sysv_msg.c revision 298835
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: stable/10/sys/kern/sysv_msg.c 298835 2016-04-30 04:02:32Z jamie $"); 52 53#include "opt_compat.h" 54#include "opt_sysvipc.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/mount.h> 66#include <sys/msg.h> 67#include <sys/racct.h> 68#include <sys/sx.h> 69#include <sys/syscall.h> 70#include <sys/syscallsubr.h> 71#include <sys/sysent.h> 72#include <sys/sysctl.h> 73#include <sys/malloc.h> 74#include <sys/jail.h> 75 76#include <security/mac/mac_framework.h> 77 78FEATURE(sysv_msg, "System V message queues support"); 79 80static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 81 82static int msginit(void); 83static int msgunload(void); 84static int sysvmsg_modload(struct module *, int, void *); 85static void msq_remove(struct msqid_kernel *); 86static struct prison *msg_find_prison(struct ucred *); 87static int msq_prison_cansee(struct prison *, struct msqid_kernel *); 88static int msg_prison_check(void *, void *); 89static int msg_prison_set(void *, void *); 90static int msg_prison_get(void *, void *); 91static int msg_prison_remove(void *, void *); 92static void msg_prison_cleanup(struct prison *); 93 94 95#ifdef MSG_DEBUG 96#define DPRINTF(a) printf a 97#else 98#define DPRINTF(a) (void)0 99#endif 100 101static void msg_freehdr(struct msg *msghdr); 102 103#ifndef MSGSSZ 104#define MSGSSZ 8 /* Each segment must be 2^N long */ 105#endif 106#ifndef MSGSEG 107#define MSGSEG 2048 /* must be less than 32767 */ 108#endif 109#define MSGMAX (MSGSSZ*MSGSEG) 110#ifndef MSGMNB 111#define MSGMNB 2048 /* max # of bytes in a queue */ 112#endif 113#ifndef MSGMNI 114#define MSGMNI 40 115#endif 116#ifndef MSGTQL 117#define MSGTQL 40 118#endif 119 120/* 121 * Based on the configuration parameters described in an SVR2 (yes, two) 122 * config(1m) man page. 123 * 124 * Each message is broken up and stored in segments that are msgssz bytes 125 * long. For efficiency reasons, this should be a power of two. Also, 126 * it doesn't make sense if it is less than 8 or greater than about 256. 127 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 128 * two between 8 and 1024 inclusive (and panic's if it isn't). 129 */ 130struct msginfo msginfo = { 131 MSGMAX, /* max chars in a message */ 132 MSGMNI, /* # of message queue identifiers */ 133 MSGMNB, /* max chars in a queue */ 134 MSGTQL, /* max messages in system */ 135 MSGSSZ, /* size of a message segment */ 136 /* (must be small power of 2 greater than 4) */ 137 MSGSEG /* number of message segments */ 138}; 139 140/* 141 * macros to convert between msqid_ds's and msqid's. 142 * (specific to this implementation) 143 */ 144#define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 145#define MSQID_IX(id) ((id) & 0xffff) 146#define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 147 148/* 149 * The rest of this file is specific to this particular implementation. 150 */ 151 152struct msgmap { 153 short next; /* next segment in buffer */ 154 /* -1 -> available */ 155 /* 0..(MSGSEG-1) -> index of next segment */ 156}; 157 158#define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 159 160static int nfree_msgmaps; /* # of free map entries */ 161static short free_msgmaps; /* head of linked list of free map entries */ 162static struct msg *free_msghdrs;/* list of free msg headers */ 163static char *msgpool; /* MSGMAX byte long msg buffer pool */ 164static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 165static struct msg *msghdrs; /* MSGTQL msg headers */ 166static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 167static struct mtx msq_mtx; /* global mutex for message queues. */ 168static unsigned msg_prison_slot;/* prison OSD slot */ 169 170static struct syscall_helper_data msg_syscalls[] = { 171 SYSCALL_INIT_HELPER(msgctl), 172 SYSCALL_INIT_HELPER(msgget), 173 SYSCALL_INIT_HELPER(msgsnd), 174 SYSCALL_INIT_HELPER(msgrcv), 175#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 176 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 177 SYSCALL_INIT_HELPER(msgsys), 178 SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl), 179#endif 180 SYSCALL_INIT_LAST 181}; 182 183#ifdef COMPAT_FREEBSD32 184#include <compat/freebsd32/freebsd32.h> 185#include <compat/freebsd32/freebsd32_ipc.h> 186#include <compat/freebsd32/freebsd32_proto.h> 187#include <compat/freebsd32/freebsd32_signal.h> 188#include <compat/freebsd32/freebsd32_syscall.h> 189#include <compat/freebsd32/freebsd32_util.h> 190 191static struct syscall_helper_data msg32_syscalls[] = { 192 SYSCALL32_INIT_HELPER(freebsd32_msgctl), 193 SYSCALL32_INIT_HELPER(freebsd32_msgsnd), 194 SYSCALL32_INIT_HELPER(freebsd32_msgrcv), 195 SYSCALL32_INIT_HELPER_COMPAT(msgget), 196 SYSCALL32_INIT_HELPER(freebsd32_msgsys), 197#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 198 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 199 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl), 200#endif 201 SYSCALL_INIT_LAST 202}; 203#endif 204 205static int 206msginit() 207{ 208 struct prison *pr; 209 void *rsv; 210 int i, error; 211 osd_method_t methods[PR_MAXMETHOD] = { 212 [PR_METHOD_CHECK] = msg_prison_check, 213 [PR_METHOD_SET] = msg_prison_set, 214 [PR_METHOD_GET] = msg_prison_get, 215 [PR_METHOD_REMOVE] = msg_prison_remove, 216 }; 217 218 TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); 219 TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); 220 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 221 TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); 222 TUNABLE_INT_FETCH("kern.ipc.msgmnb", &msginfo.msgmnb); 223 TUNABLE_INT_FETCH("kern.ipc.msgtql", &msginfo.msgtql); 224 225 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 226 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 227 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 228 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 229 M_WAITOK); 230 231 /* 232 * msginfo.msgssz should be a power of two for efficiency reasons. 233 * It is also pretty silly if msginfo.msgssz is less than 8 234 * or greater than about 256 so ... 235 */ 236 237 i = 8; 238 while (i < 1024 && i != msginfo.msgssz) 239 i <<= 1; 240 if (i != msginfo.msgssz) { 241 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 242 msginfo.msgssz)); 243 panic("msginfo.msgssz not a small power of 2"); 244 } 245 246 if (msginfo.msgseg > 32767) { 247 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 248 panic("msginfo.msgseg > 32767"); 249 } 250 251 for (i = 0; i < msginfo.msgseg; i++) { 252 if (i > 0) 253 msgmaps[i-1].next = i; 254 msgmaps[i].next = -1; /* implies entry is available */ 255 } 256 free_msgmaps = 0; 257 nfree_msgmaps = msginfo.msgseg; 258 259 for (i = 0; i < msginfo.msgtql; i++) { 260 msghdrs[i].msg_type = 0; 261 if (i > 0) 262 msghdrs[i-1].msg_next = &msghdrs[i]; 263 msghdrs[i].msg_next = NULL; 264#ifdef MAC 265 mac_sysvmsg_init(&msghdrs[i]); 266#endif 267 } 268 free_msghdrs = &msghdrs[0]; 269 270 for (i = 0; i < msginfo.msgmni; i++) { 271 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 272 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 273 msqids[i].u.msg_perm.mode = 0; 274#ifdef MAC 275 mac_sysvmsq_init(&msqids[i]); 276#endif 277 } 278 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 279 280 /* Set current prisons according to their allow.sysvipc. */ 281 msg_prison_slot = osd_jail_register(NULL, methods); 282 rsv = osd_reserve(msg_prison_slot); 283 prison_lock(&prison0); 284 (void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0); 285 prison_unlock(&prison0); 286 rsv = NULL; 287 sx_slock(&allprison_lock); 288 TAILQ_FOREACH(pr, &allprison, pr_list) { 289 if (rsv == NULL) 290 rsv = osd_reserve(msg_prison_slot); 291 prison_lock(pr); 292 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) { 293 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 294 &prison0); 295 rsv = NULL; 296 } 297 prison_unlock(pr); 298 } 299 if (rsv != NULL) 300 osd_free_reserved(rsv); 301 sx_sunlock(&allprison_lock); 302 303 error = syscall_helper_register(msg_syscalls); 304 if (error != 0) 305 return (error); 306#ifdef COMPAT_FREEBSD32 307 error = syscall32_helper_register(msg32_syscalls); 308 if (error != 0) 309 return (error); 310#endif 311 return (0); 312} 313 314static int 315msgunload() 316{ 317 struct msqid_kernel *msqkptr; 318 int msqid; 319#ifdef MAC 320 int i; 321#endif 322 323 syscall_helper_unregister(msg_syscalls); 324#ifdef COMPAT_FREEBSD32 325 syscall32_helper_unregister(msg32_syscalls); 326#endif 327 328 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 329 /* 330 * Look for an unallocated and unlocked msqid_ds. 331 * msqid_ds's can be locked by msgsnd or msgrcv while 332 * they are copying the message in/out. We can't 333 * re-use the entry until they release it. 334 */ 335 msqkptr = &msqids[msqid]; 336 if (msqkptr->u.msg_qbytes != 0 || 337 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 338 break; 339 } 340 if (msqid != msginfo.msgmni) 341 return (EBUSY); 342 343 if (msg_prison_slot != 0) 344 osd_jail_deregister(msg_prison_slot); 345#ifdef MAC 346 for (i = 0; i < msginfo.msgtql; i++) 347 mac_sysvmsg_destroy(&msghdrs[i]); 348 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 349 mac_sysvmsq_destroy(&msqids[msqid]); 350#endif 351 free(msgpool, M_MSG); 352 free(msgmaps, M_MSG); 353 free(msghdrs, M_MSG); 354 free(msqids, M_MSG); 355 mtx_destroy(&msq_mtx); 356 return (0); 357} 358 359 360static int 361sysvmsg_modload(struct module *module, int cmd, void *arg) 362{ 363 int error = 0; 364 365 switch (cmd) { 366 case MOD_LOAD: 367 error = msginit(); 368 if (error != 0) 369 msgunload(); 370 break; 371 case MOD_UNLOAD: 372 error = msgunload(); 373 break; 374 case MOD_SHUTDOWN: 375 break; 376 default: 377 error = EINVAL; 378 break; 379 } 380 return (error); 381} 382 383static moduledata_t sysvmsg_mod = { 384 "sysvmsg", 385 &sysvmsg_modload, 386 NULL 387}; 388 389DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 390MODULE_VERSION(sysvmsg, 1); 391 392static void 393msg_freehdr(msghdr) 394 struct msg *msghdr; 395{ 396 while (msghdr->msg_ts > 0) { 397 short next; 398 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 399 panic("msghdr->msg_spot out of range"); 400 next = msgmaps[msghdr->msg_spot].next; 401 msgmaps[msghdr->msg_spot].next = free_msgmaps; 402 free_msgmaps = msghdr->msg_spot; 403 nfree_msgmaps++; 404 msghdr->msg_spot = next; 405 if (msghdr->msg_ts >= msginfo.msgssz) 406 msghdr->msg_ts -= msginfo.msgssz; 407 else 408 msghdr->msg_ts = 0; 409 } 410 if (msghdr->msg_spot != -1) 411 panic("msghdr->msg_spot != -1"); 412 msghdr->msg_next = free_msghdrs; 413 free_msghdrs = msghdr; 414#ifdef MAC 415 mac_sysvmsg_cleanup(msghdr); 416#endif 417} 418 419static void 420msq_remove(struct msqid_kernel *msqkptr) 421{ 422 struct msg *msghdr; 423 424 racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1); 425 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum); 426 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes); 427 crfree(msqkptr->cred); 428 msqkptr->cred = NULL; 429 430 /* Free the message headers */ 431 msghdr = msqkptr->u.msg_first; 432 while (msghdr != NULL) { 433 struct msg *msghdr_tmp; 434 435 /* Free the segments of each message */ 436 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 437 msqkptr->u.msg_qnum--; 438 msghdr_tmp = msghdr; 439 msghdr = msghdr->msg_next; 440 msg_freehdr(msghdr_tmp); 441 } 442 443 if (msqkptr->u.msg_cbytes != 0) 444 panic("msg_cbytes is screwed up"); 445 if (msqkptr->u.msg_qnum != 0) 446 panic("msg_qnum is screwed up"); 447 448 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 449 450#ifdef MAC 451 mac_sysvmsq_cleanup(msqkptr); 452#endif 453 454 wakeup(msqkptr); 455} 456 457static struct prison * 458msg_find_prison(struct ucred *cred) 459{ 460 struct prison *pr, *rpr; 461 462 pr = cred->cr_prison; 463 prison_lock(pr); 464 rpr = osd_jail_get(pr, msg_prison_slot); 465 prison_unlock(pr); 466 return rpr; 467} 468 469static int 470msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr) 471{ 472 473 if (msqkptr->cred == NULL || 474 !(rpr == msqkptr->cred->cr_prison || 475 prison_ischild(rpr, msqkptr->cred->cr_prison))) 476 return (EINVAL); 477 return (0); 478} 479 480#ifndef _SYS_SYSPROTO_H_ 481struct msgctl_args { 482 int msqid; 483 int cmd; 484 struct msqid_ds *buf; 485}; 486#endif 487int 488sys_msgctl(td, uap) 489 struct thread *td; 490 register struct msgctl_args *uap; 491{ 492 int msqid = uap->msqid; 493 int cmd = uap->cmd; 494 struct msqid_ds msqbuf; 495 int error; 496 497 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 498 if (cmd == IPC_SET && 499 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 500 return (error); 501 error = kern_msgctl(td, msqid, cmd, &msqbuf); 502 if (cmd == IPC_STAT && error == 0) 503 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 504 return (error); 505} 506 507int 508kern_msgctl(td, msqid, cmd, msqbuf) 509 struct thread *td; 510 int msqid; 511 int cmd; 512 struct msqid_ds *msqbuf; 513{ 514 int rval, error, msqix; 515 register struct msqid_kernel *msqkptr; 516 struct prison *rpr; 517 518 rpr = msg_find_prison(td->td_ucred); 519 if (rpr == NULL) 520 return (ENOSYS); 521 522 msqix = IPCID_TO_IX(msqid); 523 524 if (msqix < 0 || msqix >= msginfo.msgmni) { 525 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 526 msginfo.msgmni)); 527 return (EINVAL); 528 } 529 530 msqkptr = &msqids[msqix]; 531 532 mtx_lock(&msq_mtx); 533 if (msqkptr->u.msg_qbytes == 0) { 534 DPRINTF(("no such msqid\n")); 535 error = EINVAL; 536 goto done2; 537 } 538 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 539 DPRINTF(("wrong sequence number\n")); 540 error = EINVAL; 541 goto done2; 542 } 543 544 error = msq_prison_cansee(rpr, msqkptr); 545 if (error != 0) { 546 DPRINTF(("requester can't see prison\n")); 547 goto done2; 548 } 549 550#ifdef MAC 551 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 552 if (error != 0) 553 goto done2; 554#endif 555 556 error = 0; 557 rval = 0; 558 559 switch (cmd) { 560 561 case IPC_RMID: 562 { 563#ifdef MAC 564 struct msg *msghdr; 565#endif 566 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 567 goto done2; 568 569#ifdef MAC 570 /* 571 * Check that the thread has MAC access permissions to 572 * individual msghdrs. Note: We need to do this in a 573 * separate loop because the actual loop alters the 574 * msq/msghdr info as it progresses, and there is no going 575 * back if half the way through we discover that the 576 * thread cannot free a certain msghdr. The msq will get 577 * into an inconsistent state. 578 */ 579 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 580 msghdr = msghdr->msg_next) { 581 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 582 if (error != 0) 583 goto done2; 584 } 585#endif 586 587 msq_remove(msqkptr); 588 } 589 590 break; 591 592 case IPC_SET: 593 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 594 goto done2; 595 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 596 error = priv_check(td, PRIV_IPC_MSGSIZE); 597 if (error) 598 goto done2; 599 } 600 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 601 DPRINTF(("can't increase msg_qbytes beyond %d" 602 "(truncating)\n", msginfo.msgmnb)); 603 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 604 } 605 if (msqbuf->msg_qbytes == 0) { 606 DPRINTF(("can't reduce msg_qbytes to 0\n")); 607 error = EINVAL; /* non-standard errno! */ 608 goto done2; 609 } 610 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 611 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 612 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 613 (msqbuf->msg_perm.mode & 0777); 614 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 615 msqkptr->u.msg_ctime = time_second; 616 break; 617 618 case IPC_STAT: 619 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 620 DPRINTF(("requester doesn't have read access\n")); 621 goto done2; 622 } 623 *msqbuf = msqkptr->u; 624 if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison) 625 msqbuf->msg_perm.key = IPC_PRIVATE; 626 break; 627 628 default: 629 DPRINTF(("invalid command %d\n", cmd)); 630 error = EINVAL; 631 goto done2; 632 } 633 634 if (error == 0) 635 td->td_retval[0] = rval; 636done2: 637 mtx_unlock(&msq_mtx); 638 return (error); 639} 640 641#ifndef _SYS_SYSPROTO_H_ 642struct msgget_args { 643 key_t key; 644 int msgflg; 645}; 646#endif 647 648int 649sys_msgget(td, uap) 650 struct thread *td; 651 register struct msgget_args *uap; 652{ 653 int msqid, error = 0; 654 int key = uap->key; 655 int msgflg = uap->msgflg; 656 struct ucred *cred = td->td_ucred; 657 register struct msqid_kernel *msqkptr = NULL; 658 659 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 660 661 if (msg_find_prison(cred) == NULL) 662 return (ENOSYS); 663 664 mtx_lock(&msq_mtx); 665 if (key != IPC_PRIVATE) { 666 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 667 msqkptr = &msqids[msqid]; 668 if (msqkptr->u.msg_qbytes != 0 && 669 msqkptr->cred != NULL && 670 msqkptr->cred->cr_prison == cred->cr_prison && 671 msqkptr->u.msg_perm.key == key) 672 break; 673 } 674 if (msqid < msginfo.msgmni) { 675 DPRINTF(("found public key\n")); 676 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 677 DPRINTF(("not exclusive\n")); 678 error = EEXIST; 679 goto done2; 680 } 681 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 682 msgflg & 0700))) { 683 DPRINTF(("requester doesn't have 0%o access\n", 684 msgflg & 0700)); 685 goto done2; 686 } 687#ifdef MAC 688 error = mac_sysvmsq_check_msqget(cred, msqkptr); 689 if (error != 0) 690 goto done2; 691#endif 692 goto found; 693 } 694 } 695 696 DPRINTF(("need to allocate the msqid_ds\n")); 697 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 698 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 699 /* 700 * Look for an unallocated and unlocked msqid_ds. 701 * msqid_ds's can be locked by msgsnd or msgrcv while 702 * they are copying the message in/out. We can't 703 * re-use the entry until they release it. 704 */ 705 msqkptr = &msqids[msqid]; 706 if (msqkptr->u.msg_qbytes == 0 && 707 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 708 break; 709 } 710 if (msqid == msginfo.msgmni) { 711 DPRINTF(("no more msqid_ds's available\n")); 712 error = ENOSPC; 713 goto done2; 714 } 715#ifdef RACCT 716 if (racct_enable) { 717 PROC_LOCK(td->td_proc); 718 error = racct_add(td->td_proc, RACCT_NMSGQ, 1); 719 PROC_UNLOCK(td->td_proc); 720 if (error != 0) { 721 error = ENOSPC; 722 goto done2; 723 } 724 } 725#endif 726 DPRINTF(("msqid %d is available\n", msqid)); 727 msqkptr->u.msg_perm.key = key; 728 msqkptr->u.msg_perm.cuid = cred->cr_uid; 729 msqkptr->u.msg_perm.uid = cred->cr_uid; 730 msqkptr->u.msg_perm.cgid = cred->cr_gid; 731 msqkptr->u.msg_perm.gid = cred->cr_gid; 732 msqkptr->u.msg_perm.mode = (msgflg & 0777); 733 msqkptr->cred = crhold(cred); 734 /* Make sure that the returned msqid is unique */ 735 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 736 msqkptr->u.msg_first = NULL; 737 msqkptr->u.msg_last = NULL; 738 msqkptr->u.msg_cbytes = 0; 739 msqkptr->u.msg_qnum = 0; 740 msqkptr->u.msg_qbytes = msginfo.msgmnb; 741 msqkptr->u.msg_lspid = 0; 742 msqkptr->u.msg_lrpid = 0; 743 msqkptr->u.msg_stime = 0; 744 msqkptr->u.msg_rtime = 0; 745 msqkptr->u.msg_ctime = time_second; 746#ifdef MAC 747 mac_sysvmsq_create(cred, msqkptr); 748#endif 749 } else { 750 DPRINTF(("didn't find it and wasn't asked to create it\n")); 751 error = ENOENT; 752 goto done2; 753 } 754 755found: 756 /* Construct the unique msqid */ 757 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 758done2: 759 mtx_unlock(&msq_mtx); 760 return (error); 761} 762 763#ifndef _SYS_SYSPROTO_H_ 764struct msgsnd_args { 765 int msqid; 766 const void *msgp; 767 size_t msgsz; 768 int msgflg; 769}; 770#endif 771int 772kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 773 struct thread *td; 774 int msqid; 775 const void *msgp; /* XXX msgp is actually mtext. */ 776 size_t msgsz; 777 int msgflg; 778 long mtype; 779{ 780 int msqix, segs_needed, error = 0; 781 register struct msqid_kernel *msqkptr; 782 register struct msg *msghdr; 783 struct prison *rpr; 784 short next; 785#ifdef RACCT 786 size_t saved_msgsz; 787#endif 788 789 rpr = msg_find_prison(td->td_ucred); 790 if (rpr == NULL) 791 return (ENOSYS); 792 793 mtx_lock(&msq_mtx); 794 msqix = IPCID_TO_IX(msqid); 795 796 if (msqix < 0 || msqix >= msginfo.msgmni) { 797 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 798 msginfo.msgmni)); 799 error = EINVAL; 800 goto done2; 801 } 802 803 msqkptr = &msqids[msqix]; 804 if (msqkptr->u.msg_qbytes == 0) { 805 DPRINTF(("no such message queue id\n")); 806 error = EINVAL; 807 goto done2; 808 } 809 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 810 DPRINTF(("wrong sequence number\n")); 811 error = EINVAL; 812 goto done2; 813 } 814 815 if ((error = msq_prison_cansee(rpr, msqkptr))) { 816 DPRINTF(("requester can't see prison\n")); 817 goto done2; 818 } 819 820 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 821 DPRINTF(("requester doesn't have write access\n")); 822 goto done2; 823 } 824 825#ifdef MAC 826 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 827 if (error != 0) 828 goto done2; 829#endif 830 831#ifdef RACCT 832 if (racct_enable) { 833 PROC_LOCK(td->td_proc); 834 if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) { 835 PROC_UNLOCK(td->td_proc); 836 error = EAGAIN; 837 goto done2; 838 } 839 saved_msgsz = msgsz; 840 if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) { 841 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 842 PROC_UNLOCK(td->td_proc); 843 error = EAGAIN; 844 goto done2; 845 } 846 PROC_UNLOCK(td->td_proc); 847 } 848#endif 849 850 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 851 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 852 msginfo.msgssz, segs_needed)); 853 for (;;) { 854 int need_more_resources = 0; 855 856 /* 857 * check msgsz 858 * (inside this loop in case msg_qbytes changes while we sleep) 859 */ 860 861 if (msgsz > msqkptr->u.msg_qbytes) { 862 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 863 error = EINVAL; 864 goto done3; 865 } 866 867 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 868 DPRINTF(("msqid is locked\n")); 869 need_more_resources = 1; 870 } 871 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 872 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 873 need_more_resources = 1; 874 } 875 if (segs_needed > nfree_msgmaps) { 876 DPRINTF(("segs_needed > nfree_msgmaps\n")); 877 need_more_resources = 1; 878 } 879 if (free_msghdrs == NULL) { 880 DPRINTF(("no more msghdrs\n")); 881 need_more_resources = 1; 882 } 883 884 if (need_more_resources) { 885 int we_own_it; 886 887 if ((msgflg & IPC_NOWAIT) != 0) { 888 DPRINTF(("need more resources but caller " 889 "doesn't want to wait\n")); 890 error = EAGAIN; 891 goto done3; 892 } 893 894 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 895 DPRINTF(("we don't own the msqid_ds\n")); 896 we_own_it = 0; 897 } else { 898 /* Force later arrivals to wait for our 899 request */ 900 DPRINTF(("we own the msqid_ds\n")); 901 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 902 we_own_it = 1; 903 } 904 DPRINTF(("msgsnd: goodnight\n")); 905 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 906 "msgsnd", hz); 907 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 908 if (we_own_it) 909 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 910 if (error == EWOULDBLOCK) { 911 DPRINTF(("msgsnd: timed out\n")); 912 continue; 913 } 914 if (error != 0) { 915 DPRINTF(("msgsnd: interrupted system call\n")); 916 error = EINTR; 917 goto done3; 918 } 919 920 /* 921 * Make sure that the msq queue still exists 922 */ 923 924 if (msqkptr->u.msg_qbytes == 0) { 925 DPRINTF(("msqid deleted\n")); 926 error = EIDRM; 927 goto done3; 928 } 929 930 } else { 931 DPRINTF(("got all the resources that we need\n")); 932 break; 933 } 934 } 935 936 /* 937 * We have the resources that we need. 938 * Make sure! 939 */ 940 941 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 942 panic("msg_perm.mode & MSG_LOCKED"); 943 if (segs_needed > nfree_msgmaps) 944 panic("segs_needed > nfree_msgmaps"); 945 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 946 panic("msgsz + msg_cbytes > msg_qbytes"); 947 if (free_msghdrs == NULL) 948 panic("no more msghdrs"); 949 950 /* 951 * Re-lock the msqid_ds in case we page-fault when copying in the 952 * message 953 */ 954 955 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 956 panic("msqid_ds is already locked"); 957 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 958 959 /* 960 * Allocate a message header 961 */ 962 963 msghdr = free_msghdrs; 964 free_msghdrs = msghdr->msg_next; 965 msghdr->msg_spot = -1; 966 msghdr->msg_ts = msgsz; 967 msghdr->msg_type = mtype; 968#ifdef MAC 969 /* 970 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 971 * immediately? Or, should it be checked just before the msg is 972 * enqueued in the msgq (as it is done now)? 973 */ 974 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 975#endif 976 977 /* 978 * Allocate space for the message 979 */ 980 981 while (segs_needed > 0) { 982 if (nfree_msgmaps <= 0) 983 panic("not enough msgmaps"); 984 if (free_msgmaps == -1) 985 panic("nil free_msgmaps"); 986 next = free_msgmaps; 987 if (next <= -1) 988 panic("next too low #1"); 989 if (next >= msginfo.msgseg) 990 panic("next out of range #1"); 991 DPRINTF(("allocating segment %d to message\n", next)); 992 free_msgmaps = msgmaps[next].next; 993 nfree_msgmaps--; 994 msgmaps[next].next = msghdr->msg_spot; 995 msghdr->msg_spot = next; 996 segs_needed--; 997 } 998 999 /* 1000 * Validate the message type 1001 */ 1002 1003 if (msghdr->msg_type < 1) { 1004 msg_freehdr(msghdr); 1005 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1006 wakeup(msqkptr); 1007 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 1008 error = EINVAL; 1009 goto done3; 1010 } 1011 1012 /* 1013 * Copy in the message body 1014 */ 1015 1016 next = msghdr->msg_spot; 1017 while (msgsz > 0) { 1018 size_t tlen; 1019 if (msgsz > msginfo.msgssz) 1020 tlen = msginfo.msgssz; 1021 else 1022 tlen = msgsz; 1023 if (next <= -1) 1024 panic("next too low #2"); 1025 if (next >= msginfo.msgseg) 1026 panic("next out of range #2"); 1027 mtx_unlock(&msq_mtx); 1028 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 1029 tlen)) != 0) { 1030 mtx_lock(&msq_mtx); 1031 DPRINTF(("error %d copying in message segment\n", 1032 error)); 1033 msg_freehdr(msghdr); 1034 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1035 wakeup(msqkptr); 1036 goto done3; 1037 } 1038 mtx_lock(&msq_mtx); 1039 msgsz -= tlen; 1040 msgp = (const char *)msgp + tlen; 1041 next = msgmaps[next].next; 1042 } 1043 if (next != -1) 1044 panic("didn't use all the msg segments"); 1045 1046 /* 1047 * We've got the message. Unlock the msqid_ds. 1048 */ 1049 1050 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1051 1052 /* 1053 * Make sure that the msqid_ds is still allocated. 1054 */ 1055 1056 if (msqkptr->u.msg_qbytes == 0) { 1057 msg_freehdr(msghdr); 1058 wakeup(msqkptr); 1059 error = EIDRM; 1060 goto done3; 1061 } 1062 1063#ifdef MAC 1064 /* 1065 * Note: Since the task/thread allocates the msghdr and usually 1066 * primes it with its own MAC label, for a majority of policies, it 1067 * won't be necessary to check whether the msghdr has access 1068 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 1069 * suffice in that case. However, this hook may be required where 1070 * individual policies derive a non-identical label for the msghdr 1071 * from the current thread label and may want to check the msghdr 1072 * enqueue permissions, along with read/write permissions to the 1073 * msgq. 1074 */ 1075 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 1076 if (error != 0) { 1077 msg_freehdr(msghdr); 1078 wakeup(msqkptr); 1079 goto done3; 1080 } 1081#endif 1082 1083 /* 1084 * Put the message into the queue 1085 */ 1086 if (msqkptr->u.msg_first == NULL) { 1087 msqkptr->u.msg_first = msghdr; 1088 msqkptr->u.msg_last = msghdr; 1089 } else { 1090 msqkptr->u.msg_last->msg_next = msghdr; 1091 msqkptr->u.msg_last = msghdr; 1092 } 1093 msqkptr->u.msg_last->msg_next = NULL; 1094 1095 msqkptr->u.msg_cbytes += msghdr->msg_ts; 1096 msqkptr->u.msg_qnum++; 1097 msqkptr->u.msg_lspid = td->td_proc->p_pid; 1098 msqkptr->u.msg_stime = time_second; 1099 1100 wakeup(msqkptr); 1101 td->td_retval[0] = 0; 1102done3: 1103#ifdef RACCT 1104 if (racct_enable && error != 0) { 1105 PROC_LOCK(td->td_proc); 1106 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 1107 racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz); 1108 PROC_UNLOCK(td->td_proc); 1109 } 1110#endif 1111done2: 1112 mtx_unlock(&msq_mtx); 1113 return (error); 1114} 1115 1116int 1117sys_msgsnd(td, uap) 1118 struct thread *td; 1119 register struct msgsnd_args *uap; 1120{ 1121 int error; 1122 long mtype; 1123 1124 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 1125 uap->msgsz, uap->msgflg)); 1126 1127 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 1128 DPRINTF(("error %d copying the message type\n", error)); 1129 return (error); 1130 } 1131 return (kern_msgsnd(td, uap->msqid, 1132 (const char *)uap->msgp + sizeof(mtype), 1133 uap->msgsz, uap->msgflg, mtype)); 1134} 1135 1136#ifndef _SYS_SYSPROTO_H_ 1137struct msgrcv_args { 1138 int msqid; 1139 void *msgp; 1140 size_t msgsz; 1141 long msgtyp; 1142 int msgflg; 1143}; 1144#endif 1145int 1146kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1147 struct thread *td; 1148 int msqid; 1149 void *msgp; /* XXX msgp is actually mtext. */ 1150 size_t msgsz; 1151 long msgtyp; 1152 int msgflg; 1153 long *mtype; 1154{ 1155 size_t len; 1156 register struct msqid_kernel *msqkptr; 1157 register struct msg *msghdr; 1158 struct prison *rpr; 1159 int msqix, error = 0; 1160 short next; 1161 1162 rpr = msg_find_prison(td->td_ucred); 1163 if (rpr == NULL) 1164 return (ENOSYS); 1165 1166 msqix = IPCID_TO_IX(msqid); 1167 1168 if (msqix < 0 || msqix >= msginfo.msgmni) { 1169 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1170 msginfo.msgmni)); 1171 return (EINVAL); 1172 } 1173 1174 msqkptr = &msqids[msqix]; 1175 mtx_lock(&msq_mtx); 1176 if (msqkptr->u.msg_qbytes == 0) { 1177 DPRINTF(("no such message queue id\n")); 1178 error = EINVAL; 1179 goto done2; 1180 } 1181 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1182 DPRINTF(("wrong sequence number\n")); 1183 error = EINVAL; 1184 goto done2; 1185 } 1186 1187 if ((error = msq_prison_cansee(rpr, msqkptr))) { 1188 DPRINTF(("requester can't see prison\n")); 1189 goto done2; 1190 } 1191 1192 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1193 DPRINTF(("requester doesn't have read access\n")); 1194 goto done2; 1195 } 1196 1197#ifdef MAC 1198 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1199 if (error != 0) 1200 goto done2; 1201#endif 1202 1203 msghdr = NULL; 1204 while (msghdr == NULL) { 1205 if (msgtyp == 0) { 1206 msghdr = msqkptr->u.msg_first; 1207 if (msghdr != NULL) { 1208 if (msgsz < msghdr->msg_ts && 1209 (msgflg & MSG_NOERROR) == 0) { 1210 DPRINTF(("first message on the queue " 1211 "is too big (want %zu, got %d)\n", 1212 msgsz, msghdr->msg_ts)); 1213 error = E2BIG; 1214 goto done2; 1215 } 1216#ifdef MAC 1217 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1218 msghdr); 1219 if (error != 0) 1220 goto done2; 1221#endif 1222 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1223 msqkptr->u.msg_first = NULL; 1224 msqkptr->u.msg_last = NULL; 1225 } else { 1226 msqkptr->u.msg_first = msghdr->msg_next; 1227 if (msqkptr->u.msg_first == NULL) 1228 panic("msg_first/last screwed up #1"); 1229 } 1230 } 1231 } else { 1232 struct msg *previous; 1233 struct msg **prev; 1234 1235 previous = NULL; 1236 prev = &(msqkptr->u.msg_first); 1237 while ((msghdr = *prev) != NULL) { 1238 /* 1239 * Is this message's type an exact match or is 1240 * this message's type less than or equal to 1241 * the absolute value of a negative msgtyp? 1242 * Note that the second half of this test can 1243 * NEVER be true if msgtyp is positive since 1244 * msg_type is always positive! 1245 */ 1246 1247 if (msgtyp == msghdr->msg_type || 1248 msghdr->msg_type <= -msgtyp) { 1249 DPRINTF(("found message type %ld, " 1250 "requested %ld\n", 1251 msghdr->msg_type, msgtyp)); 1252 if (msgsz < msghdr->msg_ts && 1253 (msgflg & MSG_NOERROR) == 0) { 1254 DPRINTF(("requested message " 1255 "on the queue is too big " 1256 "(want %zu, got %hu)\n", 1257 msgsz, msghdr->msg_ts)); 1258 error = E2BIG; 1259 goto done2; 1260 } 1261#ifdef MAC 1262 error = mac_sysvmsq_check_msgrcv( 1263 td->td_ucred, msghdr); 1264 if (error != 0) 1265 goto done2; 1266#endif 1267 *prev = msghdr->msg_next; 1268 if (msghdr == msqkptr->u.msg_last) { 1269 if (previous == NULL) { 1270 if (prev != 1271 &msqkptr->u.msg_first) 1272 panic("msg_first/last screwed up #2"); 1273 msqkptr->u.msg_first = 1274 NULL; 1275 msqkptr->u.msg_last = 1276 NULL; 1277 } else { 1278 if (prev == 1279 &msqkptr->u.msg_first) 1280 panic("msg_first/last screwed up #3"); 1281 msqkptr->u.msg_last = 1282 previous; 1283 } 1284 } 1285 break; 1286 } 1287 previous = msghdr; 1288 prev = &(msghdr->msg_next); 1289 } 1290 } 1291 1292 /* 1293 * We've either extracted the msghdr for the appropriate 1294 * message or there isn't one. 1295 * If there is one then bail out of this loop. 1296 */ 1297 1298 if (msghdr != NULL) 1299 break; 1300 1301 /* 1302 * Hmph! No message found. Does the user want to wait? 1303 */ 1304 1305 if ((msgflg & IPC_NOWAIT) != 0) { 1306 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1307 msgtyp)); 1308 /* The SVID says to return ENOMSG. */ 1309 error = ENOMSG; 1310 goto done2; 1311 } 1312 1313 /* 1314 * Wait for something to happen 1315 */ 1316 1317 DPRINTF(("msgrcv: goodnight\n")); 1318 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1319 "msgrcv", 0); 1320 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1321 1322 if (error != 0) { 1323 DPRINTF(("msgrcv: interrupted system call\n")); 1324 error = EINTR; 1325 goto done2; 1326 } 1327 1328 /* 1329 * Make sure that the msq queue still exists 1330 */ 1331 1332 if (msqkptr->u.msg_qbytes == 0 || 1333 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1334 DPRINTF(("msqid deleted\n")); 1335 error = EIDRM; 1336 goto done2; 1337 } 1338 } 1339 1340 /* 1341 * Return the message to the user. 1342 * 1343 * First, do the bookkeeping (before we risk being interrupted). 1344 */ 1345 1346 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1347 msqkptr->u.msg_qnum--; 1348 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1349 msqkptr->u.msg_rtime = time_second; 1350 1351 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1); 1352 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts); 1353 1354 /* 1355 * Make msgsz the actual amount that we'll be returning. 1356 * Note that this effectively truncates the message if it is too long 1357 * (since msgsz is never increased). 1358 */ 1359 1360 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1361 msghdr->msg_ts)); 1362 if (msgsz > msghdr->msg_ts) 1363 msgsz = msghdr->msg_ts; 1364 *mtype = msghdr->msg_type; 1365 1366 /* 1367 * Return the segments to the user 1368 */ 1369 1370 next = msghdr->msg_spot; 1371 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1372 size_t tlen; 1373 1374 if (msgsz - len > msginfo.msgssz) 1375 tlen = msginfo.msgssz; 1376 else 1377 tlen = msgsz - len; 1378 if (next <= -1) 1379 panic("next too low #3"); 1380 if (next >= msginfo.msgseg) 1381 panic("next out of range #3"); 1382 mtx_unlock(&msq_mtx); 1383 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1384 mtx_lock(&msq_mtx); 1385 if (error != 0) { 1386 DPRINTF(("error (%d) copying out message segment\n", 1387 error)); 1388 msg_freehdr(msghdr); 1389 wakeup(msqkptr); 1390 goto done2; 1391 } 1392 msgp = (char *)msgp + tlen; 1393 next = msgmaps[next].next; 1394 } 1395 1396 /* 1397 * Done, return the actual number of bytes copied out. 1398 */ 1399 1400 msg_freehdr(msghdr); 1401 wakeup(msqkptr); 1402 td->td_retval[0] = msgsz; 1403done2: 1404 mtx_unlock(&msq_mtx); 1405 return (error); 1406} 1407 1408int 1409sys_msgrcv(td, uap) 1410 struct thread *td; 1411 register struct msgrcv_args *uap; 1412{ 1413 int error; 1414 long mtype; 1415 1416 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1417 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1418 1419 if ((error = kern_msgrcv(td, uap->msqid, 1420 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1421 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1422 return (error); 1423 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1424 DPRINTF(("error %d copying the message type\n", error)); 1425 return (error); 1426} 1427 1428static int 1429sysctl_msqids(SYSCTL_HANDLER_ARGS) 1430{ 1431 struct msqid_kernel tmsqk; 1432 struct prison *pr, *rpr; 1433 int error, i; 1434 1435 pr = req->td->td_ucred->cr_prison; 1436 rpr = msg_find_prison(req->td->td_ucred); 1437 error = 0; 1438 for (i = 0; i < msginfo.msgmni; i++) { 1439 mtx_lock(&msq_mtx); 1440 if (msqids[i].u.msg_qbytes == 0 || rpr == NULL || 1441 msq_prison_cansee(rpr, &msqids[i]) != 0) 1442 bzero(&tmsqk, sizeof(tmsqk)); 1443 else { 1444 tmsqk = msqids[i]; 1445 if (tmsqk.cred->cr_prison != pr) 1446 tmsqk.u.msg_perm.key = IPC_PRIVATE; 1447 } 1448 mtx_unlock(&msq_mtx); 1449 error = SYSCTL_OUT(req, &tmsqk, sizeof(tmsqk)); 1450 if (error != 0) 1451 break; 1452 } 1453 return (error); 1454} 1455 1456SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1457 "Maximum message size"); 1458SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1459 "Number of message queue identifiers"); 1460SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1461 "Maximum number of bytes in a queue"); 1462SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1463 "Maximum number of messages in the system"); 1464SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1465 "Size of a message segment"); 1466SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1467 "Number of message segments"); 1468SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, 1469 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 1470 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1471 1472static int 1473msg_prison_check(void *obj, void *data) 1474{ 1475 struct prison *pr = obj; 1476 struct prison *prpr; 1477 struct vfsoptlist *opts = data; 1478 int error, jsys; 1479 1480 /* 1481 * sysvmsg is a jailsys integer. 1482 * It must be "disable" if the parent jail is disabled. 1483 */ 1484 error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1485 if (error != ENOENT) { 1486 if (error != 0) 1487 return (error); 1488 switch (jsys) { 1489 case JAIL_SYS_DISABLE: 1490 break; 1491 case JAIL_SYS_NEW: 1492 case JAIL_SYS_INHERIT: 1493 prison_lock(pr->pr_parent); 1494 prpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1495 prison_unlock(pr->pr_parent); 1496 if (prpr == NULL) 1497 return (EPERM); 1498 break; 1499 default: 1500 return (EINVAL); 1501 } 1502 } 1503 1504 return (0); 1505} 1506 1507static int 1508msg_prison_set(void *obj, void *data) 1509{ 1510 struct prison *pr = obj; 1511 struct prison *tpr, *orpr, *nrpr, *trpr; 1512 struct vfsoptlist *opts = data; 1513 void *rsv; 1514 int jsys, descend; 1515 1516 /* 1517 * sysvmsg controls which jail is the root of the associated msgs (this 1518 * jail or same as the parent), or if the feature is available at all. 1519 */ 1520 if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT) 1521 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0) 1522 ? JAIL_SYS_INHERIT 1523 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0) 1524 ? JAIL_SYS_DISABLE 1525 : -1; 1526 if (jsys == JAIL_SYS_DISABLE) { 1527 prison_lock(pr); 1528 orpr = osd_jail_get(pr, msg_prison_slot); 1529 if (orpr != NULL) 1530 osd_jail_del(pr, msg_prison_slot); 1531 prison_unlock(pr); 1532 if (orpr != NULL) { 1533 if (orpr == pr) 1534 msg_prison_cleanup(pr); 1535 /* Disable all child jails as well. */ 1536 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1537 prison_lock(tpr); 1538 trpr = osd_jail_get(tpr, msg_prison_slot); 1539 if (trpr != NULL) { 1540 osd_jail_del(tpr, msg_prison_slot); 1541 prison_unlock(tpr); 1542 if (trpr == tpr) 1543 msg_prison_cleanup(tpr); 1544 } else { 1545 prison_unlock(tpr); 1546 descend = 0; 1547 } 1548 } 1549 } 1550 } else if (jsys != -1) { 1551 if (jsys == JAIL_SYS_NEW) 1552 nrpr = pr; 1553 else { 1554 prison_lock(pr->pr_parent); 1555 nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1556 prison_unlock(pr->pr_parent); 1557 } 1558 rsv = osd_reserve(msg_prison_slot); 1559 prison_lock(pr); 1560 orpr = osd_jail_get(pr, msg_prison_slot); 1561 if (orpr != nrpr) 1562 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 1563 nrpr); 1564 else 1565 osd_free_reserved(rsv); 1566 prison_unlock(pr); 1567 if (orpr != nrpr) { 1568 if (orpr == pr) 1569 msg_prison_cleanup(pr); 1570 if (orpr != NULL) { 1571 /* Change child jails matching the old root, */ 1572 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1573 prison_lock(tpr); 1574 trpr = osd_jail_get(tpr, 1575 msg_prison_slot); 1576 if (trpr == orpr) { 1577 (void)osd_jail_set(tpr, 1578 msg_prison_slot, nrpr); 1579 prison_unlock(tpr); 1580 if (trpr == tpr) 1581 msg_prison_cleanup(tpr); 1582 } else { 1583 prison_unlock(tpr); 1584 descend = 0; 1585 } 1586 } 1587 } 1588 } 1589 } 1590 1591 return (0); 1592} 1593 1594static int 1595msg_prison_get(void *obj, void *data) 1596{ 1597 struct prison *pr = obj; 1598 struct prison *rpr; 1599 struct vfsoptlist *opts = data; 1600 int error, jsys; 1601 1602 /* Set sysvmsg based on the jail's root prison. */ 1603 prison_lock(pr); 1604 rpr = osd_jail_get(pr, msg_prison_slot); 1605 prison_unlock(pr); 1606 jsys = rpr == NULL ? JAIL_SYS_DISABLE 1607 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; 1608 error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1609 if (error == ENOENT) 1610 error = 0; 1611 return (error); 1612} 1613 1614static int 1615msg_prison_remove(void *obj, void *data __unused) 1616{ 1617 struct prison *pr = obj; 1618 struct prison *rpr; 1619 1620 prison_lock(pr); 1621 rpr = osd_jail_get(pr, msg_prison_slot); 1622 prison_unlock(pr); 1623 if (rpr == pr) 1624 msg_prison_cleanup(pr); 1625 return (0); 1626} 1627 1628static void 1629msg_prison_cleanup(struct prison *pr) 1630{ 1631 struct msqid_kernel *msqkptr; 1632 int i; 1633 1634 /* Remove any msqs that belong to this jail. */ 1635 mtx_lock(&msq_mtx); 1636 for (i = 0; i < msginfo.msgmni; i++) { 1637 msqkptr = &msqids[i]; 1638 if (msqkptr->u.msg_qbytes != 0 && 1639 msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr) 1640 msq_remove(msqkptr); 1641 } 1642 mtx_unlock(&msq_mtx); 1643} 1644 1645SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues"); 1646 1647#ifdef COMPAT_FREEBSD32 1648int 1649freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap) 1650{ 1651 1652#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1653 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1654 switch (uap->which) { 1655 case 0: 1656 return (freebsd7_freebsd32_msgctl(td, 1657 (struct freebsd7_freebsd32_msgctl_args *)&uap->a2)); 1658 case 2: 1659 return (freebsd32_msgsnd(td, 1660 (struct freebsd32_msgsnd_args *)&uap->a2)); 1661 case 3: 1662 return (freebsd32_msgrcv(td, 1663 (struct freebsd32_msgrcv_args *)&uap->a2)); 1664 default: 1665 return (sys_msgsys(td, (struct msgsys_args *)uap)); 1666 } 1667#else 1668 return (nosys(td, NULL)); 1669#endif 1670} 1671 1672#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1673 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1674int 1675freebsd7_freebsd32_msgctl(struct thread *td, 1676 struct freebsd7_freebsd32_msgctl_args *uap) 1677{ 1678 struct msqid_ds msqbuf; 1679 struct msqid_ds32_old msqbuf32; 1680 int error; 1681 1682 if (uap->cmd == IPC_SET) { 1683 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1684 if (error) 1685 return (error); 1686 freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1687 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1688 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1689 CP(msqbuf32, msqbuf, msg_cbytes); 1690 CP(msqbuf32, msqbuf, msg_qnum); 1691 CP(msqbuf32, msqbuf, msg_qbytes); 1692 CP(msqbuf32, msqbuf, msg_lspid); 1693 CP(msqbuf32, msqbuf, msg_lrpid); 1694 CP(msqbuf32, msqbuf, msg_stime); 1695 CP(msqbuf32, msqbuf, msg_rtime); 1696 CP(msqbuf32, msqbuf, msg_ctime); 1697 } 1698 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1699 if (error) 1700 return (error); 1701 if (uap->cmd == IPC_STAT) { 1702 bzero(&msqbuf32, sizeof(msqbuf32)); 1703 freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1704 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1705 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1706 CP(msqbuf, msqbuf32, msg_cbytes); 1707 CP(msqbuf, msqbuf32, msg_qnum); 1708 CP(msqbuf, msqbuf32, msg_qbytes); 1709 CP(msqbuf, msqbuf32, msg_lspid); 1710 CP(msqbuf, msqbuf32, msg_lrpid); 1711 CP(msqbuf, msqbuf32, msg_stime); 1712 CP(msqbuf, msqbuf32, msg_rtime); 1713 CP(msqbuf, msqbuf32, msg_ctime); 1714 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1715 } 1716 return (error); 1717} 1718#endif 1719 1720int 1721freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap) 1722{ 1723 struct msqid_ds msqbuf; 1724 struct msqid_ds32 msqbuf32; 1725 int error; 1726 1727 if (uap->cmd == IPC_SET) { 1728 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1729 if (error) 1730 return (error); 1731 freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1732 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1733 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1734 CP(msqbuf32, msqbuf, msg_cbytes); 1735 CP(msqbuf32, msqbuf, msg_qnum); 1736 CP(msqbuf32, msqbuf, msg_qbytes); 1737 CP(msqbuf32, msqbuf, msg_lspid); 1738 CP(msqbuf32, msqbuf, msg_lrpid); 1739 CP(msqbuf32, msqbuf, msg_stime); 1740 CP(msqbuf32, msqbuf, msg_rtime); 1741 CP(msqbuf32, msqbuf, msg_ctime); 1742 } 1743 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1744 if (error) 1745 return (error); 1746 if (uap->cmd == IPC_STAT) { 1747 freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1748 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1749 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1750 CP(msqbuf, msqbuf32, msg_cbytes); 1751 CP(msqbuf, msqbuf32, msg_qnum); 1752 CP(msqbuf, msqbuf32, msg_qbytes); 1753 CP(msqbuf, msqbuf32, msg_lspid); 1754 CP(msqbuf, msqbuf32, msg_lrpid); 1755 CP(msqbuf, msqbuf32, msg_stime); 1756 CP(msqbuf, msqbuf32, msg_rtime); 1757 CP(msqbuf, msqbuf32, msg_ctime); 1758 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1759 } 1760 return (error); 1761} 1762 1763int 1764freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap) 1765{ 1766 const void *msgp; 1767 long mtype; 1768 int32_t mtype32; 1769 int error; 1770 1771 msgp = PTRIN(uap->msgp); 1772 if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0) 1773 return (error); 1774 mtype = mtype32; 1775 return (kern_msgsnd(td, uap->msqid, 1776 (const char *)msgp + sizeof(mtype32), 1777 uap->msgsz, uap->msgflg, mtype)); 1778} 1779 1780int 1781freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap) 1782{ 1783 void *msgp; 1784 long mtype; 1785 int32_t mtype32; 1786 int error; 1787 1788 msgp = PTRIN(uap->msgp); 1789 if ((error = kern_msgrcv(td, uap->msqid, 1790 (char *)msgp + sizeof(mtype32), uap->msgsz, 1791 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1792 return (error); 1793 mtype32 = (int32_t)mtype; 1794 return (copyout(&mtype32, msgp, sizeof(mtype32))); 1795} 1796#endif 1797 1798#if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1799 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1800 1801/* XXX casting to (sy_call_t *) is bogus, as usual. */ 1802static sy_call_t *msgcalls[] = { 1803 (sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget, 1804 (sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv 1805}; 1806 1807/* 1808 * Entry point for all MSG calls. 1809 */ 1810int 1811sys_msgsys(td, uap) 1812 struct thread *td; 1813 /* XXX actually varargs. */ 1814 struct msgsys_args /* { 1815 int which; 1816 int a2; 1817 int a3; 1818 int a4; 1819 int a5; 1820 int a6; 1821 } */ *uap; 1822{ 1823 int error; 1824 1825 if (uap->which < 0 || 1826 uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) 1827 return (EINVAL); 1828 error = (*msgcalls[uap->which])(td, &uap->a2); 1829 return (error); 1830} 1831 1832#ifndef CP 1833#define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0) 1834#endif 1835 1836#ifndef _SYS_SYSPROTO_H_ 1837struct freebsd7_msgctl_args { 1838 int msqid; 1839 int cmd; 1840 struct msqid_ds_old *buf; 1841}; 1842#endif 1843int 1844freebsd7_msgctl(td, uap) 1845 struct thread *td; 1846 struct freebsd7_msgctl_args *uap; 1847{ 1848 struct msqid_ds_old msqold; 1849 struct msqid_ds msqbuf; 1850 int error; 1851 1852 DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd, 1853 uap->buf)); 1854 if (uap->cmd == IPC_SET) { 1855 error = copyin(uap->buf, &msqold, sizeof(msqold)); 1856 if (error) 1857 return (error); 1858 ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm); 1859 CP(msqold, msqbuf, msg_first); 1860 CP(msqold, msqbuf, msg_last); 1861 CP(msqold, msqbuf, msg_cbytes); 1862 CP(msqold, msqbuf, msg_qnum); 1863 CP(msqold, msqbuf, msg_qbytes); 1864 CP(msqold, msqbuf, msg_lspid); 1865 CP(msqold, msqbuf, msg_lrpid); 1866 CP(msqold, msqbuf, msg_stime); 1867 CP(msqold, msqbuf, msg_rtime); 1868 CP(msqold, msqbuf, msg_ctime); 1869 } 1870 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1871 if (error) 1872 return (error); 1873 if (uap->cmd == IPC_STAT) { 1874 bzero(&msqold, sizeof(msqold)); 1875 ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm); 1876 CP(msqbuf, msqold, msg_first); 1877 CP(msqbuf, msqold, msg_last); 1878 CP(msqbuf, msqold, msg_cbytes); 1879 CP(msqbuf, msqold, msg_qnum); 1880 CP(msqbuf, msqold, msg_qbytes); 1881 CP(msqbuf, msqold, msg_lspid); 1882 CP(msqbuf, msqold, msg_lrpid); 1883 CP(msqbuf, msqold, msg_stime); 1884 CP(msqbuf, msqold, msg_rtime); 1885 CP(msqbuf, msqold, msg_ctime); 1886 error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old)); 1887 } 1888 return (error); 1889} 1890 1891#undef CP 1892 1893#endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 || 1894 COMPAT_FREEBSD7 */ 1895