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 if (!SOLISTENING(so)) 728 unp_dispose(so); 729 730 UNP_LINK_WLOCK(); 731 LIST_REMOVE(unp, unp_link); 732 if (unp->unp_gcflag & UNPGC_DEAD) 733 LIST_REMOVE(unp, unp_dead); 734 unp->unp_gencnt = ++unp_gencnt; 735 --unp_count; 736 UNP_LINK_WUNLOCK(); 737 738 UNP_PCB_UNLOCK_ASSERT(unp); 739 restart: 740 if ((vp = unp->unp_vnode) != NULL) { 741 vplock = mtx_pool_find(mtxpool_sleep, vp); 742 mtx_lock(vplock); 743 } 744 UNP_PCB_LOCK(unp); 745 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 746 if (vplock) 747 mtx_unlock(vplock); 748 UNP_PCB_UNLOCK(unp); 749 goto restart; 750 } 751 if ((vp = unp->unp_vnode) != NULL) { 752 VOP_UNP_DETACH(vp); 753 unp->unp_vnode = NULL; 754 } 755 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 756 unp_disconnect(unp, unp2); 757 else 758 UNP_PCB_UNLOCK(unp); 759 760 UNP_REF_LIST_LOCK(); 761 while (!LIST_EMPTY(&unp->unp_refs)) { 762 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 763 764 unp_pcb_hold(ref); 765 UNP_REF_LIST_UNLOCK(); 766 767 MPASS(ref != unp); 768 UNP_PCB_UNLOCK_ASSERT(ref); 769 unp_drop(ref); 770 UNP_REF_LIST_LOCK(); 771 } 772 UNP_REF_LIST_UNLOCK(); 773 774 UNP_PCB_LOCK(unp); 775 local_unp_rights = unp_rights; 776 unp->unp_socket->so_pcb = NULL; 777 unp->unp_socket = NULL; 778 free(unp->unp_addr, M_SONAME); 779 unp->unp_addr = NULL; 780 if (!unp_pcb_rele(unp)) 781 UNP_PCB_UNLOCK(unp); 782 if (vp) { 783 mtx_unlock(vplock); 784 vrele(vp); 785 } 786 if (local_unp_rights) 787 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 788 789 switch (so->so_type) { 790 case SOCK_DGRAM: 791 /* 792 * Everything should have been unlinked/freed by unp_dispose() 793 * and/or unp_disconnect(). 794 */ 795 MPASS(so->so_rcv.uxdg_peeked == NULL); 796 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 797 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 798 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 799 } 800 } 801 802 static int 803 uipc_disconnect(struct socket *so) 804 { 805 struct unpcb *unp, *unp2; 806 807 unp = sotounpcb(so); 808 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 809 810 UNP_PCB_LOCK(unp); 811 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 812 unp_disconnect(unp, unp2); 813 else 814 UNP_PCB_UNLOCK(unp); 815 return (0); 816 } 817 818 static int 819 uipc_listen(struct socket *so, int backlog, struct thread *td) 820 { 821 struct unpcb *unp; 822 int error; 823 824 MPASS(so->so_type != SOCK_DGRAM); 825 826 /* 827 * Synchronize with concurrent connection attempts. 828 */ 829 error = 0; 830 unp = sotounpcb(so); 831 UNP_PCB_LOCK(unp); 832 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 833 error = EINVAL; 834 else if (unp->unp_vnode == NULL) 835 error = EDESTADDRREQ; 836 if (error != 0) { 837 UNP_PCB_UNLOCK(unp); 838 return (error); 839 } 840 841 SOCK_LOCK(so); 842 error = solisten_proto_check(so); 843 if (error == 0) { 844 cru2xt(td, &unp->unp_peercred); 845 solisten_proto(so, backlog); 846 } 847 SOCK_UNLOCK(so); 848 UNP_PCB_UNLOCK(unp); 849 return (error); 850 } 851 852 static int 853 uipc_peeraddr(struct socket *so, struct sockaddr *ret) 854 { 855 struct unpcb *unp, *unp2; 856 const struct sockaddr *sa; 857 858 unp = sotounpcb(so); 859 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 860 861 UNP_PCB_LOCK(unp); 862 unp2 = unp_pcb_lock_peer(unp); 863 if (unp2 != NULL) { 864 if (unp2->unp_addr != NULL) 865 sa = (struct sockaddr *)unp2->unp_addr; 866 else 867 sa = &sun_noname; 868 bcopy(sa, ret, sa->sa_len); 869 unp_pcb_unlock_pair(unp, unp2); 870 } else { 871 UNP_PCB_UNLOCK(unp); 872 sa = &sun_noname; 873 bcopy(sa, ret, sa->sa_len); 874 } 875 return (0); 876 } 877 878 static int 879 uipc_rcvd(struct socket *so, int flags) 880 { 881 struct unpcb *unp, *unp2; 882 struct socket *so2; 883 u_int mbcnt, sbcc; 884 885 unp = sotounpcb(so); 886 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 887 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 888 ("%s: socktype %d", __func__, so->so_type)); 889 890 /* 891 * Adjust backpressure on sender and wakeup any waiting to write. 892 * 893 * The unp lock is acquired to maintain the validity of the unp_conn 894 * pointer; no lock on unp2 is required as unp2->unp_socket will be 895 * static as long as we don't permit unp2 to disconnect from unp, 896 * which is prevented by the lock on unp. We cache values from 897 * so_rcv to avoid holding the so_rcv lock over the entire 898 * transaction on the remote so_snd. 899 */ 900 SOCKBUF_LOCK(&so->so_rcv); 901 mbcnt = so->so_rcv.sb_mbcnt; 902 sbcc = sbavail(&so->so_rcv); 903 SOCKBUF_UNLOCK(&so->so_rcv); 904 /* 905 * There is a benign race condition at this point. If we're planning to 906 * clear SB_STOP, but uipc_send is called on the connected socket at 907 * this instant, it might add data to the sockbuf and set SB_STOP. Then 908 * we would erroneously clear SB_STOP below, even though the sockbuf is 909 * full. The race is benign because the only ill effect is to allow the 910 * sockbuf to exceed its size limit, and the size limits are not 911 * strictly guaranteed anyway. 912 */ 913 UNP_PCB_LOCK(unp); 914 unp2 = unp->unp_conn; 915 if (unp2 == NULL) { 916 UNP_PCB_UNLOCK(unp); 917 return (0); 918 } 919 so2 = unp2->unp_socket; 920 SOCKBUF_LOCK(&so2->so_snd); 921 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 922 so2->so_snd.sb_flags &= ~SB_STOP; 923 sowwakeup_locked(so2); 924 UNP_PCB_UNLOCK(unp); 925 return (0); 926 } 927 928 static int 929 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 930 struct mbuf *control, struct thread *td) 931 { 932 struct unpcb *unp, *unp2; 933 struct socket *so2; 934 u_int mbcnt, sbcc; 935 int error; 936 937 unp = sotounpcb(so); 938 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 939 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 940 ("%s: socktype %d", __func__, so->so_type)); 941 942 error = 0; 943 if (flags & PRUS_OOB) { 944 error = EOPNOTSUPP; 945 goto release; 946 } 947 if (control != NULL && 948 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 949 goto release; 950 951 unp2 = NULL; 952 if ((so->so_state & SS_ISCONNECTED) == 0) { 953 if (nam != NULL) { 954 if ((error = unp_connect(so, nam, td)) != 0) 955 goto out; 956 } else { 957 error = ENOTCONN; 958 goto out; 959 } 960 } 961 962 UNP_PCB_LOCK(unp); 963 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 964 UNP_PCB_UNLOCK(unp); 965 error = ENOTCONN; 966 goto out; 967 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 968 unp_pcb_unlock_pair(unp, unp2); 969 error = EPIPE; 970 goto out; 971 } 972 UNP_PCB_UNLOCK(unp); 973 if ((so2 = unp2->unp_socket) == NULL) { 974 UNP_PCB_UNLOCK(unp2); 975 error = ENOTCONN; 976 goto out; 977 } 978 SOCKBUF_LOCK(&so2->so_rcv); 979 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 980 /* 981 * Credentials are passed only once on SOCK_STREAM and 982 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 983 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 984 */ 985 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 986 NULL, NULL); 987 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 988 } 989 990 /* 991 * Send to paired receive port and wake up readers. Don't 992 * check for space available in the receive buffer if we're 993 * attaching ancillary data; Unix domain sockets only check 994 * for space in the sending sockbuf, and that check is 995 * performed one level up the stack. At that level we cannot 996 * precisely account for the amount of buffer space used 997 * (e.g., because control messages are not yet internalized). 998 */ 999 switch (so->so_type) { 1000 case SOCK_STREAM: 1001 if (control != NULL) { 1002 sbappendcontrol_locked(&so2->so_rcv, m, 1003 control, flags); 1004 control = NULL; 1005 } else 1006 sbappend_locked(&so2->so_rcv, m, flags); 1007 break; 1008 1009 case SOCK_SEQPACKET: 1010 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1011 &sun_noname, m, control)) 1012 control = NULL; 1013 break; 1014 } 1015 1016 mbcnt = so2->so_rcv.sb_mbcnt; 1017 sbcc = sbavail(&so2->so_rcv); 1018 if (sbcc) 1019 sorwakeup_locked(so2); 1020 else 1021 SOCKBUF_UNLOCK(&so2->so_rcv); 1022 1023 /* 1024 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1025 * it would be possible for uipc_rcvd to be called at this 1026 * point, drain the receiving sockbuf, clear SB_STOP, and then 1027 * we would set SB_STOP below. That could lead to an empty 1028 * sockbuf having SB_STOP set 1029 */ 1030 SOCKBUF_LOCK(&so->so_snd); 1031 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1032 so->so_snd.sb_flags |= SB_STOP; 1033 SOCKBUF_UNLOCK(&so->so_snd); 1034 UNP_PCB_UNLOCK(unp2); 1035 m = NULL; 1036 out: 1037 /* 1038 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1039 */ 1040 if (flags & PRUS_EOF) { 1041 UNP_PCB_LOCK(unp); 1042 socantsendmore(so); 1043 unp_shutdown(unp); 1044 UNP_PCB_UNLOCK(unp); 1045 } 1046 if (control != NULL && error != 0) 1047 unp_scan(control, unp_freerights); 1048 1049 release: 1050 if (control != NULL) 1051 m_freem(control); 1052 /* 1053 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1054 * for freeing memory. 1055 */ 1056 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1057 m_freem(m); 1058 return (error); 1059 } 1060 1061 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1062 static inline bool 1063 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1064 { 1065 u_int bleft, mleft; 1066 1067 /* 1068 * Negative space may happen if send(2) is followed by 1069 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1070 */ 1071 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1072 sb->sb_mbmax < sb->uxdg_mbcnt)) 1073 return (false); 1074 1075 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1076 return (false); 1077 1078 bleft = sb->sb_hiwat - sb->uxdg_cc; 1079 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1080 1081 return (bleft >= cc && mleft >= mbcnt); 1082 } 1083 1084 /* 1085 * PF_UNIX/SOCK_DGRAM send 1086 * 1087 * Allocate a record consisting of 3 mbufs in the sequence of 1088 * from -> control -> data and append it to the socket buffer. 1089 * 1090 * The first mbuf carries sender's name and is a pkthdr that stores 1091 * overall length of datagram, its memory consumption and control length. 1092 */ 1093 #define ctllen PH_loc.thirtytwo[1] 1094 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1095 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1096 static int 1097 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1098 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1099 { 1100 struct unpcb *unp, *unp2; 1101 const struct sockaddr *from; 1102 struct socket *so2; 1103 struct sockbuf *sb; 1104 struct mbuf *f, *clast; 1105 u_int cc, ctl, mbcnt; 1106 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1107 int error; 1108 1109 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1110 1111 error = 0; 1112 f = NULL; 1113 ctl = 0; 1114 1115 if (__predict_false(flags & MSG_OOB)) { 1116 error = EOPNOTSUPP; 1117 goto out; 1118 } 1119 if (m == NULL) { 1120 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1121 error = EMSGSIZE; 1122 goto out; 1123 } 1124 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1125 if (__predict_false(m == NULL)) { 1126 error = EFAULT; 1127 goto out; 1128 } 1129 f = m_gethdr(M_WAITOK, MT_SONAME); 1130 cc = m->m_pkthdr.len; 1131 mbcnt = MSIZE + m->m_pkthdr.memlen; 1132 if (c != NULL && 1133 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1134 goto out; 1135 } else { 1136 /* pr_sosend() with mbuf usually is a kernel thread. */ 1137 1138 M_ASSERTPKTHDR(m); 1139 if (__predict_false(c != NULL)) 1140 panic("%s: control from a kernel thread", __func__); 1141 1142 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1143 error = EMSGSIZE; 1144 goto out; 1145 } 1146 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1147 error = ENOBUFS; 1148 goto out; 1149 } 1150 /* Condition the foreign mbuf to our standards. */ 1151 m_clrprotoflags(m); 1152 m_tag_delete_chain(m, NULL); 1153 m->m_pkthdr.rcvif = NULL; 1154 m->m_pkthdr.flowid = 0; 1155 m->m_pkthdr.csum_flags = 0; 1156 m->m_pkthdr.fibnum = 0; 1157 m->m_pkthdr.rsstype = 0; 1158 1159 cc = m->m_pkthdr.len; 1160 mbcnt = MSIZE; 1161 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1162 mbcnt += MSIZE; 1163 if (mb->m_flags & M_EXT) 1164 mbcnt += mb->m_ext.ext_size; 1165 } 1166 } 1167 1168 unp = sotounpcb(so); 1169 MPASS(unp); 1170 1171 /* 1172 * XXXGL: would be cool to fully remove so_snd out of the equation 1173 * and avoid this lock, which is not only extraneous, but also being 1174 * released, thus still leaving possibility for a race. We can easily 1175 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1176 * is more difficult to invent something to handle so_error. 1177 */ 1178 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1179 if (error) 1180 goto out2; 1181 SOCK_SENDBUF_LOCK(so); 1182 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1183 SOCK_SENDBUF_UNLOCK(so); 1184 error = EPIPE; 1185 goto out3; 1186 } 1187 if (so->so_error != 0) { 1188 error = so->so_error; 1189 so->so_error = 0; 1190 SOCK_SENDBUF_UNLOCK(so); 1191 goto out3; 1192 } 1193 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1194 SOCK_SENDBUF_UNLOCK(so); 1195 error = EDESTADDRREQ; 1196 goto out3; 1197 } 1198 SOCK_SENDBUF_UNLOCK(so); 1199 1200 if (addr != NULL) { 1201 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1202 goto out3; 1203 UNP_PCB_LOCK_ASSERT(unp); 1204 unp2 = unp->unp_conn; 1205 UNP_PCB_LOCK_ASSERT(unp2); 1206 } else { 1207 UNP_PCB_LOCK(unp); 1208 unp2 = unp_pcb_lock_peer(unp); 1209 if (unp2 == NULL) { 1210 UNP_PCB_UNLOCK(unp); 1211 error = ENOTCONN; 1212 goto out3; 1213 } 1214 } 1215 1216 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1217 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1218 &mbcnt); 1219 if (unp->unp_addr != NULL) 1220 from = (struct sockaddr *)unp->unp_addr; 1221 else 1222 from = &sun_noname; 1223 f->m_len = from->sa_len; 1224 MPASS(from->sa_len <= MLEN); 1225 bcopy(from, mtod(f, void *), from->sa_len); 1226 ctl += f->m_len; 1227 1228 /* 1229 * Concatenate mbufs: from -> control -> data. 1230 * Save overall cc and mbcnt in "from" mbuf. 1231 */ 1232 if (c != NULL) { 1233 #ifdef INVARIANTS 1234 struct mbuf *mc; 1235 1236 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1237 MPASS(mc == clast); 1238 #endif 1239 f->m_next = c; 1240 clast->m_next = m; 1241 c = NULL; 1242 } else 1243 f->m_next = m; 1244 m = NULL; 1245 #ifdef INVARIANTS 1246 dcc = dctl = dmbcnt = 0; 1247 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1248 if (mb->m_type == MT_DATA) 1249 dcc += mb->m_len; 1250 else 1251 dctl += mb->m_len; 1252 dmbcnt += MSIZE; 1253 if (mb->m_flags & M_EXT) 1254 dmbcnt += mb->m_ext.ext_size; 1255 } 1256 MPASS(dcc == cc); 1257 MPASS(dctl == ctl); 1258 MPASS(dmbcnt == mbcnt); 1259 #endif 1260 f->m_pkthdr.len = cc + ctl; 1261 f->m_pkthdr.memlen = mbcnt; 1262 f->m_pkthdr.ctllen = ctl; 1263 1264 /* 1265 * Destination socket buffer selection. 1266 * 1267 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1268 * destination address is supplied, create a temporary connection for 1269 * the run time of the function (see call to unp_connectat() above and 1270 * to unp_disconnect() below). We distinguish them by condition of 1271 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1272 * that condition, since, again, through the run time of this code we 1273 * are always connected. For such "unconnected" sends, the destination 1274 * buffer would be the receive buffer of destination socket so2. 1275 * 1276 * For connected sends, data lands on the send buffer of the sender's 1277 * socket "so". Then, if we just added the very first datagram 1278 * on this send buffer, we need to add the send buffer on to the 1279 * receiving socket's buffer list. We put ourselves on top of the 1280 * list. Such logic gives infrequent senders priority over frequent 1281 * senders. 1282 * 1283 * Note on byte count management. As long as event methods kevent(2), 1284 * select(2) are not protocol specific (yet), we need to maintain 1285 * meaningful values on the receive buffer. So, the receive buffer 1286 * would accumulate counters from all connected buffers potentially 1287 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1288 */ 1289 so2 = unp2->unp_socket; 1290 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1291 SOCK_RECVBUF_LOCK(so2); 1292 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1293 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1294 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1295 uxdg_clist); 1296 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1297 sb->uxdg_cc += cc + ctl; 1298 sb->uxdg_ctl += ctl; 1299 sb->uxdg_mbcnt += mbcnt; 1300 so2->so_rcv.sb_acc += cc + ctl; 1301 so2->so_rcv.sb_ccc += cc + ctl; 1302 so2->so_rcv.sb_ctl += ctl; 1303 so2->so_rcv.sb_mbcnt += mbcnt; 1304 sorwakeup_locked(so2); 1305 f = NULL; 1306 } else { 1307 soroverflow_locked(so2); 1308 error = ENOBUFS; 1309 if (f->m_next->m_type == MT_CONTROL) { 1310 c = f->m_next; 1311 f->m_next = NULL; 1312 } 1313 } 1314 1315 if (addr != NULL) 1316 unp_disconnect(unp, unp2); 1317 else 1318 unp_pcb_unlock_pair(unp, unp2); 1319 1320 td->td_ru.ru_msgsnd++; 1321 1322 out3: 1323 SOCK_IO_SEND_UNLOCK(so); 1324 out2: 1325 if (c) 1326 unp_scan(c, unp_freerights); 1327 out: 1328 if (f) 1329 m_freem(f); 1330 if (c) 1331 m_freem(c); 1332 if (m) 1333 m_freem(m); 1334 1335 return (error); 1336 } 1337 1338 /* 1339 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1340 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1341 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1342 */ 1343 static int 1344 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1345 struct uio *uio, struct mbuf **controlp, int *flagsp) 1346 { 1347 ssize_t len = 0; 1348 int error; 1349 1350 so->so_rcv.uxdg_peeked = m; 1351 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1352 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1353 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1354 SOCK_RECVBUF_UNLOCK(so); 1355 1356 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1357 if (psa != NULL) 1358 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1359 1360 m = m->m_next; 1361 KASSERT(m, ("%s: no data or control after soname", __func__)); 1362 1363 /* 1364 * With MSG_PEEK the control isn't executed, just copied. 1365 */ 1366 while (m != NULL && m->m_type == MT_CONTROL) { 1367 if (controlp != NULL) { 1368 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1369 controlp = &(*controlp)->m_next; 1370 } 1371 m = m->m_next; 1372 } 1373 KASSERT(m == NULL || m->m_type == MT_DATA, 1374 ("%s: not MT_DATA mbuf %p", __func__, m)); 1375 while (m != NULL && uio->uio_resid > 0) { 1376 len = uio->uio_resid; 1377 if (len > m->m_len) 1378 len = m->m_len; 1379 error = uiomove(mtod(m, char *), (int)len, uio); 1380 if (error) { 1381 SOCK_IO_RECV_UNLOCK(so); 1382 return (error); 1383 } 1384 if (len == m->m_len) 1385 m = m->m_next; 1386 } 1387 SOCK_IO_RECV_UNLOCK(so); 1388 1389 if (flagsp != NULL) { 1390 if (m != NULL) { 1391 if (*flagsp & MSG_TRUNC) { 1392 /* Report real length of the packet */ 1393 uio->uio_resid -= m_length(m, NULL) - len; 1394 } 1395 *flagsp |= MSG_TRUNC; 1396 } else 1397 *flagsp &= ~MSG_TRUNC; 1398 } 1399 1400 return (0); 1401 } 1402 1403 /* 1404 * PF_UNIX/SOCK_DGRAM receive 1405 */ 1406 static int 1407 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1408 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1409 { 1410 struct sockbuf *sb = NULL; 1411 struct mbuf *m; 1412 int flags, error; 1413 ssize_t len = 0; 1414 bool nonblock; 1415 1416 MPASS(mp0 == NULL); 1417 1418 if (psa != NULL) 1419 *psa = NULL; 1420 if (controlp != NULL) 1421 *controlp = NULL; 1422 1423 flags = flagsp != NULL ? *flagsp : 0; 1424 nonblock = (so->so_state & SS_NBIO) || 1425 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1426 1427 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1428 if (__predict_false(error)) 1429 return (error); 1430 1431 /* 1432 * Loop blocking while waiting for a datagram. Prioritize connected 1433 * peers over unconnected sends. Set sb to selected socket buffer 1434 * containing an mbuf on exit from the wait loop. A datagram that 1435 * had already been peeked at has top priority. 1436 */ 1437 SOCK_RECVBUF_LOCK(so); 1438 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1439 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1440 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1441 if (so->so_error) { 1442 error = so->so_error; 1443 if (!(flags & MSG_PEEK)) 1444 so->so_error = 0; 1445 SOCK_RECVBUF_UNLOCK(so); 1446 SOCK_IO_RECV_UNLOCK(so); 1447 return (error); 1448 } 1449 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1450 uio->uio_resid == 0) { 1451 SOCK_RECVBUF_UNLOCK(so); 1452 SOCK_IO_RECV_UNLOCK(so); 1453 return (0); 1454 } 1455 if (nonblock) { 1456 SOCK_RECVBUF_UNLOCK(so); 1457 SOCK_IO_RECV_UNLOCK(so); 1458 return (EWOULDBLOCK); 1459 } 1460 error = sbwait(so, SO_RCV); 1461 if (error) { 1462 SOCK_RECVBUF_UNLOCK(so); 1463 SOCK_IO_RECV_UNLOCK(so); 1464 return (error); 1465 } 1466 } 1467 1468 if (sb == NULL) 1469 sb = &so->so_rcv; 1470 else if (m == NULL) 1471 m = STAILQ_FIRST(&sb->uxdg_mb); 1472 else 1473 MPASS(m == so->so_rcv.uxdg_peeked); 1474 1475 MPASS(sb->uxdg_cc > 0); 1476 M_ASSERTPKTHDR(m); 1477 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1478 1479 if (uio->uio_td) 1480 uio->uio_td->td_ru.ru_msgrcv++; 1481 1482 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1483 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1484 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1485 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1486 } else 1487 so->so_rcv.uxdg_peeked = NULL; 1488 1489 sb->uxdg_cc -= m->m_pkthdr.len; 1490 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1491 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1492 1493 if (__predict_false(flags & MSG_PEEK)) 1494 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1495 1496 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1497 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1498 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1499 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1500 SOCK_RECVBUF_UNLOCK(so); 1501 1502 if (psa != NULL) 1503 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1504 m = m_free(m); 1505 KASSERT(m, ("%s: no data or control after soname", __func__)); 1506 1507 /* 1508 * Packet to copyout() is now in 'm' and it is disconnected from the 1509 * queue. 1510 * 1511 * Process one or more MT_CONTROL mbufs present before any data mbufs 1512 * in the first mbuf chain on the socket buffer. We call into the 1513 * unp_externalize() to perform externalization (or freeing if 1514 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1515 * without MT_DATA mbufs. 1516 */ 1517 while (m != NULL && m->m_type == MT_CONTROL) { 1518 struct mbuf *cm; 1519 1520 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1521 * it frees whole chain, so we must disconnect the mbuf. 1522 */ 1523 cm = m; m = m->m_next; cm->m_next = NULL; 1524 error = unp_externalize(cm, controlp, flags); 1525 if (error != 0) { 1526 SOCK_IO_RECV_UNLOCK(so); 1527 unp_scan(m, unp_freerights); 1528 m_freem(m); 1529 return (error); 1530 } 1531 if (controlp != NULL) { 1532 while (*controlp != NULL) 1533 controlp = &(*controlp)->m_next; 1534 } 1535 } 1536 KASSERT(m == NULL || m->m_type == MT_DATA, 1537 ("%s: not MT_DATA mbuf %p", __func__, m)); 1538 while (m != NULL && uio->uio_resid > 0) { 1539 len = uio->uio_resid; 1540 if (len > m->m_len) 1541 len = m->m_len; 1542 error = uiomove(mtod(m, char *), (int)len, uio); 1543 if (error) { 1544 SOCK_IO_RECV_UNLOCK(so); 1545 m_freem(m); 1546 return (error); 1547 } 1548 if (len == m->m_len) 1549 m = m_free(m); 1550 else { 1551 m->m_data += len; 1552 m->m_len -= len; 1553 } 1554 } 1555 SOCK_IO_RECV_UNLOCK(so); 1556 1557 if (m != NULL) { 1558 if (flagsp != NULL) { 1559 if (flags & MSG_TRUNC) { 1560 /* Report real length of the packet */ 1561 uio->uio_resid -= m_length(m, NULL); 1562 } 1563 *flagsp |= MSG_TRUNC; 1564 } 1565 m_freem(m); 1566 } else if (flagsp != NULL) 1567 *flagsp &= ~MSG_TRUNC; 1568 1569 return (0); 1570 } 1571 1572 static bool 1573 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1574 { 1575 struct mbuf *mb, *n; 1576 struct sockbuf *sb; 1577 1578 SOCK_LOCK(so); 1579 if (SOLISTENING(so)) { 1580 SOCK_UNLOCK(so); 1581 return (false); 1582 } 1583 mb = NULL; 1584 sb = &so->so_rcv; 1585 SOCKBUF_LOCK(sb); 1586 if (sb->sb_fnrdy != NULL) { 1587 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1588 if (mb == m) { 1589 *errorp = sbready(sb, m, count); 1590 break; 1591 } 1592 mb = mb->m_next; 1593 if (mb == NULL) { 1594 mb = n; 1595 if (mb != NULL) 1596 n = mb->m_nextpkt; 1597 } 1598 } 1599 } 1600 SOCKBUF_UNLOCK(sb); 1601 SOCK_UNLOCK(so); 1602 return (mb != NULL); 1603 } 1604 1605 static int 1606 uipc_ready(struct socket *so, struct mbuf *m, int count) 1607 { 1608 struct unpcb *unp, *unp2; 1609 struct socket *so2; 1610 int error, i; 1611 1612 unp = sotounpcb(so); 1613 1614 KASSERT(so->so_type == SOCK_STREAM, 1615 ("%s: unexpected socket type for %p", __func__, so)); 1616 1617 UNP_PCB_LOCK(unp); 1618 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1619 UNP_PCB_UNLOCK(unp); 1620 so2 = unp2->unp_socket; 1621 SOCKBUF_LOCK(&so2->so_rcv); 1622 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1623 sorwakeup_locked(so2); 1624 else 1625 SOCKBUF_UNLOCK(&so2->so_rcv); 1626 UNP_PCB_UNLOCK(unp2); 1627 return (error); 1628 } 1629 UNP_PCB_UNLOCK(unp); 1630 1631 /* 1632 * The receiving socket has been disconnected, but may still be valid. 1633 * In this case, the now-ready mbufs are still present in its socket 1634 * buffer, so perform an exhaustive search before giving up and freeing 1635 * the mbufs. 1636 */ 1637 UNP_LINK_RLOCK(); 1638 LIST_FOREACH(unp, &unp_shead, unp_link) { 1639 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1640 break; 1641 } 1642 UNP_LINK_RUNLOCK(); 1643 1644 if (unp == NULL) { 1645 for (i = 0; i < count; i++) 1646 m = m_free(m); 1647 error = ECONNRESET; 1648 } 1649 return (error); 1650 } 1651 1652 static int 1653 uipc_sense(struct socket *so, struct stat *sb) 1654 { 1655 struct unpcb *unp; 1656 1657 unp = sotounpcb(so); 1658 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1659 1660 sb->st_blksize = so->so_snd.sb_hiwat; 1661 sb->st_dev = NODEV; 1662 sb->st_ino = unp->unp_ino; 1663 return (0); 1664 } 1665 1666 static int 1667 uipc_shutdown(struct socket *so, enum shutdown_how how) 1668 { 1669 struct unpcb *unp = sotounpcb(so); 1670 int error; 1671 1672 SOCK_LOCK(so); 1673 if ((so->so_state & 1674 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 1675 /* 1676 * POSIX mandates us to just return ENOTCONN when shutdown(2) is 1677 * invoked on a datagram sockets, however historically we would 1678 * actually tear socket down. This is known to be leveraged by 1679 * some applications to unblock process waiting in recv(2) by 1680 * other process that it shares that socket with. Try to meet 1681 * both backward-compatibility and POSIX requirements by forcing 1682 * ENOTCONN but still flushing buffers and performing wakeup(9). 1683 * 1684 * XXXGL: it remains unknown what applications expect this 1685 * behavior and is this isolated to unix/dgram or inet/dgram or 1686 * both. See: D10351, D3039. 1687 */ 1688 error = ENOTCONN; 1689 if (so->so_type != SOCK_DGRAM) { 1690 SOCK_UNLOCK(so); 1691 return (error); 1692 } 1693 } else 1694 error = 0; 1695 if (SOLISTENING(so)) { 1696 if (how != SHUT_WR) { 1697 so->so_error = ECONNABORTED; 1698 solisten_wakeup(so); /* unlocks so */ 1699 } else 1700 SOCK_UNLOCK(so); 1701 return (0); 1702 } 1703 SOCK_UNLOCK(so); 1704 1705 switch (how) { 1706 case SHUT_RD: 1707 socantrcvmore(so); 1708 unp_dispose(so); 1709 break; 1710 case SHUT_RDWR: 1711 socantrcvmore(so); 1712 unp_dispose(so); 1713 /* FALLTHROUGH */ 1714 case SHUT_WR: 1715 UNP_PCB_LOCK(unp); 1716 socantsendmore(so); 1717 unp_shutdown(unp); 1718 UNP_PCB_UNLOCK(unp); 1719 } 1720 wakeup(&so->so_timeo); 1721 1722 return (error); 1723 } 1724 1725 static int 1726 uipc_sockaddr(struct socket *so, struct sockaddr *ret) 1727 { 1728 struct unpcb *unp; 1729 const struct sockaddr *sa; 1730 1731 unp = sotounpcb(so); 1732 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1733 1734 UNP_PCB_LOCK(unp); 1735 if (unp->unp_addr != NULL) 1736 sa = (struct sockaddr *) unp->unp_addr; 1737 else 1738 sa = &sun_noname; 1739 bcopy(sa, ret, sa->sa_len); 1740 UNP_PCB_UNLOCK(unp); 1741 return (0); 1742 } 1743 1744 static int 1745 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1746 { 1747 struct unpcb *unp; 1748 struct xucred xu; 1749 int error, optval; 1750 1751 if (sopt->sopt_level != SOL_LOCAL) 1752 return (EINVAL); 1753 1754 unp = sotounpcb(so); 1755 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1756 error = 0; 1757 switch (sopt->sopt_dir) { 1758 case SOPT_GET: 1759 switch (sopt->sopt_name) { 1760 case LOCAL_PEERCRED: 1761 UNP_PCB_LOCK(unp); 1762 if (unp->unp_flags & UNP_HAVEPC) 1763 xu = unp->unp_peercred; 1764 else { 1765 if (so->so_type == SOCK_STREAM) 1766 error = ENOTCONN; 1767 else 1768 error = EINVAL; 1769 } 1770 UNP_PCB_UNLOCK(unp); 1771 if (error == 0) 1772 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1773 break; 1774 1775 case LOCAL_CREDS: 1776 /* Unlocked read. */ 1777 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1778 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1779 break; 1780 1781 case LOCAL_CREDS_PERSISTENT: 1782 /* Unlocked read. */ 1783 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1784 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1785 break; 1786 1787 case LOCAL_CONNWAIT: 1788 /* Unlocked read. */ 1789 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1790 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1791 break; 1792 1793 default: 1794 error = EOPNOTSUPP; 1795 break; 1796 } 1797 break; 1798 1799 case SOPT_SET: 1800 switch (sopt->sopt_name) { 1801 case LOCAL_CREDS: 1802 case LOCAL_CREDS_PERSISTENT: 1803 case LOCAL_CONNWAIT: 1804 error = sooptcopyin(sopt, &optval, sizeof(optval), 1805 sizeof(optval)); 1806 if (error) 1807 break; 1808 1809 #define OPTSET(bit, exclusive) do { \ 1810 UNP_PCB_LOCK(unp); \ 1811 if (optval) { \ 1812 if ((unp->unp_flags & (exclusive)) != 0) { \ 1813 UNP_PCB_UNLOCK(unp); \ 1814 error = EINVAL; \ 1815 break; \ 1816 } \ 1817 unp->unp_flags |= (bit); \ 1818 } else \ 1819 unp->unp_flags &= ~(bit); \ 1820 UNP_PCB_UNLOCK(unp); \ 1821 } while (0) 1822 1823 switch (sopt->sopt_name) { 1824 case LOCAL_CREDS: 1825 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1826 break; 1827 1828 case LOCAL_CREDS_PERSISTENT: 1829 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1830 break; 1831 1832 case LOCAL_CONNWAIT: 1833 OPTSET(UNP_CONNWAIT, 0); 1834 break; 1835 1836 default: 1837 break; 1838 } 1839 break; 1840 #undef OPTSET 1841 default: 1842 error = ENOPROTOOPT; 1843 break; 1844 } 1845 break; 1846 1847 default: 1848 error = EOPNOTSUPP; 1849 break; 1850 } 1851 return (error); 1852 } 1853 1854 static int 1855 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1856 { 1857 1858 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1859 } 1860 1861 static int 1862 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1863 struct thread *td, bool return_locked) 1864 { 1865 struct mtx *vplock; 1866 struct sockaddr_un *soun; 1867 struct vnode *vp; 1868 struct socket *so2; 1869 struct unpcb *unp, *unp2, *unp3; 1870 struct nameidata nd; 1871 char buf[SOCK_MAXADDRLEN]; 1872 struct sockaddr *sa; 1873 cap_rights_t rights; 1874 int error, len; 1875 bool connreq; 1876 1877 if (nam->sa_family != AF_UNIX) 1878 return (EAFNOSUPPORT); 1879 if (nam->sa_len > sizeof(struct sockaddr_un)) 1880 return (EINVAL); 1881 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1882 if (len <= 0) 1883 return (EINVAL); 1884 soun = (struct sockaddr_un *)nam; 1885 bcopy(soun->sun_path, buf, len); 1886 buf[len] = 0; 1887 1888 error = 0; 1889 unp = sotounpcb(so); 1890 UNP_PCB_LOCK(unp); 1891 for (;;) { 1892 /* 1893 * Wait for connection state to stabilize. If a connection 1894 * already exists, give up. For datagram sockets, which permit 1895 * multiple consecutive connect(2) calls, upper layers are 1896 * responsible for disconnecting in advance of a subsequent 1897 * connect(2), but this is not synchronized with PCB connection 1898 * state. 1899 * 1900 * Also make sure that no threads are currently attempting to 1901 * lock the peer socket, to ensure that unp_conn cannot 1902 * transition between two valid sockets while locks are dropped. 1903 */ 1904 if (SOLISTENING(so)) 1905 error = EOPNOTSUPP; 1906 else if (unp->unp_conn != NULL) 1907 error = EISCONN; 1908 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1909 error = EALREADY; 1910 } 1911 if (error != 0) { 1912 UNP_PCB_UNLOCK(unp); 1913 return (error); 1914 } 1915 if (unp->unp_pairbusy > 0) { 1916 unp->unp_flags |= UNP_WAITING; 1917 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1918 continue; 1919 } 1920 break; 1921 } 1922 unp->unp_flags |= UNP_CONNECTING; 1923 UNP_PCB_UNLOCK(unp); 1924 1925 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1926 if (connreq) 1927 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1928 else 1929 sa = NULL; 1930 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1931 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1932 error = namei(&nd); 1933 if (error) 1934 vp = NULL; 1935 else 1936 vp = nd.ni_vp; 1937 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1938 if (error) 1939 goto bad; 1940 NDFREE_PNBUF(&nd); 1941 1942 if (vp->v_type != VSOCK) { 1943 error = ENOTSOCK; 1944 goto bad; 1945 } 1946 #ifdef MAC 1947 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1948 if (error) 1949 goto bad; 1950 #endif 1951 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1952 if (error) 1953 goto bad; 1954 1955 unp = sotounpcb(so); 1956 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1957 1958 vplock = mtx_pool_find(mtxpool_sleep, vp); 1959 mtx_lock(vplock); 1960 VOP_UNP_CONNECT(vp, &unp2); 1961 if (unp2 == NULL) { 1962 error = ECONNREFUSED; 1963 goto bad2; 1964 } 1965 so2 = unp2->unp_socket; 1966 if (so->so_type != so2->so_type) { 1967 error = EPROTOTYPE; 1968 goto bad2; 1969 } 1970 if (connreq) { 1971 if (SOLISTENING(so2)) { 1972 CURVNET_SET(so2->so_vnet); 1973 so2 = sonewconn(so2, 0); 1974 CURVNET_RESTORE(); 1975 } else 1976 so2 = NULL; 1977 if (so2 == NULL) { 1978 error = ECONNREFUSED; 1979 goto bad2; 1980 } 1981 unp3 = sotounpcb(so2); 1982 unp_pcb_lock_pair(unp2, unp3); 1983 if (unp2->unp_addr != NULL) { 1984 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1985 unp3->unp_addr = (struct sockaddr_un *) sa; 1986 sa = NULL; 1987 } 1988 1989 unp_copy_peercred(td, unp3, unp, unp2); 1990 1991 UNP_PCB_UNLOCK(unp2); 1992 unp2 = unp3; 1993 1994 /* 1995 * It is safe to block on the PCB lock here since unp2 is 1996 * nascent and cannot be connected to any other sockets. 1997 */ 1998 UNP_PCB_LOCK(unp); 1999 #ifdef MAC 2000 mac_socketpeer_set_from_socket(so, so2); 2001 mac_socketpeer_set_from_socket(so2, so); 2002 #endif 2003 } else { 2004 unp_pcb_lock_pair(unp, unp2); 2005 } 2006 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 2007 sotounpcb(so2) == unp2, 2008 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 2009 unp_connect2(so, so2, PRU_CONNECT); 2010 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2011 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2012 unp->unp_flags &= ~UNP_CONNECTING; 2013 if (!return_locked) 2014 unp_pcb_unlock_pair(unp, unp2); 2015 bad2: 2016 mtx_unlock(vplock); 2017 bad: 2018 if (vp != NULL) { 2019 /* 2020 * If we are returning locked (called via uipc_sosend_dgram()), 2021 * we need to be sure that vput() won't sleep. This is 2022 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2023 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2024 */ 2025 MPASS(!(return_locked && connreq)); 2026 vput(vp); 2027 } 2028 free(sa, M_SONAME); 2029 if (__predict_false(error)) { 2030 UNP_PCB_LOCK(unp); 2031 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2032 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2033 unp->unp_flags &= ~UNP_CONNECTING; 2034 UNP_PCB_UNLOCK(unp); 2035 } 2036 return (error); 2037 } 2038 2039 /* 2040 * Set socket peer credentials at connection time. 2041 * 2042 * The client's PCB credentials are copied from its process structure. The 2043 * server's PCB credentials are copied from the socket on which it called 2044 * listen(2). uipc_listen cached that process's credentials at the time. 2045 */ 2046 void 2047 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2048 struct unpcb *server_unp, struct unpcb *listen_unp) 2049 { 2050 cru2xt(td, &client_unp->unp_peercred); 2051 client_unp->unp_flags |= UNP_HAVEPC; 2052 2053 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2054 sizeof(server_unp->unp_peercred)); 2055 server_unp->unp_flags |= UNP_HAVEPC; 2056 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2057 } 2058 2059 static void 2060 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2061 { 2062 struct unpcb *unp; 2063 struct unpcb *unp2; 2064 2065 MPASS(so2->so_type == so->so_type); 2066 unp = sotounpcb(so); 2067 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2068 unp2 = sotounpcb(so2); 2069 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2070 2071 UNP_PCB_LOCK_ASSERT(unp); 2072 UNP_PCB_LOCK_ASSERT(unp2); 2073 KASSERT(unp->unp_conn == NULL, 2074 ("%s: socket %p is already connected", __func__, unp)); 2075 2076 unp->unp_conn = unp2; 2077 unp_pcb_hold(unp2); 2078 unp_pcb_hold(unp); 2079 switch (so->so_type) { 2080 case SOCK_DGRAM: 2081 UNP_REF_LIST_LOCK(); 2082 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2083 UNP_REF_LIST_UNLOCK(); 2084 soisconnected(so); 2085 break; 2086 2087 case SOCK_STREAM: 2088 case SOCK_SEQPACKET: 2089 KASSERT(unp2->unp_conn == NULL, 2090 ("%s: socket %p is already connected", __func__, unp2)); 2091 unp2->unp_conn = unp; 2092 if (req == PRU_CONNECT && 2093 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2094 soisconnecting(so); 2095 else 2096 soisconnected(so); 2097 soisconnected(so2); 2098 break; 2099 2100 default: 2101 panic("unp_connect2"); 2102 } 2103 } 2104 2105 static void 2106 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2107 { 2108 struct socket *so, *so2; 2109 struct mbuf *m = NULL; 2110 #ifdef INVARIANTS 2111 struct unpcb *unptmp; 2112 #endif 2113 2114 UNP_PCB_LOCK_ASSERT(unp); 2115 UNP_PCB_LOCK_ASSERT(unp2); 2116 KASSERT(unp->unp_conn == unp2, 2117 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2118 2119 unp->unp_conn = NULL; 2120 so = unp->unp_socket; 2121 so2 = unp2->unp_socket; 2122 switch (unp->unp_socket->so_type) { 2123 case SOCK_DGRAM: 2124 /* 2125 * Remove our send socket buffer from the peer's receive buffer. 2126 * Move the data to the receive buffer only if it is empty. 2127 * This is a protection against a scenario where a peer 2128 * connects, floods and disconnects, effectively blocking 2129 * sendto() from unconnected sockets. 2130 */ 2131 SOCK_RECVBUF_LOCK(so2); 2132 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2133 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2134 uxdg_clist); 2135 if (__predict_true((so2->so_rcv.sb_state & 2136 SBS_CANTRCVMORE) == 0) && 2137 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2138 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2139 &so->so_snd.uxdg_mb); 2140 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2141 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2142 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2143 } else { 2144 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2145 STAILQ_INIT(&so->so_snd.uxdg_mb); 2146 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2147 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2148 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2149 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2150 } 2151 /* Note: so may reconnect. */ 2152 so->so_snd.uxdg_cc = 0; 2153 so->so_snd.uxdg_ctl = 0; 2154 so->so_snd.uxdg_mbcnt = 0; 2155 } 2156 SOCK_RECVBUF_UNLOCK(so2); 2157 UNP_REF_LIST_LOCK(); 2158 #ifdef INVARIANTS 2159 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2160 if (unptmp == unp) 2161 break; 2162 } 2163 KASSERT(unptmp != NULL, 2164 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2165 #endif 2166 LIST_REMOVE(unp, unp_reflink); 2167 UNP_REF_LIST_UNLOCK(); 2168 if (so) { 2169 SOCK_LOCK(so); 2170 so->so_state &= ~SS_ISCONNECTED; 2171 SOCK_UNLOCK(so); 2172 } 2173 break; 2174 2175 case SOCK_STREAM: 2176 case SOCK_SEQPACKET: 2177 if (so) 2178 soisdisconnected(so); 2179 MPASS(unp2->unp_conn == unp); 2180 unp2->unp_conn = NULL; 2181 if (so2) 2182 soisdisconnected(so2); 2183 break; 2184 } 2185 2186 if (unp == unp2) { 2187 unp_pcb_rele_notlast(unp); 2188 if (!unp_pcb_rele(unp)) 2189 UNP_PCB_UNLOCK(unp); 2190 } else { 2191 if (!unp_pcb_rele(unp)) 2192 UNP_PCB_UNLOCK(unp); 2193 if (!unp_pcb_rele(unp2)) 2194 UNP_PCB_UNLOCK(unp2); 2195 } 2196 2197 if (m != NULL) { 2198 unp_scan(m, unp_freerights); 2199 m_freem(m); 2200 } 2201 } 2202 2203 /* 2204 * unp_pcblist() walks the global list of struct unpcb's to generate a 2205 * pointer list, bumping the refcount on each unpcb. It then copies them out 2206 * sequentially, validating the generation number on each to see if it has 2207 * been detached. All of this is necessary because copyout() may sleep on 2208 * disk I/O. 2209 */ 2210 static int 2211 unp_pcblist(SYSCTL_HANDLER_ARGS) 2212 { 2213 struct unpcb *unp, **unp_list; 2214 unp_gen_t gencnt; 2215 struct xunpgen *xug; 2216 struct unp_head *head; 2217 struct xunpcb *xu; 2218 u_int i; 2219 int error, n; 2220 2221 switch ((intptr_t)arg1) { 2222 case SOCK_STREAM: 2223 head = &unp_shead; 2224 break; 2225 2226 case SOCK_DGRAM: 2227 head = &unp_dhead; 2228 break; 2229 2230 case SOCK_SEQPACKET: 2231 head = &unp_sphead; 2232 break; 2233 2234 default: 2235 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2236 } 2237 2238 /* 2239 * The process of preparing the PCB list is too time-consuming and 2240 * resource-intensive to repeat twice on every request. 2241 */ 2242 if (req->oldptr == NULL) { 2243 n = unp_count; 2244 req->oldidx = 2 * (sizeof *xug) 2245 + (n + n/8) * sizeof(struct xunpcb); 2246 return (0); 2247 } 2248 2249 if (req->newptr != NULL) 2250 return (EPERM); 2251 2252 /* 2253 * OK, now we're committed to doing something. 2254 */ 2255 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2256 UNP_LINK_RLOCK(); 2257 gencnt = unp_gencnt; 2258 n = unp_count; 2259 UNP_LINK_RUNLOCK(); 2260 2261 xug->xug_len = sizeof *xug; 2262 xug->xug_count = n; 2263 xug->xug_gen = gencnt; 2264 xug->xug_sogen = so_gencnt; 2265 error = SYSCTL_OUT(req, xug, sizeof *xug); 2266 if (error) { 2267 free(xug, M_TEMP); 2268 return (error); 2269 } 2270 2271 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2272 2273 UNP_LINK_RLOCK(); 2274 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2275 unp = LIST_NEXT(unp, unp_link)) { 2276 UNP_PCB_LOCK(unp); 2277 if (unp->unp_gencnt <= gencnt) { 2278 if (cr_cansee(req->td->td_ucred, 2279 unp->unp_socket->so_cred)) { 2280 UNP_PCB_UNLOCK(unp); 2281 continue; 2282 } 2283 unp_list[i++] = unp; 2284 unp_pcb_hold(unp); 2285 } 2286 UNP_PCB_UNLOCK(unp); 2287 } 2288 UNP_LINK_RUNLOCK(); 2289 n = i; /* In case we lost some during malloc. */ 2290 2291 error = 0; 2292 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2293 for (i = 0; i < n; i++) { 2294 unp = unp_list[i]; 2295 UNP_PCB_LOCK(unp); 2296 if (unp_pcb_rele(unp)) 2297 continue; 2298 2299 if (unp->unp_gencnt <= gencnt) { 2300 xu->xu_len = sizeof *xu; 2301 xu->xu_unpp = (uintptr_t)unp; 2302 /* 2303 * XXX - need more locking here to protect against 2304 * connect/disconnect races for SMP. 2305 */ 2306 if (unp->unp_addr != NULL) 2307 bcopy(unp->unp_addr, &xu->xu_addr, 2308 unp->unp_addr->sun_len); 2309 else 2310 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2311 if (unp->unp_conn != NULL && 2312 unp->unp_conn->unp_addr != NULL) 2313 bcopy(unp->unp_conn->unp_addr, 2314 &xu->xu_caddr, 2315 unp->unp_conn->unp_addr->sun_len); 2316 else 2317 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2318 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2319 xu->unp_conn = (uintptr_t)unp->unp_conn; 2320 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2321 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2322 xu->unp_gencnt = unp->unp_gencnt; 2323 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2324 UNP_PCB_UNLOCK(unp); 2325 error = SYSCTL_OUT(req, xu, sizeof *xu); 2326 } else { 2327 UNP_PCB_UNLOCK(unp); 2328 } 2329 } 2330 free(xu, M_TEMP); 2331 if (!error) { 2332 /* 2333 * Give the user an updated idea of our state. If the 2334 * generation differs from what we told her before, she knows 2335 * that something happened while we were processing this 2336 * request, and it might be necessary to retry. 2337 */ 2338 xug->xug_gen = unp_gencnt; 2339 xug->xug_sogen = so_gencnt; 2340 xug->xug_count = unp_count; 2341 error = SYSCTL_OUT(req, xug, sizeof *xug); 2342 } 2343 free(unp_list, M_TEMP); 2344 free(xug, M_TEMP); 2345 return (error); 2346 } 2347 2348 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2349 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2350 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2351 "List of active local datagram sockets"); 2352 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2353 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2354 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2355 "List of active local stream sockets"); 2356 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2357 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2358 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2359 "List of active local seqpacket sockets"); 2360 2361 static void 2362 unp_shutdown(struct unpcb *unp) 2363 { 2364 struct unpcb *unp2; 2365 struct socket *so; 2366 2367 UNP_PCB_LOCK_ASSERT(unp); 2368 2369 unp2 = unp->unp_conn; 2370 if ((unp->unp_socket->so_type == SOCK_STREAM || 2371 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2372 so = unp2->unp_socket; 2373 if (so != NULL) 2374 socantrcvmore(so); 2375 } 2376 } 2377 2378 static void 2379 unp_drop(struct unpcb *unp) 2380 { 2381 struct socket *so; 2382 struct unpcb *unp2; 2383 2384 /* 2385 * Regardless of whether the socket's peer dropped the connection 2386 * with this socket by aborting or disconnecting, POSIX requires 2387 * that ECONNRESET is returned. 2388 */ 2389 2390 UNP_PCB_LOCK(unp); 2391 so = unp->unp_socket; 2392 if (so) 2393 so->so_error = ECONNRESET; 2394 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2395 /* Last reference dropped in unp_disconnect(). */ 2396 unp_pcb_rele_notlast(unp); 2397 unp_disconnect(unp, unp2); 2398 } else if (!unp_pcb_rele(unp)) { 2399 UNP_PCB_UNLOCK(unp); 2400 } 2401 } 2402 2403 static void 2404 unp_freerights(struct filedescent **fdep, int fdcount) 2405 { 2406 struct file *fp; 2407 int i; 2408 2409 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2410 2411 for (i = 0; i < fdcount; i++) { 2412 fp = fdep[i]->fde_file; 2413 filecaps_free(&fdep[i]->fde_caps); 2414 unp_discard(fp); 2415 } 2416 free(fdep[0], M_FILECAPS); 2417 } 2418 2419 static int 2420 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2421 { 2422 struct thread *td = curthread; /* XXX */ 2423 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2424 int i; 2425 int *fdp; 2426 struct filedesc *fdesc = td->td_proc->p_fd; 2427 struct filedescent **fdep; 2428 void *data; 2429 socklen_t clen = control->m_len, datalen; 2430 int error, newfds; 2431 u_int newlen; 2432 2433 UNP_LINK_UNLOCK_ASSERT(); 2434 2435 error = 0; 2436 if (controlp != NULL) /* controlp == NULL => free control messages */ 2437 *controlp = NULL; 2438 while (cm != NULL) { 2439 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2440 2441 data = CMSG_DATA(cm); 2442 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2443 if (cm->cmsg_level == SOL_SOCKET 2444 && cm->cmsg_type == SCM_RIGHTS) { 2445 newfds = datalen / sizeof(*fdep); 2446 if (newfds == 0) 2447 goto next; 2448 fdep = data; 2449 2450 /* If we're not outputting the descriptors free them. */ 2451 if (error || controlp == NULL) { 2452 unp_freerights(fdep, newfds); 2453 goto next; 2454 } 2455 FILEDESC_XLOCK(fdesc); 2456 2457 /* 2458 * Now change each pointer to an fd in the global 2459 * table to an integer that is the index to the local 2460 * fd table entry that we set up to point to the 2461 * global one we are transferring. 2462 */ 2463 newlen = newfds * sizeof(int); 2464 *controlp = sbcreatecontrol(NULL, newlen, 2465 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2466 2467 fdp = (int *) 2468 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2469 if ((error = fdallocn(td, 0, fdp, newfds))) { 2470 FILEDESC_XUNLOCK(fdesc); 2471 unp_freerights(fdep, newfds); 2472 m_freem(*controlp); 2473 *controlp = NULL; 2474 goto next; 2475 } 2476 for (i = 0; i < newfds; i++, fdp++) { 2477 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2478 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2479 &fdep[i]->fde_caps); 2480 unp_externalize_fp(fdep[i]->fde_file); 2481 } 2482 2483 /* 2484 * The new type indicates that the mbuf data refers to 2485 * kernel resources that may need to be released before 2486 * the mbuf is freed. 2487 */ 2488 m_chtype(*controlp, MT_EXTCONTROL); 2489 FILEDESC_XUNLOCK(fdesc); 2490 free(fdep[0], M_FILECAPS); 2491 } else { 2492 /* We can just copy anything else across. */ 2493 if (error || controlp == NULL) 2494 goto next; 2495 *controlp = sbcreatecontrol(NULL, datalen, 2496 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2497 bcopy(data, 2498 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2499 datalen); 2500 } 2501 controlp = &(*controlp)->m_next; 2502 2503 next: 2504 if (CMSG_SPACE(datalen) < clen) { 2505 clen -= CMSG_SPACE(datalen); 2506 cm = (struct cmsghdr *) 2507 ((caddr_t)cm + CMSG_SPACE(datalen)); 2508 } else { 2509 clen = 0; 2510 cm = NULL; 2511 } 2512 } 2513 2514 m_freem(control); 2515 return (error); 2516 } 2517 2518 static void 2519 unp_zone_change(void *tag) 2520 { 2521 2522 uma_zone_set_max(unp_zone, maxsockets); 2523 } 2524 2525 #ifdef INVARIANTS 2526 static void 2527 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2528 { 2529 struct unpcb *unp; 2530 2531 unp = mem; 2532 2533 KASSERT(LIST_EMPTY(&unp->unp_refs), 2534 ("%s: unpcb %p has lingering refs", __func__, unp)); 2535 KASSERT(unp->unp_socket == NULL, 2536 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2537 KASSERT(unp->unp_vnode == NULL, 2538 ("%s: unpcb %p has vnode references", __func__, unp)); 2539 KASSERT(unp->unp_conn == NULL, 2540 ("%s: unpcb %p is still connected", __func__, unp)); 2541 KASSERT(unp->unp_addr == NULL, 2542 ("%s: unpcb %p has leaked addr", __func__, unp)); 2543 } 2544 #endif 2545 2546 static void 2547 unp_init(void *arg __unused) 2548 { 2549 uma_dtor dtor; 2550 2551 #ifdef INVARIANTS 2552 dtor = unp_zdtor; 2553 #else 2554 dtor = NULL; 2555 #endif 2556 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2557 NULL, NULL, UMA_ALIGN_CACHE, 0); 2558 uma_zone_set_max(unp_zone, maxsockets); 2559 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2560 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2561 NULL, EVENTHANDLER_PRI_ANY); 2562 LIST_INIT(&unp_dhead); 2563 LIST_INIT(&unp_shead); 2564 LIST_INIT(&unp_sphead); 2565 SLIST_INIT(&unp_defers); 2566 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2567 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2568 UNP_LINK_LOCK_INIT(); 2569 UNP_DEFERRED_LOCK_INIT(); 2570 } 2571 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2572 2573 static void 2574 unp_internalize_cleanup_rights(struct mbuf *control) 2575 { 2576 struct cmsghdr *cp; 2577 struct mbuf *m; 2578 void *data; 2579 socklen_t datalen; 2580 2581 for (m = control; m != NULL; m = m->m_next) { 2582 cp = mtod(m, struct cmsghdr *); 2583 if (cp->cmsg_level != SOL_SOCKET || 2584 cp->cmsg_type != SCM_RIGHTS) 2585 continue; 2586 data = CMSG_DATA(cp); 2587 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2588 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2589 } 2590 } 2591 2592 static int 2593 unp_internalize(struct mbuf **controlp, struct thread *td, 2594 struct mbuf **clast, u_int *space, u_int *mbcnt) 2595 { 2596 struct mbuf *control, **initial_controlp; 2597 struct proc *p; 2598 struct filedesc *fdesc; 2599 struct bintime *bt; 2600 struct cmsghdr *cm; 2601 struct cmsgcred *cmcred; 2602 struct filedescent *fde, **fdep, *fdev; 2603 struct file *fp; 2604 struct timeval *tv; 2605 struct timespec *ts; 2606 void *data; 2607 socklen_t clen, datalen; 2608 int i, j, error, *fdp, oldfds; 2609 u_int newlen; 2610 2611 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2612 UNP_LINK_UNLOCK_ASSERT(); 2613 2614 p = td->td_proc; 2615 fdesc = p->p_fd; 2616 error = 0; 2617 control = *controlp; 2618 *controlp = NULL; 2619 initial_controlp = controlp; 2620 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2621 data = CMSG_DATA(cm); 2622 2623 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2624 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2625 (char *)cm + cm->cmsg_len >= (char *)data; 2626 2627 clen -= min(CMSG_SPACE(datalen), clen), 2628 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2629 data = CMSG_DATA(cm)) { 2630 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2631 switch (cm->cmsg_type) { 2632 case SCM_CREDS: 2633 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2634 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2635 cmcred = (struct cmsgcred *) 2636 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2637 cmcred->cmcred_pid = p->p_pid; 2638 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2639 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2640 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2641 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2642 CMGROUP_MAX); 2643 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2644 cmcred->cmcred_groups[i] = 2645 td->td_ucred->cr_groups[i]; 2646 break; 2647 2648 case SCM_RIGHTS: 2649 oldfds = datalen / sizeof (int); 2650 if (oldfds == 0) 2651 continue; 2652 /* On some machines sizeof pointer is bigger than 2653 * sizeof int, so we need to check if data fits into 2654 * single mbuf. We could allocate several mbufs, and 2655 * unp_externalize() should even properly handle that. 2656 * But it is not worth to complicate the code for an 2657 * insane scenario of passing over 200 file descriptors 2658 * at once. 2659 */ 2660 newlen = oldfds * sizeof(fdep[0]); 2661 if (CMSG_SPACE(newlen) > MCLBYTES) { 2662 error = EMSGSIZE; 2663 goto out; 2664 } 2665 /* 2666 * Check that all the FDs passed in refer to legal 2667 * files. If not, reject the entire operation. 2668 */ 2669 fdp = data; 2670 FILEDESC_SLOCK(fdesc); 2671 for (i = 0; i < oldfds; i++, fdp++) { 2672 fp = fget_noref(fdesc, *fdp); 2673 if (fp == NULL) { 2674 FILEDESC_SUNLOCK(fdesc); 2675 error = EBADF; 2676 goto out; 2677 } 2678 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2679 FILEDESC_SUNLOCK(fdesc); 2680 error = EOPNOTSUPP; 2681 goto out; 2682 } 2683 } 2684 2685 /* 2686 * Now replace the integer FDs with pointers to the 2687 * file structure and capability rights. 2688 */ 2689 *controlp = sbcreatecontrol(NULL, newlen, 2690 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2691 fdp = data; 2692 for (i = 0; i < oldfds; i++, fdp++) { 2693 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2694 fdp = data; 2695 for (j = 0; j < i; j++, fdp++) { 2696 fdrop(fdesc->fd_ofiles[*fdp]. 2697 fde_file, td); 2698 } 2699 FILEDESC_SUNLOCK(fdesc); 2700 error = EBADF; 2701 goto out; 2702 } 2703 } 2704 fdp = data; 2705 fdep = (struct filedescent **) 2706 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2707 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2708 M_WAITOK); 2709 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2710 fde = &fdesc->fd_ofiles[*fdp]; 2711 fdep[i] = fdev; 2712 fdep[i]->fde_file = fde->fde_file; 2713 filecaps_copy(&fde->fde_caps, 2714 &fdep[i]->fde_caps, true); 2715 unp_internalize_fp(fdep[i]->fde_file); 2716 } 2717 FILEDESC_SUNLOCK(fdesc); 2718 break; 2719 2720 case SCM_TIMESTAMP: 2721 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2722 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2723 tv = (struct timeval *) 2724 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2725 microtime(tv); 2726 break; 2727 2728 case SCM_BINTIME: 2729 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2730 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2731 bt = (struct bintime *) 2732 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2733 bintime(bt); 2734 break; 2735 2736 case SCM_REALTIME: 2737 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2738 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2739 ts = (struct timespec *) 2740 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2741 nanotime(ts); 2742 break; 2743 2744 case SCM_MONOTONIC: 2745 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2746 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2747 ts = (struct timespec *) 2748 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2749 nanouptime(ts); 2750 break; 2751 2752 default: 2753 error = EINVAL; 2754 goto out; 2755 } 2756 2757 if (space != NULL) { 2758 *space += (*controlp)->m_len; 2759 *mbcnt += MSIZE; 2760 if ((*controlp)->m_flags & M_EXT) 2761 *mbcnt += (*controlp)->m_ext.ext_size; 2762 *clast = *controlp; 2763 } 2764 controlp = &(*controlp)->m_next; 2765 } 2766 if (clen > 0) 2767 error = EINVAL; 2768 2769 out: 2770 if (error != 0 && initial_controlp != NULL) 2771 unp_internalize_cleanup_rights(*initial_controlp); 2772 m_freem(control); 2773 return (error); 2774 } 2775 2776 static struct mbuf * 2777 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2778 struct mbuf **clast, u_int *space, u_int *mbcnt) 2779 { 2780 struct mbuf *m, *n, *n_prev; 2781 const struct cmsghdr *cm; 2782 int ngroups, i, cmsgtype; 2783 size_t ctrlsz; 2784 2785 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2786 if (mode & UNP_WANTCRED_ALWAYS) { 2787 ctrlsz = SOCKCRED2SIZE(ngroups); 2788 cmsgtype = SCM_CREDS2; 2789 } else { 2790 ctrlsz = SOCKCREDSIZE(ngroups); 2791 cmsgtype = SCM_CREDS; 2792 } 2793 2794 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2795 if (m == NULL) 2796 return (control); 2797 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2798 2799 if (mode & UNP_WANTCRED_ALWAYS) { 2800 struct sockcred2 *sc; 2801 2802 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2803 sc->sc_version = 0; 2804 sc->sc_pid = td->td_proc->p_pid; 2805 sc->sc_uid = td->td_ucred->cr_ruid; 2806 sc->sc_euid = td->td_ucred->cr_uid; 2807 sc->sc_gid = td->td_ucred->cr_rgid; 2808 sc->sc_egid = td->td_ucred->cr_gid; 2809 sc->sc_ngroups = ngroups; 2810 for (i = 0; i < sc->sc_ngroups; i++) 2811 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2812 } else { 2813 struct sockcred *sc; 2814 2815 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2816 sc->sc_uid = td->td_ucred->cr_ruid; 2817 sc->sc_euid = td->td_ucred->cr_uid; 2818 sc->sc_gid = td->td_ucred->cr_rgid; 2819 sc->sc_egid = td->td_ucred->cr_gid; 2820 sc->sc_ngroups = ngroups; 2821 for (i = 0; i < sc->sc_ngroups; i++) 2822 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2823 } 2824 2825 /* 2826 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2827 * created SCM_CREDS control message (struct sockcred) has another 2828 * format. 2829 */ 2830 if (control != NULL && cmsgtype == SCM_CREDS) 2831 for (n = control, n_prev = NULL; n != NULL;) { 2832 cm = mtod(n, struct cmsghdr *); 2833 if (cm->cmsg_level == SOL_SOCKET && 2834 cm->cmsg_type == SCM_CREDS) { 2835 if (n_prev == NULL) 2836 control = n->m_next; 2837 else 2838 n_prev->m_next = n->m_next; 2839 if (space != NULL) { 2840 MPASS(*space >= n->m_len); 2841 *space -= n->m_len; 2842 MPASS(*mbcnt >= MSIZE); 2843 *mbcnt -= MSIZE; 2844 if (n->m_flags & M_EXT) { 2845 MPASS(*mbcnt >= 2846 n->m_ext.ext_size); 2847 *mbcnt -= n->m_ext.ext_size; 2848 } 2849 MPASS(clast); 2850 if (*clast == n) { 2851 MPASS(n->m_next == NULL); 2852 if (n_prev == NULL) 2853 *clast = m; 2854 else 2855 *clast = n_prev; 2856 } 2857 } 2858 n = m_free(n); 2859 } else { 2860 n_prev = n; 2861 n = n->m_next; 2862 } 2863 } 2864 2865 /* Prepend it to the head. */ 2866 m->m_next = control; 2867 if (space != NULL) { 2868 *space += m->m_len; 2869 *mbcnt += MSIZE; 2870 if (control == NULL) 2871 *clast = m; 2872 } 2873 return (m); 2874 } 2875 2876 static struct unpcb * 2877 fptounp(struct file *fp) 2878 { 2879 struct socket *so; 2880 2881 if (fp->f_type != DTYPE_SOCKET) 2882 return (NULL); 2883 if ((so = fp->f_data) == NULL) 2884 return (NULL); 2885 if (so->so_proto->pr_domain != &localdomain) 2886 return (NULL); 2887 return sotounpcb(so); 2888 } 2889 2890 static void 2891 unp_discard(struct file *fp) 2892 { 2893 struct unp_defer *dr; 2894 2895 if (unp_externalize_fp(fp)) { 2896 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2897 dr->ud_fp = fp; 2898 UNP_DEFERRED_LOCK(); 2899 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2900 UNP_DEFERRED_UNLOCK(); 2901 atomic_add_int(&unp_defers_count, 1); 2902 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2903 } else 2904 closef_nothread(fp); 2905 } 2906 2907 static void 2908 unp_process_defers(void *arg __unused, int pending) 2909 { 2910 struct unp_defer *dr; 2911 SLIST_HEAD(, unp_defer) drl; 2912 int count; 2913 2914 SLIST_INIT(&drl); 2915 for (;;) { 2916 UNP_DEFERRED_LOCK(); 2917 if (SLIST_FIRST(&unp_defers) == NULL) { 2918 UNP_DEFERRED_UNLOCK(); 2919 break; 2920 } 2921 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2922 UNP_DEFERRED_UNLOCK(); 2923 count = 0; 2924 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2925 SLIST_REMOVE_HEAD(&drl, ud_link); 2926 closef_nothread(dr->ud_fp); 2927 free(dr, M_TEMP); 2928 count++; 2929 } 2930 atomic_add_int(&unp_defers_count, -count); 2931 } 2932 } 2933 2934 static void 2935 unp_internalize_fp(struct file *fp) 2936 { 2937 struct unpcb *unp; 2938 2939 UNP_LINK_WLOCK(); 2940 if ((unp = fptounp(fp)) != NULL) { 2941 unp->unp_file = fp; 2942 unp->unp_msgcount++; 2943 } 2944 unp_rights++; 2945 UNP_LINK_WUNLOCK(); 2946 } 2947 2948 static int 2949 unp_externalize_fp(struct file *fp) 2950 { 2951 struct unpcb *unp; 2952 int ret; 2953 2954 UNP_LINK_WLOCK(); 2955 if ((unp = fptounp(fp)) != NULL) { 2956 unp->unp_msgcount--; 2957 ret = 1; 2958 } else 2959 ret = 0; 2960 unp_rights--; 2961 UNP_LINK_WUNLOCK(); 2962 return (ret); 2963 } 2964 2965 /* 2966 * unp_defer indicates whether additional work has been defered for a future 2967 * pass through unp_gc(). It is thread local and does not require explicit 2968 * synchronization. 2969 */ 2970 static int unp_marked; 2971 2972 static void 2973 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2974 { 2975 struct unpcb *unp; 2976 struct file *fp; 2977 int i; 2978 2979 /* 2980 * This function can only be called from the gc task. 2981 */ 2982 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2983 ("%s: not on gc callout", __func__)); 2984 UNP_LINK_LOCK_ASSERT(); 2985 2986 for (i = 0; i < fdcount; i++) { 2987 fp = fdep[i]->fde_file; 2988 if ((unp = fptounp(fp)) == NULL) 2989 continue; 2990 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2991 continue; 2992 unp->unp_gcrefs--; 2993 } 2994 } 2995 2996 static void 2997 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2998 { 2999 struct unpcb *unp; 3000 struct file *fp; 3001 int i; 3002 3003 /* 3004 * This function can only be called from the gc task. 3005 */ 3006 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 3007 ("%s: not on gc callout", __func__)); 3008 UNP_LINK_LOCK_ASSERT(); 3009 3010 for (i = 0; i < fdcount; i++) { 3011 fp = fdep[i]->fde_file; 3012 if ((unp = fptounp(fp)) == NULL) 3013 continue; 3014 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3015 continue; 3016 unp->unp_gcrefs++; 3017 unp_marked++; 3018 } 3019 } 3020 3021 static void 3022 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3023 { 3024 struct sockbuf *sb; 3025 3026 SOCK_LOCK_ASSERT(so); 3027 3028 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3029 return; 3030 3031 SOCK_RECVBUF_LOCK(so); 3032 switch (so->so_type) { 3033 case SOCK_DGRAM: 3034 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3035 unp_scan(so->so_rcv.uxdg_peeked, op); 3036 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3037 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3038 break; 3039 case SOCK_STREAM: 3040 case SOCK_SEQPACKET: 3041 unp_scan(so->so_rcv.sb_mb, op); 3042 break; 3043 } 3044 SOCK_RECVBUF_UNLOCK(so); 3045 } 3046 3047 static void 3048 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3049 { 3050 struct socket *so, *soa; 3051 3052 so = unp->unp_socket; 3053 SOCK_LOCK(so); 3054 if (SOLISTENING(so)) { 3055 /* 3056 * Mark all sockets in our accept queue. 3057 */ 3058 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3059 unp_scan_socket(soa, op); 3060 } else { 3061 /* 3062 * Mark all sockets we reference with RIGHTS. 3063 */ 3064 unp_scan_socket(so, op); 3065 } 3066 SOCK_UNLOCK(so); 3067 } 3068 3069 static int unp_recycled; 3070 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3071 "Number of unreachable sockets claimed by the garbage collector."); 3072 3073 static int unp_taskcount; 3074 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3075 "Number of times the garbage collector has run."); 3076 3077 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3078 "Number of active local sockets."); 3079 3080 static void 3081 unp_gc(__unused void *arg, int pending) 3082 { 3083 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3084 NULL }; 3085 struct unp_head **head; 3086 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3087 struct file *f, **unref; 3088 struct unpcb *unp, *unptmp; 3089 int i, total, unp_unreachable; 3090 3091 LIST_INIT(&unp_deadhead); 3092 unp_taskcount++; 3093 UNP_LINK_RLOCK(); 3094 /* 3095 * First determine which sockets may be in cycles. 3096 */ 3097 unp_unreachable = 0; 3098 3099 for (head = heads; *head != NULL; head++) 3100 LIST_FOREACH(unp, *head, unp_link) { 3101 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3102 ("%s: unp %p has unexpected gc flags 0x%x", 3103 __func__, unp, (unsigned int)unp->unp_gcflag)); 3104 3105 f = unp->unp_file; 3106 3107 /* 3108 * Check for an unreachable socket potentially in a 3109 * cycle. It must be in a queue as indicated by 3110 * msgcount, and this must equal the file reference 3111 * count. Note that when msgcount is 0 the file is 3112 * NULL. 3113 */ 3114 if (f != NULL && unp->unp_msgcount != 0 && 3115 refcount_load(&f->f_count) == unp->unp_msgcount) { 3116 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3117 unp->unp_gcflag |= UNPGC_DEAD; 3118 unp->unp_gcrefs = unp->unp_msgcount; 3119 unp_unreachable++; 3120 } 3121 } 3122 3123 /* 3124 * Scan all sockets previously marked as potentially being in a cycle 3125 * and remove the references each socket holds on any UNPGC_DEAD 3126 * sockets in its queue. After this step, all remaining references on 3127 * sockets marked UNPGC_DEAD should not be part of any cycle. 3128 */ 3129 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3130 unp_gc_scan(unp, unp_remove_dead_ref); 3131 3132 /* 3133 * If a socket still has a non-negative refcount, it cannot be in a 3134 * cycle. In this case increment refcount of all children iteratively. 3135 * Stop the scan once we do a complete loop without discovering 3136 * a new reachable socket. 3137 */ 3138 do { 3139 unp_marked = 0; 3140 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3141 if (unp->unp_gcrefs > 0) { 3142 unp->unp_gcflag &= ~UNPGC_DEAD; 3143 LIST_REMOVE(unp, unp_dead); 3144 KASSERT(unp_unreachable > 0, 3145 ("%s: unp_unreachable underflow.", 3146 __func__)); 3147 unp_unreachable--; 3148 unp_gc_scan(unp, unp_restore_undead_ref); 3149 } 3150 } while (unp_marked); 3151 3152 UNP_LINK_RUNLOCK(); 3153 3154 if (unp_unreachable == 0) 3155 return; 3156 3157 /* 3158 * Allocate space for a local array of dead unpcbs. 3159 * TODO: can this path be simplified by instead using the local 3160 * dead list at unp_deadhead, after taking out references 3161 * on the file object and/or unpcb and dropping the link lock? 3162 */ 3163 unref = malloc(unp_unreachable * sizeof(struct file *), 3164 M_TEMP, M_WAITOK); 3165 3166 /* 3167 * Iterate looking for sockets which have been specifically marked 3168 * as unreachable and store them locally. 3169 */ 3170 UNP_LINK_RLOCK(); 3171 total = 0; 3172 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3173 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3174 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3175 unp->unp_gcflag &= ~UNPGC_DEAD; 3176 f = unp->unp_file; 3177 if (unp->unp_msgcount == 0 || f == NULL || 3178 refcount_load(&f->f_count) != unp->unp_msgcount || 3179 !fhold(f)) 3180 continue; 3181 unref[total++] = f; 3182 KASSERT(total <= unp_unreachable, 3183 ("%s: incorrect unreachable count.", __func__)); 3184 } 3185 UNP_LINK_RUNLOCK(); 3186 3187 /* 3188 * Now flush all sockets, free'ing rights. This will free the 3189 * struct files associated with these sockets but leave each socket 3190 * with one remaining ref. 3191 */ 3192 for (i = 0; i < total; i++) { 3193 struct socket *so; 3194 3195 so = unref[i]->f_data; 3196 CURVNET_SET(so->so_vnet); 3197 socantrcvmore(so); 3198 unp_dispose(so); 3199 CURVNET_RESTORE(); 3200 } 3201 3202 /* 3203 * And finally release the sockets so they can be reclaimed. 3204 */ 3205 for (i = 0; i < total; i++) 3206 fdrop(unref[i], NULL); 3207 unp_recycled += total; 3208 free(unref, M_TEMP); 3209 } 3210 3211 /* 3212 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3213 */ 3214 static void 3215 unp_dispose(struct socket *so) 3216 { 3217 struct sockbuf *sb; 3218 struct unpcb *unp; 3219 struct mbuf *m; 3220 3221 MPASS(!SOLISTENING(so)); 3222 3223 unp = sotounpcb(so); 3224 UNP_LINK_WLOCK(); 3225 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3226 UNP_LINK_WUNLOCK(); 3227 3228 /* 3229 * Grab our special mbufs before calling sbrelease(). 3230 */ 3231 SOCK_RECVBUF_LOCK(so); 3232 switch (so->so_type) { 3233 case SOCK_DGRAM: 3234 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3235 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3236 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3237 /* Note: socket of sb may reconnect. */ 3238 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3239 } 3240 sb = &so->so_rcv; 3241 if (sb->uxdg_peeked != NULL) { 3242 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3243 m_stailqpkt); 3244 sb->uxdg_peeked = NULL; 3245 } 3246 m = STAILQ_FIRST(&sb->uxdg_mb); 3247 STAILQ_INIT(&sb->uxdg_mb); 3248 /* XXX: our shortened sbrelease() */ 3249 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3250 RLIM_INFINITY); 3251 /* 3252 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3253 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3254 * data to the socket. 3255 * We came here either through shutdown(2) or from the final 3256 * sofree(). The sofree() case is simple as it guarantees 3257 * that no more sends will happen, however we can race with 3258 * unp_disconnect() from our peer. The shutdown(2) case is 3259 * more exotic. It would call into unp_dispose() only if 3260 * socket is SS_ISCONNECTED. This is possible if we did 3261 * connect(2) on this socket and we also had it bound with 3262 * bind(2) and receive connections from other sockets. 3263 * Because uipc_shutdown() violates POSIX (see comment 3264 * there) we will end up here shutting down our receive side. 3265 * Of course this will have affect not only on the peer we 3266 * connect(2)ed to, but also on all of the peers who had 3267 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3268 */ 3269 sb->sb_state |= SBS_CANTRCVMORE; 3270 break; 3271 case SOCK_STREAM: 3272 case SOCK_SEQPACKET: 3273 sb = &so->so_rcv; 3274 m = sbcut_locked(sb, sb->sb_ccc); 3275 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3276 ("%s: ccc %u mb %p mbcnt %u", __func__, 3277 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3278 sbrelease_locked(so, SO_RCV); 3279 break; 3280 } 3281 SOCK_RECVBUF_UNLOCK(so); 3282 if (SOCK_IO_RECV_OWNED(so)) 3283 SOCK_IO_RECV_UNLOCK(so); 3284 3285 if (m != NULL) { 3286 unp_scan(m, unp_freerights); 3287 m_freem(m); 3288 } 3289 } 3290 3291 static void 3292 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3293 { 3294 struct mbuf *m; 3295 struct cmsghdr *cm; 3296 void *data; 3297 socklen_t clen, datalen; 3298 3299 while (m0 != NULL) { 3300 for (m = m0; m; m = m->m_next) { 3301 if (m->m_type != MT_CONTROL) 3302 continue; 3303 3304 cm = mtod(m, struct cmsghdr *); 3305 clen = m->m_len; 3306 3307 while (cm != NULL) { 3308 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3309 break; 3310 3311 data = CMSG_DATA(cm); 3312 datalen = (caddr_t)cm + cm->cmsg_len 3313 - (caddr_t)data; 3314 3315 if (cm->cmsg_level == SOL_SOCKET && 3316 cm->cmsg_type == SCM_RIGHTS) { 3317 (*op)(data, datalen / 3318 sizeof(struct filedescent *)); 3319 } 3320 3321 if (CMSG_SPACE(datalen) < clen) { 3322 clen -= CMSG_SPACE(datalen); 3323 cm = (struct cmsghdr *) 3324 ((caddr_t)cm + CMSG_SPACE(datalen)); 3325 } else { 3326 clen = 0; 3327 cm = NULL; 3328 } 3329 } 3330 } 3331 m0 = m0->m_nextpkt; 3332 } 3333 } 3334 3335 /* 3336 * Definitions of protocols supported in the LOCAL domain. 3337 */ 3338 static struct protosw streamproto = { 3339 .pr_type = SOCK_STREAM, 3340 .pr_flags = PR_CONNREQUIRED | PR_WANTRCVD | PR_CAPATTACH, 3341 .pr_ctloutput = &uipc_ctloutput, 3342 .pr_abort = uipc_abort, 3343 .pr_accept = uipc_peeraddr, 3344 .pr_attach = uipc_attach, 3345 .pr_bind = uipc_bind, 3346 .pr_bindat = uipc_bindat, 3347 .pr_connect = uipc_connect, 3348 .pr_connectat = uipc_connectat, 3349 .pr_connect2 = uipc_connect2, 3350 .pr_detach = uipc_detach, 3351 .pr_disconnect = uipc_disconnect, 3352 .pr_listen = uipc_listen, 3353 .pr_peeraddr = uipc_peeraddr, 3354 .pr_rcvd = uipc_rcvd, 3355 .pr_send = uipc_send, 3356 .pr_ready = uipc_ready, 3357 .pr_sense = uipc_sense, 3358 .pr_shutdown = uipc_shutdown, 3359 .pr_sockaddr = uipc_sockaddr, 3360 .pr_soreceive = soreceive_generic, 3361 .pr_close = uipc_close, 3362 }; 3363 3364 static struct protosw dgramproto = { 3365 .pr_type = SOCK_DGRAM, 3366 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH | PR_SOCKBUF, 3367 .pr_ctloutput = &uipc_ctloutput, 3368 .pr_abort = uipc_abort, 3369 .pr_accept = uipc_peeraddr, 3370 .pr_attach = uipc_attach, 3371 .pr_bind = uipc_bind, 3372 .pr_bindat = uipc_bindat, 3373 .pr_connect = uipc_connect, 3374 .pr_connectat = uipc_connectat, 3375 .pr_connect2 = uipc_connect2, 3376 .pr_detach = uipc_detach, 3377 .pr_disconnect = uipc_disconnect, 3378 .pr_peeraddr = uipc_peeraddr, 3379 .pr_sosend = uipc_sosend_dgram, 3380 .pr_sense = uipc_sense, 3381 .pr_shutdown = uipc_shutdown, 3382 .pr_sockaddr = uipc_sockaddr, 3383 .pr_soreceive = uipc_soreceive_dgram, 3384 .pr_close = uipc_close, 3385 }; 3386 3387 static struct protosw seqpacketproto = { 3388 .pr_type = SOCK_SEQPACKET, 3389 /* 3390 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3391 * due to our use of sbappendaddr. A new sbappend variants is needed 3392 * that supports both atomic record writes and control data. 3393 */ 3394 .pr_flags = PR_ADDR | PR_ATOMIC | PR_CONNREQUIRED | 3395 PR_WANTRCVD | PR_CAPATTACH, 3396 .pr_ctloutput = &uipc_ctloutput, 3397 .pr_abort = uipc_abort, 3398 .pr_accept = uipc_peeraddr, 3399 .pr_attach = uipc_attach, 3400 .pr_bind = uipc_bind, 3401 .pr_bindat = uipc_bindat, 3402 .pr_connect = uipc_connect, 3403 .pr_connectat = uipc_connectat, 3404 .pr_connect2 = uipc_connect2, 3405 .pr_detach = uipc_detach, 3406 .pr_disconnect = uipc_disconnect, 3407 .pr_listen = uipc_listen, 3408 .pr_peeraddr = uipc_peeraddr, 3409 .pr_rcvd = uipc_rcvd, 3410 .pr_send = uipc_send, 3411 .pr_sense = uipc_sense, 3412 .pr_shutdown = uipc_shutdown, 3413 .pr_sockaddr = uipc_sockaddr, 3414 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3415 .pr_close = uipc_close, 3416 }; 3417 3418 static struct domain localdomain = { 3419 .dom_family = AF_LOCAL, 3420 .dom_name = "local", 3421 .dom_externalize = unp_externalize, 3422 .dom_nprotosw = 3, 3423 .dom_protosw = { 3424 &streamproto, 3425 &dgramproto, 3426 &seqpacketproto, 3427 } 3428 }; 3429 DOMAIN_SET(local); 3430 3431 /* 3432 * A helper function called by VFS before socket-type vnode reclamation. 3433 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3434 * use count. 3435 */ 3436 void 3437 vfs_unp_reclaim(struct vnode *vp) 3438 { 3439 struct unpcb *unp; 3440 int active; 3441 struct mtx *vplock; 3442 3443 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3444 KASSERT(vp->v_type == VSOCK, 3445 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3446 3447 active = 0; 3448 vplock = mtx_pool_find(mtxpool_sleep, vp); 3449 mtx_lock(vplock); 3450 VOP_UNP_CONNECT(vp, &unp); 3451 if (unp == NULL) 3452 goto done; 3453 UNP_PCB_LOCK(unp); 3454 if (unp->unp_vnode == vp) { 3455 VOP_UNP_DETACH(vp); 3456 unp->unp_vnode = NULL; 3457 active = 1; 3458 } 3459 UNP_PCB_UNLOCK(unp); 3460 done: 3461 mtx_unlock(vplock); 3462 if (active) 3463 vunref(vp); 3464 } 3465 3466 #ifdef DDB 3467 static void 3468 db_print_indent(int indent) 3469 { 3470 int i; 3471 3472 for (i = 0; i < indent; i++) 3473 db_printf(" "); 3474 } 3475 3476 static void 3477 db_print_unpflags(int unp_flags) 3478 { 3479 int comma; 3480 3481 comma = 0; 3482 if (unp_flags & UNP_HAVEPC) { 3483 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3484 comma = 1; 3485 } 3486 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3487 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3488 comma = 1; 3489 } 3490 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3491 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3492 comma = 1; 3493 } 3494 if (unp_flags & UNP_CONNWAIT) { 3495 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3496 comma = 1; 3497 } 3498 if (unp_flags & UNP_CONNECTING) { 3499 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3500 comma = 1; 3501 } 3502 if (unp_flags & UNP_BINDING) { 3503 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3504 comma = 1; 3505 } 3506 } 3507 3508 static void 3509 db_print_xucred(int indent, struct xucred *xu) 3510 { 3511 int comma, i; 3512 3513 db_print_indent(indent); 3514 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3515 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3516 db_print_indent(indent); 3517 db_printf("cr_groups: "); 3518 comma = 0; 3519 for (i = 0; i < xu->cr_ngroups; i++) { 3520 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3521 comma = 1; 3522 } 3523 db_printf("\n"); 3524 } 3525 3526 static void 3527 db_print_unprefs(int indent, struct unp_head *uh) 3528 { 3529 struct unpcb *unp; 3530 int counter; 3531 3532 counter = 0; 3533 LIST_FOREACH(unp, uh, unp_reflink) { 3534 if (counter % 4 == 0) 3535 db_print_indent(indent); 3536 db_printf("%p ", unp); 3537 if (counter % 4 == 3) 3538 db_printf("\n"); 3539 counter++; 3540 } 3541 if (counter != 0 && counter % 4 != 0) 3542 db_printf("\n"); 3543 } 3544 3545 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3546 { 3547 struct unpcb *unp; 3548 3549 if (!have_addr) { 3550 db_printf("usage: show unpcb <addr>\n"); 3551 return; 3552 } 3553 unp = (struct unpcb *)addr; 3554 3555 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3556 unp->unp_vnode); 3557 3558 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3559 unp->unp_conn); 3560 3561 db_printf("unp_refs:\n"); 3562 db_print_unprefs(2, &unp->unp_refs); 3563 3564 /* XXXRW: Would be nice to print the full address, if any. */ 3565 db_printf("unp_addr: %p\n", unp->unp_addr); 3566 3567 db_printf("unp_gencnt: %llu\n", 3568 (unsigned long long)unp->unp_gencnt); 3569 3570 db_printf("unp_flags: %x (", unp->unp_flags); 3571 db_print_unpflags(unp->unp_flags); 3572 db_printf(")\n"); 3573 3574 db_printf("unp_peercred:\n"); 3575 db_print_xucred(2, &unp->unp_peercred); 3576 3577 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3578 } 3579 #endif 3580