1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All Rights Reserved. 6 * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved. 7 * Copyright (c) 2018 Matthew Macy 8 * Copyright (c) 2022-2025 Gleb Smirnoff <glebius@FreeBSD.org> 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /* 36 * UNIX Domain (Local) Sockets 37 * 38 * This is an implementation of UNIX (local) domain sockets. Each socket has 39 * an associated struct unpcb (UNIX protocol control block). Stream sockets 40 * may be connected to 0 or 1 other socket. Datagram sockets may be 41 * connected to 0, 1, or many other sockets. Sockets may be created and 42 * connected in pairs (socketpair(2)), or bound/connected to using the file 43 * system name space. For most purposes, only the receive socket buffer is 44 * used, as sending on one socket delivers directly to the receive socket 45 * buffer of a second socket. 46 * 47 * The implementation is substantially complicated by the fact that 48 * "ancillary data", such as file descriptors or credentials, may be passed 49 * across UNIX domain sockets. The potential for passing UNIX domain sockets 50 * over other UNIX domain sockets requires the implementation of a simple 51 * garbage collector to find and tear down cycles of disconnected sockets. 52 * 53 * TODO: 54 * RDM 55 * rethink name space problems 56 * need a proper out-of-band 57 */ 58 59 #include "opt_ddb.h" 60 61 #include <sys/param.h> 62 #include <sys/capsicum.h> 63 #include <sys/domain.h> 64 #include <sys/eventhandler.h> 65 #include <sys/fcntl.h> 66 #include <sys/file.h> 67 #include <sys/filedesc.h> 68 #include <sys/jail.h> 69 #include <sys/kernel.h> 70 #include <sys/lock.h> 71 #include <sys/malloc.h> 72 #include <sys/mbuf.h> 73 #include <sys/mount.h> 74 #include <sys/mutex.h> 75 #include <sys/namei.h> 76 #include <sys/poll.h> 77 #include <sys/proc.h> 78 #include <sys/protosw.h> 79 #include <sys/queue.h> 80 #include <sys/resourcevar.h> 81 #include <sys/rwlock.h> 82 #include <sys/socket.h> 83 #include <sys/socketvar.h> 84 #include <sys/signalvar.h> 85 #include <sys/stat.h> 86 #include <sys/sx.h> 87 #include <sys/sysctl.h> 88 #include <sys/systm.h> 89 #include <sys/taskqueue.h> 90 #include <sys/un.h> 91 #include <sys/unpcb.h> 92 #include <sys/vnode.h> 93 94 #include <net/vnet.h> 95 96 #ifdef DDB 97 #include <ddb/ddb.h> 98 #endif 99 100 #include <security/mac/mac_framework.h> 101 102 #include <vm/uma.h> 103 104 MALLOC_DECLARE(M_FILECAPS); 105 106 static struct domain localdomain; 107 108 static uma_zone_t unp_zone; 109 static unp_gen_t unp_gencnt; /* (l) */ 110 static u_int unp_count; /* (l) Count of local sockets. */ 111 static ino_t unp_ino; /* Prototype for fake inode numbers. */ 112 static int unp_rights; /* (g) File descriptors in flight. */ 113 static struct unp_head unp_shead; /* (l) List of stream sockets. */ 114 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ 115 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ 116 static struct mtx_pool *unp_vp_mtxpool; 117 118 struct unp_defer { 119 SLIST_ENTRY(unp_defer) ud_link; 120 struct file *ud_fp; 121 }; 122 static SLIST_HEAD(, unp_defer) unp_defers; 123 static int unp_defers_count; 124 125 static const struct sockaddr sun_noname = { 126 .sa_len = sizeof(sun_noname), 127 .sa_family = AF_LOCAL, 128 }; 129 130 /* 131 * Garbage collection of cyclic file descriptor/socket references occurs 132 * asynchronously in a taskqueue context in order to avoid recursion and 133 * reentrance in the UNIX domain socket, file descriptor, and socket layer 134 * code. See unp_gc() for a full description. 135 */ 136 static struct timeout_task unp_gc_task; 137 138 /* 139 * The close of unix domain sockets attached as SCM_RIGHTS is 140 * postponed to the taskqueue, to avoid arbitrary recursion depth. 141 * The attached sockets might have another sockets attached. 142 */ 143 static struct task unp_defer_task; 144 145 /* 146 * SOCK_STREAM and SOCK_SEQPACKET unix(4) sockets fully bypass the send buffer, 147 * however the notion of send buffer still makes sense with them. Its size is 148 * the amount of space that a send(2) syscall may copyin(9) before checking 149 * with the receive buffer of a peer. Although not linked anywhere yet, 150 * pointed to by a stack variable, effectively it is a buffer that needs to be 151 * sized. 152 * 153 * SOCK_DGRAM sockets really use the sendspace as the maximum datagram size, 154 * and don't really want to reserve the sendspace. Their recvspace should be 155 * large enough for at least one max-size datagram plus address. 156 */ 157 static u_long unpst_sendspace = 64*1024; 158 static u_long unpst_recvspace = 64*1024; 159 static u_long unpdg_maxdgram = 8*1024; /* support 8KB syslog msgs */ 160 static u_long unpdg_recvspace = 16*1024; 161 static u_long unpsp_sendspace = 64*1024; 162 static u_long unpsp_recvspace = 64*1024; 163 164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 165 "Local domain"); 166 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, 167 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 168 "SOCK_STREAM"); 169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, 170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 171 "SOCK_DGRAM"); 172 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, 173 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 174 "SOCK_SEQPACKET"); 175 176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 177 &unpst_sendspace, 0, "Default stream send space."); 178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 179 &unpst_recvspace, 0, "Default stream receive space."); 180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 181 &unpdg_maxdgram, 0, "Maximum datagram size."); 182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 183 &unpdg_recvspace, 0, "Default datagram receive space."); 184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, 185 &unpsp_sendspace, 0, "Default seqpacket send space."); 186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, 187 &unpsp_recvspace, 0, "Default seqpacket receive space."); 188 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, 189 "File descriptors in flight."); 190 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, 191 &unp_defers_count, 0, 192 "File descriptors deferred to taskqueue for close."); 193 194 /* 195 * Locking and synchronization: 196 * 197 * Several types of locks exist in the local domain socket implementation: 198 * - a global linkage lock 199 * - a global connection list lock 200 * - the mtxpool lock 201 * - per-unpcb mutexes 202 * 203 * The linkage lock protects the global socket lists, the generation number 204 * counter and garbage collector state. 205 * 206 * The connection list lock protects the list of referring sockets in a datagram 207 * socket PCB. This lock is also overloaded to protect a global list of 208 * sockets whose buffers contain socket references in the form of SCM_RIGHTS 209 * messages. To avoid recursion, such references are released by a dedicated 210 * thread. 211 * 212 * The mtxpool lock protects the vnode from being modified while referenced. 213 * Lock ordering rules require that it be acquired before any PCB locks. 214 * 215 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the 216 * unpcb. This includes the unp_conn field, which either links two connected 217 * PCBs together (for connected socket types) or points at the destination 218 * socket (for connectionless socket types). The operations of creating or 219 * destroying a connection therefore involve locking multiple PCBs. To avoid 220 * lock order reversals, in some cases this involves dropping a PCB lock and 221 * using a reference counter to maintain liveness. 222 * 223 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 224 * allocated in pr_attach() and freed in pr_detach(). The validity of that 225 * pointer is an invariant, so no lock is required to dereference the so_pcb 226 * pointer if a valid socket reference is held by the caller. In practice, 227 * this is always true during operations performed on a socket. Each unpcb 228 * has a back-pointer to its socket, unp_socket, which will be stable under 229 * the same circumstances. 230 * 231 * This pointer may only be safely dereferenced as long as a valid reference 232 * to the unpcb is held. Typically, this reference will be from the socket, 233 * or from another unpcb when the referring unpcb's lock is held (in order 234 * that the reference not be invalidated during use). For example, to follow 235 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee 236 * that detach is not run clearing unp_socket. 237 * 238 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 239 * protocols, bind() is a non-atomic operation, and connect() requires 240 * potential sleeping in the protocol, due to potentially waiting on local or 241 * distributed file systems. We try to separate "lookup" operations, which 242 * may sleep, and the IPC operations themselves, which typically can occur 243 * with relative atomicity as locks can be held over the entire operation. 244 * 245 * Another tricky issue is simultaneous multi-threaded or multi-process 246 * access to a single UNIX domain socket. These are handled by the flags 247 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 248 * binding, both of which involve dropping UNIX domain socket locks in order 249 * to perform namei() and other file system operations. 250 */ 251 static struct rwlock unp_link_rwlock; 252 static struct mtx unp_defers_lock; 253 254 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ 255 "unp_link_rwlock") 256 257 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 258 RA_LOCKED) 259 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 260 RA_UNLOCKED) 261 262 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) 263 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) 264 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) 265 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) 266 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ 267 RA_WLOCKED) 268 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) 269 270 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ 271 "unp_defer", NULL, MTX_DEF) 272 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) 273 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) 274 275 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); 276 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); 277 278 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 279 "unp", "unp", \ 280 MTX_DUPOK|MTX_DEF) 281 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 282 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) 283 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 284 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) 285 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 286 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) 287 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 288 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) 289 290 static int uipc_connect2(struct socket *, struct socket *); 291 static int uipc_ctloutput(struct socket *, struct sockopt *); 292 static int unp_connect(struct socket *, struct sockaddr *, 293 struct thread *); 294 static int unp_connectat(int, struct socket *, struct sockaddr *, 295 struct thread *, bool); 296 static void unp_connect2(struct socket *, struct socket *, bool); 297 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 298 static void unp_dispose(struct socket *so); 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 mchain *, 305 struct thread *); 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 void unp_addsockcred(struct thread *, struct mchain *, int); 310 static void unp_process_defers(void * __unused, int); 311 312 static void uipc_wrknl_lock(void *); 313 static void uipc_wrknl_unlock(void *); 314 static void uipc_wrknl_assert_lock(void *, int); 315 316 static void 317 unp_pcb_hold(struct unpcb *unp) 318 { 319 u_int old __unused; 320 321 old = refcount_acquire(&unp->unp_refcount); 322 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); 323 } 324 325 static __result_use_check bool 326 unp_pcb_rele(struct unpcb *unp) 327 { 328 bool ret; 329 330 UNP_PCB_LOCK_ASSERT(unp); 331 332 if ((ret = refcount_release(&unp->unp_refcount))) { 333 UNP_PCB_UNLOCK(unp); 334 UNP_PCB_LOCK_DESTROY(unp); 335 uma_zfree(unp_zone, unp); 336 } 337 return (ret); 338 } 339 340 static void 341 unp_pcb_rele_notlast(struct unpcb *unp) 342 { 343 bool ret __unused; 344 345 ret = refcount_release(&unp->unp_refcount); 346 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); 347 } 348 349 static void 350 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) 351 { 352 UNP_PCB_UNLOCK_ASSERT(unp); 353 UNP_PCB_UNLOCK_ASSERT(unp2); 354 355 if (unp == unp2) { 356 UNP_PCB_LOCK(unp); 357 } else if ((uintptr_t)unp2 > (uintptr_t)unp) { 358 UNP_PCB_LOCK(unp); 359 UNP_PCB_LOCK(unp2); 360 } else { 361 UNP_PCB_LOCK(unp2); 362 UNP_PCB_LOCK(unp); 363 } 364 } 365 366 static void 367 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) 368 { 369 UNP_PCB_UNLOCK(unp); 370 if (unp != unp2) 371 UNP_PCB_UNLOCK(unp2); 372 } 373 374 /* 375 * Try to lock the connected peer of an already locked socket. In some cases 376 * this requires that we unlock the current socket. The pairbusy counter is 377 * used to block concurrent connection attempts while the lock is dropped. The 378 * caller must be careful to revalidate PCB state. 379 */ 380 static struct unpcb * 381 unp_pcb_lock_peer(struct unpcb *unp) 382 { 383 struct unpcb *unp2; 384 385 UNP_PCB_LOCK_ASSERT(unp); 386 unp2 = unp->unp_conn; 387 if (unp2 == NULL) 388 return (NULL); 389 if (__predict_false(unp == unp2)) 390 return (unp); 391 392 UNP_PCB_UNLOCK_ASSERT(unp2); 393 394 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) 395 return (unp2); 396 if ((uintptr_t)unp2 > (uintptr_t)unp) { 397 UNP_PCB_LOCK(unp2); 398 return (unp2); 399 } 400 unp->unp_pairbusy++; 401 unp_pcb_hold(unp2); 402 UNP_PCB_UNLOCK(unp); 403 404 UNP_PCB_LOCK(unp2); 405 UNP_PCB_LOCK(unp); 406 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, 407 ("%s: socket %p was reconnected", __func__, unp)); 408 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { 409 unp->unp_flags &= ~UNP_WAITING; 410 wakeup(unp); 411 } 412 if (unp_pcb_rele(unp2)) { 413 /* unp2 is unlocked. */ 414 return (NULL); 415 } 416 if (unp->unp_conn == NULL) { 417 UNP_PCB_UNLOCK(unp2); 418 return (NULL); 419 } 420 return (unp2); 421 } 422 423 /* 424 * Try to lock peer of our socket for purposes of sending data to it. 425 */ 426 static int 427 uipc_lock_peer(struct socket *so, struct unpcb **unp2) 428 { 429 struct unpcb *unp; 430 int error; 431 432 unp = sotounpcb(so); 433 UNP_PCB_LOCK(unp); 434 *unp2 = unp_pcb_lock_peer(unp); 435 if (__predict_false(so->so_error != 0)) { 436 error = so->so_error; 437 so->so_error = 0; 438 UNP_PCB_UNLOCK(unp); 439 if (*unp2 != NULL) 440 UNP_PCB_UNLOCK(*unp2); 441 return (error); 442 } 443 if (__predict_false(*unp2 == NULL)) { 444 /* 445 * Different error code for a previously connected socket and 446 * a never connected one. The SS_ISDISCONNECTED is set in the 447 * unp_soisdisconnected() and is synchronized by the pcb lock. 448 */ 449 error = so->so_state & SS_ISDISCONNECTED ? EPIPE : ENOTCONN; 450 UNP_PCB_UNLOCK(unp); 451 return (error); 452 } 453 UNP_PCB_UNLOCK(unp); 454 455 return (0); 456 } 457 458 static void 459 uipc_abort(struct socket *so) 460 { 461 struct unpcb *unp, *unp2; 462 463 unp = sotounpcb(so); 464 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 465 UNP_PCB_UNLOCK_ASSERT(unp); 466 467 UNP_PCB_LOCK(unp); 468 unp2 = unp->unp_conn; 469 if (unp2 != NULL) { 470 unp_pcb_hold(unp2); 471 UNP_PCB_UNLOCK(unp); 472 unp_drop(unp2); 473 } else 474 UNP_PCB_UNLOCK(unp); 475 } 476 477 static int 478 uipc_attach(struct socket *so, int proto, struct thread *td) 479 { 480 u_long sendspace, recvspace; 481 struct unpcb *unp; 482 int error; 483 bool locked; 484 485 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 486 switch (so->so_type) { 487 case SOCK_DGRAM: 488 STAILQ_INIT(&so->so_rcv.uxdg_mb); 489 STAILQ_INIT(&so->so_snd.uxdg_mb); 490 TAILQ_INIT(&so->so_rcv.uxdg_conns); 491 /* 492 * Since send buffer is either bypassed or is a part 493 * of one-to-many receive buffer, we assign both space 494 * limits to unpdg_recvspace. 495 */ 496 sendspace = recvspace = unpdg_recvspace; 497 break; 498 499 case SOCK_STREAM: 500 sendspace = unpst_sendspace; 501 recvspace = unpst_recvspace; 502 goto common; 503 504 case SOCK_SEQPACKET: 505 sendspace = unpsp_sendspace; 506 recvspace = unpsp_recvspace; 507 common: 508 /* 509 * XXXGL: we need to initialize the mutex with MTX_DUPOK. 510 * Ideally, protocols that have PR_SOCKBUF should be 511 * responsible for mutex initialization officially, and then 512 * this uglyness with mtx_destroy(); mtx_init(); would go away. 513 */ 514 mtx_destroy(&so->so_rcv_mtx); 515 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF | MTX_DUPOK); 516 knlist_init(&so->so_wrsel.si_note, so, uipc_wrknl_lock, 517 uipc_wrknl_unlock, uipc_wrknl_assert_lock); 518 STAILQ_INIT(&so->so_rcv.uxst_mbq); 519 break; 520 default: 521 panic("uipc_attach"); 522 } 523 error = soreserve(so, sendspace, recvspace); 524 if (error) 525 return (error); 526 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 527 if (unp == NULL) 528 return (ENOBUFS); 529 LIST_INIT(&unp->unp_refs); 530 UNP_PCB_LOCK_INIT(unp); 531 unp->unp_socket = so; 532 so->so_pcb = unp; 533 refcount_init(&unp->unp_refcount, 1); 534 unp->unp_mode = ACCESSPERMS; 535 536 if ((locked = UNP_LINK_WOWNED()) == false) 537 UNP_LINK_WLOCK(); 538 539 unp->unp_gencnt = ++unp_gencnt; 540 unp->unp_ino = ++unp_ino; 541 unp_count++; 542 switch (so->so_type) { 543 case SOCK_STREAM: 544 LIST_INSERT_HEAD(&unp_shead, unp, unp_link); 545 break; 546 547 case SOCK_DGRAM: 548 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); 549 break; 550 551 case SOCK_SEQPACKET: 552 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); 553 break; 554 555 default: 556 panic("uipc_attach"); 557 } 558 559 if (locked == false) 560 UNP_LINK_WUNLOCK(); 561 562 return (0); 563 } 564 565 static int 566 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 567 { 568 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 569 struct vattr vattr; 570 int error, namelen; 571 struct nameidata nd; 572 struct unpcb *unp; 573 struct vnode *vp; 574 struct mount *mp; 575 cap_rights_t rights; 576 char *buf; 577 mode_t mode; 578 579 if (nam->sa_family != AF_UNIX) 580 return (EAFNOSUPPORT); 581 582 unp = sotounpcb(so); 583 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 584 585 if (soun->sun_len > sizeof(struct sockaddr_un)) 586 return (EINVAL); 587 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 588 if (namelen <= 0) 589 return (EINVAL); 590 591 /* 592 * We don't allow simultaneous bind() calls on a single UNIX domain 593 * socket, so flag in-progress operations, and return an error if an 594 * operation is already in progress. 595 * 596 * Historically, we have not allowed a socket to be rebound, so this 597 * also returns an error. Not allowing re-binding simplifies the 598 * implementation and avoids a great many possible failure modes. 599 */ 600 UNP_PCB_LOCK(unp); 601 if (unp->unp_vnode != NULL) { 602 UNP_PCB_UNLOCK(unp); 603 return (EINVAL); 604 } 605 if (unp->unp_flags & UNP_BINDING) { 606 UNP_PCB_UNLOCK(unp); 607 return (EALREADY); 608 } 609 unp->unp_flags |= UNP_BINDING; 610 mode = unp->unp_mode & ~td->td_proc->p_pd->pd_cmask; 611 UNP_PCB_UNLOCK(unp); 612 613 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 614 bcopy(soun->sun_path, buf, namelen); 615 buf[namelen] = 0; 616 617 restart: 618 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE, 619 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT)); 620 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 621 error = namei(&nd); 622 if (error) 623 goto error; 624 vp = nd.ni_vp; 625 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 626 NDFREE_PNBUF(&nd); 627 if (nd.ni_dvp == vp) 628 vrele(nd.ni_dvp); 629 else 630 vput(nd.ni_dvp); 631 if (vp != NULL) { 632 vrele(vp); 633 error = EADDRINUSE; 634 goto error; 635 } 636 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH); 637 if (error) 638 goto error; 639 goto restart; 640 } 641 VATTR_NULL(&vattr); 642 vattr.va_type = VSOCK; 643 vattr.va_mode = mode; 644 #ifdef MAC 645 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 646 &vattr); 647 #endif 648 if (error == 0) { 649 /* 650 * The prior lookup may have left LK_SHARED in cn_lkflags, 651 * and VOP_CREATE technically only requires the new vnode to 652 * be locked shared. Most filesystems will return the new vnode 653 * locked exclusive regardless, but we should explicitly 654 * specify that here since we require it and assert to that 655 * effect below. 656 */ 657 nd.ni_cnd.cn_lkflags = (nd.ni_cnd.cn_lkflags & ~LK_SHARED) | 658 LK_EXCLUSIVE; 659 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 660 } 661 NDFREE_PNBUF(&nd); 662 if (error) { 663 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true); 664 vn_finished_write(mp); 665 if (error == ERELOOKUP) 666 goto restart; 667 goto error; 668 } 669 vp = nd.ni_vp; 670 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 671 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 672 673 UNP_PCB_LOCK(unp); 674 VOP_UNP_BIND(vp, unp); 675 unp->unp_vnode = vp; 676 unp->unp_addr = soun; 677 unp->unp_flags &= ~UNP_BINDING; 678 UNP_PCB_UNLOCK(unp); 679 vref(vp); 680 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true); 681 vn_finished_write(mp); 682 free(buf, M_TEMP); 683 return (0); 684 685 error: 686 UNP_PCB_LOCK(unp); 687 unp->unp_flags &= ~UNP_BINDING; 688 UNP_PCB_UNLOCK(unp); 689 free(buf, M_TEMP); 690 return (error); 691 } 692 693 static int 694 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 695 { 696 697 return (uipc_bindat(AT_FDCWD, so, nam, td)); 698 } 699 700 static int 701 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 702 { 703 int error; 704 705 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 706 error = unp_connect(so, nam, td); 707 return (error); 708 } 709 710 static int 711 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, 712 struct thread *td) 713 { 714 int error; 715 716 KASSERT(td == curthread, ("uipc_connectat: td != curthread")); 717 error = unp_connectat(fd, so, nam, td, false); 718 return (error); 719 } 720 721 static void 722 uipc_close(struct socket *so) 723 { 724 struct unpcb *unp, *unp2; 725 struct vnode *vp = NULL; 726 struct mtx *vplock; 727 728 unp = sotounpcb(so); 729 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 730 731 vplock = NULL; 732 if ((vp = unp->unp_vnode) != NULL) { 733 vplock = mtx_pool_find(unp_vp_mtxpool, vp); 734 mtx_lock(vplock); 735 } 736 UNP_PCB_LOCK(unp); 737 if (vp && unp->unp_vnode == NULL) { 738 mtx_unlock(vplock); 739 vp = NULL; 740 } 741 if (vp != NULL) { 742 VOP_UNP_DETACH(vp); 743 unp->unp_vnode = NULL; 744 } 745 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 746 unp_disconnect(unp, unp2); 747 else 748 UNP_PCB_UNLOCK(unp); 749 if (vp) { 750 mtx_unlock(vplock); 751 vrele(vp); 752 } 753 } 754 755 static int 756 uipc_chmod(struct socket *so, mode_t mode, struct ucred *cred __unused, 757 struct thread *td __unused) 758 { 759 struct unpcb *unp; 760 int error; 761 762 if ((mode & ~ACCESSPERMS) != 0) 763 return (EINVAL); 764 765 error = 0; 766 unp = sotounpcb(so); 767 UNP_PCB_LOCK(unp); 768 if (unp->unp_vnode != NULL || (unp->unp_flags & UNP_BINDING) != 0) 769 error = EINVAL; 770 else 771 unp->unp_mode = mode; 772 UNP_PCB_UNLOCK(unp); 773 return (error); 774 } 775 776 static int 777 uipc_connect2(struct socket *so1, struct socket *so2) 778 { 779 struct unpcb *unp, *unp2; 780 781 if (so1->so_type != so2->so_type) 782 return (EPROTOTYPE); 783 784 unp = so1->so_pcb; 785 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 786 unp2 = so2->so_pcb; 787 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 788 unp_pcb_lock_pair(unp, unp2); 789 unp_connect2(so1, so2, false); 790 unp_pcb_unlock_pair(unp, unp2); 791 792 return (0); 793 } 794 795 static void 796 maybe_schedule_gc(void) 797 { 798 if (atomic_load_int(&unp_rights) != 0) 799 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); 800 } 801 802 static void 803 uipc_detach(struct socket *so) 804 { 805 struct unpcb *unp, *unp2; 806 struct mtx *vplock; 807 struct vnode *vp; 808 809 unp = sotounpcb(so); 810 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 811 812 vp = NULL; 813 vplock = NULL; 814 815 if (!SOLISTENING(so)) 816 unp_dispose(so); 817 818 UNP_LINK_WLOCK(); 819 LIST_REMOVE(unp, unp_link); 820 if (unp->unp_gcflag & UNPGC_DEAD) 821 LIST_REMOVE(unp, unp_dead); 822 unp->unp_gencnt = ++unp_gencnt; 823 --unp_count; 824 UNP_LINK_WUNLOCK(); 825 826 UNP_PCB_UNLOCK_ASSERT(unp); 827 restart: 828 if ((vp = unp->unp_vnode) != NULL) { 829 vplock = mtx_pool_find(unp_vp_mtxpool, vp); 830 mtx_lock(vplock); 831 } 832 UNP_PCB_LOCK(unp); 833 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { 834 if (vplock) 835 mtx_unlock(vplock); 836 UNP_PCB_UNLOCK(unp); 837 goto restart; 838 } 839 if ((vp = unp->unp_vnode) != NULL) { 840 VOP_UNP_DETACH(vp); 841 unp->unp_vnode = NULL; 842 } 843 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 844 unp_disconnect(unp, unp2); 845 else 846 UNP_PCB_UNLOCK(unp); 847 848 UNP_REF_LIST_LOCK(); 849 while (!LIST_EMPTY(&unp->unp_refs)) { 850 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 851 852 unp_pcb_hold(ref); 853 UNP_REF_LIST_UNLOCK(); 854 855 MPASS(ref != unp); 856 UNP_PCB_UNLOCK_ASSERT(ref); 857 unp_drop(ref); 858 UNP_REF_LIST_LOCK(); 859 } 860 UNP_REF_LIST_UNLOCK(); 861 862 UNP_PCB_LOCK(unp); 863 unp->unp_socket->so_pcb = NULL; 864 unp->unp_socket = NULL; 865 free(unp->unp_addr, M_SONAME); 866 unp->unp_addr = NULL; 867 if (!unp_pcb_rele(unp)) 868 UNP_PCB_UNLOCK(unp); 869 if (vp) { 870 mtx_unlock(vplock); 871 vrele(vp); 872 } 873 maybe_schedule_gc(); 874 875 switch (so->so_type) { 876 case SOCK_STREAM: 877 case SOCK_SEQPACKET: 878 MPASS(SOLISTENING(so) || (STAILQ_EMPTY(&so->so_rcv.uxst_mbq) && 879 so->so_rcv.uxst_peer == NULL)); 880 break; 881 case SOCK_DGRAM: 882 /* 883 * Everything should have been unlinked/freed by unp_dispose() 884 * and/or unp_disconnect(). 885 */ 886 MPASS(so->so_rcv.uxdg_peeked == NULL); 887 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb)); 888 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns)); 889 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb)); 890 } 891 } 892 893 static int 894 uipc_disconnect(struct socket *so) 895 { 896 struct unpcb *unp, *unp2; 897 898 unp = sotounpcb(so); 899 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 900 901 UNP_PCB_LOCK(unp); 902 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) 903 unp_disconnect(unp, unp2); 904 else 905 UNP_PCB_UNLOCK(unp); 906 return (0); 907 } 908 909 static void 910 uipc_fdclose(struct socket *so __unused) 911 { 912 /* 913 * Ensure that userspace can't create orphaned file descriptors without 914 * triggering garbage collection. Triggering GC from uipc_detach() is 915 * not sufficient, since that's only closed once a socket reference 916 * count drops to zero. 917 */ 918 maybe_schedule_gc(); 919 } 920 921 static int 922 uipc_listen(struct socket *so, int backlog, struct thread *td) 923 { 924 struct unpcb *unp; 925 int error; 926 927 MPASS(so->so_type != SOCK_DGRAM); 928 929 /* 930 * Synchronize with concurrent connection attempts. 931 */ 932 error = 0; 933 unp = sotounpcb(so); 934 UNP_PCB_LOCK(unp); 935 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0) 936 error = EINVAL; 937 else if (unp->unp_vnode == NULL) 938 error = EDESTADDRREQ; 939 if (error != 0) { 940 UNP_PCB_UNLOCK(unp); 941 return (error); 942 } 943 944 SOCK_LOCK(so); 945 error = solisten_proto_check(so); 946 if (error == 0) { 947 cru2xt(td, &unp->unp_peercred); 948 if (!SOLISTENING(so)) { 949 (void)chgsbsize(so->so_cred->cr_uidinfo, 950 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 951 (void)chgsbsize(so->so_cred->cr_uidinfo, 952 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 953 } 954 solisten_proto(so, backlog); 955 } 956 SOCK_UNLOCK(so); 957 UNP_PCB_UNLOCK(unp); 958 return (error); 959 } 960 961 static int 962 uipc_peeraddr(struct socket *so, struct sockaddr *ret) 963 { 964 struct unpcb *unp, *unp2; 965 const struct sockaddr *sa; 966 967 unp = sotounpcb(so); 968 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 969 970 UNP_PCB_LOCK(unp); 971 unp2 = unp_pcb_lock_peer(unp); 972 if (unp2 != NULL) { 973 if (unp2->unp_addr != NULL) 974 sa = (struct sockaddr *)unp2->unp_addr; 975 else 976 sa = &sun_noname; 977 bcopy(sa, ret, sa->sa_len); 978 unp_pcb_unlock_pair(unp, unp2); 979 } else { 980 UNP_PCB_UNLOCK(unp); 981 sa = &sun_noname; 982 bcopy(sa, ret, sa->sa_len); 983 } 984 return (0); 985 } 986 987 /* 988 * pr_sosend() called with mbuf instead of uio is a kernel thread. NFS, 989 * netgraph(4) and other subsystems can call into socket code. The 990 * function will condition the mbuf so that it can be safely put onto socket 991 * buffer and calculate its char count and mbuf count. 992 * 993 * Note: we don't support receiving control data from a kernel thread. Our 994 * pr_sosend methods have MPASS() to check that. This may change. 995 */ 996 static void 997 uipc_reset_kernel_mbuf(struct mbuf *m, struct mchain *mc) 998 { 999 1000 M_ASSERTPKTHDR(m); 1001 1002 m_clrprotoflags(m); 1003 m_tag_delete_chain(m, NULL); 1004 m->m_pkthdr.rcvif = NULL; 1005 m->m_pkthdr.flowid = 0; 1006 m->m_pkthdr.csum_flags = 0; 1007 m->m_pkthdr.fibnum = 0; 1008 m->m_pkthdr.rsstype = 0; 1009 1010 mc_init_m(mc, m); 1011 MPASS(m->m_pkthdr.len == mc->mc_len); 1012 } 1013 1014 #ifdef SOCKBUF_DEBUG 1015 static inline void 1016 uipc_stream_sbcheck(struct sockbuf *sb) 1017 { 1018 struct mbuf *d; 1019 u_int dacc, dccc, dctl, dmbcnt; 1020 bool notready = false; 1021 1022 dacc = dccc = dctl = dmbcnt = 0; 1023 STAILQ_FOREACH(d, &sb->uxst_mbq, m_stailq) { 1024 if (d == sb->uxst_fnrdy) { 1025 MPASS(d->m_flags & M_NOTREADY); 1026 notready = true; 1027 } 1028 if (d->m_type == MT_CONTROL) 1029 dctl += d->m_len; 1030 else if (d->m_type == MT_DATA) { 1031 dccc += d->m_len; 1032 if (!notready) 1033 dacc += d->m_len; 1034 } else 1035 MPASS(0); 1036 dmbcnt += MSIZE; 1037 if (d->m_flags & M_EXT) 1038 dmbcnt += d->m_ext.ext_size; 1039 if (d->m_stailq.stqe_next == NULL) 1040 MPASS(sb->uxst_mbq.stqh_last == &d->m_stailq.stqe_next); 1041 } 1042 MPASS(sb->uxst_fnrdy == NULL || notready); 1043 MPASS(dacc == sb->sb_acc); 1044 MPASS(dccc == sb->sb_ccc); 1045 MPASS(dctl == sb->sb_ctl); 1046 MPASS(dmbcnt == sb->sb_mbcnt); 1047 (void)STAILQ_EMPTY(&sb->uxst_mbq); 1048 } 1049 #define UIPC_STREAM_SBCHECK(sb) uipc_stream_sbcheck(sb) 1050 #else 1051 #define UIPC_STREAM_SBCHECK(sb) do {} while (0) 1052 #endif 1053 1054 /* 1055 * uipc_stream_sbspace() returns how much a writer can send, limited by char 1056 * count or mbuf memory use, whatever ends first. 1057 * 1058 * An obvious and legitimate reason for a socket having more data than allowed, 1059 * is lowering the limit with setsockopt(SO_RCVBUF) on already full buffer. 1060 * Also, sb_mbcnt may overcommit sb_mbmax in case if previous write observed 1061 * 'space < mbspace', but mchain allocated to hold 'space' bytes of data ended 1062 * up with 'mc_mlen > mbspace'. A typical scenario would be a full buffer with 1063 * writer trying to push in a large write, and a slow reader, that reads just 1064 * a few bytes at a time. In that case writer will keep creating new mbufs 1065 * with mc_split(). These mbufs will carry little chars, but will all point at 1066 * the same cluster, thus each adding cluster size to sb_mbcnt. This means we 1067 * will count same cluster many times potentially underutilizing socket buffer. 1068 * We aren't optimizing towards ineffective readers. Classic socket buffer had 1069 * the same "feature". 1070 */ 1071 static inline u_int 1072 uipc_stream_sbspace(struct sockbuf *sb) 1073 { 1074 u_int space, mbspace; 1075 1076 if (__predict_true(sb->sb_hiwat >= sb->sb_ccc + sb->sb_ctl)) 1077 space = sb->sb_hiwat - sb->sb_ccc - sb->sb_ctl; 1078 else 1079 return (0); 1080 if (__predict_true(sb->sb_mbmax >= sb->sb_mbcnt)) 1081 mbspace = sb->sb_mbmax - sb->sb_mbcnt; 1082 else 1083 return (0); 1084 1085 return (min(space, mbspace)); 1086 } 1087 1088 /* 1089 * UNIX version of generic sbwait() for writes. We wait on peer's receive 1090 * buffer, using our timeout. 1091 */ 1092 static int 1093 uipc_stream_sbwait(struct socket *so, sbintime_t timeo) 1094 { 1095 struct sockbuf *sb = &so->so_rcv; 1096 1097 SOCK_RECVBUF_LOCK_ASSERT(so); 1098 sb->sb_flags |= SB_WAIT; 1099 return (msleep_sbt(&sb->sb_acc, SOCK_RECVBUF_MTX(so), PSOCK | PCATCH, 1100 "sbwait", timeo, 0, 0)); 1101 } 1102 1103 static int 1104 uipc_sosend_stream_or_seqpacket(struct socket *so, struct sockaddr *addr, 1105 struct uio *uio0, struct mbuf *m, struct mbuf *c, int flags, 1106 struct thread *td) 1107 { 1108 struct unpcb *unp2; 1109 struct socket *so2; 1110 struct sockbuf *sb; 1111 struct uio *uio; 1112 struct mchain mc, cmc; 1113 size_t resid, sent; 1114 bool nonblock, eor, aio; 1115 int error; 1116 1117 MPASS((uio0 != NULL && m == NULL) || (m != NULL && uio0 == NULL)); 1118 MPASS(m == NULL || c == NULL); 1119 1120 if (__predict_false(flags & MSG_OOB)) 1121 return (EOPNOTSUPP); 1122 1123 nonblock = (so->so_state & SS_NBIO) || 1124 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1125 eor = flags & MSG_EOR; 1126 1127 mc = MCHAIN_INITIALIZER(&mc); 1128 cmc = MCHAIN_INITIALIZER(&cmc); 1129 sent = 0; 1130 aio = false; 1131 1132 if (m == NULL) { 1133 if (c != NULL && (error = unp_internalize(c, &cmc, td))) 1134 goto out; 1135 /* 1136 * This function may read more data from the uio than it would 1137 * then place on socket. That would leave uio inconsistent 1138 * upon return. Normally uio is allocated on the stack of the 1139 * syscall thread and we don't care about leaving it consistent. 1140 * However, aio(9) will allocate a uio as part of job and will 1141 * use it to track progress. We detect aio(9) checking the 1142 * SB_AIO_RUNNING flag. It is safe to check it without lock 1143 * cause it is set and cleared in the same taskqueue thread. 1144 * 1145 * This check can also produce a false positive: there is 1146 * aio(9) job and also there is a syscall we are serving now. 1147 * No sane software does that, it would leave to a mess in 1148 * the socket buffer, as aio(9) doesn't grab the I/O sx(9). 1149 * But syzkaller can create this mess. For such false positive 1150 * our goal is just don't panic or leak memory. 1151 */ 1152 if (__predict_false(so->so_snd.sb_flags & SB_AIO_RUNNING)) { 1153 uio = cloneuio(uio0); 1154 aio = true; 1155 } else { 1156 uio = uio0; 1157 resid = uio->uio_resid; 1158 } 1159 /* 1160 * Optimization for a case when our send fits into the receive 1161 * buffer - do the copyin before taking any locks, sized to our 1162 * send buffer. Later copyins will also take into account 1163 * space in the peer's receive buffer. 1164 */ 1165 error = mc_uiotomc(&mc, uio, so->so_snd.sb_hiwat, 0, M_WAITOK, 1166 eor ? M_EOR : 0); 1167 if (__predict_false(error)) 1168 goto out2; 1169 } else 1170 uipc_reset_kernel_mbuf(m, &mc); 1171 1172 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 1173 if (error) 1174 goto out2; 1175 1176 if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0)) 1177 goto out3; 1178 1179 if (unp2->unp_flags & UNP_WANTCRED_MASK) { 1180 /* 1181 * Credentials are passed only once on SOCK_STREAM and 1182 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or 1183 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). 1184 */ 1185 unp_addsockcred(td, &cmc, unp2->unp_flags); 1186 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; 1187 } 1188 1189 /* 1190 * Cycle through the data to send and available space in the peer's 1191 * receive buffer. Put a reference on the peer socket, so that it 1192 * doesn't get freed while we sbwait(). If peer goes away, we will 1193 * observe the SBS_CANTRCVMORE and our sorele() will finalize peer's 1194 * socket destruction. 1195 */ 1196 so2 = unp2->unp_socket; 1197 soref(so2); 1198 UNP_PCB_UNLOCK(unp2); 1199 sb = &so2->so_rcv; 1200 while (mc.mc_len + cmc.mc_len > 0) { 1201 struct mchain mcnext = MCHAIN_INITIALIZER(&mcnext); 1202 u_int space; 1203 1204 SOCK_RECVBUF_LOCK(so2); 1205 restart: 1206 UIPC_STREAM_SBCHECK(sb); 1207 if (__predict_false(cmc.mc_len > sb->sb_hiwat)) { 1208 SOCK_RECVBUF_UNLOCK(so2); 1209 error = EMSGSIZE; 1210 goto out4; 1211 } 1212 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) { 1213 SOCK_RECVBUF_UNLOCK(so2); 1214 error = EPIPE; 1215 goto out4; 1216 } 1217 /* 1218 * Wait on the peer socket receive buffer until we have enough 1219 * space to put at least control. The data is a stream and can 1220 * be put partially, but control is really a datagram. 1221 */ 1222 space = uipc_stream_sbspace(sb); 1223 if (space < sb->sb_lowat || space < cmc.mc_len) { 1224 if (nonblock) { 1225 if (aio) 1226 sb->uxst_flags |= UXST_PEER_AIO; 1227 SOCK_RECVBUF_UNLOCK(so2); 1228 if (aio) { 1229 SOCK_SENDBUF_LOCK(so); 1230 so->so_snd.sb_ccc = 1231 so->so_snd.sb_hiwat - space; 1232 SOCK_SENDBUF_UNLOCK(so); 1233 } 1234 error = EWOULDBLOCK; 1235 goto out4; 1236 } 1237 if ((error = uipc_stream_sbwait(so2, 1238 so->so_snd.sb_timeo)) != 0) { 1239 SOCK_RECVBUF_UNLOCK(so2); 1240 goto out4; 1241 } else 1242 goto restart; 1243 } 1244 MPASS(space >= cmc.mc_len); 1245 space -= cmc.mc_len; 1246 if (space == 0) { 1247 /* There is space only to send control. */ 1248 MPASS(!STAILQ_EMPTY(&cmc.mc_q)); 1249 mcnext = mc; 1250 mc = MCHAIN_INITIALIZER(&mc); 1251 } else if (space < mc.mc_len) { 1252 /* Not enough space. */ 1253 if (__predict_false(mc_split(&mc, &mcnext, space, 1254 M_NOWAIT) == ENOMEM)) { 1255 /* 1256 * If allocation failed use M_WAITOK and merge 1257 * the chain back. Next time mc_split() will 1258 * easily split at the same place. Only if we 1259 * race with setsockopt(SO_RCVBUF) shrinking 1260 * sb_hiwat can this happen more than once. 1261 */ 1262 SOCK_RECVBUF_UNLOCK(so2); 1263 (void)mc_split(&mc, &mcnext, space, M_WAITOK); 1264 mc_concat(&mc, &mcnext); 1265 SOCK_RECVBUF_LOCK(so2); 1266 goto restart; 1267 } 1268 MPASS(mc.mc_len == space); 1269 } 1270 if (!STAILQ_EMPTY(&cmc.mc_q)) { 1271 STAILQ_CONCAT(&sb->uxst_mbq, &cmc.mc_q); 1272 sb->sb_ctl += cmc.mc_len; 1273 sb->sb_mbcnt += cmc.mc_mlen; 1274 cmc.mc_len = 0; 1275 } 1276 sent += mc.mc_len; 1277 if (sb->uxst_fnrdy == NULL) 1278 sb->sb_acc += mc.mc_len; 1279 sb->sb_ccc += mc.mc_len; 1280 sb->sb_mbcnt += mc.mc_mlen; 1281 STAILQ_CONCAT(&sb->uxst_mbq, &mc.mc_q); 1282 UIPC_STREAM_SBCHECK(sb); 1283 space = uipc_stream_sbspace(sb); 1284 sorwakeup_locked(so2); 1285 if (!STAILQ_EMPTY(&mcnext.mc_q)) { 1286 /* 1287 * Such assignment is unsafe in general, but it is 1288 * safe with !STAILQ_EMPTY(&mcnext.mc_q). In C++ we 1289 * could reload = for STAILQs :) 1290 */ 1291 mc = mcnext; 1292 } else if (uio != NULL && uio->uio_resid > 0) { 1293 /* 1294 * Copyin sum of peer's receive buffer space and our 1295 * sb_hiwat, which is our virtual send buffer size. 1296 * See comment above unpst_sendspace declaration. 1297 * We are reading sb_hiwat locklessly, cause a) we 1298 * don't care about an application that does send(2) 1299 * and setsockopt(2) racing internally, and for an 1300 * application that does this in sequence we will see 1301 * the correct value cause sbsetopt() uses buffer lock 1302 * and we also have already acquired it at least once. 1303 */ 1304 error = mc_uiotomc(&mc, uio, space + 1305 atomic_load_int(&so->so_snd.sb_hiwat), 0, M_WAITOK, 1306 eor ? M_EOR : 0); 1307 if (__predict_false(error)) 1308 goto out4; 1309 } else 1310 mc = MCHAIN_INITIALIZER(&mc); 1311 } 1312 1313 MPASS(STAILQ_EMPTY(&mc.mc_q)); 1314 1315 td->td_ru.ru_msgsnd++; 1316 out4: 1317 sorele(so2); 1318 out3: 1319 SOCK_IO_SEND_UNLOCK(so); 1320 out2: 1321 if (aio) { 1322 freeuio(uio); 1323 uioadvance(uio0, sent); 1324 } else if (uio != NULL) 1325 uio->uio_resid = resid - sent; 1326 if (!mc_empty(&cmc)) 1327 unp_scan(mc_first(&cmc), unp_freerights); 1328 out: 1329 mc_freem(&mc); 1330 mc_freem(&cmc); 1331 1332 return (error); 1333 } 1334 1335 /* 1336 * Wakeup a writer, used by recv(2) and shutdown(2). 1337 * 1338 * @param so Points to a connected stream socket with receive buffer locked 1339 * 1340 * In a blocking mode peer is sleeping on our receive buffer, and we need just 1341 * wakeup(9) on it. But to wake up various event engines, we need to reach 1342 * over to peer's selinfo. This can be safely done as the socket buffer 1343 * receive lock is protecting us from the peer going away. 1344 */ 1345 static void 1346 uipc_wakeup_writer(struct socket *so) 1347 { 1348 struct sockbuf *sb = &so->so_rcv; 1349 struct selinfo *sel; 1350 1351 SOCK_RECVBUF_LOCK_ASSERT(so); 1352 MPASS(sb->uxst_peer != NULL); 1353 1354 sel = &sb->uxst_peer->so_wrsel; 1355 1356 if (sb->uxst_flags & UXST_PEER_SEL) { 1357 selwakeuppri(sel, PSOCK); 1358 /* 1359 * XXXGL: sowakeup() does SEL_WAITING() without locks. 1360 */ 1361 if (!SEL_WAITING(sel)) 1362 sb->uxst_flags &= ~UXST_PEER_SEL; 1363 } 1364 if (sb->sb_flags & SB_WAIT) { 1365 sb->sb_flags &= ~SB_WAIT; 1366 wakeup(&sb->sb_acc); 1367 } 1368 KNOTE_LOCKED(&sel->si_note, 0); 1369 SOCK_RECVBUF_UNLOCK(so); 1370 } 1371 1372 static void 1373 uipc_cantrcvmore(struct socket *so) 1374 { 1375 1376 SOCK_RECVBUF_LOCK(so); 1377 so->so_rcv.sb_state |= SBS_CANTRCVMORE; 1378 selwakeuppri(&so->so_rdsel, PSOCK); 1379 KNOTE_LOCKED(&so->so_rdsel.si_note, 0); 1380 if (so->so_rcv.uxst_peer != NULL) 1381 uipc_wakeup_writer(so); 1382 else 1383 SOCK_RECVBUF_UNLOCK(so); 1384 } 1385 1386 static int 1387 uipc_soreceive_stream_or_seqpacket(struct socket *so, struct sockaddr **psa, 1388 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 1389 { 1390 struct sockbuf *sb = &so->so_rcv; 1391 struct mbuf *control, *m, *first, *part, *next; 1392 u_int ctl, space, datalen, mbcnt, partlen; 1393 int error, flags; 1394 bool nonblock, waitall, peek; 1395 1396 MPASS(mp0 == NULL); 1397 1398 if (psa != NULL) 1399 *psa = NULL; 1400 if (controlp != NULL) 1401 *controlp = NULL; 1402 1403 flags = flagsp != NULL ? *flagsp : 0; 1404 nonblock = (so->so_state & SS_NBIO) || 1405 (flags & (MSG_DONTWAIT | MSG_NBIO)); 1406 peek = flags & MSG_PEEK; 1407 waitall = (flags & MSG_WAITALL) && !peek; 1408 1409 /* 1410 * This check may fail only on a socket that never went through 1411 * connect(2). We can check this locklessly, cause: a) for a new born 1412 * socket we don't care about applications that may race internally 1413 * between connect(2) and recv(2), and b) for a dying socket if we 1414 * miss update by unp_sosidisconnected(), we would still get the check 1415 * correct. For dying socket we would observe SBS_CANTRCVMORE later. 1416 */ 1417 if (__predict_false((atomic_load_short(&so->so_state) & 1418 (SS_ISCONNECTED|SS_ISDISCONNECTED)) == 0)) 1419 return (ENOTCONN); 1420 1421 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 1422 if (__predict_false(error)) 1423 return (error); 1424 1425 restart: 1426 SOCK_RECVBUF_LOCK(so); 1427 UIPC_STREAM_SBCHECK(sb); 1428 while (sb->sb_acc < sb->sb_lowat && 1429 (sb->sb_ctl == 0 || controlp == NULL)) { 1430 if (so->so_error) { 1431 error = so->so_error; 1432 if (!peek) 1433 so->so_error = 0; 1434 SOCK_RECVBUF_UNLOCK(so); 1435 SOCK_IO_RECV_UNLOCK(so); 1436 return (error); 1437 } 1438 if (sb->sb_state & SBS_CANTRCVMORE) { 1439 SOCK_RECVBUF_UNLOCK(so); 1440 SOCK_IO_RECV_UNLOCK(so); 1441 return (0); 1442 } 1443 if (nonblock) { 1444 SOCK_RECVBUF_UNLOCK(so); 1445 SOCK_IO_RECV_UNLOCK(so); 1446 return (EWOULDBLOCK); 1447 } 1448 error = sbwait(so, SO_RCV); 1449 if (error) { 1450 SOCK_RECVBUF_UNLOCK(so); 1451 SOCK_IO_RECV_UNLOCK(so); 1452 return (error); 1453 } 1454 } 1455 1456 MPASS(STAILQ_FIRST(&sb->uxst_mbq)); 1457 MPASS(sb->sb_acc > 0 || sb->sb_ctl > 0); 1458 1459 mbcnt = 0; 1460 ctl = 0; 1461 first = STAILQ_FIRST(&sb->uxst_mbq); 1462 if (first->m_type == MT_CONTROL) { 1463 control = first; 1464 STAILQ_FOREACH_FROM(first, &sb->uxst_mbq, m_stailq) { 1465 if (first->m_type != MT_CONTROL) 1466 break; 1467 ctl += first->m_len; 1468 mbcnt += MSIZE; 1469 if (first->m_flags & M_EXT) 1470 mbcnt += first->m_ext.ext_size; 1471 } 1472 } else 1473 control = NULL; 1474 1475 /* 1476 * Find split point for the next copyout. On exit from the loop, 1477 * 'next' points to the new head of the buffer STAILQ and 'datalen' 1478 * contains the amount of data we will copy out at the end. The 1479 * copyout is protected by the I/O lock only, as writers can only 1480 * append to the buffer. We need to record the socket buffer state 1481 * and do all length adjustments before dropping the socket buffer lock. 1482 */ 1483 for (space = uio->uio_resid, m = next = first, part = NULL, datalen = 0; 1484 space > 0 && m != sb->uxst_fnrdy && m->m_type == MT_DATA; 1485 m = STAILQ_NEXT(m, m_stailq)) { 1486 if (space >= m->m_len) { 1487 space -= m->m_len; 1488 datalen += m->m_len; 1489 mbcnt += MSIZE; 1490 if (m->m_flags & M_EXT) 1491 mbcnt += m->m_ext.ext_size; 1492 if (m->m_flags & M_EOR) { 1493 flags |= MSG_EOR; 1494 next = STAILQ_NEXT(m, m_stailq); 1495 break; 1496 } 1497 } else { 1498 datalen += space; 1499 partlen = space; 1500 if (!peek) { 1501 m->m_len -= partlen; 1502 m->m_data += partlen; 1503 } 1504 next = part = m; 1505 break; 1506 } 1507 next = STAILQ_NEXT(m, m_stailq); 1508 } 1509 1510 if (!peek) { 1511 if (next == NULL) 1512 STAILQ_INIT(&sb->uxst_mbq); 1513 else 1514 STAILQ_FIRST(&sb->uxst_mbq) = next; 1515 MPASS(sb->sb_acc >= datalen); 1516 sb->sb_acc -= datalen; 1517 sb->sb_ccc -= datalen; 1518 MPASS(sb->sb_ctl >= ctl); 1519 sb->sb_ctl -= ctl; 1520 MPASS(sb->sb_mbcnt >= mbcnt); 1521 sb->sb_mbcnt -= mbcnt; 1522 UIPC_STREAM_SBCHECK(sb); 1523 if (__predict_true(sb->uxst_peer != NULL)) { 1524 struct unpcb *unp2; 1525 bool aio; 1526 1527 if ((aio = sb->uxst_flags & UXST_PEER_AIO)) 1528 sb->uxst_flags &= ~UXST_PEER_AIO; 1529 1530 uipc_wakeup_writer(so); 1531 /* 1532 * XXXGL: need to go through uipc_lock_peer() after 1533 * the receive buffer lock dropped, it was protecting 1534 * us from unp_soisdisconnected(). The aio workarounds 1535 * should be refactored to the aio(4) side. 1536 */ 1537 if (aio && uipc_lock_peer(so, &unp2) == 0) { 1538 struct socket *so2 = unp2->unp_socket; 1539 1540 SOCK_SENDBUF_LOCK(so2); 1541 so2->so_snd.sb_ccc -= datalen; 1542 sowakeup_aio(so2, SO_SND); 1543 SOCK_SENDBUF_UNLOCK(so2); 1544 UNP_PCB_UNLOCK(unp2); 1545 } 1546 } else 1547 SOCK_RECVBUF_UNLOCK(so); 1548 } else 1549 SOCK_RECVBUF_UNLOCK(so); 1550 1551 while (control != NULL && control->m_type == MT_CONTROL) { 1552 if (!peek) { 1553 /* 1554 * unp_externalize() failure must abort entire read(2). 1555 * Such failure should also free the problematic 1556 * control, but link back the remaining data to the head 1557 * of the buffer, so that socket is not left in a state 1558 * where it can't progress forward with reading. 1559 * Probability of such a failure is really low, so it 1560 * is fine that we need to perform pretty complex 1561 * operation here to reconstruct the buffer. 1562 */ 1563 error = unp_externalize(control, controlp, flags); 1564 control = m_free(control); 1565 if (__predict_false(error && control != NULL)) { 1566 struct mchain cmc; 1567 1568 mc_init_m(&cmc, control); 1569 1570 SOCK_RECVBUF_LOCK(so); 1571 if (__predict_false( 1572 (sb->sb_state & SBS_CANTRCVMORE) || 1573 cmc.mc_len + sb->sb_ccc + sb->sb_ctl > 1574 sb->sb_hiwat)) { 1575 /* 1576 * While the lock was dropped and we 1577 * were failing in unp_externalize(), 1578 * the peer could has a) disconnected, 1579 * b) filled the buffer so that we 1580 * can't prepend data back. 1581 * These are two edge conditions that 1582 * we just can't handle, so lose the 1583 * data and return the error. 1584 */ 1585 SOCK_RECVBUF_UNLOCK(so); 1586 SOCK_IO_RECV_UNLOCK(so); 1587 unp_scan(mc_first(&cmc), 1588 unp_freerights); 1589 mc_freem(&cmc); 1590 return (error); 1591 } 1592 1593 UIPC_STREAM_SBCHECK(sb); 1594 /* XXXGL: STAILQ_PREPEND */ 1595 STAILQ_CONCAT(&cmc.mc_q, &sb->uxst_mbq); 1596 STAILQ_SWAP(&cmc.mc_q, &sb->uxst_mbq, mbuf); 1597 1598 sb->sb_ctl = sb->sb_acc = sb->sb_ccc = 1599 sb->sb_mbcnt = 0; 1600 STAILQ_FOREACH(m, &sb->uxst_mbq, m_stailq) { 1601 if (m->m_type == MT_DATA) { 1602 sb->sb_acc += m->m_len; 1603 sb->sb_ccc += m->m_len; 1604 } else { 1605 sb->sb_ctl += m->m_len; 1606 } 1607 sb->sb_mbcnt += MSIZE; 1608 if (m->m_flags & M_EXT) 1609 sb->sb_mbcnt += 1610 m->m_ext.ext_size; 1611 } 1612 UIPC_STREAM_SBCHECK(sb); 1613 SOCK_RECVBUF_UNLOCK(so); 1614 SOCK_IO_RECV_UNLOCK(so); 1615 return (error); 1616 } 1617 if (controlp != NULL) { 1618 while (*controlp != NULL) 1619 controlp = &(*controlp)->m_next; 1620 } 1621 } else { 1622 /* 1623 * XXXGL 1624 * 1625 * In MSG_PEEK case control is not externalized. This 1626 * means we are leaking some kernel pointers to the 1627 * userland. They are useless to a law-abiding 1628 * application, but may be useful to a malware. This 1629 * is what the historical implementation in the 1630 * soreceive_generic() did. To be improved? 1631 */ 1632 if (controlp != NULL) { 1633 *controlp = m_copym(control, 0, control->m_len, 1634 M_WAITOK); 1635 controlp = &(*controlp)->m_next; 1636 } 1637 control = STAILQ_NEXT(control, m_stailq); 1638 } 1639 } 1640 1641 for (m = first; datalen > 0; m = next) { 1642 void *data; 1643 u_int len; 1644 1645 next = STAILQ_NEXT(m, m_stailq); 1646 if (m == part) { 1647 data = peek ? 1648 mtod(m, char *) : mtod(m, char *) - partlen; 1649 len = partlen; 1650 } else { 1651 data = mtod(m, char *); 1652 len = m->m_len; 1653 } 1654 error = uiomove(data, len, uio); 1655 if (__predict_false(error)) { 1656 if (!peek) 1657 for (; m != part && datalen > 0; m = next) { 1658 next = STAILQ_NEXT(m, m_stailq); 1659 MPASS(datalen >= m->m_len); 1660 datalen -= m->m_len; 1661 m_free(m); 1662 } 1663 SOCK_IO_RECV_UNLOCK(so); 1664 return (error); 1665 } 1666 datalen -= len; 1667 if (!peek && m != part) 1668 m_free(m); 1669 } 1670 if (waitall && !(flags & MSG_EOR) && uio->uio_resid > 0) 1671 goto restart; 1672 SOCK_IO_RECV_UNLOCK(so); 1673 1674 if (flagsp != NULL) 1675 *flagsp |= flags; 1676 1677 uio->uio_td->td_ru.ru_msgrcv++; 1678 1679 return (0); 1680 } 1681 1682 static int 1683 uipc_sopoll_stream_or_seqpacket(struct socket *so, int events, 1684 struct thread *td) 1685 { 1686 struct unpcb *unp = sotounpcb(so); 1687 int revents; 1688 1689 UNP_PCB_LOCK(unp); 1690 if (SOLISTENING(so)) { 1691 /* The above check is safe, since conversion to listening uses 1692 * both protocol and socket lock. 1693 */ 1694 SOCK_LOCK(so); 1695 if (!(events & (POLLIN | POLLRDNORM))) 1696 revents = 0; 1697 else if (!TAILQ_EMPTY(&so->sol_comp)) 1698 revents = events & (POLLIN | POLLRDNORM); 1699 else if (so->so_error) 1700 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 1701 else { 1702 selrecord(td, &so->so_rdsel); 1703 revents = 0; 1704 } 1705 SOCK_UNLOCK(so); 1706 } else { 1707 if (so->so_state & SS_ISDISCONNECTED) 1708 revents = POLLHUP; 1709 else 1710 revents = 0; 1711 if (events & (POLLIN | POLLRDNORM | POLLRDHUP)) { 1712 SOCK_RECVBUF_LOCK(so); 1713 if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat || 1714 so->so_error || so->so_rerror) 1715 revents |= events & (POLLIN | POLLRDNORM); 1716 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 1717 revents |= events & 1718 (POLLIN | POLLRDNORM | POLLRDHUP); 1719 if (!(revents & (POLLIN | POLLRDNORM | POLLRDHUP))) { 1720 selrecord(td, &so->so_rdsel); 1721 so->so_rcv.sb_flags |= SB_SEL; 1722 } 1723 SOCK_RECVBUF_UNLOCK(so); 1724 } 1725 if (events & (POLLOUT | POLLWRNORM)) { 1726 struct socket *so2 = so->so_rcv.uxst_peer; 1727 1728 if (so2 != NULL) { 1729 struct sockbuf *sb = &so2->so_rcv; 1730 1731 SOCK_RECVBUF_LOCK(so2); 1732 if (uipc_stream_sbspace(sb) >= sb->sb_lowat) 1733 revents |= events & 1734 (POLLOUT | POLLWRNORM); 1735 if (sb->sb_state & SBS_CANTRCVMORE) 1736 revents |= POLLHUP; 1737 if (!(revents & (POLLOUT | POLLWRNORM))) { 1738 so2->so_rcv.uxst_flags |= UXST_PEER_SEL; 1739 selrecord(td, &so->so_wrsel); 1740 } 1741 SOCK_RECVBUF_UNLOCK(so2); 1742 } else 1743 selrecord(td, &so->so_wrsel); 1744 } 1745 } 1746 UNP_PCB_UNLOCK(unp); 1747 return (revents); 1748 } 1749 1750 static void 1751 uipc_wrknl_lock(void *arg) 1752 { 1753 struct socket *so = arg; 1754 struct unpcb *unp = sotounpcb(so); 1755 1756 retry: 1757 if (SOLISTENING(so)) { 1758 SOLISTEN_LOCK(so); 1759 } else { 1760 UNP_PCB_LOCK(unp); 1761 if (__predict_false(SOLISTENING(so))) { 1762 UNP_PCB_UNLOCK(unp); 1763 goto retry; 1764 } 1765 if (so->so_rcv.uxst_peer != NULL) 1766 SOCK_RECVBUF_LOCK(so->so_rcv.uxst_peer); 1767 } 1768 } 1769 1770 static void 1771 uipc_wrknl_unlock(void *arg) 1772 { 1773 struct socket *so = arg; 1774 struct unpcb *unp = sotounpcb(so); 1775 1776 if (SOLISTENING(so)) 1777 SOLISTEN_UNLOCK(so); 1778 else { 1779 if (so->so_rcv.uxst_peer != NULL) 1780 SOCK_RECVBUF_UNLOCK(so->so_rcv.uxst_peer); 1781 UNP_PCB_UNLOCK(unp); 1782 } 1783 } 1784 1785 static void 1786 uipc_wrknl_assert_lock(void *arg, int what) 1787 { 1788 struct socket *so = arg; 1789 1790 if (SOLISTENING(so)) { 1791 if (what == LA_LOCKED) 1792 SOLISTEN_LOCK_ASSERT(so); 1793 else 1794 SOLISTEN_UNLOCK_ASSERT(so); 1795 } else { 1796 /* 1797 * The pr_soreceive method will put a note without owning the 1798 * unp lock, so we can't assert it here. But we can safely 1799 * dereference uxst_peer pointer, since receive buffer lock 1800 * is assumed to be held here. 1801 */ 1802 if (what == LA_LOCKED && so->so_rcv.uxst_peer != NULL) 1803 SOCK_RECVBUF_LOCK_ASSERT(so->so_rcv.uxst_peer); 1804 } 1805 } 1806 1807 static void 1808 uipc_filt_sowdetach(struct knote *kn) 1809 { 1810 struct socket *so = kn->kn_fp->f_data; 1811 1812 uipc_wrknl_lock(so); 1813 knlist_remove(&so->so_wrsel.si_note, kn, 1); 1814 uipc_wrknl_unlock(so); 1815 } 1816 1817 static int 1818 uipc_filt_sowrite(struct knote *kn, long hint) 1819 { 1820 struct socket *so = kn->kn_fp->f_data, *so2; 1821 struct unpcb *unp = sotounpcb(so), *unp2 = unp->unp_conn; 1822 1823 if (SOLISTENING(so)) 1824 return (0); 1825 1826 if (unp2 == NULL) { 1827 if (so->so_state & SS_ISDISCONNECTED) { 1828 kn->kn_flags |= EV_EOF; 1829 kn->kn_fflags = so->so_error; 1830 return (1); 1831 } else 1832 return (0); 1833 } 1834 1835 so2 = unp2->unp_socket; 1836 SOCK_RECVBUF_LOCK_ASSERT(so2); 1837 kn->kn_data = uipc_stream_sbspace(&so2->so_rcv); 1838 1839 if (so2->so_rcv.sb_state & SBS_CANTRCVMORE) { 1840 kn->kn_flags |= EV_EOF; 1841 return (1); 1842 } else if (kn->kn_sfflags & NOTE_LOWAT) 1843 return (kn->kn_data >= kn->kn_sdata); 1844 else 1845 return (kn->kn_data >= so2->so_rcv.sb_lowat); 1846 } 1847 1848 static int 1849 uipc_filt_soempty(struct knote *kn, long hint) 1850 { 1851 struct socket *so = kn->kn_fp->f_data, *so2; 1852 struct unpcb *unp = sotounpcb(so), *unp2 = unp->unp_conn; 1853 1854 if (SOLISTENING(so) || unp2 == NULL) 1855 return (1); 1856 1857 so2 = unp2->unp_socket; 1858 SOCK_RECVBUF_LOCK_ASSERT(so2); 1859 kn->kn_data = uipc_stream_sbspace(&so2->so_rcv); 1860 1861 return (kn->kn_data == 0 ? 1 : 0); 1862 } 1863 1864 static const struct filterops uipc_write_filtops = { 1865 .f_isfd = 1, 1866 .f_detach = uipc_filt_sowdetach, 1867 .f_event = uipc_filt_sowrite, 1868 .f_copy = knote_triv_copy, 1869 }; 1870 static const struct filterops uipc_empty_filtops = { 1871 .f_isfd = 1, 1872 .f_detach = uipc_filt_sowdetach, 1873 .f_event = uipc_filt_soempty, 1874 .f_copy = knote_triv_copy, 1875 }; 1876 1877 static int 1878 uipc_kqfilter_stream_or_seqpacket(struct socket *so, struct knote *kn) 1879 { 1880 struct unpcb *unp = sotounpcb(so); 1881 struct knlist *knl; 1882 1883 switch (kn->kn_filter) { 1884 case EVFILT_READ: 1885 return (sokqfilter_generic(so, kn)); 1886 case EVFILT_WRITE: 1887 kn->kn_fop = &uipc_write_filtops; 1888 break; 1889 case EVFILT_EMPTY: 1890 kn->kn_fop = &uipc_empty_filtops; 1891 break; 1892 default: 1893 return (EINVAL); 1894 } 1895 1896 knl = &so->so_wrsel.si_note; 1897 UNP_PCB_LOCK(unp); 1898 if (SOLISTENING(so)) { 1899 SOLISTEN_LOCK(so); 1900 knlist_add(knl, kn, 1); 1901 SOLISTEN_UNLOCK(so); 1902 } else { 1903 struct socket *so2 = so->so_rcv.uxst_peer; 1904 1905 if (so2 != NULL) 1906 SOCK_RECVBUF_LOCK(so2); 1907 knlist_add(knl, kn, 1); 1908 if (so2 != NULL) 1909 SOCK_RECVBUF_UNLOCK(so2); 1910 } 1911 UNP_PCB_UNLOCK(unp); 1912 return (0); 1913 } 1914 1915 /* PF_UNIX/SOCK_DGRAM version of sbspace() */ 1916 static inline bool 1917 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt) 1918 { 1919 u_int bleft, mleft; 1920 1921 /* 1922 * Negative space may happen if send(2) is followed by 1923 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum. 1924 */ 1925 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc || 1926 sb->sb_mbmax < sb->uxdg_mbcnt)) 1927 return (false); 1928 1929 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) 1930 return (false); 1931 1932 bleft = sb->sb_hiwat - sb->uxdg_cc; 1933 mleft = sb->sb_mbmax - sb->uxdg_mbcnt; 1934 1935 return (bleft >= cc && mleft >= mbcnt); 1936 } 1937 1938 /* 1939 * PF_UNIX/SOCK_DGRAM send 1940 * 1941 * Allocate a record consisting of 3 mbufs in the sequence of 1942 * from -> control -> data and append it to the socket buffer. 1943 * 1944 * The first mbuf carries sender's name and is a pkthdr that stores 1945 * overall length of datagram, its memory consumption and control length. 1946 */ 1947 #define ctllen PH_loc.thirtytwo[1] 1948 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <= 1949 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen"); 1950 static int 1951 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 1952 struct mbuf *m, struct mbuf *c, int flags, struct thread *td) 1953 { 1954 struct unpcb *unp, *unp2; 1955 const struct sockaddr *from; 1956 struct socket *so2; 1957 struct sockbuf *sb; 1958 struct mchain cmc = MCHAIN_INITIALIZER(&cmc); 1959 struct mbuf *f; 1960 u_int cc, ctl, mbcnt; 1961 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused; 1962 int error; 1963 1964 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL)); 1965 1966 error = 0; 1967 f = NULL; 1968 1969 if (__predict_false(flags & MSG_OOB)) { 1970 error = EOPNOTSUPP; 1971 goto out; 1972 } 1973 if (m == NULL) { 1974 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) { 1975 error = EMSGSIZE; 1976 goto out; 1977 } 1978 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR); 1979 if (__predict_false(m == NULL)) { 1980 error = EFAULT; 1981 goto out; 1982 } 1983 f = m_gethdr(M_WAITOK, MT_SONAME); 1984 cc = m->m_pkthdr.len; 1985 mbcnt = MSIZE + m->m_pkthdr.memlen; 1986 if (c != NULL && (error = unp_internalize(c, &cmc, td))) 1987 goto out; 1988 } else { 1989 struct mchain mc; 1990 1991 uipc_reset_kernel_mbuf(m, &mc); 1992 cc = mc.mc_len; 1993 mbcnt = mc.mc_mlen; 1994 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) { 1995 error = EMSGSIZE; 1996 goto out; 1997 } 1998 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) { 1999 error = ENOBUFS; 2000 goto out; 2001 } 2002 } 2003 2004 unp = sotounpcb(so); 2005 MPASS(unp); 2006 2007 /* 2008 * XXXGL: would be cool to fully remove so_snd out of the equation 2009 * and avoid this lock, which is not only extraneous, but also being 2010 * released, thus still leaving possibility for a race. We can easily 2011 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it 2012 * is more difficult to invent something to handle so_error. 2013 */ 2014 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 2015 if (error) 2016 goto out2; 2017 SOCK_SENDBUF_LOCK(so); 2018 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 2019 SOCK_SENDBUF_UNLOCK(so); 2020 error = EPIPE; 2021 goto out3; 2022 } 2023 if (so->so_error != 0) { 2024 error = so->so_error; 2025 so->so_error = 0; 2026 SOCK_SENDBUF_UNLOCK(so); 2027 goto out3; 2028 } 2029 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) { 2030 SOCK_SENDBUF_UNLOCK(so); 2031 error = EDESTADDRREQ; 2032 goto out3; 2033 } 2034 SOCK_SENDBUF_UNLOCK(so); 2035 2036 if (addr != NULL) { 2037 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true))) 2038 goto out3; 2039 UNP_PCB_LOCK_ASSERT(unp); 2040 unp2 = unp->unp_conn; 2041 UNP_PCB_LOCK_ASSERT(unp2); 2042 } else { 2043 UNP_PCB_LOCK(unp); 2044 unp2 = unp_pcb_lock_peer(unp); 2045 if (unp2 == NULL) { 2046 UNP_PCB_UNLOCK(unp); 2047 error = ENOTCONN; 2048 goto out3; 2049 } 2050 } 2051 2052 if (unp2->unp_flags & UNP_WANTCRED_MASK) 2053 unp_addsockcred(td, &cmc, unp2->unp_flags); 2054 if (unp->unp_addr != NULL) 2055 from = (struct sockaddr *)unp->unp_addr; 2056 else 2057 from = &sun_noname; 2058 f->m_len = from->sa_len; 2059 MPASS(from->sa_len <= MLEN); 2060 bcopy(from, mtod(f, void *), from->sa_len); 2061 2062 /* 2063 * Concatenate mbufs: from -> control -> data. 2064 * Save overall cc and mbcnt in "from" mbuf. 2065 */ 2066 if (!STAILQ_EMPTY(&cmc.mc_q)) { 2067 f->m_next = mc_first(&cmc); 2068 mc_last(&cmc)->m_next = m; 2069 /* XXXGL: This is dirty as well as rollback after ENOBUFS. */ 2070 STAILQ_INIT(&cmc.mc_q); 2071 } else 2072 f->m_next = m; 2073 m = NULL; 2074 ctl = f->m_len + cmc.mc_len; 2075 mbcnt += cmc.mc_mlen; 2076 #ifdef INVARIANTS 2077 dcc = dctl = dmbcnt = 0; 2078 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) { 2079 if (mb->m_type == MT_DATA) 2080 dcc += mb->m_len; 2081 else 2082 dctl += mb->m_len; 2083 dmbcnt += MSIZE; 2084 if (mb->m_flags & M_EXT) 2085 dmbcnt += mb->m_ext.ext_size; 2086 } 2087 MPASS(dcc == cc); 2088 MPASS(dctl == ctl); 2089 MPASS(dmbcnt == mbcnt); 2090 #endif 2091 f->m_pkthdr.len = cc + ctl; 2092 f->m_pkthdr.memlen = mbcnt; 2093 f->m_pkthdr.ctllen = ctl; 2094 2095 /* 2096 * Destination socket buffer selection. 2097 * 2098 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the 2099 * destination address is supplied, create a temporary connection for 2100 * the run time of the function (see call to unp_connectat() above and 2101 * to unp_disconnect() below). We distinguish them by condition of 2102 * (addr != NULL). We intentionally avoid adding 'bool connected' for 2103 * that condition, since, again, through the run time of this code we 2104 * are always connected. For such "unconnected" sends, the destination 2105 * buffer would be the receive buffer of destination socket so2. 2106 * 2107 * For connected sends, data lands on the send buffer of the sender's 2108 * socket "so". Then, if we just added the very first datagram 2109 * on this send buffer, we need to add the send buffer on to the 2110 * receiving socket's buffer list. We put ourselves on top of the 2111 * list. Such logic gives infrequent senders priority over frequent 2112 * senders. 2113 * 2114 * Note on byte count management. As long as event methods kevent(2), 2115 * select(2) are not protocol specific (yet), we need to maintain 2116 * meaningful values on the receive buffer. So, the receive buffer 2117 * would accumulate counters from all connected buffers potentially 2118 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax. 2119 */ 2120 so2 = unp2->unp_socket; 2121 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv; 2122 SOCK_RECVBUF_LOCK(so2); 2123 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) { 2124 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb)) 2125 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd, 2126 uxdg_clist); 2127 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt); 2128 sb->uxdg_cc += cc + ctl; 2129 sb->uxdg_ctl += ctl; 2130 sb->uxdg_mbcnt += mbcnt; 2131 so2->so_rcv.sb_acc += cc + ctl; 2132 so2->so_rcv.sb_ccc += cc + ctl; 2133 so2->so_rcv.sb_ctl += ctl; 2134 so2->so_rcv.sb_mbcnt += mbcnt; 2135 sorwakeup_locked(so2); 2136 f = NULL; 2137 } else { 2138 soroverflow_locked(so2); 2139 error = ENOBUFS; 2140 if (f->m_next->m_type == MT_CONTROL) { 2141 STAILQ_FIRST(&cmc.mc_q) = f->m_next; 2142 f->m_next = NULL; 2143 } 2144 } 2145 2146 if (addr != NULL) 2147 unp_disconnect(unp, unp2); 2148 else 2149 unp_pcb_unlock_pair(unp, unp2); 2150 2151 td->td_ru.ru_msgsnd++; 2152 2153 out3: 2154 SOCK_IO_SEND_UNLOCK(so); 2155 out2: 2156 if (!mc_empty(&cmc)) 2157 unp_scan(mc_first(&cmc), unp_freerights); 2158 out: 2159 if (f) 2160 m_freem(f); 2161 mc_freem(&cmc); 2162 if (m) 2163 m_freem(m); 2164 2165 return (error); 2166 } 2167 2168 /* 2169 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK. 2170 * The mbuf has already been unlinked from the uxdg_mb of socket buffer 2171 * and needs to be linked onto uxdg_peeked of receive socket buffer. 2172 */ 2173 static int 2174 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa, 2175 struct uio *uio, struct mbuf **controlp, int *flagsp) 2176 { 2177 ssize_t len = 0; 2178 int error; 2179 2180 so->so_rcv.uxdg_peeked = m; 2181 so->so_rcv.uxdg_cc += m->m_pkthdr.len; 2182 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen; 2183 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen; 2184 SOCK_RECVBUF_UNLOCK(so); 2185 2186 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 2187 if (psa != NULL) 2188 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 2189 2190 m = m->m_next; 2191 KASSERT(m, ("%s: no data or control after soname", __func__)); 2192 2193 /* 2194 * With MSG_PEEK the control isn't executed, just copied. 2195 */ 2196 while (m != NULL && m->m_type == MT_CONTROL) { 2197 if (controlp != NULL) { 2198 *controlp = m_copym(m, 0, m->m_len, M_WAITOK); 2199 controlp = &(*controlp)->m_next; 2200 } 2201 m = m->m_next; 2202 } 2203 KASSERT(m == NULL || m->m_type == MT_DATA, 2204 ("%s: not MT_DATA mbuf %p", __func__, m)); 2205 while (m != NULL && uio->uio_resid > 0) { 2206 len = uio->uio_resid; 2207 if (len > m->m_len) 2208 len = m->m_len; 2209 error = uiomove(mtod(m, char *), (int)len, uio); 2210 if (error) { 2211 SOCK_IO_RECV_UNLOCK(so); 2212 return (error); 2213 } 2214 if (len == m->m_len) 2215 m = m->m_next; 2216 } 2217 SOCK_IO_RECV_UNLOCK(so); 2218 2219 if (flagsp != NULL) { 2220 if (m != NULL) { 2221 if (*flagsp & MSG_TRUNC) { 2222 /* Report real length of the packet */ 2223 uio->uio_resid -= m_length(m, NULL) - len; 2224 } 2225 *flagsp |= MSG_TRUNC; 2226 } else 2227 *flagsp &= ~MSG_TRUNC; 2228 } 2229 2230 return (0); 2231 } 2232 2233 /* 2234 * PF_UNIX/SOCK_DGRAM receive 2235 */ 2236 static int 2237 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 2238 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 2239 { 2240 struct sockbuf *sb = NULL; 2241 struct mbuf *m; 2242 int flags, error; 2243 ssize_t len = 0; 2244 bool nonblock; 2245 2246 MPASS(mp0 == NULL); 2247 2248 if (psa != NULL) 2249 *psa = NULL; 2250 if (controlp != NULL) 2251 *controlp = NULL; 2252 2253 flags = flagsp != NULL ? *flagsp : 0; 2254 nonblock = (so->so_state & SS_NBIO) || 2255 (flags & (MSG_DONTWAIT | MSG_NBIO)); 2256 2257 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 2258 if (__predict_false(error)) 2259 return (error); 2260 2261 /* 2262 * Loop blocking while waiting for a datagram. Prioritize connected 2263 * peers over unconnected sends. Set sb to selected socket buffer 2264 * containing an mbuf on exit from the wait loop. A datagram that 2265 * had already been peeked at has top priority. 2266 */ 2267 SOCK_RECVBUF_LOCK(so); 2268 while ((m = so->so_rcv.uxdg_peeked) == NULL && 2269 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL && 2270 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) { 2271 if (so->so_error) { 2272 error = so->so_error; 2273 if (!(flags & MSG_PEEK)) 2274 so->so_error = 0; 2275 SOCK_RECVBUF_UNLOCK(so); 2276 SOCK_IO_RECV_UNLOCK(so); 2277 return (error); 2278 } 2279 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 2280 uio->uio_resid == 0) { 2281 SOCK_RECVBUF_UNLOCK(so); 2282 SOCK_IO_RECV_UNLOCK(so); 2283 return (0); 2284 } 2285 if (nonblock) { 2286 SOCK_RECVBUF_UNLOCK(so); 2287 SOCK_IO_RECV_UNLOCK(so); 2288 return (EWOULDBLOCK); 2289 } 2290 error = sbwait(so, SO_RCV); 2291 if (error) { 2292 SOCK_RECVBUF_UNLOCK(so); 2293 SOCK_IO_RECV_UNLOCK(so); 2294 return (error); 2295 } 2296 } 2297 2298 if (sb == NULL) 2299 sb = &so->so_rcv; 2300 else if (m == NULL) 2301 m = STAILQ_FIRST(&sb->uxdg_mb); 2302 else 2303 MPASS(m == so->so_rcv.uxdg_peeked); 2304 2305 MPASS(sb->uxdg_cc > 0); 2306 M_ASSERTPKTHDR(m); 2307 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); 2308 2309 if (uio->uio_td) 2310 uio->uio_td->td_ru.ru_msgrcv++; 2311 2312 if (__predict_true(m != so->so_rcv.uxdg_peeked)) { 2313 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt); 2314 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv) 2315 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 2316 } else 2317 so->so_rcv.uxdg_peeked = NULL; 2318 2319 sb->uxdg_cc -= m->m_pkthdr.len; 2320 sb->uxdg_ctl -= m->m_pkthdr.ctllen; 2321 sb->uxdg_mbcnt -= m->m_pkthdr.memlen; 2322 2323 if (__predict_false(flags & MSG_PEEK)) 2324 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp)); 2325 2326 so->so_rcv.sb_acc -= m->m_pkthdr.len; 2327 so->so_rcv.sb_ccc -= m->m_pkthdr.len; 2328 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen; 2329 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen; 2330 SOCK_RECVBUF_UNLOCK(so); 2331 2332 if (psa != NULL) 2333 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK); 2334 m = m_free(m); 2335 KASSERT(m, ("%s: no data or control after soname", __func__)); 2336 2337 /* 2338 * Packet to copyout() is now in 'm' and it is disconnected from the 2339 * queue. 2340 * 2341 * Process one or more MT_CONTROL mbufs present before any data mbufs 2342 * in the first mbuf chain on the socket buffer. We call into the 2343 * unp_externalize() to perform externalization (or freeing if 2344 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs 2345 * without MT_DATA mbufs. 2346 */ 2347 while (m != NULL && m->m_type == MT_CONTROL) { 2348 error = unp_externalize(m, controlp, flags); 2349 m = m_free(m); 2350 if (error != 0) { 2351 SOCK_IO_RECV_UNLOCK(so); 2352 unp_scan(m, unp_freerights); 2353 m_freem(m); 2354 return (error); 2355 } 2356 if (controlp != NULL) { 2357 while (*controlp != NULL) 2358 controlp = &(*controlp)->m_next; 2359 } 2360 } 2361 KASSERT(m == NULL || m->m_type == MT_DATA, 2362 ("%s: not MT_DATA mbuf %p", __func__, m)); 2363 while (m != NULL && uio->uio_resid > 0) { 2364 len = uio->uio_resid; 2365 if (len > m->m_len) 2366 len = m->m_len; 2367 error = uiomove(mtod(m, char *), (int)len, uio); 2368 if (error) { 2369 SOCK_IO_RECV_UNLOCK(so); 2370 m_freem(m); 2371 return (error); 2372 } 2373 if (len == m->m_len) 2374 m = m_free(m); 2375 else { 2376 m->m_data += len; 2377 m->m_len -= len; 2378 } 2379 } 2380 SOCK_IO_RECV_UNLOCK(so); 2381 2382 if (m != NULL) { 2383 if (flagsp != NULL) { 2384 if (flags & MSG_TRUNC) { 2385 /* Report real length of the packet */ 2386 uio->uio_resid -= m_length(m, NULL); 2387 } 2388 *flagsp |= MSG_TRUNC; 2389 } 2390 m_freem(m); 2391 } else if (flagsp != NULL) 2392 *flagsp &= ~MSG_TRUNC; 2393 2394 return (0); 2395 } 2396 2397 static int 2398 uipc_sendfile_wait(struct socket *so, off_t need, int *space) 2399 { 2400 struct unpcb *unp2; 2401 struct socket *so2; 2402 struct sockbuf *sb; 2403 bool nonblock, sockref; 2404 int error; 2405 2406 MPASS(so->so_type == SOCK_STREAM); 2407 MPASS(need > 0); 2408 MPASS(space != NULL); 2409 2410 nonblock = so->so_state & SS_NBIO; 2411 sockref = false; 2412 2413 if (__predict_false((so->so_state & SS_ISCONNECTED) == 0)) 2414 return (ENOTCONN); 2415 2416 if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0)) 2417 return (error); 2418 2419 so2 = unp2->unp_socket; 2420 sb = &so2->so_rcv; 2421 SOCK_RECVBUF_LOCK(so2); 2422 UNP_PCB_UNLOCK(unp2); 2423 while ((*space = uipc_stream_sbspace(sb)) < need && 2424 (*space < so->so_snd.sb_hiwat / 2)) { 2425 UIPC_STREAM_SBCHECK(sb); 2426 if (nonblock) { 2427 SOCK_RECVBUF_UNLOCK(so2); 2428 return (EAGAIN); 2429 } 2430 if (!sockref) 2431 soref(so2); 2432 error = uipc_stream_sbwait(so2, so->so_snd.sb_timeo); 2433 if (error == 0 && 2434 __predict_false(sb->sb_state & SBS_CANTRCVMORE)) 2435 error = EPIPE; 2436 if (error) { 2437 SOCK_RECVBUF_UNLOCK(so2); 2438 sorele(so2); 2439 return (error); 2440 } 2441 } 2442 UIPC_STREAM_SBCHECK(sb); 2443 SOCK_RECVBUF_UNLOCK(so2); 2444 if (sockref) 2445 sorele(so2); 2446 2447 return (0); 2448 } 2449 2450 /* 2451 * Although this is a pr_send method, for unix(4) it is called only via 2452 * sendfile(2) path. This means we can be sure that mbufs are clear of 2453 * any extra flags and don't require any conditioning. 2454 */ 2455 static int 2456 uipc_sendfile(struct socket *so, int flags, struct mbuf *m, 2457 struct sockaddr *from, struct mbuf *control, struct thread *td) 2458 { 2459 struct mchain mc; 2460 struct unpcb *unp2; 2461 struct socket *so2; 2462 struct sockbuf *sb; 2463 bool notready, wakeup; 2464 int error; 2465 2466 MPASS(so->so_type == SOCK_STREAM); 2467 MPASS(from == NULL && control == NULL); 2468 KASSERT(!(m->m_flags & M_EXTPG), 2469 ("unix(4): TLS sendfile(2) not supported")); 2470 2471 notready = flags & PRUS_NOTREADY; 2472 2473 if (__predict_false((so->so_state & SS_ISCONNECTED) == 0)) { 2474 error = ENOTCONN; 2475 goto out; 2476 } 2477 2478 if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0)) 2479 goto out; 2480 2481 mc_init_m(&mc, m); 2482 2483 so2 = unp2->unp_socket; 2484 sb = &so2->so_rcv; 2485 SOCK_RECVBUF_LOCK(so2); 2486 UNP_PCB_UNLOCK(unp2); 2487 UIPC_STREAM_SBCHECK(sb); 2488 sb->sb_ccc += mc.mc_len; 2489 sb->sb_mbcnt += mc.mc_mlen; 2490 if (sb->uxst_fnrdy == NULL) { 2491 if (notready) { 2492 wakeup = false; 2493 STAILQ_FOREACH(m, &mc.mc_q, m_stailq) { 2494 if (m->m_flags & M_NOTREADY) { 2495 sb->uxst_fnrdy = m; 2496 break; 2497 } else { 2498 sb->sb_acc += m->m_len; 2499 wakeup = true; 2500 } 2501 } 2502 } else { 2503 wakeup = true; 2504 sb->sb_acc += mc.mc_len; 2505 } 2506 } else { 2507 wakeup = false; 2508 } 2509 STAILQ_CONCAT(&sb->uxst_mbq, &mc.mc_q); 2510 UIPC_STREAM_SBCHECK(sb); 2511 if (wakeup) 2512 sorwakeup_locked(so2); 2513 else 2514 SOCK_RECVBUF_UNLOCK(so2); 2515 2516 return (0); 2517 out: 2518 /* 2519 * In case of not ready data, uipc_ready() is responsible 2520 * for freeing memory. 2521 */ 2522 if (m != NULL && !notready) 2523 m_freem(m); 2524 2525 return (error); 2526 } 2527 2528 static int 2529 uipc_sbready(struct sockbuf *sb, struct mbuf *m, int count) 2530 { 2531 bool blocker; 2532 2533 /* assert locked */ 2534 2535 blocker = (sb->uxst_fnrdy == m); 2536 STAILQ_FOREACH_FROM(m, &sb->uxst_mbq, m_stailq) { 2537 if (count > 0) { 2538 MPASS(m->m_flags & M_NOTREADY); 2539 m->m_flags &= ~M_NOTREADY; 2540 if (blocker) 2541 sb->sb_acc += m->m_len; 2542 count--; 2543 } else if (m->m_flags & M_NOTREADY) 2544 break; 2545 else if (blocker) 2546 sb->sb_acc += m->m_len; 2547 } 2548 if (blocker) { 2549 sb->uxst_fnrdy = m; 2550 return (0); 2551 } else 2552 return (EINPROGRESS); 2553 } 2554 2555 static bool 2556 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) 2557 { 2558 struct mbuf *mb; 2559 struct sockbuf *sb; 2560 2561 SOCK_LOCK(so); 2562 if (SOLISTENING(so)) { 2563 SOCK_UNLOCK(so); 2564 return (false); 2565 } 2566 mb = NULL; 2567 sb = &so->so_rcv; 2568 SOCK_RECVBUF_LOCK(so); 2569 if (sb->uxst_fnrdy != NULL) { 2570 STAILQ_FOREACH(mb, &sb->uxst_mbq, m_stailq) { 2571 if (mb == m) { 2572 *errorp = uipc_sbready(sb, m, count); 2573 break; 2574 } 2575 } 2576 } 2577 SOCK_RECVBUF_UNLOCK(so); 2578 SOCK_UNLOCK(so); 2579 return (mb != NULL); 2580 } 2581 2582 static int 2583 uipc_ready(struct socket *so, struct mbuf *m, int count) 2584 { 2585 struct unpcb *unp, *unp2; 2586 int error; 2587 2588 MPASS(so->so_type == SOCK_STREAM); 2589 2590 if (__predict_true(uipc_lock_peer(so, &unp2) == 0)) { 2591 struct socket *so2; 2592 struct sockbuf *sb; 2593 2594 so2 = unp2->unp_socket; 2595 sb = &so2->so_rcv; 2596 SOCK_RECVBUF_LOCK(so2); 2597 UNP_PCB_UNLOCK(unp2); 2598 UIPC_STREAM_SBCHECK(sb); 2599 error = uipc_sbready(sb, m, count); 2600 UIPC_STREAM_SBCHECK(sb); 2601 if (error == 0) 2602 sorwakeup_locked(so2); 2603 else 2604 SOCK_RECVBUF_UNLOCK(so2); 2605 } else { 2606 /* 2607 * The receiving socket has been disconnected, but may still 2608 * be valid. In this case, the not-ready mbufs are still 2609 * present in its socket buffer, so perform an exhaustive 2610 * search before giving up and freeing the mbufs. 2611 */ 2612 UNP_LINK_RLOCK(); 2613 LIST_FOREACH(unp, &unp_shead, unp_link) { 2614 if (uipc_ready_scan(unp->unp_socket, m, count, &error)) 2615 break; 2616 } 2617 UNP_LINK_RUNLOCK(); 2618 2619 if (unp == NULL) { 2620 for (int i = 0; i < count; i++) 2621 m = m_free(m); 2622 return (ECONNRESET); 2623 } 2624 } 2625 return (error); 2626 } 2627 2628 static int 2629 uipc_sense(struct socket *so, struct stat *sb) 2630 { 2631 struct unpcb *unp; 2632 2633 unp = sotounpcb(so); 2634 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 2635 2636 sb->st_blksize = so->so_snd.sb_hiwat; 2637 sb->st_dev = NODEV; 2638 sb->st_ino = unp->unp_ino; 2639 return (0); 2640 } 2641 2642 static int 2643 uipc_shutdown(struct socket *so, enum shutdown_how how) 2644 { 2645 struct unpcb *unp = sotounpcb(so); 2646 int error; 2647 2648 SOCK_LOCK(so); 2649 if (SOLISTENING(so)) { 2650 if (how != SHUT_WR) { 2651 so->so_error = ECONNABORTED; 2652 solisten_wakeup(so); /* unlocks so */ 2653 } else 2654 SOCK_UNLOCK(so); 2655 return (ENOTCONN); 2656 } else if ((so->so_state & 2657 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) { 2658 /* 2659 * POSIX mandates us to just return ENOTCONN when shutdown(2) is 2660 * invoked on a datagram sockets, however historically we would 2661 * actually tear socket down. This is known to be leveraged by 2662 * some applications to unblock process waiting in recv(2) by 2663 * other process that it shares that socket with. Try to meet 2664 * both backward-compatibility and POSIX requirements by forcing 2665 * ENOTCONN but still flushing buffers and performing wakeup(9). 2666 * 2667 * XXXGL: it remains unknown what applications expect this 2668 * behavior and is this isolated to unix/dgram or inet/dgram or 2669 * both. See: D10351, D3039. 2670 */ 2671 error = ENOTCONN; 2672 if (so->so_type != SOCK_DGRAM) { 2673 SOCK_UNLOCK(so); 2674 return (error); 2675 } 2676 } else 2677 error = 0; 2678 SOCK_UNLOCK(so); 2679 2680 switch (how) { 2681 case SHUT_RD: 2682 if (so->so_type == SOCK_DGRAM) 2683 socantrcvmore(so); 2684 else 2685 uipc_cantrcvmore(so); 2686 unp_dispose(so); 2687 break; 2688 case SHUT_RDWR: 2689 if (so->so_type == SOCK_DGRAM) 2690 socantrcvmore(so); 2691 else 2692 uipc_cantrcvmore(so); 2693 unp_dispose(so); 2694 /* FALLTHROUGH */ 2695 case SHUT_WR: 2696 if (so->so_type == SOCK_DGRAM) { 2697 socantsendmore(so); 2698 } else { 2699 UNP_PCB_LOCK(unp); 2700 if (unp->unp_conn != NULL) 2701 uipc_cantrcvmore(unp->unp_conn->unp_socket); 2702 UNP_PCB_UNLOCK(unp); 2703 } 2704 } 2705 wakeup(&so->so_timeo); 2706 2707 return (error); 2708 } 2709 2710 static int 2711 uipc_sockaddr(struct socket *so, struct sockaddr *ret) 2712 { 2713 struct unpcb *unp; 2714 const struct sockaddr *sa; 2715 2716 unp = sotounpcb(so); 2717 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 2718 2719 UNP_PCB_LOCK(unp); 2720 if (unp->unp_addr != NULL) 2721 sa = (struct sockaddr *) unp->unp_addr; 2722 else 2723 sa = &sun_noname; 2724 bcopy(sa, ret, sa->sa_len); 2725 UNP_PCB_UNLOCK(unp); 2726 return (0); 2727 } 2728 2729 static int 2730 uipc_ctloutput(struct socket *so, struct sockopt *sopt) 2731 { 2732 struct unpcb *unp; 2733 struct xucred xu; 2734 int error, optval; 2735 2736 if (sopt->sopt_level != SOL_LOCAL) 2737 return (EINVAL); 2738 2739 unp = sotounpcb(so); 2740 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 2741 error = 0; 2742 switch (sopt->sopt_dir) { 2743 case SOPT_GET: 2744 switch (sopt->sopt_name) { 2745 case LOCAL_PEERCRED: 2746 UNP_PCB_LOCK(unp); 2747 if (unp->unp_flags & UNP_HAVEPC) 2748 xu = unp->unp_peercred; 2749 else { 2750 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 2751 error = ENOTCONN; 2752 else 2753 error = EINVAL; 2754 } 2755 UNP_PCB_UNLOCK(unp); 2756 if (error == 0) 2757 error = sooptcopyout(sopt, &xu, sizeof(xu)); 2758 break; 2759 2760 case LOCAL_CREDS: 2761 /* Unlocked read. */ 2762 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; 2763 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2764 break; 2765 2766 case LOCAL_CREDS_PERSISTENT: 2767 /* Unlocked read. */ 2768 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; 2769 error = sooptcopyout(sopt, &optval, sizeof(optval)); 2770 break; 2771 2772 default: 2773 error = EOPNOTSUPP; 2774 break; 2775 } 2776 break; 2777 2778 case SOPT_SET: 2779 switch (sopt->sopt_name) { 2780 case LOCAL_CREDS: 2781 case LOCAL_CREDS_PERSISTENT: 2782 error = sooptcopyin(sopt, &optval, sizeof(optval), 2783 sizeof(optval)); 2784 if (error) 2785 break; 2786 2787 #define OPTSET(bit, exclusive) do { \ 2788 UNP_PCB_LOCK(unp); \ 2789 if (optval) { \ 2790 if ((unp->unp_flags & (exclusive)) != 0) { \ 2791 UNP_PCB_UNLOCK(unp); \ 2792 error = EINVAL; \ 2793 break; \ 2794 } \ 2795 unp->unp_flags |= (bit); \ 2796 } else \ 2797 unp->unp_flags &= ~(bit); \ 2798 UNP_PCB_UNLOCK(unp); \ 2799 } while (0) 2800 2801 switch (sopt->sopt_name) { 2802 case LOCAL_CREDS: 2803 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); 2804 break; 2805 2806 case LOCAL_CREDS_PERSISTENT: 2807 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); 2808 break; 2809 2810 default: 2811 break; 2812 } 2813 break; 2814 #undef OPTSET 2815 default: 2816 error = ENOPROTOOPT; 2817 break; 2818 } 2819 break; 2820 2821 default: 2822 error = EOPNOTSUPP; 2823 break; 2824 } 2825 return (error); 2826 } 2827 2828 static int 2829 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 2830 { 2831 2832 return (unp_connectat(AT_FDCWD, so, nam, td, false)); 2833 } 2834 2835 static int 2836 unp_connectat(int fd, struct socket *so, struct sockaddr *nam, 2837 struct thread *td, bool return_locked) 2838 { 2839 struct mtx *vplock; 2840 struct sockaddr_un *soun; 2841 struct vnode *vp; 2842 struct socket *so2; 2843 struct unpcb *unp, *unp2, *unp3; 2844 struct nameidata nd; 2845 char buf[SOCK_MAXADDRLEN]; 2846 struct sockaddr *sa; 2847 cap_rights_t rights; 2848 int error, len; 2849 bool connreq; 2850 2851 CURVNET_ASSERT_SET(); 2852 2853 if (nam->sa_family != AF_UNIX) 2854 return (EAFNOSUPPORT); 2855 if (nam->sa_len > sizeof(struct sockaddr_un)) 2856 return (EINVAL); 2857 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 2858 if (len <= 0) 2859 return (EINVAL); 2860 soun = (struct sockaddr_un *)nam; 2861 bcopy(soun->sun_path, buf, len); 2862 buf[len] = 0; 2863 2864 error = 0; 2865 unp = sotounpcb(so); 2866 UNP_PCB_LOCK(unp); 2867 for (;;) { 2868 /* 2869 * Wait for connection state to stabilize. If a connection 2870 * already exists, give up. For datagram sockets, which permit 2871 * multiple consecutive connect(2) calls, upper layers are 2872 * responsible for disconnecting in advance of a subsequent 2873 * connect(2), but this is not synchronized with PCB connection 2874 * state. 2875 * 2876 * Also make sure that no threads are currently attempting to 2877 * lock the peer socket, to ensure that unp_conn cannot 2878 * transition between two valid sockets while locks are dropped. 2879 */ 2880 if (SOLISTENING(so)) 2881 error = EOPNOTSUPP; 2882 else if (unp->unp_conn != NULL) 2883 error = EISCONN; 2884 else if ((unp->unp_flags & UNP_CONNECTING) != 0) { 2885 error = EALREADY; 2886 } 2887 if (error != 0) { 2888 UNP_PCB_UNLOCK(unp); 2889 return (error); 2890 } 2891 if (unp->unp_pairbusy > 0) { 2892 unp->unp_flags |= UNP_WAITING; 2893 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); 2894 continue; 2895 } 2896 break; 2897 } 2898 unp->unp_flags |= UNP_CONNECTING; 2899 UNP_PCB_UNLOCK(unp); 2900 2901 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; 2902 if (connreq) 2903 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 2904 else 2905 sa = NULL; 2906 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 2907 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT)); 2908 error = namei(&nd); 2909 if (error) 2910 vp = NULL; 2911 else 2912 vp = nd.ni_vp; 2913 ASSERT_VOP_LOCKED(vp, "unp_connect"); 2914 if (error) 2915 goto bad; 2916 NDFREE_PNBUF(&nd); 2917 2918 if (vp->v_type != VSOCK) { 2919 error = ENOTSOCK; 2920 goto bad; 2921 } 2922 #ifdef MAC 2923 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 2924 if (error) 2925 goto bad; 2926 #endif 2927 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 2928 if (error) 2929 goto bad; 2930 2931 unp = sotounpcb(so); 2932 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 2933 2934 vplock = mtx_pool_find(unp_vp_mtxpool, vp); 2935 mtx_lock(vplock); 2936 VOP_UNP_CONNECT(vp, &unp2); 2937 if (unp2 == NULL) { 2938 error = ECONNREFUSED; 2939 goto bad2; 2940 } 2941 so2 = unp2->unp_socket; 2942 if (so->so_type != so2->so_type) { 2943 error = EPROTOTYPE; 2944 goto bad2; 2945 } 2946 if (connreq) { 2947 if (SOLISTENING(so2)) 2948 so2 = solisten_clone(so2); 2949 else 2950 so2 = NULL; 2951 if (so2 == NULL) { 2952 error = ECONNREFUSED; 2953 goto bad2; 2954 } 2955 if ((error = uipc_attach(so2, 0, NULL)) != 0) { 2956 sodealloc(so2); 2957 goto bad2; 2958 } 2959 unp3 = sotounpcb(so2); 2960 unp_pcb_lock_pair(unp2, unp3); 2961 if (unp2->unp_addr != NULL) { 2962 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 2963 unp3->unp_addr = (struct sockaddr_un *) sa; 2964 sa = NULL; 2965 } 2966 2967 unp_copy_peercred(td, unp3, unp, unp2); 2968 2969 UNP_PCB_UNLOCK(unp2); 2970 unp2 = unp3; 2971 2972 /* 2973 * It is safe to block on the PCB lock here since unp2 is 2974 * nascent and cannot be connected to any other sockets. 2975 */ 2976 UNP_PCB_LOCK(unp); 2977 #ifdef MAC 2978 mac_socketpeer_set_from_socket(so, so2); 2979 mac_socketpeer_set_from_socket(so2, so); 2980 #endif 2981 } else { 2982 unp_pcb_lock_pair(unp, unp2); 2983 } 2984 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && 2985 sotounpcb(so2) == unp2, 2986 ("%s: unp2 %p so2 %p", __func__, unp2, so2)); 2987 unp_connect2(so, so2, connreq); 2988 if (connreq) 2989 (void)solisten_enqueue(so2, SS_ISCONNECTED); 2990 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 2991 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 2992 unp->unp_flags &= ~UNP_CONNECTING; 2993 if (!return_locked) 2994 unp_pcb_unlock_pair(unp, unp2); 2995 bad2: 2996 mtx_unlock(vplock); 2997 bad: 2998 if (vp != NULL) { 2999 /* 3000 * If we are returning locked (called via uipc_sosend_dgram()), 3001 * we need to be sure that vput() won't sleep. This is 3002 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock. 3003 * SOCK_STREAM/SEQPACKET can't request return_locked (yet). 3004 */ 3005 MPASS(!(return_locked && connreq)); 3006 vput(vp); 3007 } 3008 free(sa, M_SONAME); 3009 if (__predict_false(error)) { 3010 UNP_PCB_LOCK(unp); 3011 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, 3012 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); 3013 unp->unp_flags &= ~UNP_CONNECTING; 3014 UNP_PCB_UNLOCK(unp); 3015 } 3016 return (error); 3017 } 3018 3019 /* 3020 * Set socket peer credentials at connection time. 3021 * 3022 * The client's PCB credentials are copied from its process structure. The 3023 * server's PCB credentials are copied from the socket on which it called 3024 * listen(2). uipc_listen cached that process's credentials at the time. 3025 */ 3026 void 3027 unp_copy_peercred(struct thread *td, struct unpcb *client_unp, 3028 struct unpcb *server_unp, struct unpcb *listen_unp) 3029 { 3030 cru2xt(td, &client_unp->unp_peercred); 3031 client_unp->unp_flags |= UNP_HAVEPC; 3032 3033 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, 3034 sizeof(server_unp->unp_peercred)); 3035 server_unp->unp_flags |= UNP_HAVEPC; 3036 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); 3037 } 3038 3039 /* 3040 * unix/stream & unix/seqpacket version of soisconnected(). 3041 * 3042 * The crucial thing we are doing here is setting up the uxst_peer linkage, 3043 * holding unp and receive buffer locks of the both sockets. The disconnect 3044 * procedure does the same. This gives as a safe way to access the peer in the 3045 * send(2) and recv(2) during the socket lifetime. 3046 * 3047 * The less important thing is event notification of the fact that a socket is 3048 * now connected. It is unusual for a software to put a socket into event 3049 * mechanism before connect(2), but is supposed to be supported. Note that 3050 * there can not be any sleeping I/O on the socket, yet, only presence in the 3051 * select/poll/kevent. 3052 * 3053 * This function can be called via two call paths: 3054 * 1) socketpair(2) - in this case socket has not been yet reported to userland 3055 * and just can't have any event notifications mechanisms set up. The 3056 * 'wakeup' boolean is always false. 3057 * 2) connect(2) of existing socket to a recent clone of a listener: 3058 * 2.1) Socket that connect(2)s will have 'wakeup' true. An application 3059 * could have already put it into event mechanism, is it shall be 3060 * reported as readable and as writable. 3061 * 2.2) Socket that was just cloned with solisten_clone(). Same as 1). 3062 */ 3063 static void 3064 unp_soisconnected(struct socket *so, bool wakeup) 3065 { 3066 struct socket *so2 = sotounpcb(so)->unp_conn->unp_socket; 3067 struct sockbuf *sb; 3068 3069 SOCK_LOCK_ASSERT(so); 3070 UNP_PCB_LOCK_ASSERT(sotounpcb(so)); 3071 UNP_PCB_LOCK_ASSERT(sotounpcb(so2)); 3072 SOCK_RECVBUF_LOCK_ASSERT(so); 3073 SOCK_RECVBUF_LOCK_ASSERT(so2); 3074 3075 MPASS(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET); 3076 MPASS((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 3077 SS_ISDISCONNECTING)) == 0); 3078 MPASS(so->so_qstate == SQ_NONE); 3079 3080 so->so_state &= ~SS_ISDISCONNECTED; 3081 so->so_state |= SS_ISCONNECTED; 3082 3083 sb = &so2->so_rcv; 3084 sb->uxst_peer = so; 3085 3086 if (wakeup) { 3087 KNOTE_LOCKED(&sb->sb_sel->si_note, 0); 3088 sb = &so->so_rcv; 3089 selwakeuppri(sb->sb_sel, PSOCK); 3090 SOCK_SENDBUF_LOCK_ASSERT(so); 3091 sb = &so->so_snd; 3092 selwakeuppri(sb->sb_sel, PSOCK); 3093 SOCK_SENDBUF_UNLOCK(so); 3094 } 3095 } 3096 3097 static void 3098 unp_connect2(struct socket *so, struct socket *so2, bool wakeup) 3099 { 3100 struct unpcb *unp; 3101 struct unpcb *unp2; 3102 3103 MPASS(so2->so_type == so->so_type); 3104 unp = sotounpcb(so); 3105 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 3106 unp2 = sotounpcb(so2); 3107 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 3108 3109 UNP_PCB_LOCK_ASSERT(unp); 3110 UNP_PCB_LOCK_ASSERT(unp2); 3111 KASSERT(unp->unp_conn == NULL, 3112 ("%s: socket %p is already connected", __func__, unp)); 3113 3114 unp->unp_conn = unp2; 3115 unp_pcb_hold(unp2); 3116 unp_pcb_hold(unp); 3117 switch (so->so_type) { 3118 case SOCK_DGRAM: 3119 UNP_REF_LIST_LOCK(); 3120 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 3121 UNP_REF_LIST_UNLOCK(); 3122 soisconnected(so); 3123 break; 3124 3125 case SOCK_STREAM: 3126 case SOCK_SEQPACKET: 3127 KASSERT(unp2->unp_conn == NULL, 3128 ("%s: socket %p is already connected", __func__, unp2)); 3129 unp2->unp_conn = unp; 3130 SOCK_LOCK(so); 3131 SOCK_LOCK(so2); 3132 if (wakeup) /* Avoid LOR with receive buffer lock. */ 3133 SOCK_SENDBUF_LOCK(so); 3134 SOCK_RECVBUF_LOCK(so); 3135 SOCK_RECVBUF_LOCK(so2); 3136 unp_soisconnected(so, wakeup); /* Will unlock send buffer. */ 3137 unp_soisconnected(so2, false); 3138 SOCK_RECVBUF_UNLOCK(so); 3139 SOCK_RECVBUF_UNLOCK(so2); 3140 SOCK_UNLOCK(so); 3141 SOCK_UNLOCK(so2); 3142 break; 3143 3144 default: 3145 panic("unp_connect2"); 3146 } 3147 } 3148 3149 static void 3150 unp_soisdisconnected(struct socket *so) 3151 { 3152 SOCK_LOCK_ASSERT(so); 3153 SOCK_RECVBUF_LOCK_ASSERT(so); 3154 MPASS(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET); 3155 MPASS(!SOLISTENING(so)); 3156 MPASS((so->so_state & (SS_ISCONNECTING | SS_ISDISCONNECTING | 3157 SS_ISDISCONNECTED)) == 0); 3158 MPASS(so->so_state & SS_ISCONNECTED); 3159 3160 so->so_state |= SS_ISDISCONNECTED; 3161 so->so_state &= ~SS_ISCONNECTED; 3162 so->so_rcv.uxst_peer = NULL; 3163 socantrcvmore_locked(so); 3164 } 3165 3166 static void 3167 unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 3168 { 3169 struct socket *so, *so2; 3170 struct mbuf *m = NULL; 3171 #ifdef INVARIANTS 3172 struct unpcb *unptmp; 3173 #endif 3174 3175 UNP_PCB_LOCK_ASSERT(unp); 3176 UNP_PCB_LOCK_ASSERT(unp2); 3177 KASSERT(unp->unp_conn == unp2, 3178 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); 3179 3180 unp->unp_conn = NULL; 3181 so = unp->unp_socket; 3182 so2 = unp2->unp_socket; 3183 switch (unp->unp_socket->so_type) { 3184 case SOCK_DGRAM: 3185 /* 3186 * Remove our send socket buffer from the peer's receive buffer. 3187 * Move the data to the receive buffer only if it is empty. 3188 * This is a protection against a scenario where a peer 3189 * connects, floods and disconnects, effectively blocking 3190 * sendto() from unconnected sockets. 3191 */ 3192 SOCK_RECVBUF_LOCK(so2); 3193 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) { 3194 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd, 3195 uxdg_clist); 3196 if (__predict_true((so2->so_rcv.sb_state & 3197 SBS_CANTRCVMORE) == 0) && 3198 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) { 3199 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb, 3200 &so->so_snd.uxdg_mb); 3201 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc; 3202 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl; 3203 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt; 3204 } else { 3205 m = STAILQ_FIRST(&so->so_snd.uxdg_mb); 3206 STAILQ_INIT(&so->so_snd.uxdg_mb); 3207 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc; 3208 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc; 3209 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl; 3210 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt; 3211 } 3212 /* Note: so may reconnect. */ 3213 so->so_snd.uxdg_cc = 0; 3214 so->so_snd.uxdg_ctl = 0; 3215 so->so_snd.uxdg_mbcnt = 0; 3216 } 3217 SOCK_RECVBUF_UNLOCK(so2); 3218 UNP_REF_LIST_LOCK(); 3219 #ifdef INVARIANTS 3220 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { 3221 if (unptmp == unp) 3222 break; 3223 } 3224 KASSERT(unptmp != NULL, 3225 ("%s: %p not found in reflist of %p", __func__, unp, unp2)); 3226 #endif 3227 LIST_REMOVE(unp, unp_reflink); 3228 UNP_REF_LIST_UNLOCK(); 3229 SOCK_LOCK(so); 3230 so->so_state &= ~SS_ISCONNECTED; 3231 SOCK_UNLOCK(so); 3232 break; 3233 3234 case SOCK_STREAM: 3235 case SOCK_SEQPACKET: 3236 SOCK_LOCK(so); 3237 SOCK_LOCK(so2); 3238 SOCK_RECVBUF_LOCK(so); 3239 SOCK_RECVBUF_LOCK(so2); 3240 unp_soisdisconnected(so); 3241 MPASS(unp2->unp_conn == unp); 3242 unp2->unp_conn = NULL; 3243 unp_soisdisconnected(so2); 3244 SOCK_UNLOCK(so); 3245 SOCK_UNLOCK(so2); 3246 break; 3247 } 3248 3249 if (unp == unp2) { 3250 unp_pcb_rele_notlast(unp); 3251 if (!unp_pcb_rele(unp)) 3252 UNP_PCB_UNLOCK(unp); 3253 } else { 3254 if (!unp_pcb_rele(unp)) 3255 UNP_PCB_UNLOCK(unp); 3256 if (!unp_pcb_rele(unp2)) 3257 UNP_PCB_UNLOCK(unp2); 3258 } 3259 3260 if (m != NULL) { 3261 unp_scan(m, unp_freerights); 3262 m_freemp(m); 3263 } 3264 } 3265 3266 /* 3267 * unp_pcblist() walks the global list of struct unpcb's to generate a 3268 * pointer list, bumping the refcount on each unpcb. It then copies them out 3269 * sequentially, validating the generation number on each to see if it has 3270 * been detached. All of this is necessary because copyout() may sleep on 3271 * disk I/O. 3272 */ 3273 static int 3274 unp_pcblist(SYSCTL_HANDLER_ARGS) 3275 { 3276 struct unpcb *unp, **unp_list; 3277 unp_gen_t gencnt; 3278 struct xunpgen *xug; 3279 struct unp_head *head; 3280 struct xunpcb *xu; 3281 u_int i; 3282 int error, n; 3283 3284 switch ((intptr_t)arg1) { 3285 case SOCK_STREAM: 3286 head = &unp_shead; 3287 break; 3288 3289 case SOCK_DGRAM: 3290 head = &unp_dhead; 3291 break; 3292 3293 case SOCK_SEQPACKET: 3294 head = &unp_sphead; 3295 break; 3296 3297 default: 3298 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); 3299 } 3300 3301 /* 3302 * The process of preparing the PCB list is too time-consuming and 3303 * resource-intensive to repeat twice on every request. 3304 */ 3305 if (req->oldptr == NULL) { 3306 n = unp_count; 3307 req->oldidx = 2 * (sizeof *xug) 3308 + (n + n/8) * sizeof(struct xunpcb); 3309 return (0); 3310 } 3311 3312 if (req->newptr != NULL) 3313 return (EPERM); 3314 3315 /* 3316 * OK, now we're committed to doing something. 3317 */ 3318 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); 3319 UNP_LINK_RLOCK(); 3320 gencnt = unp_gencnt; 3321 n = unp_count; 3322 UNP_LINK_RUNLOCK(); 3323 3324 xug->xug_len = sizeof *xug; 3325 xug->xug_count = n; 3326 xug->xug_gen = gencnt; 3327 xug->xug_sogen = so_gencnt; 3328 error = SYSCTL_OUT(req, xug, sizeof *xug); 3329 if (error) { 3330 free(xug, M_TEMP); 3331 return (error); 3332 } 3333 3334 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 3335 3336 UNP_LINK_RLOCK(); 3337 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 3338 unp = LIST_NEXT(unp, unp_link)) { 3339 UNP_PCB_LOCK(unp); 3340 if (unp->unp_gencnt <= gencnt) { 3341 if (cr_cansee(req->td->td_ucred, 3342 unp->unp_socket->so_cred)) { 3343 UNP_PCB_UNLOCK(unp); 3344 continue; 3345 } 3346 unp_list[i++] = unp; 3347 unp_pcb_hold(unp); 3348 } 3349 UNP_PCB_UNLOCK(unp); 3350 } 3351 UNP_LINK_RUNLOCK(); 3352 n = i; /* In case we lost some during malloc. */ 3353 3354 error = 0; 3355 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 3356 for (i = 0; i < n; i++) { 3357 unp = unp_list[i]; 3358 UNP_PCB_LOCK(unp); 3359 if (unp_pcb_rele(unp)) 3360 continue; 3361 3362 if (unp->unp_gencnt <= gencnt) { 3363 xu->xu_len = sizeof *xu; 3364 xu->xu_unpp = (uintptr_t)unp; 3365 /* 3366 * XXX - need more locking here to protect against 3367 * connect/disconnect races for SMP. 3368 */ 3369 if (unp->unp_addr != NULL) 3370 bcopy(unp->unp_addr, &xu->xu_addr, 3371 unp->unp_addr->sun_len); 3372 else 3373 bzero(&xu->xu_addr, sizeof(xu->xu_addr)); 3374 if (unp->unp_conn != NULL && 3375 unp->unp_conn->unp_addr != NULL) 3376 bcopy(unp->unp_conn->unp_addr, 3377 &xu->xu_caddr, 3378 unp->unp_conn->unp_addr->sun_len); 3379 else 3380 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); 3381 xu->unp_vnode = (uintptr_t)unp->unp_vnode; 3382 xu->unp_conn = (uintptr_t)unp->unp_conn; 3383 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); 3384 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); 3385 xu->unp_gencnt = unp->unp_gencnt; 3386 sotoxsocket(unp->unp_socket, &xu->xu_socket); 3387 UNP_PCB_UNLOCK(unp); 3388 error = SYSCTL_OUT(req, xu, sizeof *xu); 3389 } else { 3390 UNP_PCB_UNLOCK(unp); 3391 } 3392 } 3393 free(xu, M_TEMP); 3394 if (!error) { 3395 /* 3396 * Give the user an updated idea of our state. If the 3397 * generation differs from what we told her before, she knows 3398 * that something happened while we were processing this 3399 * request, and it might be necessary to retry. 3400 */ 3401 xug->xug_gen = unp_gencnt; 3402 xug->xug_sogen = so_gencnt; 3403 xug->xug_count = unp_count; 3404 error = SYSCTL_OUT(req, xug, sizeof *xug); 3405 } 3406 free(unp_list, M_TEMP); 3407 free(xug, M_TEMP); 3408 return (error); 3409 } 3410 3411 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, 3412 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 3413 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 3414 "List of active local datagram sockets"); 3415 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, 3416 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 3417 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 3418 "List of active local stream sockets"); 3419 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, 3420 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 3421 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", 3422 "List of active local seqpacket sockets"); 3423 3424 static void 3425 unp_drop(struct unpcb *unp) 3426 { 3427 struct socket *so; 3428 struct unpcb *unp2; 3429 3430 /* 3431 * Regardless of whether the socket's peer dropped the connection 3432 * with this socket by aborting or disconnecting, POSIX requires 3433 * that ECONNRESET is returned on next connected send(2) in case of 3434 * a SOCK_DGRAM socket and EPIPE for SOCK_STREAM. 3435 */ 3436 UNP_PCB_LOCK(unp); 3437 if ((so = unp->unp_socket) != NULL) 3438 so->so_error = 3439 so->so_proto->pr_type == SOCK_DGRAM ? ECONNRESET : EPIPE; 3440 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { 3441 /* Last reference dropped in unp_disconnect(). */ 3442 unp_pcb_rele_notlast(unp); 3443 unp_disconnect(unp, unp2); 3444 } else if (!unp_pcb_rele(unp)) { 3445 UNP_PCB_UNLOCK(unp); 3446 } 3447 } 3448 3449 static void 3450 unp_freerights(struct filedescent **fdep, int fdcount) 3451 { 3452 struct file *fp; 3453 int i; 3454 3455 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); 3456 3457 for (i = 0; i < fdcount; i++) { 3458 fp = fdep[i]->fde_file; 3459 filecaps_free(&fdep[i]->fde_caps); 3460 unp_discard(fp); 3461 } 3462 free(fdep[0], M_FILECAPS); 3463 } 3464 3465 static bool 3466 restrict_rights(struct file *fp, struct thread *td) 3467 { 3468 struct prison *prison1, *prison2; 3469 3470 prison1 = fp->f_cred->cr_prison; 3471 prison2 = td->td_ucred->cr_prison; 3472 return (prison1 != prison2 && prison1->pr_root != prison2->pr_root && 3473 prison2 != &prison0); 3474 } 3475 3476 static int 3477 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) 3478 { 3479 struct thread *td = curthread; /* XXX */ 3480 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 3481 int *fdp; 3482 struct filedesc *fdesc = td->td_proc->p_fd; 3483 struct filedescent **fdep; 3484 void *data; 3485 socklen_t clen = control->m_len, datalen; 3486 int error, fdflags, newfds; 3487 u_int newlen; 3488 3489 UNP_LINK_UNLOCK_ASSERT(); 3490 3491 fdflags = ((flags & MSG_CMSG_CLOEXEC) ? O_CLOEXEC : 0) | 3492 ((flags & MSG_CMSG_CLOFORK) ? O_CLOFORK : 0); 3493 3494 error = 0; 3495 if (controlp != NULL) /* controlp == NULL => free control messages */ 3496 *controlp = NULL; 3497 while (cm != NULL) { 3498 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len); 3499 3500 data = CMSG_DATA(cm); 3501 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 3502 if (cm->cmsg_level == SOL_SOCKET 3503 && cm->cmsg_type == SCM_RIGHTS) { 3504 newfds = datalen / sizeof(*fdep); 3505 if (newfds == 0) 3506 goto next; 3507 fdep = data; 3508 3509 /* If we're not outputting the descriptors free them. */ 3510 if (error || controlp == NULL) { 3511 unp_freerights(fdep, newfds); 3512 goto next; 3513 } 3514 FILEDESC_XLOCK(fdesc); 3515 3516 /* 3517 * Now change each pointer to an fd in the global 3518 * table to an integer that is the index to the local 3519 * fd table entry that we set up to point to the 3520 * global one we are transferring. 3521 */ 3522 newlen = newfds * sizeof(int); 3523 *controlp = sbcreatecontrol(NULL, newlen, 3524 SCM_RIGHTS, SOL_SOCKET, M_WAITOK); 3525 3526 fdp = (int *) 3527 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 3528 if ((error = fdallocn(td, 0, fdp, newfds))) { 3529 FILEDESC_XUNLOCK(fdesc); 3530 unp_freerights(fdep, newfds); 3531 m_freem(*controlp); 3532 *controlp = NULL; 3533 goto next; 3534 } 3535 for (int i = 0; i < newfds; i++, fdp++) { 3536 struct file *fp; 3537 3538 fp = fdep[i]->fde_file; 3539 _finstall(fdesc, fp, *fdp, fdflags | 3540 (restrict_rights(fp, td) ? 3541 O_RESOLVE_BENEATH : 0), &fdep[i]->fde_caps); 3542 unp_externalize_fp(fp); 3543 } 3544 3545 /* 3546 * The new type indicates that the mbuf data refers to 3547 * kernel resources that may need to be released before 3548 * the mbuf is freed. 3549 */ 3550 m_chtype(*controlp, MT_EXTCONTROL); 3551 FILEDESC_XUNLOCK(fdesc); 3552 free(fdep[0], M_FILECAPS); 3553 } else { 3554 /* We can just copy anything else across. */ 3555 if (error || controlp == NULL) 3556 goto next; 3557 *controlp = sbcreatecontrol(NULL, datalen, 3558 cm->cmsg_type, cm->cmsg_level, M_WAITOK); 3559 bcopy(data, 3560 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 3561 datalen); 3562 } 3563 controlp = &(*controlp)->m_next; 3564 3565 next: 3566 if (CMSG_SPACE(datalen) < clen) { 3567 clen -= CMSG_SPACE(datalen); 3568 cm = (struct cmsghdr *) 3569 ((caddr_t)cm + CMSG_SPACE(datalen)); 3570 } else { 3571 clen = 0; 3572 cm = NULL; 3573 } 3574 } 3575 3576 return (error); 3577 } 3578 3579 static void 3580 unp_zone_change(void *tag) 3581 { 3582 3583 uma_zone_set_max(unp_zone, maxsockets); 3584 } 3585 3586 #ifdef INVARIANTS 3587 static void 3588 unp_zdtor(void *mem, int size __unused, void *arg __unused) 3589 { 3590 struct unpcb *unp; 3591 3592 unp = mem; 3593 3594 KASSERT(LIST_EMPTY(&unp->unp_refs), 3595 ("%s: unpcb %p has lingering refs", __func__, unp)); 3596 KASSERT(unp->unp_socket == NULL, 3597 ("%s: unpcb %p has socket backpointer", __func__, unp)); 3598 KASSERT(unp->unp_vnode == NULL, 3599 ("%s: unpcb %p has vnode references", __func__, unp)); 3600 KASSERT(unp->unp_conn == NULL, 3601 ("%s: unpcb %p is still connected", __func__, unp)); 3602 KASSERT(unp->unp_addr == NULL, 3603 ("%s: unpcb %p has leaked addr", __func__, unp)); 3604 } 3605 #endif 3606 3607 static void 3608 unp_init(void *arg __unused) 3609 { 3610 uma_dtor dtor; 3611 3612 #ifdef INVARIANTS 3613 dtor = unp_zdtor; 3614 #else 3615 dtor = NULL; 3616 #endif 3617 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, 3618 NULL, NULL, UMA_ALIGN_CACHE, 0); 3619 uma_zone_set_max(unp_zone, maxsockets); 3620 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); 3621 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 3622 NULL, EVENTHANDLER_PRI_ANY); 3623 LIST_INIT(&unp_dhead); 3624 LIST_INIT(&unp_shead); 3625 LIST_INIT(&unp_sphead); 3626 SLIST_INIT(&unp_defers); 3627 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); 3628 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); 3629 UNP_LINK_LOCK_INIT(); 3630 UNP_DEFERRED_LOCK_INIT(); 3631 unp_vp_mtxpool = mtx_pool_create("unp vp mtxpool", 32, MTX_DEF); 3632 } 3633 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL); 3634 3635 static void 3636 unp_internalize_cleanup_rights(struct mbuf *control) 3637 { 3638 struct cmsghdr *cp; 3639 struct mbuf *m; 3640 void *data; 3641 socklen_t datalen; 3642 3643 for (m = control; m != NULL; m = m->m_next) { 3644 cp = mtod(m, struct cmsghdr *); 3645 if (cp->cmsg_level != SOL_SOCKET || 3646 cp->cmsg_type != SCM_RIGHTS) 3647 continue; 3648 data = CMSG_DATA(cp); 3649 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; 3650 unp_freerights(data, datalen / sizeof(struct filedesc *)); 3651 } 3652 } 3653 3654 static int 3655 unp_internalize(struct mbuf *control, struct mchain *mc, struct thread *td) 3656 { 3657 struct proc *p; 3658 struct filedesc *fdesc; 3659 struct bintime *bt; 3660 struct cmsghdr *cm; 3661 struct cmsgcred *cmcred; 3662 struct mbuf *m; 3663 struct filedescent *fde, **fdep, *fdev; 3664 struct file *fp; 3665 struct timeval *tv; 3666 struct timespec *ts; 3667 void *data; 3668 socklen_t clen, datalen; 3669 int i, j, error, *fdp, oldfds; 3670 u_int newlen; 3671 3672 MPASS(control->m_next == NULL); /* COMPAT_OLDSOCK may violate */ 3673 UNP_LINK_UNLOCK_ASSERT(); 3674 3675 p = td->td_proc; 3676 fdesc = p->p_fd; 3677 error = 0; 3678 *mc = MCHAIN_INITIALIZER(mc); 3679 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *), 3680 data = CMSG_DATA(cm); 3681 3682 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET && 3683 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) && 3684 (char *)cm + cm->cmsg_len >= (char *)data; 3685 3686 clen -= min(CMSG_SPACE(datalen), clen), 3687 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)), 3688 data = CMSG_DATA(cm)) { 3689 datalen = (char *)cm + cm->cmsg_len - (char *)data; 3690 switch (cm->cmsg_type) { 3691 case SCM_CREDS: 3692 m = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS, 3693 SOL_SOCKET, M_WAITOK); 3694 cmcred = (struct cmsgcred *) 3695 CMSG_DATA(mtod(m, struct cmsghdr *)); 3696 cmcred->cmcred_pid = p->p_pid; 3697 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 3698 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 3699 cmcred->cmcred_euid = td->td_ucred->cr_uid; 3700 _Static_assert(CMGROUP_MAX >= 1, 3701 "Room needed for the effective GID."); 3702 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups + 1, 3703 CMGROUP_MAX); 3704 cmcred->cmcred_groups[0] = td->td_ucred->cr_gid; 3705 for (i = 1; i < cmcred->cmcred_ngroups; i++) 3706 cmcred->cmcred_groups[i] = 3707 td->td_ucred->cr_groups[i - 1]; 3708 break; 3709 3710 case SCM_RIGHTS: 3711 oldfds = datalen / sizeof (int); 3712 if (oldfds == 0) 3713 continue; 3714 /* On some machines sizeof pointer is bigger than 3715 * sizeof int, so we need to check if data fits into 3716 * single mbuf. We could allocate several mbufs, and 3717 * unp_externalize() should even properly handle that. 3718 * But it is not worth to complicate the code for an 3719 * insane scenario of passing over 200 file descriptors 3720 * at once. 3721 */ 3722 newlen = oldfds * sizeof(fdep[0]); 3723 if (CMSG_SPACE(newlen) > MCLBYTES) { 3724 error = EMSGSIZE; 3725 goto out; 3726 } 3727 /* 3728 * Check that all the FDs passed in refer to legal 3729 * files. If not, reject the entire operation. 3730 */ 3731 fdp = data; 3732 FILEDESC_SLOCK(fdesc); 3733 for (i = 0; i < oldfds; i++, fdp++) { 3734 fp = fget_noref(fdesc, *fdp); 3735 if (fp == NULL) { 3736 FILEDESC_SUNLOCK(fdesc); 3737 error = EBADF; 3738 goto out; 3739 } 3740 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 3741 FILEDESC_SUNLOCK(fdesc); 3742 error = EOPNOTSUPP; 3743 goto out; 3744 } 3745 } 3746 3747 /* 3748 * Now replace the integer FDs with pointers to the 3749 * file structure and capability rights. 3750 */ 3751 m = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, 3752 SOL_SOCKET, M_WAITOK); 3753 fdp = data; 3754 for (i = 0; i < oldfds; i++, fdp++) { 3755 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { 3756 fdp = data; 3757 for (j = 0; j < i; j++, fdp++) { 3758 fdrop(fdesc->fd_ofiles[*fdp]. 3759 fde_file, td); 3760 } 3761 FILEDESC_SUNLOCK(fdesc); 3762 error = EBADF; 3763 goto out; 3764 } 3765 } 3766 fdp = data; 3767 fdep = (struct filedescent **) 3768 CMSG_DATA(mtod(m, struct cmsghdr *)); 3769 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, 3770 M_WAITOK); 3771 for (i = 0; i < oldfds; i++, fdev++, fdp++) { 3772 fde = &fdesc->fd_ofiles[*fdp]; 3773 fdep[i] = fdev; 3774 fdep[i]->fde_file = fde->fde_file; 3775 filecaps_copy(&fde->fde_caps, 3776 &fdep[i]->fde_caps, true); 3777 unp_internalize_fp(fdep[i]->fde_file); 3778 } 3779 FILEDESC_SUNLOCK(fdesc); 3780 break; 3781 3782 case SCM_TIMESTAMP: 3783 m = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP, 3784 SOL_SOCKET, M_WAITOK); 3785 tv = (struct timeval *) 3786 CMSG_DATA(mtod(m, struct cmsghdr *)); 3787 microtime(tv); 3788 break; 3789 3790 case SCM_BINTIME: 3791 m = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME, 3792 SOL_SOCKET, M_WAITOK); 3793 bt = (struct bintime *) 3794 CMSG_DATA(mtod(m, struct cmsghdr *)); 3795 bintime(bt); 3796 break; 3797 3798 case SCM_REALTIME: 3799 m = sbcreatecontrol(NULL, sizeof(*ts), SCM_REALTIME, 3800 SOL_SOCKET, M_WAITOK); 3801 ts = (struct timespec *) 3802 CMSG_DATA(mtod(m, struct cmsghdr *)); 3803 nanotime(ts); 3804 break; 3805 3806 case SCM_MONOTONIC: 3807 m = sbcreatecontrol(NULL, sizeof(*ts), SCM_MONOTONIC, 3808 SOL_SOCKET, M_WAITOK); 3809 ts = (struct timespec *) 3810 CMSG_DATA(mtod(m, struct cmsghdr *)); 3811 nanouptime(ts); 3812 break; 3813 3814 default: 3815 error = EINVAL; 3816 goto out; 3817 } 3818 3819 mc_append(mc, m); 3820 } 3821 if (clen > 0) 3822 error = EINVAL; 3823 3824 out: 3825 if (error != 0) 3826 unp_internalize_cleanup_rights(mc_first(mc)); 3827 m_freem(control); 3828 return (error); 3829 } 3830 3831 static void 3832 unp_addsockcred(struct thread *td, struct mchain *mc, int mode) 3833 { 3834 struct mbuf *m, *n, *n_prev; 3835 const struct cmsghdr *cm; 3836 int ngroups, i, cmsgtype; 3837 size_t ctrlsz; 3838 3839 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 3840 if (mode & UNP_WANTCRED_ALWAYS) { 3841 ctrlsz = SOCKCRED2SIZE(ngroups); 3842 cmsgtype = SCM_CREDS2; 3843 } else { 3844 ctrlsz = SOCKCREDSIZE(ngroups); 3845 cmsgtype = SCM_CREDS; 3846 } 3847 3848 /* XXXGL: uipc_sosend_*() need to be improved so that we can M_WAITOK */ 3849 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT); 3850 if (m == NULL) 3851 return; 3852 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL); 3853 3854 if (mode & UNP_WANTCRED_ALWAYS) { 3855 struct sockcred2 *sc; 3856 3857 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 3858 sc->sc_version = 0; 3859 sc->sc_pid = td->td_proc->p_pid; 3860 sc->sc_uid = td->td_ucred->cr_ruid; 3861 sc->sc_euid = td->td_ucred->cr_uid; 3862 sc->sc_gid = td->td_ucred->cr_rgid; 3863 sc->sc_egid = td->td_ucred->cr_gid; 3864 sc->sc_ngroups = ngroups; 3865 for (i = 0; i < sc->sc_ngroups; i++) 3866 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 3867 } else { 3868 struct sockcred *sc; 3869 3870 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *)); 3871 sc->sc_uid = td->td_ucred->cr_ruid; 3872 sc->sc_euid = td->td_ucred->cr_uid; 3873 sc->sc_gid = td->td_ucred->cr_rgid; 3874 sc->sc_egid = td->td_ucred->cr_gid; 3875 sc->sc_ngroups = ngroups; 3876 for (i = 0; i < sc->sc_ngroups; i++) 3877 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 3878 } 3879 3880 /* 3881 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 3882 * created SCM_CREDS control message (struct sockcred) has another 3883 * format. 3884 */ 3885 if (!STAILQ_EMPTY(&mc->mc_q) && cmsgtype == SCM_CREDS) 3886 STAILQ_FOREACH_SAFE(n, &mc->mc_q, m_stailq, n_prev) { 3887 cm = mtod(n, struct cmsghdr *); 3888 if (cm->cmsg_level == SOL_SOCKET && 3889 cm->cmsg_type == SCM_CREDS) { 3890 mc_remove(mc, n); 3891 m_free(n); 3892 } 3893 } 3894 3895 /* Prepend it to the head. */ 3896 mc_prepend(mc, m); 3897 } 3898 3899 static struct unpcb * 3900 fptounp(struct file *fp) 3901 { 3902 struct socket *so; 3903 3904 if (fp->f_type != DTYPE_SOCKET) 3905 return (NULL); 3906 if ((so = fp->f_data) == NULL) 3907 return (NULL); 3908 if (so->so_proto->pr_domain != &localdomain) 3909 return (NULL); 3910 return sotounpcb(so); 3911 } 3912 3913 static void 3914 unp_discard(struct file *fp) 3915 { 3916 struct unp_defer *dr; 3917 3918 if (unp_externalize_fp(fp)) { 3919 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); 3920 dr->ud_fp = fp; 3921 UNP_DEFERRED_LOCK(); 3922 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); 3923 UNP_DEFERRED_UNLOCK(); 3924 atomic_add_int(&unp_defers_count, 1); 3925 taskqueue_enqueue(taskqueue_thread, &unp_defer_task); 3926 } else 3927 closef_nothread(fp); 3928 } 3929 3930 static void 3931 unp_process_defers(void *arg __unused, int pending) 3932 { 3933 struct unp_defer *dr; 3934 SLIST_HEAD(, unp_defer) drl; 3935 int count; 3936 3937 SLIST_INIT(&drl); 3938 for (;;) { 3939 UNP_DEFERRED_LOCK(); 3940 if (SLIST_FIRST(&unp_defers) == NULL) { 3941 UNP_DEFERRED_UNLOCK(); 3942 break; 3943 } 3944 SLIST_SWAP(&unp_defers, &drl, unp_defer); 3945 UNP_DEFERRED_UNLOCK(); 3946 count = 0; 3947 while ((dr = SLIST_FIRST(&drl)) != NULL) { 3948 SLIST_REMOVE_HEAD(&drl, ud_link); 3949 closef_nothread(dr->ud_fp); 3950 free(dr, M_TEMP); 3951 count++; 3952 } 3953 atomic_add_int(&unp_defers_count, -count); 3954 } 3955 } 3956 3957 static void 3958 unp_internalize_fp(struct file *fp) 3959 { 3960 struct unpcb *unp; 3961 3962 UNP_LINK_WLOCK(); 3963 if ((unp = fptounp(fp)) != NULL) { 3964 unp->unp_file = fp; 3965 unp->unp_msgcount++; 3966 } 3967 unp_rights++; 3968 UNP_LINK_WUNLOCK(); 3969 } 3970 3971 static int 3972 unp_externalize_fp(struct file *fp) 3973 { 3974 struct unpcb *unp; 3975 int ret; 3976 3977 UNP_LINK_WLOCK(); 3978 if ((unp = fptounp(fp)) != NULL) { 3979 unp->unp_msgcount--; 3980 ret = 1; 3981 } else 3982 ret = 0; 3983 unp_rights--; 3984 UNP_LINK_WUNLOCK(); 3985 return (ret); 3986 } 3987 3988 /* 3989 * unp_defer indicates whether additional work has been defered for a future 3990 * pass through unp_gc(). It is thread local and does not require explicit 3991 * synchronization. 3992 */ 3993 static int unp_marked; 3994 3995 static void 3996 unp_remove_dead_ref(struct filedescent **fdep, int fdcount) 3997 { 3998 struct unpcb *unp; 3999 struct file *fp; 4000 int i; 4001 4002 /* 4003 * This function can only be called from the gc task. 4004 */ 4005 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 4006 ("%s: not on gc callout", __func__)); 4007 UNP_LINK_LOCK_ASSERT(); 4008 4009 for (i = 0; i < fdcount; i++) { 4010 fp = fdep[i]->fde_file; 4011 if ((unp = fptounp(fp)) == NULL) 4012 continue; 4013 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 4014 continue; 4015 unp->unp_gcrefs--; 4016 } 4017 } 4018 4019 static void 4020 unp_restore_undead_ref(struct filedescent **fdep, int fdcount) 4021 { 4022 struct unpcb *unp; 4023 struct file *fp; 4024 int i; 4025 4026 /* 4027 * This function can only be called from the gc task. 4028 */ 4029 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, 4030 ("%s: not on gc callout", __func__)); 4031 UNP_LINK_LOCK_ASSERT(); 4032 4033 for (i = 0; i < fdcount; i++) { 4034 fp = fdep[i]->fde_file; 4035 if ((unp = fptounp(fp)) == NULL) 4036 continue; 4037 if ((unp->unp_gcflag & UNPGC_DEAD) == 0) 4038 continue; 4039 unp->unp_gcrefs++; 4040 unp_marked++; 4041 } 4042 } 4043 4044 static void 4045 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int)) 4046 { 4047 struct sockbuf *sb; 4048 4049 SOCK_LOCK_ASSERT(so); 4050 4051 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS) 4052 return; 4053 4054 SOCK_RECVBUF_LOCK(so); 4055 switch (so->so_type) { 4056 case SOCK_DGRAM: 4057 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op); 4058 unp_scan(so->so_rcv.uxdg_peeked, op); 4059 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist) 4060 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op); 4061 break; 4062 case SOCK_STREAM: 4063 case SOCK_SEQPACKET: 4064 unp_scan(STAILQ_FIRST(&so->so_rcv.uxst_mbq), op); 4065 break; 4066 } 4067 SOCK_RECVBUF_UNLOCK(so); 4068 } 4069 4070 static void 4071 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) 4072 { 4073 struct socket *so, *soa; 4074 4075 so = unp->unp_socket; 4076 SOCK_LOCK(so); 4077 if (SOLISTENING(so)) { 4078 /* 4079 * Mark all sockets in our accept queue. 4080 */ 4081 TAILQ_FOREACH(soa, &so->sol_comp, so_list) 4082 unp_scan_socket(soa, op); 4083 } else { 4084 /* 4085 * Mark all sockets we reference with RIGHTS. 4086 */ 4087 unp_scan_socket(so, op); 4088 } 4089 SOCK_UNLOCK(so); 4090 } 4091 4092 static int unp_recycled; 4093 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 4094 "Number of unreachable sockets claimed by the garbage collector."); 4095 4096 static int unp_taskcount; 4097 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 4098 "Number of times the garbage collector has run."); 4099 4100 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, 4101 "Number of active local sockets."); 4102 4103 static void 4104 unp_gc(__unused void *arg, int pending) 4105 { 4106 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, 4107 NULL }; 4108 struct unp_head **head; 4109 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ 4110 struct file *f, **unref; 4111 struct unpcb *unp, *unptmp; 4112 int i, total, unp_unreachable; 4113 4114 LIST_INIT(&unp_deadhead); 4115 unp_taskcount++; 4116 UNP_LINK_RLOCK(); 4117 /* 4118 * First determine which sockets may be in cycles. 4119 */ 4120 unp_unreachable = 0; 4121 4122 for (head = heads; *head != NULL; head++) 4123 LIST_FOREACH(unp, *head, unp_link) { 4124 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, 4125 ("%s: unp %p has unexpected gc flags 0x%x", 4126 __func__, unp, (unsigned int)unp->unp_gcflag)); 4127 4128 f = unp->unp_file; 4129 4130 /* 4131 * Check for an unreachable socket potentially in a 4132 * cycle. It must be in a queue as indicated by 4133 * msgcount, and this must equal the file reference 4134 * count. Note that when msgcount is 0 the file is 4135 * NULL. 4136 */ 4137 if (f != NULL && unp->unp_msgcount != 0 && 4138 refcount_load(&f->f_count) == unp->unp_msgcount) { 4139 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); 4140 unp->unp_gcflag |= UNPGC_DEAD; 4141 unp->unp_gcrefs = unp->unp_msgcount; 4142 unp_unreachable++; 4143 } 4144 } 4145 4146 /* 4147 * Scan all sockets previously marked as potentially being in a cycle 4148 * and remove the references each socket holds on any UNPGC_DEAD 4149 * sockets in its queue. After this step, all remaining references on 4150 * sockets marked UNPGC_DEAD should not be part of any cycle. 4151 */ 4152 LIST_FOREACH(unp, &unp_deadhead, unp_dead) 4153 unp_gc_scan(unp, unp_remove_dead_ref); 4154 4155 /* 4156 * If a socket still has a non-negative refcount, it cannot be in a 4157 * cycle. In this case increment refcount of all children iteratively. 4158 * Stop the scan once we do a complete loop without discovering 4159 * a new reachable socket. 4160 */ 4161 do { 4162 unp_marked = 0; 4163 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) 4164 if (unp->unp_gcrefs > 0) { 4165 unp->unp_gcflag &= ~UNPGC_DEAD; 4166 LIST_REMOVE(unp, unp_dead); 4167 KASSERT(unp_unreachable > 0, 4168 ("%s: unp_unreachable underflow.", 4169 __func__)); 4170 unp_unreachable--; 4171 unp_gc_scan(unp, unp_restore_undead_ref); 4172 } 4173 } while (unp_marked); 4174 4175 UNP_LINK_RUNLOCK(); 4176 4177 if (unp_unreachable == 0) 4178 return; 4179 4180 /* 4181 * Allocate space for a local array of dead unpcbs. 4182 * TODO: can this path be simplified by instead using the local 4183 * dead list at unp_deadhead, after taking out references 4184 * on the file object and/or unpcb and dropping the link lock? 4185 */ 4186 unref = malloc(unp_unreachable * sizeof(struct file *), 4187 M_TEMP, M_WAITOK); 4188 4189 /* 4190 * Iterate looking for sockets which have been specifically marked 4191 * as unreachable and store them locally. 4192 */ 4193 UNP_LINK_RLOCK(); 4194 total = 0; 4195 LIST_FOREACH(unp, &unp_deadhead, unp_dead) { 4196 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, 4197 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); 4198 unp->unp_gcflag &= ~UNPGC_DEAD; 4199 f = unp->unp_file; 4200 if (unp->unp_msgcount == 0 || f == NULL || 4201 refcount_load(&f->f_count) != unp->unp_msgcount || 4202 !fhold(f)) 4203 continue; 4204 unref[total++] = f; 4205 KASSERT(total <= unp_unreachable, 4206 ("%s: incorrect unreachable count.", __func__)); 4207 } 4208 UNP_LINK_RUNLOCK(); 4209 4210 /* 4211 * Now flush all sockets, free'ing rights. This will free the 4212 * struct files associated with these sockets but leave each socket 4213 * with one remaining ref. 4214 */ 4215 for (i = 0; i < total; i++) { 4216 struct socket *so; 4217 4218 so = unref[i]->f_data; 4219 if (!SOLISTENING(so)) { 4220 CURVNET_SET(so->so_vnet); 4221 socantrcvmore(so); 4222 unp_dispose(so); 4223 CURVNET_RESTORE(); 4224 } 4225 } 4226 4227 /* 4228 * And finally release the sockets so they can be reclaimed. 4229 */ 4230 for (i = 0; i < total; i++) 4231 fdrop(unref[i], NULL); 4232 unp_recycled += total; 4233 free(unref, M_TEMP); 4234 } 4235 4236 /* 4237 * Synchronize against unp_gc, which can trip over data as we are freeing it. 4238 */ 4239 static void 4240 unp_dispose(struct socket *so) 4241 { 4242 struct sockbuf *sb; 4243 struct unpcb *unp; 4244 struct mbuf *m; 4245 int error __diagused; 4246 4247 MPASS(!SOLISTENING(so)); 4248 4249 unp = sotounpcb(so); 4250 UNP_LINK_WLOCK(); 4251 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; 4252 UNP_LINK_WUNLOCK(); 4253 4254 /* 4255 * Grab our special mbufs before calling sbrelease(). 4256 */ 4257 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 4258 MPASS(!error); 4259 SOCK_RECVBUF_LOCK(so); 4260 switch (so->so_type) { 4261 case SOCK_DGRAM: 4262 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) { 4263 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb); 4264 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist); 4265 /* Note: socket of sb may reconnect. */ 4266 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0; 4267 } 4268 sb = &so->so_rcv; 4269 if (sb->uxdg_peeked != NULL) { 4270 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked, 4271 m_stailqpkt); 4272 sb->uxdg_peeked = NULL; 4273 } 4274 m = STAILQ_FIRST(&sb->uxdg_mb); 4275 STAILQ_INIT(&sb->uxdg_mb); 4276 break; 4277 case SOCK_STREAM: 4278 case SOCK_SEQPACKET: 4279 sb = &so->so_rcv; 4280 m = STAILQ_FIRST(&sb->uxst_mbq); 4281 STAILQ_INIT(&sb->uxst_mbq); 4282 sb->sb_acc = sb->sb_ccc = sb->sb_ctl = sb->sb_mbcnt = 0; 4283 /* 4284 * Trim M_NOTREADY buffers from the free list. They are 4285 * referenced by the I/O thread. 4286 */ 4287 if (sb->uxst_fnrdy != NULL) { 4288 struct mbuf *n, *prev; 4289 4290 while (m != NULL && m->m_flags & M_NOTREADY) 4291 m = m->m_next; 4292 for (prev = n = m; n != NULL; n = n->m_next) { 4293 if (n->m_flags & M_NOTREADY) 4294 prev->m_next = n->m_next; 4295 else 4296 prev = n; 4297 } 4298 sb->uxst_fnrdy = NULL; 4299 } 4300 break; 4301 } 4302 /* 4303 * Mark sb with SBS_CANTRCVMORE. This is needed to prevent 4304 * uipc_sosend_*() or unp_disconnect() adding more data to the socket. 4305 * We came here either through shutdown(2) or from the final sofree(). 4306 * The sofree() case is simple as it guarantees that no more sends will 4307 * happen, however we can race with unp_disconnect() from our peer. 4308 * The shutdown(2) case is more exotic. It would call into 4309 * unp_dispose() only if socket is SS_ISCONNECTED. This is possible if 4310 * we did connect(2) on this socket and we also had it bound with 4311 * bind(2) and receive connections from other sockets. Because 4312 * uipc_shutdown() violates POSIX (see comment there) this applies to 4313 * SOCK_DGRAM as well. For SOCK_DGRAM this SBS_CANTRCVMORE will have 4314 * affect not only on the peer we connect(2)ed to, but also on all of 4315 * the peers who had connect(2)ed to us. Their sends would end up 4316 * with ENOBUFS. 4317 */ 4318 sb->sb_state |= SBS_CANTRCVMORE; 4319 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0, 4320 RLIM_INFINITY); 4321 SOCK_RECVBUF_UNLOCK(so); 4322 SOCK_IO_RECV_UNLOCK(so); 4323 4324 if (m != NULL) { 4325 unp_scan(m, unp_freerights); 4326 m_freemp(m); 4327 } 4328 } 4329 4330 static void 4331 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) 4332 { 4333 struct mbuf *m; 4334 struct cmsghdr *cm; 4335 void *data; 4336 socklen_t clen, datalen; 4337 4338 while (m0 != NULL) { 4339 for (m = m0; m; m = m->m_next) { 4340 if (m->m_type != MT_CONTROL) 4341 continue; 4342 4343 cm = mtod(m, struct cmsghdr *); 4344 clen = m->m_len; 4345 4346 while (cm != NULL) { 4347 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 4348 break; 4349 4350 data = CMSG_DATA(cm); 4351 datalen = (caddr_t)cm + cm->cmsg_len 4352 - (caddr_t)data; 4353 4354 if (cm->cmsg_level == SOL_SOCKET && 4355 cm->cmsg_type == SCM_RIGHTS) { 4356 (*op)(data, datalen / 4357 sizeof(struct filedescent *)); 4358 } 4359 4360 if (CMSG_SPACE(datalen) < clen) { 4361 clen -= CMSG_SPACE(datalen); 4362 cm = (struct cmsghdr *) 4363 ((caddr_t)cm + CMSG_SPACE(datalen)); 4364 } else { 4365 clen = 0; 4366 cm = NULL; 4367 } 4368 } 4369 } 4370 m0 = m0->m_nextpkt; 4371 } 4372 } 4373 4374 /* 4375 * Definitions of protocols supported in the LOCAL domain. 4376 */ 4377 static struct protosw streamproto = { 4378 .pr_type = SOCK_STREAM, 4379 .pr_flags = PR_CONNREQUIRED | PR_CAPATTACH | PR_SOCKBUF, 4380 .pr_ctloutput = &uipc_ctloutput, 4381 .pr_abort = uipc_abort, 4382 .pr_accept = uipc_peeraddr, 4383 .pr_attach = uipc_attach, 4384 .pr_bind = uipc_bind, 4385 .pr_bindat = uipc_bindat, 4386 .pr_connect = uipc_connect, 4387 .pr_connectat = uipc_connectat, 4388 .pr_connect2 = uipc_connect2, 4389 .pr_detach = uipc_detach, 4390 .pr_disconnect = uipc_disconnect, 4391 .pr_fdclose = uipc_fdclose, 4392 .pr_listen = uipc_listen, 4393 .pr_peeraddr = uipc_peeraddr, 4394 .pr_send = uipc_sendfile, 4395 .pr_sendfile_wait = uipc_sendfile_wait, 4396 .pr_ready = uipc_ready, 4397 .pr_sense = uipc_sense, 4398 .pr_shutdown = uipc_shutdown, 4399 .pr_sockaddr = uipc_sockaddr, 4400 .pr_sosend = uipc_sosend_stream_or_seqpacket, 4401 .pr_soreceive = uipc_soreceive_stream_or_seqpacket, 4402 .pr_sopoll = uipc_sopoll_stream_or_seqpacket, 4403 .pr_kqfilter = uipc_kqfilter_stream_or_seqpacket, 4404 .pr_close = uipc_close, 4405 .pr_chmod = uipc_chmod, 4406 }; 4407 4408 static struct protosw dgramproto = { 4409 .pr_type = SOCK_DGRAM, 4410 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH | PR_SOCKBUF, 4411 .pr_ctloutput = &uipc_ctloutput, 4412 .pr_abort = uipc_abort, 4413 .pr_accept = uipc_peeraddr, 4414 .pr_attach = uipc_attach, 4415 .pr_bind = uipc_bind, 4416 .pr_bindat = uipc_bindat, 4417 .pr_connect = uipc_connect, 4418 .pr_connectat = uipc_connectat, 4419 .pr_connect2 = uipc_connect2, 4420 .pr_detach = uipc_detach, 4421 .pr_disconnect = uipc_disconnect, 4422 .pr_fdclose = uipc_fdclose, 4423 .pr_peeraddr = uipc_peeraddr, 4424 .pr_sosend = uipc_sosend_dgram, 4425 .pr_sense = uipc_sense, 4426 .pr_shutdown = uipc_shutdown, 4427 .pr_sockaddr = uipc_sockaddr, 4428 .pr_soreceive = uipc_soreceive_dgram, 4429 .pr_close = uipc_close, 4430 .pr_chmod = uipc_chmod, 4431 }; 4432 4433 static struct protosw seqpacketproto = { 4434 .pr_type = SOCK_SEQPACKET, 4435 .pr_flags = PR_CONNREQUIRED | PR_CAPATTACH | PR_SOCKBUF, 4436 .pr_ctloutput = &uipc_ctloutput, 4437 .pr_abort = uipc_abort, 4438 .pr_accept = uipc_peeraddr, 4439 .pr_attach = uipc_attach, 4440 .pr_bind = uipc_bind, 4441 .pr_bindat = uipc_bindat, 4442 .pr_connect = uipc_connect, 4443 .pr_connectat = uipc_connectat, 4444 .pr_connect2 = uipc_connect2, 4445 .pr_detach = uipc_detach, 4446 .pr_disconnect = uipc_disconnect, 4447 .pr_fdclose = uipc_fdclose, 4448 .pr_listen = uipc_listen, 4449 .pr_peeraddr = uipc_peeraddr, 4450 .pr_sense = uipc_sense, 4451 .pr_shutdown = uipc_shutdown, 4452 .pr_sockaddr = uipc_sockaddr, 4453 .pr_sosend = uipc_sosend_stream_or_seqpacket, 4454 .pr_soreceive = uipc_soreceive_stream_or_seqpacket, 4455 .pr_sopoll = uipc_sopoll_stream_or_seqpacket, 4456 .pr_kqfilter = uipc_kqfilter_stream_or_seqpacket, 4457 .pr_close = uipc_close, 4458 .pr_chmod = uipc_chmod, 4459 }; 4460 4461 static struct domain localdomain = { 4462 .dom_family = AF_LOCAL, 4463 .dom_name = "local", 4464 .dom_nprotosw = 3, 4465 .dom_protosw = { 4466 &streamproto, 4467 &dgramproto, 4468 &seqpacketproto, 4469 } 4470 }; 4471 DOMAIN_SET(local); 4472 4473 /* 4474 * A helper function called by VFS before socket-type vnode reclamation. 4475 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode 4476 * use count. 4477 */ 4478 void 4479 vfs_unp_reclaim(struct vnode *vp) 4480 { 4481 struct unpcb *unp; 4482 int active; 4483 struct mtx *vplock; 4484 4485 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); 4486 KASSERT(vp->v_type == VSOCK, 4487 ("vfs_unp_reclaim: vp->v_type != VSOCK")); 4488 4489 active = 0; 4490 vplock = mtx_pool_find(unp_vp_mtxpool, vp); 4491 mtx_lock(vplock); 4492 VOP_UNP_CONNECT(vp, &unp); 4493 if (unp == NULL) 4494 goto done; 4495 UNP_PCB_LOCK(unp); 4496 if (unp->unp_vnode == vp) { 4497 VOP_UNP_DETACH(vp); 4498 unp->unp_vnode = NULL; 4499 active = 1; 4500 } 4501 UNP_PCB_UNLOCK(unp); 4502 done: 4503 mtx_unlock(vplock); 4504 if (active) 4505 vunref(vp); 4506 } 4507 4508 #ifdef DDB 4509 static void 4510 db_print_indent(int indent) 4511 { 4512 int i; 4513 4514 for (i = 0; i < indent; i++) 4515 db_printf(" "); 4516 } 4517 4518 static void 4519 db_print_unpflags(int unp_flags) 4520 { 4521 int comma; 4522 4523 comma = 0; 4524 if (unp_flags & UNP_HAVEPC) { 4525 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 4526 comma = 1; 4527 } 4528 if (unp_flags & UNP_WANTCRED_ALWAYS) { 4529 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); 4530 comma = 1; 4531 } 4532 if (unp_flags & UNP_WANTCRED_ONESHOT) { 4533 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); 4534 comma = 1; 4535 } 4536 if (unp_flags & UNP_CONNECTING) { 4537 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 4538 comma = 1; 4539 } 4540 if (unp_flags & UNP_BINDING) { 4541 db_printf("%sUNP_BINDING", comma ? ", " : ""); 4542 comma = 1; 4543 } 4544 } 4545 4546 static void 4547 db_print_xucred(int indent, struct xucred *xu) 4548 { 4549 int comma, i; 4550 4551 db_print_indent(indent); 4552 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", 4553 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); 4554 db_print_indent(indent); 4555 db_printf("cr_groups: "); 4556 comma = 0; 4557 for (i = 0; i < xu->cr_ngroups; i++) { 4558 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 4559 comma = 1; 4560 } 4561 db_printf("\n"); 4562 } 4563 4564 static void 4565 db_print_unprefs(int indent, struct unp_head *uh) 4566 { 4567 struct unpcb *unp; 4568 int counter; 4569 4570 counter = 0; 4571 LIST_FOREACH(unp, uh, unp_reflink) { 4572 if (counter % 4 == 0) 4573 db_print_indent(indent); 4574 db_printf("%p ", unp); 4575 if (counter % 4 == 3) 4576 db_printf("\n"); 4577 counter++; 4578 } 4579 if (counter != 0 && counter % 4 != 0) 4580 db_printf("\n"); 4581 } 4582 4583 DB_SHOW_COMMAND(unpcb, db_show_unpcb) 4584 { 4585 struct unpcb *unp; 4586 4587 if (!have_addr) { 4588 db_printf("usage: show unpcb <addr>\n"); 4589 return; 4590 } 4591 unp = (struct unpcb *)addr; 4592 4593 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 4594 unp->unp_vnode); 4595 4596 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, 4597 unp->unp_conn); 4598 4599 db_printf("unp_refs:\n"); 4600 db_print_unprefs(2, &unp->unp_refs); 4601 4602 /* XXXRW: Would be nice to print the full address, if any. */ 4603 db_printf("unp_addr: %p\n", unp->unp_addr); 4604 4605 db_printf("unp_gencnt: %llu\n", 4606 (unsigned long long)unp->unp_gencnt); 4607 4608 db_printf("unp_flags: %x (", unp->unp_flags); 4609 db_print_unpflags(unp->unp_flags); 4610 db_printf(")\n"); 4611 4612 db_printf("unp_peercred:\n"); 4613 db_print_xucred(2, &unp->unp_peercred); 4614 4615 db_printf("unp_refcount: %u\n", unp->unp_refcount); 4616 } 4617 #endif 4618