2650 } 2651 FILEDESC_SUNLOCK(fdesc); 2652 error = EBADF; 2653 goto out; 2654 } 2655 } 2656 fdp = data; 2657 fdep = (struct filedescent **) 2658 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2659 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2660 M_WAITOK); 2661 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2662 fde = &fdesc->fd_ofiles[*fdp]; 2663 fdep[i] = fdev; 2664 fdep[i]->fde_file = fde->fde_file; 2665 filecaps_copy(&fde->fde_caps, 2666 &fdep[i]->fde_caps, true); 2667 unp_internalize_fp(fdep[i]->fde_file); 2668 } 2669 FILEDESC_SUNLOCK(fdesc); 2670 break; 2671 2672 case SCM_TIMESTAMP: 2673 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2674 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2675 tv = (struct timeval *) 2676 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2677 microtime(tv); 2678 break; 2679 2680 case SCM_BINTIME: 2681 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2682 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2683 bt = (struct bintime *) 2684 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2685 bintime(bt); 2686 break; 2687 2688 case SCM_REALTIME: 2689 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2690 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2691 ts = (struct timespec *) 2692 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2693 nanotime(ts); 2694 break; 2695 2696 case SCM_MONOTONIC: 2697 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2698 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2699 ts = (struct timespec *) 2700 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2701 nanouptime(ts); 2702 break; 2703 2704 default: 2705 error = EINVAL; 2706 goto out; 2707 } 2708 2709 if (space != NULL) { 2710 *space += (*controlp)->m_len; 2711 *mbcnt += MSIZE; 2712 if ((*controlp)->m_flags & M_EXT) 2713 *mbcnt += (*controlp)->m_ext.ext_size; 2714 *clast = *controlp; 2715 } 2716 controlp = &(*controlp)->m_next; 2717 } 2718 if (clen > 0) 2719 error = EINVAL; 2720 2721 out: 2722 if (error != 0 && initial_controlp != NULL) 2723 unp_internalize_cleanup_rights(*initial_controlp); 2724 m_freem(control); 2725 return (error); 2726 } 2727 2728 static struct mbuf * 2729 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2730 struct mbuf **clast, u_int *space, u_int *mbcnt) 2731 { 2732 struct mbuf *m, *n, *n_prev; 2733 const struct cmsghdr *cm; 2734 int ngroups, i, cmsgtype; 2735 size_t ctrlsz; 2736 2737 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2738 if (mode & UNP_WANTCRED_ALWAYS) { 2739 ctrlsz = SOCKCRED2SIZE(ngroups); 2740 cmsgtype = SCM_CREDS2; 2741 } else { 2742 ctrlsz = SOCKCREDSIZE(ngroups); 2743 cmsgtype = SCM_CREDS; 2744 } 2745 2746 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2747 if (m == NULL) 2748 return (control); 2749 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2750 2751 if (mode & UNP_WANTCRED_ALWAYS) { 2752 struct sockcred2 *sc; 2753 2754 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2755 sc->sc_version = 0; 2756 sc->sc_pid = td->td_proc->p_pid; 2757 sc->sc_uid = td->td_ucred->cr_ruid; 2758 sc->sc_euid = td->td_ucred->cr_uid; 2759 sc->sc_gid = td->td_ucred->cr_rgid; 2760 sc->sc_egid = td->td_ucred->cr_gid; 2761 sc->sc_ngroups = ngroups; 2762 for (i = 0; i < sc->sc_ngroups; i++) 2763 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2764 } else { 2765 struct sockcred *sc; 2766 2767 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2768 sc->sc_uid = td->td_ucred->cr_ruid; 2769 sc->sc_euid = td->td_ucred->cr_uid; 2770 sc->sc_gid = td->td_ucred->cr_rgid; 2771 sc->sc_egid = td->td_ucred->cr_gid; 2772 sc->sc_ngroups = ngroups; 2773 for (i = 0; i < sc->sc_ngroups; i++) 2774 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2775 } 2776 2777 /* 2778 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2779 * created SCM_CREDS control message (struct sockcred) has another 2780 * format. 2781 */ 2782 if (control != NULL && cmsgtype == SCM_CREDS) 2783 for (n = control, n_prev = NULL; n != NULL;) { 2784 cm = mtod(n, struct cmsghdr *); 2785 if (cm->cmsg_level == SOL_SOCKET && 2786 cm->cmsg_type == SCM_CREDS) { 2787 if (n_prev == NULL) 2788 control = n->m_next; 2789 else 2790 n_prev->m_next = n->m_next; 2791 if (space != NULL) { 2792 MPASS(*space >= n->m_len); 2793 *space -= n->m_len; 2794 MPASS(*mbcnt >= MSIZE); 2795 *mbcnt -= MSIZE; 2796 if (n->m_flags & M_EXT) { 2797 MPASS(*mbcnt >= 2798 n->m_ext.ext_size); 2799 *mbcnt -= n->m_ext.ext_size; 2800 } 2801 MPASS(clast); 2802 if (*clast == n) { 2803 MPASS(n->m_next == NULL); 2804 if (n_prev == NULL) 2805 *clast = m; 2806 else 2807 *clast = n_prev; 2808 } 2809 } 2810 n = m_free(n); 2811 } else { 2812 n_prev = n; 2813 n = n->m_next; 2814 } 2815 } 2816 2817 /* Prepend it to the head. */ 2818 m->m_next = control; 2819 if (space != NULL) { 2820 *space += m->m_len; 2821 *mbcnt += MSIZE; 2822 if (control == NULL) 2823 *clast = m; 2824 } 2825 return (m); 2826 } 2827 2828 static struct unpcb * 2829 fptounp(struct file *fp) 2830 { 2831 struct socket *so; 2832 2833 if (fp->f_type != DTYPE_SOCKET) 2834 return (NULL); 2835 if ((so = fp->f_data) == NULL) 2836 return (NULL); 2837 if (so->so_proto->pr_domain != &localdomain) 2838 return (NULL); 2839 return sotounpcb(so); 2840 } 2841 2842 static void 2843 unp_discard(struct file *fp) 2844 { 2845 struct unp_defer *dr; 2846 2847 if (unp_externalize_fp(fp)) { 2848 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2849 dr->ud_fp = fp; 2850 UNP_DEFERRED_LOCK(); 2851 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2852 UNP_DEFERRED_UNLOCK(); 2853 atomic_add_int(&unp_defers_count, 1); 2854 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2855 } else 2856 closef_nothread(fp); 2857 } 2858 2859 static void 2860 unp_process_defers(void *arg __unused, int pending) 2861 { 2862 struct unp_defer *dr; 2863 SLIST_HEAD(, unp_defer) drl; 2864 int count; 2865 2866 SLIST_INIT(&drl); 2867 for (;;) { 2868 UNP_DEFERRED_LOCK(); 2869 if (SLIST_FIRST(&unp_defers) == NULL) { 2870 UNP_DEFERRED_UNLOCK(); 2871 break; 2872 } 2873 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2874 UNP_DEFERRED_UNLOCK(); 2875 count = 0; 2876 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2877 SLIST_REMOVE_HEAD(&drl, ud_link); 2878 closef_nothread(dr->ud_fp); 2879 free(dr, M_TEMP); 2880 count++; 2881 } 2882 atomic_add_int(&unp_defers_count, -count); 2883 } 2884 } 2885 2886 static void 2887 unp_internalize_fp(struct file *fp) 2888 { 2889 struct unpcb *unp; 2890 2891 UNP_LINK_WLOCK(); 2892 if ((unp = fptounp(fp)) != NULL) { 2893 unp->unp_file = fp; 2894 unp->unp_msgcount++; 2895 } 2896 unp_rights++; 2897 UNP_LINK_WUNLOCK(); 2898 } 2899 2900 static int 2901 unp_externalize_fp(struct file *fp) 2902 { 2903 struct unpcb *unp; 2904 int ret; 2905 2906 UNP_LINK_WLOCK(); 2907 if ((unp = fptounp(fp)) != NULL) { 2908 unp->unp_msgcount--; 2909 ret = 1; 2910 } else 2911 ret = 0; 2912 unp_rights--; 2913 UNP_LINK_WUNLOCK(); 2914 return (ret); 2915 } 2916 2917 /* 2918 * unp_defer indicates whether additional work has been defered for a future 2919 * pass through unp_gc(). It is thread local and does not require explicit 2920 * synchronization. 2921 */ 2922 static int unp_marked; 2923 2924 static void 2925 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2926 { 2927 struct unpcb *unp; 2928 struct file *fp; 2929 int i; 2930 2931 /* 2932 * This function can only be called from the gc task. 2933 */ 2934 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2935 ("%s: not on gc callout", __func__)); 2936 UNP_LINK_LOCK_ASSERT(); 2937 2938 for (i = 0; i < fdcount; i++) { 2939 fp = fdep[i]->fde_file; 2940 if ((unp = fptounp(fp)) == NULL) 2941 continue; 2942 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2943 continue; 2944 unp->unp_gcrefs--; 2945 } 2946 } 2947 2948 static void 2949 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2950 { 2951 struct unpcb *unp; 2952 struct file *fp; 2953 int i; 2954 2955 /* 2956 * This function can only be called from the gc task. 2957 */ 2958 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2959 ("%s: not on gc callout", __func__)); 2960 UNP_LINK_LOCK_ASSERT(); 2961 2962 for (i = 0; i < fdcount; i++) { 2963 fp = fdep[i]->fde_file; 2964 if ((unp = fptounp(fp)) == NULL) 2965 continue; 2966 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2967 continue; 2968 unp->unp_gcrefs++; 2969 unp_marked++; 2970 } 2971 } 2972 2973 static void 2974 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 2975 { 2976 struct sockbuf *sb; 2977 2978 SOCK_LOCK_ASSERT(so); 2979 2980 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 2981 return; 2982 2983 SOCK_RECVBUF_LOCK(so); 2984 switch (so->so_type) { 2985 case SOCK_DGRAM: 2986 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 2987 unp_scan(so->so_rcv.uxdg_peeked, op); 2988 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 2989 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 2990 break; 2991 case SOCK_STREAM: 2992 case SOCK_SEQPACKET: 2993 unp_scan(so->so_rcv.sb_mb, op); 2994 break; 2995 } 2996 SOCK_RECVBUF_UNLOCK(so); 2997 } 2998 2999 static void 3000 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3001 { 3002 struct socket *so, *soa; 3003 3004 so = unp->unp_socket; 3005 SOCK_LOCK(so); 3006 if (SOLISTENING(so)) { 3007 /* 3008 * Mark all sockets in our accept queue. 3009 */ 3010 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3011 unp_scan_socket(soa, op); 3012 } else { 3013 /* 3014 * Mark all sockets we reference with RIGHTS. 3015 */ 3016 unp_scan_socket(so, op); 3017 } 3018 SOCK_UNLOCK(so); 3019 } 3020 3021 static int unp_recycled; 3022 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3023 "Number of unreachable sockets claimed by the garbage collector."); 3024 3025 static int unp_taskcount; 3026 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3027 "Number of times the garbage collector has run."); 3028 3029 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3030 "Number of active local sockets."); 3031 3032 static void 3033 unp_gc(__unused void *arg, int pending) 3034 { 3035 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3036 NULL }; 3037 struct unp_head **head; 3038 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3039 struct file *f, **unref; 3040 struct unpcb *unp, *unptmp; 3041 int i, total, unp_unreachable; 3042 3043 LIST_INIT(&unp_deadhead); 3044 unp_taskcount++; 3045 UNP_LINK_RLOCK(); 3046 /* 3047 * First determine which sockets may be in cycles. 3048 */ 3049 unp_unreachable = 0; 3050 3051 for (head = heads; *head != NULL; head++) 3052 LIST_FOREACH(unp, *head, unp_link) { 3053 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3054 ("%s: unp %p has unexpected gc flags 0x%x", 3055 __func__, unp, (unsigned int)unp->unp_gcflag)); 3056 3057 f = unp->unp_file; 3058 3059 /* 3060 * Check for an unreachable socket potentially in a 3061 * cycle. It must be in a queue as indicated by 3062 * msgcount, and this must equal the file reference 3063 * count. Note that when msgcount is 0 the file is 3064 * NULL. 3065 */ 3066 if (f != NULL && unp->unp_msgcount != 0 && 3067 refcount_load(&f->f_count) == unp->unp_msgcount) { 3068 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3069 unp->unp_gcflag |= UNPGC_DEAD; 3070 unp->unp_gcrefs = unp->unp_msgcount; 3071 unp_unreachable++; 3072 } 3073 } 3074 3075 /* 3076 * Scan all sockets previously marked as potentially being in a cycle 3077 * and remove the references each socket holds on any UNPGC_DEAD 3078 * sockets in its queue. After this step, all remaining references on 3079 * sockets marked UNPGC_DEAD should not be part of any cycle. 3080 */ 3081 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3082 unp_gc_scan(unp, unp_remove_dead_ref); 3083 3084 /* 3085 * If a socket still has a non-negative refcount, it cannot be in a 3086 * cycle. In this case increment refcount of all children iteratively. 3087 * Stop the scan once we do a complete loop without discovering 3088 * a new reachable socket. 3089 */ 3090 do { 3091 unp_marked = 0; 3092 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3093 if (unp->unp_gcrefs > 0) { 3094 unp->unp_gcflag &= ~UNPGC_DEAD; 3095 LIST_REMOVE(unp, unp_dead); 3096 KASSERT(unp_unreachable > 0, 3097 ("%s: unp_unreachable underflow.", 3098 __func__)); 3099 unp_unreachable--; 3100 unp_gc_scan(unp, unp_restore_undead_ref); 3101 } 3102 } while (unp_marked); 3103 3104 UNP_LINK_RUNLOCK(); 3105 3106 if (unp_unreachable == 0) 3107 return; 3108 3109 /* 3110 * Allocate space for a local array of dead unpcbs. 3111 * TODO: can this path be simplified by instead using the local 3112 * dead list at unp_deadhead, after taking out references 3113 * on the file object and/or unpcb and dropping the link lock? 3114 */ 3115 unref = malloc(unp_unreachable * sizeof(struct file *), 3116 M_TEMP, M_WAITOK); 3117 3118 /* 3119 * Iterate looking for sockets which have been specifically marked 3120 * as unreachable and store them locally. 3121 */ 3122 UNP_LINK_RLOCK(); 3123 total = 0; 3124 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3125 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3126 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3127 unp->unp_gcflag &= ~UNPGC_DEAD; 3128 f = unp->unp_file; 3129 if (unp->unp_msgcount == 0 || f == NULL || 3130 refcount_load(&f->f_count) != unp->unp_msgcount || 3131 !fhold(f)) 3132 continue; 3133 unref[total++] = f; 3134 KASSERT(total <= unp_unreachable, 3135 ("%s: incorrect unreachable count.", __func__)); 3136 } 3137 UNP_LINK_RUNLOCK(); 3138 3139 /* 3140 * Now flush all sockets, free'ing rights. This will free the 3141 * struct files associated with these sockets but leave each socket 3142 * with one remaining ref. 3143 */ 3144 for (i = 0; i < total; i++) { 3145 struct socket *so; 3146 3147 so = unref[i]->f_data; 3148 CURVNET_SET(so->so_vnet); 3149 sorflush(so); 3150 CURVNET_RESTORE(); 3151 } 3152 3153 /* 3154 * And finally release the sockets so they can be reclaimed. 3155 */ 3156 for (i = 0; i < total; i++) 3157 fdrop(unref[i], NULL); 3158 unp_recycled += total; 3159 free(unref, M_TEMP); 3160 } 3161 3162 /* 3163 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3164 */ 3165 static void 3166 unp_dispose(struct socket *so) 3167 { 3168 struct sockbuf *sb; 3169 struct unpcb *unp; 3170 struct mbuf *m; 3171 3172 MPASS(!SOLISTENING(so)); 3173 3174 unp = sotounpcb(so); 3175 UNP_LINK_WLOCK(); 3176 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3177 UNP_LINK_WUNLOCK(); 3178 3179 /* 3180 * Grab our special mbufs before calling sbrelease(). 3181 */ 3182 SOCK_RECVBUF_LOCK(so); 3183 switch (so->so_type) { 3184 case SOCK_DGRAM: 3185 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3186 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3187 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3188 /* Note: socket of sb may reconnect. */ 3189 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3190 } 3191 sb = &so->so_rcv; 3192 if (sb->uxdg_peeked != NULL) { 3193 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3194 m_stailqpkt); 3195 sb->uxdg_peeked = NULL; 3196 } 3197 m = STAILQ_FIRST(&sb->uxdg_mb); 3198 STAILQ_INIT(&sb->uxdg_mb); 3199 /* XXX: our shortened sbrelease() */ 3200 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3201 RLIM_INFINITY); 3202 /* 3203 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3204 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3205 * data to the socket. 3206 * We are now in dom_dispose and it could be a call from 3207 * soshutdown() or from the final sofree(). The sofree() case 3208 * is simple as it guarantees that no more sends will happen, 3209 * however we can race with unp_disconnect() from our peer. 3210 * The shutdown(2) case is more exotic. It would call into 3211 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3212 * possible if we did connect(2) on this socket and we also 3213 * had it bound with bind(2) and receive connections from other 3214 * sockets. Because soshutdown() violates POSIX (see comment 3215 * there) we will end up here shutting down our receive side. 3216 * Of course this will have affect not only on the peer we 3217 * connect(2)ed to, but also on all of the peers who had 3218 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3219 */ 3220 sb->sb_state |= SBS_CANTRCVMORE; 3221 break; 3222 case SOCK_STREAM: 3223 case SOCK_SEQPACKET: 3224 sb = &so->so_rcv; 3225 m = sbcut_locked(sb, sb->sb_ccc); 3226 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3227 ("%s: ccc %u mb %p mbcnt %u", __func__, 3228 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3229 sbrelease_locked(so, SO_RCV); 3230 break; 3231 } 3232 SOCK_RECVBUF_UNLOCK(so); 3233 if (SOCK_IO_RECV_OWNED(so)) 3234 SOCK_IO_RECV_UNLOCK(so); 3235 3236 if (m != NULL) { 3237 unp_scan(m, unp_freerights); 3238 m_freem(m); 3239 } 3240 } 3241 3242 static void 3243 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3244 { 3245 struct mbuf *m; 3246 struct cmsghdr *cm; 3247 void *data; 3248 socklen_t clen, datalen; 3249 3250 while (m0 != NULL) { 3251 for (m = m0; m; m = m->m_next) { 3252 if (m->m_type != MT_CONTROL) 3253 continue; 3254 3255 cm = mtod(m, struct cmsghdr *); 3256 clen = m->m_len; 3257 3258 while (cm != NULL) { 3259 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3260 break; 3261 3262 data = CMSG_DATA(cm); 3263 datalen = (caddr_t)cm + cm->cmsg_len 3264 - (caddr_t)data; 3265 3266 if (cm->cmsg_level == SOL_SOCKET && 3267 cm->cmsg_type == SCM_RIGHTS) { 3268 (*op)(data, datalen / 3269 sizeof(struct filedescent *)); 3270 } 3271 3272 if (CMSG_SPACE(datalen) < clen) { 3273 clen -= CMSG_SPACE(datalen); 3274 cm = (struct cmsghdr *) 3275 ((caddr_t)cm + CMSG_SPACE(datalen)); 3276 } else { 3277 clen = 0; 3278 cm = NULL; 3279 } 3280 } 3281 } 3282 m0 = m0->m_nextpkt; 3283 } 3284 } 3285 3286 /* 3287 * Definitions of protocols supported in the LOCAL domain. 3288 */ 3289 static struct protosw streamproto = { 3290 .pr_type = SOCK_STREAM, 3291 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 3292 PR_CAPATTACH, 3293 .pr_ctloutput = &uipc_ctloutput, 3294 .pr_abort = uipc_abort, 3295 .pr_accept = uipc_peeraddr, 3296 .pr_attach = uipc_attach, 3297 .pr_bind = uipc_bind, 3298 .pr_bindat = uipc_bindat, 3299 .pr_connect = uipc_connect, 3300 .pr_connectat = uipc_connectat, 3301 .pr_connect2 = uipc_connect2, 3302 .pr_detach = uipc_detach, 3303 .pr_disconnect = uipc_disconnect, 3304 .pr_listen = uipc_listen, 3305 .pr_peeraddr = uipc_peeraddr, 3306 .pr_rcvd = uipc_rcvd, 3307 .pr_send = uipc_send, 3308 .pr_ready = uipc_ready, 3309 .pr_sense = uipc_sense, 3310 .pr_shutdown = uipc_shutdown, 3311 .pr_sockaddr = uipc_sockaddr, 3312 .pr_soreceive = soreceive_generic, 3313 .pr_close = uipc_close, 3314 }; 3315 3316 static struct protosw dgramproto = { 3317 .pr_type = SOCK_DGRAM, 3318 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 3319 PR_SOCKBUF, 3320 .pr_ctloutput = &uipc_ctloutput, 3321 .pr_abort = uipc_abort, 3322 .pr_accept = uipc_peeraddr, 3323 .pr_attach = uipc_attach, 3324 .pr_bind = uipc_bind, 3325 .pr_bindat = uipc_bindat, 3326 .pr_connect = uipc_connect, 3327 .pr_connectat = uipc_connectat, 3328 .pr_connect2 = uipc_connect2, 3329 .pr_detach = uipc_detach, 3330 .pr_disconnect = uipc_disconnect, 3331 .pr_peeraddr = uipc_peeraddr, 3332 .pr_sosend = uipc_sosend_dgram, 3333 .pr_sense = uipc_sense, 3334 .pr_shutdown = uipc_shutdown, 3335 .pr_sockaddr = uipc_sockaddr, 3336 .pr_soreceive = uipc_soreceive_dgram, 3337 .pr_close = uipc_close, 3338 }; 3339 3340 static struct protosw seqpacketproto = { 3341 .pr_type = SOCK_SEQPACKET, 3342 /* 3343 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3344 * due to our use of sbappendaddr. A new sbappend variants is needed 3345 * that supports both atomic record writes and control data. 3346 */ 3347 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 3348 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 3349 .pr_ctloutput = &uipc_ctloutput, 3350 .pr_abort = uipc_abort, 3351 .pr_accept = uipc_peeraddr, 3352 .pr_attach = uipc_attach, 3353 .pr_bind = uipc_bind, 3354 .pr_bindat = uipc_bindat, 3355 .pr_connect = uipc_connect, 3356 .pr_connectat = uipc_connectat, 3357 .pr_connect2 = uipc_connect2, 3358 .pr_detach = uipc_detach, 3359 .pr_disconnect = uipc_disconnect, 3360 .pr_listen = uipc_listen, 3361 .pr_peeraddr = uipc_peeraddr, 3362 .pr_rcvd = uipc_rcvd, 3363 .pr_send = uipc_send, 3364 .pr_sense = uipc_sense, 3365 .pr_shutdown = uipc_shutdown, 3366 .pr_sockaddr = uipc_sockaddr, 3367 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3368 .pr_close = uipc_close, 3369 }; 3370 3371 static struct domain localdomain = { 3372 .dom_family = AF_LOCAL, 3373 .dom_name = "local", 3374 .dom_externalize = unp_externalize, 3375 .dom_dispose = unp_dispose, 3376 .dom_nprotosw = 3, 3377 .dom_protosw = { 3378 &streamproto, 3379 &dgramproto, 3380 &seqpacketproto, 3381 } 3382 }; 3383 DOMAIN_SET(local); 3384 3385 /* 3386 * A helper function called by VFS before socket-type vnode reclamation. 3387 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3388 * use count. 3389 */ 3390 void 3391 vfs_unp_reclaim(struct vnode *vp) 3392 { 3393 struct unpcb *unp; 3394 int active; 3395 struct mtx *vplock; 3396 3397 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3398 KASSERT(vp->v_type == VSOCK, 3399 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3400 3401 active = 0; 3402 vplock = mtx_pool_find(mtxpool_sleep, vp); 3403 mtx_lock(vplock); 3404 VOP_UNP_CONNECT(vp, &unp); 3405 if (unp == NULL) 3406 goto done; 3407 UNP_PCB_LOCK(unp); 3408 if (unp->unp_vnode == vp) { 3409 VOP_UNP_DETACH(vp); 3410 unp->unp_vnode = NULL; 3411 active = 1; 3412 } 3413 UNP_PCB_UNLOCK(unp); 3414 done: 3415 mtx_unlock(vplock); 3416 if (active) 3417 vunref(vp); 3418 } 3419 3420 #ifdef DDB 3421 static void 3422 db_print_indent(int indent) 3423 { 3424 int i; 3425 3426 for (i = 0; i < indent; i++) 3427 db_printf(" "); 3428 } 3429 3430 static void 3431 db_print_unpflags(int unp_flags) 3432 { 3433 int comma; 3434 3435 comma = 0; 3436 if (unp_flags & UNP_HAVEPC) { 3437 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3438 comma = 1; 3439 } 3440 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3441 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3442 comma = 1; 3443 } 3444 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3445 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3446 comma = 1; 3447 } 3448 if (unp_flags & UNP_CONNWAIT) { 3449 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3450 comma = 1; 3451 } 3452 if (unp_flags & UNP_CONNECTING) { 3453 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3454 comma = 1; 3455 } 3456 if (unp_flags & UNP_BINDING) { 3457 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3458 comma = 1; 3459 } 3460 } 3461 3462 static void 3463 db_print_xucred(int indent, struct xucred *xu) 3464 { 3465 int comma, i; 3466 3467 db_print_indent(indent); 3468 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3469 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3470 db_print_indent(indent); 3471 db_printf("cr_groups: "); 3472 comma = 0; 3473 for (i = 0; i < xu->cr_ngroups; i++) { 3474 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3475 comma = 1; 3476 } 3477 db_printf("\n"); 3478 } 3479 3480 static void 3481 db_print_unprefs(int indent, struct unp_head *uh) 3482 { 3483 struct unpcb *unp; 3484 int counter; 3485 3486 counter = 0; 3487 LIST_FOREACH(unp, uh, unp_reflink) { 3488 if (counter % 4 == 0) 3489 db_print_indent(indent); 3490 db_printf("%p ", unp); 3491 if (counter % 4 == 3) 3492 db_printf("\n"); 3493 counter++; 3494 } 3495 if (counter != 0 && counter % 4 != 0) 3496 db_printf("\n"); 3497 } 3498 3499 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3500 { 3501 struct unpcb *unp; 3502 3503 if (!have_addr) { 3504 db_printf("usage: show unpcb <addr>\n"); 3505 return; 3506 } 3507 unp = (struct unpcb *)addr; 3508 3509 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3510 unp->unp_vnode); 3511 3512 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3513 unp->unp_conn); 3514 3515 db_printf("unp_refs:\n"); 3516 db_print_unprefs(2, &unp->unp_refs); 3517 3518 /* XXXRW: Would be nice to print the full address, if any. */ 3519 db_printf("unp_addr: %p\n", unp->unp_addr); 3520 3521 db_printf("unp_gencnt: %llu\n", 3522 (unsigned long long)unp->unp_gencnt); 3523 3524 db_printf("unp_flags: %x (", unp->unp_flags); 3525 db_print_unpflags(unp->unp_flags); 3526 db_printf(")\n"); 3527 3528 db_printf("unp_peercred:\n"); 3529 db_print_xucred(2, &unp->unp_peercred); 3530 3531 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3532 } 3533 #endif 3534