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_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/sbuf.h> 69 #include <sys/sx.h> 70 #include <sys/syscall.h> 71 #include <sys/syscallsubr.h> 72 #include <sys/sysent.h> 73 #include <sys/sysctl.h> 74 #include <sys/malloc.h> 75 #include <sys/jail.h> 76 77 #include <security/mac/mac_framework.h> 78 79 FEATURE(sysv_msg, "System V message queues support"); 80 81 static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); 82 83 static int msginit(void); 84 static int msgunload(void); 85 static int sysvmsg_modload(struct module *, int, void *); 86 static void msq_remove(struct msqid_kernel *); 87 static struct prison *msg_find_prison(struct ucred *); 88 static int msq_prison_cansee(struct prison *, struct msqid_kernel *); 89 static int msg_prison_check(void *, void *); 90 static int msg_prison_set(void *, void *); 91 static int msg_prison_get(void *, void *); 92 static int msg_prison_remove(void *, void *); 93 static void msg_prison_cleanup(struct prison *); 94 95 96 #ifdef MSG_DEBUG 97 #define DPRINTF(a) printf a 98 #else 99 #define DPRINTF(a) (void)0 100 #endif 101 102 static void msg_freehdr(struct msg *msghdr); 103 104 #ifndef MSGSSZ 105 #define MSGSSZ 8 /* Each segment must be 2^N long */ 106 #endif 107 #ifndef MSGSEG 108 #define MSGSEG 2048 /* must be less than 32767 */ 109 #endif 110 #define MSGMAX (MSGSSZ*MSGSEG) 111 #ifndef MSGMNB 112 #define MSGMNB 2048 /* max # of bytes in a queue */ 113 #endif 114 #ifndef MSGMNI 115 #define MSGMNI 40 116 #endif 117 #ifndef MSGTQL 118 #define MSGTQL 40 119 #endif 120 121 /* 122 * Based on the configuration parameters described in an SVR2 (yes, two) 123 * config(1m) man page. 124 * 125 * Each message is broken up and stored in segments that are msgssz bytes 126 * long. For efficiency reasons, this should be a power of two. Also, 127 * it doesn't make sense if it is less than 8 or greater than about 256. 128 * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of 129 * two between 8 and 1024 inclusive (and panic's if it isn't). 130 */ 131 struct msginfo msginfo = { 132 MSGMAX, /* max chars in a message */ 133 MSGMNI, /* # of message queue identifiers */ 134 MSGMNB, /* max chars in a queue */ 135 MSGTQL, /* max messages in system */ 136 MSGSSZ, /* size of a message segment */ 137 /* (must be small power of 2 greater than 4) */ 138 MSGSEG /* number of message segments */ 139 }; 140 141 /* 142 * macros to convert between msqid_ds's and msqid's. 143 * (specific to this implementation) 144 */ 145 #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) 146 #define MSQID_IX(id) ((id) & 0xffff) 147 #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) 148 149 /* 150 * The rest of this file is specific to this particular implementation. 151 */ 152 153 struct msgmap { 154 short next; /* next segment in buffer */ 155 /* -1 -> available */ 156 /* 0..(MSGSEG-1) -> index of next segment */ 157 }; 158 159 #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ 160 161 static int nfree_msgmaps; /* # of free map entries */ 162 static short free_msgmaps; /* head of linked list of free map entries */ 163 static struct msg *free_msghdrs;/* list of free msg headers */ 164 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 165 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 166 static struct msg *msghdrs; /* MSGTQL msg headers */ 167 static struct msqid_kernel *msqids; /* MSGMNI msqid_kernel struct's */ 168 static struct mtx msq_mtx; /* global mutex for message queues. */ 169 static unsigned msg_prison_slot;/* prison OSD slot */ 170 171 static struct syscall_helper_data msg_syscalls[] = { 172 SYSCALL_INIT_HELPER(msgctl), 173 SYSCALL_INIT_HELPER(msgget), 174 SYSCALL_INIT_HELPER(msgsnd), 175 SYSCALL_INIT_HELPER(msgrcv), 176 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 177 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 178 SYSCALL_INIT_HELPER(msgsys), 179 SYSCALL_INIT_HELPER_COMPAT(freebsd7_msgctl), 180 #endif 181 SYSCALL_INIT_LAST 182 }; 183 184 #ifdef COMPAT_FREEBSD32 185 #include <compat/freebsd32/freebsd32.h> 186 #include <compat/freebsd32/freebsd32_ipc.h> 187 #include <compat/freebsd32/freebsd32_proto.h> 188 #include <compat/freebsd32/freebsd32_signal.h> 189 #include <compat/freebsd32/freebsd32_syscall.h> 190 #include <compat/freebsd32/freebsd32_util.h> 191 192 static struct syscall_helper_data msg32_syscalls[] = { 193 SYSCALL32_INIT_HELPER(freebsd32_msgctl), 194 SYSCALL32_INIT_HELPER(freebsd32_msgsnd), 195 SYSCALL32_INIT_HELPER(freebsd32_msgrcv), 196 SYSCALL32_INIT_HELPER_COMPAT(msgget), 197 SYSCALL32_INIT_HELPER(freebsd32_msgsys), 198 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 199 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 200 SYSCALL32_INIT_HELPER(freebsd7_freebsd32_msgctl), 201 #endif 202 SYSCALL_INIT_LAST 203 }; 204 #endif 205 206 static int 207 msginit() 208 { 209 struct prison *pr; 210 void *rsv; 211 int i, error; 212 osd_method_t methods[PR_MAXMETHOD] = { 213 [PR_METHOD_CHECK] = msg_prison_check, 214 [PR_METHOD_SET] = msg_prison_set, 215 [PR_METHOD_GET] = msg_prison_get, 216 [PR_METHOD_REMOVE] = msg_prison_remove, 217 }; 218 219 msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; 220 msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); 221 msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); 222 msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); 223 msqids = malloc(sizeof(struct msqid_kernel) * msginfo.msgmni, M_MSG, 224 M_WAITOK); 225 226 /* 227 * msginfo.msgssz should be a power of two for efficiency reasons. 228 * It is also pretty silly if msginfo.msgssz is less than 8 229 * or greater than about 256 so ... 230 */ 231 232 i = 8; 233 while (i < 1024 && i != msginfo.msgssz) 234 i <<= 1; 235 if (i != msginfo.msgssz) { 236 DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, 237 msginfo.msgssz)); 238 panic("msginfo.msgssz not a small power of 2"); 239 } 240 241 if (msginfo.msgseg > 32767) { 242 DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); 243 panic("msginfo.msgseg > 32767"); 244 } 245 246 for (i = 0; i < msginfo.msgseg; i++) { 247 if (i > 0) 248 msgmaps[i-1].next = i; 249 msgmaps[i].next = -1; /* implies entry is available */ 250 } 251 free_msgmaps = 0; 252 nfree_msgmaps = msginfo.msgseg; 253 254 for (i = 0; i < msginfo.msgtql; i++) { 255 msghdrs[i].msg_type = 0; 256 if (i > 0) 257 msghdrs[i-1].msg_next = &msghdrs[i]; 258 msghdrs[i].msg_next = NULL; 259 #ifdef MAC 260 mac_sysvmsg_init(&msghdrs[i]); 261 #endif 262 } 263 free_msghdrs = &msghdrs[0]; 264 265 for (i = 0; i < msginfo.msgmni; i++) { 266 msqids[i].u.msg_qbytes = 0; /* implies entry is available */ 267 msqids[i].u.msg_perm.seq = 0; /* reset to a known value */ 268 msqids[i].u.msg_perm.mode = 0; 269 #ifdef MAC 270 mac_sysvmsq_init(&msqids[i]); 271 #endif 272 } 273 mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); 274 275 /* Set current prisons according to their allow.sysvipc. */ 276 msg_prison_slot = osd_jail_register(NULL, methods); 277 rsv = osd_reserve(msg_prison_slot); 278 prison_lock(&prison0); 279 (void)osd_jail_set_reserved(&prison0, msg_prison_slot, rsv, &prison0); 280 prison_unlock(&prison0); 281 rsv = NULL; 282 sx_slock(&allprison_lock); 283 TAILQ_FOREACH(pr, &allprison, pr_list) { 284 if (rsv == NULL) 285 rsv = osd_reserve(msg_prison_slot); 286 prison_lock(pr); 287 if ((pr->pr_allow & PR_ALLOW_SYSVIPC) && pr->pr_ref > 0) { 288 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 289 &prison0); 290 rsv = NULL; 291 } 292 prison_unlock(pr); 293 } 294 if (rsv != NULL) 295 osd_free_reserved(rsv); 296 sx_sunlock(&allprison_lock); 297 298 error = syscall_helper_register(msg_syscalls, SY_THR_STATIC_KLD); 299 if (error != 0) 300 return (error); 301 #ifdef COMPAT_FREEBSD32 302 error = syscall32_helper_register(msg32_syscalls, SY_THR_STATIC_KLD); 303 if (error != 0) 304 return (error); 305 #endif 306 return (0); 307 } 308 309 static int 310 msgunload() 311 { 312 struct msqid_kernel *msqkptr; 313 int msqid; 314 #ifdef MAC 315 int i; 316 #endif 317 318 syscall_helper_unregister(msg_syscalls); 319 #ifdef COMPAT_FREEBSD32 320 syscall32_helper_unregister(msg32_syscalls); 321 #endif 322 323 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 324 /* 325 * Look for an unallocated and unlocked msqid_ds. 326 * msqid_ds's can be locked by msgsnd or msgrcv while 327 * they are copying the message in/out. We can't 328 * re-use the entry until they release it. 329 */ 330 msqkptr = &msqids[msqid]; 331 if (msqkptr->u.msg_qbytes != 0 || 332 (msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 333 break; 334 } 335 if (msqid != msginfo.msgmni) 336 return (EBUSY); 337 338 if (msg_prison_slot != 0) 339 osd_jail_deregister(msg_prison_slot); 340 #ifdef MAC 341 for (i = 0; i < msginfo.msgtql; i++) 342 mac_sysvmsg_destroy(&msghdrs[i]); 343 for (msqid = 0; msqid < msginfo.msgmni; msqid++) 344 mac_sysvmsq_destroy(&msqids[msqid]); 345 #endif 346 free(msgpool, M_MSG); 347 free(msgmaps, M_MSG); 348 free(msghdrs, M_MSG); 349 free(msqids, M_MSG); 350 mtx_destroy(&msq_mtx); 351 return (0); 352 } 353 354 355 static int 356 sysvmsg_modload(struct module *module, int cmd, void *arg) 357 { 358 int error = 0; 359 360 switch (cmd) { 361 case MOD_LOAD: 362 error = msginit(); 363 if (error != 0) 364 msgunload(); 365 break; 366 case MOD_UNLOAD: 367 error = msgunload(); 368 break; 369 case MOD_SHUTDOWN: 370 break; 371 default: 372 error = EINVAL; 373 break; 374 } 375 return (error); 376 } 377 378 static moduledata_t sysvmsg_mod = { 379 "sysvmsg", 380 &sysvmsg_modload, 381 NULL 382 }; 383 384 DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); 385 MODULE_VERSION(sysvmsg, 1); 386 387 static void 388 msg_freehdr(msghdr) 389 struct msg *msghdr; 390 { 391 while (msghdr->msg_ts > 0) { 392 short next; 393 if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) 394 panic("msghdr->msg_spot out of range"); 395 next = msgmaps[msghdr->msg_spot].next; 396 msgmaps[msghdr->msg_spot].next = free_msgmaps; 397 free_msgmaps = msghdr->msg_spot; 398 nfree_msgmaps++; 399 msghdr->msg_spot = next; 400 if (msghdr->msg_ts >= msginfo.msgssz) 401 msghdr->msg_ts -= msginfo.msgssz; 402 else 403 msghdr->msg_ts = 0; 404 } 405 if (msghdr->msg_spot != -1) 406 panic("msghdr->msg_spot != -1"); 407 msghdr->msg_next = free_msghdrs; 408 free_msghdrs = msghdr; 409 #ifdef MAC 410 mac_sysvmsg_cleanup(msghdr); 411 #endif 412 } 413 414 static void 415 msq_remove(struct msqid_kernel *msqkptr) 416 { 417 struct msg *msghdr; 418 419 racct_sub_cred(msqkptr->cred, RACCT_NMSGQ, 1); 420 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, msqkptr->u.msg_qnum); 421 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msqkptr->u.msg_cbytes); 422 crfree(msqkptr->cred); 423 msqkptr->cred = NULL; 424 425 /* Free the message headers */ 426 msghdr = msqkptr->u.msg_first; 427 while (msghdr != NULL) { 428 struct msg *msghdr_tmp; 429 430 /* Free the segments of each message */ 431 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 432 msqkptr->u.msg_qnum--; 433 msghdr_tmp = msghdr; 434 msghdr = msghdr->msg_next; 435 msg_freehdr(msghdr_tmp); 436 } 437 438 if (msqkptr->u.msg_cbytes != 0) 439 panic("msg_cbytes is screwed up"); 440 if (msqkptr->u.msg_qnum != 0) 441 panic("msg_qnum is screwed up"); 442 443 msqkptr->u.msg_qbytes = 0; /* Mark it as free */ 444 445 #ifdef MAC 446 mac_sysvmsq_cleanup(msqkptr); 447 #endif 448 449 wakeup(msqkptr); 450 } 451 452 static struct prison * 453 msg_find_prison(struct ucred *cred) 454 { 455 struct prison *pr, *rpr; 456 457 pr = cred->cr_prison; 458 prison_lock(pr); 459 rpr = osd_jail_get(pr, msg_prison_slot); 460 prison_unlock(pr); 461 return rpr; 462 } 463 464 static int 465 msq_prison_cansee(struct prison *rpr, struct msqid_kernel *msqkptr) 466 { 467 468 if (msqkptr->cred == NULL || 469 !(rpr == msqkptr->cred->cr_prison || 470 prison_ischild(rpr, msqkptr->cred->cr_prison))) 471 return (EINVAL); 472 return (0); 473 } 474 475 #ifndef _SYS_SYSPROTO_H_ 476 struct msgctl_args { 477 int msqid; 478 int cmd; 479 struct msqid_ds *buf; 480 }; 481 #endif 482 int 483 sys_msgctl(td, uap) 484 struct thread *td; 485 register struct msgctl_args *uap; 486 { 487 int msqid = uap->msqid; 488 int cmd = uap->cmd; 489 struct msqid_ds msqbuf; 490 int error; 491 492 DPRINTF(("call to msgctl(%d, %d, %p)\n", msqid, cmd, uap->buf)); 493 if (cmd == IPC_SET && 494 (error = copyin(uap->buf, &msqbuf, sizeof(msqbuf))) != 0) 495 return (error); 496 error = kern_msgctl(td, msqid, cmd, &msqbuf); 497 if (cmd == IPC_STAT && error == 0) 498 error = copyout(&msqbuf, uap->buf, sizeof(struct msqid_ds)); 499 return (error); 500 } 501 502 int 503 kern_msgctl(td, msqid, cmd, msqbuf) 504 struct thread *td; 505 int msqid; 506 int cmd; 507 struct msqid_ds *msqbuf; 508 { 509 int rval, error, msqix; 510 register struct msqid_kernel *msqkptr; 511 struct prison *rpr; 512 513 rpr = msg_find_prison(td->td_ucred); 514 if (rpr == NULL) 515 return (ENOSYS); 516 517 msqix = IPCID_TO_IX(msqid); 518 519 if (msqix < 0 || msqix >= msginfo.msgmni) { 520 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 521 msginfo.msgmni)); 522 return (EINVAL); 523 } 524 525 msqkptr = &msqids[msqix]; 526 527 mtx_lock(&msq_mtx); 528 if (msqkptr->u.msg_qbytes == 0) { 529 DPRINTF(("no such msqid\n")); 530 error = EINVAL; 531 goto done2; 532 } 533 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 534 DPRINTF(("wrong sequence number\n")); 535 error = EINVAL; 536 goto done2; 537 } 538 539 error = msq_prison_cansee(rpr, msqkptr); 540 if (error != 0) { 541 DPRINTF(("requester can't see prison\n")); 542 goto done2; 543 } 544 545 #ifdef MAC 546 error = mac_sysvmsq_check_msqctl(td->td_ucred, msqkptr, cmd); 547 if (error != 0) 548 goto done2; 549 #endif 550 551 error = 0; 552 rval = 0; 553 554 switch (cmd) { 555 556 case IPC_RMID: 557 { 558 #ifdef MAC 559 struct msg *msghdr; 560 #endif 561 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 562 goto done2; 563 564 #ifdef MAC 565 /* 566 * Check that the thread has MAC access permissions to 567 * individual msghdrs. Note: We need to do this in a 568 * separate loop because the actual loop alters the 569 * msq/msghdr info as it progresses, and there is no going 570 * back if half the way through we discover that the 571 * thread cannot free a certain msghdr. The msq will get 572 * into an inconsistent state. 573 */ 574 for (msghdr = msqkptr->u.msg_first; msghdr != NULL; 575 msghdr = msghdr->msg_next) { 576 error = mac_sysvmsq_check_msgrmid(td->td_ucred, msghdr); 577 if (error != 0) 578 goto done2; 579 } 580 #endif 581 582 msq_remove(msqkptr); 583 } 584 585 break; 586 587 case IPC_SET: 588 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_M))) 589 goto done2; 590 if (msqbuf->msg_qbytes > msqkptr->u.msg_qbytes) { 591 error = priv_check(td, PRIV_IPC_MSGSIZE); 592 if (error) 593 goto done2; 594 } 595 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 596 DPRINTF(("can't increase msg_qbytes beyond %d" 597 "(truncating)\n", msginfo.msgmnb)); 598 msqbuf->msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ 599 } 600 if (msqbuf->msg_qbytes == 0) { 601 DPRINTF(("can't reduce msg_qbytes to 0\n")); 602 error = EINVAL; /* non-standard errno! */ 603 goto done2; 604 } 605 msqkptr->u.msg_perm.uid = msqbuf->msg_perm.uid; /* change the owner */ 606 msqkptr->u.msg_perm.gid = msqbuf->msg_perm.gid; /* change the owner */ 607 msqkptr->u.msg_perm.mode = (msqkptr->u.msg_perm.mode & ~0777) | 608 (msqbuf->msg_perm.mode & 0777); 609 msqkptr->u.msg_qbytes = msqbuf->msg_qbytes; 610 msqkptr->u.msg_ctime = time_second; 611 break; 612 613 case IPC_STAT: 614 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 615 DPRINTF(("requester doesn't have read access\n")); 616 goto done2; 617 } 618 *msqbuf = msqkptr->u; 619 if (td->td_ucred->cr_prison != msqkptr->cred->cr_prison) 620 msqbuf->msg_perm.key = IPC_PRIVATE; 621 break; 622 623 default: 624 DPRINTF(("invalid command %d\n", cmd)); 625 error = EINVAL; 626 goto done2; 627 } 628 629 if (error == 0) 630 td->td_retval[0] = rval; 631 done2: 632 mtx_unlock(&msq_mtx); 633 return (error); 634 } 635 636 #ifndef _SYS_SYSPROTO_H_ 637 struct msgget_args { 638 key_t key; 639 int msgflg; 640 }; 641 #endif 642 643 int 644 sys_msgget(td, uap) 645 struct thread *td; 646 register struct msgget_args *uap; 647 { 648 int msqid, error = 0; 649 int key = uap->key; 650 int msgflg = uap->msgflg; 651 struct ucred *cred = td->td_ucred; 652 register struct msqid_kernel *msqkptr = NULL; 653 654 DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 655 656 if (msg_find_prison(cred) == NULL) 657 return (ENOSYS); 658 659 mtx_lock(&msq_mtx); 660 if (key != IPC_PRIVATE) { 661 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 662 msqkptr = &msqids[msqid]; 663 if (msqkptr->u.msg_qbytes != 0 && 664 msqkptr->cred != NULL && 665 msqkptr->cred->cr_prison == cred->cr_prison && 666 msqkptr->u.msg_perm.key == key) 667 break; 668 } 669 if (msqid < msginfo.msgmni) { 670 DPRINTF(("found public key\n")); 671 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 672 DPRINTF(("not exclusive\n")); 673 error = EEXIST; 674 goto done2; 675 } 676 if ((error = ipcperm(td, &msqkptr->u.msg_perm, 677 msgflg & 0700))) { 678 DPRINTF(("requester doesn't have 0%o access\n", 679 msgflg & 0700)); 680 goto done2; 681 } 682 #ifdef MAC 683 error = mac_sysvmsq_check_msqget(cred, msqkptr); 684 if (error != 0) 685 goto done2; 686 #endif 687 goto found; 688 } 689 } 690 691 DPRINTF(("need to allocate the msqid_ds\n")); 692 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 693 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 694 /* 695 * Look for an unallocated and unlocked msqid_ds. 696 * msqid_ds's can be locked by msgsnd or msgrcv while 697 * they are copying the message in/out. We can't 698 * re-use the entry until they release it. 699 */ 700 msqkptr = &msqids[msqid]; 701 if (msqkptr->u.msg_qbytes == 0 && 702 (msqkptr->u.msg_perm.mode & MSG_LOCKED) == 0) 703 break; 704 } 705 if (msqid == msginfo.msgmni) { 706 DPRINTF(("no more msqid_ds's available\n")); 707 error = ENOSPC; 708 goto done2; 709 } 710 #ifdef RACCT 711 if (racct_enable) { 712 PROC_LOCK(td->td_proc); 713 error = racct_add(td->td_proc, RACCT_NMSGQ, 1); 714 PROC_UNLOCK(td->td_proc); 715 if (error != 0) { 716 error = ENOSPC; 717 goto done2; 718 } 719 } 720 #endif 721 DPRINTF(("msqid %d is available\n", msqid)); 722 msqkptr->u.msg_perm.key = key; 723 msqkptr->u.msg_perm.cuid = cred->cr_uid; 724 msqkptr->u.msg_perm.uid = cred->cr_uid; 725 msqkptr->u.msg_perm.cgid = cred->cr_gid; 726 msqkptr->u.msg_perm.gid = cred->cr_gid; 727 msqkptr->u.msg_perm.mode = (msgflg & 0777); 728 msqkptr->cred = crhold(cred); 729 /* Make sure that the returned msqid is unique */ 730 msqkptr->u.msg_perm.seq = (msqkptr->u.msg_perm.seq + 1) & 0x7fff; 731 msqkptr->u.msg_first = NULL; 732 msqkptr->u.msg_last = NULL; 733 msqkptr->u.msg_cbytes = 0; 734 msqkptr->u.msg_qnum = 0; 735 msqkptr->u.msg_qbytes = msginfo.msgmnb; 736 msqkptr->u.msg_lspid = 0; 737 msqkptr->u.msg_lrpid = 0; 738 msqkptr->u.msg_stime = 0; 739 msqkptr->u.msg_rtime = 0; 740 msqkptr->u.msg_ctime = time_second; 741 #ifdef MAC 742 mac_sysvmsq_create(cred, msqkptr); 743 #endif 744 } else { 745 DPRINTF(("didn't find it and wasn't asked to create it\n")); 746 error = ENOENT; 747 goto done2; 748 } 749 750 found: 751 /* Construct the unique msqid */ 752 td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqkptr->u.msg_perm); 753 done2: 754 mtx_unlock(&msq_mtx); 755 return (error); 756 } 757 758 #ifndef _SYS_SYSPROTO_H_ 759 struct msgsnd_args { 760 int msqid; 761 const void *msgp; 762 size_t msgsz; 763 int msgflg; 764 }; 765 #endif 766 int 767 kern_msgsnd(td, msqid, msgp, msgsz, msgflg, mtype) 768 struct thread *td; 769 int msqid; 770 const void *msgp; /* XXX msgp is actually mtext. */ 771 size_t msgsz; 772 int msgflg; 773 long mtype; 774 { 775 int msqix, segs_needed, error = 0; 776 register struct msqid_kernel *msqkptr; 777 register struct msg *msghdr; 778 struct prison *rpr; 779 short next; 780 #ifdef RACCT 781 size_t saved_msgsz; 782 #endif 783 784 rpr = msg_find_prison(td->td_ucred); 785 if (rpr == NULL) 786 return (ENOSYS); 787 788 mtx_lock(&msq_mtx); 789 msqix = IPCID_TO_IX(msqid); 790 791 if (msqix < 0 || msqix >= msginfo.msgmni) { 792 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 793 msginfo.msgmni)); 794 error = EINVAL; 795 goto done2; 796 } 797 798 msqkptr = &msqids[msqix]; 799 if (msqkptr->u.msg_qbytes == 0) { 800 DPRINTF(("no such message queue id\n")); 801 error = EINVAL; 802 goto done2; 803 } 804 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 805 DPRINTF(("wrong sequence number\n")); 806 error = EINVAL; 807 goto done2; 808 } 809 810 if ((error = msq_prison_cansee(rpr, msqkptr))) { 811 DPRINTF(("requester can't see prison\n")); 812 goto done2; 813 } 814 815 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_W))) { 816 DPRINTF(("requester doesn't have write access\n")); 817 goto done2; 818 } 819 820 #ifdef MAC 821 error = mac_sysvmsq_check_msqsnd(td->td_ucred, msqkptr); 822 if (error != 0) 823 goto done2; 824 #endif 825 826 #ifdef RACCT 827 if (racct_enable) { 828 PROC_LOCK(td->td_proc); 829 if (racct_add(td->td_proc, RACCT_MSGQQUEUED, 1)) { 830 PROC_UNLOCK(td->td_proc); 831 error = EAGAIN; 832 goto done2; 833 } 834 saved_msgsz = msgsz; 835 if (racct_add(td->td_proc, RACCT_MSGQSIZE, msgsz)) { 836 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 837 PROC_UNLOCK(td->td_proc); 838 error = EAGAIN; 839 goto done2; 840 } 841 PROC_UNLOCK(td->td_proc); 842 } 843 #endif 844 845 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 846 DPRINTF(("msgsz=%zu, msgssz=%d, segs_needed=%d\n", msgsz, 847 msginfo.msgssz, segs_needed)); 848 for (;;) { 849 int need_more_resources = 0; 850 851 /* 852 * check msgsz 853 * (inside this loop in case msg_qbytes changes while we sleep) 854 */ 855 856 if (msgsz > msqkptr->u.msg_qbytes) { 857 DPRINTF(("msgsz > msqkptr->u.msg_qbytes\n")); 858 error = EINVAL; 859 goto done3; 860 } 861 862 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) { 863 DPRINTF(("msqid is locked\n")); 864 need_more_resources = 1; 865 } 866 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) { 867 DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 868 need_more_resources = 1; 869 } 870 if (segs_needed > nfree_msgmaps) { 871 DPRINTF(("segs_needed > nfree_msgmaps\n")); 872 need_more_resources = 1; 873 } 874 if (free_msghdrs == NULL) { 875 DPRINTF(("no more msghdrs\n")); 876 need_more_resources = 1; 877 } 878 879 if (need_more_resources) { 880 int we_own_it; 881 882 if ((msgflg & IPC_NOWAIT) != 0) { 883 DPRINTF(("need more resources but caller " 884 "doesn't want to wait\n")); 885 error = EAGAIN; 886 goto done3; 887 } 888 889 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) { 890 DPRINTF(("we don't own the msqid_ds\n")); 891 we_own_it = 0; 892 } else { 893 /* Force later arrivals to wait for our 894 request */ 895 DPRINTF(("we own the msqid_ds\n")); 896 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 897 we_own_it = 1; 898 } 899 DPRINTF(("msgsnd: goodnight\n")); 900 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 901 "msgsnd", hz); 902 DPRINTF(("msgsnd: good morning, error=%d\n", error)); 903 if (we_own_it) 904 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 905 if (error == EWOULDBLOCK) { 906 DPRINTF(("msgsnd: timed out\n")); 907 continue; 908 } 909 if (error != 0) { 910 DPRINTF(("msgsnd: interrupted system call\n")); 911 error = EINTR; 912 goto done3; 913 } 914 915 /* 916 * Make sure that the msq queue still exists 917 */ 918 919 if (msqkptr->u.msg_qbytes == 0) { 920 DPRINTF(("msqid deleted\n")); 921 error = EIDRM; 922 goto done3; 923 } 924 925 } else { 926 DPRINTF(("got all the resources that we need\n")); 927 break; 928 } 929 } 930 931 /* 932 * We have the resources that we need. 933 * Make sure! 934 */ 935 936 if (msqkptr->u.msg_perm.mode & MSG_LOCKED) 937 panic("msg_perm.mode & MSG_LOCKED"); 938 if (segs_needed > nfree_msgmaps) 939 panic("segs_needed > nfree_msgmaps"); 940 if (msgsz + msqkptr->u.msg_cbytes > msqkptr->u.msg_qbytes) 941 panic("msgsz + msg_cbytes > msg_qbytes"); 942 if (free_msghdrs == NULL) 943 panic("no more msghdrs"); 944 945 /* 946 * Re-lock the msqid_ds in case we page-fault when copying in the 947 * message 948 */ 949 950 if ((msqkptr->u.msg_perm.mode & MSG_LOCKED) != 0) 951 panic("msqid_ds is already locked"); 952 msqkptr->u.msg_perm.mode |= MSG_LOCKED; 953 954 /* 955 * Allocate a message header 956 */ 957 958 msghdr = free_msghdrs; 959 free_msghdrs = msghdr->msg_next; 960 msghdr->msg_spot = -1; 961 msghdr->msg_ts = msgsz; 962 msghdr->msg_type = mtype; 963 #ifdef MAC 964 /* 965 * XXXMAC: Should the mac_sysvmsq_check_msgmsq check follow here 966 * immediately? Or, should it be checked just before the msg is 967 * enqueued in the msgq (as it is done now)? 968 */ 969 mac_sysvmsg_create(td->td_ucred, msqkptr, msghdr); 970 #endif 971 972 /* 973 * Allocate space for the message 974 */ 975 976 while (segs_needed > 0) { 977 if (nfree_msgmaps <= 0) 978 panic("not enough msgmaps"); 979 if (free_msgmaps == -1) 980 panic("nil free_msgmaps"); 981 next = free_msgmaps; 982 if (next <= -1) 983 panic("next too low #1"); 984 if (next >= msginfo.msgseg) 985 panic("next out of range #1"); 986 DPRINTF(("allocating segment %d to message\n", next)); 987 free_msgmaps = msgmaps[next].next; 988 nfree_msgmaps--; 989 msgmaps[next].next = msghdr->msg_spot; 990 msghdr->msg_spot = next; 991 segs_needed--; 992 } 993 994 /* 995 * Validate the message type 996 */ 997 998 if (msghdr->msg_type < 1) { 999 msg_freehdr(msghdr); 1000 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1001 wakeup(msqkptr); 1002 DPRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 1003 error = EINVAL; 1004 goto done3; 1005 } 1006 1007 /* 1008 * Copy in the message body 1009 */ 1010 1011 next = msghdr->msg_spot; 1012 while (msgsz > 0) { 1013 size_t tlen; 1014 if (msgsz > msginfo.msgssz) 1015 tlen = msginfo.msgssz; 1016 else 1017 tlen = msgsz; 1018 if (next <= -1) 1019 panic("next too low #2"); 1020 if (next >= msginfo.msgseg) 1021 panic("next out of range #2"); 1022 mtx_unlock(&msq_mtx); 1023 if ((error = copyin(msgp, &msgpool[next * msginfo.msgssz], 1024 tlen)) != 0) { 1025 mtx_lock(&msq_mtx); 1026 DPRINTF(("error %d copying in message segment\n", 1027 error)); 1028 msg_freehdr(msghdr); 1029 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1030 wakeup(msqkptr); 1031 goto done3; 1032 } 1033 mtx_lock(&msq_mtx); 1034 msgsz -= tlen; 1035 msgp = (const char *)msgp + tlen; 1036 next = msgmaps[next].next; 1037 } 1038 if (next != -1) 1039 panic("didn't use all the msg segments"); 1040 1041 /* 1042 * We've got the message. Unlock the msqid_ds. 1043 */ 1044 1045 msqkptr->u.msg_perm.mode &= ~MSG_LOCKED; 1046 1047 /* 1048 * Make sure that the msqid_ds is still allocated. 1049 */ 1050 1051 if (msqkptr->u.msg_qbytes == 0) { 1052 msg_freehdr(msghdr); 1053 wakeup(msqkptr); 1054 error = EIDRM; 1055 goto done3; 1056 } 1057 1058 #ifdef MAC 1059 /* 1060 * Note: Since the task/thread allocates the msghdr and usually 1061 * primes it with its own MAC label, for a majority of policies, it 1062 * won't be necessary to check whether the msghdr has access 1063 * permissions to the msgq. The mac_sysvmsq_check_msqsnd check would 1064 * suffice in that case. However, this hook may be required where 1065 * individual policies derive a non-identical label for the msghdr 1066 * from the current thread label and may want to check the msghdr 1067 * enqueue permissions, along with read/write permissions to the 1068 * msgq. 1069 */ 1070 error = mac_sysvmsq_check_msgmsq(td->td_ucred, msghdr, msqkptr); 1071 if (error != 0) { 1072 msg_freehdr(msghdr); 1073 wakeup(msqkptr); 1074 goto done3; 1075 } 1076 #endif 1077 1078 /* 1079 * Put the message into the queue 1080 */ 1081 if (msqkptr->u.msg_first == NULL) { 1082 msqkptr->u.msg_first = msghdr; 1083 msqkptr->u.msg_last = msghdr; 1084 } else { 1085 msqkptr->u.msg_last->msg_next = msghdr; 1086 msqkptr->u.msg_last = msghdr; 1087 } 1088 msqkptr->u.msg_last->msg_next = NULL; 1089 1090 msqkptr->u.msg_cbytes += msghdr->msg_ts; 1091 msqkptr->u.msg_qnum++; 1092 msqkptr->u.msg_lspid = td->td_proc->p_pid; 1093 msqkptr->u.msg_stime = time_second; 1094 1095 wakeup(msqkptr); 1096 td->td_retval[0] = 0; 1097 done3: 1098 #ifdef RACCT 1099 if (racct_enable && error != 0) { 1100 PROC_LOCK(td->td_proc); 1101 racct_sub(td->td_proc, RACCT_MSGQQUEUED, 1); 1102 racct_sub(td->td_proc, RACCT_MSGQSIZE, saved_msgsz); 1103 PROC_UNLOCK(td->td_proc); 1104 } 1105 #endif 1106 done2: 1107 mtx_unlock(&msq_mtx); 1108 return (error); 1109 } 1110 1111 int 1112 sys_msgsnd(td, uap) 1113 struct thread *td; 1114 register struct msgsnd_args *uap; 1115 { 1116 int error; 1117 long mtype; 1118 1119 DPRINTF(("call to msgsnd(%d, %p, %zu, %d)\n", uap->msqid, uap->msgp, 1120 uap->msgsz, uap->msgflg)); 1121 1122 if ((error = copyin(uap->msgp, &mtype, sizeof(mtype))) != 0) { 1123 DPRINTF(("error %d copying the message type\n", error)); 1124 return (error); 1125 } 1126 return (kern_msgsnd(td, uap->msqid, 1127 (const char *)uap->msgp + sizeof(mtype), 1128 uap->msgsz, uap->msgflg, mtype)); 1129 } 1130 1131 #ifndef _SYS_SYSPROTO_H_ 1132 struct msgrcv_args { 1133 int msqid; 1134 void *msgp; 1135 size_t msgsz; 1136 long msgtyp; 1137 int msgflg; 1138 }; 1139 #endif 1140 int 1141 kern_msgrcv(td, msqid, msgp, msgsz, msgtyp, msgflg, mtype) 1142 struct thread *td; 1143 int msqid; 1144 void *msgp; /* XXX msgp is actually mtext. */ 1145 size_t msgsz; 1146 long msgtyp; 1147 int msgflg; 1148 long *mtype; 1149 { 1150 size_t len; 1151 register struct msqid_kernel *msqkptr; 1152 register struct msg *msghdr; 1153 struct prison *rpr; 1154 int msqix, error = 0; 1155 short next; 1156 1157 rpr = msg_find_prison(td->td_ucred); 1158 if (rpr == NULL) 1159 return (ENOSYS); 1160 1161 msqix = IPCID_TO_IX(msqid); 1162 1163 if (msqix < 0 || msqix >= msginfo.msgmni) { 1164 DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqix, 1165 msginfo.msgmni)); 1166 return (EINVAL); 1167 } 1168 1169 msqkptr = &msqids[msqix]; 1170 mtx_lock(&msq_mtx); 1171 if (msqkptr->u.msg_qbytes == 0) { 1172 DPRINTF(("no such message queue id\n")); 1173 error = EINVAL; 1174 goto done2; 1175 } 1176 if (msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1177 DPRINTF(("wrong sequence number\n")); 1178 error = EINVAL; 1179 goto done2; 1180 } 1181 1182 if ((error = msq_prison_cansee(rpr, msqkptr))) { 1183 DPRINTF(("requester can't see prison\n")); 1184 goto done2; 1185 } 1186 1187 if ((error = ipcperm(td, &msqkptr->u.msg_perm, IPC_R))) { 1188 DPRINTF(("requester doesn't have read access\n")); 1189 goto done2; 1190 } 1191 1192 #ifdef MAC 1193 error = mac_sysvmsq_check_msqrcv(td->td_ucred, msqkptr); 1194 if (error != 0) 1195 goto done2; 1196 #endif 1197 1198 msghdr = NULL; 1199 while (msghdr == NULL) { 1200 if (msgtyp == 0) { 1201 msghdr = msqkptr->u.msg_first; 1202 if (msghdr != NULL) { 1203 if (msgsz < msghdr->msg_ts && 1204 (msgflg & MSG_NOERROR) == 0) { 1205 DPRINTF(("first message on the queue " 1206 "is too big (want %zu, got %d)\n", 1207 msgsz, msghdr->msg_ts)); 1208 error = E2BIG; 1209 goto done2; 1210 } 1211 #ifdef MAC 1212 error = mac_sysvmsq_check_msgrcv(td->td_ucred, 1213 msghdr); 1214 if (error != 0) 1215 goto done2; 1216 #endif 1217 if (msqkptr->u.msg_first == msqkptr->u.msg_last) { 1218 msqkptr->u.msg_first = NULL; 1219 msqkptr->u.msg_last = NULL; 1220 } else { 1221 msqkptr->u.msg_first = msghdr->msg_next; 1222 if (msqkptr->u.msg_first == NULL) 1223 panic("msg_first/last screwed up #1"); 1224 } 1225 } 1226 } else { 1227 struct msg *previous; 1228 struct msg **prev; 1229 1230 previous = NULL; 1231 prev = &(msqkptr->u.msg_first); 1232 while ((msghdr = *prev) != NULL) { 1233 /* 1234 * Is this message's type an exact match or is 1235 * this message's type less than or equal to 1236 * the absolute value of a negative msgtyp? 1237 * Note that the second half of this test can 1238 * NEVER be true if msgtyp is positive since 1239 * msg_type is always positive! 1240 */ 1241 1242 if (msgtyp == msghdr->msg_type || 1243 msghdr->msg_type <= -msgtyp) { 1244 DPRINTF(("found message type %ld, " 1245 "requested %ld\n", 1246 msghdr->msg_type, msgtyp)); 1247 if (msgsz < msghdr->msg_ts && 1248 (msgflg & MSG_NOERROR) == 0) { 1249 DPRINTF(("requested message " 1250 "on the queue is too big " 1251 "(want %zu, got %hu)\n", 1252 msgsz, msghdr->msg_ts)); 1253 error = E2BIG; 1254 goto done2; 1255 } 1256 #ifdef MAC 1257 error = mac_sysvmsq_check_msgrcv( 1258 td->td_ucred, msghdr); 1259 if (error != 0) 1260 goto done2; 1261 #endif 1262 *prev = msghdr->msg_next; 1263 if (msghdr == msqkptr->u.msg_last) { 1264 if (previous == NULL) { 1265 if (prev != 1266 &msqkptr->u.msg_first) 1267 panic("msg_first/last screwed up #2"); 1268 msqkptr->u.msg_first = 1269 NULL; 1270 msqkptr->u.msg_last = 1271 NULL; 1272 } else { 1273 if (prev == 1274 &msqkptr->u.msg_first) 1275 panic("msg_first/last screwed up #3"); 1276 msqkptr->u.msg_last = 1277 previous; 1278 } 1279 } 1280 break; 1281 } 1282 previous = msghdr; 1283 prev = &(msghdr->msg_next); 1284 } 1285 } 1286 1287 /* 1288 * We've either extracted the msghdr for the appropriate 1289 * message or there isn't one. 1290 * If there is one then bail out of this loop. 1291 */ 1292 1293 if (msghdr != NULL) 1294 break; 1295 1296 /* 1297 * Hmph! No message found. Does the user want to wait? 1298 */ 1299 1300 if ((msgflg & IPC_NOWAIT) != 0) { 1301 DPRINTF(("no appropriate message found (msgtyp=%ld)\n", 1302 msgtyp)); 1303 /* The SVID says to return ENOMSG. */ 1304 error = ENOMSG; 1305 goto done2; 1306 } 1307 1308 /* 1309 * Wait for something to happen 1310 */ 1311 1312 DPRINTF(("msgrcv: goodnight\n")); 1313 error = msleep(msqkptr, &msq_mtx, (PZERO - 4) | PCATCH, 1314 "msgrcv", 0); 1315 DPRINTF(("msgrcv: good morning (error=%d)\n", error)); 1316 1317 if (error != 0) { 1318 DPRINTF(("msgrcv: interrupted system call\n")); 1319 error = EINTR; 1320 goto done2; 1321 } 1322 1323 /* 1324 * Make sure that the msq queue still exists 1325 */ 1326 1327 if (msqkptr->u.msg_qbytes == 0 || 1328 msqkptr->u.msg_perm.seq != IPCID_TO_SEQ(msqid)) { 1329 DPRINTF(("msqid deleted\n")); 1330 error = EIDRM; 1331 goto done2; 1332 } 1333 } 1334 1335 /* 1336 * Return the message to the user. 1337 * 1338 * First, do the bookkeeping (before we risk being interrupted). 1339 */ 1340 1341 msqkptr->u.msg_cbytes -= msghdr->msg_ts; 1342 msqkptr->u.msg_qnum--; 1343 msqkptr->u.msg_lrpid = td->td_proc->p_pid; 1344 msqkptr->u.msg_rtime = time_second; 1345 1346 racct_sub_cred(msqkptr->cred, RACCT_MSGQQUEUED, 1); 1347 racct_sub_cred(msqkptr->cred, RACCT_MSGQSIZE, msghdr->msg_ts); 1348 1349 /* 1350 * Make msgsz the actual amount that we'll be returning. 1351 * Note that this effectively truncates the message if it is too long 1352 * (since msgsz is never increased). 1353 */ 1354 1355 DPRINTF(("found a message, msgsz=%zu, msg_ts=%hu\n", msgsz, 1356 msghdr->msg_ts)); 1357 if (msgsz > msghdr->msg_ts) 1358 msgsz = msghdr->msg_ts; 1359 *mtype = msghdr->msg_type; 1360 1361 /* 1362 * Return the segments to the user 1363 */ 1364 1365 next = msghdr->msg_spot; 1366 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1367 size_t tlen; 1368 1369 if (msgsz - len > msginfo.msgssz) 1370 tlen = msginfo.msgssz; 1371 else 1372 tlen = msgsz - len; 1373 if (next <= -1) 1374 panic("next too low #3"); 1375 if (next >= msginfo.msgseg) 1376 panic("next out of range #3"); 1377 mtx_unlock(&msq_mtx); 1378 error = copyout(&msgpool[next * msginfo.msgssz], msgp, tlen); 1379 mtx_lock(&msq_mtx); 1380 if (error != 0) { 1381 DPRINTF(("error (%d) copying out message segment\n", 1382 error)); 1383 msg_freehdr(msghdr); 1384 wakeup(msqkptr); 1385 goto done2; 1386 } 1387 msgp = (char *)msgp + tlen; 1388 next = msgmaps[next].next; 1389 } 1390 1391 /* 1392 * Done, return the actual number of bytes copied out. 1393 */ 1394 1395 msg_freehdr(msghdr); 1396 wakeup(msqkptr); 1397 td->td_retval[0] = msgsz; 1398 done2: 1399 mtx_unlock(&msq_mtx); 1400 return (error); 1401 } 1402 1403 int 1404 sys_msgrcv(td, uap) 1405 struct thread *td; 1406 register struct msgrcv_args *uap; 1407 { 1408 int error; 1409 long mtype; 1410 1411 DPRINTF(("call to msgrcv(%d, %p, %zu, %ld, %d)\n", uap->msqid, 1412 uap->msgp, uap->msgsz, uap->msgtyp, uap->msgflg)); 1413 1414 if ((error = kern_msgrcv(td, uap->msqid, 1415 (char *)uap->msgp + sizeof(mtype), uap->msgsz, 1416 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1417 return (error); 1418 if ((error = copyout(&mtype, uap->msgp, sizeof(mtype))) != 0) 1419 DPRINTF(("error %d copying the message type\n", error)); 1420 return (error); 1421 } 1422 1423 static int 1424 sysctl_msqids(SYSCTL_HANDLER_ARGS) 1425 { 1426 struct sbuf sb; 1427 struct msqid_kernel tmp, empty; 1428 struct msqid_kernel *msqkptr; 1429 struct prison *rpr; 1430 int error, i; 1431 1432 error = sysctl_wire_old_buffer(req, 0); 1433 if (error != 0) 1434 goto done; 1435 rpr = msg_find_prison(req->td->td_ucred); 1436 sbuf_new_for_sysctl(&sb, NULL, sizeof(struct msqid_kernel) * 1437 msginfo.msgmni, req); 1438 1439 bzero(&empty, sizeof(empty)); 1440 for (i = 0; i < msginfo.msgmni; i++) { 1441 msqkptr = &msqids[i]; 1442 if (msqkptr->u.msg_qbytes == 0 || rpr == NULL || 1443 msq_prison_cansee(rpr, msqkptr) != 0) { 1444 msqkptr = ∅ 1445 } else if (req->td->td_ucred->cr_prison != 1446 msqkptr->cred->cr_prison) { 1447 bcopy(msqkptr, &tmp, sizeof(tmp)); 1448 msqkptr = &tmp; 1449 msqkptr->u.msg_perm.key = IPC_PRIVATE; 1450 } 1451 1452 sbuf_bcat(&sb, msqkptr, sizeof(*msqkptr)); 1453 } 1454 error = sbuf_finish(&sb); 1455 sbuf_delete(&sb); 1456 1457 done: 1458 return (error); 1459 } 1460 1461 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, 1462 "Maximum message size"); 1463 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, 1464 "Number of message queue identifiers"); 1465 SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RDTUN, &msginfo.msgmnb, 0, 1466 "Maximum number of bytes in a queue"); 1467 SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RDTUN, &msginfo.msgtql, 0, 1468 "Maximum number of messages in the system"); 1469 SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, 1470 "Size of a message segment"); 1471 SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, 1472 "Number of message segments"); 1473 SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLTYPE_OPAQUE | CTLFLAG_RD, 1474 NULL, 0, sysctl_msqids, "", "Message queue IDs"); 1475 1476 static int 1477 msg_prison_check(void *obj, void *data) 1478 { 1479 struct prison *pr = obj; 1480 struct prison *prpr; 1481 struct vfsoptlist *opts = data; 1482 int error, jsys; 1483 1484 /* 1485 * sysvmsg is a jailsys integer. 1486 * It must be "disable" if the parent jail is disabled. 1487 */ 1488 error = vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1489 if (error != ENOENT) { 1490 if (error != 0) 1491 return (error); 1492 switch (jsys) { 1493 case JAIL_SYS_DISABLE: 1494 break; 1495 case JAIL_SYS_NEW: 1496 case JAIL_SYS_INHERIT: 1497 prison_lock(pr->pr_parent); 1498 prpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1499 prison_unlock(pr->pr_parent); 1500 if (prpr == NULL) 1501 return (EPERM); 1502 break; 1503 default: 1504 return (EINVAL); 1505 } 1506 } 1507 1508 return (0); 1509 } 1510 1511 static int 1512 msg_prison_set(void *obj, void *data) 1513 { 1514 struct prison *pr = obj; 1515 struct prison *tpr, *orpr, *nrpr, *trpr; 1516 struct vfsoptlist *opts = data; 1517 void *rsv; 1518 int jsys, descend; 1519 1520 /* 1521 * sysvmsg controls which jail is the root of the associated msgs (this 1522 * jail or same as the parent), or if the feature is available at all. 1523 */ 1524 if (vfs_copyopt(opts, "sysvmsg", &jsys, sizeof(jsys)) == ENOENT) 1525 jsys = vfs_flagopt(opts, "allow.sysvipc", NULL, 0) 1526 ? JAIL_SYS_INHERIT 1527 : vfs_flagopt(opts, "allow.nosysvipc", NULL, 0) 1528 ? JAIL_SYS_DISABLE 1529 : -1; 1530 if (jsys == JAIL_SYS_DISABLE) { 1531 prison_lock(pr); 1532 orpr = osd_jail_get(pr, msg_prison_slot); 1533 if (orpr != NULL) 1534 osd_jail_del(pr, msg_prison_slot); 1535 prison_unlock(pr); 1536 if (orpr != NULL) { 1537 if (orpr == pr) 1538 msg_prison_cleanup(pr); 1539 /* Disable all child jails as well. */ 1540 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1541 prison_lock(tpr); 1542 trpr = osd_jail_get(tpr, msg_prison_slot); 1543 if (trpr != NULL) { 1544 osd_jail_del(tpr, msg_prison_slot); 1545 prison_unlock(tpr); 1546 if (trpr == tpr) 1547 msg_prison_cleanup(tpr); 1548 } else { 1549 prison_unlock(tpr); 1550 descend = 0; 1551 } 1552 } 1553 } 1554 } else if (jsys != -1) { 1555 if (jsys == JAIL_SYS_NEW) 1556 nrpr = pr; 1557 else { 1558 prison_lock(pr->pr_parent); 1559 nrpr = osd_jail_get(pr->pr_parent, msg_prison_slot); 1560 prison_unlock(pr->pr_parent); 1561 } 1562 rsv = osd_reserve(msg_prison_slot); 1563 prison_lock(pr); 1564 orpr = osd_jail_get(pr, msg_prison_slot); 1565 if (orpr != nrpr) 1566 (void)osd_jail_set_reserved(pr, msg_prison_slot, rsv, 1567 nrpr); 1568 else 1569 osd_free_reserved(rsv); 1570 prison_unlock(pr); 1571 if (orpr != nrpr) { 1572 if (orpr == pr) 1573 msg_prison_cleanup(pr); 1574 if (orpr != NULL) { 1575 /* Change child jails matching the old root, */ 1576 FOREACH_PRISON_DESCENDANT(pr, tpr, descend) { 1577 prison_lock(tpr); 1578 trpr = osd_jail_get(tpr, 1579 msg_prison_slot); 1580 if (trpr == orpr) { 1581 (void)osd_jail_set(tpr, 1582 msg_prison_slot, nrpr); 1583 prison_unlock(tpr); 1584 if (trpr == tpr) 1585 msg_prison_cleanup(tpr); 1586 } else { 1587 prison_unlock(tpr); 1588 descend = 0; 1589 } 1590 } 1591 } 1592 } 1593 } 1594 1595 return (0); 1596 } 1597 1598 static int 1599 msg_prison_get(void *obj, void *data) 1600 { 1601 struct prison *pr = obj; 1602 struct prison *rpr; 1603 struct vfsoptlist *opts = data; 1604 int error, jsys; 1605 1606 /* Set sysvmsg based on the jail's root prison. */ 1607 prison_lock(pr); 1608 rpr = osd_jail_get(pr, msg_prison_slot); 1609 prison_unlock(pr); 1610 jsys = rpr == NULL ? JAIL_SYS_DISABLE 1611 : rpr == pr ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; 1612 error = vfs_setopt(opts, "sysvmsg", &jsys, sizeof(jsys)); 1613 if (error == ENOENT) 1614 error = 0; 1615 return (error); 1616 } 1617 1618 static int 1619 msg_prison_remove(void *obj, void *data __unused) 1620 { 1621 struct prison *pr = obj; 1622 struct prison *rpr; 1623 1624 prison_lock(pr); 1625 rpr = osd_jail_get(pr, msg_prison_slot); 1626 prison_unlock(pr); 1627 if (rpr == pr) 1628 msg_prison_cleanup(pr); 1629 return (0); 1630 } 1631 1632 static void 1633 msg_prison_cleanup(struct prison *pr) 1634 { 1635 struct msqid_kernel *msqkptr; 1636 int i; 1637 1638 /* Remove any msqs that belong to this jail. */ 1639 mtx_lock(&msq_mtx); 1640 for (i = 0; i < msginfo.msgmni; i++) { 1641 msqkptr = &msqids[i]; 1642 if (msqkptr->u.msg_qbytes != 0 && 1643 msqkptr->cred != NULL && msqkptr->cred->cr_prison == pr) 1644 msq_remove(msqkptr); 1645 } 1646 mtx_unlock(&msq_mtx); 1647 } 1648 1649 SYSCTL_JAIL_PARAM_SYS_NODE(sysvmsg, CTLFLAG_RW, "SYSV message queues"); 1650 1651 #ifdef COMPAT_FREEBSD32 1652 int 1653 freebsd32_msgsys(struct thread *td, struct freebsd32_msgsys_args *uap) 1654 { 1655 1656 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1657 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1658 switch (uap->which) { 1659 case 0: 1660 return (freebsd7_freebsd32_msgctl(td, 1661 (struct freebsd7_freebsd32_msgctl_args *)&uap->a2)); 1662 case 2: 1663 return (freebsd32_msgsnd(td, 1664 (struct freebsd32_msgsnd_args *)&uap->a2)); 1665 case 3: 1666 return (freebsd32_msgrcv(td, 1667 (struct freebsd32_msgrcv_args *)&uap->a2)); 1668 default: 1669 return (sys_msgsys(td, (struct msgsys_args *)uap)); 1670 } 1671 #else 1672 return (nosys(td, NULL)); 1673 #endif 1674 } 1675 1676 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1677 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1678 int 1679 freebsd7_freebsd32_msgctl(struct thread *td, 1680 struct freebsd7_freebsd32_msgctl_args *uap) 1681 { 1682 struct msqid_ds msqbuf; 1683 struct msqid_ds32_old msqbuf32; 1684 int error; 1685 1686 if (uap->cmd == IPC_SET) { 1687 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1688 if (error) 1689 return (error); 1690 freebsd32_ipcperm_old_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1691 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1692 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1693 CP(msqbuf32, msqbuf, msg_cbytes); 1694 CP(msqbuf32, msqbuf, msg_qnum); 1695 CP(msqbuf32, msqbuf, msg_qbytes); 1696 CP(msqbuf32, msqbuf, msg_lspid); 1697 CP(msqbuf32, msqbuf, msg_lrpid); 1698 CP(msqbuf32, msqbuf, msg_stime); 1699 CP(msqbuf32, msqbuf, msg_rtime); 1700 CP(msqbuf32, msqbuf, msg_ctime); 1701 } 1702 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1703 if (error) 1704 return (error); 1705 if (uap->cmd == IPC_STAT) { 1706 bzero(&msqbuf32, sizeof(msqbuf32)); 1707 freebsd32_ipcperm_old_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1708 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1709 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1710 CP(msqbuf, msqbuf32, msg_cbytes); 1711 CP(msqbuf, msqbuf32, msg_qnum); 1712 CP(msqbuf, msqbuf32, msg_qbytes); 1713 CP(msqbuf, msqbuf32, msg_lspid); 1714 CP(msqbuf, msqbuf32, msg_lrpid); 1715 CP(msqbuf, msqbuf32, msg_stime); 1716 CP(msqbuf, msqbuf32, msg_rtime); 1717 CP(msqbuf, msqbuf32, msg_ctime); 1718 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1719 } 1720 return (error); 1721 } 1722 #endif 1723 1724 int 1725 freebsd32_msgctl(struct thread *td, struct freebsd32_msgctl_args *uap) 1726 { 1727 struct msqid_ds msqbuf; 1728 struct msqid_ds32 msqbuf32; 1729 int error; 1730 1731 if (uap->cmd == IPC_SET) { 1732 error = copyin(uap->buf, &msqbuf32, sizeof(msqbuf32)); 1733 if (error) 1734 return (error); 1735 freebsd32_ipcperm_in(&msqbuf32.msg_perm, &msqbuf.msg_perm); 1736 PTRIN_CP(msqbuf32, msqbuf, msg_first); 1737 PTRIN_CP(msqbuf32, msqbuf, msg_last); 1738 CP(msqbuf32, msqbuf, msg_cbytes); 1739 CP(msqbuf32, msqbuf, msg_qnum); 1740 CP(msqbuf32, msqbuf, msg_qbytes); 1741 CP(msqbuf32, msqbuf, msg_lspid); 1742 CP(msqbuf32, msqbuf, msg_lrpid); 1743 CP(msqbuf32, msqbuf, msg_stime); 1744 CP(msqbuf32, msqbuf, msg_rtime); 1745 CP(msqbuf32, msqbuf, msg_ctime); 1746 } 1747 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1748 if (error) 1749 return (error); 1750 if (uap->cmd == IPC_STAT) { 1751 freebsd32_ipcperm_out(&msqbuf.msg_perm, &msqbuf32.msg_perm); 1752 PTROUT_CP(msqbuf, msqbuf32, msg_first); 1753 PTROUT_CP(msqbuf, msqbuf32, msg_last); 1754 CP(msqbuf, msqbuf32, msg_cbytes); 1755 CP(msqbuf, msqbuf32, msg_qnum); 1756 CP(msqbuf, msqbuf32, msg_qbytes); 1757 CP(msqbuf, msqbuf32, msg_lspid); 1758 CP(msqbuf, msqbuf32, msg_lrpid); 1759 CP(msqbuf, msqbuf32, msg_stime); 1760 CP(msqbuf, msqbuf32, msg_rtime); 1761 CP(msqbuf, msqbuf32, msg_ctime); 1762 error = copyout(&msqbuf32, uap->buf, sizeof(struct msqid_ds32)); 1763 } 1764 return (error); 1765 } 1766 1767 int 1768 freebsd32_msgsnd(struct thread *td, struct freebsd32_msgsnd_args *uap) 1769 { 1770 const void *msgp; 1771 long mtype; 1772 int32_t mtype32; 1773 int error; 1774 1775 msgp = PTRIN(uap->msgp); 1776 if ((error = copyin(msgp, &mtype32, sizeof(mtype32))) != 0) 1777 return (error); 1778 mtype = mtype32; 1779 return (kern_msgsnd(td, uap->msqid, 1780 (const char *)msgp + sizeof(mtype32), 1781 uap->msgsz, uap->msgflg, mtype)); 1782 } 1783 1784 int 1785 freebsd32_msgrcv(struct thread *td, struct freebsd32_msgrcv_args *uap) 1786 { 1787 void *msgp; 1788 long mtype; 1789 int32_t mtype32; 1790 int error; 1791 1792 msgp = PTRIN(uap->msgp); 1793 if ((error = kern_msgrcv(td, uap->msqid, 1794 (char *)msgp + sizeof(mtype32), uap->msgsz, 1795 uap->msgtyp, uap->msgflg, &mtype)) != 0) 1796 return (error); 1797 mtype32 = (int32_t)mtype; 1798 return (copyout(&mtype32, msgp, sizeof(mtype32))); 1799 } 1800 #endif 1801 1802 #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ 1803 defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) 1804 1805 /* XXX casting to (sy_call_t *) is bogus, as usual. */ 1806 static sy_call_t *msgcalls[] = { 1807 (sy_call_t *)freebsd7_msgctl, (sy_call_t *)sys_msgget, 1808 (sy_call_t *)sys_msgsnd, (sy_call_t *)sys_msgrcv 1809 }; 1810 1811 /* 1812 * Entry point for all MSG calls. 1813 */ 1814 int 1815 sys_msgsys(td, uap) 1816 struct thread *td; 1817 /* XXX actually varargs. */ 1818 struct msgsys_args /* { 1819 int which; 1820 int a2; 1821 int a3; 1822 int a4; 1823 int a5; 1824 int a6; 1825 } */ *uap; 1826 { 1827 int error; 1828 1829 if (uap->which < 0 || uap->which >= nitems(msgcalls)) 1830 return (EINVAL); 1831 error = (*msgcalls[uap->which])(td, &uap->a2); 1832 return (error); 1833 } 1834 1835 #ifndef CP 1836 #define CP(src, dst, fld) do { (dst).fld = (src).fld; } while (0) 1837 #endif 1838 1839 #ifndef _SYS_SYSPROTO_H_ 1840 struct freebsd7_msgctl_args { 1841 int msqid; 1842 int cmd; 1843 struct msqid_ds_old *buf; 1844 }; 1845 #endif 1846 int 1847 freebsd7_msgctl(td, uap) 1848 struct thread *td; 1849 struct freebsd7_msgctl_args *uap; 1850 { 1851 struct msqid_ds_old msqold; 1852 struct msqid_ds msqbuf; 1853 int error; 1854 1855 DPRINTF(("call to freebsd7_msgctl(%d, %d, %p)\n", uap->msqid, uap->cmd, 1856 uap->buf)); 1857 if (uap->cmd == IPC_SET) { 1858 error = copyin(uap->buf, &msqold, sizeof(msqold)); 1859 if (error) 1860 return (error); 1861 ipcperm_old2new(&msqold.msg_perm, &msqbuf.msg_perm); 1862 CP(msqold, msqbuf, msg_first); 1863 CP(msqold, msqbuf, msg_last); 1864 CP(msqold, msqbuf, msg_cbytes); 1865 CP(msqold, msqbuf, msg_qnum); 1866 CP(msqold, msqbuf, msg_qbytes); 1867 CP(msqold, msqbuf, msg_lspid); 1868 CP(msqold, msqbuf, msg_lrpid); 1869 CP(msqold, msqbuf, msg_stime); 1870 CP(msqold, msqbuf, msg_rtime); 1871 CP(msqold, msqbuf, msg_ctime); 1872 } 1873 error = kern_msgctl(td, uap->msqid, uap->cmd, &msqbuf); 1874 if (error) 1875 return (error); 1876 if (uap->cmd == IPC_STAT) { 1877 bzero(&msqold, sizeof(msqold)); 1878 ipcperm_new2old(&msqbuf.msg_perm, &msqold.msg_perm); 1879 CP(msqbuf, msqold, msg_first); 1880 CP(msqbuf, msqold, msg_last); 1881 CP(msqbuf, msqold, msg_cbytes); 1882 CP(msqbuf, msqold, msg_qnum); 1883 CP(msqbuf, msqold, msg_qbytes); 1884 CP(msqbuf, msqold, msg_lspid); 1885 CP(msqbuf, msqold, msg_lrpid); 1886 CP(msqbuf, msqold, msg_stime); 1887 CP(msqbuf, msqold, msg_rtime); 1888 CP(msqbuf, msqold, msg_ctime); 1889 error = copyout(&msqold, uap->buf, sizeof(struct msqid_ds_old)); 1890 } 1891 return (error); 1892 } 1893 1894 #undef CP 1895 1896 #endif /* COMPAT_FREEBSD4 || COMPAT_FREEBSD5 || COMPAT_FREEBSD6 || 1897 COMPAT_FREEBSD7 */ 1898