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/audit/audit.h> 78 #include <security/mac/mac_framework.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_pager.h> 87 #include <vm/vm_pageout.h> 88 #include <vm/vm_extern.h> 89 #include <vm/vm_page.h> 90 #include <vm/vnode_pager.h> 91 92 #ifdef HWPMC_HOOKS 93 #include <sys/pmckern.h> 94 #endif 95 96 int old_mlock = 0; 97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, 98 "Do not apply RLIMIT_MEMLOCK on mlockall"); 99 100 #ifdef MAP_32BIT 101 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 102 #endif 103 104 #ifndef _SYS_SYSPROTO_H_ 105 struct sbrk_args { 106 int incr; 107 }; 108 #endif 109 110 /* 111 * MPSAFE 112 */ 113 /* ARGSUSED */ 114 int 115 sys_sbrk(td, uap) 116 struct thread *td; 117 struct sbrk_args *uap; 118 { 119 /* Not yet implemented */ 120 return (EOPNOTSUPP); 121 } 122 123 #ifndef _SYS_SYSPROTO_H_ 124 struct sstk_args { 125 int incr; 126 }; 127 #endif 128 129 /* 130 * MPSAFE 131 */ 132 /* ARGSUSED */ 133 int 134 sys_sstk(td, uap) 135 struct thread *td; 136 struct sstk_args *uap; 137 { 138 /* Not yet implemented */ 139 return (EOPNOTSUPP); 140 } 141 142 #if defined(COMPAT_43) 143 #ifndef _SYS_SYSPROTO_H_ 144 struct getpagesize_args { 145 int dummy; 146 }; 147 #endif 148 149 int 150 ogetpagesize(td, uap) 151 struct thread *td; 152 struct getpagesize_args *uap; 153 { 154 /* MP SAFE */ 155 td->td_retval[0] = PAGE_SIZE; 156 return (0); 157 } 158 #endif /* COMPAT_43 */ 159 160 161 /* 162 * Memory Map (mmap) system call. Note that the file offset 163 * and address are allowed to be NOT page aligned, though if 164 * the MAP_FIXED flag it set, both must have the same remainder 165 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 166 * page-aligned, the actual mapping starts at trunc_page(addr) 167 * and the return value is adjusted up by the page offset. 168 * 169 * Generally speaking, only character devices which are themselves 170 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 171 * there would be no cache coherency between a descriptor and a VM mapping 172 * both to the same character device. 173 */ 174 #ifndef _SYS_SYSPROTO_H_ 175 struct mmap_args { 176 void *addr; 177 size_t len; 178 int prot; 179 int flags; 180 int fd; 181 long pad; 182 off_t pos; 183 }; 184 #endif 185 186 /* 187 * MPSAFE 188 */ 189 int 190 sys_mmap(td, uap) 191 struct thread *td; 192 struct mmap_args *uap; 193 { 194 struct file *fp; 195 vm_offset_t addr; 196 vm_size_t size, pageoff; 197 vm_prot_t cap_maxprot; 198 int align, error, flags, prot; 199 off_t pos; 200 struct vmspace *vms = td->td_proc->p_vmspace; 201 cap_rights_t rights; 202 203 addr = (vm_offset_t) uap->addr; 204 size = uap->len; 205 prot = uap->prot; 206 flags = uap->flags; 207 pos = uap->pos; 208 209 fp = NULL; 210 AUDIT_ARG_FD(uap->fd); 211 212 /* 213 * Ignore old flags that used to be defined but did not do anything. 214 */ 215 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 216 217 /* 218 * Enforce the constraints. 219 * Mapping of length 0 is only allowed for old binaries. 220 * Anonymous mapping shall specify -1 as filedescriptor and 221 * zero position for new code. Be nice to ancient a.out 222 * binaries and correct pos for anonymous mapping, since old 223 * ld.so sometimes issues anonymous map requests with non-zero 224 * pos. 225 */ 226 if (!SV_CURPROC_FLAG(SV_AOUT)) { 227 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) || 228 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0))) 229 return (EINVAL); 230 } else { 231 if ((flags & MAP_ANON) != 0) 232 pos = 0; 233 } 234 235 if (flags & MAP_STACK) { 236 if ((uap->fd != -1) || 237 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 238 return (EINVAL); 239 flags |= MAP_ANON; 240 pos = 0; 241 } 242 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 243 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 244 MAP_PREFAULT_READ | 245 #ifdef MAP_32BIT 246 MAP_32BIT | 247 #endif 248 MAP_ALIGNMENT_MASK)) != 0) 249 return (EINVAL); 250 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 251 return (EINVAL); 252 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 253 return (EINVAL); 254 if (prot != PROT_NONE && 255 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 256 return (EINVAL); 257 258 /* 259 * Align the file position to a page boundary, 260 * and save its page offset component. 261 */ 262 pageoff = (pos & PAGE_MASK); 263 pos -= pageoff; 264 265 /* Adjust size for rounding (on both ends). */ 266 size += pageoff; /* low end... */ 267 size = (vm_size_t) round_page(size); /* hi end */ 268 269 /* Ensure alignment is at least a page and fits in a pointer. */ 270 align = flags & MAP_ALIGNMENT_MASK; 271 if (align != 0 && align != MAP_ALIGNED_SUPER && 272 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 273 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 274 return (EINVAL); 275 276 /* 277 * Check for illegal addresses. Watch out for address wrap... Note 278 * that VM_*_ADDRESS are not constants due to casts (argh). 279 */ 280 if (flags & MAP_FIXED) { 281 /* 282 * The specified address must have the same remainder 283 * as the file offset taken modulo PAGE_SIZE, so it 284 * should be aligned after adjustment by pageoff. 285 */ 286 addr -= pageoff; 287 if (addr & PAGE_MASK) 288 return (EINVAL); 289 290 /* Address range must be all in user VM space. */ 291 if (addr < vm_map_min(&vms->vm_map) || 292 addr + size > vm_map_max(&vms->vm_map)) 293 return (EINVAL); 294 if (addr + size < addr) 295 return (EINVAL); 296 #ifdef MAP_32BIT 297 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 298 return (EINVAL); 299 } else if (flags & MAP_32BIT) { 300 /* 301 * For MAP_32BIT, override the hint if it is too high and 302 * do not bother moving the mapping past the heap (since 303 * the heap is usually above 2GB). 304 */ 305 if (addr + size > MAP_32BIT_MAX_ADDR) 306 addr = 0; 307 #endif 308 } else { 309 /* 310 * XXX for non-fixed mappings where no hint is provided or 311 * the hint would fall in the potential heap space, 312 * place it after the end of the largest possible heap. 313 * 314 * There should really be a pmap call to determine a reasonable 315 * location. 316 */ 317 if (addr == 0 || 318 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 319 addr < round_page((vm_offset_t)vms->vm_daddr + 320 lim_max(td, RLIMIT_DATA)))) 321 addr = round_page((vm_offset_t)vms->vm_daddr + 322 lim_max(td, RLIMIT_DATA)); 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 && m->valid == 0) 881 m = NULL; 882 if (m != NULL) 883 mincoreinfo = MINCORE_INCORE; 884 } 885 } 886 if (m != NULL) { 887 /* Examine other mappings to the page. */ 888 if (m->dirty == 0 && pmap_is_modified(m)) 889 vm_page_dirty(m); 890 if (m->dirty != 0) 891 mincoreinfo |= MINCORE_MODIFIED_OTHER; 892 /* 893 * The first test for PGA_REFERENCED is an 894 * optimization. The second test is 895 * required because a concurrent pmap 896 * operation could clear the last reference 897 * and set PGA_REFERENCED before the call to 898 * pmap_is_referenced(). 899 */ 900 if ((m->aflags & PGA_REFERENCED) != 0 || 901 pmap_is_referenced(m) || 902 (m->aflags & PGA_REFERENCED) != 0) 903 mincoreinfo |= MINCORE_REFERENCED_OTHER; 904 } 905 if (object != NULL) 906 VM_OBJECT_WUNLOCK(object); 907 908 /* 909 * subyte may page fault. In case it needs to modify 910 * the map, we release the lock. 911 */ 912 vm_map_unlock_read(map); 913 914 /* 915 * calculate index into user supplied byte vector 916 */ 917 vecindex = OFF_TO_IDX(addr - first_addr); 918 919 /* 920 * If we have skipped map entries, we need to make sure that 921 * the byte vector is zeroed for those skipped entries. 922 */ 923 while ((lastvecindex + 1) < vecindex) { 924 ++lastvecindex; 925 error = subyte(vec + lastvecindex, 0); 926 if (error) { 927 error = EFAULT; 928 goto done2; 929 } 930 } 931 932 /* 933 * Pass the page information to the user 934 */ 935 error = subyte(vec + vecindex, mincoreinfo); 936 if (error) { 937 error = EFAULT; 938 goto done2; 939 } 940 941 /* 942 * If the map has changed, due to the subyte, the previous 943 * output may be invalid. 944 */ 945 vm_map_lock_read(map); 946 if (timestamp != map->timestamp) 947 goto RestartScan; 948 949 lastvecindex = vecindex; 950 addr += PAGE_SIZE; 951 } 952 } 953 954 /* 955 * subyte may page fault. In case it needs to modify 956 * the map, we release the lock. 957 */ 958 vm_map_unlock_read(map); 959 960 /* 961 * Zero the last entries in the byte vector. 962 */ 963 vecindex = OFF_TO_IDX(end - first_addr); 964 while ((lastvecindex + 1) < vecindex) { 965 ++lastvecindex; 966 error = subyte(vec + lastvecindex, 0); 967 if (error) { 968 error = EFAULT; 969 goto done2; 970 } 971 } 972 973 /* 974 * If the map has changed, due to the subyte, the previous 975 * output may be invalid. 976 */ 977 vm_map_lock_read(map); 978 if (timestamp != map->timestamp) 979 goto RestartScan; 980 vm_map_unlock_read(map); 981 done2: 982 return (error); 983 } 984 985 #ifndef _SYS_SYSPROTO_H_ 986 struct mlock_args { 987 const void *addr; 988 size_t len; 989 }; 990 #endif 991 /* 992 * MPSAFE 993 */ 994 int 995 sys_mlock(td, uap) 996 struct thread *td; 997 struct mlock_args *uap; 998 { 999 1000 return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len)); 1001 } 1002 1003 int 1004 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len) 1005 { 1006 vm_offset_t addr, end, last, start; 1007 vm_size_t npages, size; 1008 vm_map_t map; 1009 unsigned long nsize; 1010 int error; 1011 1012 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0); 1013 if (error) 1014 return (error); 1015 addr = (vm_offset_t)addr0; 1016 size = len; 1017 last = addr + size; 1018 start = trunc_page(addr); 1019 end = round_page(last); 1020 if (last < addr || end < addr) 1021 return (EINVAL); 1022 npages = atop(end - start); 1023 if (npages > vm_page_max_wired) 1024 return (ENOMEM); 1025 map = &proc->p_vmspace->vm_map; 1026 PROC_LOCK(proc); 1027 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 1028 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1029 PROC_UNLOCK(proc); 1030 return (ENOMEM); 1031 } 1032 PROC_UNLOCK(proc); 1033 if (npages + vm_cnt.v_wire_count > vm_page_max_wired) 1034 return (EAGAIN); 1035 #ifdef RACCT 1036 if (racct_enable) { 1037 PROC_LOCK(proc); 1038 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1039 PROC_UNLOCK(proc); 1040 if (error != 0) 1041 return (ENOMEM); 1042 } 1043 #endif 1044 error = vm_map_wire(map, start, end, 1045 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1046 #ifdef RACCT 1047 if (racct_enable && error != KERN_SUCCESS) { 1048 PROC_LOCK(proc); 1049 racct_set(proc, RACCT_MEMLOCK, 1050 ptoa(pmap_wired_count(map->pmap))); 1051 PROC_UNLOCK(proc); 1052 } 1053 #endif 1054 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1055 } 1056 1057 #ifndef _SYS_SYSPROTO_H_ 1058 struct mlockall_args { 1059 int how; 1060 }; 1061 #endif 1062 1063 /* 1064 * MPSAFE 1065 */ 1066 int 1067 sys_mlockall(td, uap) 1068 struct thread *td; 1069 struct mlockall_args *uap; 1070 { 1071 vm_map_t map; 1072 int error; 1073 1074 map = &td->td_proc->p_vmspace->vm_map; 1075 error = priv_check(td, PRIV_VM_MLOCK); 1076 if (error) 1077 return (error); 1078 1079 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1080 return (EINVAL); 1081 1082 /* 1083 * If wiring all pages in the process would cause it to exceed 1084 * a hard resource limit, return ENOMEM. 1085 */ 1086 if (!old_mlock && uap->how & MCL_CURRENT) { 1087 PROC_LOCK(td->td_proc); 1088 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) { 1089 PROC_UNLOCK(td->td_proc); 1090 return (ENOMEM); 1091 } 1092 PROC_UNLOCK(td->td_proc); 1093 } 1094 #ifdef RACCT 1095 if (racct_enable) { 1096 PROC_LOCK(td->td_proc); 1097 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1098 PROC_UNLOCK(td->td_proc); 1099 if (error != 0) 1100 return (ENOMEM); 1101 } 1102 #endif 1103 1104 if (uap->how & MCL_FUTURE) { 1105 vm_map_lock(map); 1106 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1107 vm_map_unlock(map); 1108 error = 0; 1109 } 1110 1111 if (uap->how & MCL_CURRENT) { 1112 /* 1113 * P1003.1-2001 mandates that all currently mapped pages 1114 * will be memory resident and locked (wired) upon return 1115 * from mlockall(). vm_map_wire() will wire pages, by 1116 * calling vm_fault_wire() for each page in the region. 1117 */ 1118 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1119 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1120 error = (error == KERN_SUCCESS ? 0 : EAGAIN); 1121 } 1122 #ifdef RACCT 1123 if (racct_enable && error != KERN_SUCCESS) { 1124 PROC_LOCK(td->td_proc); 1125 racct_set(td->td_proc, RACCT_MEMLOCK, 1126 ptoa(pmap_wired_count(map->pmap))); 1127 PROC_UNLOCK(td->td_proc); 1128 } 1129 #endif 1130 1131 return (error); 1132 } 1133 1134 #ifndef _SYS_SYSPROTO_H_ 1135 struct munlockall_args { 1136 register_t dummy; 1137 }; 1138 #endif 1139 1140 /* 1141 * MPSAFE 1142 */ 1143 int 1144 sys_munlockall(td, uap) 1145 struct thread *td; 1146 struct munlockall_args *uap; 1147 { 1148 vm_map_t map; 1149 int error; 1150 1151 map = &td->td_proc->p_vmspace->vm_map; 1152 error = priv_check(td, PRIV_VM_MUNLOCK); 1153 if (error) 1154 return (error); 1155 1156 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1157 vm_map_lock(map); 1158 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1159 vm_map_unlock(map); 1160 1161 /* Forcibly unwire all pages. */ 1162 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1163 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1164 #ifdef RACCT 1165 if (racct_enable && error == KERN_SUCCESS) { 1166 PROC_LOCK(td->td_proc); 1167 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1168 PROC_UNLOCK(td->td_proc); 1169 } 1170 #endif 1171 1172 return (error); 1173 } 1174 1175 #ifndef _SYS_SYSPROTO_H_ 1176 struct munlock_args { 1177 const void *addr; 1178 size_t len; 1179 }; 1180 #endif 1181 /* 1182 * MPSAFE 1183 */ 1184 int 1185 sys_munlock(td, uap) 1186 struct thread *td; 1187 struct munlock_args *uap; 1188 { 1189 vm_offset_t addr, end, last, start; 1190 vm_size_t size; 1191 #ifdef RACCT 1192 vm_map_t map; 1193 #endif 1194 int error; 1195 1196 error = priv_check(td, PRIV_VM_MUNLOCK); 1197 if (error) 1198 return (error); 1199 addr = (vm_offset_t)uap->addr; 1200 size = uap->len; 1201 last = addr + size; 1202 start = trunc_page(addr); 1203 end = round_page(last); 1204 if (last < addr || end < addr) 1205 return (EINVAL); 1206 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1207 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1208 #ifdef RACCT 1209 if (racct_enable && error == KERN_SUCCESS) { 1210 PROC_LOCK(td->td_proc); 1211 map = &td->td_proc->p_vmspace->vm_map; 1212 racct_set(td->td_proc, RACCT_MEMLOCK, 1213 ptoa(pmap_wired_count(map->pmap))); 1214 PROC_UNLOCK(td->td_proc); 1215 } 1216 #endif 1217 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1218 } 1219 1220 /* 1221 * vm_mmap_vnode() 1222 * 1223 * Helper function for vm_mmap. Perform sanity check specific for mmap 1224 * operations on vnodes. 1225 */ 1226 int 1227 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1228 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1229 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1230 boolean_t *writecounted) 1231 { 1232 struct vattr va; 1233 vm_object_t obj; 1234 vm_offset_t foff; 1235 struct ucred *cred; 1236 int error, flags, locktype; 1237 1238 cred = td->td_ucred; 1239 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED)) 1240 locktype = LK_EXCLUSIVE; 1241 else 1242 locktype = LK_SHARED; 1243 if ((error = vget(vp, locktype, td)) != 0) 1244 return (error); 1245 AUDIT_ARG_VNODE1(vp); 1246 foff = *foffp; 1247 flags = *flagsp; 1248 obj = vp->v_object; 1249 if (vp->v_type == VREG) { 1250 /* 1251 * Get the proper underlying object 1252 */ 1253 if (obj == NULL) { 1254 error = EINVAL; 1255 goto done; 1256 } 1257 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1258 vput(vp); 1259 vp = (struct vnode *)obj->handle; 1260 /* 1261 * Bypass filesystems obey the mpsafety of the 1262 * underlying fs. Tmpfs never bypasses. 1263 */ 1264 error = vget(vp, locktype, td); 1265 if (error != 0) 1266 return (error); 1267 } 1268 if (locktype == LK_EXCLUSIVE) { 1269 *writecounted = TRUE; 1270 vnode_pager_update_writecount(obj, 0, objsize); 1271 } 1272 } else { 1273 error = EINVAL; 1274 goto done; 1275 } 1276 if ((error = VOP_GETATTR(vp, &va, cred))) 1277 goto done; 1278 #ifdef MAC 1279 /* This relies on VM_PROT_* matching PROT_*. */ 1280 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1281 if (error != 0) 1282 goto done; 1283 #endif 1284 if ((flags & MAP_SHARED) != 0) { 1285 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1286 if (prot & VM_PROT_WRITE) { 1287 error = EPERM; 1288 goto done; 1289 } 1290 *maxprotp &= ~VM_PROT_WRITE; 1291 } 1292 } 1293 /* 1294 * If it is a regular file without any references 1295 * we do not need to sync it. 1296 * Adjust object size to be the size of actual file. 1297 */ 1298 objsize = round_page(va.va_size); 1299 if (va.va_nlink == 0) 1300 flags |= MAP_NOSYNC; 1301 if (obj->type == OBJT_VNODE) { 1302 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1303 cred); 1304 if (obj == NULL) { 1305 error = ENOMEM; 1306 goto done; 1307 } 1308 } else { 1309 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, 1310 ("wrong object type")); 1311 VM_OBJECT_WLOCK(obj); 1312 vm_object_reference_locked(obj); 1313 #if VM_NRESERVLEVEL > 0 1314 vm_object_color(obj, 0); 1315 #endif 1316 VM_OBJECT_WUNLOCK(obj); 1317 } 1318 *objp = obj; 1319 *flagsp = flags; 1320 1321 vfs_mark_atime(vp, cred); 1322 1323 done: 1324 if (error != 0 && *writecounted) { 1325 *writecounted = FALSE; 1326 vnode_pager_update_writecount(obj, objsize, 0); 1327 } 1328 vput(vp); 1329 return (error); 1330 } 1331 1332 /* 1333 * vm_mmap_cdev() 1334 * 1335 * MPSAFE 1336 * 1337 * Helper function for vm_mmap. Perform sanity check specific for mmap 1338 * operations on cdevs. 1339 */ 1340 int 1341 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1342 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, 1343 vm_ooffset_t *foff, vm_object_t *objp) 1344 { 1345 vm_object_t obj; 1346 int error, flags; 1347 1348 flags = *flagsp; 1349 1350 if (dsw->d_flags & D_MMAP_ANON) { 1351 *objp = NULL; 1352 *foff = 0; 1353 *maxprotp = VM_PROT_ALL; 1354 *flagsp |= MAP_ANON; 1355 return (0); 1356 } 1357 /* 1358 * cdevs do not provide private mappings of any kind. 1359 */ 1360 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1361 (prot & VM_PROT_WRITE) != 0) 1362 return (EACCES); 1363 if (flags & (MAP_PRIVATE|MAP_COPY)) 1364 return (EINVAL); 1365 /* 1366 * Force device mappings to be shared. 1367 */ 1368 flags |= MAP_SHARED; 1369 #ifdef MAC_XXX 1370 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); 1371 if (error != 0) 1372 return (error); 1373 #endif 1374 /* 1375 * First, try d_mmap_single(). If that is not implemented 1376 * (returns ENODEV), fall back to using the device pager. 1377 * Note that d_mmap_single() must return a reference to the 1378 * object (it needs to bump the reference count of the object 1379 * it returns somehow). 1380 * 1381 * XXX assumes VM_PROT_* == PROT_* 1382 */ 1383 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); 1384 if (error != ENODEV) 1385 return (error); 1386 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, 1387 td->td_ucred); 1388 if (obj == NULL) 1389 return (EINVAL); 1390 *objp = obj; 1391 *flagsp = flags; 1392 return (0); 1393 } 1394 1395 /* 1396 * vm_mmap() 1397 * 1398 * Internal version of mmap used by exec, sys5 shared memory, and 1399 * various device drivers. Handle is either a vnode pointer, a 1400 * character device, or NULL for MAP_ANON. 1401 */ 1402 int 1403 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1404 vm_prot_t maxprot, int flags, 1405 objtype_t handle_type, void *handle, 1406 vm_ooffset_t foff) 1407 { 1408 vm_object_t object; 1409 struct thread *td = curthread; 1410 int error; 1411 boolean_t writecounted; 1412 1413 if (size == 0) 1414 return (EINVAL); 1415 1416 size = round_page(size); 1417 object = NULL; 1418 writecounted = FALSE; 1419 1420 /* 1421 * Lookup/allocate object. 1422 */ 1423 switch (handle_type) { 1424 case OBJT_DEVICE: { 1425 struct cdevsw *dsw; 1426 struct cdev *cdev; 1427 int ref; 1428 1429 cdev = handle; 1430 dsw = dev_refthread(cdev, &ref); 1431 if (dsw == NULL) 1432 return (ENXIO); 1433 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1434 dsw, &foff, &object); 1435 dev_relthread(cdev, ref); 1436 break; 1437 } 1438 case OBJT_VNODE: 1439 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1440 handle, &foff, &object, &writecounted); 1441 break; 1442 case OBJT_DEFAULT: 1443 if (handle == NULL) { 1444 error = 0; 1445 break; 1446 } 1447 /* FALLTHROUGH */ 1448 default: 1449 error = EINVAL; 1450 break; 1451 } 1452 if (error) 1453 return (error); 1454 1455 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1456 foff, writecounted, td); 1457 if (error != 0 && object != NULL) { 1458 /* 1459 * If this mapping was accounted for in the vnode's 1460 * writecount, then undo that now. 1461 */ 1462 if (writecounted) 1463 vnode_pager_release_writecount(object, 0, size); 1464 vm_object_deallocate(object); 1465 } 1466 return (error); 1467 } 1468 1469 /* 1470 * Internal version of mmap that maps a specific VM object into an 1471 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1472 */ 1473 int 1474 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1475 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1476 boolean_t writecounted, struct thread *td) 1477 { 1478 boolean_t fitit; 1479 int docow, error, findspace, rv; 1480 1481 if (map == &td->td_proc->p_vmspace->vm_map) { 1482 PROC_LOCK(td->td_proc); 1483 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) { 1484 PROC_UNLOCK(td->td_proc); 1485 return (ENOMEM); 1486 } 1487 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1488 PROC_UNLOCK(td->td_proc); 1489 return (ENOMEM); 1490 } 1491 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1492 if (ptoa(pmap_wired_count(map->pmap)) + size > 1493 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) { 1494 racct_set_force(td->td_proc, RACCT_VMEM, 1495 map->size); 1496 PROC_UNLOCK(td->td_proc); 1497 return (ENOMEM); 1498 } 1499 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1500 ptoa(pmap_wired_count(map->pmap)) + size); 1501 if (error != 0) { 1502 racct_set_force(td->td_proc, RACCT_VMEM, 1503 map->size); 1504 PROC_UNLOCK(td->td_proc); 1505 return (error); 1506 } 1507 } 1508 PROC_UNLOCK(td->td_proc); 1509 } 1510 1511 /* 1512 * We currently can only deal with page aligned file offsets. 1513 * The mmap() system call already enforces this by subtracting 1514 * the page offset from the file offset, but checking here 1515 * catches errors in device drivers (e.g. d_single_mmap() 1516 * callbacks) and other internal mapping requests (such as in 1517 * exec). 1518 */ 1519 if (foff & PAGE_MASK) 1520 return (EINVAL); 1521 1522 if ((flags & MAP_FIXED) == 0) { 1523 fitit = TRUE; 1524 *addr = round_page(*addr); 1525 } else { 1526 if (*addr != trunc_page(*addr)) 1527 return (EINVAL); 1528 fitit = FALSE; 1529 } 1530 1531 if (flags & MAP_ANON) { 1532 if (object != NULL || foff != 0) 1533 return (EINVAL); 1534 docow = 0; 1535 } else if (flags & MAP_PREFAULT_READ) 1536 docow = MAP_PREFAULT; 1537 else 1538 docow = MAP_PREFAULT_PARTIAL; 1539 1540 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1541 docow |= MAP_COPY_ON_WRITE; 1542 if (flags & MAP_NOSYNC) 1543 docow |= MAP_DISABLE_SYNCER; 1544 if (flags & MAP_NOCORE) 1545 docow |= MAP_DISABLE_COREDUMP; 1546 /* Shared memory is also shared with children. */ 1547 if (flags & MAP_SHARED) 1548 docow |= MAP_INHERIT_SHARE; 1549 if (writecounted) 1550 docow |= MAP_VN_WRITECOUNT; 1551 if (flags & MAP_STACK) { 1552 if (object != NULL) 1553 return (EINVAL); 1554 docow |= MAP_STACK_GROWS_DOWN; 1555 } 1556 if ((flags & MAP_EXCL) != 0) 1557 docow |= MAP_CHECK_EXCL; 1558 1559 if (fitit) { 1560 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1561 findspace = VMFS_SUPER_SPACE; 1562 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1563 findspace = VMFS_ALIGNED_SPACE(flags >> 1564 MAP_ALIGNMENT_SHIFT); 1565 else 1566 findspace = VMFS_OPTIMAL_SPACE; 1567 rv = vm_map_find(map, object, foff, addr, size, 1568 #ifdef MAP_32BIT 1569 flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR : 1570 #endif 1571 0, findspace, prot, maxprot, docow); 1572 } else { 1573 rv = vm_map_fixed(map, object, foff, *addr, size, 1574 prot, maxprot, docow); 1575 } 1576 1577 if (rv == KERN_SUCCESS) { 1578 /* 1579 * If the process has requested that all future mappings 1580 * be wired, then heed this. 1581 */ 1582 if (map->flags & MAP_WIREFUTURE) { 1583 vm_map_wire(map, *addr, *addr + size, 1584 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ? 1585 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES)); 1586 } 1587 } 1588 return (vm_mmap_to_errno(rv)); 1589 } 1590 1591 /* 1592 * Translate a Mach VM return code to zero on success or the appropriate errno 1593 * on failure. 1594 */ 1595 int 1596 vm_mmap_to_errno(int rv) 1597 { 1598 1599 switch (rv) { 1600 case KERN_SUCCESS: 1601 return (0); 1602 case KERN_INVALID_ADDRESS: 1603 case KERN_NO_SPACE: 1604 return (ENOMEM); 1605 case KERN_PROTECTION_FAILURE: 1606 return (EACCES); 1607 default: 1608 return (EINVAL); 1609 } 1610 } 1611