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 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$ 35 * 36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94 37 */ 38 39 /* 40 * Mapped file (mmap) interface to VM 41 */ 42 43 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include "opt_compat.h" 47 #include "opt_hwpmc_hooks.h" 48 #include "opt_vm.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/capsicum.h> 53 #include <sys/kernel.h> 54 #include <sys/lock.h> 55 #include <sys/mutex.h> 56 #include <sys/sysproto.h> 57 #include <sys/filedesc.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/procctl.h> 61 #include <sys/racct.h> 62 #include <sys/resource.h> 63 #include <sys/resourcevar.h> 64 #include <sys/rwlock.h> 65 #include <sys/sysctl.h> 66 #include <sys/vnode.h> 67 #include <sys/fcntl.h> 68 #include <sys/file.h> 69 #include <sys/mman.h> 70 #include <sys/mount.h> 71 #include <sys/conf.h> 72 #include <sys/stat.h> 73 #include <sys/syscallsubr.h> 74 #include <sys/sysent.h> 75 #include <sys/vmmeter.h> 76 77 #include <security/mac/mac_framework.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_param.h> 81 #include <vm/pmap.h> 82 #include <vm/vm_map.h> 83 #include <vm/vm_object.h> 84 #include <vm/vm_page.h> 85 #include <vm/vm_pager.h> 86 #include <vm/vm_pageout.h> 87 #include <vm/vm_extern.h> 88 #include <vm/vm_page.h> 89 #include <vm/vnode_pager.h> 90 91 #ifdef HWPMC_HOOKS 92 #include <sys/pmckern.h> 93 #endif 94 95 int old_mlock = 0; 96 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, 97 "Do not apply RLIMIT_MEMLOCK on mlockall"); 98 99 #ifdef MAP_32BIT 100 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 101 #endif 102 103 #ifndef _SYS_SYSPROTO_H_ 104 struct sbrk_args { 105 int incr; 106 }; 107 #endif 108 109 /* 110 * MPSAFE 111 */ 112 /* ARGSUSED */ 113 int 114 sys_sbrk(td, uap) 115 struct thread *td; 116 struct sbrk_args *uap; 117 { 118 /* Not yet implemented */ 119 return (EOPNOTSUPP); 120 } 121 122 #ifndef _SYS_SYSPROTO_H_ 123 struct sstk_args { 124 int incr; 125 }; 126 #endif 127 128 /* 129 * MPSAFE 130 */ 131 /* ARGSUSED */ 132 int 133 sys_sstk(td, uap) 134 struct thread *td; 135 struct sstk_args *uap; 136 { 137 /* Not yet implemented */ 138 return (EOPNOTSUPP); 139 } 140 141 #if defined(COMPAT_43) 142 #ifndef _SYS_SYSPROTO_H_ 143 struct getpagesize_args { 144 int dummy; 145 }; 146 #endif 147 148 int 149 ogetpagesize(td, uap) 150 struct thread *td; 151 struct getpagesize_args *uap; 152 { 153 /* MP SAFE */ 154 td->td_retval[0] = PAGE_SIZE; 155 return (0); 156 } 157 #endif /* COMPAT_43 */ 158 159 160 /* 161 * Memory Map (mmap) system call. Note that the file offset 162 * and address are allowed to be NOT page aligned, though if 163 * the MAP_FIXED flag it set, both must have the same remainder 164 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 165 * page-aligned, the actual mapping starts at trunc_page(addr) 166 * and the return value is adjusted up by the page offset. 167 * 168 * Generally speaking, only character devices which are themselves 169 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 170 * there would be no cache coherency between a descriptor and a VM mapping 171 * both to the same character device. 172 */ 173 #ifndef _SYS_SYSPROTO_H_ 174 struct mmap_args { 175 void *addr; 176 size_t len; 177 int prot; 178 int flags; 179 int fd; 180 long pad; 181 off_t pos; 182 }; 183 #endif 184 185 /* 186 * MPSAFE 187 */ 188 int 189 sys_mmap(td, uap) 190 struct thread *td; 191 struct mmap_args *uap; 192 { 193 struct file *fp; 194 vm_offset_t addr; 195 vm_size_t size, pageoff; 196 vm_prot_t cap_maxprot; 197 int align, error, flags, prot; 198 off_t pos; 199 struct vmspace *vms = td->td_proc->p_vmspace; 200 cap_rights_t rights; 201 202 addr = (vm_offset_t) uap->addr; 203 size = uap->len; 204 prot = uap->prot; 205 flags = uap->flags; 206 pos = uap->pos; 207 208 fp = NULL; 209 210 /* 211 * Ignore old flags that used to be defined but did not do anything. 212 */ 213 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 214 215 /* 216 * Enforce the constraints. 217 * Mapping of length 0 is only allowed for old binaries. 218 * Anonymous mapping shall specify -1 as filedescriptor and 219 * zero position for new code. Be nice to ancient a.out 220 * binaries and correct pos for anonymous mapping, since old 221 * ld.so sometimes issues anonymous map requests with non-zero 222 * pos. 223 */ 224 if (!SV_CURPROC_FLAG(SV_AOUT)) { 225 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) || 226 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0))) 227 return (EINVAL); 228 } else { 229 if ((flags & MAP_ANON) != 0) 230 pos = 0; 231 } 232 233 if (flags & MAP_STACK) { 234 if ((uap->fd != -1) || 235 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 236 return (EINVAL); 237 flags |= MAP_ANON; 238 pos = 0; 239 } 240 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 241 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 242 MAP_PREFAULT_READ | 243 #ifdef MAP_32BIT 244 MAP_32BIT | 245 #endif 246 MAP_ALIGNMENT_MASK)) != 0) 247 return (EINVAL); 248 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 249 return (EINVAL); 250 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 251 return (EINVAL); 252 if (prot != PROT_NONE && 253 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 254 return (EINVAL); 255 256 /* 257 * Align the file position to a page boundary, 258 * and save its page offset component. 259 */ 260 pageoff = (pos & PAGE_MASK); 261 pos -= pageoff; 262 263 /* Adjust size for rounding (on both ends). */ 264 size += pageoff; /* low end... */ 265 size = (vm_size_t) round_page(size); /* hi end */ 266 267 /* Ensure alignment is at least a page and fits in a pointer. */ 268 align = flags & MAP_ALIGNMENT_MASK; 269 if (align != 0 && align != MAP_ALIGNED_SUPER && 270 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 271 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 272 return (EINVAL); 273 274 /* 275 * Check for illegal addresses. Watch out for address wrap... Note 276 * that VM_*_ADDRESS are not constants due to casts (argh). 277 */ 278 if (flags & MAP_FIXED) { 279 /* 280 * The specified address must have the same remainder 281 * as the file offset taken modulo PAGE_SIZE, so it 282 * should be aligned after adjustment by pageoff. 283 */ 284 addr -= pageoff; 285 if (addr & PAGE_MASK) 286 return (EINVAL); 287 288 /* Address range must be all in user VM space. */ 289 if (addr < vm_map_min(&vms->vm_map) || 290 addr + size > vm_map_max(&vms->vm_map)) 291 return (EINVAL); 292 if (addr + size < addr) 293 return (EINVAL); 294 #ifdef MAP_32BIT 295 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 296 return (EINVAL); 297 } else if (flags & MAP_32BIT) { 298 /* 299 * For MAP_32BIT, override the hint if it is too high and 300 * do not bother moving the mapping past the heap (since 301 * the heap is usually above 2GB). 302 */ 303 if (addr + size > MAP_32BIT_MAX_ADDR) 304 addr = 0; 305 #endif 306 } else { 307 /* 308 * XXX for non-fixed mappings where no hint is provided or 309 * the hint would fall in the potential heap space, 310 * place it after the end of the largest possible heap. 311 * 312 * There should really be a pmap call to determine a reasonable 313 * location. 314 */ 315 PROC_LOCK(td->td_proc); 316 if (addr == 0 || 317 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 318 addr < round_page((vm_offset_t)vms->vm_daddr + 319 lim_max_proc(td->td_proc, RLIMIT_DATA)))) 320 addr = round_page((vm_offset_t)vms->vm_daddr + 321 lim_max_proc(td->td_proc, RLIMIT_DATA)); 322 PROC_UNLOCK(td->td_proc); 323 } 324 if (size == 0) { 325 /* 326 * Return success without mapping anything for old 327 * binaries that request a page-aligned mapping of 328 * length 0. For modern binaries, this function 329 * returns an error earlier. 330 */ 331 error = 0; 332 } else if (flags & MAP_ANON) { 333 /* 334 * Mapping blank space is trivial. 335 * 336 * This relies on VM_PROT_* matching PROT_*. 337 */ 338 error = vm_mmap_object(&vms->vm_map, &addr, size, prot, 339 VM_PROT_ALL, flags, NULL, pos, FALSE, td); 340 } else { 341 /* 342 * Mapping file, get fp for validation and don't let the 343 * descriptor disappear on us if we block. Check capability 344 * rights, but also return the maximum rights to be combined 345 * with maxprot later. 346 */ 347 cap_rights_init(&rights, CAP_MMAP); 348 if (prot & PROT_READ) 349 cap_rights_set(&rights, CAP_MMAP_R); 350 if ((flags & MAP_SHARED) != 0) { 351 if (prot & PROT_WRITE) 352 cap_rights_set(&rights, CAP_MMAP_W); 353 } 354 if (prot & PROT_EXEC) 355 cap_rights_set(&rights, CAP_MMAP_X); 356 error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp); 357 if (error != 0) 358 goto done; 359 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && 360 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) { 361 error = EINVAL; 362 goto done; 363 } 364 365 /* This relies on VM_PROT_* matching PROT_*. */ 366 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, 367 cap_maxprot, flags, pos, td); 368 } 369 370 if (error == 0) 371 td->td_retval[0] = (register_t) (addr + pageoff); 372 done: 373 if (fp) 374 fdrop(fp, td); 375 376 return (error); 377 } 378 379 #if defined(COMPAT_FREEBSD6) 380 int 381 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) 382 { 383 struct mmap_args oargs; 384 385 oargs.addr = uap->addr; 386 oargs.len = uap->len; 387 oargs.prot = uap->prot; 388 oargs.flags = uap->flags; 389 oargs.fd = uap->fd; 390 oargs.pos = uap->pos; 391 return (sys_mmap(td, &oargs)); 392 } 393 #endif 394 395 #ifdef COMPAT_43 396 #ifndef _SYS_SYSPROTO_H_ 397 struct ommap_args { 398 caddr_t addr; 399 int len; 400 int prot; 401 int flags; 402 int fd; 403 long pos; 404 }; 405 #endif 406 int 407 ommap(td, uap) 408 struct thread *td; 409 struct ommap_args *uap; 410 { 411 struct mmap_args nargs; 412 static const char cvtbsdprot[8] = { 413 0, 414 PROT_EXEC, 415 PROT_WRITE, 416 PROT_EXEC | PROT_WRITE, 417 PROT_READ, 418 PROT_EXEC | PROT_READ, 419 PROT_WRITE | PROT_READ, 420 PROT_EXEC | PROT_WRITE | PROT_READ, 421 }; 422 423 #define OMAP_ANON 0x0002 424 #define OMAP_COPY 0x0020 425 #define OMAP_SHARED 0x0010 426 #define OMAP_FIXED 0x0100 427 428 nargs.addr = uap->addr; 429 nargs.len = uap->len; 430 nargs.prot = cvtbsdprot[uap->prot & 0x7]; 431 #ifdef COMPAT_FREEBSD32 432 #if defined(__amd64__) 433 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && 434 nargs.prot != 0) 435 nargs.prot |= PROT_EXEC; 436 #endif 437 #endif 438 nargs.flags = 0; 439 if (uap->flags & OMAP_ANON) 440 nargs.flags |= MAP_ANON; 441 if (uap->flags & OMAP_COPY) 442 nargs.flags |= MAP_COPY; 443 if (uap->flags & OMAP_SHARED) 444 nargs.flags |= MAP_SHARED; 445 else 446 nargs.flags |= MAP_PRIVATE; 447 if (uap->flags & OMAP_FIXED) 448 nargs.flags |= MAP_FIXED; 449 nargs.fd = uap->fd; 450 nargs.pos = uap->pos; 451 return (sys_mmap(td, &nargs)); 452 } 453 #endif /* COMPAT_43 */ 454 455 456 #ifndef _SYS_SYSPROTO_H_ 457 struct msync_args { 458 void *addr; 459 size_t len; 460 int flags; 461 }; 462 #endif 463 /* 464 * MPSAFE 465 */ 466 int 467 sys_msync(td, uap) 468 struct thread *td; 469 struct msync_args *uap; 470 { 471 vm_offset_t addr; 472 vm_size_t size, pageoff; 473 int flags; 474 vm_map_t map; 475 int rv; 476 477 addr = (vm_offset_t) uap->addr; 478 size = uap->len; 479 flags = uap->flags; 480 481 pageoff = (addr & PAGE_MASK); 482 addr -= pageoff; 483 size += pageoff; 484 size = (vm_size_t) round_page(size); 485 if (addr + size < addr) 486 return (EINVAL); 487 488 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 489 return (EINVAL); 490 491 map = &td->td_proc->p_vmspace->vm_map; 492 493 /* 494 * Clean the pages and interpret the return value. 495 */ 496 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, 497 (flags & MS_INVALIDATE) != 0); 498 switch (rv) { 499 case KERN_SUCCESS: 500 return (0); 501 case KERN_INVALID_ADDRESS: 502 return (ENOMEM); 503 case KERN_INVALID_ARGUMENT: 504 return (EBUSY); 505 case KERN_FAILURE: 506 return (EIO); 507 default: 508 return (EINVAL); 509 } 510 } 511 512 #ifndef _SYS_SYSPROTO_H_ 513 struct munmap_args { 514 void *addr; 515 size_t len; 516 }; 517 #endif 518 /* 519 * MPSAFE 520 */ 521 int 522 sys_munmap(td, uap) 523 struct thread *td; 524 struct munmap_args *uap; 525 { 526 #ifdef HWPMC_HOOKS 527 struct pmckern_map_out pkm; 528 vm_map_entry_t entry; 529 #endif 530 vm_offset_t addr; 531 vm_size_t size, pageoff; 532 vm_map_t map; 533 534 addr = (vm_offset_t) uap->addr; 535 size = uap->len; 536 if (size == 0) 537 return (EINVAL); 538 539 pageoff = (addr & PAGE_MASK); 540 addr -= pageoff; 541 size += pageoff; 542 size = (vm_size_t) round_page(size); 543 if (addr + size < addr) 544 return (EINVAL); 545 546 /* 547 * Check for illegal addresses. Watch out for address wrap... 548 */ 549 map = &td->td_proc->p_vmspace->vm_map; 550 if (addr < vm_map_min(map) || addr + size > vm_map_max(map)) 551 return (EINVAL); 552 vm_map_lock(map); 553 #ifdef HWPMC_HOOKS 554 /* 555 * Inform hwpmc if the address range being unmapped contains 556 * an executable region. 557 */ 558 pkm.pm_address = (uintptr_t) NULL; 559 if (vm_map_lookup_entry(map, addr, &entry)) { 560 for (; 561 entry != &map->header && entry->start < addr + size; 562 entry = entry->next) { 563 if (vm_map_check_protection(map, entry->start, 564 entry->end, VM_PROT_EXECUTE) == TRUE) { 565 pkm.pm_address = (uintptr_t) addr; 566 pkm.pm_size = (size_t) size; 567 break; 568 } 569 } 570 } 571 #endif 572 vm_map_delete(map, addr, addr + size); 573 574 #ifdef HWPMC_HOOKS 575 /* downgrade the lock to prevent a LOR with the pmc-sx lock */ 576 vm_map_lock_downgrade(map); 577 if (pkm.pm_address != (uintptr_t) NULL) 578 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); 579 vm_map_unlock_read(map); 580 #else 581 vm_map_unlock(map); 582 #endif 583 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */ 584 return (0); 585 } 586 587 #ifndef _SYS_SYSPROTO_H_ 588 struct mprotect_args { 589 const void *addr; 590 size_t len; 591 int prot; 592 }; 593 #endif 594 /* 595 * MPSAFE 596 */ 597 int 598 sys_mprotect(td, uap) 599 struct thread *td; 600 struct mprotect_args *uap; 601 { 602 vm_offset_t addr; 603 vm_size_t size, pageoff; 604 vm_prot_t prot; 605 606 addr = (vm_offset_t) uap->addr; 607 size = uap->len; 608 prot = uap->prot & VM_PROT_ALL; 609 610 pageoff = (addr & PAGE_MASK); 611 addr -= pageoff; 612 size += pageoff; 613 size = (vm_size_t) round_page(size); 614 if (addr + size < addr) 615 return (EINVAL); 616 617 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr, 618 addr + size, prot, FALSE)) { 619 case KERN_SUCCESS: 620 return (0); 621 case KERN_PROTECTION_FAILURE: 622 return (EACCES); 623 case KERN_RESOURCE_SHORTAGE: 624 return (ENOMEM); 625 } 626 return (EINVAL); 627 } 628 629 #ifndef _SYS_SYSPROTO_H_ 630 struct minherit_args { 631 void *addr; 632 size_t len; 633 int inherit; 634 }; 635 #endif 636 /* 637 * MPSAFE 638 */ 639 int 640 sys_minherit(td, uap) 641 struct thread *td; 642 struct minherit_args *uap; 643 { 644 vm_offset_t addr; 645 vm_size_t size, pageoff; 646 vm_inherit_t inherit; 647 648 addr = (vm_offset_t)uap->addr; 649 size = uap->len; 650 inherit = uap->inherit; 651 652 pageoff = (addr & PAGE_MASK); 653 addr -= pageoff; 654 size += pageoff; 655 size = (vm_size_t) round_page(size); 656 if (addr + size < addr) 657 return (EINVAL); 658 659 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, 660 addr + size, inherit)) { 661 case KERN_SUCCESS: 662 return (0); 663 case KERN_PROTECTION_FAILURE: 664 return (EACCES); 665 } 666 return (EINVAL); 667 } 668 669 #ifndef _SYS_SYSPROTO_H_ 670 struct madvise_args { 671 void *addr; 672 size_t len; 673 int behav; 674 }; 675 #endif 676 677 /* 678 * MPSAFE 679 */ 680 int 681 sys_madvise(td, uap) 682 struct thread *td; 683 struct madvise_args *uap; 684 { 685 vm_offset_t start, end; 686 vm_map_t map; 687 int flags; 688 689 /* 690 * Check for our special case, advising the swap pager we are 691 * "immortal." 692 */ 693 if (uap->behav == MADV_PROTECT) { 694 flags = PPROT_SET; 695 return (kern_procctl(td, P_PID, td->td_proc->p_pid, 696 PROC_SPROTECT, &flags)); 697 } 698 699 /* 700 * Check for illegal behavior 701 */ 702 if (uap->behav < 0 || uap->behav > MADV_CORE) 703 return (EINVAL); 704 /* 705 * Check for illegal addresses. Watch out for address wrap... Note 706 * that VM_*_ADDRESS are not constants due to casts (argh). 707 */ 708 map = &td->td_proc->p_vmspace->vm_map; 709 if ((vm_offset_t)uap->addr < vm_map_min(map) || 710 (vm_offset_t)uap->addr + uap->len > vm_map_max(map)) 711 return (EINVAL); 712 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr) 713 return (EINVAL); 714 715 /* 716 * Since this routine is only advisory, we default to conservative 717 * behavior. 718 */ 719 start = trunc_page((vm_offset_t) uap->addr); 720 end = round_page((vm_offset_t) uap->addr + uap->len); 721 722 if (vm_map_madvise(map, start, end, uap->behav)) 723 return (EINVAL); 724 return (0); 725 } 726 727 #ifndef _SYS_SYSPROTO_H_ 728 struct mincore_args { 729 const void *addr; 730 size_t len; 731 char *vec; 732 }; 733 #endif 734 735 /* 736 * MPSAFE 737 */ 738 int 739 sys_mincore(td, uap) 740 struct thread *td; 741 struct mincore_args *uap; 742 { 743 vm_offset_t addr, first_addr; 744 vm_offset_t end, cend; 745 pmap_t pmap; 746 vm_map_t map; 747 char *vec; 748 int error = 0; 749 int vecindex, lastvecindex; 750 vm_map_entry_t current; 751 vm_map_entry_t entry; 752 vm_object_t object; 753 vm_paddr_t locked_pa; 754 vm_page_t m; 755 vm_pindex_t pindex; 756 int mincoreinfo; 757 unsigned int timestamp; 758 boolean_t locked; 759 760 /* 761 * Make sure that the addresses presented are valid for user 762 * mode. 763 */ 764 first_addr = addr = trunc_page((vm_offset_t) uap->addr); 765 end = addr + (vm_size_t)round_page(uap->len); 766 map = &td->td_proc->p_vmspace->vm_map; 767 if (end > vm_map_max(map) || end < addr) 768 return (ENOMEM); 769 770 /* 771 * Address of byte vector 772 */ 773 vec = uap->vec; 774 775 pmap = vmspace_pmap(td->td_proc->p_vmspace); 776 777 vm_map_lock_read(map); 778 RestartScan: 779 timestamp = map->timestamp; 780 781 if (!vm_map_lookup_entry(map, addr, &entry)) { 782 vm_map_unlock_read(map); 783 return (ENOMEM); 784 } 785 786 /* 787 * Do this on a map entry basis so that if the pages are not 788 * in the current processes address space, we can easily look 789 * up the pages elsewhere. 790 */ 791 lastvecindex = -1; 792 for (current = entry; 793 (current != &map->header) && (current->start < end); 794 current = current->next) { 795 796 /* 797 * check for contiguity 798 */ 799 if (current->end < end && 800 (entry->next == &map->header || 801 current->next->start > current->end)) { 802 vm_map_unlock_read(map); 803 return (ENOMEM); 804 } 805 806 /* 807 * ignore submaps (for now) or null objects 808 */ 809 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 810 current->object.vm_object == NULL) 811 continue; 812 813 /* 814 * limit this scan to the current map entry and the 815 * limits for the mincore call 816 */ 817 if (addr < current->start) 818 addr = current->start; 819 cend = current->end; 820 if (cend > end) 821 cend = end; 822 823 /* 824 * scan this entry one page at a time 825 */ 826 while (addr < cend) { 827 /* 828 * Check pmap first, it is likely faster, also 829 * it can provide info as to whether we are the 830 * one referencing or modifying the page. 831 */ 832 object = NULL; 833 locked_pa = 0; 834 retry: 835 m = NULL; 836 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa); 837 if (locked_pa != 0) { 838 /* 839 * The page is mapped by this process but not 840 * both accessed and modified. It is also 841 * managed. Acquire the object lock so that 842 * other mappings might be examined. 843 */ 844 m = PHYS_TO_VM_PAGE(locked_pa); 845 if (m->object != object) { 846 if (object != NULL) 847 VM_OBJECT_WUNLOCK(object); 848 object = m->object; 849 locked = VM_OBJECT_TRYWLOCK(object); 850 vm_page_unlock(m); 851 if (!locked) { 852 VM_OBJECT_WLOCK(object); 853 vm_page_lock(m); 854 goto retry; 855 } 856 } else 857 vm_page_unlock(m); 858 KASSERT(m->valid == VM_PAGE_BITS_ALL, 859 ("mincore: page %p is mapped but invalid", 860 m)); 861 } else if (mincoreinfo == 0) { 862 /* 863 * The page is not mapped by this process. If 864 * the object implements managed pages, then 865 * determine if the page is resident so that 866 * the mappings might be examined. 867 */ 868 if (current->object.vm_object != object) { 869 if (object != NULL) 870 VM_OBJECT_WUNLOCK(object); 871 object = current->object.vm_object; 872 VM_OBJECT_WLOCK(object); 873 } 874 if (object->type == OBJT_DEFAULT || 875 object->type == OBJT_SWAP || 876 object->type == OBJT_VNODE) { 877 pindex = OFF_TO_IDX(current->offset + 878 (addr - current->start)); 879 m = vm_page_lookup(object, pindex); 880 if (m == NULL && 881 vm_page_is_cached(object, pindex)) 882 mincoreinfo = MINCORE_INCORE; 883 if (m != NULL && m->valid == 0) 884 m = NULL; 885 if (m != NULL) 886 mincoreinfo = MINCORE_INCORE; 887 } 888 } 889 if (m != NULL) { 890 /* Examine other mappings to the page. */ 891 if (m->dirty == 0 && pmap_is_modified(m)) 892 vm_page_dirty(m); 893 if (m->dirty != 0) 894 mincoreinfo |= MINCORE_MODIFIED_OTHER; 895 /* 896 * The first test for PGA_REFERENCED is an 897 * optimization. The second test is 898 * required because a concurrent pmap 899 * operation could clear the last reference 900 * and set PGA_REFERENCED before the call to 901 * pmap_is_referenced(). 902 */ 903 if ((m->aflags & PGA_REFERENCED) != 0 || 904 pmap_is_referenced(m) || 905 (m->aflags & PGA_REFERENCED) != 0) 906 mincoreinfo |= MINCORE_REFERENCED_OTHER; 907 } 908 if (object != NULL) 909 VM_OBJECT_WUNLOCK(object); 910 911 /* 912 * subyte may page fault. In case it needs to modify 913 * the map, we release the lock. 914 */ 915 vm_map_unlock_read(map); 916 917 /* 918 * calculate index into user supplied byte vector 919 */ 920 vecindex = OFF_TO_IDX(addr - first_addr); 921 922 /* 923 * If we have skipped map entries, we need to make sure that 924 * the byte vector is zeroed for those skipped entries. 925 */ 926 while ((lastvecindex + 1) < vecindex) { 927 ++lastvecindex; 928 error = subyte(vec + lastvecindex, 0); 929 if (error) { 930 error = EFAULT; 931 goto done2; 932 } 933 } 934 935 /* 936 * Pass the page information to the user 937 */ 938 error = subyte(vec + vecindex, mincoreinfo); 939 if (error) { 940 error = EFAULT; 941 goto done2; 942 } 943 944 /* 945 * If the map has changed, due to the subyte, the previous 946 * output may be invalid. 947 */ 948 vm_map_lock_read(map); 949 if (timestamp != map->timestamp) 950 goto RestartScan; 951 952 lastvecindex = vecindex; 953 addr += PAGE_SIZE; 954 } 955 } 956 957 /* 958 * subyte may page fault. In case it needs to modify 959 * the map, we release the lock. 960 */ 961 vm_map_unlock_read(map); 962 963 /* 964 * Zero the last entries in the byte vector. 965 */ 966 vecindex = OFF_TO_IDX(end - first_addr); 967 while ((lastvecindex + 1) < vecindex) { 968 ++lastvecindex; 969 error = subyte(vec + lastvecindex, 0); 970 if (error) { 971 error = EFAULT; 972 goto done2; 973 } 974 } 975 976 /* 977 * If the map has changed, due to the subyte, the previous 978 * output may be invalid. 979 */ 980 vm_map_lock_read(map); 981 if (timestamp != map->timestamp) 982 goto RestartScan; 983 vm_map_unlock_read(map); 984 done2: 985 return (error); 986 } 987 988 #ifndef _SYS_SYSPROTO_H_ 989 struct mlock_args { 990 const void *addr; 991 size_t len; 992 }; 993 #endif 994 /* 995 * MPSAFE 996 */ 997 int 998 sys_mlock(td, uap) 999 struct thread *td; 1000 struct mlock_args *uap; 1001 { 1002 1003 return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len)); 1004 } 1005 1006 int 1007 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len) 1008 { 1009 vm_offset_t addr, end, last, start; 1010 vm_size_t npages, size; 1011 vm_map_t map; 1012 unsigned long nsize; 1013 int error; 1014 1015 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0); 1016 if (error) 1017 return (error); 1018 addr = (vm_offset_t)addr0; 1019 size = len; 1020 last = addr + size; 1021 start = trunc_page(addr); 1022 end = round_page(last); 1023 if (last < addr || end < addr) 1024 return (EINVAL); 1025 npages = atop(end - start); 1026 if (npages > vm_page_max_wired) 1027 return (ENOMEM); 1028 map = &proc->p_vmspace->vm_map; 1029 PROC_LOCK(proc); 1030 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 1031 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1032 PROC_UNLOCK(proc); 1033 return (ENOMEM); 1034 } 1035 PROC_UNLOCK(proc); 1036 if (npages + vm_cnt.v_wire_count > vm_page_max_wired) 1037 return (EAGAIN); 1038 #ifdef RACCT 1039 if (racct_enable) { 1040 PROC_LOCK(proc); 1041 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1042 PROC_UNLOCK(proc); 1043 if (error != 0) 1044 return (ENOMEM); 1045 } 1046 #endif 1047 error = vm_map_wire(map, start, end, 1048 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1049 #ifdef RACCT 1050 if (racct_enable && error != KERN_SUCCESS) { 1051 PROC_LOCK(proc); 1052 racct_set(proc, RACCT_MEMLOCK, 1053 ptoa(pmap_wired_count(map->pmap))); 1054 PROC_UNLOCK(proc); 1055 } 1056 #endif 1057 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1058 } 1059 1060 #ifndef _SYS_SYSPROTO_H_ 1061 struct mlockall_args { 1062 int how; 1063 }; 1064 #endif 1065 1066 /* 1067 * MPSAFE 1068 */ 1069 int 1070 sys_mlockall(td, uap) 1071 struct thread *td; 1072 struct mlockall_args *uap; 1073 { 1074 vm_map_t map; 1075 int error; 1076 1077 map = &td->td_proc->p_vmspace->vm_map; 1078 error = priv_check(td, PRIV_VM_MLOCK); 1079 if (error) 1080 return (error); 1081 1082 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1083 return (EINVAL); 1084 1085 /* 1086 * If wiring all pages in the process would cause it to exceed 1087 * a hard resource limit, return ENOMEM. 1088 */ 1089 if (!old_mlock && uap->how & MCL_CURRENT) { 1090 PROC_LOCK(td->td_proc); 1091 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) { 1092 PROC_UNLOCK(td->td_proc); 1093 return (ENOMEM); 1094 } 1095 PROC_UNLOCK(td->td_proc); 1096 } 1097 #ifdef RACCT 1098 if (racct_enable) { 1099 PROC_LOCK(td->td_proc); 1100 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1101 PROC_UNLOCK(td->td_proc); 1102 if (error != 0) 1103 return (ENOMEM); 1104 } 1105 #endif 1106 1107 if (uap->how & MCL_FUTURE) { 1108 vm_map_lock(map); 1109 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1110 vm_map_unlock(map); 1111 error = 0; 1112 } 1113 1114 if (uap->how & MCL_CURRENT) { 1115 /* 1116 * P1003.1-2001 mandates that all currently mapped pages 1117 * will be memory resident and locked (wired) upon return 1118 * from mlockall(). vm_map_wire() will wire pages, by 1119 * calling vm_fault_wire() for each page in the region. 1120 */ 1121 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1122 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1123 error = (error == KERN_SUCCESS ? 0 : EAGAIN); 1124 } 1125 #ifdef RACCT 1126 if (racct_enable && error != KERN_SUCCESS) { 1127 PROC_LOCK(td->td_proc); 1128 racct_set(td->td_proc, RACCT_MEMLOCK, 1129 ptoa(pmap_wired_count(map->pmap))); 1130 PROC_UNLOCK(td->td_proc); 1131 } 1132 #endif 1133 1134 return (error); 1135 } 1136 1137 #ifndef _SYS_SYSPROTO_H_ 1138 struct munlockall_args { 1139 register_t dummy; 1140 }; 1141 #endif 1142 1143 /* 1144 * MPSAFE 1145 */ 1146 int 1147 sys_munlockall(td, uap) 1148 struct thread *td; 1149 struct munlockall_args *uap; 1150 { 1151 vm_map_t map; 1152 int error; 1153 1154 map = &td->td_proc->p_vmspace->vm_map; 1155 error = priv_check(td, PRIV_VM_MUNLOCK); 1156 if (error) 1157 return (error); 1158 1159 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1160 vm_map_lock(map); 1161 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1162 vm_map_unlock(map); 1163 1164 /* Forcibly unwire all pages. */ 1165 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1166 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1167 #ifdef RACCT 1168 if (racct_enable && error == KERN_SUCCESS) { 1169 PROC_LOCK(td->td_proc); 1170 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1171 PROC_UNLOCK(td->td_proc); 1172 } 1173 #endif 1174 1175 return (error); 1176 } 1177 1178 #ifndef _SYS_SYSPROTO_H_ 1179 struct munlock_args { 1180 const void *addr; 1181 size_t len; 1182 }; 1183 #endif 1184 /* 1185 * MPSAFE 1186 */ 1187 int 1188 sys_munlock(td, uap) 1189 struct thread *td; 1190 struct munlock_args *uap; 1191 { 1192 vm_offset_t addr, end, last, start; 1193 vm_size_t size; 1194 #ifdef RACCT 1195 vm_map_t map; 1196 #endif 1197 int error; 1198 1199 error = priv_check(td, PRIV_VM_MUNLOCK); 1200 if (error) 1201 return (error); 1202 addr = (vm_offset_t)uap->addr; 1203 size = uap->len; 1204 last = addr + size; 1205 start = trunc_page(addr); 1206 end = round_page(last); 1207 if (last < addr || end < addr) 1208 return (EINVAL); 1209 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1210 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1211 #ifdef RACCT 1212 if (racct_enable && error == KERN_SUCCESS) { 1213 PROC_LOCK(td->td_proc); 1214 map = &td->td_proc->p_vmspace->vm_map; 1215 racct_set(td->td_proc, RACCT_MEMLOCK, 1216 ptoa(pmap_wired_count(map->pmap))); 1217 PROC_UNLOCK(td->td_proc); 1218 } 1219 #endif 1220 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1221 } 1222 1223 /* 1224 * vm_mmap_vnode() 1225 * 1226 * Helper function for vm_mmap. Perform sanity check specific for mmap 1227 * operations on vnodes. 1228 */ 1229 int 1230 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1231 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1232 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1233 boolean_t *writecounted) 1234 { 1235 struct vattr va; 1236 vm_object_t obj; 1237 vm_offset_t foff; 1238 struct ucred *cred; 1239 int error, flags, locktype; 1240 1241 cred = td->td_ucred; 1242 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED)) 1243 locktype = LK_EXCLUSIVE; 1244 else 1245 locktype = LK_SHARED; 1246 if ((error = vget(vp, locktype, td)) != 0) 1247 return (error); 1248 foff = *foffp; 1249 flags = *flagsp; 1250 obj = vp->v_object; 1251 if (vp->v_type == VREG) { 1252 /* 1253 * Get the proper underlying object 1254 */ 1255 if (obj == NULL) { 1256 error = EINVAL; 1257 goto done; 1258 } 1259 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1260 vput(vp); 1261 vp = (struct vnode *)obj->handle; 1262 /* 1263 * Bypass filesystems obey the mpsafety of the 1264 * underlying fs. Tmpfs never bypasses. 1265 */ 1266 error = vget(vp, locktype, td); 1267 if (error != 0) 1268 return (error); 1269 } 1270 if (locktype == LK_EXCLUSIVE) { 1271 *writecounted = TRUE; 1272 vnode_pager_update_writecount(obj, 0, objsize); 1273 } 1274 } else { 1275 error = EINVAL; 1276 goto done; 1277 } 1278 if ((error = VOP_GETATTR(vp, &va, cred))) 1279 goto done; 1280 #ifdef MAC 1281 /* This relies on VM_PROT_* matching PROT_*. */ 1282 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1283 if (error != 0) 1284 goto done; 1285 #endif 1286 if ((flags & MAP_SHARED) != 0) { 1287 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1288 if (prot & VM_PROT_WRITE) { 1289 error = EPERM; 1290 goto done; 1291 } 1292 *maxprotp &= ~VM_PROT_WRITE; 1293 } 1294 } 1295 /* 1296 * If it is a regular file without any references 1297 * we do not need to sync it. 1298 * Adjust object size to be the size of actual file. 1299 */ 1300 objsize = round_page(va.va_size); 1301 if (va.va_nlink == 0) 1302 flags |= MAP_NOSYNC; 1303 if (obj->type == OBJT_VNODE) { 1304 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1305 cred); 1306 if (obj == NULL) { 1307 error = ENOMEM; 1308 goto done; 1309 } 1310 } else { 1311 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, 1312 ("wrong object type")); 1313 VM_OBJECT_WLOCK(obj); 1314 vm_object_reference_locked(obj); 1315 #if VM_NRESERVLEVEL > 0 1316 vm_object_color(obj, 0); 1317 #endif 1318 VM_OBJECT_WUNLOCK(obj); 1319 } 1320 *objp = obj; 1321 *flagsp = flags; 1322 1323 vfs_mark_atime(vp, cred); 1324 1325 done: 1326 if (error != 0 && *writecounted) { 1327 *writecounted = FALSE; 1328 vnode_pager_update_writecount(obj, objsize, 0); 1329 } 1330 vput(vp); 1331 return (error); 1332 } 1333 1334 /* 1335 * vm_mmap_cdev() 1336 * 1337 * MPSAFE 1338 * 1339 * Helper function for vm_mmap. Perform sanity check specific for mmap 1340 * operations on cdevs. 1341 */ 1342 int 1343 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1344 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, 1345 vm_ooffset_t *foff, vm_object_t *objp) 1346 { 1347 vm_object_t obj; 1348 int error, flags; 1349 1350 flags = *flagsp; 1351 1352 if (dsw->d_flags & D_MMAP_ANON) { 1353 *objp = NULL; 1354 *foff = 0; 1355 *maxprotp = VM_PROT_ALL; 1356 *flagsp |= MAP_ANON; 1357 return (0); 1358 } 1359 /* 1360 * cdevs do not provide private mappings of any kind. 1361 */ 1362 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1363 (prot & VM_PROT_WRITE) != 0) 1364 return (EACCES); 1365 if (flags & (MAP_PRIVATE|MAP_COPY)) 1366 return (EINVAL); 1367 /* 1368 * Force device mappings to be shared. 1369 */ 1370 flags |= MAP_SHARED; 1371 #ifdef MAC_XXX 1372 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); 1373 if (error != 0) 1374 return (error); 1375 #endif 1376 /* 1377 * First, try d_mmap_single(). If that is not implemented 1378 * (returns ENODEV), fall back to using the device pager. 1379 * Note that d_mmap_single() must return a reference to the 1380 * object (it needs to bump the reference count of the object 1381 * it returns somehow). 1382 * 1383 * XXX assumes VM_PROT_* == PROT_* 1384 */ 1385 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); 1386 if (error != ENODEV) 1387 return (error); 1388 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, 1389 td->td_ucred); 1390 if (obj == NULL) 1391 return (EINVAL); 1392 *objp = obj; 1393 *flagsp = flags; 1394 return (0); 1395 } 1396 1397 /* 1398 * vm_mmap() 1399 * 1400 * Internal version of mmap used by exec, sys5 shared memory, and 1401 * various device drivers. Handle is either a vnode pointer, a 1402 * character device, or NULL for MAP_ANON. 1403 */ 1404 int 1405 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1406 vm_prot_t maxprot, int flags, 1407 objtype_t handle_type, void *handle, 1408 vm_ooffset_t foff) 1409 { 1410 vm_object_t object; 1411 struct thread *td = curthread; 1412 int error; 1413 boolean_t writecounted; 1414 1415 if (size == 0) 1416 return (EINVAL); 1417 1418 size = round_page(size); 1419 writecounted = FALSE; 1420 1421 /* 1422 * Lookup/allocate object. 1423 */ 1424 switch (handle_type) { 1425 case OBJT_DEVICE: { 1426 struct cdevsw *dsw; 1427 struct cdev *cdev; 1428 int ref; 1429 1430 cdev = handle; 1431 dsw = dev_refthread(cdev, &ref); 1432 if (dsw == NULL) 1433 return (ENXIO); 1434 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1435 dsw, &foff, &object); 1436 dev_relthread(cdev, ref); 1437 break; 1438 } 1439 case OBJT_VNODE: 1440 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1441 handle, &foff, &object, &writecounted); 1442 break; 1443 case OBJT_DEFAULT: 1444 if (handle == NULL) { 1445 error = 0; 1446 break; 1447 } 1448 /* FALLTHROUGH */ 1449 default: 1450 error = EINVAL; 1451 break; 1452 } 1453 if (error) 1454 return (error); 1455 1456 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1457 foff, writecounted, td); 1458 if (error != 0 && object != NULL) { 1459 /* 1460 * If this mapping was accounted for in the vnode's 1461 * writecount, then undo that now. 1462 */ 1463 if (writecounted) 1464 vnode_pager_release_writecount(object, 0, size); 1465 vm_object_deallocate(object); 1466 } 1467 return (error); 1468 } 1469 1470 /* 1471 * Internal version of mmap that maps a specific VM object into an 1472 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1473 */ 1474 int 1475 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1476 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1477 boolean_t writecounted, struct thread *td) 1478 { 1479 boolean_t fitit; 1480 int docow, error, findspace, rv; 1481 1482 if (map == &td->td_proc->p_vmspace->vm_map) { 1483 PROC_LOCK(td->td_proc); 1484 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) { 1485 PROC_UNLOCK(td->td_proc); 1486 return (ENOMEM); 1487 } 1488 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1489 PROC_UNLOCK(td->td_proc); 1490 return (ENOMEM); 1491 } 1492 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1493 if (ptoa(pmap_wired_count(map->pmap)) + size > 1494 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) { 1495 racct_set_force(td->td_proc, RACCT_VMEM, 1496 map->size); 1497 PROC_UNLOCK(td->td_proc); 1498 return (ENOMEM); 1499 } 1500 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1501 ptoa(pmap_wired_count(map->pmap)) + size); 1502 if (error != 0) { 1503 racct_set_force(td->td_proc, RACCT_VMEM, 1504 map->size); 1505 PROC_UNLOCK(td->td_proc); 1506 return (error); 1507 } 1508 } 1509 PROC_UNLOCK(td->td_proc); 1510 } 1511 1512 /* 1513 * We currently can only deal with page aligned file offsets. 1514 * The mmap() system call already enforces this by subtracting 1515 * the page offset from the file offset, but checking here 1516 * catches errors in device drivers (e.g. d_single_mmap() 1517 * callbacks) and other internal mapping requests (such as in 1518 * exec). 1519 */ 1520 if (foff & PAGE_MASK) 1521 return (EINVAL); 1522 1523 if ((flags & MAP_FIXED) == 0) { 1524 fitit = TRUE; 1525 *addr = round_page(*addr); 1526 } else { 1527 if (*addr != trunc_page(*addr)) 1528 return (EINVAL); 1529 fitit = FALSE; 1530 } 1531 1532 if (flags & MAP_ANON) { 1533 if (object != NULL || foff != 0) 1534 return (EINVAL); 1535 docow = 0; 1536 } else if (flags & MAP_PREFAULT_READ) 1537 docow = MAP_PREFAULT; 1538 else 1539 docow = MAP_PREFAULT_PARTIAL; 1540 1541 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1542 docow |= MAP_COPY_ON_WRITE; 1543 if (flags & MAP_NOSYNC) 1544 docow |= MAP_DISABLE_SYNCER; 1545 if (flags & MAP_NOCORE) 1546 docow |= MAP_DISABLE_COREDUMP; 1547 /* Shared memory is also shared with children. */ 1548 if (flags & MAP_SHARED) 1549 docow |= MAP_INHERIT_SHARE; 1550 if (writecounted) 1551 docow |= MAP_VN_WRITECOUNT; 1552 if (flags & MAP_STACK) { 1553 if (object != NULL) 1554 return (EINVAL); 1555 docow |= MAP_STACK_GROWS_DOWN; 1556 } 1557 if ((flags & MAP_EXCL) != 0) 1558 docow |= MAP_CHECK_EXCL; 1559 1560 if (fitit) { 1561 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1562 findspace = VMFS_SUPER_SPACE; 1563 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1564 findspace = VMFS_ALIGNED_SPACE(flags >> 1565 MAP_ALIGNMENT_SHIFT); 1566 else 1567 findspace = VMFS_OPTIMAL_SPACE; 1568 rv = vm_map_find(map, object, foff, addr, size, 1569 #ifdef MAP_32BIT 1570 flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR : 1571 #endif 1572 0, findspace, prot, maxprot, docow); 1573 } else { 1574 rv = vm_map_fixed(map, object, foff, *addr, size, 1575 prot, maxprot, docow); 1576 } 1577 1578 if (rv == KERN_SUCCESS) { 1579 /* 1580 * If the process has requested that all future mappings 1581 * be wired, then heed this. 1582 */ 1583 if (map->flags & MAP_WIREFUTURE) { 1584 vm_map_wire(map, *addr, *addr + size, 1585 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ? 1586 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES)); 1587 } 1588 } 1589 return (vm_mmap_to_errno(rv)); 1590 } 1591 1592 /* 1593 * Translate a Mach VM return code to zero on success or the appropriate errno 1594 * on failure. 1595 */ 1596 int 1597 vm_mmap_to_errno(int rv) 1598 { 1599 1600 switch (rv) { 1601 case KERN_SUCCESS: 1602 return (0); 1603 case KERN_INVALID_ADDRESS: 1604 case KERN_NO_SPACE: 1605 return (ENOMEM); 1606 case KERN_PROTECTION_FAILURE: 1607 return (EACCES); 1608 default: 1609 return (EINVAL); 1610 } 1611 } 1612