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 * $FreeBSD$ 42 */ 43 44 /* 45 * Mapped file (mmap) interface to VM 46 */ 47 48 #include "opt_bleed.h" 49 #include "opt_compat.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/kernel.h> 54 #include <sys/lock.h> 55 #include <sys/mutex.h> 56 #include <sys/sysproto.h> 57 #include <sys/filedesc.h> 58 #include <sys/proc.h> 59 #include <sys/vnode.h> 60 #include <sys/fcntl.h> 61 #include <sys/file.h> 62 #include <sys/mman.h> 63 #include <sys/conf.h> 64 #include <sys/stat.h> 65 #include <sys/vmmeter.h> 66 #include <sys/sysctl.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_param.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 #ifndef _SYS_SYSPROTO_H_ 81 struct sbrk_args { 82 int incr; 83 }; 84 #endif 85 86 static int max_proc_mmap; 87 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); 88 89 /* 90 * Set the maximum number of vm_map_entry structures per process. Roughly 91 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 92 * of our KVM malloc space still results in generous limits. We want a 93 * default that is good enough to prevent the kernel running out of resources 94 * if attacked from compromised user account but generous enough such that 95 * multi-threaded processes are not unduly inconvenienced. 96 */ 97 98 static void vmmapentry_rsrc_init __P((void *)); 99 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL) 100 101 static void 102 vmmapentry_rsrc_init(dummy) 103 void *dummy; 104 { 105 max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry); 106 max_proc_mmap /= 100; 107 } 108 109 /* 110 * MPSAFE 111 */ 112 /* ARGSUSED */ 113 int 114 sbrk(td, uap) 115 struct thread *td; 116 struct sbrk_args *uap; 117 { 118 /* Not yet implemented */ 119 /* mtx_lock(&Giant); */ 120 /* mtx_unlock(&Giant); */ 121 return (EOPNOTSUPP); 122 } 123 124 #ifndef _SYS_SYSPROTO_H_ 125 struct sstk_args { 126 int incr; 127 }; 128 #endif 129 130 /* 131 * MPSAFE 132 */ 133 /* ARGSUSED */ 134 int 135 sstk(td, uap) 136 struct thread *td; 137 struct sstk_args *uap; 138 { 139 /* Not yet implemented */ 140 /* mtx_lock(&Giant); */ 141 /* mtx_unlock(&Giant); */ 142 return (EOPNOTSUPP); 143 } 144 145 #if defined(COMPAT_43) || defined(COMPAT_SUNOS) 146 #ifndef _SYS_SYSPROTO_H_ 147 struct getpagesize_args { 148 int dummy; 149 }; 150 #endif 151 152 /* ARGSUSED */ 153 int 154 ogetpagesize(td, uap) 155 struct thread *td; 156 struct getpagesize_args *uap; 157 { 158 /* MP SAFE */ 159 td->td_retval[0] = PAGE_SIZE; 160 return (0); 161 } 162 #endif /* COMPAT_43 || COMPAT_SUNOS */ 163 164 165 /* 166 * Memory Map (mmap) system call. Note that the file offset 167 * and address are allowed to be NOT page aligned, though if 168 * the MAP_FIXED flag it set, both must have the same remainder 169 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 170 * page-aligned, the actual mapping starts at trunc_page(addr) 171 * and the return value is adjusted up by the page offset. 172 * 173 * Generally speaking, only character devices which are themselves 174 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 175 * there would be no cache coherency between a descriptor and a VM mapping 176 * both to the same character device. 177 * 178 * Block devices can be mmap'd no matter what they represent. Cache coherency 179 * is maintained as long as you do not write directly to the underlying 180 * character device. 181 */ 182 #ifndef _SYS_SYSPROTO_H_ 183 struct mmap_args { 184 void *addr; 185 size_t len; 186 int prot; 187 int flags; 188 int fd; 189 long pad; 190 off_t pos; 191 }; 192 #endif 193 194 /* 195 * MPSAFE 196 */ 197 int 198 mmap(td, uap) 199 struct thread *td; 200 struct mmap_args *uap; 201 { 202 struct filedesc *fdp = td->td_proc->p_fd; 203 struct file *fp = NULL; 204 struct vnode *vp; 205 vm_offset_t addr; 206 vm_size_t size, pageoff; 207 vm_prot_t prot, maxprot; 208 void *handle; 209 int flags, error; 210 int disablexworkaround; 211 off_t pos; 212 struct vmspace *vms = td->td_proc->p_vmspace; 213 vm_object_t obj; 214 215 addr = (vm_offset_t) uap->addr; 216 size = uap->len; 217 prot = uap->prot & VM_PROT_ALL; 218 flags = uap->flags; 219 pos = uap->pos; 220 221 /* make sure mapping fits into numeric range etc */ 222 if ((ssize_t) uap->len < 0 || 223 ((flags & MAP_ANON) && uap->fd != -1)) 224 return (EINVAL); 225 226 if (flags & MAP_STACK) { 227 if ((uap->fd != -1) || 228 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 229 return (EINVAL); 230 flags |= MAP_ANON; 231 pos = 0; 232 } 233 234 /* 235 * Align the file position to a page boundary, 236 * and save its page offset component. 237 */ 238 pageoff = (pos & PAGE_MASK); 239 pos -= pageoff; 240 241 /* Adjust size for rounding (on both ends). */ 242 size += pageoff; /* low end... */ 243 size = (vm_size_t) round_page(size); /* hi end */ 244 245 /* 246 * Check for illegal addresses. Watch out for address wrap... Note 247 * that VM_*_ADDRESS are not constants due to casts (argh). 248 */ 249 if (flags & MAP_FIXED) { 250 /* 251 * The specified address must have the same remainder 252 * as the file offset taken modulo PAGE_SIZE, so it 253 * should be aligned after adjustment by pageoff. 254 */ 255 addr -= pageoff; 256 if (addr & PAGE_MASK) 257 return (EINVAL); 258 /* Address range must be all in user VM space. */ 259 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 260 return (EINVAL); 261 #ifndef i386 262 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 263 return (EINVAL); 264 #endif 265 if (addr + size < addr) 266 return (EINVAL); 267 } 268 /* 269 * XXX for non-fixed mappings where no hint is provided or 270 * the hint would fall in the potential heap space, 271 * place it after the end of the largest possible heap. 272 * 273 * There should really be a pmap call to determine a reasonable 274 * location. 275 */ 276 else if (addr == 0 || 277 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 278 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) 279 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 280 281 mtx_lock(&Giant); /* syscall marked mp-safe but isn't */ 282 if (flags & MAP_ANON) { 283 /* 284 * Mapping blank space is trivial. 285 */ 286 handle = NULL; 287 maxprot = VM_PROT_ALL; 288 pos = 0; 289 } else { 290 /* 291 * Mapping file, get fp for validation. Obtain vnode and make 292 * sure it is of appropriate type. 293 */ 294 if (((unsigned) uap->fd) >= fdp->fd_nfiles || 295 (fp = fdp->fd_ofiles[uap->fd]) == NULL) { 296 error = EBADF; 297 goto done2; 298 } 299 if (fp->f_type != DTYPE_VNODE) { 300 error = EINVAL; 301 goto done2; 302 } 303 304 /* 305 * don't let the descriptor disappear on us if we block 306 */ 307 fhold(fp); 308 309 /* 310 * POSIX shared-memory objects are defined to have 311 * kernel persistence, and are not defined to support 312 * read(2)/write(2) -- or even open(2). Thus, we can 313 * use MAP_ASYNC to trade on-disk coherence for speed. 314 * The shm_open(3) library routine turns on the FPOSIXSHM 315 * flag to request this behavior. 316 */ 317 if (fp->f_flag & FPOSIXSHM) 318 flags |= MAP_NOSYNC; 319 vp = (struct vnode *) fp->f_data; 320 if (vp->v_type != VREG && vp->v_type != VCHR) { 321 error = EINVAL; 322 goto done; 323 } 324 if (vp->v_type == VREG) { 325 /* 326 * Get the proper underlying object 327 */ 328 if (VOP_GETVOBJECT(vp, &obj) != 0) { 329 error = EINVAL; 330 goto done; 331 } 332 vp = (struct vnode*)obj->handle; 333 } 334 /* 335 * XXX hack to handle use of /dev/zero to map anon memory (ala 336 * SunOS). 337 */ 338 if ((vp->v_type == VCHR) && 339 (vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON)) { 340 handle = NULL; 341 maxprot = VM_PROT_ALL; 342 flags |= MAP_ANON; 343 pos = 0; 344 } else { 345 /* 346 * cdevs does not provide private mappings of any kind. 347 */ 348 /* 349 * However, for XIG X server to continue to work, 350 * we should allow the superuser to do it anyway. 351 * We only allow it at securelevel < 1. 352 * (Because the XIG X server writes directly to video 353 * memory via /dev/mem, it should never work at any 354 * other securelevel. 355 * XXX this will have to go 356 */ 357 if (securelevel_ge(td->td_proc->p_ucred, 1)) 358 disablexworkaround = 1; 359 else 360 disablexworkaround = suser_td(td); 361 if (vp->v_type == VCHR && disablexworkaround && 362 (flags & (MAP_PRIVATE|MAP_COPY))) { 363 error = EINVAL; 364 goto done; 365 } 366 /* 367 * Ensure that file and memory protections are 368 * compatible. Note that we only worry about 369 * writability if mapping is shared; in this case, 370 * current and max prot are dictated by the open file. 371 * XXX use the vnode instead? Problem is: what 372 * credentials do we use for determination? What if 373 * proc does a setuid? 374 */ 375 maxprot = VM_PROT_EXECUTE; /* ??? */ 376 if (fp->f_flag & FREAD) { 377 maxprot |= VM_PROT_READ; 378 } else if (prot & PROT_READ) { 379 error = EACCES; 380 goto done; 381 } 382 /* 383 * If we are sharing potential changes (either via 384 * MAP_SHARED or via the implicit sharing of character 385 * device mappings), and we are trying to get write 386 * permission although we opened it without asking 387 * for it, bail out. Check for superuser, only if 388 * we're at securelevel < 1, to allow the XIG X server 389 * to continue to work. 390 */ 391 392 if ((flags & MAP_SHARED) != 0 || 393 (vp->v_type == VCHR && disablexworkaround)) { 394 if ((fp->f_flag & FWRITE) != 0) { 395 struct vattr va; 396 if ((error = 397 VOP_GETATTR(vp, &va, 398 td->td_proc->p_ucred, td))) { 399 goto done; 400 } 401 if ((va.va_flags & 402 (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) { 403 maxprot |= VM_PROT_WRITE; 404 } else if (prot & PROT_WRITE) { 405 error = EPERM; 406 goto done; 407 } 408 } else if ((prot & PROT_WRITE) != 0) { 409 error = EACCES; 410 goto done; 411 } 412 } else { 413 maxprot |= VM_PROT_WRITE; 414 } 415 416 handle = (void *)vp; 417 } 418 } 419 420 /* 421 * Do not allow more then a certain number of vm_map_entry structures 422 * per process. Scale with the number of rforks sharing the map 423 * to make the limit reasonable for threads. 424 */ 425 if (max_proc_mmap && 426 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) { 427 error = ENOMEM; 428 goto done; 429 } 430 431 mtx_unlock(&Giant); 432 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 433 flags, handle, pos); 434 if (error == 0) 435 td->td_retval[0] = (register_t) (addr + pageoff); 436 mtx_lock(&Giant); 437 done: 438 if (fp) 439 fdrop(fp, td); 440 done2: 441 mtx_unlock(&Giant); 442 return (error); 443 } 444 445 #ifdef COMPAT_43 446 #ifndef _SYS_SYSPROTO_H_ 447 struct ommap_args { 448 caddr_t addr; 449 int len; 450 int prot; 451 int flags; 452 int fd; 453 long pos; 454 }; 455 #endif 456 int 457 ommap(td, uap) 458 struct thread *td; 459 struct ommap_args *uap; 460 { 461 struct mmap_args nargs; 462 static const char cvtbsdprot[8] = { 463 0, 464 PROT_EXEC, 465 PROT_WRITE, 466 PROT_EXEC | PROT_WRITE, 467 PROT_READ, 468 PROT_EXEC | PROT_READ, 469 PROT_WRITE | PROT_READ, 470 PROT_EXEC | PROT_WRITE | PROT_READ, 471 }; 472 473 #define OMAP_ANON 0x0002 474 #define OMAP_COPY 0x0020 475 #define OMAP_SHARED 0x0010 476 #define OMAP_FIXED 0x0100 477 478 nargs.addr = uap->addr; 479 nargs.len = uap->len; 480 nargs.prot = cvtbsdprot[uap->prot & 0x7]; 481 nargs.flags = 0; 482 if (uap->flags & OMAP_ANON) 483 nargs.flags |= MAP_ANON; 484 if (uap->flags & OMAP_COPY) 485 nargs.flags |= MAP_COPY; 486 if (uap->flags & OMAP_SHARED) 487 nargs.flags |= MAP_SHARED; 488 else 489 nargs.flags |= MAP_PRIVATE; 490 if (uap->flags & OMAP_FIXED) 491 nargs.flags |= MAP_FIXED; 492 nargs.fd = uap->fd; 493 nargs.pos = uap->pos; 494 return (mmap(td, &nargs)); 495 } 496 #endif /* COMPAT_43 */ 497 498 499 #ifndef _SYS_SYSPROTO_H_ 500 struct msync_args { 501 void *addr; 502 int len; 503 int flags; 504 }; 505 #endif 506 /* 507 * MPSAFE 508 */ 509 int 510 msync(td, uap) 511 struct thread *td; 512 struct msync_args *uap; 513 { 514 vm_offset_t addr; 515 vm_size_t size, pageoff; 516 int flags; 517 vm_map_t map; 518 int rv; 519 520 addr = (vm_offset_t) uap->addr; 521 size = uap->len; 522 flags = uap->flags; 523 524 pageoff = (addr & PAGE_MASK); 525 addr -= pageoff; 526 size += pageoff; 527 size = (vm_size_t) round_page(size); 528 if (addr + size < addr) 529 return(EINVAL); 530 531 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 532 return (EINVAL); 533 534 mtx_lock(&Giant); 535 536 map = &td->td_proc->p_vmspace->vm_map; 537 538 /* 539 * XXX Gak! If size is zero we are supposed to sync "all modified 540 * pages with the region containing addr". Unfortunately, we don't 541 * really keep track of individual mmaps so we approximate by flushing 542 * the range of the map entry containing addr. This can be incorrect 543 * if the region splits or is coalesced with a neighbor. 544 */ 545 if (size == 0) { 546 vm_map_entry_t entry; 547 548 vm_map_lock_read(map); 549 rv = vm_map_lookup_entry(map, addr, &entry); 550 vm_map_unlock_read(map); 551 if (rv == FALSE) { 552 rv = -1; 553 goto done2; 554 } 555 addr = entry->start; 556 size = entry->end - entry->start; 557 } 558 559 /* 560 * Clean the pages and interpret the return value. 561 */ 562 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 563 (flags & MS_INVALIDATE) != 0); 564 565 done2: 566 mtx_unlock(&Giant); 567 568 switch (rv) { 569 case KERN_SUCCESS: 570 return(0); 571 case KERN_INVALID_ADDRESS: 572 return (EINVAL); /* Sun returns ENOMEM? */ 573 case KERN_FAILURE: 574 return (EIO); 575 default: 576 return (EINVAL); 577 } 578 } 579 580 #ifndef _SYS_SYSPROTO_H_ 581 struct munmap_args { 582 void *addr; 583 size_t len; 584 }; 585 #endif 586 /* 587 * MPSAFE 588 */ 589 int 590 munmap(td, uap) 591 struct thread *td; 592 struct munmap_args *uap; 593 { 594 vm_offset_t addr; 595 vm_size_t size, pageoff; 596 vm_map_t map; 597 598 addr = (vm_offset_t) uap->addr; 599 size = uap->len; 600 601 pageoff = (addr & PAGE_MASK); 602 addr -= pageoff; 603 size += pageoff; 604 size = (vm_size_t) round_page(size); 605 if (addr + size < addr) 606 return(EINVAL); 607 608 if (size == 0) 609 return (0); 610 611 /* 612 * Check for illegal addresses. Watch out for address wrap... Note 613 * that VM_*_ADDRESS are not constants due to casts (argh). 614 */ 615 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 616 return (EINVAL); 617 #ifndef i386 618 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 619 return (EINVAL); 620 #endif 621 mtx_lock(&Giant); 622 map = &td->td_proc->p_vmspace->vm_map; 623 /* 624 * Make sure entire range is allocated. 625 */ 626 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) { 627 mtx_unlock(&Giant); 628 return (EINVAL); 629 } 630 /* returns nothing but KERN_SUCCESS anyway */ 631 (void) vm_map_remove(map, addr, addr + size); 632 mtx_unlock(&Giant); 633 return (0); 634 } 635 636 #if 0 637 void 638 munmapfd(td, fd) 639 struct thread *td; 640 int fd; 641 { 642 /* 643 * XXX should unmap any regions mapped to this file 644 */ 645 td->td_proc->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED; 646 } 647 #endif 648 649 #ifndef _SYS_SYSPROTO_H_ 650 struct mprotect_args { 651 const void *addr; 652 size_t len; 653 int prot; 654 }; 655 #endif 656 /* 657 * MPSAFE 658 */ 659 int 660 mprotect(td, uap) 661 struct thread *td; 662 struct mprotect_args *uap; 663 { 664 vm_offset_t addr; 665 vm_size_t size, pageoff; 666 vm_prot_t prot; 667 int ret; 668 669 addr = (vm_offset_t) uap->addr; 670 size = uap->len; 671 prot = uap->prot & VM_PROT_ALL; 672 #if defined(VM_PROT_READ_IS_EXEC) 673 if (prot & VM_PROT_READ) 674 prot |= VM_PROT_EXECUTE; 675 #endif 676 677 pageoff = (addr & PAGE_MASK); 678 addr -= pageoff; 679 size += pageoff; 680 size = (vm_size_t) round_page(size); 681 if (addr + size < addr) 682 return(EINVAL); 683 684 mtx_lock(&Giant); 685 ret = vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr, 686 addr + size, prot, FALSE); 687 mtx_unlock(&Giant); 688 switch (ret) { 689 case KERN_SUCCESS: 690 return (0); 691 case KERN_PROTECTION_FAILURE: 692 return (EACCES); 693 } 694 return (EINVAL); 695 } 696 697 #ifndef _SYS_SYSPROTO_H_ 698 struct minherit_args { 699 void *addr; 700 size_t len; 701 int inherit; 702 }; 703 #endif 704 /* 705 * MPSAFE 706 */ 707 int 708 minherit(td, uap) 709 struct thread *td; 710 struct minherit_args *uap; 711 { 712 vm_offset_t addr; 713 vm_size_t size, pageoff; 714 vm_inherit_t inherit; 715 int ret; 716 717 addr = (vm_offset_t)uap->addr; 718 size = uap->len; 719 inherit = uap->inherit; 720 721 pageoff = (addr & PAGE_MASK); 722 addr -= pageoff; 723 size += pageoff; 724 size = (vm_size_t) round_page(size); 725 if (addr + size < addr) 726 return(EINVAL); 727 728 mtx_lock(&Giant); 729 ret = vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, addr+size, 730 inherit); 731 mtx_unlock(&Giant); 732 733 switch (ret) { 734 case KERN_SUCCESS: 735 return (0); 736 case KERN_PROTECTION_FAILURE: 737 return (EACCES); 738 } 739 return (EINVAL); 740 } 741 742 #ifndef _SYS_SYSPROTO_H_ 743 struct madvise_args { 744 void *addr; 745 size_t len; 746 int behav; 747 }; 748 #endif 749 750 /* 751 * MPSAFE 752 */ 753 /* ARGSUSED */ 754 int 755 madvise(td, uap) 756 struct thread *td; 757 struct madvise_args *uap; 758 { 759 vm_offset_t start, end; 760 int ret; 761 762 /* 763 * Check for illegal behavior 764 */ 765 if (uap->behav < 0 || uap->behav > MADV_CORE) 766 return (EINVAL); 767 /* 768 * Check for illegal addresses. Watch out for address wrap... Note 769 * that VM_*_ADDRESS are not constants due to casts (argh). 770 */ 771 if (VM_MAXUSER_ADDRESS > 0 && 772 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS) 773 return (EINVAL); 774 #ifndef i386 775 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS) 776 return (EINVAL); 777 #endif 778 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) 779 return (EINVAL); 780 781 /* 782 * Since this routine is only advisory, we default to conservative 783 * behavior. 784 */ 785 start = trunc_page((vm_offset_t) uap->addr); 786 end = round_page((vm_offset_t) uap->addr + uap->len); 787 788 mtx_lock(&Giant); 789 ret = vm_map_madvise(&td->td_proc->p_vmspace->vm_map, start, end, uap->behav); 790 mtx_unlock(&Giant); 791 return (ret ? EINVAL : 0); 792 } 793 794 #ifndef _SYS_SYSPROTO_H_ 795 struct mincore_args { 796 const void *addr; 797 size_t len; 798 char *vec; 799 }; 800 #endif 801 802 /* 803 * MPSAFE 804 */ 805 /* ARGSUSED */ 806 int 807 mincore(td, uap) 808 struct thread *td; 809 struct mincore_args *uap; 810 { 811 vm_offset_t addr, first_addr; 812 vm_offset_t end, cend; 813 pmap_t pmap; 814 vm_map_t map; 815 char *vec; 816 int error = 0; 817 int vecindex, lastvecindex; 818 vm_map_entry_t current; 819 vm_map_entry_t entry; 820 int mincoreinfo; 821 unsigned int timestamp; 822 823 /* 824 * Make sure that the addresses presented are valid for user 825 * mode. 826 */ 827 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 828 end = addr + (vm_size_t)round_page(uap->len); 829 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS) 830 return (EINVAL); 831 if (end < addr) 832 return (EINVAL); 833 834 /* 835 * Address of byte vector 836 */ 837 vec = uap->vec; 838 839 mtx_lock(&Giant); 840 map = &td->td_proc->p_vmspace->vm_map; 841 pmap = vmspace_pmap(td->td_proc->p_vmspace); 842 843 vm_map_lock_read(map); 844 RestartScan: 845 timestamp = map->timestamp; 846 847 if (!vm_map_lookup_entry(map, addr, &entry)) 848 entry = entry->next; 849 850 /* 851 * Do this on a map entry basis so that if the pages are not 852 * in the current processes address space, we can easily look 853 * up the pages elsewhere. 854 */ 855 lastvecindex = -1; 856 for (current = entry; 857 (current != &map->header) && (current->start < end); 858 current = current->next) { 859 860 /* 861 * ignore submaps (for now) or null objects 862 */ 863 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 864 current->object.vm_object == NULL) 865 continue; 866 867 /* 868 * limit this scan to the current map entry and the 869 * limits for the mincore call 870 */ 871 if (addr < current->start) 872 addr = current->start; 873 cend = current->end; 874 if (cend > end) 875 cend = end; 876 877 /* 878 * scan this entry one page at a time 879 */ 880 while (addr < cend) { 881 /* 882 * Check pmap first, it is likely faster, also 883 * it can provide info as to whether we are the 884 * one referencing or modifying the page. 885 */ 886 mincoreinfo = pmap_mincore(pmap, addr); 887 if (!mincoreinfo) { 888 vm_pindex_t pindex; 889 vm_ooffset_t offset; 890 vm_page_t m; 891 /* 892 * calculate the page index into the object 893 */ 894 offset = current->offset + (addr - current->start); 895 pindex = OFF_TO_IDX(offset); 896 m = vm_page_lookup(current->object.vm_object, 897 pindex); 898 /* 899 * if the page is resident, then gather information about 900 * it. 901 */ 902 if (m) { 903 mincoreinfo = MINCORE_INCORE; 904 if (m->dirty || 905 pmap_is_modified(m)) 906 mincoreinfo |= MINCORE_MODIFIED_OTHER; 907 if ((m->flags & PG_REFERENCED) || 908 pmap_ts_referenced(m)) { 909 vm_page_flag_set(m, PG_REFERENCED); 910 mincoreinfo |= MINCORE_REFERENCED_OTHER; 911 } 912 } 913 } 914 915 /* 916 * subyte may page fault. In case it needs to modify 917 * the map, we release the lock. 918 */ 919 vm_map_unlock_read(map); 920 921 /* 922 * calculate index into user supplied byte vector 923 */ 924 vecindex = OFF_TO_IDX(addr - first_addr); 925 926 /* 927 * If we have skipped map entries, we need to make sure that 928 * the byte vector is zeroed for those skipped entries. 929 */ 930 while ((lastvecindex + 1) < vecindex) { 931 error = subyte( vec + lastvecindex, 0); 932 if (error) { 933 error = EFAULT; 934 goto done2; 935 } 936 ++lastvecindex; 937 } 938 939 /* 940 * Pass the page information to the user 941 */ 942 error = subyte( vec + vecindex, mincoreinfo); 943 if (error) { 944 error = EFAULT; 945 goto done2; 946 } 947 948 /* 949 * If the map has changed, due to the subyte, the previous 950 * output may be invalid. 951 */ 952 vm_map_lock_read(map); 953 if (timestamp != map->timestamp) 954 goto RestartScan; 955 956 lastvecindex = vecindex; 957 addr += PAGE_SIZE; 958 } 959 } 960 961 /* 962 * subyte may page fault. In case it needs to modify 963 * the map, we release the lock. 964 */ 965 vm_map_unlock_read(map); 966 967 /* 968 * Zero the last entries in the byte vector. 969 */ 970 vecindex = OFF_TO_IDX(end - first_addr); 971 while ((lastvecindex + 1) < vecindex) { 972 error = subyte( vec + lastvecindex, 0); 973 if (error) { 974 error = EFAULT; 975 goto done2; 976 } 977 ++lastvecindex; 978 } 979 980 /* 981 * If the map has changed, due to the subyte, the previous 982 * output may be invalid. 983 */ 984 vm_map_lock_read(map); 985 if (timestamp != map->timestamp) 986 goto RestartScan; 987 vm_map_unlock_read(map); 988 done2: 989 mtx_unlock(&Giant); 990 return (error); 991 } 992 993 #ifndef _SYS_SYSPROTO_H_ 994 struct mlock_args { 995 const void *addr; 996 size_t len; 997 }; 998 #endif 999 /* 1000 * MPSAFE 1001 */ 1002 int 1003 mlock(td, uap) 1004 struct thread *td; 1005 struct mlock_args *uap; 1006 { 1007 vm_offset_t addr; 1008 vm_size_t size, pageoff; 1009 int error; 1010 1011 addr = (vm_offset_t) uap->addr; 1012 size = uap->len; 1013 1014 pageoff = (addr & PAGE_MASK); 1015 addr -= pageoff; 1016 size += pageoff; 1017 size = (vm_size_t) round_page(size); 1018 1019 /* disable wrap around */ 1020 if (addr + size < addr) 1021 return (EINVAL); 1022 1023 if (atop(size) + cnt.v_wire_count > vm_page_max_wired) 1024 return (EAGAIN); 1025 1026 #ifdef pmap_wired_count 1027 if (size + ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))) > 1028 td->td_proc->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 1029 return (ENOMEM); 1030 #else 1031 error = suser_td(td); 1032 if (error) 1033 return (error); 1034 #endif 1035 1036 mtx_lock(&Giant); 1037 error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr, 1038 addr + size, FALSE); 1039 mtx_unlock(&Giant); 1040 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1041 } 1042 1043 #ifndef _SYS_SYSPROTO_H_ 1044 struct mlockall_args { 1045 int how; 1046 }; 1047 #endif 1048 1049 /* 1050 * MPSAFE 1051 */ 1052 int 1053 mlockall(td, uap) 1054 struct thread *td; 1055 struct mlockall_args *uap; 1056 { 1057 /* mtx_lock(&Giant); */ 1058 /* mtx_unlock(&Giant); */ 1059 return 0; 1060 } 1061 1062 #ifndef _SYS_SYSPROTO_H_ 1063 struct mlockall_args { 1064 int how; 1065 }; 1066 #endif 1067 1068 /* 1069 * MPSAFE 1070 */ 1071 int 1072 munlockall(td, uap) 1073 struct thread *td; 1074 struct munlockall_args *uap; 1075 { 1076 /* mtx_lock(&Giant); */ 1077 /* mtx_unlock(&Giant); */ 1078 return 0; 1079 } 1080 1081 #ifndef _SYS_SYSPROTO_H_ 1082 struct munlock_args { 1083 const void *addr; 1084 size_t len; 1085 }; 1086 #endif 1087 /* 1088 * MPSAFE 1089 */ 1090 int 1091 munlock(td, uap) 1092 struct thread *td; 1093 struct munlock_args *uap; 1094 { 1095 vm_offset_t addr; 1096 vm_size_t size, pageoff; 1097 int error; 1098 1099 addr = (vm_offset_t) uap->addr; 1100 size = uap->len; 1101 1102 pageoff = (addr & PAGE_MASK); 1103 addr -= pageoff; 1104 size += pageoff; 1105 size = (vm_size_t) round_page(size); 1106 1107 /* disable wrap around */ 1108 if (addr + size < addr) 1109 return (EINVAL); 1110 1111 #ifndef pmap_wired_count 1112 error = suser_td(td); 1113 if (error) 1114 return (error); 1115 #endif 1116 1117 mtx_lock(&Giant); 1118 error = vm_map_user_pageable(&td->td_proc->p_vmspace->vm_map, addr, 1119 addr + size, TRUE); 1120 mtx_unlock(&Giant); 1121 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1122 } 1123 1124 /* 1125 * vm_mmap() 1126 * 1127 * MPSAFE 1128 * 1129 * Internal version of mmap. Currently used by mmap, exec, and sys5 1130 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON. 1131 */ 1132 int 1133 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1134 vm_prot_t maxprot, int flags, 1135 void *handle, 1136 vm_ooffset_t foff) 1137 { 1138 boolean_t fitit; 1139 vm_object_t object; 1140 struct vnode *vp = NULL; 1141 objtype_t type; 1142 int rv = KERN_SUCCESS; 1143 vm_ooffset_t objsize; 1144 int docow; 1145 struct thread *td = curthread; 1146 1147 if (size == 0) 1148 return (0); 1149 1150 objsize = size = round_page(size); 1151 1152 /* 1153 * We currently can only deal with page aligned file offsets. 1154 * The check is here rather than in the syscall because the 1155 * kernel calls this function internally for other mmaping 1156 * operations (such as in exec) and non-aligned offsets will 1157 * cause pmap inconsistencies...so we want to be sure to 1158 * disallow this in all cases. 1159 */ 1160 if (foff & PAGE_MASK) 1161 return (EINVAL); 1162 1163 if ((flags & MAP_FIXED) == 0) { 1164 fitit = TRUE; 1165 *addr = round_page(*addr); 1166 mtx_lock(&Giant); 1167 } else { 1168 if (*addr != trunc_page(*addr)) 1169 return (EINVAL); 1170 fitit = FALSE; 1171 mtx_lock(&Giant); 1172 (void) vm_map_remove(map, *addr, *addr + size); 1173 } 1174 1175 /* 1176 * Lookup/allocate object. 1177 */ 1178 if (flags & MAP_ANON) { 1179 type = OBJT_DEFAULT; 1180 /* 1181 * Unnamed anonymous regions always start at 0. 1182 */ 1183 if (handle == 0) 1184 foff = 0; 1185 } else { 1186 vp = (struct vnode *) handle; 1187 if (vp->v_type == VCHR) { 1188 type = OBJT_DEVICE; 1189 handle = (void *)(intptr_t)vp->v_rdev; 1190 } else { 1191 struct vattr vat; 1192 int error; 1193 1194 error = VOP_GETATTR(vp, &vat, td->td_proc->p_ucred, td); 1195 if (error) { 1196 mtx_unlock(&Giant); 1197 return (error); 1198 } 1199 objsize = round_page(vat.va_size); 1200 type = OBJT_VNODE; 1201 /* 1202 * if it is a regular file without any references 1203 * we do not need to sync it. 1204 */ 1205 if (vp->v_type == VREG && vat.va_nlink == 0) { 1206 flags |= MAP_NOSYNC; 1207 } 1208 } 1209 } 1210 1211 if (handle == NULL) { 1212 object = NULL; 1213 docow = 0; 1214 } else { 1215 object = vm_pager_allocate(type, 1216 handle, objsize, prot, foff); 1217 if (object == NULL) { 1218 mtx_unlock(&Giant); 1219 return (type == OBJT_DEVICE ? EINVAL : ENOMEM); 1220 } 1221 docow = MAP_PREFAULT_PARTIAL; 1222 } 1223 1224 /* 1225 * Force device mappings to be shared. 1226 */ 1227 if (type == OBJT_DEVICE || type == OBJT_PHYS) { 1228 flags &= ~(MAP_PRIVATE|MAP_COPY); 1229 flags |= MAP_SHARED; 1230 } 1231 1232 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1233 docow |= MAP_COPY_ON_WRITE; 1234 if (flags & MAP_NOSYNC) 1235 docow |= MAP_DISABLE_SYNCER; 1236 if (flags & MAP_NOCORE) 1237 docow |= MAP_DISABLE_COREDUMP; 1238 1239 #if defined(VM_PROT_READ_IS_EXEC) 1240 if (prot & VM_PROT_READ) 1241 prot |= VM_PROT_EXECUTE; 1242 1243 if (maxprot & VM_PROT_READ) 1244 maxprot |= VM_PROT_EXECUTE; 1245 #endif 1246 1247 if (fitit) 1248 *addr = pmap_addr_hint(object, *addr, size); 1249 1250 if (flags & MAP_STACK) 1251 rv = vm_map_stack (map, *addr, size, prot, 1252 maxprot, docow); 1253 else 1254 rv = vm_map_find(map, object, foff, addr, size, fitit, 1255 prot, maxprot, docow); 1256 1257 if (rv != KERN_SUCCESS) { 1258 /* 1259 * Lose the object reference. Will destroy the 1260 * object if it's an unnamed anonymous mapping 1261 * or named anonymous without other references. 1262 */ 1263 vm_object_deallocate(object); 1264 } else if (flags & MAP_SHARED) { 1265 /* 1266 * Shared memory is also shared with children. 1267 */ 1268 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1269 if (rv != KERN_SUCCESS) 1270 (void) vm_map_remove(map, *addr, *addr + size); 1271 } 1272 mtx_unlock(&Giant); 1273 switch (rv) { 1274 case KERN_SUCCESS: 1275 return (0); 1276 case KERN_INVALID_ADDRESS: 1277 case KERN_NO_SPACE: 1278 return (ENOMEM); 1279 case KERN_PROTECTION_FAILURE: 1280 return (EACCES); 1281 default: 1282 return (EINVAL); 1283 } 1284 } 1285