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 #include "opt_mac.h" 55 56 #include <sys/param.h> 57 #include <sys/systm.h> 58 #include <sys/sysproto.h> 59 #include <sys/kernel.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/lock.h> 63 #include <sys/mutex.h> 64 #include <sys/module.h> 65 #include <sys/msg.h> 66 #include <sys/syscall.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysent.h> 69 #include <sys/sysctl.h> 70 #include <sys/malloc.h> 71 #include <sys/jail.h> 72 73 #include <security/mac/mac_framework.h> 74 75 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 76 77 static void msginit(void); 78 static int msgunload(void); 79 static int sysvmsg_modload(struct module *, int, void *); 80 81 #ifdef MSG_DEBUG 82 #define DPRINTF(a) printf a 83 #else 84 #define DPRINTF(a) 85 #endif 86 87 static void msg_freehdr(struct msg *msghdr); 88 89 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 90 static sy_call_t *msgcalls[] = { 91 (sy_call_t *)msgctl, (sy_call_t *)msgget, 92 (sy_call_t *)msgsnd, (sy_call_t *)msgrcv 93 }; 94 95 #ifndef MSGSSZ 96 #define MSGSSZ 8 /* Each segment must be 2^N long */ 97 #endif 98 #ifndef MSGSEG 99 #define MSGSEG 2048 /* must be less than 32767 */ 100 #endif 101 #define MSGMAX (MSGSSZ*MSGSEG) 102 #ifndef MSGMNB 103 #define MSGMNB 2048 /* max # of bytes in a queue */ 104 #endif 105 #ifndef MSGMNI 106 #define MSGMNI 40 107 #endif 108 #ifndef MSGTQL 109 #define MSGTQL 40 110 #endif 111 112 /* 113 * Based on the configuration parameters described in an SVR2 (yes, two) 114 * config(1m) man page. 115 * 116 * Each message is broken up and stored in segments that are msgssz bytes 117 * long. For efficiency reasons, this should be a power of two. Also, 118 * it doesn't make sense if it is less than 8 or greater than about 256. 119 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 120 * two between 8 and 1024 inclusive (and panic's if it isn't). 121 */ 122 struct msginfo msginfo = { 123 MSGMAX, /* max chars in a message */ 124 MSGMNI, /* # of message queue identifiers */ 125 MSGMNB, /* max chars in a queue */ 126 MSGTQL, /* max messages in system */ 127 MSGSSZ, /* size of a message segment */ 128 /* (must be small power of 2 greater than 4) */ 129 MSGSEG /* number of message segments */ 130 }; 131 132 /* 133 * macros to convert between msqid_ds's and msqid's. 134 * (specific to this implementation) 135 */ 136 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 137 #define MSQID_IX(id) ((id) & 0xffff) 138 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 139 140 /* 141 * The rest of this file is specific to this particular implementation. 142 */ 143 144 struct msgmap { 145 short next; /* next segment in buffer */ 146 /* -1 -> available */ 147 /* 0..(MSGSEG-1) -> index of next segment */ 148 }; 149 150 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 151 152 static int nfree_msgmaps; /* # of free map entries */ 153 static short free_msgmaps; /* head of linked list of free map entries */ 154 static struct msg *free_msghdrs;/* list of free msg headers */ 155 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 156 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 157 static struct msg *msghdrs; /* MSGTQL msg headers */ 158 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 159 static struct mtx msq_mtx; /* global mutex for message queues. */ 160 161 static void 162 msginit() 163 { 164 register int i; 165 166 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 167 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 168 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 169 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 170 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 171 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 172 173 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 174 if (msgpool == NULL) 175 panic("msgpool is NULL"); 176 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 177 if (msgmaps == NULL) 178 panic("msgmaps is NULL"); 179 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 180 if (msghdrs == NULL) 181 panic("msghdrs is NULL"); 182 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 183 M_WAITOK); 184 if (msqids == NULL) 185 panic("msqids is NULL"); 186 187 /* 188 * msginfo.msgssz should be a power of two for efficiency reasons. 189 * It is also pretty silly if msginfo.msgssz is less than 8 190 * or greater than about 256 so ... 191 */ 192 193 i = 8; 194 while (i < 1024 && i != msginfo.msgssz) 195 i <<= 1; 196 if (i != msginfo.msgssz) { 197 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 198 msginfo.msgssz)); 199 panic("msginfo.msgssz not a small power of 2"); 200 } 201 202 if (msginfo.msgseg > 32767) { 203 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 204 panic("msginfo.msgseg > 32767"); 205 } 206 207 if (msgmaps == NULL) 208 panic("msgmaps is NULL"); 209 210 for (i = 0; i < msginfo.msgseg; i++) { 211 if (i > 0) 212 msgmaps[i-1].next = i; 213 msgmaps[i].next = -1; /* implies entry is available */ 214 } 215 free_msgmaps = 0; 216 nfree_msgmaps = msginfo.msgseg; 217 218 if (msghdrs == NULL) 219 panic("msghdrs is NULL"); 220 221 for (i = 0; i < msginfo.msgtql; i++) { 222 msghdrs[i].msg_type = 0; 223 if (i > 0) 224 msghdrs[i-1].msg_next = &msghdrs[i]; 225 msghdrs[i].msg_next = NULL; 226 #ifdef MAC 227 mac_init_sysv_msgmsg(&msghdrs[i]); 228 #endif 229 } 230 free_msghdrs = &msghdrs[0]; 231 232 if (msqids == NULL) 233 panic("msqids is NULL"); 234 235 for (i = 0; i < msginfo.msgmni; i++) { 236 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 237 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 238 msqids[i].u.msg_perm.mode = 0; 239 #ifdef MAC 240 mac_init_sysv_msgqueue(&msqids[i]); 241 #endif 242 } 243 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 244 } 245 246 static int 247 msgunload() 248 { 249 struct msqid_kernel *msqkptr; 250 int msqid; 251 #ifdef MAC 252 int i; 253 #endif 254 255 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 256 /* 257 * Look for an unallocated and unlocked msqid_ds. 258 * msqid_ds's can be locked by msgsnd or msgrcv while 259 * they are copying the message in/out. We can't 260 * re-use the entry until they release it. 261 */ 262 msqkptr = &msqids[msqid]; 263 if (msqkptr->u.msg_qbytes != 0 || 264 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 265 break; 266 } 267 if (msqid != msginfo.msgmni) 268 return (EBUSY); 269 270 #ifdef MAC 271 for (i = 0; i < msginfo.msgtql; i++) 272 mac_destroy_sysv_msgmsg(&msghdrs[i]); 273 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 274 mac_destroy_sysv_msgqueue(&msqids[msqid]); 275 #endif 276 free(msgpool, M_MSG); 277 free(msgmaps, M_MSG); 278 free(msghdrs, M_MSG); 279 free(msqids, M_MSG); 280 mtx_destroy(&msq_mtx); 281 return (0); 282 } 283 284 285 static int 286 sysvmsg_modload(struct module *module, int cmd, void *arg) 287 { 288 int error = 0; 289 290 switch (cmd) { 291 case MOD_LOAD: 292 msginit(); 293 break; 294 case MOD_UNLOAD: 295 error = msgunload(); 296 break; 297 case MOD_SHUTDOWN: 298 break; 299 default: 300 error = EINVAL; 301 break; 302 } 303 return (error); 304 } 305 306 static moduledata_t sysvmsg_mod = { 307 "sysvmsg", 308 &sysvmsg_modload, 309 NULL 310 }; 311 312 SYSCALL_MODULE_HELPER(msgsys); 313 SYSCALL_MODULE_HELPER(msgctl); 314 SYSCALL_MODULE_HELPER(msgget); 315 SYSCALL_MODULE_HELPER(msgsnd); 316 SYSCALL_MODULE_HELPER(msgrcv); 317 318 DECLARE_MODULE(sysvmsg, sysvmsg_mod, 319 SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 320 MODULE_VERSION(sysvmsg, 1); 321 322 /* 323 * Entry point for all MSG calls. 324 */ 325 int 326 msgsys(td, uap) 327 struct thread *td; 328 /* XXX actually varargs. */ 329 struct msgsys_args /* { 330 int which; 331 int a2; 332 int a3; 333 int a4; 334 int a5; 335 int a6; 336 } */ *uap; 337 { 338 int error; 339 340 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 341 return (ENOSYS); 342 if (uap->which < 0 || 343 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 344 return (EINVAL); 345 error = (*msgcalls[uap->which])(td, &uap->a2); 346 return (error); 347 } 348 349 static void 350 msg_freehdr(msghdr) 351 struct msg *msghdr; 352 { 353 while (msghdr->msg_ts > 0) { 354 short next; 355 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 356 panic("msghdr->msg_spot out of range"); 357 next = msgmaps[msghdr->msg_spot].next; 358 msgmaps[msghdr->msg_spot].next = free_msgmaps; 359 free_msgmaps = msghdr->msg_spot; 360 nfree_msgmaps++; 361 msghdr->msg_spot = next; 362 if (msghdr->msg_ts >= msginfo.msgssz) 363 msghdr->msg_ts -= msginfo.msgssz; 364 else 365 msghdr->msg_ts = 0; 366 } 367 if (msghdr->msg_spot != -1) 368 panic("msghdr->msg_spot != -1"); 369 msghdr->msg_next = free_msghdrs; 370 free_msghdrs = msghdr; 371 #ifdef MAC 372 mac_cleanup_sysv_msgmsg(msghdr); 373 #endif 374 } 375 376 #ifndef _SYS_SYSPROTO_H_ 377 struct msgctl_args { 378 int msqid; 379 int cmd; 380 struct msqid_ds *buf; 381 }; 382 #endif 383 int 384 msgctl(td, uap) 385 struct thread *td; 386 register struct msgctl_args *uap; 387 { 388 int msqid = uap->msqid; 389 int cmd = uap->cmd; 390 struct msqid_ds msqbuf; 391 int error; 392 393 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 394 if (cmd == IPC_SET && 395 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 396 return (error); 397 error = kern_msgctl(td, msqid, cmd, &msqbuf); 398 if (cmd == IPC_STAT && error == 0) 399 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 400 return (error); 401 } 402 403 int 404 kern_msgctl(td, msqid, cmd, msqbuf) 405 struct thread *td; 406 int msqid; 407 int cmd; 408 struct msqid_ds *msqbuf; 409 { 410 int rval, error, msqix; 411 register struct msqid_kernel *msqkptr; 412 413 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 414 return (ENOSYS); 415 416 msqix = IPCID_TO_IX(msqid); 417 418 if (msqix < 0 || msqix >= msginfo.msgmni) { 419 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 420 msginfo.msgmni)); 421 return (EINVAL); 422 } 423 424 msqkptr = &msqids[msqix]; 425 426 mtx_lock(&msq_mtx); 427 if (msqkptr->u.msg_qbytes == 0) { 428 DPRINTF(("no such msqid\n")); 429 error = EINVAL; 430 goto done2; 431 } 432 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 433 DPRINTF(("wrong sequence number\n")); 434 error = EINVAL; 435 goto done2; 436 } 437 #ifdef MAC 438 error = mac_check_sysv_msqctl(td->td_ucred, msqkptr, cmd); 439 if (error != 0) 440 goto done2; 441 #endif 442 443 error = 0; 444 rval = 0; 445 446 switch (cmd) { 447 448 case IPC_RMID: 449 { 450 struct msg *msghdr; 451 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 452 goto done2; 453 454 #ifdef MAC 455 /* 456 * Check that the thread has MAC access permissions to 457 * individual msghdrs. Note: We need to do this in a 458 * separate loop because the actual loop alters the 459 * msq/msghdr info as it progresses, and there is no going 460 * back if half the way through we discover that the 461 * thread cannot free a certain msghdr. The msq will get 462 * into an inconsistent state. 463 */ 464 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 465 msghdr = msghdr->msg_next) { 466 error = mac_check_sysv_msgrmid(td->td_ucred, msghdr); 467 if (error != 0) 468 goto done2; 469 } 470 #endif 471 472 /* Free the message headers */ 473 msghdr = msqkptr->u.msg_first; 474 while (msghdr != NULL) { 475 struct msg *msghdr_tmp; 476 477 /* Free the segments of each message */ 478 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 479 msqkptr->u.msg_qnum--; 480 msghdr_tmp = msghdr; 481 msghdr = msghdr->msg_next; 482 msg_freehdr(msghdr_tmp); 483 } 484 485 if (msqkptr->u.msg_cbytes != 0) 486 panic("msg_cbytes is screwed up"); 487 if (msqkptr->u.msg_qnum != 0) 488 panic("msg_qnum is screwed up"); 489 490 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 491 492 #ifdef MAC 493 mac_cleanup_sysv_msgqueue(msqkptr); 494 #endif 495 496 wakeup(msqkptr); 497 } 498 499 break; 500 501 case IPC_SET: 502 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 503 goto done2; 504 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 505 error = priv_check(td, PRIV_IPC_MSGSIZE); 506 if (error) 507 goto done2; 508 } 509 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 510 DPRINTF(("can't increase msg_qbytes beyond %d" 511 "(truncating)\n", msginfo.msgmnb)); 512 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 513 } 514 if (msqbuf->msg_qbytes == 0) { 515 DPRINTF(("can't reduce msg_qbytes to 0\n")); 516 error = EINVAL; /* non-standard errno! */ 517 goto done2; 518 } 519 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 520 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 521 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 522 (msqbuf->msg_perm.mode & 0777); 523 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 524 msqkptr->u.msg_ctime = time_second; 525 break; 526 527 case IPC_STAT: 528 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 529 DPRINTF(("requester doesn't have read access\n")); 530 goto done2; 531 } 532 *msqbuf = msqkptr->u; 533 break; 534 535 default: 536 DPRINTF(("invalid command %d\n", cmd)); 537 error = EINVAL; 538 goto done2; 539 } 540 541 if (error == 0) 542 td->td_retval[0] = rval; 543 done2: 544 mtx_unlock(&msq_mtx); 545 return (error); 546 } 547 548 #ifndef _SYS_SYSPROTO_H_ 549 struct msgget_args { 550 key_t key; 551 int msgflg; 552 }; 553 #endif 554 int 555 msgget(td, uap) 556 struct thread *td; 557 register struct msgget_args *uap; 558 { 559 int msqid, error = 0; 560 int key = uap->key; 561 int msgflg = uap->msgflg; 562 struct ucred *cred = td->td_ucred; 563 register struct msqid_kernel *msqkptr = NULL; 564 565 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 566 567 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 568 return (ENOSYS); 569 570 mtx_lock(&msq_mtx); 571 if (key != IPC_PRIVATE) { 572 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 573 msqkptr = &msqids[msqid]; 574 if (msqkptr->u.msg_qbytes != 0 && 575 msqkptr->u.msg_perm.key == key) 576 break; 577 } 578 if (msqid < msginfo.msgmni) { 579 DPRINTF(("found public key\n")); 580 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 581 DPRINTF(("not exclusive\n")); 582 error = EEXIST; 583 goto done2; 584 } 585 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 586 msgflg & 0700))) { 587 DPRINTF(("requester doesn't have 0%o access\n", 588 msgflg & 0700)); 589 goto done2; 590 } 591 #ifdef MAC 592 error = mac_check_sysv_msqget(cred, msqkptr); 593 if (error != 0) 594 goto done2; 595 #endif 596 goto found; 597 } 598 } 599 600 DPRINTF(("need to allocate the msqid_ds\n")); 601 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 602 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 603 /* 604 * Look for an unallocated and unlocked msqid_ds. 605 * msqid_ds's can be locked by msgsnd or msgrcv while 606 * they are copying the message in/out. We can't 607 * re-use the entry until they release it. 608 */ 609 msqkptr = &msqids[msqid]; 610 if (msqkptr->u.msg_qbytes == 0 && 611 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 612 break; 613 } 614 if (msqid == msginfo.msgmni) { 615 DPRINTF(("no more msqid_ds's available\n")); 616 error = ENOSPC; 617 goto done2; 618 } 619 DPRINTF(("msqid %d is available\n", msqid)); 620 msqkptr->u.msg_perm.key = key; 621 msqkptr->u.msg_perm.cuid = cred->cr_uid; 622 msqkptr->u.msg_perm.uid = cred->cr_uid; 623 msqkptr->u.msg_perm.cgid = cred->cr_gid; 624 msqkptr->u.msg_perm.gid = cred->cr_gid; 625 msqkptr->u.msg_perm.mode = (msgflg & 0777); 626 /* Make sure that the returned msqid is unique */ 627 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 628 msqkptr->u.msg_first = NULL; 629 msqkptr->u.msg_last = NULL; 630 msqkptr->u.msg_cbytes = 0; 631 msqkptr->u.msg_qnum = 0; 632 msqkptr->u.msg_qbytes = msginfo.msgmnb; 633 msqkptr->u.msg_lspid = 0; 634 msqkptr->u.msg_lrpid = 0; 635 msqkptr->u.msg_stime = 0; 636 msqkptr->u.msg_rtime = 0; 637 msqkptr->u.msg_ctime = time_second; 638 #ifdef MAC 639 mac_create_sysv_msgqueue(cred, msqkptr); 640 #endif 641 } else { 642 DPRINTF(("didn't find it and wasn't asked to create it\n")); 643 error = ENOENT; 644 goto done2; 645 } 646 647 found: 648 /* Construct the unique msqid */ 649 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 650 done2: 651 mtx_unlock(&msq_mtx); 652 return (error); 653 } 654 655 #ifndef _SYS_SYSPROTO_H_ 656 struct msgsnd_args { 657 int msqid; 658 const void *msgp; 659 size_t msgsz; 660 int msgflg; 661 }; 662 #endif 663 int 664 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 665 struct thread *td; 666 int msqid; 667 const void *msgp; /* XXX msgp is actually mtext. */ 668 size_t msgsz; 669 int msgflg; 670 long mtype; 671 { 672 int msqix, segs_needed, error = 0; 673 register struct msqid_kernel *msqkptr; 674 register struct msg *msghdr; 675 short next; 676 677 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 678 return (ENOSYS); 679 680 mtx_lock(&msq_mtx); 681 msqix = IPCID_TO_IX(msqid); 682 683 if (msqix < 0 || msqix >= msginfo.msgmni) { 684 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 685 msginfo.msgmni)); 686 error = EINVAL; 687 goto done2; 688 } 689 690 msqkptr = &msqids[msqix]; 691 if (msqkptr->u.msg_qbytes == 0) { 692 DPRINTF(("no such message queue id\n")); 693 error = EINVAL; 694 goto done2; 695 } 696 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 697 DPRINTF(("wrong sequence number\n")); 698 error = EINVAL; 699 goto done2; 700 } 701 702 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 703 DPRINTF(("requester doesn't have write access\n")); 704 goto done2; 705 } 706 707 #ifdef MAC 708 error = mac_check_sysv_msqsnd(td->td_ucred, msqkptr); 709 if (error != 0) 710 goto done2; 711 #endif 712 713 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 714 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 715 msginfo.msgssz, segs_needed)); 716 for (;;) { 717 int need_more_resources = 0; 718 719 /* 720 * check msgsz 721 * (inside this loop in case msg_qbytes changes while we sleep) 722 */ 723 724 if (msgsz > msqkptr->u.msg_qbytes) { 725 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 726 error = EINVAL; 727 goto done2; 728 } 729 730 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 731 DPRINTF(("msqid is locked\n")); 732 need_more_resources = 1; 733 } 734 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 735 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 736 need_more_resources = 1; 737 } 738 if (segs_needed > nfree_msgmaps) { 739 DPRINTF(("segs_needed > nfree_msgmaps\n")); 740 need_more_resources = 1; 741 } 742 if (free_msghdrs == NULL) { 743 DPRINTF(("no more msghdrs\n")); 744 need_more_resources = 1; 745 } 746 747 if (need_more_resources) { 748 int we_own_it; 749 750 if ((msgflg & IPC_NOWAIT) != 0) { 751 DPRINTF(("need more resources but caller " 752 "doesn't want to wait\n")); 753 error = EAGAIN; 754 goto done2; 755 } 756 757 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 758 DPRINTF(("we don't own the msqid_ds\n")); 759 we_own_it = 0; 760 } else { 761 /* Force later arrivals to wait for our 762 request */ 763 DPRINTF(("we own the msqid_ds\n")); 764 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 765 we_own_it = 1; 766 } 767 DPRINTF(("msgsnd: goodnight\n")); 768 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 769 "msgsnd", hz); 770 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 771 if (we_own_it) 772 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 773 if (error == EWOULDBLOCK) { 774 DPRINTF(("msgsnd: timed out\n")); 775 continue; 776 } 777 if (error != 0) { 778 DPRINTF(("msgsnd: interrupted system call\n")); 779 error = EINTR; 780 goto done2; 781 } 782 783 /* 784 * Make sure that the msq queue still exists 785 */ 786 787 if (msqkptr->u.msg_qbytes == 0) { 788 DPRINTF(("msqid deleted\n")); 789 error = EIDRM; 790 goto done2; 791 } 792 793 } else { 794 DPRINTF(("got all the resources that we need\n")); 795 break; 796 } 797 } 798 799 /* 800 * We have the resources that we need. 801 * Make sure! 802 */ 803 804 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 805 panic("msg_perm.mode & MSG_LOCKED"); 806 if (segs_needed > nfree_msgmaps) 807 panic("segs_needed > nfree_msgmaps"); 808 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 809 panic("msgsz + msg_cbytes > msg_qbytes"); 810 if (free_msghdrs == NULL) 811 panic("no more msghdrs"); 812 813 /* 814 * Re-lock the msqid_ds in case we page-fault when copying in the 815 * message 816 */ 817 818 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 819 panic("msqid_ds is already locked"); 820 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 821 822 /* 823 * Allocate a message header 824 */ 825 826 msghdr = free_msghdrs; 827 free_msghdrs = msghdr->msg_next; 828 msghdr->msg_spot = -1; 829 msghdr->msg_ts = msgsz; 830 msghdr->msg_type = mtype; 831 #ifdef MAC 832 /* 833 * XXXMAC: Should the mac_check_sysv_msgmsq check follow here 834 * immediately? Or, should it be checked just before the msg is 835 * enqueued in the msgq (as it is done now)? 836 */ 837 mac_create_sysv_msgmsg(td->td_ucred, msqkptr, msghdr); 838 #endif 839 840 /* 841 * Allocate space for the message 842 */ 843 844 while (segs_needed > 0) { 845 if (nfree_msgmaps <= 0) 846 panic("not enough msgmaps"); 847 if (free_msgmaps == -1) 848 panic("nil free_msgmaps"); 849 next = free_msgmaps; 850 if (next <= -1) 851 panic("next too low #1"); 852 if (next >= msginfo.msgseg) 853 panic("next out of range #1"); 854 DPRINTF(("allocating segment %d to message\n", next)); 855 free_msgmaps = msgmaps[next].next; 856 nfree_msgmaps--; 857 msgmaps[next].next = msghdr->msg_spot; 858 msghdr->msg_spot = next; 859 segs_needed--; 860 } 861 862 /* 863 * Validate the message type 864 */ 865 866 if (msghdr->msg_type < 1) { 867 msg_freehdr(msghdr); 868 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 869 wakeup(msqkptr); 870 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 871 error = EINVAL; 872 goto done2; 873 } 874 875 /* 876 * Copy in the message body 877 */ 878 879 next = msghdr->msg_spot; 880 while (msgsz > 0) { 881 size_t tlen; 882 if (msgsz > msginfo.msgssz) 883 tlen = msginfo.msgssz; 884 else 885 tlen = msgsz; 886 if (next <= -1) 887 panic("next too low #2"); 888 if (next >= msginfo.msgseg) 889 panic("next out of range #2"); 890 mtx_unlock(&msq_mtx); 891 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 892 tlen)) != 0) { 893 mtx_lock(&msq_mtx); 894 DPRINTF(("error %d copying in message segment\n", 895 error)); 896 msg_freehdr(msghdr); 897 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 898 wakeup(msqkptr); 899 goto done2; 900 } 901 mtx_lock(&msq_mtx); 902 msgsz -= tlen; 903 msgp = (const char *)msgp + tlen; 904 next = msgmaps[next].next; 905 } 906 if (next != -1) 907 panic("didn't use all the msg segments"); 908 909 /* 910 * We've got the message. Unlock the msqid_ds. 911 */ 912 913 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 914 915 /* 916 * Make sure that the msqid_ds is still allocated. 917 */ 918 919 if (msqkptr->u.msg_qbytes == 0) { 920 msg_freehdr(msghdr); 921 wakeup(msqkptr); 922 error = EIDRM; 923 goto done2; 924 } 925 926 #ifdef MAC 927 /* 928 * Note: Since the task/thread allocates the msghdr and usually 929 * primes it with its own MAC label, for a majority of policies, it 930 * won't be necessary to check whether the msghdr has access 931 * permissions to the msgq. The mac_check_sysv_msqsnd check would 932 * suffice in that case. However, this hook may be required where 933 * individual policies derive a non-identical label for the msghdr 934 * from the current thread label and may want to check the msghdr 935 * enqueue permissions, along with read/write permissions to the 936 * msgq. 937 */ 938 error = mac_check_sysv_msgmsq(td->td_ucred, msghdr, msqkptr); 939 if (error != 0) { 940 msg_freehdr(msghdr); 941 wakeup(msqkptr); 942 goto done2; 943 } 944 #endif 945 946 /* 947 * Put the message into the queue 948 */ 949 if (msqkptr->u.msg_first == NULL) { 950 msqkptr->u.msg_first = msghdr; 951 msqkptr->u.msg_last = msghdr; 952 } else { 953 msqkptr->u.msg_last->msg_next = msghdr; 954 msqkptr->u.msg_last = msghdr; 955 } 956 msqkptr->u.msg_last->msg_next = NULL; 957 958 msqkptr->u.msg_cbytes += msghdr->msg_ts; 959 msqkptr->u.msg_qnum++; 960 msqkptr->u.msg_lspid = td->td_proc->p_pid; 961 msqkptr->u.msg_stime = time_second; 962 963 wakeup(msqkptr); 964 td->td_retval[0] = 0; 965 done2: 966 mtx_unlock(&msq_mtx); 967 return (error); 968 } 969 970 int 971 msgsnd(td, uap) 972 struct thread *td; 973 register struct msgsnd_args *uap; 974 { 975 int error; 976 long mtype; 977 978 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 979 uap->msgsz, uap->msgflg)); 980 981 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 982 DPRINTF(("error %d copying the message type\n", error)); 983 return (error); 984 } 985 return (kern_msgsnd(td, uap->msqid, 986 (const char *)uap->msgp + sizeof(mtype), 987 uap->msgsz, uap->msgflg, mtype)); 988 } 989 990 #ifndef _SYS_SYSPROTO_H_ 991 struct msgrcv_args { 992 int msqid; 993 void *msgp; 994 size_t msgsz; 995 long msgtyp; 996 int msgflg; 997 }; 998 #endif 999 int 1000 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1001 struct thread *td; 1002 int msqid; 1003 void *msgp; /* XXX msgp is actually mtext. */ 1004 size_t msgsz; 1005 long msgtyp; 1006 int msgflg; 1007 long *mtype; 1008 { 1009 size_t len; 1010 register struct msqid_kernel *msqkptr; 1011 register struct msg *msghdr; 1012 int msqix, error = 0; 1013 short next; 1014 1015 if (!jail_sysvipc_allowed && jailed(td->td_ucred)) 1016 return (ENOSYS); 1017 1018 msqix = IPCID_TO_IX(msqid); 1019 1020 if (msqix < 0 || msqix >= msginfo.msgmni) { 1021 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1022 msginfo.msgmni)); 1023 return (EINVAL); 1024 } 1025 1026 msqkptr = &msqids[msqix]; 1027 mtx_lock(&msq_mtx); 1028 if (msqkptr->u.msg_qbytes == 0) { 1029 DPRINTF(("no such message queue id\n")); 1030 error = EINVAL; 1031 goto done2; 1032 } 1033 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1034 DPRINTF(("wrong sequence number\n")); 1035 error = EINVAL; 1036 goto done2; 1037 } 1038 1039 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1040 DPRINTF(("requester doesn't have read access\n")); 1041 goto done2; 1042 } 1043 1044 #ifdef MAC 1045 error = mac_check_sysv_msqrcv(td->td_ucred, msqkptr); 1046 if (error != 0) 1047 goto done2; 1048 #endif 1049 1050 msghdr = NULL; 1051 while (msghdr == NULL) { 1052 if (msgtyp == 0) { 1053 msghdr = msqkptr->u.msg_first; 1054 if (msghdr != NULL) { 1055 if (msgsz < msghdr->msg_ts && 1056 (msgflg & MSG_NOERROR) == 0) { 1057 DPRINTF(("first message on the queue " 1058 "is too big (want %zu, got %d)\n", 1059 msgsz, msghdr->msg_ts)); 1060 error = E2BIG; 1061 goto done2; 1062 } 1063 #ifdef MAC 1064 error = mac_check_sysv_msgrcv(td->td_ucred, 1065 msghdr); 1066 if (error != 0) 1067 goto done2; 1068 #endif 1069 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1070 msqkptr->u.msg_first = NULL; 1071 msqkptr->u.msg_last = NULL; 1072 } else { 1073 msqkptr->u.msg_first = msghdr->msg_next; 1074 if (msqkptr->u.msg_first == NULL) 1075 panic("msg_first/last screwed up #1"); 1076 } 1077 } 1078 } else { 1079 struct msg *previous; 1080 struct msg **prev; 1081 1082 previous = NULL; 1083 prev = &(msqkptr->u.msg_first); 1084 while ((msghdr = *prev) != NULL) { 1085 /* 1086 * Is this message's type an exact match or is 1087 * this message's type less than or equal to 1088 * the absolute value of a negative msgtyp? 1089 * Note that the second half of this test can 1090 * NEVER be true if msgtyp is positive since 1091 * msg_type is always positive! 1092 */ 1093 1094 if (msgtyp == msghdr->msg_type || 1095 msghdr->msg_type <= -msgtyp) { 1096 DPRINTF(("found message type %ld, " 1097 "requested %ld\n", 1098 msghdr->msg_type, msgtyp)); 1099 if (msgsz < msghdr->msg_ts && 1100 (msgflg & MSG_NOERROR) == 0) { 1101 DPRINTF(("requested message " 1102 "on the queue is too big " 1103 "(want %zu, got %hu)\n", 1104 msgsz, msghdr->msg_ts)); 1105 error = E2BIG; 1106 goto done2; 1107 } 1108 #ifdef MAC 1109 error = mac_check_sysv_msgrcv( 1110 td->td_ucred, msghdr); 1111 if (error != 0) 1112 goto done2; 1113 #endif 1114 *prev = msghdr->msg_next; 1115 if (msghdr == msqkptr->u.msg_last) { 1116 if (previous == NULL) { 1117 if (prev != 1118 &msqkptr->u.msg_first) 1119 panic("msg_first/last screwed up #2"); 1120 msqkptr->u.msg_first = 1121 NULL; 1122 msqkptr->u.msg_last = 1123 NULL; 1124 } else { 1125 if (prev == 1126 &msqkptr->u.msg_first) 1127 panic("msg_first/last screwed up #3"); 1128 msqkptr->u.msg_last = 1129 previous; 1130 } 1131 } 1132 break; 1133 } 1134 previous = msghdr; 1135 prev = &(msghdr->msg_next); 1136 } 1137 } 1138 1139 /* 1140 * We've either extracted the msghdr for the appropriate 1141 * message or there isn't one. 1142 * If there is one then bail out of this loop. 1143 */ 1144 1145 if (msghdr != NULL) 1146 break; 1147 1148 /* 1149 * Hmph! No message found. Does the user want to wait? 1150 */ 1151 1152 if ((msgflg & IPC_NOWAIT) != 0) { 1153 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1154 msgtyp)); 1155 /* The SVID says to return ENOMSG. */ 1156 error = ENOMSG; 1157 goto done2; 1158 } 1159 1160 /* 1161 * Wait for something to happen 1162 */ 1163 1164 DPRINTF(("msgrcv: goodnight\n")); 1165 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1166 "msgrcv", 0); 1167 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1168 1169 if (error != 0) { 1170 DPRINTF(("msgrcv: interrupted system call\n")); 1171 error = EINTR; 1172 goto done2; 1173 } 1174 1175 /* 1176 * Make sure that the msq queue still exists 1177 */ 1178 1179 if (msqkptr->u.msg_qbytes == 0 || 1180 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1181 DPRINTF(("msqid deleted\n")); 1182 error = EIDRM; 1183 goto done2; 1184 } 1185 } 1186 1187 /* 1188 * Return the message to the user. 1189 * 1190 * First, do the bookkeeping (before we risk being interrupted). 1191 */ 1192 1193 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1194 msqkptr->u.msg_qnum--; 1195 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1196 msqkptr->u.msg_rtime = time_second; 1197 1198 /* 1199 * Make msgsz the actual amount that we'll be returning. 1200 * Note that this effectively truncates the message if it is too long 1201 * (since msgsz is never increased). 1202 */ 1203 1204 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1205 msghdr->msg_ts)); 1206 if (msgsz > msghdr->msg_ts) 1207 msgsz = msghdr->msg_ts; 1208 *mtype = msghdr->msg_type; 1209 1210 /* 1211 * Return the segments to the user 1212 */ 1213 1214 next = msghdr->msg_spot; 1215 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1216 size_t tlen; 1217 1218 if (msgsz - len > msginfo.msgssz) 1219 tlen = msginfo.msgssz; 1220 else 1221 tlen = msgsz - len; 1222 if (next <= -1) 1223 panic("next too low #3"); 1224 if (next >= msginfo.msgseg) 1225 panic("next out of range #3"); 1226 mtx_unlock(&msq_mtx); 1227 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1228 mtx_lock(&msq_mtx); 1229 if (error != 0) { 1230 DPRINTF(("error (%d) copying out message segment\n", 1231 error)); 1232 msg_freehdr(msghdr); 1233 wakeup(msqkptr); 1234 goto done2; 1235 } 1236 msgp = (char *)msgp + tlen; 1237 next = msgmaps[next].next; 1238 } 1239 1240 /* 1241 * Done, return the actual number of bytes copied out. 1242 */ 1243 1244 msg_freehdr(msghdr); 1245 wakeup(msqkptr); 1246 td->td_retval[0] = msgsz; 1247 done2: 1248 mtx_unlock(&msq_mtx); 1249 return (error); 1250 } 1251 1252 int 1253 msgrcv(td, uap) 1254 struct thread *td; 1255 register struct msgrcv_args *uap; 1256 { 1257 int error; 1258 long mtype; 1259 1260 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1261 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1262 1263 if ((error = kern_msgrcv(td, uap->msqid, 1264 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1265 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1266 return (error); 1267 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1268 DPRINTF(("error %d copying the message type\n", error)); 1269 return (error); 1270 } 1271 1272 static int 1273 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1274 { 1275 1276 return (SYSCTL_OUT(req, msqids, 1277 sizeof(struct msqid_kernel) * msginfo.msgmni)); 1278 } 1279 1280 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1281 "Maximum message size"); 1282 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1283 "Number of message queue identifiers"); 1284 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1285 "Maximum number of bytes in a queue"); 1286 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1287 "Maximum number of messages in the system"); 1288 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1289 "Size of a message segment"); 1290 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1291 "Number of message segments"); 1292 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, 1293 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1294