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