1 /* $NetBSD: sysv_msg.c,v 1.58 2008/09/19 11:21:33 rmind 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 1.58 2008/09/19 11:21:33 rmind 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___msgctl13(struct lwp *l, const struct sys___msgctl13_args *uap, register_t *retval) 396 { 397 /* { 398 syscallarg(int) msqid; 399 syscallarg(int) cmd; 400 syscallarg(struct msqid_ds *) buf; 401 } */ 402 struct msqid_ds msqbuf; 403 int cmd, error; 404 405 cmd = SCARG(uap, cmd); 406 407 if (cmd == IPC_SET) { 408 error = copyin(SCARG(uap, buf), &msqbuf, sizeof(msqbuf)); 409 if (error) 410 return (error); 411 } 412 413 error = msgctl1(l, SCARG(uap, msqid), cmd, 414 (cmd == IPC_SET || cmd == IPC_STAT) ? &msqbuf : NULL); 415 416 if (error == 0 && cmd == IPC_STAT) 417 error = copyout(&msqbuf, SCARG(uap, buf), sizeof(msqbuf)); 418 419 return (error); 420 } 421 422 int 423 msgctl1(struct lwp *l, int msqid, int cmd, struct msqid_ds *msqbuf) 424 { 425 kauth_cred_t cred = l->l_cred; 426 struct msqid_ds *msqptr; 427 kmsq_t *msq; 428 int error = 0, ix; 429 430 MSG_PRINTF(("call to msgctl1(%d, %d)\n", msqid, cmd)); 431 432 ix = IPCID_TO_IX(msqid); 433 434 mutex_enter(&msgmutex); 435 436 if (ix < 0 || ix >= msginfo.msgmni) { 437 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", ix, 438 msginfo.msgmni)); 439 error = EINVAL; 440 goto unlock; 441 } 442 443 msq = &msqs[ix]; 444 msqptr = &msq->msq_u; 445 446 if (msqptr->msg_qbytes == 0) { 447 MSG_PRINTF(("no such msqid\n")); 448 error = EINVAL; 449 goto unlock; 450 } 451 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqid)) { 452 MSG_PRINTF(("wrong sequence number\n")); 453 error = EINVAL; 454 goto unlock; 455 } 456 457 switch (cmd) { 458 case IPC_RMID: 459 { 460 struct __msg *msghdr; 461 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M)) != 0) 462 break; 463 /* Free the message headers */ 464 msghdr = msqptr->_msg_first; 465 while (msghdr != NULL) { 466 struct __msg *msghdr_tmp; 467 468 /* Free the segments of each message */ 469 msqptr->_msg_cbytes -= msghdr->msg_ts; 470 msqptr->msg_qnum--; 471 msghdr_tmp = msghdr; 472 msghdr = msghdr->msg_next; 473 msg_freehdr(msghdr_tmp); 474 } 475 KASSERT(msqptr->_msg_cbytes == 0); 476 KASSERT(msqptr->msg_qnum == 0); 477 478 /* Mark it as free */ 479 msqptr->msg_qbytes = 0; 480 cv_broadcast(&msq->msq_cv); 481 } 482 break; 483 484 case IPC_SET: 485 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_M))) 486 break; 487 if (msqbuf->msg_qbytes > msqptr->msg_qbytes && 488 kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 489 NULL) != 0) { 490 error = EPERM; 491 break; 492 } 493 if (msqbuf->msg_qbytes > msginfo.msgmnb) { 494 MSG_PRINTF(("can't increase msg_qbytes beyond %d " 495 "(truncating)\n", msginfo.msgmnb)); 496 /* silently restrict qbytes to system limit */ 497 msqbuf->msg_qbytes = msginfo.msgmnb; 498 } 499 if (msqbuf->msg_qbytes == 0) { 500 MSG_PRINTF(("can't reduce msg_qbytes to 0\n")); 501 error = EINVAL; /* XXX non-standard errno! */ 502 break; 503 } 504 msqptr->msg_perm.uid = msqbuf->msg_perm.uid; 505 msqptr->msg_perm.gid = msqbuf->msg_perm.gid; 506 msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) | 507 (msqbuf->msg_perm.mode & 0777); 508 msqptr->msg_qbytes = msqbuf->msg_qbytes; 509 msqptr->msg_ctime = time_second; 510 break; 511 512 case IPC_STAT: 513 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 514 MSG_PRINTF(("requester doesn't have read access\n")); 515 break; 516 } 517 memcpy(msqbuf, msqptr, sizeof(struct msqid_ds)); 518 break; 519 520 default: 521 MSG_PRINTF(("invalid command %d\n", cmd)); 522 error = EINVAL; 523 break; 524 } 525 526 unlock: 527 mutex_exit(&msgmutex); 528 return (error); 529 } 530 531 int 532 sys_msgget(struct lwp *l, const struct sys_msgget_args *uap, register_t *retval) 533 { 534 /* { 535 syscallarg(key_t) key; 536 syscallarg(int) msgflg; 537 } */ 538 int msqid, error = 0; 539 int key = SCARG(uap, key); 540 int msgflg = SCARG(uap, msgflg); 541 kauth_cred_t cred = l->l_cred; 542 struct msqid_ds *msqptr = NULL; 543 kmsq_t *msq; 544 545 mutex_enter(&msgmutex); 546 547 MSG_PRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); 548 549 if (key != IPC_PRIVATE) { 550 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 551 msq = &msqs[msqid]; 552 msqptr = &msq->msq_u; 553 if (msqptr->msg_qbytes != 0 && 554 msqptr->msg_perm._key == key) 555 break; 556 } 557 if (msqid < msginfo.msgmni) { 558 MSG_PRINTF(("found public key\n")); 559 if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { 560 MSG_PRINTF(("not exclusive\n")); 561 error = EEXIST; 562 goto unlock; 563 } 564 if ((error = ipcperm(cred, &msqptr->msg_perm, 565 msgflg & 0700 ))) { 566 MSG_PRINTF(("requester doesn't have 0%o access\n", 567 msgflg & 0700)); 568 goto unlock; 569 } 570 goto found; 571 } 572 } 573 574 MSG_PRINTF(("need to allocate the msqid_ds\n")); 575 if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { 576 for (msqid = 0; msqid < msginfo.msgmni; msqid++) { 577 /* 578 * Look for an unallocated and unlocked msqid_ds. 579 * msqid_ds's can be locked by msgsnd or msgrcv while 580 * they are copying the message in/out. We can't 581 * re-use the entry until they release it. 582 */ 583 msq = &msqs[msqid]; 584 msqptr = &msq->msq_u; 585 if (msqptr->msg_qbytes == 0 && 586 (msqptr->msg_perm.mode & MSG_LOCKED) == 0) 587 break; 588 } 589 if (msqid == msginfo.msgmni) { 590 MSG_PRINTF(("no more msqid_ds's available\n")); 591 error = ENOSPC; 592 goto unlock; 593 } 594 MSG_PRINTF(("msqid %d is available\n", msqid)); 595 msqptr->msg_perm._key = key; 596 msqptr->msg_perm.cuid = kauth_cred_geteuid(cred); 597 msqptr->msg_perm.uid = kauth_cred_geteuid(cred); 598 msqptr->msg_perm.cgid = kauth_cred_getegid(cred); 599 msqptr->msg_perm.gid = kauth_cred_getegid(cred); 600 msqptr->msg_perm.mode = (msgflg & 0777); 601 /* Make sure that the returned msqid is unique */ 602 msqptr->msg_perm._seq++; 603 msqptr->_msg_first = NULL; 604 msqptr->_msg_last = NULL; 605 msqptr->_msg_cbytes = 0; 606 msqptr->msg_qnum = 0; 607 msqptr->msg_qbytes = msginfo.msgmnb; 608 msqptr->msg_lspid = 0; 609 msqptr->msg_lrpid = 0; 610 msqptr->msg_stime = 0; 611 msqptr->msg_rtime = 0; 612 msqptr->msg_ctime = time_second; 613 } else { 614 MSG_PRINTF(("didn't find it and wasn't asked to create it\n")); 615 error = ENOENT; 616 goto unlock; 617 } 618 619 found: 620 /* Construct the unique msqid */ 621 *retval = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm); 622 623 unlock: 624 mutex_exit(&msgmutex); 625 return (error); 626 } 627 628 int 629 sys_msgsnd(struct lwp *l, const struct sys_msgsnd_args *uap, register_t *retval) 630 { 631 /* { 632 syscallarg(int) msqid; 633 syscallarg(const void *) msgp; 634 syscallarg(size_t) msgsz; 635 syscallarg(int) msgflg; 636 } */ 637 638 return msgsnd1(l, SCARG(uap, msqid), SCARG(uap, msgp), 639 SCARG(uap, msgsz), SCARG(uap, msgflg), sizeof(long), copyin); 640 } 641 642 int 643 msgsnd1(struct lwp *l, int msqidr, const char *user_msgp, size_t msgsz, 644 int msgflg, size_t typesz, copyin_t fetch_type) 645 { 646 int segs_needed, error = 0, msqid; 647 kauth_cred_t cred = l->l_cred; 648 struct msqid_ds *msqptr; 649 struct __msg *msghdr; 650 kmsq_t *msq; 651 short next; 652 653 MSG_PRINTF(("call to msgsnd(%d, %p, %lld, %d)\n", msqid, user_msgp, 654 (long long)msgsz, msgflg)); 655 restart: 656 msqid = IPCID_TO_IX(msqidr); 657 658 mutex_enter(&msgmutex); 659 /* In case of reallocation, we will wait for completion */ 660 while (__predict_false(msg_realloc_state)) 661 cv_wait(&msg_realloc_cv, &msgmutex); 662 663 if (msqid < 0 || msqid >= msginfo.msgmni) { 664 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 665 msginfo.msgmni)); 666 error = EINVAL; 667 goto unlock; 668 } 669 670 msq = &msqs[msqid]; 671 msqptr = &msq->msq_u; 672 673 if (msqptr->msg_qbytes == 0) { 674 MSG_PRINTF(("no such message queue id\n")); 675 error = EINVAL; 676 goto unlock; 677 } 678 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 679 MSG_PRINTF(("wrong sequence number\n")); 680 error = EINVAL; 681 goto unlock; 682 } 683 684 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_W))) { 685 MSG_PRINTF(("requester doesn't have write access\n")); 686 goto unlock; 687 } 688 689 segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; 690 MSG_PRINTF(("msgsz=%lld, msgssz=%d, segs_needed=%d\n", 691 (long long)msgsz, msginfo.msgssz, segs_needed)); 692 for (;;) { 693 int need_more_resources = 0; 694 695 /* 696 * check msgsz [cannot be negative since it is unsigned] 697 * (inside this loop in case msg_qbytes changes while we sleep) 698 */ 699 700 if (msgsz > msqptr->msg_qbytes) { 701 MSG_PRINTF(("msgsz > msqptr->msg_qbytes\n")); 702 error = EINVAL; 703 goto unlock; 704 } 705 706 if (msqptr->msg_perm.mode & MSG_LOCKED) { 707 MSG_PRINTF(("msqid is locked\n")); 708 need_more_resources = 1; 709 } 710 if (msgsz + msqptr->_msg_cbytes > msqptr->msg_qbytes) { 711 MSG_PRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); 712 need_more_resources = 1; 713 } 714 if (segs_needed > nfree_msgmaps) { 715 MSG_PRINTF(("segs_needed > nfree_msgmaps\n")); 716 need_more_resources = 1; 717 } 718 if (free_msghdrs == NULL) { 719 MSG_PRINTF(("no more msghdrs\n")); 720 need_more_resources = 1; 721 } 722 723 if (need_more_resources) { 724 int we_own_it; 725 726 if ((msgflg & IPC_NOWAIT) != 0) { 727 MSG_PRINTF(("need more resources but caller " 728 "doesn't want to wait\n")); 729 error = EAGAIN; 730 goto unlock; 731 } 732 733 if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { 734 MSG_PRINTF(("we don't own the msqid_ds\n")); 735 we_own_it = 0; 736 } else { 737 /* Force later arrivals to wait for our 738 request */ 739 MSG_PRINTF(("we own the msqid_ds\n")); 740 msqptr->msg_perm.mode |= MSG_LOCKED; 741 we_own_it = 1; 742 } 743 744 msg_waiters++; 745 MSG_PRINTF(("goodnight\n")); 746 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 747 MSG_PRINTF(("good morning, error=%d\n", error)); 748 msg_waiters--; 749 750 if (we_own_it) 751 msqptr->msg_perm.mode &= ~MSG_LOCKED; 752 753 /* 754 * In case of such state, notify reallocator and 755 * restart the call. 756 */ 757 if (msg_realloc_state) { 758 cv_broadcast(&msg_realloc_cv); 759 mutex_exit(&msgmutex); 760 goto restart; 761 } 762 763 if (error != 0) { 764 MSG_PRINTF(("msgsnd: interrupted system " 765 "call\n")); 766 error = EINTR; 767 goto unlock; 768 } 769 770 /* 771 * Make sure that the msq queue still exists 772 */ 773 774 if (msqptr->msg_qbytes == 0) { 775 MSG_PRINTF(("msqid deleted\n")); 776 error = EIDRM; 777 goto unlock; 778 } 779 } else { 780 MSG_PRINTF(("got all the resources that we need\n")); 781 break; 782 } 783 } 784 785 /* 786 * We have the resources that we need. 787 * Make sure! 788 */ 789 790 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 791 KASSERT(segs_needed <= nfree_msgmaps); 792 KASSERT(msgsz + msqptr->_msg_cbytes <= msqptr->msg_qbytes); 793 KASSERT(free_msghdrs != NULL); 794 795 /* 796 * Re-lock the msqid_ds in case we page-fault when copying in the 797 * message 798 */ 799 800 KASSERT((msqptr->msg_perm.mode & MSG_LOCKED) == 0); 801 msqptr->msg_perm.mode |= MSG_LOCKED; 802 803 /* 804 * Allocate a message header 805 */ 806 807 msghdr = free_msghdrs; 808 free_msghdrs = msghdr->msg_next; 809 msghdr->msg_spot = -1; 810 msghdr->msg_ts = msgsz; 811 812 /* 813 * Allocate space for the message 814 */ 815 816 while (segs_needed > 0) { 817 KASSERT(nfree_msgmaps > 0); 818 KASSERT(free_msgmaps != -1); 819 KASSERT(free_msgmaps < msginfo.msgseg); 820 821 next = free_msgmaps; 822 MSG_PRINTF(("allocating segment %d to message\n", next)); 823 free_msgmaps = msgmaps[next].next; 824 nfree_msgmaps--; 825 msgmaps[next].next = msghdr->msg_spot; 826 msghdr->msg_spot = next; 827 segs_needed--; 828 } 829 830 /* 831 * Copy in the message type 832 */ 833 mutex_exit(&msgmutex); 834 error = (*fetch_type)(user_msgp, &msghdr->msg_type, typesz); 835 mutex_enter(&msgmutex); 836 if (error != 0) { 837 MSG_PRINTF(("error %d copying the message type\n", error)); 838 msg_freehdr(msghdr); 839 msqptr->msg_perm.mode &= ~MSG_LOCKED; 840 cv_broadcast(&msq->msq_cv); 841 goto unlock; 842 } 843 user_msgp += typesz; 844 845 /* 846 * Validate the message type 847 */ 848 849 if (msghdr->msg_type < 1) { 850 msg_freehdr(msghdr); 851 msqptr->msg_perm.mode &= ~MSG_LOCKED; 852 cv_broadcast(&msq->msq_cv); 853 MSG_PRINTF(("mtype (%ld) < 1\n", msghdr->msg_type)); 854 error = EINVAL; 855 goto unlock; 856 } 857 858 /* 859 * Copy in the message body 860 */ 861 862 next = msghdr->msg_spot; 863 while (msgsz > 0) { 864 size_t tlen; 865 KASSERT(next > -1); 866 KASSERT(next < msginfo.msgseg); 867 868 if (msgsz > msginfo.msgssz) 869 tlen = msginfo.msgssz; 870 else 871 tlen = msgsz; 872 mutex_exit(&msgmutex); 873 error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen); 874 mutex_enter(&msgmutex); 875 if (error != 0) { 876 MSG_PRINTF(("error %d copying in message segment\n", 877 error)); 878 msg_freehdr(msghdr); 879 msqptr->msg_perm.mode &= ~MSG_LOCKED; 880 cv_broadcast(&msq->msq_cv); 881 goto unlock; 882 } 883 msgsz -= tlen; 884 user_msgp += tlen; 885 next = msgmaps[next].next; 886 } 887 KASSERT(next == -1); 888 889 /* 890 * We've got the message. Unlock the msqid_ds. 891 */ 892 893 msqptr->msg_perm.mode &= ~MSG_LOCKED; 894 895 /* 896 * Make sure that the msqid_ds is still allocated. 897 */ 898 899 if (msqptr->msg_qbytes == 0) { 900 msg_freehdr(msghdr); 901 cv_broadcast(&msq->msq_cv); 902 error = EIDRM; 903 goto unlock; 904 } 905 906 /* 907 * Put the message into the queue 908 */ 909 910 if (msqptr->_msg_first == NULL) { 911 msqptr->_msg_first = msghdr; 912 msqptr->_msg_last = msghdr; 913 } else { 914 msqptr->_msg_last->msg_next = msghdr; 915 msqptr->_msg_last = msghdr; 916 } 917 msqptr->_msg_last->msg_next = NULL; 918 919 msqptr->_msg_cbytes += msghdr->msg_ts; 920 msqptr->msg_qnum++; 921 msqptr->msg_lspid = l->l_proc->p_pid; 922 msqptr->msg_stime = time_second; 923 924 cv_broadcast(&msq->msq_cv); 925 926 unlock: 927 mutex_exit(&msgmutex); 928 return error; 929 } 930 931 int 932 sys_msgrcv(struct lwp *l, const struct sys_msgrcv_args *uap, register_t *retval) 933 { 934 /* { 935 syscallarg(int) msqid; 936 syscallarg(void *) msgp; 937 syscallarg(size_t) msgsz; 938 syscallarg(long) msgtyp; 939 syscallarg(int) msgflg; 940 } */ 941 942 return msgrcv1(l, SCARG(uap, msqid), SCARG(uap, msgp), 943 SCARG(uap, msgsz), SCARG(uap, msgtyp), SCARG(uap, msgflg), 944 sizeof(long), copyout, retval); 945 } 946 947 int 948 msgrcv1(struct lwp *l, int msqidr, char *user_msgp, size_t msgsz, long msgtyp, 949 int msgflg, size_t typesz, copyout_t put_type, register_t *retval) 950 { 951 size_t len; 952 kauth_cred_t cred = l->l_cred; 953 struct msqid_ds *msqptr; 954 struct __msg *msghdr; 955 int error = 0, msqid; 956 kmsq_t *msq; 957 short next; 958 959 MSG_PRINTF(("call to msgrcv(%d, %p, %lld, %ld, %d)\n", msqid, 960 user_msgp, (long long)msgsz, msgtyp, msgflg)); 961 restart: 962 msqid = IPCID_TO_IX(msqidr); 963 964 mutex_enter(&msgmutex); 965 /* In case of reallocation, we will wait for completion */ 966 while (__predict_false(msg_realloc_state)) 967 cv_wait(&msg_realloc_cv, &msgmutex); 968 969 if (msqid < 0 || msqid >= msginfo.msgmni) { 970 MSG_PRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, 971 msginfo.msgmni)); 972 error = EINVAL; 973 goto unlock; 974 } 975 976 msq = &msqs[msqid]; 977 msqptr = &msq->msq_u; 978 979 if (msqptr->msg_qbytes == 0) { 980 MSG_PRINTF(("no such message queue id\n")); 981 error = EINVAL; 982 goto unlock; 983 } 984 if (msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 985 MSG_PRINTF(("wrong sequence number\n")); 986 error = EINVAL; 987 goto unlock; 988 } 989 990 if ((error = ipcperm(cred, &msqptr->msg_perm, IPC_R))) { 991 MSG_PRINTF(("requester doesn't have read access\n")); 992 goto unlock; 993 } 994 995 msghdr = NULL; 996 while (msghdr == NULL) { 997 if (msgtyp == 0) { 998 msghdr = msqptr->_msg_first; 999 if (msghdr != NULL) { 1000 if (msgsz < msghdr->msg_ts && 1001 (msgflg & MSG_NOERROR) == 0) { 1002 MSG_PRINTF(("first msg on the queue " 1003 "is too big (want %lld, got %d)\n", 1004 (long long)msgsz, msghdr->msg_ts)); 1005 error = E2BIG; 1006 goto unlock; 1007 } 1008 if (msqptr->_msg_first == msqptr->_msg_last) { 1009 msqptr->_msg_first = NULL; 1010 msqptr->_msg_last = NULL; 1011 } else { 1012 msqptr->_msg_first = msghdr->msg_next; 1013 KASSERT(msqptr->_msg_first != NULL); 1014 } 1015 } 1016 } else { 1017 struct __msg *previous; 1018 struct __msg **prev; 1019 1020 for (previous = NULL, prev = &msqptr->_msg_first; 1021 (msghdr = *prev) != NULL; 1022 previous = msghdr, prev = &msghdr->msg_next) { 1023 /* 1024 * Is this message's type an exact match or is 1025 * this message's type less than or equal to 1026 * the absolute value of a negative msgtyp? 1027 * Note that the second half of this test can 1028 * NEVER be true if msgtyp is positive since 1029 * msg_type is always positive! 1030 */ 1031 1032 if (msgtyp != msghdr->msg_type && 1033 msghdr->msg_type > -msgtyp) 1034 continue; 1035 1036 MSG_PRINTF(("found message type %ld, requested %ld\n", 1037 msghdr->msg_type, msgtyp)); 1038 if (msgsz < msghdr->msg_ts && 1039 (msgflg & MSG_NOERROR) == 0) { 1040 MSG_PRINTF(("requested message on the queue " 1041 "is too big (want %lld, got %d)\n", 1042 (long long)msgsz, msghdr->msg_ts)); 1043 error = E2BIG; 1044 goto unlock; 1045 } 1046 *prev = msghdr->msg_next; 1047 if (msghdr != msqptr->_msg_last) 1048 break; 1049 if (previous == NULL) { 1050 KASSERT(prev == &msqptr->_msg_first); 1051 msqptr->_msg_first = NULL; 1052 msqptr->_msg_last = NULL; 1053 } else { 1054 KASSERT(prev != &msqptr->_msg_first); 1055 msqptr->_msg_last = previous; 1056 } 1057 break; 1058 } 1059 } 1060 1061 /* 1062 * We've either extracted the msghdr for the appropriate 1063 * message or there isn't one. 1064 * If there is one then bail out of this loop. 1065 */ 1066 if (msghdr != NULL) 1067 break; 1068 1069 /* 1070 * Hmph! No message found. Does the user want to wait? 1071 */ 1072 1073 if ((msgflg & IPC_NOWAIT) != 0) { 1074 MSG_PRINTF(("no appropriate message found (msgtyp=%ld)\n", 1075 msgtyp)); 1076 error = ENOMSG; 1077 goto unlock; 1078 } 1079 1080 /* 1081 * Wait for something to happen 1082 */ 1083 1084 msg_waiters++; 1085 MSG_PRINTF(("msgrcv: goodnight\n")); 1086 error = cv_wait_sig(&msq->msq_cv, &msgmutex); 1087 MSG_PRINTF(("msgrcv: good morning (error=%d)\n", error)); 1088 msg_waiters--; 1089 1090 /* 1091 * In case of such state, notify reallocator and 1092 * restart the call. 1093 */ 1094 if (msg_realloc_state) { 1095 cv_broadcast(&msg_realloc_cv); 1096 mutex_exit(&msgmutex); 1097 goto restart; 1098 } 1099 1100 if (error != 0) { 1101 MSG_PRINTF(("msgsnd: interrupted system call\n")); 1102 error = EINTR; 1103 goto unlock; 1104 } 1105 1106 /* 1107 * Make sure that the msq queue still exists 1108 */ 1109 1110 if (msqptr->msg_qbytes == 0 || 1111 msqptr->msg_perm._seq != IPCID_TO_SEQ(msqidr)) { 1112 MSG_PRINTF(("msqid deleted\n")); 1113 error = EIDRM; 1114 goto unlock; 1115 } 1116 } 1117 1118 /* 1119 * Return the message to the user. 1120 * 1121 * First, do the bookkeeping (before we risk being interrupted). 1122 */ 1123 1124 msqptr->_msg_cbytes -= msghdr->msg_ts; 1125 msqptr->msg_qnum--; 1126 msqptr->msg_lrpid = l->l_proc->p_pid; 1127 msqptr->msg_rtime = time_second; 1128 1129 /* 1130 * Make msgsz the actual amount that we'll be returning. 1131 * Note that this effectively truncates the message if it is too long 1132 * (since msgsz is never increased). 1133 */ 1134 1135 MSG_PRINTF(("found a message, msgsz=%lld, msg_ts=%d\n", 1136 (long long)msgsz, msghdr->msg_ts)); 1137 if (msgsz > msghdr->msg_ts) 1138 msgsz = msghdr->msg_ts; 1139 1140 /* 1141 * Return the type to the user. 1142 */ 1143 mutex_exit(&msgmutex); 1144 error = (*put_type)(&msghdr->msg_type, user_msgp, typesz); 1145 mutex_enter(&msgmutex); 1146 if (error != 0) { 1147 MSG_PRINTF(("error (%d) copying out message type\n", error)); 1148 msg_freehdr(msghdr); 1149 cv_broadcast(&msq->msq_cv); 1150 goto unlock; 1151 } 1152 user_msgp += typesz; 1153 1154 /* 1155 * Return the segments to the user 1156 */ 1157 1158 next = msghdr->msg_spot; 1159 for (len = 0; len < msgsz; len += msginfo.msgssz) { 1160 size_t tlen; 1161 KASSERT(next > -1); 1162 KASSERT(next < msginfo.msgseg); 1163 1164 if (msgsz - len > msginfo.msgssz) 1165 tlen = msginfo.msgssz; 1166 else 1167 tlen = msgsz - len; 1168 mutex_exit(&msgmutex); 1169 error = (*put_type)(&msgpool[next * msginfo.msgssz], 1170 user_msgp, tlen); 1171 mutex_enter(&msgmutex); 1172 if (error != 0) { 1173 MSG_PRINTF(("error (%d) copying out message segment\n", 1174 error)); 1175 msg_freehdr(msghdr); 1176 cv_broadcast(&msq->msq_cv); 1177 goto unlock; 1178 } 1179 user_msgp += tlen; 1180 next = msgmaps[next].next; 1181 } 1182 1183 /* 1184 * Done, return the actual number of bytes copied out. 1185 */ 1186 1187 msg_freehdr(msghdr); 1188 cv_broadcast(&msq->msq_cv); 1189 *retval = msgsz; 1190 1191 unlock: 1192 mutex_exit(&msgmutex); 1193 return error; 1194 } 1195 1196 /* 1197 * Sysctl initialization and nodes. 1198 */ 1199 1200 static int 1201 sysctl_ipc_msgmni(SYSCTLFN_ARGS) 1202 { 1203 int newsize, error; 1204 struct sysctlnode node; 1205 node = *rnode; 1206 node.sysctl_data = &newsize; 1207 1208 newsize = msginfo.msgmni; 1209 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1210 if (error || newp == NULL) 1211 return error; 1212 1213 sysctl_unlock(); 1214 error = msgrealloc(newsize, msginfo.msgseg); 1215 sysctl_relock(); 1216 return error; 1217 } 1218 1219 static int 1220 sysctl_ipc_msgseg(SYSCTLFN_ARGS) 1221 { 1222 int newsize, error; 1223 struct sysctlnode node; 1224 node = *rnode; 1225 node.sysctl_data = &newsize; 1226 1227 newsize = msginfo.msgseg; 1228 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1229 if (error || newp == NULL) 1230 return error; 1231 1232 sysctl_unlock(); 1233 error = msgrealloc(msginfo.msgmni, newsize); 1234 sysctl_relock(); 1235 return error; 1236 } 1237 1238 SYSCTL_SETUP(sysctl_ipc_msg_setup, "sysctl kern.ipc subtree setup") 1239 { 1240 const struct sysctlnode *node = NULL; 1241 1242 sysctl_createv(clog, 0, NULL, NULL, 1243 CTLFLAG_PERMANENT, 1244 CTLTYPE_NODE, "kern", NULL, 1245 NULL, 0, NULL, 0, 1246 CTL_KERN, CTL_EOL); 1247 sysctl_createv(clog, 0, NULL, &node, 1248 CTLFLAG_PERMANENT, 1249 CTLTYPE_NODE, "ipc", 1250 SYSCTL_DESCR("SysV IPC options"), 1251 NULL, 0, NULL, 0, 1252 CTL_KERN, KERN_SYSVIPC, CTL_EOL); 1253 1254 if (node == NULL) 1255 return; 1256 1257 sysctl_createv(clog, 0, &node, NULL, 1258 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1259 CTLTYPE_INT, "msgmni", 1260 SYSCTL_DESCR("Max number of message queue identifiers"), 1261 sysctl_ipc_msgmni, 0, &msginfo.msgmni, 0, 1262 CTL_CREATE, CTL_EOL); 1263 sysctl_createv(clog, 0, &node, NULL, 1264 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1265 CTLTYPE_INT, "msgseg", 1266 SYSCTL_DESCR("Max number of number of message segments"), 1267 sysctl_ipc_msgseg, 0, &msginfo.msgseg, 0, 1268 CTL_CREATE, CTL_EOL); 1269 } 1270