1 /* $NetBSD: sysv_msg.c,v 1.59 2009/01/11 02:45:53 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2006, 2007 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Implementation of SVID messages 35 * 36 * Author: Daniel Boulet 37 * 38 * Copyright 1993 Daniel Boulet and RTMX Inc. 39 * 40 * This system call was implemented by Daniel Boulet under contract from RTMX. 41 * 42 * Redistribution and use in source forms, with and without modification, 43 * are permitted provided that this entire comment appears intact. 44 * 45 * Redistribution in binary form may occur without any restrictions. 46 * Obviously, it would be nice if you gave credit where credit is due 47 * but requiring it would be too onerous. 48 * 49 * This software is provided ``AS IS'' without any warranties of any kind. 50 */ 51 52 #include <sys/cdefs.h> 53 __KERNEL_RCSID(0, "$NetBSD: sysv_msg.c,v 1.59 2009/01/11 02:45:53 christos Exp $"); 54 55 #define SYSVMSG 56 57 #include <sys/param.h> 58 #include <sys/kernel.h> 59 #include <sys/msg.h> 60 #include <sys/sysctl.h> 61 #include <sys/mount.h> /* XXX for <sys/syscallargs.h> */ 62 #include <sys/syscallargs.h> 63 #include <sys/kauth.h> 64 65 #define MSG_DEBUG 66 #undef MSG_DEBUG_OK 67 68 #ifdef MSG_DEBUG_OK 69 #define MSG_PRINTF(a) printf a 70 #else 71 #define MSG_PRINTF(a) 72 #endif 73 74 static int nfree_msgmaps; /* # of free map entries */ 75 static short free_msgmaps; /* head of linked list of free map entries */ 76 static struct __msg *free_msghdrs; /* list of free msg headers */ 77 static char *msgpool; /* MSGMAX byte long msg buffer pool */ 78 static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ 79 static struct __msg *msghdrs; /* MSGTQL msg headers */ 80 81 kmsq_t *msqs; /* MSGMNI msqid_ds struct's */ 82 kmutex_t msgmutex; /* subsystem lock */ 83 84 static u_int msg_waiters = 0; /* total number of msgrcv waiters */ 85 static bool msg_realloc_state; 86 static kcondvar_t msg_realloc_cv; 87 88 static void msg_freehdr(struct __msg *); 89 90 void 91 msginit(void) 92 { 93 int i, sz; 94 vaddr_t v; 95 96 /* 97 * msginfo.msgssz should be a power of two for efficiency reasons. 98 * It is also pretty silly if msginfo.msgssz is less than 8 99 * or greater than about 256 so ... 100 */ 101 102 i = 8; 103 while (i < 1024 && i != msginfo.msgssz) 104 i <<= 1; 105 if (i != msginfo.msgssz) { 106 panic("msginfo.msgssz = %d, not a small power of 2", 107 msginfo.msgssz); 108 } 109 110 if (msginfo.msgseg > 32767) { 111 panic("msginfo.msgseg = %d > 32767", msginfo.msgseg); 112 } 113 114 /* Allocate the wired memory for our structures */ 115 sz = ALIGN(msginfo.msgmax) + 116 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 117 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 118 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 119 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 120 UVM_KMF_WIRED|UVM_KMF_ZERO); 121 if (v == 0) 122 panic("sysv_msg: cannot allocate memory"); 123 msgpool = (void *)v; 124 msgmaps = (void *)((uintptr_t)msgpool + ALIGN(msginfo.msgmax)); 125 msghdrs = (void *)((uintptr_t)msgmaps + 126 ALIGN(msginfo.msgseg * sizeof(struct msgmap))); 127 msqs = (void *)((uintptr_t)msghdrs + 128 ALIGN(msginfo.msgtql * sizeof(struct __msg))); 129 130 for (i = 0; i < (msginfo.msgseg - 1); i++) 131 msgmaps[i].next = i + 1; 132 msgmaps[msginfo.msgseg - 1].next = -1; 133 134 free_msgmaps = 0; 135 nfree_msgmaps = msginfo.msgseg; 136 137 for (i = 0; i < (msginfo.msgtql - 1); i++) { 138 msghdrs[i].msg_type = 0; 139 msghdrs[i].msg_next = &msghdrs[i + 1]; 140 } 141 i = msginfo.msgtql - 1; 142 msghdrs[i].msg_type = 0; 143 msghdrs[i].msg_next = NULL; 144 free_msghdrs = &msghdrs[0]; 145 146 for (i = 0; i < msginfo.msgmni; i++) { 147 cv_init(&msqs[i].msq_cv, "msgwait"); 148 /* Implies entry is available */ 149 msqs[i].msq_u.msg_qbytes = 0; 150 /* Reset to a known value */ 151 msqs[i].msq_u.msg_perm._seq = 0; 152 } 153 154 mutex_init(&msgmutex, MUTEX_DEFAULT, IPL_NONE); 155 cv_init(&msg_realloc_cv, "msgrealc"); 156 msg_realloc_state = false; 157 } 158 159 static int 160 msgrealloc(int newmsgmni, int newmsgseg) 161 { 162 struct msgmap *new_msgmaps; 163 struct __msg *new_msghdrs, *new_free_msghdrs; 164 char *old_msgpool, *new_msgpool; 165 kmsq_t *new_msqs; 166 vaddr_t v; 167 int i, sz, msqid, newmsgmax, new_nfree_msgmaps; 168 short new_free_msgmaps; 169 170 if (newmsgmni < 1 || newmsgseg < 1) 171 return EINVAL; 172 173 /* Allocate the wired memory for our structures */ 174 newmsgmax = msginfo.msgssz * newmsgseg; 175 sz = ALIGN(newmsgmax) + 176 ALIGN(newmsgseg * sizeof(struct msgmap)) + 177 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 178 ALIGN(newmsgmni * sizeof(kmsq_t)); 179 v = uvm_km_alloc(kernel_map, round_page(sz), 0, 180 UVM_KMF_WIRED|UVM_KMF_ZERO); 181 if (v == 0) 182 return ENOMEM; 183 184 mutex_enter(&msgmutex); 185 if (msg_realloc_state) { 186 mutex_exit(&msgmutex); 187 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 188 return EBUSY; 189 } 190 msg_realloc_state = true; 191 if (msg_waiters) { 192 /* 193 * Mark reallocation state, wake-up all waiters, 194 * and wait while they will all exit. 195 */ 196 for (i = 0; i < msginfo.msgmni; i++) 197 cv_broadcast(&msqs[i].msq_cv); 198 while (msg_waiters) 199 cv_wait(&msg_realloc_cv, &msgmutex); 200 } 201 old_msgpool = msgpool; 202 203 /* We cannot reallocate less memory than we use */ 204 i = 0; 205 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 206 struct msqid_ds *mptr; 207 kmsq_t *msq; 208 209 msq = &msqs[msqid]; 210 mptr = &msq->msq_u; 211 if (mptr->msg_qbytes || (mptr->msg_perm.mode & MSG_LOCKED)) 212 i = msqid; 213 } 214 if (i >= newmsgmni || (msginfo.msgseg - nfree_msgmaps) > newmsgseg) { 215 mutex_exit(&msgmutex); 216 uvm_km_free(kernel_map, v, sz, UVM_KMF_WIRED); 217 return EBUSY; 218 } 219 220 new_msgpool = (void *)v; 221 new_msgmaps = (void *)((uintptr_t)new_msgpool + ALIGN(newmsgmax)); 222 new_msghdrs = (void *)((uintptr_t)new_msgmaps + 223 ALIGN(newmsgseg * sizeof(struct msgmap))); 224 new_msqs = (void *)((uintptr_t)new_msghdrs + 225 ALIGN(msginfo.msgtql * sizeof(struct __msg))); 226 227 /* Initialize the structures */ 228 for (i = 0; i < (newmsgseg - 1); i++) 229 new_msgmaps[i].next = i + 1; 230 new_msgmaps[newmsgseg - 1].next = -1; 231 new_free_msgmaps = 0; 232 new_nfree_msgmaps = newmsgseg; 233 234 for (i = 0; i < (msginfo.msgtql - 1); i++) { 235 new_msghdrs[i].msg_type = 0; 236 new_msghdrs[i].msg_next = &new_msghdrs[i + 1]; 237 } 238 i = msginfo.msgtql - 1; 239 new_msghdrs[i].msg_type = 0; 240 new_msghdrs[i].msg_next = NULL; 241 new_free_msghdrs = &new_msghdrs[0]; 242 243 for (i = 0; i < newmsgmni; i++) { 244 new_msqs[i].msq_u.msg_qbytes = 0; 245 new_msqs[i].msq_u.msg_perm._seq = 0; 246 cv_init(&new_msqs[i].msq_cv, "msgwait"); 247 } 248 249 /* 250 * Copy all message queue identifiers, mesage headers and buffer 251 * pools to the new memory location. 252 */ 253 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 254 struct __msg *nmsghdr, *msghdr, *pmsghdr; 255 struct msqid_ds *nmptr, *mptr; 256 kmsq_t *nmsq, *msq; 257 258 msq = &msqs[msqid]; 259 mptr = &msq->msq_u; 260 261 if (mptr->msg_qbytes == 0 && 262 (mptr->msg_perm.mode & MSG_LOCKED) == 0) 263 continue; 264 265 nmsq = &new_msqs[msqid]; 266 nmptr = &nmsq->msq_u; 267 memcpy(nmptr, mptr, sizeof(struct msqid_ds)); 268 269 /* 270 * Go through the message headers, and and copy each 271 * one by taking the new ones, and thus defragmenting. 272 */ 273 nmsghdr = pmsghdr = NULL; 274 msghdr = mptr->_msg_first; 275 while (msghdr) { 276 short nnext = 0, next; 277 u_short msgsz, segcnt; 278 279 /* Take an entry from the new list of free msghdrs */ 280 nmsghdr = new_free_msghdrs; 281 KASSERT(nmsghdr != NULL); 282 new_free_msghdrs = nmsghdr->msg_next; 283 284 nmsghdr->msg_next = NULL; 285 if (pmsghdr) { 286 pmsghdr->msg_next = nmsghdr; 287 } else { 288 nmptr->_msg_first = nmsghdr; 289 pmsghdr = nmsghdr; 290 } 291 nmsghdr->msg_ts = msghdr->msg_ts; 292 nmsghdr->msg_spot = -1; 293 294 /* Compute the amount of segments and reserve them */ 295 msgsz = msghdr->msg_ts; 296 segcnt = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 297 if (segcnt == 0) 298 continue; 299 while (segcnt--) { 300 nnext = new_free_msgmaps; 301 new_free_msgmaps = new_msgmaps[nnext].next; 302 new_nfree_msgmaps--; 303 new_msgmaps[nnext].next = nmsghdr->msg_spot; 304 nmsghdr->msg_spot = nnext; 305 } 306 307 /* Copy all segments */ 308 KASSERT(nnext == nmsghdr->msg_spot); 309 next = msghdr->msg_spot; 310 while (msgsz > 0) { 311 size_t tlen; 312 313 if (msgsz >= msginfo.msgssz) { 314 tlen = msginfo.msgssz; 315 msgsz -= msginfo.msgssz; 316 } else { 317 tlen = msgsz; 318 msgsz = 0; 319 } 320 321 /* Copy the message buffer */ 322 memcpy(&new_msgpool[nnext * msginfo.msgssz], 323 &msgpool[next * msginfo.msgssz], tlen); 324 325 /* Next entry of the map */ 326 nnext = msgmaps[nnext].next; 327 next = msgmaps[next].next; 328 } 329 330 /* Next message header */ 331 msghdr = msghdr->msg_next; 332 } 333 nmptr->_msg_last = nmsghdr; 334 } 335 KASSERT((msginfo.msgseg - nfree_msgmaps) == 336 (newmsgseg - new_nfree_msgmaps)); 337 338 sz = ALIGN(msginfo.msgmax) + 339 ALIGN(msginfo.msgseg * sizeof(struct msgmap)) + 340 ALIGN(msginfo.msgtql * sizeof(struct __msg)) + 341 ALIGN(msginfo.msgmni * sizeof(kmsq_t)); 342 343 for (i = 0; i < msginfo.msgmni; i++) 344 cv_destroy(&msqs[i].msq_cv); 345 346 /* Set the pointers and update the new values */ 347 msgpool = new_msgpool; 348 msgmaps = new_msgmaps; 349 msghdrs = new_msghdrs; 350 msqs = new_msqs; 351 352 free_msghdrs = new_free_msghdrs; 353 free_msgmaps = new_free_msgmaps; 354 nfree_msgmaps = new_nfree_msgmaps; 355 msginfo.msgmni = newmsgmni; 356 msginfo.msgseg = newmsgseg; 357 msginfo.msgmax = newmsgmax; 358 359 /* Reallocation completed - notify all waiters, if any */ 360 msg_realloc_state = false; 361 cv_broadcast(&msg_realloc_cv); 362 mutex_exit(&msgmutex); 363 364 uvm_km_free(kernel_map, (vaddr_t)old_msgpool, sz, UVM_KMF_WIRED); 365 return 0; 366 } 367 368 static void 369 msg_freehdr(struct __msg *msghdr) 370 { 371 372 KASSERT(mutex_owned(&msgmutex)); 373 374 while (msghdr->msg_ts > 0) { 375 short next; 376 KASSERT(msghdr->msg_spot >= 0); 377 KASSERT(msghdr->msg_spot < msginfo.msgseg); 378 379 next = msgmaps[msghdr->msg_spot].next; 380 msgmaps[msghdr->msg_spot].next = free_msgmaps; 381 free_msgmaps = msghdr->msg_spot; 382 nfree_msgmaps++; 383 msghdr->msg_spot = next; 384 if (msghdr->msg_ts >= msginfo.msgssz) 385 msghdr->msg_ts -= msginfo.msgssz; 386 else 387 msghdr->msg_ts = 0; 388 } 389 KASSERT(msghdr->msg_spot == -1); 390 msghdr->msg_next = free_msghdrs; 391 free_msghdrs = msghdr; 392 } 393 394 int 395 sys___msgctl50(struct lwp *l, const struct sys___msgctl50_args *uap, 396 register_t *retval) 397 { 398 /* { 399 syscallarg(int) msqid; 400 syscallarg(int) cmd; 401 syscallarg(struct msqid_ds *) buf; 402 } */ 403 struct msqid_ds msqbuf; 404 int cmd, error; 405 406 cmd = SCARG(uap, cmd); 407 408 if (cmd == IPC_SET) { 409 error = copyin(SCARG(uap, buf), &msqbuf, sizeof(msqbuf)); 410 if (error) 411 return (error); 412 } 413 414 error = msgctl1(l, SCARG(uap, msqid), cmd, 415 (cmd == IPC_SET || cmd == IPC_STAT) ? &msqbuf : NULL); 416 417 if (error == 0 && cmd == IPC_STAT) 418 error = copyout(&msqbuf, SCARG(uap, buf), sizeof(msqbuf)); 419 420 return (error); 421 } 422 423 int 424 msgctl1(struct lwp *l, int msqid, int cmd, struct msqid_ds *msqbuf) 425 { 426 kauth_cred_t cred = l->l_cred; 427 struct msqid_ds *msqptr; 428 kmsq_t *msq; 429 int error = 0, ix; 430 431 MSG_PRINTF(("call to msgctl1(%d, %d)\n", msqid, cmd)); 432 433 ix = IPCID_TO_IX(msqid); 434 435 mutex_enter(&msgmutex); 436 437 if (ix < 0 || ix >= msginfo.msgmni) { 438 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", ix, 439 msginfo.msgmni)); 440 error = EINVAL; 441 goto unlock; 442 } 443 444 msq = &msqs[ix]; 445 msqptr = &msq->msq_u; 446 447 if (msqptr->msg_qbytes == 0) { 448 MSG_PRINTF(("no such msqid\n")); 449 error = EINVAL; 450 goto unlock; 451 } 452 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqid)) { 453 MSG_PRINTF(("wrong sequence number\n")); 454 error = EINVAL; 455 goto unlock; 456 } 457 458 switch (cmd) { 459 case IPC_RMID: 460 { 461 struct __msg *msghdr; 462 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M)) != 0) 463 break; 464 /* Free the message headers */ 465 msghdr = msqptr->_msg_first; 466 while (msghdr != NULL) { 467 struct __msg *msghdr_tmp; 468 469 /* Free the segments of each message */ 470 msqptr->_msg_cbytes -= msghdr->msg_ts; 471 msqptr->msg_qnum--; 472 msghdr_tmp = msghdr; 473 msghdr = msghdr->msg_next; 474 msg_freehdr(msghdr_tmp); 475 } 476 KASSERT(msqptr->_msg_cbytes == 0); 477 KASSERT(msqptr->msg_qnum == 0); 478 479 /* Mark it as free */ 480 msqptr->msg_qbytes = 0; 481 cv_broadcast(&msq->msq_cv); 482 } 483 break; 484 485 case IPC_SET: 486 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M))) 487 break; 488 if (msqbuf->msg_qbytes > msqptr->msg_qbytes && 489 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 490 NULL) != 0) { 491 error = EPERM; 492 break; 493 } 494 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 495 MSG_PRINTF(("can't increase msg_qbytes beyond %d " 496 "(truncating)\n", msginfo.msgmnb)); 497 /* silently restrict qbytes to system limit */ 498 msqbuf->msg_qbytes = msginfo.msgmnb; 499 } 500 if (msqbuf->msg_qbytes == 0) { 501 MSG_PRINTF(("can't reduce msg_qbytes to 0\n")); 502 error = EINVAL; /* XXX non-standard errno! */ 503 break; 504 } 505 msqptr->msg_perm.uid = msqbuf->msg_perm.uid; 506 msqptr->msg_perm.gid = msqbuf->msg_perm.gid; 507 msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) | 508 (msqbuf->msg_perm.mode & 0777); 509 msqptr->msg_qbytes = msqbuf->msg_qbytes; 510 msqptr->msg_ctime = time_second; 511 break; 512 513 case IPC_STAT: 514 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 515 MSG_PRINTF(("requester doesn't have read access\n")); 516 break; 517 } 518 memcpy(msqbuf, msqptr, sizeof(struct msqid_ds)); 519 break; 520 521 default: 522 MSG_PRINTF(("invalid command %d\n", cmd)); 523 error = EINVAL; 524 break; 525 } 526 527 unlock: 528 mutex_exit(&msgmutex); 529 return (error); 530 } 531 532 int 533 sys_msgget(struct lwp *l, const struct sys_msgget_args *uap, register_t *retval) 534 { 535 /* { 536 syscallarg(key_t) key; 537 syscallarg(int) msgflg; 538 } */ 539 int msqid, error = 0; 540 int key = SCARG(uap, key); 541 int msgflg = SCARG(uap, msgflg); 542 kauth_cred_t cred = l->l_cred; 543 struct msqid_ds *msqptr = NULL; 544 kmsq_t *msq; 545 546 mutex_enter(&msgmutex); 547 548 MSG_PRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 549 550 if (key != IPC_PRIVATE) { 551 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 552 msq = &msqs[msqid]; 553 msqptr = &msq->msq_u; 554 if (msqptr->msg_qbytes != 0 && 555 msqptr->msg_perm._key == key) 556 break; 557 } 558 if (msqid < msginfo.msgmni) { 559 MSG_PRINTF(("found public key\n")); 560 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 561 MSG_PRINTF(("not exclusive\n")); 562 error = EEXIST; 563 goto unlock; 564 } 565 if ((error = ipcperm(cred, &msqptr->msg_perm, 566 msgflg & 0700 ))) { 567 MSG_PRINTF(("requester doesn't have 0%o access\n", 568 msgflg & 0700)); 569 goto unlock; 570 } 571 goto found; 572 } 573 } 574 575 MSG_PRINTF(("need to allocate the msqid_ds\n")); 576 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 577 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 578 /* 579 * Look for an unallocated and unlocked msqid_ds. 580 * msqid_ds's can be locked by msgsnd or msgrcv while 581 * they are copying the message in/out. We can't 582 * re-use the entry until they release it. 583 */ 584 msq = &msqs[msqid]; 585 msqptr = &msq->msq_u; 586 if (msqptr->msg_qbytes == 0 && 587 (msqptr->msg_perm.mode & MSG_LOCKED) == 0) 588 break; 589 } 590 if (msqid == msginfo.msgmni) { 591 MSG_PRINTF(("no more msqid_ds's available\n")); 592 error = ENOSPC; 593 goto unlock; 594 } 595 MSG_PRINTF(("msqid %d is available\n", msqid)); 596 msqptr->msg_perm._key = key; 597 msqptr->msg_perm.cuid = kauth_cred_geteuid(cred); 598 msqptr->msg_perm.uid = kauth_cred_geteuid(cred); 599 msqptr->msg_perm.cgid = kauth_cred_getegid(cred); 600 msqptr->msg_perm.gid = kauth_cred_getegid(cred); 601 msqptr->msg_perm.mode = (msgflg & 0777); 602 /* Make sure that the returned msqid is unique */ 603 msqptr->msg_perm._seq++; 604 msqptr->_msg_first = NULL; 605 msqptr->_msg_last = NULL; 606 msqptr->_msg_cbytes = 0; 607 msqptr->msg_qnum = 0; 608 msqptr->msg_qbytes = msginfo.msgmnb; 609 msqptr->msg_lspid = 0; 610 msqptr->msg_lrpid = 0; 611 msqptr->msg_stime = 0; 612 msqptr->msg_rtime = 0; 613 msqptr->msg_ctime = time_second; 614 } else { 615 MSG_PRINTF(("didn't find it and wasn't asked to create it\n")); 616 error = ENOENT; 617 goto unlock; 618 } 619 620 found: 621 /* Construct the unique msqid */ 622 *retval = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm); 623 624 unlock: 625 mutex_exit(&msgmutex); 626 return (error); 627 } 628 629 int 630 sys_msgsnd(struct lwp *l, const struct sys_msgsnd_args *uap, register_t *retval) 631 { 632 /* { 633 syscallarg(int) msqid; 634 syscallarg(const void *) msgp; 635 syscallarg(size_t) msgsz; 636 syscallarg(int) msgflg; 637 } */ 638 639 return msgsnd1(l, SCARG(uap, msqid), SCARG(uap, msgp), 640 SCARG(uap, msgsz), SCARG(uap, msgflg), sizeof(long), copyin); 641 } 642 643 int 644 msgsnd1(struct lwp *l, int msqidr, const char *user_msgp, size_t msgsz, 645 int msgflg, size_t typesz, copyin_t fetch_type) 646 { 647 int segs_needed, error = 0, msqid; 648 kauth_cred_t cred = l->l_cred; 649 struct msqid_ds *msqptr; 650 struct __msg *msghdr; 651 kmsq_t *msq; 652 short next; 653 654 MSG_PRINTF(("call to msgsnd(%d, %p, %lld, %d)\n", msqid, user_msgp, 655 (long long)msgsz, msgflg)); 656 restart: 657 msqid = IPCID_TO_IX(msqidr); 658 659 mutex_enter(&msgmutex); 660 /* In case of reallocation, we will wait for completion */ 661 while (__predict_false(msg_realloc_state)) 662 cv_wait(&msg_realloc_cv, &msgmutex); 663 664 if (msqid < 0 || msqid >= msginfo.msgmni) { 665 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 666 msginfo.msgmni)); 667 error = EINVAL; 668 goto unlock; 669 } 670 671 msq = &msqs[msqid]; 672 msqptr = &msq->msq_u; 673 674 if (msqptr->msg_qbytes == 0) { 675 MSG_PRINTF(("no such message queue id\n")); 676 error = EINVAL; 677 goto unlock; 678 } 679 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 680 MSG_PRINTF(("wrong sequence number\n")); 681 error = EINVAL; 682 goto unlock; 683 } 684 685 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_W))) { 686 MSG_PRINTF(("requester doesn't have write access\n")); 687 goto unlock; 688 } 689 690 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 691 MSG_PRINTF(("msgsz=%lld, msgssz=%d, segs_needed=%d\n", 692 (long long)msgsz, msginfo.msgssz, segs_needed)); 693 for (;;) { 694 int need_more_resources = 0; 695 696 /* 697 * check msgsz [cannot be negative since it is unsigned] 698 * (inside this loop in case msg_qbytes changes while we sleep) 699 */ 700 701 if (msgsz > msqptr->msg_qbytes) { 702 MSG_PRINTF(("msgsz > msqptr->msg_qbytes\n")); 703 error = EINVAL; 704 goto unlock; 705 } 706 707 if (msqptr->msg_perm.mode & MSG_LOCKED) { 708 MSG_PRINTF(("msqid is locked\n")); 709 need_more_resources = 1; 710 } 711 if (msgsz + msqptr->_msg_cbytes > msqptr->msg_qbytes) { 712 MSG_PRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 713 need_more_resources = 1; 714 } 715 if (segs_needed > nfree_msgmaps) { 716 MSG_PRINTF(("segs_needed > nfree_msgmaps\n")); 717 need_more_resources = 1; 718 } 719 if (free_msghdrs == NULL) { 720 MSG_PRINTF(("no more msghdrs\n")); 721 need_more_resources = 1; 722 } 723 724 if (need_more_resources) { 725 int we_own_it; 726 727 if ((msgflg & IPC_NOWAIT) != 0) { 728 MSG_PRINTF(("need more resources but caller " 729 "doesn't want to wait\n")); 730 error = EAGAIN; 731 goto unlock; 732 } 733 734 if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { 735 MSG_PRINTF(("we don't own the msqid_ds\n")); 736 we_own_it = 0; 737 } else { 738 /* Force later arrivals to wait for our 739 request */ 740 MSG_PRINTF(("we own the msqid_ds\n")); 741 msqptr->msg_perm.mode |= MSG_LOCKED; 742 we_own_it = 1; 743 } 744 745 msg_waiters++; 746 MSG_PRINTF(("goodnight\n")); 747 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 748 MSG_PRINTF(("good morning, error=%d\n", error)); 749 msg_waiters--; 750 751 if (we_own_it) 752 msqptr->msg_perm.mode &= ~MSG_LOCKED; 753 754 /* 755 * In case of such state, notify reallocator and 756 * restart the call. 757 */ 758 if (msg_realloc_state) { 759 cv_broadcast(&msg_realloc_cv); 760 mutex_exit(&msgmutex); 761 goto restart; 762 } 763 764 if (error != 0) { 765 MSG_PRINTF(("msgsnd: interrupted system " 766 "call\n")); 767 error = EINTR; 768 goto unlock; 769 } 770 771 /* 772 * Make sure that the msq queue still exists 773 */ 774 775 if (msqptr->msg_qbytes == 0) { 776 MSG_PRINTF(("msqid deleted\n")); 777 error = EIDRM; 778 goto unlock; 779 } 780 } else { 781 MSG_PRINTF(("got all the resources that we need\n")); 782 break; 783 } 784 } 785 786 /* 787 * We have the resources that we need. 788 * Make sure! 789 */ 790 791 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 792 KASSERT(segs_needed <= nfree_msgmaps); 793 KASSERT(msgsz + msqptr->_msg_cbytes <= msqptr->msg_qbytes); 794 KASSERT(free_msghdrs != NULL); 795 796 /* 797 * Re-lock the msqid_ds in case we page-fault when copying in the 798 * message 799 */ 800 801 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 802 msqptr->msg_perm.mode |= MSG_LOCKED; 803 804 /* 805 * Allocate a message header 806 */ 807 808 msghdr = free_msghdrs; 809 free_msghdrs = msghdr->msg_next; 810 msghdr->msg_spot = -1; 811 msghdr->msg_ts = msgsz; 812 813 /* 814 * Allocate space for the message 815 */ 816 817 while (segs_needed > 0) { 818 KASSERT(nfree_msgmaps > 0); 819 KASSERT(free_msgmaps != -1); 820 KASSERT(free_msgmaps < msginfo.msgseg); 821 822 next = free_msgmaps; 823 MSG_PRINTF(("allocating segment %d to message\n", next)); 824 free_msgmaps = msgmaps[next].next; 825 nfree_msgmaps--; 826 msgmaps[next].next = msghdr->msg_spot; 827 msghdr->msg_spot = next; 828 segs_needed--; 829 } 830 831 /* 832 * Copy in the message type 833 */ 834 mutex_exit(&msgmutex); 835 error = (*fetch_type)(user_msgp, &msghdr->msg_type, typesz); 836 mutex_enter(&msgmutex); 837 if (error != 0) { 838 MSG_PRINTF(("error %d copying the message type\n", error)); 839 msg_freehdr(msghdr); 840 msqptr->msg_perm.mode &= ~MSG_LOCKED; 841 cv_broadcast(&msq->msq_cv); 842 goto unlock; 843 } 844 user_msgp += typesz; 845 846 /* 847 * Validate the message type 848 */ 849 850 if (msghdr->msg_type < 1) { 851 msg_freehdr(msghdr); 852 msqptr->msg_perm.mode &= ~MSG_LOCKED; 853 cv_broadcast(&msq->msq_cv); 854 MSG_PRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 855 error = EINVAL; 856 goto unlock; 857 } 858 859 /* 860 * Copy in the message body 861 */ 862 863 next = msghdr->msg_spot; 864 while (msgsz > 0) { 865 size_t tlen; 866 KASSERT(next > -1); 867 KASSERT(next < msginfo.msgseg); 868 869 if (msgsz > msginfo.msgssz) 870 tlen = msginfo.msgssz; 871 else 872 tlen = msgsz; 873 mutex_exit(&msgmutex); 874 error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen); 875 mutex_enter(&msgmutex); 876 if (error != 0) { 877 MSG_PRINTF(("error %d copying in message segment\n", 878 error)); 879 msg_freehdr(msghdr); 880 msqptr->msg_perm.mode &= ~MSG_LOCKED; 881 cv_broadcast(&msq->msq_cv); 882 goto unlock; 883 } 884 msgsz -= tlen; 885 user_msgp += tlen; 886 next = msgmaps[next].next; 887 } 888 KASSERT(next == -1); 889 890 /* 891 * We've got the message. Unlock the msqid_ds. 892 */ 893 894 msqptr->msg_perm.mode &= ~MSG_LOCKED; 895 896 /* 897 * Make sure that the msqid_ds is still allocated. 898 */ 899 900 if (msqptr->msg_qbytes == 0) { 901 msg_freehdr(msghdr); 902 cv_broadcast(&msq->msq_cv); 903 error = EIDRM; 904 goto unlock; 905 } 906 907 /* 908 * Put the message into the queue 909 */ 910 911 if (msqptr->_msg_first == NULL) { 912 msqptr->_msg_first = msghdr; 913 msqptr->_msg_last = msghdr; 914 } else { 915 msqptr->_msg_last->msg_next = msghdr; 916 msqptr->_msg_last = msghdr; 917 } 918 msqptr->_msg_last->msg_next = NULL; 919 920 msqptr->_msg_cbytes += msghdr->msg_ts; 921 msqptr->msg_qnum++; 922 msqptr->msg_lspid = l->l_proc->p_pid; 923 msqptr->msg_stime = time_second; 924 925 cv_broadcast(&msq->msq_cv); 926 927 unlock: 928 mutex_exit(&msgmutex); 929 return error; 930 } 931 932 int 933 sys_msgrcv(struct lwp *l, const struct sys_msgrcv_args *uap, register_t *retval) 934 { 935 /* { 936 syscallarg(int) msqid; 937 syscallarg(void *) msgp; 938 syscallarg(size_t) msgsz; 939 syscallarg(long) msgtyp; 940 syscallarg(int) msgflg; 941 } */ 942 943 return msgrcv1(l, SCARG(uap, msqid), SCARG(uap, msgp), 944 SCARG(uap, msgsz), SCARG(uap, msgtyp), SCARG(uap, msgflg), 945 sizeof(long), copyout, retval); 946 } 947 948 int 949 msgrcv1(struct lwp *l, int msqidr, char *user_msgp, size_t msgsz, long msgtyp, 950 int msgflg, size_t typesz, copyout_t put_type, register_t *retval) 951 { 952 size_t len; 953 kauth_cred_t cred = l->l_cred; 954 struct msqid_ds *msqptr; 955 struct __msg *msghdr; 956 int error = 0, msqid; 957 kmsq_t *msq; 958 short next; 959 960 MSG_PRINTF(("call to msgrcv(%d, %p, %lld, %ld, %d)\n", msqid, 961 user_msgp, (long long)msgsz, msgtyp, msgflg)); 962 restart: 963 msqid = IPCID_TO_IX(msqidr); 964 965 mutex_enter(&msgmutex); 966 /* In case of reallocation, we will wait for completion */ 967 while (__predict_false(msg_realloc_state)) 968 cv_wait(&msg_realloc_cv, &msgmutex); 969 970 if (msqid < 0 || msqid >= msginfo.msgmni) { 971 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 972 msginfo.msgmni)); 973 error = EINVAL; 974 goto unlock; 975 } 976 977 msq = &msqs[msqid]; 978 msqptr = &msq->msq_u; 979 980 if (msqptr->msg_qbytes == 0) { 981 MSG_PRINTF(("no such message queue id\n")); 982 error = EINVAL; 983 goto unlock; 984 } 985 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 986 MSG_PRINTF(("wrong sequence number\n")); 987 error = EINVAL; 988 goto unlock; 989 } 990 991 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 992 MSG_PRINTF(("requester doesn't have read access\n")); 993 goto unlock; 994 } 995 996 msghdr = NULL; 997 while (msghdr == NULL) { 998 if (msgtyp == 0) { 999 msghdr = msqptr->_msg_first; 1000 if (msghdr != NULL) { 1001 if (msgsz < msghdr->msg_ts && 1002 (msgflg & MSG_NOERROR) == 0) { 1003 MSG_PRINTF(("first msg on the queue " 1004 "is too big (want %lld, got %d)\n", 1005 (long long)msgsz, msghdr->msg_ts)); 1006 error = E2BIG; 1007 goto unlock; 1008 } 1009 if (msqptr->_msg_first == msqptr->_msg_last) { 1010 msqptr->_msg_first = NULL; 1011 msqptr->_msg_last = NULL; 1012 } else { 1013 msqptr->_msg_first = msghdr->msg_next; 1014 KASSERT(msqptr->_msg_first != NULL); 1015 } 1016 } 1017 } else { 1018 struct __msg *previous; 1019 struct __msg **prev; 1020 1021 for (previous = NULL, prev = &msqptr->_msg_first; 1022 (msghdr = *prev) != NULL; 1023 previous = msghdr, prev = &msghdr->msg_next) { 1024 /* 1025 * Is this message's type an exact match or is 1026 * this message's type less than or equal to 1027 * the absolute value of a negative msgtyp? 1028 * Note that the second half of this test can 1029 * NEVER be true if msgtyp is positive since 1030 * msg_type is always positive! 1031 */ 1032 1033 if (msgtyp != msghdr->msg_type && 1034 msghdr->msg_type > -msgtyp) 1035 continue; 1036 1037 MSG_PRINTF(("found message type %ld, requested %ld\n", 1038 msghdr->msg_type, msgtyp)); 1039 if (msgsz < msghdr->msg_ts && 1040 (msgflg & MSG_NOERROR) == 0) { 1041 MSG_PRINTF(("requested message on the queue " 1042 "is too big (want %lld, got %d)\n", 1043 (long long)msgsz, msghdr->msg_ts)); 1044 error = E2BIG; 1045 goto unlock; 1046 } 1047 *prev = msghdr->msg_next; 1048 if (msghdr != msqptr->_msg_last) 1049 break; 1050 if (previous == NULL) { 1051 KASSERT(prev == &msqptr->_msg_first); 1052 msqptr->_msg_first = NULL; 1053 msqptr->_msg_last = NULL; 1054 } else { 1055 KASSERT(prev != &msqptr->_msg_first); 1056 msqptr->_msg_last = previous; 1057 } 1058 break; 1059 } 1060 } 1061 1062 /* 1063 * We've either extracted the msghdr for the appropriate 1064 * message or there isn't one. 1065 * If there is one then bail out of this loop. 1066 */ 1067 if (msghdr != NULL) 1068 break; 1069 1070 /* 1071 * Hmph! No message found. Does the user want to wait? 1072 */ 1073 1074 if ((msgflg & IPC_NOWAIT) != 0) { 1075 MSG_PRINTF(("no appropriate message found (msgtyp=%ld)\n", 1076 msgtyp)); 1077 error = ENOMSG; 1078 goto unlock; 1079 } 1080 1081 /* 1082 * Wait for something to happen 1083 */ 1084 1085 msg_waiters++; 1086 MSG_PRINTF(("msgrcv: goodnight\n")); 1087 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 1088 MSG_PRINTF(("msgrcv: good morning (error=%d)\n", error)); 1089 msg_waiters--; 1090 1091 /* 1092 * In case of such state, notify reallocator and 1093 * restart the call. 1094 */ 1095 if (msg_realloc_state) { 1096 cv_broadcast(&msg_realloc_cv); 1097 mutex_exit(&msgmutex); 1098 goto restart; 1099 } 1100 1101 if (error != 0) { 1102 MSG_PRINTF(("msgsnd: interrupted system call\n")); 1103 error = EINTR; 1104 goto unlock; 1105 } 1106 1107 /* 1108 * Make sure that the msq queue still exists 1109 */ 1110 1111 if (msqptr->msg_qbytes == 0 || 1112 msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 1113 MSG_PRINTF(("msqid deleted\n")); 1114 error = EIDRM; 1115 goto unlock; 1116 } 1117 } 1118 1119 /* 1120 * Return the message to the user. 1121 * 1122 * First, do the bookkeeping (before we risk being interrupted). 1123 */ 1124 1125 msqptr->_msg_cbytes -= msghdr->msg_ts; 1126 msqptr->msg_qnum--; 1127 msqptr->msg_lrpid = l->l_proc->p_pid; 1128 msqptr->msg_rtime = time_second; 1129 1130 /* 1131 * Make msgsz the actual amount that we'll be returning. 1132 * Note that this effectively truncates the message if it is too long 1133 * (since msgsz is never increased). 1134 */ 1135 1136 MSG_PRINTF(("found a message, msgsz=%lld, msg_ts=%d\n", 1137 (long long)msgsz, msghdr->msg_ts)); 1138 if (msgsz > msghdr->msg_ts) 1139 msgsz = msghdr->msg_ts; 1140 1141 /* 1142 * Return the type to the user. 1143 */ 1144 mutex_exit(&msgmutex); 1145 error = (*put_type)(&msghdr->msg_type, user_msgp, typesz); 1146 mutex_enter(&msgmutex); 1147 if (error != 0) { 1148 MSG_PRINTF(("error (%d) copying out message type\n", error)); 1149 msg_freehdr(msghdr); 1150 cv_broadcast(&msq->msq_cv); 1151 goto unlock; 1152 } 1153 user_msgp += typesz; 1154 1155 /* 1156 * Return the segments to the user 1157 */ 1158 1159 next = msghdr->msg_spot; 1160 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1161 size_t tlen; 1162 KASSERT(next > -1); 1163 KASSERT(next < msginfo.msgseg); 1164 1165 if (msgsz - len > msginfo.msgssz) 1166 tlen = msginfo.msgssz; 1167 else 1168 tlen = msgsz - len; 1169 mutex_exit(&msgmutex); 1170 error = (*put_type)(&msgpool[next * msginfo.msgssz], 1171 user_msgp, tlen); 1172 mutex_enter(&msgmutex); 1173 if (error != 0) { 1174 MSG_PRINTF(("error (%d) copying out message segment\n", 1175 error)); 1176 msg_freehdr(msghdr); 1177 cv_broadcast(&msq->msq_cv); 1178 goto unlock; 1179 } 1180 user_msgp += tlen; 1181 next = msgmaps[next].next; 1182 } 1183 1184 /* 1185 * Done, return the actual number of bytes copied out. 1186 */ 1187 1188 msg_freehdr(msghdr); 1189 cv_broadcast(&msq->msq_cv); 1190 *retval = msgsz; 1191 1192 unlock: 1193 mutex_exit(&msgmutex); 1194 return error; 1195 } 1196 1197 /* 1198 * Sysctl initialization and nodes. 1199 */ 1200 1201 static int 1202 sysctl_ipc_msgmni(SYSCTLFN_ARGS) 1203 { 1204 int newsize, error; 1205 struct sysctlnode node; 1206 node = *rnode; 1207 node.sysctl_data = &newsize; 1208 1209 newsize = msginfo.msgmni; 1210 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1211 if (error || newp == NULL) 1212 return error; 1213 1214 sysctl_unlock(); 1215 error = msgrealloc(newsize, msginfo.msgseg); 1216 sysctl_relock(); 1217 return error; 1218 } 1219 1220 static int 1221 sysctl_ipc_msgseg(SYSCTLFN_ARGS) 1222 { 1223 int newsize, error; 1224 struct sysctlnode node; 1225 node = *rnode; 1226 node.sysctl_data = &newsize; 1227 1228 newsize = msginfo.msgseg; 1229 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1230 if (error || newp == NULL) 1231 return error; 1232 1233 sysctl_unlock(); 1234 error = msgrealloc(msginfo.msgmni, newsize); 1235 sysctl_relock(); 1236 return error; 1237 } 1238 1239 SYSCTL_SETUP(sysctl_ipc_msg_setup, "sysctl kern.ipc subtree setup") 1240 { 1241 const struct sysctlnode *node = NULL; 1242 1243 sysctl_createv(clog, 0, NULL, NULL, 1244 CTLFLAG_PERMANENT, 1245 CTLTYPE_NODE, "kern", NULL, 1246 NULL, 0, NULL, 0, 1247 CTL_KERN, CTL_EOL); 1248 sysctl_createv(clog, 0, NULL, &node, 1249 CTLFLAG_PERMANENT, 1250 CTLTYPE_NODE, "ipc", 1251 SYSCTL_DESCR("SysV IPC options"), 1252 NULL, 0, NULL, 0, 1253 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1254 1255 if (node == NULL) 1256 return; 1257 1258 sysctl_createv(clog, 0, &node, NULL, 1259 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1260 CTLTYPE_INT, "msgmni", 1261 SYSCTL_DESCR("Max number of message queue identifiers"), 1262 sysctl_ipc_msgmni, 0, &msginfo.msgmni, 0, 1263 CTL_CREATE, CTL_EOL); 1264 sysctl_createv(clog, 0, &node, NULL, 1265 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1266 CTLTYPE_INT, "msgseg", 1267 SYSCTL_DESCR("Max number of number of message segments"), 1268 sysctl_ipc_msgseg, 0, &msginfo.msgseg, 0, 1269 CTL_CREATE, CTL_EOL); 1270 } 1271