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