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 1.7 2003/07/24 01:41:27 dillon Exp $ 43 */ 44 45 /* 46 * Mapped file (mmap) interface to VM 47 */ 48 49 #include "opt_compat.h" 50 51 #include <sys/param.h> 52 #include <sys/kernel.h> 53 #include <sys/systm.h> 54 #include <sys/sysproto.h> 55 #include <sys/filedesc.h> 56 #include <sys/proc.h> 57 #include <sys/resource.h> 58 #include <sys/resourcevar.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 <sys/lock.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_map.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_page.h> 75 #include <vm/vm_pager.h> 76 #include <vm/vm_pageout.h> 77 #include <vm/vm_extern.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_kern.h> 80 81 #include <sys/file2.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 __P((void *)); 96 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL) 97 98 static void 99 vmmapentry_rsrc_init(dummy) 100 void *dummy; 101 { 102 max_proc_mmap = vm_kmem_size / 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 #if defined(COMPAT_43) || defined(COMPAT_SUNOS) 126 127 /* 128 * getpagesize_args(int dummy) 129 */ 130 /* ARGSUSED */ 131 int 132 ogetpagesize(struct getpagesize_args *uap) 133 { 134 struct proc *p = curproc; 135 p->p_retval[0] = PAGE_SIZE; 136 return (0); 137 } 138 #endif /* COMPAT_43 || COMPAT_SUNOS */ 139 140 141 /* 142 * mmap_args(void *addr, size_t len, int prot, int flags, int fd, 143 * long pad, off_t pos) 144 * 145 * Memory Map (mmap) system call. Note that the file offset 146 * and address are allowed to be NOT page aligned, though if 147 * the MAP_FIXED flag it set, both must have the same remainder 148 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 149 * page-aligned, the actual mapping starts at trunc_page(addr) 150 * and the return value is adjusted up by the page offset. 151 * 152 * Generally speaking, only character devices which are themselves 153 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 154 * there would be no cache coherency between a descriptor and a VM mapping 155 * both to the same character device. 156 * 157 * Block devices can be mmap'd no matter what they represent. Cache coherency 158 * is maintained as long as you do not write directly to the underlying 159 * character device. 160 */ 161 162 int 163 mmap(struct mmap_args *uap) 164 { 165 struct thread *td = curthread; 166 struct proc *p = td->td_proc; 167 struct filedesc *fdp = p->p_fd; 168 struct file *fp = NULL; 169 struct vnode *vp; 170 vm_offset_t addr; 171 vm_size_t size, pageoff; 172 vm_prot_t prot, maxprot; 173 void *handle; 174 int flags, error; 175 int disablexworkaround; 176 off_t pos; 177 struct vmspace *vms = p->p_vmspace; 178 vm_object_t obj; 179 180 KKASSERT(p); 181 182 addr = (vm_offset_t) uap->addr; 183 size = uap->len; 184 prot = uap->prot & VM_PROT_ALL; 185 flags = uap->flags; 186 pos = uap->pos; 187 188 /* make sure mapping fits into numeric range etc */ 189 if ((ssize_t) uap->len < 0 || 190 ((flags & MAP_ANON) && uap->fd != -1)) 191 return (EINVAL); 192 193 if (flags & MAP_STACK) { 194 if ((uap->fd != -1) || 195 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 196 return (EINVAL); 197 flags |= MAP_ANON; 198 pos = 0; 199 } 200 201 /* 202 * Align the file position to a page boundary, 203 * and save its page offset component. 204 */ 205 pageoff = (pos & PAGE_MASK); 206 pos -= pageoff; 207 208 /* Adjust size for rounding (on both ends). */ 209 size += pageoff; /* low end... */ 210 size = (vm_size_t) round_page(size); /* hi end */ 211 212 /* 213 * Check for illegal addresses. Watch out for address wrap... Note 214 * that VM_*_ADDRESS are not constants due to casts (argh). 215 */ 216 if (flags & MAP_FIXED) { 217 /* 218 * The specified address must have the same remainder 219 * as the file offset taken modulo PAGE_SIZE, so it 220 * should be aligned after adjustment by pageoff. 221 */ 222 addr -= pageoff; 223 if (addr & PAGE_MASK) 224 return (EINVAL); 225 /* Address range must be all in user VM space. */ 226 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 227 return (EINVAL); 228 #ifndef i386 229 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 230 return (EINVAL); 231 #endif 232 if (addr + size < addr) 233 return (EINVAL); 234 } 235 /* 236 * XXX for non-fixed mappings where no hint is provided or 237 * the hint would fall in the potential heap space, 238 * place it after the end of the largest possible heap. 239 * 240 * There should really be a pmap call to determine a reasonable 241 * location. 242 */ 243 else if (addr == 0 || 244 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 245 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) 246 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 247 248 if (flags & MAP_ANON) { 249 /* 250 * Mapping blank space is trivial. 251 */ 252 handle = NULL; 253 maxprot = VM_PROT_ALL; 254 pos = 0; 255 } else { 256 /* 257 * Mapping file, get fp for validation. Obtain vnode and make 258 * sure it is of appropriate type. 259 */ 260 if (((unsigned) uap->fd) >= fdp->fd_nfiles || 261 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 262 return (EBADF); 263 if (fp->f_type != DTYPE_VNODE) 264 return (EINVAL); 265 /* 266 * POSIX shared-memory objects are defined to have 267 * kernel persistence, and are not defined to support 268 * read(2)/write(2) -- or even open(2). Thus, we can 269 * use MAP_ASYNC to trade on-disk coherence for speed. 270 * The shm_open(3) library routine turns on the FPOSIXSHM 271 * flag to request this behavior. 272 */ 273 if (fp->f_flag & FPOSIXSHM) 274 flags |= MAP_NOSYNC; 275 vp = (struct vnode *) fp->f_data; 276 if (vp->v_type != VREG && vp->v_type != VCHR) 277 return (EINVAL); 278 if (vp->v_type == VREG) { 279 /* 280 * Get the proper underlying object 281 */ 282 if (VOP_GETVOBJECT(vp, &obj) != 0) 283 return (EINVAL); 284 vp = (struct vnode*)obj->handle; 285 } 286 287 /* 288 * don't let the descriptor disappear on us if we block 289 */ 290 fhold(fp); 291 292 /* 293 * XXX hack to handle use of /dev/zero to map anon memory (ala 294 * SunOS). 295 */ 296 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 297 handle = NULL; 298 maxprot = VM_PROT_ALL; 299 flags |= MAP_ANON; 300 pos = 0; 301 } else { 302 /* 303 * cdevs does not provide private mappings of any kind. 304 */ 305 /* 306 * However, for XIG X server to continue to work, 307 * we should allow the superuser to do it anyway. 308 * We only allow it at securelevel < 1. 309 * (Because the XIG X server writes directly to video 310 * memory via /dev/mem, it should never work at any 311 * other securelevel. 312 * XXX this will have to go 313 */ 314 if (securelevel >= 1) 315 disablexworkaround = 1; 316 else 317 disablexworkaround = suser(td); 318 if (vp->v_type == VCHR && disablexworkaround && 319 (flags & (MAP_PRIVATE|MAP_COPY))) { 320 error = EINVAL; 321 goto done; 322 } 323 /* 324 * Ensure that file and memory protections are 325 * compatible. Note that we only worry about 326 * writability if mapping is shared; in this case, 327 * current and max prot are dictated by the open file. 328 * XXX use the vnode instead? Problem is: what 329 * credentials do we use for determination? What if 330 * proc does a setuid? 331 */ 332 maxprot = VM_PROT_EXECUTE; /* ??? */ 333 if (fp->f_flag & FREAD) { 334 maxprot |= VM_PROT_READ; 335 } else if (prot & PROT_READ) { 336 error = EACCES; 337 goto done; 338 } 339 /* 340 * If we are sharing potential changes (either via 341 * MAP_SHARED or via the implicit sharing of character 342 * device mappings), and we are trying to get write 343 * permission although we opened it without asking 344 * for it, bail out. Check for superuser, only if 345 * we're at securelevel < 1, to allow the XIG X server 346 * to continue to work. 347 */ 348 349 if ((flags & MAP_SHARED) != 0 || 350 (vp->v_type == VCHR && disablexworkaround)) { 351 if ((fp->f_flag & FWRITE) != 0) { 352 struct vattr va; 353 if ((error = VOP_GETATTR(vp, &va, td))) { 354 goto done; 355 } 356 if ((va.va_flags & 357 (IMMUTABLE|APPEND)) == 0) { 358 maxprot |= VM_PROT_WRITE; 359 } else if (prot & PROT_WRITE) { 360 error = EPERM; 361 goto done; 362 } 363 } else if ((prot & PROT_WRITE) != 0) { 364 error = EACCES; 365 goto done; 366 } 367 } else { 368 maxprot |= VM_PROT_WRITE; 369 } 370 handle = (void *)vp; 371 } 372 } 373 374 /* 375 * Do not allow more then a certain number of vm_map_entry structures 376 * per process. Scale with the number of rforks sharing the map 377 * to make the limit reasonable for threads. 378 */ 379 if (max_proc_mmap && 380 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) { 381 error = ENOMEM; 382 goto done; 383 } 384 385 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 386 flags, handle, pos); 387 if (error == 0) 388 p->p_retval[0] = (register_t) (addr + pageoff); 389 done: 390 if (fp) 391 fdrop(fp, td); 392 return (error); 393 } 394 395 #ifdef COMPAT_43 396 /* 397 * ommap_args(caddr_t addr, int len, int prot, int flags, int fd, long pos) 398 */ 399 int 400 ommap(struct ommap_args *uap) 401 { 402 struct mmap_args nargs; 403 static const char cvtbsdprot[8] = { 404 0, 405 PROT_EXEC, 406 PROT_WRITE, 407 PROT_EXEC | PROT_WRITE, 408 PROT_READ, 409 PROT_EXEC | PROT_READ, 410 PROT_WRITE | PROT_READ, 411 PROT_EXEC | PROT_WRITE | PROT_READ, 412 }; 413 414 #define OMAP_ANON 0x0002 415 #define OMAP_COPY 0x0020 416 #define OMAP_SHARED 0x0010 417 #define OMAP_FIXED 0x0100 418 #define OMAP_INHERIT 0x0800 419 420 nargs.addr = uap->addr; 421 nargs.len = uap->len; 422 nargs.prot = cvtbsdprot[uap->prot & 0x7]; 423 nargs.flags = 0; 424 if (uap->flags & OMAP_ANON) 425 nargs.flags |= MAP_ANON; 426 if (uap->flags & OMAP_COPY) 427 nargs.flags |= MAP_COPY; 428 if (uap->flags & OMAP_SHARED) 429 nargs.flags |= MAP_SHARED; 430 else 431 nargs.flags |= MAP_PRIVATE; 432 if (uap->flags & OMAP_FIXED) 433 nargs.flags |= MAP_FIXED; 434 if (uap->flags & OMAP_INHERIT) 435 nargs.flags |= MAP_INHERIT; 436 nargs.fd = uap->fd; 437 nargs.pos = uap->pos; 438 return (mmap(&nargs)); 439 } 440 #endif /* COMPAT_43 */ 441 442 443 /* 444 * msync_args(void *addr, int len, int flags) 445 */ 446 int 447 msync(struct msync_args *uap) 448 { 449 struct proc *p = curproc; 450 vm_offset_t addr; 451 vm_size_t size, pageoff; 452 int flags; 453 vm_map_t map; 454 int rv; 455 456 addr = (vm_offset_t) uap->addr; 457 size = uap->len; 458 flags = uap->flags; 459 460 pageoff = (addr & PAGE_MASK); 461 addr -= pageoff; 462 size += pageoff; 463 size = (vm_size_t) round_page(size); 464 if (addr + size < addr) 465 return(EINVAL); 466 467 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 468 return (EINVAL); 469 470 map = &p->p_vmspace->vm_map; 471 472 /* 473 * XXX Gak! If size is zero we are supposed to sync "all modified 474 * pages with the region containing addr". Unfortunately, we don't 475 * really keep track of individual mmaps so we approximate by flushing 476 * the range of the map entry containing addr. This can be incorrect 477 * if the region splits or is coalesced with a neighbor. 478 */ 479 if (size == 0) { 480 vm_map_entry_t entry; 481 482 vm_map_lock_read(map); 483 rv = vm_map_lookup_entry(map, addr, &entry); 484 vm_map_unlock_read(map); 485 if (rv == FALSE) 486 return (EINVAL); 487 addr = entry->start; 488 size = entry->end - entry->start; 489 } 490 491 /* 492 * Clean the pages and interpret the return value. 493 */ 494 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 495 (flags & MS_INVALIDATE) != 0); 496 497 switch (rv) { 498 case KERN_SUCCESS: 499 break; 500 case KERN_INVALID_ADDRESS: 501 return (EINVAL); /* Sun returns ENOMEM? */ 502 case KERN_FAILURE: 503 return (EIO); 504 default: 505 return (EINVAL); 506 } 507 508 return (0); 509 } 510 511 /* 512 * munmap_args(void *addr, size_t len) 513 */ 514 int 515 munmap(struct munmap_args *uap) 516 { 517 struct proc *p = curproc; 518 vm_offset_t addr; 519 vm_size_t size, pageoff; 520 vm_map_t map; 521 522 addr = (vm_offset_t) uap->addr; 523 size = uap->len; 524 525 pageoff = (addr & PAGE_MASK); 526 addr -= pageoff; 527 size += pageoff; 528 size = (vm_size_t) round_page(size); 529 if (addr + size < addr) 530 return(EINVAL); 531 532 if (size == 0) 533 return (0); 534 535 /* 536 * Check for illegal addresses. Watch out for address wrap... Note 537 * that VM_*_ADDRESS are not constants due to casts (argh). 538 */ 539 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS) 540 return (EINVAL); 541 #ifndef i386 542 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS) 543 return (EINVAL); 544 #endif 545 map = &p->p_vmspace->vm_map; 546 /* 547 * Make sure entire range is allocated. 548 */ 549 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) 550 return (EINVAL); 551 /* returns nothing but KERN_SUCCESS anyway */ 552 (void) vm_map_remove(map, addr, addr + size); 553 return (0); 554 } 555 556 #if 0 557 void 558 munmapfd(p, fd) 559 struct proc *p; 560 int fd; 561 { 562 /* 563 * XXX should unmap any regions mapped to this file 564 */ 565 p->p_fd->fd_ofileflags[fd] &= ~UF_MAPPED; 566 } 567 #endif 568 569 /* 570 * mprotect_args(const void *addr, size_t len, int prot) 571 */ 572 int 573 mprotect(struct mprotect_args *uap) 574 { 575 struct proc *p = curproc; 576 vm_offset_t addr; 577 vm_size_t size, pageoff; 578 register vm_prot_t prot; 579 580 addr = (vm_offset_t) uap->addr; 581 size = uap->len; 582 prot = uap->prot & VM_PROT_ALL; 583 #if defined(VM_PROT_READ_IS_EXEC) 584 if (prot & VM_PROT_READ) 585 prot |= VM_PROT_EXECUTE; 586 #endif 587 588 pageoff = (addr & PAGE_MASK); 589 addr -= pageoff; 590 size += pageoff; 591 size = (vm_size_t) round_page(size); 592 if (addr + size < addr) 593 return(EINVAL); 594 595 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, 596 FALSE)) { 597 case KERN_SUCCESS: 598 return (0); 599 case KERN_PROTECTION_FAILURE: 600 return (EACCES); 601 } 602 return (EINVAL); 603 } 604 605 /* 606 * minherit_args(void *addr, size_t len, int inherit) 607 */ 608 int 609 minherit(struct minherit_args *uap) 610 { 611 struct proc *p = curproc; 612 vm_offset_t addr; 613 vm_size_t size, pageoff; 614 register vm_inherit_t inherit; 615 616 addr = (vm_offset_t)uap->addr; 617 size = uap->len; 618 inherit = uap->inherit; 619 620 pageoff = (addr & PAGE_MASK); 621 addr -= pageoff; 622 size += pageoff; 623 size = (vm_size_t) round_page(size); 624 if (addr + size < addr) 625 return(EINVAL); 626 627 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size, 628 inherit)) { 629 case KERN_SUCCESS: 630 return (0); 631 case KERN_PROTECTION_FAILURE: 632 return (EACCES); 633 } 634 return (EINVAL); 635 } 636 637 /* 638 * madvise_args(void *addr, size_t len, int behav) 639 */ 640 /* ARGSUSED */ 641 int 642 madvise(struct madvise_args *uap) 643 { 644 struct proc *p = curproc; 645 vm_offset_t start, end; 646 647 /* 648 * Check for illegal behavior 649 */ 650 if (uap->behav < 0 || uap->behav > MADV_CORE) 651 return (EINVAL); 652 /* 653 * Check for illegal addresses. Watch out for address wrap... Note 654 * that VM_*_ADDRESS are not constants due to casts (argh). 655 */ 656 if (VM_MAXUSER_ADDRESS > 0 && 657 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS) 658 return (EINVAL); 659 #ifndef i386 660 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS) 661 return (EINVAL); 662 #endif 663 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) 664 return (EINVAL); 665 666 /* 667 * Since this routine is only advisory, we default to conservative 668 * behavior. 669 */ 670 start = trunc_page((vm_offset_t) uap->addr); 671 end = round_page((vm_offset_t) uap->addr + uap->len); 672 673 if (vm_map_madvise(&p->p_vmspace->vm_map, start, end, uap->behav)) 674 return (EINVAL); 675 return (0); 676 } 677 678 /* 679 * mincore_args(const void *addr, size_t len, char *vec) 680 */ 681 /* ARGSUSED */ 682 int 683 mincore(struct mincore_args *uap) 684 { 685 struct proc *p = curproc; 686 vm_offset_t addr, first_addr; 687 vm_offset_t end, cend; 688 pmap_t pmap; 689 vm_map_t map; 690 char *vec; 691 int error; 692 int vecindex, lastvecindex; 693 register vm_map_entry_t current; 694 vm_map_entry_t entry; 695 int mincoreinfo; 696 unsigned int timestamp; 697 698 /* 699 * Make sure that the addresses presented are valid for user 700 * mode. 701 */ 702 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 703 end = addr + (vm_size_t)round_page(uap->len); 704 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS) 705 return (EINVAL); 706 if (end < addr) 707 return (EINVAL); 708 709 /* 710 * Address of byte vector 711 */ 712 vec = uap->vec; 713 714 map = &p->p_vmspace->vm_map; 715 pmap = vmspace_pmap(p->p_vmspace); 716 717 vm_map_lock_read(map); 718 RestartScan: 719 timestamp = map->timestamp; 720 721 if (!vm_map_lookup_entry(map, addr, &entry)) 722 entry = entry->next; 723 724 /* 725 * Do this on a map entry basis so that if the pages are not 726 * in the current processes address space, we can easily look 727 * up the pages elsewhere. 728 */ 729 lastvecindex = -1; 730 for(current = entry; 731 (current != &map->header) && (current->start < end); 732 current = current->next) { 733 734 /* 735 * ignore submaps (for now) or null objects 736 */ 737 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 738 current->object.vm_object == NULL) 739 continue; 740 741 /* 742 * limit this scan to the current map entry and the 743 * limits for the mincore call 744 */ 745 if (addr < current->start) 746 addr = current->start; 747 cend = current->end; 748 if (cend > end) 749 cend = end; 750 751 /* 752 * scan this entry one page at a time 753 */ 754 while(addr < cend) { 755 /* 756 * Check pmap first, it is likely faster, also 757 * it can provide info as to whether we are the 758 * one referencing or modifying the page. 759 */ 760 mincoreinfo = pmap_mincore(pmap, addr); 761 if (!mincoreinfo) { 762 vm_pindex_t pindex; 763 vm_ooffset_t offset; 764 vm_page_t m; 765 /* 766 * calculate the page index into the object 767 */ 768 offset = current->offset + (addr - current->start); 769 pindex = OFF_TO_IDX(offset); 770 m = vm_page_lookup(current->object.vm_object, 771 pindex); 772 /* 773 * if the page is resident, then gather information about 774 * it. 775 */ 776 if (m) { 777 mincoreinfo = MINCORE_INCORE; 778 if (m->dirty || 779 pmap_is_modified(m)) 780 mincoreinfo |= MINCORE_MODIFIED_OTHER; 781 if ((m->flags & PG_REFERENCED) || 782 pmap_ts_referenced(m)) { 783 vm_page_flag_set(m, PG_REFERENCED); 784 mincoreinfo |= MINCORE_REFERENCED_OTHER; 785 } 786 } 787 } 788 789 /* 790 * subyte may page fault. In case it needs to modify 791 * the map, we release the lock. 792 */ 793 vm_map_unlock_read(map); 794 795 /* 796 * calculate index into user supplied byte vector 797 */ 798 vecindex = OFF_TO_IDX(addr - first_addr); 799 800 /* 801 * If we have skipped map entries, we need to make sure that 802 * the byte vector is zeroed for those skipped entries. 803 */ 804 while((lastvecindex + 1) < vecindex) { 805 error = subyte( vec + lastvecindex, 0); 806 if (error) { 807 return (EFAULT); 808 } 809 ++lastvecindex; 810 } 811 812 /* 813 * Pass the page information to the user 814 */ 815 error = subyte( vec + vecindex, mincoreinfo); 816 if (error) { 817 return (EFAULT); 818 } 819 820 /* 821 * If the map has changed, due to the subyte, the previous 822 * output may be invalid. 823 */ 824 vm_map_lock_read(map); 825 if (timestamp != map->timestamp) 826 goto RestartScan; 827 828 lastvecindex = vecindex; 829 addr += PAGE_SIZE; 830 } 831 } 832 833 /* 834 * subyte may page fault. In case it needs to modify 835 * the map, we release the lock. 836 */ 837 vm_map_unlock_read(map); 838 839 /* 840 * Zero the last entries in the byte vector. 841 */ 842 vecindex = OFF_TO_IDX(end - first_addr); 843 while((lastvecindex + 1) < vecindex) { 844 error = subyte( vec + lastvecindex, 0); 845 if (error) { 846 return (EFAULT); 847 } 848 ++lastvecindex; 849 } 850 851 /* 852 * If the map has changed, due to the subyte, the previous 853 * output may be invalid. 854 */ 855 vm_map_lock_read(map); 856 if (timestamp != map->timestamp) 857 goto RestartScan; 858 vm_map_unlock_read(map); 859 860 return (0); 861 } 862 863 /* 864 * mlock_args(const void *addr, size_t len) 865 */ 866 int 867 mlock(struct mlock_args *uap) 868 { 869 vm_offset_t addr; 870 vm_size_t size, pageoff; 871 int error; 872 struct proc *p = curproc; 873 874 addr = (vm_offset_t) uap->addr; 875 size = uap->len; 876 877 pageoff = (addr & PAGE_MASK); 878 addr -= pageoff; 879 size += pageoff; 880 size = (vm_size_t) round_page(size); 881 882 /* disable wrap around */ 883 if (addr + size < addr) 884 return (EINVAL); 885 886 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) 887 return (EAGAIN); 888 889 #ifdef pmap_wired_count 890 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 891 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 892 return (ENOMEM); 893 #else 894 error = suser_cred(p->p_ucred, 0); 895 if (error) 896 return (error); 897 #endif 898 899 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, FALSE); 900 return (error == KERN_SUCCESS ? 0 : ENOMEM); 901 } 902 903 /* 904 * mlockall_args(int how) 905 */ 906 int 907 mlockall(struct mlockall_args *uap) 908 { 909 return 0; 910 } 911 912 /* 913 * mlockall_args(int how) 914 */ 915 int 916 munlockall(struct munlockall_args *uap) 917 { 918 return 0; 919 } 920 921 /* 922 * munlock_args(const void *addr, size_t len) 923 */ 924 int 925 munlock(struct munlock_args *uap) 926 { 927 struct thread *td = curthread; 928 struct proc *p = td->td_proc; 929 vm_offset_t addr; 930 vm_size_t size, pageoff; 931 int error; 932 933 addr = (vm_offset_t) uap->addr; 934 size = uap->len; 935 936 pageoff = (addr & PAGE_MASK); 937 addr -= pageoff; 938 size += pageoff; 939 size = (vm_size_t) round_page(size); 940 941 /* disable wrap around */ 942 if (addr + size < addr) 943 return (EINVAL); 944 945 #ifndef pmap_wired_count 946 error = suser(td); 947 if (error) 948 return (error); 949 #endif 950 951 error = vm_map_user_pageable(&p->p_vmspace->vm_map, addr, addr + size, TRUE); 952 return (error == KERN_SUCCESS ? 0 : ENOMEM); 953 } 954 955 /* 956 * Internal version of mmap. 957 * Currently used by mmap, exec, and sys5 shared memory. 958 * Handle is either a vnode pointer or NULL for MAP_ANON. 959 */ 960 int 961 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 962 vm_prot_t maxprot, int flags, 963 void *handle, 964 vm_ooffset_t foff) 965 { 966 boolean_t fitit; 967 vm_object_t object; 968 struct vnode *vp = NULL; 969 objtype_t type; 970 int rv = KERN_SUCCESS; 971 vm_ooffset_t objsize; 972 int docow; 973 struct thread *td = curthread; /* XXX */ 974 struct proc *p = td->td_proc; 975 976 KKASSERT(p); 977 978 if (size == 0) 979 return (0); 980 981 objsize = size = round_page(size); 982 983 if (p->p_vmspace->vm_map.size + size > 984 p->p_rlimit[RLIMIT_VMEM].rlim_cur) { 985 return(ENOMEM); 986 } 987 988 /* 989 * We currently can only deal with page aligned file offsets. 990 * The check is here rather than in the syscall because the 991 * kernel calls this function internally for other mmaping 992 * operations (such as in exec) and non-aligned offsets will 993 * cause pmap inconsistencies...so we want to be sure to 994 * disallow this in all cases. 995 */ 996 if (foff & PAGE_MASK) 997 return (EINVAL); 998 999 if ((flags & MAP_FIXED) == 0) { 1000 fitit = TRUE; 1001 *addr = round_page(*addr); 1002 } else { 1003 if (*addr != trunc_page(*addr)) 1004 return (EINVAL); 1005 fitit = FALSE; 1006 (void) vm_map_remove(map, *addr, *addr + size); 1007 } 1008 1009 /* 1010 * Lookup/allocate object. 1011 */ 1012 if (flags & MAP_ANON) { 1013 type = OBJT_DEFAULT; 1014 /* 1015 * Unnamed anonymous regions always start at 0. 1016 */ 1017 if (handle == 0) 1018 foff = 0; 1019 } else { 1020 vp = (struct vnode *) handle; 1021 if (vp->v_type == VCHR) { 1022 type = OBJT_DEVICE; 1023 handle = (void *)(intptr_t)vp->v_rdev; 1024 } else { 1025 struct vattr vat; 1026 int error; 1027 1028 error = VOP_GETATTR(vp, &vat, td); 1029 if (error) 1030 return (error); 1031 objsize = round_page(vat.va_size); 1032 type = OBJT_VNODE; 1033 /* 1034 * if it is a regular file without any references 1035 * we do not need to sync it. 1036 */ 1037 if (vp->v_type == VREG && vat.va_nlink == 0) { 1038 flags |= MAP_NOSYNC; 1039 } 1040 } 1041 } 1042 1043 if (handle == NULL) { 1044 object = NULL; 1045 docow = 0; 1046 } else { 1047 object = vm_pager_allocate(type, 1048 handle, objsize, prot, foff); 1049 if (object == NULL) 1050 return (type == OBJT_DEVICE ? EINVAL : ENOMEM); 1051 docow = MAP_PREFAULT_PARTIAL; 1052 } 1053 1054 /* 1055 * Force device mappings to be shared. 1056 */ 1057 if (type == OBJT_DEVICE || type == OBJT_PHYS) { 1058 flags &= ~(MAP_PRIVATE|MAP_COPY); 1059 flags |= MAP_SHARED; 1060 } 1061 1062 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1063 docow |= MAP_COPY_ON_WRITE; 1064 if (flags & MAP_NOSYNC) 1065 docow |= MAP_DISABLE_SYNCER; 1066 if (flags & MAP_NOCORE) 1067 docow |= MAP_DISABLE_COREDUMP; 1068 1069 #if defined(VM_PROT_READ_IS_EXEC) 1070 if (prot & VM_PROT_READ) 1071 prot |= VM_PROT_EXECUTE; 1072 1073 if (maxprot & VM_PROT_READ) 1074 maxprot |= VM_PROT_EXECUTE; 1075 #endif 1076 1077 if (fitit) { 1078 *addr = pmap_addr_hint(object, *addr, size); 1079 } 1080 1081 if (flags & MAP_STACK) 1082 rv = vm_map_stack (map, *addr, size, prot, 1083 maxprot, docow); 1084 else 1085 rv = vm_map_find(map, object, foff, addr, size, fitit, 1086 prot, maxprot, docow); 1087 1088 if (rv != KERN_SUCCESS) { 1089 /* 1090 * Lose the object reference. Will destroy the 1091 * object if it's an unnamed anonymous mapping 1092 * or named anonymous without other references. 1093 */ 1094 vm_object_deallocate(object); 1095 goto out; 1096 } 1097 1098 /* 1099 * Shared memory is also shared with children. 1100 */ 1101 if (flags & (MAP_SHARED|MAP_INHERIT)) { 1102 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1103 if (rv != KERN_SUCCESS) { 1104 (void) vm_map_remove(map, *addr, *addr + size); 1105 goto out; 1106 } 1107 } 1108 out: 1109 switch (rv) { 1110 case KERN_SUCCESS: 1111 return (0); 1112 case KERN_INVALID_ADDRESS: 1113 case KERN_NO_SPACE: 1114 return (ENOMEM); 1115 case KERN_PROTECTION_FAILURE: 1116 return (EACCES); 1117 default: 1118 return (EINVAL); 1119 } 1120 } 1121