1 /* 2 * (MPSAFE) 3 * 4 * Copyright (c) 1988 University of Utah. 5 * Copyright (c) 1991, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the Systems Programming Group of the University of Utah Computer 10 * Science Department. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ 37 * 38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 39 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $ 40 */ 41 42 /* 43 * Mapped file (mmap) interface to VM 44 */ 45 46 #include <sys/param.h> 47 #include <sys/kernel.h> 48 #include <sys/systm.h> 49 #include <sys/sysproto.h> 50 #include <sys/filedesc.h> 51 #include <sys/kern_syscall.h> 52 #include <sys/proc.h> 53 #include <sys/priv.h> 54 #include <sys/resource.h> 55 #include <sys/resourcevar.h> 56 #include <sys/vnode.h> 57 #include <sys/fcntl.h> 58 #include <sys/file.h> 59 #include <sys/mman.h> 60 #include <sys/conf.h> 61 #include <sys/stat.h> 62 #include <sys/vmmeter.h> 63 #include <sys/sysctl.h> 64 65 #include <vm/vm.h> 66 #include <vm/vm_param.h> 67 #include <sys/lock.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_map.h> 70 #include <vm/vm_object.h> 71 #include <vm/vm_page.h> 72 #include <vm/vm_pager.h> 73 #include <vm/vm_pageout.h> 74 #include <vm/vm_extern.h> 75 #include <vm/vm_page.h> 76 #include <vm/vm_kern.h> 77 78 #include <sys/file2.h> 79 #include <sys/thread.h> 80 #include <sys/thread2.h> 81 82 static int max_proc_mmap; 83 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, ""); 84 int vkernel_enable; 85 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, ""); 86 87 /* 88 * Set the maximum number of vm_map_entry structures per process. Roughly 89 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100 90 * of our KVM malloc space still results in generous limits. We want a 91 * default that is good enough to prevent the kernel running out of resources 92 * if attacked from compromised user account but generous enough such that 93 * multi-threaded processes are not unduly inconvenienced. 94 */ 95 96 static void vmmapentry_rsrc_init (void *); 97 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL) 98 99 static void 100 vmmapentry_rsrc_init(void *dummy) 101 { 102 max_proc_mmap = KvaSize / sizeof(struct vm_map_entry); 103 max_proc_mmap /= 100; 104 } 105 106 /* 107 * MPSAFE 108 */ 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 * MPSAFE 120 */ 121 int 122 sys_sstk(struct sstk_args *uap) 123 { 124 /* Not yet implemented */ 125 return (EOPNOTSUPP); 126 } 127 128 /* 129 * mmap_args(void *addr, size_t len, int prot, int flags, int fd, 130 * long pad, off_t pos) 131 * 132 * Memory Map (mmap) system call. Note that the file offset 133 * and address are allowed to be NOT page aligned, though if 134 * the MAP_FIXED flag it set, both must have the same remainder 135 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 136 * page-aligned, the actual mapping starts at trunc_page(addr) 137 * and the return value is adjusted up by the page offset. 138 * 139 * Generally speaking, only character devices which are themselves 140 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 141 * there would be no cache coherency between a descriptor and a VM mapping 142 * both to the same character device. 143 * 144 * Block devices can be mmap'd no matter what they represent. Cache coherency 145 * is maintained as long as you do not write directly to the underlying 146 * character device. 147 * 148 * No requirements 149 */ 150 int 151 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen, 152 int uprot, int uflags, int fd, off_t upos, void **res) 153 { 154 struct thread *td = curthread; 155 struct proc *p = td->td_proc; 156 struct file *fp = NULL; 157 struct vnode *vp; 158 vm_offset_t addr; 159 vm_offset_t tmpaddr; 160 vm_size_t size, pageoff; 161 vm_prot_t prot, maxprot; 162 void *handle; 163 int flags, error; 164 off_t pos; 165 vm_object_t obj; 166 167 KKASSERT(p); 168 169 addr = (vm_offset_t) uaddr; 170 size = ulen; 171 prot = uprot & VM_PROT_ALL; 172 flags = uflags; 173 pos = upos; 174 175 /* 176 * Make sure mapping fits into numeric range etc. 177 * 178 * NOTE: We support the full unsigned range for size now. 179 */ 180 if (((flags & MAP_ANON) && (fd != -1 || pos != 0))) 181 return (EINVAL); 182 183 if (size == 0) 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 * Get a hint of where to map. It also provides mmap offset 251 * randomization if enabled. 252 */ 253 addr = vm_map_hint(p, addr, prot); 254 } 255 256 if (flags & MAP_ANON) { 257 /* 258 * Mapping blank space is trivial. 259 */ 260 handle = NULL; 261 maxprot = VM_PROT_ALL; 262 } else { 263 /* 264 * Mapping file, get fp for validation. Obtain vnode and make 265 * sure it is of appropriate type. 266 */ 267 fp = holdfp(p->p_fd, fd, -1); 268 if (fp == NULL) 269 return (EBADF); 270 if (fp->f_type != DTYPE_VNODE) { 271 error = EINVAL; 272 goto done; 273 } 274 /* 275 * POSIX shared-memory objects are defined to have 276 * kernel persistence, and are not defined to support 277 * read(2)/write(2) -- or even open(2). Thus, we can 278 * use MAP_ASYNC to trade on-disk coherence for speed. 279 * The shm_open(3) library routine turns on the FPOSIXSHM 280 * flag to request this behavior. 281 */ 282 if (fp->f_flag & FPOSIXSHM) 283 flags |= MAP_NOSYNC; 284 vp = (struct vnode *) fp->f_data; 285 286 /* 287 * Validate the vnode for the operation. 288 */ 289 switch(vp->v_type) { 290 case VREG: 291 /* 292 * Get the proper underlying object 293 */ 294 if ((obj = vp->v_object) == NULL) { 295 error = EINVAL; 296 goto done; 297 } 298 KKASSERT((struct vnode *)obj->handle == vp); 299 break; 300 case VCHR: 301 /* 302 * Make sure a device has not been revoked. 303 * Mappability is handled by the device layer. 304 */ 305 if (vp->v_rdev == NULL) { 306 error = EBADF; 307 goto done; 308 } 309 break; 310 default: 311 /* 312 * Nothing else is mappable. 313 */ 314 error = EINVAL; 315 goto done; 316 } 317 318 /* 319 * XXX hack to handle use of /dev/zero to map anon memory (ala 320 * SunOS). 321 */ 322 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) { 323 handle = NULL; 324 maxprot = VM_PROT_ALL; 325 flags |= MAP_ANON; 326 pos = 0; 327 } else { 328 /* 329 * cdevs does not provide private mappings of any kind. 330 */ 331 if (vp->v_type == VCHR && 332 (flags & (MAP_PRIVATE|MAP_COPY))) { 333 error = EINVAL; 334 goto done; 335 } 336 /* 337 * Ensure that file and memory protections are 338 * compatible. Note that we only worry about 339 * writability if mapping is shared; in this case, 340 * current and max prot are dictated by the open file. 341 * XXX use the vnode instead? Problem is: what 342 * credentials do we use for determination? What if 343 * proc does a setuid? 344 */ 345 maxprot = VM_PROT_EXECUTE; /* ??? */ 346 if (fp->f_flag & FREAD) { 347 maxprot |= VM_PROT_READ; 348 } else if (prot & PROT_READ) { 349 error = EACCES; 350 goto done; 351 } 352 /* 353 * If we are sharing potential changes (either via 354 * MAP_SHARED or via the implicit sharing of character 355 * device mappings), and we are trying to get write 356 * permission although we opened it without asking 357 * for it, bail out. Check for superuser, only if 358 * we're at securelevel < 1, to allow the XIG X server 359 * to continue to work. 360 */ 361 if ((flags & MAP_SHARED) != 0 || vp->v_type == VCHR) { 362 if ((fp->f_flag & FWRITE) != 0) { 363 struct vattr va; 364 if ((error = VOP_GETATTR(vp, &va))) { 365 goto done; 366 } 367 if ((va.va_flags & 368 (IMMUTABLE|APPEND)) == 0) { 369 maxprot |= VM_PROT_WRITE; 370 } else if (prot & PROT_WRITE) { 371 error = EPERM; 372 goto done; 373 } 374 } else if ((prot & PROT_WRITE) != 0) { 375 error = EACCES; 376 goto done; 377 } 378 } else { 379 maxprot |= VM_PROT_WRITE; 380 } 381 handle = (void *)vp; 382 } 383 } 384 385 lwkt_gettoken(&vms->vm_map.token); 386 387 /* 388 * Do not allow more then a certain number of vm_map_entry structures 389 * per process. Scale with the number of rforks sharing the map 390 * to make the limit reasonable for threads. 391 */ 392 if (max_proc_mmap && 393 vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) { 394 error = ENOMEM; 395 lwkt_reltoken(&vms->vm_map.token); 396 goto done; 397 } 398 399 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot, 400 flags, handle, pos); 401 if (error == 0) 402 *res = (void *)(addr + pageoff); 403 404 lwkt_reltoken(&vms->vm_map.token); 405 done: 406 if (fp) 407 fdrop(fp); 408 409 return (error); 410 } 411 412 /* 413 * mmap system call handler 414 * 415 * No requirements. 416 */ 417 int 418 sys_mmap(struct mmap_args *uap) 419 { 420 int error; 421 422 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len, 423 uap->prot, uap->flags, 424 uap->fd, uap->pos, &uap->sysmsg_resultp); 425 426 return (error); 427 } 428 429 /* 430 * msync system call handler 431 * 432 * msync_args(void *addr, size_t len, int flags) 433 * 434 * No requirements 435 */ 436 int 437 sys_msync(struct msync_args *uap) 438 { 439 struct proc *p = curproc; 440 vm_offset_t addr; 441 vm_offset_t tmpaddr; 442 vm_size_t size, pageoff; 443 int flags; 444 vm_map_t map; 445 int rv; 446 447 addr = (vm_offset_t) uap->addr; 448 size = uap->len; 449 flags = uap->flags; 450 451 pageoff = (addr & PAGE_MASK); 452 addr -= pageoff; 453 size += pageoff; 454 size = (vm_size_t) round_page(size); 455 if (size < uap->len) /* wrap */ 456 return(EINVAL); 457 tmpaddr = addr + size; /* workaround gcc4 opt */ 458 if (tmpaddr < addr) /* wrap */ 459 return(EINVAL); 460 461 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 462 return (EINVAL); 463 464 map = &p->p_vmspace->vm_map; 465 466 /* 467 * map->token serializes extracting the address range for size == 0 468 * msyncs with the vm_map_clean call; if the token were not held 469 * across the two calls, an intervening munmap/mmap pair, for example, 470 * could cause msync to occur on a wrong region. 471 */ 472 lwkt_gettoken(&map->token); 473 474 /* 475 * XXX Gak! If size is zero we are supposed to sync "all modified 476 * pages with the region containing addr". Unfortunately, we don't 477 * really keep track of individual mmaps so we approximate by flushing 478 * the range of the map entry containing addr. This can be incorrect 479 * if the region splits or is coalesced with a neighbor. 480 */ 481 if (size == 0) { 482 vm_map_entry_t entry; 483 484 vm_map_lock_read(map); 485 rv = vm_map_lookup_entry(map, addr, &entry); 486 if (rv == FALSE) { 487 vm_map_unlock_read(map); 488 rv = KERN_INVALID_ADDRESS; 489 goto done; 490 } 491 addr = entry->start; 492 size = entry->end - entry->start; 493 vm_map_unlock_read(map); 494 } 495 496 /* 497 * Clean the pages and interpret the return value. 498 */ 499 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0, 500 (flags & MS_INVALIDATE) != 0); 501 done: 502 lwkt_reltoken(&map->token); 503 504 switch (rv) { 505 case KERN_SUCCESS: 506 break; 507 case KERN_INVALID_ADDRESS: 508 return (EINVAL); /* Sun returns ENOMEM? */ 509 case KERN_FAILURE: 510 return (EIO); 511 default: 512 return (EINVAL); 513 } 514 515 return (0); 516 } 517 518 /* 519 * munmap system call handler 520 * 521 * munmap_args(void *addr, size_t len) 522 * 523 * No requirements 524 */ 525 int 526 sys_munmap(struct munmap_args *uap) 527 { 528 struct proc *p = curproc; 529 vm_offset_t addr; 530 vm_offset_t tmpaddr; 531 vm_size_t size, pageoff; 532 vm_map_t map; 533 534 addr = (vm_offset_t) uap->addr; 535 size = uap->len; 536 537 pageoff = (addr & PAGE_MASK); 538 addr -= pageoff; 539 size += pageoff; 540 size = (vm_size_t) round_page(size); 541 if (size < uap->len) /* wrap */ 542 return(EINVAL); 543 tmpaddr = addr + size; /* workaround gcc4 opt */ 544 if (tmpaddr < addr) /* wrap */ 545 return(EINVAL); 546 547 if (size == 0) 548 return (0); 549 550 /* 551 * Check for illegal addresses. Watch out for address wrap... Note 552 * that VM_*_ADDRESS are not constants due to casts (argh). 553 */ 554 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS) 555 return (EINVAL); 556 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS) 557 return (EINVAL); 558 559 map = &p->p_vmspace->vm_map; 560 561 /* map->token serializes between the map check and the actual unmap */ 562 lwkt_gettoken(&map->token); 563 564 /* 565 * Make sure entire range is allocated. 566 */ 567 if (!vm_map_check_protection(map, addr, addr + size, 568 VM_PROT_NONE, FALSE)) { 569 lwkt_reltoken(&map->token); 570 return (EINVAL); 571 } 572 /* returns nothing but KERN_SUCCESS anyway */ 573 vm_map_remove(map, addr, addr + size); 574 lwkt_reltoken(&map->token); 575 return (0); 576 } 577 578 /* 579 * mprotect_args(const void *addr, size_t len, int prot) 580 * 581 * No requirements. 582 */ 583 int 584 sys_mprotect(struct mprotect_args *uap) 585 { 586 struct proc *p = curproc; 587 vm_offset_t addr; 588 vm_offset_t tmpaddr; 589 vm_size_t size, pageoff; 590 vm_prot_t prot; 591 int error; 592 593 addr = (vm_offset_t) uap->addr; 594 size = uap->len; 595 prot = uap->prot & VM_PROT_ALL; 596 #if defined(VM_PROT_READ_IS_EXEC) 597 if (prot & VM_PROT_READ) 598 prot |= VM_PROT_EXECUTE; 599 #endif 600 601 pageoff = (addr & PAGE_MASK); 602 addr -= pageoff; 603 size += pageoff; 604 size = (vm_size_t) round_page(size); 605 if (size < uap->len) /* wrap */ 606 return(EINVAL); 607 tmpaddr = addr + size; /* workaround gcc4 opt */ 608 if (tmpaddr < addr) /* wrap */ 609 return(EINVAL); 610 611 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 return (error); 624 } 625 626 /* 627 * minherit system call handler 628 * 629 * minherit_args(void *addr, size_t len, int inherit) 630 * 631 * No requirements. 632 */ 633 int 634 sys_minherit(struct minherit_args *uap) 635 { 636 struct proc *p = curproc; 637 vm_offset_t addr; 638 vm_offset_t tmpaddr; 639 vm_size_t size, pageoff; 640 vm_inherit_t inherit; 641 int error; 642 643 addr = (vm_offset_t)uap->addr; 644 size = uap->len; 645 inherit = uap->inherit; 646 647 pageoff = (addr & PAGE_MASK); 648 addr -= pageoff; 649 size += pageoff; 650 size = (vm_size_t) round_page(size); 651 if (size < uap->len) /* wrap */ 652 return(EINVAL); 653 tmpaddr = addr + size; /* workaround gcc4 opt */ 654 if (tmpaddr < addr) /* wrap */ 655 return(EINVAL); 656 657 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, 658 addr + size, inherit)) { 659 case KERN_SUCCESS: 660 error = 0; 661 break; 662 case KERN_PROTECTION_FAILURE: 663 error = EACCES; 664 break; 665 default: 666 error = EINVAL; 667 break; 668 } 669 return (error); 670 } 671 672 /* 673 * madvise system call handler 674 * 675 * madvise_args(void *addr, size_t len, int behav) 676 * 677 * No requirements. 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 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end, 711 uap->behav, 0); 712 return (error); 713 } 714 715 /* 716 * mcontrol system call handler 717 * 718 * mcontrol_args(void *addr, size_t len, int behav, off_t value) 719 * 720 * No requirements 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 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end, 754 uap->behav, uap->value); 755 return (error); 756 } 757 758 759 /* 760 * mincore system call handler 761 * 762 * mincore_args(const void *addr, size_t len, char *vec) 763 * 764 * No requirements 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 lwkt_gettoken(&map->token); 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 * XXX vm_token - legacy for pmap_ts_referenced 873 * in i386 and vkernel pmap code. 874 */ 875 lwkt_gettoken(&vm_token); 876 vm_object_hold(current->object.vm_object); 877 m = vm_page_lookup(current->object.vm_object, 878 pindex); 879 if (m && m->valid) { 880 mincoreinfo = MINCORE_INCORE; 881 if (m->dirty || 882 pmap_is_modified(m)) 883 mincoreinfo |= MINCORE_MODIFIED_OTHER; 884 if ((m->flags & PG_REFERENCED) || 885 pmap_ts_referenced(m)) { 886 vm_page_flag_set(m, PG_REFERENCED); 887 mincoreinfo |= MINCORE_REFERENCED_OTHER; 888 } 889 } 890 vm_object_drop(current->object.vm_object); 891 lwkt_reltoken(&vm_token); 892 } 893 894 /* 895 * subyte may page fault. In case it needs to modify 896 * the map, we release the lock. 897 */ 898 vm_map_unlock_read(map); 899 900 /* 901 * calculate index into user supplied byte vector 902 */ 903 vecindex = OFF_TO_IDX(addr - first_addr); 904 905 /* 906 * If we have skipped map entries, we need to make sure that 907 * the byte vector is zeroed for those skipped entries. 908 */ 909 while((lastvecindex + 1) < vecindex) { 910 error = subyte( vec + lastvecindex, 0); 911 if (error) { 912 error = EFAULT; 913 goto done; 914 } 915 ++lastvecindex; 916 } 917 918 /* 919 * Pass the page information to the user 920 */ 921 error = subyte( vec + vecindex, mincoreinfo); 922 if (error) { 923 error = EFAULT; 924 goto done; 925 } 926 927 /* 928 * If the map has changed, due to the subyte, the previous 929 * output may be invalid. 930 */ 931 vm_map_lock_read(map); 932 if (timestamp != map->timestamp) 933 goto RestartScan; 934 935 lastvecindex = vecindex; 936 addr += PAGE_SIZE; 937 } 938 } 939 940 /* 941 * subyte may page fault. In case it needs to modify 942 * the map, we release the lock. 943 */ 944 vm_map_unlock_read(map); 945 946 /* 947 * Zero the last entries in the byte vector. 948 */ 949 vecindex = OFF_TO_IDX(end - first_addr); 950 while((lastvecindex + 1) < vecindex) { 951 error = subyte( vec + lastvecindex, 0); 952 if (error) { 953 error = EFAULT; 954 goto done; 955 } 956 ++lastvecindex; 957 } 958 959 /* 960 * If the map has changed, due to the subyte, the previous 961 * output may be invalid. 962 */ 963 vm_map_lock_read(map); 964 if (timestamp != map->timestamp) 965 goto RestartScan; 966 vm_map_unlock_read(map); 967 968 error = 0; 969 done: 970 lwkt_reltoken(&map->token); 971 return (error); 972 } 973 974 /* 975 * mlock system call handler 976 * 977 * mlock_args(const void *addr, size_t len) 978 * 979 * No requirements 980 */ 981 int 982 sys_mlock(struct mlock_args *uap) 983 { 984 vm_offset_t addr; 985 vm_offset_t tmpaddr; 986 vm_size_t size, pageoff; 987 struct thread *td = curthread; 988 struct proc *p = td->td_proc; 989 int error; 990 991 addr = (vm_offset_t) uap->addr; 992 size = uap->len; 993 994 pageoff = (addr & PAGE_MASK); 995 addr -= pageoff; 996 size += pageoff; 997 size = (vm_size_t) round_page(size); 998 if (size < uap->len) /* wrap */ 999 return(EINVAL); 1000 tmpaddr = addr + size; /* workaround gcc4 opt */ 1001 if (tmpaddr < addr) /* wrap */ 1002 return (EINVAL); 1003 1004 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired) 1005 return (EAGAIN); 1006 1007 /* 1008 * We do not need to synchronize against other threads updating ucred; 1009 * they update p->ucred, which is synchronized into td_ucred ourselves. 1010 */ 1011 #ifdef pmap_wired_count 1012 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) > 1013 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) { 1014 return (ENOMEM); 1015 } 1016 #else 1017 error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0); 1018 if (error) { 1019 return (error); 1020 } 1021 #endif 1022 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE); 1023 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1024 } 1025 1026 /* 1027 * mlockall(int how) 1028 * 1029 * No requirements 1030 */ 1031 int 1032 sys_mlockall(struct mlockall_args *uap) 1033 { 1034 #ifdef _P1003_1B_VISIBLE 1035 struct thread *td = curthread; 1036 struct proc *p = td->td_proc; 1037 vm_map_t map = &p->p_vmspace->vm_map; 1038 vm_map_entry_t entry; 1039 int how = uap->how; 1040 int rc = KERN_SUCCESS; 1041 1042 if (((how & MCL_CURRENT) == 0) && ((how & MCL_FUTURE) == 0)) 1043 return (EINVAL); 1044 1045 rc = priv_check_cred(td->td_ucred, PRIV_ROOT, 0); 1046 if (rc) 1047 return (rc); 1048 1049 vm_map_lock(map); 1050 do { 1051 if (how & MCL_CURRENT) { 1052 for(entry = map->header.next; 1053 entry != &map->header; 1054 entry = entry->next); 1055 1056 rc = ENOSYS; 1057 break; 1058 } 1059 1060 if (how & MCL_FUTURE) 1061 map->flags |= MAP_WIREFUTURE; 1062 } while(0); 1063 vm_map_unlock(map); 1064 1065 return (rc); 1066 #else /* !_P1003_1B_VISIBLE */ 1067 return (ENOSYS); 1068 #endif /* _P1003_1B_VISIBLE */ 1069 } 1070 1071 /* 1072 * munlockall(void) 1073 * 1074 * Unwire all user-wired map entries, cancel MCL_FUTURE. 1075 * 1076 * No requirements 1077 */ 1078 int 1079 sys_munlockall(struct munlockall_args *uap) 1080 { 1081 struct thread *td = curthread; 1082 struct proc *p = td->td_proc; 1083 vm_map_t map = &p->p_vmspace->vm_map; 1084 vm_map_entry_t entry; 1085 int rc = KERN_SUCCESS; 1086 1087 vm_map_lock(map); 1088 1089 /* Clear MAP_WIREFUTURE to cancel mlockall(MCL_FUTURE) */ 1090 map->flags &= ~MAP_WIREFUTURE; 1091 1092 retry: 1093 for (entry = map->header.next; 1094 entry != &map->header; 1095 entry = entry->next) { 1096 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) 1097 continue; 1098 1099 /* 1100 * If we encounter an in-transition entry, we release the 1101 * map lock and retry the scan; we do not decrement any 1102 * wired_count more than once because we do not touch 1103 * any entries with MAP_ENTRY_USER_WIRED not set. 1104 * 1105 * There is a potential interleaving with concurrent 1106 * mlockall()s here -- if we abort a scan, an mlockall() 1107 * could start, wire a number of entries before our 1108 * current position in, and then stall itself on this 1109 * or any other in-transition entry. If that occurs, when 1110 * we resume, we will unwire those entries. 1111 */ 1112 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) { 1113 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP; 1114 ++mycpu->gd_cnt.v_intrans_coll; 1115 ++mycpu->gd_cnt.v_intrans_wait; 1116 vm_map_transition_wait(map); 1117 goto retry; 1118 } 1119 1120 KASSERT(entry->wired_count > 0, 1121 ("wired_count was 0 with USER_WIRED set! %p", entry)); 1122 1123 /* Drop wired count, if it hits zero, unwire the entry */ 1124 entry->eflags &= ~MAP_ENTRY_USER_WIRED; 1125 entry->wired_count--; 1126 if (entry->wired_count == 0) 1127 vm_fault_unwire(map, entry); 1128 } 1129 1130 map->timestamp++; 1131 vm_map_unlock(map); 1132 1133 return (rc); 1134 } 1135 1136 /* 1137 * munlock system call handler 1138 * 1139 * munlock_args(const void *addr, size_t len) 1140 * 1141 * No requirements 1142 */ 1143 int 1144 sys_munlock(struct munlock_args *uap) 1145 { 1146 struct thread *td = curthread; 1147 struct proc *p = td->td_proc; 1148 vm_offset_t addr; 1149 vm_offset_t tmpaddr; 1150 vm_size_t size, pageoff; 1151 int error; 1152 1153 addr = (vm_offset_t) uap->addr; 1154 size = uap->len; 1155 1156 pageoff = (addr & PAGE_MASK); 1157 addr -= pageoff; 1158 size += pageoff; 1159 size = (vm_size_t) round_page(size); 1160 1161 tmpaddr = addr + size; 1162 if (tmpaddr < addr) /* wrap */ 1163 return (EINVAL); 1164 1165 #ifndef pmap_wired_count 1166 error = priv_check(td, PRIV_ROOT); 1167 if (error) 1168 return (error); 1169 #endif 1170 1171 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE); 1172 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1173 } 1174 1175 /* 1176 * Internal version of mmap. 1177 * Currently used by mmap, exec, and sys5 shared memory. 1178 * Handle is either a vnode pointer or NULL for MAP_ANON. 1179 * 1180 * No requirements 1181 */ 1182 int 1183 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1184 vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff) 1185 { 1186 boolean_t fitit; 1187 vm_object_t object; 1188 vm_offset_t eaddr; 1189 vm_size_t esize; 1190 vm_size_t align; 1191 struct vnode *vp; 1192 struct thread *td = curthread; 1193 struct proc *p; 1194 int rv = KERN_SUCCESS; 1195 off_t objsize; 1196 int docow; 1197 1198 if (size == 0) 1199 return (0); 1200 1201 objsize = round_page(size); 1202 if (objsize < size) 1203 return (EINVAL); 1204 size = objsize; 1205 1206 lwkt_gettoken(&map->token); 1207 1208 /* 1209 * XXX messy code, fixme 1210 * 1211 * NOTE: Overflow checks require discrete statements or GCC4 1212 * will optimize it out. 1213 */ 1214 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) { 1215 esize = map->size + size; /* workaround gcc4 opt */ 1216 if (esize < map->size || 1217 esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) { 1218 lwkt_reltoken(&map->token); 1219 return(ENOMEM); 1220 } 1221 } 1222 1223 /* 1224 * We currently can only deal with page aligned file offsets. 1225 * The check is here rather than in the syscall because the 1226 * kernel calls this function internally for other mmaping 1227 * operations (such as in exec) and non-aligned offsets will 1228 * cause pmap inconsistencies...so we want to be sure to 1229 * disallow this in all cases. 1230 * 1231 * NOTE: Overflow checks require discrete statements or GCC4 1232 * will optimize it out. 1233 */ 1234 if (foff & PAGE_MASK) { 1235 lwkt_reltoken(&map->token); 1236 return (EINVAL); 1237 } 1238 1239 /* 1240 * Handle alignment. For large memory maps it is possible 1241 * that the MMU can optimize the page table so align anything 1242 * that is a multiple of SEG_SIZE to SEG_SIZE. 1243 * 1244 * Also align any large mapping (bigger than 16x SG_SIZE) to a 1245 * SEG_SIZE address boundary. 1246 */ 1247 if (flags & MAP_SIZEALIGN) { 1248 align = size; 1249 if ((align ^ (align - 1)) != (align << 1) - 1) { 1250 lwkt_reltoken(&map->token); 1251 return (EINVAL); 1252 } 1253 } else if ((flags & MAP_FIXED) == 0 && 1254 ((size & SEG_MASK) == 0 || size > SEG_SIZE * 16)) { 1255 align = SEG_SIZE; 1256 } else { 1257 align = PAGE_SIZE; 1258 } 1259 1260 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) { 1261 fitit = TRUE; 1262 *addr = round_page(*addr); 1263 } else { 1264 if (*addr != trunc_page(*addr)) { 1265 lwkt_reltoken(&map->token); 1266 return (EINVAL); 1267 } 1268 eaddr = *addr + size; 1269 if (eaddr < *addr) { 1270 lwkt_reltoken(&map->token); 1271 return (EINVAL); 1272 } 1273 fitit = FALSE; 1274 if ((flags & MAP_TRYFIXED) == 0) 1275 vm_map_remove(map, *addr, *addr + size); 1276 } 1277 1278 /* 1279 * Lookup/allocate object. 1280 */ 1281 if (flags & MAP_ANON) { 1282 /* 1283 * Unnamed anonymous regions always start at 0. 1284 */ 1285 if (handle) { 1286 /* 1287 * Default memory object 1288 */ 1289 object = default_pager_alloc(handle, objsize, 1290 prot, foff); 1291 if (object == NULL) { 1292 lwkt_reltoken(&map->token); 1293 return(ENOMEM); 1294 } 1295 docow = MAP_PREFAULT_PARTIAL; 1296 } else { 1297 /* 1298 * Implicit single instance of a default memory 1299 * object, so we don't need a VM object yet. 1300 */ 1301 foff = 0; 1302 object = NULL; 1303 docow = 0; 1304 } 1305 vp = NULL; 1306 } else { 1307 vp = (struct vnode *)handle; 1308 if (vp->v_type == VCHR) { 1309 /* 1310 * Device mappings (device size unknown?). 1311 * Force them to be shared. 1312 */ 1313 handle = (void *)(intptr_t)vp->v_rdev; 1314 object = dev_pager_alloc(handle, objsize, prot, foff); 1315 if (object == NULL) { 1316 lwkt_reltoken(&map->token); 1317 return(EINVAL); 1318 } 1319 docow = MAP_PREFAULT_PARTIAL; 1320 flags &= ~(MAP_PRIVATE|MAP_COPY); 1321 flags |= MAP_SHARED; 1322 } else { 1323 /* 1324 * Regular file mapping (typically). The attribute 1325 * check is for the link count test only. Mmapble 1326 * vnodes must already have a VM object assigned. 1327 */ 1328 struct vattr vat; 1329 int error; 1330 1331 error = VOP_GETATTR(vp, &vat); 1332 if (error) { 1333 lwkt_reltoken(&map->token); 1334 return (error); 1335 } 1336 docow = MAP_PREFAULT_PARTIAL; 1337 object = vnode_pager_reference(vp); 1338 if (object == NULL && vp->v_type == VREG) { 1339 lwkt_reltoken(&map->token); 1340 kprintf("Warning: cannot mmap vnode %p, no " 1341 "object\n", vp); 1342 return(EINVAL); 1343 } 1344 1345 /* 1346 * If it is a regular file without any references 1347 * we do not need to sync it. 1348 */ 1349 if (vp->v_type == VREG && vat.va_nlink == 0) { 1350 flags |= MAP_NOSYNC; 1351 } 1352 } 1353 } 1354 1355 /* 1356 * Deal with the adjusted flags 1357 */ 1358 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1359 docow |= MAP_COPY_ON_WRITE; 1360 if (flags & MAP_NOSYNC) 1361 docow |= MAP_DISABLE_SYNCER; 1362 if (flags & MAP_NOCORE) 1363 docow |= MAP_DISABLE_COREDUMP; 1364 1365 #if defined(VM_PROT_READ_IS_EXEC) 1366 if (prot & VM_PROT_READ) 1367 prot |= VM_PROT_EXECUTE; 1368 1369 if (maxprot & VM_PROT_READ) 1370 maxprot |= VM_PROT_EXECUTE; 1371 #endif 1372 1373 /* 1374 * This may place the area in its own page directory if (size) is 1375 * large enough, otherwise it typically returns its argument. 1376 */ 1377 if (fitit) { 1378 *addr = pmap_addr_hint(object, *addr, size); 1379 } 1380 1381 /* 1382 * Stack mappings need special attention. 1383 * 1384 * Mappings that use virtual page tables will default to storing 1385 * the page table at offset 0. 1386 */ 1387 if (flags & MAP_STACK) { 1388 rv = vm_map_stack(map, *addr, size, flags, 1389 prot, maxprot, docow); 1390 } else if (flags & MAP_VPAGETABLE) { 1391 rv = vm_map_find(map, object, foff, addr, size, align, 1392 fitit, VM_MAPTYPE_VPAGETABLE, 1393 prot, maxprot, docow); 1394 } else { 1395 rv = vm_map_find(map, object, foff, addr, size, align, 1396 fitit, VM_MAPTYPE_NORMAL, 1397 prot, maxprot, docow); 1398 } 1399 1400 if (rv != KERN_SUCCESS) { 1401 /* 1402 * Lose the object reference. Will destroy the 1403 * object if it's an unnamed anonymous mapping 1404 * or named anonymous without other references. 1405 */ 1406 vm_object_deallocate(object); 1407 goto out; 1408 } 1409 1410 /* 1411 * Shared memory is also shared with children. 1412 */ 1413 if (flags & (MAP_SHARED|MAP_INHERIT)) { 1414 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE); 1415 if (rv != KERN_SUCCESS) { 1416 vm_map_remove(map, *addr, *addr + size); 1417 goto out; 1418 } 1419 } 1420 1421 /* If a process has marked all future mappings for wiring, do so */ 1422 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE)) 1423 vm_map_unwire(map, *addr, *addr + size, FALSE); 1424 1425 /* 1426 * Set the access time on the vnode 1427 */ 1428 if (vp != NULL) 1429 vn_mark_atime(vp, td); 1430 out: 1431 lwkt_reltoken(&map->token); 1432 1433 switch (rv) { 1434 case KERN_SUCCESS: 1435 return (0); 1436 case KERN_INVALID_ADDRESS: 1437 case KERN_NO_SPACE: 1438 return (ENOMEM); 1439 case KERN_PROTECTION_FAILURE: 1440 return (EACCES); 1441 default: 1442 return (EINVAL); 1443 } 1444 } 1445