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