2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1988 University of Utah.
5 * Copyright (c) 1991, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
42 * Mapped file (mmap) interface to VM
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include "opt_hwpmc_hooks.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
58 #include <sys/filedesc.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>
71 #include <sys/mount.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/vmmeter.h>
77 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
78 #include <machine/md_var.h>
81 #include <security/audit/audit.h>
82 #include <security/mac/mac_framework.h>
85 #include <vm/vm_param.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_pageout.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_page.h>
94 #include <vm/vnode_pager.h>
97 #include <sys/pmckern.h>
101 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
102 "Do not apply RLIMIT_MEMLOCK on mlockall");
103 static int mincore_mapped = 1;
104 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
105 "mincore reports mappings, not residency");
106 static int imply_prot_max = 0;
107 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
108 "Imply maximum page permissions in mmap() when none are specified");
111 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
114 #ifndef _SYS_SYSPROTO_H_
121 sys_sbrk(struct thread *td, struct sbrk_args *uap)
123 /* Not yet implemented */
127 #ifndef _SYS_SYSPROTO_H_
134 sys_sstk(struct thread *td, struct sstk_args *uap)
136 /* Not yet implemented */
140 #if defined(COMPAT_43)
142 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
145 td->td_retval[0] = PAGE_SIZE;
148 #endif /* COMPAT_43 */
152 * Memory Map (mmap) system call. Note that the file offset
153 * and address are allowed to be NOT page aligned, though if
154 * the MAP_FIXED flag it set, both must have the same remainder
155 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
156 * page-aligned, the actual mapping starts at trunc_page(addr)
157 * and the return value is adjusted up by the page offset.
159 * Generally speaking, only character devices which are themselves
160 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
161 * there would be no cache coherency between a descriptor and a VM mapping
162 * both to the same character device.
164 #ifndef _SYS_SYSPROTO_H_
177 sys_mmap(struct thread *td, struct mmap_args *uap)
180 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
181 uap->flags, uap->fd, uap->pos));
185 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
192 vm_size_t pageoff, size;
193 vm_prot_t cap_maxprot;
194 int align, error, max_prot;
197 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
199 max_prot = PROT_MAX_EXTRACT(prot);
200 prot = PROT_EXTRACT(prot);
201 if (max_prot != 0 && (max_prot & prot) != prot)
207 * Always honor PROT_MAX if set. If not, default to all
208 * permissions unless we're implying maximum permissions.
210 * XXX: should be tunable per process and ABI.
213 max_prot = (imply_prot_max && prot != PROT_NONE) ?
222 * Ignore old flags that used to be defined but did not do anything.
224 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
227 * Enforce the constraints.
228 * Mapping of length 0 is only allowed for old binaries.
229 * Anonymous mapping shall specify -1 as filedescriptor and
230 * zero position for new code. Be nice to ancient a.out
231 * binaries and correct pos for anonymous mapping, since old
232 * ld.so sometimes issues anonymous map requests with non-zero
235 if (!SV_CURPROC_FLAG(SV_AOUT)) {
236 if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
237 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
240 if ((flags & MAP_ANON) != 0)
244 if (flags & MAP_STACK) {
246 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
251 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
252 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
253 MAP_PREFAULT_READ | MAP_GUARD |
257 MAP_ALIGNMENT_MASK)) != 0)
259 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
261 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
263 if (prot != PROT_NONE &&
264 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
266 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
267 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
271 MAP_ALIGNMENT_MASK)) != 0))
275 * Align the file position to a page boundary,
276 * and save its page offset component.
278 pageoff = (pos & PAGE_MASK);
281 /* Compute size from len by rounding (on both ends). */
282 size = len + pageoff; /* low end... */
283 size = round_page(size); /* hi end */
284 /* Check for rounding up to zero. */
288 /* Ensure alignment is at least a page and fits in a pointer. */
289 align = flags & MAP_ALIGNMENT_MASK;
290 if (align != 0 && align != MAP_ALIGNED_SUPER &&
291 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
292 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
296 * Check for illegal addresses. Watch out for address wrap... Note
297 * that VM_*_ADDRESS are not constants due to casts (argh).
299 if (flags & MAP_FIXED) {
301 * The specified address must have the same remainder
302 * as the file offset taken modulo PAGE_SIZE, so it
303 * should be aligned after adjustment by pageoff.
306 if (addr & PAGE_MASK)
309 /* Address range must be all in user VM space. */
310 if (addr < vm_map_min(&vms->vm_map) ||
311 addr + size > vm_map_max(&vms->vm_map))
313 if (addr + size < addr)
316 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
318 } else if (flags & MAP_32BIT) {
320 * For MAP_32BIT, override the hint if it is too high and
321 * do not bother moving the mapping past the heap (since
322 * the heap is usually above 2GB).
324 if (addr + size > MAP_32BIT_MAX_ADDR)
329 * XXX for non-fixed mappings where no hint is provided or
330 * the hint would fall in the potential heap space,
331 * place it after the end of the largest possible heap.
333 * There should really be a pmap call to determine a reasonable
337 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
338 addr < round_page((vm_offset_t)vms->vm_daddr +
339 lim_max(td, RLIMIT_DATA))))
340 addr = round_page((vm_offset_t)vms->vm_daddr +
341 lim_max(td, RLIMIT_DATA));
345 * Return success without mapping anything for old
346 * binaries that request a page-aligned mapping of
347 * length 0. For modern binaries, this function
348 * returns an error earlier.
351 } else if ((flags & MAP_GUARD) != 0) {
352 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
353 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
354 } else if ((flags & MAP_ANON) != 0) {
356 * Mapping blank space is trivial.
358 * This relies on VM_PROT_* matching PROT_*.
360 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
361 max_prot, flags, NULL, pos, FALSE, td);
364 * Mapping file, get fp for validation and don't let the
365 * descriptor disappear on us if we block. Check capability
366 * rights, but also return the maximum rights to be combined
367 * with maxprot later.
369 cap_rights_init(&rights, CAP_MMAP);
370 if (prot & PROT_READ)
371 cap_rights_set(&rights, CAP_MMAP_R);
372 if ((flags & MAP_SHARED) != 0) {
373 if (prot & PROT_WRITE)
374 cap_rights_set(&rights, CAP_MMAP_W);
376 if (prot & PROT_EXEC)
377 cap_rights_set(&rights, CAP_MMAP_X);
378 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
381 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
382 p->p_osrel >= P_OSREL_MAP_FSTRICT) {
387 /* This relies on VM_PROT_* matching PROT_*. */
388 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
389 max_prot & cap_maxprot, flags, pos, td);
393 td->td_retval[0] = (register_t) (addr + pageoff);
401 #if defined(COMPAT_FREEBSD6)
403 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
406 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
407 uap->flags, uap->fd, uap->pos));
412 #ifndef _SYS_SYSPROTO_H_
423 ommap(struct thread *td, struct ommap_args *uap)
425 static const char cvtbsdprot[8] = {
429 PROT_EXEC | PROT_WRITE,
431 PROT_EXEC | PROT_READ,
432 PROT_WRITE | PROT_READ,
433 PROT_EXEC | PROT_WRITE | PROT_READ,
437 #define OMAP_ANON 0x0002
438 #define OMAP_COPY 0x0020
439 #define OMAP_SHARED 0x0010
440 #define OMAP_FIXED 0x0100
442 prot = cvtbsdprot[uap->prot & 0x7];
443 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
444 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
449 if (uap->flags & OMAP_ANON)
451 if (uap->flags & OMAP_COPY)
453 if (uap->flags & OMAP_SHARED)
456 flags |= MAP_PRIVATE;
457 if (uap->flags & OMAP_FIXED)
459 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
462 #endif /* COMPAT_43 */
465 #ifndef _SYS_SYSPROTO_H_
473 sys_msync(struct thread *td, struct msync_args *uap)
476 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
480 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
488 pageoff = (addr & PAGE_MASK);
491 size = (vm_size_t) round_page(size);
492 if (addr + size < addr)
495 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
498 map = &td->td_proc->p_vmspace->vm_map;
501 * Clean the pages and interpret the return value.
503 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
504 (flags & MS_INVALIDATE) != 0);
508 case KERN_INVALID_ADDRESS:
510 case KERN_INVALID_ARGUMENT:
519 #ifndef _SYS_SYSPROTO_H_
526 sys_munmap(struct thread *td, struct munmap_args *uap)
529 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
533 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
536 struct pmckern_map_out pkm;
537 vm_map_entry_t entry;
548 pageoff = (addr & PAGE_MASK);
551 size = (vm_size_t) round_page(size);
552 if (addr + size < addr)
556 * Check for illegal addresses. Watch out for address wrap...
558 map = &td->td_proc->p_vmspace->vm_map;
559 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
564 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
567 * Inform hwpmc if the address range being unmapped contains
568 * an executable region.
570 pkm.pm_address = (uintptr_t) NULL;
571 if (vm_map_lookup_entry(map, addr, &entry)) {
572 for (; entry->start < addr + size;
573 entry = entry->next) {
574 if (vm_map_check_protection(map, entry->start,
575 entry->end, VM_PROT_EXECUTE) == TRUE) {
576 pkm.pm_address = (uintptr_t) addr;
577 pkm.pm_size = (size_t) size;
584 vm_map_delete(map, addr, addr + size);
587 if (__predict_false(pmc_handled)) {
588 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
589 vm_map_lock_downgrade(map);
590 if (pkm.pm_address != (uintptr_t) NULL)
591 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
592 vm_map_unlock_read(map);
597 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
601 #ifndef _SYS_SYSPROTO_H_
602 struct mprotect_args {
609 sys_mprotect(struct thread *td, struct mprotect_args *uap)
612 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
616 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
620 int vm_error, max_prot;
623 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
625 max_prot = PROT_MAX_EXTRACT(prot);
626 prot = PROT_EXTRACT(prot);
627 pageoff = (addr & PAGE_MASK);
630 size = (vm_size_t) round_page(size);
631 #ifdef COMPAT_FREEBSD32
632 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
633 if (((addr + size) & 0xffffffff) < addr)
637 if (addr + size < addr)
640 vm_error = KERN_SUCCESS;
642 if ((max_prot & prot) != prot)
644 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
645 addr, addr + size, max_prot, TRUE);
647 if (vm_error == KERN_SUCCESS)
648 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
649 addr, addr + size, prot, FALSE);
654 case KERN_PROTECTION_FAILURE:
656 case KERN_RESOURCE_SHORTAGE:
662 #ifndef _SYS_SYSPROTO_H_
663 struct minherit_args {
670 sys_minherit(struct thread *td, struct minherit_args *uap)
673 vm_size_t size, pageoff;
674 vm_inherit_t inherit;
676 addr = (vm_offset_t)uap->addr;
678 inherit = uap->inherit;
680 pageoff = (addr & PAGE_MASK);
683 size = (vm_size_t) round_page(size);
684 if (addr + size < addr)
687 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
688 addr + size, inherit)) {
691 case KERN_PROTECTION_FAILURE:
697 #ifndef _SYS_SYSPROTO_H_
698 struct madvise_args {
706 sys_madvise(struct thread *td, struct madvise_args *uap)
709 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
713 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
716 vm_offset_t addr, end, start;
720 * Check for our special case, advising the swap pager we are
723 if (behav == MADV_PROTECT) {
725 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
726 PROC_SPROTECT, &flags));
730 * Check for illegal addresses. Watch out for address wrap... Note
731 * that VM_*_ADDRESS are not constants due to casts (argh).
733 map = &td->td_proc->p_vmspace->vm_map;
735 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
737 if ((addr + len) < addr)
741 * Since this routine is only advisory, we default to conservative
744 start = trunc_page(addr);
745 end = round_page(addr + len);
748 * vm_map_madvise() checks for illegal values of behav.
750 return (vm_map_madvise(map, start, end, behav));
753 #ifndef _SYS_SYSPROTO_H_
754 struct mincore_args {
762 sys_mincore(struct thread *td, struct mincore_args *uap)
765 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
769 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
771 vm_offset_t addr, first_addr;
772 vm_offset_t end, cend;
776 int vecindex, lastvecindex;
777 vm_map_entry_t current;
778 vm_map_entry_t entry;
780 vm_paddr_t locked_pa;
784 unsigned int timestamp;
788 * Make sure that the addresses presented are valid for user
791 first_addr = addr = trunc_page(addr0);
792 end = addr + (vm_size_t)round_page(len);
793 map = &td->td_proc->p_vmspace->vm_map;
794 if (end > vm_map_max(map) || end < addr)
797 pmap = vmspace_pmap(td->td_proc->p_vmspace);
799 vm_map_lock_read(map);
801 timestamp = map->timestamp;
803 if (!vm_map_lookup_entry(map, addr, &entry)) {
804 vm_map_unlock_read(map);
809 * Do this on a map entry basis so that if the pages are not
810 * in the current processes address space, we can easily look
811 * up the pages elsewhere.
814 for (current = entry; current->start < end; current = current->next) {
817 * check for contiguity
819 if (current->end < end && current->next->start > current->end) {
820 vm_map_unlock_read(map);
825 * ignore submaps (for now) or null objects
827 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
828 current->object.vm_object == NULL)
832 * limit this scan to the current map entry and the
833 * limits for the mincore call
835 if (addr < current->start)
836 addr = current->start;
842 * scan this entry one page at a time
844 while (addr < cend) {
846 * Check pmap first, it is likely faster, also
847 * it can provide info as to whether we are the
848 * one referencing or modifying the page.
854 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
855 if (mincore_mapped) {
857 * We only care about this pmap's
858 * mapping of the page, if any.
860 if (locked_pa != 0) {
861 vm_page_unlock(PHYS_TO_VM_PAGE(
864 } else if (locked_pa != 0) {
866 * The page is mapped by this process but not
867 * both accessed and modified. It is also
868 * managed. Acquire the object lock so that
869 * other mappings might be examined.
871 m = PHYS_TO_VM_PAGE(locked_pa);
872 if (m->object != object) {
874 VM_OBJECT_WUNLOCK(object);
876 locked = VM_OBJECT_TRYWLOCK(object);
879 VM_OBJECT_WLOCK(object);
885 KASSERT(m->valid == VM_PAGE_BITS_ALL,
886 ("mincore: page %p is mapped but invalid",
888 } else if (mincoreinfo == 0) {
890 * The page is not mapped by this process. If
891 * the object implements managed pages, then
892 * determine if the page is resident so that
893 * the mappings might be examined.
895 if (current->object.vm_object != object) {
897 VM_OBJECT_WUNLOCK(object);
898 object = current->object.vm_object;
899 VM_OBJECT_WLOCK(object);
901 if (object->type == OBJT_DEFAULT ||
902 object->type == OBJT_SWAP ||
903 object->type == OBJT_VNODE) {
904 pindex = OFF_TO_IDX(current->offset +
905 (addr - current->start));
906 m = vm_page_lookup(object, pindex);
907 if (m != NULL && m->valid == 0)
910 mincoreinfo = MINCORE_INCORE;
914 /* Examine other mappings to the page. */
915 if (m->dirty == 0 && pmap_is_modified(m))
918 mincoreinfo |= MINCORE_MODIFIED_OTHER;
920 * The first test for PGA_REFERENCED is an
921 * optimization. The second test is
922 * required because a concurrent pmap
923 * operation could clear the last reference
924 * and set PGA_REFERENCED before the call to
925 * pmap_is_referenced().
927 if ((m->aflags & PGA_REFERENCED) != 0 ||
928 pmap_is_referenced(m) ||
929 (m->aflags & PGA_REFERENCED) != 0)
930 mincoreinfo |= MINCORE_REFERENCED_OTHER;
933 VM_OBJECT_WUNLOCK(object);
936 * subyte may page fault. In case it needs to modify
937 * the map, we release the lock.
939 vm_map_unlock_read(map);
942 * calculate index into user supplied byte vector
944 vecindex = atop(addr - first_addr);
947 * If we have skipped map entries, we need to make sure that
948 * the byte vector is zeroed for those skipped entries.
950 while ((lastvecindex + 1) < vecindex) {
952 error = subyte(vec + lastvecindex, 0);
960 * Pass the page information to the user
962 error = subyte(vec + vecindex, mincoreinfo);
969 * If the map has changed, due to the subyte, the previous
970 * output may be invalid.
972 vm_map_lock_read(map);
973 if (timestamp != map->timestamp)
976 lastvecindex = vecindex;
982 * subyte may page fault. In case it needs to modify
983 * the map, we release the lock.
985 vm_map_unlock_read(map);
988 * Zero the last entries in the byte vector.
990 vecindex = atop(end - first_addr);
991 while ((lastvecindex + 1) < vecindex) {
993 error = subyte(vec + lastvecindex, 0);
1001 * If the map has changed, due to the subyte, the previous
1002 * output may be invalid.
1004 vm_map_lock_read(map);
1005 if (timestamp != map->timestamp)
1007 vm_map_unlock_read(map);
1012 #ifndef _SYS_SYSPROTO_H_
1019 sys_mlock(struct thread *td, struct mlock_args *uap)
1022 return (kern_mlock(td->td_proc, td->td_ucred,
1023 __DECONST(uintptr_t, uap->addr), uap->len));
1027 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1029 vm_offset_t addr, end, last, start;
1030 vm_size_t npages, size;
1032 unsigned long nsize;
1035 error = priv_check_cred(cred, PRIV_VM_MLOCK);
1041 start = trunc_page(addr);
1042 end = round_page(last);
1043 if (last < addr || end < addr)
1045 npages = atop(end - start);
1046 if (npages > vm_page_max_user_wired)
1048 map = &proc->p_vmspace->vm_map;
1050 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1051 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1059 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1065 error = vm_map_wire(map, start, end,
1066 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1068 if (racct_enable && error != KERN_SUCCESS) {
1070 racct_set(proc, RACCT_MEMLOCK,
1071 ptoa(pmap_wired_count(map->pmap)));
1075 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1078 #ifndef _SYS_SYSPROTO_H_
1079 struct mlockall_args {
1085 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1090 map = &td->td_proc->p_vmspace->vm_map;
1091 error = priv_check(td, PRIV_VM_MLOCK);
1095 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1099 * If wiring all pages in the process would cause it to exceed
1100 * a hard resource limit, return ENOMEM.
1102 if (!old_mlock && uap->how & MCL_CURRENT) {
1103 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1108 PROC_LOCK(td->td_proc);
1109 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1110 PROC_UNLOCK(td->td_proc);
1116 if (uap->how & MCL_FUTURE) {
1118 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1123 if (uap->how & MCL_CURRENT) {
1125 * P1003.1-2001 mandates that all currently mapped pages
1126 * will be memory resident and locked (wired) upon return
1127 * from mlockall(). vm_map_wire() will wire pages, by
1128 * calling vm_fault_wire() for each page in the region.
1130 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1131 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1132 if (error == KERN_SUCCESS)
1134 else if (error == KERN_RESOURCE_SHORTAGE)
1140 if (racct_enable && error != KERN_SUCCESS) {
1141 PROC_LOCK(td->td_proc);
1142 racct_set(td->td_proc, RACCT_MEMLOCK,
1143 ptoa(pmap_wired_count(map->pmap)));
1144 PROC_UNLOCK(td->td_proc);
1151 #ifndef _SYS_SYSPROTO_H_
1152 struct munlockall_args {
1158 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1163 map = &td->td_proc->p_vmspace->vm_map;
1164 error = priv_check(td, PRIV_VM_MUNLOCK);
1168 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1170 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1173 /* Forcibly unwire all pages. */
1174 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1175 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1177 if (racct_enable && error == KERN_SUCCESS) {
1178 PROC_LOCK(td->td_proc);
1179 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1180 PROC_UNLOCK(td->td_proc);
1187 #ifndef _SYS_SYSPROTO_H_
1188 struct munlock_args {
1194 sys_munlock(struct thread *td, struct munlock_args *uap)
1197 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1201 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1203 vm_offset_t addr, end, last, start;
1209 error = priv_check(td, PRIV_VM_MUNLOCK);
1214 start = trunc_page(addr);
1215 end = round_page(last);
1216 if (last < addr || end < addr)
1218 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1219 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1221 if (racct_enable && error == KERN_SUCCESS) {
1222 PROC_LOCK(td->td_proc);
1223 map = &td->td_proc->p_vmspace->vm_map;
1224 racct_set(td->td_proc, RACCT_MEMLOCK,
1225 ptoa(pmap_wired_count(map->pmap)));
1226 PROC_UNLOCK(td->td_proc);
1229 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1235 * Helper function for vm_mmap. Perform sanity check specific for mmap
1236 * operations on vnodes.
1239 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1240 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1241 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1242 boolean_t *writecounted)
1251 cred = td->td_ucred;
1252 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1253 (*flagsp & MAP_SHARED) != 0;
1254 if ((error = vget(vp, LK_SHARED, td)) != 0)
1256 AUDIT_ARG_VNODE1(vp);
1260 if (vp->v_type == VREG) {
1262 * Get the proper underlying object
1268 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1270 vp = (struct vnode *)obj->handle;
1272 * Bypass filesystems obey the mpsafety of the
1273 * underlying fs. Tmpfs never bypasses.
1275 error = vget(vp, LK_SHARED, td);
1280 *writecounted = TRUE;
1281 vnode_pager_update_writecount(obj, 0, objsize);
1287 if ((error = VOP_GETATTR(vp, &va, cred)))
1290 /* This relies on VM_PROT_* matching PROT_*. */
1291 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1295 if ((flags & MAP_SHARED) != 0) {
1296 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1297 if (prot & VM_PROT_WRITE) {
1301 *maxprotp &= ~VM_PROT_WRITE;
1305 * If it is a regular file without any references
1306 * we do not need to sync it.
1307 * Adjust object size to be the size of actual file.
1309 objsize = round_page(va.va_size);
1310 if (va.va_nlink == 0)
1311 flags |= MAP_NOSYNC;
1312 if (obj->type == OBJT_VNODE) {
1313 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1320 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1321 ("wrong object type"));
1322 VM_OBJECT_WLOCK(obj);
1323 vm_object_reference_locked(obj);
1324 #if VM_NRESERVLEVEL > 0
1325 vm_object_color(obj, 0);
1327 VM_OBJECT_WUNLOCK(obj);
1332 vfs_mark_atime(vp, cred);
1335 if (error != 0 && *writecounted) {
1336 *writecounted = FALSE;
1337 vnode_pager_update_writecount(obj, objsize, 0);
1346 * Helper function for vm_mmap. Perform sanity check specific for mmap
1347 * operations on cdevs.
1350 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1351 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1352 vm_ooffset_t *foff, vm_object_t *objp)
1359 if (dsw->d_flags & D_MMAP_ANON) {
1362 *maxprotp = VM_PROT_ALL;
1363 *flagsp |= MAP_ANON;
1367 * cdevs do not provide private mappings of any kind.
1369 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1370 (prot & VM_PROT_WRITE) != 0)
1372 if (flags & (MAP_PRIVATE|MAP_COPY))
1375 * Force device mappings to be shared.
1377 flags |= MAP_SHARED;
1379 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1384 * First, try d_mmap_single(). If that is not implemented
1385 * (returns ENODEV), fall back to using the device pager.
1386 * Note that d_mmap_single() must return a reference to the
1387 * object (it needs to bump the reference count of the object
1388 * it returns somehow).
1390 * XXX assumes VM_PROT_* == PROT_*
1392 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1393 if (error != ENODEV)
1395 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1407 * Internal version of mmap used by exec, sys5 shared memory, and
1408 * various device drivers. Handle is either a vnode pointer, a
1409 * character device, or NULL for MAP_ANON.
1412 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1413 vm_prot_t maxprot, int flags,
1414 objtype_t handle_type, void *handle,
1418 struct thread *td = curthread;
1420 boolean_t writecounted;
1425 size = round_page(size);
1427 writecounted = FALSE;
1430 * Lookup/allocate object.
1432 switch (handle_type) {
1439 dsw = dev_refthread(cdev, &ref);
1442 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1443 dsw, &foff, &object);
1444 dev_relthread(cdev, ref);
1448 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1449 handle, &foff, &object, &writecounted);
1452 if (handle == NULL) {
1464 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1465 foff, writecounted, td);
1466 if (error != 0 && object != NULL) {
1468 * If this mapping was accounted for in the vnode's
1469 * writecount, then undo that now.
1472 vnode_pager_release_writecount(object, 0, size);
1473 vm_object_deallocate(object);
1479 * Internal version of mmap that maps a specific VM object into an
1480 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1483 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1484 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1485 boolean_t writecounted, struct thread *td)
1487 boolean_t curmap, fitit;
1488 vm_offset_t max_addr;
1489 int docow, error, findspace, rv;
1491 curmap = map == &td->td_proc->p_vmspace->vm_map;
1493 RACCT_PROC_LOCK(td->td_proc);
1494 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1495 RACCT_PROC_UNLOCK(td->td_proc);
1498 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1499 RACCT_PROC_UNLOCK(td->td_proc);
1502 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1503 if (ptoa(pmap_wired_count(map->pmap)) + size >
1504 lim_cur(td, RLIMIT_MEMLOCK)) {
1505 racct_set_force(td->td_proc, RACCT_VMEM,
1507 RACCT_PROC_UNLOCK(td->td_proc);
1510 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1511 ptoa(pmap_wired_count(map->pmap)) + size);
1513 racct_set_force(td->td_proc, RACCT_VMEM,
1515 RACCT_PROC_UNLOCK(td->td_proc);
1519 RACCT_PROC_UNLOCK(td->td_proc);
1523 * We currently can only deal with page aligned file offsets.
1524 * The mmap() system call already enforces this by subtracting
1525 * the page offset from the file offset, but checking here
1526 * catches errors in device drivers (e.g. d_single_mmap()
1527 * callbacks) and other internal mapping requests (such as in
1530 if (foff & PAGE_MASK)
1533 if ((flags & MAP_FIXED) == 0) {
1535 *addr = round_page(*addr);
1537 if (*addr != trunc_page(*addr))
1542 if (flags & MAP_ANON) {
1543 if (object != NULL || foff != 0)
1546 } else if (flags & MAP_PREFAULT_READ)
1547 docow = MAP_PREFAULT;
1549 docow = MAP_PREFAULT_PARTIAL;
1551 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1552 docow |= MAP_COPY_ON_WRITE;
1553 if (flags & MAP_NOSYNC)
1554 docow |= MAP_DISABLE_SYNCER;
1555 if (flags & MAP_NOCORE)
1556 docow |= MAP_DISABLE_COREDUMP;
1557 /* Shared memory is also shared with children. */
1558 if (flags & MAP_SHARED)
1559 docow |= MAP_INHERIT_SHARE;
1561 docow |= MAP_VN_WRITECOUNT;
1562 if (flags & MAP_STACK) {
1565 docow |= MAP_STACK_GROWS_DOWN;
1567 if ((flags & MAP_EXCL) != 0)
1568 docow |= MAP_CHECK_EXCL;
1569 if ((flags & MAP_GUARD) != 0)
1570 docow |= MAP_CREATE_GUARD;
1573 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1574 findspace = VMFS_SUPER_SPACE;
1575 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1576 findspace = VMFS_ALIGNED_SPACE(flags >>
1577 MAP_ALIGNMENT_SHIFT);
1579 findspace = VMFS_OPTIMAL_SPACE;
1582 if ((flags & MAP_32BIT) != 0)
1583 max_addr = MAP_32BIT_MAX_ADDR;
1586 rv = vm_map_find_min(map, object, foff, addr, size,
1587 round_page((vm_offset_t)td->td_proc->p_vmspace->
1588 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1589 findspace, prot, maxprot, docow);
1591 rv = vm_map_find(map, object, foff, addr, size,
1592 max_addr, findspace, prot, maxprot, docow);
1595 rv = vm_map_fixed(map, object, foff, *addr, size,
1596 prot, maxprot, docow);
1599 if (rv == KERN_SUCCESS) {
1601 * If the process has requested that all future mappings
1602 * be wired, then heed this.
1604 if ((map->flags & MAP_WIREFUTURE) != 0) {
1606 if ((map->flags & MAP_WIREFUTURE) != 0)
1607 (void)vm_map_wire_locked(map, *addr,
1608 *addr + size, VM_MAP_WIRE_USER |
1609 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1610 VM_MAP_WIRE_NOHOLES));
1614 return (vm_mmap_to_errno(rv));
1618 * Translate a Mach VM return code to zero on success or the appropriate errno
1622 vm_mmap_to_errno(int rv)
1628 case KERN_INVALID_ADDRESS:
1631 case KERN_PROTECTION_FAILURE: