1 /*- 2 * Implementation of SVID messages 3 * 4 * Author: Daniel Boulet 5 * 6 * Copyright 1993 Daniel Boulet and RTMX Inc. 7 * 8 * This system call was implemented by Daniel Boulet under contract from RTMX. 9 * 10 * Redistribution and use in source forms, with and without modification, 11 * are permitted provided that this entire comment appears intact. 12 * 13 * Redistribution in binary form may occur without any restrictions. 14 * Obviously, it would be nice if you gave credit where credit is due 15 * but requiring it would be too onerous. 16 * 17 * This software is provided ``AS IS'' without any warranties of any kind. 18 */ 19 /*- 20 * Copyright (c) 2003-2005 McAfee, Inc. 21 * All rights reserved. 22 * 23 * This software was developed for the FreeBSD Project in part by McAfee 24 * Research, the Security Research Division of McAfee, Inc under DARPA/SPAWAR 25 * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS research 26 * program. 27 * 28 * Redistribution and use in source and binary forms, with or without 29 * modification, are permitted provided that the following conditions 30 * are met: 31 * 1. Redistributions of source code must retain the above copyright 32 * notice, this list of conditions and the following disclaimer. 33 * 2. Redistributions in binary form must reproduce the above copyright 34 * notice, this list of conditions and the following disclaimer in the 35 * documentation and/or other materials provided with the distribution. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_sysvipc.h" 54 55 #include <sys/param.h> 56 #include <sys/systm.h> 57 #include <sys/sysproto.h> 58 #include <sys/kernel.h> 59 #include <sys/priv.h> 60 #include <sys/proc.h> 61 #include <sys/lock.h> 62 #include <sys/mutex.h> 63 #include <sys/module.h> 64 #include <sys/msg.h> 65 #include <sys/syscall.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/sysent.h> 68 #include <sys/sysctl.h> 69 #include <sys/malloc.h> 70 #include <sys/jail.h> 71 72 #include <security/mac/mac_framework.h> 73 74 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 75 76 static void msginit(void); 77 static int msgunload(void); 78 static int sysvmsg_modload(struct module *, int, void *); 79 80 #ifdef MSG_DEBUG 81 #define DPRINTF(a) printf a 82 #else 83 #define DPRINTF(a) (void)0 84 #endif 85 86 static void msg_freehdr(struct msg *msghdr); 87 88 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 89 static sy_call_t *msgcalls[] = { 90 (sy_call_t *)msgctl, (sy_call_t *)msgget, 91 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 92 }; 93 94 #ifndef MSGSSZ 95 #define MSGSSZ 8 /* Each segment must be 2^N long */ 96 #endif 97 #ifndef MSGSEG 98 #define MSGSEG 2048 /* must be less than 32767 */ 99 #endif 100 #define MSGMAX (MSGSSZ*MSGSEG) 101 #ifndef MSGMNB 102 #define MSGMNB 2048 /* max # of bytes in a queue */ 103 #endif 104 #ifndef MSGMNI 105 #define MSGMNI 40 106 #endif 107 #ifndef MSGTQL 108 #define MSGTQL 40 109 #endif 110 111 /* 112 * Based on the configuration parameters described in an SVR2 (yes, two) 113 * config(1m) man page. 114 * 115 * Each message is broken up and stored in segments that are msgssz bytes 116 * long. For efficiency reasons, this should be a power of two. Also, 117 * it doesn't make sense if it is less than 8 or greater than about 256. 118 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 119 * two between 8 and 1024 inclusive (and panic's if it isn't). 120 */ 121 struct msginfo msginfo = { 122 MSGMAX, /* max chars in a message */ 123 MSGMNI, /* # of message queue identifiers */ 124 MSGMNB, /* max chars in a queue */ 125 MSGTQL, /* max messages in system */ 126 MSGSSZ, /* size of a message segment */ 127 /* (must be small power of 2 greater than 4) */ 128 MSGSEG /* number of message segments */ 129 }; 130 131 /* 132 * macros to convert between msqid_ds's and msqid's. 133 * (specific to this implementation) 134 */ 135 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 136 #define MSQID_IX(id) ((id) & 0xffff) 137 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 138 139 /* 140 * The rest of this file is specific to this particular implementation. 141 */ 142 143 struct msgmap { 144 short next; /* next segment in buffer */ 145 /* -1 -> available */ 146 /* 0..(MSGSEG-1) -> index of next segment */ 147 }; 148 149 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 150 151 static int nfree_msgmaps; /* # of free map entries */ 152 static short free_msgmaps; /* head of linked list of free map entries */ 153 static struct msg *free_msghdrs;/* list of free msg headers */ 154 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 155 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 156 static struct msg *msghdrs; /* MSGTQL msg headers */ 157 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 158 static struct mtx msq_mtx; /* global mutex for message queues. */ 159 160 static void 161 msginit() 162 { 163 register int i; 164 165 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 166 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 167 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 168 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 169 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 170 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 171 172 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 173 if (msgpool == NULL) 174 panic("msgpool is NULL"); 175 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 176 if (msgmaps == NULL) 177 panic("msgmaps is NULL"); 178 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 179 if (msghdrs == NULL) 180 panic("msghdrs is NULL"); 181 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 182 M_WAITOK); 183 if (msqids == NULL) 184 panic("msqids is NULL"); 185 186 /* 187 * msginfo.msgssz should be a power of two for efficiency reasons. 188 * It is also pretty silly if msginfo.msgssz is less than 8 189 * or greater than about 256 so ... 190 */ 191 192 i = 8; 193 while (i < 1024 && i != msginfo.msgssz) 194 i <<= 1; 195 if (i != msginfo.msgssz) { 196 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 197 msginfo.msgssz)); 198 panic("msginfo.msgssz not a small power of 2"); 199 } 200 201 if (msginfo.msgseg > 32767) { 202 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 203 panic("msginfo.msgseg > 32767"); 204 } 205 206 if (msgmaps == NULL) 207 panic("msgmaps is NULL"); 208 209 for (i = 0; i < msginfo.msgseg; i++) { 210 if (i > 0) 211 msgmaps[i-1].next = i; 212 msgmaps[i].next = -1; /* implies entry is available */ 213 } 214 free_msgmaps = 0; 215 nfree_msgmaps = msginfo.msgseg; 216 217 if (msghdrs == NULL) 218 panic("msghdrs is NULL"); 219 220 for (i = 0; i < msginfo.msgtql; i++) { 221 msghdrs[i].msg_type = 0; 222 if (i > 0) 223 msghdrs[i-1].msg_next = &msghdrs[i]; 224 msghdrs[i].msg_next = NULL; 225 #ifdef MAC 226 mac_sysvmsg_init(&msghdrs[i]); 227 #endif 228 } 229 free_msghdrs = &msghdrs[0]; 230 231 if (msqids == NULL) 232 panic("msqids is NULL"); 233 234 for (i = 0; i < msginfo.msgmni; i++) { 235 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 236 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 237 msqids[i].u.msg_perm.mode = 0; 238 #ifdef MAC 239 mac_sysvmsq_init(&msqids[i]); 240 #endif 241 } 242 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 243 } 244 245 static int 246 msgunload() 247 { 248 struct msqid_kernel *msqkptr; 249 int msqid; 250 #ifdef MAC 251 int i; 252 #endif 253 254 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 255 /* 256 * Look for an unallocated and unlocked msqid_ds. 257 * msqid_ds's can be locked by msgsnd or msgrcv while 258 * they are copying the message in/out. We can't 259 * re-use the entry until they release it. 260 */ 261 msqkptr = &msqids[msqid]; 262 if (msqkptr->u.msg_qbytes != 0 || 263 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 264 break; 265 } 266 if (msqid != msginfo.msgmni) 267 return (EBUSY); 268 269 #ifdef MAC 270 for (i = 0; i < msginfo.msgtql; i++) 271 mac_sysvmsg_destroy(&msghdrs[i]); 272 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 273 mac_sysvmsq_destroy(&msqids[msqid]); 274 #endif 275 free(msgpool, M_MSG); 276 free(msgmaps, M_MSG); 277 free(msghdrs, M_MSG); 278 free(msqids, M_MSG); 279 mtx_destroy(&msq_mtx); 280 return (0); 281 } 282 283 284 static int 285 sysvmsg_modload(struct module *module, int cmd, void *arg) 286 { 287 int error = 0; 288 289 switch (cmd) { 290 case MOD_LOAD: 291 msginit(); 292 break; 293 case MOD_UNLOAD: 294 error = msgunload(); 295 break; 296 case MOD_SHUTDOWN: 297 break; 298 default: 299 error = EINVAL; 300 break; 301 } 302 return (error); 303 } 304 305 static moduledata_t sysvmsg_mod = { 306 "sysvmsg", 307 &sysvmsg_modload, 308 NULL 309 }; 310 311 SYSCALL_MODULE_HELPER(msgsys); 312 SYSCALL_MODULE_HELPER(msgctl); 313 SYSCALL_MODULE_HELPER(msgget); 314 SYSCALL_MODULE_HELPER(msgsnd); 315 SYSCALL_MODULE_HELPER(msgrcv); 316 317 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 318 MODULE_VERSION(sysvmsg, 1); 319 320 /* 321 * Entry point for all MSG calls. 322 */ 323 int 324 msgsys(td, uap) 325 struct thread *td; 326 /* XXX actually varargs. */ 327 struct msgsys_args /* { 328 int which; 329 int a2; 330 int a3; 331 int a4; 332 int a5; 333 int a6; 334 } */ *uap; 335 { 336 int error; 337 338 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 339 return (ENOSYS); 340 if (uap->which < 0 || 341 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 342 return (EINVAL); 343 error = (*msgcalls[uap->which])(td, &uap->a2); 344 return (error); 345 } 346 347 static void 348 msg_freehdr(msghdr) 349 struct msg *msghdr; 350 { 351 while (msghdr->msg_ts > 0) { 352 short next; 353 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 354 panic("msghdr->msg_spot out of range"); 355 next = msgmaps[msghdr->msg_spot].next; 356 msgmaps[msghdr->msg_spot].next = free_msgmaps; 357 free_msgmaps = msghdr->msg_spot; 358 nfree_msgmaps++; 359 msghdr->msg_spot = next; 360 if (msghdr->msg_ts >= msginfo.msgssz) 361 msghdr->msg_ts -= msginfo.msgssz; 362 else 363 msghdr->msg_ts = 0; 364 } 365 if (msghdr->msg_spot != -1) 366 panic("msghdr->msg_spot != -1"); 367 msghdr->msg_next = free_msghdrs; 368 free_msghdrs = msghdr; 369 #ifdef MAC 370 mac_sysvmsg_cleanup(msghdr); 371 #endif 372 } 373 374 #ifndef _SYS_SYSPROTO_H_ 375 struct msgctl_args { 376 int msqid; 377 int cmd; 378 struct msqid_ds *buf; 379 }; 380 #endif 381 int 382 msgctl(td, uap) 383 struct thread *td; 384 register struct msgctl_args *uap; 385 { 386 int msqid = uap->msqid; 387 int cmd = uap->cmd; 388 struct msqid_ds msqbuf; 389 int error; 390 391 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 392 if (cmd == IPC_SET && 393 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 394 return (error); 395 error = kern_msgctl(td, msqid, cmd, &msqbuf); 396 if (cmd == IPC_STAT && error == 0) 397 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 398 return (error); 399 } 400 401 int 402 kern_msgctl(td, msqid, cmd, msqbuf) 403 struct thread *td; 404 int msqid; 405 int cmd; 406 struct msqid_ds *msqbuf; 407 { 408 int rval, error, msqix; 409 register struct msqid_kernel *msqkptr; 410 411 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 412 return (ENOSYS); 413 414 msqix = IPCID_TO_IX(msqid); 415 416 if (msqix < 0 || msqix >= msginfo.msgmni) { 417 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 418 msginfo.msgmni)); 419 return (EINVAL); 420 } 421 422 msqkptr = &msqids[msqix]; 423 424 mtx_lock(&msq_mtx); 425 if (msqkptr->u.msg_qbytes == 0) { 426 DPRINTF(("no such msqid\n")); 427 error = EINVAL; 428 goto done2; 429 } 430 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 431 DPRINTF(("wrong sequence number\n")); 432 error = EINVAL; 433 goto done2; 434 } 435 #ifdef MAC 436 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 437 if (error != 0) 438 goto done2; 439 #endif 440 441 error = 0; 442 rval = 0; 443 444 switch (cmd) { 445 446 case IPC_RMID: 447 { 448 struct msg *msghdr; 449 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 450 goto done2; 451 452 #ifdef MAC 453 /* 454 * Check that the thread has MAC access permissions to 455 * individual msghdrs. Note: We need to do this in a 456 * separate loop because the actual loop alters the 457 * msq/msghdr info as it progresses, and there is no going 458 * back if half the way through we discover that the 459 * thread cannot free a certain msghdr. The msq will get 460 * into an inconsistent state. 461 */ 462 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 463 msghdr = msghdr->msg_next) { 464 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 465 if (error != 0) 466 goto done2; 467 } 468 #endif 469 470 /* Free the message headers */ 471 msghdr = msqkptr->u.msg_first; 472 while (msghdr != NULL) { 473 struct msg *msghdr_tmp; 474 475 /* Free the segments of each message */ 476 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 477 msqkptr->u.msg_qnum--; 478 msghdr_tmp = msghdr; 479 msghdr = msghdr->msg_next; 480 msg_freehdr(msghdr_tmp); 481 } 482 483 if (msqkptr->u.msg_cbytes != 0) 484 panic("msg_cbytes is screwed up"); 485 if (msqkptr->u.msg_qnum != 0) 486 panic("msg_qnum is screwed up"); 487 488 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 489 490 #ifdef MAC 491 mac_sysvmsq_cleanup(msqkptr); 492 #endif 493 494 wakeup(msqkptr); 495 } 496 497 break; 498 499 case IPC_SET: 500 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 501 goto done2; 502 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 503 error = priv_check(td, PRIV_IPC_MSGSIZE); 504 if (error) 505 goto done2; 506 } 507 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 508 DPRINTF(("can't increase msg_qbytes beyond %d" 509 "(truncating)\n", msginfo.msgmnb)); 510 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 511 } 512 if (msqbuf->msg_qbytes == 0) { 513 DPRINTF(("can't reduce msg_qbytes to 0\n")); 514 error = EINVAL; /* non-standard errno! */ 515 goto done2; 516 } 517 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 518 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 519 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 520 (msqbuf->msg_perm.mode & 0777); 521 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 522 msqkptr->u.msg_ctime = time_second; 523 break; 524 525 case IPC_STAT: 526 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 527 DPRINTF(("requester doesn't have read access\n")); 528 goto done2; 529 } 530 *msqbuf = msqkptr->u; 531 break; 532 533 default: 534 DPRINTF(("invalid command %d\n", cmd)); 535 error = EINVAL; 536 goto done2; 537 } 538 539 if (error == 0) 540 td->td_retval[0] = rval; 541 done2: 542 mtx_unlock(&msq_mtx); 543 return (error); 544 } 545 546 #ifndef _SYS_SYSPROTO_H_ 547 struct msgget_args { 548 key_t key; 549 int msgflg; 550 }; 551 #endif 552 int 553 msgget(td, uap) 554 struct thread *td; 555 register struct msgget_args *uap; 556 { 557 int msqid, error = 0; 558 int key = uap->key; 559 int msgflg = uap->msgflg; 560 struct ucred *cred = td->td_ucred; 561 register struct msqid_kernel *msqkptr = NULL; 562 563 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 564 565 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 566 return (ENOSYS); 567 568 mtx_lock(&msq_mtx); 569 if (key != IPC_PRIVATE) { 570 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 571 msqkptr = &msqids[msqid]; 572 if (msqkptr->u.msg_qbytes != 0 && 573 msqkptr->u.msg_perm.key == key) 574 break; 575 } 576 if (msqid < msginfo.msgmni) { 577 DPRINTF(("found public key\n")); 578 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 579 DPRINTF(("not exclusive\n")); 580 error = EEXIST; 581 goto done2; 582 } 583 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 584 msgflg & 0700))) { 585 DPRINTF(("requester doesn't have 0%o access\n", 586 msgflg & 0700)); 587 goto done2; 588 } 589 #ifdef MAC 590 error = mac_sysvmsq_check_msqget(cred, msqkptr); 591 if (error != 0) 592 goto done2; 593 #endif 594 goto found; 595 } 596 } 597 598 DPRINTF(("need to allocate the msqid_ds\n")); 599 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 600 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 601 /* 602 * Look for an unallocated and unlocked msqid_ds. 603 * msqid_ds's can be locked by msgsnd or msgrcv while 604 * they are copying the message in/out. We can't 605 * re-use the entry until they release it. 606 */ 607 msqkptr = &msqids[msqid]; 608 if (msqkptr->u.msg_qbytes == 0 && 609 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 610 break; 611 } 612 if (msqid == msginfo.msgmni) { 613 DPRINTF(("no more msqid_ds's available\n")); 614 error = ENOSPC; 615 goto done2; 616 } 617 DPRINTF(("msqid %d is available\n", msqid)); 618 msqkptr->u.msg_perm.key = key; 619 msqkptr->u.msg_perm.cuid = cred->cr_uid; 620 msqkptr->u.msg_perm.uid = cred->cr_uid; 621 msqkptr->u.msg_perm.cgid = cred->cr_gid; 622 msqkptr->u.msg_perm.gid = cred->cr_gid; 623 msqkptr->u.msg_perm.mode = (msgflg & 0777); 624 /* Make sure that the returned msqid is unique */ 625 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 626 msqkptr->u.msg_first = NULL; 627 msqkptr->u.msg_last = NULL; 628 msqkptr->u.msg_cbytes = 0; 629 msqkptr->u.msg_qnum = 0; 630 msqkptr->u.msg_qbytes = msginfo.msgmnb; 631 msqkptr->u.msg_lspid = 0; 632 msqkptr->u.msg_lrpid = 0; 633 msqkptr->u.msg_stime = 0; 634 msqkptr->u.msg_rtime = 0; 635 msqkptr->u.msg_ctime = time_second; 636 #ifdef MAC 637 mac_sysvmsq_create(cred, msqkptr); 638 #endif 639 } else { 640 DPRINTF(("didn't find it and wasn't asked to create it\n")); 641 error = ENOENT; 642 goto done2; 643 } 644 645 found: 646 /* Construct the unique msqid */ 647 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 648 done2: 649 mtx_unlock(&msq_mtx); 650 return (error); 651 } 652 653 #ifndef _SYS_SYSPROTO_H_ 654 struct msgsnd_args { 655 int msqid; 656 const void *msgp; 657 size_t msgsz; 658 int msgflg; 659 }; 660 #endif 661 int 662 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 663 struct thread *td; 664 int msqid; 665 const void *msgp; /* XXX msgp is actually mtext. */ 666 size_t msgsz; 667 int msgflg; 668 long mtype; 669 { 670 int msqix, segs_needed, error = 0; 671 register struct msqid_kernel *msqkptr; 672 register struct msg *msghdr; 673 short next; 674 675 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 676 return (ENOSYS); 677 678 mtx_lock(&msq_mtx); 679 msqix = IPCID_TO_IX(msqid); 680 681 if (msqix < 0 || msqix >= msginfo.msgmni) { 682 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 683 msginfo.msgmni)); 684 error = EINVAL; 685 goto done2; 686 } 687 688 msqkptr = &msqids[msqix]; 689 if (msqkptr->u.msg_qbytes == 0) { 690 DPRINTF(("no such message queue id\n")); 691 error = EINVAL; 692 goto done2; 693 } 694 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 695 DPRINTF(("wrong sequence number\n")); 696 error = EINVAL; 697 goto done2; 698 } 699 700 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 701 DPRINTF(("requester doesn't have write access\n")); 702 goto done2; 703 } 704 705 #ifdef MAC 706 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 707 if (error != 0) 708 goto done2; 709 #endif 710 711 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 712 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 713 msginfo.msgssz, segs_needed)); 714 for (;;) { 715 int need_more_resources = 0; 716 717 /* 718 * check msgsz 719 * (inside this loop in case msg_qbytes changes while we sleep) 720 */ 721 722 if (msgsz > msqkptr->u.msg_qbytes) { 723 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 724 error = EINVAL; 725 goto done2; 726 } 727 728 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 729 DPRINTF(("msqid is locked\n")); 730 need_more_resources = 1; 731 } 732 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 733 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 734 need_more_resources = 1; 735 } 736 if (segs_needed > nfree_msgmaps) { 737 DPRINTF(("segs_needed > nfree_msgmaps\n")); 738 need_more_resources = 1; 739 } 740 if (free_msghdrs == NULL) { 741 DPRINTF(("no more msghdrs\n")); 742 need_more_resources = 1; 743 } 744 745 if (need_more_resources) { 746 int we_own_it; 747 748 if ((msgflg & IPC_NOWAIT) != 0) { 749 DPRINTF(("need more resources but caller " 750 "doesn't want to wait\n")); 751 error = EAGAIN; 752 goto done2; 753 } 754 755 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 756 DPRINTF(("we don't own the msqid_ds\n")); 757 we_own_it = 0; 758 } else { 759 /* Force later arrivals to wait for our 760 request */ 761 DPRINTF(("we own the msqid_ds\n")); 762 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 763 we_own_it = 1; 764 } 765 DPRINTF(("msgsnd: goodnight\n")); 766 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 767 "msgsnd", hz); 768 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 769 if (we_own_it) 770 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 771 if (error == EWOULDBLOCK) { 772 DPRINTF(("msgsnd: timed out\n")); 773 continue; 774 } 775 if (error != 0) { 776 DPRINTF(("msgsnd: interrupted system call\n")); 777 error = EINTR; 778 goto done2; 779 } 780 781 /* 782 * Make sure that the msq queue still exists 783 */ 784 785 if (msqkptr->u.msg_qbytes == 0) { 786 DPRINTF(("msqid deleted\n")); 787 error = EIDRM; 788 goto done2; 789 } 790 791 } else { 792 DPRINTF(("got all the resources that we need\n")); 793 break; 794 } 795 } 796 797 /* 798 * We have the resources that we need. 799 * Make sure! 800 */ 801 802 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 803 panic("msg_perm.mode & MSG_LOCKED"); 804 if (segs_needed > nfree_msgmaps) 805 panic("segs_needed > nfree_msgmaps"); 806 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 807 panic("msgsz + msg_cbytes > msg_qbytes"); 808 if (free_msghdrs == NULL) 809 panic("no more msghdrs"); 810 811 /* 812 * Re-lock the msqid_ds in case we page-fault when copying in the 813 * message 814 */ 815 816 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 817 panic("msqid_ds is already locked"); 818 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 819 820 /* 821 * Allocate a message header 822 */ 823 824 msghdr = free_msghdrs; 825 free_msghdrs = msghdr->msg_next; 826 msghdr->msg_spot = -1; 827 msghdr->msg_ts = msgsz; 828 msghdr->msg_type = mtype; 829 #ifdef MAC 830 /* 831 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 832 * immediately? Or, should it be checked just before the msg is 833 * enqueued in the msgq (as it is done now)? 834 */ 835 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 836 #endif 837 838 /* 839 * Allocate space for the message 840 */ 841 842 while (segs_needed > 0) { 843 if (nfree_msgmaps <= 0) 844 panic("not enough msgmaps"); 845 if (free_msgmaps == -1) 846 panic("nil free_msgmaps"); 847 next = free_msgmaps; 848 if (next <= -1) 849 panic("next too low #1"); 850 if (next >= msginfo.msgseg) 851 panic("next out of range #1"); 852 DPRINTF(("allocating segment %d to message\n", next)); 853 free_msgmaps = msgmaps[next].next; 854 nfree_msgmaps--; 855 msgmaps[next].next = msghdr->msg_spot; 856 msghdr->msg_spot = next; 857 segs_needed--; 858 } 859 860 /* 861 * Validate the message type 862 */ 863 864 if (msghdr->msg_type < 1) { 865 msg_freehdr(msghdr); 866 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 867 wakeup(msqkptr); 868 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 869 error = EINVAL; 870 goto done2; 871 } 872 873 /* 874 * Copy in the message body 875 */ 876 877 next = msghdr->msg_spot; 878 while (msgsz > 0) { 879 size_t tlen; 880 if (msgsz > msginfo.msgssz) 881 tlen = msginfo.msgssz; 882 else 883 tlen = msgsz; 884 if (next <= -1) 885 panic("next too low #2"); 886 if (next >= msginfo.msgseg) 887 panic("next out of range #2"); 888 mtx_unlock(&msq_mtx); 889 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 890 tlen)) != 0) { 891 mtx_lock(&msq_mtx); 892 DPRINTF(("error %d copying in message segment\n", 893 error)); 894 msg_freehdr(msghdr); 895 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 896 wakeup(msqkptr); 897 goto done2; 898 } 899 mtx_lock(&msq_mtx); 900 msgsz -= tlen; 901 msgp = (const char *)msgp + tlen; 902 next = msgmaps[next].next; 903 } 904 if (next != -1) 905 panic("didn't use all the msg segments"); 906 907 /* 908 * We've got the message. Unlock the msqid_ds. 909 */ 910 911 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 912 913 /* 914 * Make sure that the msqid_ds is still allocated. 915 */ 916 917 if (msqkptr->u.msg_qbytes == 0) { 918 msg_freehdr(msghdr); 919 wakeup(msqkptr); 920 error = EIDRM; 921 goto done2; 922 } 923 924 #ifdef MAC 925 /* 926 * Note: Since the task/thread allocates the msghdr and usually 927 * primes it with its own MAC label, for a majority of policies, it 928 * won't be necessary to check whether the msghdr has access 929 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 930 * suffice in that case. However, this hook may be required where 931 * individual policies derive a non-identical label for the msghdr 932 * from the current thread label and may want to check the msghdr 933 * enqueue permissions, along with read/write permissions to the 934 * msgq. 935 */ 936 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 937 if (error != 0) { 938 msg_freehdr(msghdr); 939 wakeup(msqkptr); 940 goto done2; 941 } 942 #endif 943 944 /* 945 * Put the message into the queue 946 */ 947 if (msqkptr->u.msg_first == NULL) { 948 msqkptr->u.msg_first = msghdr; 949 msqkptr->u.msg_last = msghdr; 950 } else { 951 msqkptr->u.msg_last->msg_next = msghdr; 952 msqkptr->u.msg_last = msghdr; 953 } 954 msqkptr->u.msg_last->msg_next = NULL; 955 956 msqkptr->u.msg_cbytes += msghdr->msg_ts; 957 msqkptr->u.msg_qnum++; 958 msqkptr->u.msg_lspid = td->td_proc->p_pid; 959 msqkptr->u.msg_stime = time_second; 960 961 wakeup(msqkptr); 962 td->td_retval[0] = 0; 963 done2: 964 mtx_unlock(&msq_mtx); 965 return (error); 966 } 967 968 int 969 msgsnd(td, uap) 970 struct thread *td; 971 register struct msgsnd_args *uap; 972 { 973 int error; 974 long mtype; 975 976 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 977 uap->msgsz, uap->msgflg)); 978 979 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 980 DPRINTF(("error %d copying the message type\n", error)); 981 return (error); 982 } 983 return (kern_msgsnd(td, uap->msqid, 984 (const char *)uap->msgp + sizeof(mtype), 985 uap->msgsz, uap->msgflg, mtype)); 986 } 987 988 #ifndef _SYS_SYSPROTO_H_ 989 struct msgrcv_args { 990 int msqid; 991 void *msgp; 992 size_t msgsz; 993 long msgtyp; 994 int msgflg; 995 }; 996 #endif 997 int 998 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 999 struct thread *td; 1000 int msqid; 1001 void *msgp; /* XXX msgp is actually mtext. */ 1002 size_t msgsz; 1003 long msgtyp; 1004 int msgflg; 1005 long *mtype; 1006 { 1007 size_t len; 1008 register struct msqid_kernel *msqkptr; 1009 register struct msg *msghdr; 1010 int msqix, error = 0; 1011 short next; 1012 1013 if (!prison_allow(td->td_ucred, PR_ALLOW_SYSVIPC)) 1014 return (ENOSYS); 1015 1016 msqix = IPCID_TO_IX(msqid); 1017 1018 if (msqix < 0 || msqix >= msginfo.msgmni) { 1019 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1020 msginfo.msgmni)); 1021 return (EINVAL); 1022 } 1023 1024 msqkptr = &msqids[msqix]; 1025 mtx_lock(&msq_mtx); 1026 if (msqkptr->u.msg_qbytes == 0) { 1027 DPRINTF(("no such message queue id\n")); 1028 error = EINVAL; 1029 goto done2; 1030 } 1031 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1032 DPRINTF(("wrong sequence number\n")); 1033 error = EINVAL; 1034 goto done2; 1035 } 1036 1037 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1038 DPRINTF(("requester doesn't have read access\n")); 1039 goto done2; 1040 } 1041 1042 #ifdef MAC 1043 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1044 if (error != 0) 1045 goto done2; 1046 #endif 1047 1048 msghdr = NULL; 1049 while (msghdr == NULL) { 1050 if (msgtyp == 0) { 1051 msghdr = msqkptr->u.msg_first; 1052 if (msghdr != NULL) { 1053 if (msgsz < msghdr->msg_ts && 1054 (msgflg & MSG_NOERROR) == 0) { 1055 DPRINTF(("first message on the queue " 1056 "is too big (want %zu, got %d)\n", 1057 msgsz, msghdr->msg_ts)); 1058 error = E2BIG; 1059 goto done2; 1060 } 1061 #ifdef MAC 1062 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1063 msghdr); 1064 if (error != 0) 1065 goto done2; 1066 #endif 1067 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1068 msqkptr->u.msg_first = NULL; 1069 msqkptr->u.msg_last = NULL; 1070 } else { 1071 msqkptr->u.msg_first = msghdr->msg_next; 1072 if (msqkptr->u.msg_first == NULL) 1073 panic("msg_first/last screwed up #1"); 1074 } 1075 } 1076 } else { 1077 struct msg *previous; 1078 struct msg **prev; 1079 1080 previous = NULL; 1081 prev = &(msqkptr->u.msg_first); 1082 while ((msghdr = *prev) != NULL) { 1083 /* 1084 * Is this message's type an exact match or is 1085 * this message's type less than or equal to 1086 * the absolute value of a negative msgtyp? 1087 * Note that the second half of this test can 1088 * NEVER be true if msgtyp is positive since 1089 * msg_type is always positive! 1090 */ 1091 1092 if (msgtyp == msghdr->msg_type || 1093 msghdr->msg_type <= -msgtyp) { 1094 DPRINTF(("found message type %ld, " 1095 "requested %ld\n", 1096 msghdr->msg_type, msgtyp)); 1097 if (msgsz < msghdr->msg_ts && 1098 (msgflg & MSG_NOERROR) == 0) { 1099 DPRINTF(("requested message " 1100 "on the queue is too big " 1101 "(want %zu, got %hu)\n", 1102 msgsz, msghdr->msg_ts)); 1103 error = E2BIG; 1104 goto done2; 1105 } 1106 #ifdef MAC 1107 error = mac_sysvmsq_check_msgrcv( 1108 td->td_ucred, msghdr); 1109 if (error != 0) 1110 goto done2; 1111 #endif 1112 *prev = msghdr->msg_next; 1113 if (msghdr == msqkptr->u.msg_last) { 1114 if (previous == NULL) { 1115 if (prev != 1116 &msqkptr->u.msg_first) 1117 panic("msg_first/last screwed up #2"); 1118 msqkptr->u.msg_first = 1119 NULL; 1120 msqkptr->u.msg_last = 1121 NULL; 1122 } else { 1123 if (prev == 1124 &msqkptr->u.msg_first) 1125 panic("msg_first/last screwed up #3"); 1126 msqkptr->u.msg_last = 1127 previous; 1128 } 1129 } 1130 break; 1131 } 1132 previous = msghdr; 1133 prev = &(msghdr->msg_next); 1134 } 1135 } 1136 1137 /* 1138 * We've either extracted the msghdr for the appropriate 1139 * message or there isn't one. 1140 * If there is one then bail out of this loop. 1141 */ 1142 1143 if (msghdr != NULL) 1144 break; 1145 1146 /* 1147 * Hmph! No message found. Does the user want to wait? 1148 */ 1149 1150 if ((msgflg & IPC_NOWAIT) != 0) { 1151 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1152 msgtyp)); 1153 /* The SVID says to return ENOMSG. */ 1154 error = ENOMSG; 1155 goto done2; 1156 } 1157 1158 /* 1159 * Wait for something to happen 1160 */ 1161 1162 DPRINTF(("msgrcv: goodnight\n")); 1163 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1164 "msgrcv", 0); 1165 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1166 1167 if (error != 0) { 1168 DPRINTF(("msgrcv: interrupted system call\n")); 1169 error = EINTR; 1170 goto done2; 1171 } 1172 1173 /* 1174 * Make sure that the msq queue still exists 1175 */ 1176 1177 if (msqkptr->u.msg_qbytes == 0 || 1178 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1179 DPRINTF(("msqid deleted\n")); 1180 error = EIDRM; 1181 goto done2; 1182 } 1183 } 1184 1185 /* 1186 * Return the message to the user. 1187 * 1188 * First, do the bookkeeping (before we risk being interrupted). 1189 */ 1190 1191 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1192 msqkptr->u.msg_qnum--; 1193 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1194 msqkptr->u.msg_rtime = time_second; 1195 1196 /* 1197 * Make msgsz the actual amount that we'll be returning. 1198 * Note that this effectively truncates the message if it is too long 1199 * (since msgsz is never increased). 1200 */ 1201 1202 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1203 msghdr->msg_ts)); 1204 if (msgsz > msghdr->msg_ts) 1205 msgsz = msghdr->msg_ts; 1206 *mtype = msghdr->msg_type; 1207 1208 /* 1209 * Return the segments to the user 1210 */ 1211 1212 next = msghdr->msg_spot; 1213 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1214 size_t tlen; 1215 1216 if (msgsz - len > msginfo.msgssz) 1217 tlen = msginfo.msgssz; 1218 else 1219 tlen = msgsz - len; 1220 if (next <= -1) 1221 panic("next too low #3"); 1222 if (next >= msginfo.msgseg) 1223 panic("next out of range #3"); 1224 mtx_unlock(&msq_mtx); 1225 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1226 mtx_lock(&msq_mtx); 1227 if (error != 0) { 1228 DPRINTF(("error (%d) copying out message segment\n", 1229 error)); 1230 msg_freehdr(msghdr); 1231 wakeup(msqkptr); 1232 goto done2; 1233 } 1234 msgp = (char *)msgp + tlen; 1235 next = msgmaps[next].next; 1236 } 1237 1238 /* 1239 * Done, return the actual number of bytes copied out. 1240 */ 1241 1242 msg_freehdr(msghdr); 1243 wakeup(msqkptr); 1244 td->td_retval[0] = msgsz; 1245 done2: 1246 mtx_unlock(&msq_mtx); 1247 return (error); 1248 } 1249 1250 int 1251 msgrcv(td, uap) 1252 struct thread *td; 1253 register struct msgrcv_args *uap; 1254 { 1255 int error; 1256 long mtype; 1257 1258 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1259 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1260 1261 if ((error = kern_msgrcv(td, uap->msqid, 1262 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1263 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1264 return (error); 1265 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1266 DPRINTF(("error %d copying the message type\n", error)); 1267 return (error); 1268 } 1269 1270 static int 1271 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1272 { 1273 1274 return (SYSCTL_OUT(req, msqids, 1275 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1276 } 1277 1278 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1279 "Maximum message size"); 1280 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1281 "Number of message queue identifiers"); 1282 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1283 "Maximum number of bytes in a queue"); 1284 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1285 "Maximum number of messages in the system"); 1286 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1287 "Size of a message segment"); 1288 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1289 "Number of message segments"); 1290 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1291 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1292