2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
40 * Mapped file (mmap) interface to VM
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/capsicum.h>
53 #include <sys/kernel.h>
55 #include <sys/mutex.h>
56 #include <sys/sysproto.h>
57 #include <sys/filedesc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
70 #include <sys/mount.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_param.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
93 #include <sys/pmckern.h>
97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
98 "Do not apply RLIMIT_MEMLOCK on mlockall");
101 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
104 #ifndef _SYS_SYSPROTO_H_
111 sys_sbrk(struct thread *td, struct sbrk_args *uap)
113 /* Not yet implemented */
117 #ifndef _SYS_SYSPROTO_H_
124 sys_sstk(struct thread *td, struct sstk_args *uap)
126 /* Not yet implemented */
130 #if defined(COMPAT_43)
131 #ifndef _SYS_SYSPROTO_H_
132 struct getpagesize_args {
138 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
141 td->td_retval[0] = PAGE_SIZE;
144 #endif /* COMPAT_43 */
148 * Memory Map (mmap) system call. Note that the file offset
149 * and address are allowed to be NOT page aligned, though if
150 * the MAP_FIXED flag it set, both must have the same remainder
151 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
152 * page-aligned, the actual mapping starts at trunc_page(addr)
153 * and the return value is adjusted up by the page offset.
155 * Generally speaking, only character devices which are themselves
156 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
157 * there would be no cache coherency between a descriptor and a VM mapping
158 * both to the same character device.
160 #ifndef _SYS_SYSPROTO_H_
173 sys_mmap(struct thread *td, struct mmap_args *uap)
176 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
177 uap->flags, uap->fd, uap->pos));
181 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
188 vm_prot_t cap_maxprot;
192 vms = td->td_proc->p_vmspace;
198 * Ignore old flags that used to be defined but did not do anything.
200 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
203 * Enforce the constraints.
204 * Mapping of length 0 is only allowed for old binaries.
205 * Anonymous mapping shall specify -1 as filedescriptor and
206 * zero position for new code. Be nice to ancient a.out
207 * binaries and correct pos for anonymous mapping, since old
208 * ld.so sometimes issues anonymous map requests with non-zero
211 if (!SV_CURPROC_FLAG(SV_AOUT)) {
212 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
213 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
216 if ((flags & MAP_ANON) != 0)
220 if (flags & MAP_STACK) {
222 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
227 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
228 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
233 MAP_ALIGNMENT_MASK)) != 0)
235 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
237 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
239 if (prot != PROT_NONE &&
240 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
244 * Align the file position to a page boundary,
245 * and save its page offset component.
247 pageoff = (pos & PAGE_MASK);
250 /* Adjust size for rounding (on both ends). */
251 size += pageoff; /* low end... */
252 size = (vm_size_t) round_page(size); /* hi end */
254 /* Ensure alignment is at least a page and fits in a pointer. */
255 align = flags & MAP_ALIGNMENT_MASK;
256 if (align != 0 && align != MAP_ALIGNED_SUPER &&
257 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
258 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
262 * Check for illegal addresses. Watch out for address wrap... Note
263 * that VM_*_ADDRESS are not constants due to casts (argh).
265 if (flags & MAP_FIXED) {
267 * The specified address must have the same remainder
268 * as the file offset taken modulo PAGE_SIZE, so it
269 * should be aligned after adjustment by pageoff.
272 if (addr & PAGE_MASK)
275 /* Address range must be all in user VM space. */
276 if (addr < vm_map_min(&vms->vm_map) ||
277 addr + size > vm_map_max(&vms->vm_map))
279 if (addr + size < addr)
282 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
284 } else if (flags & MAP_32BIT) {
286 * For MAP_32BIT, override the hint if it is too high and
287 * do not bother moving the mapping past the heap (since
288 * the heap is usually above 2GB).
290 if (addr + size > MAP_32BIT_MAX_ADDR)
295 * XXX for non-fixed mappings where no hint is provided or
296 * the hint would fall in the potential heap space,
297 * place it after the end of the largest possible heap.
299 * There should really be a pmap call to determine a reasonable
303 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
304 addr < round_page((vm_offset_t)vms->vm_daddr +
305 lim_max(td, RLIMIT_DATA))))
306 addr = round_page((vm_offset_t)vms->vm_daddr +
307 lim_max(td, RLIMIT_DATA));
311 * Return success without mapping anything for old
312 * binaries that request a page-aligned mapping of
313 * length 0. For modern binaries, this function
314 * returns an error earlier.
317 } else if (flags & MAP_ANON) {
319 * Mapping blank space is trivial.
321 * This relies on VM_PROT_* matching PROT_*.
323 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
324 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
327 * Mapping file, get fp for validation and don't let the
328 * descriptor disappear on us if we block. Check capability
329 * rights, but also return the maximum rights to be combined
330 * with maxprot later.
332 cap_rights_init(&rights, CAP_MMAP);
333 if (prot & PROT_READ)
334 cap_rights_set(&rights, CAP_MMAP_R);
335 if ((flags & MAP_SHARED) != 0) {
336 if (prot & PROT_WRITE)
337 cap_rights_set(&rights, CAP_MMAP_W);
339 if (prot & PROT_EXEC)
340 cap_rights_set(&rights, CAP_MMAP_X);
341 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
344 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
345 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
350 /* This relies on VM_PROT_* matching PROT_*. */
351 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
352 cap_maxprot, flags, pos, td);
356 td->td_retval[0] = (register_t) (addr + pageoff);
364 #if defined(COMPAT_FREEBSD6)
366 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
369 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
370 uap->flags, uap->fd, uap->pos));
375 #ifndef _SYS_SYSPROTO_H_
386 ommap(struct thread *td, struct ommap_args *uap)
388 static const char cvtbsdprot[8] = {
392 PROT_EXEC | PROT_WRITE,
394 PROT_EXEC | PROT_READ,
395 PROT_WRITE | PROT_READ,
396 PROT_EXEC | PROT_WRITE | PROT_READ,
400 #define OMAP_ANON 0x0002
401 #define OMAP_COPY 0x0020
402 #define OMAP_SHARED 0x0010
403 #define OMAP_FIXED 0x0100
405 prot = cvtbsdprot[uap->prot & 0x7];
406 #ifdef COMPAT_FREEBSD32
407 #if defined(__amd64__)
408 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
414 if (uap->flags & OMAP_ANON)
416 if (uap->flags & OMAP_COPY)
418 if (uap->flags & OMAP_SHARED)
421 flags |= MAP_PRIVATE;
422 if (uap->flags & OMAP_FIXED)
424 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
427 #endif /* COMPAT_43 */
430 #ifndef _SYS_SYSPROTO_H_
438 sys_msync(struct thread *td, struct msync_args *uap)
441 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
445 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
453 pageoff = (addr & PAGE_MASK);
456 size = (vm_size_t) round_page(size);
457 if (addr + size < addr)
460 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
463 map = &td->td_proc->p_vmspace->vm_map;
466 * Clean the pages and interpret the return value.
468 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
469 (flags & MS_INVALIDATE) != 0);
473 case KERN_INVALID_ADDRESS:
475 case KERN_INVALID_ARGUMENT:
484 #ifndef _SYS_SYSPROTO_H_
491 sys_munmap(struct thread *td, struct munmap_args *uap)
494 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
498 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
501 struct pmckern_map_out pkm;
502 vm_map_entry_t entry;
513 pageoff = (addr & PAGE_MASK);
516 size = (vm_size_t) round_page(size);
517 if (addr + size < addr)
521 * Check for illegal addresses. Watch out for address wrap...
523 map = &td->td_proc->p_vmspace->vm_map;
524 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
529 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
532 * Inform hwpmc if the address range being unmapped contains
533 * an executable region.
535 pkm.pm_address = (uintptr_t) NULL;
536 if (vm_map_lookup_entry(map, addr, &entry)) {
538 entry != &map->header && entry->start < addr + size;
539 entry = entry->next) {
540 if (vm_map_check_protection(map, entry->start,
541 entry->end, VM_PROT_EXECUTE) == TRUE) {
542 pkm.pm_address = (uintptr_t) addr;
543 pkm.pm_size = (size_t) size;
550 vm_map_delete(map, addr, addr + size);
553 if (__predict_false(pmc_handled)) {
554 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
555 vm_map_lock_downgrade(map);
556 if (pkm.pm_address != (uintptr_t) NULL)
557 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
558 vm_map_unlock_read(map);
563 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
567 #ifndef _SYS_SYSPROTO_H_
568 struct mprotect_args {
575 sys_mprotect(struct thread *td, struct mprotect_args *uap)
578 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
582 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
588 prot = (prot & VM_PROT_ALL);
589 pageoff = (addr & PAGE_MASK);
592 size = (vm_size_t) round_page(size);
593 if (addr + size < addr)
596 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
597 addr + size, prot, FALSE)) {
600 case KERN_PROTECTION_FAILURE:
602 case KERN_RESOURCE_SHORTAGE:
608 #ifndef _SYS_SYSPROTO_H_
609 struct minherit_args {
616 sys_minherit(struct thread *td, struct minherit_args *uap)
619 vm_size_t size, pageoff;
620 vm_inherit_t inherit;
622 addr = (vm_offset_t)uap->addr;
624 inherit = uap->inherit;
626 pageoff = (addr & PAGE_MASK);
629 size = (vm_size_t) round_page(size);
630 if (addr + size < addr)
633 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
634 addr + size, inherit)) {
637 case KERN_PROTECTION_FAILURE:
643 #ifndef _SYS_SYSPROTO_H_
644 struct madvise_args {
652 sys_madvise(struct thread *td, struct madvise_args *uap)
655 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
659 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
662 vm_offset_t addr, end, start;
666 * Check for our special case, advising the swap pager we are
669 if (behav == MADV_PROTECT) {
671 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
672 PROC_SPROTECT, &flags));
676 * Check for illegal behavior
678 if (behav < 0 || behav > MADV_CORE)
681 * Check for illegal addresses. Watch out for address wrap... Note
682 * that VM_*_ADDRESS are not constants due to casts (argh).
684 map = &td->td_proc->p_vmspace->vm_map;
686 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
688 if ((addr + len) < addr)
692 * Since this routine is only advisory, we default to conservative
695 start = trunc_page(addr);
696 end = round_page(addr + len);
698 if (vm_map_madvise(map, start, end, behav))
703 #ifndef _SYS_SYSPROTO_H_
704 struct mincore_args {
712 sys_mincore(struct thread *td, struct mincore_args *uap)
715 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
719 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
721 vm_offset_t addr, first_addr;
722 vm_offset_t end, cend;
726 int vecindex, lastvecindex;
727 vm_map_entry_t current;
728 vm_map_entry_t entry;
730 vm_paddr_t locked_pa;
734 unsigned int timestamp;
738 * Make sure that the addresses presented are valid for user
741 first_addr = addr = trunc_page(addr0);
742 end = addr + (vm_size_t)round_page(len);
743 map = &td->td_proc->p_vmspace->vm_map;
744 if (end > vm_map_max(map) || end < addr)
747 pmap = vmspace_pmap(td->td_proc->p_vmspace);
749 vm_map_lock_read(map);
751 timestamp = map->timestamp;
753 if (!vm_map_lookup_entry(map, addr, &entry)) {
754 vm_map_unlock_read(map);
759 * Do this on a map entry basis so that if the pages are not
760 * in the current processes address space, we can easily look
761 * up the pages elsewhere.
764 for (current = entry;
765 (current != &map->header) && (current->start < end);
766 current = current->next) {
769 * check for contiguity
771 if (current->end < end &&
772 (entry->next == &map->header ||
773 current->next->start > current->end)) {
774 vm_map_unlock_read(map);
779 * ignore submaps (for now) or null objects
781 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
782 current->object.vm_object == NULL)
786 * limit this scan to the current map entry and the
787 * limits for the mincore call
789 if (addr < current->start)
790 addr = current->start;
796 * scan this entry one page at a time
798 while (addr < cend) {
800 * Check pmap first, it is likely faster, also
801 * it can provide info as to whether we are the
802 * one referencing or modifying the page.
808 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
809 if (locked_pa != 0) {
811 * The page is mapped by this process but not
812 * both accessed and modified. It is also
813 * managed. Acquire the object lock so that
814 * other mappings might be examined.
816 m = PHYS_TO_VM_PAGE(locked_pa);
817 if (m->object != object) {
819 VM_OBJECT_WUNLOCK(object);
821 locked = VM_OBJECT_TRYWLOCK(object);
824 VM_OBJECT_WLOCK(object);
830 KASSERT(m->valid == VM_PAGE_BITS_ALL,
831 ("mincore: page %p is mapped but invalid",
833 } else if (mincoreinfo == 0) {
835 * The page is not mapped by this process. If
836 * the object implements managed pages, then
837 * determine if the page is resident so that
838 * the mappings might be examined.
840 if (current->object.vm_object != object) {
842 VM_OBJECT_WUNLOCK(object);
843 object = current->object.vm_object;
844 VM_OBJECT_WLOCK(object);
846 if (object->type == OBJT_DEFAULT ||
847 object->type == OBJT_SWAP ||
848 object->type == OBJT_VNODE) {
849 pindex = OFF_TO_IDX(current->offset +
850 (addr - current->start));
851 m = vm_page_lookup(object, pindex);
852 if (m != NULL && m->valid == 0)
855 mincoreinfo = MINCORE_INCORE;
859 /* Examine other mappings to the page. */
860 if (m->dirty == 0 && pmap_is_modified(m))
863 mincoreinfo |= MINCORE_MODIFIED_OTHER;
865 * The first test for PGA_REFERENCED is an
866 * optimization. The second test is
867 * required because a concurrent pmap
868 * operation could clear the last reference
869 * and set PGA_REFERENCED before the call to
870 * pmap_is_referenced().
872 if ((m->aflags & PGA_REFERENCED) != 0 ||
873 pmap_is_referenced(m) ||
874 (m->aflags & PGA_REFERENCED) != 0)
875 mincoreinfo |= MINCORE_REFERENCED_OTHER;
878 VM_OBJECT_WUNLOCK(object);
881 * subyte may page fault. In case it needs to modify
882 * the map, we release the lock.
884 vm_map_unlock_read(map);
887 * calculate index into user supplied byte vector
889 vecindex = atop(addr - first_addr);
892 * If we have skipped map entries, we need to make sure that
893 * the byte vector is zeroed for those skipped entries.
895 while ((lastvecindex + 1) < vecindex) {
897 error = subyte(vec + lastvecindex, 0);
905 * Pass the page information to the user
907 error = subyte(vec + vecindex, mincoreinfo);
914 * If the map has changed, due to the subyte, the previous
915 * output may be invalid.
917 vm_map_lock_read(map);
918 if (timestamp != map->timestamp)
921 lastvecindex = vecindex;
927 * subyte may page fault. In case it needs to modify
928 * the map, we release the lock.
930 vm_map_unlock_read(map);
933 * Zero the last entries in the byte vector.
935 vecindex = atop(end - first_addr);
936 while ((lastvecindex + 1) < vecindex) {
938 error = subyte(vec + lastvecindex, 0);
946 * If the map has changed, due to the subyte, the previous
947 * output may be invalid.
949 vm_map_lock_read(map);
950 if (timestamp != map->timestamp)
952 vm_map_unlock_read(map);
957 #ifndef _SYS_SYSPROTO_H_
964 sys_mlock(struct thread *td, struct mlock_args *uap)
967 return (kern_mlock(td->td_proc, td->td_ucred,
968 __DECONST(uintptr_t, uap->addr), uap->len));
972 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
974 vm_offset_t addr, end, last, start;
975 vm_size_t npages, size;
980 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
986 start = trunc_page(addr);
987 end = round_page(last);
988 if (last < addr || end < addr)
990 npages = atop(end - start);
991 if (npages > vm_page_max_wired)
993 map = &proc->p_vmspace->vm_map;
995 nsize = ptoa(npages + pmap_wired_count(map->pmap));
996 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1001 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1006 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1012 error = vm_map_wire(map, start, end,
1013 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1015 if (racct_enable && error != KERN_SUCCESS) {
1017 racct_set(proc, RACCT_MEMLOCK,
1018 ptoa(pmap_wired_count(map->pmap)));
1022 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1025 #ifndef _SYS_SYSPROTO_H_
1026 struct mlockall_args {
1032 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1037 map = &td->td_proc->p_vmspace->vm_map;
1038 error = priv_check(td, PRIV_VM_MLOCK);
1042 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1046 * If wiring all pages in the process would cause it to exceed
1047 * a hard resource limit, return ENOMEM.
1049 if (!old_mlock && uap->how & MCL_CURRENT) {
1050 PROC_LOCK(td->td_proc);
1051 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1052 PROC_UNLOCK(td->td_proc);
1055 PROC_UNLOCK(td->td_proc);
1059 PROC_LOCK(td->td_proc);
1060 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1061 PROC_UNLOCK(td->td_proc);
1067 if (uap->how & MCL_FUTURE) {
1069 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1074 if (uap->how & MCL_CURRENT) {
1076 * P1003.1-2001 mandates that all currently mapped pages
1077 * will be memory resident and locked (wired) upon return
1078 * from mlockall(). vm_map_wire() will wire pages, by
1079 * calling vm_fault_wire() for each page in the region.
1081 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1082 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1083 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1086 if (racct_enable && error != KERN_SUCCESS) {
1087 PROC_LOCK(td->td_proc);
1088 racct_set(td->td_proc, RACCT_MEMLOCK,
1089 ptoa(pmap_wired_count(map->pmap)));
1090 PROC_UNLOCK(td->td_proc);
1097 #ifndef _SYS_SYSPROTO_H_
1098 struct munlockall_args {
1104 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1109 map = &td->td_proc->p_vmspace->vm_map;
1110 error = priv_check(td, PRIV_VM_MUNLOCK);
1114 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1116 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1119 /* Forcibly unwire all pages. */
1120 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1121 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1123 if (racct_enable && error == KERN_SUCCESS) {
1124 PROC_LOCK(td->td_proc);
1125 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1126 PROC_UNLOCK(td->td_proc);
1133 #ifndef _SYS_SYSPROTO_H_
1134 struct munlock_args {
1140 sys_munlock(struct thread *td, struct munlock_args *uap)
1143 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1147 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1149 vm_offset_t addr, end, last, start;
1155 error = priv_check(td, PRIV_VM_MUNLOCK);
1160 start = trunc_page(addr);
1161 end = round_page(last);
1162 if (last < addr || end < addr)
1164 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1165 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1167 if (racct_enable && error == KERN_SUCCESS) {
1168 PROC_LOCK(td->td_proc);
1169 map = &td->td_proc->p_vmspace->vm_map;
1170 racct_set(td->td_proc, RACCT_MEMLOCK,
1171 ptoa(pmap_wired_count(map->pmap)));
1172 PROC_UNLOCK(td->td_proc);
1175 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1181 * Helper function for vm_mmap. Perform sanity check specific for mmap
1182 * operations on vnodes.
1185 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1186 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1187 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1188 boolean_t *writecounted)
1194 int error, flags, locktype;
1196 cred = td->td_ucred;
1197 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1198 locktype = LK_EXCLUSIVE;
1200 locktype = LK_SHARED;
1201 if ((error = vget(vp, locktype, td)) != 0)
1203 AUDIT_ARG_VNODE1(vp);
1207 if (vp->v_type == VREG) {
1209 * Get the proper underlying object
1215 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1217 vp = (struct vnode *)obj->handle;
1219 * Bypass filesystems obey the mpsafety of the
1220 * underlying fs. Tmpfs never bypasses.
1222 error = vget(vp, locktype, td);
1226 if (locktype == LK_EXCLUSIVE) {
1227 *writecounted = TRUE;
1228 vnode_pager_update_writecount(obj, 0, objsize);
1234 if ((error = VOP_GETATTR(vp, &va, cred)))
1237 /* This relies on VM_PROT_* matching PROT_*. */
1238 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1242 if ((flags & MAP_SHARED) != 0) {
1243 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1244 if (prot & VM_PROT_WRITE) {
1248 *maxprotp &= ~VM_PROT_WRITE;
1252 * If it is a regular file without any references
1253 * we do not need to sync it.
1254 * Adjust object size to be the size of actual file.
1256 objsize = round_page(va.va_size);
1257 if (va.va_nlink == 0)
1258 flags |= MAP_NOSYNC;
1259 if (obj->type == OBJT_VNODE) {
1260 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1267 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1268 ("wrong object type"));
1269 VM_OBJECT_WLOCK(obj);
1270 vm_object_reference_locked(obj);
1271 #if VM_NRESERVLEVEL > 0
1272 vm_object_color(obj, 0);
1274 VM_OBJECT_WUNLOCK(obj);
1279 vfs_mark_atime(vp, cred);
1282 if (error != 0 && *writecounted) {
1283 *writecounted = FALSE;
1284 vnode_pager_update_writecount(obj, objsize, 0);
1293 * Helper function for vm_mmap. Perform sanity check specific for mmap
1294 * operations on cdevs.
1297 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1298 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1299 vm_ooffset_t *foff, vm_object_t *objp)
1306 if (dsw->d_flags & D_MMAP_ANON) {
1309 *maxprotp = VM_PROT_ALL;
1310 *flagsp |= MAP_ANON;
1314 * cdevs do not provide private mappings of any kind.
1316 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1317 (prot & VM_PROT_WRITE) != 0)
1319 if (flags & (MAP_PRIVATE|MAP_COPY))
1322 * Force device mappings to be shared.
1324 flags |= MAP_SHARED;
1326 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1331 * First, try d_mmap_single(). If that is not implemented
1332 * (returns ENODEV), fall back to using the device pager.
1333 * Note that d_mmap_single() must return a reference to the
1334 * object (it needs to bump the reference count of the object
1335 * it returns somehow).
1337 * XXX assumes VM_PROT_* == PROT_*
1339 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1340 if (error != ENODEV)
1342 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1354 * Internal version of mmap used by exec, sys5 shared memory, and
1355 * various device drivers. Handle is either a vnode pointer, a
1356 * character device, or NULL for MAP_ANON.
1359 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1360 vm_prot_t maxprot, int flags,
1361 objtype_t handle_type, void *handle,
1365 struct thread *td = curthread;
1367 boolean_t writecounted;
1372 size = round_page(size);
1374 writecounted = FALSE;
1377 * Lookup/allocate object.
1379 switch (handle_type) {
1386 dsw = dev_refthread(cdev, &ref);
1389 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1390 dsw, &foff, &object);
1391 dev_relthread(cdev, ref);
1395 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1396 handle, &foff, &object, &writecounted);
1399 if (handle == NULL) {
1411 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1412 foff, writecounted, td);
1413 if (error != 0 && object != NULL) {
1415 * If this mapping was accounted for in the vnode's
1416 * writecount, then undo that now.
1419 vnode_pager_release_writecount(object, 0, size);
1420 vm_object_deallocate(object);
1426 * Internal version of mmap that maps a specific VM object into an
1427 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1430 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1431 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1432 boolean_t writecounted, struct thread *td)
1435 int docow, error, findspace, rv;
1437 if (map == &td->td_proc->p_vmspace->vm_map) {
1438 PROC_LOCK(td->td_proc);
1439 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1440 PROC_UNLOCK(td->td_proc);
1443 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1444 PROC_UNLOCK(td->td_proc);
1447 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1448 if (ptoa(pmap_wired_count(map->pmap)) + size >
1449 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1450 racct_set_force(td->td_proc, RACCT_VMEM,
1452 PROC_UNLOCK(td->td_proc);
1455 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1456 ptoa(pmap_wired_count(map->pmap)) + size);
1458 racct_set_force(td->td_proc, RACCT_VMEM,
1460 PROC_UNLOCK(td->td_proc);
1464 PROC_UNLOCK(td->td_proc);
1468 * We currently can only deal with page aligned file offsets.
1469 * The mmap() system call already enforces this by subtracting
1470 * the page offset from the file offset, but checking here
1471 * catches errors in device drivers (e.g. d_single_mmap()
1472 * callbacks) and other internal mapping requests (such as in
1475 if (foff & PAGE_MASK)
1478 if ((flags & MAP_FIXED) == 0) {
1480 *addr = round_page(*addr);
1482 if (*addr != trunc_page(*addr))
1487 if (flags & MAP_ANON) {
1488 if (object != NULL || foff != 0)
1491 } else if (flags & MAP_PREFAULT_READ)
1492 docow = MAP_PREFAULT;
1494 docow = MAP_PREFAULT_PARTIAL;
1496 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1497 docow |= MAP_COPY_ON_WRITE;
1498 if (flags & MAP_NOSYNC)
1499 docow |= MAP_DISABLE_SYNCER;
1500 if (flags & MAP_NOCORE)
1501 docow |= MAP_DISABLE_COREDUMP;
1502 /* Shared memory is also shared with children. */
1503 if (flags & MAP_SHARED)
1504 docow |= MAP_INHERIT_SHARE;
1506 docow |= MAP_VN_WRITECOUNT;
1507 if (flags & MAP_STACK) {
1510 docow |= MAP_STACK_GROWS_DOWN;
1512 if ((flags & MAP_EXCL) != 0)
1513 docow |= MAP_CHECK_EXCL;
1516 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1517 findspace = VMFS_SUPER_SPACE;
1518 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1519 findspace = VMFS_ALIGNED_SPACE(flags >>
1520 MAP_ALIGNMENT_SHIFT);
1522 findspace = VMFS_OPTIMAL_SPACE;
1523 rv = vm_map_find(map, object, foff, addr, size,
1525 flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1527 0, findspace, prot, maxprot, docow);
1529 rv = vm_map_fixed(map, object, foff, *addr, size,
1530 prot, maxprot, docow);
1533 if (rv == KERN_SUCCESS) {
1535 * If the process has requested that all future mappings
1536 * be wired, then heed this.
1538 if (map->flags & MAP_WIREFUTURE) {
1539 vm_map_wire(map, *addr, *addr + size,
1540 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1541 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1544 return (vm_mmap_to_errno(rv));
1548 * Translate a Mach VM return code to zero on success or the appropriate errno
1552 vm_mmap_to_errno(int rv)
1558 case KERN_INVALID_ADDRESS:
1561 case KERN_PROTECTION_FAILURE: