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 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 34 */ 35 36 /* 37 * UNIX Domain (Local) Sockets 38 * 39 * This is an implementation of UNIX (local) domain sockets. Each socket has 40 * an associated struct unpcb (UNIX protocol control block). Stream sockets 41 * may be connected to 0 or 1 other socket. Datagram sockets may be 42 * connected to 0, 1, or many other sockets. Sockets may be created and 43 * connected in pairs (socketpair(2)), or bound/connected to using the file 44 * system name space. For most purposes, only the receive socket buffer is 45 * used, as sending on one socket delivers directly to the receive socket 46 * buffer of a second socket. 47 * 48 * The implementation is substantially complicated by the fact that 49 * "ancillary data", such as file descriptors or credentials, may be passed 50 * across UNIX domain sockets. The potential for passing UNIX domain sockets 51 * over other UNIX domain sockets requires the implementation of a simple 52 * garbage collector to find and tear down cycles of disconnected sockets. 53 * 54 * TODO: 55 * RDM 56 * rethink name space problems 57 * need a proper out-of-band 58 */ 59 60 #include <sys/cdefs.h> 61 __FBSDID("$FreeBSD$"); 62 63 #include "opt_ddb.h" 64 65 #include <sys/param.h> 66 #include <sys/capsicum.h> 67 #include <sys/domain.h> 68 #include <sys/eventhandler.h> 69 #include <sys/fcntl.h> 70 #include <sys/file.h> 71 #include <sys/filedesc.h> 72 #include <sys/kernel.h> 73 #include <sys/lock.h> 74 #include <sys/malloc.h> 75 #include <sys/mbuf.h> 76 #include <sys/mount.h> 77 #include <sys/mutex.h> 78 #include <sys/namei.h> 79 #include <sys/proc.h> 80 #include <sys/protosw.h> 81 #include <sys/queue.h> 82 #include <sys/resourcevar.h> 83 #include <sys/rwlock.h> 84 #include <sys/socket.h> 85 #include <sys/socketvar.h> 86 #include <sys/signalvar.h> 87 #include <sys/stat.h> 88 #include <sys/sx.h> 89 #include <sys/sysctl.h> 90 #include <sys/systm.h> 91 #include <sys/taskqueue.h> 92 #include <sys/un.h> 93 #include <sys/unpcb.h> 94 #include <sys/vnode.h> 95 96 #include <net/vnet.h> 97 98 #ifdef DDB 99 #include <ddb/ddb.h> 100 #endif 101 102 #include <security/mac/mac_framework.h> 103 104 #include <vm/uma.h> 105 106 MALLOC_DECLARE(M_FILECAPS); 107 108 static struct domain localdomain; 109 110 static uma_zone_t unp_zone; 111 static unp_gen_t unp_gencnt; /* (l) */ 112 static u_int unp_count; /* (l) Count of local sockets. */ 113 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 114 static int unp_rights; /* (g) File descriptors in flight. */ 115 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 116 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 117 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 118 119 struct unp_defer { 120 SLIST_ENTRY(unp_defer) ud_link; 121 struct file *ud_fp; 122 }; 123 static SLIST_HEAD(, unp_defer) unp_defers; 124 static int unp_defers_count; 125 126 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 127 128 /* 129 * Garbage collection of cyclic file descriptor/socket references occurs 130 * asynchronously in a taskqueue context in order to avoid recursion and 131 * reentrance in the UNIX domain socket, file descriptor, and socket layer 132 * code. See unp_gc() for a full description. 133 */ 134 static struct timeout_task unp_gc_task; 135 136 /* 137 * The close of unix domain sockets attached as SCM_RIGHTS is 138 * postponed to the taskqueue, to avoid arbitrary recursion depth. 139 * The attached sockets might have another sockets attached. 140 */ 141 static struct task unp_defer_task; 142 143 /* 144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 145 * stream sockets, although the total for sender and receiver is actually 146 * only PIPSIZ. 147 * 148 * Datagram sockets really use the sendspace as the maximum datagram size, 149 * and don't really want to reserve the sendspace. Their recvspace should be 150 * large enough for at least one max-size datagram plus address. 151 */ 152 #ifndef PIPSIZ 153 #define PIPSIZ 8192 154 #endif 155 static u_long unpst_sendspace = PIPSIZ; 156 static u_long unpst_recvspace = PIPSIZ; 157 static u_long unpdg_maxdgram = 2*1024; 158 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */ 159 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ 160 static u_long unpsp_recvspace = PIPSIZ; 161 162 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 163 "Local domain"); 164 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 165 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 166 "SOCK_STREAM"); 167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 169 "SOCK_DGRAM"); 170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 172 "SOCK_SEQPACKET"); 173 174 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 175 &unpst_sendspace, 0, "Default stream send space."); 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 177 &unpst_recvspace, 0, "Default stream receive space."); 178 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 179 &unpdg_maxdgram, 0, "Maximum datagram size."); 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 181 &unpdg_recvspace, 0, "Default datagram receive space."); 182 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 183 &unpsp_sendspace, 0, "Default seqpacket send space."); 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 185 &unpsp_recvspace, 0, "Default seqpacket receive space."); 186 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 187 "File descriptors in flight."); 188 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 189 &unp_defers_count, 0, 190 "File descriptors deferred to taskqueue for close."); 191 192 /* 193 * Locking and synchronization: 194 * 195 * Several types of locks exist in the local domain socket implementation: 196 * - a global linkage lock 197 * - a global connection list lock 198 * - the mtxpool lock 199 * - per-unpcb mutexes 200 * 201 * The linkage lock protects the global socket lists, the generation number 202 * counter and garbage collector state. 203 * 204 * The connection list lock protects the list of referring sockets in a datagram 205 * socket PCB. This lock is also overloaded to protect a global list of 206 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 207 * messages. To avoid recursion, such references are released by a dedicated 208 * thread. 209 * 210 * The mtxpool lock protects the vnode from being modified while referenced. 211 * Lock ordering rules require that it be acquired before any PCB locks. 212 * 213 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 214 * unpcb. This includes the unp_conn field, which either links two connected 215 * PCBs together (for connected socket types) or points at the destination 216 * socket (for connectionless socket types). The operations of creating or 217 * destroying a connection therefore involve locking multiple PCBs. To avoid 218 * lock order reversals, in some cases this involves dropping a PCB lock and 219 * using a reference counter to maintain liveness. 220 * 221 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 222 * allocated in pr_attach() and freed in pr_detach(). The validity of that 223 * pointer is an invariant, so no lock is required to dereference the so_pcb 224 * pointer if a valid socket reference is held by the caller. In practice, 225 * this is always true during operations performed on a socket. Each unpcb 226 * has a back-pointer to its socket, unp_socket, which will be stable under 227 * the same circumstances. 228 * 229 * This pointer may only be safely dereferenced as long as a valid reference 230 * to the unpcb is held. Typically, this reference will be from the socket, 231 * or from another unpcb when the referring unpcb's lock is held (in order 232 * that the reference not be invalidated during use). For example, to follow 233 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 234 * that detach is not run clearing unp_socket. 235 * 236 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 237 * protocols, bind() is a non-atomic operation, and connect() requires 238 * potential sleeping in the protocol, due to potentially waiting on local or 239 * distributed file systems. We try to separate "lookup" operations, which 240 * may sleep, and the IPC operations themselves, which typically can occur 241 * with relative atomicity as locks can be held over the entire operation. 242 * 243 * Another tricky issue is simultaneous multi-threaded or multi-process 244 * access to a single UNIX domain socket. These are handled by the flags 245 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 246 * binding, both of which involve dropping UNIX domain socket locks in order 247 * to perform namei() and other file system operations. 248 */ 249 static struct rwlock unp_link_rwlock; 250 static struct mtx unp_defers_lock; 251 252 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 253 "unp_link_rwlock") 254 255 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 256 RA_LOCKED) 257 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 258 RA_UNLOCKED) 259 260 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 261 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 262 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 263 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 264 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 265 RA_WLOCKED) 266 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 267 268 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 269 "unp_defer", NULL, MTX_DEF) 270 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 271 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 272 273 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 274 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 275 276 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 277 "unp", "unp", \ 278 MTX_DUPOK|MTX_DEF) 279 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 280 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 281 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 282 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 283 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 284 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 285 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 286 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 287 288 static int uipc_connect2(struct socket *, struct socket *); 289 static int uipc_ctloutput(struct socket *, struct sockopt *); 290 static int unp_connect(struct socket *, struct sockaddr *, 291 struct thread *); 292 static int unp_connectat(int, struct socket *, struct sockaddr *, 293 struct thread *, bool); 294 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how; 295 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how); 296 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 297 static void unp_dispose(struct socket *so); 298 static void unp_shutdown(struct unpcb *); 299 static void unp_drop(struct unpcb *); 300 static void unp_gc(__unused void *, int); 301 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); 302 static void unp_discard(struct file *); 303 static void unp_freerights(struct filedescent **, int); 304 static int unp_internalize(struct mbuf **, struct thread *, 305 struct mbuf **, u_int *, u_int *); 306 static void unp_internalize_fp(struct file *); 307 static int unp_externalize(struct mbuf *, struct mbuf **, int); 308 static int unp_externalize_fp(struct file *); 309 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, 310 int, struct mbuf **, u_int *, u_int *); 311 static void unp_process_defers(void * __unused, int); 312 313 static void 314 unp_pcb_hold(struct unpcb *unp) 315 { 316 u_int old __unused; 317 318 old = refcount_acquire(&unp->unp_refcount); 319 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 320 } 321 322 static __result_use_check bool 323 unp_pcb_rele(struct unpcb *unp) 324 { 325 bool ret; 326 327 UNP_PCB_LOCK_ASSERT(unp); 328 329 if ((ret = refcount_release(&unp->unp_refcount))) { 330 UNP_PCB_UNLOCK(unp); 331 UNP_PCB_LOCK_DESTROY(unp); 332 uma_zfree(unp_zone, unp); 333 } 334 return (ret); 335 } 336 337 static void 338 unp_pcb_rele_notlast(struct unpcb *unp) 339 { 340 bool ret __unused; 341 342 ret = refcount_release(&unp->unp_refcount); 343 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 344 } 345 346 static void 347 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 348 { 349 UNP_PCB_UNLOCK_ASSERT(unp); 350 UNP_PCB_UNLOCK_ASSERT(unp2); 351 352 if (unp == unp2) { 353 UNP_PCB_LOCK(unp); 354 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 355 UNP_PCB_LOCK(unp); 356 UNP_PCB_LOCK(unp2); 357 } else { 358 UNP_PCB_LOCK(unp2); 359 UNP_PCB_LOCK(unp); 360 } 361 } 362 363 static void 364 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 365 { 366 UNP_PCB_UNLOCK(unp); 367 if (unp != unp2) 368 UNP_PCB_UNLOCK(unp2); 369 } 370 371 /* 372 * Try to lock the connected peer of an already locked socket. In some cases 373 * this requires that we unlock the current socket. The pairbusy counter is 374 * used to block concurrent connection attempts while the lock is dropped. The 375 * caller must be careful to revalidate PCB state. 376 */ 377 static struct unpcb * 378 unp_pcb_lock_peer(struct unpcb *unp) 379 { 380 struct unpcb *unp2; 381 382 UNP_PCB_LOCK_ASSERT(unp); 383 unp2 = unp->unp_conn; 384 if (unp2 == NULL) 385 return (NULL); 386 if (__predict_false(unp == unp2)) 387 return (unp); 388 389 UNP_PCB_UNLOCK_ASSERT(unp2); 390 391 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 392 return (unp2); 393 if ((uintptr_t)unp2 > (uintptr_t)unp) { 394 UNP_PCB_LOCK(unp2); 395 return (unp2); 396 } 397 unp->unp_pairbusy++; 398 unp_pcb_hold(unp2); 399 UNP_PCB_UNLOCK(unp); 400 401 UNP_PCB_LOCK(unp2); 402 UNP_PCB_LOCK(unp); 403 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 404 ("%s: socket %p was reconnected", __func__, unp)); 405 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 406 unp->unp_flags &= ~UNP_WAITING; 407 wakeup(unp); 408 } 409 if (unp_pcb_rele(unp2)) { 410 /* unp2 is unlocked. */ 411 return (NULL); 412 } 413 if (unp->unp_conn == NULL) { 414 UNP_PCB_UNLOCK(unp2); 415 return (NULL); 416 } 417 return (unp2); 418 } 419 420 static void 421 uipc_abort(struct socket *so) 422 { 423 struct unpcb *unp, *unp2; 424 425 unp = sotounpcb(so); 426 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 427 UNP_PCB_UNLOCK_ASSERT(unp); 428 429 UNP_PCB_LOCK(unp); 430 unp2 = unp->unp_conn; 431 if (unp2 != NULL) { 432 unp_pcb_hold(unp2); 433 UNP_PCB_UNLOCK(unp); 434 unp_drop(unp2); 435 } else 436 UNP_PCB_UNLOCK(unp); 437 } 438 439 static int 440 uipc_accept(struct socket *so, struct sockaddr **nam) 441 { 442 struct unpcb *unp, *unp2; 443 const struct sockaddr *sa; 444 445 /* 446 * Pass back name of connected socket, if it was bound and we are 447 * still connected (our peer may have closed already!). 448 */ 449 unp = sotounpcb(so); 450 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 451 452 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 453 UNP_PCB_LOCK(unp); 454 unp2 = unp_pcb_lock_peer(unp); 455 if (unp2 != NULL && unp2->unp_addr != NULL) 456 sa = (struct sockaddr *)unp2->unp_addr; 457 else 458 sa = &sun_noname; 459 bcopy(sa, *nam, sa->sa_len); 460 if (unp2 != NULL) 461 unp_pcb_unlock_pair(unp, unp2); 462 else 463 UNP_PCB_UNLOCK(unp); 464 return (0); 465 } 466 467 static int 468 uipc_attach(struct socket *so, int proto, struct thread *td) 469 { 470 u_long sendspace, recvspace; 471 struct unpcb *unp; 472 int error; 473 bool locked; 474 475 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 476 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 477 switch (so->so_type) { 478 case SOCK_STREAM: 479 sendspace = unpst_sendspace; 480 recvspace = unpst_recvspace; 481 break; 482 483 case SOCK_DGRAM: 484 STAILQ_INIT(&so->so_rcv.uxdg_mb); 485 STAILQ_INIT(&so->so_snd.uxdg_mb); 486 TAILQ_INIT(&so->so_rcv.uxdg_conns); 487 /* 488 * Since send buffer is either bypassed or is a part 489 * of one-to-many receive buffer, we assign both space 490 * limits to unpdg_recvspace. 491 */ 492 sendspace = recvspace = unpdg_recvspace; 493 break; 494 495 case SOCK_SEQPACKET: 496 sendspace = unpsp_sendspace; 497 recvspace = unpsp_recvspace; 498 break; 499 500 default: 501 panic("uipc_attach"); 502 } 503 error = soreserve(so, sendspace, recvspace); 504 if (error) 505 return (error); 506 } 507 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 508 if (unp == NULL) 509 return (ENOBUFS); 510 LIST_INIT(&unp->unp_refs); 511 UNP_PCB_LOCK_INIT(unp); 512 unp->unp_socket = so; 513 so->so_pcb = unp; 514 refcount_init(&unp->unp_refcount, 1); 515 516 if ((locked = UNP_LINK_WOWNED()) == false) 517 UNP_LINK_WLOCK(); 518 519 unp->unp_gencnt = ++unp_gencnt; 520 unp->unp_ino = ++unp_ino; 521 unp_count++; 522 switch (so->so_type) { 523 case SOCK_STREAM: 524 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 525 break; 526 527 case SOCK_DGRAM: 528 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 529 break; 530 531 case SOCK_SEQPACKET: 532 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 533 break; 534 535 default: 536 panic("uipc_attach"); 537 } 538 539 if (locked == false) 540 UNP_LINK_WUNLOCK(); 541 542 return (0); 543 } 544 545 static int 546 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 547 { 548 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 549 struct vattr vattr; 550 int error, namelen; 551 struct nameidata nd; 552 struct unpcb *unp; 553 struct vnode *vp; 554 struct mount *mp; 555 cap_rights_t rights; 556 char *buf; 557 558 if (nam->sa_family != AF_UNIX) 559 return (EAFNOSUPPORT); 560 561 unp = sotounpcb(so); 562 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 563 564 if (soun->sun_len > sizeof(struct sockaddr_un)) 565 return (EINVAL); 566 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 567 if (namelen <= 0) 568 return (EINVAL); 569 570 /* 571 * We don't allow simultaneous bind() calls on a single UNIX domain 572 * socket, so flag in-progress operations, and return an error if an 573 * operation is already in progress. 574 * 575 * Historically, we have not allowed a socket to be rebound, so this 576 * also returns an error. Not allowing re-binding simplifies the 577 * implementation and avoids a great many possible failure modes. 578 */ 579 UNP_PCB_LOCK(unp); 580 if (unp->unp_vnode != NULL) { 581 UNP_PCB_UNLOCK(unp); 582 return (EINVAL); 583 } 584 if (unp->unp_flags & UNP_BINDING) { 585 UNP_PCB_UNLOCK(unp); 586 return (EALREADY); 587 } 588 unp->unp_flags |= UNP_BINDING; 589 UNP_PCB_UNLOCK(unp); 590 591 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 592 bcopy(soun->sun_path, buf, namelen); 593 buf[namelen] = 0; 594 595 restart: 596 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, 597 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 598 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 599 error = namei(&nd); 600 if (error) 601 goto error; 602 vp = nd.ni_vp; 603 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 604 NDFREE_PNBUF(&nd); 605 if (nd.ni_dvp == vp) 606 vrele(nd.ni_dvp); 607 else 608 vput(nd.ni_dvp); 609 if (vp != NULL) { 610 vrele(vp); 611 error = EADDRINUSE; 612 goto error; 613 } 614 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 615 if (error) 616 goto error; 617 goto restart; 618 } 619 VATTR_NULL(&vattr); 620 vattr.va_type = VSOCK; 621 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask); 622 #ifdef MAC 623 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 624 &vattr); 625 #endif 626 if (error == 0) 627 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 628 NDFREE_PNBUF(&nd); 629 if (error) { 630 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 631 vn_finished_write(mp); 632 if (error == ERELOOKUP) 633 goto restart; 634 goto error; 635 } 636 vp = nd.ni_vp; 637 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 638 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 639 640 UNP_PCB_LOCK(unp); 641 VOP_UNP_BIND(vp, unp); 642 unp->unp_vnode = vp; 643 unp->unp_addr = soun; 644 unp->unp_flags &= ~UNP_BINDING; 645 UNP_PCB_UNLOCK(unp); 646 vref(vp); 647 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 648 vn_finished_write(mp); 649 free(buf, M_TEMP); 650 return (0); 651 652 error: 653 UNP_PCB_LOCK(unp); 654 unp->unp_flags &= ~UNP_BINDING; 655 UNP_PCB_UNLOCK(unp); 656 free(buf, M_TEMP); 657 return (error); 658 } 659 660 static int 661 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 662 { 663 664 return (uipc_bindat(AT_FDCWD, so, nam, td)); 665 } 666 667 static int 668 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 669 { 670 int error; 671 672 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 673 error = unp_connect(so, nam, td); 674 return (error); 675 } 676 677 static int 678 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 679 struct thread *td) 680 { 681 int error; 682 683 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 684 error = unp_connectat(fd, so, nam, td, false); 685 return (error); 686 } 687 688 static void 689 uipc_close(struct socket *so) 690 { 691 struct unpcb *unp, *unp2; 692 struct vnode *vp = NULL; 693 struct mtx *vplock; 694 695 unp = sotounpcb(so); 696 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 697 698 vplock = NULL; 699 if ((vp = unp->unp_vnode) != NULL) { 700 vplock = mtx_pool_find(mtxpool_sleep, vp); 701 mtx_lock(vplock); 702 } 703 UNP_PCB_LOCK(unp); 704 if (vp && unp->unp_vnode == NULL) { 705 mtx_unlock(vplock); 706 vp = NULL; 707 } 708 if (vp != NULL) { 709 VOP_UNP_DETACH(vp); 710 unp->unp_vnode = NULL; 711 } 712 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 713 unp_disconnect(unp, unp2); 714 else 715 UNP_PCB_UNLOCK(unp); 716 if (vp) { 717 mtx_unlock(vplock); 718 vrele(vp); 719 } 720 } 721 722 static int 723 uipc_connect2(struct socket *so1, struct socket *so2) 724 { 725 struct unpcb *unp, *unp2; 726 727 if (so1->so_type != so2->so_type) 728 return (EPROTOTYPE); 729 730 unp = so1->so_pcb; 731 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 732 unp2 = so2->so_pcb; 733 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 734 unp_pcb_lock_pair(unp, unp2); 735 unp_connect2(so1, so2, PRU_CONNECT2); 736 unp_pcb_unlock_pair(unp, unp2); 737 738 return (0); 739 } 740 741 static void 742 uipc_detach(struct socket *so) 743 { 744 struct unpcb *unp, *unp2; 745 struct mtx *vplock; 746 struct vnode *vp; 747 int local_unp_rights; 748 749 unp = sotounpcb(so); 750 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 751 752 vp = NULL; 753 vplock = NULL; 754 755 UNP_LINK_WLOCK(); 756 LIST_REMOVE(unp, unp_link); 757 if (unp->unp_gcflag & UNPGC_DEAD) 758 LIST_REMOVE(unp, unp_dead); 759 unp->unp_gencnt = ++unp_gencnt; 760 --unp_count; 761 UNP_LINK_WUNLOCK(); 762 763 UNP_PCB_UNLOCK_ASSERT(unp); 764 restart: 765 if ((vp = unp->unp_vnode) != NULL) { 766 vplock = mtx_pool_find(mtxpool_sleep, vp); 767 mtx_lock(vplock); 768 } 769 UNP_PCB_LOCK(unp); 770 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 771 if (vplock) 772 mtx_unlock(vplock); 773 UNP_PCB_UNLOCK(unp); 774 goto restart; 775 } 776 if ((vp = unp->unp_vnode) != NULL) { 777 VOP_UNP_DETACH(vp); 778 unp->unp_vnode = NULL; 779 } 780 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 781 unp_disconnect(unp, unp2); 782 else 783 UNP_PCB_UNLOCK(unp); 784 785 UNP_REF_LIST_LOCK(); 786 while (!LIST_EMPTY(&unp->unp_refs)) { 787 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 788 789 unp_pcb_hold(ref); 790 UNP_REF_LIST_UNLOCK(); 791 792 MPASS(ref != unp); 793 UNP_PCB_UNLOCK_ASSERT(ref); 794 unp_drop(ref); 795 UNP_REF_LIST_LOCK(); 796 } 797 UNP_REF_LIST_UNLOCK(); 798 799 UNP_PCB_LOCK(unp); 800 local_unp_rights = unp_rights; 801 unp->unp_socket->so_pcb = NULL; 802 unp->unp_socket = NULL; 803 free(unp->unp_addr, M_SONAME); 804 unp->unp_addr = NULL; 805 if (!unp_pcb_rele(unp)) 806 UNP_PCB_UNLOCK(unp); 807 if (vp) { 808 mtx_unlock(vplock); 809 vrele(vp); 810 } 811 if (local_unp_rights) 812 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 813 814 switch (so->so_type) { 815 case SOCK_DGRAM: 816 /* 817 * Everything should have been unlinked/freed by unp_dispose() 818 * and/or unp_disconnect(). 819 */ 820 MPASS(so->so_rcv.uxdg_peeked == NULL); 821 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 822 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 823 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 824 } 825 } 826 827 static int 828 uipc_disconnect(struct socket *so) 829 { 830 struct unpcb *unp, *unp2; 831 832 unp = sotounpcb(so); 833 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 834 835 UNP_PCB_LOCK(unp); 836 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 837 unp_disconnect(unp, unp2); 838 else 839 UNP_PCB_UNLOCK(unp); 840 return (0); 841 } 842 843 static int 844 uipc_listen(struct socket *so, int backlog, struct thread *td) 845 { 846 struct unpcb *unp; 847 int error; 848 849 MPASS(so->so_type != SOCK_DGRAM); 850 851 /* 852 * Synchronize with concurrent connection attempts. 853 */ 854 error = 0; 855 unp = sotounpcb(so); 856 UNP_PCB_LOCK(unp); 857 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 858 error = EINVAL; 859 else if (unp->unp_vnode == NULL) 860 error = EDESTADDRREQ; 861 if (error != 0) { 862 UNP_PCB_UNLOCK(unp); 863 return (error); 864 } 865 866 SOCK_LOCK(so); 867 error = solisten_proto_check(so); 868 if (error == 0) { 869 cru2xt(td, &unp->unp_peercred); 870 solisten_proto(so, backlog); 871 } 872 SOCK_UNLOCK(so); 873 UNP_PCB_UNLOCK(unp); 874 return (error); 875 } 876 877 static int 878 uipc_peeraddr(struct socket *so, struct sockaddr **nam) 879 { 880 struct unpcb *unp, *unp2; 881 const struct sockaddr *sa; 882 883 unp = sotounpcb(so); 884 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 885 886 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 887 UNP_LINK_RLOCK(); 888 /* 889 * XXX: It seems that this test always fails even when connection is 890 * established. So, this else clause is added as workaround to 891 * return PF_LOCAL sockaddr. 892 */ 893 unp2 = unp->unp_conn; 894 if (unp2 != NULL) { 895 UNP_PCB_LOCK(unp2); 896 if (unp2->unp_addr != NULL) 897 sa = (struct sockaddr *) unp2->unp_addr; 898 else 899 sa = &sun_noname; 900 bcopy(sa, *nam, sa->sa_len); 901 UNP_PCB_UNLOCK(unp2); 902 } else { 903 sa = &sun_noname; 904 bcopy(sa, *nam, sa->sa_len); 905 } 906 UNP_LINK_RUNLOCK(); 907 return (0); 908 } 909 910 static int 911 uipc_rcvd(struct socket *so, int flags) 912 { 913 struct unpcb *unp, *unp2; 914 struct socket *so2; 915 u_int mbcnt, sbcc; 916 917 unp = sotounpcb(so); 918 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 919 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 920 ("%s: socktype %d", __func__, so->so_type)); 921 922 /* 923 * Adjust backpressure on sender and wakeup any waiting to write. 924 * 925 * The unp lock is acquired to maintain the validity of the unp_conn 926 * pointer; no lock on unp2 is required as unp2->unp_socket will be 927 * static as long as we don't permit unp2 to disconnect from unp, 928 * which is prevented by the lock on unp. We cache values from 929 * so_rcv to avoid holding the so_rcv lock over the entire 930 * transaction on the remote so_snd. 931 */ 932 SOCKBUF_LOCK(&so->so_rcv); 933 mbcnt = so->so_rcv.sb_mbcnt; 934 sbcc = sbavail(&so->so_rcv); 935 SOCKBUF_UNLOCK(&so->so_rcv); 936 /* 937 * There is a benign race condition at this point. If we're planning to 938 * clear SB_STOP, but uipc_send is called on the connected socket at 939 * this instant, it might add data to the sockbuf and set SB_STOP. Then 940 * we would erroneously clear SB_STOP below, even though the sockbuf is 941 * full. The race is benign because the only ill effect is to allow the 942 * sockbuf to exceed its size limit, and the size limits are not 943 * strictly guaranteed anyway. 944 */ 945 UNP_PCB_LOCK(unp); 946 unp2 = unp->unp_conn; 947 if (unp2 == NULL) { 948 UNP_PCB_UNLOCK(unp); 949 return (0); 950 } 951 so2 = unp2->unp_socket; 952 SOCKBUF_LOCK(&so2->so_snd); 953 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) 954 so2->so_snd.sb_flags &= ~SB_STOP; 955 sowwakeup_locked(so2); 956 UNP_PCB_UNLOCK(unp); 957 return (0); 958 } 959 960 static int 961 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 962 struct mbuf *control, struct thread *td) 963 { 964 struct unpcb *unp, *unp2; 965 struct socket *so2; 966 u_int mbcnt, sbcc; 967 int error; 968 969 unp = sotounpcb(so); 970 KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); 971 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, 972 ("%s: socktype %d", __func__, so->so_type)); 973 974 error = 0; 975 if (flags & PRUS_OOB) { 976 error = EOPNOTSUPP; 977 goto release; 978 } 979 if (control != NULL && 980 (error = unp_internalize(&control, td, NULL, NULL, NULL))) 981 goto release; 982 983 unp2 = NULL; 984 if ((so->so_state & SS_ISCONNECTED) == 0) { 985 if (nam != NULL) { 986 if ((error = unp_connect(so, nam, td)) != 0) 987 goto out; 988 } else { 989 error = ENOTCONN; 990 goto out; 991 } 992 } 993 994 UNP_PCB_LOCK(unp); 995 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { 996 UNP_PCB_UNLOCK(unp); 997 error = ENOTCONN; 998 goto out; 999 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1000 unp_pcb_unlock_pair(unp, unp2); 1001 error = EPIPE; 1002 goto out; 1003 } 1004 UNP_PCB_UNLOCK(unp); 1005 if ((so2 = unp2->unp_socket) == NULL) { 1006 UNP_PCB_UNLOCK(unp2); 1007 error = ENOTCONN; 1008 goto out; 1009 } 1010 SOCKBUF_LOCK(&so2->so_rcv); 1011 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 1012 /* 1013 * Credentials are passed only once on SOCK_STREAM and 1014 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 1015 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 1016 */ 1017 control = unp_addsockcred(td, control, unp2->unp_flags, NULL, 1018 NULL, NULL); 1019 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 1020 } 1021 1022 /* 1023 * Send to paired receive port and wake up readers. Don't 1024 * check for space available in the receive buffer if we're 1025 * attaching ancillary data; Unix domain sockets only check 1026 * for space in the sending sockbuf, and that check is 1027 * performed one level up the stack. At that level we cannot 1028 * precisely account for the amount of buffer space used 1029 * (e.g., because control messages are not yet internalized). 1030 */ 1031 switch (so->so_type) { 1032 case SOCK_STREAM: 1033 if (control != NULL) { 1034 sbappendcontrol_locked(&so2->so_rcv, m, 1035 control, flags); 1036 control = NULL; 1037 } else 1038 sbappend_locked(&so2->so_rcv, m, flags); 1039 break; 1040 1041 case SOCK_SEQPACKET: 1042 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, 1043 &sun_noname, m, control)) 1044 control = NULL; 1045 break; 1046 } 1047 1048 mbcnt = so2->so_rcv.sb_mbcnt; 1049 sbcc = sbavail(&so2->so_rcv); 1050 if (sbcc) 1051 sorwakeup_locked(so2); 1052 else 1053 SOCKBUF_UNLOCK(&so2->so_rcv); 1054 1055 /* 1056 * The PCB lock on unp2 protects the SB_STOP flag. Without it, 1057 * it would be possible for uipc_rcvd to be called at this 1058 * point, drain the receiving sockbuf, clear SB_STOP, and then 1059 * we would set SB_STOP below. That could lead to an empty 1060 * sockbuf having SB_STOP set 1061 */ 1062 SOCKBUF_LOCK(&so->so_snd); 1063 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) 1064 so->so_snd.sb_flags |= SB_STOP; 1065 SOCKBUF_UNLOCK(&so->so_snd); 1066 UNP_PCB_UNLOCK(unp2); 1067 m = NULL; 1068 out: 1069 /* 1070 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown. 1071 */ 1072 if (flags & PRUS_EOF) { 1073 UNP_PCB_LOCK(unp); 1074 socantsendmore(so); 1075 unp_shutdown(unp); 1076 UNP_PCB_UNLOCK(unp); 1077 } 1078 if (control != NULL && error != 0) 1079 unp_scan(control, unp_freerights); 1080 1081 release: 1082 if (control != NULL) 1083 m_freem(control); 1084 /* 1085 * In case of PRUS_NOTREADY, uipc_ready() is responsible 1086 * for freeing memory. 1087 */ 1088 if (m != NULL && (flags & PRUS_NOTREADY) == 0) 1089 m_freem(m); 1090 return (error); 1091 } 1092 1093 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1094 static inline bool 1095 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1096 { 1097 u_int bleft, mleft; 1098 1099 MPASS(sb->sb_hiwat >= sb->uxdg_cc); 1100 MPASS(sb->sb_mbmax >= sb->uxdg_mbcnt); 1101 1102 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1103 return (false); 1104 1105 bleft = sb->sb_hiwat - sb->uxdg_cc; 1106 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1107 1108 return (bleft >= cc && mleft >= mbcnt); 1109 } 1110 1111 /* 1112 * PF_UNIX/SOCK_DGRAM send 1113 * 1114 * Allocate a record consisting of 3 mbufs in the sequence of 1115 * from -> control -> data and append it to the socket buffer. 1116 * 1117 * The first mbuf carries sender's name and is a pkthdr that stores 1118 * overall length of datagram, its memory consumption and control length. 1119 */ 1120 #define ctllen PH_loc.thirtytwo[1] 1121 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1122 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1123 static int 1124 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1125 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1126 { 1127 struct unpcb *unp, *unp2; 1128 const struct sockaddr *from; 1129 struct socket *so2; 1130 struct sockbuf *sb; 1131 struct mbuf *f, *clast; 1132 u_int cc, ctl, mbcnt; 1133 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1134 int error; 1135 1136 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1137 1138 error = 0; 1139 f = NULL; 1140 ctl = 0; 1141 1142 if (__predict_false(flags & MSG_OOB)) { 1143 error = EOPNOTSUPP; 1144 goto out; 1145 } 1146 if (m == NULL) { 1147 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1148 error = EMSGSIZE; 1149 goto out; 1150 } 1151 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1152 if (__predict_false(m == NULL)) { 1153 error = EFAULT; 1154 goto out; 1155 } 1156 f = m_gethdr(M_WAITOK, MT_SONAME); 1157 cc = m->m_pkthdr.len; 1158 mbcnt = MSIZE + m->m_pkthdr.memlen; 1159 if (c != NULL && 1160 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt))) 1161 goto out; 1162 } else { 1163 /* pr_sosend() with mbuf usually is a kernel thread. */ 1164 1165 M_ASSERTPKTHDR(m); 1166 if (__predict_false(c != NULL)) 1167 panic("%s: control from a kernel thread", __func__); 1168 1169 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1170 error = EMSGSIZE; 1171 goto out; 1172 } 1173 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1174 error = ENOBUFS; 1175 goto out; 1176 } 1177 /* Condition the foreign mbuf to our standards. */ 1178 m_clrprotoflags(m); 1179 m_tag_delete_chain(m, NULL); 1180 m->m_pkthdr.rcvif = NULL; 1181 m->m_pkthdr.flowid = 0; 1182 m->m_pkthdr.csum_flags = 0; 1183 m->m_pkthdr.fibnum = 0; 1184 m->m_pkthdr.rsstype = 0; 1185 1186 cc = m->m_pkthdr.len; 1187 mbcnt = MSIZE; 1188 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) { 1189 mbcnt += MSIZE; 1190 if (mb->m_flags & M_EXT) 1191 mbcnt += mb->m_ext.ext_size; 1192 } 1193 } 1194 1195 unp = sotounpcb(so); 1196 MPASS(unp); 1197 1198 /* 1199 * XXXGL: would be cool to fully remove so_snd out of the equation 1200 * and avoid this lock, which is not only extraneous, but also being 1201 * released, thus still leaving possibility for a race. We can easily 1202 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 1203 * is more difficult to invent something to handle so_error. 1204 */ 1205 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1206 if (error) 1207 goto out2; 1208 SOCK_SENDBUF_LOCK(so); 1209 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 1210 SOCK_SENDBUF_UNLOCK(so); 1211 error = EPIPE; 1212 goto out3; 1213 } 1214 if (so->so_error != 0) { 1215 error = so->so_error; 1216 so->so_error = 0; 1217 SOCK_SENDBUF_UNLOCK(so); 1218 goto out3; 1219 } 1220 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 1221 SOCK_SENDBUF_UNLOCK(so); 1222 error = EDESTADDRREQ; 1223 goto out3; 1224 } 1225 SOCK_SENDBUF_UNLOCK(so); 1226 1227 if (addr != NULL) { 1228 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 1229 goto out3; 1230 UNP_PCB_LOCK_ASSERT(unp); 1231 unp2 = unp->unp_conn; 1232 UNP_PCB_LOCK_ASSERT(unp2); 1233 } else { 1234 UNP_PCB_LOCK(unp); 1235 unp2 = unp_pcb_lock_peer(unp); 1236 if (unp2 == NULL) { 1237 UNP_PCB_UNLOCK(unp); 1238 error = ENOTCONN; 1239 goto out3; 1240 } 1241 } 1242 1243 if (unp2->unp_flags & UNP_WANTCRED_MASK) 1244 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl, 1245 &mbcnt); 1246 if (unp->unp_addr != NULL) 1247 from = (struct sockaddr *)unp->unp_addr; 1248 else 1249 from = &sun_noname; 1250 f->m_len = from->sa_len; 1251 MPASS(from->sa_len <= MLEN); 1252 bcopy(from, mtod(f, void *), from->sa_len); 1253 ctl += f->m_len; 1254 1255 /* 1256 * Concatenate mbufs: from -> control -> data. 1257 * Save overall cc and mbcnt in "from" mbuf. 1258 */ 1259 if (c != NULL) { 1260 #ifdef INVARIANTS 1261 struct mbuf *mc; 1262 1263 for (mc = c; mc->m_next != NULL; mc = mc->m_next); 1264 MPASS(mc == clast); 1265 #endif 1266 f->m_next = c; 1267 clast->m_next = m; 1268 c = NULL; 1269 } else 1270 f->m_next = m; 1271 m = NULL; 1272 #ifdef INVARIANTS 1273 dcc = dctl = dmbcnt = 0; 1274 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 1275 if (mb->m_type == MT_DATA) 1276 dcc += mb->m_len; 1277 else 1278 dctl += mb->m_len; 1279 dmbcnt += MSIZE; 1280 if (mb->m_flags & M_EXT) 1281 dmbcnt += mb->m_ext.ext_size; 1282 } 1283 MPASS(dcc == cc); 1284 MPASS(dctl == ctl); 1285 MPASS(dmbcnt == mbcnt); 1286 #endif 1287 f->m_pkthdr.len = cc + ctl; 1288 f->m_pkthdr.memlen = mbcnt; 1289 f->m_pkthdr.ctllen = ctl; 1290 1291 /* 1292 * Destination socket buffer selection. 1293 * 1294 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 1295 * destination address is supplied, create a temporary connection for 1296 * the run time of the function (see call to unp_connectat() above and 1297 * to unp_disconnect() below). We distinguish them by condition of 1298 * (addr != NULL). We intentionally avoid adding 'bool connected' for 1299 * that condition, since, again, through the run time of this code we 1300 * are always connected. For such "unconnected" sends, the destination 1301 * buffer would be the receive buffer of destination socket so2. 1302 * 1303 * For connected sends, data lands on the send buffer of the sender's 1304 * socket "so". Then, if we just added the very first datagram 1305 * on this send buffer, we need to add the send buffer on to the 1306 * receiving socket's buffer list. We put ourselves on top of the 1307 * list. Such logic gives infrequent senders priority over frequent 1308 * senders. 1309 * 1310 * Note on byte count management. As long as event methods kevent(2), 1311 * select(2) are not protocol specific (yet), we need to maintain 1312 * meaningful values on the receive buffer. So, the receive buffer 1313 * would accumulate counters from all connected buffers potentially 1314 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 1315 */ 1316 so2 = unp2->unp_socket; 1317 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 1318 SOCK_RECVBUF_LOCK(so2); 1319 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 1320 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 1321 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 1322 uxdg_clist); 1323 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 1324 sb->uxdg_cc += cc + ctl; 1325 sb->uxdg_ctl += ctl; 1326 sb->uxdg_mbcnt += mbcnt; 1327 so2->so_rcv.sb_acc += cc + ctl; 1328 so2->so_rcv.sb_ccc += cc + ctl; 1329 so2->so_rcv.sb_ctl += ctl; 1330 so2->so_rcv.sb_mbcnt += mbcnt; 1331 sorwakeup_locked(so2); 1332 f = NULL; 1333 } else { 1334 soroverflow_locked(so2); 1335 error = (so->so_state & SS_NBIO) ? EAGAIN : ENOBUFS; 1336 } 1337 1338 if (addr != NULL) 1339 unp_disconnect(unp, unp2); 1340 else 1341 unp_pcb_unlock_pair(unp, unp2); 1342 1343 td->td_ru.ru_msgsnd++; 1344 1345 out3: 1346 SOCK_IO_SEND_UNLOCK(so); 1347 out2: 1348 if (c) 1349 unp_scan(c, unp_freerights); 1350 out: 1351 if (f) 1352 m_freem(f); 1353 if (c) 1354 m_freem(c); 1355 if (m) 1356 m_freem(m); 1357 1358 return (error); 1359 } 1360 1361 /* 1362 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1363 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1364 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1365 */ 1366 static int 1367 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1368 struct uio *uio, struct mbuf **controlp, int *flagsp) 1369 { 1370 ssize_t len = 0; 1371 int error; 1372 1373 so->so_rcv.uxdg_peeked = m; 1374 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1375 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1376 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1377 SOCK_RECVBUF_UNLOCK(so); 1378 1379 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1380 if (psa != NULL) 1381 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1382 1383 m = m->m_next; 1384 KASSERT(m, ("%s: no data or control after soname", __func__)); 1385 1386 /* 1387 * With MSG_PEEK the control isn't executed, just copied. 1388 */ 1389 while (m != NULL && m->m_type == MT_CONTROL) { 1390 if (controlp != NULL) { 1391 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1392 controlp = &(*controlp)->m_next; 1393 } 1394 m = m->m_next; 1395 } 1396 KASSERT(m == NULL || m->m_type == MT_DATA, 1397 ("%s: not MT_DATA mbuf %p", __func__, m)); 1398 while (m != NULL && uio->uio_resid > 0) { 1399 len = uio->uio_resid; 1400 if (len > m->m_len) 1401 len = m->m_len; 1402 error = uiomove(mtod(m, char *), (int)len, uio); 1403 if (error) { 1404 SOCK_IO_RECV_UNLOCK(so); 1405 return (error); 1406 } 1407 if (len == m->m_len) 1408 m = m->m_next; 1409 } 1410 SOCK_IO_RECV_UNLOCK(so); 1411 1412 if (flagsp != NULL) { 1413 if (m != NULL) { 1414 if (*flagsp & MSG_TRUNC) { 1415 /* Report real length of the packet */ 1416 uio->uio_resid -= m_length(m, NULL) - len; 1417 } 1418 *flagsp |= MSG_TRUNC; 1419 } else 1420 *flagsp &= ~MSG_TRUNC; 1421 } 1422 1423 return (0); 1424 } 1425 1426 /* 1427 * PF_UNIX/SOCK_DGRAM receive 1428 */ 1429 static int 1430 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1431 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1432 { 1433 struct sockbuf *sb = NULL; 1434 struct mbuf *m; 1435 int flags, error; 1436 ssize_t len = 0; 1437 bool nonblock; 1438 1439 MPASS(mp0 == NULL); 1440 1441 if (psa != NULL) 1442 *psa = NULL; 1443 if (controlp != NULL) 1444 *controlp = NULL; 1445 1446 flags = flagsp != NULL ? *flagsp : 0; 1447 nonblock = (so->so_state & SS_NBIO) || 1448 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1449 1450 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1451 if (__predict_false(error)) 1452 return (error); 1453 1454 /* 1455 * Loop blocking while waiting for a datagram. Prioritize connected 1456 * peers over unconnected sends. Set sb to selected socket buffer 1457 * containing an mbuf on exit from the wait loop. A datagram that 1458 * had already been peeked at has top priority. 1459 */ 1460 SOCK_RECVBUF_LOCK(so); 1461 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1462 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1463 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1464 if (so->so_error) { 1465 error = so->so_error; 1466 so->so_error = 0; 1467 SOCK_RECVBUF_UNLOCK(so); 1468 SOCK_IO_RECV_UNLOCK(so); 1469 return (error); 1470 } 1471 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1472 uio->uio_resid == 0) { 1473 SOCK_RECVBUF_UNLOCK(so); 1474 SOCK_IO_RECV_UNLOCK(so); 1475 return (0); 1476 } 1477 if (nonblock) { 1478 SOCK_RECVBUF_UNLOCK(so); 1479 SOCK_IO_RECV_UNLOCK(so); 1480 return (EWOULDBLOCK); 1481 } 1482 error = sbwait(so, SO_RCV); 1483 if (error) { 1484 SOCK_RECVBUF_UNLOCK(so); 1485 SOCK_IO_RECV_UNLOCK(so); 1486 return (error); 1487 } 1488 } 1489 1490 if (sb == NULL) 1491 sb = &so->so_rcv; 1492 else if (m == NULL) 1493 m = STAILQ_FIRST(&sb->uxdg_mb); 1494 else 1495 MPASS(m == so->so_rcv.uxdg_peeked); 1496 1497 MPASS(sb->uxdg_cc > 0); 1498 M_ASSERTPKTHDR(m); 1499 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1500 1501 if (uio->uio_td) 1502 uio->uio_td->td_ru.ru_msgrcv++; 1503 1504 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1505 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1506 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1507 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1508 } else 1509 so->so_rcv.uxdg_peeked = NULL; 1510 1511 sb->uxdg_cc -= m->m_pkthdr.len; 1512 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1513 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1514 1515 if (__predict_false(flags & MSG_PEEK)) 1516 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1517 1518 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1519 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1520 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1521 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1522 SOCK_RECVBUF_UNLOCK(so); 1523 1524 if (psa != NULL) 1525 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1526 m = m_free(m); 1527 KASSERT(m, ("%s: no data or control after soname", __func__)); 1528 1529 /* 1530 * Packet to copyout() is now in 'm' and it is disconnected from the 1531 * queue. 1532 * 1533 * Process one or more MT_CONTROL mbufs present before any data mbufs 1534 * in the first mbuf chain on the socket buffer. We call into the 1535 * unp_externalize() to perform externalization (or freeing if 1536 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1537 * without MT_DATA mbufs. 1538 */ 1539 while (m != NULL && m->m_type == MT_CONTROL) { 1540 struct mbuf *cm; 1541 1542 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1543 * it frees whole chain, so we must disconnect the mbuf. 1544 */ 1545 cm = m; m = m->m_next; cm->m_next = NULL; 1546 error = unp_externalize(cm, controlp, flags); 1547 if (error != 0) { 1548 SOCK_IO_RECV_UNLOCK(so); 1549 unp_scan(m, unp_freerights); 1550 m_freem(m); 1551 return (error); 1552 } 1553 if (controlp != NULL) { 1554 while (*controlp != NULL) 1555 controlp = &(*controlp)->m_next; 1556 } 1557 } 1558 KASSERT(m == NULL || m->m_type == MT_DATA, 1559 ("%s: not MT_DATA mbuf %p", __func__, m)); 1560 while (m != NULL && uio->uio_resid > 0) { 1561 len = uio->uio_resid; 1562 if (len > m->m_len) 1563 len = m->m_len; 1564 error = uiomove(mtod(m, char *), (int)len, uio); 1565 if (error) { 1566 SOCK_IO_RECV_UNLOCK(so); 1567 m_freem(m); 1568 return (error); 1569 } 1570 if (len == m->m_len) 1571 m = m_free(m); 1572 else { 1573 m->m_data += len; 1574 m->m_len -= len; 1575 } 1576 } 1577 SOCK_IO_RECV_UNLOCK(so); 1578 1579 if (m != NULL) { 1580 if (flagsp != NULL) { 1581 if (flags & MSG_TRUNC) { 1582 /* Report real length of the packet */ 1583 uio->uio_resid -= m_length(m, NULL); 1584 } 1585 *flagsp |= MSG_TRUNC; 1586 } 1587 m_freem(m); 1588 } else if (flagsp != NULL) 1589 *flagsp &= ~MSG_TRUNC; 1590 1591 return (0); 1592 } 1593 1594 static bool 1595 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1596 { 1597 struct mbuf *mb, *n; 1598 struct sockbuf *sb; 1599 1600 SOCK_LOCK(so); 1601 if (SOLISTENING(so)) { 1602 SOCK_UNLOCK(so); 1603 return (false); 1604 } 1605 mb = NULL; 1606 sb = &so->so_rcv; 1607 SOCKBUF_LOCK(sb); 1608 if (sb->sb_fnrdy != NULL) { 1609 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1610 if (mb == m) { 1611 *errorp = sbready(sb, m, count); 1612 break; 1613 } 1614 mb = mb->m_next; 1615 if (mb == NULL) { 1616 mb = n; 1617 if (mb != NULL) 1618 n = mb->m_nextpkt; 1619 } 1620 } 1621 } 1622 SOCKBUF_UNLOCK(sb); 1623 SOCK_UNLOCK(so); 1624 return (mb != NULL); 1625 } 1626 1627 static int 1628 uipc_ready(struct socket *so, struct mbuf *m, int count) 1629 { 1630 struct unpcb *unp, *unp2; 1631 struct socket *so2; 1632 int error, i; 1633 1634 unp = sotounpcb(so); 1635 1636 KASSERT(so->so_type == SOCK_STREAM, 1637 ("%s: unexpected socket type for %p", __func__, so)); 1638 1639 UNP_PCB_LOCK(unp); 1640 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1641 UNP_PCB_UNLOCK(unp); 1642 so2 = unp2->unp_socket; 1643 SOCKBUF_LOCK(&so2->so_rcv); 1644 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1645 sorwakeup_locked(so2); 1646 else 1647 SOCKBUF_UNLOCK(&so2->so_rcv); 1648 UNP_PCB_UNLOCK(unp2); 1649 return (error); 1650 } 1651 UNP_PCB_UNLOCK(unp); 1652 1653 /* 1654 * The receiving socket has been disconnected, but may still be valid. 1655 * In this case, the now-ready mbufs are still present in its socket 1656 * buffer, so perform an exhaustive search before giving up and freeing 1657 * the mbufs. 1658 */ 1659 UNP_LINK_RLOCK(); 1660 LIST_FOREACH(unp, &unp_shead, unp_link) { 1661 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1662 break; 1663 } 1664 UNP_LINK_RUNLOCK(); 1665 1666 if (unp == NULL) { 1667 for (i = 0; i < count; i++) 1668 m = m_free(m); 1669 error = ECONNRESET; 1670 } 1671 return (error); 1672 } 1673 1674 static int 1675 uipc_sense(struct socket *so, struct stat *sb) 1676 { 1677 struct unpcb *unp; 1678 1679 unp = sotounpcb(so); 1680 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1681 1682 sb->st_blksize = so->so_snd.sb_hiwat; 1683 sb->st_dev = NODEV; 1684 sb->st_ino = unp->unp_ino; 1685 return (0); 1686 } 1687 1688 static int 1689 uipc_shutdown(struct socket *so) 1690 { 1691 struct unpcb *unp; 1692 1693 unp = sotounpcb(so); 1694 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1695 1696 UNP_PCB_LOCK(unp); 1697 socantsendmore(so); 1698 unp_shutdown(unp); 1699 UNP_PCB_UNLOCK(unp); 1700 return (0); 1701 } 1702 1703 static int 1704 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1705 { 1706 struct unpcb *unp; 1707 const struct sockaddr *sa; 1708 1709 unp = sotounpcb(so); 1710 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1711 1712 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1713 UNP_PCB_LOCK(unp); 1714 if (unp->unp_addr != NULL) 1715 sa = (struct sockaddr *) unp->unp_addr; 1716 else 1717 sa = &sun_noname; 1718 bcopy(sa, *nam, sa->sa_len); 1719 UNP_PCB_UNLOCK(unp); 1720 return (0); 1721 } 1722 1723 static int 1724 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1725 { 1726 struct unpcb *unp; 1727 struct xucred xu; 1728 int error, optval; 1729 1730 if (sopt->sopt_level != SOL_LOCAL) 1731 return (EINVAL); 1732 1733 unp = sotounpcb(so); 1734 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1735 error = 0; 1736 switch (sopt->sopt_dir) { 1737 case SOPT_GET: 1738 switch (sopt->sopt_name) { 1739 case LOCAL_PEERCRED: 1740 UNP_PCB_LOCK(unp); 1741 if (unp->unp_flags & UNP_HAVEPC) 1742 xu = unp->unp_peercred; 1743 else { 1744 if (so->so_type == SOCK_STREAM) 1745 error = ENOTCONN; 1746 else 1747 error = EINVAL; 1748 } 1749 UNP_PCB_UNLOCK(unp); 1750 if (error == 0) 1751 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1752 break; 1753 1754 case LOCAL_CREDS: 1755 /* Unlocked read. */ 1756 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1757 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1758 break; 1759 1760 case LOCAL_CREDS_PERSISTENT: 1761 /* Unlocked read. */ 1762 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1763 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1764 break; 1765 1766 case LOCAL_CONNWAIT: 1767 /* Unlocked read. */ 1768 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1769 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1770 break; 1771 1772 default: 1773 error = EOPNOTSUPP; 1774 break; 1775 } 1776 break; 1777 1778 case SOPT_SET: 1779 switch (sopt->sopt_name) { 1780 case LOCAL_CREDS: 1781 case LOCAL_CREDS_PERSISTENT: 1782 case LOCAL_CONNWAIT: 1783 error = sooptcopyin(sopt, &optval, sizeof(optval), 1784 sizeof(optval)); 1785 if (error) 1786 break; 1787 1788 #define OPTSET(bit, exclusive) do { \ 1789 UNP_PCB_LOCK(unp); \ 1790 if (optval) { \ 1791 if ((unp->unp_flags & (exclusive)) != 0) { \ 1792 UNP_PCB_UNLOCK(unp); \ 1793 error = EINVAL; \ 1794 break; \ 1795 } \ 1796 unp->unp_flags |= (bit); \ 1797 } else \ 1798 unp->unp_flags &= ~(bit); \ 1799 UNP_PCB_UNLOCK(unp); \ 1800 } while (0) 1801 1802 switch (sopt->sopt_name) { 1803 case LOCAL_CREDS: 1804 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1805 break; 1806 1807 case LOCAL_CREDS_PERSISTENT: 1808 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1809 break; 1810 1811 case LOCAL_CONNWAIT: 1812 OPTSET(UNP_CONNWAIT, 0); 1813 break; 1814 1815 default: 1816 break; 1817 } 1818 break; 1819 #undef OPTSET 1820 default: 1821 error = ENOPROTOOPT; 1822 break; 1823 } 1824 break; 1825 1826 default: 1827 error = EOPNOTSUPP; 1828 break; 1829 } 1830 return (error); 1831 } 1832 1833 static int 1834 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1835 { 1836 1837 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1838 } 1839 1840 static int 1841 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1842 struct thread *td, bool return_locked) 1843 { 1844 struct mtx *vplock; 1845 struct sockaddr_un *soun; 1846 struct vnode *vp; 1847 struct socket *so2; 1848 struct unpcb *unp, *unp2, *unp3; 1849 struct nameidata nd; 1850 char buf[SOCK_MAXADDRLEN]; 1851 struct sockaddr *sa; 1852 cap_rights_t rights; 1853 int error, len; 1854 bool connreq; 1855 1856 if (nam->sa_family != AF_UNIX) 1857 return (EAFNOSUPPORT); 1858 if (nam->sa_len > sizeof(struct sockaddr_un)) 1859 return (EINVAL); 1860 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1861 if (len <= 0) 1862 return (EINVAL); 1863 soun = (struct sockaddr_un *)nam; 1864 bcopy(soun->sun_path, buf, len); 1865 buf[len] = 0; 1866 1867 error = 0; 1868 unp = sotounpcb(so); 1869 UNP_PCB_LOCK(unp); 1870 for (;;) { 1871 /* 1872 * Wait for connection state to stabilize. If a connection 1873 * already exists, give up. For datagram sockets, which permit 1874 * multiple consecutive connect(2) calls, upper layers are 1875 * responsible for disconnecting in advance of a subsequent 1876 * connect(2), but this is not synchronized with PCB connection 1877 * state. 1878 * 1879 * Also make sure that no threads are currently attempting to 1880 * lock the peer socket, to ensure that unp_conn cannot 1881 * transition between two valid sockets while locks are dropped. 1882 */ 1883 if (SOLISTENING(so)) 1884 error = EOPNOTSUPP; 1885 else if (unp->unp_conn != NULL) 1886 error = EISCONN; 1887 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1888 error = EALREADY; 1889 } 1890 if (error != 0) { 1891 UNP_PCB_UNLOCK(unp); 1892 return (error); 1893 } 1894 if (unp->unp_pairbusy > 0) { 1895 unp->unp_flags |= UNP_WAITING; 1896 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1897 continue; 1898 } 1899 break; 1900 } 1901 unp->unp_flags |= UNP_CONNECTING; 1902 UNP_PCB_UNLOCK(unp); 1903 1904 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1905 if (connreq) 1906 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1907 else 1908 sa = NULL; 1909 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1910 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1911 error = namei(&nd); 1912 if (error) 1913 vp = NULL; 1914 else 1915 vp = nd.ni_vp; 1916 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1917 NDFREE_NOTHING(&nd); 1918 if (error) 1919 goto bad; 1920 1921 if (vp->v_type != VSOCK) { 1922 error = ENOTSOCK; 1923 goto bad; 1924 } 1925 #ifdef MAC 1926 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1927 if (error) 1928 goto bad; 1929 #endif 1930 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1931 if (error) 1932 goto bad; 1933 1934 unp = sotounpcb(so); 1935 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1936 1937 vplock = mtx_pool_find(mtxpool_sleep, vp); 1938 mtx_lock(vplock); 1939 VOP_UNP_CONNECT(vp, &unp2); 1940 if (unp2 == NULL) { 1941 error = ECONNREFUSED; 1942 goto bad2; 1943 } 1944 so2 = unp2->unp_socket; 1945 if (so->so_type != so2->so_type) { 1946 error = EPROTOTYPE; 1947 goto bad2; 1948 } 1949 if (connreq) { 1950 if (SOLISTENING(so2)) { 1951 CURVNET_SET(so2->so_vnet); 1952 so2 = sonewconn(so2, 0); 1953 CURVNET_RESTORE(); 1954 } else 1955 so2 = NULL; 1956 if (so2 == NULL) { 1957 error = ECONNREFUSED; 1958 goto bad2; 1959 } 1960 unp3 = sotounpcb(so2); 1961 unp_pcb_lock_pair(unp2, unp3); 1962 if (unp2->unp_addr != NULL) { 1963 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1964 unp3->unp_addr = (struct sockaddr_un *) sa; 1965 sa = NULL; 1966 } 1967 1968 unp_copy_peercred(td, unp3, unp, unp2); 1969 1970 UNP_PCB_UNLOCK(unp2); 1971 unp2 = unp3; 1972 1973 /* 1974 * It is safe to block on the PCB lock here since unp2 is 1975 * nascent and cannot be connected to any other sockets. 1976 */ 1977 UNP_PCB_LOCK(unp); 1978 #ifdef MAC 1979 mac_socketpeer_set_from_socket(so, so2); 1980 mac_socketpeer_set_from_socket(so2, so); 1981 #endif 1982 } else { 1983 unp_pcb_lock_pair(unp, unp2); 1984 } 1985 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1986 sotounpcb(so2) == unp2, 1987 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1988 unp_connect2(so, so2, PRU_CONNECT); 1989 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1990 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1991 unp->unp_flags &= ~UNP_CONNECTING; 1992 if (!return_locked) 1993 unp_pcb_unlock_pair(unp, unp2); 1994 bad2: 1995 mtx_unlock(vplock); 1996 bad: 1997 if (vp != NULL) { 1998 /* 1999 * If we are returning locked (called via uipc_sosend_dgram()), 2000 * we need to be sure that vput() won't sleep. This is 2001 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2002 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2003 */ 2004 MPASS(!(return_locked && connreq)); 2005 vput(vp); 2006 } 2007 free(sa, M_SONAME); 2008 if (__predict_false(error)) { 2009 UNP_PCB_LOCK(unp); 2010 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2011 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2012 unp->unp_flags &= ~UNP_CONNECTING; 2013 UNP_PCB_UNLOCK(unp); 2014 } 2015 return (error); 2016 } 2017 2018 /* 2019 * Set socket peer credentials at connection time. 2020 * 2021 * The client's PCB credentials are copied from its process structure. The 2022 * server's PCB credentials are copied from the socket on which it called 2023 * listen(2). uipc_listen cached that process's credentials at the time. 2024 */ 2025 void 2026 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2027 struct unpcb *server_unp, struct unpcb *listen_unp) 2028 { 2029 cru2xt(td, &client_unp->unp_peercred); 2030 client_unp->unp_flags |= UNP_HAVEPC; 2031 2032 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2033 sizeof(server_unp->unp_peercred)); 2034 server_unp->unp_flags |= UNP_HAVEPC; 2035 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2036 } 2037 2038 static void 2039 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2040 { 2041 struct unpcb *unp; 2042 struct unpcb *unp2; 2043 2044 MPASS(so2->so_type == so->so_type); 2045 unp = sotounpcb(so); 2046 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2047 unp2 = sotounpcb(so2); 2048 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2049 2050 UNP_PCB_LOCK_ASSERT(unp); 2051 UNP_PCB_LOCK_ASSERT(unp2); 2052 KASSERT(unp->unp_conn == NULL, 2053 ("%s: socket %p is already connected", __func__, unp)); 2054 2055 unp->unp_conn = unp2; 2056 unp_pcb_hold(unp2); 2057 unp_pcb_hold(unp); 2058 switch (so->so_type) { 2059 case SOCK_DGRAM: 2060 UNP_REF_LIST_LOCK(); 2061 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2062 UNP_REF_LIST_UNLOCK(); 2063 soisconnected(so); 2064 break; 2065 2066 case SOCK_STREAM: 2067 case SOCK_SEQPACKET: 2068 KASSERT(unp2->unp_conn == NULL, 2069 ("%s: socket %p is already connected", __func__, unp2)); 2070 unp2->unp_conn = unp; 2071 if (req == PRU_CONNECT && 2072 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2073 soisconnecting(so); 2074 else 2075 soisconnected(so); 2076 soisconnected(so2); 2077 break; 2078 2079 default: 2080 panic("unp_connect2"); 2081 } 2082 } 2083 2084 static void 2085 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2086 { 2087 struct socket *so, *so2; 2088 struct mbuf *m = NULL; 2089 #ifdef INVARIANTS 2090 struct unpcb *unptmp; 2091 #endif 2092 2093 UNP_PCB_LOCK_ASSERT(unp); 2094 UNP_PCB_LOCK_ASSERT(unp2); 2095 KASSERT(unp->unp_conn == unp2, 2096 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2097 2098 unp->unp_conn = NULL; 2099 so = unp->unp_socket; 2100 so2 = unp2->unp_socket; 2101 switch (unp->unp_socket->so_type) { 2102 case SOCK_DGRAM: 2103 /* 2104 * Remove our send socket buffer from the peer's receive buffer. 2105 * Move the data to the receive buffer only if it is empty. 2106 * This is a protection against a scenario where a peer 2107 * connects, floods and disconnects, effectively blocking 2108 * sendto() from unconnected sockets. 2109 */ 2110 SOCK_RECVBUF_LOCK(so2); 2111 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2112 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2113 uxdg_clist); 2114 if (__predict_true((so2->so_rcv.sb_state & 2115 SBS_CANTRCVMORE) == 0) && 2116 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2117 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2118 &so->so_snd.uxdg_mb); 2119 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2120 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2121 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2122 } else { 2123 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2124 STAILQ_INIT(&so->so_snd.uxdg_mb); 2125 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2126 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2127 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2128 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2129 } 2130 /* Note: so may reconnect. */ 2131 so->so_snd.uxdg_cc = 0; 2132 so->so_snd.uxdg_ctl = 0; 2133 so->so_snd.uxdg_mbcnt = 0; 2134 } 2135 SOCK_RECVBUF_UNLOCK(so2); 2136 UNP_REF_LIST_LOCK(); 2137 #ifdef INVARIANTS 2138 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2139 if (unptmp == unp) 2140 break; 2141 } 2142 KASSERT(unptmp != NULL, 2143 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2144 #endif 2145 LIST_REMOVE(unp, unp_reflink); 2146 UNP_REF_LIST_UNLOCK(); 2147 if (so) { 2148 SOCK_LOCK(so); 2149 so->so_state &= ~SS_ISCONNECTED; 2150 SOCK_UNLOCK(so); 2151 } 2152 break; 2153 2154 case SOCK_STREAM: 2155 case SOCK_SEQPACKET: 2156 if (so) 2157 soisdisconnected(so); 2158 MPASS(unp2->unp_conn == unp); 2159 unp2->unp_conn = NULL; 2160 if (so2) 2161 soisdisconnected(so2); 2162 break; 2163 } 2164 2165 if (unp == unp2) { 2166 unp_pcb_rele_notlast(unp); 2167 if (!unp_pcb_rele(unp)) 2168 UNP_PCB_UNLOCK(unp); 2169 } else { 2170 if (!unp_pcb_rele(unp)) 2171 UNP_PCB_UNLOCK(unp); 2172 if (!unp_pcb_rele(unp2)) 2173 UNP_PCB_UNLOCK(unp2); 2174 } 2175 2176 if (m != NULL) { 2177 unp_scan(m, unp_freerights); 2178 m_freem(m); 2179 } 2180 } 2181 2182 /* 2183 * unp_pcblist() walks the global list of struct unpcb's to generate a 2184 * pointer list, bumping the refcount on each unpcb. It then copies them out 2185 * sequentially, validating the generation number on each to see if it has 2186 * been detached. All of this is necessary because copyout() may sleep on 2187 * disk I/O. 2188 */ 2189 static int 2190 unp_pcblist(SYSCTL_HANDLER_ARGS) 2191 { 2192 struct unpcb *unp, **unp_list; 2193 unp_gen_t gencnt; 2194 struct xunpgen *xug; 2195 struct unp_head *head; 2196 struct xunpcb *xu; 2197 u_int i; 2198 int error, n; 2199 2200 switch ((intptr_t)arg1) { 2201 case SOCK_STREAM: 2202 head = &unp_shead; 2203 break; 2204 2205 case SOCK_DGRAM: 2206 head = &unp_dhead; 2207 break; 2208 2209 case SOCK_SEQPACKET: 2210 head = &unp_sphead; 2211 break; 2212 2213 default: 2214 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2215 } 2216 2217 /* 2218 * The process of preparing the PCB list is too time-consuming and 2219 * resource-intensive to repeat twice on every request. 2220 */ 2221 if (req->oldptr == NULL) { 2222 n = unp_count; 2223 req->oldidx = 2 * (sizeof *xug) 2224 + (n + n/8) * sizeof(struct xunpcb); 2225 return (0); 2226 } 2227 2228 if (req->newptr != NULL) 2229 return (EPERM); 2230 2231 /* 2232 * OK, now we're committed to doing something. 2233 */ 2234 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2235 UNP_LINK_RLOCK(); 2236 gencnt = unp_gencnt; 2237 n = unp_count; 2238 UNP_LINK_RUNLOCK(); 2239 2240 xug->xug_len = sizeof *xug; 2241 xug->xug_count = n; 2242 xug->xug_gen = gencnt; 2243 xug->xug_sogen = so_gencnt; 2244 error = SYSCTL_OUT(req, xug, sizeof *xug); 2245 if (error) { 2246 free(xug, M_TEMP); 2247 return (error); 2248 } 2249 2250 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2251 2252 UNP_LINK_RLOCK(); 2253 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2254 unp = LIST_NEXT(unp, unp_link)) { 2255 UNP_PCB_LOCK(unp); 2256 if (unp->unp_gencnt <= gencnt) { 2257 if (cr_cansee(req->td->td_ucred, 2258 unp->unp_socket->so_cred)) { 2259 UNP_PCB_UNLOCK(unp); 2260 continue; 2261 } 2262 unp_list[i++] = unp; 2263 unp_pcb_hold(unp); 2264 } 2265 UNP_PCB_UNLOCK(unp); 2266 } 2267 UNP_LINK_RUNLOCK(); 2268 n = i; /* In case we lost some during malloc. */ 2269 2270 error = 0; 2271 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2272 for (i = 0; i < n; i++) { 2273 unp = unp_list[i]; 2274 UNP_PCB_LOCK(unp); 2275 if (unp_pcb_rele(unp)) 2276 continue; 2277 2278 if (unp->unp_gencnt <= gencnt) { 2279 xu->xu_len = sizeof *xu; 2280 xu->xu_unpp = (uintptr_t)unp; 2281 /* 2282 * XXX - need more locking here to protect against 2283 * connect/disconnect races for SMP. 2284 */ 2285 if (unp->unp_addr != NULL) 2286 bcopy(unp->unp_addr, &xu->xu_addr, 2287 unp->unp_addr->sun_len); 2288 else 2289 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2290 if (unp->unp_conn != NULL && 2291 unp->unp_conn->unp_addr != NULL) 2292 bcopy(unp->unp_conn->unp_addr, 2293 &xu->xu_caddr, 2294 unp->unp_conn->unp_addr->sun_len); 2295 else 2296 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2297 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2298 xu->unp_conn = (uintptr_t)unp->unp_conn; 2299 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2300 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2301 xu->unp_gencnt = unp->unp_gencnt; 2302 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2303 UNP_PCB_UNLOCK(unp); 2304 error = SYSCTL_OUT(req, xu, sizeof *xu); 2305 } else { 2306 UNP_PCB_UNLOCK(unp); 2307 } 2308 } 2309 free(xu, M_TEMP); 2310 if (!error) { 2311 /* 2312 * Give the user an updated idea of our state. If the 2313 * generation differs from what we told her before, she knows 2314 * that something happened while we were processing this 2315 * request, and it might be necessary to retry. 2316 */ 2317 xug->xug_gen = unp_gencnt; 2318 xug->xug_sogen = so_gencnt; 2319 xug->xug_count = unp_count; 2320 error = SYSCTL_OUT(req, xug, sizeof *xug); 2321 } 2322 free(unp_list, M_TEMP); 2323 free(xug, M_TEMP); 2324 return (error); 2325 } 2326 2327 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2328 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2329 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2330 "List of active local datagram sockets"); 2331 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2332 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2333 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2334 "List of active local stream sockets"); 2335 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2336 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2337 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2338 "List of active local seqpacket sockets"); 2339 2340 static void 2341 unp_shutdown(struct unpcb *unp) 2342 { 2343 struct unpcb *unp2; 2344 struct socket *so; 2345 2346 UNP_PCB_LOCK_ASSERT(unp); 2347 2348 unp2 = unp->unp_conn; 2349 if ((unp->unp_socket->so_type == SOCK_STREAM || 2350 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2351 so = unp2->unp_socket; 2352 if (so != NULL) 2353 socantrcvmore(so); 2354 } 2355 } 2356 2357 static void 2358 unp_drop(struct unpcb *unp) 2359 { 2360 struct socket *so; 2361 struct unpcb *unp2; 2362 2363 /* 2364 * Regardless of whether the socket's peer dropped the connection 2365 * with this socket by aborting or disconnecting, POSIX requires 2366 * that ECONNRESET is returned. 2367 */ 2368 2369 UNP_PCB_LOCK(unp); 2370 so = unp->unp_socket; 2371 if (so) 2372 so->so_error = ECONNRESET; 2373 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2374 /* Last reference dropped in unp_disconnect(). */ 2375 unp_pcb_rele_notlast(unp); 2376 unp_disconnect(unp, unp2); 2377 } else if (!unp_pcb_rele(unp)) { 2378 UNP_PCB_UNLOCK(unp); 2379 } 2380 } 2381 2382 static void 2383 unp_freerights(struct filedescent **fdep, int fdcount) 2384 { 2385 struct file *fp; 2386 int i; 2387 2388 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2389 2390 for (i = 0; i < fdcount; i++) { 2391 fp = fdep[i]->fde_file; 2392 filecaps_free(&fdep[i]->fde_caps); 2393 unp_discard(fp); 2394 } 2395 free(fdep[0], M_FILECAPS); 2396 } 2397 2398 static int 2399 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2400 { 2401 struct thread *td = curthread; /* XXX */ 2402 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2403 int i; 2404 int *fdp; 2405 struct filedesc *fdesc = td->td_proc->p_fd; 2406 struct filedescent **fdep; 2407 void *data; 2408 socklen_t clen = control->m_len, datalen; 2409 int error, newfds; 2410 u_int newlen; 2411 2412 UNP_LINK_UNLOCK_ASSERT(); 2413 2414 error = 0; 2415 if (controlp != NULL) /* controlp == NULL => free control messages */ 2416 *controlp = NULL; 2417 while (cm != NULL) { 2418 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2419 2420 data = CMSG_DATA(cm); 2421 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2422 if (cm->cmsg_level == SOL_SOCKET 2423 && cm->cmsg_type == SCM_RIGHTS) { 2424 newfds = datalen / sizeof(*fdep); 2425 if (newfds == 0) 2426 goto next; 2427 fdep = data; 2428 2429 /* If we're not outputting the descriptors free them. */ 2430 if (error || controlp == NULL) { 2431 unp_freerights(fdep, newfds); 2432 goto next; 2433 } 2434 FILEDESC_XLOCK(fdesc); 2435 2436 /* 2437 * Now change each pointer to an fd in the global 2438 * table to an integer that is the index to the local 2439 * fd table entry that we set up to point to the 2440 * global one we are transferring. 2441 */ 2442 newlen = newfds * sizeof(int); 2443 *controlp = sbcreatecontrol(NULL, newlen, 2444 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2445 2446 fdp = (int *) 2447 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2448 if ((error = fdallocn(td, 0, fdp, newfds))) { 2449 FILEDESC_XUNLOCK(fdesc); 2450 unp_freerights(fdep, newfds); 2451 m_freem(*controlp); 2452 *controlp = NULL; 2453 goto next; 2454 } 2455 for (i = 0; i < newfds; i++, fdp++) { 2456 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2457 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2458 &fdep[i]->fde_caps); 2459 unp_externalize_fp(fdep[i]->fde_file); 2460 } 2461 2462 /* 2463 * The new type indicates that the mbuf data refers to 2464 * kernel resources that may need to be released before 2465 * the mbuf is freed. 2466 */ 2467 m_chtype(*controlp, MT_EXTCONTROL); 2468 FILEDESC_XUNLOCK(fdesc); 2469 free(fdep[0], M_FILECAPS); 2470 } else { 2471 /* We can just copy anything else across. */ 2472 if (error || controlp == NULL) 2473 goto next; 2474 *controlp = sbcreatecontrol(NULL, datalen, 2475 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2476 bcopy(data, 2477 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2478 datalen); 2479 } 2480 controlp = &(*controlp)->m_next; 2481 2482 next: 2483 if (CMSG_SPACE(datalen) < clen) { 2484 clen -= CMSG_SPACE(datalen); 2485 cm = (struct cmsghdr *) 2486 ((caddr_t)cm + CMSG_SPACE(datalen)); 2487 } else { 2488 clen = 0; 2489 cm = NULL; 2490 } 2491 } 2492 2493 m_freem(control); 2494 return (error); 2495 } 2496 2497 static void 2498 unp_zone_change(void *tag) 2499 { 2500 2501 uma_zone_set_max(unp_zone, maxsockets); 2502 } 2503 2504 #ifdef INVARIANTS 2505 static void 2506 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2507 { 2508 struct unpcb *unp; 2509 2510 unp = mem; 2511 2512 KASSERT(LIST_EMPTY(&unp->unp_refs), 2513 ("%s: unpcb %p has lingering refs", __func__, unp)); 2514 KASSERT(unp->unp_socket == NULL, 2515 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2516 KASSERT(unp->unp_vnode == NULL, 2517 ("%s: unpcb %p has vnode references", __func__, unp)); 2518 KASSERT(unp->unp_conn == NULL, 2519 ("%s: unpcb %p is still connected", __func__, unp)); 2520 KASSERT(unp->unp_addr == NULL, 2521 ("%s: unpcb %p has leaked addr", __func__, unp)); 2522 } 2523 #endif 2524 2525 static void 2526 unp_init(void *arg __unused) 2527 { 2528 uma_dtor dtor; 2529 2530 #ifdef INVARIANTS 2531 dtor = unp_zdtor; 2532 #else 2533 dtor = NULL; 2534 #endif 2535 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2536 NULL, NULL, UMA_ALIGN_CACHE, 0); 2537 uma_zone_set_max(unp_zone, maxsockets); 2538 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2539 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2540 NULL, EVENTHANDLER_PRI_ANY); 2541 LIST_INIT(&unp_dhead); 2542 LIST_INIT(&unp_shead); 2543 LIST_INIT(&unp_sphead); 2544 SLIST_INIT(&unp_defers); 2545 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2546 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2547 UNP_LINK_LOCK_INIT(); 2548 UNP_DEFERRED_LOCK_INIT(); 2549 } 2550 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2551 2552 static void 2553 unp_internalize_cleanup_rights(struct mbuf *control) 2554 { 2555 struct cmsghdr *cp; 2556 struct mbuf *m; 2557 void *data; 2558 socklen_t datalen; 2559 2560 for (m = control; m != NULL; m = m->m_next) { 2561 cp = mtod(m, struct cmsghdr *); 2562 if (cp->cmsg_level != SOL_SOCKET || 2563 cp->cmsg_type != SCM_RIGHTS) 2564 continue; 2565 data = CMSG_DATA(cp); 2566 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2567 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2568 } 2569 } 2570 2571 static int 2572 unp_internalize(struct mbuf **controlp, struct thread *td, 2573 struct mbuf **clast, u_int *space, u_int *mbcnt) 2574 { 2575 struct mbuf *control, **initial_controlp; 2576 struct proc *p; 2577 struct filedesc *fdesc; 2578 struct bintime *bt; 2579 struct cmsghdr *cm; 2580 struct cmsgcred *cmcred; 2581 struct filedescent *fde, **fdep, *fdev; 2582 struct file *fp; 2583 struct timeval *tv; 2584 struct timespec *ts; 2585 void *data; 2586 socklen_t clen, datalen; 2587 int i, j, error, *fdp, oldfds; 2588 u_int newlen; 2589 2590 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2591 UNP_LINK_UNLOCK_ASSERT(); 2592 2593 p = td->td_proc; 2594 fdesc = p->p_fd; 2595 error = 0; 2596 control = *controlp; 2597 *controlp = NULL; 2598 initial_controlp = controlp; 2599 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2600 data = CMSG_DATA(cm); 2601 2602 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2603 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2604 (char *)cm + cm->cmsg_len >= (char *)data; 2605 2606 clen -= min(CMSG_SPACE(datalen), clen), 2607 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2608 data = CMSG_DATA(cm)) { 2609 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2610 switch (cm->cmsg_type) { 2611 case SCM_CREDS: 2612 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2613 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2614 cmcred = (struct cmsgcred *) 2615 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2616 cmcred->cmcred_pid = p->p_pid; 2617 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2618 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2619 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2620 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2621 CMGROUP_MAX); 2622 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2623 cmcred->cmcred_groups[i] = 2624 td->td_ucred->cr_groups[i]; 2625 break; 2626 2627 case SCM_RIGHTS: 2628 oldfds = datalen / sizeof (int); 2629 if (oldfds == 0) 2630 continue; 2631 /* On some machines sizeof pointer is bigger than 2632 * sizeof int, so we need to check if data fits into 2633 * single mbuf. We could allocate several mbufs, and 2634 * unp_externalize() should even properly handle that. 2635 * But it is not worth to complicate the code for an 2636 * insane scenario of passing over 200 file descriptors 2637 * at once. 2638 */ 2639 newlen = oldfds * sizeof(fdep[0]); 2640 if (CMSG_SPACE(newlen) > MCLBYTES) { 2641 error = EMSGSIZE; 2642 goto out; 2643 } 2644 /* 2645 * Check that all the FDs passed in refer to legal 2646 * files. If not, reject the entire operation. 2647 */ 2648 fdp = data; 2649 FILEDESC_SLOCK(fdesc); 2650 for (i = 0; i < oldfds; i++, fdp++) { 2651 fp = fget_noref(fdesc, *fdp); 2652 if (fp == NULL) { 2653 FILEDESC_SUNLOCK(fdesc); 2654 error = EBADF; 2655 goto out; 2656 } 2657 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2658 FILEDESC_SUNLOCK(fdesc); 2659 error = EOPNOTSUPP; 2660 goto out; 2661 } 2662 } 2663 2664 /* 2665 * Now replace the integer FDs with pointers to the 2666 * file structure and capability rights. 2667 */ 2668 *controlp = sbcreatecontrol(NULL, newlen, 2669 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2670 fdp = data; 2671 for (i = 0; i < oldfds; i++, fdp++) { 2672 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2673 fdp = data; 2674 for (j = 0; j < i; j++, fdp++) { 2675 fdrop(fdesc->fd_ofiles[*fdp]. 2676 fde_file, td); 2677 } 2678 FILEDESC_SUNLOCK(fdesc); 2679 error = EBADF; 2680 goto out; 2681 } 2682 } 2683 fdp = data; 2684 fdep = (struct filedescent **) 2685 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2686 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2687 M_WAITOK); 2688 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2689 fde = &fdesc->fd_ofiles[*fdp]; 2690 fdep[i] = fdev; 2691 fdep[i]->fde_file = fde->fde_file; 2692 filecaps_copy(&fde->fde_caps, 2693 &fdep[i]->fde_caps, true); 2694 unp_internalize_fp(fdep[i]->fde_file); 2695 } 2696 FILEDESC_SUNLOCK(fdesc); 2697 break; 2698 2699 case SCM_TIMESTAMP: 2700 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2701 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2702 tv = (struct timeval *) 2703 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2704 microtime(tv); 2705 break; 2706 2707 case SCM_BINTIME: 2708 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2709 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2710 bt = (struct bintime *) 2711 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2712 bintime(bt); 2713 break; 2714 2715 case SCM_REALTIME: 2716 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2717 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2718 ts = (struct timespec *) 2719 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2720 nanotime(ts); 2721 break; 2722 2723 case SCM_MONOTONIC: 2724 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2725 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2726 ts = (struct timespec *) 2727 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2728 nanouptime(ts); 2729 break; 2730 2731 default: 2732 error = EINVAL; 2733 goto out; 2734 } 2735 2736 if (space != NULL) { 2737 *space += (*controlp)->m_len; 2738 *mbcnt += MSIZE; 2739 if ((*controlp)->m_flags & M_EXT) 2740 *mbcnt += (*controlp)->m_ext.ext_size; 2741 *clast = *controlp; 2742 } 2743 controlp = &(*controlp)->m_next; 2744 } 2745 if (clen > 0) 2746 error = EINVAL; 2747 2748 out: 2749 if (error != 0 && initial_controlp != NULL) 2750 unp_internalize_cleanup_rights(*initial_controlp); 2751 m_freem(control); 2752 return (error); 2753 } 2754 2755 static struct mbuf * 2756 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2757 struct mbuf **clast, u_int *space, u_int *mbcnt) 2758 { 2759 struct mbuf *m, *n, *n_prev; 2760 const struct cmsghdr *cm; 2761 int ngroups, i, cmsgtype; 2762 size_t ctrlsz; 2763 2764 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2765 if (mode & UNP_WANTCRED_ALWAYS) { 2766 ctrlsz = SOCKCRED2SIZE(ngroups); 2767 cmsgtype = SCM_CREDS2; 2768 } else { 2769 ctrlsz = SOCKCREDSIZE(ngroups); 2770 cmsgtype = SCM_CREDS; 2771 } 2772 2773 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2774 if (m == NULL) 2775 return (control); 2776 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2777 2778 if (mode & UNP_WANTCRED_ALWAYS) { 2779 struct sockcred2 *sc; 2780 2781 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2782 sc->sc_version = 0; 2783 sc->sc_pid = td->td_proc->p_pid; 2784 sc->sc_uid = td->td_ucred->cr_ruid; 2785 sc->sc_euid = td->td_ucred->cr_uid; 2786 sc->sc_gid = td->td_ucred->cr_rgid; 2787 sc->sc_egid = td->td_ucred->cr_gid; 2788 sc->sc_ngroups = ngroups; 2789 for (i = 0; i < sc->sc_ngroups; i++) 2790 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2791 } else { 2792 struct sockcred *sc; 2793 2794 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2795 sc->sc_uid = td->td_ucred->cr_ruid; 2796 sc->sc_euid = td->td_ucred->cr_uid; 2797 sc->sc_gid = td->td_ucred->cr_rgid; 2798 sc->sc_egid = td->td_ucred->cr_gid; 2799 sc->sc_ngroups = ngroups; 2800 for (i = 0; i < sc->sc_ngroups; i++) 2801 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2802 } 2803 2804 /* 2805 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2806 * created SCM_CREDS control message (struct sockcred) has another 2807 * format. 2808 */ 2809 if (control != NULL && cmsgtype == SCM_CREDS) 2810 for (n = control, n_prev = NULL; n != NULL;) { 2811 cm = mtod(n, struct cmsghdr *); 2812 if (cm->cmsg_level == SOL_SOCKET && 2813 cm->cmsg_type == SCM_CREDS) { 2814 if (n_prev == NULL) 2815 control = n->m_next; 2816 else 2817 n_prev->m_next = n->m_next; 2818 if (space != NULL) { 2819 MPASS(*space >= n->m_len); 2820 *space -= n->m_len; 2821 MPASS(*mbcnt >= MSIZE); 2822 *mbcnt -= MSIZE; 2823 if (n->m_flags & M_EXT) { 2824 MPASS(*mbcnt >= 2825 n->m_ext.ext_size); 2826 *mbcnt -= n->m_ext.ext_size; 2827 } 2828 MPASS(clast); 2829 if (*clast == n) { 2830 MPASS(n->m_next == NULL); 2831 if (n_prev == NULL) 2832 *clast = m; 2833 else 2834 *clast = n_prev; 2835 } 2836 } 2837 n = m_free(n); 2838 } else { 2839 n_prev = n; 2840 n = n->m_next; 2841 } 2842 } 2843 2844 /* Prepend it to the head. */ 2845 m->m_next = control; 2846 if (space != NULL) { 2847 *space += m->m_len; 2848 *mbcnt += MSIZE; 2849 if (control == NULL) 2850 *clast = m; 2851 } 2852 return (m); 2853 } 2854 2855 static struct unpcb * 2856 fptounp(struct file *fp) 2857 { 2858 struct socket *so; 2859 2860 if (fp->f_type != DTYPE_SOCKET) 2861 return (NULL); 2862 if ((so = fp->f_data) == NULL) 2863 return (NULL); 2864 if (so->so_proto->pr_domain != &localdomain) 2865 return (NULL); 2866 return sotounpcb(so); 2867 } 2868 2869 static void 2870 unp_discard(struct file *fp) 2871 { 2872 struct unp_defer *dr; 2873 2874 if (unp_externalize_fp(fp)) { 2875 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2876 dr->ud_fp = fp; 2877 UNP_DEFERRED_LOCK(); 2878 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2879 UNP_DEFERRED_UNLOCK(); 2880 atomic_add_int(&unp_defers_count, 1); 2881 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2882 } else 2883 closef_nothread(fp); 2884 } 2885 2886 static void 2887 unp_process_defers(void *arg __unused, int pending) 2888 { 2889 struct unp_defer *dr; 2890 SLIST_HEAD(, unp_defer) drl; 2891 int count; 2892 2893 SLIST_INIT(&drl); 2894 for (;;) { 2895 UNP_DEFERRED_LOCK(); 2896 if (SLIST_FIRST(&unp_defers) == NULL) { 2897 UNP_DEFERRED_UNLOCK(); 2898 break; 2899 } 2900 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2901 UNP_DEFERRED_UNLOCK(); 2902 count = 0; 2903 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2904 SLIST_REMOVE_HEAD(&drl, ud_link); 2905 closef_nothread(dr->ud_fp); 2906 free(dr, M_TEMP); 2907 count++; 2908 } 2909 atomic_add_int(&unp_defers_count, -count); 2910 } 2911 } 2912 2913 static void 2914 unp_internalize_fp(struct file *fp) 2915 { 2916 struct unpcb *unp; 2917 2918 UNP_LINK_WLOCK(); 2919 if ((unp = fptounp(fp)) != NULL) { 2920 unp->unp_file = fp; 2921 unp->unp_msgcount++; 2922 } 2923 unp_rights++; 2924 UNP_LINK_WUNLOCK(); 2925 } 2926 2927 static int 2928 unp_externalize_fp(struct file *fp) 2929 { 2930 struct unpcb *unp; 2931 int ret; 2932 2933 UNP_LINK_WLOCK(); 2934 if ((unp = fptounp(fp)) != NULL) { 2935 unp->unp_msgcount--; 2936 ret = 1; 2937 } else 2938 ret = 0; 2939 unp_rights--; 2940 UNP_LINK_WUNLOCK(); 2941 return (ret); 2942 } 2943 2944 /* 2945 * unp_defer indicates whether additional work has been defered for a future 2946 * pass through unp_gc(). It is thread local and does not require explicit 2947 * synchronization. 2948 */ 2949 static int unp_marked; 2950 2951 static void 2952 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2953 { 2954 struct unpcb *unp; 2955 struct file *fp; 2956 int i; 2957 2958 /* 2959 * This function can only be called from the gc task. 2960 */ 2961 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2962 ("%s: not on gc callout", __func__)); 2963 UNP_LINK_LOCK_ASSERT(); 2964 2965 for (i = 0; i < fdcount; i++) { 2966 fp = fdep[i]->fde_file; 2967 if ((unp = fptounp(fp)) == NULL) 2968 continue; 2969 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2970 continue; 2971 unp->unp_gcrefs--; 2972 } 2973 } 2974 2975 static void 2976 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2977 { 2978 struct unpcb *unp; 2979 struct file *fp; 2980 int i; 2981 2982 /* 2983 * This function can only be called from the gc task. 2984 */ 2985 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2986 ("%s: not on gc callout", __func__)); 2987 UNP_LINK_LOCK_ASSERT(); 2988 2989 for (i = 0; i < fdcount; i++) { 2990 fp = fdep[i]->fde_file; 2991 if ((unp = fptounp(fp)) == NULL) 2992 continue; 2993 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2994 continue; 2995 unp->unp_gcrefs++; 2996 unp_marked++; 2997 } 2998 } 2999 3000 static void 3001 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3002 { 3003 struct sockbuf *sb; 3004 3005 SOCK_LOCK_ASSERT(so); 3006 3007 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3008 return; 3009 3010 SOCK_RECVBUF_LOCK(so); 3011 switch (so->so_type) { 3012 case SOCK_DGRAM: 3013 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3014 unp_scan(so->so_rcv.uxdg_peeked, op); 3015 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3016 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3017 break; 3018 case SOCK_STREAM: 3019 case SOCK_SEQPACKET: 3020 unp_scan(so->so_rcv.sb_mb, op); 3021 break; 3022 } 3023 SOCK_RECVBUF_UNLOCK(so); 3024 } 3025 3026 static void 3027 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3028 { 3029 struct socket *so, *soa; 3030 3031 so = unp->unp_socket; 3032 SOCK_LOCK(so); 3033 if (SOLISTENING(so)) { 3034 /* 3035 * Mark all sockets in our accept queue. 3036 */ 3037 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3038 unp_scan_socket(soa, op); 3039 } else { 3040 /* 3041 * Mark all sockets we reference with RIGHTS. 3042 */ 3043 unp_scan_socket(so, op); 3044 } 3045 SOCK_UNLOCK(so); 3046 } 3047 3048 static int unp_recycled; 3049 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3050 "Number of unreachable sockets claimed by the garbage collector."); 3051 3052 static int unp_taskcount; 3053 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3054 "Number of times the garbage collector has run."); 3055 3056 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3057 "Number of active local sockets."); 3058 3059 static void 3060 unp_gc(__unused void *arg, int pending) 3061 { 3062 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3063 NULL }; 3064 struct unp_head **head; 3065 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3066 struct file *f, **unref; 3067 struct unpcb *unp, *unptmp; 3068 int i, total, unp_unreachable; 3069 3070 LIST_INIT(&unp_deadhead); 3071 unp_taskcount++; 3072 UNP_LINK_RLOCK(); 3073 /* 3074 * First determine which sockets may be in cycles. 3075 */ 3076 unp_unreachable = 0; 3077 3078 for (head = heads; *head != NULL; head++) 3079 LIST_FOREACH(unp, *head, unp_link) { 3080 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3081 ("%s: unp %p has unexpected gc flags 0x%x", 3082 __func__, unp, (unsigned int)unp->unp_gcflag)); 3083 3084 f = unp->unp_file; 3085 3086 /* 3087 * Check for an unreachable socket potentially in a 3088 * cycle. It must be in a queue as indicated by 3089 * msgcount, and this must equal the file reference 3090 * count. Note that when msgcount is 0 the file is 3091 * NULL. 3092 */ 3093 if (f != NULL && unp->unp_msgcount != 0 && 3094 refcount_load(&f->f_count) == unp->unp_msgcount) { 3095 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3096 unp->unp_gcflag |= UNPGC_DEAD; 3097 unp->unp_gcrefs = unp->unp_msgcount; 3098 unp_unreachable++; 3099 } 3100 } 3101 3102 /* 3103 * Scan all sockets previously marked as potentially being in a cycle 3104 * and remove the references each socket holds on any UNPGC_DEAD 3105 * sockets in its queue. After this step, all remaining references on 3106 * sockets marked UNPGC_DEAD should not be part of any cycle. 3107 */ 3108 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3109 unp_gc_scan(unp, unp_remove_dead_ref); 3110 3111 /* 3112 * If a socket still has a non-negative refcount, it cannot be in a 3113 * cycle. In this case increment refcount of all children iteratively. 3114 * Stop the scan once we do a complete loop without discovering 3115 * a new reachable socket. 3116 */ 3117 do { 3118 unp_marked = 0; 3119 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3120 if (unp->unp_gcrefs > 0) { 3121 unp->unp_gcflag &= ~UNPGC_DEAD; 3122 LIST_REMOVE(unp, unp_dead); 3123 KASSERT(unp_unreachable > 0, 3124 ("%s: unp_unreachable underflow.", 3125 __func__)); 3126 unp_unreachable--; 3127 unp_gc_scan(unp, unp_restore_undead_ref); 3128 } 3129 } while (unp_marked); 3130 3131 UNP_LINK_RUNLOCK(); 3132 3133 if (unp_unreachable == 0) 3134 return; 3135 3136 /* 3137 * Allocate space for a local array of dead unpcbs. 3138 * TODO: can this path be simplified by instead using the local 3139 * dead list at unp_deadhead, after taking out references 3140 * on the file object and/or unpcb and dropping the link lock? 3141 */ 3142 unref = malloc(unp_unreachable * sizeof(struct file *), 3143 M_TEMP, M_WAITOK); 3144 3145 /* 3146 * Iterate looking for sockets which have been specifically marked 3147 * as unreachable and store them locally. 3148 */ 3149 UNP_LINK_RLOCK(); 3150 total = 0; 3151 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3152 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3153 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3154 unp->unp_gcflag &= ~UNPGC_DEAD; 3155 f = unp->unp_file; 3156 if (unp->unp_msgcount == 0 || f == NULL || 3157 refcount_load(&f->f_count) != unp->unp_msgcount || 3158 !fhold(f)) 3159 continue; 3160 unref[total++] = f; 3161 KASSERT(total <= unp_unreachable, 3162 ("%s: incorrect unreachable count.", __func__)); 3163 } 3164 UNP_LINK_RUNLOCK(); 3165 3166 /* 3167 * Now flush all sockets, free'ing rights. This will free the 3168 * struct files associated with these sockets but leave each socket 3169 * with one remaining ref. 3170 */ 3171 for (i = 0; i < total; i++) { 3172 struct socket *so; 3173 3174 so = unref[i]->f_data; 3175 CURVNET_SET(so->so_vnet); 3176 sorflush(so); 3177 CURVNET_RESTORE(); 3178 } 3179 3180 /* 3181 * And finally release the sockets so they can be reclaimed. 3182 */ 3183 for (i = 0; i < total; i++) 3184 fdrop(unref[i], NULL); 3185 unp_recycled += total; 3186 free(unref, M_TEMP); 3187 } 3188 3189 /* 3190 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3191 */ 3192 static void 3193 unp_dispose(struct socket *so) 3194 { 3195 struct sockbuf *sb; 3196 struct unpcb *unp; 3197 struct mbuf *m; 3198 3199 MPASS(!SOLISTENING(so)); 3200 3201 unp = sotounpcb(so); 3202 UNP_LINK_WLOCK(); 3203 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3204 UNP_LINK_WUNLOCK(); 3205 3206 /* 3207 * Grab our special mbufs before calling sbrelease(). 3208 */ 3209 SOCK_RECVBUF_LOCK(so); 3210 switch (so->so_type) { 3211 case SOCK_DGRAM: 3212 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3213 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3214 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3215 /* Note: socket of sb may reconnect. */ 3216 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3217 } 3218 sb = &so->so_rcv; 3219 if (sb->uxdg_peeked != NULL) { 3220 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3221 m_stailqpkt); 3222 sb->uxdg_peeked = NULL; 3223 } 3224 m = STAILQ_FIRST(&sb->uxdg_mb); 3225 STAILQ_INIT(&sb->uxdg_mb); 3226 /* XXX: our shortened sbrelease() */ 3227 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3228 RLIM_INFINITY); 3229 /* 3230 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3231 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3232 * data to the socket. 3233 * We are now in dom_dispose and it could be a call from 3234 * soshutdown() or from the final sofree(). The sofree() case 3235 * is simple as it guarantees that no more sends will happen, 3236 * however we can race with unp_disconnect() from our peer. 3237 * The shutdown(2) case is more exotic. It would call into 3238 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3239 * possible if we did connect(2) on this socket and we also 3240 * had it bound with bind(2) and receive connections from other 3241 * sockets. Because soshutdown() violates POSIX (see comment 3242 * there) we will end up here shutting down our receive side. 3243 * Of course this will have affect not only on the peer we 3244 * connect(2)ed to, but also on all of the peers who had 3245 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3246 */ 3247 sb->sb_state |= SBS_CANTRCVMORE; 3248 break; 3249 case SOCK_STREAM: 3250 case SOCK_SEQPACKET: 3251 sb = &so->so_rcv; 3252 m = sbcut_locked(sb, sb->sb_ccc); 3253 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3254 ("%s: ccc %u mb %p mbcnt %u", __func__, 3255 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3256 sbrelease_locked(so, SO_RCV); 3257 break; 3258 } 3259 SOCK_RECVBUF_UNLOCK(so); 3260 if (SOCK_IO_RECV_OWNED(so)) 3261 SOCK_IO_RECV_UNLOCK(so); 3262 3263 if (m != NULL) { 3264 unp_scan(m, unp_freerights); 3265 m_freem(m); 3266 } 3267 } 3268 3269 static void 3270 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3271 { 3272 struct mbuf *m; 3273 struct cmsghdr *cm; 3274 void *data; 3275 socklen_t clen, datalen; 3276 3277 while (m0 != NULL) { 3278 for (m = m0; m; m = m->m_next) { 3279 if (m->m_type != MT_CONTROL) 3280 continue; 3281 3282 cm = mtod(m, struct cmsghdr *); 3283 clen = m->m_len; 3284 3285 while (cm != NULL) { 3286 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3287 break; 3288 3289 data = CMSG_DATA(cm); 3290 datalen = (caddr_t)cm + cm->cmsg_len 3291 - (caddr_t)data; 3292 3293 if (cm->cmsg_level == SOL_SOCKET && 3294 cm->cmsg_type == SCM_RIGHTS) { 3295 (*op)(data, datalen / 3296 sizeof(struct filedescent *)); 3297 } 3298 3299 if (CMSG_SPACE(datalen) < clen) { 3300 clen -= CMSG_SPACE(datalen); 3301 cm = (struct cmsghdr *) 3302 ((caddr_t)cm + CMSG_SPACE(datalen)); 3303 } else { 3304 clen = 0; 3305 cm = NULL; 3306 } 3307 } 3308 } 3309 m0 = m0->m_nextpkt; 3310 } 3311 } 3312 3313 /* 3314 * Definitions of protocols supported in the LOCAL domain. 3315 */ 3316 static struct protosw streamproto = { 3317 .pr_type = SOCK_STREAM, 3318 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 3319 PR_CAPATTACH, 3320 .pr_ctloutput = &uipc_ctloutput, 3321 .pr_abort = uipc_abort, 3322 .pr_accept = uipc_accept, 3323 .pr_attach = uipc_attach, 3324 .pr_bind = uipc_bind, 3325 .pr_bindat = uipc_bindat, 3326 .pr_connect = uipc_connect, 3327 .pr_connectat = uipc_connectat, 3328 .pr_connect2 = uipc_connect2, 3329 .pr_detach = uipc_detach, 3330 .pr_disconnect = uipc_disconnect, 3331 .pr_listen = uipc_listen, 3332 .pr_peeraddr = uipc_peeraddr, 3333 .pr_rcvd = uipc_rcvd, 3334 .pr_send = uipc_send, 3335 .pr_ready = uipc_ready, 3336 .pr_sense = uipc_sense, 3337 .pr_shutdown = uipc_shutdown, 3338 .pr_sockaddr = uipc_sockaddr, 3339 .pr_soreceive = soreceive_generic, 3340 .pr_close = uipc_close, 3341 }; 3342 3343 static struct protosw dgramproto = { 3344 .pr_type = SOCK_DGRAM, 3345 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 3346 PR_SOCKBUF, 3347 .pr_ctloutput = &uipc_ctloutput, 3348 .pr_abort = uipc_abort, 3349 .pr_accept = uipc_accept, 3350 .pr_attach = uipc_attach, 3351 .pr_bind = uipc_bind, 3352 .pr_bindat = uipc_bindat, 3353 .pr_connect = uipc_connect, 3354 .pr_connectat = uipc_connectat, 3355 .pr_connect2 = uipc_connect2, 3356 .pr_detach = uipc_detach, 3357 .pr_disconnect = uipc_disconnect, 3358 .pr_peeraddr = uipc_peeraddr, 3359 .pr_sosend = uipc_sosend_dgram, 3360 .pr_sense = uipc_sense, 3361 .pr_shutdown = uipc_shutdown, 3362 .pr_sockaddr = uipc_sockaddr, 3363 .pr_soreceive = uipc_soreceive_dgram, 3364 .pr_close = uipc_close, 3365 }; 3366 3367 static struct protosw seqpacketproto = { 3368 .pr_type = SOCK_SEQPACKET, 3369 /* 3370 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3371 * due to our use of sbappendaddr. A new sbappend variants is needed 3372 * that supports both atomic record writes and control data. 3373 */ 3374 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 3375 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 3376 .pr_ctloutput = &uipc_ctloutput, 3377 .pr_abort = uipc_abort, 3378 .pr_accept = uipc_accept, 3379 .pr_attach = uipc_attach, 3380 .pr_bind = uipc_bind, 3381 .pr_bindat = uipc_bindat, 3382 .pr_connect = uipc_connect, 3383 .pr_connectat = uipc_connectat, 3384 .pr_connect2 = uipc_connect2, 3385 .pr_detach = uipc_detach, 3386 .pr_disconnect = uipc_disconnect, 3387 .pr_listen = uipc_listen, 3388 .pr_peeraddr = uipc_peeraddr, 3389 .pr_rcvd = uipc_rcvd, 3390 .pr_send = uipc_send, 3391 .pr_sense = uipc_sense, 3392 .pr_shutdown = uipc_shutdown, 3393 .pr_sockaddr = uipc_sockaddr, 3394 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3395 .pr_close = uipc_close, 3396 }; 3397 3398 static struct domain localdomain = { 3399 .dom_family = AF_LOCAL, 3400 .dom_name = "local", 3401 .dom_externalize = unp_externalize, 3402 .dom_dispose = unp_dispose, 3403 .dom_nprotosw = 3, 3404 .dom_protosw = { 3405 &streamproto, 3406 &dgramproto, 3407 &seqpacketproto, 3408 } 3409 }; 3410 DOMAIN_SET(local); 3411 3412 /* 3413 * A helper function called by VFS before socket-type vnode reclamation. 3414 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3415 * use count. 3416 */ 3417 void 3418 vfs_unp_reclaim(struct vnode *vp) 3419 { 3420 struct unpcb *unp; 3421 int active; 3422 struct mtx *vplock; 3423 3424 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3425 KASSERT(vp->v_type == VSOCK, 3426 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3427 3428 active = 0; 3429 vplock = mtx_pool_find(mtxpool_sleep, vp); 3430 mtx_lock(vplock); 3431 VOP_UNP_CONNECT(vp, &unp); 3432 if (unp == NULL) 3433 goto done; 3434 UNP_PCB_LOCK(unp); 3435 if (unp->unp_vnode == vp) { 3436 VOP_UNP_DETACH(vp); 3437 unp->unp_vnode = NULL; 3438 active = 1; 3439 } 3440 UNP_PCB_UNLOCK(unp); 3441 done: 3442 mtx_unlock(vplock); 3443 if (active) 3444 vunref(vp); 3445 } 3446 3447 #ifdef DDB 3448 static void 3449 db_print_indent(int indent) 3450 { 3451 int i; 3452 3453 for (i = 0; i < indent; i++) 3454 db_printf(" "); 3455 } 3456 3457 static void 3458 db_print_unpflags(int unp_flags) 3459 { 3460 int comma; 3461 3462 comma = 0; 3463 if (unp_flags & UNP_HAVEPC) { 3464 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3465 comma = 1; 3466 } 3467 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3468 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3469 comma = 1; 3470 } 3471 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3472 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3473 comma = 1; 3474 } 3475 if (unp_flags & UNP_CONNWAIT) { 3476 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3477 comma = 1; 3478 } 3479 if (unp_flags & UNP_CONNECTING) { 3480 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3481 comma = 1; 3482 } 3483 if (unp_flags & UNP_BINDING) { 3484 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3485 comma = 1; 3486 } 3487 } 3488 3489 static void 3490 db_print_xucred(int indent, struct xucred *xu) 3491 { 3492 int comma, i; 3493 3494 db_print_indent(indent); 3495 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3496 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3497 db_print_indent(indent); 3498 db_printf("cr_groups: "); 3499 comma = 0; 3500 for (i = 0; i < xu->cr_ngroups; i++) { 3501 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3502 comma = 1; 3503 } 3504 db_printf("\n"); 3505 } 3506 3507 static void 3508 db_print_unprefs(int indent, struct unp_head *uh) 3509 { 3510 struct unpcb *unp; 3511 int counter; 3512 3513 counter = 0; 3514 LIST_FOREACH(unp, uh, unp_reflink) { 3515 if (counter % 4 == 0) 3516 db_print_indent(indent); 3517 db_printf("%p ", unp); 3518 if (counter % 4 == 3) 3519 db_printf("\n"); 3520 counter++; 3521 } 3522 if (counter != 0 && counter % 4 != 0) 3523 db_printf("\n"); 3524 } 3525 3526 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3527 { 3528 struct unpcb *unp; 3529 3530 if (!have_addr) { 3531 db_printf("usage: show unpcb <addr>\n"); 3532 return; 3533 } 3534 unp = (struct unpcb *)addr; 3535 3536 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3537 unp->unp_vnode); 3538 3539 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3540 unp->unp_conn); 3541 3542 db_printf("unp_refs:\n"); 3543 db_print_unprefs(2, &unp->unp_refs); 3544 3545 /* XXXRW: Would be nice to print the full address, if any. */ 3546 db_printf("unp_addr: %p\n", unp->unp_addr); 3547 3548 db_printf("unp_gencnt: %llu\n", 3549 (unsigned long long)unp->unp_gencnt); 3550 3551 db_printf("unp_flags: %x (", unp->unp_flags); 3552 db_print_unpflags(unp->unp_flags); 3553 db_printf(")\n"); 3554 3555 db_printf("unp_peercred:\n"); 3556 db_print_xucred(2, &unp->unp_peercred); 3557 3558 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3559 } 3560 #endif 3561