1 /* $NetBSD: uvm_mmap.c,v 1.128 2009/03/29 01:02:51 mrg 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 1.128 2009/03/29 01:02:51 mrg 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 #if NVERIEXEC > 0 506 if (handle != NULL) { 507 /* 508 * Check if the file can be executed indirectly. 509 * 510 * XXX: This gives false warnings about "Incorrect access type" 511 * XXX: if the mapping is not executable. Harmless, but will be 512 * XXX: fixed as part of other changes. 513 */ 514 if (veriexec_verify(l, handle, "(mmap)", VERIEXEC_INDIRECT, 515 NULL)) { 516 /* 517 * Don't allow executable mappings if we can't 518 * indirectly execute the file. 519 */ 520 if (prot & VM_PROT_EXECUTE) { 521 if (fp != NULL) 522 fd_putfile(fd); 523 return (EPERM); 524 } 525 526 /* 527 * Strip the executable bit from 'maxprot' to make sure 528 * it can't be made executable later. 529 */ 530 maxprot &= ~VM_PROT_EXECUTE; 531 } 532 } 533 #endif /* NVERIEXEC > 0 */ 534 535 #ifdef PAX_MPROTECT 536 pax_mprotect(l, &prot, &maxprot); 537 #endif /* PAX_MPROTECT */ 538 539 #ifdef PAX_ASLR 540 pax_aslr(l, &addr, orig_addr, flags); 541 #endif /* PAX_ASLR */ 542 543 /* 544 * now let kernel internal function uvm_mmap do the work. 545 */ 546 547 error = uvm_mmap(&p->p_vmspace->vm_map, &addr, size, prot, maxprot, 548 flags, handle, pos, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur); 549 550 if (error == 0) 551 /* remember to add offset */ 552 *retval = (register_t)(addr + pageoff); 553 554 if (fp != NULL) 555 fd_putfile(fd); 556 557 return (error); 558 } 559 560 /* 561 * sys___msync13: the msync system call (a front-end for flush) 562 */ 563 564 int 565 sys___msync13(struct lwp *l, const struct sys___msync13_args *uap, register_t *retval) 566 { 567 /* { 568 syscallarg(void *) addr; 569 syscallarg(size_t) len; 570 syscallarg(int) flags; 571 } */ 572 struct proc *p = l->l_proc; 573 vaddr_t addr; 574 vsize_t size, pageoff; 575 struct vm_map *map; 576 int error, rv, flags, uvmflags; 577 578 /* 579 * extract syscall args from the uap 580 */ 581 582 addr = (vaddr_t)SCARG(uap, addr); 583 size = (vsize_t)SCARG(uap, len); 584 flags = SCARG(uap, flags); 585 586 /* sanity check flags */ 587 if ((flags & ~(MS_ASYNC | MS_SYNC | MS_INVALIDATE)) != 0 || 588 (flags & (MS_ASYNC | MS_SYNC | MS_INVALIDATE)) == 0 || 589 (flags & (MS_ASYNC | MS_SYNC)) == (MS_ASYNC | MS_SYNC)) 590 return (EINVAL); 591 if ((flags & (MS_ASYNC | MS_SYNC)) == 0) 592 flags |= MS_SYNC; 593 594 /* 595 * align the address to a page boundary and adjust the size accordingly. 596 */ 597 598 pageoff = (addr & PAGE_MASK); 599 addr -= pageoff; 600 size += pageoff; 601 size = (vsize_t)round_page(size); 602 603 error = range_test(addr, size, false); 604 if (error) 605 return error; 606 607 /* 608 * get map 609 */ 610 611 map = &p->p_vmspace->vm_map; 612 613 /* 614 * XXXCDC: do we really need this semantic? 615 * 616 * XXX Gak! If size is zero we are supposed to sync "all modified 617 * pages with the region containing addr". Unfortunately, we 618 * don't really keep track of individual mmaps so we approximate 619 * by flushing the range of the map entry containing addr. 620 * This can be incorrect if the region splits or is coalesced 621 * with a neighbor. 622 */ 623 624 if (size == 0) { 625 struct vm_map_entry *entry; 626 627 vm_map_lock_read(map); 628 rv = uvm_map_lookup_entry(map, addr, &entry); 629 if (rv == true) { 630 addr = entry->start; 631 size = entry->end - entry->start; 632 } 633 vm_map_unlock_read(map); 634 if (rv == false) 635 return (EINVAL); 636 } 637 638 /* 639 * translate MS_ flags into PGO_ flags 640 */ 641 642 uvmflags = PGO_CLEANIT; 643 if (flags & MS_INVALIDATE) 644 uvmflags |= PGO_FREE; 645 if (flags & MS_SYNC) 646 uvmflags |= PGO_SYNCIO; 647 648 error = uvm_map_clean(map, addr, addr+size, uvmflags); 649 return error; 650 } 651 652 /* 653 * sys_munmap: unmap a users memory 654 */ 655 656 int 657 sys_munmap(struct lwp *l, const struct sys_munmap_args *uap, register_t *retval) 658 { 659 /* { 660 syscallarg(void *) addr; 661 syscallarg(size_t) len; 662 } */ 663 struct proc *p = l->l_proc; 664 vaddr_t addr; 665 vsize_t size, pageoff; 666 struct vm_map *map; 667 struct vm_map_entry *dead_entries; 668 int error; 669 670 /* 671 * get syscall args. 672 */ 673 674 addr = (vaddr_t)SCARG(uap, addr); 675 size = (vsize_t)SCARG(uap, len); 676 677 /* 678 * align the address to a page boundary and adjust the size accordingly. 679 */ 680 681 pageoff = (addr & PAGE_MASK); 682 addr -= pageoff; 683 size += pageoff; 684 size = (vsize_t)round_page(size); 685 686 if (size == 0) 687 return (0); 688 689 error = range_test(addr, size, false); 690 if (error) 691 return error; 692 693 map = &p->p_vmspace->vm_map; 694 695 /* 696 * interesting system call semantic: make sure entire range is 697 * allocated before allowing an unmap. 698 */ 699 700 vm_map_lock(map); 701 #if 0 702 if (!uvm_map_checkprot(map, addr, addr + size, VM_PROT_NONE)) { 703 vm_map_unlock(map); 704 return (EINVAL); 705 } 706 #endif 707 uvm_unmap_remove(map, addr, addr + size, &dead_entries, NULL, 0); 708 vm_map_unlock(map); 709 if (dead_entries != NULL) 710 uvm_unmap_detach(dead_entries, 0); 711 return (0); 712 } 713 714 /* 715 * sys_mprotect: the mprotect system call 716 */ 717 718 int 719 sys_mprotect(struct lwp *l, const struct sys_mprotect_args *uap, register_t *retval) 720 { 721 /* { 722 syscallarg(void *) addr; 723 syscallarg(size_t) len; 724 syscallarg(int) prot; 725 } */ 726 struct proc *p = l->l_proc; 727 vaddr_t addr; 728 vsize_t size, pageoff; 729 vm_prot_t prot; 730 int error; 731 732 /* 733 * extract syscall args from uap 734 */ 735 736 addr = (vaddr_t)SCARG(uap, addr); 737 size = (vsize_t)SCARG(uap, len); 738 prot = SCARG(uap, prot) & VM_PROT_ALL; 739 740 /* 741 * align the address to a page boundary and adjust the size accordingly. 742 */ 743 744 pageoff = (addr & PAGE_MASK); 745 addr -= pageoff; 746 size += pageoff; 747 size = round_page(size); 748 749 error = range_test(addr, size, false); 750 if (error) 751 return error; 752 753 error = uvm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot, 754 false); 755 return error; 756 } 757 758 /* 759 * sys_minherit: the minherit system call 760 */ 761 762 int 763 sys_minherit(struct lwp *l, const struct sys_minherit_args *uap, register_t *retval) 764 { 765 /* { 766 syscallarg(void *) addr; 767 syscallarg(int) len; 768 syscallarg(int) inherit; 769 } */ 770 struct proc *p = l->l_proc; 771 vaddr_t addr; 772 vsize_t size, pageoff; 773 vm_inherit_t inherit; 774 int error; 775 776 addr = (vaddr_t)SCARG(uap, addr); 777 size = (vsize_t)SCARG(uap, len); 778 inherit = SCARG(uap, inherit); 779 780 /* 781 * align the address to a page boundary and adjust the size accordingly. 782 */ 783 784 pageoff = (addr & PAGE_MASK); 785 addr -= pageoff; 786 size += pageoff; 787 size = (vsize_t)round_page(size); 788 789 error = range_test(addr, size, false); 790 if (error) 791 return error; 792 793 error = uvm_map_inherit(&p->p_vmspace->vm_map, addr, addr + size, 794 inherit); 795 return error; 796 } 797 798 /* 799 * sys_madvise: give advice about memory usage. 800 */ 801 802 /* ARGSUSED */ 803 int 804 sys_madvise(struct lwp *l, const struct sys_madvise_args *uap, register_t *retval) 805 { 806 /* { 807 syscallarg(void *) addr; 808 syscallarg(size_t) len; 809 syscallarg(int) behav; 810 } */ 811 struct proc *p = l->l_proc; 812 vaddr_t addr; 813 vsize_t size, pageoff; 814 int advice, error; 815 816 addr = (vaddr_t)SCARG(uap, addr); 817 size = (vsize_t)SCARG(uap, len); 818 advice = SCARG(uap, behav); 819 820 /* 821 * align the address to a page boundary, and adjust the size accordingly 822 */ 823 824 pageoff = (addr & PAGE_MASK); 825 addr -= pageoff; 826 size += pageoff; 827 size = (vsize_t)round_page(size); 828 829 error = range_test(addr, size, false); 830 if (error) 831 return error; 832 833 switch (advice) { 834 case MADV_NORMAL: 835 case MADV_RANDOM: 836 case MADV_SEQUENTIAL: 837 error = uvm_map_advice(&p->p_vmspace->vm_map, addr, addr + size, 838 advice); 839 break; 840 841 case MADV_WILLNEED: 842 843 /* 844 * Activate all these pages, pre-faulting them in if 845 * necessary. 846 */ 847 /* 848 * XXX IMPLEMENT ME. 849 * Should invent a "weak" mode for uvm_fault() 850 * which would only do the PGO_LOCKED pgo_get(). 851 */ 852 853 return (0); 854 855 case MADV_DONTNEED: 856 857 /* 858 * Deactivate all these pages. We don't need them 859 * any more. We don't, however, toss the data in 860 * the pages. 861 */ 862 863 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size, 864 PGO_DEACTIVATE); 865 break; 866 867 case MADV_FREE: 868 869 /* 870 * These pages contain no valid data, and may be 871 * garbage-collected. Toss all resources, including 872 * any swap space in use. 873 */ 874 875 error = uvm_map_clean(&p->p_vmspace->vm_map, addr, addr + size, 876 PGO_FREE); 877 break; 878 879 case MADV_SPACEAVAIL: 880 881 /* 882 * XXXMRG What is this? I think it's: 883 * 884 * Ensure that we have allocated backing-store 885 * for these pages. 886 * 887 * This is going to require changes to the page daemon, 888 * as it will free swap space allocated to pages in core. 889 * There's also what to do for device/file/anonymous memory. 890 */ 891 892 return (EINVAL); 893 894 default: 895 return (EINVAL); 896 } 897 898 return error; 899 } 900 901 /* 902 * sys_mlock: memory lock 903 */ 904 905 int 906 sys_mlock(struct lwp *l, const struct sys_mlock_args *uap, register_t *retval) 907 { 908 /* { 909 syscallarg(const void *) addr; 910 syscallarg(size_t) len; 911 } */ 912 struct proc *p = l->l_proc; 913 vaddr_t addr; 914 vsize_t size, pageoff; 915 int error; 916 917 /* 918 * extract syscall args from uap 919 */ 920 921 addr = (vaddr_t)SCARG(uap, addr); 922 size = (vsize_t)SCARG(uap, len); 923 924 /* 925 * align the address to a page boundary and adjust the size accordingly 926 */ 927 928 pageoff = (addr & PAGE_MASK); 929 addr -= pageoff; 930 size += pageoff; 931 size = (vsize_t)round_page(size); 932 933 error = range_test(addr, size, false); 934 if (error) 935 return error; 936 937 if (atop(size) + uvmexp.wired > uvmexp.wiredmax) 938 return (EAGAIN); 939 940 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 941 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) 942 return (EAGAIN); 943 944 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, false, 945 0); 946 if (error == EFAULT) 947 error = ENOMEM; 948 return error; 949 } 950 951 /* 952 * sys_munlock: unlock wired pages 953 */ 954 955 int 956 sys_munlock(struct lwp *l, const struct sys_munlock_args *uap, register_t *retval) 957 { 958 /* { 959 syscallarg(const void *) addr; 960 syscallarg(size_t) len; 961 } */ 962 struct proc *p = l->l_proc; 963 vaddr_t addr; 964 vsize_t size, pageoff; 965 int error; 966 967 /* 968 * extract syscall args from uap 969 */ 970 971 addr = (vaddr_t)SCARG(uap, addr); 972 size = (vsize_t)SCARG(uap, len); 973 974 /* 975 * align the address to a page boundary, and adjust the size accordingly 976 */ 977 978 pageoff = (addr & PAGE_MASK); 979 addr -= pageoff; 980 size += pageoff; 981 size = (vsize_t)round_page(size); 982 983 error = range_test(addr, size, false); 984 if (error) 985 return error; 986 987 error = uvm_map_pageable(&p->p_vmspace->vm_map, addr, addr+size, true, 988 0); 989 if (error == EFAULT) 990 error = ENOMEM; 991 return error; 992 } 993 994 /* 995 * sys_mlockall: lock all pages mapped into an address space. 996 */ 997 998 int 999 sys_mlockall(struct lwp *l, const struct sys_mlockall_args *uap, register_t *retval) 1000 { 1001 /* { 1002 syscallarg(int) flags; 1003 } */ 1004 struct proc *p = l->l_proc; 1005 int error, flags; 1006 1007 flags = SCARG(uap, flags); 1008 1009 if (flags == 0 || 1010 (flags & ~(MCL_CURRENT|MCL_FUTURE)) != 0) 1011 return (EINVAL); 1012 1013 error = uvm_map_pageable_all(&p->p_vmspace->vm_map, flags, 1014 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur); 1015 return (error); 1016 } 1017 1018 /* 1019 * sys_munlockall: unlock all pages mapped into an address space. 1020 */ 1021 1022 int 1023 sys_munlockall(struct lwp *l, const void *v, register_t *retval) 1024 { 1025 struct proc *p = l->l_proc; 1026 1027 (void) uvm_map_pageable_all(&p->p_vmspace->vm_map, 0, 0); 1028 return (0); 1029 } 1030 1031 /* 1032 * uvm_mmap: internal version of mmap 1033 * 1034 * - used by sys_mmap and various framebuffers 1035 * - handle is a vnode pointer or NULL for MAP_ANON 1036 * - caller must page-align the file offset 1037 */ 1038 1039 int 1040 uvm_mmap(struct vm_map *map, vaddr_t *addr, vsize_t size, vm_prot_t prot, vm_prot_t maxprot, int flags, void *handle, voff_t foff, vsize_t locklimit) 1041 { 1042 struct uvm_object *uobj; 1043 struct vnode *vp; 1044 vaddr_t align = 0; 1045 int error; 1046 int advice = UVM_ADV_NORMAL; 1047 uvm_flag_t uvmflag = 0; 1048 bool needwritemap; 1049 1050 /* 1051 * check params 1052 */ 1053 1054 if (size == 0) 1055 return(0); 1056 if (foff & PAGE_MASK) 1057 return(EINVAL); 1058 if ((prot & maxprot) != prot) 1059 return(EINVAL); 1060 1061 /* 1062 * for non-fixed mappings, round off the suggested address. 1063 * for fixed mappings, check alignment and zap old mappings. 1064 */ 1065 1066 if ((flags & MAP_FIXED) == 0) { 1067 *addr = round_page(*addr); 1068 } else { 1069 if (*addr & PAGE_MASK) 1070 return(EINVAL); 1071 uvmflag |= UVM_FLAG_FIXED; 1072 (void) uvm_unmap(map, *addr, *addr + size); 1073 } 1074 1075 /* 1076 * Try to see if any requested alignment can even be attemped. 1077 * Make sure we can express the alignment (asking for a >= 4GB 1078 * alignment on an ILP32 architecure make no sense) and the 1079 * alignment is at least for a page sized quanitiy. If the 1080 * request was for a fixed mapping, make sure supplied address 1081 * adheres to the request alignment. 1082 */ 1083 align = (flags & MAP_ALIGNMENT_MASK) >> MAP_ALIGNMENT_SHIFT; 1084 if (align) { 1085 if (align >= sizeof(vaddr_t) * NBBY) 1086 return(EINVAL); 1087 align = 1L << align; 1088 if (align < PAGE_SIZE) 1089 return(EINVAL); 1090 if (align >= vm_map_max(map)) 1091 return(ENOMEM); 1092 if (flags & MAP_FIXED) { 1093 if ((*addr & (align-1)) != 0) 1094 return(EINVAL); 1095 align = 0; 1096 } 1097 } 1098 1099 /* 1100 * check resource limits 1101 */ 1102 1103 if (!VM_MAP_IS_KERNEL(map) && 1104 (((rlim_t)curproc->p_vmspace->vm_map.size + (rlim_t)size) > 1105 curproc->p_rlimit[RLIMIT_AS].rlim_cur)) 1106 return ENOMEM; 1107 1108 /* 1109 * handle anon vs. non-anon mappings. for non-anon mappings attach 1110 * to underlying vm object. 1111 */ 1112 1113 if (flags & MAP_ANON) { 1114 KASSERT(handle == NULL); 1115 foff = UVM_UNKNOWN_OFFSET; 1116 uobj = NULL; 1117 if ((flags & MAP_SHARED) == 0) 1118 /* XXX: defer amap create */ 1119 uvmflag |= UVM_FLAG_COPYONW; 1120 else 1121 /* shared: create amap now */ 1122 uvmflag |= UVM_FLAG_OVERLAY; 1123 1124 } else { 1125 KASSERT(handle != NULL); 1126 vp = (struct vnode *)handle; 1127 1128 /* 1129 * Don't allow mmap for EXEC if the file system 1130 * is mounted NOEXEC. 1131 */ 1132 if ((prot & PROT_EXEC) != 0 && 1133 (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) 1134 return (EACCES); 1135 1136 if (vp->v_type != VCHR) { 1137 error = VOP_MMAP(vp, prot, curlwp->l_cred); 1138 if (error) { 1139 return error; 1140 } 1141 vref(vp); 1142 uobj = &vp->v_uobj; 1143 1144 /* 1145 * If the vnode is being mapped with PROT_EXEC, 1146 * then mark it as text. 1147 */ 1148 if (prot & PROT_EXEC) { 1149 vn_markexec(vp); 1150 } 1151 } else { 1152 int i = maxprot; 1153 1154 /* 1155 * XXX Some devices don't like to be mapped with 1156 * XXX PROT_EXEC or PROT_WRITE, but we don't really 1157 * XXX have a better way of handling this, right now 1158 */ 1159 do { 1160 uobj = udv_attach((void *) &vp->v_rdev, 1161 (flags & MAP_SHARED) ? i : 1162 (i & ~VM_PROT_WRITE), foff, size); 1163 i--; 1164 } while ((uobj == NULL) && (i > 0)); 1165 advice = UVM_ADV_RANDOM; 1166 } 1167 if (uobj == NULL) 1168 return((vp->v_type == VREG) ? ENOMEM : EINVAL); 1169 if ((flags & MAP_SHARED) == 0) { 1170 uvmflag |= UVM_FLAG_COPYONW; 1171 } 1172 1173 /* 1174 * Set vnode flags to indicate the new kinds of mapping. 1175 * We take the vnode lock in exclusive mode here to serialize 1176 * with direct I/O. 1177 * 1178 * Safe to check for these flag values without a lock, as 1179 * long as a reference to the vnode is held. 1180 */ 1181 needwritemap = (vp->v_iflag & VI_WRMAP) == 0 && 1182 (flags & MAP_SHARED) != 0 && 1183 (maxprot & VM_PROT_WRITE) != 0; 1184 if ((vp->v_vflag & VV_MAPPED) == 0 || needwritemap) { 1185 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1186 vp->v_vflag |= VV_MAPPED; 1187 if (needwritemap) { 1188 mutex_enter(&vp->v_interlock); 1189 vp->v_iflag |= VI_WRMAP; 1190 mutex_exit(&vp->v_interlock); 1191 } 1192 VOP_UNLOCK(vp, 0); 1193 } 1194 } 1195 1196 uvmflag = UVM_MAPFLAG(prot, maxprot, 1197 (flags & MAP_SHARED) ? UVM_INH_SHARE : UVM_INH_COPY, 1198 advice, uvmflag); 1199 error = uvm_map(map, addr, size, uobj, foff, align, uvmflag); 1200 if (error) { 1201 if (uobj) 1202 uobj->pgops->pgo_detach(uobj); 1203 return error; 1204 } 1205 1206 /* 1207 * POSIX 1003.1b -- if our address space was configured 1208 * to lock all future mappings, wire the one we just made. 1209 * 1210 * Also handle the MAP_WIRED flag here. 1211 */ 1212 1213 if (prot == VM_PROT_NONE) { 1214 1215 /* 1216 * No more work to do in this case. 1217 */ 1218 1219 return (0); 1220 } 1221 if ((flags & MAP_WIRED) != 0 || (map->flags & VM_MAP_WIREFUTURE) != 0) { 1222 vm_map_lock(map); 1223 if (atop(size) + uvmexp.wired > uvmexp.wiredmax || 1224 (locklimit != 0 && 1225 size + ptoa(pmap_wired_count(vm_map_pmap(map))) > 1226 locklimit)) { 1227 vm_map_unlock(map); 1228 uvm_unmap(map, *addr, *addr + size); 1229 return ENOMEM; 1230 } 1231 1232 /* 1233 * uvm_map_pageable() always returns the map unlocked. 1234 */ 1235 1236 error = uvm_map_pageable(map, *addr, *addr + size, 1237 false, UVM_LK_ENTER); 1238 if (error) { 1239 uvm_unmap(map, *addr, *addr + size); 1240 return error; 1241 } 1242 return (0); 1243 } 1244 return 0; 1245 } 1246 1247 vaddr_t 1248 uvm_default_mapaddr(struct proc *p, vaddr_t base, vsize_t sz) 1249 { 1250 1251 return VM_DEFAULT_ADDRESS(base, sz); 1252 } 1253