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