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 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 35 */ 36 37 /* 38 * UNIX Domain (Local) Sockets 39 * 40 * This is an implementation of UNIX (local) domain sockets. Each socket has 41 * an associated struct unpcb (UNIX protocol control block). Stream sockets 42 * may be connected to 0 or 1 other socket. Datagram sockets may be 43 * connected to 0, 1, or many other sockets. Sockets may be created and 44 * connected in pairs (socketpair(2)), or bound/connected to using the file 45 * system name space. For most purposes, only the receive socket buffer is 46 * used, as sending on one socket delivers directly to the receive socket 47 * buffer of a second socket. 48 * 49 * The implementation is substantially complicated by the fact that 50 * "ancillary data", such as file descriptors or credentials, may be passed 51 * across UNIX domain sockets. The potential for passing UNIX domain sockets 52 * over other UNIX domain sockets requires the implementation of a simple 53 * garbage collector to find and tear down cycles of disconnected sockets. 54 * 55 * TODO: 56 * RDM 57 * rethink name space problems 58 * need a proper out-of-band 59 */ 60 61 #include <sys/cdefs.h> 62 #include "opt_ddb.h" 63 64 #include <sys/param.h> 65 #include <sys/capsicum.h> 66 #include <sys/domain.h> 67 #include <sys/eventhandler.h> 68 #include <sys/fcntl.h> 69 #include <sys/file.h> 70 #include <sys/filedesc.h> 71 #include <sys/kernel.h> 72 #include <sys/lock.h> 73 #include <sys/malloc.h> 74 #include <sys/mbuf.h> 75 #include <sys/mount.h> 76 #include <sys/mutex.h> 77 #include <sys/namei.h> 78 #include <sys/proc.h> 79 #include <sys/protosw.h> 80 #include <sys/queue.h> 81 #include <sys/resourcevar.h> 82 #include <sys/rwlock.h> 83 #include <sys/socket.h> 84 #include <sys/socketvar.h> 85 #include <sys/signalvar.h> 86 #include <sys/stat.h> 87 #include <sys/sx.h> 88 #include <sys/sysctl.h> 89 #include <sys/systm.h> 90 #include <sys/taskqueue.h> 91 #include <sys/un.h> 92 #include <sys/unpcb.h> 93 #include <sys/vnode.h> 94 95 #include <net/vnet.h> 96 97 #ifdef DDB 98 #include <ddb/ddb.h> 99 #endif 100 101 #include <security/mac/mac_framework.h> 102 103 #include <vm/uma.h> 104 105 MALLOC_DECLARE(M_FILECAPS); 106 107 static struct domain localdomain; 108 109 static uma_zone_t unp_zone; 110 static unp_gen_t unp_gencnt; /* (l) */ 111 static u_int unp_count; /* (l) Count of local sockets. */ 112 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 113 static int unp_rights; /* (g) File descriptors in flight. */ 114 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 115 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 116 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 117 118 struct unp_defer { 119 SLIST_ENTRY(unp_defer) ud_link; 120 struct file *ud_fp; 121 }; 122 static SLIST_HEAD(, unp_defer) unp_defers; 123 static int unp_defers_count; 124 125 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 126 127 /* 128 * Garbage collection of cyclic file descriptor/socket references occurs 129 * asynchronously in a taskqueue context in order to avoid recursion and 130 * reentrance in the UNIX domain socket, file descriptor, and socket layer 131 * code. See unp_gc() for a full description. 132 */ 133 static struct timeout_task unp_gc_task; 134 135 /* 136 * The close of unix domain sockets attached as SCM_RIGHTS is 137 * postponed to the taskqueue, to avoid arbitrary recursion depth. 138 * The attached sockets might have another sockets attached. 139 */ 140 static struct task unp_defer_task; 141 142 /* 143 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 144 * stream sockets, although the total for sender and receiver is actually 145 * only PIPSIZ. 146 * 147 * Datagram sockets really use the sendspace as the maximum datagram size, 148 * and don't really want to reserve the sendspace. Their recvspace should be 149 * large enough for at least one max-size datagram plus address. 150 */ 151 #ifndef PIPSIZ 152 #define PIPSIZ 8192 153 #endif 154 static u_long unpst_sendspace = PIPSIZ; 155 static u_long unpst_recvspace = PIPSIZ; 156 static u_long unpdg_maxdgram = 2*1024; 157 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */ 158 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 159 static u_long unpsp_recvspace = PIPSIZ; 160 161 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 162 "Local domain"); 163 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 164 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 165 "SOCK_STREAM"); 166 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 167 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 168 "SOCK_DGRAM"); 169 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 171 "SOCK_SEQPACKET"); 172 173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 174 &unpst_sendspace, 0, "Default stream send space."); 175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 176 &unpst_recvspace, 0, "Default stream receive space."); 177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 178 &unpdg_maxdgram, 0, "Maximum datagram size."); 179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 180 &unpdg_recvspace, 0, "Default datagram receive space."); 181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 182 &unpsp_sendspace, 0, "Default seqpacket send space."); 183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 184 &unpsp_recvspace, 0, "Default seqpacket receive space."); 185 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 186 "File descriptors in flight."); 187 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 188 &unp_defers_count, 0, 189 "File descriptors deferred to taskqueue for close."); 190 191 /* 192 * Locking and synchronization: 193 * 194 * Several types of locks exist in the local domain socket implementation: 195 * - a global linkage lock 196 * - a global connection list lock 197 * - the mtxpool lock 198 * - per-unpcb mutexes 199 * 200 * The linkage lock protects the global socket lists, the generation number 201 * counter and garbage collector state. 202 * 203 * The connection list lock protects the list of referring sockets in a datagram 204 * socket PCB. This lock is also overloaded to protect a global list of 205 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 206 * messages. To avoid recursion, such references are released by a dedicated 207 * thread. 208 * 209 * The mtxpool lock protects the vnode from being modified while referenced. 210 * Lock ordering rules require that it be acquired before any PCB locks. 211 * 212 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 213 * unpcb. This includes the unp_conn field, which either links two connected 214 * PCBs together (for connected socket types) or points at the destination 215 * socket (for connectionless socket types). The operations of creating or 216 * destroying a connection therefore involve locking multiple PCBs. To avoid 217 * lock order reversals, in some cases this involves dropping a PCB lock and 218 * using a reference counter to maintain liveness. 219 * 220 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 221 * allocated in pr_attach() and freed in pr_detach(). The validity of that 222 * pointer is an invariant, so no lock is required to dereference the so_pcb 223 * pointer if a valid socket reference is held by the caller. In practice, 224 * this is always true during operations performed on a socket. Each unpcb 225 * has a back-pointer to its socket, unp_socket, which will be stable under 226 * the same circumstances. 227 * 228 * This pointer may only be safely dereferenced as long as a valid reference 229 * to the unpcb is held. Typically, this reference will be from the socket, 230 * or from another unpcb when the referring unpcb's lock is held (in order 231 * that the reference not be invalidated during use). For example, to follow 232 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 233 * that detach is not run clearing unp_socket. 234 * 235 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 236 * protocols, bind() is a non-atomic operation, and connect() requires 237 * potential sleeping in the protocol, due to potentially waiting on local or 238 * distributed file systems. We try to separate "lookup" operations, which 239 * may sleep, and the IPC operations themselves, which typically can occur 240 * with relative atomicity as locks can be held over the entire operation. 241 * 242 * Another tricky issue is simultaneous multi-threaded or multi-process 243 * access to a single UNIX domain socket. These are handled by the flags 244 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 245 * binding, both of which involve dropping UNIX domain socket locks in order 246 * to perform namei() and other file system operations. 247 */ 248 static struct rwlock unp_link_rwlock; 249 static struct mtx unp_defers_lock; 250 251 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 252 "unp_link_rwlock") 253 254 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 255 RA_LOCKED) 256 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 257 RA_UNLOCKED) 258 259 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 260 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 261 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 262 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 263 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 264 RA_WLOCKED) 265 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 266 267 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 268 "unp_defer", NULL, MTX_DEF) 269 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 270 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 271 272 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 273 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 274 275 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 276 "unp", "unp", \ 277 MTX_DUPOK|MTX_DEF) 278 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 279 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 280 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 281 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 282 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 283 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 284 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 285 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 286 287 static int uipc_connect2(struct socket *, struct socket *); 288 static int uipc_ctloutput(struct socket *, struct sockopt *); 289 static int unp_connect(struct socket *, struct sockaddr *, 290 struct thread *); 291 static int unp_connectat(int, struct socket *, struct sockaddr *, 292 struct thread *, bool); 293 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how; 294 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how); 295 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 296 static void unp_dispose(struct socket *so); 297 static void unp_shutdown(struct unpcb *); 298 static void unp_drop(struct unpcb *); 299 static void unp_gc(__unused void *, int); 300 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 301 static void unp_discard(struct file *); 302 static void unp_freerights(struct filedescent **, int); 303 static int unp_internalize(struct mbuf **, struct thread *, 304 struct mbuf **, u_int *, u_int *); 305 static void unp_internalize_fp(struct file *); 306 static int unp_externalize(struct mbuf *, struct mbuf **, int); 307 static int unp_externalize_fp(struct file *); 308 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 309 int, struct mbuf **, u_int *, u_int *); 310 static void unp_process_defers(void * __unused, int); 311 312 static void 313 unp_pcb_hold(struct unpcb *unp) 314 { 315 u_int old __unused; 316 317 old = refcount_acquire(&unp->unp_refcount); 318 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 319 } 320 321 static __result_use_check bool 322 unp_pcb_rele(struct unpcb *unp) 323 { 324 bool ret; 325 326 UNP_PCB_LOCK_ASSERT(unp); 327 328 if ((ret = refcount_release(&unp->unp_refcount))) { 329 UNP_PCB_UNLOCK(unp); 330 UNP_PCB_LOCK_DESTROY(unp); 331 uma_zfree(unp_zone, unp); 332 } 333 return (ret); 334 } 335 336 static void 337 unp_pcb_rele_notlast(struct unpcb *unp) 338 { 339 bool ret __unused; 340 341 ret = refcount_release(&unp->unp_refcount); 342 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 343 } 344 345 static void 346 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 347 { 348 UNP_PCB_UNLOCK_ASSERT(unp); 349 UNP_PCB_UNLOCK_ASSERT(unp2); 350 351 if (unp == unp2) { 352 UNP_PCB_LOCK(unp); 353 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 354 UNP_PCB_LOCK(unp); 355 UNP_PCB_LOCK(unp2); 356 } else { 357 UNP_PCB_LOCK(unp2); 358 UNP_PCB_LOCK(unp); 359 } 360 } 361 362 static void 363 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 364 { 365 UNP_PCB_UNLOCK(unp); 366 if (unp != unp2) 367 UNP_PCB_UNLOCK(unp2); 368 } 369 370 /* 371 * Try to lock the connected peer of an already locked socket. In some cases 372 * this requires that we unlock the current socket. The pairbusy counter is 373 * used to block concurrent connection attempts while the lock is dropped. The 374 * caller must be careful to revalidate PCB state. 375 */ 376 static struct unpcb * 377 unp_pcb_lock_peer(struct unpcb *unp) 378 { 379 struct unpcb *unp2; 380 381 UNP_PCB_LOCK_ASSERT(unp); 382 unp2 = unp->unp_conn; 383 if (unp2 == NULL) 384 return (NULL); 385 if (__predict_false(unp == unp2)) 386 return (unp); 387 388 UNP_PCB_UNLOCK_ASSERT(unp2); 389 390 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 391 return (unp2); 392 if ((uintptr_t)unp2 > (uintptr_t)unp) { 393 UNP_PCB_LOCK(unp2); 394 return (unp2); 395 } 396 unp->unp_pairbusy++; 397 unp_pcb_hold(unp2); 398 UNP_PCB_UNLOCK(unp); 399 400 UNP_PCB_LOCK(unp2); 401 UNP_PCB_LOCK(unp); 402 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 403 ("%s: socket %p was reconnected", __func__, unp)); 404 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 405 unp->unp_flags &= ~UNP_WAITING; 406 wakeup(unp); 407 } 408 if (unp_pcb_rele(unp2)) { 409 /* unp2 is unlocked. */ 410 return (NULL); 411 } 412 if (unp->unp_conn == NULL) { 413 UNP_PCB_UNLOCK(unp2); 414 return (NULL); 415 } 416 return (unp2); 417 } 418 419 static void 420 uipc_abort(struct socket *so) 421 { 422 struct unpcb *unp, *unp2; 423 424 unp = sotounpcb(so); 425 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 426 UNP_PCB_UNLOCK_ASSERT(unp); 427 428 UNP_PCB_LOCK(unp); 429 unp2 = unp->unp_conn; 430 if (unp2 != NULL) { 431 unp_pcb_hold(unp2); 432 UNP_PCB_UNLOCK(unp); 433 unp_drop(unp2); 434 } else 435 UNP_PCB_UNLOCK(unp); 436 } 437 438 static int 439 uipc_accept(struct socket *so, struct sockaddr **nam) 440 { 441 struct unpcb *unp, *unp2; 442 const struct sockaddr *sa; 443 444 /* 445 * Pass back name of connected socket, if it was bound and we are 446 * still connected (our peer may have closed already!). 447 */ 448 unp = sotounpcb(so); 449 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 450 451 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 452 UNP_PCB_LOCK(unp); 453 unp2 = unp_pcb_lock_peer(unp); 454 if (unp2 != NULL && unp2->unp_addr != NULL) 455 sa = (struct sockaddr *)unp2->unp_addr; 456 else 457 sa = &sun_noname; 458 bcopy(sa, *nam, sa->sa_len); 459 if (unp2 != NULL) 460 unp_pcb_unlock_pair(unp, unp2); 461 else 462 UNP_PCB_UNLOCK(unp); 463 return (0); 464 } 465 466 static int 467 uipc_attach(struct socket *so, int proto, struct thread *td) 468 { 469 u_long sendspace, recvspace; 470 struct unpcb *unp; 471 int error; 472 bool locked; 473 474 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 475 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 476 switch (so->so_type) { 477 case SOCK_STREAM: 478 sendspace = unpst_sendspace; 479 recvspace = unpst_recvspace; 480 break; 481 482 case SOCK_DGRAM: 483 STAILQ_INIT(&so->so_rcv.uxdg_mb); 484 STAILQ_INIT(&so->so_snd.uxdg_mb); 485 TAILQ_INIT(&so->so_rcv.uxdg_conns); 486 /* 487 * Since send buffer is either bypassed or is a part 488 * of one-to-many receive buffer, we assign both space 489 * limits to unpdg_recvspace. 490 */ 491 sendspace = recvspace = unpdg_recvspace; 492 break; 493 494 case SOCK_SEQPACKET: 495 sendspace = unpsp_sendspace; 496 recvspace = unpsp_recvspace; 497 break; 498 499 default: 500 panic("uipc_attach"); 501 } 502 error = soreserve(so, sendspace, recvspace); 503 if (error) 504 return (error); 505 } 506 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 507 if (unp == NULL) 508 return (ENOBUFS); 509 LIST_INIT(&unp->unp_refs); 510 UNP_PCB_LOCK_INIT(unp); 511 unp->unp_socket = so; 512 so->so_pcb = unp; 513 refcount_init(&unp->unp_refcount, 1); 514 515 if ((locked = UNP_LINK_WOWNED()) == false) 516 UNP_LINK_WLOCK(); 517 518 unp->unp_gencnt = ++unp_gencnt; 519 unp->unp_ino = ++unp_ino; 520 unp_count++; 521 switch (so->so_type) { 522 case SOCK_STREAM: 523 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 524 break; 525 526 case SOCK_DGRAM: 527 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 528 break; 529 530 case SOCK_SEQPACKET: 531 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 532 break; 533 534 default: 535 panic("uipc_attach"); 536 } 537 538 if (locked == false) 539 UNP_LINK_WUNLOCK(); 540 541 return (0); 542 } 543 544 static int 545 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 546 { 547 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 548 struct vattr vattr; 549 int error, namelen; 550 struct nameidata nd; 551 struct unpcb *unp; 552 struct vnode *vp; 553 struct mount *mp; 554 cap_rights_t rights; 555 char *buf; 556 557 if (nam->sa_family != AF_UNIX) 558 return (EAFNOSUPPORT); 559 560 unp = sotounpcb(so); 561 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 562 563 if (soun->sun_len > sizeof(struct sockaddr_un)) 564 return (EINVAL); 565 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 566 if (namelen <= 0) 567 return (EINVAL); 568 569 /* 570 * We don't allow simultaneous bind() calls on a single UNIX domain 571 * socket, so flag in-progress operations, and return an error if an 572 * operation is already in progress. 573 * 574 * Historically, we have not allowed a socket to be rebound, so this 575 * also returns an error. Not allowing re-binding simplifies the 576 * implementation and avoids a great many possible failure modes. 577 */ 578 UNP_PCB_LOCK(unp); 579 if (unp->unp_vnode != NULL) { 580 UNP_PCB_UNLOCK(unp); 581 return (EINVAL); 582 } 583 if (unp->unp_flags & UNP_BINDING) { 584 UNP_PCB_UNLOCK(unp); 585 return (EALREADY); 586 } 587 unp->unp_flags |= UNP_BINDING; 588 UNP_PCB_UNLOCK(unp); 589 590 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 591 bcopy(soun->sun_path, buf, namelen); 592 buf[namelen] = 0; 593 594 restart: 595 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, 596 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 597 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 598 error = namei(&nd); 599 if (error) 600 goto error; 601 vp = nd.ni_vp; 602 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 603 NDFREE_PNBUF(&nd); 604 if (nd.ni_dvp == vp) 605 vrele(nd.ni_dvp); 606 else 607 vput(nd.ni_dvp); 608 if (vp != NULL) { 609 vrele(vp); 610 error = EADDRINUSE; 611 goto error; 612 } 613 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); 614 if (error) 615 goto error; 616 goto restart; 617 } 618 VATTR_NULL(&vattr); 619 vattr.va_type = VSOCK; 620 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 621 #ifdef MAC 622 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 623 &vattr); 624 #endif 625 if (error == 0) 626 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 627 NDFREE_PNBUF(&nd); 628 if (error) { 629 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 630 vn_finished_write(mp); 631 if (error == ERELOOKUP) 632 goto restart; 633 goto error; 634 } 635 vp = nd.ni_vp; 636 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 637 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 638 639 UNP_PCB_LOCK(unp); 640 VOP_UNP_BIND(vp, unp); 641 unp->unp_vnode = vp; 642 unp->unp_addr = soun; 643 unp->unp_flags &= ~UNP_BINDING; 644 UNP_PCB_UNLOCK(unp); 645 vref(vp); 646 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 647 vn_finished_write(mp); 648 free(buf, M_TEMP); 649 return (0); 650 651 error: 652 UNP_PCB_LOCK(unp); 653 unp->unp_flags &= ~UNP_BINDING; 654 UNP_PCB_UNLOCK(unp); 655 free(buf, M_TEMP); 656 return (error); 657 } 658 659 static int 660 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 661 { 662 663 return (uipc_bindat(AT_FDCWD, so, nam, td)); 664 } 665 666 static int 667 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 668 { 669 int error; 670 671 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 672 error = unp_connect(so, nam, td); 673 return (error); 674 } 675 676 static int 677 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 678 struct thread *td) 679 { 680 int error; 681 682 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 683 error = unp_connectat(fd, so, nam, td, false); 684 return (error); 685 } 686 687 static void 688 uipc_close(struct socket *so) 689 { 690 struct unpcb *unp, *unp2; 691 struct vnode *vp = NULL; 692 struct mtx *vplock; 693 694 unp = sotounpcb(so); 695 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 696 697 vplock = NULL; 698 if ((vp = unp->unp_vnode) != NULL) { 699 vplock = mtx_pool_find(mtxpool_sleep, vp); 700 mtx_lock(vplock); 701 } 702 UNP_PCB_LOCK(unp); 703 if (vp && unp->unp_vnode == NULL) { 704 mtx_unlock(vplock); 705 vp = NULL; 706 } 707 if (vp != NULL) { 708 VOP_UNP_DETACH(vp); 709 unp->unp_vnode = NULL; 710 } 711 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 712 unp_disconnect(unp, unp2); 713 else 714 UNP_PCB_UNLOCK(unp); 715 if (vp) { 716 mtx_unlock(vplock); 717 vrele(vp); 718 } 719 } 720 721 static int 722 uipc_connect2(struct socket *so1, struct socket *so2) 723 { 724 struct unpcb *unp, *unp2; 725 726 if (so1->so_type != so2->so_type) 727 return (EPROTOTYPE); 728 729 unp = so1->so_pcb; 730 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 731 unp2 = so2->so_pcb; 732 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 733 unp_pcb_lock_pair(unp, unp2); 734 unp_connect2(so1, so2, PRU_CONNECT2); 735 unp_pcb_unlock_pair(unp, unp2); 736 737 return (0); 738 } 739 740 static void 741 uipc_detach(struct socket *so) 742 { 743 struct unpcb *unp, *unp2; 744 struct mtx *vplock; 745 struct vnode *vp; 746 int local_unp_rights; 747 748 unp = sotounpcb(so); 749 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 750 751 vp = NULL; 752 vplock = NULL; 753 754 UNP_LINK_WLOCK(); 755 LIST_REMOVE(unp, unp_link); 756 if (unp->unp_gcflag & UNPGC_DEAD) 757 LIST_REMOVE(unp, unp_dead); 758 unp->unp_gencnt = ++unp_gencnt; 759 --unp_count; 760 UNP_LINK_WUNLOCK(); 761 762 UNP_PCB_UNLOCK_ASSERT(unp); 763 restart: 764 if ((vp = unp->unp_vnode) != NULL) { 765 vplock = mtx_pool_find(mtxpool_sleep, vp); 766 mtx_lock(vplock); 767 } 768 UNP_PCB_LOCK(unp); 769 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 770 if (vplock) 771 mtx_unlock(vplock); 772 UNP_PCB_UNLOCK(unp); 773 goto restart; 774 } 775 if ((vp = unp->unp_vnode) != NULL) { 776 VOP_UNP_DETACH(vp); 777 unp->unp_vnode = NULL; 778 } 779 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 780 unp_disconnect(unp, unp2); 781 else 782 UNP_PCB_UNLOCK(unp); 783 784 UNP_REF_LIST_LOCK(); 785 while (!LIST_EMPTY(&unp->unp_refs)) { 786 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 787 788 unp_pcb_hold(ref); 789 UNP_REF_LIST_UNLOCK(); 790 791 MPASS(ref != unp); 792 UNP_PCB_UNLOCK_ASSERT(ref); 793 unp_drop(ref); 794 UNP_REF_LIST_LOCK(); 795 } 796 UNP_REF_LIST_UNLOCK(); 797 798 UNP_PCB_LOCK(unp); 799 local_unp_rights = unp_rights; 800 unp->unp_socket->so_pcb = NULL; 801 unp->unp_socket = NULL; 802 free(unp->unp_addr, M_SONAME); 803 unp->unp_addr = NULL; 804 if (!unp_pcb_rele(unp)) 805 UNP_PCB_UNLOCK(unp); 806 if (vp) { 807 mtx_unlock(vplock); 808 vrele(vp); 809 } 810 if (local_unp_rights) 811 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 812 813 switch (so->so_type) { 814 case SOCK_DGRAM: 815 /* 816 * Everything should have been unlinked/freed by unp_dispose() 817 * and/or unp_disconnect(). 818 */ 819 MPASS(so->so_rcv.uxdg_peeked == NULL); 820 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 821 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 822 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 823 } 824 } 825 826 static int 827 uipc_disconnect(struct socket *so) 828 { 829 struct unpcb *unp, *unp2; 830 831 unp = sotounpcb(so); 832 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 833 834 UNP_PCB_LOCK(unp); 835 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 836 unp_disconnect(unp, unp2); 837 else 838 UNP_PCB_UNLOCK(unp); 839 return (0); 840 } 841 842 static int 843 uipc_listen(struct socket *so, int backlog, struct thread *td) 844 { 845 struct unpcb *unp; 846 int error; 847 848 MPASS(so->so_type != SOCK_DGRAM); 849 850 /* 851 * Synchronize with concurrent connection attempts. 852 */ 853 error = 0; 854 unp = sotounpcb(so); 855 UNP_PCB_LOCK(unp); 856 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 857 error = EINVAL; 858 else if (unp->unp_vnode == NULL) 859 error = EDESTADDRREQ; 860 if (error != 0) { 861 UNP_PCB_UNLOCK(unp); 862 return (error); 863 } 864 865 SOCK_LOCK(so); 866 error = solisten_proto_check(so); 867 if (error == 0) { 868 cru2xt(td, &unp->unp_peercred); 869 solisten_proto(so, backlog); 870 } 871 SOCK_UNLOCK(so); 872 UNP_PCB_UNLOCK(unp); 873 return (error); 874 } 875 876 static int 877 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 878 { 879 struct unpcb *unp, *unp2; 880 const struct sockaddr *sa; 881 882 unp = sotounpcb(so); 883 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 884 885 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 886 887 UNP_PCB_LOCK(unp); 888 unp2 = unp_pcb_lock_peer(unp); 889 if (unp2 != NULL) { 890 if (unp2->unp_addr != NULL) 891 sa = (struct sockaddr *)unp2->unp_addr; 892 else 893 sa = &sun_noname; 894 bcopy(sa, *nam, sa->sa_len); 895 unp_pcb_unlock_pair(unp, unp2); 896 } else { 897 sa = &sun_noname; 898 bcopy(sa, *nam, sa->sa_len); 899 UNP_PCB_UNLOCK(unp); 900 } 901 return (0); 902 } 903 904 static int 905 uipc_rcvd(struct socket *so, int flags) 906 { 907 struct unpcb *unp, *unp2; 908 struct socket *so2; 909 u_int mbcnt, sbcc; 910 911 unp = sotounpcb(so); 912 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 913 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 914 ("%s: socktype %d", __func__, so->so_type)); 915 916 /* 917 * Adjust backpressure on sender and wakeup any waiting to write. 918 * 919 * The unp lock is acquired to maintain the validity of the unp_conn 920 * pointer; no lock on unp2 is required as unp2->unp_socket will be 921 * static as long as we don't permit unp2 to disconnect from unp, 922 * which is prevented by the lock on unp. We cache values from 923 * so_rcv to avoid holding the so_rcv lock over the entire 924 * transaction on the remote so_snd. 925 */ 926 SOCKBUF_LOCK(&so->so_rcv); 927 mbcnt = so->so_rcv.sb_mbcnt; 928 sbcc = sbavail(&so->so_rcv); 929 SOCKBUF_UNLOCK(&so->so_rcv); 930 /* 931 * There is a benign race condition at this point. If we're planning to 932 * clear SB_STOP, but uipc_send is called on the connected socket at 933 * this instant, it might add data to the sockbuf and set SB_STOP. Then 934 * we would erroneously clear SB_STOP below, even though the sockbuf is 935 * full. The race is benign because the only ill effect is to allow the 936 * sockbuf to exceed its size limit, and the size limits are not 937 * strictly guaranteed anyway. 938 */ 939 UNP_PCB_LOCK(unp); 940 unp2 = unp->unp_conn; 941 if (unp2 == NULL) { 942 UNP_PCB_UNLOCK(unp); 943 return (0); 944 } 945 so2 = unp2->unp_socket; 946 SOCKBUF_LOCK(&so2->so_snd); 947 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 948 so2->so_snd.sb_flags &= ~SB_STOP; 949 sowwakeup_locked(so2); 950 UNP_PCB_UNLOCK(unp); 951 return (0); 952 } 953 954 static int 955 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 956 struct mbuf *control, struct thread *td) 957 { 958 struct unpcb *unp, *unp2; 959 struct socket *so2; 960 u_int mbcnt, sbcc; 961 int error; 962 963 unp = sotounpcb(so); 964 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 965 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 966 ("%s: socktype %d", __func__, so->so_type)); 967 968 error = 0; 969 if (flags & PRUS_OOB) { 970 error = EOPNOTSUPP; 971 goto release; 972 } 973 if (control != NULL && 974 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 975 goto release; 976 977 unp2 = NULL; 978 if ((so->so_state & SS_ISCONNECTED) == 0) { 979 if (nam != NULL) { 980 if ((error = unp_connect(so, nam, td)) != 0) 981 goto out; 982 } else { 983 error = ENOTCONN; 984 goto out; 985 } 986 } 987 988 UNP_PCB_LOCK(unp); 989 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 990 UNP_PCB_UNLOCK(unp); 991 error = ENOTCONN; 992 goto out; 993 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 994 unp_pcb_unlock_pair(unp, unp2); 995 error = EPIPE; 996 goto out; 997 } 998 UNP_PCB_UNLOCK(unp); 999 if ((so2 = unp2->unp_socket) == NULL) { 1000 UNP_PCB_UNLOCK(unp2); 1001 error = ENOTCONN; 1002 goto out; 1003 } 1004 SOCKBUF_LOCK(&so2->so_rcv); 1005 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 1006 /* 1007 * Credentials are passed only once on SOCK_STREAM and 1008 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 1009 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 1010 */ 1011 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 1012 NULL, NULL); 1013 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 1014 } 1015 1016 /* 1017 * Send to paired receive port and wake up readers. Don't 1018 * check for space available in the receive buffer if we're 1019 * attaching ancillary data; Unix domain sockets only check 1020 * for space in the sending sockbuf, and that check is 1021 * performed one level up the stack. At that level we cannot 1022 * precisely account for the amount of buffer space used 1023 * (e.g., because control messages are not yet internalized). 1024 */ 1025 switch (so->so_type) { 1026 case SOCK_STREAM: 1027 if (control != NULL) { 1028 sbappendcontrol_locked(&so2->so_rcv, m, 1029 control, flags); 1030 control = NULL; 1031 } else 1032 sbappend_locked(&so2->so_rcv, m, flags); 1033 break; 1034 1035 case SOCK_SEQPACKET: 1036 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1037 &sun_noname, m, control)) 1038 control = NULL; 1039 break; 1040 } 1041 1042 mbcnt = so2->so_rcv.sb_mbcnt; 1043 sbcc = sbavail(&so2->so_rcv); 1044 if (sbcc) 1045 sorwakeup_locked(so2); 1046 else 1047 SOCKBUF_UNLOCK(&so2->so_rcv); 1048 1049 /* 1050 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1051 * it would be possible for uipc_rcvd to be called at this 1052 * point, drain the receiving sockbuf, clear SB_STOP, and then 1053 * we would set SB_STOP below. That could lead to an empty 1054 * sockbuf having SB_STOP set 1055 */ 1056 SOCKBUF_LOCK(&so->so_snd); 1057 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1058 so->so_snd.sb_flags |= SB_STOP; 1059 SOCKBUF_UNLOCK(&so->so_snd); 1060 UNP_PCB_UNLOCK(unp2); 1061 m = NULL; 1062 out: 1063 /* 1064 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1065 */ 1066 if (flags & PRUS_EOF) { 1067 UNP_PCB_LOCK(unp); 1068 socantsendmore(so); 1069 unp_shutdown(unp); 1070 UNP_PCB_UNLOCK(unp); 1071 } 1072 if (control != NULL && error != 0) 1073 unp_scan(control, unp_freerights); 1074 1075 release: 1076 if (control != NULL) 1077 m_freem(control); 1078 /* 1079 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1080 * for freeing memory. 1081 */ 1082 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1083 m_freem(m); 1084 return (error); 1085 } 1086 1087 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1088 static inline bool 1089 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1090 { 1091 u_int bleft, mleft; 1092 1093 /* 1094 * Negative space may happen if send(2) is followed by 1095 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1096 */ 1097 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1098 sb->sb_mbmax < sb->uxdg_mbcnt)) 1099 return (false); 1100 1101 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1102 return (false); 1103 1104 bleft = sb->sb_hiwat - sb->uxdg_cc; 1105 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1106 1107 return (bleft >= cc && mleft >= mbcnt); 1108 } 1109 1110 /* 1111 * PF_UNIX/SOCK_DGRAM send 1112 * 1113 * Allocate a record consisting of 3 mbufs in the sequence of 1114 * from -> control -> data and append it to the socket buffer. 1115 * 1116 * The first mbuf carries sender's name and is a pkthdr that stores 1117 * overall length of datagram, its memory consumption and control length. 1118 */ 1119 #define ctllen PH_loc.thirtytwo[1] 1120 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1121 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1122 static int 1123 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1124 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1125 { 1126 struct unpcb *unp, *unp2; 1127 const struct sockaddr *from; 1128 struct socket *so2; 1129 struct sockbuf *sb; 1130 struct mbuf *f, *clast; 1131 u_int cc, ctl, mbcnt; 1132 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1133 int error; 1134 1135 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1136 1137 error = 0; 1138 f = NULL; 1139 ctl = 0; 1140 1141 if (__predict_false(flags & MSG_OOB)) { 1142 error = EOPNOTSUPP; 1143 goto out; 1144 } 1145 if (m == NULL) { 1146 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1147 error = EMSGSIZE; 1148 goto out; 1149 } 1150 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1151 if (__predict_false(m == NULL)) { 1152 error = EFAULT; 1153 goto out; 1154 } 1155 f = m_gethdr(M_WAITOK, MT_SONAME); 1156 cc = m->m_pkthdr.len; 1157 mbcnt = MSIZE + m->m_pkthdr.memlen; 1158 if (c != NULL && 1159 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1160 goto out; 1161 } else { 1162 /* pr_sosend() with mbuf usually is a kernel thread. */ 1163 1164 M_ASSERTPKTHDR(m); 1165 if (__predict_false(c != NULL)) 1166 panic("%s: control from a kernel thread", __func__); 1167 1168 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1169 error = EMSGSIZE; 1170 goto out; 1171 } 1172 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1173 error = ENOBUFS; 1174 goto out; 1175 } 1176 /* Condition the foreign mbuf to our standards. */ 1177 m_clrprotoflags(m); 1178 m_tag_delete_chain(m, NULL); 1179 m->m_pkthdr.rcvif = NULL; 1180 m->m_pkthdr.flowid = 0; 1181 m->m_pkthdr.csum_flags = 0; 1182 m->m_pkthdr.fibnum = 0; 1183 m->m_pkthdr.rsstype = 0; 1184 1185 cc = m->m_pkthdr.len; 1186 mbcnt = MSIZE; 1187 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1188 mbcnt += MSIZE; 1189 if (mb->m_flags & M_EXT) 1190 mbcnt += mb->m_ext.ext_size; 1191 } 1192 } 1193 1194 unp = sotounpcb(so); 1195 MPASS(unp); 1196 1197 /* 1198 * XXXGL: would be cool to fully remove so_snd out of the equation 1199 * and avoid this lock, which is not only extraneous, but also being 1200 * released, thus still leaving possibility for a race. We can easily 1201 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1202 * is more difficult to invent something to handle so_error. 1203 */ 1204 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1205 if (error) 1206 goto out2; 1207 SOCK_SENDBUF_LOCK(so); 1208 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1209 SOCK_SENDBUF_UNLOCK(so); 1210 error = EPIPE; 1211 goto out3; 1212 } 1213 if (so->so_error != 0) { 1214 error = so->so_error; 1215 so->so_error = 0; 1216 SOCK_SENDBUF_UNLOCK(so); 1217 goto out3; 1218 } 1219 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1220 SOCK_SENDBUF_UNLOCK(so); 1221 error = EDESTADDRREQ; 1222 goto out3; 1223 } 1224 SOCK_SENDBUF_UNLOCK(so); 1225 1226 if (addr != NULL) { 1227 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1228 goto out3; 1229 UNP_PCB_LOCK_ASSERT(unp); 1230 unp2 = unp->unp_conn; 1231 UNP_PCB_LOCK_ASSERT(unp2); 1232 } else { 1233 UNP_PCB_LOCK(unp); 1234 unp2 = unp_pcb_lock_peer(unp); 1235 if (unp2 == NULL) { 1236 UNP_PCB_UNLOCK(unp); 1237 error = ENOTCONN; 1238 goto out3; 1239 } 1240 } 1241 1242 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1243 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1244 &mbcnt); 1245 if (unp->unp_addr != NULL) 1246 from = (struct sockaddr *)unp->unp_addr; 1247 else 1248 from = &sun_noname; 1249 f->m_len = from->sa_len; 1250 MPASS(from->sa_len <= MLEN); 1251 bcopy(from, mtod(f, void *), from->sa_len); 1252 ctl += f->m_len; 1253 1254 /* 1255 * Concatenate mbufs: from -> control -> data. 1256 * Save overall cc and mbcnt in "from" mbuf. 1257 */ 1258 if (c != NULL) { 1259 #ifdef INVARIANTS 1260 struct mbuf *mc; 1261 1262 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1263 MPASS(mc == clast); 1264 #endif 1265 f->m_next = c; 1266 clast->m_next = m; 1267 c = NULL; 1268 } else 1269 f->m_next = m; 1270 m = NULL; 1271 #ifdef INVARIANTS 1272 dcc = dctl = dmbcnt = 0; 1273 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1274 if (mb->m_type == MT_DATA) 1275 dcc += mb->m_len; 1276 else 1277 dctl += mb->m_len; 1278 dmbcnt += MSIZE; 1279 if (mb->m_flags & M_EXT) 1280 dmbcnt += mb->m_ext.ext_size; 1281 } 1282 MPASS(dcc == cc); 1283 MPASS(dctl == ctl); 1284 MPASS(dmbcnt == mbcnt); 1285 #endif 1286 f->m_pkthdr.len = cc + ctl; 1287 f->m_pkthdr.memlen = mbcnt; 1288 f->m_pkthdr.ctllen = ctl; 1289 1290 /* 1291 * Destination socket buffer selection. 1292 * 1293 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1294 * destination address is supplied, create a temporary connection for 1295 * the run time of the function (see call to unp_connectat() above and 1296 * to unp_disconnect() below). We distinguish them by condition of 1297 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1298 * that condition, since, again, through the run time of this code we 1299 * are always connected. For such "unconnected" sends, the destination 1300 * buffer would be the receive buffer of destination socket so2. 1301 * 1302 * For connected sends, data lands on the send buffer of the sender's 1303 * socket "so". Then, if we just added the very first datagram 1304 * on this send buffer, we need to add the send buffer on to the 1305 * receiving socket's buffer list. We put ourselves on top of the 1306 * list. Such logic gives infrequent senders priority over frequent 1307 * senders. 1308 * 1309 * Note on byte count management. As long as event methods kevent(2), 1310 * select(2) are not protocol specific (yet), we need to maintain 1311 * meaningful values on the receive buffer. So, the receive buffer 1312 * would accumulate counters from all connected buffers potentially 1313 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1314 */ 1315 so2 = unp2->unp_socket; 1316 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1317 SOCK_RECVBUF_LOCK(so2); 1318 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1319 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1320 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1321 uxdg_clist); 1322 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1323 sb->uxdg_cc += cc + ctl; 1324 sb->uxdg_ctl += ctl; 1325 sb->uxdg_mbcnt += mbcnt; 1326 so2->so_rcv.sb_acc += cc + ctl; 1327 so2->so_rcv.sb_ccc += cc + ctl; 1328 so2->so_rcv.sb_ctl += ctl; 1329 so2->so_rcv.sb_mbcnt += mbcnt; 1330 sorwakeup_locked(so2); 1331 f = NULL; 1332 } else { 1333 soroverflow_locked(so2); 1334 error = ENOBUFS; 1335 if (f->m_next->m_type == MT_CONTROL) 1336 unp_scan(f->m_next, unp_freerights); 1337 } 1338 1339 if (addr != NULL) 1340 unp_disconnect(unp, unp2); 1341 else 1342 unp_pcb_unlock_pair(unp, unp2); 1343 1344 td->td_ru.ru_msgsnd++; 1345 1346 out3: 1347 SOCK_IO_SEND_UNLOCK(so); 1348 out2: 1349 if (c) 1350 unp_scan(c, unp_freerights); 1351 out: 1352 if (f) 1353 m_freem(f); 1354 if (c) 1355 m_freem(c); 1356 if (m) 1357 m_freem(m); 1358 1359 return (error); 1360 } 1361 1362 /* 1363 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1364 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1365 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1366 */ 1367 static int 1368 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1369 struct uio *uio, struct mbuf **controlp, int *flagsp) 1370 { 1371 ssize_t len = 0; 1372 int error; 1373 1374 so->so_rcv.uxdg_peeked = m; 1375 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1376 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1377 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1378 SOCK_RECVBUF_UNLOCK(so); 1379 1380 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1381 if (psa != NULL) 1382 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1383 1384 m = m->m_next; 1385 KASSERT(m, ("%s: no data or control after soname", __func__)); 1386 1387 /* 1388 * With MSG_PEEK the control isn't executed, just copied. 1389 */ 1390 while (m != NULL && m->m_type == MT_CONTROL) { 1391 if (controlp != NULL) { 1392 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1393 controlp = &(*controlp)->m_next; 1394 } 1395 m = m->m_next; 1396 } 1397 KASSERT(m == NULL || m->m_type == MT_DATA, 1398 ("%s: not MT_DATA mbuf %p", __func__, m)); 1399 while (m != NULL && uio->uio_resid > 0) { 1400 len = uio->uio_resid; 1401 if (len > m->m_len) 1402 len = m->m_len; 1403 error = uiomove(mtod(m, char *), (int)len, uio); 1404 if (error) { 1405 SOCK_IO_RECV_UNLOCK(so); 1406 return (error); 1407 } 1408 if (len == m->m_len) 1409 m = m->m_next; 1410 } 1411 SOCK_IO_RECV_UNLOCK(so); 1412 1413 if (flagsp != NULL) { 1414 if (m != NULL) { 1415 if (*flagsp & MSG_TRUNC) { 1416 /* Report real length of the packet */ 1417 uio->uio_resid -= m_length(m, NULL) - len; 1418 } 1419 *flagsp |= MSG_TRUNC; 1420 } else 1421 *flagsp &= ~MSG_TRUNC; 1422 } 1423 1424 return (0); 1425 } 1426 1427 /* 1428 * PF_UNIX/SOCK_DGRAM receive 1429 */ 1430 static int 1431 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1432 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1433 { 1434 struct sockbuf *sb = NULL; 1435 struct mbuf *m; 1436 int flags, error; 1437 ssize_t len = 0; 1438 bool nonblock; 1439 1440 MPASS(mp0 == NULL); 1441 1442 if (psa != NULL) 1443 *psa = NULL; 1444 if (controlp != NULL) 1445 *controlp = NULL; 1446 1447 flags = flagsp != NULL ? *flagsp : 0; 1448 nonblock = (so->so_state & SS_NBIO) || 1449 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1450 1451 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1452 if (__predict_false(error)) 1453 return (error); 1454 1455 /* 1456 * Loop blocking while waiting for a datagram. Prioritize connected 1457 * peers over unconnected sends. Set sb to selected socket buffer 1458 * containing an mbuf on exit from the wait loop. A datagram that 1459 * had already been peeked at has top priority. 1460 */ 1461 SOCK_RECVBUF_LOCK(so); 1462 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1463 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1464 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1465 if (so->so_error) { 1466 error = so->so_error; 1467 so->so_error = 0; 1468 SOCK_RECVBUF_UNLOCK(so); 1469 SOCK_IO_RECV_UNLOCK(so); 1470 return (error); 1471 } 1472 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1473 uio->uio_resid == 0) { 1474 SOCK_RECVBUF_UNLOCK(so); 1475 SOCK_IO_RECV_UNLOCK(so); 1476 return (0); 1477 } 1478 if (nonblock) { 1479 SOCK_RECVBUF_UNLOCK(so); 1480 SOCK_IO_RECV_UNLOCK(so); 1481 return (EWOULDBLOCK); 1482 } 1483 error = sbwait(so, SO_RCV); 1484 if (error) { 1485 SOCK_RECVBUF_UNLOCK(so); 1486 SOCK_IO_RECV_UNLOCK(so); 1487 return (error); 1488 } 1489 } 1490 1491 if (sb == NULL) 1492 sb = &so->so_rcv; 1493 else if (m == NULL) 1494 m = STAILQ_FIRST(&sb->uxdg_mb); 1495 else 1496 MPASS(m == so->so_rcv.uxdg_peeked); 1497 1498 MPASS(sb->uxdg_cc > 0); 1499 M_ASSERTPKTHDR(m); 1500 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1501 1502 if (uio->uio_td) 1503 uio->uio_td->td_ru.ru_msgrcv++; 1504 1505 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1506 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1507 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1508 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1509 } else 1510 so->so_rcv.uxdg_peeked = NULL; 1511 1512 sb->uxdg_cc -= m->m_pkthdr.len; 1513 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1514 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1515 1516 if (__predict_false(flags & MSG_PEEK)) 1517 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1518 1519 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1520 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1521 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1522 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1523 SOCK_RECVBUF_UNLOCK(so); 1524 1525 if (psa != NULL) 1526 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1527 m = m_free(m); 1528 KASSERT(m, ("%s: no data or control after soname", __func__)); 1529 1530 /* 1531 * Packet to copyout() is now in 'm' and it is disconnected from the 1532 * queue. 1533 * 1534 * Process one or more MT_CONTROL mbufs present before any data mbufs 1535 * in the first mbuf chain on the socket buffer. We call into the 1536 * unp_externalize() to perform externalization (or freeing if 1537 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1538 * without MT_DATA mbufs. 1539 */ 1540 while (m != NULL && m->m_type == MT_CONTROL) { 1541 struct mbuf *cm; 1542 1543 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1544 * it frees whole chain, so we must disconnect the mbuf. 1545 */ 1546 cm = m; m = m->m_next; cm->m_next = NULL; 1547 error = unp_externalize(cm, controlp, flags); 1548 if (error != 0) { 1549 SOCK_IO_RECV_UNLOCK(so); 1550 unp_scan(m, unp_freerights); 1551 m_freem(m); 1552 return (error); 1553 } 1554 if (controlp != NULL) { 1555 while (*controlp != NULL) 1556 controlp = &(*controlp)->m_next; 1557 } 1558 } 1559 KASSERT(m == NULL || m->m_type == MT_DATA, 1560 ("%s: not MT_DATA mbuf %p", __func__, m)); 1561 while (m != NULL && uio->uio_resid > 0) { 1562 len = uio->uio_resid; 1563 if (len > m->m_len) 1564 len = m->m_len; 1565 error = uiomove(mtod(m, char *), (int)len, uio); 1566 if (error) { 1567 SOCK_IO_RECV_UNLOCK(so); 1568 m_freem(m); 1569 return (error); 1570 } 1571 if (len == m->m_len) 1572 m = m_free(m); 1573 else { 1574 m->m_data += len; 1575 m->m_len -= len; 1576 } 1577 } 1578 SOCK_IO_RECV_UNLOCK(so); 1579 1580 if (m != NULL) { 1581 if (flagsp != NULL) { 1582 if (flags & MSG_TRUNC) { 1583 /* Report real length of the packet */ 1584 uio->uio_resid -= m_length(m, NULL); 1585 } 1586 *flagsp |= MSG_TRUNC; 1587 } 1588 m_freem(m); 1589 } else if (flagsp != NULL) 1590 *flagsp &= ~MSG_TRUNC; 1591 1592 return (0); 1593 } 1594 1595 static bool 1596 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1597 { 1598 struct mbuf *mb, *n; 1599 struct sockbuf *sb; 1600 1601 SOCK_LOCK(so); 1602 if (SOLISTENING(so)) { 1603 SOCK_UNLOCK(so); 1604 return (false); 1605 } 1606 mb = NULL; 1607 sb = &so->so_rcv; 1608 SOCKBUF_LOCK(sb); 1609 if (sb->sb_fnrdy != NULL) { 1610 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1611 if (mb == m) { 1612 *errorp = sbready(sb, m, count); 1613 break; 1614 } 1615 mb = mb->m_next; 1616 if (mb == NULL) { 1617 mb = n; 1618 if (mb != NULL) 1619 n = mb->m_nextpkt; 1620 } 1621 } 1622 } 1623 SOCKBUF_UNLOCK(sb); 1624 SOCK_UNLOCK(so); 1625 return (mb != NULL); 1626 } 1627 1628 static int 1629 uipc_ready(struct socket *so, struct mbuf *m, int count) 1630 { 1631 struct unpcb *unp, *unp2; 1632 struct socket *so2; 1633 int error, i; 1634 1635 unp = sotounpcb(so); 1636 1637 KASSERT(so->so_type == SOCK_STREAM, 1638 ("%s: unexpected socket type for %p", __func__, so)); 1639 1640 UNP_PCB_LOCK(unp); 1641 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1642 UNP_PCB_UNLOCK(unp); 1643 so2 = unp2->unp_socket; 1644 SOCKBUF_LOCK(&so2->so_rcv); 1645 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1646 sorwakeup_locked(so2); 1647 else 1648 SOCKBUF_UNLOCK(&so2->so_rcv); 1649 UNP_PCB_UNLOCK(unp2); 1650 return (error); 1651 } 1652 UNP_PCB_UNLOCK(unp); 1653 1654 /* 1655 * The receiving socket has been disconnected, but may still be valid. 1656 * In this case, the now-ready mbufs are still present in its socket 1657 * buffer, so perform an exhaustive search before giving up and freeing 1658 * the mbufs. 1659 */ 1660 UNP_LINK_RLOCK(); 1661 LIST_FOREACH(unp, &unp_shead, unp_link) { 1662 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1663 break; 1664 } 1665 UNP_LINK_RUNLOCK(); 1666 1667 if (unp == NULL) { 1668 for (i = 0; i < count; i++) 1669 m = m_free(m); 1670 error = ECONNRESET; 1671 } 1672 return (error); 1673 } 1674 1675 static int 1676 uipc_sense(struct socket *so, struct stat *sb) 1677 { 1678 struct unpcb *unp; 1679 1680 unp = sotounpcb(so); 1681 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1682 1683 sb->st_blksize = so->so_snd.sb_hiwat; 1684 sb->st_dev = NODEV; 1685 sb->st_ino = unp->unp_ino; 1686 return (0); 1687 } 1688 1689 static int 1690 uipc_shutdown(struct socket *so) 1691 { 1692 struct unpcb *unp; 1693 1694 unp = sotounpcb(so); 1695 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1696 1697 UNP_PCB_LOCK(unp); 1698 socantsendmore(so); 1699 unp_shutdown(unp); 1700 UNP_PCB_UNLOCK(unp); 1701 return (0); 1702 } 1703 1704 static int 1705 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1706 { 1707 struct unpcb *unp; 1708 const struct sockaddr *sa; 1709 1710 unp = sotounpcb(so); 1711 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1712 1713 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1714 UNP_PCB_LOCK(unp); 1715 if (unp->unp_addr != NULL) 1716 sa = (struct sockaddr *) unp->unp_addr; 1717 else 1718 sa = &sun_noname; 1719 bcopy(sa, *nam, sa->sa_len); 1720 UNP_PCB_UNLOCK(unp); 1721 return (0); 1722 } 1723 1724 static int 1725 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1726 { 1727 struct unpcb *unp; 1728 struct xucred xu; 1729 int error, optval; 1730 1731 if (sopt->sopt_level != SOL_LOCAL) 1732 return (EINVAL); 1733 1734 unp = sotounpcb(so); 1735 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1736 error = 0; 1737 switch (sopt->sopt_dir) { 1738 case SOPT_GET: 1739 switch (sopt->sopt_name) { 1740 case LOCAL_PEERCRED: 1741 UNP_PCB_LOCK(unp); 1742 if (unp->unp_flags & UNP_HAVEPC) 1743 xu = unp->unp_peercred; 1744 else { 1745 if (so->so_type == SOCK_STREAM) 1746 error = ENOTCONN; 1747 else 1748 error = EINVAL; 1749 } 1750 UNP_PCB_UNLOCK(unp); 1751 if (error == 0) 1752 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1753 break; 1754 1755 case LOCAL_CREDS: 1756 /* Unlocked read. */ 1757 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1758 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1759 break; 1760 1761 case LOCAL_CREDS_PERSISTENT: 1762 /* Unlocked read. */ 1763 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1764 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1765 break; 1766 1767 case LOCAL_CONNWAIT: 1768 /* Unlocked read. */ 1769 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1770 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1771 break; 1772 1773 default: 1774 error = EOPNOTSUPP; 1775 break; 1776 } 1777 break; 1778 1779 case SOPT_SET: 1780 switch (sopt->sopt_name) { 1781 case LOCAL_CREDS: 1782 case LOCAL_CREDS_PERSISTENT: 1783 case LOCAL_CONNWAIT: 1784 error = sooptcopyin(sopt, &optval, sizeof(optval), 1785 sizeof(optval)); 1786 if (error) 1787 break; 1788 1789 #define OPTSET(bit, exclusive) do { \ 1790 UNP_PCB_LOCK(unp); \ 1791 if (optval) { \ 1792 if ((unp->unp_flags & (exclusive)) != 0) { \ 1793 UNP_PCB_UNLOCK(unp); \ 1794 error = EINVAL; \ 1795 break; \ 1796 } \ 1797 unp->unp_flags |= (bit); \ 1798 } else \ 1799 unp->unp_flags &= ~(bit); \ 1800 UNP_PCB_UNLOCK(unp); \ 1801 } while (0) 1802 1803 switch (sopt->sopt_name) { 1804 case LOCAL_CREDS: 1805 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1806 break; 1807 1808 case LOCAL_CREDS_PERSISTENT: 1809 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1810 break; 1811 1812 case LOCAL_CONNWAIT: 1813 OPTSET(UNP_CONNWAIT, 0); 1814 break; 1815 1816 default: 1817 break; 1818 } 1819 break; 1820 #undef OPTSET 1821 default: 1822 error = ENOPROTOOPT; 1823 break; 1824 } 1825 break; 1826 1827 default: 1828 error = EOPNOTSUPP; 1829 break; 1830 } 1831 return (error); 1832 } 1833 1834 static int 1835 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1836 { 1837 1838 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1839 } 1840 1841 static int 1842 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1843 struct thread *td, bool return_locked) 1844 { 1845 struct mtx *vplock; 1846 struct sockaddr_un *soun; 1847 struct vnode *vp; 1848 struct socket *so2; 1849 struct unpcb *unp, *unp2, *unp3; 1850 struct nameidata nd; 1851 char buf[SOCK_MAXADDRLEN]; 1852 struct sockaddr *sa; 1853 cap_rights_t rights; 1854 int error, len; 1855 bool connreq; 1856 1857 if (nam->sa_family != AF_UNIX) 1858 return (EAFNOSUPPORT); 1859 if (nam->sa_len > sizeof(struct sockaddr_un)) 1860 return (EINVAL); 1861 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1862 if (len <= 0) 1863 return (EINVAL); 1864 soun = (struct sockaddr_un *)nam; 1865 bcopy(soun->sun_path, buf, len); 1866 buf[len] = 0; 1867 1868 error = 0; 1869 unp = sotounpcb(so); 1870 UNP_PCB_LOCK(unp); 1871 for (;;) { 1872 /* 1873 * Wait for connection state to stabilize. If a connection 1874 * already exists, give up. For datagram sockets, which permit 1875 * multiple consecutive connect(2) calls, upper layers are 1876 * responsible for disconnecting in advance of a subsequent 1877 * connect(2), but this is not synchronized with PCB connection 1878 * state. 1879 * 1880 * Also make sure that no threads are currently attempting to 1881 * lock the peer socket, to ensure that unp_conn cannot 1882 * transition between two valid sockets while locks are dropped. 1883 */ 1884 if (SOLISTENING(so)) 1885 error = EOPNOTSUPP; 1886 else if (unp->unp_conn != NULL) 1887 error = EISCONN; 1888 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1889 error = EALREADY; 1890 } 1891 if (error != 0) { 1892 UNP_PCB_UNLOCK(unp); 1893 return (error); 1894 } 1895 if (unp->unp_pairbusy > 0) { 1896 unp->unp_flags |= UNP_WAITING; 1897 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1898 continue; 1899 } 1900 break; 1901 } 1902 unp->unp_flags |= UNP_CONNECTING; 1903 UNP_PCB_UNLOCK(unp); 1904 1905 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1906 if (connreq) 1907 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1908 else 1909 sa = NULL; 1910 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1911 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1912 error = namei(&nd); 1913 if (error) 1914 vp = NULL; 1915 else 1916 vp = nd.ni_vp; 1917 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1918 if (error) 1919 goto bad; 1920 NDFREE_PNBUF(&nd); 1921 1922 if (vp->v_type != VSOCK) { 1923 error = ENOTSOCK; 1924 goto bad; 1925 } 1926 #ifdef MAC 1927 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1928 if (error) 1929 goto bad; 1930 #endif 1931 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1932 if (error) 1933 goto bad; 1934 1935 unp = sotounpcb(so); 1936 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1937 1938 vplock = mtx_pool_find(mtxpool_sleep, vp); 1939 mtx_lock(vplock); 1940 VOP_UNP_CONNECT(vp, &unp2); 1941 if (unp2 == NULL) { 1942 error = ECONNREFUSED; 1943 goto bad2; 1944 } 1945 so2 = unp2->unp_socket; 1946 if (so->so_type != so2->so_type) { 1947 error = EPROTOTYPE; 1948 goto bad2; 1949 } 1950 if (connreq) { 1951 if (SOLISTENING(so2)) { 1952 CURVNET_SET(so2->so_vnet); 1953 so2 = sonewconn(so2, 0); 1954 CURVNET_RESTORE(); 1955 } else 1956 so2 = NULL; 1957 if (so2 == NULL) { 1958 error = ECONNREFUSED; 1959 goto bad2; 1960 } 1961 unp3 = sotounpcb(so2); 1962 unp_pcb_lock_pair(unp2, unp3); 1963 if (unp2->unp_addr != NULL) { 1964 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1965 unp3->unp_addr = (struct sockaddr_un *) sa; 1966 sa = NULL; 1967 } 1968 1969 unp_copy_peercred(td, unp3, unp, unp2); 1970 1971 UNP_PCB_UNLOCK(unp2); 1972 unp2 = unp3; 1973 1974 /* 1975 * It is safe to block on the PCB lock here since unp2 is 1976 * nascent and cannot be connected to any other sockets. 1977 */ 1978 UNP_PCB_LOCK(unp); 1979 #ifdef MAC 1980 mac_socketpeer_set_from_socket(so, so2); 1981 mac_socketpeer_set_from_socket(so2, so); 1982 #endif 1983 } else { 1984 unp_pcb_lock_pair(unp, unp2); 1985 } 1986 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1987 sotounpcb(so2) == unp2, 1988 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1989 unp_connect2(so, so2, PRU_CONNECT); 1990 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1991 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1992 unp->unp_flags &= ~UNP_CONNECTING; 1993 if (!return_locked) 1994 unp_pcb_unlock_pair(unp, unp2); 1995 bad2: 1996 mtx_unlock(vplock); 1997 bad: 1998 if (vp != NULL) { 1999 /* 2000 * If we are returning locked (called via uipc_sosend_dgram()), 2001 * we need to be sure that vput() won't sleep. This is 2002 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2003 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2004 */ 2005 MPASS(!(return_locked && connreq)); 2006 vput(vp); 2007 } 2008 free(sa, M_SONAME); 2009 if (__predict_false(error)) { 2010 UNP_PCB_LOCK(unp); 2011 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2012 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2013 unp->unp_flags &= ~UNP_CONNECTING; 2014 UNP_PCB_UNLOCK(unp); 2015 } 2016 return (error); 2017 } 2018 2019 /* 2020 * Set socket peer credentials at connection time. 2021 * 2022 * The client's PCB credentials are copied from its process structure. The 2023 * server's PCB credentials are copied from the socket on which it called 2024 * listen(2). uipc_listen cached that process's credentials at the time. 2025 */ 2026 void 2027 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2028 struct unpcb *server_unp, struct unpcb *listen_unp) 2029 { 2030 cru2xt(td, &client_unp->unp_peercred); 2031 client_unp->unp_flags |= UNP_HAVEPC; 2032 2033 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2034 sizeof(server_unp->unp_peercred)); 2035 server_unp->unp_flags |= UNP_HAVEPC; 2036 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2037 } 2038 2039 static void 2040 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2041 { 2042 struct unpcb *unp; 2043 struct unpcb *unp2; 2044 2045 MPASS(so2->so_type == so->so_type); 2046 unp = sotounpcb(so); 2047 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2048 unp2 = sotounpcb(so2); 2049 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2050 2051 UNP_PCB_LOCK_ASSERT(unp); 2052 UNP_PCB_LOCK_ASSERT(unp2); 2053 KASSERT(unp->unp_conn == NULL, 2054 ("%s: socket %p is already connected", __func__, unp)); 2055 2056 unp->unp_conn = unp2; 2057 unp_pcb_hold(unp2); 2058 unp_pcb_hold(unp); 2059 switch (so->so_type) { 2060 case SOCK_DGRAM: 2061 UNP_REF_LIST_LOCK(); 2062 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2063 UNP_REF_LIST_UNLOCK(); 2064 soisconnected(so); 2065 break; 2066 2067 case SOCK_STREAM: 2068 case SOCK_SEQPACKET: 2069 KASSERT(unp2->unp_conn == NULL, 2070 ("%s: socket %p is already connected", __func__, unp2)); 2071 unp2->unp_conn = unp; 2072 if (req == PRU_CONNECT && 2073 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2074 soisconnecting(so); 2075 else 2076 soisconnected(so); 2077 soisconnected(so2); 2078 break; 2079 2080 default: 2081 panic("unp_connect2"); 2082 } 2083 } 2084 2085 static void 2086 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2087 { 2088 struct socket *so, *so2; 2089 struct mbuf *m = NULL; 2090 #ifdef INVARIANTS 2091 struct unpcb *unptmp; 2092 #endif 2093 2094 UNP_PCB_LOCK_ASSERT(unp); 2095 UNP_PCB_LOCK_ASSERT(unp2); 2096 KASSERT(unp->unp_conn == unp2, 2097 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2098 2099 unp->unp_conn = NULL; 2100 so = unp->unp_socket; 2101 so2 = unp2->unp_socket; 2102 switch (unp->unp_socket->so_type) { 2103 case SOCK_DGRAM: 2104 /* 2105 * Remove our send socket buffer from the peer's receive buffer. 2106 * Move the data to the receive buffer only if it is empty. 2107 * This is a protection against a scenario where a peer 2108 * connects, floods and disconnects, effectively blocking 2109 * sendto() from unconnected sockets. 2110 */ 2111 SOCK_RECVBUF_LOCK(so2); 2112 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2113 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2114 uxdg_clist); 2115 if (__predict_true((so2->so_rcv.sb_state & 2116 SBS_CANTRCVMORE) == 0) && 2117 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2118 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2119 &so->so_snd.uxdg_mb); 2120 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2121 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2122 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2123 } else { 2124 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2125 STAILQ_INIT(&so->so_snd.uxdg_mb); 2126 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2127 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2128 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2129 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2130 } 2131 /* Note: so may reconnect. */ 2132 so->so_snd.uxdg_cc = 0; 2133 so->so_snd.uxdg_ctl = 0; 2134 so->so_snd.uxdg_mbcnt = 0; 2135 } 2136 SOCK_RECVBUF_UNLOCK(so2); 2137 UNP_REF_LIST_LOCK(); 2138 #ifdef INVARIANTS 2139 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2140 if (unptmp == unp) 2141 break; 2142 } 2143 KASSERT(unptmp != NULL, 2144 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2145 #endif 2146 LIST_REMOVE(unp, unp_reflink); 2147 UNP_REF_LIST_UNLOCK(); 2148 if (so) { 2149 SOCK_LOCK(so); 2150 so->so_state &= ~SS_ISCONNECTED; 2151 SOCK_UNLOCK(so); 2152 } 2153 break; 2154 2155 case SOCK_STREAM: 2156 case SOCK_SEQPACKET: 2157 if (so) 2158 soisdisconnected(so); 2159 MPASS(unp2->unp_conn == unp); 2160 unp2->unp_conn = NULL; 2161 if (so2) 2162 soisdisconnected(so2); 2163 break; 2164 } 2165 2166 if (unp == unp2) { 2167 unp_pcb_rele_notlast(unp); 2168 if (!unp_pcb_rele(unp)) 2169 UNP_PCB_UNLOCK(unp); 2170 } else { 2171 if (!unp_pcb_rele(unp)) 2172 UNP_PCB_UNLOCK(unp); 2173 if (!unp_pcb_rele(unp2)) 2174 UNP_PCB_UNLOCK(unp2); 2175 } 2176 2177 if (m != NULL) { 2178 unp_scan(m, unp_freerights); 2179 m_freem(m); 2180 } 2181 } 2182 2183 /* 2184 * unp_pcblist() walks the global list of struct unpcb's to generate a 2185 * pointer list, bumping the refcount on each unpcb. It then copies them out 2186 * sequentially, validating the generation number on each to see if it has 2187 * been detached. All of this is necessary because copyout() may sleep on 2188 * disk I/O. 2189 */ 2190 static int 2191 unp_pcblist(SYSCTL_HANDLER_ARGS) 2192 { 2193 struct unpcb *unp, **unp_list; 2194 unp_gen_t gencnt; 2195 struct xunpgen *xug; 2196 struct unp_head *head; 2197 struct xunpcb *xu; 2198 u_int i; 2199 int error, n; 2200 2201 switch ((intptr_t)arg1) { 2202 case SOCK_STREAM: 2203 head = &unp_shead; 2204 break; 2205 2206 case SOCK_DGRAM: 2207 head = &unp_dhead; 2208 break; 2209 2210 case SOCK_SEQPACKET: 2211 head = &unp_sphead; 2212 break; 2213 2214 default: 2215 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2216 } 2217 2218 /* 2219 * The process of preparing the PCB list is too time-consuming and 2220 * resource-intensive to repeat twice on every request. 2221 */ 2222 if (req->oldptr == NULL) { 2223 n = unp_count; 2224 req->oldidx = 2 * (sizeof *xug) 2225 + (n + n/8) * sizeof(struct xunpcb); 2226 return (0); 2227 } 2228 2229 if (req->newptr != NULL) 2230 return (EPERM); 2231 2232 /* 2233 * OK, now we're committed to doing something. 2234 */ 2235 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2236 UNP_LINK_RLOCK(); 2237 gencnt = unp_gencnt; 2238 n = unp_count; 2239 UNP_LINK_RUNLOCK(); 2240 2241 xug->xug_len = sizeof *xug; 2242 xug->xug_count = n; 2243 xug->xug_gen = gencnt; 2244 xug->xug_sogen = so_gencnt; 2245 error = SYSCTL_OUT(req, xug, sizeof *xug); 2246 if (error) { 2247 free(xug, M_TEMP); 2248 return (error); 2249 } 2250 2251 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2252 2253 UNP_LINK_RLOCK(); 2254 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2255 unp = LIST_NEXT(unp, unp_link)) { 2256 UNP_PCB_LOCK(unp); 2257 if (unp->unp_gencnt <= gencnt) { 2258 if (cr_cansee(req->td->td_ucred, 2259 unp->unp_socket->so_cred)) { 2260 UNP_PCB_UNLOCK(unp); 2261 continue; 2262 } 2263 unp_list[i++] = unp; 2264 unp_pcb_hold(unp); 2265 } 2266 UNP_PCB_UNLOCK(unp); 2267 } 2268 UNP_LINK_RUNLOCK(); 2269 n = i; /* In case we lost some during malloc. */ 2270 2271 error = 0; 2272 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2273 for (i = 0; i < n; i++) { 2274 unp = unp_list[i]; 2275 UNP_PCB_LOCK(unp); 2276 if (unp_pcb_rele(unp)) 2277 continue; 2278 2279 if (unp->unp_gencnt <= gencnt) { 2280 xu->xu_len = sizeof *xu; 2281 xu->xu_unpp = (uintptr_t)unp; 2282 /* 2283 * XXX - need more locking here to protect against 2284 * connect/disconnect races for SMP. 2285 */ 2286 if (unp->unp_addr != NULL) 2287 bcopy(unp->unp_addr, &xu->xu_addr, 2288 unp->unp_addr->sun_len); 2289 else 2290 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2291 if (unp->unp_conn != NULL && 2292 unp->unp_conn->unp_addr != NULL) 2293 bcopy(unp->unp_conn->unp_addr, 2294 &xu->xu_caddr, 2295 unp->unp_conn->unp_addr->sun_len); 2296 else 2297 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2298 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2299 xu->unp_conn = (uintptr_t)unp->unp_conn; 2300 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2301 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2302 xu->unp_gencnt = unp->unp_gencnt; 2303 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2304 UNP_PCB_UNLOCK(unp); 2305 error = SYSCTL_OUT(req, xu, sizeof *xu); 2306 } else { 2307 UNP_PCB_UNLOCK(unp); 2308 } 2309 } 2310 free(xu, M_TEMP); 2311 if (!error) { 2312 /* 2313 * Give the user an updated idea of our state. If the 2314 * generation differs from what we told her before, she knows 2315 * that something happened while we were processing this 2316 * request, and it might be necessary to retry. 2317 */ 2318 xug->xug_gen = unp_gencnt; 2319 xug->xug_sogen = so_gencnt; 2320 xug->xug_count = unp_count; 2321 error = SYSCTL_OUT(req, xug, sizeof *xug); 2322 } 2323 free(unp_list, M_TEMP); 2324 free(xug, M_TEMP); 2325 return (error); 2326 } 2327 2328 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2329 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2330 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2331 "List of active local datagram sockets"); 2332 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2333 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2334 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2335 "List of active local stream sockets"); 2336 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2337 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2338 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2339 "List of active local seqpacket sockets"); 2340 2341 static void 2342 unp_shutdown(struct unpcb *unp) 2343 { 2344 struct unpcb *unp2; 2345 struct socket *so; 2346 2347 UNP_PCB_LOCK_ASSERT(unp); 2348 2349 unp2 = unp->unp_conn; 2350 if ((unp->unp_socket->so_type == SOCK_STREAM || 2351 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2352 so = unp2->unp_socket; 2353 if (so != NULL) 2354 socantrcvmore(so); 2355 } 2356 } 2357 2358 static void 2359 unp_drop(struct unpcb *unp) 2360 { 2361 struct socket *so; 2362 struct unpcb *unp2; 2363 2364 /* 2365 * Regardless of whether the socket's peer dropped the connection 2366 * with this socket by aborting or disconnecting, POSIX requires 2367 * that ECONNRESET is returned. 2368 */ 2369 2370 UNP_PCB_LOCK(unp); 2371 so = unp->unp_socket; 2372 if (so) 2373 so->so_error = ECONNRESET; 2374 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2375 /* Last reference dropped in unp_disconnect(). */ 2376 unp_pcb_rele_notlast(unp); 2377 unp_disconnect(unp, unp2); 2378 } else if (!unp_pcb_rele(unp)) { 2379 UNP_PCB_UNLOCK(unp); 2380 } 2381 } 2382 2383 static void 2384 unp_freerights(struct filedescent **fdep, int fdcount) 2385 { 2386 struct file *fp; 2387 int i; 2388 2389 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2390 2391 for (i = 0; i < fdcount; i++) { 2392 fp = fdep[i]->fde_file; 2393 filecaps_free(&fdep[i]->fde_caps); 2394 unp_discard(fp); 2395 } 2396 free(fdep[0], M_FILECAPS); 2397 } 2398 2399 static int 2400 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2401 { 2402 struct thread *td = curthread; /* XXX */ 2403 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2404 int i; 2405 int *fdp; 2406 struct filedesc *fdesc = td->td_proc->p_fd; 2407 struct filedescent **fdep; 2408 void *data; 2409 socklen_t clen = control->m_len, datalen; 2410 int error, newfds; 2411 u_int newlen; 2412 2413 UNP_LINK_UNLOCK_ASSERT(); 2414 2415 error = 0; 2416 if (controlp != NULL) /* controlp == NULL => free control messages */ 2417 *controlp = NULL; 2418 while (cm != NULL) { 2419 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2420 2421 data = CMSG_DATA(cm); 2422 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2423 if (cm->cmsg_level == SOL_SOCKET 2424 && cm->cmsg_type == SCM_RIGHTS) { 2425 newfds = datalen / sizeof(*fdep); 2426 if (newfds == 0) 2427 goto next; 2428 fdep = data; 2429 2430 /* If we're not outputting the descriptors free them. */ 2431 if (error || controlp == NULL) { 2432 unp_freerights(fdep, newfds); 2433 goto next; 2434 } 2435 FILEDESC_XLOCK(fdesc); 2436 2437 /* 2438 * Now change each pointer to an fd in the global 2439 * table to an integer that is the index to the local 2440 * fd table entry that we set up to point to the 2441 * global one we are transferring. 2442 */ 2443 newlen = newfds * sizeof(int); 2444 *controlp = sbcreatecontrol(NULL, newlen, 2445 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2446 2447 fdp = (int *) 2448 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2449 if ((error = fdallocn(td, 0, fdp, newfds))) { 2450 FILEDESC_XUNLOCK(fdesc); 2451 unp_freerights(fdep, newfds); 2452 m_freem(*controlp); 2453 *controlp = NULL; 2454 goto next; 2455 } 2456 for (i = 0; i < newfds; i++, fdp++) { 2457 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2458 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2459 &fdep[i]->fde_caps); 2460 unp_externalize_fp(fdep[i]->fde_file); 2461 } 2462 2463 /* 2464 * The new type indicates that the mbuf data refers to 2465 * kernel resources that may need to be released before 2466 * the mbuf is freed. 2467 */ 2468 m_chtype(*controlp, MT_EXTCONTROL); 2469 FILEDESC_XUNLOCK(fdesc); 2470 free(fdep[0], M_FILECAPS); 2471 } else { 2472 /* We can just copy anything else across. */ 2473 if (error || controlp == NULL) 2474 goto next; 2475 *controlp = sbcreatecontrol(NULL, datalen, 2476 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2477 bcopy(data, 2478 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2479 datalen); 2480 } 2481 controlp = &(*controlp)->m_next; 2482 2483 next: 2484 if (CMSG_SPACE(datalen) < clen) { 2485 clen -= CMSG_SPACE(datalen); 2486 cm = (struct cmsghdr *) 2487 ((caddr_t)cm + CMSG_SPACE(datalen)); 2488 } else { 2489 clen = 0; 2490 cm = NULL; 2491 } 2492 } 2493 2494 m_freem(control); 2495 return (error); 2496 } 2497 2498 static void 2499 unp_zone_change(void *tag) 2500 { 2501 2502 uma_zone_set_max(unp_zone, maxsockets); 2503 } 2504 2505 #ifdef INVARIANTS 2506 static void 2507 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2508 { 2509 struct unpcb *unp; 2510 2511 unp = mem; 2512 2513 KASSERT(LIST_EMPTY(&unp->unp_refs), 2514 ("%s: unpcb %p has lingering refs", __func__, unp)); 2515 KASSERT(unp->unp_socket == NULL, 2516 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2517 KASSERT(unp->unp_vnode == NULL, 2518 ("%s: unpcb %p has vnode references", __func__, unp)); 2519 KASSERT(unp->unp_conn == NULL, 2520 ("%s: unpcb %p is still connected", __func__, unp)); 2521 KASSERT(unp->unp_addr == NULL, 2522 ("%s: unpcb %p has leaked addr", __func__, unp)); 2523 } 2524 #endif 2525 2526 static void 2527 unp_init(void *arg __unused) 2528 { 2529 uma_dtor dtor; 2530 2531 #ifdef INVARIANTS 2532 dtor = unp_zdtor; 2533 #else 2534 dtor = NULL; 2535 #endif 2536 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2537 NULL, NULL, UMA_ALIGN_CACHE, 0); 2538 uma_zone_set_max(unp_zone, maxsockets); 2539 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2540 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2541 NULL, EVENTHANDLER_PRI_ANY); 2542 LIST_INIT(&unp_dhead); 2543 LIST_INIT(&unp_shead); 2544 LIST_INIT(&unp_sphead); 2545 SLIST_INIT(&unp_defers); 2546 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2547 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2548 UNP_LINK_LOCK_INIT(); 2549 UNP_DEFERRED_LOCK_INIT(); 2550 } 2551 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2552 2553 static void 2554 unp_internalize_cleanup_rights(struct mbuf *control) 2555 { 2556 struct cmsghdr *cp; 2557 struct mbuf *m; 2558 void *data; 2559 socklen_t datalen; 2560 2561 for (m = control; m != NULL; m = m->m_next) { 2562 cp = mtod(m, struct cmsghdr *); 2563 if (cp->cmsg_level != SOL_SOCKET || 2564 cp->cmsg_type != SCM_RIGHTS) 2565 continue; 2566 data = CMSG_DATA(cp); 2567 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2568 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2569 } 2570 } 2571 2572 static int 2573 unp_internalize(struct mbuf **controlp, struct thread *td, 2574 struct mbuf **clast, u_int *space, u_int *mbcnt) 2575 { 2576 struct mbuf *control, **initial_controlp; 2577 struct proc *p; 2578 struct filedesc *fdesc; 2579 struct bintime *bt; 2580 struct cmsghdr *cm; 2581 struct cmsgcred *cmcred; 2582 struct filedescent *fde, **fdep, *fdev; 2583 struct file *fp; 2584 struct timeval *tv; 2585 struct timespec *ts; 2586 void *data; 2587 socklen_t clen, datalen; 2588 int i, j, error, *fdp, oldfds; 2589 u_int newlen; 2590 2591 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2592 UNP_LINK_UNLOCK_ASSERT(); 2593 2594 p = td->td_proc; 2595 fdesc = p->p_fd; 2596 error = 0; 2597 control = *controlp; 2598 *controlp = NULL; 2599 initial_controlp = controlp; 2600 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2601 data = CMSG_DATA(cm); 2602 2603 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2604 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2605 (char *)cm + cm->cmsg_len >= (char *)data; 2606 2607 clen -= min(CMSG_SPACE(datalen), clen), 2608 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2609 data = CMSG_DATA(cm)) { 2610 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2611 switch (cm->cmsg_type) { 2612 case SCM_CREDS: 2613 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2614 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2615 cmcred = (struct cmsgcred *) 2616 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2617 cmcred->cmcred_pid = p->p_pid; 2618 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2619 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2620 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2621 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2622 CMGROUP_MAX); 2623 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2624 cmcred->cmcred_groups[i] = 2625 td->td_ucred->cr_groups[i]; 2626 break; 2627 2628 case SCM_RIGHTS: 2629 oldfds = datalen / sizeof (int); 2630 if (oldfds == 0) 2631 continue; 2632 /* On some machines sizeof pointer is bigger than 2633 * sizeof int, so we need to check if data fits into 2634 * single mbuf. We could allocate several mbufs, and 2635 * unp_externalize() should even properly handle that. 2636 * But it is not worth to complicate the code for an 2637 * insane scenario of passing over 200 file descriptors 2638 * at once. 2639 */ 2640 newlen = oldfds * sizeof(fdep[0]); 2641 if (CMSG_SPACE(newlen) > MCLBYTES) { 2642 error = EMSGSIZE; 2643 goto out; 2644 } 2645 /* 2646 * Check that all the FDs passed in refer to legal 2647 * files. If not, reject the entire operation. 2648 */ 2649 fdp = data; 2650 FILEDESC_SLOCK(fdesc); 2651 for (i = 0; i < oldfds; i++, fdp++) { 2652 fp = fget_noref(fdesc, *fdp); 2653 if (fp == NULL) { 2654 FILEDESC_SUNLOCK(fdesc); 2655 error = EBADF; 2656 goto out; 2657 } 2658 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2659 FILEDESC_SUNLOCK(fdesc); 2660 error = EOPNOTSUPP; 2661 goto out; 2662 } 2663 } 2664 2665 /* 2666 * Now replace the integer FDs with pointers to the 2667 * file structure and capability rights. 2668 */ 2669 *controlp = sbcreatecontrol(NULL, newlen, 2670 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2671 fdp = data; 2672 for (i = 0; i < oldfds; i++, fdp++) { 2673 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2674 fdp = data; 2675 for (j = 0; j < i; j++, fdp++) { 2676 fdrop(fdesc->fd_ofiles[*fdp]. 2677 fde_file, td); 2678 } 2679 FILEDESC_SUNLOCK(fdesc); 2680 error = EBADF; 2681 goto out; 2682 } 2683 } 2684 fdp = data; 2685 fdep = (struct filedescent **) 2686 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2687 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2688 M_WAITOK); 2689 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2690 fde = &fdesc->fd_ofiles[*fdp]; 2691 fdep[i] = fdev; 2692 fdep[i]->fde_file = fde->fde_file; 2693 filecaps_copy(&fde->fde_caps, 2694 &fdep[i]->fde_caps, true); 2695 unp_internalize_fp(fdep[i]->fde_file); 2696 } 2697 FILEDESC_SUNLOCK(fdesc); 2698 break; 2699 2700 case SCM_TIMESTAMP: 2701 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2702 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2703 tv = (struct timeval *) 2704 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2705 microtime(tv); 2706 break; 2707 2708 case SCM_BINTIME: 2709 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2710 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2711 bt = (struct bintime *) 2712 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2713 bintime(bt); 2714 break; 2715 2716 case SCM_REALTIME: 2717 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2718 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2719 ts = (struct timespec *) 2720 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2721 nanotime(ts); 2722 break; 2723 2724 case SCM_MONOTONIC: 2725 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2726 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2727 ts = (struct timespec *) 2728 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2729 nanouptime(ts); 2730 break; 2731 2732 default: 2733 error = EINVAL; 2734 goto out; 2735 } 2736 2737 if (space != NULL) { 2738 *space += (*controlp)->m_len; 2739 *mbcnt += MSIZE; 2740 if ((*controlp)->m_flags & M_EXT) 2741 *mbcnt += (*controlp)->m_ext.ext_size; 2742 *clast = *controlp; 2743 } 2744 controlp = &(*controlp)->m_next; 2745 } 2746 if (clen > 0) 2747 error = EINVAL; 2748 2749 out: 2750 if (error != 0 && initial_controlp != NULL) 2751 unp_internalize_cleanup_rights(*initial_controlp); 2752 m_freem(control); 2753 return (error); 2754 } 2755 2756 static struct mbuf * 2757 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2758 struct mbuf **clast, u_int *space, u_int *mbcnt) 2759 { 2760 struct mbuf *m, *n, *n_prev; 2761 const struct cmsghdr *cm; 2762 int ngroups, i, cmsgtype; 2763 size_t ctrlsz; 2764 2765 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2766 if (mode & UNP_WANTCRED_ALWAYS) { 2767 ctrlsz = SOCKCRED2SIZE(ngroups); 2768 cmsgtype = SCM_CREDS2; 2769 } else { 2770 ctrlsz = SOCKCREDSIZE(ngroups); 2771 cmsgtype = SCM_CREDS; 2772 } 2773 2774 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2775 if (m == NULL) 2776 return (control); 2777 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2778 2779 if (mode & UNP_WANTCRED_ALWAYS) { 2780 struct sockcred2 *sc; 2781 2782 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2783 sc->sc_version = 0; 2784 sc->sc_pid = td->td_proc->p_pid; 2785 sc->sc_uid = td->td_ucred->cr_ruid; 2786 sc->sc_euid = td->td_ucred->cr_uid; 2787 sc->sc_gid = td->td_ucred->cr_rgid; 2788 sc->sc_egid = td->td_ucred->cr_gid; 2789 sc->sc_ngroups = ngroups; 2790 for (i = 0; i < sc->sc_ngroups; i++) 2791 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2792 } else { 2793 struct sockcred *sc; 2794 2795 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2796 sc->sc_uid = td->td_ucred->cr_ruid; 2797 sc->sc_euid = td->td_ucred->cr_uid; 2798 sc->sc_gid = td->td_ucred->cr_rgid; 2799 sc->sc_egid = td->td_ucred->cr_gid; 2800 sc->sc_ngroups = ngroups; 2801 for (i = 0; i < sc->sc_ngroups; i++) 2802 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2803 } 2804 2805 /* 2806 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2807 * created SCM_CREDS control message (struct sockcred) has another 2808 * format. 2809 */ 2810 if (control != NULL && cmsgtype == SCM_CREDS) 2811 for (n = control, n_prev = NULL; n != NULL;) { 2812 cm = mtod(n, struct cmsghdr *); 2813 if (cm->cmsg_level == SOL_SOCKET && 2814 cm->cmsg_type == SCM_CREDS) { 2815 if (n_prev == NULL) 2816 control = n->m_next; 2817 else 2818 n_prev->m_next = n->m_next; 2819 if (space != NULL) { 2820 MPASS(*space >= n->m_len); 2821 *space -= n->m_len; 2822 MPASS(*mbcnt >= MSIZE); 2823 *mbcnt -= MSIZE; 2824 if (n->m_flags & M_EXT) { 2825 MPASS(*mbcnt >= 2826 n->m_ext.ext_size); 2827 *mbcnt -= n->m_ext.ext_size; 2828 } 2829 MPASS(clast); 2830 if (*clast == n) { 2831 MPASS(n->m_next == NULL); 2832 if (n_prev == NULL) 2833 *clast = m; 2834 else 2835 *clast = n_prev; 2836 } 2837 } 2838 n = m_free(n); 2839 } else { 2840 n_prev = n; 2841 n = n->m_next; 2842 } 2843 } 2844 2845 /* Prepend it to the head. */ 2846 m->m_next = control; 2847 if (space != NULL) { 2848 *space += m->m_len; 2849 *mbcnt += MSIZE; 2850 if (control == NULL) 2851 *clast = m; 2852 } 2853 return (m); 2854 } 2855 2856 static struct unpcb * 2857 fptounp(struct file *fp) 2858 { 2859 struct socket *so; 2860 2861 if (fp->f_type != DTYPE_SOCKET) 2862 return (NULL); 2863 if ((so = fp->f_data) == NULL) 2864 return (NULL); 2865 if (so->so_proto->pr_domain != &localdomain) 2866 return (NULL); 2867 return sotounpcb(so); 2868 } 2869 2870 static void 2871 unp_discard(struct file *fp) 2872 { 2873 struct unp_defer *dr; 2874 2875 if (unp_externalize_fp(fp)) { 2876 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2877 dr->ud_fp = fp; 2878 UNP_DEFERRED_LOCK(); 2879 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2880 UNP_DEFERRED_UNLOCK(); 2881 atomic_add_int(&unp_defers_count, 1); 2882 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2883 } else 2884 closef_nothread(fp); 2885 } 2886 2887 static void 2888 unp_process_defers(void *arg __unused, int pending) 2889 { 2890 struct unp_defer *dr; 2891 SLIST_HEAD(, unp_defer) drl; 2892 int count; 2893 2894 SLIST_INIT(&drl); 2895 for (;;) { 2896 UNP_DEFERRED_LOCK(); 2897 if (SLIST_FIRST(&unp_defers) == NULL) { 2898 UNP_DEFERRED_UNLOCK(); 2899 break; 2900 } 2901 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2902 UNP_DEFERRED_UNLOCK(); 2903 count = 0; 2904 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2905 SLIST_REMOVE_HEAD(&drl, ud_link); 2906 closef_nothread(dr->ud_fp); 2907 free(dr, M_TEMP); 2908 count++; 2909 } 2910 atomic_add_int(&unp_defers_count, -count); 2911 } 2912 } 2913 2914 static void 2915 unp_internalize_fp(struct file *fp) 2916 { 2917 struct unpcb *unp; 2918 2919 UNP_LINK_WLOCK(); 2920 if ((unp = fptounp(fp)) != NULL) { 2921 unp->unp_file = fp; 2922 unp->unp_msgcount++; 2923 } 2924 unp_rights++; 2925 UNP_LINK_WUNLOCK(); 2926 } 2927 2928 static int 2929 unp_externalize_fp(struct file *fp) 2930 { 2931 struct unpcb *unp; 2932 int ret; 2933 2934 UNP_LINK_WLOCK(); 2935 if ((unp = fptounp(fp)) != NULL) { 2936 unp->unp_msgcount--; 2937 ret = 1; 2938 } else 2939 ret = 0; 2940 unp_rights--; 2941 UNP_LINK_WUNLOCK(); 2942 return (ret); 2943 } 2944 2945 /* 2946 * unp_defer indicates whether additional work has been defered for a future 2947 * pass through unp_gc(). It is thread local and does not require explicit 2948 * synchronization. 2949 */ 2950 static int unp_marked; 2951 2952 static void 2953 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2954 { 2955 struct unpcb *unp; 2956 struct file *fp; 2957 int i; 2958 2959 /* 2960 * This function can only be called from the gc task. 2961 */ 2962 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2963 ("%s: not on gc callout", __func__)); 2964 UNP_LINK_LOCK_ASSERT(); 2965 2966 for (i = 0; i < fdcount; i++) { 2967 fp = fdep[i]->fde_file; 2968 if ((unp = fptounp(fp)) == NULL) 2969 continue; 2970 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2971 continue; 2972 unp->unp_gcrefs--; 2973 } 2974 } 2975 2976 static void 2977 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2978 { 2979 struct unpcb *unp; 2980 struct file *fp; 2981 int i; 2982 2983 /* 2984 * This function can only be called from the gc task. 2985 */ 2986 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2987 ("%s: not on gc callout", __func__)); 2988 UNP_LINK_LOCK_ASSERT(); 2989 2990 for (i = 0; i < fdcount; i++) { 2991 fp = fdep[i]->fde_file; 2992 if ((unp = fptounp(fp)) == NULL) 2993 continue; 2994 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2995 continue; 2996 unp->unp_gcrefs++; 2997 unp_marked++; 2998 } 2999 } 3000 3001 static void 3002 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3003 { 3004 struct sockbuf *sb; 3005 3006 SOCK_LOCK_ASSERT(so); 3007 3008 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3009 return; 3010 3011 SOCK_RECVBUF_LOCK(so); 3012 switch (so->so_type) { 3013 case SOCK_DGRAM: 3014 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3015 unp_scan(so->so_rcv.uxdg_peeked, op); 3016 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3017 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3018 break; 3019 case SOCK_STREAM: 3020 case SOCK_SEQPACKET: 3021 unp_scan(so->so_rcv.sb_mb, op); 3022 break; 3023 } 3024 SOCK_RECVBUF_UNLOCK(so); 3025 } 3026 3027 static void 3028 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3029 { 3030 struct socket *so, *soa; 3031 3032 so = unp->unp_socket; 3033 SOCK_LOCK(so); 3034 if (SOLISTENING(so)) { 3035 /* 3036 * Mark all sockets in our accept queue. 3037 */ 3038 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3039 unp_scan_socket(soa, op); 3040 } else { 3041 /* 3042 * Mark all sockets we reference with RIGHTS. 3043 */ 3044 unp_scan_socket(so, op); 3045 } 3046 SOCK_UNLOCK(so); 3047 } 3048 3049 static int unp_recycled; 3050 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3051 "Number of unreachable sockets claimed by the garbage collector."); 3052 3053 static int unp_taskcount; 3054 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3055 "Number of times the garbage collector has run."); 3056 3057 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3058 "Number of active local sockets."); 3059 3060 static void 3061 unp_gc(__unused void *arg, int pending) 3062 { 3063 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3064 NULL }; 3065 struct unp_head **head; 3066 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3067 struct file *f, **unref; 3068 struct unpcb *unp, *unptmp; 3069 int i, total, unp_unreachable; 3070 3071 LIST_INIT(&unp_deadhead); 3072 unp_taskcount++; 3073 UNP_LINK_RLOCK(); 3074 /* 3075 * First determine which sockets may be in cycles. 3076 */ 3077 unp_unreachable = 0; 3078 3079 for (head = heads; *head != NULL; head++) 3080 LIST_FOREACH(unp, *head, unp_link) { 3081 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3082 ("%s: unp %p has unexpected gc flags 0x%x", 3083 __func__, unp, (unsigned int)unp->unp_gcflag)); 3084 3085 f = unp->unp_file; 3086 3087 /* 3088 * Check for an unreachable socket potentially in a 3089 * cycle. It must be in a queue as indicated by 3090 * msgcount, and this must equal the file reference 3091 * count. Note that when msgcount is 0 the file is 3092 * NULL. 3093 */ 3094 if (f != NULL && unp->unp_msgcount != 0 && 3095 refcount_load(&f->f_count) == unp->unp_msgcount) { 3096 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3097 unp->unp_gcflag |= UNPGC_DEAD; 3098 unp->unp_gcrefs = unp->unp_msgcount; 3099 unp_unreachable++; 3100 } 3101 } 3102 3103 /* 3104 * Scan all sockets previously marked as potentially being in a cycle 3105 * and remove the references each socket holds on any UNPGC_DEAD 3106 * sockets in its queue. After this step, all remaining references on 3107 * sockets marked UNPGC_DEAD should not be part of any cycle. 3108 */ 3109 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3110 unp_gc_scan(unp, unp_remove_dead_ref); 3111 3112 /* 3113 * If a socket still has a non-negative refcount, it cannot be in a 3114 * cycle. In this case increment refcount of all children iteratively. 3115 * Stop the scan once we do a complete loop without discovering 3116 * a new reachable socket. 3117 */ 3118 do { 3119 unp_marked = 0; 3120 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3121 if (unp->unp_gcrefs > 0) { 3122 unp->unp_gcflag &= ~UNPGC_DEAD; 3123 LIST_REMOVE(unp, unp_dead); 3124 KASSERT(unp_unreachable > 0, 3125 ("%s: unp_unreachable underflow.", 3126 __func__)); 3127 unp_unreachable--; 3128 unp_gc_scan(unp, unp_restore_undead_ref); 3129 } 3130 } while (unp_marked); 3131 3132 UNP_LINK_RUNLOCK(); 3133 3134 if (unp_unreachable == 0) 3135 return; 3136 3137 /* 3138 * Allocate space for a local array of dead unpcbs. 3139 * TODO: can this path be simplified by instead using the local 3140 * dead list at unp_deadhead, after taking out references 3141 * on the file object and/or unpcb and dropping the link lock? 3142 */ 3143 unref = malloc(unp_unreachable * sizeof(struct file *), 3144 M_TEMP, M_WAITOK); 3145 3146 /* 3147 * Iterate looking for sockets which have been specifically marked 3148 * as unreachable and store them locally. 3149 */ 3150 UNP_LINK_RLOCK(); 3151 total = 0; 3152 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3153 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3154 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3155 unp->unp_gcflag &= ~UNPGC_DEAD; 3156 f = unp->unp_file; 3157 if (unp->unp_msgcount == 0 || f == NULL || 3158 refcount_load(&f->f_count) != unp->unp_msgcount || 3159 !fhold(f)) 3160 continue; 3161 unref[total++] = f; 3162 KASSERT(total <= unp_unreachable, 3163 ("%s: incorrect unreachable count.", __func__)); 3164 } 3165 UNP_LINK_RUNLOCK(); 3166 3167 /* 3168 * Now flush all sockets, free'ing rights. This will free the 3169 * struct files associated with these sockets but leave each socket 3170 * with one remaining ref. 3171 */ 3172 for (i = 0; i < total; i++) { 3173 struct socket *so; 3174 3175 so = unref[i]->f_data; 3176 CURVNET_SET(so->so_vnet); 3177 sorflush(so); 3178 CURVNET_RESTORE(); 3179 } 3180 3181 /* 3182 * And finally release the sockets so they can be reclaimed. 3183 */ 3184 for (i = 0; i < total; i++) 3185 fdrop(unref[i], NULL); 3186 unp_recycled += total; 3187 free(unref, M_TEMP); 3188 } 3189 3190 /* 3191 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3192 */ 3193 static void 3194 unp_dispose(struct socket *so) 3195 { 3196 struct sockbuf *sb; 3197 struct unpcb *unp; 3198 struct mbuf *m; 3199 3200 MPASS(!SOLISTENING(so)); 3201 3202 unp = sotounpcb(so); 3203 UNP_LINK_WLOCK(); 3204 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3205 UNP_LINK_WUNLOCK(); 3206 3207 /* 3208 * Grab our special mbufs before calling sbrelease(). 3209 */ 3210 SOCK_RECVBUF_LOCK(so); 3211 switch (so->so_type) { 3212 case SOCK_DGRAM: 3213 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3214 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3215 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3216 /* Note: socket of sb may reconnect. */ 3217 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3218 } 3219 sb = &so->so_rcv; 3220 if (sb->uxdg_peeked != NULL) { 3221 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3222 m_stailqpkt); 3223 sb->uxdg_peeked = NULL; 3224 } 3225 m = STAILQ_FIRST(&sb->uxdg_mb); 3226 STAILQ_INIT(&sb->uxdg_mb); 3227 /* XXX: our shortened sbrelease() */ 3228 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3229 RLIM_INFINITY); 3230 /* 3231 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3232 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3233 * data to the socket. 3234 * We are now in dom_dispose and it could be a call from 3235 * soshutdown() or from the final sofree(). The sofree() case 3236 * is simple as it guarantees that no more sends will happen, 3237 * however we can race with unp_disconnect() from our peer. 3238 * The shutdown(2) case is more exotic. It would call into 3239 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3240 * possible if we did connect(2) on this socket and we also 3241 * had it bound with bind(2) and receive connections from other 3242 * sockets. Because soshutdown() violates POSIX (see comment 3243 * there) we will end up here shutting down our receive side. 3244 * Of course this will have affect not only on the peer we 3245 * connect(2)ed to, but also on all of the peers who had 3246 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3247 */ 3248 sb->sb_state |= SBS_CANTRCVMORE; 3249 break; 3250 case SOCK_STREAM: 3251 case SOCK_SEQPACKET: 3252 sb = &so->so_rcv; 3253 m = sbcut_locked(sb, sb->sb_ccc); 3254 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3255 ("%s: ccc %u mb %p mbcnt %u", __func__, 3256 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3257 sbrelease_locked(so, SO_RCV); 3258 break; 3259 } 3260 SOCK_RECVBUF_UNLOCK(so); 3261 if (SOCK_IO_RECV_OWNED(so)) 3262 SOCK_IO_RECV_UNLOCK(so); 3263 3264 if (m != NULL) { 3265 unp_scan(m, unp_freerights); 3266 m_freem(m); 3267 } 3268 } 3269 3270 static void 3271 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3272 { 3273 struct mbuf *m; 3274 struct cmsghdr *cm; 3275 void *data; 3276 socklen_t clen, datalen; 3277 3278 while (m0 != NULL) { 3279 for (m = m0; m; m = m->m_next) { 3280 if (m->m_type != MT_CONTROL) 3281 continue; 3282 3283 cm = mtod(m, struct cmsghdr *); 3284 clen = m->m_len; 3285 3286 while (cm != NULL) { 3287 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3288 break; 3289 3290 data = CMSG_DATA(cm); 3291 datalen = (caddr_t)cm + cm->cmsg_len 3292 - (caddr_t)data; 3293 3294 if (cm->cmsg_level == SOL_SOCKET && 3295 cm->cmsg_type == SCM_RIGHTS) { 3296 (*op)(data, datalen / 3297 sizeof(struct filedescent *)); 3298 } 3299 3300 if (CMSG_SPACE(datalen) < clen) { 3301 clen -= CMSG_SPACE(datalen); 3302 cm = (struct cmsghdr *) 3303 ((caddr_t)cm + CMSG_SPACE(datalen)); 3304 } else { 3305 clen = 0; 3306 cm = NULL; 3307 } 3308 } 3309 } 3310 m0 = m0->m_nextpkt; 3311 } 3312 } 3313 3314 /* 3315 * Definitions of protocols supported in the LOCAL domain. 3316 */ 3317 static struct protosw streamproto = { 3318 .pr_type = SOCK_STREAM, 3319 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 3320 PR_CAPATTACH, 3321 .pr_ctloutput = &uipc_ctloutput, 3322 .pr_abort = uipc_abort, 3323 .pr_accept = uipc_accept, 3324 .pr_attach = uipc_attach, 3325 .pr_bind = uipc_bind, 3326 .pr_bindat = uipc_bindat, 3327 .pr_connect = uipc_connect, 3328 .pr_connectat = uipc_connectat, 3329 .pr_connect2 = uipc_connect2, 3330 .pr_detach = uipc_detach, 3331 .pr_disconnect = uipc_disconnect, 3332 .pr_listen = uipc_listen, 3333 .pr_peeraddr = uipc_peeraddr, 3334 .pr_rcvd = uipc_rcvd, 3335 .pr_send = uipc_send, 3336 .pr_ready = uipc_ready, 3337 .pr_sense = uipc_sense, 3338 .pr_shutdown = uipc_shutdown, 3339 .pr_sockaddr = uipc_sockaddr, 3340 .pr_soreceive = soreceive_generic, 3341 .pr_close = uipc_close, 3342 }; 3343 3344 static struct protosw dgramproto = { 3345 .pr_type = SOCK_DGRAM, 3346 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 3347 PR_SOCKBUF, 3348 .pr_ctloutput = &uipc_ctloutput, 3349 .pr_abort = uipc_abort, 3350 .pr_accept = uipc_accept, 3351 .pr_attach = uipc_attach, 3352 .pr_bind = uipc_bind, 3353 .pr_bindat = uipc_bindat, 3354 .pr_connect = uipc_connect, 3355 .pr_connectat = uipc_connectat, 3356 .pr_connect2 = uipc_connect2, 3357 .pr_detach = uipc_detach, 3358 .pr_disconnect = uipc_disconnect, 3359 .pr_peeraddr = uipc_peeraddr, 3360 .pr_sosend = uipc_sosend_dgram, 3361 .pr_sense = uipc_sense, 3362 .pr_shutdown = uipc_shutdown, 3363 .pr_sockaddr = uipc_sockaddr, 3364 .pr_soreceive = uipc_soreceive_dgram, 3365 .pr_close = uipc_close, 3366 }; 3367 3368 static struct protosw seqpacketproto = { 3369 .pr_type = SOCK_SEQPACKET, 3370 /* 3371 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3372 * due to our use of sbappendaddr. A new sbappend variants is needed 3373 * that supports both atomic record writes and control data. 3374 */ 3375 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 3376 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 3377 .pr_ctloutput = &uipc_ctloutput, 3378 .pr_abort = uipc_abort, 3379 .pr_accept = uipc_accept, 3380 .pr_attach = uipc_attach, 3381 .pr_bind = uipc_bind, 3382 .pr_bindat = uipc_bindat, 3383 .pr_connect = uipc_connect, 3384 .pr_connectat = uipc_connectat, 3385 .pr_connect2 = uipc_connect2, 3386 .pr_detach = uipc_detach, 3387 .pr_disconnect = uipc_disconnect, 3388 .pr_listen = uipc_listen, 3389 .pr_peeraddr = uipc_peeraddr, 3390 .pr_rcvd = uipc_rcvd, 3391 .pr_send = uipc_send, 3392 .pr_sense = uipc_sense, 3393 .pr_shutdown = uipc_shutdown, 3394 .pr_sockaddr = uipc_sockaddr, 3395 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3396 .pr_close = uipc_close, 3397 }; 3398 3399 static struct domain localdomain = { 3400 .dom_family = AF_LOCAL, 3401 .dom_name = "local", 3402 .dom_externalize = unp_externalize, 3403 .dom_dispose = unp_dispose, 3404 .dom_nprotosw = 3, 3405 .dom_protosw = { 3406 &streamproto, 3407 &dgramproto, 3408 &seqpacketproto, 3409 } 3410 }; 3411 DOMAIN_SET(local); 3412 3413 /* 3414 * A helper function called by VFS before socket-type vnode reclamation. 3415 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3416 * use count. 3417 */ 3418 void 3419 vfs_unp_reclaim(struct vnode *vp) 3420 { 3421 struct unpcb *unp; 3422 int active; 3423 struct mtx *vplock; 3424 3425 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3426 KASSERT(vp->v_type == VSOCK, 3427 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3428 3429 active = 0; 3430 vplock = mtx_pool_find(mtxpool_sleep, vp); 3431 mtx_lock(vplock); 3432 VOP_UNP_CONNECT(vp, &unp); 3433 if (unp == NULL) 3434 goto done; 3435 UNP_PCB_LOCK(unp); 3436 if (unp->unp_vnode == vp) { 3437 VOP_UNP_DETACH(vp); 3438 unp->unp_vnode = NULL; 3439 active = 1; 3440 } 3441 UNP_PCB_UNLOCK(unp); 3442 done: 3443 mtx_unlock(vplock); 3444 if (active) 3445 vunref(vp); 3446 } 3447 3448 #ifdef DDB 3449 static void 3450 db_print_indent(int indent) 3451 { 3452 int i; 3453 3454 for (i = 0; i < indent; i++) 3455 db_printf(" "); 3456 } 3457 3458 static void 3459 db_print_unpflags(int unp_flags) 3460 { 3461 int comma; 3462 3463 comma = 0; 3464 if (unp_flags & UNP_HAVEPC) { 3465 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3466 comma = 1; 3467 } 3468 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3469 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3470 comma = 1; 3471 } 3472 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3473 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3474 comma = 1; 3475 } 3476 if (unp_flags & UNP_CONNWAIT) { 3477 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3478 comma = 1; 3479 } 3480 if (unp_flags & UNP_CONNECTING) { 3481 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3482 comma = 1; 3483 } 3484 if (unp_flags & UNP_BINDING) { 3485 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3486 comma = 1; 3487 } 3488 } 3489 3490 static void 3491 db_print_xucred(int indent, struct xucred *xu) 3492 { 3493 int comma, i; 3494 3495 db_print_indent(indent); 3496 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3497 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3498 db_print_indent(indent); 3499 db_printf("cr_groups: "); 3500 comma = 0; 3501 for (i = 0; i < xu->cr_ngroups; i++) { 3502 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3503 comma = 1; 3504 } 3505 db_printf("\n"); 3506 } 3507 3508 static void 3509 db_print_unprefs(int indent, struct unp_head *uh) 3510 { 3511 struct unpcb *unp; 3512 int counter; 3513 3514 counter = 0; 3515 LIST_FOREACH(unp, uh, unp_reflink) { 3516 if (counter % 4 == 0) 3517 db_print_indent(indent); 3518 db_printf("%p ", unp); 3519 if (counter % 4 == 3) 3520 db_printf("\n"); 3521 counter++; 3522 } 3523 if (counter != 0 && counter % 4 != 0) 3524 db_printf("\n"); 3525 } 3526 3527 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3528 { 3529 struct unpcb *unp; 3530 3531 if (!have_addr) { 3532 db_printf("usage: show unpcb <addr>\n"); 3533 return; 3534 } 3535 unp = (struct unpcb *)addr; 3536 3537 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3538 unp->unp_vnode); 3539 3540 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3541 unp->unp_conn); 3542 3543 db_printf("unp_refs:\n"); 3544 db_print_unprefs(2, &unp->unp_refs); 3545 3546 /* XXXRW: Would be nice to print the full address, if any. */ 3547 db_printf("unp_addr: %p\n", unp->unp_addr); 3548 3549 db_printf("unp_gencnt: %llu\n", 3550 (unsigned long long)unp->unp_gencnt); 3551 3552 db_printf("unp_flags: %x (", unp->unp_flags); 3553 db_print_unpflags(unp->unp_flags); 3554 db_printf(")\n"); 3555 3556 db_printf("unp_peercred:\n"); 3557 db_print_xucred(2, &unp->unp_peercred); 3558 3559 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3560 } 3561 #endif 3562