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 7.26 (Berkeley) 06/03/91 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 "namei.h" 20 #include "vnode.h" 21 #include "file.h" 22 #include "stat.h" 23 #include "mbuf.h" 24 25 /* 26 * Unix communications domain. 27 * 28 * TODO: 29 * SEQPACKET, RDM 30 * rethink name space problems 31 * need a proper out-of-band 32 */ 33 struct sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX }; 34 ino_t unp_ino; /* prototype for fake inode numbers */ 35 36 /*ARGSUSED*/ 37 uipc_usrreq(so, req, m, nam, control) 38 struct socket *so; 39 int req; 40 struct mbuf *m, *nam, *control; 41 { 42 struct unpcb *unp = sotounpcb(so); 43 register struct socket *so2; 44 register int error = 0; 45 struct proc *p = curproc; /* XXX */ 46 47 if (req == PRU_CONTROL) 48 return (EOPNOTSUPP); 49 if (req != PRU_SEND && control && control->m_len) { 50 error = EOPNOTSUPP; 51 goto release; 52 } 53 if (unp == 0 && req != PRU_ATTACH) { 54 error = EINVAL; 55 goto release; 56 } 57 switch (req) { 58 59 case PRU_ATTACH: 60 if (unp) { 61 error = EISCONN; 62 break; 63 } 64 error = unp_attach(so); 65 break; 66 67 case PRU_DETACH: 68 unp_detach(unp); 69 break; 70 71 case PRU_BIND: 72 error = unp_bind(unp, nam, p); 73 break; 74 75 case PRU_LISTEN: 76 if (unp->unp_vnode == 0) 77 error = EINVAL; 78 break; 79 80 case PRU_CONNECT: 81 error = unp_connect(so, nam, p); 82 break; 83 84 case PRU_CONNECT2: 85 error = unp_connect2(so, (struct socket *)nam); 86 break; 87 88 case PRU_DISCONNECT: 89 unp_disconnect(unp); 90 break; 91 92 case PRU_ACCEPT: 93 /* 94 * Pass back name of connected socket, 95 * if it was bound and we are still connected 96 * (our peer may have closed already!). 97 */ 98 if (unp->unp_conn && unp->unp_conn->unp_addr) { 99 nam->m_len = unp->unp_conn->unp_addr->m_len; 100 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 101 mtod(nam, caddr_t), (unsigned)nam->m_len); 102 } else { 103 nam->m_len = sizeof(sun_noname); 104 *(mtod(nam, struct sockaddr *)) = sun_noname; 105 } 106 break; 107 108 case PRU_SHUTDOWN: 109 socantsendmore(so); 110 unp_shutdown(unp); 111 break; 112 113 case PRU_RCVD: 114 switch (so->so_type) { 115 116 case SOCK_DGRAM: 117 panic("uipc 1"); 118 /*NOTREACHED*/ 119 120 case SOCK_STREAM: 121 #define rcv (&so->so_rcv) 122 #define snd (&so2->so_snd) 123 if (unp->unp_conn == 0) 124 break; 125 so2 = unp->unp_conn->unp_socket; 126 /* 127 * Adjust backpressure on sender 128 * and wakeup any waiting to write. 129 */ 130 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 131 unp->unp_mbcnt = rcv->sb_mbcnt; 132 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc; 133 unp->unp_cc = rcv->sb_cc; 134 sowwakeup(so2); 135 #undef snd 136 #undef rcv 137 break; 138 139 default: 140 panic("uipc 2"); 141 } 142 break; 143 144 case PRU_SEND: 145 if (control && (error = unp_internalize(control, p))) 146 break; 147 switch (so->so_type) { 148 149 case SOCK_DGRAM: { 150 struct sockaddr *from; 151 152 if (nam) { 153 if (unp->unp_conn) { 154 error = EISCONN; 155 break; 156 } 157 error = unp_connect(so, nam, p); 158 if (error) 159 break; 160 } else { 161 if (unp->unp_conn == 0) { 162 error = ENOTCONN; 163 break; 164 } 165 } 166 so2 = unp->unp_conn->unp_socket; 167 if (unp->unp_addr) 168 from = mtod(unp->unp_addr, struct sockaddr *); 169 else 170 from = &sun_noname; 171 if (sbappendaddr(&so2->so_rcv, from, m, control)) { 172 sorwakeup(so2); 173 m = 0; 174 control = 0; 175 } else 176 error = ENOBUFS; 177 if (nam) 178 unp_disconnect(unp); 179 break; 180 } 181 182 case SOCK_STREAM: 183 #define rcv (&so2->so_rcv) 184 #define snd (&so->so_snd) 185 if (so->so_state & SS_CANTSENDMORE) { 186 error = EPIPE; 187 break; 188 } 189 if (unp->unp_conn == 0) 190 panic("uipc 3"); 191 so2 = unp->unp_conn->unp_socket; 192 /* 193 * Send to paired receive port, and then reduce 194 * send buffer hiwater marks to maintain backpressure. 195 * Wake up readers. 196 */ 197 if (control) { 198 if (sbappendcontrol(rcv, m, control)) 199 control = 0; 200 } else 201 sbappend(rcv, m); 202 snd->sb_mbmax -= 203 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 204 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 205 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc; 206 unp->unp_conn->unp_cc = rcv->sb_cc; 207 sorwakeup(so2); 208 m = 0; 209 #undef snd 210 #undef rcv 211 break; 212 213 default: 214 panic("uipc 4"); 215 } 216 break; 217 218 case PRU_ABORT: 219 unp_drop(unp, ECONNABORTED); 220 break; 221 222 case PRU_SENSE: 223 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 224 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { 225 so2 = unp->unp_conn->unp_socket; 226 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 227 } 228 ((struct stat *) m)->st_dev = NODEV; 229 if (unp->unp_ino == 0) 230 unp->unp_ino = unp_ino++; 231 ((struct stat *) m)->st_ino = unp->unp_ino; 232 return (0); 233 234 case PRU_RCVOOB: 235 return (EOPNOTSUPP); 236 237 case PRU_SENDOOB: 238 error = EOPNOTSUPP; 239 break; 240 241 case PRU_SOCKADDR: 242 if (unp->unp_addr) { 243 nam->m_len = unp->unp_addr->m_len; 244 bcopy(mtod(unp->unp_addr, caddr_t), 245 mtod(nam, caddr_t), (unsigned)nam->m_len); 246 } else 247 nam->m_len = 0; 248 break; 249 250 case PRU_PEERADDR: 251 if (unp->unp_conn && unp->unp_conn->unp_addr) { 252 nam->m_len = unp->unp_conn->unp_addr->m_len; 253 bcopy(mtod(unp->unp_conn->unp_addr, caddr_t), 254 mtod(nam, caddr_t), (unsigned)nam->m_len); 255 } else 256 nam->m_len = 0; 257 break; 258 259 case PRU_SLOWTIMO: 260 break; 261 262 default: 263 panic("piusrreq"); 264 } 265 release: 266 if (control) 267 m_freem(control); 268 if (m) 269 m_freem(m); 270 return (error); 271 } 272 273 /* 274 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 275 * for stream sockets, although the total for sender and receiver is 276 * actually only PIPSIZ. 277 * Datagram sockets really use the sendspace as the maximum datagram size, 278 * and don't really want to reserve the sendspace. Their recvspace should 279 * be large enough for at least one max-size datagram plus address. 280 */ 281 #define PIPSIZ 4096 282 u_long unpst_sendspace = PIPSIZ; 283 u_long unpst_recvspace = PIPSIZ; 284 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 285 u_long unpdg_recvspace = 4*1024; 286 287 int unp_rights; /* file descriptors in flight */ 288 289 unp_attach(so) 290 struct socket *so; 291 { 292 register struct mbuf *m; 293 register struct unpcb *unp; 294 int error; 295 296 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 297 switch (so->so_type) { 298 299 case SOCK_STREAM: 300 error = soreserve(so, unpst_sendspace, unpst_recvspace); 301 break; 302 303 case SOCK_DGRAM: 304 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 305 break; 306 } 307 if (error) 308 return (error); 309 } 310 m = m_getclr(M_DONTWAIT, MT_PCB); 311 if (m == NULL) 312 return (ENOBUFS); 313 unp = mtod(m, struct unpcb *); 314 so->so_pcb = (caddr_t)unp; 315 unp->unp_socket = so; 316 return (0); 317 } 318 319 unp_detach(unp) 320 register struct unpcb *unp; 321 { 322 323 if (unp->unp_vnode) { 324 unp->unp_vnode->v_socket = 0; 325 vrele(unp->unp_vnode); 326 unp->unp_vnode = 0; 327 } 328 if (unp->unp_conn) 329 unp_disconnect(unp); 330 while (unp->unp_refs) 331 unp_drop(unp->unp_refs, ECONNRESET); 332 soisdisconnected(unp->unp_socket); 333 unp->unp_socket->so_pcb = 0; 334 m_freem(unp->unp_addr); 335 (void) m_free(dtom(unp)); 336 if (unp_rights) 337 unp_gc(); 338 } 339 340 unp_bind(unp, nam, p) 341 struct unpcb *unp; 342 struct mbuf *nam; 343 struct proc *p; 344 { 345 struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 346 register struct vnode *vp; 347 register struct nameidata *ndp; 348 struct vattr vattr; 349 int error; 350 struct nameidata nd; 351 352 ndp = &nd; 353 ndp->ni_dirp = soun->sun_path; 354 if (unp->unp_vnode != NULL) 355 return (EINVAL); 356 if (nam->m_len == MLEN) { 357 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 358 return (EINVAL); 359 } else 360 *(mtod(nam, caddr_t) + nam->m_len) = 0; 361 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 362 ndp->ni_nameiop = CREATE | FOLLOW | LOCKPARENT; 363 ndp->ni_segflg = UIO_SYSSPACE; 364 if (error = namei(ndp, p)) 365 return (error); 366 vp = ndp->ni_vp; 367 if (vp != NULL) { 368 VOP_ABORTOP(ndp); 369 if (ndp->ni_dvp == vp) 370 vrele(ndp->ni_dvp); 371 else 372 vput(ndp->ni_dvp); 373 vrele(vp); 374 return (EADDRINUSE); 375 } 376 VATTR_NULL(&vattr); 377 vattr.va_type = VSOCK; 378 vattr.va_mode = 0777; 379 if (error = VOP_CREATE(ndp, &vattr, p)) 380 return (error); 381 vp = ndp->ni_vp; 382 vp->v_socket = unp->unp_socket; 383 unp->unp_vnode = vp; 384 unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL); 385 VOP_UNLOCK(vp); 386 return (0); 387 } 388 389 unp_connect(so, nam, p) 390 struct socket *so; 391 struct mbuf *nam; 392 struct proc *p; 393 { 394 register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *); 395 register struct vnode *vp; 396 register struct socket *so2, *so3; 397 register struct nameidata *ndp; 398 struct unpcb *unp2, *unp3; 399 int error; 400 struct nameidata nd; 401 402 ndp = &nd; 403 ndp->ni_dirp = soun->sun_path; 404 if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) { /* XXX */ 405 if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0) 406 return (EMSGSIZE); 407 } else 408 *(mtod(nam, caddr_t) + nam->m_len) = 0; 409 ndp->ni_nameiop = LOOKUP | FOLLOW | LOCKLEAF; 410 ndp->ni_segflg = UIO_SYSSPACE; 411 if (error = namei(ndp, p)) 412 return (error); 413 vp = ndp->ni_vp; 414 if (vp->v_type != VSOCK) { 415 error = ENOTSOCK; 416 goto bad; 417 } 418 if (error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) 419 goto bad; 420 so2 = vp->v_socket; 421 if (so2 == 0) { 422 error = ECONNREFUSED; 423 goto bad; 424 } 425 if (so->so_type != so2->so_type) { 426 error = EPROTOTYPE; 427 goto bad; 428 } 429 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 430 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 431 (so3 = sonewconn(so2, 0)) == 0) { 432 error = ECONNREFUSED; 433 goto bad; 434 } 435 unp2 = sotounpcb(so2); 436 unp3 = sotounpcb(so3); 437 if (unp2->unp_addr) 438 unp3->unp_addr = 439 m_copy(unp2->unp_addr, 0, (int)M_COPYALL); 440 so2 = so3; 441 } 442 error = unp_connect2(so, so2); 443 bad: 444 vput(vp); 445 return (error); 446 } 447 448 unp_connect2(so, so2) 449 register struct socket *so; 450 register struct socket *so2; 451 { 452 register struct unpcb *unp = sotounpcb(so); 453 register struct unpcb *unp2; 454 455 if (so2->so_type != so->so_type) 456 return (EPROTOTYPE); 457 unp2 = sotounpcb(so2); 458 unp->unp_conn = unp2; 459 switch (so->so_type) { 460 461 case SOCK_DGRAM: 462 unp->unp_nextref = unp2->unp_refs; 463 unp2->unp_refs = unp; 464 soisconnected(so); 465 break; 466 467 case SOCK_STREAM: 468 unp2->unp_conn = unp; 469 soisconnected(so); 470 soisconnected(so2); 471 break; 472 473 default: 474 panic("unp_connect2"); 475 } 476 return (0); 477 } 478 479 unp_disconnect(unp) 480 struct unpcb *unp; 481 { 482 register struct unpcb *unp2 = unp->unp_conn; 483 484 if (unp2 == 0) 485 return; 486 unp->unp_conn = 0; 487 switch (unp->unp_socket->so_type) { 488 489 case SOCK_DGRAM: 490 if (unp2->unp_refs == unp) 491 unp2->unp_refs = unp->unp_nextref; 492 else { 493 unp2 = unp2->unp_refs; 494 for (;;) { 495 if (unp2 == 0) 496 panic("unp_disconnect"); 497 if (unp2->unp_nextref == unp) 498 break; 499 unp2 = unp2->unp_nextref; 500 } 501 unp2->unp_nextref = unp->unp_nextref; 502 } 503 unp->unp_nextref = 0; 504 unp->unp_socket->so_state &= ~SS_ISCONNECTED; 505 break; 506 507 case SOCK_STREAM: 508 soisdisconnected(unp->unp_socket); 509 unp2->unp_conn = 0; 510 soisdisconnected(unp2->unp_socket); 511 break; 512 } 513 } 514 515 #ifdef notdef 516 unp_abort(unp) 517 struct unpcb *unp; 518 { 519 520 unp_detach(unp); 521 } 522 #endif 523 524 unp_shutdown(unp) 525 struct unpcb *unp; 526 { 527 struct socket *so; 528 529 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && 530 (so = unp->unp_conn->unp_socket)) 531 socantrcvmore(so); 532 } 533 534 unp_drop(unp, errno) 535 struct unpcb *unp; 536 int errno; 537 { 538 struct socket *so = unp->unp_socket; 539 540 so->so_error = errno; 541 unp_disconnect(unp); 542 if (so->so_head) { 543 so->so_pcb = (caddr_t) 0; 544 m_freem(unp->unp_addr); 545 (void) m_free(dtom(unp)); 546 sofree(so); 547 } 548 } 549 550 #ifdef notdef 551 unp_drain() 552 { 553 554 } 555 #endif 556 557 unp_externalize(rights) 558 struct mbuf *rights; 559 { 560 struct proc *p = curproc; /* XXX */ 561 register int i; 562 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *); 563 register struct file **rp = (struct file **)(cm + 1); 564 register struct file *fp; 565 int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int); 566 int f; 567 568 if (fdavail(p, newfds)) { 569 for (i = 0; i < newfds; i++) { 570 fp = *rp; 571 unp_discard(fp); 572 *rp++ = 0; 573 } 574 return (EMSGSIZE); 575 } 576 for (i = 0; i < newfds; i++) { 577 if (fdalloc(p, 0, &f)) 578 panic("unp_externalize"); 579 fp = *rp; 580 p->p_fd->fd_ofiles[f] = fp; 581 fp->f_msgcount--; 582 unp_rights--; 583 *(int *)rp++ = f; 584 } 585 return (0); 586 } 587 588 unp_internalize(control, p) 589 struct mbuf *control; 590 struct proc *p; 591 { 592 struct filedesc *fdp = p->p_fd; 593 register struct cmsghdr *cm = mtod(control, struct cmsghdr *); 594 register struct file **rp; 595 register struct file *fp; 596 register int i, fd; 597 int oldfds; 598 599 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 600 cm->cmsg_len != control->m_len) 601 return (EINVAL); 602 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int); 603 rp = (struct file **)(cm + 1); 604 for (i = 0; i < oldfds; i++) { 605 fd = *(int *)rp++; 606 if ((unsigned)fd >= fdp->fd_nfiles || 607 fdp->fd_ofiles[fd] == NULL) 608 return (EBADF); 609 } 610 rp = (struct file **)(cm + 1); 611 for (i = 0; i < oldfds; i++) { 612 fp = fdp->fd_ofiles[*(int *)rp]; 613 *rp++ = fp; 614 fp->f_count++; 615 fp->f_msgcount++; 616 unp_rights++; 617 } 618 return (0); 619 } 620 621 int unp_defer, unp_gcing; 622 int unp_mark(); 623 extern struct domain unixdomain; 624 625 unp_gc() 626 { 627 register struct file *fp; 628 register struct socket *so; 629 630 if (unp_gcing) 631 return; 632 unp_gcing = 1; 633 restart: 634 unp_defer = 0; 635 for (fp = filehead; fp; fp = fp->f_filef) 636 fp->f_flag &= ~(FMARK|FDEFER); 637 do { 638 for (fp = filehead; fp; fp = fp->f_filef) { 639 if (fp->f_count == 0) 640 continue; 641 if (fp->f_flag & FDEFER) { 642 fp->f_flag &= ~FDEFER; 643 unp_defer--; 644 } else { 645 if (fp->f_flag & FMARK) 646 continue; 647 if (fp->f_count == fp->f_msgcount) 648 continue; 649 fp->f_flag |= FMARK; 650 } 651 if (fp->f_type != DTYPE_SOCKET || 652 (so = (struct socket *)fp->f_data) == 0) 653 continue; 654 if (so->so_proto->pr_domain != &unixdomain || 655 (so->so_proto->pr_flags&PR_RIGHTS) == 0) 656 continue; 657 #ifdef notdef 658 if (so->so_rcv.sb_flags & SB_LOCK) { 659 /* 660 * This is problematical; it's not clear 661 * we need to wait for the sockbuf to be 662 * unlocked (on a uniprocessor, at least), 663 * and it's also not clear what to do 664 * if sbwait returns an error due to receipt 665 * of a signal. If sbwait does return 666 * an error, we'll go into an infinite 667 * loop. Delete all of this for now. 668 */ 669 (void) sbwait(&so->so_rcv); 670 goto restart; 671 } 672 #endif 673 unp_scan(so->so_rcv.sb_mb, unp_mark); 674 } 675 } while (unp_defer); 676 for (fp = filehead; fp; fp = fp->f_filef) { 677 if (fp->f_count == 0) 678 continue; 679 if (fp->f_count == fp->f_msgcount && (fp->f_flag & FMARK) == 0) 680 while (fp->f_msgcount) 681 unp_discard(fp); 682 } 683 unp_gcing = 0; 684 } 685 686 unp_dispose(m) 687 struct mbuf *m; 688 { 689 int unp_discard(); 690 691 if (m) 692 unp_scan(m, unp_discard); 693 } 694 695 unp_scan(m0, op) 696 register struct mbuf *m0; 697 int (*op)(); 698 { 699 register struct mbuf *m; 700 register struct file **rp; 701 register struct cmsghdr *cm; 702 register int i; 703 int qfds; 704 705 while (m0) { 706 for (m = m0; m; m = m->m_next) 707 if (m->m_type == MT_CONTROL && 708 m->m_len >= sizeof(*cm)) { 709 cm = mtod(m, struct cmsghdr *); 710 if (cm->cmsg_level != SOL_SOCKET || 711 cm->cmsg_type != SCM_RIGHTS) 712 continue; 713 qfds = (cm->cmsg_len - sizeof *cm) 714 / sizeof (struct file *); 715 rp = (struct file **)(cm + 1); 716 for (i = 0; i < qfds; i++) 717 (*op)(*rp++); 718 break; /* XXX, but saves time */ 719 } 720 m0 = m0->m_act; 721 } 722 } 723 724 unp_mark(fp) 725 struct file *fp; 726 { 727 728 if (fp->f_flag & FMARK) 729 return; 730 unp_defer++; 731 fp->f_flag |= (FMARK|FDEFER); 732 } 733 734 unp_discard(fp) 735 struct file *fp; 736 { 737 738 fp->f_msgcount--; 739 unp_rights--; 740 (void) closef(fp); 741 } 742