1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991 Regents of the University of California. 3 * All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)uipc_usrreq.c 7.24 (Berkeley) 03/24/91 8 */ 9 10 #include "param.h" 11 #include "proc.h" 12 #include "filedesc.h" 13 #include "domain.h" 14 #include "protosw.h" 15 #include "socket.h" 16 #include "socketvar.h" 17 #include "unpcb.h" 18 #include "un.h" 19 #include "vnode.h" 20 #include "file.h" 21 #include "stat.h" 22 #include "mbuf.h" 23 24 /* 25 * Unix communications domain. 26 * 27 * TODO: 28 * SEQPACKET, RDM 29 * rethink name space problems 30 * need a proper out-of-band 31 */ 32 struct sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX }; 33 ino_t unp_ino; /* prototype for fake inode numbers */ 34 35 /*ARGSUSED*/ 36 uipc_usrreq(so, req, m, nam, control) 37 struct socket *so; 38 int req; 39 struct mbuf *m, *nam, *control; 40 { 41 struct unpcb *unp = sotounpcb(so); 42 register struct socket *so2; 43 register int error = 0; 44 45 if (req == PRU_CONTROL) 46 return (EOPNOTSUPP); 47 if (req != PRU_SEND && control && control->m_len) { 48 error = EOPNOTSUPP; 49 goto release; 50 } 51 if (unp == 0 && req != PRU_ATTACH) { 52 error = EINVAL; 53 goto release; 54 } 55 switch (req) { 56 57 case PRU_ATTACH: 58 if (unp) { 59 error = EISCONN; 60 break; 61 } 62 error = unp_attach(so); 63 break; 64 65 case PRU_DETACH: 66 unp_detach(unp); 67 break; 68 69 case PRU_BIND: 70 error = unp_bind(unp, nam); 71 break; 72 73 case PRU_LISTEN: 74 if (unp->unp_vnode == 0) 75 error = EINVAL; 76 break; 77 78 case PRU_CONNECT: 79 error = unp_connect(so, nam); 80 break; 81 82 case PRU_CONNECT2: 83 error = unp_connect2(so, (struct socket *)nam); 84 break; 85 86 case PRU_DISCONNECT: 87 unp_disconnect(unp); 88 break; 89 90 case PRU_ACCEPT: 91 /* 92 * Pass back name of connected socket, 93 * if it was bound and we are still connected 94 * (our peer may have closed already!). 95 */ 96 if (unp->unp_conn && unp->unp_conn->unp_addr) { 97 nam->m_len = unp->unp_conn->unp_addr->m_len; 98 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 99 mtod(nam, caddr_t), (unsigned)nam->m_len); 100 } else { 101 nam->m_len = sizeof(sun_noname); 102 *(mtod(nam, struct sockaddr *)) = sun_noname; 103 } 104 break; 105 106 case PRU_SHUTDOWN: 107 socantsendmore(so); 108 unp_shutdown(unp); 109 break; 110 111 case PRU_RCVD: 112 switch (so->so_type) { 113 114 case SOCK_DGRAM: 115 panic("uipc 1"); 116 /*NOTREACHED*/ 117 118 case SOCK_STREAM: 119 #define rcv (&so->so_rcv) 120 #define snd (&so2->so_snd) 121 if (unp->unp_conn == 0) 122 break; 123 so2 = unp->unp_conn->unp_socket; 124 /* 125 * Adjust backpressure on sender 126 * and wakeup any waiting to write. 127 */ 128 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 129 unp->unp_mbcnt = rcv->sb_mbcnt; 130 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 131 unp->unp_cc = rcv->sb_cc; 132 sowwakeup(so2); 133 #undef snd 134 #undef rcv 135 break; 136 137 default: 138 panic("uipc 2"); 139 } 140 break; 141 142 case PRU_SEND: 143 if (control && (error = unp_internalize(control))) 144 break; 145 switch (so->so_type) { 146 147 case SOCK_DGRAM: { 148 struct sockaddr *from; 149 150 if (nam) { 151 if (unp->unp_conn) { 152 error = EISCONN; 153 break; 154 } 155 error = unp_connect(so, nam); 156 if (error) 157 break; 158 } else { 159 if (unp->unp_conn == 0) { 160 error = ENOTCONN; 161 break; 162 } 163 } 164 so2 = unp->unp_conn->unp_socket; 165 if (unp->unp_addr) 166 from = mtod(unp->unp_addr, struct sockaddr *); 167 else 168 from = &sun_noname; 169 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 170 sorwakeup(so2); 171 m = 0; 172 control = 0; 173 } else 174 error = ENOBUFS; 175 if (nam) 176 unp_disconnect(unp); 177 break; 178 } 179 180 case SOCK_STREAM: 181 #define rcv (&so2->so_rcv) 182 #define snd (&so->so_snd) 183 if (so->so_state & SS_CANTSENDMORE) { 184 error = EPIPE; 185 break; 186 } 187 if (unp->unp_conn == 0) 188 panic("uipc 3"); 189 so2 = unp->unp_conn->unp_socket; 190 /* 191 * Send to paired receive port, and then reduce 192 * send buffer hiwater marks to maintain backpressure. 193 * Wake up readers. 194 */ 195 if (control) { 196 if (sbappendcontrol(rcv, m, control)) 197 control = 0; 198 } else 199 sbappend(rcv, m); 200 snd->sb_mbmax -= 201 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 202 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 203 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 204 unp->unp_conn->unp_cc = rcv->sb_cc; 205 sorwakeup(so2); 206 m = 0; 207 #undef snd 208 #undef rcv 209 break; 210 211 default: 212 panic("uipc 4"); 213 } 214 break; 215 216 case PRU_ABORT: 217 unp_drop(unp, ECONNABORTED); 218 break; 219 220 case PRU_SENSE: 221 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 222 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 223 so2 = unp->unp_conn->unp_socket; 224 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 225 } 226 ((struct stat *) m)->st_dev = NODEV; 227 if (unp->unp_ino == 0) 228 unp->unp_ino = unp_ino++; 229 ((struct stat *) m)->st_ino = unp->unp_ino; 230 return (0); 231 232 case PRU_RCVOOB: 233 return (EOPNOTSUPP); 234 235 case PRU_SENDOOB: 236 error = EOPNOTSUPP; 237 break; 238 239 case PRU_SOCKADDR: 240 if (unp->unp_addr) { 241 nam->m_len = unp->unp_addr->m_len; 242 bcopy(mtod(unp->unp_addr, caddr_t), 243 mtod(nam, caddr_t), (unsigned)nam->m_len); 244 } else 245 nam->m_len = 0; 246 break; 247 248 case PRU_PEERADDR: 249 if (unp->unp_conn && unp->unp_conn->unp_addr) { 250 nam->m_len = unp->unp_conn->unp_addr->m_len; 251 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 252 mtod(nam, caddr_t), (unsigned)nam->m_len); 253 } else 254 nam->m_len = 0; 255 break; 256 257 case PRU_SLOWTIMO: 258 break; 259 260 default: 261 panic("piusrreq"); 262 } 263 release: 264 if (control) 265 m_freem(control); 266 if (m) 267 m_freem(m); 268 return (error); 269 } 270 271 /* 272 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 273 * for stream sockets, although the total for sender and receiver is 274 * actually only PIPSIZ. 275 * Datagram sockets really use the sendspace as the maximum datagram size, 276 * and don't really want to reserve the sendspace. Their recvspace should 277 * be large enough for at least one max-size datagram plus address. 278 */ 279 #define PIPSIZ 4096 280 u_long unpst_sendspace = PIPSIZ; 281 u_long unpst_recvspace = PIPSIZ; 282 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 283 u_long unpdg_recvspace = 4*1024; 284 285 int unp_rights; /* file descriptors in flight */ 286 287 unp_attach(so) 288 struct socket *so; 289 { 290 register struct mbuf *m; 291 register struct unpcb *unp; 292 int error; 293 294 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 295 switch (so->so_type) { 296 297 case SOCK_STREAM: 298 error = soreserve(so, unpst_sendspace, unpst_recvspace); 299 break; 300 301 case SOCK_DGRAM: 302 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 303 break; 304 } 305 if (error) 306 return (error); 307 } 308 m = m_getclr(M_DONTWAIT, MT_PCB); 309 if (m == NULL) 310 return (ENOBUFS); 311 unp = mtod(m, struct unpcb *); 312 so->so_pcb = (caddr_t)unp; 313 unp->unp_socket = so; 314 return (0); 315 } 316 317 unp_detach(unp) 318 register struct unpcb *unp; 319 { 320 321 if (unp->unp_vnode) { 322 unp->unp_vnode->v_socket = 0; 323 vrele(unp->unp_vnode); 324 unp->unp_vnode = 0; 325 } 326 if (unp->unp_conn) 327 unp_disconnect(unp); 328 while (unp->unp_refs) 329 unp_drop(unp->unp_refs, ECONNRESET); 330 soisdisconnected(unp->unp_socket); 331 unp->unp_socket->so_pcb = 0; 332 m_freem(unp->unp_addr); 333 (void) m_free(dtom(unp)); 334 if (unp_rights) 335 unp_gc(); 336 } 337 338 unp_bind(unp, nam) 339 struct unpcb *unp; 340 struct mbuf *nam; 341 { 342 struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 343 register struct vnode *vp; 344 register struct nameidata *ndp; 345 struct vattr vattr; 346 int error; 347 struct nameidata nd; 348 349 ndp = &nd; 350 ndp->ni_dirp = soun->sun_path; 351 if (unp->unp_vnode != NULL) 352 return (EINVAL); 353 if (nam->m_len == MLEN) { 354 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 355 return (EINVAL); 356 } else 357 *(mtod(nam, caddr_t) + nam->m_len) = 0; 358 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 359 ndp->ni_nameiop = CREATE | FOLLOW | LOCKPARENT; 360 ndp->ni_segflg = UIO_SYSSPACE; 361 if (error = namei(ndp, curproc)) /* XXX */ 362 return (error); 363 vp = ndp->ni_vp; 364 if (vp != NULL) { 365 VOP_ABORTOP(ndp); 366 if (ndp->ni_dvp == vp) 367 vrele(ndp->ni_dvp); 368 else 369 vput(ndp->ni_dvp); 370 vrele(vp); 371 return (EADDRINUSE); 372 } 373 VATTR_NULL(&vattr); 374 vattr.va_type = VSOCK; 375 vattr.va_mode = 0777; 376 if (error = VOP_CREATE(ndp, &vattr)) 377 return (error); 378 vp = ndp->ni_vp; 379 vp->v_socket = unp->unp_socket; 380 unp->unp_vnode = vp; 381 unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL); 382 VOP_UNLOCK(vp); 383 return (0); 384 } 385 386 unp_connect(so, nam) 387 struct socket *so; 388 struct mbuf *nam; 389 { 390 register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 391 register struct vnode *vp; 392 register struct socket *so2, *so3; 393 register struct nameidata *ndp; 394 struct unpcb *unp2, *unp3; 395 int error; 396 struct nameidata nd; 397 398 ndp = &nd; 399 ndp->ni_dirp = soun->sun_path; 400 if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) { /* XXX */ 401 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 402 return (EMSGSIZE); 403 } else 404 *(mtod(nam, caddr_t) + nam->m_len) = 0; 405 ndp->ni_nameiop = LOOKUP | FOLLOW | LOCKLEAF; 406 ndp->ni_segflg = UIO_SYSSPACE; 407 if (error = namei(ndp, curproc)) /* XXX */ 408 return (error); 409 vp = ndp->ni_vp; 410 if (vp->v_type != VSOCK) { 411 error = ENOTSOCK; 412 goto bad; 413 } 414 if (error = VOP_ACCESS(vp, VWRITE, curproc->p_ucred)) 415 goto bad; 416 so2 = vp->v_socket; 417 if (so2 == 0) { 418 error = ECONNREFUSED; 419 goto bad; 420 } 421 if (so->so_type != so2->so_type) { 422 error = EPROTOTYPE; 423 goto bad; 424 } 425 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 426 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 427 (so3 = sonewconn(so2, 0)) == 0) { 428 error = ECONNREFUSED; 429 goto bad; 430 } 431 unp2 = sotounpcb(so2); 432 unp3 = sotounpcb(so3); 433 if (unp2->unp_addr) 434 unp3->unp_addr = 435 m_copy(unp2->unp_addr, 0, (int)M_COPYALL); 436 so2 = so3; 437 } 438 error = unp_connect2(so, so2); 439 bad: 440 vput(vp); 441 return (error); 442 } 443 444 unp_connect2(so, so2) 445 register struct socket *so; 446 register struct socket *so2; 447 { 448 register struct unpcb *unp = sotounpcb(so); 449 register struct unpcb *unp2; 450 451 if (so2->so_type != so->so_type) 452 return (EPROTOTYPE); 453 unp2 = sotounpcb(so2); 454 unp->unp_conn = unp2; 455 switch (so->so_type) { 456 457 case SOCK_DGRAM: 458 unp->unp_nextref = unp2->unp_refs; 459 unp2->unp_refs = unp; 460 soisconnected(so); 461 break; 462 463 case SOCK_STREAM: 464 unp2->unp_conn = unp; 465 soisconnected(so); 466 soisconnected(so2); 467 break; 468 469 default: 470 panic("unp_connect2"); 471 } 472 return (0); 473 } 474 475 unp_disconnect(unp) 476 struct unpcb *unp; 477 { 478 register struct unpcb *unp2 = unp->unp_conn; 479 480 if (unp2 == 0) 481 return; 482 unp->unp_conn = 0; 483 switch (unp->unp_socket->so_type) { 484 485 case SOCK_DGRAM: 486 if (unp2->unp_refs == unp) 487 unp2->unp_refs = unp->unp_nextref; 488 else { 489 unp2 = unp2->unp_refs; 490 for (;;) { 491 if (unp2 == 0) 492 panic("unp_disconnect"); 493 if (unp2->unp_nextref == unp) 494 break; 495 unp2 = unp2->unp_nextref; 496 } 497 unp2->unp_nextref = unp->unp_nextref; 498 } 499 unp->unp_nextref = 0; 500 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 501 break; 502 503 case SOCK_STREAM: 504 soisdisconnected(unp->unp_socket); 505 unp2->unp_conn = 0; 506 soisdisconnected(unp2->unp_socket); 507 break; 508 } 509 } 510 511 #ifdef notdef 512 unp_abort(unp) 513 struct unpcb *unp; 514 { 515 516 unp_detach(unp); 517 } 518 #endif 519 520 unp_shutdown(unp) 521 struct unpcb *unp; 522 { 523 struct socket *so; 524 525 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 526 (so = unp->unp_conn->unp_socket)) 527 socantrcvmore(so); 528 } 529 530 unp_drop(unp, errno) 531 struct unpcb *unp; 532 int errno; 533 { 534 struct socket *so = unp->unp_socket; 535 536 so->so_error = errno; 537 unp_disconnect(unp); 538 if (so->so_head) { 539 so->so_pcb = (caddr_t) 0; 540 m_freem(unp->unp_addr); 541 (void) m_free(dtom(unp)); 542 sofree(so); 543 } 544 } 545 546 #ifdef notdef 547 unp_drain() 548 { 549 550 } 551 #endif 552 553 unp_externalize(rights) 554 struct mbuf *rights; 555 { 556 struct proc *p = curproc; /* XXX */ 557 register int i; 558 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 559 register struct file **rp = (struct file **)(cm + 1); 560 register struct file *fp; 561 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 562 int f; 563 564 if (fdavail(p, newfds)) { 565 for (i = 0; i < newfds; i++) { 566 fp = *rp; 567 unp_discard(fp); 568 *rp++ = 0; 569 } 570 return (EMSGSIZE); 571 } 572 for (i = 0; i < newfds; i++) { 573 if (fdalloc(p, 0, &f)) 574 panic("unp_externalize"); 575 fp = *rp; 576 p->p_fd->fd_ofiles[f] = fp; 577 fp->f_msgcount--; 578 unp_rights--; 579 *(int *)rp++ = f; 580 } 581 return (0); 582 } 583 584 unp_internalize(control) 585 struct mbuf *control; 586 { 587 struct filedesc *fdp = curproc->p_fd; /* XXX */ 588 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 589 register struct file **rp; 590 register struct file *fp; 591 register int i, fd; 592 int oldfds; 593 594 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 595 cm->cmsg_len != control->m_len) 596 return (EINVAL); 597 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 598 rp = (struct file **)(cm + 1); 599 for (i = 0; i < oldfds; i++) { 600 fd = *(int *)rp++; 601 if ((unsigned)fd >= fdp->fd_nfiles || 602 fdp->fd_ofiles[fd] == NULL) 603 return (EBADF); 604 } 605 rp = (struct file **)(cm + 1); 606 for (i = 0; i < oldfds; i++) { 607 fp = fdp->fd_ofiles[*(int *)rp]; 608 *rp++ = fp; 609 fp->f_count++; 610 fp->f_msgcount++; 611 unp_rights++; 612 } 613 return (0); 614 } 615 616 int unp_defer, unp_gcing; 617 int unp_mark(); 618 extern struct domain unixdomain; 619 620 unp_gc() 621 { 622 register struct file *fp; 623 register struct socket *so; 624 625 if (unp_gcing) 626 return; 627 unp_gcing = 1; 628 restart: 629 unp_defer = 0; 630 for (fp = file; fp < fileNFILE; fp++) 631 fp->f_flag &= ~(FMARK|FDEFER); 632 do { 633 for (fp = file; fp < fileNFILE; fp++) { 634 if (fp->f_count == 0) 635 continue; 636 if (fp->f_flag & FDEFER) { 637 fp->f_flag &= ~FDEFER; 638 unp_defer--; 639 } else { 640 if (fp->f_flag & FMARK) 641 continue; 642 if (fp->f_count == fp->f_msgcount) 643 continue; 644 fp->f_flag |= FMARK; 645 } 646 if (fp->f_type != DTYPE_SOCKET || 647 (so = (struct socket *)fp->f_data) == 0) 648 continue; 649 if (so->so_proto->pr_domain != &unixdomain || 650 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 651 continue; 652 #ifdef notdef 653 if (so->so_rcv.sb_flags & SB_LOCK) { 654 /* 655 * This is problematical; it's not clear 656 * we need to wait for the sockbuf to be 657 * unlocked (on a uniprocessor, at least), 658 * and it's also not clear what to do 659 * if sbwait returns an error due to receipt 660 * of a signal. If sbwait does return 661 * an error, we'll go into an infinite 662 * loop. Delete all of this for now. 663 */ 664 (void) sbwait(&so->so_rcv); 665 goto restart; 666 } 667 #endif 668 unp_scan(so->so_rcv.sb_mb, unp_mark); 669 } 670 } while (unp_defer); 671 for (fp = file; fp < fileNFILE; fp++) { 672 if (fp->f_count == 0) 673 continue; 674 if (fp->f_count == fp->f_msgcount && (fp->f_flag & FMARK) == 0) 675 while (fp->f_msgcount) 676 unp_discard(fp); 677 } 678 unp_gcing = 0; 679 } 680 681 unp_dispose(m) 682 struct mbuf *m; 683 { 684 int unp_discard(); 685 686 if (m) 687 unp_scan(m, unp_discard); 688 } 689 690 unp_scan(m0, op) 691 register struct mbuf *m0; 692 int (*op)(); 693 { 694 register struct mbuf *m; 695 register struct file **rp; 696 register struct cmsghdr *cm; 697 register int i; 698 int qfds; 699 700 while (m0) { 701 for (m = m0; m; m = m->m_next) 702 if (m->m_type == MT_CONTROL && 703 m->m_len >= sizeof(*cm)) { 704 cm = mtod(m, struct cmsghdr *); 705 if (cm->cmsg_level != SOL_SOCKET || 706 cm->cmsg_type != SCM_RIGHTS) 707 continue; 708 qfds = (cm->cmsg_len - sizeof *cm) 709 / sizeof (struct file *); 710 rp = (struct file **)(cm + 1); 711 for (i = 0; i < qfds; i++) 712 (*op)(*rp++); 713 break; /* XXX, but saves time */ 714 } 715 m0 = m0->m_act; 716 } 717 } 718 719 unp_mark(fp) 720 struct file *fp; 721 { 722 723 if (fp->f_flag & FMARK) 724 return; 725 unp_defer++; 726 fp->f_flag |= (FMARK|FDEFER); 727 } 728 729 unp_discard(fp) 730 struct file *fp; 731 { 732 733 fp->f_msgcount--; 734 unp_rights--; 735 (void) closef(fp); 736 } 737