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 1.39 2007/04/30 07:18:57 dillon Exp $ 43 */ 44 45 /* 46 * Mapped file (mmap) interface to VM 47 */ 48 49 #include <sys/param.h> 50 #include <sys/kernel.h> 51 #include <sys/systm.h> 52 #include <sys/sysproto.h> 53 #include <sys/filedesc.h> 54 #include <sys/kern_syscall.h> 55 #include <sys/proc.h> 56 #include <sys/priv.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 #include <sys/thread2.h> 83 #include <sys/mplock2.h> 84 85 static int max_proc_mmap; 86 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); 87 int vkernel_enable; 88 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, ""); 89 90 /* 91 * Set the maximum number of vm_map_entry structures per process. Roughly 92 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 93 * of our KVM malloc space still results in generous limits. We want a 94 * default that is good enough to prevent the kernel running out of resources 95 * if attacked from compromised user account but generous enough such that 96 * multi-threaded processes are not unduly inconvenienced. 97 */ 98 99 static void vmmapentry_rsrc_init (void *); 100 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL) 101 102 static void 103 vmmapentry_rsrc_init(void *dummy) 104 { 105 max_proc_mmap = KvaSize / sizeof(struct vm_map_entry); 106 max_proc_mmap /= 100; 107 } 108 109 /* 110 * MPSAFE 111 */ 112 int 113 sys_sbrk(struct sbrk_args *uap) 114 { 115 /* Not yet implemented */ 116 return (EOPNOTSUPP); 117 } 118 119 /* 120 * sstk_args(int incr) 121 * 122 * MPSAFE 123 */ 124 int 125 sys_sstk(struct sstk_args *uap) 126 { 127 /* Not yet implemented */ 128 return (EOPNOTSUPP); 129 } 130 131 /* 132 * mmap_args(void *addr, size_t len, int prot, int flags, int fd, 133 * long pad, off_t pos) 134 * 135 * Memory Map (mmap) system call. Note that the file offset 136 * and address are allowed to be NOT page aligned, though if 137 * the MAP_FIXED flag it set, both must have the same remainder 138 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 139 * page-aligned, the actual mapping starts at trunc_page(addr) 140 * and the return value is adjusted up by the page offset. 141 * 142 * Generally speaking, only character devices which are themselves 143 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 144 * there would be no cache coherency between a descriptor and a VM mapping 145 * both to the same character device. 146 * 147 * Block devices can be mmap'd no matter what they represent. Cache coherency 148 * is maintained as long as you do not write directly to the underlying 149 * character device. 150 */ 151 152 int 153 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen, 154 int uprot, int uflags, int fd, off_t upos, void **res) 155 { 156 struct thread *td = curthread; 157 struct proc *p = td->td_proc; 158 struct file *fp = NULL; 159 struct vnode *vp; 160 vm_offset_t addr; 161 vm_offset_t tmpaddr; 162 vm_size_t size, pageoff; 163 vm_prot_t prot, maxprot; 164 void *handle; 165 int flags, error; 166 int disablexworkaround; 167 off_t pos; 168 vm_object_t obj; 169 170 KKASSERT(p); 171 172 addr = (vm_offset_t) uaddr; 173 size = ulen; 174 prot = uprot & VM_PROT_ALL; 175 flags = uflags; 176 pos = upos; 177 178 /* 179 * Make sure mapping fits into numeric range etc. 180 * 181 * NOTE: We support the full unsigned range for size now. 182 */ 183 if (((flags & MAP_ANON) && fd != -1)) 184 return (EINVAL); 185 186 if (flags & MAP_STACK) { 187 if ((fd != -1) || 188 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 189 return (EINVAL); 190 flags |= MAP_ANON; 191 pos = 0; 192 } 193 194 /* 195 * Virtual page tables cannot be used with MAP_STACK. Apart from 196 * it not making any sense, the aux union is used by both 197 * types. 198 * 199 * Because the virtual page table is stored in the backing object 200 * and might be updated by the kernel, the mapping must be R+W. 201 */ 202 if (flags & MAP_VPAGETABLE) { 203 if (vkernel_enable == 0) 204 return (EOPNOTSUPP); 205 if (flags & MAP_STACK) 206 return (EINVAL); 207 if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE)) 208 return (EINVAL); 209 } 210 211 /* 212 * Align the file position to a page boundary, 213 * and save its page offset component. 214 */ 215 pageoff = (pos & PAGE_MASK); 216 pos -= pageoff; 217 218 /* Adjust size for rounding (on both ends). */ 219 size += pageoff; /* low end... */ 220 size = (vm_size_t) round_page(size); /* hi end */ 221 if (size < ulen) /* wrap */ 222 return(EINVAL); 223 224 /* 225 * Check for illegal addresses. Watch out for address wrap... Note 226 * that VM_*_ADDRESS are not constants due to casts (argh). 227 */ 228 if (flags & (MAP_FIXED | MAP_TRYFIXED)) { 229 /* 230 * The specified address must have the same remainder 231 * as the file offset taken modulo PAGE_SIZE, so it 232 * should be aligned after adjustment by pageoff. 233 */ 234 addr -= pageoff; 235 if (addr & PAGE_MASK) 236 return (EINVAL); 237 238 /* 239 * Address range must be all in user VM space and not wrap. 240 */ 241 tmpaddr = addr + size; 242 if (tmpaddr < addr) 243 return (EINVAL); 244 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) 245 return (EINVAL); 246 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) 247 return (EINVAL); 248 } else { 249 /* 250 * Set a reasonable start point for the hint if it was 251 * not specified or if it falls within the heap space. 252 * Hinted mmap()s do not allocate out of the heap space. 253 */ 254 if (addr == 0 || 255 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 256 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz))) 257 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz); 258 } 259 260 if (flags & MAP_ANON) { 261 /* 262 * Mapping blank space is trivial. 263 */ 264 handle = NULL; 265 maxprot = VM_PROT_ALL; 266 pos = 0; 267 } else { 268 /* 269 * Mapping file, get fp for validation. Obtain vnode and make 270 * sure it is of appropriate type. 271 */ 272 fp = holdfp(p->p_fd, fd, -1); 273 if (fp == NULL) 274 return (EBADF); 275 if (fp->f_type != DTYPE_VNODE) { 276 error = EINVAL; 277 goto done; 278 } 279 /* 280 * POSIX shared-memory objects are defined to have 281 * kernel persistence, and are not defined to support 282 * read(2)/write(2) -- or even open(2). Thus, we can 283 * use MAP_ASYNC to trade on-disk coherence for speed. 284 * The shm_open(3) library routine turns on the FPOSIXSHM 285 * flag to request this behavior. 286 */ 287 if (fp->f_flag & FPOSIXSHM) 288 flags |= MAP_NOSYNC; 289 vp = (struct vnode *) fp->f_data; 290 291 /* 292 * Validate the vnode for the operation. 293 */ 294 switch(vp->v_type) { 295 case VREG: 296 /* 297 * Get the proper underlying object 298 */ 299 if ((obj = vp->v_object) == NULL) { 300 error = EINVAL; 301 goto done; 302 } 303 KKASSERT((struct vnode *)obj->handle == vp); 304 break; 305 case VCHR: 306 /* 307 * Make sure a device has not been revoked. 308 * Mappability is handled by the device layer. 309 */ 310 if (vp->v_rdev == NULL) { 311 error = EBADF; 312 goto done; 313 } 314 break; 315 default: 316 /* 317 * Nothing else is mappable. 318 */ 319 error = EINVAL; 320 goto done; 321 } 322 323 /* 324 * XXX hack to handle use of /dev/zero to map anon memory (ala 325 * SunOS). 326 */ 327 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 328 handle = NULL; 329 maxprot = VM_PROT_ALL; 330 flags |= MAP_ANON; 331 pos = 0; 332 } else { 333 /* 334 * cdevs does not provide private mappings of any kind. 335 */ 336 /* 337 * However, for XIG X server to continue to work, 338 * we should allow the superuser to do it anyway. 339 * We only allow it at securelevel < 1. 340 * (Because the XIG X server writes directly to video 341 * memory via /dev/mem, it should never work at any 342 * other securelevel. 343 * XXX this will have to go 344 */ 345 if (securelevel >= 1) 346 disablexworkaround = 1; 347 else 348 disablexworkaround = priv_check(td, PRIV_ROOT); 349 if (vp->v_type == VCHR && disablexworkaround && 350 (flags & (MAP_PRIVATE|MAP_COPY))) { 351 error = EINVAL; 352 goto done; 353 } 354 /* 355 * Ensure that file and memory protections are 356 * compatible. Note that we only worry about 357 * writability if mapping is shared; in this case, 358 * current and max prot are dictated by the open file. 359 * XXX use the vnode instead? Problem is: what 360 * credentials do we use for determination? What if 361 * proc does a setuid? 362 */ 363 maxprot = VM_PROT_EXECUTE; /* ??? */ 364 if (fp->f_flag & FREAD) { 365 maxprot |= VM_PROT_READ; 366 } else if (prot & PROT_READ) { 367 error = EACCES; 368 goto done; 369 } 370 /* 371 * If we are sharing potential changes (either via 372 * MAP_SHARED or via the implicit sharing of character 373 * device mappings), and we are trying to get write 374 * permission although we opened it without asking 375 * for it, bail out. Check for superuser, only if 376 * we're at securelevel < 1, to allow the XIG X server 377 * to continue to work. 378 */ 379 380 if ((flags & MAP_SHARED) != 0 || 381 (vp->v_type == VCHR && disablexworkaround)) { 382 if ((fp->f_flag & FWRITE) != 0) { 383 struct vattr va; 384 if ((error = VOP_GETATTR(vp, &va))) { 385 goto done; 386 } 387 if ((va.va_flags & 388 (IMMUTABLE|APPEND)) == 0) { 389 maxprot |= VM_PROT_WRITE; 390 } else if (prot & PROT_WRITE) { 391 error = EPERM; 392 goto done; 393 } 394 } else if ((prot & PROT_WRITE) != 0) { 395 error = EACCES; 396 goto done; 397 } 398 } else { 399 maxprot |= VM_PROT_WRITE; 400 } 401 handle = (void *)vp; 402 } 403 } 404 405 /* 406 * Do not allow more then a certain number of vm_map_entry structures 407 * per process. Scale with the number of rforks sharing the map 408 * to make the limit reasonable for threads. 409 */ 410 if (max_proc_mmap && 411 vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) { 412 error = ENOMEM; 413 goto done; 414 } 415 416 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 417 flags, handle, pos); 418 if (error == 0) 419 *res = (void *)(addr + pageoff); 420 done: 421 if (fp) 422 fdrop(fp); 423 return (error); 424 } 425 426 /* 427 * MPALMOSTSAFE 428 */ 429 int 430 sys_mmap(struct mmap_args *uap) 431 { 432 int error; 433 434 get_mplock(); 435 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len, 436 uap->prot, uap->flags, 437 uap->fd, uap->pos, &uap->sysmsg_resultp); 438 rel_mplock(); 439 440 return (error); 441 } 442 443 /* 444 * msync_args(void *addr, size_t len, int flags) 445 * 446 * MPALMOSTSAFE 447 */ 448 int 449 sys_msync(struct msync_args *uap) 450 { 451 struct proc *p = curproc; 452 vm_offset_t addr; 453 vm_offset_t tmpaddr; 454 vm_size_t size, pageoff; 455 int flags; 456 vm_map_t map; 457 int rv; 458 459 addr = (vm_offset_t) uap->addr; 460 size = uap->len; 461 flags = uap->flags; 462 463 pageoff = (addr & PAGE_MASK); 464 addr -= pageoff; 465 size += pageoff; 466 size = (vm_size_t) round_page(size); 467 if (size < uap->len) /* wrap */ 468 return(EINVAL); 469 tmpaddr = addr + size; /* workaround gcc4 opt */ 470 if (tmpaddr < addr) /* wrap */ 471 return(EINVAL); 472 473 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 474 return (EINVAL); 475 476 get_mplock(); 477 map = &p->p_vmspace->vm_map; 478 479 /* 480 * XXX Gak! If size is zero we are supposed to sync "all modified 481 * pages with the region containing addr". Unfortunately, we don't 482 * really keep track of individual mmaps so we approximate by flushing 483 * the range of the map entry containing addr. This can be incorrect 484 * if the region splits or is coalesced with a neighbor. 485 */ 486 if (size == 0) { 487 vm_map_entry_t entry; 488 489 vm_map_lock_read(map); 490 rv = vm_map_lookup_entry(map, addr, &entry); 491 if (rv == FALSE) { 492 vm_map_unlock_read(map); 493 rv = KERN_INVALID_ADDRESS; 494 goto done; 495 } 496 addr = entry->start; 497 size = entry->end - entry->start; 498 vm_map_unlock_read(map); 499 } 500 501 /* 502 * Clean the pages and interpret the return value. 503 */ 504 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 505 (flags & MS_INVALIDATE) != 0); 506 done: 507 rel_mplock(); 508 509 switch (rv) { 510 case KERN_SUCCESS: 511 break; 512 case KERN_INVALID_ADDRESS: 513 return (EINVAL); /* Sun returns ENOMEM? */ 514 case KERN_FAILURE: 515 return (EIO); 516 default: 517 return (EINVAL); 518 } 519 520 return (0); 521 } 522 523 /* 524 * munmap_args(void *addr, size_t len) 525 * 526 * MPALMOSTSAFE 527 */ 528 int 529 sys_munmap(struct munmap_args *uap) 530 { 531 struct proc *p = curproc; 532 vm_offset_t addr; 533 vm_offset_t tmpaddr; 534 vm_size_t size, pageoff; 535 vm_map_t map; 536 537 addr = (vm_offset_t) uap->addr; 538 size = uap->len; 539 540 pageoff = (addr & PAGE_MASK); 541 addr -= pageoff; 542 size += pageoff; 543 size = (vm_size_t) round_page(size); 544 if (size < uap->len) /* wrap */ 545 return(EINVAL); 546 tmpaddr = addr + size; /* workaround gcc4 opt */ 547 if (tmpaddr < addr) /* wrap */ 548 return(EINVAL); 549 550 if (size == 0) 551 return (0); 552 553 /* 554 * Check for illegal addresses. Watch out for address wrap... Note 555 * that VM_*_ADDRESS are not constants due to casts (argh). 556 */ 557 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) 558 return (EINVAL); 559 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) 560 return (EINVAL); 561 562 get_mplock(); 563 map = &p->p_vmspace->vm_map; 564 /* 565 * Make sure entire range is allocated. 566 */ 567 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) { 568 rel_mplock(); 569 return (EINVAL); 570 } 571 /* returns nothing but KERN_SUCCESS anyway */ 572 vm_map_remove(map, addr, addr + size); 573 rel_mplock(); 574 return (0); 575 } 576 577 /* 578 * mprotect_args(const void *addr, size_t len, int prot) 579 * 580 * MPALMOSTSAFE 581 */ 582 int 583 sys_mprotect(struct mprotect_args *uap) 584 { 585 struct proc *p = curproc; 586 vm_offset_t addr; 587 vm_offset_t tmpaddr; 588 vm_size_t size, pageoff; 589 vm_prot_t prot; 590 int error; 591 592 addr = (vm_offset_t) uap->addr; 593 size = uap->len; 594 prot = uap->prot & VM_PROT_ALL; 595 #if defined(VM_PROT_READ_IS_EXEC) 596 if (prot & VM_PROT_READ) 597 prot |= VM_PROT_EXECUTE; 598 #endif 599 600 pageoff = (addr & PAGE_MASK); 601 addr -= pageoff; 602 size += pageoff; 603 size = (vm_size_t) round_page(size); 604 if (size < uap->len) /* wrap */ 605 return(EINVAL); 606 tmpaddr = addr + size; /* workaround gcc4 opt */ 607 if (tmpaddr < addr) /* wrap */ 608 return(EINVAL); 609 610 get_mplock(); 611 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, 612 prot, FALSE)) { 613 case KERN_SUCCESS: 614 error = 0; 615 break; 616 case KERN_PROTECTION_FAILURE: 617 error = EACCES; 618 break; 619 default: 620 error = EINVAL; 621 break; 622 } 623 rel_mplock(); 624 return (error); 625 } 626 627 /* 628 * minherit_args(void *addr, size_t len, int inherit) 629 * 630 * MPALMOSTSAFE 631 */ 632 int 633 sys_minherit(struct minherit_args *uap) 634 { 635 struct proc *p = curproc; 636 vm_offset_t addr; 637 vm_offset_t tmpaddr; 638 vm_size_t size, pageoff; 639 vm_inherit_t inherit; 640 int error; 641 642 addr = (vm_offset_t)uap->addr; 643 size = uap->len; 644 inherit = uap->inherit; 645 646 pageoff = (addr & PAGE_MASK); 647 addr -= pageoff; 648 size += pageoff; 649 size = (vm_size_t) round_page(size); 650 if (size < uap->len) /* wrap */ 651 return(EINVAL); 652 tmpaddr = addr + size; /* workaround gcc4 opt */ 653 if (tmpaddr < addr) /* wrap */ 654 return(EINVAL); 655 656 get_mplock(); 657 658 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, 659 addr + size, inherit)) { 660 case KERN_SUCCESS: 661 error = 0; 662 break; 663 case KERN_PROTECTION_FAILURE: 664 error = EACCES; 665 break; 666 default: 667 error = EINVAL; 668 break; 669 } 670 rel_mplock(); 671 return (error); 672 } 673 674 /* 675 * madvise_args(void *addr, size_t len, int behav) 676 * 677 * MPALMOSTSAFE 678 */ 679 int 680 sys_madvise(struct madvise_args *uap) 681 { 682 struct proc *p = curproc; 683 vm_offset_t start, end; 684 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len; 685 int error; 686 687 /* 688 * Check for illegal behavior 689 */ 690 if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END) 691 return (EINVAL); 692 /* 693 * Check for illegal addresses. Watch out for address wrap... Note 694 * that VM_*_ADDRESS are not constants due to casts (argh). 695 */ 696 if (tmpaddr < (vm_offset_t)uap->addr) 697 return (EINVAL); 698 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) 699 return (EINVAL); 700 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS) 701 return (EINVAL); 702 703 /* 704 * Since this routine is only advisory, we default to conservative 705 * behavior. 706 */ 707 start = trunc_page((vm_offset_t)uap->addr); 708 end = round_page(tmpaddr); 709 710 get_mplock(); 711 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end, 712 uap->behav, 0); 713 rel_mplock(); 714 return (error); 715 } 716 717 /* 718 * mcontrol_args(void *addr, size_t len, int behav, off_t value) 719 * 720 * MPALMOSTSAFE 721 */ 722 int 723 sys_mcontrol(struct mcontrol_args *uap) 724 { 725 struct proc *p = curproc; 726 vm_offset_t start, end; 727 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len; 728 int error; 729 730 /* 731 * Check for illegal behavior 732 */ 733 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END) 734 return (EINVAL); 735 /* 736 * Check for illegal addresses. Watch out for address wrap... Note 737 * that VM_*_ADDRESS are not constants due to casts (argh). 738 */ 739 if (tmpaddr < (vm_offset_t) uap->addr) 740 return (EINVAL); 741 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) 742 return (EINVAL); 743 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS) 744 return (EINVAL); 745 746 /* 747 * Since this routine is only advisory, we default to conservative 748 * behavior. 749 */ 750 start = trunc_page((vm_offset_t)uap->addr); 751 end = round_page(tmpaddr); 752 753 get_mplock(); 754 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end, 755 uap->behav, uap->value); 756 rel_mplock(); 757 return (error); 758 } 759 760 761 /* 762 * mincore_args(const void *addr, size_t len, char *vec) 763 * 764 * MPALMOSTSAFE 765 */ 766 int 767 sys_mincore(struct mincore_args *uap) 768 { 769 struct proc *p = curproc; 770 vm_offset_t addr, first_addr; 771 vm_offset_t end, cend; 772 pmap_t pmap; 773 vm_map_t map; 774 char *vec; 775 int error; 776 int vecindex, lastvecindex; 777 vm_map_entry_t current; 778 vm_map_entry_t entry; 779 int mincoreinfo; 780 unsigned int timestamp; 781 782 /* 783 * Make sure that the addresses presented are valid for user 784 * mode. 785 */ 786 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 787 end = addr + (vm_size_t)round_page(uap->len); 788 if (end < addr) 789 return (EINVAL); 790 if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS) 791 return (EINVAL); 792 793 /* 794 * Address of byte vector 795 */ 796 vec = uap->vec; 797 798 map = &p->p_vmspace->vm_map; 799 pmap = vmspace_pmap(p->p_vmspace); 800 801 get_mplock(); 802 vm_map_lock_read(map); 803 RestartScan: 804 timestamp = map->timestamp; 805 806 if (!vm_map_lookup_entry(map, addr, &entry)) 807 entry = entry->next; 808 809 /* 810 * Do this on a map entry basis so that if the pages are not 811 * in the current processes address space, we can easily look 812 * up the pages elsewhere. 813 */ 814 lastvecindex = -1; 815 for(current = entry; 816 (current != &map->header) && (current->start < end); 817 current = current->next) { 818 819 /* 820 * ignore submaps (for now) or null objects 821 */ 822 if (current->maptype != VM_MAPTYPE_NORMAL && 823 current->maptype != VM_MAPTYPE_VPAGETABLE) { 824 continue; 825 } 826 if (current->object.vm_object == NULL) 827 continue; 828 829 /* 830 * limit this scan to the current map entry and the 831 * limits for the mincore call 832 */ 833 if (addr < current->start) 834 addr = current->start; 835 cend = current->end; 836 if (cend > end) 837 cend = end; 838 839 /* 840 * scan this entry one page at a time 841 */ 842 while (addr < cend) { 843 /* 844 * Check pmap first, it is likely faster, also 845 * it can provide info as to whether we are the 846 * one referencing or modifying the page. 847 * 848 * If we have to check the VM object, only mess 849 * around with normal maps. Do not mess around 850 * with virtual page tables (XXX). 851 */ 852 mincoreinfo = pmap_mincore(pmap, addr); 853 if (mincoreinfo == 0 && 854 current->maptype == VM_MAPTYPE_NORMAL) { 855 vm_pindex_t pindex; 856 vm_ooffset_t offset; 857 vm_page_t m; 858 859 /* 860 * calculate the page index into the object 861 */ 862 offset = current->offset + (addr - current->start); 863 pindex = OFF_TO_IDX(offset); 864 865 /* 866 * if the page is resident, then gather 867 * information about it. spl protection is 868 * required to maintain the object 869 * association. And XXX what if the page is 870 * busy? What's the deal with that? 871 */ 872 crit_enter(); 873 m = vm_page_lookup(current->object.vm_object, 874 pindex); 875 if (m && m->valid) { 876 mincoreinfo = MINCORE_INCORE; 877 if (m->dirty || 878 pmap_is_modified(m)) 879 mincoreinfo |= MINCORE_MODIFIED_OTHER; 880 if ((m->flags & PG_REFERENCED) || 881 pmap_ts_referenced(m)) { 882 vm_page_flag_set(m, PG_REFERENCED); 883 mincoreinfo |= MINCORE_REFERENCED_OTHER; 884 } 885 } 886 crit_exit(); 887 } 888 889 /* 890 * subyte may page fault. In case it needs to modify 891 * the map, we release the lock. 892 */ 893 vm_map_unlock_read(map); 894 895 /* 896 * calculate index into user supplied byte vector 897 */ 898 vecindex = OFF_TO_IDX(addr - first_addr); 899 900 /* 901 * If we have skipped map entries, we need to make sure that 902 * the byte vector is zeroed for those skipped entries. 903 */ 904 while((lastvecindex + 1) < vecindex) { 905 error = subyte( vec + lastvecindex, 0); 906 if (error) { 907 error = EFAULT; 908 goto done; 909 } 910 ++lastvecindex; 911 } 912 913 /* 914 * Pass the page information to the user 915 */ 916 error = subyte( vec + vecindex, mincoreinfo); 917 if (error) { 918 error = EFAULT; 919 goto done; 920 } 921 922 /* 923 * If the map has changed, due to the subyte, the previous 924 * output may be invalid. 925 */ 926 vm_map_lock_read(map); 927 if (timestamp != map->timestamp) 928 goto RestartScan; 929 930 lastvecindex = vecindex; 931 addr += PAGE_SIZE; 932 } 933 } 934 935 /* 936 * subyte may page fault. In case it needs to modify 937 * the map, we release the lock. 938 */ 939 vm_map_unlock_read(map); 940 941 /* 942 * Zero the last entries in the byte vector. 943 */ 944 vecindex = OFF_TO_IDX(end - first_addr); 945 while((lastvecindex + 1) < vecindex) { 946 error = subyte( vec + lastvecindex, 0); 947 if (error) { 948 error = EFAULT; 949 goto done; 950 } 951 ++lastvecindex; 952 } 953 954 /* 955 * If the map has changed, due to the subyte, the previous 956 * output may be invalid. 957 */ 958 vm_map_lock_read(map); 959 if (timestamp != map->timestamp) 960 goto RestartScan; 961 vm_map_unlock_read(map); 962 963 error = 0; 964 done: 965 rel_mplock(); 966 return (error); 967 } 968 969 /* 970 * mlock_args(const void *addr, size_t len) 971 * 972 * MPALMOSTSAFE 973 */ 974 int 975 sys_mlock(struct mlock_args *uap) 976 { 977 vm_offset_t addr; 978 vm_offset_t tmpaddr; 979 vm_size_t size, pageoff; 980 struct thread *td = curthread; 981 struct proc *p = td->td_proc; 982 int error; 983 984 addr = (vm_offset_t) uap->addr; 985 size = uap->len; 986 987 pageoff = (addr & PAGE_MASK); 988 addr -= pageoff; 989 size += pageoff; 990 size = (vm_size_t) round_page(size); 991 if (size < uap->len) /* wrap */ 992 return(EINVAL); 993 tmpaddr = addr + size; /* workaround gcc4 opt */ 994 if (tmpaddr < addr) /* wrap */ 995 return (EINVAL); 996 997 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) 998 return (EAGAIN); 999 1000 get_mplock(); 1001 #ifdef pmap_wired_count 1002 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 1003 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) { 1004 rel_mplock(); 1005 return (ENOMEM); 1006 } 1007 #else 1008 error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0); 1009 if (error) { 1010 rel_mplock(); 1011 return (error); 1012 } 1013 #endif 1014 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE); 1015 rel_mplock(); 1016 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1017 } 1018 1019 /* 1020 * mlockall_args(int how) 1021 * 1022 * Dummy routine, doesn't actually do anything. 1023 * 1024 * MPSAFE 1025 */ 1026 int 1027 sys_mlockall(struct mlockall_args *uap) 1028 { 1029 return 0; 1030 } 1031 1032 /* 1033 * munlockall_args(void) 1034 * 1035 * Dummy routine, doesn't actually do anything. 1036 * 1037 * MPSAFE 1038 */ 1039 int 1040 sys_munlockall(struct munlockall_args *uap) 1041 { 1042 return 0; 1043 } 1044 1045 /* 1046 * munlock_args(const void *addr, size_t len) 1047 * 1048 * MPALMOSTSAFE 1049 */ 1050 int 1051 sys_munlock(struct munlock_args *uap) 1052 { 1053 struct thread *td = curthread; 1054 struct proc *p = td->td_proc; 1055 vm_offset_t addr; 1056 vm_offset_t tmpaddr; 1057 vm_size_t size, pageoff; 1058 int error; 1059 1060 addr = (vm_offset_t) uap->addr; 1061 size = uap->len; 1062 1063 pageoff = (addr & PAGE_MASK); 1064 addr -= pageoff; 1065 size += pageoff; 1066 size = (vm_size_t) round_page(size); 1067 1068 tmpaddr = addr + size; 1069 if (tmpaddr < addr) /* wrap */ 1070 return (EINVAL); 1071 1072 #ifndef pmap_wired_count 1073 error = priv_check(td, PRIV_ROOT); 1074 if (error) 1075 return (error); 1076 #endif 1077 1078 get_mplock(); 1079 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE); 1080 rel_mplock(); 1081 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1082 } 1083 1084 /* 1085 * Internal version of mmap. 1086 * Currently used by mmap, exec, and sys5 shared memory. 1087 * Handle is either a vnode pointer or NULL for MAP_ANON. 1088 */ 1089 int 1090 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1091 vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff) 1092 { 1093 boolean_t fitit; 1094 vm_object_t object; 1095 vm_offset_t eaddr; 1096 vm_size_t esize; 1097 struct vnode *vp; 1098 struct thread *td = curthread; 1099 struct proc *p; 1100 int rv = KERN_SUCCESS; 1101 off_t objsize; 1102 int docow; 1103 1104 if (size == 0) 1105 return (0); 1106 1107 objsize = round_page(size); 1108 if (objsize < size) 1109 return (EINVAL); 1110 size = objsize; 1111 1112 /* 1113 * XXX messy code, fixme 1114 * 1115 * NOTE: Overflow checks require discrete statements or GCC4 1116 * will optimize it out. 1117 */ 1118 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) { 1119 esize = map->size + size; /* workaround gcc4 opt */ 1120 if (esize < map->size || 1121 esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) { 1122 return(ENOMEM); 1123 } 1124 } 1125 1126 /* 1127 * We currently can only deal with page aligned file offsets. 1128 * The check is here rather than in the syscall because the 1129 * kernel calls this function internally for other mmaping 1130 * operations (such as in exec) and non-aligned offsets will 1131 * cause pmap inconsistencies...so we want to be sure to 1132 * disallow this in all cases. 1133 * 1134 * NOTE: Overflow checks require discrete statements or GCC4 1135 * will optimize it out. 1136 */ 1137 if (foff & PAGE_MASK) 1138 return (EINVAL); 1139 1140 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) { 1141 fitit = TRUE; 1142 *addr = round_page(*addr); 1143 } else { 1144 if (*addr != trunc_page(*addr)) 1145 return (EINVAL); 1146 eaddr = *addr + size; 1147 if (eaddr < *addr) 1148 return (EINVAL); 1149 fitit = FALSE; 1150 if ((flags & MAP_TRYFIXED) == 0) 1151 vm_map_remove(map, *addr, *addr + size); 1152 } 1153 1154 /* 1155 * Lookup/allocate object. 1156 */ 1157 if (flags & MAP_ANON) { 1158 /* 1159 * Unnamed anonymous regions always start at 0. 1160 */ 1161 if (handle) { 1162 /* 1163 * Default memory object 1164 */ 1165 object = default_pager_alloc(handle, objsize, 1166 prot, foff); 1167 if (object == NULL) 1168 return(ENOMEM); 1169 docow = MAP_PREFAULT_PARTIAL; 1170 } else { 1171 /* 1172 * Implicit single instance of a default memory 1173 * object, so we don't need a VM object yet. 1174 */ 1175 foff = 0; 1176 object = NULL; 1177 docow = 0; 1178 } 1179 vp = NULL; 1180 } else { 1181 vp = (struct vnode *)handle; 1182 if (vp->v_type == VCHR) { 1183 /* 1184 * Device mappings (device size unknown?). 1185 * Force them to be shared. 1186 */ 1187 handle = (void *)(intptr_t)vp->v_rdev; 1188 object = dev_pager_alloc(handle, objsize, prot, foff); 1189 if (object == NULL) 1190 return(EINVAL); 1191 docow = MAP_PREFAULT_PARTIAL; 1192 flags &= ~(MAP_PRIVATE|MAP_COPY); 1193 flags |= MAP_SHARED; 1194 } else { 1195 /* 1196 * Regular file mapping (typically). The attribute 1197 * check is for the link count test only. Mmapble 1198 * vnodes must already have a VM object assigned. 1199 */ 1200 struct vattr vat; 1201 int error; 1202 1203 error = VOP_GETATTR(vp, &vat); 1204 if (error) 1205 return (error); 1206 docow = MAP_PREFAULT_PARTIAL; 1207 object = vnode_pager_reference(vp); 1208 if (object == NULL && vp->v_type == VREG) { 1209 kprintf("Warning: cannot mmap vnode %p, no " 1210 "object\n", vp); 1211 return(EINVAL); 1212 } 1213 1214 /* 1215 * If it is a regular file without any references 1216 * we do not need to sync it. 1217 */ 1218 if (vp->v_type == VREG && vat.va_nlink == 0) { 1219 flags |= MAP_NOSYNC; 1220 } 1221 } 1222 } 1223 1224 /* 1225 * Deal with the adjusted flags 1226 */ 1227 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1228 docow |= MAP_COPY_ON_WRITE; 1229 if (flags & MAP_NOSYNC) 1230 docow |= MAP_DISABLE_SYNCER; 1231 if (flags & MAP_NOCORE) 1232 docow |= MAP_DISABLE_COREDUMP; 1233 1234 #if defined(VM_PROT_READ_IS_EXEC) 1235 if (prot & VM_PROT_READ) 1236 prot |= VM_PROT_EXECUTE; 1237 1238 if (maxprot & VM_PROT_READ) 1239 maxprot |= VM_PROT_EXECUTE; 1240 #endif 1241 1242 /* 1243 * This may place the area in its own page directory if (size) is 1244 * large enough, otherwise it typically returns its argument. 1245 */ 1246 if (fitit) { 1247 *addr = pmap_addr_hint(object, *addr, size); 1248 } 1249 1250 /* 1251 * Stack mappings need special attention. 1252 * 1253 * Mappings that use virtual page tables will default to storing 1254 * the page table at offset 0. 1255 */ 1256 if (flags & MAP_STACK) { 1257 rv = vm_map_stack(map, *addr, size, flags, 1258 prot, maxprot, docow); 1259 } else if (flags & MAP_VPAGETABLE) { 1260 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE, 1261 fitit, VM_MAPTYPE_VPAGETABLE, 1262 prot, maxprot, docow); 1263 } else { 1264 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE, 1265 fitit, VM_MAPTYPE_NORMAL, 1266 prot, maxprot, docow); 1267 } 1268 1269 if (rv != KERN_SUCCESS) { 1270 /* 1271 * Lose the object reference. Will destroy the 1272 * object if it's an unnamed anonymous mapping 1273 * or named anonymous without other references. 1274 */ 1275 vm_object_deallocate(object); 1276 goto out; 1277 } 1278 1279 /* 1280 * Shared memory is also shared with children. 1281 */ 1282 if (flags & (MAP_SHARED|MAP_INHERIT)) { 1283 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1284 if (rv != KERN_SUCCESS) { 1285 vm_map_remove(map, *addr, *addr + size); 1286 goto out; 1287 } 1288 } 1289 1290 /* 1291 * Set the access time on the vnode 1292 */ 1293 if (vp != NULL) 1294 vn_mark_atime(vp, td); 1295 out: 1296 switch (rv) { 1297 case KERN_SUCCESS: 1298 return (0); 1299 case KERN_INVALID_ADDRESS: 1300 case KERN_NO_SPACE: 1301 return (ENOMEM); 1302 case KERN_PROTECTION_FAILURE: 1303 return (EACCES); 1304 default: 1305 return (EINVAL); 1306 } 1307 } 1308