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>
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
82 #include <vm/vm_param.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>
94 #include <sys/pmckern.h>
98 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
99 "Do not apply RLIMIT_MEMLOCK on mlockall");
100 static int mincore_mapped = 1;
101 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
102 "mincore reports mappings, not residency");
105 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
108 #ifndef _SYS_SYSPROTO_H_
115 sys_sbrk(struct thread *td, struct sbrk_args *uap)
117 /* Not yet implemented */
121 #ifndef _SYS_SYSPROTO_H_
128 sys_sstk(struct thread *td, struct sstk_args *uap)
130 /* Not yet implemented */
134 #if defined(COMPAT_43)
135 #ifndef _SYS_SYSPROTO_H_
136 struct getpagesize_args {
142 ogetpagesize(struct thread *td, struct getpagesize_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 size, int prot, int flags,
193 vm_prot_t cap_maxprot;
204 * Ignore old flags that used to be defined but did not do anything.
206 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
209 * Enforce the constraints.
210 * Mapping of length 0 is only allowed for old binaries.
211 * Anonymous mapping shall specify -1 as filedescriptor and
212 * zero position for new code. Be nice to ancient a.out
213 * binaries and correct pos for anonymous mapping, since old
214 * ld.so sometimes issues anonymous map requests with non-zero
217 if (!SV_CURPROC_FLAG(SV_AOUT)) {
218 if ((size == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
219 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
222 if ((flags & MAP_ANON) != 0)
226 if (flags & MAP_STACK) {
228 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
233 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
234 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
235 MAP_PREFAULT_READ | MAP_GUARD |
239 MAP_ALIGNMENT_MASK)) != 0)
241 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
243 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
245 if (prot != PROT_NONE &&
246 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
248 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
249 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
253 MAP_ALIGNMENT_MASK)) != 0))
257 * Align the file position to a page boundary,
258 * and save its page offset component.
260 pageoff = (pos & PAGE_MASK);
263 /* Adjust size for rounding (on both ends). */
264 size += pageoff; /* low end... */
265 size = (vm_size_t) round_page(size); /* hi end */
267 /* Ensure alignment is at least a page and fits in a pointer. */
268 align = flags & MAP_ALIGNMENT_MASK;
269 if (align != 0 && align != MAP_ALIGNED_SUPER &&
270 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
271 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
275 * Check for illegal addresses. Watch out for address wrap... Note
276 * that VM_*_ADDRESS are not constants due to casts (argh).
278 if (flags & MAP_FIXED) {
280 * The specified address must have the same remainder
281 * as the file offset taken modulo PAGE_SIZE, so it
282 * should be aligned after adjustment by pageoff.
285 if (addr & PAGE_MASK)
288 /* Address range must be all in user VM space. */
289 if (addr < vm_map_min(&vms->vm_map) ||
290 addr + size > vm_map_max(&vms->vm_map))
292 if (addr + size < addr)
295 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
297 } else if (flags & MAP_32BIT) {
299 * For MAP_32BIT, override the hint if it is too high and
300 * do not bother moving the mapping past the heap (since
301 * the heap is usually above 2GB).
303 if (addr + size > MAP_32BIT_MAX_ADDR)
308 * XXX for non-fixed mappings where no hint is provided or
309 * the hint would fall in the potential heap space,
310 * place it after the end of the largest possible heap.
312 * There should really be a pmap call to determine a reasonable
316 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
317 addr < round_page((vm_offset_t)vms->vm_daddr +
318 lim_max(td, RLIMIT_DATA))))
319 addr = round_page((vm_offset_t)vms->vm_daddr +
320 lim_max(td, RLIMIT_DATA));
324 * Return success without mapping anything for old
325 * binaries that request a page-aligned mapping of
326 * length 0. For modern binaries, this function
327 * returns an error earlier.
330 } else if ((flags & MAP_GUARD) != 0) {
331 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
332 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
333 } else if ((flags & MAP_ANON) != 0) {
335 * Mapping blank space is trivial.
337 * This relies on VM_PROT_* matching PROT_*.
339 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
340 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
343 * Mapping file, get fp for validation and don't let the
344 * descriptor disappear on us if we block. Check capability
345 * rights, but also return the maximum rights to be combined
346 * with maxprot later.
348 cap_rights_init(&rights, CAP_MMAP);
349 if (prot & PROT_READ)
350 cap_rights_set(&rights, CAP_MMAP_R);
351 if ((flags & MAP_SHARED) != 0) {
352 if (prot & PROT_WRITE)
353 cap_rights_set(&rights, CAP_MMAP_W);
355 if (prot & PROT_EXEC)
356 cap_rights_set(&rights, CAP_MMAP_X);
357 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
360 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
361 p->p_osrel >= P_OSREL_MAP_FSTRICT) {
366 /* This relies on VM_PROT_* matching PROT_*. */
367 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
368 cap_maxprot, flags, pos, td);
372 td->td_retval[0] = (register_t) (addr + pageoff);
380 #if defined(COMPAT_FREEBSD6)
382 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
385 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
386 uap->flags, uap->fd, uap->pos));
391 #ifndef _SYS_SYSPROTO_H_
402 ommap(struct thread *td, struct ommap_args *uap)
404 static const char cvtbsdprot[8] = {
408 PROT_EXEC | PROT_WRITE,
410 PROT_EXEC | PROT_READ,
411 PROT_WRITE | PROT_READ,
412 PROT_EXEC | PROT_WRITE | PROT_READ,
416 #define OMAP_ANON 0x0002
417 #define OMAP_COPY 0x0020
418 #define OMAP_SHARED 0x0010
419 #define OMAP_FIXED 0x0100
421 prot = cvtbsdprot[uap->prot & 0x7];
422 #ifdef COMPAT_FREEBSD32
423 #if defined(__amd64__)
424 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
430 if (uap->flags & OMAP_ANON)
432 if (uap->flags & OMAP_COPY)
434 if (uap->flags & OMAP_SHARED)
437 flags |= MAP_PRIVATE;
438 if (uap->flags & OMAP_FIXED)
440 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
443 #endif /* COMPAT_43 */
446 #ifndef _SYS_SYSPROTO_H_
454 sys_msync(struct thread *td, struct msync_args *uap)
457 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
461 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
469 pageoff = (addr & PAGE_MASK);
472 size = (vm_size_t) round_page(size);
473 if (addr + size < addr)
476 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
479 map = &td->td_proc->p_vmspace->vm_map;
482 * Clean the pages and interpret the return value.
484 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
485 (flags & MS_INVALIDATE) != 0);
489 case KERN_INVALID_ADDRESS:
491 case KERN_INVALID_ARGUMENT:
500 #ifndef _SYS_SYSPROTO_H_
507 sys_munmap(struct thread *td, struct munmap_args *uap)
510 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
514 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
517 struct pmckern_map_out pkm;
518 vm_map_entry_t entry;
529 pageoff = (addr & PAGE_MASK);
532 size = (vm_size_t) round_page(size);
533 if (addr + size < addr)
537 * Check for illegal addresses. Watch out for address wrap...
539 map = &td->td_proc->p_vmspace->vm_map;
540 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
545 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
548 * Inform hwpmc if the address range being unmapped contains
549 * an executable region.
551 pkm.pm_address = (uintptr_t) NULL;
552 if (vm_map_lookup_entry(map, addr, &entry)) {
553 for (; entry->start < addr + size;
554 entry = entry->next) {
555 if (vm_map_check_protection(map, entry->start,
556 entry->end, VM_PROT_EXECUTE) == TRUE) {
557 pkm.pm_address = (uintptr_t) addr;
558 pkm.pm_size = (size_t) size;
565 vm_map_delete(map, addr, addr + size);
568 if (__predict_false(pmc_handled)) {
569 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
570 vm_map_lock_downgrade(map);
571 if (pkm.pm_address != (uintptr_t) NULL)
572 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
573 vm_map_unlock_read(map);
578 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
582 #ifndef _SYS_SYSPROTO_H_
583 struct mprotect_args {
590 sys_mprotect(struct thread *td, struct mprotect_args *uap)
593 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
597 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
603 prot = (prot & VM_PROT_ALL);
604 pageoff = (addr & PAGE_MASK);
607 size = (vm_size_t) round_page(size);
608 #ifdef COMPAT_FREEBSD32
609 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
610 if (((addr + size) & 0xffffffff) < addr)
614 if (addr + size < addr)
617 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
618 addr + size, prot, FALSE)) {
621 case KERN_PROTECTION_FAILURE:
623 case KERN_RESOURCE_SHORTAGE:
629 #ifndef _SYS_SYSPROTO_H_
630 struct minherit_args {
637 sys_minherit(struct thread *td, struct minherit_args *uap)
640 vm_size_t size, pageoff;
641 vm_inherit_t inherit;
643 addr = (vm_offset_t)uap->addr;
645 inherit = uap->inherit;
647 pageoff = (addr & PAGE_MASK);
650 size = (vm_size_t) round_page(size);
651 if (addr + size < addr)
654 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
655 addr + size, inherit)) {
658 case KERN_PROTECTION_FAILURE:
664 #ifndef _SYS_SYSPROTO_H_
665 struct madvise_args {
673 sys_madvise(struct thread *td, struct madvise_args *uap)
676 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
680 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
683 vm_offset_t addr, end, start;
687 * Check for our special case, advising the swap pager we are
690 if (behav == MADV_PROTECT) {
692 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
693 PROC_SPROTECT, &flags));
697 * Check for illegal addresses. Watch out for address wrap... Note
698 * that VM_*_ADDRESS are not constants due to casts (argh).
700 map = &td->td_proc->p_vmspace->vm_map;
702 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
704 if ((addr + len) < addr)
708 * Since this routine is only advisory, we default to conservative
711 start = trunc_page(addr);
712 end = round_page(addr + len);
715 * vm_map_madvise() checks for illegal values of behav.
717 return (vm_map_madvise(map, start, end, behav));
720 #ifndef _SYS_SYSPROTO_H_
721 struct mincore_args {
729 sys_mincore(struct thread *td, struct mincore_args *uap)
732 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
736 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
738 vm_offset_t addr, first_addr;
739 vm_offset_t end, cend;
743 int vecindex, lastvecindex;
744 vm_map_entry_t current;
745 vm_map_entry_t entry;
747 vm_paddr_t locked_pa;
751 unsigned int timestamp;
755 * Make sure that the addresses presented are valid for user
758 first_addr = addr = trunc_page(addr0);
759 end = addr + (vm_size_t)round_page(len);
760 map = &td->td_proc->p_vmspace->vm_map;
761 if (end > vm_map_max(map) || end < addr)
764 pmap = vmspace_pmap(td->td_proc->p_vmspace);
766 vm_map_lock_read(map);
768 timestamp = map->timestamp;
770 if (!vm_map_lookup_entry(map, addr, &entry)) {
771 vm_map_unlock_read(map);
776 * Do this on a map entry basis so that if the pages are not
777 * in the current processes address space, we can easily look
778 * up the pages elsewhere.
781 for (current = entry; current->start < end; current = current->next) {
784 * check for contiguity
786 if (current->end < end && current->next->start > current->end) {
787 vm_map_unlock_read(map);
792 * ignore submaps (for now) or null objects
794 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
795 current->object.vm_object == NULL)
799 * limit this scan to the current map entry and the
800 * limits for the mincore call
802 if (addr < current->start)
803 addr = current->start;
809 * scan this entry one page at a time
811 while (addr < cend) {
813 * Check pmap first, it is likely faster, also
814 * it can provide info as to whether we are the
815 * one referencing or modifying the page.
821 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
822 if (mincore_mapped) {
824 * We only care about this pmap's
825 * mapping of the page, if any.
827 if (locked_pa != 0) {
828 vm_page_unlock(PHYS_TO_VM_PAGE(
831 } else if (locked_pa != 0) {
833 * The page is mapped by this process but not
834 * both accessed and modified. It is also
835 * managed. Acquire the object lock so that
836 * other mappings might be examined.
838 m = PHYS_TO_VM_PAGE(locked_pa);
839 if (m->object != object) {
841 VM_OBJECT_WUNLOCK(object);
843 locked = VM_OBJECT_TRYWLOCK(object);
846 VM_OBJECT_WLOCK(object);
852 KASSERT(m->valid == VM_PAGE_BITS_ALL,
853 ("mincore: page %p is mapped but invalid",
855 } else if (mincoreinfo == 0) {
857 * The page is not mapped by this process. If
858 * the object implements managed pages, then
859 * determine if the page is resident so that
860 * the mappings might be examined.
862 if (current->object.vm_object != object) {
864 VM_OBJECT_WUNLOCK(object);
865 object = current->object.vm_object;
866 VM_OBJECT_WLOCK(object);
868 if (object->type == OBJT_DEFAULT ||
869 object->type == OBJT_SWAP ||
870 object->type == OBJT_VNODE) {
871 pindex = OFF_TO_IDX(current->offset +
872 (addr - current->start));
873 m = vm_page_lookup(object, pindex);
874 if (m != NULL && m->valid == 0)
877 mincoreinfo = MINCORE_INCORE;
881 /* Examine other mappings to the page. */
882 if (m->dirty == 0 && pmap_is_modified(m))
885 mincoreinfo |= MINCORE_MODIFIED_OTHER;
887 * The first test for PGA_REFERENCED is an
888 * optimization. The second test is
889 * required because a concurrent pmap
890 * operation could clear the last reference
891 * and set PGA_REFERENCED before the call to
892 * pmap_is_referenced().
894 if ((m->aflags & PGA_REFERENCED) != 0 ||
895 pmap_is_referenced(m) ||
896 (m->aflags & PGA_REFERENCED) != 0)
897 mincoreinfo |= MINCORE_REFERENCED_OTHER;
900 VM_OBJECT_WUNLOCK(object);
903 * subyte may page fault. In case it needs to modify
904 * the map, we release the lock.
906 vm_map_unlock_read(map);
909 * calculate index into user supplied byte vector
911 vecindex = atop(addr - first_addr);
914 * If we have skipped map entries, we need to make sure that
915 * the byte vector is zeroed for those skipped entries.
917 while ((lastvecindex + 1) < vecindex) {
919 error = subyte(vec + lastvecindex, 0);
927 * Pass the page information to the user
929 error = subyte(vec + vecindex, mincoreinfo);
936 * If the map has changed, due to the subyte, the previous
937 * output may be invalid.
939 vm_map_lock_read(map);
940 if (timestamp != map->timestamp)
943 lastvecindex = vecindex;
949 * subyte may page fault. In case it needs to modify
950 * the map, we release the lock.
952 vm_map_unlock_read(map);
955 * Zero the last entries in the byte vector.
957 vecindex = atop(end - first_addr);
958 while ((lastvecindex + 1) < vecindex) {
960 error = subyte(vec + lastvecindex, 0);
968 * If the map has changed, due to the subyte, the previous
969 * output may be invalid.
971 vm_map_lock_read(map);
972 if (timestamp != map->timestamp)
974 vm_map_unlock_read(map);
979 #ifndef _SYS_SYSPROTO_H_
986 sys_mlock(struct thread *td, struct mlock_args *uap)
989 return (kern_mlock(td->td_proc, td->td_ucred,
990 __DECONST(uintptr_t, uap->addr), uap->len));
994 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
996 vm_offset_t addr, end, last, start;
997 vm_size_t npages, size;
1002 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1008 start = trunc_page(addr);
1009 end = round_page(last);
1010 if (last < addr || end < addr)
1012 npages = atop(end - start);
1013 if (npages > vm_page_max_wired)
1015 map = &proc->p_vmspace->vm_map;
1017 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1018 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1023 if (npages + vm_wire_count() > vm_page_max_wired)
1028 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1034 error = vm_map_wire(map, start, end,
1035 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1037 if (racct_enable && error != KERN_SUCCESS) {
1039 racct_set(proc, RACCT_MEMLOCK,
1040 ptoa(pmap_wired_count(map->pmap)));
1044 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1047 #ifndef _SYS_SYSPROTO_H_
1048 struct mlockall_args {
1054 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1059 map = &td->td_proc->p_vmspace->vm_map;
1060 error = priv_check(td, PRIV_VM_MLOCK);
1064 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1068 * If wiring all pages in the process would cause it to exceed
1069 * a hard resource limit, return ENOMEM.
1071 if (!old_mlock && uap->how & MCL_CURRENT) {
1072 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1077 PROC_LOCK(td->td_proc);
1078 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1079 PROC_UNLOCK(td->td_proc);
1085 if (uap->how & MCL_FUTURE) {
1087 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1092 if (uap->how & MCL_CURRENT) {
1094 * P1003.1-2001 mandates that all currently mapped pages
1095 * will be memory resident and locked (wired) upon return
1096 * from mlockall(). vm_map_wire() will wire pages, by
1097 * calling vm_fault_wire() for each page in the region.
1099 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1100 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1101 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1104 if (racct_enable && error != KERN_SUCCESS) {
1105 PROC_LOCK(td->td_proc);
1106 racct_set(td->td_proc, RACCT_MEMLOCK,
1107 ptoa(pmap_wired_count(map->pmap)));
1108 PROC_UNLOCK(td->td_proc);
1115 #ifndef _SYS_SYSPROTO_H_
1116 struct munlockall_args {
1122 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1127 map = &td->td_proc->p_vmspace->vm_map;
1128 error = priv_check(td, PRIV_VM_MUNLOCK);
1132 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1134 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1137 /* Forcibly unwire all pages. */
1138 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1139 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1141 if (racct_enable && error == KERN_SUCCESS) {
1142 PROC_LOCK(td->td_proc);
1143 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1144 PROC_UNLOCK(td->td_proc);
1151 #ifndef _SYS_SYSPROTO_H_
1152 struct munlock_args {
1158 sys_munlock(struct thread *td, struct munlock_args *uap)
1161 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1165 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1167 vm_offset_t addr, end, last, start;
1173 error = priv_check(td, PRIV_VM_MUNLOCK);
1178 start = trunc_page(addr);
1179 end = round_page(last);
1180 if (last < addr || end < addr)
1182 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1183 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1185 if (racct_enable && error == KERN_SUCCESS) {
1186 PROC_LOCK(td->td_proc);
1187 map = &td->td_proc->p_vmspace->vm_map;
1188 racct_set(td->td_proc, RACCT_MEMLOCK,
1189 ptoa(pmap_wired_count(map->pmap)));
1190 PROC_UNLOCK(td->td_proc);
1193 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1199 * Helper function for vm_mmap. Perform sanity check specific for mmap
1200 * operations on vnodes.
1203 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1204 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1205 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1206 boolean_t *writecounted)
1215 cred = td->td_ucred;
1216 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1217 (*flagsp & MAP_SHARED) != 0;
1218 if ((error = vget(vp, LK_SHARED, td)) != 0)
1220 AUDIT_ARG_VNODE1(vp);
1224 if (vp->v_type == VREG) {
1226 * Get the proper underlying object
1232 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1234 vp = (struct vnode *)obj->handle;
1236 * Bypass filesystems obey the mpsafety of the
1237 * underlying fs. Tmpfs never bypasses.
1239 error = vget(vp, LK_SHARED, td);
1244 *writecounted = TRUE;
1245 vnode_pager_update_writecount(obj, 0, objsize);
1251 if ((error = VOP_GETATTR(vp, &va, cred)))
1254 /* This relies on VM_PROT_* matching PROT_*. */
1255 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1259 if ((flags & MAP_SHARED) != 0) {
1260 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1261 if (prot & VM_PROT_WRITE) {
1265 *maxprotp &= ~VM_PROT_WRITE;
1269 * If it is a regular file without any references
1270 * we do not need to sync it.
1271 * Adjust object size to be the size of actual file.
1273 objsize = round_page(va.va_size);
1274 if (va.va_nlink == 0)
1275 flags |= MAP_NOSYNC;
1276 if (obj->type == OBJT_VNODE) {
1277 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1284 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1285 ("wrong object type"));
1286 VM_OBJECT_WLOCK(obj);
1287 vm_object_reference_locked(obj);
1288 #if VM_NRESERVLEVEL > 0
1289 vm_object_color(obj, 0);
1291 VM_OBJECT_WUNLOCK(obj);
1296 vfs_mark_atime(vp, cred);
1299 if (error != 0 && *writecounted) {
1300 *writecounted = FALSE;
1301 vnode_pager_update_writecount(obj, objsize, 0);
1310 * Helper function for vm_mmap. Perform sanity check specific for mmap
1311 * operations on cdevs.
1314 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1315 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1316 vm_ooffset_t *foff, vm_object_t *objp)
1323 if (dsw->d_flags & D_MMAP_ANON) {
1326 *maxprotp = VM_PROT_ALL;
1327 *flagsp |= MAP_ANON;
1331 * cdevs do not provide private mappings of any kind.
1333 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1334 (prot & VM_PROT_WRITE) != 0)
1336 if (flags & (MAP_PRIVATE|MAP_COPY))
1339 * Force device mappings to be shared.
1341 flags |= MAP_SHARED;
1343 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1348 * First, try d_mmap_single(). If that is not implemented
1349 * (returns ENODEV), fall back to using the device pager.
1350 * Note that d_mmap_single() must return a reference to the
1351 * object (it needs to bump the reference count of the object
1352 * it returns somehow).
1354 * XXX assumes VM_PROT_* == PROT_*
1356 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1357 if (error != ENODEV)
1359 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1371 * Internal version of mmap used by exec, sys5 shared memory, and
1372 * various device drivers. Handle is either a vnode pointer, a
1373 * character device, or NULL for MAP_ANON.
1376 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1377 vm_prot_t maxprot, int flags,
1378 objtype_t handle_type, void *handle,
1382 struct thread *td = curthread;
1384 boolean_t writecounted;
1389 size = round_page(size);
1391 writecounted = FALSE;
1394 * Lookup/allocate object.
1396 switch (handle_type) {
1403 dsw = dev_refthread(cdev, &ref);
1406 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1407 dsw, &foff, &object);
1408 dev_relthread(cdev, ref);
1412 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1413 handle, &foff, &object, &writecounted);
1416 if (handle == NULL) {
1428 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1429 foff, writecounted, td);
1430 if (error != 0 && object != NULL) {
1432 * If this mapping was accounted for in the vnode's
1433 * writecount, then undo that now.
1436 vnode_pager_release_writecount(object, 0, size);
1437 vm_object_deallocate(object);
1443 * Internal version of mmap that maps a specific VM object into an
1444 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1447 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1448 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1449 boolean_t writecounted, struct thread *td)
1451 boolean_t curmap, fitit;
1452 vm_offset_t max_addr;
1453 int docow, error, findspace, rv;
1455 curmap = map == &td->td_proc->p_vmspace->vm_map;
1457 RACCT_PROC_LOCK(td->td_proc);
1458 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1459 RACCT_PROC_UNLOCK(td->td_proc);
1462 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1463 RACCT_PROC_UNLOCK(td->td_proc);
1466 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1467 if (ptoa(pmap_wired_count(map->pmap)) + size >
1468 lim_cur(td, RLIMIT_MEMLOCK)) {
1469 racct_set_force(td->td_proc, RACCT_VMEM,
1471 RACCT_PROC_UNLOCK(td->td_proc);
1474 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1475 ptoa(pmap_wired_count(map->pmap)) + size);
1477 racct_set_force(td->td_proc, RACCT_VMEM,
1479 RACCT_PROC_UNLOCK(td->td_proc);
1483 RACCT_PROC_UNLOCK(td->td_proc);
1487 * We currently can only deal with page aligned file offsets.
1488 * The mmap() system call already enforces this by subtracting
1489 * the page offset from the file offset, but checking here
1490 * catches errors in device drivers (e.g. d_single_mmap()
1491 * callbacks) and other internal mapping requests (such as in
1494 if (foff & PAGE_MASK)
1497 if ((flags & MAP_FIXED) == 0) {
1499 *addr = round_page(*addr);
1501 if (*addr != trunc_page(*addr))
1506 if (flags & MAP_ANON) {
1507 if (object != NULL || foff != 0)
1510 } else if (flags & MAP_PREFAULT_READ)
1511 docow = MAP_PREFAULT;
1513 docow = MAP_PREFAULT_PARTIAL;
1515 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1516 docow |= MAP_COPY_ON_WRITE;
1517 if (flags & MAP_NOSYNC)
1518 docow |= MAP_DISABLE_SYNCER;
1519 if (flags & MAP_NOCORE)
1520 docow |= MAP_DISABLE_COREDUMP;
1521 /* Shared memory is also shared with children. */
1522 if (flags & MAP_SHARED)
1523 docow |= MAP_INHERIT_SHARE;
1525 docow |= MAP_VN_WRITECOUNT;
1526 if (flags & MAP_STACK) {
1529 docow |= MAP_STACK_GROWS_DOWN;
1531 if ((flags & MAP_EXCL) != 0)
1532 docow |= MAP_CHECK_EXCL;
1533 if ((flags & MAP_GUARD) != 0)
1534 docow |= MAP_CREATE_GUARD;
1537 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1538 findspace = VMFS_SUPER_SPACE;
1539 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1540 findspace = VMFS_ALIGNED_SPACE(flags >>
1541 MAP_ALIGNMENT_SHIFT);
1543 findspace = VMFS_OPTIMAL_SPACE;
1546 if ((flags & MAP_32BIT) != 0)
1547 max_addr = MAP_32BIT_MAX_ADDR;
1550 rv = vm_map_find_min(map, object, foff, addr, size,
1551 round_page((vm_offset_t)td->td_proc->p_vmspace->
1552 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1553 findspace, prot, maxprot, docow);
1555 rv = vm_map_find(map, object, foff, addr, size,
1556 max_addr, findspace, prot, maxprot, docow);
1559 rv = vm_map_fixed(map, object, foff, *addr, size,
1560 prot, maxprot, docow);
1563 if (rv == KERN_SUCCESS) {
1565 * If the process has requested that all future mappings
1566 * be wired, then heed this.
1568 if (map->flags & MAP_WIREFUTURE) {
1569 vm_map_wire(map, *addr, *addr + size,
1570 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1571 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1574 return (vm_mmap_to_errno(rv));
1578 * Translate a Mach VM return code to zero on success or the appropriate errno
1582 vm_mmap_to_errno(int rv)
1588 case KERN_INVALID_ADDRESS:
1591 case KERN_PROTECTION_FAILURE: