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 __FBSDID("$FreeBSD$"); 63 64 #include "opt_ddb.h" 65 66 #include <sys/param.h> 67 #include <sys/capsicum.h> 68 #include <sys/domain.h> 69 #include <sys/eventhandler.h> 70 #include <sys/fcntl.h> 71 #include <sys/file.h> 72 #include <sys/filedesc.h> 73 #include <sys/kernel.h> 74 #include <sys/lock.h> 75 #include <sys/malloc.h> 76 #include <sys/mbuf.h> 77 #include <sys/mount.h> 78 #include <sys/mutex.h> 79 #include <sys/namei.h> 80 #include <sys/proc.h> 81 #include <sys/protosw.h> 82 #include <sys/queue.h> 83 #include <sys/resourcevar.h> 84 #include <sys/rwlock.h> 85 #include <sys/socket.h> 86 #include <sys/socketvar.h> 87 #include <sys/signalvar.h> 88 #include <sys/stat.h> 89 #include <sys/sx.h> 90 #include <sys/sysctl.h> 91 #include <sys/systm.h> 92 #include <sys/taskqueue.h> 93 #include <sys/un.h> 94 #include <sys/unpcb.h> 95 #include <sys/vnode.h> 96 97 #include <net/vnet.h> 98 99 #ifdef DDB 100 #include <ddb/ddb.h> 101 #endif 102 103 #include <security/mac/mac_framework.h> 104 105 #include <vm/uma.h> 106 107 MALLOC_DECLARE(M_FILECAPS); 108 109 static struct domain localdomain; 110 111 static uma_zone_t unp_zone; 112 static unp_gen_t unp_gencnt; /* (l) */ 113 static u_int unp_count; /* (l) Count of local sockets. */ 114 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 115 static int unp_rights; /* (g) File descriptors in flight. */ 116 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 117 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 118 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 119 120 struct unp_defer { 121 SLIST_ENTRY(unp_defer) ud_link; 122 struct file *ud_fp; 123 }; 124 static SLIST_HEAD(, unp_defer) unp_defers; 125 static int unp_defers_count; 126 127 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 128 129 /* 130 * Garbage collection of cyclic file descriptor/socket references occurs 131 * asynchronously in a taskqueue context in order to avoid recursion and 132 * reentrance in the UNIX domain socket, file descriptor, and socket layer 133 * code. See unp_gc() for a full description. 134 */ 135 static struct timeout_task unp_gc_task; 136 137 /* 138 * The close of unix domain sockets attached as SCM_RIGHTS is 139 * postponed to the taskqueue, to avoid arbitrary recursion depth. 140 * The attached sockets might have another sockets attached. 141 */ 142 static struct task unp_defer_task; 143 144 /* 145 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 146 * stream sockets, although the total for sender and receiver is actually 147 * only PIPSIZ. 148 * 149 * Datagram sockets really use the sendspace as the maximum datagram size, 150 * and don't really want to reserve the sendspace. Their recvspace should be 151 * large enough for at least one max-size datagram plus address. 152 */ 153 #ifndef PIPSIZ 154 #define PIPSIZ 8192 155 #endif 156 static u_long unpst_sendspace = PIPSIZ; 157 static u_long unpst_recvspace = PIPSIZ; 158 static u_long unpdg_maxdgram = 2*1024; 159 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */ 160 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 161 static u_long unpsp_recvspace = PIPSIZ; 162 163 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 164 "Local domain"); 165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 166 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 167 "SOCK_STREAM"); 168 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 170 "SOCK_DGRAM"); 171 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 173 "SOCK_SEQPACKET"); 174 175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 176 &unpst_sendspace, 0, "Default stream send space."); 177 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 178 &unpst_recvspace, 0, "Default stream receive space."); 179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 180 &unpdg_maxdgram, 0, "Maximum datagram size."); 181 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 182 &unpdg_recvspace, 0, "Default datagram receive space."); 183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 184 &unpsp_sendspace, 0, "Default seqpacket send space."); 185 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 186 &unpsp_recvspace, 0, "Default seqpacket receive space."); 187 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 188 "File descriptors in flight."); 189 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 190 &unp_defers_count, 0, 191 "File descriptors deferred to taskqueue for close."); 192 193 /* 194 * Locking and synchronization: 195 * 196 * Several types of locks exist in the local domain socket implementation: 197 * - a global linkage lock 198 * - a global connection list lock 199 * - the mtxpool lock 200 * - per-unpcb mutexes 201 * 202 * The linkage lock protects the global socket lists, the generation number 203 * counter and garbage collector state. 204 * 205 * The connection list lock protects the list of referring sockets in a datagram 206 * socket PCB. This lock is also overloaded to protect a global list of 207 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 208 * messages. To avoid recursion, such references are released by a dedicated 209 * thread. 210 * 211 * The mtxpool lock protects the vnode from being modified while referenced. 212 * Lock ordering rules require that it be acquired before any PCB locks. 213 * 214 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 215 * unpcb. This includes the unp_conn field, which either links two connected 216 * PCBs together (for connected socket types) or points at the destination 217 * socket (for connectionless socket types). The operations of creating or 218 * destroying a connection therefore involve locking multiple PCBs. To avoid 219 * lock order reversals, in some cases this involves dropping a PCB lock and 220 * using a reference counter to maintain liveness. 221 * 222 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 223 * allocated in pr_attach() and freed in pr_detach(). The validity of that 224 * pointer is an invariant, so no lock is required to dereference the so_pcb 225 * pointer if a valid socket reference is held by the caller. In practice, 226 * this is always true during operations performed on a socket. Each unpcb 227 * has a back-pointer to its socket, unp_socket, which will be stable under 228 * the same circumstances. 229 * 230 * This pointer may only be safely dereferenced as long as a valid reference 231 * to the unpcb is held. Typically, this reference will be from the socket, 232 * or from another unpcb when the referring unpcb's lock is held (in order 233 * that the reference not be invalidated during use). For example, to follow 234 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 235 * that detach is not run clearing unp_socket. 236 * 237 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 238 * protocols, bind() is a non-atomic operation, and connect() requires 239 * potential sleeping in the protocol, due to potentially waiting on local or 240 * distributed file systems. We try to separate "lookup" operations, which 241 * may sleep, and the IPC operations themselves, which typically can occur 242 * with relative atomicity as locks can be held over the entire operation. 243 * 244 * Another tricky issue is simultaneous multi-threaded or multi-process 245 * access to a single UNIX domain socket. These are handled by the flags 246 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 247 * binding, both of which involve dropping UNIX domain socket locks in order 248 * to perform namei() and other file system operations. 249 */ 250 static struct rwlock unp_link_rwlock; 251 static struct mtx unp_defers_lock; 252 253 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 254 "unp_link_rwlock") 255 256 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 257 RA_LOCKED) 258 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 259 RA_UNLOCKED) 260 261 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 262 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 263 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 264 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 265 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 266 RA_WLOCKED) 267 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 268 269 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 270 "unp_defer", NULL, MTX_DEF) 271 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 272 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 273 274 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 275 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 276 277 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 278 "unp", "unp", \ 279 MTX_DUPOK|MTX_DEF) 280 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 281 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 282 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 283 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 284 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 285 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 286 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 287 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 288 289 static int uipc_connect2(struct socket *, struct socket *); 290 static int uipc_ctloutput(struct socket *, struct sockopt *); 291 static int unp_connect(struct socket *, struct sockaddr *, 292 struct thread *); 293 static int unp_connectat(int, struct socket *, struct sockaddr *, 294 struct thread *, bool); 295 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how; 296 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how); 297 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 298 static void unp_dispose(struct socket *so); 299 static void unp_shutdown(struct unpcb *); 300 static void unp_drop(struct unpcb *); 301 static void unp_gc(__unused void *, int); 302 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 303 static void unp_discard(struct file *); 304 static void unp_freerights(struct filedescent **, int); 305 static int unp_internalize(struct mbuf **, struct thread *, 306 struct mbuf **, u_int *, u_int *); 307 static void unp_internalize_fp(struct file *); 308 static int unp_externalize(struct mbuf *, struct mbuf **, int); 309 static int unp_externalize_fp(struct file *); 310 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 311 int, struct mbuf **, u_int *, u_int *); 312 static void unp_process_defers(void * __unused, int); 313 314 static void 315 unp_pcb_hold(struct unpcb *unp) 316 { 317 u_int old __unused; 318 319 old = refcount_acquire(&unp->unp_refcount); 320 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 321 } 322 323 static __result_use_check bool 324 unp_pcb_rele(struct unpcb *unp) 325 { 326 bool ret; 327 328 UNP_PCB_LOCK_ASSERT(unp); 329 330 if ((ret = refcount_release(&unp->unp_refcount))) { 331 UNP_PCB_UNLOCK(unp); 332 UNP_PCB_LOCK_DESTROY(unp); 333 uma_zfree(unp_zone, unp); 334 } 335 return (ret); 336 } 337 338 static void 339 unp_pcb_rele_notlast(struct unpcb *unp) 340 { 341 bool ret __unused; 342 343 ret = refcount_release(&unp->unp_refcount); 344 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 345 } 346 347 static void 348 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 349 { 350 UNP_PCB_UNLOCK_ASSERT(unp); 351 UNP_PCB_UNLOCK_ASSERT(unp2); 352 353 if (unp == unp2) { 354 UNP_PCB_LOCK(unp); 355 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 356 UNP_PCB_LOCK(unp); 357 UNP_PCB_LOCK(unp2); 358 } else { 359 UNP_PCB_LOCK(unp2); 360 UNP_PCB_LOCK(unp); 361 } 362 } 363 364 static void 365 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 366 { 367 UNP_PCB_UNLOCK(unp); 368 if (unp != unp2) 369 UNP_PCB_UNLOCK(unp2); 370 } 371 372 /* 373 * Try to lock the connected peer of an already locked socket. In some cases 374 * this requires that we unlock the current socket. The pairbusy counter is 375 * used to block concurrent connection attempts while the lock is dropped. The 376 * caller must be careful to revalidate PCB state. 377 */ 378 static struct unpcb * 379 unp_pcb_lock_peer(struct unpcb *unp) 380 { 381 struct unpcb *unp2; 382 383 UNP_PCB_LOCK_ASSERT(unp); 384 unp2 = unp->unp_conn; 385 if (unp2 == NULL) 386 return (NULL); 387 if (__predict_false(unp == unp2)) 388 return (unp); 389 390 UNP_PCB_UNLOCK_ASSERT(unp2); 391 392 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 393 return (unp2); 394 if ((uintptr_t)unp2 > (uintptr_t)unp) { 395 UNP_PCB_LOCK(unp2); 396 return (unp2); 397 } 398 unp->unp_pairbusy++; 399 unp_pcb_hold(unp2); 400 UNP_PCB_UNLOCK(unp); 401 402 UNP_PCB_LOCK(unp2); 403 UNP_PCB_LOCK(unp); 404 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 405 ("%s: socket %p was reconnected", __func__, unp)); 406 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 407 unp->unp_flags &= ~UNP_WAITING; 408 wakeup(unp); 409 } 410 if (unp_pcb_rele(unp2)) { 411 /* unp2 is unlocked. */ 412 return (NULL); 413 } 414 if (unp->unp_conn == NULL) { 415 UNP_PCB_UNLOCK(unp2); 416 return (NULL); 417 } 418 return (unp2); 419 } 420 421 static void 422 uipc_abort(struct socket *so) 423 { 424 struct unpcb *unp, *unp2; 425 426 unp = sotounpcb(so); 427 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 428 UNP_PCB_UNLOCK_ASSERT(unp); 429 430 UNP_PCB_LOCK(unp); 431 unp2 = unp->unp_conn; 432 if (unp2 != NULL) { 433 unp_pcb_hold(unp2); 434 UNP_PCB_UNLOCK(unp); 435 unp_drop(unp2); 436 } else 437 UNP_PCB_UNLOCK(unp); 438 } 439 440 static int 441 uipc_accept(struct socket *so, struct sockaddr **nam) 442 { 443 struct unpcb *unp, *unp2; 444 const struct sockaddr *sa; 445 446 /* 447 * Pass back name of connected socket, if it was bound and we are 448 * still connected (our peer may have closed already!). 449 */ 450 unp = sotounpcb(so); 451 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 452 453 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 454 UNP_PCB_LOCK(unp); 455 unp2 = unp_pcb_lock_peer(unp); 456 if (unp2 != NULL && unp2->unp_addr != NULL) 457 sa = (struct sockaddr *)unp2->unp_addr; 458 else 459 sa = &sun_noname; 460 bcopy(sa, *nam, sa->sa_len); 461 if (unp2 != NULL) 462 unp_pcb_unlock_pair(unp, unp2); 463 else 464 UNP_PCB_UNLOCK(unp); 465 return (0); 466 } 467 468 static int 469 uipc_attach(struct socket *so, int proto, struct thread *td) 470 { 471 u_long sendspace, recvspace; 472 struct unpcb *unp; 473 int error; 474 bool locked; 475 476 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 477 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 478 switch (so->so_type) { 479 case SOCK_STREAM: 480 sendspace = unpst_sendspace; 481 recvspace = unpst_recvspace; 482 break; 483 484 case SOCK_DGRAM: 485 STAILQ_INIT(&so->so_rcv.uxdg_mb); 486 STAILQ_INIT(&so->so_snd.uxdg_mb); 487 TAILQ_INIT(&so->so_rcv.uxdg_conns); 488 /* 489 * Since send buffer is either bypassed or is a part 490 * of one-to-many receive buffer, we assign both space 491 * limits to unpdg_recvspace. 492 */ 493 sendspace = recvspace = unpdg_recvspace; 494 break; 495 496 case SOCK_SEQPACKET: 497 sendspace = unpsp_sendspace; 498 recvspace = unpsp_recvspace; 499 break; 500 501 default: 502 panic("uipc_attach"); 503 } 504 error = soreserve(so, sendspace, recvspace); 505 if (error) 506 return (error); 507 } 508 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 509 if (unp == NULL) 510 return (ENOBUFS); 511 LIST_INIT(&unp->unp_refs); 512 UNP_PCB_LOCK_INIT(unp); 513 unp->unp_socket = so; 514 so->so_pcb = unp; 515 refcount_init(&unp->unp_refcount, 1); 516 517 if ((locked = UNP_LINK_WOWNED()) == false) 518 UNP_LINK_WLOCK(); 519 520 unp->unp_gencnt = ++unp_gencnt; 521 unp->unp_ino = ++unp_ino; 522 unp_count++; 523 switch (so->so_type) { 524 case SOCK_STREAM: 525 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 526 break; 527 528 case SOCK_DGRAM: 529 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 530 break; 531 532 case SOCK_SEQPACKET: 533 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 534 break; 535 536 default: 537 panic("uipc_attach"); 538 } 539 540 if (locked == false) 541 UNP_LINK_WUNLOCK(); 542 543 return (0); 544 } 545 546 static int 547 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 548 { 549 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 550 struct vattr vattr; 551 int error, namelen; 552 struct nameidata nd; 553 struct unpcb *unp; 554 struct vnode *vp; 555 struct mount *mp; 556 cap_rights_t rights; 557 char *buf; 558 559 if (nam->sa_family != AF_UNIX) 560 return (EAFNOSUPPORT); 561 562 unp = sotounpcb(so); 563 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 564 565 if (soun->sun_len > sizeof(struct sockaddr_un)) 566 return (EINVAL); 567 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 568 if (namelen <= 0) 569 return (EINVAL); 570 571 /* 572 * We don't allow simultaneous bind() calls on a single UNIX domain 573 * socket, so flag in-progress operations, and return an error if an 574 * operation is already in progress. 575 * 576 * Historically, we have not allowed a socket to be rebound, so this 577 * also returns an error. Not allowing re-binding simplifies the 578 * implementation and avoids a great many possible failure modes. 579 */ 580 UNP_PCB_LOCK(unp); 581 if (unp->unp_vnode != NULL) { 582 UNP_PCB_UNLOCK(unp); 583 return (EINVAL); 584 } 585 if (unp->unp_flags & UNP_BINDING) { 586 UNP_PCB_UNLOCK(unp); 587 return (EALREADY); 588 } 589 unp->unp_flags |= UNP_BINDING; 590 UNP_PCB_UNLOCK(unp); 591 592 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 593 bcopy(soun->sun_path, buf, namelen); 594 buf[namelen] = 0; 595 596 restart: 597 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, 598 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 599 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 600 error = namei(&nd); 601 if (error) 602 goto error; 603 vp = nd.ni_vp; 604 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 605 NDFREE_PNBUF(&nd); 606 if (nd.ni_dvp == vp) 607 vrele(nd.ni_dvp); 608 else 609 vput(nd.ni_dvp); 610 if (vp != NULL) { 611 vrele(vp); 612 error = EADDRINUSE; 613 goto error; 614 } 615 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); 616 if (error) 617 goto error; 618 goto restart; 619 } 620 VATTR_NULL(&vattr); 621 vattr.va_type = VSOCK; 622 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 623 #ifdef MAC 624 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 625 &vattr); 626 #endif 627 if (error == 0) 628 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 629 NDFREE_PNBUF(&nd); 630 if (error) { 631 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 632 vn_finished_write(mp); 633 if (error == ERELOOKUP) 634 goto restart; 635 goto error; 636 } 637 vp = nd.ni_vp; 638 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 639 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 640 641 UNP_PCB_LOCK(unp); 642 VOP_UNP_BIND(vp, unp); 643 unp->unp_vnode = vp; 644 unp->unp_addr = soun; 645 unp->unp_flags &= ~UNP_BINDING; 646 UNP_PCB_UNLOCK(unp); 647 vref(vp); 648 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 649 vn_finished_write(mp); 650 free(buf, M_TEMP); 651 return (0); 652 653 error: 654 UNP_PCB_LOCK(unp); 655 unp->unp_flags &= ~UNP_BINDING; 656 UNP_PCB_UNLOCK(unp); 657 free(buf, M_TEMP); 658 return (error); 659 } 660 661 static int 662 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 663 { 664 665 return (uipc_bindat(AT_FDCWD, so, nam, td)); 666 } 667 668 static int 669 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 670 { 671 int error; 672 673 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 674 error = unp_connect(so, nam, td); 675 return (error); 676 } 677 678 static int 679 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 680 struct thread *td) 681 { 682 int error; 683 684 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 685 error = unp_connectat(fd, so, nam, td, false); 686 return (error); 687 } 688 689 static void 690 uipc_close(struct socket *so) 691 { 692 struct unpcb *unp, *unp2; 693 struct vnode *vp = NULL; 694 struct mtx *vplock; 695 696 unp = sotounpcb(so); 697 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 698 699 vplock = NULL; 700 if ((vp = unp->unp_vnode) != NULL) { 701 vplock = mtx_pool_find(mtxpool_sleep, vp); 702 mtx_lock(vplock); 703 } 704 UNP_PCB_LOCK(unp); 705 if (vp && unp->unp_vnode == NULL) { 706 mtx_unlock(vplock); 707 vp = NULL; 708 } 709 if (vp != NULL) { 710 VOP_UNP_DETACH(vp); 711 unp->unp_vnode = NULL; 712 } 713 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 714 unp_disconnect(unp, unp2); 715 else 716 UNP_PCB_UNLOCK(unp); 717 if (vp) { 718 mtx_unlock(vplock); 719 vrele(vp); 720 } 721 } 722 723 static int 724 uipc_connect2(struct socket *so1, struct socket *so2) 725 { 726 struct unpcb *unp, *unp2; 727 728 if (so1->so_type != so2->so_type) 729 return (EPROTOTYPE); 730 731 unp = so1->so_pcb; 732 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 733 unp2 = so2->so_pcb; 734 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 735 unp_pcb_lock_pair(unp, unp2); 736 unp_connect2(so1, so2, PRU_CONNECT2); 737 unp_pcb_unlock_pair(unp, unp2); 738 739 return (0); 740 } 741 742 static void 743 uipc_detach(struct socket *so) 744 { 745 struct unpcb *unp, *unp2; 746 struct mtx *vplock; 747 struct vnode *vp; 748 int local_unp_rights; 749 750 unp = sotounpcb(so); 751 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 752 753 vp = NULL; 754 vplock = NULL; 755 756 UNP_LINK_WLOCK(); 757 LIST_REMOVE(unp, unp_link); 758 if (unp->unp_gcflag & UNPGC_DEAD) 759 LIST_REMOVE(unp, unp_dead); 760 unp->unp_gencnt = ++unp_gencnt; 761 --unp_count; 762 UNP_LINK_WUNLOCK(); 763 764 UNP_PCB_UNLOCK_ASSERT(unp); 765 restart: 766 if ((vp = unp->unp_vnode) != NULL) { 767 vplock = mtx_pool_find(mtxpool_sleep, vp); 768 mtx_lock(vplock); 769 } 770 UNP_PCB_LOCK(unp); 771 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 772 if (vplock) 773 mtx_unlock(vplock); 774 UNP_PCB_UNLOCK(unp); 775 goto restart; 776 } 777 if ((vp = unp->unp_vnode) != NULL) { 778 VOP_UNP_DETACH(vp); 779 unp->unp_vnode = NULL; 780 } 781 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 782 unp_disconnect(unp, unp2); 783 else 784 UNP_PCB_UNLOCK(unp); 785 786 UNP_REF_LIST_LOCK(); 787 while (!LIST_EMPTY(&unp->unp_refs)) { 788 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 789 790 unp_pcb_hold(ref); 791 UNP_REF_LIST_UNLOCK(); 792 793 MPASS(ref != unp); 794 UNP_PCB_UNLOCK_ASSERT(ref); 795 unp_drop(ref); 796 UNP_REF_LIST_LOCK(); 797 } 798 UNP_REF_LIST_UNLOCK(); 799 800 UNP_PCB_LOCK(unp); 801 local_unp_rights = unp_rights; 802 unp->unp_socket->so_pcb = NULL; 803 unp->unp_socket = NULL; 804 free(unp->unp_addr, M_SONAME); 805 unp->unp_addr = NULL; 806 if (!unp_pcb_rele(unp)) 807 UNP_PCB_UNLOCK(unp); 808 if (vp) { 809 mtx_unlock(vplock); 810 vrele(vp); 811 } 812 if (local_unp_rights) 813 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 814 815 switch (so->so_type) { 816 case SOCK_DGRAM: 817 /* 818 * Everything should have been unlinked/freed by unp_dispose() 819 * and/or unp_disconnect(). 820 */ 821 MPASS(so->so_rcv.uxdg_peeked == NULL); 822 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 823 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 824 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 825 } 826 } 827 828 static int 829 uipc_disconnect(struct socket *so) 830 { 831 struct unpcb *unp, *unp2; 832 833 unp = sotounpcb(so); 834 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 835 836 UNP_PCB_LOCK(unp); 837 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 838 unp_disconnect(unp, unp2); 839 else 840 UNP_PCB_UNLOCK(unp); 841 return (0); 842 } 843 844 static int 845 uipc_listen(struct socket *so, int backlog, struct thread *td) 846 { 847 struct unpcb *unp; 848 int error; 849 850 MPASS(so->so_type != SOCK_DGRAM); 851 852 /* 853 * Synchronize with concurrent connection attempts. 854 */ 855 error = 0; 856 unp = sotounpcb(so); 857 UNP_PCB_LOCK(unp); 858 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 859 error = EINVAL; 860 else if (unp->unp_vnode == NULL) 861 error = EDESTADDRREQ; 862 if (error != 0) { 863 UNP_PCB_UNLOCK(unp); 864 return (error); 865 } 866 867 SOCK_LOCK(so); 868 error = solisten_proto_check(so); 869 if (error == 0) { 870 cru2xt(td, &unp->unp_peercred); 871 solisten_proto(so, backlog); 872 } 873 SOCK_UNLOCK(so); 874 UNP_PCB_UNLOCK(unp); 875 return (error); 876 } 877 878 static int 879 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 880 { 881 struct unpcb *unp, *unp2; 882 const struct sockaddr *sa; 883 884 unp = sotounpcb(so); 885 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 886 887 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 888 UNP_LINK_RLOCK(); 889 /* 890 * XXX: It seems that this test always fails even when connection is 891 * established. So, this else clause is added as workaround to 892 * return PF_LOCAL sockaddr. 893 */ 894 unp2 = unp->unp_conn; 895 if (unp2 != NULL) { 896 UNP_PCB_LOCK(unp2); 897 if (unp2->unp_addr != NULL) 898 sa = (struct sockaddr *) unp2->unp_addr; 899 else 900 sa = &sun_noname; 901 bcopy(sa, *nam, sa->sa_len); 902 UNP_PCB_UNLOCK(unp2); 903 } else { 904 sa = &sun_noname; 905 bcopy(sa, *nam, sa->sa_len); 906 } 907 UNP_LINK_RUNLOCK(); 908 return (0); 909 } 910 911 static int 912 uipc_rcvd(struct socket *so, int flags) 913 { 914 struct unpcb *unp, *unp2; 915 struct socket *so2; 916 u_int mbcnt, sbcc; 917 918 unp = sotounpcb(so); 919 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 920 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 921 ("%s: socktype %d", __func__, so->so_type)); 922 923 /* 924 * Adjust backpressure on sender and wakeup any waiting to write. 925 * 926 * The unp lock is acquired to maintain the validity of the unp_conn 927 * pointer; no lock on unp2 is required as unp2->unp_socket will be 928 * static as long as we don't permit unp2 to disconnect from unp, 929 * which is prevented by the lock on unp. We cache values from 930 * so_rcv to avoid holding the so_rcv lock over the entire 931 * transaction on the remote so_snd. 932 */ 933 SOCKBUF_LOCK(&so->so_rcv); 934 mbcnt = so->so_rcv.sb_mbcnt; 935 sbcc = sbavail(&so->so_rcv); 936 SOCKBUF_UNLOCK(&so->so_rcv); 937 /* 938 * There is a benign race condition at this point. If we're planning to 939 * clear SB_STOP, but uipc_send is called on the connected socket at 940 * this instant, it might add data to the sockbuf and set SB_STOP. Then 941 * we would erroneously clear SB_STOP below, even though the sockbuf is 942 * full. The race is benign because the only ill effect is to allow the 943 * sockbuf to exceed its size limit, and the size limits are not 944 * strictly guaranteed anyway. 945 */ 946 UNP_PCB_LOCK(unp); 947 unp2 = unp->unp_conn; 948 if (unp2 == NULL) { 949 UNP_PCB_UNLOCK(unp); 950 return (0); 951 } 952 so2 = unp2->unp_socket; 953 SOCKBUF_LOCK(&so2->so_snd); 954 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 955 so2->so_snd.sb_flags &= ~SB_STOP; 956 sowwakeup_locked(so2); 957 UNP_PCB_UNLOCK(unp); 958 return (0); 959 } 960 961 static int 962 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 963 struct mbuf *control, struct thread *td) 964 { 965 struct unpcb *unp, *unp2; 966 struct socket *so2; 967 u_int mbcnt, sbcc; 968 int error; 969 970 unp = sotounpcb(so); 971 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 972 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 973 ("%s: socktype %d", __func__, so->so_type)); 974 975 error = 0; 976 if (flags & PRUS_OOB) { 977 error = EOPNOTSUPP; 978 goto release; 979 } 980 if (control != NULL && 981 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 982 goto release; 983 984 unp2 = NULL; 985 if ((so->so_state & SS_ISCONNECTED) == 0) { 986 if (nam != NULL) { 987 if ((error = unp_connect(so, nam, td)) != 0) 988 goto out; 989 } else { 990 error = ENOTCONN; 991 goto out; 992 } 993 } 994 995 UNP_PCB_LOCK(unp); 996 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 997 UNP_PCB_UNLOCK(unp); 998 error = ENOTCONN; 999 goto out; 1000 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1001 unp_pcb_unlock_pair(unp, unp2); 1002 error = EPIPE; 1003 goto out; 1004 } 1005 UNP_PCB_UNLOCK(unp); 1006 if ((so2 = unp2->unp_socket) == NULL) { 1007 UNP_PCB_UNLOCK(unp2); 1008 error = ENOTCONN; 1009 goto out; 1010 } 1011 SOCKBUF_LOCK(&so2->so_rcv); 1012 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 1013 /* 1014 * Credentials are passed only once on SOCK_STREAM and 1015 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 1016 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 1017 */ 1018 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 1019 NULL, NULL); 1020 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 1021 } 1022 1023 /* 1024 * Send to paired receive port and wake up readers. Don't 1025 * check for space available in the receive buffer if we're 1026 * attaching ancillary data; Unix domain sockets only check 1027 * for space in the sending sockbuf, and that check is 1028 * performed one level up the stack. At that level we cannot 1029 * precisely account for the amount of buffer space used 1030 * (e.g., because control messages are not yet internalized). 1031 */ 1032 switch (so->so_type) { 1033 case SOCK_STREAM: 1034 if (control != NULL) { 1035 sbappendcontrol_locked(&so2->so_rcv, m, 1036 control, flags); 1037 control = NULL; 1038 } else 1039 sbappend_locked(&so2->so_rcv, m, flags); 1040 break; 1041 1042 case SOCK_SEQPACKET: 1043 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1044 &sun_noname, m, control)) 1045 control = NULL; 1046 break; 1047 } 1048 1049 mbcnt = so2->so_rcv.sb_mbcnt; 1050 sbcc = sbavail(&so2->so_rcv); 1051 if (sbcc) 1052 sorwakeup_locked(so2); 1053 else 1054 SOCKBUF_UNLOCK(&so2->so_rcv); 1055 1056 /* 1057 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1058 * it would be possible for uipc_rcvd to be called at this 1059 * point, drain the receiving sockbuf, clear SB_STOP, and then 1060 * we would set SB_STOP below. That could lead to an empty 1061 * sockbuf having SB_STOP set 1062 */ 1063 SOCKBUF_LOCK(&so->so_snd); 1064 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1065 so->so_snd.sb_flags |= SB_STOP; 1066 SOCKBUF_UNLOCK(&so->so_snd); 1067 UNP_PCB_UNLOCK(unp2); 1068 m = NULL; 1069 out: 1070 /* 1071 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1072 */ 1073 if (flags & PRUS_EOF) { 1074 UNP_PCB_LOCK(unp); 1075 socantsendmore(so); 1076 unp_shutdown(unp); 1077 UNP_PCB_UNLOCK(unp); 1078 } 1079 if (control != NULL && error != 0) 1080 unp_scan(control, unp_freerights); 1081 1082 release: 1083 if (control != NULL) 1084 m_freem(control); 1085 /* 1086 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1087 * for freeing memory. 1088 */ 1089 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1090 m_freem(m); 1091 return (error); 1092 } 1093 1094 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1095 static inline bool 1096 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1097 { 1098 u_int bleft, mleft; 1099 1100 /* 1101 * Negative space may happen if send(2) is followed by 1102 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1103 */ 1104 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1105 sb->sb_mbmax < sb->uxdg_mbcnt)) 1106 return (false); 1107 1108 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1109 return (false); 1110 1111 bleft = sb->sb_hiwat - sb->uxdg_cc; 1112 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1113 1114 return (bleft >= cc && mleft >= mbcnt); 1115 } 1116 1117 /* 1118 * PF_UNIX/SOCK_DGRAM send 1119 * 1120 * Allocate a record consisting of 3 mbufs in the sequence of 1121 * from -> control -> data and append it to the socket buffer. 1122 * 1123 * The first mbuf carries sender's name and is a pkthdr that stores 1124 * overall length of datagram, its memory consumption and control length. 1125 */ 1126 #define ctllen PH_loc.thirtytwo[1] 1127 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1128 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1129 static int 1130 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1131 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1132 { 1133 struct unpcb *unp, *unp2; 1134 const struct sockaddr *from; 1135 struct socket *so2; 1136 struct sockbuf *sb; 1137 struct mbuf *f, *clast; 1138 u_int cc, ctl, mbcnt; 1139 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1140 int error; 1141 1142 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1143 1144 error = 0; 1145 f = NULL; 1146 ctl = 0; 1147 1148 if (__predict_false(flags & MSG_OOB)) { 1149 error = EOPNOTSUPP; 1150 goto out; 1151 } 1152 if (m == NULL) { 1153 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1154 error = EMSGSIZE; 1155 goto out; 1156 } 1157 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1158 if (__predict_false(m == NULL)) { 1159 error = EFAULT; 1160 goto out; 1161 } 1162 f = m_gethdr(M_WAITOK, MT_SONAME); 1163 cc = m->m_pkthdr.len; 1164 mbcnt = MSIZE + m->m_pkthdr.memlen; 1165 if (c != NULL && 1166 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1167 goto out; 1168 } else { 1169 /* pr_sosend() with mbuf usually is a kernel thread. */ 1170 1171 M_ASSERTPKTHDR(m); 1172 if (__predict_false(c != NULL)) 1173 panic("%s: control from a kernel thread", __func__); 1174 1175 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1176 error = EMSGSIZE; 1177 goto out; 1178 } 1179 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1180 error = ENOBUFS; 1181 goto out; 1182 } 1183 /* Condition the foreign mbuf to our standards. */ 1184 m_clrprotoflags(m); 1185 m_tag_delete_chain(m, NULL); 1186 m->m_pkthdr.rcvif = NULL; 1187 m->m_pkthdr.flowid = 0; 1188 m->m_pkthdr.csum_flags = 0; 1189 m->m_pkthdr.fibnum = 0; 1190 m->m_pkthdr.rsstype = 0; 1191 1192 cc = m->m_pkthdr.len; 1193 mbcnt = MSIZE; 1194 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1195 mbcnt += MSIZE; 1196 if (mb->m_flags & M_EXT) 1197 mbcnt += mb->m_ext.ext_size; 1198 } 1199 } 1200 1201 unp = sotounpcb(so); 1202 MPASS(unp); 1203 1204 /* 1205 * XXXGL: would be cool to fully remove so_snd out of the equation 1206 * and avoid this lock, which is not only extraneous, but also being 1207 * released, thus still leaving possibility for a race. We can easily 1208 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1209 * is more difficult to invent something to handle so_error. 1210 */ 1211 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1212 if (error) 1213 goto out2; 1214 SOCK_SENDBUF_LOCK(so); 1215 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1216 SOCK_SENDBUF_UNLOCK(so); 1217 error = EPIPE; 1218 goto out3; 1219 } 1220 if (so->so_error != 0) { 1221 error = so->so_error; 1222 so->so_error = 0; 1223 SOCK_SENDBUF_UNLOCK(so); 1224 goto out3; 1225 } 1226 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1227 SOCK_SENDBUF_UNLOCK(so); 1228 error = EDESTADDRREQ; 1229 goto out3; 1230 } 1231 SOCK_SENDBUF_UNLOCK(so); 1232 1233 if (addr != NULL) { 1234 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1235 goto out3; 1236 UNP_PCB_LOCK_ASSERT(unp); 1237 unp2 = unp->unp_conn; 1238 UNP_PCB_LOCK_ASSERT(unp2); 1239 } else { 1240 UNP_PCB_LOCK(unp); 1241 unp2 = unp_pcb_lock_peer(unp); 1242 if (unp2 == NULL) { 1243 UNP_PCB_UNLOCK(unp); 1244 error = ENOTCONN; 1245 goto out3; 1246 } 1247 } 1248 1249 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1250 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1251 &mbcnt); 1252 if (unp->unp_addr != NULL) 1253 from = (struct sockaddr *)unp->unp_addr; 1254 else 1255 from = &sun_noname; 1256 f->m_len = from->sa_len; 1257 MPASS(from->sa_len <= MLEN); 1258 bcopy(from, mtod(f, void *), from->sa_len); 1259 ctl += f->m_len; 1260 1261 /* 1262 * Concatenate mbufs: from -> control -> data. 1263 * Save overall cc and mbcnt in "from" mbuf. 1264 */ 1265 if (c != NULL) { 1266 #ifdef INVARIANTS 1267 struct mbuf *mc; 1268 1269 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1270 MPASS(mc == clast); 1271 #endif 1272 f->m_next = c; 1273 clast->m_next = m; 1274 c = NULL; 1275 } else 1276 f->m_next = m; 1277 m = NULL; 1278 #ifdef INVARIANTS 1279 dcc = dctl = dmbcnt = 0; 1280 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1281 if (mb->m_type == MT_DATA) 1282 dcc += mb->m_len; 1283 else 1284 dctl += mb->m_len; 1285 dmbcnt += MSIZE; 1286 if (mb->m_flags & M_EXT) 1287 dmbcnt += mb->m_ext.ext_size; 1288 } 1289 MPASS(dcc == cc); 1290 MPASS(dctl == ctl); 1291 MPASS(dmbcnt == mbcnt); 1292 #endif 1293 f->m_pkthdr.len = cc + ctl; 1294 f->m_pkthdr.memlen = mbcnt; 1295 f->m_pkthdr.ctllen = ctl; 1296 1297 /* 1298 * Destination socket buffer selection. 1299 * 1300 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1301 * destination address is supplied, create a temporary connection for 1302 * the run time of the function (see call to unp_connectat() above and 1303 * to unp_disconnect() below). We distinguish them by condition of 1304 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1305 * that condition, since, again, through the run time of this code we 1306 * are always connected. For such "unconnected" sends, the destination 1307 * buffer would be the receive buffer of destination socket so2. 1308 * 1309 * For connected sends, data lands on the send buffer of the sender's 1310 * socket "so". Then, if we just added the very first datagram 1311 * on this send buffer, we need to add the send buffer on to the 1312 * receiving socket's buffer list. We put ourselves on top of the 1313 * list. Such logic gives infrequent senders priority over frequent 1314 * senders. 1315 * 1316 * Note on byte count management. As long as event methods kevent(2), 1317 * select(2) are not protocol specific (yet), we need to maintain 1318 * meaningful values on the receive buffer. So, the receive buffer 1319 * would accumulate counters from all connected buffers potentially 1320 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1321 */ 1322 so2 = unp2->unp_socket; 1323 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1324 SOCK_RECVBUF_LOCK(so2); 1325 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1326 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1327 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1328 uxdg_clist); 1329 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1330 sb->uxdg_cc += cc + ctl; 1331 sb->uxdg_ctl += ctl; 1332 sb->uxdg_mbcnt += mbcnt; 1333 so2->so_rcv.sb_acc += cc + ctl; 1334 so2->so_rcv.sb_ccc += cc + ctl; 1335 so2->so_rcv.sb_ctl += ctl; 1336 so2->so_rcv.sb_mbcnt += mbcnt; 1337 sorwakeup_locked(so2); 1338 f = NULL; 1339 } else { 1340 soroverflow_locked(so2); 1341 error = ENOBUFS; 1342 if (f->m_next->m_type == MT_CONTROL) 1343 unp_scan(f->m_next, unp_freerights); 1344 } 1345 1346 if (addr != NULL) 1347 unp_disconnect(unp, unp2); 1348 else 1349 unp_pcb_unlock_pair(unp, unp2); 1350 1351 td->td_ru.ru_msgsnd++; 1352 1353 out3: 1354 SOCK_IO_SEND_UNLOCK(so); 1355 out2: 1356 if (c) 1357 unp_scan(c, unp_freerights); 1358 out: 1359 if (f) 1360 m_freem(f); 1361 if (c) 1362 m_freem(c); 1363 if (m) 1364 m_freem(m); 1365 1366 return (error); 1367 } 1368 1369 /* 1370 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1371 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1372 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1373 */ 1374 static int 1375 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1376 struct uio *uio, struct mbuf **controlp, int *flagsp) 1377 { 1378 ssize_t len = 0; 1379 int error; 1380 1381 so->so_rcv.uxdg_peeked = m; 1382 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1383 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1384 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1385 SOCK_RECVBUF_UNLOCK(so); 1386 1387 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1388 if (psa != NULL) 1389 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1390 1391 m = m->m_next; 1392 KASSERT(m, ("%s: no data or control after soname", __func__)); 1393 1394 /* 1395 * With MSG_PEEK the control isn't executed, just copied. 1396 */ 1397 while (m != NULL && m->m_type == MT_CONTROL) { 1398 if (controlp != NULL) { 1399 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1400 controlp = &(*controlp)->m_next; 1401 } 1402 m = m->m_next; 1403 } 1404 KASSERT(m == NULL || m->m_type == MT_DATA, 1405 ("%s: not MT_DATA mbuf %p", __func__, m)); 1406 while (m != NULL && uio->uio_resid > 0) { 1407 len = uio->uio_resid; 1408 if (len > m->m_len) 1409 len = m->m_len; 1410 error = uiomove(mtod(m, char *), (int)len, uio); 1411 if (error) { 1412 SOCK_IO_RECV_UNLOCK(so); 1413 return (error); 1414 } 1415 if (len == m->m_len) 1416 m = m->m_next; 1417 } 1418 SOCK_IO_RECV_UNLOCK(so); 1419 1420 if (flagsp != NULL) { 1421 if (m != NULL) { 1422 if (*flagsp & MSG_TRUNC) { 1423 /* Report real length of the packet */ 1424 uio->uio_resid -= m_length(m, NULL) - len; 1425 } 1426 *flagsp |= MSG_TRUNC; 1427 } else 1428 *flagsp &= ~MSG_TRUNC; 1429 } 1430 1431 return (0); 1432 } 1433 1434 /* 1435 * PF_UNIX/SOCK_DGRAM receive 1436 */ 1437 static int 1438 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1439 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1440 { 1441 struct sockbuf *sb = NULL; 1442 struct mbuf *m; 1443 int flags, error; 1444 ssize_t len = 0; 1445 bool nonblock; 1446 1447 MPASS(mp0 == NULL); 1448 1449 if (psa != NULL) 1450 *psa = NULL; 1451 if (controlp != NULL) 1452 *controlp = NULL; 1453 1454 flags = flagsp != NULL ? *flagsp : 0; 1455 nonblock = (so->so_state & SS_NBIO) || 1456 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1457 1458 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1459 if (__predict_false(error)) 1460 return (error); 1461 1462 /* 1463 * Loop blocking while waiting for a datagram. Prioritize connected 1464 * peers over unconnected sends. Set sb to selected socket buffer 1465 * containing an mbuf on exit from the wait loop. A datagram that 1466 * had already been peeked at has top priority. 1467 */ 1468 SOCK_RECVBUF_LOCK(so); 1469 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1470 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1471 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1472 if (so->so_error) { 1473 error = so->so_error; 1474 so->so_error = 0; 1475 SOCK_RECVBUF_UNLOCK(so); 1476 SOCK_IO_RECV_UNLOCK(so); 1477 return (error); 1478 } 1479 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1480 uio->uio_resid == 0) { 1481 SOCK_RECVBUF_UNLOCK(so); 1482 SOCK_IO_RECV_UNLOCK(so); 1483 return (0); 1484 } 1485 if (nonblock) { 1486 SOCK_RECVBUF_UNLOCK(so); 1487 SOCK_IO_RECV_UNLOCK(so); 1488 return (EWOULDBLOCK); 1489 } 1490 error = sbwait(so, SO_RCV); 1491 if (error) { 1492 SOCK_RECVBUF_UNLOCK(so); 1493 SOCK_IO_RECV_UNLOCK(so); 1494 return (error); 1495 } 1496 } 1497 1498 if (sb == NULL) 1499 sb = &so->so_rcv; 1500 else if (m == NULL) 1501 m = STAILQ_FIRST(&sb->uxdg_mb); 1502 else 1503 MPASS(m == so->so_rcv.uxdg_peeked); 1504 1505 MPASS(sb->uxdg_cc > 0); 1506 M_ASSERTPKTHDR(m); 1507 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1508 1509 if (uio->uio_td) 1510 uio->uio_td->td_ru.ru_msgrcv++; 1511 1512 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1513 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1514 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1515 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1516 } else 1517 so->so_rcv.uxdg_peeked = NULL; 1518 1519 sb->uxdg_cc -= m->m_pkthdr.len; 1520 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1521 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1522 1523 if (__predict_false(flags & MSG_PEEK)) 1524 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1525 1526 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1527 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1528 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1529 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1530 SOCK_RECVBUF_UNLOCK(so); 1531 1532 if (psa != NULL) 1533 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1534 m = m_free(m); 1535 KASSERT(m, ("%s: no data or control after soname", __func__)); 1536 1537 /* 1538 * Packet to copyout() is now in 'm' and it is disconnected from the 1539 * queue. 1540 * 1541 * Process one or more MT_CONTROL mbufs present before any data mbufs 1542 * in the first mbuf chain on the socket buffer. We call into the 1543 * unp_externalize() to perform externalization (or freeing if 1544 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1545 * without MT_DATA mbufs. 1546 */ 1547 while (m != NULL && m->m_type == MT_CONTROL) { 1548 struct mbuf *cm; 1549 1550 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1551 * it frees whole chain, so we must disconnect the mbuf. 1552 */ 1553 cm = m; m = m->m_next; cm->m_next = NULL; 1554 error = unp_externalize(cm, controlp, flags); 1555 if (error != 0) { 1556 SOCK_IO_RECV_UNLOCK(so); 1557 unp_scan(m, unp_freerights); 1558 m_freem(m); 1559 return (error); 1560 } 1561 if (controlp != NULL) { 1562 while (*controlp != NULL) 1563 controlp = &(*controlp)->m_next; 1564 } 1565 } 1566 KASSERT(m == NULL || m->m_type == MT_DATA, 1567 ("%s: not MT_DATA mbuf %p", __func__, m)); 1568 while (m != NULL && uio->uio_resid > 0) { 1569 len = uio->uio_resid; 1570 if (len > m->m_len) 1571 len = m->m_len; 1572 error = uiomove(mtod(m, char *), (int)len, uio); 1573 if (error) { 1574 SOCK_IO_RECV_UNLOCK(so); 1575 m_freem(m); 1576 return (error); 1577 } 1578 if (len == m->m_len) 1579 m = m_free(m); 1580 else { 1581 m->m_data += len; 1582 m->m_len -= len; 1583 } 1584 } 1585 SOCK_IO_RECV_UNLOCK(so); 1586 1587 if (m != NULL) { 1588 if (flagsp != NULL) { 1589 if (flags & MSG_TRUNC) { 1590 /* Report real length of the packet */ 1591 uio->uio_resid -= m_length(m, NULL); 1592 } 1593 *flagsp |= MSG_TRUNC; 1594 } 1595 m_freem(m); 1596 } else if (flagsp != NULL) 1597 *flagsp &= ~MSG_TRUNC; 1598 1599 return (0); 1600 } 1601 1602 static bool 1603 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1604 { 1605 struct mbuf *mb, *n; 1606 struct sockbuf *sb; 1607 1608 SOCK_LOCK(so); 1609 if (SOLISTENING(so)) { 1610 SOCK_UNLOCK(so); 1611 return (false); 1612 } 1613 mb = NULL; 1614 sb = &so->so_rcv; 1615 SOCKBUF_LOCK(sb); 1616 if (sb->sb_fnrdy != NULL) { 1617 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1618 if (mb == m) { 1619 *errorp = sbready(sb, m, count); 1620 break; 1621 } 1622 mb = mb->m_next; 1623 if (mb == NULL) { 1624 mb = n; 1625 if (mb != NULL) 1626 n = mb->m_nextpkt; 1627 } 1628 } 1629 } 1630 SOCKBUF_UNLOCK(sb); 1631 SOCK_UNLOCK(so); 1632 return (mb != NULL); 1633 } 1634 1635 static int 1636 uipc_ready(struct socket *so, struct mbuf *m, int count) 1637 { 1638 struct unpcb *unp, *unp2; 1639 struct socket *so2; 1640 int error, i; 1641 1642 unp = sotounpcb(so); 1643 1644 KASSERT(so->so_type == SOCK_STREAM, 1645 ("%s: unexpected socket type for %p", __func__, so)); 1646 1647 UNP_PCB_LOCK(unp); 1648 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1649 UNP_PCB_UNLOCK(unp); 1650 so2 = unp2->unp_socket; 1651 SOCKBUF_LOCK(&so2->so_rcv); 1652 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1653 sorwakeup_locked(so2); 1654 else 1655 SOCKBUF_UNLOCK(&so2->so_rcv); 1656 UNP_PCB_UNLOCK(unp2); 1657 return (error); 1658 } 1659 UNP_PCB_UNLOCK(unp); 1660 1661 /* 1662 * The receiving socket has been disconnected, but may still be valid. 1663 * In this case, the now-ready mbufs are still present in its socket 1664 * buffer, so perform an exhaustive search before giving up and freeing 1665 * the mbufs. 1666 */ 1667 UNP_LINK_RLOCK(); 1668 LIST_FOREACH(unp, &unp_shead, unp_link) { 1669 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1670 break; 1671 } 1672 UNP_LINK_RUNLOCK(); 1673 1674 if (unp == NULL) { 1675 for (i = 0; i < count; i++) 1676 m = m_free(m); 1677 error = ECONNRESET; 1678 } 1679 return (error); 1680 } 1681 1682 static int 1683 uipc_sense(struct socket *so, struct stat *sb) 1684 { 1685 struct unpcb *unp; 1686 1687 unp = sotounpcb(so); 1688 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1689 1690 sb->st_blksize = so->so_snd.sb_hiwat; 1691 sb->st_dev = NODEV; 1692 sb->st_ino = unp->unp_ino; 1693 return (0); 1694 } 1695 1696 static int 1697 uipc_shutdown(struct socket *so) 1698 { 1699 struct unpcb *unp; 1700 1701 unp = sotounpcb(so); 1702 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1703 1704 UNP_PCB_LOCK(unp); 1705 socantsendmore(so); 1706 unp_shutdown(unp); 1707 UNP_PCB_UNLOCK(unp); 1708 return (0); 1709 } 1710 1711 static int 1712 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1713 { 1714 struct unpcb *unp; 1715 const struct sockaddr *sa; 1716 1717 unp = sotounpcb(so); 1718 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1719 1720 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1721 UNP_PCB_LOCK(unp); 1722 if (unp->unp_addr != NULL) 1723 sa = (struct sockaddr *) unp->unp_addr; 1724 else 1725 sa = &sun_noname; 1726 bcopy(sa, *nam, sa->sa_len); 1727 UNP_PCB_UNLOCK(unp); 1728 return (0); 1729 } 1730 1731 static int 1732 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1733 { 1734 struct unpcb *unp; 1735 struct xucred xu; 1736 int error, optval; 1737 1738 if (sopt->sopt_level != SOL_LOCAL) 1739 return (EINVAL); 1740 1741 unp = sotounpcb(so); 1742 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1743 error = 0; 1744 switch (sopt->sopt_dir) { 1745 case SOPT_GET: 1746 switch (sopt->sopt_name) { 1747 case LOCAL_PEERCRED: 1748 UNP_PCB_LOCK(unp); 1749 if (unp->unp_flags & UNP_HAVEPC) 1750 xu = unp->unp_peercred; 1751 else { 1752 if (so->so_type == SOCK_STREAM) 1753 error = ENOTCONN; 1754 else 1755 error = EINVAL; 1756 } 1757 UNP_PCB_UNLOCK(unp); 1758 if (error == 0) 1759 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1760 break; 1761 1762 case LOCAL_CREDS: 1763 /* Unlocked read. */ 1764 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1765 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1766 break; 1767 1768 case LOCAL_CREDS_PERSISTENT: 1769 /* Unlocked read. */ 1770 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1771 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1772 break; 1773 1774 case LOCAL_CONNWAIT: 1775 /* Unlocked read. */ 1776 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1777 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1778 break; 1779 1780 default: 1781 error = EOPNOTSUPP; 1782 break; 1783 } 1784 break; 1785 1786 case SOPT_SET: 1787 switch (sopt->sopt_name) { 1788 case LOCAL_CREDS: 1789 case LOCAL_CREDS_PERSISTENT: 1790 case LOCAL_CONNWAIT: 1791 error = sooptcopyin(sopt, &optval, sizeof(optval), 1792 sizeof(optval)); 1793 if (error) 1794 break; 1795 1796 #define OPTSET(bit, exclusive) do { \ 1797 UNP_PCB_LOCK(unp); \ 1798 if (optval) { \ 1799 if ((unp->unp_flags & (exclusive)) != 0) { \ 1800 UNP_PCB_UNLOCK(unp); \ 1801 error = EINVAL; \ 1802 break; \ 1803 } \ 1804 unp->unp_flags |= (bit); \ 1805 } else \ 1806 unp->unp_flags &= ~(bit); \ 1807 UNP_PCB_UNLOCK(unp); \ 1808 } while (0) 1809 1810 switch (sopt->sopt_name) { 1811 case LOCAL_CREDS: 1812 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1813 break; 1814 1815 case LOCAL_CREDS_PERSISTENT: 1816 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1817 break; 1818 1819 case LOCAL_CONNWAIT: 1820 OPTSET(UNP_CONNWAIT, 0); 1821 break; 1822 1823 default: 1824 break; 1825 } 1826 break; 1827 #undef OPTSET 1828 default: 1829 error = ENOPROTOOPT; 1830 break; 1831 } 1832 break; 1833 1834 default: 1835 error = EOPNOTSUPP; 1836 break; 1837 } 1838 return (error); 1839 } 1840 1841 static int 1842 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1843 { 1844 1845 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1846 } 1847 1848 static int 1849 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1850 struct thread *td, bool return_locked) 1851 { 1852 struct mtx *vplock; 1853 struct sockaddr_un *soun; 1854 struct vnode *vp; 1855 struct socket *so2; 1856 struct unpcb *unp, *unp2, *unp3; 1857 struct nameidata nd; 1858 char buf[SOCK_MAXADDRLEN]; 1859 struct sockaddr *sa; 1860 cap_rights_t rights; 1861 int error, len; 1862 bool connreq; 1863 1864 if (nam->sa_family != AF_UNIX) 1865 return (EAFNOSUPPORT); 1866 if (nam->sa_len > sizeof(struct sockaddr_un)) 1867 return (EINVAL); 1868 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1869 if (len <= 0) 1870 return (EINVAL); 1871 soun = (struct sockaddr_un *)nam; 1872 bcopy(soun->sun_path, buf, len); 1873 buf[len] = 0; 1874 1875 error = 0; 1876 unp = sotounpcb(so); 1877 UNP_PCB_LOCK(unp); 1878 for (;;) { 1879 /* 1880 * Wait for connection state to stabilize. If a connection 1881 * already exists, give up. For datagram sockets, which permit 1882 * multiple consecutive connect(2) calls, upper layers are 1883 * responsible for disconnecting in advance of a subsequent 1884 * connect(2), but this is not synchronized with PCB connection 1885 * state. 1886 * 1887 * Also make sure that no threads are currently attempting to 1888 * lock the peer socket, to ensure that unp_conn cannot 1889 * transition between two valid sockets while locks are dropped. 1890 */ 1891 if (SOLISTENING(so)) 1892 error = EOPNOTSUPP; 1893 else if (unp->unp_conn != NULL) 1894 error = EISCONN; 1895 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1896 error = EALREADY; 1897 } 1898 if (error != 0) { 1899 UNP_PCB_UNLOCK(unp); 1900 return (error); 1901 } 1902 if (unp->unp_pairbusy > 0) { 1903 unp->unp_flags |= UNP_WAITING; 1904 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1905 continue; 1906 } 1907 break; 1908 } 1909 unp->unp_flags |= UNP_CONNECTING; 1910 UNP_PCB_UNLOCK(unp); 1911 1912 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1913 if (connreq) 1914 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1915 else 1916 sa = NULL; 1917 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1918 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1919 error = namei(&nd); 1920 if (error) 1921 vp = NULL; 1922 else 1923 vp = nd.ni_vp; 1924 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1925 if (error) 1926 goto bad; 1927 NDFREE_PNBUF(&nd); 1928 1929 if (vp->v_type != VSOCK) { 1930 error = ENOTSOCK; 1931 goto bad; 1932 } 1933 #ifdef MAC 1934 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1935 if (error) 1936 goto bad; 1937 #endif 1938 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1939 if (error) 1940 goto bad; 1941 1942 unp = sotounpcb(so); 1943 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1944 1945 vplock = mtx_pool_find(mtxpool_sleep, vp); 1946 mtx_lock(vplock); 1947 VOP_UNP_CONNECT(vp, &unp2); 1948 if (unp2 == NULL) { 1949 error = ECONNREFUSED; 1950 goto bad2; 1951 } 1952 so2 = unp2->unp_socket; 1953 if (so->so_type != so2->so_type) { 1954 error = EPROTOTYPE; 1955 goto bad2; 1956 } 1957 if (connreq) { 1958 if (SOLISTENING(so2)) { 1959 CURVNET_SET(so2->so_vnet); 1960 so2 = sonewconn(so2, 0); 1961 CURVNET_RESTORE(); 1962 } else 1963 so2 = NULL; 1964 if (so2 == NULL) { 1965 error = ECONNREFUSED; 1966 goto bad2; 1967 } 1968 unp3 = sotounpcb(so2); 1969 unp_pcb_lock_pair(unp2, unp3); 1970 if (unp2->unp_addr != NULL) { 1971 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1972 unp3->unp_addr = (struct sockaddr_un *) sa; 1973 sa = NULL; 1974 } 1975 1976 unp_copy_peercred(td, unp3, unp, unp2); 1977 1978 UNP_PCB_UNLOCK(unp2); 1979 unp2 = unp3; 1980 1981 /* 1982 * It is safe to block on the PCB lock here since unp2 is 1983 * nascent and cannot be connected to any other sockets. 1984 */ 1985 UNP_PCB_LOCK(unp); 1986 #ifdef MAC 1987 mac_socketpeer_set_from_socket(so, so2); 1988 mac_socketpeer_set_from_socket(so2, so); 1989 #endif 1990 } else { 1991 unp_pcb_lock_pair(unp, unp2); 1992 } 1993 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1994 sotounpcb(so2) == unp2, 1995 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1996 unp_connect2(so, so2, PRU_CONNECT); 1997 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1998 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1999 unp->unp_flags &= ~UNP_CONNECTING; 2000 if (!return_locked) 2001 unp_pcb_unlock_pair(unp, unp2); 2002 bad2: 2003 mtx_unlock(vplock); 2004 bad: 2005 if (vp != NULL) { 2006 /* 2007 * If we are returning locked (called via uipc_sosend_dgram()), 2008 * we need to be sure that vput() won't sleep. This is 2009 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2010 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2011 */ 2012 MPASS(!(return_locked && connreq)); 2013 vput(vp); 2014 } 2015 free(sa, M_SONAME); 2016 if (__predict_false(error)) { 2017 UNP_PCB_LOCK(unp); 2018 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2019 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2020 unp->unp_flags &= ~UNP_CONNECTING; 2021 UNP_PCB_UNLOCK(unp); 2022 } 2023 return (error); 2024 } 2025 2026 /* 2027 * Set socket peer credentials at connection time. 2028 * 2029 * The client's PCB credentials are copied from its process structure. The 2030 * server's PCB credentials are copied from the socket on which it called 2031 * listen(2). uipc_listen cached that process's credentials at the time. 2032 */ 2033 void 2034 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2035 struct unpcb *server_unp, struct unpcb *listen_unp) 2036 { 2037 cru2xt(td, &client_unp->unp_peercred); 2038 client_unp->unp_flags |= UNP_HAVEPC; 2039 2040 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2041 sizeof(server_unp->unp_peercred)); 2042 server_unp->unp_flags |= UNP_HAVEPC; 2043 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2044 } 2045 2046 static void 2047 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2048 { 2049 struct unpcb *unp; 2050 struct unpcb *unp2; 2051 2052 MPASS(so2->so_type == so->so_type); 2053 unp = sotounpcb(so); 2054 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2055 unp2 = sotounpcb(so2); 2056 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2057 2058 UNP_PCB_LOCK_ASSERT(unp); 2059 UNP_PCB_LOCK_ASSERT(unp2); 2060 KASSERT(unp->unp_conn == NULL, 2061 ("%s: socket %p is already connected", __func__, unp)); 2062 2063 unp->unp_conn = unp2; 2064 unp_pcb_hold(unp2); 2065 unp_pcb_hold(unp); 2066 switch (so->so_type) { 2067 case SOCK_DGRAM: 2068 UNP_REF_LIST_LOCK(); 2069 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2070 UNP_REF_LIST_UNLOCK(); 2071 soisconnected(so); 2072 break; 2073 2074 case SOCK_STREAM: 2075 case SOCK_SEQPACKET: 2076 KASSERT(unp2->unp_conn == NULL, 2077 ("%s: socket %p is already connected", __func__, unp2)); 2078 unp2->unp_conn = unp; 2079 if (req == PRU_CONNECT && 2080 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2081 soisconnecting(so); 2082 else 2083 soisconnected(so); 2084 soisconnected(so2); 2085 break; 2086 2087 default: 2088 panic("unp_connect2"); 2089 } 2090 } 2091 2092 static void 2093 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2094 { 2095 struct socket *so, *so2; 2096 struct mbuf *m = NULL; 2097 #ifdef INVARIANTS 2098 struct unpcb *unptmp; 2099 #endif 2100 2101 UNP_PCB_LOCK_ASSERT(unp); 2102 UNP_PCB_LOCK_ASSERT(unp2); 2103 KASSERT(unp->unp_conn == unp2, 2104 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2105 2106 unp->unp_conn = NULL; 2107 so = unp->unp_socket; 2108 so2 = unp2->unp_socket; 2109 switch (unp->unp_socket->so_type) { 2110 case SOCK_DGRAM: 2111 /* 2112 * Remove our send socket buffer from the peer's receive buffer. 2113 * Move the data to the receive buffer only if it is empty. 2114 * This is a protection against a scenario where a peer 2115 * connects, floods and disconnects, effectively blocking 2116 * sendto() from unconnected sockets. 2117 */ 2118 SOCK_RECVBUF_LOCK(so2); 2119 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2120 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2121 uxdg_clist); 2122 if (__predict_true((so2->so_rcv.sb_state & 2123 SBS_CANTRCVMORE) == 0) && 2124 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2125 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2126 &so->so_snd.uxdg_mb); 2127 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2128 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2129 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2130 } else { 2131 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2132 STAILQ_INIT(&so->so_snd.uxdg_mb); 2133 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2134 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2135 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2136 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2137 } 2138 /* Note: so may reconnect. */ 2139 so->so_snd.uxdg_cc = 0; 2140 so->so_snd.uxdg_ctl = 0; 2141 so->so_snd.uxdg_mbcnt = 0; 2142 } 2143 SOCK_RECVBUF_UNLOCK(so2); 2144 UNP_REF_LIST_LOCK(); 2145 #ifdef INVARIANTS 2146 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2147 if (unptmp == unp) 2148 break; 2149 } 2150 KASSERT(unptmp != NULL, 2151 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2152 #endif 2153 LIST_REMOVE(unp, unp_reflink); 2154 UNP_REF_LIST_UNLOCK(); 2155 if (so) { 2156 SOCK_LOCK(so); 2157 so->so_state &= ~SS_ISCONNECTED; 2158 SOCK_UNLOCK(so); 2159 } 2160 break; 2161 2162 case SOCK_STREAM: 2163 case SOCK_SEQPACKET: 2164 if (so) 2165 soisdisconnected(so); 2166 MPASS(unp2->unp_conn == unp); 2167 unp2->unp_conn = NULL; 2168 if (so2) 2169 soisdisconnected(so2); 2170 break; 2171 } 2172 2173 if (unp == unp2) { 2174 unp_pcb_rele_notlast(unp); 2175 if (!unp_pcb_rele(unp)) 2176 UNP_PCB_UNLOCK(unp); 2177 } else { 2178 if (!unp_pcb_rele(unp)) 2179 UNP_PCB_UNLOCK(unp); 2180 if (!unp_pcb_rele(unp2)) 2181 UNP_PCB_UNLOCK(unp2); 2182 } 2183 2184 if (m != NULL) { 2185 unp_scan(m, unp_freerights); 2186 m_freem(m); 2187 } 2188 } 2189 2190 /* 2191 * unp_pcblist() walks the global list of struct unpcb's to generate a 2192 * pointer list, bumping the refcount on each unpcb. It then copies them out 2193 * sequentially, validating the generation number on each to see if it has 2194 * been detached. All of this is necessary because copyout() may sleep on 2195 * disk I/O. 2196 */ 2197 static int 2198 unp_pcblist(SYSCTL_HANDLER_ARGS) 2199 { 2200 struct unpcb *unp, **unp_list; 2201 unp_gen_t gencnt; 2202 struct xunpgen *xug; 2203 struct unp_head *head; 2204 struct xunpcb *xu; 2205 u_int i; 2206 int error, n; 2207 2208 switch ((intptr_t)arg1) { 2209 case SOCK_STREAM: 2210 head = &unp_shead; 2211 break; 2212 2213 case SOCK_DGRAM: 2214 head = &unp_dhead; 2215 break; 2216 2217 case SOCK_SEQPACKET: 2218 head = &unp_sphead; 2219 break; 2220 2221 default: 2222 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2223 } 2224 2225 /* 2226 * The process of preparing the PCB list is too time-consuming and 2227 * resource-intensive to repeat twice on every request. 2228 */ 2229 if (req->oldptr == NULL) { 2230 n = unp_count; 2231 req->oldidx = 2 * (sizeof *xug) 2232 + (n + n/8) * sizeof(struct xunpcb); 2233 return (0); 2234 } 2235 2236 if (req->newptr != NULL) 2237 return (EPERM); 2238 2239 /* 2240 * OK, now we're committed to doing something. 2241 */ 2242 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2243 UNP_LINK_RLOCK(); 2244 gencnt = unp_gencnt; 2245 n = unp_count; 2246 UNP_LINK_RUNLOCK(); 2247 2248 xug->xug_len = sizeof *xug; 2249 xug->xug_count = n; 2250 xug->xug_gen = gencnt; 2251 xug->xug_sogen = so_gencnt; 2252 error = SYSCTL_OUT(req, xug, sizeof *xug); 2253 if (error) { 2254 free(xug, M_TEMP); 2255 return (error); 2256 } 2257 2258 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2259 2260 UNP_LINK_RLOCK(); 2261 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2262 unp = LIST_NEXT(unp, unp_link)) { 2263 UNP_PCB_LOCK(unp); 2264 if (unp->unp_gencnt <= gencnt) { 2265 if (cr_cansee(req->td->td_ucred, 2266 unp->unp_socket->so_cred)) { 2267 UNP_PCB_UNLOCK(unp); 2268 continue; 2269 } 2270 unp_list[i++] = unp; 2271 unp_pcb_hold(unp); 2272 } 2273 UNP_PCB_UNLOCK(unp); 2274 } 2275 UNP_LINK_RUNLOCK(); 2276 n = i; /* In case we lost some during malloc. */ 2277 2278 error = 0; 2279 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2280 for (i = 0; i < n; i++) { 2281 unp = unp_list[i]; 2282 UNP_PCB_LOCK(unp); 2283 if (unp_pcb_rele(unp)) 2284 continue; 2285 2286 if (unp->unp_gencnt <= gencnt) { 2287 xu->xu_len = sizeof *xu; 2288 xu->xu_unpp = (uintptr_t)unp; 2289 /* 2290 * XXX - need more locking here to protect against 2291 * connect/disconnect races for SMP. 2292 */ 2293 if (unp->unp_addr != NULL) 2294 bcopy(unp->unp_addr, &xu->xu_addr, 2295 unp->unp_addr->sun_len); 2296 else 2297 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2298 if (unp->unp_conn != NULL && 2299 unp->unp_conn->unp_addr != NULL) 2300 bcopy(unp->unp_conn->unp_addr, 2301 &xu->xu_caddr, 2302 unp->unp_conn->unp_addr->sun_len); 2303 else 2304 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2305 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2306 xu->unp_conn = (uintptr_t)unp->unp_conn; 2307 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2308 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2309 xu->unp_gencnt = unp->unp_gencnt; 2310 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2311 UNP_PCB_UNLOCK(unp); 2312 error = SYSCTL_OUT(req, xu, sizeof *xu); 2313 } else { 2314 UNP_PCB_UNLOCK(unp); 2315 } 2316 } 2317 free(xu, M_TEMP); 2318 if (!error) { 2319 /* 2320 * Give the user an updated idea of our state. If the 2321 * generation differs from what we told her before, she knows 2322 * that something happened while we were processing this 2323 * request, and it might be necessary to retry. 2324 */ 2325 xug->xug_gen = unp_gencnt; 2326 xug->xug_sogen = so_gencnt; 2327 xug->xug_count = unp_count; 2328 error = SYSCTL_OUT(req, xug, sizeof *xug); 2329 } 2330 free(unp_list, M_TEMP); 2331 free(xug, M_TEMP); 2332 return (error); 2333 } 2334 2335 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2336 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2337 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2338 "List of active local datagram sockets"); 2339 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2340 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2341 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2342 "List of active local stream sockets"); 2343 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2344 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2345 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2346 "List of active local seqpacket sockets"); 2347 2348 static void 2349 unp_shutdown(struct unpcb *unp) 2350 { 2351 struct unpcb *unp2; 2352 struct socket *so; 2353 2354 UNP_PCB_LOCK_ASSERT(unp); 2355 2356 unp2 = unp->unp_conn; 2357 if ((unp->unp_socket->so_type == SOCK_STREAM || 2358 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2359 so = unp2->unp_socket; 2360 if (so != NULL) 2361 socantrcvmore(so); 2362 } 2363 } 2364 2365 static void 2366 unp_drop(struct unpcb *unp) 2367 { 2368 struct socket *so; 2369 struct unpcb *unp2; 2370 2371 /* 2372 * Regardless of whether the socket's peer dropped the connection 2373 * with this socket by aborting or disconnecting, POSIX requires 2374 * that ECONNRESET is returned. 2375 */ 2376 2377 UNP_PCB_LOCK(unp); 2378 so = unp->unp_socket; 2379 if (so) 2380 so->so_error = ECONNRESET; 2381 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2382 /* Last reference dropped in unp_disconnect(). */ 2383 unp_pcb_rele_notlast(unp); 2384 unp_disconnect(unp, unp2); 2385 } else if (!unp_pcb_rele(unp)) { 2386 UNP_PCB_UNLOCK(unp); 2387 } 2388 } 2389 2390 static void 2391 unp_freerights(struct filedescent **fdep, int fdcount) 2392 { 2393 struct file *fp; 2394 int i; 2395 2396 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2397 2398 for (i = 0; i < fdcount; i++) { 2399 fp = fdep[i]->fde_file; 2400 filecaps_free(&fdep[i]->fde_caps); 2401 unp_discard(fp); 2402 } 2403 free(fdep[0], M_FILECAPS); 2404 } 2405 2406 static int 2407 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2408 { 2409 struct thread *td = curthread; /* XXX */ 2410 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2411 int i; 2412 int *fdp; 2413 struct filedesc *fdesc = td->td_proc->p_fd; 2414 struct filedescent **fdep; 2415 void *data; 2416 socklen_t clen = control->m_len, datalen; 2417 int error, newfds; 2418 u_int newlen; 2419 2420 UNP_LINK_UNLOCK_ASSERT(); 2421 2422 error = 0; 2423 if (controlp != NULL) /* controlp == NULL => free control messages */ 2424 *controlp = NULL; 2425 while (cm != NULL) { 2426 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2427 2428 data = CMSG_DATA(cm); 2429 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2430 if (cm->cmsg_level == SOL_SOCKET 2431 && cm->cmsg_type == SCM_RIGHTS) { 2432 newfds = datalen / sizeof(*fdep); 2433 if (newfds == 0) 2434 goto next; 2435 fdep = data; 2436 2437 /* If we're not outputting the descriptors free them. */ 2438 if (error || controlp == NULL) { 2439 unp_freerights(fdep, newfds); 2440 goto next; 2441 } 2442 FILEDESC_XLOCK(fdesc); 2443 2444 /* 2445 * Now change each pointer to an fd in the global 2446 * table to an integer that is the index to the local 2447 * fd table entry that we set up to point to the 2448 * global one we are transferring. 2449 */ 2450 newlen = newfds * sizeof(int); 2451 *controlp = sbcreatecontrol(NULL, newlen, 2452 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2453 2454 fdp = (int *) 2455 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2456 if ((error = fdallocn(td, 0, fdp, newfds))) { 2457 FILEDESC_XUNLOCK(fdesc); 2458 unp_freerights(fdep, newfds); 2459 m_freem(*controlp); 2460 *controlp = NULL; 2461 goto next; 2462 } 2463 for (i = 0; i < newfds; i++, fdp++) { 2464 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2465 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2466 &fdep[i]->fde_caps); 2467 unp_externalize_fp(fdep[i]->fde_file); 2468 } 2469 2470 /* 2471 * The new type indicates that the mbuf data refers to 2472 * kernel resources that may need to be released before 2473 * the mbuf is freed. 2474 */ 2475 m_chtype(*controlp, MT_EXTCONTROL); 2476 FILEDESC_XUNLOCK(fdesc); 2477 free(fdep[0], M_FILECAPS); 2478 } else { 2479 /* We can just copy anything else across. */ 2480 if (error || controlp == NULL) 2481 goto next; 2482 *controlp = sbcreatecontrol(NULL, datalen, 2483 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2484 bcopy(data, 2485 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2486 datalen); 2487 } 2488 controlp = &(*controlp)->m_next; 2489 2490 next: 2491 if (CMSG_SPACE(datalen) < clen) { 2492 clen -= CMSG_SPACE(datalen); 2493 cm = (struct cmsghdr *) 2494 ((caddr_t)cm + CMSG_SPACE(datalen)); 2495 } else { 2496 clen = 0; 2497 cm = NULL; 2498 } 2499 } 2500 2501 m_freem(control); 2502 return (error); 2503 } 2504 2505 static void 2506 unp_zone_change(void *tag) 2507 { 2508 2509 uma_zone_set_max(unp_zone, maxsockets); 2510 } 2511 2512 #ifdef INVARIANTS 2513 static void 2514 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2515 { 2516 struct unpcb *unp; 2517 2518 unp = mem; 2519 2520 KASSERT(LIST_EMPTY(&unp->unp_refs), 2521 ("%s: unpcb %p has lingering refs", __func__, unp)); 2522 KASSERT(unp->unp_socket == NULL, 2523 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2524 KASSERT(unp->unp_vnode == NULL, 2525 ("%s: unpcb %p has vnode references", __func__, unp)); 2526 KASSERT(unp->unp_conn == NULL, 2527 ("%s: unpcb %p is still connected", __func__, unp)); 2528 KASSERT(unp->unp_addr == NULL, 2529 ("%s: unpcb %p has leaked addr", __func__, unp)); 2530 } 2531 #endif 2532 2533 static void 2534 unp_init(void *arg __unused) 2535 { 2536 uma_dtor dtor; 2537 2538 #ifdef INVARIANTS 2539 dtor = unp_zdtor; 2540 #else 2541 dtor = NULL; 2542 #endif 2543 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2544 NULL, NULL, UMA_ALIGN_CACHE, 0); 2545 uma_zone_set_max(unp_zone, maxsockets); 2546 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2547 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2548 NULL, EVENTHANDLER_PRI_ANY); 2549 LIST_INIT(&unp_dhead); 2550 LIST_INIT(&unp_shead); 2551 LIST_INIT(&unp_sphead); 2552 SLIST_INIT(&unp_defers); 2553 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2554 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2555 UNP_LINK_LOCK_INIT(); 2556 UNP_DEFERRED_LOCK_INIT(); 2557 } 2558 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2559 2560 static void 2561 unp_internalize_cleanup_rights(struct mbuf *control) 2562 { 2563 struct cmsghdr *cp; 2564 struct mbuf *m; 2565 void *data; 2566 socklen_t datalen; 2567 2568 for (m = control; m != NULL; m = m->m_next) { 2569 cp = mtod(m, struct cmsghdr *); 2570 if (cp->cmsg_level != SOL_SOCKET || 2571 cp->cmsg_type != SCM_RIGHTS) 2572 continue; 2573 data = CMSG_DATA(cp); 2574 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2575 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2576 } 2577 } 2578 2579 static int 2580 unp_internalize(struct mbuf **controlp, struct thread *td, 2581 struct mbuf **clast, u_int *space, u_int *mbcnt) 2582 { 2583 struct mbuf *control, **initial_controlp; 2584 struct proc *p; 2585 struct filedesc *fdesc; 2586 struct bintime *bt; 2587 struct cmsghdr *cm; 2588 struct cmsgcred *cmcred; 2589 struct filedescent *fde, **fdep, *fdev; 2590 struct file *fp; 2591 struct timeval *tv; 2592 struct timespec *ts; 2593 void *data; 2594 socklen_t clen, datalen; 2595 int i, j, error, *fdp, oldfds; 2596 u_int newlen; 2597 2598 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2599 UNP_LINK_UNLOCK_ASSERT(); 2600 2601 p = td->td_proc; 2602 fdesc = p->p_fd; 2603 error = 0; 2604 control = *controlp; 2605 *controlp = NULL; 2606 initial_controlp = controlp; 2607 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2608 data = CMSG_DATA(cm); 2609 2610 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2611 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2612 (char *)cm + cm->cmsg_len >= (char *)data; 2613 2614 clen -= min(CMSG_SPACE(datalen), clen), 2615 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2616 data = CMSG_DATA(cm)) { 2617 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2618 switch (cm->cmsg_type) { 2619 case SCM_CREDS: 2620 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2621 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2622 cmcred = (struct cmsgcred *) 2623 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2624 cmcred->cmcred_pid = p->p_pid; 2625 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2626 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2627 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2628 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2629 CMGROUP_MAX); 2630 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2631 cmcred->cmcred_groups[i] = 2632 td->td_ucred->cr_groups[i]; 2633 break; 2634 2635 case SCM_RIGHTS: 2636 oldfds = datalen / sizeof (int); 2637 if (oldfds == 0) 2638 continue; 2639 /* On some machines sizeof pointer is bigger than 2640 * sizeof int, so we need to check if data fits into 2641 * single mbuf. We could allocate several mbufs, and 2642 * unp_externalize() should even properly handle that. 2643 * But it is not worth to complicate the code for an 2644 * insane scenario of passing over 200 file descriptors 2645 * at once. 2646 */ 2647 newlen = oldfds * sizeof(fdep[0]); 2648 if (CMSG_SPACE(newlen) > MCLBYTES) { 2649 error = EMSGSIZE; 2650 goto out; 2651 } 2652 /* 2653 * Check that all the FDs passed in refer to legal 2654 * files. If not, reject the entire operation. 2655 */ 2656 fdp = data; 2657 FILEDESC_SLOCK(fdesc); 2658 for (i = 0; i < oldfds; i++, fdp++) { 2659 fp = fget_noref(fdesc, *fdp); 2660 if (fp == NULL) { 2661 FILEDESC_SUNLOCK(fdesc); 2662 error = EBADF; 2663 goto out; 2664 } 2665 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2666 FILEDESC_SUNLOCK(fdesc); 2667 error = EOPNOTSUPP; 2668 goto out; 2669 } 2670 } 2671 2672 /* 2673 * Now replace the integer FDs with pointers to the 2674 * file structure and capability rights. 2675 */ 2676 *controlp = sbcreatecontrol(NULL, newlen, 2677 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2678 fdp = data; 2679 for (i = 0; i < oldfds; i++, fdp++) { 2680 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2681 fdp = data; 2682 for (j = 0; j < i; j++, fdp++) { 2683 fdrop(fdesc->fd_ofiles[*fdp]. 2684 fde_file, td); 2685 } 2686 FILEDESC_SUNLOCK(fdesc); 2687 error = EBADF; 2688 goto out; 2689 } 2690 } 2691 fdp = data; 2692 fdep = (struct filedescent **) 2693 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2694 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2695 M_WAITOK); 2696 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2697 fde = &fdesc->fd_ofiles[*fdp]; 2698 fdep[i] = fdev; 2699 fdep[i]->fde_file = fde->fde_file; 2700 filecaps_copy(&fde->fde_caps, 2701 &fdep[i]->fde_caps, true); 2702 unp_internalize_fp(fdep[i]->fde_file); 2703 } 2704 FILEDESC_SUNLOCK(fdesc); 2705 break; 2706 2707 case SCM_TIMESTAMP: 2708 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2709 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2710 tv = (struct timeval *) 2711 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2712 microtime(tv); 2713 break; 2714 2715 case SCM_BINTIME: 2716 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2717 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2718 bt = (struct bintime *) 2719 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2720 bintime(bt); 2721 break; 2722 2723 case SCM_REALTIME: 2724 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2725 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2726 ts = (struct timespec *) 2727 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2728 nanotime(ts); 2729 break; 2730 2731 case SCM_MONOTONIC: 2732 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2733 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2734 ts = (struct timespec *) 2735 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2736 nanouptime(ts); 2737 break; 2738 2739 default: 2740 error = EINVAL; 2741 goto out; 2742 } 2743 2744 if (space != NULL) { 2745 *space += (*controlp)->m_len; 2746 *mbcnt += MSIZE; 2747 if ((*controlp)->m_flags & M_EXT) 2748 *mbcnt += (*controlp)->m_ext.ext_size; 2749 *clast = *controlp; 2750 } 2751 controlp = &(*controlp)->m_next; 2752 } 2753 if (clen > 0) 2754 error = EINVAL; 2755 2756 out: 2757 if (error != 0 && initial_controlp != NULL) 2758 unp_internalize_cleanup_rights(*initial_controlp); 2759 m_freem(control); 2760 return (error); 2761 } 2762 2763 static struct mbuf * 2764 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2765 struct mbuf **clast, u_int *space, u_int *mbcnt) 2766 { 2767 struct mbuf *m, *n, *n_prev; 2768 const struct cmsghdr *cm; 2769 int ngroups, i, cmsgtype; 2770 size_t ctrlsz; 2771 2772 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2773 if (mode & UNP_WANTCRED_ALWAYS) { 2774 ctrlsz = SOCKCRED2SIZE(ngroups); 2775 cmsgtype = SCM_CREDS2; 2776 } else { 2777 ctrlsz = SOCKCREDSIZE(ngroups); 2778 cmsgtype = SCM_CREDS; 2779 } 2780 2781 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2782 if (m == NULL) 2783 return (control); 2784 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2785 2786 if (mode & UNP_WANTCRED_ALWAYS) { 2787 struct sockcred2 *sc; 2788 2789 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2790 sc->sc_version = 0; 2791 sc->sc_pid = td->td_proc->p_pid; 2792 sc->sc_uid = td->td_ucred->cr_ruid; 2793 sc->sc_euid = td->td_ucred->cr_uid; 2794 sc->sc_gid = td->td_ucred->cr_rgid; 2795 sc->sc_egid = td->td_ucred->cr_gid; 2796 sc->sc_ngroups = ngroups; 2797 for (i = 0; i < sc->sc_ngroups; i++) 2798 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2799 } else { 2800 struct sockcred *sc; 2801 2802 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2803 sc->sc_uid = td->td_ucred->cr_ruid; 2804 sc->sc_euid = td->td_ucred->cr_uid; 2805 sc->sc_gid = td->td_ucred->cr_rgid; 2806 sc->sc_egid = td->td_ucred->cr_gid; 2807 sc->sc_ngroups = ngroups; 2808 for (i = 0; i < sc->sc_ngroups; i++) 2809 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2810 } 2811 2812 /* 2813 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2814 * created SCM_CREDS control message (struct sockcred) has another 2815 * format. 2816 */ 2817 if (control != NULL && cmsgtype == SCM_CREDS) 2818 for (n = control, n_prev = NULL; n != NULL;) { 2819 cm = mtod(n, struct cmsghdr *); 2820 if (cm->cmsg_level == SOL_SOCKET && 2821 cm->cmsg_type == SCM_CREDS) { 2822 if (n_prev == NULL) 2823 control = n->m_next; 2824 else 2825 n_prev->m_next = n->m_next; 2826 if (space != NULL) { 2827 MPASS(*space >= n->m_len); 2828 *space -= n->m_len; 2829 MPASS(*mbcnt >= MSIZE); 2830 *mbcnt -= MSIZE; 2831 if (n->m_flags & M_EXT) { 2832 MPASS(*mbcnt >= 2833 n->m_ext.ext_size); 2834 *mbcnt -= n->m_ext.ext_size; 2835 } 2836 MPASS(clast); 2837 if (*clast == n) { 2838 MPASS(n->m_next == NULL); 2839 if (n_prev == NULL) 2840 *clast = m; 2841 else 2842 *clast = n_prev; 2843 } 2844 } 2845 n = m_free(n); 2846 } else { 2847 n_prev = n; 2848 n = n->m_next; 2849 } 2850 } 2851 2852 /* Prepend it to the head. */ 2853 m->m_next = control; 2854 if (space != NULL) { 2855 *space += m->m_len; 2856 *mbcnt += MSIZE; 2857 if (control == NULL) 2858 *clast = m; 2859 } 2860 return (m); 2861 } 2862 2863 static struct unpcb * 2864 fptounp(struct file *fp) 2865 { 2866 struct socket *so; 2867 2868 if (fp->f_type != DTYPE_SOCKET) 2869 return (NULL); 2870 if ((so = fp->f_data) == NULL) 2871 return (NULL); 2872 if (so->so_proto->pr_domain != &localdomain) 2873 return (NULL); 2874 return sotounpcb(so); 2875 } 2876 2877 static void 2878 unp_discard(struct file *fp) 2879 { 2880 struct unp_defer *dr; 2881 2882 if (unp_externalize_fp(fp)) { 2883 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2884 dr->ud_fp = fp; 2885 UNP_DEFERRED_LOCK(); 2886 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2887 UNP_DEFERRED_UNLOCK(); 2888 atomic_add_int(&unp_defers_count, 1); 2889 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2890 } else 2891 closef_nothread(fp); 2892 } 2893 2894 static void 2895 unp_process_defers(void *arg __unused, int pending) 2896 { 2897 struct unp_defer *dr; 2898 SLIST_HEAD(, unp_defer) drl; 2899 int count; 2900 2901 SLIST_INIT(&drl); 2902 for (;;) { 2903 UNP_DEFERRED_LOCK(); 2904 if (SLIST_FIRST(&unp_defers) == NULL) { 2905 UNP_DEFERRED_UNLOCK(); 2906 break; 2907 } 2908 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2909 UNP_DEFERRED_UNLOCK(); 2910 count = 0; 2911 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2912 SLIST_REMOVE_HEAD(&drl, ud_link); 2913 closef_nothread(dr->ud_fp); 2914 free(dr, M_TEMP); 2915 count++; 2916 } 2917 atomic_add_int(&unp_defers_count, -count); 2918 } 2919 } 2920 2921 static void 2922 unp_internalize_fp(struct file *fp) 2923 { 2924 struct unpcb *unp; 2925 2926 UNP_LINK_WLOCK(); 2927 if ((unp = fptounp(fp)) != NULL) { 2928 unp->unp_file = fp; 2929 unp->unp_msgcount++; 2930 } 2931 unp_rights++; 2932 UNP_LINK_WUNLOCK(); 2933 } 2934 2935 static int 2936 unp_externalize_fp(struct file *fp) 2937 { 2938 struct unpcb *unp; 2939 int ret; 2940 2941 UNP_LINK_WLOCK(); 2942 if ((unp = fptounp(fp)) != NULL) { 2943 unp->unp_msgcount--; 2944 ret = 1; 2945 } else 2946 ret = 0; 2947 unp_rights--; 2948 UNP_LINK_WUNLOCK(); 2949 return (ret); 2950 } 2951 2952 /* 2953 * unp_defer indicates whether additional work has been defered for a future 2954 * pass through unp_gc(). It is thread local and does not require explicit 2955 * synchronization. 2956 */ 2957 static int unp_marked; 2958 2959 static void 2960 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2961 { 2962 struct unpcb *unp; 2963 struct file *fp; 2964 int i; 2965 2966 /* 2967 * This function can only be called from the gc task. 2968 */ 2969 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2970 ("%s: not on gc callout", __func__)); 2971 UNP_LINK_LOCK_ASSERT(); 2972 2973 for (i = 0; i < fdcount; i++) { 2974 fp = fdep[i]->fde_file; 2975 if ((unp = fptounp(fp)) == NULL) 2976 continue; 2977 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2978 continue; 2979 unp->unp_gcrefs--; 2980 } 2981 } 2982 2983 static void 2984 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2985 { 2986 struct unpcb *unp; 2987 struct file *fp; 2988 int i; 2989 2990 /* 2991 * This function can only be called from the gc task. 2992 */ 2993 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2994 ("%s: not on gc callout", __func__)); 2995 UNP_LINK_LOCK_ASSERT(); 2996 2997 for (i = 0; i < fdcount; i++) { 2998 fp = fdep[i]->fde_file; 2999 if ((unp = fptounp(fp)) == NULL) 3000 continue; 3001 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3002 continue; 3003 unp->unp_gcrefs++; 3004 unp_marked++; 3005 } 3006 } 3007 3008 static void 3009 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3010 { 3011 struct sockbuf *sb; 3012 3013 SOCK_LOCK_ASSERT(so); 3014 3015 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3016 return; 3017 3018 SOCK_RECVBUF_LOCK(so); 3019 switch (so->so_type) { 3020 case SOCK_DGRAM: 3021 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3022 unp_scan(so->so_rcv.uxdg_peeked, op); 3023 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3024 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3025 break; 3026 case SOCK_STREAM: 3027 case SOCK_SEQPACKET: 3028 unp_scan(so->so_rcv.sb_mb, op); 3029 break; 3030 } 3031 SOCK_RECVBUF_UNLOCK(so); 3032 } 3033 3034 static void 3035 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3036 { 3037 struct socket *so, *soa; 3038 3039 so = unp->unp_socket; 3040 SOCK_LOCK(so); 3041 if (SOLISTENING(so)) { 3042 /* 3043 * Mark all sockets in our accept queue. 3044 */ 3045 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3046 unp_scan_socket(soa, op); 3047 } else { 3048 /* 3049 * Mark all sockets we reference with RIGHTS. 3050 */ 3051 unp_scan_socket(so, op); 3052 } 3053 SOCK_UNLOCK(so); 3054 } 3055 3056 static int unp_recycled; 3057 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3058 "Number of unreachable sockets claimed by the garbage collector."); 3059 3060 static int unp_taskcount; 3061 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3062 "Number of times the garbage collector has run."); 3063 3064 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3065 "Number of active local sockets."); 3066 3067 static void 3068 unp_gc(__unused void *arg, int pending) 3069 { 3070 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3071 NULL }; 3072 struct unp_head **head; 3073 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3074 struct file *f, **unref; 3075 struct unpcb *unp, *unptmp; 3076 int i, total, unp_unreachable; 3077 3078 LIST_INIT(&unp_deadhead); 3079 unp_taskcount++; 3080 UNP_LINK_RLOCK(); 3081 /* 3082 * First determine which sockets may be in cycles. 3083 */ 3084 unp_unreachable = 0; 3085 3086 for (head = heads; *head != NULL; head++) 3087 LIST_FOREACH(unp, *head, unp_link) { 3088 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3089 ("%s: unp %p has unexpected gc flags 0x%x", 3090 __func__, unp, (unsigned int)unp->unp_gcflag)); 3091 3092 f = unp->unp_file; 3093 3094 /* 3095 * Check for an unreachable socket potentially in a 3096 * cycle. It must be in a queue as indicated by 3097 * msgcount, and this must equal the file reference 3098 * count. Note that when msgcount is 0 the file is 3099 * NULL. 3100 */ 3101 if (f != NULL && unp->unp_msgcount != 0 && 3102 refcount_load(&f->f_count) == unp->unp_msgcount) { 3103 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3104 unp->unp_gcflag |= UNPGC_DEAD; 3105 unp->unp_gcrefs = unp->unp_msgcount; 3106 unp_unreachable++; 3107 } 3108 } 3109 3110 /* 3111 * Scan all sockets previously marked as potentially being in a cycle 3112 * and remove the references each socket holds on any UNPGC_DEAD 3113 * sockets in its queue. After this step, all remaining references on 3114 * sockets marked UNPGC_DEAD should not be part of any cycle. 3115 */ 3116 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3117 unp_gc_scan(unp, unp_remove_dead_ref); 3118 3119 /* 3120 * If a socket still has a non-negative refcount, it cannot be in a 3121 * cycle. In this case increment refcount of all children iteratively. 3122 * Stop the scan once we do a complete loop without discovering 3123 * a new reachable socket. 3124 */ 3125 do { 3126 unp_marked = 0; 3127 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3128 if (unp->unp_gcrefs > 0) { 3129 unp->unp_gcflag &= ~UNPGC_DEAD; 3130 LIST_REMOVE(unp, unp_dead); 3131 KASSERT(unp_unreachable > 0, 3132 ("%s: unp_unreachable underflow.", 3133 __func__)); 3134 unp_unreachable--; 3135 unp_gc_scan(unp, unp_restore_undead_ref); 3136 } 3137 } while (unp_marked); 3138 3139 UNP_LINK_RUNLOCK(); 3140 3141 if (unp_unreachable == 0) 3142 return; 3143 3144 /* 3145 * Allocate space for a local array of dead unpcbs. 3146 * TODO: can this path be simplified by instead using the local 3147 * dead list at unp_deadhead, after taking out references 3148 * on the file object and/or unpcb and dropping the link lock? 3149 */ 3150 unref = malloc(unp_unreachable * sizeof(struct file *), 3151 M_TEMP, M_WAITOK); 3152 3153 /* 3154 * Iterate looking for sockets which have been specifically marked 3155 * as unreachable and store them locally. 3156 */ 3157 UNP_LINK_RLOCK(); 3158 total = 0; 3159 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3160 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3161 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3162 unp->unp_gcflag &= ~UNPGC_DEAD; 3163 f = unp->unp_file; 3164 if (unp->unp_msgcount == 0 || f == NULL || 3165 refcount_load(&f->f_count) != unp->unp_msgcount || 3166 !fhold(f)) 3167 continue; 3168 unref[total++] = f; 3169 KASSERT(total <= unp_unreachable, 3170 ("%s: incorrect unreachable count.", __func__)); 3171 } 3172 UNP_LINK_RUNLOCK(); 3173 3174 /* 3175 * Now flush all sockets, free'ing rights. This will free the 3176 * struct files associated with these sockets but leave each socket 3177 * with one remaining ref. 3178 */ 3179 for (i = 0; i < total; i++) { 3180 struct socket *so; 3181 3182 so = unref[i]->f_data; 3183 CURVNET_SET(so->so_vnet); 3184 sorflush(so); 3185 CURVNET_RESTORE(); 3186 } 3187 3188 /* 3189 * And finally release the sockets so they can be reclaimed. 3190 */ 3191 for (i = 0; i < total; i++) 3192 fdrop(unref[i], NULL); 3193 unp_recycled += total; 3194 free(unref, M_TEMP); 3195 } 3196 3197 /* 3198 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3199 */ 3200 static void 3201 unp_dispose(struct socket *so) 3202 { 3203 struct sockbuf *sb; 3204 struct unpcb *unp; 3205 struct mbuf *m; 3206 3207 MPASS(!SOLISTENING(so)); 3208 3209 unp = sotounpcb(so); 3210 UNP_LINK_WLOCK(); 3211 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3212 UNP_LINK_WUNLOCK(); 3213 3214 /* 3215 * Grab our special mbufs before calling sbrelease(). 3216 */ 3217 SOCK_RECVBUF_LOCK(so); 3218 switch (so->so_type) { 3219 case SOCK_DGRAM: 3220 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3221 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3222 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3223 /* Note: socket of sb may reconnect. */ 3224 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3225 } 3226 sb = &so->so_rcv; 3227 if (sb->uxdg_peeked != NULL) { 3228 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3229 m_stailqpkt); 3230 sb->uxdg_peeked = NULL; 3231 } 3232 m = STAILQ_FIRST(&sb->uxdg_mb); 3233 STAILQ_INIT(&sb->uxdg_mb); 3234 /* XXX: our shortened sbrelease() */ 3235 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3236 RLIM_INFINITY); 3237 /* 3238 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3239 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3240 * data to the socket. 3241 * We are now in dom_dispose and it could be a call from 3242 * soshutdown() or from the final sofree(). The sofree() case 3243 * is simple as it guarantees that no more sends will happen, 3244 * however we can race with unp_disconnect() from our peer. 3245 * The shutdown(2) case is more exotic. It would call into 3246 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3247 * possible if we did connect(2) on this socket and we also 3248 * had it bound with bind(2) and receive connections from other 3249 * sockets. Because soshutdown() violates POSIX (see comment 3250 * there) we will end up here shutting down our receive side. 3251 * Of course this will have affect not only on the peer we 3252 * connect(2)ed to, but also on all of the peers who had 3253 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3254 */ 3255 sb->sb_state |= SBS_CANTRCVMORE; 3256 break; 3257 case SOCK_STREAM: 3258 case SOCK_SEQPACKET: 3259 sb = &so->so_rcv; 3260 m = sbcut_locked(sb, sb->sb_ccc); 3261 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3262 ("%s: ccc %u mb %p mbcnt %u", __func__, 3263 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3264 sbrelease_locked(so, SO_RCV); 3265 break; 3266 } 3267 SOCK_RECVBUF_UNLOCK(so); 3268 if (SOCK_IO_RECV_OWNED(so)) 3269 SOCK_IO_RECV_UNLOCK(so); 3270 3271 if (m != NULL) { 3272 unp_scan(m, unp_freerights); 3273 m_freem(m); 3274 } 3275 } 3276 3277 static void 3278 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3279 { 3280 struct mbuf *m; 3281 struct cmsghdr *cm; 3282 void *data; 3283 socklen_t clen, datalen; 3284 3285 while (m0 != NULL) { 3286 for (m = m0; m; m = m->m_next) { 3287 if (m->m_type != MT_CONTROL) 3288 continue; 3289 3290 cm = mtod(m, struct cmsghdr *); 3291 clen = m->m_len; 3292 3293 while (cm != NULL) { 3294 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3295 break; 3296 3297 data = CMSG_DATA(cm); 3298 datalen = (caddr_t)cm + cm->cmsg_len 3299 - (caddr_t)data; 3300 3301 if (cm->cmsg_level == SOL_SOCKET && 3302 cm->cmsg_type == SCM_RIGHTS) { 3303 (*op)(data, datalen / 3304 sizeof(struct filedescent *)); 3305 } 3306 3307 if (CMSG_SPACE(datalen) < clen) { 3308 clen -= CMSG_SPACE(datalen); 3309 cm = (struct cmsghdr *) 3310 ((caddr_t)cm + CMSG_SPACE(datalen)); 3311 } else { 3312 clen = 0; 3313 cm = NULL; 3314 } 3315 } 3316 } 3317 m0 = m0->m_nextpkt; 3318 } 3319 } 3320 3321 /* 3322 * Definitions of protocols supported in the LOCAL domain. 3323 */ 3324 static struct protosw streamproto = { 3325 .pr_type = SOCK_STREAM, 3326 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 3327 PR_CAPATTACH, 3328 .pr_ctloutput = &uipc_ctloutput, 3329 .pr_abort = uipc_abort, 3330 .pr_accept = uipc_accept, 3331 .pr_attach = uipc_attach, 3332 .pr_bind = uipc_bind, 3333 .pr_bindat = uipc_bindat, 3334 .pr_connect = uipc_connect, 3335 .pr_connectat = uipc_connectat, 3336 .pr_connect2 = uipc_connect2, 3337 .pr_detach = uipc_detach, 3338 .pr_disconnect = uipc_disconnect, 3339 .pr_listen = uipc_listen, 3340 .pr_peeraddr = uipc_peeraddr, 3341 .pr_rcvd = uipc_rcvd, 3342 .pr_send = uipc_send, 3343 .pr_ready = uipc_ready, 3344 .pr_sense = uipc_sense, 3345 .pr_shutdown = uipc_shutdown, 3346 .pr_sockaddr = uipc_sockaddr, 3347 .pr_soreceive = soreceive_generic, 3348 .pr_close = uipc_close, 3349 }; 3350 3351 static struct protosw dgramproto = { 3352 .pr_type = SOCK_DGRAM, 3353 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 3354 PR_SOCKBUF, 3355 .pr_ctloutput = &uipc_ctloutput, 3356 .pr_abort = uipc_abort, 3357 .pr_accept = uipc_accept, 3358 .pr_attach = uipc_attach, 3359 .pr_bind = uipc_bind, 3360 .pr_bindat = uipc_bindat, 3361 .pr_connect = uipc_connect, 3362 .pr_connectat = uipc_connectat, 3363 .pr_connect2 = uipc_connect2, 3364 .pr_detach = uipc_detach, 3365 .pr_disconnect = uipc_disconnect, 3366 .pr_peeraddr = uipc_peeraddr, 3367 .pr_sosend = uipc_sosend_dgram, 3368 .pr_sense = uipc_sense, 3369 .pr_shutdown = uipc_shutdown, 3370 .pr_sockaddr = uipc_sockaddr, 3371 .pr_soreceive = uipc_soreceive_dgram, 3372 .pr_close = uipc_close, 3373 }; 3374 3375 static struct protosw seqpacketproto = { 3376 .pr_type = SOCK_SEQPACKET, 3377 /* 3378 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3379 * due to our use of sbappendaddr. A new sbappend variants is needed 3380 * that supports both atomic record writes and control data. 3381 */ 3382 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 3383 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 3384 .pr_ctloutput = &uipc_ctloutput, 3385 .pr_abort = uipc_abort, 3386 .pr_accept = uipc_accept, 3387 .pr_attach = uipc_attach, 3388 .pr_bind = uipc_bind, 3389 .pr_bindat = uipc_bindat, 3390 .pr_connect = uipc_connect, 3391 .pr_connectat = uipc_connectat, 3392 .pr_connect2 = uipc_connect2, 3393 .pr_detach = uipc_detach, 3394 .pr_disconnect = uipc_disconnect, 3395 .pr_listen = uipc_listen, 3396 .pr_peeraddr = uipc_peeraddr, 3397 .pr_rcvd = uipc_rcvd, 3398 .pr_send = uipc_send, 3399 .pr_sense = uipc_sense, 3400 .pr_shutdown = uipc_shutdown, 3401 .pr_sockaddr = uipc_sockaddr, 3402 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3403 .pr_close = uipc_close, 3404 }; 3405 3406 static struct domain localdomain = { 3407 .dom_family = AF_LOCAL, 3408 .dom_name = "local", 3409 .dom_externalize = unp_externalize, 3410 .dom_dispose = unp_dispose, 3411 .dom_nprotosw = 3, 3412 .dom_protosw = { 3413 &streamproto, 3414 &dgramproto, 3415 &seqpacketproto, 3416 } 3417 }; 3418 DOMAIN_SET(local); 3419 3420 /* 3421 * A helper function called by VFS before socket-type vnode reclamation. 3422 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3423 * use count. 3424 */ 3425 void 3426 vfs_unp_reclaim(struct vnode *vp) 3427 { 3428 struct unpcb *unp; 3429 int active; 3430 struct mtx *vplock; 3431 3432 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3433 KASSERT(vp->v_type == VSOCK, 3434 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3435 3436 active = 0; 3437 vplock = mtx_pool_find(mtxpool_sleep, vp); 3438 mtx_lock(vplock); 3439 VOP_UNP_CONNECT(vp, &unp); 3440 if (unp == NULL) 3441 goto done; 3442 UNP_PCB_LOCK(unp); 3443 if (unp->unp_vnode == vp) { 3444 VOP_UNP_DETACH(vp); 3445 unp->unp_vnode = NULL; 3446 active = 1; 3447 } 3448 UNP_PCB_UNLOCK(unp); 3449 done: 3450 mtx_unlock(vplock); 3451 if (active) 3452 vunref(vp); 3453 } 3454 3455 #ifdef DDB 3456 static void 3457 db_print_indent(int indent) 3458 { 3459 int i; 3460 3461 for (i = 0; i < indent; i++) 3462 db_printf(" "); 3463 } 3464 3465 static void 3466 db_print_unpflags(int unp_flags) 3467 { 3468 int comma; 3469 3470 comma = 0; 3471 if (unp_flags & UNP_HAVEPC) { 3472 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3473 comma = 1; 3474 } 3475 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3476 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3477 comma = 1; 3478 } 3479 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3480 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3481 comma = 1; 3482 } 3483 if (unp_flags & UNP_CONNWAIT) { 3484 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3485 comma = 1; 3486 } 3487 if (unp_flags & UNP_CONNECTING) { 3488 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3489 comma = 1; 3490 } 3491 if (unp_flags & UNP_BINDING) { 3492 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3493 comma = 1; 3494 } 3495 } 3496 3497 static void 3498 db_print_xucred(int indent, struct xucred *xu) 3499 { 3500 int comma, i; 3501 3502 db_print_indent(indent); 3503 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3504 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3505 db_print_indent(indent); 3506 db_printf("cr_groups: "); 3507 comma = 0; 3508 for (i = 0; i < xu->cr_ngroups; i++) { 3509 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3510 comma = 1; 3511 } 3512 db_printf("\n"); 3513 } 3514 3515 static void 3516 db_print_unprefs(int indent, struct unp_head *uh) 3517 { 3518 struct unpcb *unp; 3519 int counter; 3520 3521 counter = 0; 3522 LIST_FOREACH(unp, uh, unp_reflink) { 3523 if (counter % 4 == 0) 3524 db_print_indent(indent); 3525 db_printf("%p ", unp); 3526 if (counter % 4 == 3) 3527 db_printf("\n"); 3528 counter++; 3529 } 3530 if (counter != 0 && counter % 4 != 0) 3531 db_printf("\n"); 3532 } 3533 3534 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3535 { 3536 struct unpcb *unp; 3537 3538 if (!have_addr) { 3539 db_printf("usage: show unpcb <addr>\n"); 3540 return; 3541 } 3542 unp = (struct unpcb *)addr; 3543 3544 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3545 unp->unp_vnode); 3546 3547 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3548 unp->unp_conn); 3549 3550 db_printf("unp_refs:\n"); 3551 db_print_unprefs(2, &unp->unp_refs); 3552 3553 /* XXXRW: Would be nice to print the full address, if any. */ 3554 db_printf("unp_addr: %p\n", unp->unp_addr); 3555 3556 db_printf("unp_gencnt: %llu\n", 3557 (unsigned long long)unp->unp_gencnt); 3558 3559 db_printf("unp_flags: %x (", unp->unp_flags); 3560 db_print_unpflags(unp->unp_flags); 3561 db_printf(")\n"); 3562 3563 db_printf("unp_peercred:\n"); 3564 db_print_xucred(2, &unp->unp_peercred); 3565 3566 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3567 } 3568 #endif 3569