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,
191 vm_size_t pageoff, size;
192 vm_prot_t cap_maxprot;
193 int align, error, max_prot;
196 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
198 max_prot = PROT_MAX_EXTRACT(prot);
199 prot = PROT_EXTRACT(prot);
200 if (max_prot != 0 && (max_prot & prot) != prot)
203 * Always honor PROT_MAX if set. If not, default to all
204 * permissions unless we're implying maximum permissions.
206 * XXX: should be tunable per process and ABI.
209 max_prot = (imply_prot_max && prot != PROT_NONE) ?
212 vms = td->td_proc->p_vmspace;
218 * Ignore old flags that used to be defined but did not do anything.
220 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
223 * Enforce the constraints.
224 * Mapping of length 0 is only allowed for old binaries.
225 * Anonymous mapping shall specify -1 as filedescriptor and
226 * zero position for new code. Be nice to ancient a.out
227 * binaries and correct pos for anonymous mapping, since old
228 * ld.so sometimes issues anonymous map requests with non-zero
231 if (!SV_CURPROC_FLAG(SV_AOUT)) {
232 if ((len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
233 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
236 if ((flags & MAP_ANON) != 0)
240 if (flags & MAP_STACK) {
242 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
247 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
248 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
249 MAP_PREFAULT_READ | MAP_GUARD |
253 MAP_ALIGNMENT_MASK)) != 0)
255 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
257 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
259 if (prot != PROT_NONE &&
260 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
262 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
263 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
267 MAP_ALIGNMENT_MASK)) != 0))
271 * Align the file position to a page boundary,
272 * and save its page offset component.
274 pageoff = (pos & PAGE_MASK);
277 /* Compute size from len by rounding (on both ends). */
278 size = len + pageoff; /* low end... */
279 size = round_page(size); /* hi end */
280 /* Check for rounding up to zero. */
284 /* Ensure alignment is at least a page and fits in a pointer. */
285 align = flags & MAP_ALIGNMENT_MASK;
286 if (align != 0 && align != MAP_ALIGNED_SUPER &&
287 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
288 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
292 * Check for illegal addresses. Watch out for address wrap... Note
293 * that VM_*_ADDRESS are not constants due to casts (argh).
295 if (flags & MAP_FIXED) {
297 * The specified address must have the same remainder
298 * as the file offset taken modulo PAGE_SIZE, so it
299 * should be aligned after adjustment by pageoff.
302 if (addr & PAGE_MASK)
305 /* Address range must be all in user VM space. */
306 if (addr < vm_map_min(&vms->vm_map) ||
307 addr + size > vm_map_max(&vms->vm_map))
309 if (addr + size < addr)
312 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
314 } else if (flags & MAP_32BIT) {
316 * For MAP_32BIT, override the hint if it is too high and
317 * do not bother moving the mapping past the heap (since
318 * the heap is usually above 2GB).
320 if (addr + size > MAP_32BIT_MAX_ADDR)
325 * XXX for non-fixed mappings where no hint is provided or
326 * the hint would fall in the potential heap space,
327 * place it after the end of the largest possible heap.
329 * There should really be a pmap call to determine a reasonable
333 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
334 addr < round_page((vm_offset_t)vms->vm_daddr +
335 lim_max(td, RLIMIT_DATA))))
336 addr = round_page((vm_offset_t)vms->vm_daddr +
337 lim_max(td, RLIMIT_DATA));
341 * Return success without mapping anything for old
342 * binaries that request a page-aligned mapping of
343 * length 0. For modern binaries, this function
344 * returns an error earlier.
347 } else if ((flags & MAP_GUARD) != 0) {
348 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
349 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
350 } else if ((flags & MAP_ANON) != 0) {
352 * Mapping blank space is trivial.
354 * This relies on VM_PROT_* matching PROT_*.
356 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
357 max_prot, flags, NULL, pos, FALSE, td);
360 * Mapping file, get fp for validation and don't let the
361 * descriptor disappear on us if we block. Check capability
362 * rights, but also return the maximum rights to be combined
363 * with maxprot later.
365 cap_rights_init(&rights, CAP_MMAP);
366 if (prot & PROT_READ)
367 cap_rights_set(&rights, CAP_MMAP_R);
368 if ((flags & MAP_SHARED) != 0) {
369 if (prot & PROT_WRITE)
370 cap_rights_set(&rights, CAP_MMAP_W);
372 if (prot & PROT_EXEC)
373 cap_rights_set(&rights, CAP_MMAP_X);
374 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
377 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
378 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
383 /* This relies on VM_PROT_* matching PROT_*. */
384 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
385 max_prot & cap_maxprot, flags, pos, td);
389 td->td_retval[0] = (register_t) (addr + pageoff);
397 #if defined(COMPAT_FREEBSD6)
399 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
402 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
403 uap->flags, uap->fd, uap->pos));
408 #ifndef _SYS_SYSPROTO_H_
419 ommap(struct thread *td, struct ommap_args *uap)
421 static const char cvtbsdprot[8] = {
425 PROT_EXEC | PROT_WRITE,
427 PROT_EXEC | PROT_READ,
428 PROT_WRITE | PROT_READ,
429 PROT_EXEC | PROT_WRITE | PROT_READ,
433 #define OMAP_ANON 0x0002
434 #define OMAP_COPY 0x0020
435 #define OMAP_SHARED 0x0010
436 #define OMAP_FIXED 0x0100
438 prot = cvtbsdprot[uap->prot & 0x7];
439 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
440 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
445 if (uap->flags & OMAP_ANON)
447 if (uap->flags & OMAP_COPY)
449 if (uap->flags & OMAP_SHARED)
452 flags |= MAP_PRIVATE;
453 if (uap->flags & OMAP_FIXED)
455 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
458 #endif /* COMPAT_43 */
461 #ifndef _SYS_SYSPROTO_H_
469 sys_msync(struct thread *td, struct msync_args *uap)
472 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
476 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
484 pageoff = (addr & PAGE_MASK);
487 size = (vm_size_t) round_page(size);
488 if (addr + size < addr)
491 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
494 map = &td->td_proc->p_vmspace->vm_map;
497 * Clean the pages and interpret the return value.
499 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
500 (flags & MS_INVALIDATE) != 0);
504 case KERN_INVALID_ADDRESS:
506 case KERN_INVALID_ARGUMENT:
515 #ifndef _SYS_SYSPROTO_H_
522 sys_munmap(struct thread *td, struct munmap_args *uap)
525 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
529 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
532 struct pmckern_map_out pkm;
533 vm_map_entry_t entry;
544 pageoff = (addr & PAGE_MASK);
547 size = (vm_size_t) round_page(size);
548 if (addr + size < addr)
552 * Check for illegal addresses. Watch out for address wrap...
554 map = &td->td_proc->p_vmspace->vm_map;
555 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
560 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
563 * Inform hwpmc if the address range being unmapped contains
564 * an executable region.
566 pkm.pm_address = (uintptr_t) NULL;
567 if (vm_map_lookup_entry(map, addr, &entry)) {
568 for (; entry->start < addr + size;
569 entry = entry->next) {
570 if (vm_map_check_protection(map, entry->start,
571 entry->end, VM_PROT_EXECUTE) == TRUE) {
572 pkm.pm_address = (uintptr_t) addr;
573 pkm.pm_size = (size_t) size;
580 vm_map_delete(map, addr, addr + size);
583 if (__predict_false(pmc_handled)) {
584 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
585 vm_map_lock_downgrade(map);
586 if (pkm.pm_address != (uintptr_t) NULL)
587 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
588 vm_map_unlock_read(map);
593 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
597 #ifndef _SYS_SYSPROTO_H_
598 struct mprotect_args {
605 sys_mprotect(struct thread *td, struct mprotect_args *uap)
608 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
612 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
616 int vm_error, max_prot;
619 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
621 max_prot = PROT_MAX_EXTRACT(prot);
622 prot = PROT_EXTRACT(prot);
623 pageoff = (addr & PAGE_MASK);
626 size = (vm_size_t) round_page(size);
627 #ifdef COMPAT_FREEBSD32
628 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
629 if (((addr + size) & 0xffffffff) < addr)
633 if (addr + size < addr)
636 vm_error = KERN_SUCCESS;
638 if ((max_prot & prot) != prot)
640 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
641 addr, addr + size, max_prot, TRUE);
643 if (vm_error == KERN_SUCCESS)
644 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
645 addr, addr + size, prot, FALSE);
650 case KERN_PROTECTION_FAILURE:
652 case KERN_RESOURCE_SHORTAGE:
658 #ifndef _SYS_SYSPROTO_H_
659 struct minherit_args {
666 sys_minherit(struct thread *td, struct minherit_args *uap)
669 vm_size_t size, pageoff;
670 vm_inherit_t inherit;
672 addr = (vm_offset_t)uap->addr;
674 inherit = uap->inherit;
676 pageoff = (addr & PAGE_MASK);
679 size = (vm_size_t) round_page(size);
680 if (addr + size < addr)
683 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
684 addr + size, inherit)) {
687 case KERN_PROTECTION_FAILURE:
693 #ifndef _SYS_SYSPROTO_H_
694 struct madvise_args {
702 sys_madvise(struct thread *td, struct madvise_args *uap)
705 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
709 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
712 vm_offset_t addr, end, start;
716 * Check for our special case, advising the swap pager we are
719 if (behav == MADV_PROTECT) {
721 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
722 PROC_SPROTECT, &flags));
726 * Check for illegal addresses. Watch out for address wrap... Note
727 * that VM_*_ADDRESS are not constants due to casts (argh).
729 map = &td->td_proc->p_vmspace->vm_map;
731 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
733 if ((addr + len) < addr)
737 * Since this routine is only advisory, we default to conservative
740 start = trunc_page(addr);
741 end = round_page(addr + len);
744 * vm_map_madvise() checks for illegal values of behav.
746 return (vm_map_madvise(map, start, end, behav));
749 #ifndef _SYS_SYSPROTO_H_
750 struct mincore_args {
758 sys_mincore(struct thread *td, struct mincore_args *uap)
761 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
765 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
767 vm_offset_t addr, first_addr;
768 vm_offset_t end, cend;
772 int vecindex, lastvecindex;
773 vm_map_entry_t current;
774 vm_map_entry_t entry;
776 vm_paddr_t locked_pa;
780 unsigned int timestamp;
784 * Make sure that the addresses presented are valid for user
787 first_addr = addr = trunc_page(addr0);
788 end = addr + (vm_size_t)round_page(len);
789 map = &td->td_proc->p_vmspace->vm_map;
790 if (end > vm_map_max(map) || end < addr)
793 pmap = vmspace_pmap(td->td_proc->p_vmspace);
795 vm_map_lock_read(map);
797 timestamp = map->timestamp;
799 if (!vm_map_lookup_entry(map, addr, &entry)) {
800 vm_map_unlock_read(map);
805 * Do this on a map entry basis so that if the pages are not
806 * in the current processes address space, we can easily look
807 * up the pages elsewhere.
810 for (current = entry; current->start < end; current = current->next) {
813 * check for contiguity
815 if (current->end < end && current->next->start > current->end) {
816 vm_map_unlock_read(map);
821 * ignore submaps (for now) or null objects
823 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
824 current->object.vm_object == NULL)
828 * limit this scan to the current map entry and the
829 * limits for the mincore call
831 if (addr < current->start)
832 addr = current->start;
838 * scan this entry one page at a time
840 while (addr < cend) {
842 * Check pmap first, it is likely faster, also
843 * it can provide info as to whether we are the
844 * one referencing or modifying the page.
850 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
851 if (mincore_mapped) {
853 * We only care about this pmap's
854 * mapping of the page, if any.
856 if (locked_pa != 0) {
857 vm_page_unlock(PHYS_TO_VM_PAGE(
860 } else if (locked_pa != 0) {
862 * The page is mapped by this process but not
863 * both accessed and modified. It is also
864 * managed. Acquire the object lock so that
865 * other mappings might be examined.
867 m = PHYS_TO_VM_PAGE(locked_pa);
868 if (m->object != object) {
870 VM_OBJECT_WUNLOCK(object);
872 locked = VM_OBJECT_TRYWLOCK(object);
875 VM_OBJECT_WLOCK(object);
881 KASSERT(m->valid == VM_PAGE_BITS_ALL,
882 ("mincore: page %p is mapped but invalid",
884 } else if (mincoreinfo == 0) {
886 * The page is not mapped by this process. If
887 * the object implements managed pages, then
888 * determine if the page is resident so that
889 * the mappings might be examined.
891 if (current->object.vm_object != object) {
893 VM_OBJECT_WUNLOCK(object);
894 object = current->object.vm_object;
895 VM_OBJECT_WLOCK(object);
897 if (object->type == OBJT_DEFAULT ||
898 object->type == OBJT_SWAP ||
899 object->type == OBJT_VNODE) {
900 pindex = OFF_TO_IDX(current->offset +
901 (addr - current->start));
902 m = vm_page_lookup(object, pindex);
903 if (m != NULL && m->valid == 0)
906 mincoreinfo = MINCORE_INCORE;
910 /* Examine other mappings to the page. */
911 if (m->dirty == 0 && pmap_is_modified(m))
914 mincoreinfo |= MINCORE_MODIFIED_OTHER;
916 * The first test for PGA_REFERENCED is an
917 * optimization. The second test is
918 * required because a concurrent pmap
919 * operation could clear the last reference
920 * and set PGA_REFERENCED before the call to
921 * pmap_is_referenced().
923 if ((m->aflags & PGA_REFERENCED) != 0 ||
924 pmap_is_referenced(m) ||
925 (m->aflags & PGA_REFERENCED) != 0)
926 mincoreinfo |= MINCORE_REFERENCED_OTHER;
929 VM_OBJECT_WUNLOCK(object);
932 * subyte may page fault. In case it needs to modify
933 * the map, we release the lock.
935 vm_map_unlock_read(map);
938 * calculate index into user supplied byte vector
940 vecindex = atop(addr - first_addr);
943 * If we have skipped map entries, we need to make sure that
944 * the byte vector is zeroed for those skipped entries.
946 while ((lastvecindex + 1) < vecindex) {
948 error = subyte(vec + lastvecindex, 0);
956 * Pass the page information to the user
958 error = subyte(vec + vecindex, mincoreinfo);
965 * If the map has changed, due to the subyte, the previous
966 * output may be invalid.
968 vm_map_lock_read(map);
969 if (timestamp != map->timestamp)
972 lastvecindex = vecindex;
978 * subyte may page fault. In case it needs to modify
979 * the map, we release the lock.
981 vm_map_unlock_read(map);
984 * Zero the last entries in the byte vector.
986 vecindex = atop(end - first_addr);
987 while ((lastvecindex + 1) < vecindex) {
989 error = subyte(vec + lastvecindex, 0);
997 * If the map has changed, due to the subyte, the previous
998 * output may be invalid.
1000 vm_map_lock_read(map);
1001 if (timestamp != map->timestamp)
1003 vm_map_unlock_read(map);
1008 #ifndef _SYS_SYSPROTO_H_
1015 sys_mlock(struct thread *td, struct mlock_args *uap)
1018 return (kern_mlock(td->td_proc, td->td_ucred,
1019 __DECONST(uintptr_t, uap->addr), uap->len));
1023 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1025 vm_offset_t addr, end, last, start;
1026 vm_size_t npages, size;
1028 unsigned long nsize;
1031 error = priv_check_cred(cred, PRIV_VM_MLOCK);
1037 start = trunc_page(addr);
1038 end = round_page(last);
1039 if (last < addr || end < addr)
1041 npages = atop(end - start);
1042 if (npages > vm_page_max_user_wired)
1044 map = &proc->p_vmspace->vm_map;
1046 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1047 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1055 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1061 error = vm_map_wire(map, start, end,
1062 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1064 if (racct_enable && error != KERN_SUCCESS) {
1066 racct_set(proc, RACCT_MEMLOCK,
1067 ptoa(pmap_wired_count(map->pmap)));
1071 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1074 #ifndef _SYS_SYSPROTO_H_
1075 struct mlockall_args {
1081 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1086 map = &td->td_proc->p_vmspace->vm_map;
1087 error = priv_check(td, PRIV_VM_MLOCK);
1091 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1095 * If wiring all pages in the process would cause it to exceed
1096 * a hard resource limit, return ENOMEM.
1098 if (!old_mlock && uap->how & MCL_CURRENT) {
1099 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1104 PROC_LOCK(td->td_proc);
1105 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1106 PROC_UNLOCK(td->td_proc);
1112 if (uap->how & MCL_FUTURE) {
1114 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1119 if (uap->how & MCL_CURRENT) {
1121 * P1003.1-2001 mandates that all currently mapped pages
1122 * will be memory resident and locked (wired) upon return
1123 * from mlockall(). vm_map_wire() will wire pages, by
1124 * calling vm_fault_wire() for each page in the region.
1126 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1127 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1128 if (error == KERN_SUCCESS)
1130 else if (error == KERN_RESOURCE_SHORTAGE)
1136 if (racct_enable && error != KERN_SUCCESS) {
1137 PROC_LOCK(td->td_proc);
1138 racct_set(td->td_proc, RACCT_MEMLOCK,
1139 ptoa(pmap_wired_count(map->pmap)));
1140 PROC_UNLOCK(td->td_proc);
1147 #ifndef _SYS_SYSPROTO_H_
1148 struct munlockall_args {
1154 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1159 map = &td->td_proc->p_vmspace->vm_map;
1160 error = priv_check(td, PRIV_VM_MUNLOCK);
1164 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1166 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1169 /* Forcibly unwire all pages. */
1170 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1171 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1173 if (racct_enable && error == KERN_SUCCESS) {
1174 PROC_LOCK(td->td_proc);
1175 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1176 PROC_UNLOCK(td->td_proc);
1183 #ifndef _SYS_SYSPROTO_H_
1184 struct munlock_args {
1190 sys_munlock(struct thread *td, struct munlock_args *uap)
1193 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1197 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1199 vm_offset_t addr, end, last, start;
1205 error = priv_check(td, PRIV_VM_MUNLOCK);
1210 start = trunc_page(addr);
1211 end = round_page(last);
1212 if (last < addr || end < addr)
1214 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1215 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1217 if (racct_enable && error == KERN_SUCCESS) {
1218 PROC_LOCK(td->td_proc);
1219 map = &td->td_proc->p_vmspace->vm_map;
1220 racct_set(td->td_proc, RACCT_MEMLOCK,
1221 ptoa(pmap_wired_count(map->pmap)));
1222 PROC_UNLOCK(td->td_proc);
1225 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1231 * Helper function for vm_mmap. Perform sanity check specific for mmap
1232 * operations on vnodes.
1235 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1236 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1237 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1238 boolean_t *writecounted)
1247 cred = td->td_ucred;
1248 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1249 (*flagsp & MAP_SHARED) != 0;
1250 if ((error = vget(vp, LK_SHARED, td)) != 0)
1252 AUDIT_ARG_VNODE1(vp);
1256 if (vp->v_type == VREG) {
1258 * Get the proper underlying object
1264 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1266 vp = (struct vnode *)obj->handle;
1268 * Bypass filesystems obey the mpsafety of the
1269 * underlying fs. Tmpfs never bypasses.
1271 error = vget(vp, LK_SHARED, td);
1276 *writecounted = TRUE;
1277 vnode_pager_update_writecount(obj, 0, objsize);
1283 if ((error = VOP_GETATTR(vp, &va, cred)))
1286 /* This relies on VM_PROT_* matching PROT_*. */
1287 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1291 if ((flags & MAP_SHARED) != 0) {
1292 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1293 if (prot & VM_PROT_WRITE) {
1297 *maxprotp &= ~VM_PROT_WRITE;
1301 * If it is a regular file without any references
1302 * we do not need to sync it.
1303 * Adjust object size to be the size of actual file.
1305 objsize = round_page(va.va_size);
1306 if (va.va_nlink == 0)
1307 flags |= MAP_NOSYNC;
1308 if (obj->type == OBJT_VNODE) {
1309 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1316 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1317 ("wrong object type"));
1318 VM_OBJECT_WLOCK(obj);
1319 vm_object_reference_locked(obj);
1320 #if VM_NRESERVLEVEL > 0
1321 vm_object_color(obj, 0);
1323 VM_OBJECT_WUNLOCK(obj);
1328 vfs_mark_atime(vp, cred);
1331 if (error != 0 && *writecounted) {
1332 *writecounted = FALSE;
1333 vnode_pager_update_writecount(obj, objsize, 0);
1342 * Helper function for vm_mmap. Perform sanity check specific for mmap
1343 * operations on cdevs.
1346 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1347 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1348 vm_ooffset_t *foff, vm_object_t *objp)
1355 if (dsw->d_flags & D_MMAP_ANON) {
1358 *maxprotp = VM_PROT_ALL;
1359 *flagsp |= MAP_ANON;
1363 * cdevs do not provide private mappings of any kind.
1365 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1366 (prot & VM_PROT_WRITE) != 0)
1368 if (flags & (MAP_PRIVATE|MAP_COPY))
1371 * Force device mappings to be shared.
1373 flags |= MAP_SHARED;
1375 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1380 * First, try d_mmap_single(). If that is not implemented
1381 * (returns ENODEV), fall back to using the device pager.
1382 * Note that d_mmap_single() must return a reference to the
1383 * object (it needs to bump the reference count of the object
1384 * it returns somehow).
1386 * XXX assumes VM_PROT_* == PROT_*
1388 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1389 if (error != ENODEV)
1391 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1403 * Internal version of mmap used by exec, sys5 shared memory, and
1404 * various device drivers. Handle is either a vnode pointer, a
1405 * character device, or NULL for MAP_ANON.
1408 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1409 vm_prot_t maxprot, int flags,
1410 objtype_t handle_type, void *handle,
1414 struct thread *td = curthread;
1416 boolean_t writecounted;
1421 size = round_page(size);
1423 writecounted = FALSE;
1426 * Lookup/allocate object.
1428 switch (handle_type) {
1435 dsw = dev_refthread(cdev, &ref);
1438 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1439 dsw, &foff, &object);
1440 dev_relthread(cdev, ref);
1444 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1445 handle, &foff, &object, &writecounted);
1448 if (handle == NULL) {
1460 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1461 foff, writecounted, td);
1462 if (error != 0 && object != NULL) {
1464 * If this mapping was accounted for in the vnode's
1465 * writecount, then undo that now.
1468 vnode_pager_release_writecount(object, 0, size);
1469 vm_object_deallocate(object);
1475 * Internal version of mmap that maps a specific VM object into an
1476 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1479 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1480 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1481 boolean_t writecounted, struct thread *td)
1483 boolean_t curmap, fitit;
1484 vm_offset_t max_addr;
1485 int docow, error, findspace, rv;
1487 curmap = map == &td->td_proc->p_vmspace->vm_map;
1489 RACCT_PROC_LOCK(td->td_proc);
1490 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1491 RACCT_PROC_UNLOCK(td->td_proc);
1494 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1495 RACCT_PROC_UNLOCK(td->td_proc);
1498 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1499 if (ptoa(pmap_wired_count(map->pmap)) + size >
1500 lim_cur(td, RLIMIT_MEMLOCK)) {
1501 racct_set_force(td->td_proc, RACCT_VMEM,
1503 RACCT_PROC_UNLOCK(td->td_proc);
1506 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1507 ptoa(pmap_wired_count(map->pmap)) + size);
1509 racct_set_force(td->td_proc, RACCT_VMEM,
1511 RACCT_PROC_UNLOCK(td->td_proc);
1515 RACCT_PROC_UNLOCK(td->td_proc);
1519 * We currently can only deal with page aligned file offsets.
1520 * The mmap() system call already enforces this by subtracting
1521 * the page offset from the file offset, but checking here
1522 * catches errors in device drivers (e.g. d_single_mmap()
1523 * callbacks) and other internal mapping requests (such as in
1526 if (foff & PAGE_MASK)
1529 if ((flags & MAP_FIXED) == 0) {
1531 *addr = round_page(*addr);
1533 if (*addr != trunc_page(*addr))
1538 if (flags & MAP_ANON) {
1539 if (object != NULL || foff != 0)
1542 } else if (flags & MAP_PREFAULT_READ)
1543 docow = MAP_PREFAULT;
1545 docow = MAP_PREFAULT_PARTIAL;
1547 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1548 docow |= MAP_COPY_ON_WRITE;
1549 if (flags & MAP_NOSYNC)
1550 docow |= MAP_DISABLE_SYNCER;
1551 if (flags & MAP_NOCORE)
1552 docow |= MAP_DISABLE_COREDUMP;
1553 /* Shared memory is also shared with children. */
1554 if (flags & MAP_SHARED)
1555 docow |= MAP_INHERIT_SHARE;
1557 docow |= MAP_VN_WRITECOUNT;
1558 if (flags & MAP_STACK) {
1561 docow |= MAP_STACK_GROWS_DOWN;
1563 if ((flags & MAP_EXCL) != 0)
1564 docow |= MAP_CHECK_EXCL;
1565 if ((flags & MAP_GUARD) != 0)
1566 docow |= MAP_CREATE_GUARD;
1569 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1570 findspace = VMFS_SUPER_SPACE;
1571 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1572 findspace = VMFS_ALIGNED_SPACE(flags >>
1573 MAP_ALIGNMENT_SHIFT);
1575 findspace = VMFS_OPTIMAL_SPACE;
1578 if ((flags & MAP_32BIT) != 0)
1579 max_addr = MAP_32BIT_MAX_ADDR;
1582 rv = vm_map_find_min(map, object, foff, addr, size,
1583 round_page((vm_offset_t)td->td_proc->p_vmspace->
1584 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1585 findspace, prot, maxprot, docow);
1587 rv = vm_map_find(map, object, foff, addr, size,
1588 max_addr, findspace, prot, maxprot, docow);
1591 rv = vm_map_fixed(map, object, foff, *addr, size,
1592 prot, maxprot, docow);
1595 if (rv == KERN_SUCCESS) {
1597 * If the process has requested that all future mappings
1598 * be wired, then heed this.
1600 if ((map->flags & MAP_WIREFUTURE) != 0) {
1602 if ((map->flags & MAP_WIREFUTURE) != 0)
1603 (void)vm_map_wire_locked(map, *addr,
1604 *addr + size, VM_MAP_WIRE_USER |
1605 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1606 VM_MAP_WIRE_NOHOLES));
1610 return (vm_mmap_to_errno(rv));
1614 * Translate a Mach VM return code to zero on success or the appropriate errno
1618 vm_mmap_to_errno(int rv)
1624 case KERN_INVALID_ADDRESS:
1627 case KERN_PROTECTION_FAILURE: