uipc_usrreq.c revision 174988
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. 4 * Copyright (c) 2004-2007 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 4. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 32 */ 33 34/* 35 * UNIX Domain (Local) Sockets 36 * 37 * This is an implementation of UNIX (local) domain sockets. Each socket has 38 * an associated struct unpcb (UNIX protocol control block). Stream sockets 39 * may be connected to 0 or 1 other socket. Datagram sockets may be 40 * connected to 0, 1, or many other sockets. Sockets may be created and 41 * connected in pairs (socketpair(2)), or bound/connected to using the file 42 * system name space. For most purposes, only the receive socket buffer is 43 * used, as sending on one socket delivers directly to the receive socket 44 * buffer of a second socket. 45 * 46 * The implementation is substantially complicated by the fact that 47 * "ancillary data", such as file descriptors or credentials, may be passed 48 * across UNIX domain sockets. The potential for passing UNIX domain sockets 49 * over other UNIX domain sockets requires the implementation of a simple 50 * garbage collector to find and tear down cycles of disconnected sockets. 51 * 52 * TODO: 53 * SEQPACKET, RDM 54 * rethink name space problems 55 * need a proper out-of-band 56 * lock pushdown 57 */ 58 59#include <sys/cdefs.h> 60__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 174988 2007-12-30 01:42:15Z jeff $"); 61 62#include "opt_ddb.h" 63#include "opt_mac.h" 64 65#include <sys/param.h> 66#include <sys/domain.h> 67#include <sys/fcntl.h> 68#include <sys/malloc.h> /* XXX must be before <sys/file.h> */ 69#include <sys/eventhandler.h> 70#include <sys/file.h> 71#include <sys/filedesc.h> 72#include <sys/jail.h> 73#include <sys/kernel.h> 74#include <sys/lock.h> 75#include <sys/mbuf.h> 76#include <sys/mount.h> 77#include <sys/mutex.h> 78#include <sys/namei.h> 79#include <sys/proc.h> 80#include <sys/protosw.h> 81#include <sys/resourcevar.h> 82#include <sys/rwlock.h> 83#include <sys/socket.h> 84#include <sys/socketvar.h> 85#include <sys/signalvar.h> 86#include <sys/stat.h> 87#include <sys/sx.h> 88#include <sys/sysctl.h> 89#include <sys/systm.h> 90#include <sys/taskqueue.h> 91#include <sys/un.h> 92#include <sys/unpcb.h> 93#include <sys/vnode.h> 94 95#ifdef DDB 96#include <ddb/ddb.h> 97#endif 98 99#include <security/mac/mac_framework.h> 100 101#include <vm/uma.h> 102 103static uma_zone_t unp_zone; 104static unp_gen_t unp_gencnt; 105static u_int unp_count; /* Count of local sockets. */ 106static ino_t unp_ino; /* Prototype for fake inode numbers. */ 107static int unp_rights; /* File descriptors in flight. */ 108static struct unp_head unp_shead; /* List of local stream sockets. */ 109static struct unp_head unp_dhead; /* List of local datagram sockets. */ 110 111static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; 112 113/* 114 * Garbage collection of cyclic file descriptor/socket references occurs 115 * asynchronously in a taskqueue context in order to avoid recursion and 116 * reentrance in the UNIX domain socket, file descriptor, and socket layer 117 * code. See unp_gc() for a full description. 118 */ 119static struct task unp_gc_task; 120 121/* 122 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for 123 * stream sockets, although the total for sender and receiver is actually 124 * only PIPSIZ. 125 * 126 * Datagram sockets really use the sendspace as the maximum datagram size, 127 * and don't really want to reserve the sendspace. Their recvspace should be 128 * large enough for at least one max-size datagram plus address. 129 */ 130#ifndef PIPSIZ 131#define PIPSIZ 8192 132#endif 133static u_long unpst_sendspace = PIPSIZ; 134static u_long unpst_recvspace = PIPSIZ; 135static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 136static u_long unpdg_recvspace = 4*1024; 137 138SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain"); 139SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM"); 140SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM"); 141 142SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, 143 &unpst_sendspace, 0, ""); 144SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, 145 &unpst_recvspace, 0, ""); 146SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, 147 &unpdg_sendspace, 0, ""); 148SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, 149 &unpdg_recvspace, 0, ""); 150SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, ""); 151 152/*- 153 * Locking and synchronization: 154 * 155 * The global UNIX domain socket rwlock (unp_global_rwlock) protects all 156 * global variables, including the linked lists tracking the set of allocated 157 * UNIX domain sockets. The global rwlock also serves to prevent deadlock 158 * when more than one PCB lock is acquired at a time (i.e., during 159 * connect()). Finally, the global rwlock protects uncounted references from 160 * vnodes to sockets bound to those vnodes: to safely dereference the 161 * v_socket pointer, the global rwlock must be held while a full reference is 162 * acquired. 163 * 164 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, 165 * allocated in pru_attach() and freed in pru_detach(). The validity of that 166 * pointer is an invariant, so no lock is required to dereference the so_pcb 167 * pointer if a valid socket reference is held by the caller. In practice, 168 * this is always true during operations performed on a socket. Each unpcb 169 * has a back-pointer to its socket, unp_socket, which will be stable under 170 * the same circumstances. 171 * 172 * This pointer may only be safely dereferenced as long as a valid reference 173 * to the unpcb is held. Typically, this reference will be from the socket, 174 * or from another unpcb when the referring unpcb's lock is held (in order 175 * that the reference not be invalidated during use). For example, to follow 176 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn, 177 * as unp_socket remains valid as long as the reference to unp_conn is valid. 178 * 179 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual 180 * atomic reads without the lock may be performed "lockless", but more 181 * complex reads and read-modify-writes require the mutex to be held. No 182 * lock order is defined between unpcb locks -- multiple unpcb locks may be 183 * acquired at the same time only when holding the global UNIX domain socket 184 * rwlock exclusively, which prevents deadlocks. 185 * 186 * Blocking with UNIX domain sockets is a tricky issue: unlike most network 187 * protocols, bind() is a non-atomic operation, and connect() requires 188 * potential sleeping in the protocol, due to potentially waiting on local or 189 * distributed file systems. We try to separate "lookup" operations, which 190 * may sleep, and the IPC operations themselves, which typically can occur 191 * with relative atomicity as locks can be held over the entire operation. 192 * 193 * Another tricky issue is simultaneous multi-threaded or multi-process 194 * access to a single UNIX domain socket. These are handled by the flags 195 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or 196 * binding, both of which involve dropping UNIX domain socket locks in order 197 * to perform namei() and other file system operations. 198 */ 199static struct rwlock unp_global_rwlock; 200 201#define UNP_GLOBAL_LOCK_INIT() rw_init(&unp_global_rwlock, \ 202 "unp_global_rwlock") 203 204#define UNP_GLOBAL_LOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 205 RA_LOCKED) 206#define UNP_GLOBAL_UNLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 207 RA_UNLOCKED) 208 209#define UNP_GLOBAL_WLOCK() rw_wlock(&unp_global_rwlock) 210#define UNP_GLOBAL_WUNLOCK() rw_wunlock(&unp_global_rwlock) 211#define UNP_GLOBAL_WLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 212 RA_WLOCKED) 213#define UNP_GLOBAL_WOWNED() rw_wowned(&unp_global_rwlock) 214 215#define UNP_GLOBAL_RLOCK() rw_rlock(&unp_global_rwlock) 216#define UNP_GLOBAL_RUNLOCK() rw_runlock(&unp_global_rwlock) 217#define UNP_GLOBAL_RLOCK_ASSERT() rw_assert(&unp_global_rwlock, \ 218 RA_RLOCKED) 219 220#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ 221 "unp_mtx", "unp_mtx", \ 222 MTX_DUPOK|MTX_DEF|MTX_RECURSE) 223#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) 224#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) 225#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) 226#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) 227 228static int unp_connect(struct socket *, struct sockaddr *, 229 struct thread *); 230static int unp_connect2(struct socket *so, struct socket *so2, int); 231static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); 232static void unp_shutdown(struct unpcb *); 233static void unp_drop(struct unpcb *, int); 234static void unp_gc(__unused void *, int); 235static void unp_scan(struct mbuf *, void (*)(struct file *)); 236static void unp_discard(struct file *); 237static void unp_freerights(struct file **, int); 238static int unp_internalize(struct mbuf **, struct thread *); 239static void unp_internalize_fp(struct file *); 240static void unp_externalize_fp(struct file *); 241static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); 242 243/* 244 * Definitions of protocols supported in the LOCAL domain. 245 */ 246static struct domain localdomain; 247static struct protosw localsw[] = { 248{ 249 .pr_type = SOCK_STREAM, 250 .pr_domain = &localdomain, 251 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, 252 .pr_ctloutput = &uipc_ctloutput, 253 .pr_usrreqs = &uipc_usrreqs 254}, 255{ 256 .pr_type = SOCK_DGRAM, 257 .pr_domain = &localdomain, 258 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, 259 .pr_usrreqs = &uipc_usrreqs 260}, 261}; 262 263static struct domain localdomain = { 264 .dom_family = AF_LOCAL, 265 .dom_name = "local", 266 .dom_init = unp_init, 267 .dom_externalize = unp_externalize, 268 .dom_dispose = unp_dispose, 269 .dom_protosw = localsw, 270 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])] 271}; 272DOMAIN_SET(local); 273 274static void 275uipc_abort(struct socket *so) 276{ 277 struct unpcb *unp, *unp2; 278 279 unp = sotounpcb(so); 280 KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); 281 282 UNP_GLOBAL_WLOCK(); 283 UNP_PCB_LOCK(unp); 284 unp2 = unp->unp_conn; 285 if (unp2 != NULL) { 286 UNP_PCB_LOCK(unp2); 287 unp_drop(unp2, ECONNABORTED); 288 UNP_PCB_UNLOCK(unp2); 289 } 290 UNP_PCB_UNLOCK(unp); 291 UNP_GLOBAL_WUNLOCK(); 292} 293 294static int 295uipc_accept(struct socket *so, struct sockaddr **nam) 296{ 297 struct unpcb *unp, *unp2; 298 const struct sockaddr *sa; 299 300 /* 301 * Pass back name of connected socket, if it was bound and we are 302 * still connected (our peer may have closed already!). 303 */ 304 unp = sotounpcb(so); 305 KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); 306 307 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 308 UNP_GLOBAL_RLOCK(); 309 unp2 = unp->unp_conn; 310 if (unp2 != NULL && unp2->unp_addr != NULL) { 311 UNP_PCB_LOCK(unp2); 312 sa = (struct sockaddr *) unp2->unp_addr; 313 bcopy(sa, *nam, sa->sa_len); 314 UNP_PCB_UNLOCK(unp2); 315 } else { 316 sa = &sun_noname; 317 bcopy(sa, *nam, sa->sa_len); 318 } 319 UNP_GLOBAL_RUNLOCK(); 320 return (0); 321} 322 323static int 324uipc_attach(struct socket *so, int proto, struct thread *td) 325{ 326 u_long sendspace, recvspace; 327 struct unpcb *unp; 328 int error, locked; 329 330 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); 331 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 332 switch (so->so_type) { 333 case SOCK_STREAM: 334 sendspace = unpst_sendspace; 335 recvspace = unpst_recvspace; 336 break; 337 338 case SOCK_DGRAM: 339 sendspace = unpdg_sendspace; 340 recvspace = unpdg_recvspace; 341 break; 342 343 default: 344 panic("uipc_attach"); 345 } 346 error = soreserve(so, sendspace, recvspace); 347 if (error) 348 return (error); 349 } 350 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); 351 if (unp == NULL) 352 return (ENOBUFS); 353 LIST_INIT(&unp->unp_refs); 354 UNP_PCB_LOCK_INIT(unp); 355 unp->unp_socket = so; 356 so->so_pcb = unp; 357 unp->unp_refcount = 1; 358 359 /* 360 * uipc_attach() may be called indirectly from within the UNIX domain 361 * socket code via sonewconn() in unp_connect(). Since rwlocks can 362 * not be recursed, we do the closest thing. 363 */ 364 locked = 0; 365 if (!UNP_GLOBAL_WOWNED()) { 366 UNP_GLOBAL_WLOCK(); 367 locked = 1; 368 } 369 unp->unp_gencnt = ++unp_gencnt; 370 unp_count++; 371 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead, 372 unp, unp_link); 373 if (locked) 374 UNP_GLOBAL_WUNLOCK(); 375 376 return (0); 377} 378 379static int 380uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 381{ 382 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 383 struct vattr vattr; 384 int error, namelen, vfslocked; 385 struct nameidata nd; 386 struct unpcb *unp; 387 struct vnode *vp; 388 struct mount *mp; 389 char *buf; 390 391 unp = sotounpcb(so); 392 KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); 393 394 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); 395 if (namelen <= 0) 396 return (EINVAL); 397 398 /* 399 * We don't allow simultaneous bind() calls on a single UNIX domain 400 * socket, so flag in-progress operations, and return an error if an 401 * operation is already in progress. 402 * 403 * Historically, we have not allowed a socket to be rebound, so this 404 * also returns an error. Not allowing re-binding simplifies the 405 * implementation and avoids a great many possible failure modes. 406 */ 407 UNP_PCB_LOCK(unp); 408 if (unp->unp_vnode != NULL) { 409 UNP_PCB_UNLOCK(unp); 410 return (EINVAL); 411 } 412 if (unp->unp_flags & UNP_BINDING) { 413 UNP_PCB_UNLOCK(unp); 414 return (EALREADY); 415 } 416 unp->unp_flags |= UNP_BINDING; 417 UNP_PCB_UNLOCK(unp); 418 419 buf = malloc(namelen + 1, M_TEMP, M_WAITOK); 420 strlcpy(buf, soun->sun_path, namelen + 1); 421 422restart: 423 vfslocked = 0; 424 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME, 425 UIO_SYSSPACE, buf, td); 426/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 427 error = namei(&nd); 428 if (error) 429 goto error; 430 vp = nd.ni_vp; 431 vfslocked = NDHASGIANT(&nd); 432 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { 433 NDFREE(&nd, NDF_ONLY_PNBUF); 434 if (nd.ni_dvp == vp) 435 vrele(nd.ni_dvp); 436 else 437 vput(nd.ni_dvp); 438 if (vp != NULL) { 439 vrele(vp); 440 error = EADDRINUSE; 441 goto error; 442 } 443 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); 444 if (error) 445 goto error; 446 VFS_UNLOCK_GIANT(vfslocked); 447 goto restart; 448 } 449 VATTR_NULL(&vattr); 450 vattr.va_type = VSOCK; 451 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); 452#ifdef MAC 453 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, 454 &vattr); 455#endif 456 if (error == 0) { 457 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE); 458 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 459 } 460 NDFREE(&nd, NDF_ONLY_PNBUF); 461 vput(nd.ni_dvp); 462 if (error) { 463 vn_finished_write(mp); 464 goto error; 465 } 466 vp = nd.ni_vp; 467 ASSERT_VOP_ELOCKED(vp, "uipc_bind"); 468 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); 469 470 UNP_GLOBAL_WLOCK(); 471 UNP_PCB_LOCK(unp); 472 vp->v_socket = unp->unp_socket; 473 unp->unp_vnode = vp; 474 unp->unp_addr = soun; 475 unp->unp_flags &= ~UNP_BINDING; 476 UNP_PCB_UNLOCK(unp); 477 UNP_GLOBAL_WUNLOCK(); 478 VOP_UNLOCK(vp, 0, td); 479 vn_finished_write(mp); 480 VFS_UNLOCK_GIANT(vfslocked); 481 free(buf, M_TEMP); 482 return (0); 483 484error: 485 VFS_UNLOCK_GIANT(vfslocked); 486 UNP_PCB_LOCK(unp); 487 unp->unp_flags &= ~UNP_BINDING; 488 UNP_PCB_UNLOCK(unp); 489 free(buf, M_TEMP); 490 return (error); 491} 492 493static int 494uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 495{ 496 int error; 497 498 KASSERT(td == curthread, ("uipc_connect: td != curthread")); 499 UNP_GLOBAL_WLOCK(); 500 error = unp_connect(so, nam, td); 501 UNP_GLOBAL_WUNLOCK(); 502 return (error); 503} 504 505static void 506uipc_close(struct socket *so) 507{ 508 struct unpcb *unp, *unp2; 509 510 unp = sotounpcb(so); 511 KASSERT(unp != NULL, ("uipc_close: unp == NULL")); 512 513 UNP_GLOBAL_WLOCK(); 514 UNP_PCB_LOCK(unp); 515 unp2 = unp->unp_conn; 516 if (unp2 != NULL) { 517 UNP_PCB_LOCK(unp2); 518 unp_disconnect(unp, unp2); 519 UNP_PCB_UNLOCK(unp2); 520 } 521 UNP_PCB_UNLOCK(unp); 522 UNP_GLOBAL_WUNLOCK(); 523} 524 525int 526uipc_connect2(struct socket *so1, struct socket *so2) 527{ 528 struct unpcb *unp, *unp2; 529 int error; 530 531 UNP_GLOBAL_WLOCK(); 532 unp = so1->so_pcb; 533 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); 534 UNP_PCB_LOCK(unp); 535 unp2 = so2->so_pcb; 536 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); 537 UNP_PCB_LOCK(unp2); 538 error = unp_connect2(so1, so2, PRU_CONNECT2); 539 UNP_PCB_UNLOCK(unp2); 540 UNP_PCB_UNLOCK(unp); 541 UNP_GLOBAL_WUNLOCK(); 542 return (error); 543} 544 545/* control is EOPNOTSUPP */ 546 547static void 548uipc_detach(struct socket *so) 549{ 550 struct unpcb *unp, *unp2; 551 struct sockaddr_un *saved_unp_addr; 552 struct vnode *vp; 553 int freeunp, local_unp_rights; 554 555 unp = sotounpcb(so); 556 KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); 557 558 UNP_GLOBAL_WLOCK(); 559 UNP_PCB_LOCK(unp); 560 561 LIST_REMOVE(unp, unp_link); 562 unp->unp_gencnt = ++unp_gencnt; 563 --unp_count; 564 565 /* 566 * XXXRW: Should assert vp->v_socket == so. 567 */ 568 if ((vp = unp->unp_vnode) != NULL) { 569 unp->unp_vnode->v_socket = NULL; 570 unp->unp_vnode = NULL; 571 } 572 unp2 = unp->unp_conn; 573 if (unp2 != NULL) { 574 UNP_PCB_LOCK(unp2); 575 unp_disconnect(unp, unp2); 576 UNP_PCB_UNLOCK(unp2); 577 } 578 579 /* 580 * We hold the global lock, so it's OK to acquire multiple pcb locks 581 * at a time. 582 */ 583 while (!LIST_EMPTY(&unp->unp_refs)) { 584 struct unpcb *ref = LIST_FIRST(&unp->unp_refs); 585 586 UNP_PCB_LOCK(ref); 587 unp_drop(ref, ECONNRESET); 588 UNP_PCB_UNLOCK(ref); 589 } 590 local_unp_rights = unp_rights; 591 UNP_GLOBAL_WUNLOCK(); 592 unp->unp_socket->so_pcb = NULL; 593 saved_unp_addr = unp->unp_addr; 594 unp->unp_addr = NULL; 595 unp->unp_refcount--; 596 freeunp = (unp->unp_refcount == 0); 597 if (saved_unp_addr != NULL) 598 FREE(saved_unp_addr, M_SONAME); 599 if (freeunp) { 600 UNP_PCB_LOCK_DESTROY(unp); 601 uma_zfree(unp_zone, unp); 602 } else 603 UNP_PCB_UNLOCK(unp); 604 if (vp) { 605 int vfslocked; 606 607 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 608 vrele(vp); 609 VFS_UNLOCK_GIANT(vfslocked); 610 } 611 if (local_unp_rights) 612 taskqueue_enqueue(taskqueue_thread, &unp_gc_task); 613} 614 615static int 616uipc_disconnect(struct socket *so) 617{ 618 struct unpcb *unp, *unp2; 619 620 unp = sotounpcb(so); 621 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); 622 623 UNP_GLOBAL_WLOCK(); 624 UNP_PCB_LOCK(unp); 625 unp2 = unp->unp_conn; 626 if (unp2 != NULL) { 627 UNP_PCB_LOCK(unp2); 628 unp_disconnect(unp, unp2); 629 UNP_PCB_UNLOCK(unp2); 630 } 631 UNP_PCB_UNLOCK(unp); 632 UNP_GLOBAL_WUNLOCK(); 633 return (0); 634} 635 636static int 637uipc_listen(struct socket *so, int backlog, struct thread *td) 638{ 639 struct unpcb *unp; 640 int error; 641 642 unp = sotounpcb(so); 643 KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); 644 645 UNP_PCB_LOCK(unp); 646 if (unp->unp_vnode == NULL) { 647 UNP_PCB_UNLOCK(unp); 648 return (EINVAL); 649 } 650 651 SOCK_LOCK(so); 652 error = solisten_proto_check(so); 653 if (error == 0) { 654 cru2x(td->td_ucred, &unp->unp_peercred); 655 unp->unp_flags |= UNP_HAVEPCCACHED; 656 solisten_proto(so, backlog); 657 } 658 SOCK_UNLOCK(so); 659 UNP_PCB_UNLOCK(unp); 660 return (error); 661} 662 663static int 664uipc_peeraddr(struct socket *so, struct sockaddr **nam) 665{ 666 struct unpcb *unp, *unp2; 667 const struct sockaddr *sa; 668 669 unp = sotounpcb(so); 670 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); 671 672 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 673 UNP_PCB_LOCK(unp); 674 /* 675 * XXX: It seems that this test always fails even when connection is 676 * established. So, this else clause is added as workaround to 677 * return PF_LOCAL sockaddr. 678 */ 679 unp2 = unp->unp_conn; 680 if (unp2 != NULL) { 681 UNP_PCB_LOCK(unp2); 682 if (unp2->unp_addr != NULL) 683 sa = (struct sockaddr *) unp->unp_conn->unp_addr; 684 else 685 sa = &sun_noname; 686 bcopy(sa, *nam, sa->sa_len); 687 UNP_PCB_UNLOCK(unp2); 688 } else { 689 sa = &sun_noname; 690 bcopy(sa, *nam, sa->sa_len); 691 } 692 UNP_PCB_UNLOCK(unp); 693 return (0); 694} 695 696static int 697uipc_rcvd(struct socket *so, int flags) 698{ 699 struct unpcb *unp, *unp2; 700 struct socket *so2; 701 u_int mbcnt, sbcc; 702 u_long newhiwat; 703 704 unp = sotounpcb(so); 705 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL")); 706 707 if (so->so_type == SOCK_DGRAM) 708 panic("uipc_rcvd DGRAM?"); 709 710 if (so->so_type != SOCK_STREAM) 711 panic("uipc_rcvd unknown socktype"); 712 713 /* 714 * Adjust backpressure on sender and wakeup any waiting to write. 715 * 716 * The unp lock is acquired to maintain the validity of the unp_conn 717 * pointer; no lock on unp2 is required as unp2->unp_socket will be 718 * static as long as we don't permit unp2 to disconnect from unp, 719 * which is prevented by the lock on unp. We cache values from 720 * so_rcv to avoid holding the so_rcv lock over the entire 721 * transaction on the remote so_snd. 722 */ 723 SOCKBUF_LOCK(&so->so_rcv); 724 mbcnt = so->so_rcv.sb_mbcnt; 725 sbcc = so->so_rcv.sb_cc; 726 SOCKBUF_UNLOCK(&so->so_rcv); 727 UNP_PCB_LOCK(unp); 728 unp2 = unp->unp_conn; 729 if (unp2 == NULL) { 730 UNP_PCB_UNLOCK(unp); 731 return (0); 732 } 733 so2 = unp2->unp_socket; 734 SOCKBUF_LOCK(&so2->so_snd); 735 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt; 736 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc; 737 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, 738 newhiwat, RLIM_INFINITY); 739 sowwakeup_locked(so2); 740 unp->unp_mbcnt = mbcnt; 741 unp->unp_cc = sbcc; 742 UNP_PCB_UNLOCK(unp); 743 return (0); 744} 745 746/* pru_rcvoob is EOPNOTSUPP */ 747 748static int 749uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 750 struct mbuf *control, struct thread *td) 751{ 752 struct unpcb *unp, *unp2; 753 struct socket *so2; 754 u_int mbcnt, sbcc; 755 u_long newhiwat; 756 int error = 0; 757 758 unp = sotounpcb(so); 759 KASSERT(unp != NULL, ("uipc_send: unp == NULL")); 760 761 if (flags & PRUS_OOB) { 762 error = EOPNOTSUPP; 763 goto release; 764 } 765 766 if (control != NULL && (error = unp_internalize(&control, td))) 767 goto release; 768 769 if ((nam != NULL) || (flags & PRUS_EOF)) 770 UNP_GLOBAL_WLOCK(); 771 else 772 UNP_GLOBAL_RLOCK(); 773 774 switch (so->so_type) { 775 case SOCK_DGRAM: 776 { 777 const struct sockaddr *from; 778 779 unp2 = unp->unp_conn; 780 if (nam != NULL) { 781 UNP_GLOBAL_WLOCK_ASSERT(); 782 if (unp2 != NULL) { 783 error = EISCONN; 784 break; 785 } 786 error = unp_connect(so, nam, td); 787 if (error) 788 break; 789 unp2 = unp->unp_conn; 790 } 791 /* 792 * Because connect() and send() are non-atomic in a sendto() 793 * with a target address, it's possible that the socket will 794 * have disconnected before the send() can run. In that case 795 * return the slightly counter-intuitive but otherwise 796 * correct error that the socket is not connected. 797 */ 798 if (unp2 == NULL) { 799 error = ENOTCONN; 800 break; 801 } 802 /* Lockless read. */ 803 if (unp2->unp_flags & UNP_WANTCRED) 804 control = unp_addsockcred(td, control); 805 UNP_PCB_LOCK(unp); 806 if (unp->unp_addr != NULL) 807 from = (struct sockaddr *)unp->unp_addr; 808 else 809 from = &sun_noname; 810 so2 = unp2->unp_socket; 811 SOCKBUF_LOCK(&so2->so_rcv); 812 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { 813 sorwakeup_locked(so2); 814 m = NULL; 815 control = NULL; 816 } else { 817 SOCKBUF_UNLOCK(&so2->so_rcv); 818 error = ENOBUFS; 819 } 820 if (nam != NULL) { 821 UNP_GLOBAL_WLOCK_ASSERT(); 822 UNP_PCB_LOCK(unp2); 823 unp_disconnect(unp, unp2); 824 UNP_PCB_UNLOCK(unp2); 825 } 826 UNP_PCB_UNLOCK(unp); 827 break; 828 } 829 830 case SOCK_STREAM: 831 /* 832 * Connect if not connected yet. 833 * 834 * Note: A better implementation would complain if not equal 835 * to the peer's address. 836 */ 837 if ((so->so_state & SS_ISCONNECTED) == 0) { 838 if (nam != NULL) { 839 UNP_GLOBAL_WLOCK_ASSERT(); 840 error = unp_connect(so, nam, td); 841 if (error) 842 break; /* XXX */ 843 } else { 844 error = ENOTCONN; 845 break; 846 } 847 } 848 849 /* Lockless read. */ 850 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 851 error = EPIPE; 852 break; 853 } 854 /* 855 * Because connect() and send() are non-atomic in a sendto() 856 * with a target address, it's possible that the socket will 857 * have disconnected before the send() can run. In that case 858 * return the slightly counter-intuitive but otherwise 859 * correct error that the socket is not connected. 860 * 861 * Locking here must be done carefully: the global lock 862 * prevents interconnections between unpcbs from changing, so 863 * we can traverse from unp to unp2 without acquiring unp's 864 * lock. Socket buffer locks follow unpcb locks, so we can 865 * acquire both remote and lock socket buffer locks. 866 */ 867 unp2 = unp->unp_conn; 868 if (unp2 == NULL) { 869 error = ENOTCONN; 870 break; 871 } 872 so2 = unp2->unp_socket; 873 UNP_PCB_LOCK(unp2); 874 SOCKBUF_LOCK(&so2->so_rcv); 875 if (unp2->unp_flags & UNP_WANTCRED) { 876 /* 877 * Credentials are passed only once on SOCK_STREAM. 878 */ 879 unp2->unp_flags &= ~UNP_WANTCRED; 880 control = unp_addsockcred(td, control); 881 } 882 /* 883 * Send to paired receive port, and then reduce send buffer 884 * hiwater marks to maintain backpressure. Wake up readers. 885 */ 886 if (control != NULL) { 887 if (sbappendcontrol_locked(&so2->so_rcv, m, control)) 888 control = NULL; 889 } else 890 sbappend_locked(&so2->so_rcv, m); 891 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt; 892 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt; 893 sbcc = so2->so_rcv.sb_cc; 894 sorwakeup_locked(so2); 895 896 SOCKBUF_LOCK(&so->so_snd); 897 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc); 898 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, 899 newhiwat, RLIM_INFINITY); 900 so->so_snd.sb_mbmax -= mbcnt; 901 SOCKBUF_UNLOCK(&so->so_snd); 902 unp2->unp_cc = sbcc; 903 UNP_PCB_UNLOCK(unp2); 904 m = NULL; 905 break; 906 907 default: 908 panic("uipc_send unknown socktype"); 909 } 910 911 /* 912 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN. 913 */ 914 if (flags & PRUS_EOF) { 915 UNP_PCB_LOCK(unp); 916 socantsendmore(so); 917 unp_shutdown(unp); 918 UNP_PCB_UNLOCK(unp); 919 } 920 921 if ((nam != NULL) || (flags & PRUS_EOF)) 922 UNP_GLOBAL_WUNLOCK(); 923 else 924 UNP_GLOBAL_RUNLOCK(); 925 926 if (control != NULL && error != 0) 927 unp_dispose(control); 928 929release: 930 if (control != NULL) 931 m_freem(control); 932 if (m != NULL) 933 m_freem(m); 934 return (error); 935} 936 937static int 938uipc_sense(struct socket *so, struct stat *sb) 939{ 940 struct unpcb *unp, *unp2; 941 struct socket *so2; 942 943 unp = sotounpcb(so); 944 KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); 945 946 sb->st_blksize = so->so_snd.sb_hiwat; 947 UNP_GLOBAL_RLOCK(); 948 UNP_PCB_LOCK(unp); 949 unp2 = unp->unp_conn; 950 if (so->so_type == SOCK_STREAM && unp2 != NULL) { 951 so2 = unp2->unp_socket; 952 sb->st_blksize += so2->so_rcv.sb_cc; 953 } 954 sb->st_dev = NODEV; 955 if (unp->unp_ino == 0) 956 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; 957 sb->st_ino = unp->unp_ino; 958 UNP_PCB_UNLOCK(unp); 959 UNP_GLOBAL_RUNLOCK(); 960 return (0); 961} 962 963static int 964uipc_shutdown(struct socket *so) 965{ 966 struct unpcb *unp; 967 968 unp = sotounpcb(so); 969 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); 970 971 UNP_GLOBAL_WLOCK(); 972 UNP_PCB_LOCK(unp); 973 socantsendmore(so); 974 unp_shutdown(unp); 975 UNP_PCB_UNLOCK(unp); 976 UNP_GLOBAL_WUNLOCK(); 977 return (0); 978} 979 980static int 981uipc_sockaddr(struct socket *so, struct sockaddr **nam) 982{ 983 struct unpcb *unp; 984 const struct sockaddr *sa; 985 986 unp = sotounpcb(so); 987 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); 988 989 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 990 UNP_PCB_LOCK(unp); 991 if (unp->unp_addr != NULL) 992 sa = (struct sockaddr *) unp->unp_addr; 993 else 994 sa = &sun_noname; 995 bcopy(sa, *nam, sa->sa_len); 996 UNP_PCB_UNLOCK(unp); 997 return (0); 998} 999 1000struct pr_usrreqs uipc_usrreqs = { 1001 .pru_abort = uipc_abort, 1002 .pru_accept = uipc_accept, 1003 .pru_attach = uipc_attach, 1004 .pru_bind = uipc_bind, 1005 .pru_connect = uipc_connect, 1006 .pru_connect2 = uipc_connect2, 1007 .pru_detach = uipc_detach, 1008 .pru_disconnect = uipc_disconnect, 1009 .pru_listen = uipc_listen, 1010 .pru_peeraddr = uipc_peeraddr, 1011 .pru_rcvd = uipc_rcvd, 1012 .pru_send = uipc_send, 1013 .pru_sense = uipc_sense, 1014 .pru_shutdown = uipc_shutdown, 1015 .pru_sockaddr = uipc_sockaddr, 1016 .pru_close = uipc_close, 1017}; 1018 1019int 1020uipc_ctloutput(struct socket *so, struct sockopt *sopt) 1021{ 1022 struct unpcb *unp; 1023 struct xucred xu; 1024 int error, optval; 1025 1026 if (sopt->sopt_level != 0) 1027 return (EINVAL); 1028 1029 unp = sotounpcb(so); 1030 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); 1031 error = 0; 1032 switch (sopt->sopt_dir) { 1033 case SOPT_GET: 1034 switch (sopt->sopt_name) { 1035 case LOCAL_PEERCRED: 1036 UNP_PCB_LOCK(unp); 1037 if (unp->unp_flags & UNP_HAVEPC) 1038 xu = unp->unp_peercred; 1039 else { 1040 if (so->so_type == SOCK_STREAM) 1041 error = ENOTCONN; 1042 else 1043 error = EINVAL; 1044 } 1045 UNP_PCB_UNLOCK(unp); 1046 if (error == 0) 1047 error = sooptcopyout(sopt, &xu, sizeof(xu)); 1048 break; 1049 1050 case LOCAL_CREDS: 1051 /* Unocked read. */ 1052 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0; 1053 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1054 break; 1055 1056 case LOCAL_CONNWAIT: 1057 /* Unocked read. */ 1058 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; 1059 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1060 break; 1061 1062 default: 1063 error = EOPNOTSUPP; 1064 break; 1065 } 1066 break; 1067 1068 case SOPT_SET: 1069 switch (sopt->sopt_name) { 1070 case LOCAL_CREDS: 1071 case LOCAL_CONNWAIT: 1072 error = sooptcopyin(sopt, &optval, sizeof(optval), 1073 sizeof(optval)); 1074 if (error) 1075 break; 1076 1077#define OPTSET(bit) do { \ 1078 UNP_PCB_LOCK(unp); \ 1079 if (optval) \ 1080 unp->unp_flags |= bit; \ 1081 else \ 1082 unp->unp_flags &= ~bit; \ 1083 UNP_PCB_UNLOCK(unp); \ 1084} while (0) 1085 1086 switch (sopt->sopt_name) { 1087 case LOCAL_CREDS: 1088 OPTSET(UNP_WANTCRED); 1089 break; 1090 1091 case LOCAL_CONNWAIT: 1092 OPTSET(UNP_CONNWAIT); 1093 break; 1094 1095 default: 1096 break; 1097 } 1098 break; 1099#undef OPTSET 1100 default: 1101 error = ENOPROTOOPT; 1102 break; 1103 } 1104 break; 1105 1106 default: 1107 error = EOPNOTSUPP; 1108 break; 1109 } 1110 return (error); 1111} 1112 1113static int 1114unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1115{ 1116 struct sockaddr_un *soun = (struct sockaddr_un *)nam; 1117 struct vnode *vp; 1118 struct socket *so2, *so3; 1119 struct unpcb *unp, *unp2, *unp3; 1120 int error, len, vfslocked; 1121 struct nameidata nd; 1122 char buf[SOCK_MAXADDRLEN]; 1123 struct sockaddr *sa; 1124 1125 UNP_GLOBAL_WLOCK_ASSERT(); 1126 UNP_GLOBAL_WUNLOCK(); 1127 1128 unp = sotounpcb(so); 1129 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1130 1131 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); 1132 if (len <= 0) 1133 return (EINVAL); 1134 strlcpy(buf, soun->sun_path, len + 1); 1135 1136 UNP_PCB_LOCK(unp); 1137 if (unp->unp_flags & UNP_CONNECTING) { 1138 UNP_PCB_UNLOCK(unp); 1139 return (EALREADY); 1140 } 1141 unp->unp_flags |= UNP_CONNECTING; 1142 UNP_PCB_UNLOCK(unp); 1143 1144 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); 1145 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, 1146 td); 1147 error = namei(&nd); 1148 if (error) 1149 vp = NULL; 1150 else 1151 vp = nd.ni_vp; 1152 ASSERT_VOP_LOCKED(vp, "unp_connect"); 1153 vfslocked = NDHASGIANT(&nd); 1154 NDFREE(&nd, NDF_ONLY_PNBUF); 1155 if (error) 1156 goto bad; 1157 1158 if (vp->v_type != VSOCK) { 1159 error = ENOTSOCK; 1160 goto bad; 1161 } 1162#ifdef MAC 1163 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); 1164 if (error) 1165 goto bad; 1166#endif 1167 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); 1168 if (error) 1169 goto bad; 1170 VFS_UNLOCK_GIANT(vfslocked); 1171 1172 unp = sotounpcb(so); 1173 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1174 1175 /* 1176 * Lock global lock for two reasons: make sure v_socket is stable, 1177 * and to protect simultaneous locking of multiple pcbs. 1178 */ 1179 UNP_GLOBAL_WLOCK(); 1180 so2 = vp->v_socket; 1181 if (so2 == NULL) { 1182 error = ECONNREFUSED; 1183 goto bad2; 1184 } 1185 if (so->so_type != so2->so_type) { 1186 error = EPROTOTYPE; 1187 goto bad2; 1188 } 1189 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 1190 if (so2->so_options & SO_ACCEPTCONN) { 1191 /* 1192 * We can't drop the global lock here or 'so2' may 1193 * become invalid. As a result, we need to handle 1194 * possibly lock recursion in uipc_attach. 1195 */ 1196 so3 = sonewconn(so2, 0); 1197 } else 1198 so3 = NULL; 1199 if (so3 == NULL) { 1200 error = ECONNREFUSED; 1201 goto bad2; 1202 } 1203 unp = sotounpcb(so); 1204 unp2 = sotounpcb(so2); 1205 unp3 = sotounpcb(so3); 1206 UNP_PCB_LOCK(unp); 1207 UNP_PCB_LOCK(unp2); 1208 UNP_PCB_LOCK(unp3); 1209 if (unp2->unp_addr != NULL) { 1210 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); 1211 unp3->unp_addr = (struct sockaddr_un *) sa; 1212 sa = NULL; 1213 } 1214 /* 1215 * unp_peercred management: 1216 * 1217 * The connecter's (client's) credentials are copied from its 1218 * process structure at the time of connect() (which is now). 1219 */ 1220 cru2x(td->td_ucred, &unp3->unp_peercred); 1221 unp3->unp_flags |= UNP_HAVEPC; 1222 /* 1223 * The receiver's (server's) credentials are copied from the 1224 * unp_peercred member of socket on which the former called 1225 * listen(); uipc_listen() cached that process's credentials 1226 * at that time so we can use them now. 1227 */ 1228 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED, 1229 ("unp_connect: listener without cached peercred")); 1230 memcpy(&unp->unp_peercred, &unp2->unp_peercred, 1231 sizeof(unp->unp_peercred)); 1232 unp->unp_flags |= UNP_HAVEPC; 1233 if (unp2->unp_flags & UNP_WANTCRED) 1234 unp3->unp_flags |= UNP_WANTCRED; 1235 UNP_PCB_UNLOCK(unp3); 1236 UNP_PCB_UNLOCK(unp2); 1237 UNP_PCB_UNLOCK(unp); 1238#ifdef MAC 1239 SOCK_LOCK(so); 1240 mac_socketpeer_set_from_socket(so, so3); 1241 mac_socketpeer_set_from_socket(so3, so); 1242 SOCK_UNLOCK(so); 1243#endif 1244 1245 so2 = so3; 1246 } 1247 unp = sotounpcb(so); 1248 KASSERT(unp != NULL, ("unp_connect: unp == NULL")); 1249 unp2 = sotounpcb(so2); 1250 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL")); 1251 UNP_PCB_LOCK(unp); 1252 UNP_PCB_LOCK(unp2); 1253 error = unp_connect2(so, so2, PRU_CONNECT); 1254 UNP_PCB_UNLOCK(unp2); 1255 UNP_PCB_UNLOCK(unp); 1256bad2: 1257 UNP_GLOBAL_WUNLOCK(); 1258 if (vfslocked) 1259 /* 1260 * Giant has been previously acquired. This means filesystem 1261 * isn't MPSAFE. Do it once again. 1262 */ 1263 mtx_lock(&Giant); 1264bad: 1265 if (vp != NULL) 1266 vput(vp); 1267 VFS_UNLOCK_GIANT(vfslocked); 1268 free(sa, M_SONAME); 1269 UNP_GLOBAL_WLOCK(); 1270 UNP_PCB_LOCK(unp); 1271 unp->unp_flags &= ~UNP_CONNECTING; 1272 UNP_PCB_UNLOCK(unp); 1273 return (error); 1274} 1275 1276static int 1277unp_connect2(struct socket *so, struct socket *so2, int req) 1278{ 1279 struct unpcb *unp; 1280 struct unpcb *unp2; 1281 1282 unp = sotounpcb(so); 1283 KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); 1284 unp2 = sotounpcb(so2); 1285 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); 1286 1287 UNP_GLOBAL_WLOCK_ASSERT(); 1288 UNP_PCB_LOCK_ASSERT(unp); 1289 UNP_PCB_LOCK_ASSERT(unp2); 1290 1291 if (so2->so_type != so->so_type) 1292 return (EPROTOTYPE); 1293 unp->unp_conn = unp2; 1294 1295 switch (so->so_type) { 1296 case SOCK_DGRAM: 1297 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); 1298 soisconnected(so); 1299 break; 1300 1301 case SOCK_STREAM: 1302 unp2->unp_conn = unp; 1303 if (req == PRU_CONNECT && 1304 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1305 soisconnecting(so); 1306 else 1307 soisconnected(so); 1308 soisconnected(so2); 1309 break; 1310 1311 default: 1312 panic("unp_connect2"); 1313 } 1314 return (0); 1315} 1316 1317static void 1318unp_disconnect(struct unpcb *unp, struct unpcb *unp2) 1319{ 1320 struct socket *so; 1321 1322 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL")); 1323 1324 UNP_GLOBAL_WLOCK_ASSERT(); 1325 UNP_PCB_LOCK_ASSERT(unp); 1326 UNP_PCB_LOCK_ASSERT(unp2); 1327 1328 unp->unp_conn = NULL; 1329 switch (unp->unp_socket->so_type) { 1330 case SOCK_DGRAM: 1331 LIST_REMOVE(unp, unp_reflink); 1332 so = unp->unp_socket; 1333 SOCK_LOCK(so); 1334 so->so_state &= ~SS_ISCONNECTED; 1335 SOCK_UNLOCK(so); 1336 break; 1337 1338 case SOCK_STREAM: 1339 soisdisconnected(unp->unp_socket); 1340 unp2->unp_conn = NULL; 1341 soisdisconnected(unp2->unp_socket); 1342 break; 1343 } 1344} 1345 1346/* 1347 * unp_pcblist() walks the global list of struct unpcb's to generate a 1348 * pointer list, bumping the refcount on each unpcb. It then copies them out 1349 * sequentially, validating the generation number on each to see if it has 1350 * been detached. All of this is necessary because copyout() may sleep on 1351 * disk I/O. 1352 */ 1353static int 1354unp_pcblist(SYSCTL_HANDLER_ARGS) 1355{ 1356 int error, i, n; 1357 int freeunp; 1358 struct unpcb *unp, **unp_list; 1359 unp_gen_t gencnt; 1360 struct xunpgen *xug; 1361 struct unp_head *head; 1362 struct xunpcb *xu; 1363 1364 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead); 1365 1366 /* 1367 * The process of preparing the PCB list is too time-consuming and 1368 * resource-intensive to repeat twice on every request. 1369 */ 1370 if (req->oldptr == NULL) { 1371 n = unp_count; 1372 req->oldidx = 2 * (sizeof *xug) 1373 + (n + n/8) * sizeof(struct xunpcb); 1374 return (0); 1375 } 1376 1377 if (req->newptr != NULL) 1378 return (EPERM); 1379 1380 /* 1381 * OK, now we're committed to doing something. 1382 */ 1383 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); 1384 UNP_GLOBAL_RLOCK(); 1385 gencnt = unp_gencnt; 1386 n = unp_count; 1387 UNP_GLOBAL_RUNLOCK(); 1388 1389 xug->xug_len = sizeof *xug; 1390 xug->xug_count = n; 1391 xug->xug_gen = gencnt; 1392 xug->xug_sogen = so_gencnt; 1393 error = SYSCTL_OUT(req, xug, sizeof *xug); 1394 if (error) { 1395 free(xug, M_TEMP); 1396 return (error); 1397 } 1398 1399 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); 1400 1401 UNP_GLOBAL_RLOCK(); 1402 for (unp = LIST_FIRST(head), i = 0; unp && i < n; 1403 unp = LIST_NEXT(unp, unp_link)) { 1404 UNP_PCB_LOCK(unp); 1405 if (unp->unp_gencnt <= gencnt) { 1406 if (cr_cansee(req->td->td_ucred, 1407 unp->unp_socket->so_cred)) { 1408 UNP_PCB_UNLOCK(unp); 1409 continue; 1410 } 1411 unp_list[i++] = unp; 1412 unp->unp_refcount++; 1413 } 1414 UNP_PCB_UNLOCK(unp); 1415 } 1416 UNP_GLOBAL_RUNLOCK(); 1417 n = i; /* In case we lost some during malloc. */ 1418 1419 error = 0; 1420 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); 1421 for (i = 0; i < n; i++) { 1422 unp = unp_list[i]; 1423 UNP_PCB_LOCK(unp); 1424 unp->unp_refcount--; 1425 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) { 1426 xu->xu_len = sizeof *xu; 1427 xu->xu_unpp = unp; 1428 /* 1429 * XXX - need more locking here to protect against 1430 * connect/disconnect races for SMP. 1431 */ 1432 if (unp->unp_addr != NULL) 1433 bcopy(unp->unp_addr, &xu->xu_addr, 1434 unp->unp_addr->sun_len); 1435 if (unp->unp_conn != NULL && 1436 unp->unp_conn->unp_addr != NULL) 1437 bcopy(unp->unp_conn->unp_addr, 1438 &xu->xu_caddr, 1439 unp->unp_conn->unp_addr->sun_len); 1440 bcopy(unp, &xu->xu_unp, sizeof *unp); 1441 sotoxsocket(unp->unp_socket, &xu->xu_socket); 1442 UNP_PCB_UNLOCK(unp); 1443 error = SYSCTL_OUT(req, xu, sizeof *xu); 1444 } else { 1445 freeunp = (unp->unp_refcount == 0); 1446 UNP_PCB_UNLOCK(unp); 1447 if (freeunp) { 1448 UNP_PCB_LOCK_DESTROY(unp); 1449 uma_zfree(unp_zone, unp); 1450 } 1451 } 1452 } 1453 free(xu, M_TEMP); 1454 if (!error) { 1455 /* 1456 * Give the user an updated idea of our state. If the 1457 * generation differs from what we told her before, she knows 1458 * that something happened while we were processing this 1459 * request, and it might be necessary to retry. 1460 */ 1461 xug->xug_gen = unp_gencnt; 1462 xug->xug_sogen = so_gencnt; 1463 xug->xug_count = unp_count; 1464 error = SYSCTL_OUT(req, xug, sizeof *xug); 1465 } 1466 free(unp_list, M_TEMP); 1467 free(xug, M_TEMP); 1468 return (error); 1469} 1470 1471SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD, 1472 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", 1473 "List of active local datagram sockets"); 1474SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD, 1475 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", 1476 "List of active local stream sockets"); 1477 1478static void 1479unp_shutdown(struct unpcb *unp) 1480{ 1481 struct unpcb *unp2; 1482 struct socket *so; 1483 1484 UNP_GLOBAL_WLOCK_ASSERT(); 1485 UNP_PCB_LOCK_ASSERT(unp); 1486 1487 unp2 = unp->unp_conn; 1488 if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) { 1489 so = unp2->unp_socket; 1490 if (so != NULL) 1491 socantrcvmore(so); 1492 } 1493} 1494 1495static void 1496unp_drop(struct unpcb *unp, int errno) 1497{ 1498 struct socket *so = unp->unp_socket; 1499 struct unpcb *unp2; 1500 1501 UNP_GLOBAL_WLOCK_ASSERT(); 1502 UNP_PCB_LOCK_ASSERT(unp); 1503 1504 so->so_error = errno; 1505 unp2 = unp->unp_conn; 1506 if (unp2 == NULL) 1507 return; 1508 1509 UNP_PCB_LOCK(unp2); 1510 unp_disconnect(unp, unp2); 1511 UNP_PCB_UNLOCK(unp2); 1512} 1513 1514static void 1515unp_freerights(struct file **rp, int fdcount) 1516{ 1517 int i; 1518 struct file *fp; 1519 1520 for (i = 0; i < fdcount; i++) { 1521 /* 1522 * Zero the pointer before calling unp_discard since it may 1523 * end up in unp_gc().. 1524 * 1525 * XXXRW: This is less true than it used to be. 1526 */ 1527 fp = *rp; 1528 *rp++ = NULL; 1529 unp_discard(fp); 1530 } 1531} 1532 1533int 1534unp_externalize(struct mbuf *control, struct mbuf **controlp) 1535{ 1536 struct thread *td = curthread; /* XXX */ 1537 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1538 int i; 1539 int *fdp; 1540 struct file **rp; 1541 struct file *fp; 1542 void *data; 1543 socklen_t clen = control->m_len, datalen; 1544 int error, newfds; 1545 int f; 1546 u_int newlen; 1547 1548 UNP_GLOBAL_UNLOCK_ASSERT(); 1549 1550 error = 0; 1551 if (controlp != NULL) /* controlp == NULL => free control messages */ 1552 *controlp = NULL; 1553 1554 while (cm != NULL) { 1555 if (sizeof(*cm) > clen || cm->cmsg_len > clen) { 1556 error = EINVAL; 1557 break; 1558 } 1559 1560 data = CMSG_DATA(cm); 1561 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1562 1563 if (cm->cmsg_level == SOL_SOCKET 1564 && cm->cmsg_type == SCM_RIGHTS) { 1565 newfds = datalen / sizeof(struct file *); 1566 rp = data; 1567 1568 /* If we're not outputting the descriptors free them. */ 1569 if (error || controlp == NULL) { 1570 unp_freerights(rp, newfds); 1571 goto next; 1572 } 1573 FILEDESC_XLOCK(td->td_proc->p_fd); 1574 /* if the new FD's will not fit free them. */ 1575 if (!fdavail(td, newfds)) { 1576 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1577 error = EMSGSIZE; 1578 unp_freerights(rp, newfds); 1579 goto next; 1580 } 1581 /* 1582 * Now change each pointer to an fd in the global 1583 * table to an integer that is the index to the local 1584 * fd table entry that we set up to point to the 1585 * global one we are transferring. 1586 */ 1587 newlen = newfds * sizeof(int); 1588 *controlp = sbcreatecontrol(NULL, newlen, 1589 SCM_RIGHTS, SOL_SOCKET); 1590 if (*controlp == NULL) { 1591 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1592 error = E2BIG; 1593 unp_freerights(rp, newfds); 1594 goto next; 1595 } 1596 1597 fdp = (int *) 1598 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1599 for (i = 0; i < newfds; i++) { 1600 if (fdalloc(td, 0, &f)) 1601 panic("unp_externalize fdalloc failed"); 1602 fp = *rp++; 1603 td->td_proc->p_fd->fd_ofiles[f] = fp; 1604 unp_externalize_fp(fp); 1605 *fdp++ = f; 1606 } 1607 FILEDESC_XUNLOCK(td->td_proc->p_fd); 1608 } else { 1609 /* We can just copy anything else across. */ 1610 if (error || controlp == NULL) 1611 goto next; 1612 *controlp = sbcreatecontrol(NULL, datalen, 1613 cm->cmsg_type, cm->cmsg_level); 1614 if (*controlp == NULL) { 1615 error = ENOBUFS; 1616 goto next; 1617 } 1618 bcopy(data, 1619 CMSG_DATA(mtod(*controlp, struct cmsghdr *)), 1620 datalen); 1621 } 1622 1623 controlp = &(*controlp)->m_next; 1624 1625next: 1626 if (CMSG_SPACE(datalen) < clen) { 1627 clen -= CMSG_SPACE(datalen); 1628 cm = (struct cmsghdr *) 1629 ((caddr_t)cm + CMSG_SPACE(datalen)); 1630 } else { 1631 clen = 0; 1632 cm = NULL; 1633 } 1634 } 1635 1636 m_freem(control); 1637 1638 return (error); 1639} 1640 1641static void 1642unp_zone_change(void *tag) 1643{ 1644 1645 uma_zone_set_max(unp_zone, maxsockets); 1646} 1647 1648void 1649unp_init(void) 1650{ 1651 1652 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL, 1653 NULL, NULL, UMA_ALIGN_PTR, 0); 1654 if (unp_zone == NULL) 1655 panic("unp_init"); 1656 uma_zone_set_max(unp_zone, maxsockets); 1657 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, 1658 NULL, EVENTHANDLER_PRI_ANY); 1659 LIST_INIT(&unp_dhead); 1660 LIST_INIT(&unp_shead); 1661 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL); 1662 UNP_GLOBAL_LOCK_INIT(); 1663} 1664 1665static int 1666unp_internalize(struct mbuf **controlp, struct thread *td) 1667{ 1668 struct mbuf *control = *controlp; 1669 struct proc *p = td->td_proc; 1670 struct filedesc *fdescp = p->p_fd; 1671 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 1672 struct cmsgcred *cmcred; 1673 struct file **rp; 1674 struct file *fp; 1675 struct timeval *tv; 1676 int i, fd, *fdp; 1677 void *data; 1678 socklen_t clen = control->m_len, datalen; 1679 int error, oldfds; 1680 u_int newlen; 1681 1682 UNP_GLOBAL_UNLOCK_ASSERT(); 1683 1684 error = 0; 1685 *controlp = NULL; 1686 1687 while (cm != NULL) { 1688 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET 1689 || cm->cmsg_len > clen) { 1690 error = EINVAL; 1691 goto out; 1692 } 1693 1694 data = CMSG_DATA(cm); 1695 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; 1696 1697 switch (cm->cmsg_type) { 1698 /* 1699 * Fill in credential information. 1700 */ 1701 case SCM_CREDS: 1702 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), 1703 SCM_CREDS, SOL_SOCKET); 1704 if (*controlp == NULL) { 1705 error = ENOBUFS; 1706 goto out; 1707 } 1708 1709 cmcred = (struct cmsgcred *) 1710 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1711 cmcred->cmcred_pid = p->p_pid; 1712 cmcred->cmcred_uid = td->td_ucred->cr_ruid; 1713 cmcred->cmcred_gid = td->td_ucred->cr_rgid; 1714 cmcred->cmcred_euid = td->td_ucred->cr_uid; 1715 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, 1716 CMGROUP_MAX); 1717 for (i = 0; i < cmcred->cmcred_ngroups; i++) 1718 cmcred->cmcred_groups[i] = 1719 td->td_ucred->cr_groups[i]; 1720 break; 1721 1722 case SCM_RIGHTS: 1723 oldfds = datalen / sizeof (int); 1724 /* 1725 * Check that all the FDs passed in refer to legal 1726 * files. If not, reject the entire operation. 1727 */ 1728 fdp = data; 1729 FILEDESC_SLOCK(fdescp); 1730 for (i = 0; i < oldfds; i++) { 1731 fd = *fdp++; 1732 if ((unsigned)fd >= fdescp->fd_nfiles || 1733 fdescp->fd_ofiles[fd] == NULL) { 1734 FILEDESC_SUNLOCK(fdescp); 1735 error = EBADF; 1736 goto out; 1737 } 1738 fp = fdescp->fd_ofiles[fd]; 1739 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { 1740 FILEDESC_SUNLOCK(fdescp); 1741 error = EOPNOTSUPP; 1742 goto out; 1743 } 1744 1745 } 1746 1747 /* 1748 * Now replace the integer FDs with pointers to 1749 * the associated global file table entry.. 1750 */ 1751 newlen = oldfds * sizeof(struct file *); 1752 *controlp = sbcreatecontrol(NULL, newlen, 1753 SCM_RIGHTS, SOL_SOCKET); 1754 if (*controlp == NULL) { 1755 FILEDESC_SUNLOCK(fdescp); 1756 error = E2BIG; 1757 goto out; 1758 } 1759 1760 fdp = data; 1761 rp = (struct file **) 1762 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1763 for (i = 0; i < oldfds; i++) { 1764 fp = fdescp->fd_ofiles[*fdp++]; 1765 *rp++ = fp; 1766 fhold(fp); 1767 unp_internalize_fp(fp); 1768 } 1769 FILEDESC_SUNLOCK(fdescp); 1770 break; 1771 1772 case SCM_TIMESTAMP: 1773 *controlp = sbcreatecontrol(NULL, sizeof(*tv), 1774 SCM_TIMESTAMP, SOL_SOCKET); 1775 if (*controlp == NULL) { 1776 error = ENOBUFS; 1777 goto out; 1778 } 1779 tv = (struct timeval *) 1780 CMSG_DATA(mtod(*controlp, struct cmsghdr *)); 1781 microtime(tv); 1782 break; 1783 1784 default: 1785 error = EINVAL; 1786 goto out; 1787 } 1788 1789 controlp = &(*controlp)->m_next; 1790 1791 if (CMSG_SPACE(datalen) < clen) { 1792 clen -= CMSG_SPACE(datalen); 1793 cm = (struct cmsghdr *) 1794 ((caddr_t)cm + CMSG_SPACE(datalen)); 1795 } else { 1796 clen = 0; 1797 cm = NULL; 1798 } 1799 } 1800 1801out: 1802 m_freem(control); 1803 1804 return (error); 1805} 1806 1807static struct mbuf * 1808unp_addsockcred(struct thread *td, struct mbuf *control) 1809{ 1810 struct mbuf *m, *n, *n_prev; 1811 struct sockcred *sc; 1812 const struct cmsghdr *cm; 1813 int ngroups; 1814 int i; 1815 1816 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); 1817 1818 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); 1819 if (m == NULL) 1820 return (control); 1821 1822 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); 1823 sc->sc_uid = td->td_ucred->cr_ruid; 1824 sc->sc_euid = td->td_ucred->cr_uid; 1825 sc->sc_gid = td->td_ucred->cr_rgid; 1826 sc->sc_egid = td->td_ucred->cr_gid; 1827 sc->sc_ngroups = ngroups; 1828 for (i = 0; i < sc->sc_ngroups; i++) 1829 sc->sc_groups[i] = td->td_ucred->cr_groups[i]; 1830 1831 /* 1832 * Unlink SCM_CREDS control messages (struct cmsgcred), since just 1833 * created SCM_CREDS control message (struct sockcred) has another 1834 * format. 1835 */ 1836 if (control != NULL) 1837 for (n = control, n_prev = NULL; n != NULL;) { 1838 cm = mtod(n, struct cmsghdr *); 1839 if (cm->cmsg_level == SOL_SOCKET && 1840 cm->cmsg_type == SCM_CREDS) { 1841 if (n_prev == NULL) 1842 control = n->m_next; 1843 else 1844 n_prev->m_next = n->m_next; 1845 n = m_free(n); 1846 } else { 1847 n_prev = n; 1848 n = n->m_next; 1849 } 1850 } 1851 1852 /* Prepend it to the head. */ 1853 m->m_next = control; 1854 1855 return (m); 1856} 1857 1858static struct unpcb * 1859fptounp(struct file *fp) 1860{ 1861 struct socket *so; 1862 1863 if (fp->f_type != DTYPE_SOCKET) 1864 return (NULL); 1865 if ((so = fp->f_data) == NULL) 1866 return (NULL); 1867 if (so->so_proto->pr_domain != &localdomain) 1868 return (NULL); 1869 return sotounpcb(so); 1870} 1871 1872static void 1873unp_discard(struct file *fp) 1874{ 1875 1876 unp_externalize_fp(fp); 1877 (void) closef(fp, (struct thread *)NULL); 1878} 1879 1880static void 1881unp_internalize_fp(struct file *fp) 1882{ 1883 struct unpcb *unp; 1884 1885 UNP_GLOBAL_WLOCK(); 1886 if ((unp = fptounp(fp)) != NULL) { 1887 unp->unp_file = fp; 1888 unp->unp_msgcount++; 1889 } 1890 unp_rights++; 1891 UNP_GLOBAL_WUNLOCK(); 1892} 1893 1894static void 1895unp_externalize_fp(struct file *fp) 1896{ 1897 struct unpcb *unp; 1898 1899 UNP_GLOBAL_WLOCK(); 1900 if ((unp = fptounp(fp)) != NULL) 1901 unp->unp_msgcount--; 1902 unp_rights--; 1903 UNP_GLOBAL_WUNLOCK(); 1904} 1905 1906/* 1907 * unp_defer indicates whether additional work has been defered for a future 1908 * pass through unp_gc(). It is thread local and does not require explicit 1909 * synchronization. 1910 */ 1911static int unp_marked; 1912static int unp_unreachable; 1913 1914static void 1915unp_accessable(struct file *fp) 1916{ 1917 struct unpcb *unp; 1918 1919 unp = fptounp(fp); 1920 if (fp == NULL) 1921 return; 1922 if (unp->unp_gcflag & UNPGC_REF) 1923 return; 1924 unp->unp_gcflag &= ~UNPGC_DEAD; 1925 unp->unp_gcflag |= UNPGC_REF; 1926 unp_marked++; 1927} 1928 1929static void 1930unp_gc_process(struct unpcb *unp) 1931{ 1932 struct socket *soa; 1933 struct socket *so; 1934 struct file *fp; 1935 1936 /* Already processed. */ 1937 if (unp->unp_gcflag & UNPGC_SCANNED) 1938 return; 1939 fp = unp->unp_file; 1940 /* 1941 * Check for a socket potentially in a cycle. It must be in a 1942 * queue as indicated by msgcount, and this must equal the file 1943 * reference count. Note that when msgcount is 0 the file is NULL. 1944 */ 1945 if (unp->unp_msgcount != 0 && fp->f_count != 0 && 1946 fp->f_count == unp->unp_msgcount) { 1947 unp->unp_gcflag |= UNPGC_DEAD; 1948 unp_unreachable++; 1949 return; 1950 } 1951 /* 1952 * Mark all sockets we reference with RIGHTS. 1953 */ 1954 so = unp->unp_socket; 1955 SOCKBUF_LOCK(&so->so_rcv); 1956 unp_scan(so->so_rcv.sb_mb, unp_accessable); 1957 SOCKBUF_UNLOCK(&so->so_rcv); 1958 /* 1959 * Mark all sockets in our accept queue. 1960 */ 1961 ACCEPT_LOCK(); 1962 TAILQ_FOREACH(soa, &so->so_comp, so_list) { 1963 SOCKBUF_LOCK(&soa->so_rcv); 1964 unp_scan(soa->so_rcv.sb_mb, unp_accessable); 1965 SOCKBUF_UNLOCK(&soa->so_rcv); 1966 } 1967 ACCEPT_UNLOCK(); 1968 unp->unp_gcflag |= UNPGC_SCANNED; 1969} 1970 1971static int unp_recycled; 1972SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, ""); 1973 1974static int unp_taskcount; 1975SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, ""); 1976 1977static void 1978unp_gc(__unused void *arg, int pending) 1979{ 1980 struct unp_head *heads[] = { &unp_dhead, &unp_shead, NULL }; 1981 struct unp_head **head; 1982 struct file **unref; 1983 struct unpcb *unp; 1984 int i; 1985 1986 unp_taskcount++; 1987 UNP_GLOBAL_RLOCK(); 1988 /* 1989 * First clear all gc flags from previous runs. 1990 */ 1991 for (head = heads; *head != NULL; head++) 1992 LIST_FOREACH(unp, *head, unp_link) 1993 unp->unp_gcflag &= ~(UNPGC_REF|UNPGC_DEAD); 1994 /* 1995 * Scan marking all reachable sockets with UNPGC_REF. Once a socket 1996 * is reachable all of the sockets it references are reachable. 1997 * Stop the scan once we do a complete loop without discovering 1998 * a new reachable socket. 1999 */ 2000 do { 2001 unp_unreachable = 0; 2002 unp_marked = 0; 2003 for (head = heads; *head != NULL; head++) 2004 LIST_FOREACH(unp, *head, unp_link) 2005 unp_gc_process(unp); 2006 } while (unp_marked); 2007 UNP_GLOBAL_RUNLOCK(); 2008 if (unp_unreachable == 0) 2009 return; 2010 /* 2011 * Allocate space for a local list of dead unpcbs. 2012 */ 2013 unref = malloc(unp_unreachable * sizeof(struct file *), 2014 M_TEMP, M_WAITOK); 2015 /* 2016 * Iterate looking for sockets which have been specifically marked 2017 * as as unreachable and store them locally. 2018 */ 2019 UNP_GLOBAL_RLOCK(); 2020 for (i = 0, head = heads; *head != NULL; head++) 2021 LIST_FOREACH(unp, *head, unp_link) 2022 if (unp->unp_gcflag & UNPGC_DEAD) { 2023 unref[i++] = unp->unp_file; 2024 KASSERT(unp->unp_file != NULL, 2025 ("unp_gc: Invalid unpcb.")); 2026 KASSERT(i <= unp_unreachable, 2027 ("unp_gc: incorrect unreachable count.")); 2028 } 2029 UNP_GLOBAL_RUNLOCK(); 2030 /* 2031 * All further operation is now done on a local list. We first ref 2032 * all sockets to avoid closing them until all are flushed. 2033 */ 2034 for (i = 0; i < unp_unreachable; i++) 2035 fhold(unref[i]); 2036 /* 2037 * Now flush all sockets, free'ing rights. This will free the 2038 * struct files associated with these sockets but leave each socket 2039 * with one remaining ref. 2040 */ 2041 for (i = 0; i < unp_unreachable; i++) 2042 sorflush(unref[i]->f_data); 2043 /* 2044 * And finally release the sockets so they can be reclaimed. 2045 */ 2046 for (i = 0; i < unp_unreachable; i++) 2047 fdrop(unref[i], NULL); 2048 unp_recycled += unp_unreachable; 2049 free(unref, M_TEMP); 2050} 2051 2052void 2053unp_dispose(struct mbuf *m) 2054{ 2055 2056 if (m) 2057 unp_scan(m, unp_discard); 2058} 2059 2060static void 2061unp_scan(struct mbuf *m0, void (*op)(struct file *)) 2062{ 2063 struct mbuf *m; 2064 struct file **rp; 2065 struct cmsghdr *cm; 2066 void *data; 2067 int i; 2068 socklen_t clen, datalen; 2069 int qfds; 2070 2071 while (m0 != NULL) { 2072 for (m = m0; m; m = m->m_next) { 2073 if (m->m_type != MT_CONTROL) 2074 continue; 2075 2076 cm = mtod(m, struct cmsghdr *); 2077 clen = m->m_len; 2078 2079 while (cm != NULL) { 2080 if (sizeof(*cm) > clen || cm->cmsg_len > clen) 2081 break; 2082 2083 data = CMSG_DATA(cm); 2084 datalen = (caddr_t)cm + cm->cmsg_len 2085 - (caddr_t)data; 2086 2087 if (cm->cmsg_level == SOL_SOCKET && 2088 cm->cmsg_type == SCM_RIGHTS) { 2089 qfds = datalen / sizeof (struct file *); 2090 rp = data; 2091 for (i = 0; i < qfds; i++) 2092 (*op)(*rp++); 2093 } 2094 2095 if (CMSG_SPACE(datalen) < clen) { 2096 clen -= CMSG_SPACE(datalen); 2097 cm = (struct cmsghdr *) 2098 ((caddr_t)cm + CMSG_SPACE(datalen)); 2099 } else { 2100 clen = 0; 2101 cm = NULL; 2102 } 2103 } 2104 } 2105 m0 = m0->m_act; 2106 } 2107} 2108 2109#ifdef DDB 2110static void 2111db_print_indent(int indent) 2112{ 2113 int i; 2114 2115 for (i = 0; i < indent; i++) 2116 db_printf(" "); 2117} 2118 2119static void 2120db_print_unpflags(int unp_flags) 2121{ 2122 int comma; 2123 2124 comma = 0; 2125 if (unp_flags & UNP_HAVEPC) { 2126 db_printf("%sUNP_HAVEPC", comma ? ", " : ""); 2127 comma = 1; 2128 } 2129 if (unp_flags & UNP_HAVEPCCACHED) { 2130 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : ""); 2131 comma = 1; 2132 } 2133 if (unp_flags & UNP_WANTCRED) { 2134 db_printf("%sUNP_WANTCRED", comma ? ", " : ""); 2135 comma = 1; 2136 } 2137 if (unp_flags & UNP_CONNWAIT) { 2138 db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); 2139 comma = 1; 2140 } 2141 if (unp_flags & UNP_CONNECTING) { 2142 db_printf("%sUNP_CONNECTING", comma ? ", " : ""); 2143 comma = 1; 2144 } 2145 if (unp_flags & UNP_BINDING) { 2146 db_printf("%sUNP_BINDING", comma ? ", " : ""); 2147 comma = 1; 2148 } 2149} 2150 2151static void 2152db_print_xucred(int indent, struct xucred *xu) 2153{ 2154 int comma, i; 2155 2156 db_print_indent(indent); 2157 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n", 2158 xu->cr_version, xu->cr_uid, xu->cr_ngroups); 2159 db_print_indent(indent); 2160 db_printf("cr_groups: "); 2161 comma = 0; 2162 for (i = 0; i < xu->cr_ngroups; i++) { 2163 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); 2164 comma = 1; 2165 } 2166 db_printf("\n"); 2167} 2168 2169static void 2170db_print_unprefs(int indent, struct unp_head *uh) 2171{ 2172 struct unpcb *unp; 2173 int counter; 2174 2175 counter = 0; 2176 LIST_FOREACH(unp, uh, unp_reflink) { 2177 if (counter % 4 == 0) 2178 db_print_indent(indent); 2179 db_printf("%p ", unp); 2180 if (counter % 4 == 3) 2181 db_printf("\n"); 2182 counter++; 2183 } 2184 if (counter != 0 && counter % 4 != 0) 2185 db_printf("\n"); 2186} 2187 2188DB_SHOW_COMMAND(unpcb, db_show_unpcb) 2189{ 2190 struct unpcb *unp; 2191 2192 if (!have_addr) { 2193 db_printf("usage: show unpcb <addr>\n"); 2194 return; 2195 } 2196 unp = (struct unpcb *)addr; 2197 2198 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, 2199 unp->unp_vnode); 2200 2201 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino, 2202 unp->unp_conn); 2203 2204 db_printf("unp_refs:\n"); 2205 db_print_unprefs(2, &unp->unp_refs); 2206 2207 /* XXXRW: Would be nice to print the full address, if any. */ 2208 db_printf("unp_addr: %p\n", unp->unp_addr); 2209 2210 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n", 2211 unp->unp_cc, unp->unp_mbcnt, 2212 (unsigned long long)unp->unp_gencnt); 2213 2214 db_printf("unp_flags: %x (", unp->unp_flags); 2215 db_print_unpflags(unp->unp_flags); 2216 db_printf(")\n"); 2217 2218 db_printf("unp_peercred:\n"); 2219 db_print_xucred(2, &unp->unp_peercred); 2220 2221 db_printf("unp_refcount: %u\n", unp->unp_refcount); 2222} 2223#endif 2224