1 /* 2 * Copyright (c) 1988 University of Utah. 3 * Copyright (c) 1991, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the Systems Programming Group of the University of Utah Computer 8 * Science Department. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ 39 * 40 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 41 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $ 42 * $DragonFly: src/sys/vm/vm_mmap.c,v 1.20 2004/05/13 17:40:19 dillon Exp $ 43 */ 44 45 /* 46 * Mapped file (mmap) interface to VM 47 */ 48 49 #include <sys/param.h> 50 #include <sys/kernel.h> 51 #include <sys/systm.h> 52 #include <sys/sysproto.h> 53 #include <sys/filedesc.h> 54 #include <sys/kern_syscall.h> 55 #include <sys/proc.h> 56 #include <sys/resource.h> 57 #include <sys/resourcevar.h> 58 #include <sys/vnode.h> 59 #include <sys/fcntl.h> 60 #include <sys/file.h> 61 #include <sys/mman.h> 62 #include <sys/conf.h> 63 #include <sys/stat.h> 64 #include <sys/vmmeter.h> 65 #include <sys/sysctl.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_param.h> 69 #include <sys/lock.h> 70 #include <vm/pmap.h> 71 #include <vm/vm_map.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_page.h> 74 #include <vm/vm_pager.h> 75 #include <vm/vm_pageout.h> 76 #include <vm/vm_extern.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_kern.h> 79 80 #include <sys/file2.h> 81 82 static int max_proc_mmap; 83 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); 84 85 /* 86 * Set the maximum number of vm_map_entry structures per process. Roughly 87 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 88 * of our KVM malloc space still results in generous limits. We want a 89 * default that is good enough to prevent the kernel running out of resources 90 * if attacked from compromised user account but generous enough such that 91 * multi-threaded processes are not unduly inconvenienced. 92 */ 93 94 static void vmmapentry_rsrc_init (void *); 95 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL) 96 97 static void 98 vmmapentry_rsrc_init(void *dummy) 99 { 100 max_proc_mmap = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 101 sizeof(struct vm_map_entry); 102 max_proc_mmap /= 100; 103 } 104 105 /* ARGSUSED */ 106 int 107 sbrk(struct sbrk_args *uap) 108 { 109 /* Not yet implemented */ 110 return (EOPNOTSUPP); 111 } 112 113 /* 114 * sstk_args(int incr) 115 */ 116 /* ARGSUSED */ 117 int 118 sstk(struct sstk_args *uap) 119 { 120 /* Not yet implemented */ 121 return (EOPNOTSUPP); 122 } 123 124 /* 125 * mmap_args(void *addr, size_t len, int prot, int flags, int fd, 126 * long pad, off_t pos) 127 * 128 * Memory Map (mmap) system call. Note that the file offset 129 * and address are allowed to be NOT page aligned, though if 130 * the MAP_FIXED flag it set, both must have the same remainder 131 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 132 * page-aligned, the actual mapping starts at trunc_page(addr) 133 * and the return value is adjusted up by the page offset. 134 * 135 * Generally speaking, only character devices which are themselves 136 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 137 * there would be no cache coherency between a descriptor and a VM mapping 138 * both to the same character device. 139 * 140 * Block devices can be mmap'd no matter what they represent. Cache coherency 141 * is maintained as long as you do not write directly to the underlying 142 * character device. 143 */ 144 145 int 146 kern_mmap(caddr_t uaddr, size_t ulen, int uprot, int uflags, int fd, 147 off_t upos, void **res) 148 { 149 struct thread *td = curthread; 150 struct proc *p = td->td_proc; 151 struct filedesc *fdp = p->p_fd; 152 struct file *fp = NULL; 153 struct vnode *vp; 154 vm_offset_t addr; 155 vm_size_t size, pageoff; 156 vm_prot_t prot, maxprot; 157 void *handle; 158 int flags, error; 159 int disablexworkaround; 160 off_t pos; 161 struct vmspace *vms = p->p_vmspace; 162 vm_object_t obj; 163 164 KKASSERT(p); 165 166 addr = (vm_offset_t) uaddr; 167 size = ulen; 168 prot = uprot & VM_PROT_ALL; 169 flags = uflags; 170 pos = upos; 171 172 /* make sure mapping fits into numeric range etc */ 173 if ((ssize_t) ulen < 0 || 174 ((flags & MAP_ANON) && fd != -1)) 175 return (EINVAL); 176 177 if (flags & MAP_STACK) { 178 if ((fd != -1) || 179 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 180 return (EINVAL); 181 flags |= MAP_ANON; 182 pos = 0; 183 } 184 185 /* 186 * Align the file position to a page boundary, 187 * and save its page offset component. 188 */ 189 pageoff = (pos & PAGE_MASK); 190 pos -= pageoff; 191 192 /* Adjust size for rounding (on both ends). */ 193 size += pageoff; /* low end... */ 194 size = (vm_size_t) round_page(size); /* hi end */ 195 196 /* 197 * Check for illegal addresses. Watch out for address wrap... Note 198 * that VM_*_ADDRESS are not constants due to casts (argh). 199 */ 200 if (flags & MAP_FIXED) { 201 /* 202 * The specified address must have the same remainder 203 * as the file offset taken modulo PAGE_SIZE, so it 204 * should be aligned after adjustment by pageoff. 205 */ 206 addr -= pageoff; 207 if (addr & PAGE_MASK) 208 return (EINVAL); 209 /* Address range must be all in user VM space. */ 210 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 211 return (EINVAL); 212 #ifndef i386 213 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 214 return (EINVAL); 215 #endif 216 if (addr + size < addr) 217 return (EINVAL); 218 } 219 /* 220 * XXX for non-fixed mappings where no hint is provided or 221 * the hint would fall in the potential heap space, 222 * place it after the end of the largest possible heap. 223 * 224 * There should really be a pmap call to determine a reasonable 225 * location. 226 */ 227 else if (addr == 0 || 228 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 229 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) 230 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 231 232 if (flags & MAP_ANON) { 233 /* 234 * Mapping blank space is trivial. 235 */ 236 handle = NULL; 237 maxprot = VM_PROT_ALL; 238 pos = 0; 239 } else { 240 /* 241 * Mapping file, get fp for validation. Obtain vnode and make 242 * sure it is of appropriate type. 243 */ 244 if (((unsigned) fd) >= fdp->fd_nfiles || 245 (fp = fdp->fd_ofiles[fd]) == NULL) 246 return (EBADF); 247 if (fp->f_type != DTYPE_VNODE) 248 return (EINVAL); 249 /* 250 * POSIX shared-memory objects are defined to have 251 * kernel persistence, and are not defined to support 252 * read(2)/write(2) -- or even open(2). Thus, we can 253 * use MAP_ASYNC to trade on-disk coherence for speed. 254 * The shm_open(3) library routine turns on the FPOSIXSHM 255 * flag to request this behavior. 256 */ 257 if (fp->f_flag & FPOSIXSHM) 258 flags |= MAP_NOSYNC; 259 vp = (struct vnode *) fp->f_data; 260 if (vp->v_type != VREG && vp->v_type != VCHR) 261 return (EINVAL); 262 if (vp->v_type == VREG) { 263 /* 264 * Get the proper underlying object 265 */ 266 if (VOP_GETVOBJECT(vp, &obj) != 0) 267 return (EINVAL); 268 vp = (struct vnode*)obj->handle; 269 } 270 271 /* 272 * don't let the descriptor disappear on us if we block 273 */ 274 fhold(fp); 275 276 /* 277 * XXX hack to handle use of /dev/zero to map anon memory (ala 278 * SunOS). 279 */ 280 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 281 handle = NULL; 282 maxprot = VM_PROT_ALL; 283 flags |= MAP_ANON; 284 pos = 0; 285 } else { 286 /* 287 * cdevs does not provide private mappings of any kind. 288 */ 289 /* 290 * However, for XIG X server to continue to work, 291 * we should allow the superuser to do it anyway. 292 * We only allow it at securelevel < 1. 293 * (Because the XIG X server writes directly to video 294 * memory via /dev/mem, it should never work at any 295 * other securelevel. 296 * XXX this will have to go 297 */ 298 if (securelevel >= 1) 299 disablexworkaround = 1; 300 else 301 disablexworkaround = suser(td); 302 if (vp->v_type == VCHR && disablexworkaround && 303 (flags & (MAP_PRIVATE|MAP_COPY))) { 304 error = EINVAL; 305 goto done; 306 } 307 /* 308 * Ensure that file and memory protections are 309 * compatible. Note that we only worry about 310 * writability if mapping is shared; in this case, 311 * current and max prot are dictated by the open file. 312 * XXX use the vnode instead? Problem is: what 313 * credentials do we use for determination? What if 314 * proc does a setuid? 315 */ 316 maxprot = VM_PROT_EXECUTE; /* ??? */ 317 if (fp->f_flag & FREAD) { 318 maxprot |= VM_PROT_READ; 319 } else if (prot & PROT_READ) { 320 error = EACCES; 321 goto done; 322 } 323 /* 324 * If we are sharing potential changes (either via 325 * MAP_SHARED or via the implicit sharing of character 326 * device mappings), and we are trying to get write 327 * permission although we opened it without asking 328 * for it, bail out. Check for superuser, only if 329 * we're at securelevel < 1, to allow the XIG X server 330 * to continue to work. 331 */ 332 333 if ((flags & MAP_SHARED) != 0 || 334 (vp->v_type == VCHR && disablexworkaround)) { 335 if ((fp->f_flag & FWRITE) != 0) { 336 struct vattr va; 337 if ((error = VOP_GETATTR(vp, &va, td))) { 338 goto done; 339 } 340 if ((va.va_flags & 341 (IMMUTABLE|APPEND)) == 0) { 342 maxprot |= VM_PROT_WRITE; 343 } else if (prot & PROT_WRITE) { 344 error = EPERM; 345 goto done; 346 } 347 } else if ((prot & PROT_WRITE) != 0) { 348 error = EACCES; 349 goto done; 350 } 351 } else { 352 maxprot |= VM_PROT_WRITE; 353 } 354 handle = (void *)vp; 355 } 356 } 357 358 /* 359 * Do not allow more then a certain number of vm_map_entry structures 360 * per process. Scale with the number of rforks sharing the map 361 * to make the limit reasonable for threads. 362 */ 363 if (max_proc_mmap && 364 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) { 365 error = ENOMEM; 366 goto done; 367 } 368 369 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 370 flags, handle, pos); 371 if (error == 0) 372 *res = (void *)(addr + pageoff); 373 done: 374 if (fp) 375 fdrop(fp, td); 376 return (error); 377 } 378 379 int 380 mmap(struct mmap_args *uap) 381 { 382 int error; 383 384 error = kern_mmap(uap->addr, uap->len, uap->prot, uap->flags, 385 uap->fd, uap->pos, &uap->sysmsg_resultp); 386 387 return (error); 388 } 389 390 /* 391 * msync_args(void *addr, int len, int flags) 392 */ 393 int 394 msync(struct msync_args *uap) 395 { 396 struct proc *p = curproc; 397 vm_offset_t addr; 398 vm_size_t size, pageoff; 399 int flags; 400 vm_map_t map; 401 int rv; 402 403 addr = (vm_offset_t) uap->addr; 404 size = uap->len; 405 flags = uap->flags; 406 407 pageoff = (addr & PAGE_MASK); 408 addr -= pageoff; 409 size += pageoff; 410 size = (vm_size_t) round_page(size); 411 if (addr + size < addr) 412 return(EINVAL); 413 414 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 415 return (EINVAL); 416 417 map = &p->p_vmspace->vm_map; 418 419 /* 420 * XXX Gak! If size is zero we are supposed to sync "all modified 421 * pages with the region containing addr". Unfortunately, we don't 422 * really keep track of individual mmaps so we approximate by flushing 423 * the range of the map entry containing addr. This can be incorrect 424 * if the region splits or is coalesced with a neighbor. 425 */ 426 if (size == 0) { 427 vm_map_entry_t entry; 428 429 vm_map_lock_read(map); 430 rv = vm_map_lookup_entry(map, addr, &entry); 431 vm_map_unlock_read(map); 432 if (rv == FALSE) 433 return (EINVAL); 434 addr = entry->start; 435 size = entry->end - entry->start; 436 } 437 438 /* 439 * Clean the pages and interpret the return value. 440 */ 441 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 442 (flags & MS_INVALIDATE) != 0); 443 444 switch (rv) { 445 case KERN_SUCCESS: 446 break; 447 case KERN_INVALID_ADDRESS: 448 return (EINVAL); /* Sun returns ENOMEM? */ 449 case KERN_FAILURE: 450 return (EIO); 451 default: 452 return (EINVAL); 453 } 454 455 return (0); 456 } 457 458 /* 459 * munmap_args(void *addr, size_t len) 460 */ 461 int 462 munmap(struct munmap_args *uap) 463 { 464 struct proc *p = curproc; 465 vm_offset_t addr; 466 vm_size_t size, pageoff; 467 vm_map_t map; 468 469 addr = (vm_offset_t) uap->addr; 470 size = uap->len; 471 472 pageoff = (addr & PAGE_MASK); 473 addr -= pageoff; 474 size += pageoff; 475 size = (vm_size_t) round_page(size); 476 if (addr + size < addr) 477 return(EINVAL); 478 479 if (size == 0) 480 return (0); 481 482 /* 483 * Check for illegal addresses. Watch out for address wrap... Note 484 * that VM_*_ADDRESS are not constants due to casts (argh). 485 */ 486 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 487 return (EINVAL); 488 #ifndef i386 489 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 490 return (EINVAL); 491 #endif 492 map = &p->p_vmspace->vm_map; 493 /* 494 * Make sure entire range is allocated. 495 */ 496 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) 497 return (EINVAL); 498 /* returns nothing but KERN_SUCCESS anyway */ 499 (void) vm_map_remove(map, addr, addr + size); 500 return (0); 501 } 502 503 #if 0 504 void 505 munmapfd(p, fd) 506 struct proc *p; 507 int fd; 508 { 509 /* 510 * XXX should unmap any regions mapped to this file 511 */ 512 p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED; 513 } 514 #endif 515 516 /* 517 * mprotect_args(const void *addr, size_t len, int prot) 518 */ 519 int 520 mprotect(struct mprotect_args *uap) 521 { 522 struct proc *p = curproc; 523 vm_offset_t addr; 524 vm_size_t size, pageoff; 525 vm_prot_t prot; 526 527 addr = (vm_offset_t) uap->addr; 528 size = uap->len; 529 prot = uap->prot & VM_PROT_ALL; 530 #if defined(VM_PROT_READ_IS_EXEC) 531 if (prot & VM_PROT_READ) 532 prot |= VM_PROT_EXECUTE; 533 #endif 534 535 pageoff = (addr & PAGE_MASK); 536 addr -= pageoff; 537 size += pageoff; 538 size = (vm_size_t) round_page(size); 539 if (addr + size < addr) 540 return(EINVAL); 541 542 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, 543 FALSE)) { 544 case KERN_SUCCESS: 545 return (0); 546 case KERN_PROTECTION_FAILURE: 547 return (EACCES); 548 } 549 return (EINVAL); 550 } 551 552 /* 553 * minherit_args(void *addr, size_t len, int inherit) 554 */ 555 int 556 minherit(struct minherit_args *uap) 557 { 558 struct proc *p = curproc; 559 vm_offset_t addr; 560 vm_size_t size, pageoff; 561 vm_inherit_t inherit; 562 563 addr = (vm_offset_t)uap->addr; 564 size = uap->len; 565 inherit = uap->inherit; 566 567 pageoff = (addr & PAGE_MASK); 568 addr -= pageoff; 569 size += pageoff; 570 size = (vm_size_t) round_page(size); 571 if (addr + size < addr) 572 return(EINVAL); 573 574 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size, 575 inherit)) { 576 case KERN_SUCCESS: 577 return (0); 578 case KERN_PROTECTION_FAILURE: 579 return (EACCES); 580 } 581 return (EINVAL); 582 } 583 584 /* 585 * madvise_args(void *addr, size_t len, int behav) 586 */ 587 /* ARGSUSED */ 588 int 589 madvise(struct madvise_args *uap) 590 { 591 struct proc *p = curproc; 592 vm_offset_t start, end; 593 594 /* 595 * Check for illegal behavior 596 */ 597 if (uap->behav < 0 || uap->behav > MADV_CORE) 598 return (EINVAL); 599 /* 600 * Check for illegal addresses. Watch out for address wrap... Note 601 * that VM_*_ADDRESS are not constants due to casts (argh). 602 */ 603 if (VM_MAXUSER_ADDRESS > 0 && 604 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS) 605 return (EINVAL); 606 #ifndef i386 607 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS) 608 return (EINVAL); 609 #endif 610 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) 611 return (EINVAL); 612 613 /* 614 * Since this routine is only advisory, we default to conservative 615 * behavior. 616 */ 617 start = trunc_page((vm_offset_t) uap->addr); 618 end = round_page((vm_offset_t) uap->addr + uap->len); 619 620 if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav)) 621 return (EINVAL); 622 return (0); 623 } 624 625 /* 626 * mincore_args(const void *addr, size_t len, char *vec) 627 */ 628 /* ARGSUSED */ 629 int 630 mincore(struct mincore_args *uap) 631 { 632 struct proc *p = curproc; 633 vm_offset_t addr, first_addr; 634 vm_offset_t end, cend; 635 pmap_t pmap; 636 vm_map_t map; 637 char *vec; 638 int error; 639 int vecindex, lastvecindex; 640 vm_map_entry_t current; 641 vm_map_entry_t entry; 642 int mincoreinfo; 643 unsigned int timestamp; 644 645 /* 646 * Make sure that the addresses presented are valid for user 647 * mode. 648 */ 649 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 650 end = addr + (vm_size_t)round_page(uap->len); 651 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS) 652 return (EINVAL); 653 if (end < addr) 654 return (EINVAL); 655 656 /* 657 * Address of byte vector 658 */ 659 vec = uap->vec; 660 661 map = &p->p_vmspace->vm_map; 662 pmap = vmspace_pmap(p->p_vmspace); 663 664 vm_map_lock_read(map); 665 RestartScan: 666 timestamp = map->timestamp; 667 668 if (!vm_map_lookup_entry(map, addr, &entry)) 669 entry = entry->next; 670 671 /* 672 * Do this on a map entry basis so that if the pages are not 673 * in the current processes address space, we can easily look 674 * up the pages elsewhere. 675 */ 676 lastvecindex = -1; 677 for(current = entry; 678 (current != &map->header) && (current->start < end); 679 current = current->next) { 680 681 /* 682 * ignore submaps (for now) or null objects 683 */ 684 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 685 current->object.vm_object == NULL) 686 continue; 687 688 /* 689 * limit this scan to the current map entry and the 690 * limits for the mincore call 691 */ 692 if (addr < current->start) 693 addr = current->start; 694 cend = current->end; 695 if (cend > end) 696 cend = end; 697 698 /* 699 * scan this entry one page at a time 700 */ 701 while (addr < cend) { 702 /* 703 * Check pmap first, it is likely faster, also 704 * it can provide info as to whether we are the 705 * one referencing or modifying the page. 706 */ 707 mincoreinfo = pmap_mincore(pmap, addr); 708 if (!mincoreinfo) { 709 vm_pindex_t pindex; 710 vm_ooffset_t offset; 711 vm_page_t m; 712 int s; 713 714 /* 715 * calculate the page index into the object 716 */ 717 offset = current->offset + (addr - current->start); 718 pindex = OFF_TO_IDX(offset); 719 720 /* 721 * if the page is resident, then gather 722 * information about it. spl protection is 723 * required to maintain the object 724 * association. And XXX what if the page is 725 * busy? What's the deal with that? 726 */ 727 s = splvm(); 728 m = vm_page_lookup(current->object.vm_object, 729 pindex); 730 if (m && m->valid) { 731 mincoreinfo = MINCORE_INCORE; 732 if (m->dirty || 733 pmap_is_modified(m)) 734 mincoreinfo |= MINCORE_MODIFIED_OTHER; 735 if ((m->flags & PG_REFERENCED) || 736 pmap_ts_referenced(m)) { 737 vm_page_flag_set(m, PG_REFERENCED); 738 mincoreinfo |= MINCORE_REFERENCED_OTHER; 739 } 740 } 741 splx(s); 742 } 743 744 /* 745 * subyte may page fault. In case it needs to modify 746 * the map, we release the lock. 747 */ 748 vm_map_unlock_read(map); 749 750 /* 751 * calculate index into user supplied byte vector 752 */ 753 vecindex = OFF_TO_IDX(addr - first_addr); 754 755 /* 756 * If we have skipped map entries, we need to make sure that 757 * the byte vector is zeroed for those skipped entries. 758 */ 759 while((lastvecindex + 1) < vecindex) { 760 error = subyte( vec + lastvecindex, 0); 761 if (error) { 762 return (EFAULT); 763 } 764 ++lastvecindex; 765 } 766 767 /* 768 * Pass the page information to the user 769 */ 770 error = subyte( vec + vecindex, mincoreinfo); 771 if (error) { 772 return (EFAULT); 773 } 774 775 /* 776 * If the map has changed, due to the subyte, the previous 777 * output may be invalid. 778 */ 779 vm_map_lock_read(map); 780 if (timestamp != map->timestamp) 781 goto RestartScan; 782 783 lastvecindex = vecindex; 784 addr += PAGE_SIZE; 785 } 786 } 787 788 /* 789 * subyte may page fault. In case it needs to modify 790 * the map, we release the lock. 791 */ 792 vm_map_unlock_read(map); 793 794 /* 795 * Zero the last entries in the byte vector. 796 */ 797 vecindex = OFF_TO_IDX(end - first_addr); 798 while((lastvecindex + 1) < vecindex) { 799 error = subyte( vec + lastvecindex, 0); 800 if (error) { 801 return (EFAULT); 802 } 803 ++lastvecindex; 804 } 805 806 /* 807 * If the map has changed, due to the subyte, the previous 808 * output may be invalid. 809 */ 810 vm_map_lock_read(map); 811 if (timestamp != map->timestamp) 812 goto RestartScan; 813 vm_map_unlock_read(map); 814 815 return (0); 816 } 817 818 /* 819 * mlock_args(const void *addr, size_t len) 820 */ 821 int 822 mlock(struct mlock_args *uap) 823 { 824 vm_offset_t addr; 825 vm_size_t size, pageoff; 826 int error; 827 struct proc *p = curproc; 828 829 addr = (vm_offset_t) uap->addr; 830 size = uap->len; 831 832 pageoff = (addr & PAGE_MASK); 833 addr -= pageoff; 834 size += pageoff; 835 size = (vm_size_t) round_page(size); 836 837 /* disable wrap around */ 838 if (addr + size < addr) 839 return (EINVAL); 840 841 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) 842 return (EAGAIN); 843 844 #ifdef pmap_wired_count 845 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 846 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 847 return (ENOMEM); 848 #else 849 error = suser_cred(p->p_ucred, 0); 850 if (error) 851 return (error); 852 #endif 853 854 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE); 855 return (error == KERN_SUCCESS ? 0 : ENOMEM); 856 } 857 858 /* 859 * mlockall_args(int how) 860 */ 861 int 862 mlockall(struct mlockall_args *uap) 863 { 864 return 0; 865 } 866 867 /* 868 * munlockall_args(void) 869 */ 870 int 871 munlockall(struct munlockall_args *uap) 872 { 873 return 0; 874 } 875 876 /* 877 * munlock_args(const void *addr, size_t len) 878 */ 879 int 880 munlock(struct munlock_args *uap) 881 { 882 struct thread *td = curthread; 883 struct proc *p = td->td_proc; 884 vm_offset_t addr; 885 vm_size_t size, pageoff; 886 int error; 887 888 addr = (vm_offset_t) uap->addr; 889 size = uap->len; 890 891 pageoff = (addr & PAGE_MASK); 892 addr -= pageoff; 893 size += pageoff; 894 size = (vm_size_t) round_page(size); 895 896 /* disable wrap around */ 897 if (addr + size < addr) 898 return (EINVAL); 899 900 #ifndef pmap_wired_count 901 error = suser(td); 902 if (error) 903 return (error); 904 #endif 905 906 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE); 907 return (error == KERN_SUCCESS ? 0 : ENOMEM); 908 } 909 910 /* 911 * Internal version of mmap. 912 * Currently used by mmap, exec, and sys5 shared memory. 913 * Handle is either a vnode pointer or NULL for MAP_ANON. 914 */ 915 int 916 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 917 vm_prot_t maxprot, int flags, 918 void *handle, 919 vm_ooffset_t foff) 920 { 921 boolean_t fitit; 922 vm_object_t object; 923 struct vnode *vp = NULL; 924 objtype_t type; 925 int rv = KERN_SUCCESS; 926 vm_ooffset_t objsize; 927 int docow; 928 struct thread *td = curthread; /* XXX */ 929 struct proc *p = td->td_proc; 930 931 KKASSERT(p); 932 933 if (size == 0) 934 return (0); 935 936 objsize = size = round_page(size); 937 938 if (p->p_vmspace->vm_map.size + size > 939 p->p_rlimit[RLIMIT_VMEM].rlim_cur) { 940 return(ENOMEM); 941 } 942 943 /* 944 * We currently can only deal with page aligned file offsets. 945 * The check is here rather than in the syscall because the 946 * kernel calls this function internally for other mmaping 947 * operations (such as in exec) and non-aligned offsets will 948 * cause pmap inconsistencies...so we want to be sure to 949 * disallow this in all cases. 950 */ 951 if (foff & PAGE_MASK) 952 return (EINVAL); 953 954 if ((flags & MAP_FIXED) == 0) { 955 fitit = TRUE; 956 *addr = round_page(*addr); 957 } else { 958 if (*addr != trunc_page(*addr)) 959 return (EINVAL); 960 fitit = FALSE; 961 (void) vm_map_remove(map, *addr, *addr + size); 962 } 963 964 /* 965 * Lookup/allocate object. 966 */ 967 if (flags & MAP_ANON) { 968 type = OBJT_DEFAULT; 969 /* 970 * Unnamed anonymous regions always start at 0. 971 */ 972 if (handle == 0) 973 foff = 0; 974 } else { 975 vp = (struct vnode *) handle; 976 if (vp->v_type == VCHR) { 977 type = OBJT_DEVICE; 978 handle = (void *)(intptr_t)vp->v_rdev; 979 } else { 980 struct vattr vat; 981 int error; 982 983 error = VOP_GETATTR(vp, &vat, td); 984 if (error) 985 return (error); 986 objsize = round_page(vat.va_size); 987 type = OBJT_VNODE; 988 /* 989 * if it is a regular file without any references 990 * we do not need to sync it. 991 */ 992 if (vp->v_type == VREG && vat.va_nlink == 0) { 993 flags |= MAP_NOSYNC; 994 } 995 } 996 } 997 998 if (handle == NULL) { 999 object = NULL; 1000 docow = 0; 1001 } else { 1002 object = vm_pager_allocate(type, 1003 handle, objsize, prot, foff); 1004 if (object == NULL) 1005 return (type == OBJT_DEVICE ? EINVAL : ENOMEM); 1006 docow = MAP_PREFAULT_PARTIAL; 1007 } 1008 1009 /* 1010 * Force device mappings to be shared. 1011 */ 1012 if (type == OBJT_DEVICE || type == OBJT_PHYS) { 1013 flags &= ~(MAP_PRIVATE|MAP_COPY); 1014 flags |= MAP_SHARED; 1015 } 1016 1017 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1018 docow |= MAP_COPY_ON_WRITE; 1019 if (flags & MAP_NOSYNC) 1020 docow |= MAP_DISABLE_SYNCER; 1021 if (flags & MAP_NOCORE) 1022 docow |= MAP_DISABLE_COREDUMP; 1023 1024 #if defined(VM_PROT_READ_IS_EXEC) 1025 if (prot & VM_PROT_READ) 1026 prot |= VM_PROT_EXECUTE; 1027 1028 if (maxprot & VM_PROT_READ) 1029 maxprot |= VM_PROT_EXECUTE; 1030 #endif 1031 1032 if (fitit) { 1033 *addr = pmap_addr_hint(object, *addr, size); 1034 } 1035 1036 if (flags & MAP_STACK) 1037 rv = vm_map_stack (map, *addr, size, prot, 1038 maxprot, docow); 1039 else 1040 rv = vm_map_find(map, object, foff, addr, size, fitit, 1041 prot, maxprot, docow); 1042 1043 if (rv != KERN_SUCCESS) { 1044 /* 1045 * Lose the object reference. Will destroy the 1046 * object if it's an unnamed anonymous mapping 1047 * or named anonymous without other references. 1048 */ 1049 vm_object_deallocate(object); 1050 goto out; 1051 } 1052 1053 /* 1054 * Shared memory is also shared with children. 1055 */ 1056 if (flags & (MAP_SHARED|MAP_INHERIT)) { 1057 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1058 if (rv != KERN_SUCCESS) { 1059 (void) vm_map_remove(map, *addr, *addr + size); 1060 goto out; 1061 } 1062 } 1063 out: 1064 switch (rv) { 1065 case KERN_SUCCESS: 1066 return (0); 1067 case KERN_INVALID_ADDRESS: 1068 case KERN_NO_SPACE: 1069 return (ENOMEM); 1070 case KERN_PROTECTION_FAILURE: 1071 return (EACCES); 1072 default: 1073 return (EINVAL); 1074 } 1075 } 1076