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 | V_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 if (f->m_next->m_type == MT_CONTROL) 1342 unp_scan(f->m_next, unp_freerights); 1343 } 1344 1345 if (addr != NULL) 1346 unp_disconnect(unp, unp2); 1347 else 1348 unp_pcb_unlock_pair(unp, unp2); 1349 1350 td->td_ru.ru_msgsnd++; 1351 1352 out3: 1353 SOCK_IO_SEND_UNLOCK(so); 1354 out2: 1355 if (c) 1356 unp_scan(c, unp_freerights); 1357 out: 1358 if (f) 1359 m_freem(f); 1360 if (c) 1361 m_freem(c); 1362 if (m) 1363 m_freem(m); 1364 1365 return (error); 1366 } 1367 1368 /* 1369 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 1370 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 1371 * and needs to be linked onto uxdg_peeked of receive socket buffer. 1372 */ 1373 static int 1374 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 1375 struct uio *uio, struct mbuf **controlp, int *flagsp) 1376 { 1377 ssize_t len = 0; 1378 int error; 1379 1380 so->so_rcv.uxdg_peeked = m; 1381 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 1382 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 1383 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 1384 SOCK_RECVBUF_UNLOCK(so); 1385 1386 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1387 if (psa != NULL) 1388 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1389 1390 m = m->m_next; 1391 KASSERT(m, ("%s: no data or control after soname", __func__)); 1392 1393 /* 1394 * With MSG_PEEK the control isn't executed, just copied. 1395 */ 1396 while (m != NULL && m->m_type == MT_CONTROL) { 1397 if (controlp != NULL) { 1398 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 1399 controlp = &(*controlp)->m_next; 1400 } 1401 m = m->m_next; 1402 } 1403 KASSERT(m == NULL || m->m_type == MT_DATA, 1404 ("%s: not MT_DATA mbuf %p", __func__, m)); 1405 while (m != NULL && uio->uio_resid > 0) { 1406 len = uio->uio_resid; 1407 if (len > m->m_len) 1408 len = m->m_len; 1409 error = uiomove(mtod(m, char *), (int)len, uio); 1410 if (error) { 1411 SOCK_IO_RECV_UNLOCK(so); 1412 return (error); 1413 } 1414 if (len == m->m_len) 1415 m = m->m_next; 1416 } 1417 SOCK_IO_RECV_UNLOCK(so); 1418 1419 if (flagsp != NULL) { 1420 if (m != NULL) { 1421 if (*flagsp & MSG_TRUNC) { 1422 /* Report real length of the packet */ 1423 uio->uio_resid -= m_length(m, NULL) - len; 1424 } 1425 *flagsp |= MSG_TRUNC; 1426 } else 1427 *flagsp &= ~MSG_TRUNC; 1428 } 1429 1430 return (0); 1431 } 1432 1433 /* 1434 * PF_UNIX/SOCK_DGRAM receive 1435 */ 1436 static int 1437 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 1438 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1439 { 1440 struct sockbuf *sb = NULL; 1441 struct mbuf *m; 1442 int flags, error; 1443 ssize_t len = 0; 1444 bool nonblock; 1445 1446 MPASS(mp0 == NULL); 1447 1448 if (psa != NULL) 1449 *psa = NULL; 1450 if (controlp != NULL) 1451 *controlp = NULL; 1452 1453 flags = flagsp != NULL ? *flagsp : 0; 1454 nonblock = (so->so_state & SS_NBIO) || 1455 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1456 1457 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1458 if (__predict_false(error)) 1459 return (error); 1460 1461 /* 1462 * Loop blocking while waiting for a datagram. Prioritize connected 1463 * peers over unconnected sends. Set sb to selected socket buffer 1464 * containing an mbuf on exit from the wait loop. A datagram that 1465 * had already been peeked at has top priority. 1466 */ 1467 SOCK_RECVBUF_LOCK(so); 1468 while ((m = so->so_rcv.uxdg_peeked) == NULL && 1469 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 1470 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 1471 if (so->so_error) { 1472 error = so->so_error; 1473 so->so_error = 0; 1474 SOCK_RECVBUF_UNLOCK(so); 1475 SOCK_IO_RECV_UNLOCK(so); 1476 return (error); 1477 } 1478 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 1479 uio->uio_resid == 0) { 1480 SOCK_RECVBUF_UNLOCK(so); 1481 SOCK_IO_RECV_UNLOCK(so); 1482 return (0); 1483 } 1484 if (nonblock) { 1485 SOCK_RECVBUF_UNLOCK(so); 1486 SOCK_IO_RECV_UNLOCK(so); 1487 return (EWOULDBLOCK); 1488 } 1489 error = sbwait(so, SO_RCV); 1490 if (error) { 1491 SOCK_RECVBUF_UNLOCK(so); 1492 SOCK_IO_RECV_UNLOCK(so); 1493 return (error); 1494 } 1495 } 1496 1497 if (sb == NULL) 1498 sb = &so->so_rcv; 1499 else if (m == NULL) 1500 m = STAILQ_FIRST(&sb->uxdg_mb); 1501 else 1502 MPASS(m == so->so_rcv.uxdg_peeked); 1503 1504 MPASS(sb->uxdg_cc > 0); 1505 M_ASSERTPKTHDR(m); 1506 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 1507 1508 if (uio->uio_td) 1509 uio->uio_td->td_ru.ru_msgrcv++; 1510 1511 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 1512 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 1513 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 1514 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 1515 } else 1516 so->so_rcv.uxdg_peeked = NULL; 1517 1518 sb->uxdg_cc -= m->m_pkthdr.len; 1519 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 1520 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 1521 1522 if (__predict_false(flags & MSG_PEEK)) 1523 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 1524 1525 so->so_rcv.sb_acc -= m->m_pkthdr.len; 1526 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 1527 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 1528 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 1529 SOCK_RECVBUF_UNLOCK(so); 1530 1531 if (psa != NULL) 1532 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 1533 m = m_free(m); 1534 KASSERT(m, ("%s: no data or control after soname", __func__)); 1535 1536 /* 1537 * Packet to copyout() is now in 'm' and it is disconnected from the 1538 * queue. 1539 * 1540 * Process one or more MT_CONTROL mbufs present before any data mbufs 1541 * in the first mbuf chain on the socket buffer. We call into the 1542 * unp_externalize() to perform externalization (or freeing if 1543 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 1544 * without MT_DATA mbufs. 1545 */ 1546 while (m != NULL && m->m_type == MT_CONTROL) { 1547 struct mbuf *cm; 1548 1549 /* XXXGL: unp_externalize() is also dom_externalize() KBI and 1550 * it frees whole chain, so we must disconnect the mbuf. 1551 */ 1552 cm = m; m = m->m_next; cm->m_next = NULL; 1553 error = unp_externalize(cm, controlp, flags); 1554 if (error != 0) { 1555 SOCK_IO_RECV_UNLOCK(so); 1556 unp_scan(m, unp_freerights); 1557 m_freem(m); 1558 return (error); 1559 } 1560 if (controlp != NULL) { 1561 while (*controlp != NULL) 1562 controlp = &(*controlp)->m_next; 1563 } 1564 } 1565 KASSERT(m == NULL || m->m_type == MT_DATA, 1566 ("%s: not MT_DATA mbuf %p", __func__, m)); 1567 while (m != NULL && uio->uio_resid > 0) { 1568 len = uio->uio_resid; 1569 if (len > m->m_len) 1570 len = m->m_len; 1571 error = uiomove(mtod(m, char *), (int)len, uio); 1572 if (error) { 1573 SOCK_IO_RECV_UNLOCK(so); 1574 m_freem(m); 1575 return (error); 1576 } 1577 if (len == m->m_len) 1578 m = m_free(m); 1579 else { 1580 m->m_data += len; 1581 m->m_len -= len; 1582 } 1583 } 1584 SOCK_IO_RECV_UNLOCK(so); 1585 1586 if (m != NULL) { 1587 if (flagsp != NULL) { 1588 if (flags & MSG_TRUNC) { 1589 /* Report real length of the packet */ 1590 uio->uio_resid -= m_length(m, NULL); 1591 } 1592 *flagsp |= MSG_TRUNC; 1593 } 1594 m_freem(m); 1595 } else if (flagsp != NULL) 1596 *flagsp &= ~MSG_TRUNC; 1597 1598 return (0); 1599 } 1600 1601 static bool 1602 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 1603 { 1604 struct mbuf *mb, *n; 1605 struct sockbuf *sb; 1606 1607 SOCK_LOCK(so); 1608 if (SOLISTENING(so)) { 1609 SOCK_UNLOCK(so); 1610 return (false); 1611 } 1612 mb = NULL; 1613 sb = &so->so_rcv; 1614 SOCKBUF_LOCK(sb); 1615 if (sb->sb_fnrdy != NULL) { 1616 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { 1617 if (mb == m) { 1618 *errorp = sbready(sb, m, count); 1619 break; 1620 } 1621 mb = mb->m_next; 1622 if (mb == NULL) { 1623 mb = n; 1624 if (mb != NULL) 1625 n = mb->m_nextpkt; 1626 } 1627 } 1628 } 1629 SOCKBUF_UNLOCK(sb); 1630 SOCK_UNLOCK(so); 1631 return (mb != NULL); 1632 } 1633 1634 static int 1635 uipc_ready(struct socket *so, struct mbuf *m, int count) 1636 { 1637 struct unpcb *unp, *unp2; 1638 struct socket *so2; 1639 int error, i; 1640 1641 unp = sotounpcb(so); 1642 1643 KASSERT(so->so_type == SOCK_STREAM, 1644 ("%s: unexpected socket type for %p", __func__, so)); 1645 1646 UNP_PCB_LOCK(unp); 1647 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 1648 UNP_PCB_UNLOCK(unp); 1649 so2 = unp2->unp_socket; 1650 SOCKBUF_LOCK(&so2->so_rcv); 1651 if ((error = sbready(&so2->so_rcv, m, count)) == 0) 1652 sorwakeup_locked(so2); 1653 else 1654 SOCKBUF_UNLOCK(&so2->so_rcv); 1655 UNP_PCB_UNLOCK(unp2); 1656 return (error); 1657 } 1658 UNP_PCB_UNLOCK(unp); 1659 1660 /* 1661 * The receiving socket has been disconnected, but may still be valid. 1662 * In this case, the now-ready mbufs are still present in its socket 1663 * buffer, so perform an exhaustive search before giving up and freeing 1664 * the mbufs. 1665 */ 1666 UNP_LINK_RLOCK(); 1667 LIST_FOREACH(unp, &unp_shead, unp_link) { 1668 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 1669 break; 1670 } 1671 UNP_LINK_RUNLOCK(); 1672 1673 if (unp == NULL) { 1674 for (i = 0; i < count; i++) 1675 m = m_free(m); 1676 error = ECONNRESET; 1677 } 1678 return (error); 1679 } 1680 1681 static int 1682 uipc_sense(struct socket *so, struct stat *sb) 1683 { 1684 struct unpcb *unp; 1685 1686 unp = sotounpcb(so); 1687 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 1688 1689 sb->st_blksize = so->so_snd.sb_hiwat; 1690 sb->st_dev = NODEV; 1691 sb->st_ino = unp->unp_ino; 1692 return (0); 1693 } 1694 1695 static int 1696 uipc_shutdown(struct socket *so) 1697 { 1698 struct unpcb *unp; 1699 1700 unp = sotounpcb(so); 1701 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 1702 1703 UNP_PCB_LOCK(unp); 1704 socantsendmore(so); 1705 unp_shutdown(unp); 1706 UNP_PCB_UNLOCK(unp); 1707 return (0); 1708 } 1709 1710 static int 1711 uipc_sockaddr(struct socket *so, struct sockaddr **nam) 1712 { 1713 struct unpcb *unp; 1714 const struct sockaddr *sa; 1715 1716 unp = sotounpcb(so); 1717 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 1718 1719 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1720 UNP_PCB_LOCK(unp); 1721 if (unp->unp_addr != NULL) 1722 sa = (struct sockaddr *) unp->unp_addr; 1723 else 1724 sa = &sun_noname; 1725 bcopy(sa, *nam, sa->sa_len); 1726 UNP_PCB_UNLOCK(unp); 1727 return (0); 1728 } 1729 1730 static int 1731 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1732 { 1733 struct unpcb *unp; 1734 struct xucred xu; 1735 int error, optval; 1736 1737 if (sopt->sopt_level != SOL_LOCAL) 1738 return (EINVAL); 1739 1740 unp = sotounpcb(so); 1741 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1742 error = 0; 1743 switch (sopt->sopt_dir) { 1744 case SOPT_GET: 1745 switch (sopt->sopt_name) { 1746 case LOCAL_PEERCRED: 1747 UNP_PCB_LOCK(unp); 1748 if (unp->unp_flags & UNP_HAVEPC) 1749 xu = unp->unp_peercred; 1750 else { 1751 if (so->so_type == SOCK_STREAM) 1752 error = ENOTCONN; 1753 else 1754 error = EINVAL; 1755 } 1756 UNP_PCB_UNLOCK(unp); 1757 if (error == 0) 1758 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1759 break; 1760 1761 case LOCAL_CREDS: 1762 /* Unlocked read. */ 1763 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 1764 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1765 break; 1766 1767 case LOCAL_CREDS_PERSISTENT: 1768 /* Unlocked read. */ 1769 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 1770 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1771 break; 1772 1773 case LOCAL_CONNWAIT: 1774 /* Unlocked read. */ 1775 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1776 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1777 break; 1778 1779 default: 1780 error = EOPNOTSUPP; 1781 break; 1782 } 1783 break; 1784 1785 case SOPT_SET: 1786 switch (sopt->sopt_name) { 1787 case LOCAL_CREDS: 1788 case LOCAL_CREDS_PERSISTENT: 1789 case LOCAL_CONNWAIT: 1790 error = sooptcopyin(sopt, &optval, sizeof(optval), 1791 sizeof(optval)); 1792 if (error) 1793 break; 1794 1795 #define OPTSET(bit, exclusive) do { \ 1796 UNP_PCB_LOCK(unp); \ 1797 if (optval) { \ 1798 if ((unp->unp_flags & (exclusive)) != 0) { \ 1799 UNP_PCB_UNLOCK(unp); \ 1800 error = EINVAL; \ 1801 break; \ 1802 } \ 1803 unp->unp_flags |= (bit); \ 1804 } else \ 1805 unp->unp_flags &= ~(bit); \ 1806 UNP_PCB_UNLOCK(unp); \ 1807 } while (0) 1808 1809 switch (sopt->sopt_name) { 1810 case LOCAL_CREDS: 1811 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 1812 break; 1813 1814 case LOCAL_CREDS_PERSISTENT: 1815 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 1816 break; 1817 1818 case LOCAL_CONNWAIT: 1819 OPTSET(UNP_CONNWAIT, 0); 1820 break; 1821 1822 default: 1823 break; 1824 } 1825 break; 1826 #undef OPTSET 1827 default: 1828 error = ENOPROTOOPT; 1829 break; 1830 } 1831 break; 1832 1833 default: 1834 error = EOPNOTSUPP; 1835 break; 1836 } 1837 return (error); 1838 } 1839 1840 static int 1841 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1842 { 1843 1844 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 1845 } 1846 1847 static int 1848 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 1849 struct thread *td, bool return_locked) 1850 { 1851 struct mtx *vplock; 1852 struct sockaddr_un *soun; 1853 struct vnode *vp; 1854 struct socket *so2; 1855 struct unpcb *unp, *unp2, *unp3; 1856 struct nameidata nd; 1857 char buf[SOCK_MAXADDRLEN]; 1858 struct sockaddr *sa; 1859 cap_rights_t rights; 1860 int error, len; 1861 bool connreq; 1862 1863 if (nam->sa_family != AF_UNIX) 1864 return (EAFNOSUPPORT); 1865 if (nam->sa_len > sizeof(struct sockaddr_un)) 1866 return (EINVAL); 1867 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1868 if (len <= 0) 1869 return (EINVAL); 1870 soun = (struct sockaddr_un *)nam; 1871 bcopy(soun->sun_path, buf, len); 1872 buf[len] = 0; 1873 1874 error = 0; 1875 unp = sotounpcb(so); 1876 UNP_PCB_LOCK(unp); 1877 for (;;) { 1878 /* 1879 * Wait for connection state to stabilize. If a connection 1880 * already exists, give up. For datagram sockets, which permit 1881 * multiple consecutive connect(2) calls, upper layers are 1882 * responsible for disconnecting in advance of a subsequent 1883 * connect(2), but this is not synchronized with PCB connection 1884 * state. 1885 * 1886 * Also make sure that no threads are currently attempting to 1887 * lock the peer socket, to ensure that unp_conn cannot 1888 * transition between two valid sockets while locks are dropped. 1889 */ 1890 if (SOLISTENING(so)) 1891 error = EOPNOTSUPP; 1892 else if (unp->unp_conn != NULL) 1893 error = EISCONN; 1894 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 1895 error = EALREADY; 1896 } 1897 if (error != 0) { 1898 UNP_PCB_UNLOCK(unp); 1899 return (error); 1900 } 1901 if (unp->unp_pairbusy > 0) { 1902 unp->unp_flags |= UNP_WAITING; 1903 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 1904 continue; 1905 } 1906 break; 1907 } 1908 unp->unp_flags |= UNP_CONNECTING; 1909 UNP_PCB_UNLOCK(unp); 1910 1911 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 1912 if (connreq) 1913 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1914 else 1915 sa = NULL; 1916 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 1917 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 1918 error = namei(&nd); 1919 if (error) 1920 vp = NULL; 1921 else 1922 vp = nd.ni_vp; 1923 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1924 if (error) 1925 goto bad; 1926 NDFREE_PNBUF(&nd); 1927 1928 if (vp->v_type != VSOCK) { 1929 error = ENOTSOCK; 1930 goto bad; 1931 } 1932 #ifdef MAC 1933 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1934 if (error) 1935 goto bad; 1936 #endif 1937 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1938 if (error) 1939 goto bad; 1940 1941 unp = sotounpcb(so); 1942 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1943 1944 vplock = mtx_pool_find(mtxpool_sleep, vp); 1945 mtx_lock(vplock); 1946 VOP_UNP_CONNECT(vp, &unp2); 1947 if (unp2 == NULL) { 1948 error = ECONNREFUSED; 1949 goto bad2; 1950 } 1951 so2 = unp2->unp_socket; 1952 if (so->so_type != so2->so_type) { 1953 error = EPROTOTYPE; 1954 goto bad2; 1955 } 1956 if (connreq) { 1957 if (SOLISTENING(so2)) { 1958 CURVNET_SET(so2->so_vnet); 1959 so2 = sonewconn(so2, 0); 1960 CURVNET_RESTORE(); 1961 } else 1962 so2 = NULL; 1963 if (so2 == NULL) { 1964 error = ECONNREFUSED; 1965 goto bad2; 1966 } 1967 unp3 = sotounpcb(so2); 1968 unp_pcb_lock_pair(unp2, unp3); 1969 if (unp2->unp_addr != NULL) { 1970 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1971 unp3->unp_addr = (struct sockaddr_un *) sa; 1972 sa = NULL; 1973 } 1974 1975 unp_copy_peercred(td, unp3, unp, unp2); 1976 1977 UNP_PCB_UNLOCK(unp2); 1978 unp2 = unp3; 1979 1980 /* 1981 * It is safe to block on the PCB lock here since unp2 is 1982 * nascent and cannot be connected to any other sockets. 1983 */ 1984 UNP_PCB_LOCK(unp); 1985 #ifdef MAC 1986 mac_socketpeer_set_from_socket(so, so2); 1987 mac_socketpeer_set_from_socket(so2, so); 1988 #endif 1989 } else { 1990 unp_pcb_lock_pair(unp, unp2); 1991 } 1992 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 1993 sotounpcb(so2) == unp2, 1994 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 1995 unp_connect2(so, so2, PRU_CONNECT); 1996 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 1997 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 1998 unp->unp_flags &= ~UNP_CONNECTING; 1999 if (!return_locked) 2000 unp_pcb_unlock_pair(unp, unp2); 2001 bad2: 2002 mtx_unlock(vplock); 2003 bad: 2004 if (vp != NULL) { 2005 /* 2006 * If we are returning locked (called via uipc_sosend_dgram()), 2007 * we need to be sure that vput() won't sleep. This is 2008 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 2009 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 2010 */ 2011 MPASS(!(return_locked && connreq)); 2012 vput(vp); 2013 } 2014 free(sa, M_SONAME); 2015 if (__predict_false(error)) { 2016 UNP_PCB_LOCK(unp); 2017 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2018 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2019 unp->unp_flags &= ~UNP_CONNECTING; 2020 UNP_PCB_UNLOCK(unp); 2021 } 2022 return (error); 2023 } 2024 2025 /* 2026 * Set socket peer credentials at connection time. 2027 * 2028 * The client's PCB credentials are copied from its process structure. The 2029 * server's PCB credentials are copied from the socket on which it called 2030 * listen(2). uipc_listen cached that process's credentials at the time. 2031 */ 2032 void 2033 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 2034 struct unpcb *server_unp, struct unpcb *listen_unp) 2035 { 2036 cru2xt(td, &client_unp->unp_peercred); 2037 client_unp->unp_flags |= UNP_HAVEPC; 2038 2039 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 2040 sizeof(server_unp->unp_peercred)); 2041 server_unp->unp_flags |= UNP_HAVEPC; 2042 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 2043 } 2044 2045 static void 2046 unp_connect2(struct socket *so, struct socket *so2, conn2_how req) 2047 { 2048 struct unpcb *unp; 2049 struct unpcb *unp2; 2050 2051 MPASS(so2->so_type == so->so_type); 2052 unp = sotounpcb(so); 2053 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 2054 unp2 = sotounpcb(so2); 2055 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 2056 2057 UNP_PCB_LOCK_ASSERT(unp); 2058 UNP_PCB_LOCK_ASSERT(unp2); 2059 KASSERT(unp->unp_conn == NULL, 2060 ("%s: socket %p is already connected", __func__, unp)); 2061 2062 unp->unp_conn = unp2; 2063 unp_pcb_hold(unp2); 2064 unp_pcb_hold(unp); 2065 switch (so->so_type) { 2066 case SOCK_DGRAM: 2067 UNP_REF_LIST_LOCK(); 2068 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 2069 UNP_REF_LIST_UNLOCK(); 2070 soisconnected(so); 2071 break; 2072 2073 case SOCK_STREAM: 2074 case SOCK_SEQPACKET: 2075 KASSERT(unp2->unp_conn == NULL, 2076 ("%s: socket %p is already connected", __func__, unp2)); 2077 unp2->unp_conn = unp; 2078 if (req == PRU_CONNECT && 2079 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 2080 soisconnecting(so); 2081 else 2082 soisconnected(so); 2083 soisconnected(so2); 2084 break; 2085 2086 default: 2087 panic("unp_connect2"); 2088 } 2089 } 2090 2091 static void 2092 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 2093 { 2094 struct socket *so, *so2; 2095 struct mbuf *m = NULL; 2096 #ifdef INVARIANTS 2097 struct unpcb *unptmp; 2098 #endif 2099 2100 UNP_PCB_LOCK_ASSERT(unp); 2101 UNP_PCB_LOCK_ASSERT(unp2); 2102 KASSERT(unp->unp_conn == unp2, 2103 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 2104 2105 unp->unp_conn = NULL; 2106 so = unp->unp_socket; 2107 so2 = unp2->unp_socket; 2108 switch (unp->unp_socket->so_type) { 2109 case SOCK_DGRAM: 2110 /* 2111 * Remove our send socket buffer from the peer's receive buffer. 2112 * Move the data to the receive buffer only if it is empty. 2113 * This is a protection against a scenario where a peer 2114 * connects, floods and disconnects, effectively blocking 2115 * sendto() from unconnected sockets. 2116 */ 2117 SOCK_RECVBUF_LOCK(so2); 2118 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 2119 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 2120 uxdg_clist); 2121 if (__predict_true((so2->so_rcv.sb_state & 2122 SBS_CANTRCVMORE) == 0) && 2123 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 2124 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 2125 &so->so_snd.uxdg_mb); 2126 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 2127 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 2128 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 2129 } else { 2130 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 2131 STAILQ_INIT(&so->so_snd.uxdg_mb); 2132 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 2133 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 2134 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 2135 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 2136 } 2137 /* Note: so may reconnect. */ 2138 so->so_snd.uxdg_cc = 0; 2139 so->so_snd.uxdg_ctl = 0; 2140 so->so_snd.uxdg_mbcnt = 0; 2141 } 2142 SOCK_RECVBUF_UNLOCK(so2); 2143 UNP_REF_LIST_LOCK(); 2144 #ifdef INVARIANTS 2145 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 2146 if (unptmp == unp) 2147 break; 2148 } 2149 KASSERT(unptmp != NULL, 2150 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 2151 #endif 2152 LIST_REMOVE(unp, unp_reflink); 2153 UNP_REF_LIST_UNLOCK(); 2154 if (so) { 2155 SOCK_LOCK(so); 2156 so->so_state &= ~SS_ISCONNECTED; 2157 SOCK_UNLOCK(so); 2158 } 2159 break; 2160 2161 case SOCK_STREAM: 2162 case SOCK_SEQPACKET: 2163 if (so) 2164 soisdisconnected(so); 2165 MPASS(unp2->unp_conn == unp); 2166 unp2->unp_conn = NULL; 2167 if (so2) 2168 soisdisconnected(so2); 2169 break; 2170 } 2171 2172 if (unp == unp2) { 2173 unp_pcb_rele_notlast(unp); 2174 if (!unp_pcb_rele(unp)) 2175 UNP_PCB_UNLOCK(unp); 2176 } else { 2177 if (!unp_pcb_rele(unp)) 2178 UNP_PCB_UNLOCK(unp); 2179 if (!unp_pcb_rele(unp2)) 2180 UNP_PCB_UNLOCK(unp2); 2181 } 2182 2183 if (m != NULL) { 2184 unp_scan(m, unp_freerights); 2185 m_freem(m); 2186 } 2187 } 2188 2189 /* 2190 * unp_pcblist() walks the global list of struct unpcb's to generate a 2191 * pointer list, bumping the refcount on each unpcb. It then copies them out 2192 * sequentially, validating the generation number on each to see if it has 2193 * been detached. All of this is necessary because copyout() may sleep on 2194 * disk I/O. 2195 */ 2196 static int 2197 unp_pcblist(SYSCTL_HANDLER_ARGS) 2198 { 2199 struct unpcb *unp, **unp_list; 2200 unp_gen_t gencnt; 2201 struct xunpgen *xug; 2202 struct unp_head *head; 2203 struct xunpcb *xu; 2204 u_int i; 2205 int error, n; 2206 2207 switch ((intptr_t)arg1) { 2208 case SOCK_STREAM: 2209 head = &unp_shead; 2210 break; 2211 2212 case SOCK_DGRAM: 2213 head = &unp_dhead; 2214 break; 2215 2216 case SOCK_SEQPACKET: 2217 head = &unp_sphead; 2218 break; 2219 2220 default: 2221 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 2222 } 2223 2224 /* 2225 * The process of preparing the PCB list is too time-consuming and 2226 * resource-intensive to repeat twice on every request. 2227 */ 2228 if (req->oldptr == NULL) { 2229 n = unp_count; 2230 req->oldidx = 2 * (sizeof *xug) 2231 + (n + n/8) * sizeof(struct xunpcb); 2232 return (0); 2233 } 2234 2235 if (req->newptr != NULL) 2236 return (EPERM); 2237 2238 /* 2239 * OK, now we're committed to doing something. 2240 */ 2241 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 2242 UNP_LINK_RLOCK(); 2243 gencnt = unp_gencnt; 2244 n = unp_count; 2245 UNP_LINK_RUNLOCK(); 2246 2247 xug->xug_len = sizeof *xug; 2248 xug->xug_count = n; 2249 xug->xug_gen = gencnt; 2250 xug->xug_sogen = so_gencnt; 2251 error = SYSCTL_OUT(req, xug, sizeof *xug); 2252 if (error) { 2253 free(xug, M_TEMP); 2254 return (error); 2255 } 2256 2257 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 2258 2259 UNP_LINK_RLOCK(); 2260 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 2261 unp = LIST_NEXT(unp, unp_link)) { 2262 UNP_PCB_LOCK(unp); 2263 if (unp->unp_gencnt <= gencnt) { 2264 if (cr_cansee(req->td->td_ucred, 2265 unp->unp_socket->so_cred)) { 2266 UNP_PCB_UNLOCK(unp); 2267 continue; 2268 } 2269 unp_list[i++] = unp; 2270 unp_pcb_hold(unp); 2271 } 2272 UNP_PCB_UNLOCK(unp); 2273 } 2274 UNP_LINK_RUNLOCK(); 2275 n = i; /* In case we lost some during malloc. */ 2276 2277 error = 0; 2278 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 2279 for (i = 0; i < n; i++) { 2280 unp = unp_list[i]; 2281 UNP_PCB_LOCK(unp); 2282 if (unp_pcb_rele(unp)) 2283 continue; 2284 2285 if (unp->unp_gencnt <= gencnt) { 2286 xu->xu_len = sizeof *xu; 2287 xu->xu_unpp = (uintptr_t)unp; 2288 /* 2289 * XXX - need more locking here to protect against 2290 * connect/disconnect races for SMP. 2291 */ 2292 if (unp->unp_addr != NULL) 2293 bcopy(unp->unp_addr, &xu->xu_addr, 2294 unp->unp_addr->sun_len); 2295 else 2296 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 2297 if (unp->unp_conn != NULL && 2298 unp->unp_conn->unp_addr != NULL) 2299 bcopy(unp->unp_conn->unp_addr, 2300 &xu->xu_caddr, 2301 unp->unp_conn->unp_addr->sun_len); 2302 else 2303 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 2304 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 2305 xu->unp_conn = (uintptr_t)unp->unp_conn; 2306 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 2307 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 2308 xu->unp_gencnt = unp->unp_gencnt; 2309 sotoxsocket(unp->unp_socket, &xu->xu_socket); 2310 UNP_PCB_UNLOCK(unp); 2311 error = SYSCTL_OUT(req, xu, sizeof *xu); 2312 } else { 2313 UNP_PCB_UNLOCK(unp); 2314 } 2315 } 2316 free(xu, M_TEMP); 2317 if (!error) { 2318 /* 2319 * Give the user an updated idea of our state. If the 2320 * generation differs from what we told her before, she knows 2321 * that something happened while we were processing this 2322 * request, and it might be necessary to retry. 2323 */ 2324 xug->xug_gen = unp_gencnt; 2325 xug->xug_sogen = so_gencnt; 2326 xug->xug_count = unp_count; 2327 error = SYSCTL_OUT(req, xug, sizeof *xug); 2328 } 2329 free(unp_list, M_TEMP); 2330 free(xug, M_TEMP); 2331 return (error); 2332 } 2333 2334 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 2335 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2336 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 2337 "List of active local datagram sockets"); 2338 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 2339 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2340 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 2341 "List of active local stream sockets"); 2342 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 2343 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2344 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 2345 "List of active local seqpacket sockets"); 2346 2347 static void 2348 unp_shutdown(struct unpcb *unp) 2349 { 2350 struct unpcb *unp2; 2351 struct socket *so; 2352 2353 UNP_PCB_LOCK_ASSERT(unp); 2354 2355 unp2 = unp->unp_conn; 2356 if ((unp->unp_socket->so_type == SOCK_STREAM || 2357 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { 2358 so = unp2->unp_socket; 2359 if (so != NULL) 2360 socantrcvmore(so); 2361 } 2362 } 2363 2364 static void 2365 unp_drop(struct unpcb *unp) 2366 { 2367 struct socket *so; 2368 struct unpcb *unp2; 2369 2370 /* 2371 * Regardless of whether the socket's peer dropped the connection 2372 * with this socket by aborting or disconnecting, POSIX requires 2373 * that ECONNRESET is returned. 2374 */ 2375 2376 UNP_PCB_LOCK(unp); 2377 so = unp->unp_socket; 2378 if (so) 2379 so->so_error = ECONNRESET; 2380 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 2381 /* Last reference dropped in unp_disconnect(). */ 2382 unp_pcb_rele_notlast(unp); 2383 unp_disconnect(unp, unp2); 2384 } else if (!unp_pcb_rele(unp)) { 2385 UNP_PCB_UNLOCK(unp); 2386 } 2387 } 2388 2389 static void 2390 unp_freerights(struct filedescent **fdep, int fdcount) 2391 { 2392 struct file *fp; 2393 int i; 2394 2395 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 2396 2397 for (i = 0; i < fdcount; i++) { 2398 fp = fdep[i]->fde_file; 2399 filecaps_free(&fdep[i]->fde_caps); 2400 unp_discard(fp); 2401 } 2402 free(fdep[0], M_FILECAPS); 2403 } 2404 2405 static int 2406 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 2407 { 2408 struct thread *td = curthread; /* XXX */ 2409 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2410 int i; 2411 int *fdp; 2412 struct filedesc *fdesc = td->td_proc->p_fd; 2413 struct filedescent **fdep; 2414 void *data; 2415 socklen_t clen = control->m_len, datalen; 2416 int error, newfds; 2417 u_int newlen; 2418 2419 UNP_LINK_UNLOCK_ASSERT(); 2420 2421 error = 0; 2422 if (controlp != NULL) /* controlp == NULL => free control messages */ 2423 *controlp = NULL; 2424 while (cm != NULL) { 2425 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 2426 2427 data = CMSG_DATA(cm); 2428 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 2429 if (cm->cmsg_level == SOL_SOCKET 2430 && cm->cmsg_type == SCM_RIGHTS) { 2431 newfds = datalen / sizeof(*fdep); 2432 if (newfds == 0) 2433 goto next; 2434 fdep = data; 2435 2436 /* If we're not outputting the descriptors free them. */ 2437 if (error || controlp == NULL) { 2438 unp_freerights(fdep, newfds); 2439 goto next; 2440 } 2441 FILEDESC_XLOCK(fdesc); 2442 2443 /* 2444 * Now change each pointer to an fd in the global 2445 * table to an integer that is the index to the local 2446 * fd table entry that we set up to point to the 2447 * global one we are transferring. 2448 */ 2449 newlen = newfds * sizeof(int); 2450 *controlp = sbcreatecontrol(NULL, newlen, 2451 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2452 2453 fdp = (int *) 2454 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2455 if ((error = fdallocn(td, 0, fdp, newfds))) { 2456 FILEDESC_XUNLOCK(fdesc); 2457 unp_freerights(fdep, newfds); 2458 m_freem(*controlp); 2459 *controlp = NULL; 2460 goto next; 2461 } 2462 for (i = 0; i < newfds; i++, fdp++) { 2463 _finstall(fdesc, fdep[i]->fde_file, *fdp, 2464 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, 2465 &fdep[i]->fde_caps); 2466 unp_externalize_fp(fdep[i]->fde_file); 2467 } 2468 2469 /* 2470 * The new type indicates that the mbuf data refers to 2471 * kernel resources that may need to be released before 2472 * the mbuf is freed. 2473 */ 2474 m_chtype(*controlp, MT_EXTCONTROL); 2475 FILEDESC_XUNLOCK(fdesc); 2476 free(fdep[0], M_FILECAPS); 2477 } else { 2478 /* We can just copy anything else across. */ 2479 if (error || controlp == NULL) 2480 goto next; 2481 *controlp = sbcreatecontrol(NULL, datalen, 2482 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 2483 bcopy(data, 2484 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 2485 datalen); 2486 } 2487 controlp = &(*controlp)->m_next; 2488 2489 next: 2490 if (CMSG_SPACE(datalen) < clen) { 2491 clen -= CMSG_SPACE(datalen); 2492 cm = (struct cmsghdr *) 2493 ((caddr_t)cm + CMSG_SPACE(datalen)); 2494 } else { 2495 clen = 0; 2496 cm = NULL; 2497 } 2498 } 2499 2500 m_freem(control); 2501 return (error); 2502 } 2503 2504 static void 2505 unp_zone_change(void *tag) 2506 { 2507 2508 uma_zone_set_max(unp_zone, maxsockets); 2509 } 2510 2511 #ifdef INVARIANTS 2512 static void 2513 unp_zdtor(void *mem, int size __unused, void *arg __unused) 2514 { 2515 struct unpcb *unp; 2516 2517 unp = mem; 2518 2519 KASSERT(LIST_EMPTY(&unp->unp_refs), 2520 ("%s: unpcb %p has lingering refs", __func__, unp)); 2521 KASSERT(unp->unp_socket == NULL, 2522 ("%s: unpcb %p has socket backpointer", __func__, unp)); 2523 KASSERT(unp->unp_vnode == NULL, 2524 ("%s: unpcb %p has vnode references", __func__, unp)); 2525 KASSERT(unp->unp_conn == NULL, 2526 ("%s: unpcb %p is still connected", __func__, unp)); 2527 KASSERT(unp->unp_addr == NULL, 2528 ("%s: unpcb %p has leaked addr", __func__, unp)); 2529 } 2530 #endif 2531 2532 static void 2533 unp_init(void *arg __unused) 2534 { 2535 uma_dtor dtor; 2536 2537 #ifdef INVARIANTS 2538 dtor = unp_zdtor; 2539 #else 2540 dtor = NULL; 2541 #endif 2542 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 2543 NULL, NULL, UMA_ALIGN_CACHE, 0); 2544 uma_zone_set_max(unp_zone, maxsockets); 2545 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 2546 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 2547 NULL, EVENTHANDLER_PRI_ANY); 2548 LIST_INIT(&unp_dhead); 2549 LIST_INIT(&unp_shead); 2550 LIST_INIT(&unp_sphead); 2551 SLIST_INIT(&unp_defers); 2552 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 2553 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 2554 UNP_LINK_LOCK_INIT(); 2555 UNP_DEFERRED_LOCK_INIT(); 2556 } 2557 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 2558 2559 static void 2560 unp_internalize_cleanup_rights(struct mbuf *control) 2561 { 2562 struct cmsghdr *cp; 2563 struct mbuf *m; 2564 void *data; 2565 socklen_t datalen; 2566 2567 for (m = control; m != NULL; m = m->m_next) { 2568 cp = mtod(m, struct cmsghdr *); 2569 if (cp->cmsg_level != SOL_SOCKET || 2570 cp->cmsg_type != SCM_RIGHTS) 2571 continue; 2572 data = CMSG_DATA(cp); 2573 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 2574 unp_freerights(data, datalen / sizeof(struct filedesc *)); 2575 } 2576 } 2577 2578 static int 2579 unp_internalize(struct mbuf **controlp, struct thread *td, 2580 struct mbuf **clast, u_int *space, u_int *mbcnt) 2581 { 2582 struct mbuf *control, **initial_controlp; 2583 struct proc *p; 2584 struct filedesc *fdesc; 2585 struct bintime *bt; 2586 struct cmsghdr *cm; 2587 struct cmsgcred *cmcred; 2588 struct filedescent *fde, **fdep, *fdev; 2589 struct file *fp; 2590 struct timeval *tv; 2591 struct timespec *ts; 2592 void *data; 2593 socklen_t clen, datalen; 2594 int i, j, error, *fdp, oldfds; 2595 u_int newlen; 2596 2597 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 2598 UNP_LINK_UNLOCK_ASSERT(); 2599 2600 p = td->td_proc; 2601 fdesc = p->p_fd; 2602 error = 0; 2603 control = *controlp; 2604 *controlp = NULL; 2605 initial_controlp = controlp; 2606 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 2607 data = CMSG_DATA(cm); 2608 2609 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 2610 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 2611 (char *)cm + cm->cmsg_len >= (char *)data; 2612 2613 clen -= min(CMSG_SPACE(datalen), clen), 2614 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 2615 data = CMSG_DATA(cm)) { 2616 datalen = (char *)cm + cm->cmsg_len - (char *)data; 2617 switch (cm->cmsg_type) { 2618 case SCM_CREDS: 2619 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 2620 SCM_CREDS, SOL_SOCKET, M_WAITOK); 2621 cmcred = (struct cmsgcred *) 2622 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2623 cmcred->cmcred_pid = p->p_pid; 2624 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 2625 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 2626 cmcred->cmcred_euid = td->td_ucred->cr_uid; 2627 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 2628 CMGROUP_MAX); 2629 for (i = 0; i < cmcred->cmcred_ngroups; i++) 2630 cmcred->cmcred_groups[i] = 2631 td->td_ucred->cr_groups[i]; 2632 break; 2633 2634 case SCM_RIGHTS: 2635 oldfds = datalen / sizeof (int); 2636 if (oldfds == 0) 2637 continue; 2638 /* On some machines sizeof pointer is bigger than 2639 * sizeof int, so we need to check if data fits into 2640 * single mbuf. We could allocate several mbufs, and 2641 * unp_externalize() should even properly handle that. 2642 * But it is not worth to complicate the code for an 2643 * insane scenario of passing over 200 file descriptors 2644 * at once. 2645 */ 2646 newlen = oldfds * sizeof(fdep[0]); 2647 if (CMSG_SPACE(newlen) > MCLBYTES) { 2648 error = EMSGSIZE; 2649 goto out; 2650 } 2651 /* 2652 * Check that all the FDs passed in refer to legal 2653 * files. If not, reject the entire operation. 2654 */ 2655 fdp = data; 2656 FILEDESC_SLOCK(fdesc); 2657 for (i = 0; i < oldfds; i++, fdp++) { 2658 fp = fget_noref(fdesc, *fdp); 2659 if (fp == NULL) { 2660 FILEDESC_SUNLOCK(fdesc); 2661 error = EBADF; 2662 goto out; 2663 } 2664 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 2665 FILEDESC_SUNLOCK(fdesc); 2666 error = EOPNOTSUPP; 2667 goto out; 2668 } 2669 } 2670 2671 /* 2672 * Now replace the integer FDs with pointers to the 2673 * file structure and capability rights. 2674 */ 2675 *controlp = sbcreatecontrol(NULL, newlen, 2676 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 2677 fdp = data; 2678 for (i = 0; i < oldfds; i++, fdp++) { 2679 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 2680 fdp = data; 2681 for (j = 0; j < i; j++, fdp++) { 2682 fdrop(fdesc->fd_ofiles[*fdp]. 2683 fde_file, td); 2684 } 2685 FILEDESC_SUNLOCK(fdesc); 2686 error = EBADF; 2687 goto out; 2688 } 2689 } 2690 fdp = data; 2691 fdep = (struct filedescent **) 2692 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2693 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 2694 M_WAITOK); 2695 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 2696 fde = &fdesc->fd_ofiles[*fdp]; 2697 fdep[i] = fdev; 2698 fdep[i]->fde_file = fde->fde_file; 2699 filecaps_copy(&fde->fde_caps, 2700 &fdep[i]->fde_caps, true); 2701 unp_internalize_fp(fdep[i]->fde_file); 2702 } 2703 FILEDESC_SUNLOCK(fdesc); 2704 break; 2705 2706 case SCM_TIMESTAMP: 2707 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 2708 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK); 2709 tv = (struct timeval *) 2710 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2711 microtime(tv); 2712 break; 2713 2714 case SCM_BINTIME: 2715 *controlp = sbcreatecontrol(NULL, sizeof(*bt), 2716 SCM_BINTIME, SOL_SOCKET, M_WAITOK); 2717 bt = (struct bintime *) 2718 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2719 bintime(bt); 2720 break; 2721 2722 case SCM_REALTIME: 2723 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2724 SCM_REALTIME, SOL_SOCKET, M_WAITOK); 2725 ts = (struct timespec *) 2726 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2727 nanotime(ts); 2728 break; 2729 2730 case SCM_MONOTONIC: 2731 *controlp = sbcreatecontrol(NULL, sizeof(*ts), 2732 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK); 2733 ts = (struct timespec *) 2734 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 2735 nanouptime(ts); 2736 break; 2737 2738 default: 2739 error = EINVAL; 2740 goto out; 2741 } 2742 2743 if (space != NULL) { 2744 *space += (*controlp)->m_len; 2745 *mbcnt += MSIZE; 2746 if ((*controlp)->m_flags & M_EXT) 2747 *mbcnt += (*controlp)->m_ext.ext_size; 2748 *clast = *controlp; 2749 } 2750 controlp = &(*controlp)->m_next; 2751 } 2752 if (clen > 0) 2753 error = EINVAL; 2754 2755 out: 2756 if (error != 0 && initial_controlp != NULL) 2757 unp_internalize_cleanup_rights(*initial_controlp); 2758 m_freem(control); 2759 return (error); 2760 } 2761 2762 static struct mbuf * 2763 unp_addsockcred(struct thread *td, struct mbuf *control, int mode, 2764 struct mbuf **clast, u_int *space, u_int *mbcnt) 2765 { 2766 struct mbuf *m, *n, *n_prev; 2767 const struct cmsghdr *cm; 2768 int ngroups, i, cmsgtype; 2769 size_t ctrlsz; 2770 2771 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 2772 if (mode & UNP_WANTCRED_ALWAYS) { 2773 ctrlsz = SOCKCRED2SIZE(ngroups); 2774 cmsgtype = SCM_CREDS2; 2775 } else { 2776 ctrlsz = SOCKCREDSIZE(ngroups); 2777 cmsgtype = SCM_CREDS; 2778 } 2779 2780 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 2781 if (m == NULL) 2782 return (control); 2783 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 2784 2785 if (mode & UNP_WANTCRED_ALWAYS) { 2786 struct sockcred2 *sc; 2787 2788 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2789 sc->sc_version = 0; 2790 sc->sc_pid = td->td_proc->p_pid; 2791 sc->sc_uid = td->td_ucred->cr_ruid; 2792 sc->sc_euid = td->td_ucred->cr_uid; 2793 sc->sc_gid = td->td_ucred->cr_rgid; 2794 sc->sc_egid = td->td_ucred->cr_gid; 2795 sc->sc_ngroups = ngroups; 2796 for (i = 0; i < sc->sc_ngroups; i++) 2797 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2798 } else { 2799 struct sockcred *sc; 2800 2801 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 2802 sc->sc_uid = td->td_ucred->cr_ruid; 2803 sc->sc_euid = td->td_ucred->cr_uid; 2804 sc->sc_gid = td->td_ucred->cr_rgid; 2805 sc->sc_egid = td->td_ucred->cr_gid; 2806 sc->sc_ngroups = ngroups; 2807 for (i = 0; i < sc->sc_ngroups; i++) 2808 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 2809 } 2810 2811 /* 2812 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 2813 * created SCM_CREDS control message (struct sockcred) has another 2814 * format. 2815 */ 2816 if (control != NULL && cmsgtype == SCM_CREDS) 2817 for (n = control, n_prev = NULL; n != NULL;) { 2818 cm = mtod(n, struct cmsghdr *); 2819 if (cm->cmsg_level == SOL_SOCKET && 2820 cm->cmsg_type == SCM_CREDS) { 2821 if (n_prev == NULL) 2822 control = n->m_next; 2823 else 2824 n_prev->m_next = n->m_next; 2825 if (space != NULL) { 2826 MPASS(*space >= n->m_len); 2827 *space -= n->m_len; 2828 MPASS(*mbcnt >= MSIZE); 2829 *mbcnt -= MSIZE; 2830 if (n->m_flags & M_EXT) { 2831 MPASS(*mbcnt >= 2832 n->m_ext.ext_size); 2833 *mbcnt -= n->m_ext.ext_size; 2834 } 2835 MPASS(clast); 2836 if (*clast == n) { 2837 MPASS(n->m_next == NULL); 2838 if (n_prev == NULL) 2839 *clast = m; 2840 else 2841 *clast = n_prev; 2842 } 2843 } 2844 n = m_free(n); 2845 } else { 2846 n_prev = n; 2847 n = n->m_next; 2848 } 2849 } 2850 2851 /* Prepend it to the head. */ 2852 m->m_next = control; 2853 if (space != NULL) { 2854 *space += m->m_len; 2855 *mbcnt += MSIZE; 2856 if (control == NULL) 2857 *clast = m; 2858 } 2859 return (m); 2860 } 2861 2862 static struct unpcb * 2863 fptounp(struct file *fp) 2864 { 2865 struct socket *so; 2866 2867 if (fp->f_type != DTYPE_SOCKET) 2868 return (NULL); 2869 if ((so = fp->f_data) == NULL) 2870 return (NULL); 2871 if (so->so_proto->pr_domain != &localdomain) 2872 return (NULL); 2873 return sotounpcb(so); 2874 } 2875 2876 static void 2877 unp_discard(struct file *fp) 2878 { 2879 struct unp_defer *dr; 2880 2881 if (unp_externalize_fp(fp)) { 2882 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 2883 dr->ud_fp = fp; 2884 UNP_DEFERRED_LOCK(); 2885 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 2886 UNP_DEFERRED_UNLOCK(); 2887 atomic_add_int(&unp_defers_count, 1); 2888 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 2889 } else 2890 closef_nothread(fp); 2891 } 2892 2893 static void 2894 unp_process_defers(void *arg __unused, int pending) 2895 { 2896 struct unp_defer *dr; 2897 SLIST_HEAD(, unp_defer) drl; 2898 int count; 2899 2900 SLIST_INIT(&drl); 2901 for (;;) { 2902 UNP_DEFERRED_LOCK(); 2903 if (SLIST_FIRST(&unp_defers) == NULL) { 2904 UNP_DEFERRED_UNLOCK(); 2905 break; 2906 } 2907 SLIST_SWAP(&unp_defers, &drl, unp_defer); 2908 UNP_DEFERRED_UNLOCK(); 2909 count = 0; 2910 while ((dr = SLIST_FIRST(&drl)) != NULL) { 2911 SLIST_REMOVE_HEAD(&drl, ud_link); 2912 closef_nothread(dr->ud_fp); 2913 free(dr, M_TEMP); 2914 count++; 2915 } 2916 atomic_add_int(&unp_defers_count, -count); 2917 } 2918 } 2919 2920 static void 2921 unp_internalize_fp(struct file *fp) 2922 { 2923 struct unpcb *unp; 2924 2925 UNP_LINK_WLOCK(); 2926 if ((unp = fptounp(fp)) != NULL) { 2927 unp->unp_file = fp; 2928 unp->unp_msgcount++; 2929 } 2930 unp_rights++; 2931 UNP_LINK_WUNLOCK(); 2932 } 2933 2934 static int 2935 unp_externalize_fp(struct file *fp) 2936 { 2937 struct unpcb *unp; 2938 int ret; 2939 2940 UNP_LINK_WLOCK(); 2941 if ((unp = fptounp(fp)) != NULL) { 2942 unp->unp_msgcount--; 2943 ret = 1; 2944 } else 2945 ret = 0; 2946 unp_rights--; 2947 UNP_LINK_WUNLOCK(); 2948 return (ret); 2949 } 2950 2951 /* 2952 * unp_defer indicates whether additional work has been defered for a future 2953 * pass through unp_gc(). It is thread local and does not require explicit 2954 * synchronization. 2955 */ 2956 static int unp_marked; 2957 2958 static void 2959 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 2960 { 2961 struct unpcb *unp; 2962 struct file *fp; 2963 int i; 2964 2965 /* 2966 * This function can only be called from the gc task. 2967 */ 2968 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2969 ("%s: not on gc callout", __func__)); 2970 UNP_LINK_LOCK_ASSERT(); 2971 2972 for (i = 0; i < fdcount; i++) { 2973 fp = fdep[i]->fde_file; 2974 if ((unp = fptounp(fp)) == NULL) 2975 continue; 2976 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 2977 continue; 2978 unp->unp_gcrefs--; 2979 } 2980 } 2981 2982 static void 2983 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 2984 { 2985 struct unpcb *unp; 2986 struct file *fp; 2987 int i; 2988 2989 /* 2990 * This function can only be called from the gc task. 2991 */ 2992 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 2993 ("%s: not on gc callout", __func__)); 2994 UNP_LINK_LOCK_ASSERT(); 2995 2996 for (i = 0; i < fdcount; i++) { 2997 fp = fdep[i]->fde_file; 2998 if ((unp = fptounp(fp)) == NULL) 2999 continue; 3000 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 3001 continue; 3002 unp->unp_gcrefs++; 3003 unp_marked++; 3004 } 3005 } 3006 3007 static void 3008 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 3009 { 3010 struct sockbuf *sb; 3011 3012 SOCK_LOCK_ASSERT(so); 3013 3014 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 3015 return; 3016 3017 SOCK_RECVBUF_LOCK(so); 3018 switch (so->so_type) { 3019 case SOCK_DGRAM: 3020 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 3021 unp_scan(so->so_rcv.uxdg_peeked, op); 3022 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 3023 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 3024 break; 3025 case SOCK_STREAM: 3026 case SOCK_SEQPACKET: 3027 unp_scan(so->so_rcv.sb_mb, op); 3028 break; 3029 } 3030 SOCK_RECVBUF_UNLOCK(so); 3031 } 3032 3033 static void 3034 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 3035 { 3036 struct socket *so, *soa; 3037 3038 so = unp->unp_socket; 3039 SOCK_LOCK(so); 3040 if (SOLISTENING(so)) { 3041 /* 3042 * Mark all sockets in our accept queue. 3043 */ 3044 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 3045 unp_scan_socket(soa, op); 3046 } else { 3047 /* 3048 * Mark all sockets we reference with RIGHTS. 3049 */ 3050 unp_scan_socket(so, op); 3051 } 3052 SOCK_UNLOCK(so); 3053 } 3054 3055 static int unp_recycled; 3056 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 3057 "Number of unreachable sockets claimed by the garbage collector."); 3058 3059 static int unp_taskcount; 3060 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 3061 "Number of times the garbage collector has run."); 3062 3063 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 3064 "Number of active local sockets."); 3065 3066 static void 3067 unp_gc(__unused void *arg, int pending) 3068 { 3069 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 3070 NULL }; 3071 struct unp_head **head; 3072 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 3073 struct file *f, **unref; 3074 struct unpcb *unp, *unptmp; 3075 int i, total, unp_unreachable; 3076 3077 LIST_INIT(&unp_deadhead); 3078 unp_taskcount++; 3079 UNP_LINK_RLOCK(); 3080 /* 3081 * First determine which sockets may be in cycles. 3082 */ 3083 unp_unreachable = 0; 3084 3085 for (head = heads; *head != NULL; head++) 3086 LIST_FOREACH(unp, *head, unp_link) { 3087 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 3088 ("%s: unp %p has unexpected gc flags 0x%x", 3089 __func__, unp, (unsigned int)unp->unp_gcflag)); 3090 3091 f = unp->unp_file; 3092 3093 /* 3094 * Check for an unreachable socket potentially in a 3095 * cycle. It must be in a queue as indicated by 3096 * msgcount, and this must equal the file reference 3097 * count. Note that when msgcount is 0 the file is 3098 * NULL. 3099 */ 3100 if (f != NULL && unp->unp_msgcount != 0 && 3101 refcount_load(&f->f_count) == unp->unp_msgcount) { 3102 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 3103 unp->unp_gcflag |= UNPGC_DEAD; 3104 unp->unp_gcrefs = unp->unp_msgcount; 3105 unp_unreachable++; 3106 } 3107 } 3108 3109 /* 3110 * Scan all sockets previously marked as potentially being in a cycle 3111 * and remove the references each socket holds on any UNPGC_DEAD 3112 * sockets in its queue. After this step, all remaining references on 3113 * sockets marked UNPGC_DEAD should not be part of any cycle. 3114 */ 3115 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 3116 unp_gc_scan(unp, unp_remove_dead_ref); 3117 3118 /* 3119 * If a socket still has a non-negative refcount, it cannot be in a 3120 * cycle. In this case increment refcount of all children iteratively. 3121 * Stop the scan once we do a complete loop without discovering 3122 * a new reachable socket. 3123 */ 3124 do { 3125 unp_marked = 0; 3126 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 3127 if (unp->unp_gcrefs > 0) { 3128 unp->unp_gcflag &= ~UNPGC_DEAD; 3129 LIST_REMOVE(unp, unp_dead); 3130 KASSERT(unp_unreachable > 0, 3131 ("%s: unp_unreachable underflow.", 3132 __func__)); 3133 unp_unreachable--; 3134 unp_gc_scan(unp, unp_restore_undead_ref); 3135 } 3136 } while (unp_marked); 3137 3138 UNP_LINK_RUNLOCK(); 3139 3140 if (unp_unreachable == 0) 3141 return; 3142 3143 /* 3144 * Allocate space for a local array of dead unpcbs. 3145 * TODO: can this path be simplified by instead using the local 3146 * dead list at unp_deadhead, after taking out references 3147 * on the file object and/or unpcb and dropping the link lock? 3148 */ 3149 unref = malloc(unp_unreachable * sizeof(struct file *), 3150 M_TEMP, M_WAITOK); 3151 3152 /* 3153 * Iterate looking for sockets which have been specifically marked 3154 * as unreachable and store them locally. 3155 */ 3156 UNP_LINK_RLOCK(); 3157 total = 0; 3158 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 3159 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 3160 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 3161 unp->unp_gcflag &= ~UNPGC_DEAD; 3162 f = unp->unp_file; 3163 if (unp->unp_msgcount == 0 || f == NULL || 3164 refcount_load(&f->f_count) != unp->unp_msgcount || 3165 !fhold(f)) 3166 continue; 3167 unref[total++] = f; 3168 KASSERT(total <= unp_unreachable, 3169 ("%s: incorrect unreachable count.", __func__)); 3170 } 3171 UNP_LINK_RUNLOCK(); 3172 3173 /* 3174 * Now flush all sockets, free'ing rights. This will free the 3175 * struct files associated with these sockets but leave each socket 3176 * with one remaining ref. 3177 */ 3178 for (i = 0; i < total; i++) { 3179 struct socket *so; 3180 3181 so = unref[i]->f_data; 3182 CURVNET_SET(so->so_vnet); 3183 sorflush(so); 3184 CURVNET_RESTORE(); 3185 } 3186 3187 /* 3188 * And finally release the sockets so they can be reclaimed. 3189 */ 3190 for (i = 0; i < total; i++) 3191 fdrop(unref[i], NULL); 3192 unp_recycled += total; 3193 free(unref, M_TEMP); 3194 } 3195 3196 /* 3197 * Synchronize against unp_gc, which can trip over data as we are freeing it. 3198 */ 3199 static void 3200 unp_dispose(struct socket *so) 3201 { 3202 struct sockbuf *sb; 3203 struct unpcb *unp; 3204 struct mbuf *m; 3205 3206 MPASS(!SOLISTENING(so)); 3207 3208 unp = sotounpcb(so); 3209 UNP_LINK_WLOCK(); 3210 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 3211 UNP_LINK_WUNLOCK(); 3212 3213 /* 3214 * Grab our special mbufs before calling sbrelease(). 3215 */ 3216 SOCK_RECVBUF_LOCK(so); 3217 switch (so->so_type) { 3218 case SOCK_DGRAM: 3219 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 3220 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 3221 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 3222 /* Note: socket of sb may reconnect. */ 3223 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 3224 } 3225 sb = &so->so_rcv; 3226 if (sb->uxdg_peeked != NULL) { 3227 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 3228 m_stailqpkt); 3229 sb->uxdg_peeked = NULL; 3230 } 3231 m = STAILQ_FIRST(&sb->uxdg_mb); 3232 STAILQ_INIT(&sb->uxdg_mb); 3233 /* XXX: our shortened sbrelease() */ 3234 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 3235 RLIM_INFINITY); 3236 /* 3237 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to 3238 * prevent uipc_sosend_dgram() or unp_disconnect() adding more 3239 * data to the socket. 3240 * We are now in dom_dispose and it could be a call from 3241 * soshutdown() or from the final sofree(). The sofree() case 3242 * is simple as it guarantees that no more sends will happen, 3243 * however we can race with unp_disconnect() from our peer. 3244 * The shutdown(2) case is more exotic. It would call into 3245 * dom_dispose() only if socket is SS_ISCONNECTED. This is 3246 * possible if we did connect(2) on this socket and we also 3247 * had it bound with bind(2) and receive connections from other 3248 * sockets. Because soshutdown() violates POSIX (see comment 3249 * there) we will end up here shutting down our receive side. 3250 * Of course this will have affect not only on the peer we 3251 * connect(2)ed to, but also on all of the peers who had 3252 * connect(2)ed to us. Their sends would end up with ENOBUFS. 3253 */ 3254 sb->sb_state |= SBS_CANTRCVMORE; 3255 break; 3256 case SOCK_STREAM: 3257 case SOCK_SEQPACKET: 3258 sb = &so->so_rcv; 3259 m = sbcut_locked(sb, sb->sb_ccc); 3260 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0, 3261 ("%s: ccc %u mb %p mbcnt %u", __func__, 3262 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt)); 3263 sbrelease_locked(so, SO_RCV); 3264 break; 3265 } 3266 SOCK_RECVBUF_UNLOCK(so); 3267 if (SOCK_IO_RECV_OWNED(so)) 3268 SOCK_IO_RECV_UNLOCK(so); 3269 3270 if (m != NULL) { 3271 unp_scan(m, unp_freerights); 3272 m_freem(m); 3273 } 3274 } 3275 3276 static void 3277 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 3278 { 3279 struct mbuf *m; 3280 struct cmsghdr *cm; 3281 void *data; 3282 socklen_t clen, datalen; 3283 3284 while (m0 != NULL) { 3285 for (m = m0; m; m = m->m_next) { 3286 if (m->m_type != MT_CONTROL) 3287 continue; 3288 3289 cm = mtod(m, struct cmsghdr *); 3290 clen = m->m_len; 3291 3292 while (cm != NULL) { 3293 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 3294 break; 3295 3296 data = CMSG_DATA(cm); 3297 datalen = (caddr_t)cm + cm->cmsg_len 3298 - (caddr_t)data; 3299 3300 if (cm->cmsg_level == SOL_SOCKET && 3301 cm->cmsg_type == SCM_RIGHTS) { 3302 (*op)(data, datalen / 3303 sizeof(struct filedescent *)); 3304 } 3305 3306 if (CMSG_SPACE(datalen) < clen) { 3307 clen -= CMSG_SPACE(datalen); 3308 cm = (struct cmsghdr *) 3309 ((caddr_t)cm + CMSG_SPACE(datalen)); 3310 } else { 3311 clen = 0; 3312 cm = NULL; 3313 } 3314 } 3315 } 3316 m0 = m0->m_nextpkt; 3317 } 3318 } 3319 3320 /* 3321 * Definitions of protocols supported in the LOCAL domain. 3322 */ 3323 static struct protosw streamproto = { 3324 .pr_type = SOCK_STREAM, 3325 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS| 3326 PR_CAPATTACH, 3327 .pr_ctloutput = &uipc_ctloutput, 3328 .pr_abort = uipc_abort, 3329 .pr_accept = uipc_accept, 3330 .pr_attach = uipc_attach, 3331 .pr_bind = uipc_bind, 3332 .pr_bindat = uipc_bindat, 3333 .pr_connect = uipc_connect, 3334 .pr_connectat = uipc_connectat, 3335 .pr_connect2 = uipc_connect2, 3336 .pr_detach = uipc_detach, 3337 .pr_disconnect = uipc_disconnect, 3338 .pr_listen = uipc_listen, 3339 .pr_peeraddr = uipc_peeraddr, 3340 .pr_rcvd = uipc_rcvd, 3341 .pr_send = uipc_send, 3342 .pr_ready = uipc_ready, 3343 .pr_sense = uipc_sense, 3344 .pr_shutdown = uipc_shutdown, 3345 .pr_sockaddr = uipc_sockaddr, 3346 .pr_soreceive = soreceive_generic, 3347 .pr_close = uipc_close, 3348 }; 3349 3350 static struct protosw dgramproto = { 3351 .pr_type = SOCK_DGRAM, 3352 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH | 3353 PR_SOCKBUF, 3354 .pr_ctloutput = &uipc_ctloutput, 3355 .pr_abort = uipc_abort, 3356 .pr_accept = uipc_accept, 3357 .pr_attach = uipc_attach, 3358 .pr_bind = uipc_bind, 3359 .pr_bindat = uipc_bindat, 3360 .pr_connect = uipc_connect, 3361 .pr_connectat = uipc_connectat, 3362 .pr_connect2 = uipc_connect2, 3363 .pr_detach = uipc_detach, 3364 .pr_disconnect = uipc_disconnect, 3365 .pr_peeraddr = uipc_peeraddr, 3366 .pr_sosend = uipc_sosend_dgram, 3367 .pr_sense = uipc_sense, 3368 .pr_shutdown = uipc_shutdown, 3369 .pr_sockaddr = uipc_sockaddr, 3370 .pr_soreceive = uipc_soreceive_dgram, 3371 .pr_close = uipc_close, 3372 }; 3373 3374 static struct protosw seqpacketproto = { 3375 .pr_type = SOCK_SEQPACKET, 3376 /* 3377 * XXXRW: For now, PR_ADDR because soreceive will bump into them 3378 * due to our use of sbappendaddr. A new sbappend variants is needed 3379 * that supports both atomic record writes and control data. 3380 */ 3381 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED| 3382 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH, 3383 .pr_ctloutput = &uipc_ctloutput, 3384 .pr_abort = uipc_abort, 3385 .pr_accept = uipc_accept, 3386 .pr_attach = uipc_attach, 3387 .pr_bind = uipc_bind, 3388 .pr_bindat = uipc_bindat, 3389 .pr_connect = uipc_connect, 3390 .pr_connectat = uipc_connectat, 3391 .pr_connect2 = uipc_connect2, 3392 .pr_detach = uipc_detach, 3393 .pr_disconnect = uipc_disconnect, 3394 .pr_listen = uipc_listen, 3395 .pr_peeraddr = uipc_peeraddr, 3396 .pr_rcvd = uipc_rcvd, 3397 .pr_send = uipc_send, 3398 .pr_sense = uipc_sense, 3399 .pr_shutdown = uipc_shutdown, 3400 .pr_sockaddr = uipc_sockaddr, 3401 .pr_soreceive = soreceive_generic, /* XXX: or...? */ 3402 .pr_close = uipc_close, 3403 }; 3404 3405 static struct domain localdomain = { 3406 .dom_family = AF_LOCAL, 3407 .dom_name = "local", 3408 .dom_externalize = unp_externalize, 3409 .dom_dispose = unp_dispose, 3410 .dom_nprotosw = 3, 3411 .dom_protosw = { 3412 &streamproto, 3413 &dgramproto, 3414 &seqpacketproto, 3415 } 3416 }; 3417 DOMAIN_SET(local); 3418 3419 /* 3420 * A helper function called by VFS before socket-type vnode reclamation. 3421 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 3422 * use count. 3423 */ 3424 void 3425 vfs_unp_reclaim(struct vnode *vp) 3426 { 3427 struct unpcb *unp; 3428 int active; 3429 struct mtx *vplock; 3430 3431 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 3432 KASSERT(vp->v_type == VSOCK, 3433 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 3434 3435 active = 0; 3436 vplock = mtx_pool_find(mtxpool_sleep, vp); 3437 mtx_lock(vplock); 3438 VOP_UNP_CONNECT(vp, &unp); 3439 if (unp == NULL) 3440 goto done; 3441 UNP_PCB_LOCK(unp); 3442 if (unp->unp_vnode == vp) { 3443 VOP_UNP_DETACH(vp); 3444 unp->unp_vnode = NULL; 3445 active = 1; 3446 } 3447 UNP_PCB_UNLOCK(unp); 3448 done: 3449 mtx_unlock(vplock); 3450 if (active) 3451 vunref(vp); 3452 } 3453 3454 #ifdef DDB 3455 static void 3456 db_print_indent(int indent) 3457 { 3458 int i; 3459 3460 for (i = 0; i < indent; i++) 3461 db_printf(" "); 3462 } 3463 3464 static void 3465 db_print_unpflags(int unp_flags) 3466 { 3467 int comma; 3468 3469 comma = 0; 3470 if (unp_flags & UNP_HAVEPC) { 3471 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 3472 comma = 1; 3473 } 3474 if (unp_flags & UNP_WANTCRED_ALWAYS) { 3475 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 3476 comma = 1; 3477 } 3478 if (unp_flags & UNP_WANTCRED_ONESHOT) { 3479 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 3480 comma = 1; 3481 } 3482 if (unp_flags & UNP_CONNWAIT) { 3483 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 3484 comma = 1; 3485 } 3486 if (unp_flags & UNP_CONNECTING) { 3487 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 3488 comma = 1; 3489 } 3490 if (unp_flags & UNP_BINDING) { 3491 db_printf("%sUNP_BINDING", comma ? ", " : ""); 3492 comma = 1; 3493 } 3494 } 3495 3496 static void 3497 db_print_xucred(int indent, struct xucred *xu) 3498 { 3499 int comma, i; 3500 3501 db_print_indent(indent); 3502 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 3503 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 3504 db_print_indent(indent); 3505 db_printf("cr_groups: "); 3506 comma = 0; 3507 for (i = 0; i < xu->cr_ngroups; i++) { 3508 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 3509 comma = 1; 3510 } 3511 db_printf("\n"); 3512 } 3513 3514 static void 3515 db_print_unprefs(int indent, struct unp_head *uh) 3516 { 3517 struct unpcb *unp; 3518 int counter; 3519 3520 counter = 0; 3521 LIST_FOREACH(unp, uh, unp_reflink) { 3522 if (counter % 4 == 0) 3523 db_print_indent(indent); 3524 db_printf("%p ", unp); 3525 if (counter % 4 == 3) 3526 db_printf("\n"); 3527 counter++; 3528 } 3529 if (counter != 0 && counter % 4 != 0) 3530 db_printf("\n"); 3531 } 3532 3533 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 3534 { 3535 struct unpcb *unp; 3536 3537 if (!have_addr) { 3538 db_printf("usage: show unpcb <addr>\n"); 3539 return; 3540 } 3541 unp = (struct unpcb *)addr; 3542 3543 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 3544 unp->unp_vnode); 3545 3546 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 3547 unp->unp_conn); 3548 3549 db_printf("unp_refs:\n"); 3550 db_print_unprefs(2, &unp->unp_refs); 3551 3552 /* XXXRW: Would be nice to print the full address, if any. */ 3553 db_printf("unp_addr: %p\n", unp->unp_addr); 3554 3555 db_printf("unp_gencnt: %llu\n", 3556 (unsigned long long)unp->unp_gencnt); 3557 3558 db_printf("unp_flags: %x (", unp->unp_flags); 3559 db_print_unpflags(unp->unp_flags); 3560 db_printf(")\n"); 3561 3562 db_printf("unp_peercred:\n"); 3563 db_print_xucred(2, &unp->unp_peercred); 3564 3565 db_printf("unp_refcount: %u\n", unp->unp_refcount); 3566 } 3567 #endif 3568