1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 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 * 3. 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 */ 40 41 /* 42 * Mapped file (mmap) interface to VM 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include "opt_hwpmc_hooks.h" 49 #include "opt_vm.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/capsicum.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/mutex.h> 57 #include <sys/sysproto.h> 58 #include <sys/filedesc.h> 59 #include <sys/priv.h> 60 #include <sys/proc.h> 61 #include <sys/procctl.h> 62 #include <sys/racct.h> 63 #include <sys/resource.h> 64 #include <sys/resourcevar.h> 65 #include <sys/rwlock.h> 66 #include <sys/sysctl.h> 67 #include <sys/vnode.h> 68 #include <sys/fcntl.h> 69 #include <sys/file.h> 70 #include <sys/mman.h> 71 #include <sys/mount.h> 72 #include <sys/conf.h> 73 #include <sys/stat.h> 74 #include <sys/syscallsubr.h> 75 #include <sys/sysent.h> 76 #include <sys/vmmeter.h> 77 78 #include <security/audit/audit.h> 79 #include <security/mac/mac_framework.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_object.h> 86 #include <vm/vm_page.h> 87 #include <vm/vm_pager.h> 88 #include <vm/vm_pageout.h> 89 #include <vm/vm_extern.h> 90 #include <vm/vm_page.h> 91 #include <vm/vnode_pager.h> 92 93 #ifdef HWPMC_HOOKS 94 #include <sys/pmckern.h> 95 #endif 96 97 int old_mlock = 0; 98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0, 99 "Do not apply RLIMIT_MEMLOCK on mlockall"); 100 101 #ifdef MAP_32BIT 102 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31) 103 #endif 104 105 #ifndef _SYS_SYSPROTO_H_ 106 struct sbrk_args { 107 int incr; 108 }; 109 #endif 110 111 int 112 sys_sbrk(struct thread *td, struct sbrk_args *uap) 113 { 114 /* Not yet implemented */ 115 return (EOPNOTSUPP); 116 } 117 118 #ifndef _SYS_SYSPROTO_H_ 119 struct sstk_args { 120 int incr; 121 }; 122 #endif 123 124 int 125 sys_sstk(struct thread *td, struct sstk_args *uap) 126 { 127 /* Not yet implemented */ 128 return (EOPNOTSUPP); 129 } 130 131 #if defined(COMPAT_43) 132 #ifndef _SYS_SYSPROTO_H_ 133 struct getpagesize_args { 134 int dummy; 135 }; 136 #endif 137 138 int 139 ogetpagesize(struct thread *td, struct getpagesize_args *uap) 140 { 141 142 td->td_retval[0] = PAGE_SIZE; 143 return (0); 144 } 145 #endif /* COMPAT_43 */ 146 147 148 /* 149 * Memory Map (mmap) system call. Note that the file offset 150 * and address are allowed to be NOT page aligned, though if 151 * the MAP_FIXED flag it set, both must have the same remainder 152 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not 153 * page-aligned, the actual mapping starts at trunc_page(addr) 154 * and the return value is adjusted up by the page offset. 155 * 156 * Generally speaking, only character devices which are themselves 157 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise 158 * there would be no cache coherency between a descriptor and a VM mapping 159 * both to the same character device. 160 */ 161 #ifndef _SYS_SYSPROTO_H_ 162 struct mmap_args { 163 void *addr; 164 size_t len; 165 int prot; 166 int flags; 167 int fd; 168 long pad; 169 off_t pos; 170 }; 171 #endif 172 173 int 174 sys_mmap(struct thread *td, struct mmap_args *uap) 175 { 176 177 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 178 uap->flags, uap->fd, uap->pos)); 179 } 180 181 int 182 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags, 183 int fd, off_t pos) 184 { 185 struct vmspace *vms; 186 struct file *fp; 187 vm_offset_t addr; 188 vm_size_t pageoff; 189 vm_prot_t cap_maxprot; 190 int align, error; 191 cap_rights_t rights; 192 193 vms = td->td_proc->p_vmspace; 194 fp = NULL; 195 AUDIT_ARG_FD(fd); 196 addr = addr0; 197 198 /* 199 * Ignore old flags that used to be defined but did not do anything. 200 */ 201 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040); 202 203 /* 204 * Enforce the constraints. 205 * Mapping of length 0 is only allowed for old binaries. 206 * Anonymous mapping shall specify -1 as filedescriptor and 207 * zero position for new code. Be nice to ancient a.out 208 * binaries and correct pos for anonymous mapping, since old 209 * ld.so sometimes issues anonymous map requests with non-zero 210 * pos. 211 */ 212 if (!SV_CURPROC_FLAG(SV_AOUT)) { 213 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) || 214 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0))) 215 return (EINVAL); 216 } else { 217 if ((flags & MAP_ANON) != 0) 218 pos = 0; 219 } 220 221 if (flags & MAP_STACK) { 222 if ((fd != -1) || 223 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE))) 224 return (EINVAL); 225 flags |= MAP_ANON; 226 pos = 0; 227 } 228 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE | 229 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE | 230 MAP_PREFAULT_READ | MAP_GUARD | 231 #ifdef MAP_32BIT 232 MAP_32BIT | 233 #endif 234 MAP_ALIGNMENT_MASK)) != 0) 235 return (EINVAL); 236 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL) 237 return (EINVAL); 238 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE)) 239 return (EINVAL); 240 if (prot != PROT_NONE && 241 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0) 242 return (EINVAL); 243 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 || 244 pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT | 245 MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0)) 246 return (EINVAL); 247 248 /* 249 * Align the file position to a page boundary, 250 * and save its page offset component. 251 */ 252 pageoff = (pos & PAGE_MASK); 253 pos -= pageoff; 254 255 /* Adjust size for rounding (on both ends). */ 256 size += pageoff; /* low end... */ 257 size = (vm_size_t) round_page(size); /* hi end */ 258 259 /* Ensure alignment is at least a page and fits in a pointer. */ 260 align = flags & MAP_ALIGNMENT_MASK; 261 if (align != 0 && align != MAP_ALIGNED_SUPER && 262 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY || 263 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT)) 264 return (EINVAL); 265 266 /* 267 * Check for illegal addresses. Watch out for address wrap... Note 268 * that VM_*_ADDRESS are not constants due to casts (argh). 269 */ 270 if (flags & MAP_FIXED) { 271 /* 272 * The specified address must have the same remainder 273 * as the file offset taken modulo PAGE_SIZE, so it 274 * should be aligned after adjustment by pageoff. 275 */ 276 addr -= pageoff; 277 if (addr & PAGE_MASK) 278 return (EINVAL); 279 280 /* Address range must be all in user VM space. */ 281 if (addr < vm_map_min(&vms->vm_map) || 282 addr + size > vm_map_max(&vms->vm_map)) 283 return (EINVAL); 284 if (addr + size < addr) 285 return (EINVAL); 286 #ifdef MAP_32BIT 287 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR) 288 return (EINVAL); 289 } else if (flags & MAP_32BIT) { 290 /* 291 * For MAP_32BIT, override the hint if it is too high and 292 * do not bother moving the mapping past the heap (since 293 * the heap is usually above 2GB). 294 */ 295 if (addr + size > MAP_32BIT_MAX_ADDR) 296 addr = 0; 297 #endif 298 } else { 299 /* 300 * XXX for non-fixed mappings where no hint is provided or 301 * the hint would fall in the potential heap space, 302 * place it after the end of the largest possible heap. 303 * 304 * There should really be a pmap call to determine a reasonable 305 * location. 306 */ 307 if (addr == 0 || 308 (addr >= round_page((vm_offset_t)vms->vm_taddr) && 309 addr < round_page((vm_offset_t)vms->vm_daddr + 310 lim_max(td, RLIMIT_DATA)))) 311 addr = round_page((vm_offset_t)vms->vm_daddr + 312 lim_max(td, RLIMIT_DATA)); 313 } 314 if (size == 0) { 315 /* 316 * Return success without mapping anything for old 317 * binaries that request a page-aligned mapping of 318 * length 0. For modern binaries, this function 319 * returns an error earlier. 320 */ 321 error = 0; 322 } else if ((flags & MAP_GUARD) != 0) { 323 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE, 324 VM_PROT_NONE, flags, NULL, pos, FALSE, td); 325 } else if ((flags & MAP_ANON) != 0) { 326 /* 327 * Mapping blank space is trivial. 328 * 329 * This relies on VM_PROT_* matching PROT_*. 330 */ 331 error = vm_mmap_object(&vms->vm_map, &addr, size, prot, 332 VM_PROT_ALL, flags, NULL, pos, FALSE, td); 333 } else { 334 /* 335 * Mapping file, get fp for validation and don't let the 336 * descriptor disappear on us if we block. Check capability 337 * rights, but also return the maximum rights to be combined 338 * with maxprot later. 339 */ 340 cap_rights_init(&rights, CAP_MMAP); 341 if (prot & PROT_READ) 342 cap_rights_set(&rights, CAP_MMAP_R); 343 if ((flags & MAP_SHARED) != 0) { 344 if (prot & PROT_WRITE) 345 cap_rights_set(&rights, CAP_MMAP_W); 346 } 347 if (prot & PROT_EXEC) 348 cap_rights_set(&rights, CAP_MMAP_X); 349 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp); 350 if (error != 0) 351 goto done; 352 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 && 353 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) { 354 error = EINVAL; 355 goto done; 356 } 357 358 /* This relies on VM_PROT_* matching PROT_*. */ 359 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot, 360 cap_maxprot, flags, pos, td); 361 } 362 363 if (error == 0) 364 td->td_retval[0] = (register_t) (addr + pageoff); 365 done: 366 if (fp) 367 fdrop(fp, td); 368 369 return (error); 370 } 371 372 #if defined(COMPAT_FREEBSD6) 373 int 374 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap) 375 { 376 377 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot, 378 uap->flags, uap->fd, uap->pos)); 379 } 380 #endif 381 382 #ifdef COMPAT_43 383 #ifndef _SYS_SYSPROTO_H_ 384 struct ommap_args { 385 caddr_t addr; 386 int len; 387 int prot; 388 int flags; 389 int fd; 390 long pos; 391 }; 392 #endif 393 int 394 ommap(struct thread *td, struct ommap_args *uap) 395 { 396 static const char cvtbsdprot[8] = { 397 0, 398 PROT_EXEC, 399 PROT_WRITE, 400 PROT_EXEC | PROT_WRITE, 401 PROT_READ, 402 PROT_EXEC | PROT_READ, 403 PROT_WRITE | PROT_READ, 404 PROT_EXEC | PROT_WRITE | PROT_READ, 405 }; 406 int flags, prot; 407 408 #define OMAP_ANON 0x0002 409 #define OMAP_COPY 0x0020 410 #define OMAP_SHARED 0x0010 411 #define OMAP_FIXED 0x0100 412 413 prot = cvtbsdprot[uap->prot & 0x7]; 414 #ifdef COMPAT_FREEBSD32 415 #if defined(__amd64__) 416 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) && 417 prot != 0) 418 prot |= PROT_EXEC; 419 #endif 420 #endif 421 flags = 0; 422 if (uap->flags & OMAP_ANON) 423 flags |= MAP_ANON; 424 if (uap->flags & OMAP_COPY) 425 flags |= MAP_COPY; 426 if (uap->flags & OMAP_SHARED) 427 flags |= MAP_SHARED; 428 else 429 flags |= MAP_PRIVATE; 430 if (uap->flags & OMAP_FIXED) 431 flags |= MAP_FIXED; 432 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags, 433 uap->fd, uap->pos)); 434 } 435 #endif /* COMPAT_43 */ 436 437 438 #ifndef _SYS_SYSPROTO_H_ 439 struct msync_args { 440 void *addr; 441 size_t len; 442 int flags; 443 }; 444 #endif 445 int 446 sys_msync(struct thread *td, struct msync_args *uap) 447 { 448 449 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags)); 450 } 451 452 int 453 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags) 454 { 455 vm_offset_t addr; 456 vm_size_t pageoff; 457 vm_map_t map; 458 int rv; 459 460 addr = addr0; 461 pageoff = (addr & PAGE_MASK); 462 addr -= pageoff; 463 size += pageoff; 464 size = (vm_size_t) round_page(size); 465 if (addr + size < addr) 466 return (EINVAL); 467 468 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE)) 469 return (EINVAL); 470 471 map = &td->td_proc->p_vmspace->vm_map; 472 473 /* 474 * Clean the pages and interpret the return value. 475 */ 476 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0, 477 (flags & MS_INVALIDATE) != 0); 478 switch (rv) { 479 case KERN_SUCCESS: 480 return (0); 481 case KERN_INVALID_ADDRESS: 482 return (ENOMEM); 483 case KERN_INVALID_ARGUMENT: 484 return (EBUSY); 485 case KERN_FAILURE: 486 return (EIO); 487 default: 488 return (EINVAL); 489 } 490 } 491 492 #ifndef _SYS_SYSPROTO_H_ 493 struct munmap_args { 494 void *addr; 495 size_t len; 496 }; 497 #endif 498 int 499 sys_munmap(struct thread *td, struct munmap_args *uap) 500 { 501 502 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len)); 503 } 504 505 int 506 kern_munmap(struct thread *td, uintptr_t addr0, size_t size) 507 { 508 #ifdef HWPMC_HOOKS 509 struct pmckern_map_out pkm; 510 vm_map_entry_t entry; 511 bool pmc_handled; 512 #endif 513 vm_offset_t addr; 514 vm_size_t pageoff; 515 vm_map_t map; 516 517 if (size == 0) 518 return (EINVAL); 519 520 addr = addr0; 521 pageoff = (addr & PAGE_MASK); 522 addr -= pageoff; 523 size += pageoff; 524 size = (vm_size_t) round_page(size); 525 if (addr + size < addr) 526 return (EINVAL); 527 528 /* 529 * Check for illegal addresses. Watch out for address wrap... 530 */ 531 map = &td->td_proc->p_vmspace->vm_map; 532 if (addr < vm_map_min(map) || addr + size > vm_map_max(map)) 533 return (EINVAL); 534 vm_map_lock(map); 535 #ifdef HWPMC_HOOKS 536 pmc_handled = false; 537 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) { 538 pmc_handled = true; 539 /* 540 * Inform hwpmc if the address range being unmapped contains 541 * an executable region. 542 */ 543 pkm.pm_address = (uintptr_t) NULL; 544 if (vm_map_lookup_entry(map, addr, &entry)) { 545 for (; entry->start < addr + size; 546 entry = entry->next) { 547 if (vm_map_check_protection(map, entry->start, 548 entry->end, VM_PROT_EXECUTE) == TRUE) { 549 pkm.pm_address = (uintptr_t) addr; 550 pkm.pm_size = (size_t) size; 551 break; 552 } 553 } 554 } 555 } 556 #endif 557 vm_map_delete(map, addr, addr + size); 558 559 #ifdef HWPMC_HOOKS 560 if (__predict_false(pmc_handled)) { 561 /* downgrade the lock to prevent a LOR with the pmc-sx lock */ 562 vm_map_lock_downgrade(map); 563 if (pkm.pm_address != (uintptr_t) NULL) 564 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm); 565 vm_map_unlock_read(map); 566 } else 567 #endif 568 vm_map_unlock(map); 569 570 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */ 571 return (0); 572 } 573 574 #ifndef _SYS_SYSPROTO_H_ 575 struct mprotect_args { 576 const void *addr; 577 size_t len; 578 int prot; 579 }; 580 #endif 581 int 582 sys_mprotect(struct thread *td, struct mprotect_args *uap) 583 { 584 585 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot)); 586 } 587 588 int 589 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot) 590 { 591 vm_offset_t addr; 592 vm_size_t pageoff; 593 594 addr = addr0; 595 prot = (prot & VM_PROT_ALL); 596 pageoff = (addr & PAGE_MASK); 597 addr -= pageoff; 598 size += pageoff; 599 size = (vm_size_t) round_page(size); 600 if (addr + size < addr) 601 return (EINVAL); 602 603 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr, 604 addr + size, prot, FALSE)) { 605 case KERN_SUCCESS: 606 return (0); 607 case KERN_PROTECTION_FAILURE: 608 return (EACCES); 609 case KERN_RESOURCE_SHORTAGE: 610 return (ENOMEM); 611 } 612 return (EINVAL); 613 } 614 615 #ifndef _SYS_SYSPROTO_H_ 616 struct minherit_args { 617 void *addr; 618 size_t len; 619 int inherit; 620 }; 621 #endif 622 int 623 sys_minherit(struct thread *td, struct minherit_args *uap) 624 { 625 vm_offset_t addr; 626 vm_size_t size, pageoff; 627 vm_inherit_t inherit; 628 629 addr = (vm_offset_t)uap->addr; 630 size = uap->len; 631 inherit = uap->inherit; 632 633 pageoff = (addr & PAGE_MASK); 634 addr -= pageoff; 635 size += pageoff; 636 size = (vm_size_t) round_page(size); 637 if (addr + size < addr) 638 return (EINVAL); 639 640 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr, 641 addr + size, inherit)) { 642 case KERN_SUCCESS: 643 return (0); 644 case KERN_PROTECTION_FAILURE: 645 return (EACCES); 646 } 647 return (EINVAL); 648 } 649 650 #ifndef _SYS_SYSPROTO_H_ 651 struct madvise_args { 652 void *addr; 653 size_t len; 654 int behav; 655 }; 656 #endif 657 658 int 659 sys_madvise(struct thread *td, struct madvise_args *uap) 660 { 661 662 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav)); 663 } 664 665 int 666 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav) 667 { 668 vm_map_t map; 669 vm_offset_t addr, end, start; 670 int flags; 671 672 /* 673 * Check for our special case, advising the swap pager we are 674 * "immortal." 675 */ 676 if (behav == MADV_PROTECT) { 677 flags = PPROT_SET; 678 return (kern_procctl(td, P_PID, td->td_proc->p_pid, 679 PROC_SPROTECT, &flags)); 680 } 681 682 /* 683 * Check for illegal behavior 684 */ 685 if (behav < 0 || behav > MADV_CORE) 686 return (EINVAL); 687 /* 688 * Check for illegal addresses. Watch out for address wrap... Note 689 * that VM_*_ADDRESS are not constants due to casts (argh). 690 */ 691 map = &td->td_proc->p_vmspace->vm_map; 692 addr = addr0; 693 if (addr < vm_map_min(map) || addr + len > vm_map_max(map)) 694 return (EINVAL); 695 if ((addr + len) < addr) 696 return (EINVAL); 697 698 /* 699 * Since this routine is only advisory, we default to conservative 700 * behavior. 701 */ 702 start = trunc_page(addr); 703 end = round_page(addr + len); 704 705 if (vm_map_madvise(map, start, end, behav)) 706 return (EINVAL); 707 return (0); 708 } 709 710 #ifndef _SYS_SYSPROTO_H_ 711 struct mincore_args { 712 const void *addr; 713 size_t len; 714 char *vec; 715 }; 716 #endif 717 718 int 719 sys_mincore(struct thread *td, struct mincore_args *uap) 720 { 721 722 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec)); 723 } 724 725 int 726 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec) 727 { 728 vm_offset_t addr, first_addr; 729 vm_offset_t end, cend; 730 pmap_t pmap; 731 vm_map_t map; 732 int error = 0; 733 int vecindex, lastvecindex; 734 vm_map_entry_t current; 735 vm_map_entry_t entry; 736 vm_object_t object; 737 vm_paddr_t locked_pa; 738 vm_page_t m; 739 vm_pindex_t pindex; 740 int mincoreinfo; 741 unsigned int timestamp; 742 boolean_t locked; 743 744 /* 745 * Make sure that the addresses presented are valid for user 746 * mode. 747 */ 748 first_addr = addr = trunc_page(addr0); 749 end = addr + (vm_size_t)round_page(len); 750 map = &td->td_proc->p_vmspace->vm_map; 751 if (end > vm_map_max(map) || end < addr) 752 return (ENOMEM); 753 754 pmap = vmspace_pmap(td->td_proc->p_vmspace); 755 756 vm_map_lock_read(map); 757 RestartScan: 758 timestamp = map->timestamp; 759 760 if (!vm_map_lookup_entry(map, addr, &entry)) { 761 vm_map_unlock_read(map); 762 return (ENOMEM); 763 } 764 765 /* 766 * Do this on a map entry basis so that if the pages are not 767 * in the current processes address space, we can easily look 768 * up the pages elsewhere. 769 */ 770 lastvecindex = -1; 771 for (current = entry; current->start < end; current = current->next) { 772 773 /* 774 * check for contiguity 775 */ 776 if (current->end < end && current->next->start > current->end) { 777 vm_map_unlock_read(map); 778 return (ENOMEM); 779 } 780 781 /* 782 * ignore submaps (for now) or null objects 783 */ 784 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) || 785 current->object.vm_object == NULL) 786 continue; 787 788 /* 789 * limit this scan to the current map entry and the 790 * limits for the mincore call 791 */ 792 if (addr < current->start) 793 addr = current->start; 794 cend = current->end; 795 if (cend > end) 796 cend = end; 797 798 /* 799 * scan this entry one page at a time 800 */ 801 while (addr < cend) { 802 /* 803 * Check pmap first, it is likely faster, also 804 * it can provide info as to whether we are the 805 * one referencing or modifying the page. 806 */ 807 object = NULL; 808 locked_pa = 0; 809 retry: 810 m = NULL; 811 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa); 812 if (locked_pa != 0) { 813 /* 814 * The page is mapped by this process but not 815 * both accessed and modified. It is also 816 * managed. Acquire the object lock so that 817 * other mappings might be examined. 818 */ 819 m = PHYS_TO_VM_PAGE(locked_pa); 820 if (m->object != object) { 821 if (object != NULL) 822 VM_OBJECT_WUNLOCK(object); 823 object = m->object; 824 locked = VM_OBJECT_TRYWLOCK(object); 825 vm_page_unlock(m); 826 if (!locked) { 827 VM_OBJECT_WLOCK(object); 828 vm_page_lock(m); 829 goto retry; 830 } 831 } else 832 vm_page_unlock(m); 833 KASSERT(m->valid == VM_PAGE_BITS_ALL, 834 ("mincore: page %p is mapped but invalid", 835 m)); 836 } else if (mincoreinfo == 0) { 837 /* 838 * The page is not mapped by this process. If 839 * the object implements managed pages, then 840 * determine if the page is resident so that 841 * the mappings might be examined. 842 */ 843 if (current->object.vm_object != object) { 844 if (object != NULL) 845 VM_OBJECT_WUNLOCK(object); 846 object = current->object.vm_object; 847 VM_OBJECT_WLOCK(object); 848 } 849 if (object->type == OBJT_DEFAULT || 850 object->type == OBJT_SWAP || 851 object->type == OBJT_VNODE) { 852 pindex = OFF_TO_IDX(current->offset + 853 (addr - current->start)); 854 m = vm_page_lookup(object, pindex); 855 if (m != NULL && m->valid == 0) 856 m = NULL; 857 if (m != NULL) 858 mincoreinfo = MINCORE_INCORE; 859 } 860 } 861 if (m != NULL) { 862 /* Examine other mappings to the page. */ 863 if (m->dirty == 0 && pmap_is_modified(m)) 864 vm_page_dirty(m); 865 if (m->dirty != 0) 866 mincoreinfo |= MINCORE_MODIFIED_OTHER; 867 /* 868 * The first test for PGA_REFERENCED is an 869 * optimization. The second test is 870 * required because a concurrent pmap 871 * operation could clear the last reference 872 * and set PGA_REFERENCED before the call to 873 * pmap_is_referenced(). 874 */ 875 if ((m->aflags & PGA_REFERENCED) != 0 || 876 pmap_is_referenced(m) || 877 (m->aflags & PGA_REFERENCED) != 0) 878 mincoreinfo |= MINCORE_REFERENCED_OTHER; 879 } 880 if (object != NULL) 881 VM_OBJECT_WUNLOCK(object); 882 883 /* 884 * subyte may page fault. In case it needs to modify 885 * the map, we release the lock. 886 */ 887 vm_map_unlock_read(map); 888 889 /* 890 * calculate index into user supplied byte vector 891 */ 892 vecindex = atop(addr - first_addr); 893 894 /* 895 * If we have skipped map entries, we need to make sure that 896 * the byte vector is zeroed for those skipped entries. 897 */ 898 while ((lastvecindex + 1) < vecindex) { 899 ++lastvecindex; 900 error = subyte(vec + lastvecindex, 0); 901 if (error) { 902 error = EFAULT; 903 goto done2; 904 } 905 } 906 907 /* 908 * Pass the page information to the user 909 */ 910 error = subyte(vec + vecindex, mincoreinfo); 911 if (error) { 912 error = EFAULT; 913 goto done2; 914 } 915 916 /* 917 * If the map has changed, due to the subyte, the previous 918 * output may be invalid. 919 */ 920 vm_map_lock_read(map); 921 if (timestamp != map->timestamp) 922 goto RestartScan; 923 924 lastvecindex = vecindex; 925 addr += PAGE_SIZE; 926 } 927 } 928 929 /* 930 * subyte may page fault. In case it needs to modify 931 * the map, we release the lock. 932 */ 933 vm_map_unlock_read(map); 934 935 /* 936 * Zero the last entries in the byte vector. 937 */ 938 vecindex = atop(end - first_addr); 939 while ((lastvecindex + 1) < vecindex) { 940 ++lastvecindex; 941 error = subyte(vec + lastvecindex, 0); 942 if (error) { 943 error = EFAULT; 944 goto done2; 945 } 946 } 947 948 /* 949 * If the map has changed, due to the subyte, the previous 950 * output may be invalid. 951 */ 952 vm_map_lock_read(map); 953 if (timestamp != map->timestamp) 954 goto RestartScan; 955 vm_map_unlock_read(map); 956 done2: 957 return (error); 958 } 959 960 #ifndef _SYS_SYSPROTO_H_ 961 struct mlock_args { 962 const void *addr; 963 size_t len; 964 }; 965 #endif 966 int 967 sys_mlock(struct thread *td, struct mlock_args *uap) 968 { 969 970 return (kern_mlock(td->td_proc, td->td_ucred, 971 __DECONST(uintptr_t, uap->addr), uap->len)); 972 } 973 974 int 975 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len) 976 { 977 vm_offset_t addr, end, last, start; 978 vm_size_t npages, size; 979 vm_map_t map; 980 unsigned long nsize; 981 int error; 982 983 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0); 984 if (error) 985 return (error); 986 addr = addr0; 987 size = len; 988 last = addr + size; 989 start = trunc_page(addr); 990 end = round_page(last); 991 if (last < addr || end < addr) 992 return (EINVAL); 993 npages = atop(end - start); 994 if (npages > vm_page_max_wired) 995 return (ENOMEM); 996 map = &proc->p_vmspace->vm_map; 997 PROC_LOCK(proc); 998 nsize = ptoa(npages + pmap_wired_count(map->pmap)); 999 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) { 1000 PROC_UNLOCK(proc); 1001 return (ENOMEM); 1002 } 1003 PROC_UNLOCK(proc); 1004 if (npages + vm_wire_count() > vm_page_max_wired) 1005 return (EAGAIN); 1006 #ifdef RACCT 1007 if (racct_enable) { 1008 PROC_LOCK(proc); 1009 error = racct_set(proc, RACCT_MEMLOCK, nsize); 1010 PROC_UNLOCK(proc); 1011 if (error != 0) 1012 return (ENOMEM); 1013 } 1014 #endif 1015 error = vm_map_wire(map, start, end, 1016 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1017 #ifdef RACCT 1018 if (racct_enable && error != KERN_SUCCESS) { 1019 PROC_LOCK(proc); 1020 racct_set(proc, RACCT_MEMLOCK, 1021 ptoa(pmap_wired_count(map->pmap))); 1022 PROC_UNLOCK(proc); 1023 } 1024 #endif 1025 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1026 } 1027 1028 #ifndef _SYS_SYSPROTO_H_ 1029 struct mlockall_args { 1030 int how; 1031 }; 1032 #endif 1033 1034 int 1035 sys_mlockall(struct thread *td, struct mlockall_args *uap) 1036 { 1037 vm_map_t map; 1038 int error; 1039 1040 map = &td->td_proc->p_vmspace->vm_map; 1041 error = priv_check(td, PRIV_VM_MLOCK); 1042 if (error) 1043 return (error); 1044 1045 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0)) 1046 return (EINVAL); 1047 1048 /* 1049 * If wiring all pages in the process would cause it to exceed 1050 * a hard resource limit, return ENOMEM. 1051 */ 1052 if (!old_mlock && uap->how & MCL_CURRENT) { 1053 PROC_LOCK(td->td_proc); 1054 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) { 1055 PROC_UNLOCK(td->td_proc); 1056 return (ENOMEM); 1057 } 1058 PROC_UNLOCK(td->td_proc); 1059 } 1060 #ifdef RACCT 1061 if (racct_enable) { 1062 PROC_LOCK(td->td_proc); 1063 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size); 1064 PROC_UNLOCK(td->td_proc); 1065 if (error != 0) 1066 return (ENOMEM); 1067 } 1068 #endif 1069 1070 if (uap->how & MCL_FUTURE) { 1071 vm_map_lock(map); 1072 vm_map_modflags(map, MAP_WIREFUTURE, 0); 1073 vm_map_unlock(map); 1074 error = 0; 1075 } 1076 1077 if (uap->how & MCL_CURRENT) { 1078 /* 1079 * P1003.1-2001 mandates that all currently mapped pages 1080 * will be memory resident and locked (wired) upon return 1081 * from mlockall(). vm_map_wire() will wire pages, by 1082 * calling vm_fault_wire() for each page in the region. 1083 */ 1084 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map), 1085 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1086 error = (error == KERN_SUCCESS ? 0 : EAGAIN); 1087 } 1088 #ifdef RACCT 1089 if (racct_enable && error != KERN_SUCCESS) { 1090 PROC_LOCK(td->td_proc); 1091 racct_set(td->td_proc, RACCT_MEMLOCK, 1092 ptoa(pmap_wired_count(map->pmap))); 1093 PROC_UNLOCK(td->td_proc); 1094 } 1095 #endif 1096 1097 return (error); 1098 } 1099 1100 #ifndef _SYS_SYSPROTO_H_ 1101 struct munlockall_args { 1102 register_t dummy; 1103 }; 1104 #endif 1105 1106 int 1107 sys_munlockall(struct thread *td, struct munlockall_args *uap) 1108 { 1109 vm_map_t map; 1110 int error; 1111 1112 map = &td->td_proc->p_vmspace->vm_map; 1113 error = priv_check(td, PRIV_VM_MUNLOCK); 1114 if (error) 1115 return (error); 1116 1117 /* Clear the MAP_WIREFUTURE flag from this vm_map. */ 1118 vm_map_lock(map); 1119 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1120 vm_map_unlock(map); 1121 1122 /* Forcibly unwire all pages. */ 1123 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map), 1124 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK); 1125 #ifdef RACCT 1126 if (racct_enable && error == KERN_SUCCESS) { 1127 PROC_LOCK(td->td_proc); 1128 racct_set(td->td_proc, RACCT_MEMLOCK, 0); 1129 PROC_UNLOCK(td->td_proc); 1130 } 1131 #endif 1132 1133 return (error); 1134 } 1135 1136 #ifndef _SYS_SYSPROTO_H_ 1137 struct munlock_args { 1138 const void *addr; 1139 size_t len; 1140 }; 1141 #endif 1142 int 1143 sys_munlock(struct thread *td, struct munlock_args *uap) 1144 { 1145 1146 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len)); 1147 } 1148 1149 int 1150 kern_munlock(struct thread *td, uintptr_t addr0, size_t size) 1151 { 1152 vm_offset_t addr, end, last, start; 1153 #ifdef RACCT 1154 vm_map_t map; 1155 #endif 1156 int error; 1157 1158 error = priv_check(td, PRIV_VM_MUNLOCK); 1159 if (error) 1160 return (error); 1161 addr = addr0; 1162 last = addr + size; 1163 start = trunc_page(addr); 1164 end = round_page(last); 1165 if (last < addr || end < addr) 1166 return (EINVAL); 1167 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end, 1168 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 1169 #ifdef RACCT 1170 if (racct_enable && error == KERN_SUCCESS) { 1171 PROC_LOCK(td->td_proc); 1172 map = &td->td_proc->p_vmspace->vm_map; 1173 racct_set(td->td_proc, RACCT_MEMLOCK, 1174 ptoa(pmap_wired_count(map->pmap))); 1175 PROC_UNLOCK(td->td_proc); 1176 } 1177 #endif 1178 return (error == KERN_SUCCESS ? 0 : ENOMEM); 1179 } 1180 1181 /* 1182 * vm_mmap_vnode() 1183 * 1184 * Helper function for vm_mmap. Perform sanity check specific for mmap 1185 * operations on vnodes. 1186 */ 1187 int 1188 vm_mmap_vnode(struct thread *td, vm_size_t objsize, 1189 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp, 1190 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp, 1191 boolean_t *writecounted) 1192 { 1193 struct vattr va; 1194 vm_object_t obj; 1195 vm_ooffset_t foff; 1196 struct ucred *cred; 1197 int error, flags, locktype; 1198 1199 cred = td->td_ucred; 1200 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED)) 1201 locktype = LK_EXCLUSIVE; 1202 else 1203 locktype = LK_SHARED; 1204 if ((error = vget(vp, locktype, td)) != 0) 1205 return (error); 1206 AUDIT_ARG_VNODE1(vp); 1207 foff = *foffp; 1208 flags = *flagsp; 1209 obj = vp->v_object; 1210 if (vp->v_type == VREG) { 1211 /* 1212 * Get the proper underlying object 1213 */ 1214 if (obj == NULL) { 1215 error = EINVAL; 1216 goto done; 1217 } 1218 if (obj->type == OBJT_VNODE && obj->handle != vp) { 1219 vput(vp); 1220 vp = (struct vnode *)obj->handle; 1221 /* 1222 * Bypass filesystems obey the mpsafety of the 1223 * underlying fs. Tmpfs never bypasses. 1224 */ 1225 error = vget(vp, locktype, td); 1226 if (error != 0) 1227 return (error); 1228 } 1229 if (locktype == LK_EXCLUSIVE) { 1230 *writecounted = TRUE; 1231 vnode_pager_update_writecount(obj, 0, objsize); 1232 } 1233 } else { 1234 error = EINVAL; 1235 goto done; 1236 } 1237 if ((error = VOP_GETATTR(vp, &va, cred))) 1238 goto done; 1239 #ifdef MAC 1240 /* This relies on VM_PROT_* matching PROT_*. */ 1241 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags); 1242 if (error != 0) 1243 goto done; 1244 #endif 1245 if ((flags & MAP_SHARED) != 0) { 1246 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) { 1247 if (prot & VM_PROT_WRITE) { 1248 error = EPERM; 1249 goto done; 1250 } 1251 *maxprotp &= ~VM_PROT_WRITE; 1252 } 1253 } 1254 /* 1255 * If it is a regular file without any references 1256 * we do not need to sync it. 1257 * Adjust object size to be the size of actual file. 1258 */ 1259 objsize = round_page(va.va_size); 1260 if (va.va_nlink == 0) 1261 flags |= MAP_NOSYNC; 1262 if (obj->type == OBJT_VNODE) { 1263 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, 1264 cred); 1265 if (obj == NULL) { 1266 error = ENOMEM; 1267 goto done; 1268 } 1269 } else { 1270 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP, 1271 ("wrong object type")); 1272 VM_OBJECT_WLOCK(obj); 1273 vm_object_reference_locked(obj); 1274 #if VM_NRESERVLEVEL > 0 1275 vm_object_color(obj, 0); 1276 #endif 1277 VM_OBJECT_WUNLOCK(obj); 1278 } 1279 *objp = obj; 1280 *flagsp = flags; 1281 1282 vfs_mark_atime(vp, cred); 1283 1284 done: 1285 if (error != 0 && *writecounted) { 1286 *writecounted = FALSE; 1287 vnode_pager_update_writecount(obj, objsize, 0); 1288 } 1289 vput(vp); 1290 return (error); 1291 } 1292 1293 /* 1294 * vm_mmap_cdev() 1295 * 1296 * Helper function for vm_mmap. Perform sanity check specific for mmap 1297 * operations on cdevs. 1298 */ 1299 int 1300 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1301 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw, 1302 vm_ooffset_t *foff, vm_object_t *objp) 1303 { 1304 vm_object_t obj; 1305 int error, flags; 1306 1307 flags = *flagsp; 1308 1309 if (dsw->d_flags & D_MMAP_ANON) { 1310 *objp = NULL; 1311 *foff = 0; 1312 *maxprotp = VM_PROT_ALL; 1313 *flagsp |= MAP_ANON; 1314 return (0); 1315 } 1316 /* 1317 * cdevs do not provide private mappings of any kind. 1318 */ 1319 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1320 (prot & VM_PROT_WRITE) != 0) 1321 return (EACCES); 1322 if (flags & (MAP_PRIVATE|MAP_COPY)) 1323 return (EINVAL); 1324 /* 1325 * Force device mappings to be shared. 1326 */ 1327 flags |= MAP_SHARED; 1328 #ifdef MAC_XXX 1329 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot); 1330 if (error != 0) 1331 return (error); 1332 #endif 1333 /* 1334 * First, try d_mmap_single(). If that is not implemented 1335 * (returns ENODEV), fall back to using the device pager. 1336 * Note that d_mmap_single() must return a reference to the 1337 * object (it needs to bump the reference count of the object 1338 * it returns somehow). 1339 * 1340 * XXX assumes VM_PROT_* == PROT_* 1341 */ 1342 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot); 1343 if (error != ENODEV) 1344 return (error); 1345 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff, 1346 td->td_ucred); 1347 if (obj == NULL) 1348 return (EINVAL); 1349 *objp = obj; 1350 *flagsp = flags; 1351 return (0); 1352 } 1353 1354 /* 1355 * vm_mmap() 1356 * 1357 * Internal version of mmap used by exec, sys5 shared memory, and 1358 * various device drivers. Handle is either a vnode pointer, a 1359 * character device, or NULL for MAP_ANON. 1360 */ 1361 int 1362 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1363 vm_prot_t maxprot, int flags, 1364 objtype_t handle_type, void *handle, 1365 vm_ooffset_t foff) 1366 { 1367 vm_object_t object; 1368 struct thread *td = curthread; 1369 int error; 1370 boolean_t writecounted; 1371 1372 if (size == 0) 1373 return (EINVAL); 1374 1375 size = round_page(size); 1376 object = NULL; 1377 writecounted = FALSE; 1378 1379 /* 1380 * Lookup/allocate object. 1381 */ 1382 switch (handle_type) { 1383 case OBJT_DEVICE: { 1384 struct cdevsw *dsw; 1385 struct cdev *cdev; 1386 int ref; 1387 1388 cdev = handle; 1389 dsw = dev_refthread(cdev, &ref); 1390 if (dsw == NULL) 1391 return (ENXIO); 1392 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev, 1393 dsw, &foff, &object); 1394 dev_relthread(cdev, ref); 1395 break; 1396 } 1397 case OBJT_VNODE: 1398 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, 1399 handle, &foff, &object, &writecounted); 1400 break; 1401 case OBJT_DEFAULT: 1402 if (handle == NULL) { 1403 error = 0; 1404 break; 1405 } 1406 /* FALLTHROUGH */ 1407 default: 1408 error = EINVAL; 1409 break; 1410 } 1411 if (error) 1412 return (error); 1413 1414 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1415 foff, writecounted, td); 1416 if (error != 0 && object != NULL) { 1417 /* 1418 * If this mapping was accounted for in the vnode's 1419 * writecount, then undo that now. 1420 */ 1421 if (writecounted) 1422 vnode_pager_release_writecount(object, 0, size); 1423 vm_object_deallocate(object); 1424 } 1425 return (error); 1426 } 1427 1428 /* 1429 * Internal version of mmap that maps a specific VM object into an 1430 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap. 1431 */ 1432 int 1433 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, 1434 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff, 1435 boolean_t writecounted, struct thread *td) 1436 { 1437 boolean_t curmap, fitit; 1438 vm_offset_t max_addr; 1439 int docow, error, findspace, rv; 1440 1441 curmap = map == &td->td_proc->p_vmspace->vm_map; 1442 if (curmap) { 1443 PROC_LOCK(td->td_proc); 1444 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) { 1445 PROC_UNLOCK(td->td_proc); 1446 return (ENOMEM); 1447 } 1448 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) { 1449 PROC_UNLOCK(td->td_proc); 1450 return (ENOMEM); 1451 } 1452 if (!old_mlock && map->flags & MAP_WIREFUTURE) { 1453 if (ptoa(pmap_wired_count(map->pmap)) + size > 1454 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) { 1455 racct_set_force(td->td_proc, RACCT_VMEM, 1456 map->size); 1457 PROC_UNLOCK(td->td_proc); 1458 return (ENOMEM); 1459 } 1460 error = racct_set(td->td_proc, RACCT_MEMLOCK, 1461 ptoa(pmap_wired_count(map->pmap)) + size); 1462 if (error != 0) { 1463 racct_set_force(td->td_proc, RACCT_VMEM, 1464 map->size); 1465 PROC_UNLOCK(td->td_proc); 1466 return (error); 1467 } 1468 } 1469 PROC_UNLOCK(td->td_proc); 1470 } 1471 1472 /* 1473 * We currently can only deal with page aligned file offsets. 1474 * The mmap() system call already enforces this by subtracting 1475 * the page offset from the file offset, but checking here 1476 * catches errors in device drivers (e.g. d_single_mmap() 1477 * callbacks) and other internal mapping requests (such as in 1478 * exec). 1479 */ 1480 if (foff & PAGE_MASK) 1481 return (EINVAL); 1482 1483 if ((flags & MAP_FIXED) == 0) { 1484 fitit = TRUE; 1485 *addr = round_page(*addr); 1486 } else { 1487 if (*addr != trunc_page(*addr)) 1488 return (EINVAL); 1489 fitit = FALSE; 1490 } 1491 1492 if (flags & MAP_ANON) { 1493 if (object != NULL || foff != 0) 1494 return (EINVAL); 1495 docow = 0; 1496 } else if (flags & MAP_PREFAULT_READ) 1497 docow = MAP_PREFAULT; 1498 else 1499 docow = MAP_PREFAULT_PARTIAL; 1500 1501 if ((flags & (MAP_ANON|MAP_SHARED)) == 0) 1502 docow |= MAP_COPY_ON_WRITE; 1503 if (flags & MAP_NOSYNC) 1504 docow |= MAP_DISABLE_SYNCER; 1505 if (flags & MAP_NOCORE) 1506 docow |= MAP_DISABLE_COREDUMP; 1507 /* Shared memory is also shared with children. */ 1508 if (flags & MAP_SHARED) 1509 docow |= MAP_INHERIT_SHARE; 1510 if (writecounted) 1511 docow |= MAP_VN_WRITECOUNT; 1512 if (flags & MAP_STACK) { 1513 if (object != NULL) 1514 return (EINVAL); 1515 docow |= MAP_STACK_GROWS_DOWN; 1516 } 1517 if ((flags & MAP_EXCL) != 0) 1518 docow |= MAP_CHECK_EXCL; 1519 if ((flags & MAP_GUARD) != 0) 1520 docow |= MAP_CREATE_GUARD; 1521 1522 if (fitit) { 1523 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER) 1524 findspace = VMFS_SUPER_SPACE; 1525 else if ((flags & MAP_ALIGNMENT_MASK) != 0) 1526 findspace = VMFS_ALIGNED_SPACE(flags >> 1527 MAP_ALIGNMENT_SHIFT); 1528 else 1529 findspace = VMFS_OPTIMAL_SPACE; 1530 max_addr = 0; 1531 #ifdef MAP_32BIT 1532 if ((flags & MAP_32BIT) != 0) 1533 max_addr = MAP_32BIT_MAX_ADDR; 1534 #endif 1535 if (curmap) { 1536 rv = vm_map_find_min(map, object, foff, addr, size, 1537 round_page((vm_offset_t)td->td_proc->p_vmspace-> 1538 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr, 1539 findspace, prot, maxprot, docow); 1540 } else { 1541 rv = vm_map_find(map, object, foff, addr, size, 1542 max_addr, findspace, prot, maxprot, docow); 1543 } 1544 } else { 1545 rv = vm_map_fixed(map, object, foff, *addr, size, 1546 prot, maxprot, docow); 1547 } 1548 1549 if (rv == KERN_SUCCESS) { 1550 /* 1551 * If the process has requested that all future mappings 1552 * be wired, then heed this. 1553 */ 1554 if (map->flags & MAP_WIREFUTURE) { 1555 vm_map_wire(map, *addr, *addr + size, 1556 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ? 1557 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES)); 1558 } 1559 } 1560 return (vm_mmap_to_errno(rv)); 1561 } 1562 1563 /* 1564 * Translate a Mach VM return code to zero on success or the appropriate errno 1565 * on failure. 1566 */ 1567 int 1568 vm_mmap_to_errno(int rv) 1569 { 1570 1571 switch (rv) { 1572 case KERN_SUCCESS: 1573 return (0); 1574 case KERN_INVALID_ADDRESS: 1575 case KERN_NO_SPACE: 1576 return (ENOMEM); 1577 case KERN_PROTECTION_FAILURE: 1578 return (EACCES); 1579 default: 1580 return (EINVAL); 1581 } 1582 } 1583