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 1.23 2006/01/13 20:45:30 swildner 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 #include <sys/thread2.h> 82 83 static int max_proc_mmap; 84 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); 85 86 /* 87 * Set the maximum number of vm_map_entry structures per process. Roughly 88 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 89 * of our KVM malloc space still results in generous limits. We want a 90 * default that is good enough to prevent the kernel running out of resources 91 * if attacked from compromised user account but generous enough such that 92 * multi-threaded processes are not unduly inconvenienced. 93 */ 94 95 static void vmmapentry_rsrc_init (void *); 96 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL) 97 98 static void 99 vmmapentry_rsrc_init(void *dummy) 100 { 101 max_proc_mmap = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 102 sizeof(struct vm_map_entry); 103 max_proc_mmap /= 100; 104 } 105 106 /* ARGSUSED */ 107 int 108 sbrk(struct sbrk_args *uap) 109 { 110 /* Not yet implemented */ 111 return (EOPNOTSUPP); 112 } 113 114 /* 115 * sstk_args(int incr) 116 */ 117 /* ARGSUSED */ 118 int 119 sstk(struct sstk_args *uap) 120 { 121 /* Not yet implemented */ 122 return (EOPNOTSUPP); 123 } 124 125 /* 126 * mmap_args(void *addr, size_t len, int prot, int flags, int fd, 127 * long pad, off_t pos) 128 * 129 * Memory Map (mmap) system call. Note that the file offset 130 * and address are allowed to be NOT page aligned, though if 131 * the MAP_FIXED flag it set, both must have the same remainder 132 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 133 * page-aligned, the actual mapping starts at trunc_page(addr) 134 * and the return value is adjusted up by the page offset. 135 * 136 * Generally speaking, only character devices which are themselves 137 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 138 * there would be no cache coherency between a descriptor and a VM mapping 139 * both to the same character device. 140 * 141 * Block devices can be mmap'd no matter what they represent. Cache coherency 142 * is maintained as long as you do not write directly to the underlying 143 * character device. 144 */ 145 146 int 147 kern_mmap(caddr_t uaddr, size_t ulen, int uprot, int uflags, int fd, 148 off_t upos, void **res) 149 { 150 struct thread *td = curthread; 151 struct proc *p = td->td_proc; 152 struct filedesc *fdp = p->p_fd; 153 struct file *fp = NULL; 154 struct vnode *vp; 155 vm_offset_t addr; 156 vm_size_t size, pageoff; 157 vm_prot_t prot, maxprot; 158 void *handle; 159 int flags, error; 160 int disablexworkaround; 161 off_t pos; 162 struct vmspace *vms = p->p_vmspace; 163 vm_object_t obj; 164 165 KKASSERT(p); 166 167 addr = (vm_offset_t) uaddr; 168 size = ulen; 169 prot = uprot & VM_PROT_ALL; 170 flags = uflags; 171 pos = upos; 172 173 /* make sure mapping fits into numeric range etc */ 174 if ((ssize_t) ulen < 0 || 175 ((flags & MAP_ANON) && fd != -1)) 176 return (EINVAL); 177 178 if (flags & MAP_STACK) { 179 if ((fd != -1) || 180 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 181 return (EINVAL); 182 flags |= MAP_ANON; 183 pos = 0; 184 } 185 186 /* 187 * Align the file position to a page boundary, 188 * and save its page offset component. 189 */ 190 pageoff = (pos & PAGE_MASK); 191 pos -= pageoff; 192 193 /* Adjust size for rounding (on both ends). */ 194 size += pageoff; /* low end... */ 195 size = (vm_size_t) round_page(size); /* hi end */ 196 197 /* 198 * Check for illegal addresses. Watch out for address wrap... Note 199 * that VM_*_ADDRESS are not constants due to casts (argh). 200 */ 201 if (flags & MAP_FIXED) { 202 /* 203 * The specified address must have the same remainder 204 * as the file offset taken modulo PAGE_SIZE, so it 205 * should be aligned after adjustment by pageoff. 206 */ 207 addr -= pageoff; 208 if (addr & PAGE_MASK) 209 return (EINVAL); 210 /* Address range must be all in user VM space. */ 211 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 212 return (EINVAL); 213 #ifndef i386 214 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 215 return (EINVAL); 216 #endif 217 if (addr + size < addr) 218 return (EINVAL); 219 } 220 /* 221 * XXX for non-fixed mappings where no hint is provided or 222 * the hint would fall in the potential heap space, 223 * place it after the end of the largest possible heap. 224 * 225 * There should really be a pmap call to determine a reasonable 226 * location. 227 */ 228 else if (addr == 0 || 229 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 230 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) 231 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 232 233 if (flags & MAP_ANON) { 234 /* 235 * Mapping blank space is trivial. 236 */ 237 handle = NULL; 238 maxprot = VM_PROT_ALL; 239 pos = 0; 240 } else { 241 /* 242 * Mapping file, get fp for validation. Obtain vnode and make 243 * sure it is of appropriate type. 244 */ 245 if (((unsigned) fd) >= fdp->fd_nfiles || 246 (fp = fdp->fd_files[fd].fp) == NULL) 247 return (EBADF); 248 if (fp->f_type != DTYPE_VNODE) 249 return (EINVAL); 250 /* 251 * POSIX shared-memory objects are defined to have 252 * kernel persistence, and are not defined to support 253 * read(2)/write(2) -- or even open(2). Thus, we can 254 * use MAP_ASYNC to trade on-disk coherence for speed. 255 * The shm_open(3) library routine turns on the FPOSIXSHM 256 * flag to request this behavior. 257 */ 258 if (fp->f_flag & FPOSIXSHM) 259 flags |= MAP_NOSYNC; 260 vp = (struct vnode *) fp->f_data; 261 if (vp->v_type != VREG && vp->v_type != VCHR) 262 return (EINVAL); 263 if (vp->v_type == VREG) { 264 /* 265 * Get the proper underlying object 266 */ 267 if (VOP_GETVOBJECT(vp, &obj) != 0) 268 return (EINVAL); 269 vp = (struct vnode*)obj->handle; 270 } 271 272 /* 273 * don't let the descriptor disappear on us if we block 274 */ 275 fhold(fp); 276 277 /* 278 * XXX hack to handle use of /dev/zero to map anon memory (ala 279 * SunOS). 280 */ 281 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 282 handle = NULL; 283 maxprot = VM_PROT_ALL; 284 flags |= MAP_ANON; 285 pos = 0; 286 } else { 287 /* 288 * cdevs does not provide private mappings of any kind. 289 */ 290 /* 291 * However, for XIG X server to continue to work, 292 * we should allow the superuser to do it anyway. 293 * We only allow it at securelevel < 1. 294 * (Because the XIG X server writes directly to video 295 * memory via /dev/mem, it should never work at any 296 * other securelevel. 297 * XXX this will have to go 298 */ 299 if (securelevel >= 1) 300 disablexworkaround = 1; 301 else 302 disablexworkaround = suser(td); 303 if (vp->v_type == VCHR && disablexworkaround && 304 (flags & (MAP_PRIVATE|MAP_COPY))) { 305 error = EINVAL; 306 goto done; 307 } 308 /* 309 * Ensure that file and memory protections are 310 * compatible. Note that we only worry about 311 * writability if mapping is shared; in this case, 312 * current and max prot are dictated by the open file. 313 * XXX use the vnode instead? Problem is: what 314 * credentials do we use for determination? What if 315 * proc does a setuid? 316 */ 317 maxprot = VM_PROT_EXECUTE; /* ??? */ 318 if (fp->f_flag & FREAD) { 319 maxprot |= VM_PROT_READ; 320 } else if (prot & PROT_READ) { 321 error = EACCES; 322 goto done; 323 } 324 /* 325 * If we are sharing potential changes (either via 326 * MAP_SHARED or via the implicit sharing of character 327 * device mappings), and we are trying to get write 328 * permission although we opened it without asking 329 * for it, bail out. Check for superuser, only if 330 * we're at securelevel < 1, to allow the XIG X server 331 * to continue to work. 332 */ 333 334 if ((flags & MAP_SHARED) != 0 || 335 (vp->v_type == VCHR && disablexworkaround)) { 336 if ((fp->f_flag & FWRITE) != 0) { 337 struct vattr va; 338 if ((error = VOP_GETATTR(vp, &va, td))) { 339 goto done; 340 } 341 if ((va.va_flags & 342 (IMMUTABLE|APPEND)) == 0) { 343 maxprot |= VM_PROT_WRITE; 344 } else if (prot & PROT_WRITE) { 345 error = EPERM; 346 goto done; 347 } 348 } else if ((prot & PROT_WRITE) != 0) { 349 error = EACCES; 350 goto done; 351 } 352 } else { 353 maxprot |= VM_PROT_WRITE; 354 } 355 handle = (void *)vp; 356 } 357 } 358 359 /* 360 * Do not allow more then a certain number of vm_map_entry structures 361 * per process. Scale with the number of rforks sharing the map 362 * to make the limit reasonable for threads. 363 */ 364 if (max_proc_mmap && 365 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) { 366 error = ENOMEM; 367 goto done; 368 } 369 370 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 371 flags, handle, pos); 372 if (error == 0) 373 *res = (void *)(addr + pageoff); 374 done: 375 if (fp) 376 fdrop(fp, td); 377 return (error); 378 } 379 380 int 381 mmap(struct mmap_args *uap) 382 { 383 int error; 384 385 error = kern_mmap(uap->addr, uap->len, uap->prot, uap->flags, 386 uap->fd, uap->pos, &uap->sysmsg_resultp); 387 388 return (error); 389 } 390 391 /* 392 * msync_args(void *addr, int len, int flags) 393 */ 394 int 395 msync(struct msync_args *uap) 396 { 397 struct proc *p = curproc; 398 vm_offset_t addr; 399 vm_size_t size, pageoff; 400 int flags; 401 vm_map_t map; 402 int rv; 403 404 addr = (vm_offset_t) uap->addr; 405 size = uap->len; 406 flags = uap->flags; 407 408 pageoff = (addr & PAGE_MASK); 409 addr -= pageoff; 410 size += pageoff; 411 size = (vm_size_t) round_page(size); 412 if (addr + size < addr) 413 return(EINVAL); 414 415 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 416 return (EINVAL); 417 418 map = &p->p_vmspace->vm_map; 419 420 /* 421 * XXX Gak! If size is zero we are supposed to sync "all modified 422 * pages with the region containing addr". Unfortunately, we don't 423 * really keep track of individual mmaps so we approximate by flushing 424 * the range of the map entry containing addr. This can be incorrect 425 * if the region splits or is coalesced with a neighbor. 426 */ 427 if (size == 0) { 428 vm_map_entry_t entry; 429 430 vm_map_lock_read(map); 431 rv = vm_map_lookup_entry(map, addr, &entry); 432 vm_map_unlock_read(map); 433 if (rv == FALSE) 434 return (EINVAL); 435 addr = entry->start; 436 size = entry->end - entry->start; 437 } 438 439 /* 440 * Clean the pages and interpret the return value. 441 */ 442 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 443 (flags & MS_INVALIDATE) != 0); 444 445 switch (rv) { 446 case KERN_SUCCESS: 447 break; 448 case KERN_INVALID_ADDRESS: 449 return (EINVAL); /* Sun returns ENOMEM? */ 450 case KERN_FAILURE: 451 return (EIO); 452 default: 453 return (EINVAL); 454 } 455 456 return (0); 457 } 458 459 /* 460 * munmap_args(void *addr, size_t len) 461 */ 462 int 463 munmap(struct munmap_args *uap) 464 { 465 struct proc *p = curproc; 466 vm_offset_t addr; 467 vm_size_t size, pageoff; 468 vm_map_t map; 469 470 addr = (vm_offset_t) uap->addr; 471 size = uap->len; 472 473 pageoff = (addr & PAGE_MASK); 474 addr -= pageoff; 475 size += pageoff; 476 size = (vm_size_t) round_page(size); 477 if (addr + size < addr) 478 return(EINVAL); 479 480 if (size == 0) 481 return (0); 482 483 /* 484 * Check for illegal addresses. Watch out for address wrap... Note 485 * that VM_*_ADDRESS are not constants due to casts (argh). 486 */ 487 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 488 return (EINVAL); 489 #ifndef i386 490 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 491 return (EINVAL); 492 #endif 493 map = &p->p_vmspace->vm_map; 494 /* 495 * Make sure entire range is allocated. 496 */ 497 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) 498 return (EINVAL); 499 /* returns nothing but KERN_SUCCESS anyway */ 500 vm_map_remove(map, addr, addr + size); 501 return (0); 502 } 503 504 #if 0 505 void 506 munmapfd(struct proc *p, int fd) 507 { 508 /* 509 * XXX should unmap any regions mapped to this file 510 */ 511 p->p_fd->fd_files[fd].fileflags &= ~UF_MAPPED; 512 } 513 #endif 514 515 /* 516 * mprotect_args(const void *addr, size_t len, int prot) 517 */ 518 int 519 mprotect(struct mprotect_args *uap) 520 { 521 struct proc *p = curproc; 522 vm_offset_t addr; 523 vm_size_t size, pageoff; 524 vm_prot_t prot; 525 526 addr = (vm_offset_t) uap->addr; 527 size = uap->len; 528 prot = uap->prot & VM_PROT_ALL; 529 #if defined(VM_PROT_READ_IS_EXEC) 530 if (prot & VM_PROT_READ) 531 prot |= VM_PROT_EXECUTE; 532 #endif 533 534 pageoff = (addr & PAGE_MASK); 535 addr -= pageoff; 536 size += pageoff; 537 size = (vm_size_t) round_page(size); 538 if (addr + size < addr) 539 return(EINVAL); 540 541 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, 542 FALSE)) { 543 case KERN_SUCCESS: 544 return (0); 545 case KERN_PROTECTION_FAILURE: 546 return (EACCES); 547 } 548 return (EINVAL); 549 } 550 551 /* 552 * minherit_args(void *addr, size_t len, int inherit) 553 */ 554 int 555 minherit(struct minherit_args *uap) 556 { 557 struct proc *p = curproc; 558 vm_offset_t addr; 559 vm_size_t size, pageoff; 560 vm_inherit_t inherit; 561 562 addr = (vm_offset_t)uap->addr; 563 size = uap->len; 564 inherit = uap->inherit; 565 566 pageoff = (addr & PAGE_MASK); 567 addr -= pageoff; 568 size += pageoff; 569 size = (vm_size_t) round_page(size); 570 if (addr + size < addr) 571 return(EINVAL); 572 573 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size, 574 inherit)) { 575 case KERN_SUCCESS: 576 return (0); 577 case KERN_PROTECTION_FAILURE: 578 return (EACCES); 579 } 580 return (EINVAL); 581 } 582 583 /* 584 * madvise_args(void *addr, size_t len, int behav) 585 */ 586 /* ARGSUSED */ 587 int 588 madvise(struct madvise_args *uap) 589 { 590 struct proc *p = curproc; 591 vm_offset_t start, end; 592 593 /* 594 * Check for illegal behavior 595 */ 596 if (uap->behav < 0 || uap->behav > MADV_CORE) 597 return (EINVAL); 598 /* 599 * Check for illegal addresses. Watch out for address wrap... Note 600 * that VM_*_ADDRESS are not constants due to casts (argh). 601 */ 602 if (VM_MAXUSER_ADDRESS > 0 && 603 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS) 604 return (EINVAL); 605 #ifndef i386 606 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS) 607 return (EINVAL); 608 #endif 609 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) 610 return (EINVAL); 611 612 /* 613 * Since this routine is only advisory, we default to conservative 614 * behavior. 615 */ 616 start = trunc_page((vm_offset_t) uap->addr); 617 end = round_page((vm_offset_t) uap->addr + uap->len); 618 619 if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav)) 620 return (EINVAL); 621 return (0); 622 } 623 624 /* 625 * mincore_args(const void *addr, size_t len, char *vec) 626 */ 627 /* ARGSUSED */ 628 int 629 mincore(struct mincore_args *uap) 630 { 631 struct proc *p = curproc; 632 vm_offset_t addr, first_addr; 633 vm_offset_t end, cend; 634 pmap_t pmap; 635 vm_map_t map; 636 char *vec; 637 int error; 638 int vecindex, lastvecindex; 639 vm_map_entry_t current; 640 vm_map_entry_t entry; 641 int mincoreinfo; 642 unsigned int timestamp; 643 644 /* 645 * Make sure that the addresses presented are valid for user 646 * mode. 647 */ 648 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 649 end = addr + (vm_size_t)round_page(uap->len); 650 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS) 651 return (EINVAL); 652 if (end < addr) 653 return (EINVAL); 654 655 /* 656 * Address of byte vector 657 */ 658 vec = uap->vec; 659 660 map = &p->p_vmspace->vm_map; 661 pmap = vmspace_pmap(p->p_vmspace); 662 663 vm_map_lock_read(map); 664 RestartScan: 665 timestamp = map->timestamp; 666 667 if (!vm_map_lookup_entry(map, addr, &entry)) 668 entry = entry->next; 669 670 /* 671 * Do this on a map entry basis so that if the pages are not 672 * in the current processes address space, we can easily look 673 * up the pages elsewhere. 674 */ 675 lastvecindex = -1; 676 for(current = entry; 677 (current != &map->header) && (current->start < end); 678 current = current->next) { 679 680 /* 681 * ignore submaps (for now) or null objects 682 */ 683 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 684 current->object.vm_object == NULL) 685 continue; 686 687 /* 688 * limit this scan to the current map entry and the 689 * limits for the mincore call 690 */ 691 if (addr < current->start) 692 addr = current->start; 693 cend = current->end; 694 if (cend > end) 695 cend = end; 696 697 /* 698 * scan this entry one page at a time 699 */ 700 while (addr < cend) { 701 /* 702 * Check pmap first, it is likely faster, also 703 * it can provide info as to whether we are the 704 * one referencing or modifying the page. 705 */ 706 mincoreinfo = pmap_mincore(pmap, addr); 707 if (!mincoreinfo) { 708 vm_pindex_t pindex; 709 vm_ooffset_t offset; 710 vm_page_t m; 711 712 /* 713 * calculate the page index into the object 714 */ 715 offset = current->offset + (addr - current->start); 716 pindex = OFF_TO_IDX(offset); 717 718 /* 719 * if the page is resident, then gather 720 * information about it. spl protection is 721 * required to maintain the object 722 * association. And XXX what if the page is 723 * busy? What's the deal with that? 724 */ 725 crit_enter(); 726 m = vm_page_lookup(current->object.vm_object, 727 pindex); 728 if (m && m->valid) { 729 mincoreinfo = MINCORE_INCORE; 730 if (m->dirty || 731 pmap_is_modified(m)) 732 mincoreinfo |= MINCORE_MODIFIED_OTHER; 733 if ((m->flags & PG_REFERENCED) || 734 pmap_ts_referenced(m)) { 735 vm_page_flag_set(m, PG_REFERENCED); 736 mincoreinfo |= MINCORE_REFERENCED_OTHER; 737 } 738 } 739 crit_exit(); 740 } 741 742 /* 743 * subyte may page fault. In case it needs to modify 744 * the map, we release the lock. 745 */ 746 vm_map_unlock_read(map); 747 748 /* 749 * calculate index into user supplied byte vector 750 */ 751 vecindex = OFF_TO_IDX(addr - first_addr); 752 753 /* 754 * If we have skipped map entries, we need to make sure that 755 * the byte vector is zeroed for those skipped entries. 756 */ 757 while((lastvecindex + 1) < vecindex) { 758 error = subyte( vec + lastvecindex, 0); 759 if (error) { 760 return (EFAULT); 761 } 762 ++lastvecindex; 763 } 764 765 /* 766 * Pass the page information to the user 767 */ 768 error = subyte( vec + vecindex, mincoreinfo); 769 if (error) { 770 return (EFAULT); 771 } 772 773 /* 774 * If the map has changed, due to the subyte, the previous 775 * output may be invalid. 776 */ 777 vm_map_lock_read(map); 778 if (timestamp != map->timestamp) 779 goto RestartScan; 780 781 lastvecindex = vecindex; 782 addr += PAGE_SIZE; 783 } 784 } 785 786 /* 787 * subyte may page fault. In case it needs to modify 788 * the map, we release the lock. 789 */ 790 vm_map_unlock_read(map); 791 792 /* 793 * Zero the last entries in the byte vector. 794 */ 795 vecindex = OFF_TO_IDX(end - first_addr); 796 while((lastvecindex + 1) < vecindex) { 797 error = subyte( vec + lastvecindex, 0); 798 if (error) { 799 return (EFAULT); 800 } 801 ++lastvecindex; 802 } 803 804 /* 805 * If the map has changed, due to the subyte, the previous 806 * output may be invalid. 807 */ 808 vm_map_lock_read(map); 809 if (timestamp != map->timestamp) 810 goto RestartScan; 811 vm_map_unlock_read(map); 812 813 return (0); 814 } 815 816 /* 817 * mlock_args(const void *addr, size_t len) 818 */ 819 int 820 mlock(struct mlock_args *uap) 821 { 822 vm_offset_t addr; 823 vm_size_t size, pageoff; 824 int error; 825 struct proc *p = curproc; 826 827 addr = (vm_offset_t) uap->addr; 828 size = uap->len; 829 830 pageoff = (addr & PAGE_MASK); 831 addr -= pageoff; 832 size += pageoff; 833 size = (vm_size_t) round_page(size); 834 835 /* disable wrap around */ 836 if (addr + size < addr) 837 return (EINVAL); 838 839 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) 840 return (EAGAIN); 841 842 #ifdef pmap_wired_count 843 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 844 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 845 return (ENOMEM); 846 #else 847 error = suser_cred(p->p_ucred, 0); 848 if (error) 849 return (error); 850 #endif 851 852 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE); 853 return (error == KERN_SUCCESS ? 0 : ENOMEM); 854 } 855 856 /* 857 * mlockall_args(int how) 858 */ 859 int 860 mlockall(struct mlockall_args *uap) 861 { 862 return 0; 863 } 864 865 /* 866 * munlockall_args(void) 867 */ 868 int 869 munlockall(struct munlockall_args *uap) 870 { 871 return 0; 872 } 873 874 /* 875 * munlock_args(const void *addr, size_t len) 876 */ 877 int 878 munlock(struct munlock_args *uap) 879 { 880 struct thread *td = curthread; 881 struct proc *p = td->td_proc; 882 vm_offset_t addr; 883 vm_size_t size, pageoff; 884 int error; 885 886 addr = (vm_offset_t) uap->addr; 887 size = uap->len; 888 889 pageoff = (addr & PAGE_MASK); 890 addr -= pageoff; 891 size += pageoff; 892 size = (vm_size_t) round_page(size); 893 894 /* disable wrap around */ 895 if (addr + size < addr) 896 return (EINVAL); 897 898 #ifndef pmap_wired_count 899 error = suser(td); 900 if (error) 901 return (error); 902 #endif 903 904 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE); 905 return (error == KERN_SUCCESS ? 0 : ENOMEM); 906 } 907 908 /* 909 * Internal version of mmap. 910 * Currently used by mmap, exec, and sys5 shared memory. 911 * Handle is either a vnode pointer or NULL for MAP_ANON. 912 */ 913 int 914 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 915 vm_prot_t maxprot, int flags, 916 void *handle, 917 vm_ooffset_t foff) 918 { 919 boolean_t fitit; 920 vm_object_t object; 921 struct vnode *vp = NULL; 922 objtype_t type; 923 int rv = KERN_SUCCESS; 924 vm_ooffset_t objsize; 925 int docow; 926 struct thread *td = curthread; /* XXX */ 927 struct proc *p = td->td_proc; 928 929 KKASSERT(p); 930 931 if (size == 0) 932 return (0); 933 934 objsize = size = round_page(size); 935 936 if (p->p_vmspace->vm_map.size + size > 937 p->p_rlimit[RLIMIT_VMEM].rlim_cur) { 938 return(ENOMEM); 939 } 940 941 /* 942 * We currently can only deal with page aligned file offsets. 943 * The check is here rather than in the syscall because the 944 * kernel calls this function internally for other mmaping 945 * operations (such as in exec) and non-aligned offsets will 946 * cause pmap inconsistencies...so we want to be sure to 947 * disallow this in all cases. 948 */ 949 if (foff & PAGE_MASK) 950 return (EINVAL); 951 952 if ((flags & MAP_FIXED) == 0) { 953 fitit = TRUE; 954 *addr = round_page(*addr); 955 } else { 956 if (*addr != trunc_page(*addr)) 957 return (EINVAL); 958 fitit = FALSE; 959 vm_map_remove(map, *addr, *addr + size); 960 } 961 962 /* 963 * Lookup/allocate object. 964 */ 965 if (flags & MAP_ANON) { 966 type = OBJT_DEFAULT; 967 /* 968 * Unnamed anonymous regions always start at 0. 969 */ 970 if (handle == 0) 971 foff = 0; 972 } else { 973 vp = (struct vnode *) handle; 974 if (vp->v_type == VCHR) { 975 type = OBJT_DEVICE; 976 handle = (void *)(intptr_t)vp->v_rdev; 977 } else { 978 struct vattr vat; 979 int error; 980 981 error = VOP_GETATTR(vp, &vat, td); 982 if (error) 983 return (error); 984 objsize = round_page(vat.va_size); 985 type = OBJT_VNODE; 986 /* 987 * if it is a regular file without any references 988 * we do not need to sync it. 989 */ 990 if (vp->v_type == VREG && vat.va_nlink == 0) { 991 flags |= MAP_NOSYNC; 992 } 993 } 994 } 995 996 if (handle == NULL) { 997 object = NULL; 998 docow = 0; 999 } else { 1000 object = vm_pager_allocate(type, 1001 handle, objsize, prot, foff); 1002 if (object == NULL) 1003 return (type == OBJT_DEVICE ? EINVAL : ENOMEM); 1004 docow = MAP_PREFAULT_PARTIAL; 1005 } 1006 1007 /* 1008 * Force device mappings to be shared. 1009 */ 1010 if (type == OBJT_DEVICE || type == OBJT_PHYS) { 1011 flags &= ~(MAP_PRIVATE|MAP_COPY); 1012 flags |= MAP_SHARED; 1013 } 1014 1015 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1016 docow |= MAP_COPY_ON_WRITE; 1017 if (flags & MAP_NOSYNC) 1018 docow |= MAP_DISABLE_SYNCER; 1019 if (flags & MAP_NOCORE) 1020 docow |= MAP_DISABLE_COREDUMP; 1021 1022 #if defined(VM_PROT_READ_IS_EXEC) 1023 if (prot & VM_PROT_READ) 1024 prot |= VM_PROT_EXECUTE; 1025 1026 if (maxprot & VM_PROT_READ) 1027 maxprot |= VM_PROT_EXECUTE; 1028 #endif 1029 1030 if (fitit) { 1031 *addr = pmap_addr_hint(object, *addr, size); 1032 } 1033 1034 if (flags & MAP_STACK) 1035 rv = vm_map_stack (map, *addr, size, prot, 1036 maxprot, docow); 1037 else 1038 rv = vm_map_find(map, object, foff, addr, size, fitit, 1039 prot, maxprot, docow); 1040 1041 if (rv != KERN_SUCCESS) { 1042 /* 1043 * Lose the object reference. Will destroy the 1044 * object if it's an unnamed anonymous mapping 1045 * or named anonymous without other references. 1046 */ 1047 vm_object_deallocate(object); 1048 goto out; 1049 } 1050 1051 /* 1052 * Shared memory is also shared with children. 1053 */ 1054 if (flags & (MAP_SHARED|MAP_INHERIT)) { 1055 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1056 if (rv != KERN_SUCCESS) { 1057 vm_map_remove(map, *addr, *addr + size); 1058 goto out; 1059 } 1060 } 1061 out: 1062 switch (rv) { 1063 case KERN_SUCCESS: 1064 return (0); 1065 case KERN_INVALID_ADDRESS: 1066 case KERN_NO_SPACE: 1067 return (ENOMEM); 1068 case KERN_PROTECTION_FAILURE: 1069 return (EACCES); 1070 default: 1071 return (EINVAL); 1072 } 1073 } 1074