1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #include <sys/types.h> 28 #include <sys/param.h> 29 #include <sys/signal.h> 30 #include <sys/cmn_err.h> 31 32 #include <sys/stropts.h> 33 #include <sys/socket.h> 34 #include <sys/socketvar.h> 35 #include <sys/sockio.h> 36 #include <sys/strsubr.h> 37 #include <sys/strsun.h> 38 #include <sys/atomic.h> 39 #include <sys/tihdr.h> 40 41 #include <fs/sockfs/sockcommon.h> 42 #include <fs/sockfs/socktpi.h> 43 #include <fs/sockfs/sodirect.h> 44 #include <sys/ddi.h> 45 #include <inet/ip.h> 46 #include <sys/time.h> 47 #include <sys/cmn_err.h> 48 49 #ifdef SOCK_TEST 50 extern int do_useracc; 51 extern clock_t sock_test_timelimit; 52 #endif /* SOCK_TEST */ 53 54 #define MBLK_PULL_LEN 64 55 uint32_t so_mblk_pull_len = MBLK_PULL_LEN; 56 57 #ifdef DEBUG 58 boolean_t so_debug_length = B_FALSE; 59 static boolean_t so_check_length(sonode_t *so); 60 #endif 61 62 int 63 so_acceptq_enqueue_locked(struct sonode *so, struct sonode *nso) 64 { 65 ASSERT(MUTEX_HELD(&so->so_acceptq_lock)); 66 ASSERT(nso->so_acceptq_next == NULL); 67 68 *so->so_acceptq_tail = nso; 69 so->so_acceptq_tail = &nso->so_acceptq_next; 70 so->so_acceptq_len++; 71 72 /* Wakeup a single consumer */ 73 cv_signal(&so->so_acceptq_cv); 74 75 return (so->so_acceptq_len); 76 } 77 78 /* 79 * int so_acceptq_enqueue(struct sonode *so, struct sonode *nso) 80 * 81 * Enqueue an incoming connection on a listening socket. 82 * 83 * Arguments: 84 * so - listening socket 85 * nso - new connection 86 * 87 * Returns: 88 * Number of queued connections, including the new connection 89 */ 90 int 91 so_acceptq_enqueue(struct sonode *so, struct sonode *nso) 92 { 93 int conns; 94 95 mutex_enter(&so->so_acceptq_lock); 96 conns = so_acceptq_enqueue_locked(so, nso); 97 mutex_exit(&so->so_acceptq_lock); 98 99 return (conns); 100 } 101 102 static int 103 so_acceptq_dequeue_locked(struct sonode *so, boolean_t dontblock, 104 struct sonode **nsop) 105 { 106 struct sonode *nso = NULL; 107 108 *nsop = NULL; 109 ASSERT(MUTEX_HELD(&so->so_acceptq_lock)); 110 while ((nso = so->so_acceptq_head) == NULL) { 111 /* 112 * No need to check so_error here, because it is not 113 * possible for a listening socket to be reset or otherwise 114 * disconnected. 115 * 116 * So now we just need check if it's ok to wait. 117 */ 118 if (dontblock) 119 return (EWOULDBLOCK); 120 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING)) 121 return (EINTR); 122 123 if (cv_wait_sig_swap(&so->so_acceptq_cv, 124 &so->so_acceptq_lock) == 0) 125 return (EINTR); 126 } 127 128 ASSERT(nso != NULL); 129 so->so_acceptq_head = nso->so_acceptq_next; 130 nso->so_acceptq_next = NULL; 131 132 if (so->so_acceptq_head == NULL) { 133 ASSERT(so->so_acceptq_tail == &nso->so_acceptq_next); 134 so->so_acceptq_tail = &so->so_acceptq_head; 135 } 136 ASSERT(so->so_acceptq_len > 0); 137 --so->so_acceptq_len; 138 139 *nsop = nso; 140 141 return (0); 142 } 143 144 /* 145 * int so_acceptq_dequeue(struct sonode *, boolean_t, struct sonode **) 146 * 147 * Pulls a connection off of the accept queue. 148 * 149 * Arguments: 150 * so - listening socket 151 * dontblock - indicate whether it's ok to sleep if there are no 152 * connections on the queue 153 * nsop - Value-return argument 154 * 155 * Return values: 156 * 0 when a connection is successfully dequeued, in which case nsop 157 * is set to point to the new connection. Upon failure a non-zero 158 * value is returned, and the value of nsop is set to NULL. 159 * 160 * Note: 161 * so_acceptq_dequeue() may return prematurly if the socket is falling 162 * back to TPI. 163 */ 164 int 165 so_acceptq_dequeue(struct sonode *so, boolean_t dontblock, 166 struct sonode **nsop) 167 { 168 int error; 169 170 mutex_enter(&so->so_acceptq_lock); 171 error = so_acceptq_dequeue_locked(so, dontblock, nsop); 172 mutex_exit(&so->so_acceptq_lock); 173 174 return (error); 175 } 176 177 /* 178 * void so_acceptq_flush(struct sonode *so, boolean_t doclose) 179 * 180 * Removes all pending connections from a listening socket, and 181 * frees the associated resources. 182 * 183 * Arguments 184 * so - listening socket 185 * doclose - make a close downcall for each socket on the accept queue 186 * (Note, only SCTP and SDP sockets rely on this) 187 * 188 * Return values: 189 * None. 190 * 191 * Note: 192 * The caller has to ensure that no calls to so_acceptq_enqueue() or 193 * so_acceptq_dequeue() occur while the accept queue is being flushed. 194 * So either the socket needs to be in a state where no operations 195 * would come in, or so_lock needs to be obtained. 196 */ 197 void 198 so_acceptq_flush(struct sonode *so, boolean_t doclose) 199 { 200 struct sonode *nso; 201 202 while ((nso = so->so_acceptq_head) != NULL) { 203 so->so_acceptq_head = nso->so_acceptq_next; 204 nso->so_acceptq_next = NULL; 205 206 if (doclose) { 207 socket_close(nso, 0, CRED()); 208 } else { 209 /* 210 * Since the socket is on the accept queue, there can 211 * only be one reference. We drop the reference and 212 * just blow off the socket. 213 */ 214 ASSERT(nso->so_count == 1); 215 nso->so_count--; 216 } 217 socket_destroy(nso); 218 } 219 220 so->so_acceptq_head = NULL; 221 so->so_acceptq_tail = &so->so_acceptq_head; 222 so->so_acceptq_len = 0; 223 } 224 225 int 226 so_wait_connected_locked(struct sonode *so, boolean_t nonblock, 227 sock_connid_t id) 228 { 229 ASSERT(MUTEX_HELD(&so->so_lock)); 230 231 /* 232 * The protocol has notified us that a connection attempt is being 233 * made, so before we wait for a notification to arrive we must 234 * clear out any errors associated with earlier connection attempts. 235 */ 236 if (so->so_error != 0 && SOCK_CONNID_LT(so->so_proto_connid, id)) 237 so->so_error = 0; 238 239 while (SOCK_CONNID_LT(so->so_proto_connid, id)) { 240 if (nonblock) 241 return (EINPROGRESS); 242 243 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING)) 244 return (EINTR); 245 246 if (cv_wait_sig_swap(&so->so_state_cv, &so->so_lock) == 0) 247 return (EINTR); 248 } 249 250 if (so->so_error != 0) 251 return (sogeterr(so, B_TRUE)); 252 /* 253 * Under normal circumstances, so_error should contain an error 254 * in case the connect failed. However, it is possible for another 255 * thread to come in a consume the error, so generate a sensible 256 * error in that case. 257 */ 258 if ((so->so_state & SS_ISCONNECTED) == 0) 259 return (ECONNREFUSED); 260 261 return (0); 262 } 263 264 /* 265 * int so_wait_connected(struct sonode *so, boolean_t nonblock, 266 * sock_connid_t id) 267 * 268 * Wait until the socket is connected or an error has occured. 269 * 270 * Arguments: 271 * so - socket 272 * nonblock - indicate whether it's ok to sleep if the connection has 273 * not yet been established 274 * gen - generation number that was returned by the protocol 275 * when the operation was started 276 * 277 * Returns: 278 * 0 if the connection attempt was successful, or an error indicating why 279 * the connection attempt failed. 280 */ 281 int 282 so_wait_connected(struct sonode *so, boolean_t nonblock, sock_connid_t id) 283 { 284 int error; 285 286 mutex_enter(&so->so_lock); 287 error = so_wait_connected_locked(so, nonblock, id); 288 mutex_exit(&so->so_lock); 289 290 return (error); 291 } 292 293 int 294 so_snd_wait_qnotfull_locked(struct sonode *so, boolean_t dontblock) 295 { 296 int error; 297 298 ASSERT(MUTEX_HELD(&so->so_lock)); 299 while (so->so_snd_qfull) { 300 if (so->so_state & SS_CANTSENDMORE) 301 return (EPIPE); 302 if (dontblock) 303 return (EWOULDBLOCK); 304 305 if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING)) 306 return (EINTR); 307 308 if (so->so_sndtimeo == 0) { 309 /* 310 * Zero means disable timeout. 311 */ 312 error = cv_wait_sig(&so->so_snd_cv, &so->so_lock); 313 } else { 314 clock_t now; 315 316 time_to_wait(&now, so->so_sndtimeo); 317 error = cv_timedwait_sig(&so->so_snd_cv, &so->so_lock, 318 now); 319 } 320 if (error == 0) 321 return (EINTR); 322 else if (error == -1) 323 return (EAGAIN); 324 } 325 return (0); 326 } 327 328 /* 329 * int so_wait_sendbuf(struct sonode *so, boolean_t dontblock) 330 * 331 * Wait for the transport to notify us about send buffers becoming 332 * available. 333 */ 334 int 335 so_snd_wait_qnotfull(struct sonode *so, boolean_t dontblock) 336 { 337 int error = 0; 338 339 mutex_enter(&so->so_lock); 340 if (so->so_snd_qfull) { 341 so->so_snd_wakeup = B_TRUE; 342 error = so_snd_wait_qnotfull_locked(so, dontblock); 343 so->so_snd_wakeup = B_FALSE; 344 } 345 mutex_exit(&so->so_lock); 346 347 return (error); 348 } 349 350 void 351 so_snd_qfull(struct sonode *so) 352 { 353 mutex_enter(&so->so_lock); 354 so->so_snd_qfull = B_TRUE; 355 mutex_exit(&so->so_lock); 356 } 357 358 void 359 so_snd_qnotfull(struct sonode *so) 360 { 361 mutex_enter(&so->so_lock); 362 so->so_snd_qfull = B_FALSE; 363 /* wake up everyone waiting for buffers */ 364 cv_broadcast(&so->so_snd_cv); 365 mutex_exit(&so->so_lock); 366 } 367 368 /* 369 * Change the process/process group to which SIGIO is sent. 370 */ 371 int 372 socket_chgpgrp(struct sonode *so, pid_t pid) 373 { 374 int error; 375 376 ASSERT(MUTEX_HELD(&so->so_lock)); 377 if (pid != 0) { 378 /* 379 * Permissions check by sending signal 0. 380 * Note that when kill fails it does a 381 * set_errno causing the system call to fail. 382 */ 383 error = kill(pid, 0); 384 if (error != 0) { 385 return (error); 386 } 387 } 388 so->so_pgrp = pid; 389 return (0); 390 } 391 392 393 /* 394 * Generate a SIGIO, for 'writable' events include siginfo structure, 395 * for read events just send the signal. 396 */ 397 /*ARGSUSED*/ 398 static void 399 socket_sigproc(proc_t *proc, int event) 400 { 401 k_siginfo_t info; 402 403 ASSERT(event & (SOCKETSIG_WRITE | SOCKETSIG_READ | SOCKETSIG_URG)); 404 405 if (event & SOCKETSIG_WRITE) { 406 info.si_signo = SIGPOLL; 407 info.si_code = POLL_OUT; 408 info.si_errno = 0; 409 info.si_fd = 0; 410 info.si_band = 0; 411 sigaddq(proc, NULL, &info, KM_NOSLEEP); 412 } 413 if (event & SOCKETSIG_READ) { 414 sigtoproc(proc, NULL, SIGPOLL); 415 } 416 if (event & SOCKETSIG_URG) { 417 sigtoproc(proc, NULL, SIGURG); 418 } 419 } 420 421 void 422 socket_sendsig(struct sonode *so, int event) 423 { 424 proc_t *proc; 425 426 ASSERT(MUTEX_HELD(&so->so_lock)); 427 428 if (so->so_pgrp == 0 || (!(so->so_state & SS_ASYNC) && 429 event != SOCKETSIG_URG)) { 430 return; 431 } 432 433 dprint(3, ("sending sig %d to %d\n", event, so->so_pgrp)); 434 435 if (so->so_pgrp > 0) { 436 /* 437 * XXX This unfortunately still generates 438 * a signal when a fd is closed but 439 * the proc is active. 440 */ 441 mutex_enter(&pidlock); 442 proc = prfind(so->so_pgrp); 443 if (proc == NULL) { 444 mutex_exit(&pidlock); 445 return; 446 } 447 mutex_enter(&proc->p_lock); 448 mutex_exit(&pidlock); 449 socket_sigproc(proc, event); 450 mutex_exit(&proc->p_lock); 451 } else { 452 /* 453 * Send to process group. Hold pidlock across 454 * calls to socket_sigproc(). 455 */ 456 pid_t pgrp = -so->so_pgrp; 457 458 mutex_enter(&pidlock); 459 proc = pgfind(pgrp); 460 while (proc != NULL) { 461 mutex_enter(&proc->p_lock); 462 socket_sigproc(proc, event); 463 mutex_exit(&proc->p_lock); 464 proc = proc->p_pglink; 465 } 466 mutex_exit(&pidlock); 467 } 468 } 469 470 #define MIN(a, b) ((a) < (b) ? (a) : (b)) 471 /* Copy userdata into a new mblk_t */ 472 mblk_t * 473 socopyinuio(uio_t *uiop, ssize_t iosize, size_t wroff, ssize_t maxblk, 474 size_t tail_len, int *errorp, cred_t *cr) 475 { 476 mblk_t *head = NULL, **tail = &head; 477 478 ASSERT(iosize == INFPSZ || iosize > 0); 479 480 if (iosize == INFPSZ || iosize > uiop->uio_resid) 481 iosize = uiop->uio_resid; 482 483 if (maxblk == INFPSZ) 484 maxblk = iosize; 485 486 /* Nothing to do in these cases, so we're done */ 487 if (iosize < 0 || maxblk < 0 || (maxblk == 0 && iosize > 0)) 488 goto done; 489 490 /* 491 * We will enter the loop below if iosize is 0; it will allocate an 492 * empty message block and call uiomove(9F) which will just return. 493 * We could avoid that with an extra check but would only slow 494 * down the much more likely case where iosize is larger than 0. 495 */ 496 do { 497 ssize_t blocksize; 498 mblk_t *mp; 499 500 blocksize = MIN(iosize, maxblk); 501 ASSERT(blocksize >= 0); 502 if (is_system_labeled()) 503 mp = allocb_cred(wroff + blocksize + tail_len, 504 cr, curproc->p_pid); 505 else 506 mp = allocb(wroff + blocksize + tail_len, BPRI_MED); 507 if (mp == NULL) { 508 *errorp = ENOMEM; 509 return (head); 510 } 511 mp->b_rptr += wroff; 512 mp->b_wptr = mp->b_rptr + blocksize; 513 514 *tail = mp; 515 tail = &mp->b_cont; 516 517 /* uiomove(9F) either returns 0 or EFAULT */ 518 if ((*errorp = uiomove(mp->b_rptr, (size_t)blocksize, 519 UIO_WRITE, uiop)) != 0) { 520 ASSERT(*errorp != ENOMEM); 521 freemsg(head); 522 return (NULL); 523 } 524 525 iosize -= blocksize; 526 } while (iosize > 0); 527 528 done: 529 *errorp = 0; 530 return (head); 531 } 532 533 mblk_t * 534 socopyoutuio(mblk_t *mp, struct uio *uiop, ssize_t max_read, int *errorp) 535 { 536 int error; 537 ptrdiff_t n; 538 mblk_t *nmp; 539 540 ASSERT(mp->b_wptr >= mp->b_rptr); 541 542 /* 543 * max_read is the offset of the oobmark and read can not go pass 544 * the oobmark. 545 */ 546 if (max_read == INFPSZ || max_read > uiop->uio_resid) 547 max_read = uiop->uio_resid; 548 549 do { 550 if ((n = MIN(max_read, MBLKL(mp))) != 0) { 551 ASSERT(n > 0); 552 553 error = uiomove(mp->b_rptr, n, UIO_READ, uiop); 554 if (error != 0) { 555 freemsg(mp); 556 *errorp = error; 557 return (NULL); 558 } 559 } 560 561 mp->b_rptr += n; 562 max_read -= n; 563 while (mp != NULL && (mp->b_rptr >= mp->b_wptr)) { 564 /* 565 * get rid of zero length mblks 566 */ 567 nmp = mp; 568 mp = mp->b_cont; 569 freeb(nmp); 570 } 571 } while (mp != NULL && max_read > 0); 572 573 *errorp = 0; 574 return (mp); 575 } 576 577 static void 578 so_prepend_msg(struct sonode *so, mblk_t *mp, mblk_t *last_tail) 579 { 580 ASSERT(last_tail != NULL); 581 mp->b_next = so->so_rcv_q_head; 582 mp->b_prev = last_tail; 583 ASSERT(!(DB_FLAGS(mp) & DBLK_UIOA)); 584 585 if (so->so_rcv_q_head == NULL) { 586 ASSERT(so->so_rcv_q_last_head == NULL); 587 so->so_rcv_q_last_head = mp; 588 #ifdef DEBUG 589 } else { 590 ASSERT(!(DB_FLAGS(so->so_rcv_q_head) & DBLK_UIOA)); 591 #endif 592 } 593 so->so_rcv_q_head = mp; 594 595 #ifdef DEBUG 596 if (so_debug_length) { 597 mutex_enter(&so->so_lock); 598 ASSERT(so_check_length(so)); 599 mutex_exit(&so->so_lock); 600 } 601 #endif 602 } 603 604 /* 605 * Move a mblk chain (mp_head, mp_last_head) to the sonode's rcv queue so it 606 * can be processed by so_dequeue_msg(). 607 */ 608 void 609 so_process_new_message(struct sonode *so, mblk_t *mp_head, mblk_t *mp_last_head) 610 { 611 ASSERT(mp_head->b_prev != NULL); 612 if (so->so_rcv_q_head == NULL) { 613 so->so_rcv_q_head = mp_head; 614 so->so_rcv_q_last_head = mp_last_head; 615 ASSERT(so->so_rcv_q_last_head->b_prev != NULL); 616 } else { 617 boolean_t flag_equal = ((DB_FLAGS(mp_head) & DBLK_UIOA) == 618 (DB_FLAGS(so->so_rcv_q_last_head) & DBLK_UIOA)); 619 620 if (mp_head->b_next == NULL && 621 DB_TYPE(mp_head) == M_DATA && 622 DB_TYPE(so->so_rcv_q_last_head) == M_DATA && flag_equal) { 623 so->so_rcv_q_last_head->b_prev->b_cont = mp_head; 624 so->so_rcv_q_last_head->b_prev = mp_head->b_prev; 625 mp_head->b_prev = NULL; 626 } else if (flag_equal && (DB_FLAGS(mp_head) & DBLK_UIOA)) { 627 /* 628 * Append to last_head if more than one mblks, and both 629 * mp_head and last_head are I/OAT mblks. 630 */ 631 ASSERT(mp_head->b_next != NULL); 632 so->so_rcv_q_last_head->b_prev->b_cont = mp_head; 633 so->so_rcv_q_last_head->b_prev = mp_head->b_prev; 634 mp_head->b_prev = NULL; 635 636 so->so_rcv_q_last_head->b_next = mp_head->b_next; 637 mp_head->b_next = NULL; 638 so->so_rcv_q_last_head = mp_last_head; 639 } else { 640 #ifdef DEBUG 641 { 642 mblk_t *tmp_mblk; 643 tmp_mblk = mp_head; 644 while (tmp_mblk != NULL) { 645 ASSERT(tmp_mblk->b_prev != NULL); 646 tmp_mblk = tmp_mblk->b_next; 647 } 648 } 649 #endif 650 so->so_rcv_q_last_head->b_next = mp_head; 651 so->so_rcv_q_last_head = mp_last_head; 652 } 653 } 654 } 655 656 /* 657 * Check flow control on a given sonode. Must have so_lock held, and 658 * this function will release the hold. 659 */ 660 661 static void 662 so_check_flow_control(struct sonode *so) 663 { 664 ASSERT(MUTEX_HELD(&so->so_lock)); 665 666 if (so->so_flowctrld && so->so_rcv_queued < so->so_rcvlowat) { 667 so->so_flowctrld = B_FALSE; 668 mutex_exit(&so->so_lock); 669 /* 670 * Open up flow control. SCTP does not have any downcalls, and 671 * it will clr flow ctrl in sosctp_recvmsg(). 672 */ 673 if (so->so_downcalls != NULL && 674 so->so_downcalls->sd_clr_flowctrl != NULL) { 675 (*so->so_downcalls->sd_clr_flowctrl) 676 (so->so_proto_handle); 677 } 678 } else { 679 mutex_exit(&so->so_lock); 680 } 681 } 682 683 int 684 so_dequeue_msg(struct sonode *so, mblk_t **mctlp, struct uio *uiop, 685 rval_t *rvalp, int flags) 686 { 687 mblk_t *mp, *nmp; 688 mblk_t *savemp, *savemptail; 689 mblk_t *new_msg_head; 690 mblk_t *new_msg_last_head; 691 mblk_t *last_tail; 692 boolean_t partial_read; 693 boolean_t reset_atmark = B_FALSE; 694 int more = 0; 695 int error; 696 ssize_t oobmark; 697 sodirect_t *sodp = so->so_direct; 698 699 partial_read = B_FALSE; 700 *mctlp = NULL; 701 again: 702 mutex_enter(&so->so_lock); 703 again1: 704 #ifdef DEBUG 705 if (so_debug_length) { 706 ASSERT(so_check_length(so)); 707 } 708 #endif 709 if (so->so_state & SS_RCVATMARK) { 710 /* Check whether the caller is OK to read past the mark */ 711 if (flags & MSG_NOMARK) { 712 mutex_exit(&so->so_lock); 713 return (EWOULDBLOCK); 714 } 715 reset_atmark = B_TRUE; 716 } 717 /* 718 * First move messages from the dump area to processing area 719 */ 720 if (sodp != NULL) { 721 if (sodp->sod_enabled) { 722 if (sodp->sod_uioa.uioa_state & UIOA_ALLOC) { 723 /* nothing to uioamove */ 724 sodp = NULL; 725 } else if (sodp->sod_uioa.uioa_state & UIOA_INIT) { 726 sodp->sod_uioa.uioa_state &= UIOA_CLR; 727 sodp->sod_uioa.uioa_state |= UIOA_ENABLED; 728 /* 729 * try to uioamove() the data that 730 * has already queued. 731 */ 732 sod_uioa_so_init(so, sodp, uiop); 733 } 734 } else { 735 sodp = NULL; 736 } 737 } 738 new_msg_head = so->so_rcv_head; 739 new_msg_last_head = so->so_rcv_last_head; 740 so->so_rcv_head = NULL; 741 so->so_rcv_last_head = NULL; 742 oobmark = so->so_oobmark; 743 /* 744 * We can release the lock as there can only be one reader 745 */ 746 mutex_exit(&so->so_lock); 747 748 if (new_msg_head != NULL) { 749 so_process_new_message(so, new_msg_head, new_msg_last_head); 750 } 751 savemp = savemptail = NULL; 752 rvalp->r_val1 = 0; 753 error = 0; 754 mp = so->so_rcv_q_head; 755 756 if (mp != NULL && 757 (so->so_rcv_timer_tid == 0 || 758 so->so_rcv_queued >= so->so_rcv_thresh)) { 759 partial_read = B_FALSE; 760 761 if (flags & MSG_PEEK) { 762 if ((nmp = dupmsg(mp)) == NULL && 763 (nmp = copymsg(mp)) == NULL) { 764 size_t size = msgsize(mp); 765 766 error = strwaitbuf(size, BPRI_HI); 767 if (error) { 768 return (error); 769 } 770 goto again; 771 } 772 mp = nmp; 773 } else { 774 ASSERT(mp->b_prev != NULL); 775 last_tail = mp->b_prev; 776 mp->b_prev = NULL; 777 so->so_rcv_q_head = mp->b_next; 778 if (so->so_rcv_q_head == NULL) { 779 so->so_rcv_q_last_head = NULL; 780 } 781 mp->b_next = NULL; 782 } 783 784 ASSERT(mctlp != NULL); 785 /* 786 * First process PROTO or PCPROTO blocks, if any. 787 */ 788 if (DB_TYPE(mp) != M_DATA) { 789 *mctlp = mp; 790 savemp = mp; 791 savemptail = mp; 792 ASSERT(DB_TYPE(mp) == M_PROTO || 793 DB_TYPE(mp) == M_PCPROTO); 794 while (mp->b_cont != NULL && 795 DB_TYPE(mp->b_cont) != M_DATA) { 796 ASSERT(DB_TYPE(mp->b_cont) == M_PROTO || 797 DB_TYPE(mp->b_cont) == M_PCPROTO); 798 mp = mp->b_cont; 799 savemptail = mp; 800 } 801 mp = savemptail->b_cont; 802 savemptail->b_cont = NULL; 803 } 804 805 ASSERT(DB_TYPE(mp) == M_DATA); 806 /* 807 * Now process DATA blocks, if any. Note that for sodirect 808 * enabled socket, uio_resid can be 0. 809 */ 810 if (uiop->uio_resid >= 0) { 811 ssize_t copied = 0; 812 813 if (sodp != NULL && (DB_FLAGS(mp) & DBLK_UIOA)) { 814 mutex_enter(&so->so_lock); 815 ASSERT(uiop == (uio_t *)&sodp->sod_uioa); 816 copied = sod_uioa_mblk(so, mp); 817 if (copied > 0) 818 partial_read = B_TRUE; 819 mutex_exit(&so->so_lock); 820 /* mark this mblk as processed */ 821 mp = NULL; 822 } else { 823 ssize_t oldresid = uiop->uio_resid; 824 825 if (MBLKL(mp) < so_mblk_pull_len) { 826 if (pullupmsg(mp, -1) == 1) { 827 last_tail = mp; 828 } 829 } 830 /* 831 * Can not read beyond the oobmark 832 */ 833 mp = socopyoutuio(mp, uiop, 834 oobmark == 0 ? INFPSZ : oobmark, &error); 835 if (error != 0) { 836 freemsg(*mctlp); 837 *mctlp = NULL; 838 more = 0; 839 goto done; 840 } 841 ASSERT(oldresid >= uiop->uio_resid); 842 copied = oldresid - uiop->uio_resid; 843 if (oldresid > uiop->uio_resid) 844 partial_read = B_TRUE; 845 } 846 ASSERT(copied >= 0); 847 if (copied > 0 && !(flags & MSG_PEEK)) { 848 mutex_enter(&so->so_lock); 849 so->so_rcv_queued -= copied; 850 ASSERT(so->so_oobmark >= 0); 851 if (so->so_oobmark > 0) { 852 so->so_oobmark -= copied; 853 ASSERT(so->so_oobmark >= 0); 854 if (so->so_oobmark == 0) { 855 ASSERT(so->so_state & 856 SS_OOBPEND); 857 so->so_oobmark = 0; 858 so->so_state |= SS_RCVATMARK; 859 } 860 } 861 /* 862 * so_check_flow_control() will drop 863 * so->so_lock. 864 */ 865 so_check_flow_control(so); 866 } 867 } 868 if (mp != NULL) { /* more data blocks in msg */ 869 more |= MOREDATA; 870 if ((flags & (MSG_PEEK|MSG_TRUNC))) { 871 if (flags & MSG_PEEK) { 872 freemsg(mp); 873 } else { 874 unsigned int msize = msgdsize(mp); 875 876 freemsg(mp); 877 mutex_enter(&so->so_lock); 878 so->so_rcv_queued -= msize; 879 /* 880 * so_check_flow_control() will drop 881 * so->so_lock. 882 */ 883 so_check_flow_control(so); 884 } 885 } else if (partial_read && !somsghasdata(mp)) { 886 /* 887 * Avoid queuing a zero-length tail part of 888 * a message. partial_read == 1 indicates that 889 * we read some of the message. 890 */ 891 freemsg(mp); 892 more &= ~MOREDATA; 893 } else { 894 if (savemp != NULL && 895 (flags & MSG_DUPCTRL)) { 896 mblk_t *nmp; 897 /* 898 * There should only be non data mblks 899 */ 900 ASSERT(DB_TYPE(savemp) != M_DATA && 901 DB_TYPE(savemptail) != M_DATA); 902 try_again: 903 if ((nmp = dupmsg(savemp)) == NULL && 904 (nmp = copymsg(savemp)) == NULL) { 905 906 size_t size = msgsize(savemp); 907 908 error = strwaitbuf(size, 909 BPRI_HI); 910 if (error != 0) { 911 /* 912 * In case we 913 * cannot copy 914 * control data 915 * free the remaining 916 * data. 917 */ 918 freemsg(mp); 919 goto done; 920 } 921 goto try_again; 922 } 923 924 ASSERT(nmp != NULL); 925 ASSERT(DB_TYPE(nmp) != M_DATA); 926 savemptail->b_cont = mp; 927 *mctlp = nmp; 928 mp = savemp; 929 } 930 /* 931 * putback mp 932 */ 933 so_prepend_msg(so, mp, last_tail); 934 } 935 } 936 937 /* fast check so_rcv_head if there is more data */ 938 if (partial_read && !(so->so_state & SS_RCVATMARK) && 939 *mctlp == NULL && uiop->uio_resid > 0 && 940 !(flags & MSG_PEEK) && so->so_rcv_head != NULL) { 941 goto again; 942 } 943 } else if (!partial_read) { 944 mutex_enter(&so->so_lock); 945 if (so->so_error != 0) { 946 error = sogeterr(so, !(flags & MSG_PEEK)); 947 mutex_exit(&so->so_lock); 948 return (error); 949 } 950 /* 951 * No pending data. Return right away for nonblocking 952 * socket, otherwise sleep waiting for data. 953 */ 954 if (!(so->so_state & SS_CANTRCVMORE) && uiop->uio_resid > 0) { 955 if ((uiop->uio_fmode & (FNDELAY|FNONBLOCK)) || 956 (flags & MSG_DONTWAIT)) { 957 error = EWOULDBLOCK; 958 } else { 959 if (so->so_state & (SS_CLOSING | 960 SS_FALLBACK_PENDING)) { 961 mutex_exit(&so->so_lock); 962 error = EINTR; 963 goto done; 964 } 965 966 if (so->so_rcv_head != NULL) { 967 goto again1; 968 } 969 so->so_rcv_wakeup = B_TRUE; 970 so->so_rcv_wanted = uiop->uio_resid; 971 if (so->so_rcvtimeo == 0) { 972 /* 973 * Zero means disable timeout. 974 */ 975 error = cv_wait_sig(&so->so_rcv_cv, 976 &so->so_lock); 977 } else { 978 clock_t now; 979 time_to_wait(&now, so->so_rcvtimeo); 980 error = cv_timedwait_sig(&so->so_rcv_cv, 981 &so->so_lock, now); 982 } 983 so->so_rcv_wakeup = B_FALSE; 984 so->so_rcv_wanted = 0; 985 986 if (error == 0) { 987 error = EINTR; 988 } else if (error == -1) { 989 error = EAGAIN; 990 } else { 991 goto again1; 992 } 993 } 994 } 995 mutex_exit(&so->so_lock); 996 } 997 if (reset_atmark && partial_read && !(flags & MSG_PEEK)) { 998 /* 999 * We are passed the mark, update state 1000 * 4.3BSD and 4.4BSD clears the mark when peeking across it. 1001 * The draft Posix socket spec states that the mark should 1002 * not be cleared when peeking. We follow the latter. 1003 */ 1004 mutex_enter(&so->so_lock); 1005 ASSERT(so_verify_oobstate(so)); 1006 so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK); 1007 freemsg(so->so_oobmsg); 1008 so->so_oobmsg = NULL; 1009 ASSERT(so_verify_oobstate(so)); 1010 mutex_exit(&so->so_lock); 1011 } 1012 ASSERT(so->so_rcv_wakeup == B_FALSE); 1013 done: 1014 if (sodp != NULL) { 1015 mutex_enter(&so->so_lock); 1016 if (sodp->sod_enabled && 1017 (sodp->sod_uioa.uioa_state & UIOA_ENABLED)) { 1018 SOD_UIOAFINI(sodp); 1019 if (sodp->sod_uioa.uioa_mbytes > 0) { 1020 ASSERT(so->so_rcv_q_head != NULL || 1021 so->so_rcv_head != NULL); 1022 so->so_rcv_queued -= sod_uioa_mblk(so, NULL); 1023 if (error == EWOULDBLOCK) 1024 error = 0; 1025 } 1026 } 1027 mutex_exit(&so->so_lock); 1028 } 1029 #ifdef DEBUG 1030 if (so_debug_length) { 1031 mutex_enter(&so->so_lock); 1032 ASSERT(so_check_length(so)); 1033 mutex_exit(&so->so_lock); 1034 } 1035 #endif 1036 rvalp->r_val1 = more; 1037 ASSERT(MUTEX_NOT_HELD(&so->so_lock)); 1038 return (error); 1039 } 1040 1041 /* 1042 * Enqueue data from the protocol on the socket's rcv queue. 1043 * 1044 * We try to hook new M_DATA mblks onto an existing chain, however, 1045 * that cannot be done if the existing chain has already been 1046 * processed by I/OAT. Non-M_DATA mblks are just linked together via 1047 * b_next. In all cases the b_prev of the enqueued mblk is set to 1048 * point to the last mblk in its b_cont chain. 1049 */ 1050 void 1051 so_enqueue_msg(struct sonode *so, mblk_t *mp, size_t msg_size) 1052 { 1053 ASSERT(MUTEX_HELD(&so->so_lock)); 1054 1055 #ifdef DEBUG 1056 if (so_debug_length) { 1057 ASSERT(so_check_length(so)); 1058 } 1059 #endif 1060 so->so_rcv_queued += msg_size; 1061 1062 if (so->so_rcv_head == NULL) { 1063 ASSERT(so->so_rcv_last_head == NULL); 1064 so->so_rcv_head = mp; 1065 so->so_rcv_last_head = mp; 1066 } else if ((DB_TYPE(mp) == M_DATA && 1067 DB_TYPE(so->so_rcv_last_head) == M_DATA) && 1068 ((DB_FLAGS(mp) & DBLK_UIOA) == 1069 (DB_FLAGS(so->so_rcv_last_head) & DBLK_UIOA))) { 1070 /* Added to the end */ 1071 ASSERT(so->so_rcv_last_head != NULL); 1072 ASSERT(so->so_rcv_last_head->b_prev != NULL); 1073 so->so_rcv_last_head->b_prev->b_cont = mp; 1074 } else { 1075 /* Start a new end */ 1076 so->so_rcv_last_head->b_next = mp; 1077 so->so_rcv_last_head = mp; 1078 } 1079 while (mp->b_cont != NULL) 1080 mp = mp->b_cont; 1081 1082 so->so_rcv_last_head->b_prev = mp; 1083 #ifdef DEBUG 1084 if (so_debug_length) { 1085 ASSERT(so_check_length(so)); 1086 } 1087 #endif 1088 } 1089 1090 /* 1091 * Return B_TRUE if there is data in the message, B_FALSE otherwise. 1092 */ 1093 boolean_t 1094 somsghasdata(mblk_t *mp) 1095 { 1096 for (; mp; mp = mp->b_cont) 1097 if (mp->b_datap->db_type == M_DATA) { 1098 ASSERT(mp->b_wptr >= mp->b_rptr); 1099 if (mp->b_wptr > mp->b_rptr) 1100 return (B_TRUE); 1101 } 1102 return (B_FALSE); 1103 } 1104 1105 /* 1106 * Flush the read side of sockfs. 1107 * 1108 * The caller must be sure that a reader is not already active when the 1109 * buffer is being flushed. 1110 */ 1111 void 1112 so_rcv_flush(struct sonode *so) 1113 { 1114 mblk_t *mp; 1115 1116 ASSERT(MUTEX_HELD(&so->so_lock)); 1117 1118 if (so->so_oobmsg != NULL) { 1119 freemsg(so->so_oobmsg); 1120 so->so_oobmsg = NULL; 1121 so->so_oobmark = 0; 1122 so->so_state &= 1123 ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA|SS_RCVATMARK); 1124 } 1125 1126 /* 1127 * Free messages sitting in the send and recv queue 1128 */ 1129 while (so->so_rcv_q_head != NULL) { 1130 mp = so->so_rcv_q_head; 1131 so->so_rcv_q_head = mp->b_next; 1132 mp->b_next = mp->b_prev = NULL; 1133 freemsg(mp); 1134 } 1135 while (so->so_rcv_head != NULL) { 1136 mp = so->so_rcv_head; 1137 so->so_rcv_head = mp->b_next; 1138 mp->b_next = mp->b_prev = NULL; 1139 freemsg(mp); 1140 } 1141 so->so_rcv_queued = 0; 1142 so->so_rcv_q_head = NULL; 1143 so->so_rcv_q_last_head = NULL; 1144 so->so_rcv_head = NULL; 1145 so->so_rcv_last_head = NULL; 1146 } 1147 1148 /* 1149 * Handle recv* calls that set MSG_OOB or MSG_OOB together with MSG_PEEK. 1150 */ 1151 int 1152 sorecvoob(struct sonode *so, struct nmsghdr *msg, struct uio *uiop, int flags, 1153 boolean_t oob_inline) 1154 { 1155 mblk_t *mp, *nmp; 1156 int error; 1157 1158 dprintso(so, 1, ("sorecvoob(%p, %p, 0x%x)\n", (void *)so, (void *)msg, 1159 flags)); 1160 1161 if (msg != NULL) { 1162 /* 1163 * There is never any oob data with addresses or control since 1164 * the T_EXDATA_IND does not carry any options. 1165 */ 1166 msg->msg_controllen = 0; 1167 msg->msg_namelen = 0; 1168 msg->msg_flags = 0; 1169 } 1170 1171 mutex_enter(&so->so_lock); 1172 ASSERT(so_verify_oobstate(so)); 1173 if (oob_inline || 1174 (so->so_state & (SS_OOBPEND|SS_HADOOBDATA)) != SS_OOBPEND) { 1175 dprintso(so, 1, ("sorecvoob: inline or data consumed\n")); 1176 mutex_exit(&so->so_lock); 1177 return (EINVAL); 1178 } 1179 if (!(so->so_state & SS_HAVEOOBDATA)) { 1180 dprintso(so, 1, ("sorecvoob: no data yet\n")); 1181 mutex_exit(&so->so_lock); 1182 return (EWOULDBLOCK); 1183 } 1184 ASSERT(so->so_oobmsg != NULL); 1185 mp = so->so_oobmsg; 1186 if (flags & MSG_PEEK) { 1187 /* 1188 * Since recv* can not return ENOBUFS we can not use dupmsg. 1189 * Instead we revert to the consolidation private 1190 * allocb_wait plus bcopy. 1191 */ 1192 mblk_t *mp1; 1193 1194 mp1 = allocb_wait(msgdsize(mp), BPRI_MED, STR_NOSIG, NULL); 1195 ASSERT(mp1); 1196 1197 while (mp != NULL) { 1198 ssize_t size; 1199 1200 size = MBLKL(mp); 1201 bcopy(mp->b_rptr, mp1->b_wptr, size); 1202 mp1->b_wptr += size; 1203 ASSERT(mp1->b_wptr <= mp1->b_datap->db_lim); 1204 mp = mp->b_cont; 1205 } 1206 mp = mp1; 1207 } else { 1208 /* 1209 * Update the state indicating that the data has been consumed. 1210 * Keep SS_OOBPEND set until data is consumed past the mark. 1211 */ 1212 so->so_oobmsg = NULL; 1213 so->so_state ^= SS_HAVEOOBDATA|SS_HADOOBDATA; 1214 } 1215 ASSERT(so_verify_oobstate(so)); 1216 mutex_exit(&so->so_lock); 1217 1218 error = 0; 1219 nmp = mp; 1220 while (nmp != NULL && uiop->uio_resid > 0) { 1221 ssize_t n = MBLKL(nmp); 1222 1223 n = MIN(n, uiop->uio_resid); 1224 if (n > 0) 1225 error = uiomove(nmp->b_rptr, n, 1226 UIO_READ, uiop); 1227 if (error) 1228 break; 1229 nmp = nmp->b_cont; 1230 } 1231 ASSERT(mp->b_next == NULL && mp->b_prev == NULL); 1232 freemsg(mp); 1233 return (error); 1234 } 1235 1236 /* 1237 * Allocate and initializ sonode 1238 */ 1239 /* ARGSUSED */ 1240 struct sonode * 1241 socket_sonode_create(struct sockparams *sp, int family, int type, 1242 int protocol, int version, int sflags, int *errorp, struct cred *cr) 1243 { 1244 sonode_t *so; 1245 int kmflags; 1246 1247 /* 1248 * Choose the right set of sonodeops based on the upcall and 1249 * down call version that the protocol has provided 1250 */ 1251 if (SOCK_UC_VERSION != sp->sp_smod_info->smod_uc_version || 1252 SOCK_DC_VERSION != sp->sp_smod_info->smod_dc_version) { 1253 /* 1254 * mismatch 1255 */ 1256 #ifdef DEBUG 1257 cmn_err(CE_CONT, "protocol and socket module version mismatch"); 1258 #endif 1259 *errorp = EINVAL; 1260 return (NULL); 1261 } 1262 1263 kmflags = (sflags & SOCKET_NOSLEEP) ? KM_NOSLEEP : KM_SLEEP; 1264 1265 so = kmem_cache_alloc(socket_cache, kmflags); 1266 if (so == NULL) { 1267 *errorp = ENOMEM; 1268 return (NULL); 1269 } 1270 1271 sonode_init(so, sp, family, type, protocol, &so_sonodeops); 1272 1273 if (version == SOV_DEFAULT) 1274 version = so_default_version; 1275 1276 so->so_version = (short)version; 1277 1278 /* 1279 * set the default values to be INFPSZ 1280 * if a protocol desires it can change the value later 1281 */ 1282 so->so_proto_props.sopp_rxhiwat = SOCKET_RECVHIWATER; 1283 so->so_proto_props.sopp_rxlowat = SOCKET_RECVLOWATER; 1284 so->so_proto_props.sopp_maxpsz = INFPSZ; 1285 so->so_proto_props.sopp_maxblk = INFPSZ; 1286 1287 return (so); 1288 } 1289 1290 int 1291 socket_init_common(struct sonode *so, struct sonode *pso, int flags, cred_t *cr) 1292 { 1293 int error = 0; 1294 1295 if (pso != NULL) { 1296 /* 1297 * We have a passive open, so inherit basic state from 1298 * the parent (listener). 1299 * 1300 * No need to grab the new sonode's lock, since there is no 1301 * one that can have a reference to it. 1302 */ 1303 mutex_enter(&pso->so_lock); 1304 1305 so->so_state |= SS_ISCONNECTED | (pso->so_state & SS_ASYNC); 1306 so->so_pgrp = pso->so_pgrp; 1307 so->so_rcvtimeo = pso->so_rcvtimeo; 1308 so->so_sndtimeo = pso->so_sndtimeo; 1309 so->so_xpg_rcvbuf = pso->so_xpg_rcvbuf; 1310 /* 1311 * Make note of the socket level options. TCP and IP level 1312 * options are already inherited. We could do all this after 1313 * accept is successful but doing it here simplifies code and 1314 * no harm done for error case. 1315 */ 1316 so->so_options = pso->so_options & (SO_DEBUG|SO_REUSEADDR| 1317 SO_KEEPALIVE|SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK| 1318 SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER); 1319 so->so_proto_props = pso->so_proto_props; 1320 so->so_mode = pso->so_mode; 1321 so->so_pollev = pso->so_pollev & SO_POLLEV_ALWAYS; 1322 1323 mutex_exit(&pso->so_lock); 1324 } else { 1325 struct sockparams *sp = so->so_sockparams; 1326 sock_upcalls_t *upcalls_to_use; 1327 1328 /* 1329 * Based on the version number select the right upcalls to 1330 * pass down. Currently we only have one version so choose 1331 * default 1332 */ 1333 upcalls_to_use = &so_upcalls; 1334 1335 /* active open, so create a lower handle */ 1336 so->so_proto_handle = 1337 sp->sp_smod_info->smod_proto_create_func(so->so_family, 1338 so->so_type, so->so_protocol, &so->so_downcalls, 1339 &so->so_mode, &error, flags, cr); 1340 1341 if (so->so_proto_handle == NULL) { 1342 ASSERT(error != 0); 1343 /* 1344 * To be safe; if a lower handle cannot be created, and 1345 * the proto does not give a reason why, assume there 1346 * was a lack of memory. 1347 */ 1348 return ((error == 0) ? ENOMEM : error); 1349 } 1350 ASSERT(so->so_downcalls != NULL); 1351 ASSERT(so->so_downcalls->sd_send != NULL || 1352 so->so_downcalls->sd_send_uio != NULL); 1353 if (so->so_downcalls->sd_recv_uio != NULL) { 1354 ASSERT(so->so_downcalls->sd_poll != NULL); 1355 so->so_pollev |= SO_POLLEV_ALWAYS; 1356 } 1357 1358 (*so->so_downcalls->sd_activate)(so->so_proto_handle, 1359 (sock_upper_handle_t)so, upcalls_to_use, 0, cr); 1360 1361 /* Wildcard */ 1362 1363 /* 1364 * FIXME No need for this, the protocol can deal with it in 1365 * sd_create(). Should update ICMP. 1366 */ 1367 if (so->so_protocol != so->so_sockparams->sp_protocol) { 1368 int protocol = so->so_protocol; 1369 int error; 1370 /* 1371 * Issue SO_PROTOTYPE setsockopt. 1372 */ 1373 error = socket_setsockopt(so, SOL_SOCKET, SO_PROTOTYPE, 1374 &protocol, (t_uscalar_t)sizeof (protocol), cr); 1375 if (error) { 1376 (void) (*so->so_downcalls->sd_close) 1377 (so->so_proto_handle, 0, cr); 1378 1379 mutex_enter(&so->so_lock); 1380 so_rcv_flush(so); 1381 mutex_exit(&so->so_lock); 1382 /* 1383 * Setsockopt often fails with ENOPROTOOPT but 1384 * socket() should fail with 1385 * EPROTONOSUPPORT/EPROTOTYPE. 1386 */ 1387 return (EPROTONOSUPPORT); 1388 } 1389 } 1390 } 1391 1392 if (uioasync.enabled) 1393 sod_sock_init(so); 1394 1395 return (0); 1396 } 1397 1398 /* 1399 * int socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode, 1400 * struct cred *cr, int32_t *rvalp) 1401 * 1402 * Handle ioctls that manipulate basic socket state; non-blocking, 1403 * async, etc. 1404 * 1405 * Returns: 1406 * < 0 - ioctl was not handle 1407 * >= 0 - ioctl was handled, if > 0, then it is an errno 1408 * 1409 * Notes: 1410 * Assumes the standard receive buffer is used to obtain info for 1411 * NREAD. 1412 */ 1413 /* ARGSUSED */ 1414 int 1415 socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode, 1416 struct cred *cr, int32_t *rvalp) 1417 { 1418 switch (cmd) { 1419 case SIOCSQPTR: 1420 /* 1421 * SIOCSQPTR is valid only when helper stream is created 1422 * by the protocol. 1423 */ 1424 1425 return (EOPNOTSUPP); 1426 case FIONBIO: { 1427 int32_t value; 1428 1429 if (so_copyin((void *)arg, &value, sizeof (int32_t), 1430 (mode & (int)FKIOCTL))) 1431 return (EFAULT); 1432 1433 mutex_enter(&so->so_lock); 1434 if (value) { 1435 so->so_state |= SS_NDELAY; 1436 } else { 1437 so->so_state &= ~SS_NDELAY; 1438 } 1439 mutex_exit(&so->so_lock); 1440 return (0); 1441 } 1442 case FIOASYNC: { 1443 int32_t value; 1444 1445 if (so_copyin((void *)arg, &value, sizeof (int32_t), 1446 (mode & (int)FKIOCTL))) 1447 return (EFAULT); 1448 1449 mutex_enter(&so->so_lock); 1450 1451 if (value) { 1452 /* Turn on SIGIO */ 1453 so->so_state |= SS_ASYNC; 1454 } else { 1455 /* Turn off SIGIO */ 1456 so->so_state &= ~SS_ASYNC; 1457 } 1458 mutex_exit(&so->so_lock); 1459 1460 return (0); 1461 } 1462 1463 case SIOCSPGRP: 1464 case FIOSETOWN: { 1465 int error; 1466 pid_t pid; 1467 1468 if (so_copyin((void *)arg, &pid, sizeof (pid_t), 1469 (mode & (int)FKIOCTL))) 1470 return (EFAULT); 1471 1472 mutex_enter(&so->so_lock); 1473 error = (pid != so->so_pgrp) ? socket_chgpgrp(so, pid) : 0; 1474 mutex_exit(&so->so_lock); 1475 return (error); 1476 } 1477 case SIOCGPGRP: 1478 case FIOGETOWN: 1479 if (so_copyout(&so->so_pgrp, (void *)arg, 1480 sizeof (pid_t), (mode & (int)FKIOCTL))) 1481 return (EFAULT); 1482 1483 return (0); 1484 case SIOCATMARK: { 1485 int retval; 1486 1487 /* 1488 * Only protocols that support urgent data can handle ATMARK. 1489 */ 1490 if ((so->so_mode & SM_EXDATA) == 0) 1491 return (EINVAL); 1492 1493 /* 1494 * If the protocol is maintaining its own buffer, then the 1495 * request must be passed down. 1496 */ 1497 if (so->so_downcalls->sd_recv_uio != NULL) 1498 return (-1); 1499 1500 retval = (so->so_state & SS_RCVATMARK) != 0; 1501 1502 if (so_copyout(&retval, (void *)arg, sizeof (int), 1503 (mode & (int)FKIOCTL))) { 1504 return (EFAULT); 1505 } 1506 return (0); 1507 } 1508 1509 case FIONREAD: { 1510 int retval; 1511 1512 /* 1513 * If the protocol is maintaining its own buffer, then the 1514 * request must be passed down. 1515 */ 1516 if (so->so_downcalls->sd_recv_uio != NULL) 1517 return (-1); 1518 1519 retval = MIN(so->so_rcv_queued, INT_MAX); 1520 1521 if (so_copyout(&retval, (void *)arg, 1522 sizeof (retval), (mode & (int)FKIOCTL))) { 1523 return (EFAULT); 1524 } 1525 return (0); 1526 } 1527 1528 case _I_GETPEERCRED: { 1529 int error = 0; 1530 1531 if ((mode & FKIOCTL) == 0) 1532 return (EINVAL); 1533 1534 mutex_enter(&so->so_lock); 1535 if ((so->so_mode & SM_CONNREQUIRED) == 0) { 1536 error = ENOTSUP; 1537 } else if ((so->so_state & SS_ISCONNECTED) == 0) { 1538 error = ENOTCONN; 1539 } else if (so->so_peercred != NULL) { 1540 k_peercred_t *kp = (k_peercred_t *)arg; 1541 kp->pc_cr = so->so_peercred; 1542 kp->pc_cpid = so->so_cpid; 1543 crhold(so->so_peercred); 1544 } else { 1545 error = EINVAL; 1546 } 1547 mutex_exit(&so->so_lock); 1548 return (error); 1549 } 1550 default: 1551 return (-1); 1552 } 1553 } 1554 1555 /* 1556 * Handle the I_NREAD STREAM ioctl. 1557 */ 1558 static int 1559 so_strioc_nread(struct sonode *so, intptr_t arg, int mode, int32_t *rvalp) 1560 { 1561 size_t size = 0; 1562 int retval; 1563 int count = 0; 1564 mblk_t *mp; 1565 1566 if (so->so_downcalls == NULL || 1567 so->so_downcalls->sd_recv_uio != NULL) 1568 return (EINVAL); 1569 1570 mutex_enter(&so->so_lock); 1571 /* Wait for reader to get out of the way. */ 1572 while (so->so_flag & SOREADLOCKED) { 1573 /* 1574 * If reader is waiting for data, then there should be nothing 1575 * on the rcv queue. 1576 */ 1577 if (so->so_rcv_wakeup) 1578 goto out; 1579 1580 so->so_flag |= SOWANT; 1581 /* Do a timed sleep, in case the reader goes to sleep. */ 1582 (void) cv_timedwait(&so->so_state_cv, &so->so_lock, 1583 lbolt + drv_usectohz(10)); 1584 } 1585 1586 /* 1587 * Since we are holding so_lock no new reader will come in, and the 1588 * protocol will not be able to enqueue data. So it's safe to walk 1589 * both rcv queues. 1590 */ 1591 mp = so->so_rcv_q_head; 1592 if (mp != NULL) { 1593 size = msgdsize(so->so_rcv_q_head); 1594 for (; mp != NULL; mp = mp->b_next) 1595 count++; 1596 } else { 1597 /* 1598 * In case the processing list was empty, get the size of the 1599 * next msg in line. 1600 */ 1601 size = msgdsize(so->so_rcv_head); 1602 } 1603 1604 for (mp = so->so_rcv_head; mp != NULL; mp = mp->b_next) 1605 count++; 1606 out: 1607 mutex_exit(&so->so_lock); 1608 1609 /* 1610 * Drop down from size_t to the "int" required by the 1611 * interface. Cap at INT_MAX. 1612 */ 1613 retval = MIN(size, INT_MAX); 1614 if (so_copyout(&retval, (void *)arg, sizeof (retval), 1615 (mode & (int)FKIOCTL))) { 1616 return (EFAULT); 1617 } else { 1618 *rvalp = count; 1619 return (0); 1620 } 1621 } 1622 1623 /* 1624 * Process STREAM ioctls. 1625 * 1626 * Returns: 1627 * < 0 - ioctl was not handle 1628 * >= 0 - ioctl was handled, if > 0, then it is an errno 1629 */ 1630 int 1631 socket_strioc_common(struct sonode *so, int cmd, intptr_t arg, int mode, 1632 struct cred *cr, int32_t *rvalp) 1633 { 1634 int retval; 1635 1636 /* Only STREAM iotcls are handled here */ 1637 if ((cmd & 0xffffff00U) != STR) 1638 return (-1); 1639 1640 switch (cmd) { 1641 case I_CANPUT: 1642 /* 1643 * We return an error for I_CANPUT so that isastream(3C) will 1644 * not report the socket as being a STREAM. 1645 */ 1646 return (EOPNOTSUPP); 1647 case I_NREAD: 1648 /* Avoid doing a fallback for I_NREAD. */ 1649 return (so_strioc_nread(so, arg, mode, rvalp)); 1650 case I_LOOK: 1651 /* Avoid doing a fallback for I_LOOK. */ 1652 if (so_copyout("sockmod", (void *)arg, strlen("sockmod") + 1, 1653 (mode & (int)FKIOCTL))) { 1654 return (EFAULT); 1655 } 1656 return (0); 1657 default: 1658 break; 1659 } 1660 1661 /* 1662 * Try to fall back to TPI, and if successful, reissue the ioctl. 1663 */ 1664 if ((retval = so_tpi_fallback(so, cr)) == 0) { 1665 /* Reissue the ioctl */ 1666 ASSERT(so->so_rcv_q_head == NULL); 1667 return (SOP_IOCTL(so, cmd, arg, mode, cr, rvalp)); 1668 } else { 1669 return (retval); 1670 } 1671 } 1672 1673 /* 1674 * This is called for all socket types to verify that the buffer size is large 1675 * enough for the option, and if we can, handle the request as well. Most 1676 * options will be forwarded to the protocol. 1677 */ 1678 int 1679 socket_getopt_common(struct sonode *so, int level, int option_name, 1680 void *optval, socklen_t *optlenp, int flags) 1681 { 1682 if (level != SOL_SOCKET) 1683 return (-1); 1684 1685 switch (option_name) { 1686 case SO_ERROR: 1687 case SO_DOMAIN: 1688 case SO_TYPE: 1689 case SO_ACCEPTCONN: { 1690 int32_t value; 1691 socklen_t optlen = *optlenp; 1692 1693 if (optlen < (t_uscalar_t)sizeof (int32_t)) { 1694 return (EINVAL); 1695 } 1696 1697 switch (option_name) { 1698 case SO_ERROR: 1699 mutex_enter(&so->so_lock); 1700 value = sogeterr(so, B_TRUE); 1701 mutex_exit(&so->so_lock); 1702 break; 1703 case SO_DOMAIN: 1704 value = so->so_family; 1705 break; 1706 case SO_TYPE: 1707 value = so->so_type; 1708 break; 1709 case SO_ACCEPTCONN: 1710 if (so->so_state & SS_ACCEPTCONN) 1711 value = SO_ACCEPTCONN; 1712 else 1713 value = 0; 1714 break; 1715 } 1716 1717 bcopy(&value, optval, sizeof (value)); 1718 *optlenp = sizeof (value); 1719 1720 return (0); 1721 } 1722 case SO_SNDTIMEO: 1723 case SO_RCVTIMEO: { 1724 clock_t value; 1725 socklen_t optlen = *optlenp; 1726 1727 if (get_udatamodel() == DATAMODEL_NONE || 1728 get_udatamodel() == DATAMODEL_NATIVE) { 1729 if (optlen < sizeof (struct timeval)) 1730 return (EINVAL); 1731 } else { 1732 if (optlen < sizeof (struct timeval32)) 1733 return (EINVAL); 1734 } 1735 if (option_name == SO_RCVTIMEO) 1736 value = drv_hztousec(so->so_rcvtimeo); 1737 else 1738 value = drv_hztousec(so->so_sndtimeo); 1739 1740 if (get_udatamodel() == DATAMODEL_NONE || 1741 get_udatamodel() == DATAMODEL_NATIVE) { 1742 ((struct timeval *)(optval))->tv_sec = 1743 value / (1000 * 1000); 1744 ((struct timeval *)(optval))->tv_usec = 1745 value % (1000 * 1000); 1746 *optlenp = sizeof (struct timeval); 1747 } else { 1748 ((struct timeval32 *)(optval))->tv_sec = 1749 value / (1000 * 1000); 1750 ((struct timeval32 *)(optval))->tv_usec = 1751 value % (1000 * 1000); 1752 *optlenp = sizeof (struct timeval32); 1753 } 1754 return (0); 1755 } 1756 case SO_DEBUG: 1757 case SO_REUSEADDR: 1758 case SO_KEEPALIVE: 1759 case SO_DONTROUTE: 1760 case SO_BROADCAST: 1761 case SO_USELOOPBACK: 1762 case SO_OOBINLINE: 1763 case SO_SNDBUF: 1764 #ifdef notyet 1765 case SO_SNDLOWAT: 1766 case SO_RCVLOWAT: 1767 #endif /* notyet */ 1768 case SO_DGRAM_ERRIND: { 1769 socklen_t optlen = *optlenp; 1770 1771 if (optlen < (t_uscalar_t)sizeof (int32_t)) 1772 return (EINVAL); 1773 break; 1774 } 1775 case SO_RCVBUF: { 1776 socklen_t optlen = *optlenp; 1777 1778 if (optlen < (t_uscalar_t)sizeof (int32_t)) 1779 return (EINVAL); 1780 1781 if ((flags & _SOGETSOCKOPT_XPG4_2) && so->so_xpg_rcvbuf != 0) { 1782 /* 1783 * XXX If SO_RCVBUF has been set and this is an 1784 * XPG 4.2 application then do not ask the transport 1785 * since the transport might adjust the value and not 1786 * return exactly what was set by the application. 1787 * For non-XPG 4.2 application we return the value 1788 * that the transport is actually using. 1789 */ 1790 *(int32_t *)optval = so->so_xpg_rcvbuf; 1791 *optlenp = sizeof (so->so_xpg_rcvbuf); 1792 return (0); 1793 } 1794 /* 1795 * If the option has not been set then get a default 1796 * value from the transport. 1797 */ 1798 break; 1799 } 1800 case SO_LINGER: { 1801 socklen_t optlen = *optlenp; 1802 1803 if (optlen < (t_uscalar_t)sizeof (struct linger)) 1804 return (EINVAL); 1805 break; 1806 } 1807 case SO_SND_BUFINFO: { 1808 socklen_t optlen = *optlenp; 1809 1810 if (optlen < (t_uscalar_t)sizeof (struct so_snd_bufinfo)) 1811 return (EINVAL); 1812 ((struct so_snd_bufinfo *)(optval))->sbi_wroff = 1813 (so->so_proto_props).sopp_wroff; 1814 ((struct so_snd_bufinfo *)(optval))->sbi_maxblk = 1815 (so->so_proto_props).sopp_maxblk; 1816 ((struct so_snd_bufinfo *)(optval))->sbi_maxpsz = 1817 (so->so_proto_props).sopp_maxpsz; 1818 ((struct so_snd_bufinfo *)(optval))->sbi_tail = 1819 (so->so_proto_props).sopp_tail; 1820 *optlenp = sizeof (struct so_snd_bufinfo); 1821 return (0); 1822 } 1823 default: 1824 break; 1825 } 1826 1827 /* Unknown Option */ 1828 return (-1); 1829 } 1830 1831 void 1832 socket_sonode_destroy(struct sonode *so) 1833 { 1834 sonode_fini(so); 1835 kmem_cache_free(socket_cache, so); 1836 } 1837 1838 int 1839 so_zcopy_wait(struct sonode *so) 1840 { 1841 int error = 0; 1842 1843 mutex_enter(&so->so_lock); 1844 while (!(so->so_copyflag & STZCNOTIFY)) { 1845 if (so->so_state & SS_CLOSING) { 1846 mutex_exit(&so->so_lock); 1847 return (EINTR); 1848 } 1849 if (cv_wait_sig(&so->so_copy_cv, &so->so_lock) == 0) { 1850 error = EINTR; 1851 break; 1852 } 1853 } 1854 so->so_copyflag &= ~STZCNOTIFY; 1855 mutex_exit(&so->so_lock); 1856 return (error); 1857 } 1858 1859 void 1860 so_timer_callback(void *arg) 1861 { 1862 struct sonode *so = (struct sonode *)arg; 1863 1864 mutex_enter(&so->so_lock); 1865 1866 so->so_rcv_timer_tid = 0; 1867 if (so->so_rcv_queued > 0) { 1868 so_notify_data(so, so->so_rcv_queued); 1869 } else { 1870 mutex_exit(&so->so_lock); 1871 } 1872 } 1873 1874 #ifdef DEBUG 1875 /* 1876 * Verify that the length stored in so_rcv_queued and the length of data blocks 1877 * queued is same. 1878 */ 1879 static boolean_t 1880 so_check_length(sonode_t *so) 1881 { 1882 mblk_t *mp = so->so_rcv_q_head; 1883 int len = 0; 1884 1885 ASSERT(MUTEX_HELD(&so->so_lock)); 1886 1887 if (mp != NULL) { 1888 len = msgdsize(mp); 1889 while ((mp = mp->b_next) != NULL) 1890 len += msgdsize(mp); 1891 } 1892 mp = so->so_rcv_head; 1893 if (mp != NULL) { 1894 len += msgdsize(mp); 1895 while ((mp = mp->b_next) != NULL) 1896 len += msgdsize(mp); 1897 } 1898 return ((len == so->so_rcv_queued) ? B_TRUE : B_FALSE); 1899 } 1900 #endif 1901 1902 int 1903 so_get_mod_version(struct sockparams *sp) 1904 { 1905 ASSERT(sp != NULL && sp->sp_smod_info != NULL); 1906 return (sp->sp_smod_info->smod_version); 1907 } 1908 1909 /* 1910 * so_start_fallback() 1911 * 1912 * Block new socket operations from coming in, and wait for active operations 1913 * to complete. Threads that are sleeping will be woken up so they can get 1914 * out of the way. 1915 * 1916 * The caller must be a reader on so_fallback_rwlock. 1917 */ 1918 static boolean_t 1919 so_start_fallback(struct sonode *so) 1920 { 1921 ASSERT(RW_READ_HELD(&so->so_fallback_rwlock)); 1922 1923 mutex_enter(&so->so_lock); 1924 if (so->so_state & SS_FALLBACK_PENDING) { 1925 mutex_exit(&so->so_lock); 1926 return (B_FALSE); 1927 } 1928 so->so_state |= SS_FALLBACK_PENDING; 1929 /* 1930 * Poke all threads that might be sleeping. Any operation that comes 1931 * in after the cv_broadcast will observe the fallback pending flag 1932 * which cause the call to return where it would normally sleep. 1933 */ 1934 cv_broadcast(&so->so_state_cv); /* threads in connect() */ 1935 cv_broadcast(&so->so_rcv_cv); /* threads in recvmsg() */ 1936 cv_broadcast(&so->so_snd_cv); /* threads in sendmsg() */ 1937 mutex_enter(&so->so_acceptq_lock); 1938 cv_broadcast(&so->so_acceptq_cv); /* threads in accept() */ 1939 mutex_exit(&so->so_acceptq_lock); 1940 mutex_exit(&so->so_lock); 1941 1942 /* 1943 * The main reason for the rw_tryupgrade call is to provide 1944 * observability during the fallback process. We want to 1945 * be able to see if there are pending operations. 1946 */ 1947 if (rw_tryupgrade(&so->so_fallback_rwlock) == 0) { 1948 /* 1949 * It is safe to drop and reaquire the fallback lock, because 1950 * we are guaranteed that another fallback cannot take place. 1951 */ 1952 rw_exit(&so->so_fallback_rwlock); 1953 DTRACE_PROBE1(pending__ops__wait, (struct sonode *), so); 1954 rw_enter(&so->so_fallback_rwlock, RW_WRITER); 1955 DTRACE_PROBE1(pending__ops__complete, (struct sonode *), so); 1956 } 1957 1958 return (B_TRUE); 1959 } 1960 1961 /* 1962 * so_end_fallback() 1963 * 1964 * Allow socket opertions back in. 1965 * 1966 * The caller must be a writer on so_fallback_rwlock. 1967 */ 1968 static void 1969 so_end_fallback(struct sonode *so) 1970 { 1971 ASSERT(RW_ISWRITER(&so->so_fallback_rwlock)); 1972 1973 mutex_enter(&so->so_lock); 1974 so->so_state &= ~(SS_FALLBACK_PENDING|SS_FALLBACK_DRAIN); 1975 mutex_exit(&so->so_lock); 1976 1977 rw_downgrade(&so->so_fallback_rwlock); 1978 } 1979 1980 /* 1981 * so_quiesced_cb() 1982 * 1983 * Callback passed to the protocol during fallback. It is called once 1984 * the endpoint is quiescent. 1985 * 1986 * No requests from the user, no notifications from the protocol, so it 1987 * is safe to synchronize the state. Data can also be moved without 1988 * risk for reordering. 1989 * 1990 * We do not need to hold so_lock, since there can be only one thread 1991 * operating on the sonode. 1992 */ 1993 static void 1994 so_quiesced_cb(sock_upper_handle_t sock_handle, queue_t *q, 1995 struct T_capability_ack *tcap, struct sockaddr *laddr, socklen_t laddrlen, 1996 struct sockaddr *faddr, socklen_t faddrlen, short opts) 1997 { 1998 struct sonode *so = (struct sonode *)sock_handle; 1999 boolean_t atmark; 2000 2001 sotpi_update_state(so, tcap, laddr, laddrlen, faddr, faddrlen, opts); 2002 2003 /* 2004 * Some protocols do not quiece the data path during fallback. Once 2005 * we set the SS_FALLBACK_DRAIN flag any attempt to queue data will 2006 * fail and the protocol is responsible for saving the data for later 2007 * delivery (i.e., once the fallback has completed). 2008 */ 2009 mutex_enter(&so->so_lock); 2010 so->so_state |= SS_FALLBACK_DRAIN; 2011 SOCKET_TIMER_CANCEL(so); 2012 mutex_exit(&so->so_lock); 2013 2014 if (so->so_rcv_head != NULL) { 2015 if (so->so_rcv_q_last_head == NULL) 2016 so->so_rcv_q_head = so->so_rcv_head; 2017 else 2018 so->so_rcv_q_last_head->b_next = so->so_rcv_head; 2019 so->so_rcv_q_last_head = so->so_rcv_last_head; 2020 } 2021 2022 atmark = (so->so_state & SS_RCVATMARK) != 0; 2023 /* 2024 * Clear any OOB state having to do with pending data. The TPI 2025 * code path will set the appropriate oob state when we move the 2026 * oob data to the STREAM head. We leave SS_HADOOBDATA since the oob 2027 * data has already been consumed. 2028 */ 2029 so->so_state &= ~(SS_RCVATMARK|SS_OOBPEND|SS_HAVEOOBDATA); 2030 2031 ASSERT(so->so_oobmsg != NULL || so->so_oobmark <= so->so_rcv_queued); 2032 2033 /* 2034 * Move data to the STREAM head. 2035 */ 2036 while (so->so_rcv_q_head != NULL) { 2037 mblk_t *mp = so->so_rcv_q_head; 2038 size_t mlen = msgdsize(mp); 2039 2040 so->so_rcv_q_head = mp->b_next; 2041 mp->b_next = NULL; 2042 mp->b_prev = NULL; 2043 2044 /* 2045 * Send T_EXDATA_IND if we are at the oob mark. 2046 */ 2047 if (atmark) { 2048 struct T_exdata_ind *tei; 2049 mblk_t *mp1 = SOTOTPI(so)->sti_exdata_mp; 2050 2051 SOTOTPI(so)->sti_exdata_mp = NULL; 2052 ASSERT(mp1 != NULL); 2053 mp1->b_datap->db_type = M_PROTO; 2054 tei = (struct T_exdata_ind *)mp1->b_rptr; 2055 tei->PRIM_type = T_EXDATA_IND; 2056 tei->MORE_flag = 0; 2057 mp1->b_wptr = (uchar_t *)&tei[1]; 2058 2059 if (IS_SO_OOB_INLINE(so)) { 2060 mp1->b_cont = mp; 2061 } else { 2062 ASSERT(so->so_oobmsg != NULL); 2063 mp1->b_cont = so->so_oobmsg; 2064 so->so_oobmsg = NULL; 2065 2066 /* process current mp next time around */ 2067 mp->b_next = so->so_rcv_q_head; 2068 so->so_rcv_q_head = mp; 2069 mlen = 0; 2070 } 2071 mp = mp1; 2072 2073 /* we have consumed the oob mark */ 2074 atmark = B_FALSE; 2075 } else if (so->so_oobmark > 0) { 2076 /* 2077 * Check if the OOB mark is within the current 2078 * mblk chain. In that case we have to split it up. 2079 */ 2080 if (so->so_oobmark < mlen) { 2081 mblk_t *urg_mp = mp; 2082 2083 atmark = B_TRUE; 2084 mp = NULL; 2085 mlen = so->so_oobmark; 2086 2087 /* 2088 * It is assumed that the OOB mark does 2089 * not land within a mblk. 2090 */ 2091 do { 2092 so->so_oobmark -= MBLKL(urg_mp); 2093 mp = urg_mp; 2094 urg_mp = urg_mp->b_cont; 2095 } while (so->so_oobmark > 0); 2096 mp->b_cont = NULL; 2097 if (urg_mp != NULL) { 2098 urg_mp->b_next = so->so_rcv_q_head; 2099 so->so_rcv_q_head = urg_mp; 2100 } 2101 } else { 2102 so->so_oobmark -= mlen; 2103 if (so->so_oobmark == 0) 2104 atmark = B_TRUE; 2105 } 2106 } 2107 2108 /* 2109 * Queue data on the STREAM head. 2110 */ 2111 so->so_rcv_queued -= mlen; 2112 putnext(q, mp); 2113 } 2114 so->so_rcv_head = NULL; 2115 so->so_rcv_last_head = NULL; 2116 so->so_rcv_q_head = NULL; 2117 so->so_rcv_q_last_head = NULL; 2118 2119 /* 2120 * Check if the oob byte is at the end of the data stream, or if the 2121 * oob byte has not yet arrived. In the latter case we have to send a 2122 * SIGURG and a mark indicator to the STREAM head. The mark indicator 2123 * is needed to guarantee correct behavior for SIOCATMARK. See block 2124 * comment in socktpi.h for more details. 2125 */ 2126 if (atmark || so->so_oobmark > 0) { 2127 mblk_t *mp; 2128 2129 if (atmark && so->so_oobmsg != NULL) { 2130 struct T_exdata_ind *tei; 2131 2132 mp = SOTOTPI(so)->sti_exdata_mp; 2133 SOTOTPI(so)->sti_exdata_mp = NULL; 2134 ASSERT(mp != NULL); 2135 mp->b_datap->db_type = M_PROTO; 2136 tei = (struct T_exdata_ind *)mp->b_rptr; 2137 tei->PRIM_type = T_EXDATA_IND; 2138 tei->MORE_flag = 0; 2139 mp->b_wptr = (uchar_t *)&tei[1]; 2140 2141 mp->b_cont = so->so_oobmsg; 2142 so->so_oobmsg = NULL; 2143 2144 putnext(q, mp); 2145 } else { 2146 /* Send up the signal */ 2147 mp = SOTOTPI(so)->sti_exdata_mp; 2148 SOTOTPI(so)->sti_exdata_mp = NULL; 2149 ASSERT(mp != NULL); 2150 DB_TYPE(mp) = M_PCSIG; 2151 *mp->b_wptr++ = (uchar_t)SIGURG; 2152 putnext(q, mp); 2153 2154 /* Send up the mark indicator */ 2155 mp = SOTOTPI(so)->sti_urgmark_mp; 2156 SOTOTPI(so)->sti_urgmark_mp = NULL; 2157 mp->b_flag = atmark ? MSGMARKNEXT : MSGNOTMARKNEXT; 2158 putnext(q, mp); 2159 2160 so->so_oobmark = 0; 2161 } 2162 } 2163 2164 if (SOTOTPI(so)->sti_exdata_mp != NULL) { 2165 freeb(SOTOTPI(so)->sti_exdata_mp); 2166 SOTOTPI(so)->sti_exdata_mp = NULL; 2167 } 2168 2169 if (SOTOTPI(so)->sti_urgmark_mp != NULL) { 2170 freeb(SOTOTPI(so)->sti_urgmark_mp); 2171 SOTOTPI(so)->sti_urgmark_mp = NULL; 2172 } 2173 2174 ASSERT(so->so_oobmark == 0); 2175 ASSERT(so->so_rcv_queued == 0); 2176 } 2177 2178 #ifdef DEBUG 2179 /* 2180 * Do an integrity check of the sonode. This should be done if a 2181 * fallback fails after sonode has initially been converted to use 2182 * TPI and subsequently have to be reverted. 2183 * 2184 * Failure to pass the integrity check will panic the system. 2185 */ 2186 void 2187 so_integrity_check(struct sonode *cur, struct sonode *orig) 2188 { 2189 VERIFY(cur->so_vnode == orig->so_vnode); 2190 VERIFY(cur->so_ops == orig->so_ops); 2191 /* 2192 * For so_state we can only VERIFY the state flags in CHECK_STATE. 2193 * The other state flags might be affected by a notification from the 2194 * protocol. 2195 */ 2196 #define CHECK_STATE (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_NDELAY|SS_NONBLOCK| \ 2197 SS_ASYNC|SS_ACCEPTCONN|SS_SAVEDEOR|SS_RCVATMARK|SS_OOBPEND| \ 2198 SS_HAVEOOBDATA|SS_HADOOBDATA|SS_SENTLASTREADSIG|SS_SENTLASTWRITESIG) 2199 VERIFY((cur->so_state & (orig->so_state & CHECK_STATE)) == 2200 (orig->so_state & CHECK_STATE)); 2201 VERIFY(cur->so_mode == orig->so_mode); 2202 VERIFY(cur->so_flag == orig->so_flag); 2203 VERIFY(cur->so_count == orig->so_count); 2204 /* Cannot VERIFY so_proto_connid; proto can update it */ 2205 VERIFY(cur->so_sockparams == orig->so_sockparams); 2206 /* an error might have been recorded, but it can not be lost */ 2207 VERIFY(cur->so_error != 0 || orig->so_error == 0); 2208 VERIFY(cur->so_family == orig->so_family); 2209 VERIFY(cur->so_type == orig->so_type); 2210 VERIFY(cur->so_protocol == orig->so_protocol); 2211 VERIFY(cur->so_version == orig->so_version); 2212 /* New conns might have arrived, but none should have been lost */ 2213 VERIFY(cur->so_acceptq_len >= orig->so_acceptq_len); 2214 VERIFY(cur->so_acceptq_head == orig->so_acceptq_head); 2215 VERIFY(cur->so_backlog == orig->so_backlog); 2216 /* New OOB migth have arrived, but mark should not have been lost */ 2217 VERIFY(cur->so_oobmark >= orig->so_oobmark); 2218 /* Cannot VERIFY so_oobmsg; the proto might have sent up a new one */ 2219 VERIFY(cur->so_pgrp == orig->so_pgrp); 2220 VERIFY(cur->so_peercred == orig->so_peercred); 2221 VERIFY(cur->so_cpid == orig->so_cpid); 2222 VERIFY(cur->so_zoneid == orig->so_zoneid); 2223 /* New data migth have arrived, but none should have been lost */ 2224 VERIFY(cur->so_rcv_queued >= orig->so_rcv_queued); 2225 VERIFY(cur->so_rcv_q_head == orig->so_rcv_q_head); 2226 VERIFY(cur->so_rcv_head == orig->so_rcv_head); 2227 VERIFY(cur->so_proto_handle == orig->so_proto_handle); 2228 VERIFY(cur->so_downcalls == orig->so_downcalls); 2229 /* Cannot VERIFY so_proto_props; they can be updated by proto */ 2230 } 2231 #endif 2232 2233 /* 2234 * so_tpi_fallback() 2235 * 2236 * This is the fallback initation routine; things start here. 2237 * 2238 * Basic strategy: 2239 * o Block new socket operations from coming in 2240 * o Allocate/initate info needed by TPI 2241 * o Quiesce the connection, at which point we sync 2242 * state and move data 2243 * o Change operations (sonodeops) associated with the socket 2244 * o Unblock threads waiting for the fallback to finish 2245 */ 2246 int 2247 so_tpi_fallback(struct sonode *so, struct cred *cr) 2248 { 2249 int error; 2250 queue_t *q; 2251 struct sockparams *sp; 2252 struct sockparams *newsp = NULL; 2253 so_proto_fallback_func_t fbfunc; 2254 boolean_t direct; 2255 struct sonode *nso; 2256 #ifdef DEBUG 2257 struct sonode origso; 2258 #endif 2259 error = 0; 2260 sp = so->so_sockparams; 2261 fbfunc = sp->sp_smod_info->smod_proto_fallback_func; 2262 2263 /* 2264 * Fallback can only happen if there is a device associated 2265 * with the sonode, and the socket module has a fallback function. 2266 */ 2267 if (!SOCKPARAMS_HAS_DEVICE(sp) || fbfunc == NULL) 2268 return (EINVAL); 2269 2270 /* 2271 * Initiate fallback; upon success we know that no new requests 2272 * will come in from the user. 2273 */ 2274 if (!so_start_fallback(so)) 2275 return (EAGAIN); 2276 #ifdef DEBUG 2277 /* 2278 * Make a copy of the sonode in case we need to make an integrity 2279 * check later on. 2280 */ 2281 bcopy(so, &origso, sizeof (*so)); 2282 #endif 2283 2284 sp->sp_stats.sps_nfallback.value.ui64++; 2285 2286 newsp = sockparams_hold_ephemeral_bydev(so->so_family, so->so_type, 2287 so->so_protocol, so->so_sockparams->sp_sdev_info.sd_devpath, 2288 KM_SLEEP, &error); 2289 if (error != 0) 2290 goto out; 2291 2292 if (so->so_direct != NULL) { 2293 sodirect_t *sodp = so->so_direct; 2294 mutex_enter(&so->so_lock); 2295 2296 so->so_direct->sod_enabled = B_FALSE; 2297 so->so_state &= ~SS_SODIRECT; 2298 ASSERT(sodp->sod_uioafh == NULL); 2299 mutex_exit(&so->so_lock); 2300 } 2301 2302 /* Turn sonode into a TPI socket */ 2303 error = sotpi_convert_sonode(so, newsp, &direct, &q, cr); 2304 if (error != 0) 2305 goto out; 2306 2307 2308 /* 2309 * Now tell the protocol to start using TPI. so_quiesced_cb be 2310 * called once it's safe to synchronize state. 2311 */ 2312 DTRACE_PROBE1(proto__fallback__begin, struct sonode *, so); 2313 error = (*fbfunc)(so->so_proto_handle, q, direct, so_quiesced_cb); 2314 DTRACE_PROBE1(proto__fallback__end, struct sonode *, so); 2315 2316 if (error != 0) { 2317 /* protocol was unable to do a fallback, revert the sonode */ 2318 sotpi_revert_sonode(so, cr); 2319 goto out; 2320 } 2321 2322 /* 2323 * Walk the accept queue and notify the proto that they should 2324 * fall back to TPI. The protocol will send up the T_CONN_IND. 2325 */ 2326 nso = so->so_acceptq_head; 2327 while (nso != NULL) { 2328 int rval; 2329 2330 DTRACE_PROBE1(proto__fallback__begin, struct sonode *, nso); 2331 rval = (*fbfunc)(nso->so_proto_handle, NULL, direct, NULL); 2332 DTRACE_PROBE1(proto__fallback__end, struct sonode *, nso); 2333 if (rval != 0) { 2334 zcmn_err(getzoneid(), CE_WARN, 2335 "Failed to convert socket in accept queue to TPI. " 2336 "Pid = %d\n", curproc->p_pid); 2337 } 2338 nso = nso->so_acceptq_next; 2339 } 2340 2341 /* 2342 * Now flush the acceptq, this will destroy all sockets. They will 2343 * be recreated in sotpi_accept(). 2344 */ 2345 so_acceptq_flush(so, B_FALSE); 2346 2347 mutex_enter(&so->so_lock); 2348 so->so_state |= SS_FALLBACK_COMP; 2349 mutex_exit(&so->so_lock); 2350 2351 /* 2352 * Swap the sonode ops. Socket opertations that come in once this 2353 * is done will proceed without blocking. 2354 */ 2355 so->so_ops = &sotpi_sonodeops; 2356 2357 /* 2358 * Wake up any threads stuck in poll. This is needed since the poll 2359 * head changes when the fallback happens (moves from the sonode to 2360 * the STREAMS head). 2361 */ 2362 pollwakeup(&so->so_poll_list, POLLERR); 2363 out: 2364 so_end_fallback(so); 2365 2366 if (error != 0) { 2367 #ifdef DEBUG 2368 so_integrity_check(so, &origso); 2369 #endif 2370 zcmn_err(getzoneid(), CE_WARN, 2371 "Failed to convert socket to TPI (err=%d). Pid = %d\n", 2372 error, curproc->p_pid); 2373 if (newsp != NULL) 2374 SOCKPARAMS_DEC_REF(newsp); 2375 } 2376 2377 return (error); 2378 } 2379