1 /* $NetBSD: uvm_mmap.c,v 1.126 2008/06/03 21:48:27 ad Exp $ */ 2 3 /* 4 * Copyright (c) 1997 Charles D. Cranor and Washington University. 5 * Copyright (c) 1991, 1993 The Regents of the University of California. 6 * Copyright (c) 1988 University of Utah. 7 * 8 * All rights reserved. 9 * 10 * This code is derived from software contributed to Berkeley by 11 * the Systems Programming Group of the University of Utah Computer 12 * Science Department. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. All advertising materials mentioning features or use of this software 23 * must display the following acknowledgement: 24 * This product includes software developed by the Charles D. Cranor, 25 * Washington University, University of California, Berkeley and 26 * its contributors. 27 * 4. Neither the name of the University nor the names of its contributors 28 * may be used to endorse or promote products derived from this software 29 * without specific prior written permission. 30 * 31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 41 * SUCH DAMAGE. 42 * 43 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ 44 * @(#)vm_mmap.c 8.5 (Berkeley) 5/19/94 45 * from: Id: uvm_mmap.c,v 1.1.2.14 1998/01/05 21:04:26 chuck Exp 46 */ 47 48 /* 49 * uvm_mmap.c: system call interface into VM system, plus kernel vm_mmap 50 * function. 51 */ 52 53 #include <sys/cdefs.h> 54 __KERNEL_RCSID(0, "$NetBSD: uvm_mmap.c,v 1.126 2008/06/03 21:48:27 ad Exp $"); 55 56 #include "opt_compat_netbsd.h" 57 #include "opt_pax.h" 58 #include "veriexec.h" 59 60 #include <sys/param.h> 61 #include <sys/systm.h> 62 #include <sys/file.h> 63 #include <sys/filedesc.h> 64 #include <sys/resourcevar.h> 65 #include <sys/mman.h> 66 #include <sys/mount.h> 67 #include <sys/proc.h> 68 #include <sys/malloc.h> 69 #include <sys/vnode.h> 70 #include <sys/conf.h> 71 #include <sys/stat.h> 72 73 #if NVERIEXEC > 0 74 #include <sys/verified_exec.h> 75 #endif /* NVERIEXEC > 0 */ 76 77 #ifdef PAX_MPROTECT 78 #include <sys/pax.h> 79 #endif /* PAX_MPROTECT */ 80 81 #include <miscfs/specfs/specdev.h> 82 83 #include <sys/syscallargs.h> 84 85 #include <uvm/uvm.h> 86 #include <uvm/uvm_device.h> 87 88 #ifndef COMPAT_ZERODEV 89 #define COMPAT_ZERODEV(dev) (0) 90 #endif 91 92 static int 93 range_test(vaddr_t addr, vsize_t size, bool ismmap) 94 { 95 vaddr_t vm_min_address = VM_MIN_ADDRESS; 96 vaddr_t vm_max_address = VM_MAXUSER_ADDRESS; 97 vaddr_t eaddr = addr + size; 98 99 if (addr < vm_min_address) 100 return EINVAL; 101 if (eaddr > vm_max_address) 102 return ismmap ? EFBIG : EINVAL; 103 if (addr > eaddr) /* no wrapping! */ 104 return ismmap ? EOVERFLOW : EINVAL; 105 return 0; 106 } 107 108 /* 109 * unimplemented VM system calls: 110 */ 111 112 /* 113 * sys_sbrk: sbrk system call. 114 */ 115 116 /* ARGSUSED */ 117 int 118 sys_sbrk(struct lwp *l, const struct sys_sbrk_args *uap, register_t *retval) 119 { 120 /* { 121 syscallarg(intptr_t) incr; 122 } */ 123 124 return (ENOSYS); 125 } 126 127 /* 128 * sys_sstk: sstk system call. 129 */ 130 131 /* ARGSUSED */ 132 int 133 sys_sstk(struct lwp *l, const struct sys_sstk_args *uap, register_t *retval) 134 { 135 /* { 136 syscallarg(int) incr; 137 } */ 138 139 return (ENOSYS); 140 } 141 142 /* 143 * sys_mincore: determine if pages are in core or not. 144 */ 145 146 /* ARGSUSED */ 147 int 148 sys_mincore(struct lwp *l, const struct sys_mincore_args *uap, register_t *retval) 149 { 150 /* { 151 syscallarg(void *) addr; 152 syscallarg(size_t) len; 153 syscallarg(char *) vec; 154 } */ 155 struct proc *p = l->l_proc; 156 struct vm_page *pg; 157 char *vec, pgi; 158 struct uvm_object *uobj; 159 struct vm_amap *amap; 160 struct vm_anon *anon; 161 struct vm_map_entry *entry; 162 vaddr_t start, end, lim; 163 struct vm_map *map; 164 vsize_t len; 165 int error = 0, npgs; 166 167 map = &p->p_vmspace->vm_map; 168 169 start = (vaddr_t)SCARG(uap, addr); 170 len = SCARG(uap, len); 171 vec = SCARG(uap, vec); 172 173 if (start & PAGE_MASK) 174 return (EINVAL); 175 len = round_page(len); 176 end = start + len; 177 if (end <= start) 178 return (EINVAL); 179 180 /* 181 * Lock down vec, so our returned status isn't outdated by 182 * storing the status byte for a page. 183 */ 184 185 npgs = len >> PAGE_SHIFT; 186 error = uvm_vslock(p->p_vmspace, vec, npgs, VM_PROT_WRITE); 187 if (error) { 188 return error; 189 } 190 vm_map_lock_read(map); 191 192 if (uvm_map_lookup_entry(map, start, &entry) == false) { 193 error = ENOMEM; 194 goto out; 195 } 196 197 for (/* nothing */; 198 entry != &map->header && entry->start < end; 199 entry = entry->next) { 200 KASSERT(!UVM_ET_ISSUBMAP(entry)); 201 KASSERT(start >= entry->start); 202 203 /* Make sure there are no holes. */ 204 if (entry->end < end && 205 (entry->next == &map->header || 206 entry->next->start > entry->end)) { 207 error = ENOMEM; 208 goto out; 209 } 210 211 lim = end < entry->end ? end : entry->end; 212 213 /* 214 * Special case for objects with no "real" pages. Those 215 * are always considered resident (mapped devices). 216 */ 217 218 if (UVM_ET_ISOBJ(entry)) { 219 KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)); 220 if (UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) { 221 for (/* nothing */; start < lim; 222 start += PAGE_SIZE, vec++) 223 subyte(vec, 1); 224 continue; 225 } 226 } 227 228 amap = entry->aref.ar_amap; /* top layer */ 229 uobj = entry->object.uvm_obj; /* bottom layer */ 230 231 if (amap != NULL) 232 amap_lock(amap); 233 if (uobj != NULL) 234 mutex_enter(&uobj->vmobjlock); 235 236 for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) { 237 pgi = 0; 238 if (amap != NULL) { 239 /* Check the top layer first. */ 240 anon = amap_lookup(&entry->aref, 241 start - entry->start); 242 /* Don't need to lock anon here. */ 243 if (anon != NULL && anon->an_page != NULL) { 244 245 /* 246 * Anon has the page for this entry 247 * offset. 248 */ 249 250 pgi = 1; 251 } 252 } 253 if (uobj != NULL && pgi == 0) { 254 /* Check the bottom layer. */ 255 pg = uvm_pagelookup(uobj, 256 entry->offset + (start - entry->start)); 257 if (pg != NULL) { 258 259 /* 260 * Object has the page for this entry 261 * offset. 262 */ 263 264 pgi = 1; 265 } 266 } 267 (void) subyte(vec, pgi); 268 } 269 if (uobj != NULL) 270 mutex_exit(&uobj->vmobjlock); 271 if (amap != NULL) 272 amap_unlock(amap); 273 } 274 275 out: 276 vm_map_unlock_read(map); 277 uvm_vsunlock(p->p_vmspace, SCARG(uap, vec), npgs); 278 return (error); 279 } 280 281 /* 282 * sys_mmap: mmap system call. 283 * 284 * => file offset and address may not be page aligned 285 * - if MAP_FIXED, offset and address must have remainder mod PAGE_SIZE 286 * - if address isn't page aligned the mapping starts at trunc_page(addr) 287 * and the return value is adjusted up by the page offset. 288 */ 289 290 int 291 sys_mmap(struct lwp *l, const struct sys_mmap_args *uap, register_t *retval) 292 { 293 /* { 294 syscallarg(void *) addr; 295 syscallarg(size_t) len; 296 syscallarg(int) prot; 297 syscallarg(int) flags; 298 syscallarg(int) fd; 299 syscallarg(long) pad; 300 syscallarg(off_t) pos; 301 } */ 302 struct proc *p = l->l_proc; 303 vaddr_t addr; 304 struct vattr va; 305 off_t pos; 306 vsize_t size, pageoff; 307 vm_prot_t prot, maxprot; 308 int flags, fd; 309 vaddr_t defaddr; 310 struct file *fp = NULL; 311 struct vnode *vp; 312 void *handle; 313 int error; 314 #ifdef PAX_ASLR 315 vaddr_t orig_addr; 316 #endif /* PAX_ASLR */ 317 318 /* 319 * first, extract syscall args from the uap. 320 */ 321 322 addr = (vaddr_t)SCARG(uap, addr); 323 size = (vsize_t)SCARG(uap, len); 324 prot = SCARG(uap, prot) & VM_PROT_ALL; 325 flags = SCARG(uap, flags); 326 fd = SCARG(uap, fd); 327 pos = SCARG(uap, pos); 328 329 #ifdef PAX_ASLR 330 orig_addr = addr; 331 #endif /* PAX_ASLR */ 332 333 /* 334 * Fixup the old deprecated MAP_COPY into MAP_PRIVATE, and 335 * validate the flags. 336 */ 337 if (flags & MAP_COPY) 338 flags = (flags & ~MAP_COPY) | MAP_PRIVATE; 339 if ((flags & (MAP_SHARED|MAP_PRIVATE)) == (MAP_SHARED|MAP_PRIVATE)) 340 return (EINVAL); 341 342 /* 343 * align file position and save offset. adjust size. 344 */ 345 346 pageoff = (pos & PAGE_MASK); 347 pos -= pageoff; 348 size += pageoff; /* add offset */ 349 size = (vsize_t)round_page(size); /* round up */ 350 351 /* 352 * now check (MAP_FIXED) or get (!MAP_FIXED) the "addr" 353 */ 354 if (flags & MAP_FIXED) { 355 356 /* ensure address and file offset are aligned properly */ 357 addr -= pageoff; 358 if (addr & PAGE_MASK) 359 return (EINVAL); 360 361 error = range_test(addr, size, true); 362 if (error) 363 return error; 364 } else if (addr == 0 || !(flags & MAP_TRYFIXED)) { 365 366 /* 367 * not fixed: make sure we skip over the largest 368 * possible heap for non-topdown mapping arrangements. 369 * we will refine our guess later (e.g. to account for 370 * VAC, etc) 371 */ 372 373 defaddr = p->p_emul->e_vm_default_addr(p, 374 (vaddr_t)p->p_vmspace->vm_daddr, size); 375 376 if (addr == 0 || 377 !(p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN)) 378 addr = MAX(addr, defaddr); 379 else 380 addr = MIN(addr, defaddr); 381 } 382 383 /* 384 * check for file mappings (i.e. not anonymous) and verify file. 385 */ 386 387 if ((flags & MAP_ANON) == 0) { 388 if ((fp = fd_getfile(fd)) == NULL) 389 return (EBADF); 390 if (fp->f_type != DTYPE_VNODE) { 391 fd_putfile(fd); 392 return (ENODEV); /* only mmap vnodes! */ 393 } 394 vp = fp->f_data; /* convert to vnode */ 395 if (vp->v_type != VREG && vp->v_type != VCHR && 396 vp->v_type != VBLK) { 397 fd_putfile(fd); 398 return (ENODEV); /* only REG/CHR/BLK support mmap */ 399 } 400 if (vp->v_type != VCHR && pos < 0) { 401 fd_putfile(fd); 402 return (EINVAL); 403 } 404 if (vp->v_type != VCHR && (pos + size) < pos) { 405 fd_putfile(fd); 406 return (EOVERFLOW); /* no offset wrapping */ 407 } 408 409 /* special case: catch SunOS style /dev/zero */ 410 if (vp->v_type == VCHR 411 && (vp->v_rdev == zerodev || COMPAT_ZERODEV(vp->v_rdev))) { 412 flags |= MAP_ANON; 413 fd_putfile(fd); 414 fp = NULL; 415 goto is_anon; 416 } 417 418 /* 419 * Old programs may not select a specific sharing type, so 420 * default to an appropriate one. 421 * 422 * XXX: how does MAP_ANON fit in the picture? 423 */ 424 if ((flags & (MAP_SHARED|MAP_PRIVATE)) == 0) { 425 #if defined(DEBUG) 426 printf("WARNING: defaulted mmap() share type to " 427 "%s (pid %d command %s)\n", vp->v_type == VCHR ? 428 "MAP_SHARED" : "MAP_PRIVATE", p->p_pid, 429 p->p_comm); 430 #endif 431 if (vp->v_type == VCHR) 432 flags |= MAP_SHARED; /* for a device */ 433 else 434 flags |= MAP_PRIVATE; /* for a file */ 435 } 436 437 /* 438 * MAP_PRIVATE device mappings don't make sense (and aren't 439 * supported anyway). However, some programs rely on this, 440 * so just change it to MAP_SHARED. 441 */ 442 if (vp->v_type == VCHR && (flags & MAP_PRIVATE) != 0) { 443 flags = (flags & ~MAP_PRIVATE) | MAP_SHARED; 444 } 445 446 /* 447 * now check protection 448 */ 449 450 maxprot = VM_PROT_EXECUTE; 451 452 /* check read access */ 453 if (fp->f_flag & FREAD) 454 maxprot |= VM_PROT_READ; 455 else if (prot & PROT_READ) { 456 fd_putfile(fd); 457 return (EACCES); 458 } 459 460 /* check write access, shared case first */ 461 if (flags & MAP_SHARED) { 462 /* 463 * if the file is writable, only add PROT_WRITE to 464 * maxprot if the file is not immutable, append-only. 465 * otherwise, if we have asked for PROT_WRITE, return 466 * EPERM. 467 */ 468 if (fp->f_flag & FWRITE) { 469 if ((error = 470 VOP_GETATTR(vp, &va, l->l_cred))) { 471 fd_putfile(fd); 472 return (error); 473 } 474 if ((va.va_flags & 475 (SF_SNAPSHOT|IMMUTABLE|APPEND)) == 0) 476 maxprot |= VM_PROT_WRITE; 477 else if (prot & PROT_WRITE) { 478 fd_putfile(fd); 479 return (EPERM); 480 } 481 } 482 else if (prot & PROT_WRITE) { 483 fd_putfile(fd); 484 return (EACCES); 485 } 486 } else { 487 /* MAP_PRIVATE mappings can always write to */ 488 maxprot |= VM_PROT_WRITE; 489 } 490 handle = vp; 491 492 } else { /* MAP_ANON case */ 493 /* 494 * XXX What do we do about (MAP_SHARED|MAP_PRIVATE) == 0? 495 */ 496 if (fd != -1) 497 return (EINVAL); 498 499 is_anon: /* label for SunOS style /dev/zero */ 500 handle = NULL; 501 maxprot = VM_PROT_ALL; 502 pos = 0; 503 } 504 505 /* 506 * XXX (in)sanity check. We don't do proper datasize checking 507 * XXX for anonymous (or private writable) mmap(). However, 508 * XXX know that if we're trying to allocate more than the amount 509 * XXX remaining under our current data size limit, _that_ should 510 * XXX be disallowed. 511 */ 512 if ((flags & MAP_ANON) != 0 || 513 ((flags & MAP_PRIVATE) != 0 && (prot & PROT_WRITE) != 0)) { 514 if (size > 515 (p->p_rlimit[RLIMIT_DATA].rlim_cur - 516 ctob(p->p_vmspace->vm_dsize))) { 517 if (fp != NULL) 518 fd_putfile(fd); 519 return (ENOMEM); 520 } 521 } 522 523 #if NVERIEXEC > 0 524 if (handle != NULL) { 525 /* 526 * Check if the file can be executed indirectly. 527 * 528 * XXX: This gives false warnings about "Incorrect access type" 529 * XXX: if the mapping is not executable. Harmless, but will be 530 * XXX: fixed as part of other changes. 531 */ 532 if (veriexec_verify(l, handle, "(mmap)", VERIEXEC_INDIRECT, 533 NULL)) { 534 /* 535 * Don't allow executable mappings if we can't 536 * indirectly execute the file. 537 */ 538 if (prot & VM_PROT_EXECUTE) { 539 if (fp != NULL) 540 fd_putfile(fd); 541 return (EPERM); 542 } 543 544 /* 545 * Strip the executable bit from 'maxprot' to make sure 546 * it can't be made executable later. 547 */ 548 maxprot &= ~VM_PROT_EXECUTE; 549 } 550 } 551 #endif /* NVERIEXEC > 0 */ 552 553 #ifdef PAX_MPROTECT 554 pax_mprotect(l, &prot, &maxprot); 555 #endif /* PAX_MPROTECT */ 556 557 #ifdef PAX_ASLR 558 pax_aslr(l, &addr, orig_addr, flags); 559 #endif /* PAX_ASLR */ 560 561 /* 562 * now let kernel internal function uvm_mmap do the work. 563 */ 564 565 error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot, 566 flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur); 567 568 if (error == 0) 569 /* remember to add offset */ 570 *retval = (register_t)(addr + pageoff); 571 572 if (fp != NULL) 573 fd_putfile(fd); 574 575 return (error); 576 } 577 578 /* 579 * sys___msync13: the msync system call (a front-end for flush) 580 */ 581 582 int 583 sys___msync13(struct lwp *l, const struct sys___msync13_args *uap, register_t *retval) 584 { 585 /* { 586 syscallarg(void *) addr; 587 syscallarg(size_t) len; 588 syscallarg(int) flags; 589 } */ 590 struct proc *p = l->l_proc; 591 vaddr_t addr; 592 vsize_t size, pageoff; 593 struct vm_map *map; 594 int error, rv, flags, uvmflags; 595 596 /* 597 * extract syscall args from the uap 598 */ 599 600 addr = (vaddr_t)SCARG(uap, addr); 601 size = (vsize_t)SCARG(uap, len); 602 flags = SCARG(uap, flags); 603 604 /* sanity check flags */ 605 if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 || 606 (flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 || 607 (flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC)) 608 return (EINVAL); 609 if ((flags & (MS_ASYNC | MS_SYNC)) == 0) 610 flags |= MS_SYNC; 611 612 /* 613 * align the address to a page boundary and adjust the size accordingly. 614 */ 615 616 pageoff = (addr & PAGE_MASK); 617 addr -= pageoff; 618 size += pageoff; 619 size = (vsize_t)round_page(size); 620 621 error = range_test(addr, size, false); 622 if (error) 623 return error; 624 625 /* 626 * get map 627 */ 628 629 map = &p->p_vmspace->vm_map; 630 631 /* 632 * XXXCDC: do we really need this semantic? 633 * 634 * XXX Gak! If size is zero we are supposed to sync "all modified 635 * pages with the region containing addr". Unfortunately, we 636 * don't really keep track of individual mmaps so we approximate 637 * by flushing the range of the map entry containing addr. 638 * This can be incorrect if the region splits or is coalesced 639 * with a neighbor. 640 */ 641 642 if (size == 0) { 643 struct vm_map_entry *entry; 644 645 vm_map_lock_read(map); 646 rv = uvm_map_lookup_entry(map, addr, &entry); 647 if (rv == true) { 648 addr = entry->start; 649 size = entry->end - entry->start; 650 } 651 vm_map_unlock_read(map); 652 if (rv == false) 653 return (EINVAL); 654 } 655 656 /* 657 * translate MS_ flags into PGO_ flags 658 */ 659 660 uvmflags = PGO_CLEANIT; 661 if (flags & MS_INVALIDATE) 662 uvmflags |= PGO_FREE; 663 if (flags & MS_SYNC) 664 uvmflags |= PGO_SYNCIO; 665 666 error = uvm_map_clean(map, addr, addr+size, uvmflags); 667 return error; 668 } 669 670 /* 671 * sys_munmap: unmap a users memory 672 */ 673 674 int 675 sys_munmap(struct lwp *l, const struct sys_munmap_args *uap, register_t *retval) 676 { 677 /* { 678 syscallarg(void *) addr; 679 syscallarg(size_t) len; 680 } */ 681 struct proc *p = l->l_proc; 682 vaddr_t addr; 683 vsize_t size, pageoff; 684 struct vm_map *map; 685 struct vm_map_entry *dead_entries; 686 int error; 687 688 /* 689 * get syscall args. 690 */ 691 692 addr = (vaddr_t)SCARG(uap, addr); 693 size = (vsize_t)SCARG(uap, len); 694 695 /* 696 * align the address to a page boundary and adjust the size accordingly. 697 */ 698 699 pageoff = (addr & PAGE_MASK); 700 addr -= pageoff; 701 size += pageoff; 702 size = (vsize_t)round_page(size); 703 704 if (size == 0) 705 return (0); 706 707 error = range_test(addr, size, false); 708 if (error) 709 return error; 710 711 map = &p->p_vmspace->vm_map; 712 713 /* 714 * interesting system call semantic: make sure entire range is 715 * allocated before allowing an unmap. 716 */ 717 718 vm_map_lock(map); 719 #if 0 720 if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) { 721 vm_map_unlock(map); 722 return (EINVAL); 723 } 724 #endif 725 uvm_unmap_remove(map, addr, addr + size, &dead_entries, NULL, 0); 726 vm_map_unlock(map); 727 if (dead_entries != NULL) 728 uvm_unmap_detach(dead_entries, 0); 729 return (0); 730 } 731 732 /* 733 * sys_mprotect: the mprotect system call 734 */ 735 736 int 737 sys_mprotect(struct lwp *l, const struct sys_mprotect_args *uap, register_t *retval) 738 { 739 /* { 740 syscallarg(void *) addr; 741 syscallarg(size_t) len; 742 syscallarg(int) prot; 743 } */ 744 struct proc *p = l->l_proc; 745 vaddr_t addr; 746 vsize_t size, pageoff; 747 vm_prot_t prot; 748 int error; 749 750 /* 751 * extract syscall args from uap 752 */ 753 754 addr = (vaddr_t)SCARG(uap, addr); 755 size = (vsize_t)SCARG(uap, len); 756 prot = SCARG(uap, prot) & VM_PROT_ALL; 757 758 /* 759 * align the address to a page boundary and adjust the size accordingly. 760 */ 761 762 pageoff = (addr & PAGE_MASK); 763 addr -= pageoff; 764 size += pageoff; 765 size = round_page(size); 766 767 error = range_test(addr, size, false); 768 if (error) 769 return error; 770 771 error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, 772 false); 773 return error; 774 } 775 776 /* 777 * sys_minherit: the minherit system call 778 */ 779 780 int 781 sys_minherit(struct lwp *l, const struct sys_minherit_args *uap, register_t *retval) 782 { 783 /* { 784 syscallarg(void *) addr; 785 syscallarg(int) len; 786 syscallarg(int) inherit; 787 } */ 788 struct proc *p = l->l_proc; 789 vaddr_t addr; 790 vsize_t size, pageoff; 791 vm_inherit_t inherit; 792 int error; 793 794 addr = (vaddr_t)SCARG(uap, addr); 795 size = (vsize_t)SCARG(uap, len); 796 inherit = SCARG(uap, inherit); 797 798 /* 799 * align the address to a page boundary and adjust the size accordingly. 800 */ 801 802 pageoff = (addr & PAGE_MASK); 803 addr -= pageoff; 804 size += pageoff; 805 size = (vsize_t)round_page(size); 806 807 error = range_test(addr, size, false); 808 if (error) 809 return error; 810 811 error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size, 812 inherit); 813 return error; 814 } 815 816 /* 817 * sys_madvise: give advice about memory usage. 818 */ 819 820 /* ARGSUSED */ 821 int 822 sys_madvise(struct lwp *l, const struct sys_madvise_args *uap, register_t *retval) 823 { 824 /* { 825 syscallarg(void *) addr; 826 syscallarg(size_t) len; 827 syscallarg(int) behav; 828 } */ 829 struct proc *p = l->l_proc; 830 vaddr_t addr; 831 vsize_t size, pageoff; 832 int advice, error; 833 834 addr = (vaddr_t)SCARG(uap, addr); 835 size = (vsize_t)SCARG(uap, len); 836 advice = SCARG(uap, behav); 837 838 /* 839 * align the address to a page boundary, and adjust the size accordingly 840 */ 841 842 pageoff = (addr & PAGE_MASK); 843 addr -= pageoff; 844 size += pageoff; 845 size = (vsize_t)round_page(size); 846 847 error = range_test(addr, size, false); 848 if (error) 849 return error; 850 851 switch (advice) { 852 case MADV_NORMAL: 853 case MADV_RANDOM: 854 case MADV_SEQUENTIAL: 855 error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size, 856 advice); 857 break; 858 859 case MADV_WILLNEED: 860 861 /* 862 * Activate all these pages, pre-faulting them in if 863 * necessary. 864 */ 865 /* 866 * XXX IMPLEMENT ME. 867 * Should invent a "weak" mode for uvm_fault() 868 * which would only do the PGO_LOCKED pgo_get(). 869 */ 870 871 return (0); 872 873 case MADV_DONTNEED: 874 875 /* 876 * Deactivate all these pages. We don't need them 877 * any more. We don't, however, toss the data in 878 * the pages. 879 */ 880 881 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size, 882 PGO_DEACTIVATE); 883 break; 884 885 case MADV_FREE: 886 887 /* 888 * These pages contain no valid data, and may be 889 * garbage-collected. Toss all resources, including 890 * any swap space in use. 891 */ 892 893 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size, 894 PGO_FREE); 895 break; 896 897 case MADV_SPACEAVAIL: 898 899 /* 900 * XXXMRG What is this? I think it's: 901 * 902 * Ensure that we have allocated backing-store 903 * for these pages. 904 * 905 * This is going to require changes to the page daemon, 906 * as it will free swap space allocated to pages in core. 907 * There's also what to do for device/file/anonymous memory. 908 */ 909 910 return (EINVAL); 911 912 default: 913 return (EINVAL); 914 } 915 916 return error; 917 } 918 919 /* 920 * sys_mlock: memory lock 921 */ 922 923 int 924 sys_mlock(struct lwp *l, const struct sys_mlock_args *uap, register_t *retval) 925 { 926 /* { 927 syscallarg(const void *) addr; 928 syscallarg(size_t) len; 929 } */ 930 struct proc *p = l->l_proc; 931 vaddr_t addr; 932 vsize_t size, pageoff; 933 int error; 934 935 /* 936 * extract syscall args from uap 937 */ 938 939 addr = (vaddr_t)SCARG(uap, addr); 940 size = (vsize_t)SCARG(uap, len); 941 942 /* 943 * align the address to a page boundary and adjust the size accordingly 944 */ 945 946 pageoff = (addr & PAGE_MASK); 947 addr -= pageoff; 948 size += pageoff; 949 size = (vsize_t)round_page(size); 950 951 error = range_test(addr, size, false); 952 if (error) 953 return error; 954 955 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) 956 return (EAGAIN); 957 958 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 959 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 960 return (EAGAIN); 961 962 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, false, 963 0); 964 if (error == EFAULT) 965 error = ENOMEM; 966 return error; 967 } 968 969 /* 970 * sys_munlock: unlock wired pages 971 */ 972 973 int 974 sys_munlock(struct lwp *l, const struct sys_munlock_args *uap, register_t *retval) 975 { 976 /* { 977 syscallarg(const void *) addr; 978 syscallarg(size_t) len; 979 } */ 980 struct proc *p = l->l_proc; 981 vaddr_t addr; 982 vsize_t size, pageoff; 983 int error; 984 985 /* 986 * extract syscall args from uap 987 */ 988 989 addr = (vaddr_t)SCARG(uap, addr); 990 size = (vsize_t)SCARG(uap, len); 991 992 /* 993 * align the address to a page boundary, and adjust the size accordingly 994 */ 995 996 pageoff = (addr & PAGE_MASK); 997 addr -= pageoff; 998 size += pageoff; 999 size = (vsize_t)round_page(size); 1000 1001 error = range_test(addr, size, false); 1002 if (error) 1003 return error; 1004 1005 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, true, 1006 0); 1007 if (error == EFAULT) 1008 error = ENOMEM; 1009 return error; 1010 } 1011 1012 /* 1013 * sys_mlockall: lock all pages mapped into an address space. 1014 */ 1015 1016 int 1017 sys_mlockall(struct lwp *l, const struct sys_mlockall_args *uap, register_t *retval) 1018 { 1019 /* { 1020 syscallarg(int) flags; 1021 } */ 1022 struct proc *p = l->l_proc; 1023 int error, flags; 1024 1025 flags = SCARG(uap, flags); 1026 1027 if (flags == 0 || 1028 (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0) 1029 return (EINVAL); 1030 1031 error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags, 1032 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur); 1033 return (error); 1034 } 1035 1036 /* 1037 * sys_munlockall: unlock all pages mapped into an address space. 1038 */ 1039 1040 int 1041 sys_munlockall(struct lwp *l, const void *v, register_t *retval) 1042 { 1043 struct proc *p = l->l_proc; 1044 1045 (void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0); 1046 return (0); 1047 } 1048 1049 /* 1050 * uvm_mmap: internal version of mmap 1051 * 1052 * - used by sys_mmap and various framebuffers 1053 * - handle is a vnode pointer or NULL for MAP_ANON 1054 * - caller must page-align the file offset 1055 */ 1056 1057 int 1058 uvm_mmap(map, addr, size, prot, maxprot, flags, handle, foff, locklimit) 1059 struct vm_map *map; 1060 vaddr_t *addr; 1061 vsize_t size; 1062 vm_prot_t prot, maxprot; 1063 int flags; 1064 void *handle; 1065 voff_t foff; 1066 vsize_t locklimit; 1067 { 1068 struct uvm_object *uobj; 1069 struct vnode *vp; 1070 vaddr_t align = 0; 1071 int error; 1072 int advice = UVM_ADV_NORMAL; 1073 uvm_flag_t uvmflag = 0; 1074 bool needwritemap; 1075 1076 /* 1077 * check params 1078 */ 1079 1080 if (size == 0) 1081 return(0); 1082 if (foff & PAGE_MASK) 1083 return(EINVAL); 1084 if ((prot & maxprot) != prot) 1085 return(EINVAL); 1086 1087 /* 1088 * for non-fixed mappings, round off the suggested address. 1089 * for fixed mappings, check alignment and zap old mappings. 1090 */ 1091 1092 if ((flags & MAP_FIXED) == 0) { 1093 *addr = round_page(*addr); 1094 } else { 1095 if (*addr & PAGE_MASK) 1096 return(EINVAL); 1097 uvmflag |= UVM_FLAG_FIXED; 1098 (void) uvm_unmap(map, *addr, *addr + size); 1099 } 1100 1101 /* 1102 * Try to see if any requested alignment can even be attemped. 1103 * Make sure we can express the alignment (asking for a >= 4GB 1104 * alignment on an ILP32 architecure make no sense) and the 1105 * alignment is at least for a page sized quanitiy. If the 1106 * request was for a fixed mapping, make sure supplied address 1107 * adheres to the request alignment. 1108 */ 1109 align = (flags & MAP_ALIGNMENT_MASK) >> MAP_ALIGNMENT_SHIFT; 1110 if (align) { 1111 if (align >= sizeof(vaddr_t) * NBBY) 1112 return(EINVAL); 1113 align = 1L << align; 1114 if (align < PAGE_SIZE) 1115 return(EINVAL); 1116 if (align >= vm_map_max(map)) 1117 return(ENOMEM); 1118 if (flags & MAP_FIXED) { 1119 if ((*addr & (align-1)) != 0) 1120 return(EINVAL); 1121 align = 0; 1122 } 1123 } 1124 1125 /* 1126 * handle anon vs. non-anon mappings. for non-anon mappings attach 1127 * to underlying vm object. 1128 */ 1129 1130 if (flags & MAP_ANON) { 1131 KASSERT(handle == NULL); 1132 foff = UVM_UNKNOWN_OFFSET; 1133 uobj = NULL; 1134 if ((flags & MAP_SHARED) == 0) 1135 /* XXX: defer amap create */ 1136 uvmflag |= UVM_FLAG_COPYONW; 1137 else 1138 /* shared: create amap now */ 1139 uvmflag |= UVM_FLAG_OVERLAY; 1140 1141 } else { 1142 KASSERT(handle != NULL); 1143 vp = (struct vnode *)handle; 1144 1145 /* 1146 * Don't allow mmap for EXEC if the file system 1147 * is mounted NOEXEC. 1148 */ 1149 if ((prot & PROT_EXEC) != 0 && 1150 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) 1151 return (EACCES); 1152 1153 if (vp->v_type != VCHR) { 1154 error = VOP_MMAP(vp, prot, curlwp->l_cred); 1155 if (error) { 1156 return error; 1157 } 1158 vref(vp); 1159 uobj = &vp->v_uobj; 1160 1161 /* 1162 * If the vnode is being mapped with PROT_EXEC, 1163 * then mark it as text. 1164 */ 1165 if (prot & PROT_EXEC) { 1166 vn_markexec(vp); 1167 } 1168 } else { 1169 int i = maxprot; 1170 1171 /* 1172 * XXX Some devices don't like to be mapped with 1173 * XXX PROT_EXEC or PROT_WRITE, but we don't really 1174 * XXX have a better way of handling this, right now 1175 */ 1176 do { 1177 uobj = udv_attach((void *) &vp->v_rdev, 1178 (flags & MAP_SHARED) ? i : 1179 (i & ~VM_PROT_WRITE), foff, size); 1180 i--; 1181 } while ((uobj == NULL) && (i > 0)); 1182 advice = UVM_ADV_RANDOM; 1183 } 1184 if (uobj == NULL) 1185 return((vp->v_type == VREG) ? ENOMEM : EINVAL); 1186 if ((flags & MAP_SHARED) == 0) { 1187 uvmflag |= UVM_FLAG_COPYONW; 1188 } 1189 1190 /* 1191 * Set vnode flags to indicate the new kinds of mapping. 1192 * We take the vnode lock in exclusive mode here to serialize 1193 * with direct I/O. 1194 * 1195 * Safe to check for these flag values without a lock, as 1196 * long as a reference to the vnode is held. 1197 */ 1198 needwritemap = (vp->v_iflag & VI_WRMAP) == 0 && 1199 (flags & MAP_SHARED) != 0 && 1200 (maxprot & VM_PROT_WRITE) != 0; 1201 if ((vp->v_vflag & VV_MAPPED) == 0 || needwritemap) { 1202 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1203 vp->v_vflag |= VV_MAPPED; 1204 if (needwritemap) { 1205 mutex_enter(&vp->v_interlock); 1206 vp->v_iflag |= VI_WRMAP; 1207 mutex_exit(&vp->v_interlock); 1208 } 1209 VOP_UNLOCK(vp, 0); 1210 } 1211 } 1212 1213 uvmflag = UVM_MAPFLAG(prot, maxprot, 1214 (flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY, 1215 advice, uvmflag); 1216 error = uvm_map(map, addr, size, uobj, foff, align, uvmflag); 1217 if (error) { 1218 if (uobj) 1219 uobj->pgops->pgo_detach(uobj); 1220 return error; 1221 } 1222 1223 /* 1224 * POSIX 1003.1b -- if our address space was configured 1225 * to lock all future mappings, wire the one we just made. 1226 * 1227 * Also handle the MAP_WIRED flag here. 1228 */ 1229 1230 if (prot == VM_PROT_NONE) { 1231 1232 /* 1233 * No more work to do in this case. 1234 */ 1235 1236 return (0); 1237 } 1238 if ((flags & MAP_WIRED) != 0 || (map->flags & VM_MAP_WIREFUTURE) != 0) { 1239 vm_map_lock(map); 1240 if (atop(size) + uvmexp.wired > uvmexp.wiredmax || 1241 (locklimit != 0 && 1242 size + ptoa(pmap_wired_count(vm_map_pmap(map))) > 1243 locklimit)) { 1244 vm_map_unlock(map); 1245 uvm_unmap(map, *addr, *addr + size); 1246 return ENOMEM; 1247 } 1248 1249 /* 1250 * uvm_map_pageable() always returns the map unlocked. 1251 */ 1252 1253 error = uvm_map_pageable(map, *addr, *addr + size, 1254 false, UVM_LK_ENTER); 1255 if (error) { 1256 uvm_unmap(map, *addr, *addr + size); 1257 return error; 1258 } 1259 return (0); 1260 } 1261 return 0; 1262 } 1263 1264 vaddr_t 1265 uvm_default_mapaddr(struct proc *p, vaddr_t base, vsize_t sz) 1266 { 1267 1268 return VM_DEFAULT_ADDRESS(base, sz); 1269 } 1270