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_compat.h"
49 #include "opt_hwpmc_hooks.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/capsicum.h>
55 #include <sys/kernel.h>
57 #include <sys/mutex.h>
58 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
72 #include <sys/mount.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
83 #include <vm/vm_param.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_pageout.h>
90 #include <vm/vm_extern.h>
91 #include <vm/vm_page.h>
92 #include <vm/vnode_pager.h>
95 #include <sys/pmckern.h>
99 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
100 "Do not apply RLIMIT_MEMLOCK on mlockall");
103 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
106 #ifndef _SYS_SYSPROTO_H_
113 sys_sbrk(struct thread *td, struct sbrk_args *uap)
115 /* Not yet implemented */
119 #ifndef _SYS_SYSPROTO_H_
126 sys_sstk(struct thread *td, struct sstk_args *uap)
128 /* Not yet implemented */
132 #if defined(COMPAT_43)
133 #ifndef _SYS_SYSPROTO_H_
134 struct getpagesize_args {
140 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
143 td->td_retval[0] = PAGE_SIZE;
146 #endif /* COMPAT_43 */
150 * Memory Map (mmap) system call. Note that the file offset
151 * and address are allowed to be NOT page aligned, though if
152 * the MAP_FIXED flag it set, both must have the same remainder
153 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
154 * page-aligned, the actual mapping starts at trunc_page(addr)
155 * and the return value is adjusted up by the page offset.
157 * Generally speaking, only character devices which are themselves
158 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
159 * there would be no cache coherency between a descriptor and a VM mapping
160 * both to the same character device.
162 #ifndef _SYS_SYSPROTO_H_
175 sys_mmap(struct thread *td, struct mmap_args *uap)
178 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
179 uap->flags, uap->fd, uap->pos));
183 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
190 vm_prot_t cap_maxprot;
194 vms = td->td_proc->p_vmspace;
200 * Ignore old flags that used to be defined but did not do anything.
202 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
205 * Enforce the constraints.
206 * Mapping of length 0 is only allowed for old binaries.
207 * Anonymous mapping shall specify -1 as filedescriptor and
208 * zero position for new code. Be nice to ancient a.out
209 * binaries and correct pos for anonymous mapping, since old
210 * ld.so sometimes issues anonymous map requests with non-zero
213 if (!SV_CURPROC_FLAG(SV_AOUT)) {
214 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
215 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
218 if ((flags & MAP_ANON) != 0)
222 if (flags & MAP_STACK) {
224 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
229 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
230 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
231 MAP_PREFAULT_READ | MAP_GUARD |
235 MAP_ALIGNMENT_MASK)) != 0)
237 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
239 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
241 if (prot != PROT_NONE &&
242 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
244 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
245 pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
246 MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
250 * Align the file position to a page boundary,
251 * and save its page offset component.
253 pageoff = (pos & PAGE_MASK);
256 /* Adjust size for rounding (on both ends). */
257 size += pageoff; /* low end... */
258 size = (vm_size_t) round_page(size); /* hi end */
260 /* Ensure alignment is at least a page and fits in a pointer. */
261 align = flags & MAP_ALIGNMENT_MASK;
262 if (align != 0 && align != MAP_ALIGNED_SUPER &&
263 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
264 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
268 * Check for illegal addresses. Watch out for address wrap... Note
269 * that VM_*_ADDRESS are not constants due to casts (argh).
271 if (flags & MAP_FIXED) {
273 * The specified address must have the same remainder
274 * as the file offset taken modulo PAGE_SIZE, so it
275 * should be aligned after adjustment by pageoff.
278 if (addr & PAGE_MASK)
281 /* Address range must be all in user VM space. */
282 if (addr < vm_map_min(&vms->vm_map) ||
283 addr + size > vm_map_max(&vms->vm_map))
285 if (addr + size < addr)
288 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
290 } else if (flags & MAP_32BIT) {
292 * For MAP_32BIT, override the hint if it is too high and
293 * do not bother moving the mapping past the heap (since
294 * the heap is usually above 2GB).
296 if (addr + size > MAP_32BIT_MAX_ADDR)
301 * XXX for non-fixed mappings where no hint is provided or
302 * the hint would fall in the potential heap space,
303 * place it after the end of the largest possible heap.
305 * There should really be a pmap call to determine a reasonable
309 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
310 addr < round_page((vm_offset_t)vms->vm_daddr +
311 lim_max(td, RLIMIT_DATA))))
312 addr = round_page((vm_offset_t)vms->vm_daddr +
313 lim_max(td, RLIMIT_DATA));
317 * Return success without mapping anything for old
318 * binaries that request a page-aligned mapping of
319 * length 0. For modern binaries, this function
320 * returns an error earlier.
323 } else if ((flags & MAP_GUARD) != 0) {
324 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
325 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
326 } else if ((flags & MAP_ANON) != 0) {
328 * Mapping blank space is trivial.
330 * This relies on VM_PROT_* matching PROT_*.
332 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
333 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
336 * Mapping file, get fp for validation and don't let the
337 * descriptor disappear on us if we block. Check capability
338 * rights, but also return the maximum rights to be combined
339 * with maxprot later.
341 cap_rights_init(&rights, CAP_MMAP);
342 if (prot & PROT_READ)
343 cap_rights_set(&rights, CAP_MMAP_R);
344 if ((flags & MAP_SHARED) != 0) {
345 if (prot & PROT_WRITE)
346 cap_rights_set(&rights, CAP_MMAP_W);
348 if (prot & PROT_EXEC)
349 cap_rights_set(&rights, CAP_MMAP_X);
350 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
353 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
354 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
359 /* This relies on VM_PROT_* matching PROT_*. */
360 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
361 cap_maxprot, flags, pos, td);
365 td->td_retval[0] = (register_t) (addr + pageoff);
373 #if defined(COMPAT_FREEBSD6)
375 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
378 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
379 uap->flags, uap->fd, uap->pos));
384 #ifndef _SYS_SYSPROTO_H_
395 ommap(struct thread *td, struct ommap_args *uap)
397 static const char cvtbsdprot[8] = {
401 PROT_EXEC | PROT_WRITE,
403 PROT_EXEC | PROT_READ,
404 PROT_WRITE | PROT_READ,
405 PROT_EXEC | PROT_WRITE | PROT_READ,
409 #define OMAP_ANON 0x0002
410 #define OMAP_COPY 0x0020
411 #define OMAP_SHARED 0x0010
412 #define OMAP_FIXED 0x0100
414 prot = cvtbsdprot[uap->prot & 0x7];
415 #ifdef COMPAT_FREEBSD32
416 #if defined(__amd64__)
417 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
423 if (uap->flags & OMAP_ANON)
425 if (uap->flags & OMAP_COPY)
427 if (uap->flags & OMAP_SHARED)
430 flags |= MAP_PRIVATE;
431 if (uap->flags & OMAP_FIXED)
433 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
436 #endif /* COMPAT_43 */
439 #ifndef _SYS_SYSPROTO_H_
447 sys_msync(struct thread *td, struct msync_args *uap)
450 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
454 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
462 pageoff = (addr & PAGE_MASK);
465 size = (vm_size_t) round_page(size);
466 if (addr + size < addr)
469 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
472 map = &td->td_proc->p_vmspace->vm_map;
475 * Clean the pages and interpret the return value.
477 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
478 (flags & MS_INVALIDATE) != 0);
482 case KERN_INVALID_ADDRESS:
484 case KERN_INVALID_ARGUMENT:
493 #ifndef _SYS_SYSPROTO_H_
500 sys_munmap(struct thread *td, struct munmap_args *uap)
503 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
507 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
510 struct pmckern_map_out pkm;
511 vm_map_entry_t entry;
522 pageoff = (addr & PAGE_MASK);
525 size = (vm_size_t) round_page(size);
526 if (addr + size < addr)
530 * Check for illegal addresses. Watch out for address wrap...
532 map = &td->td_proc->p_vmspace->vm_map;
533 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
538 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
541 * Inform hwpmc if the address range being unmapped contains
542 * an executable region.
544 pkm.pm_address = (uintptr_t) NULL;
545 if (vm_map_lookup_entry(map, addr, &entry)) {
547 entry != &map->header && entry->start < addr + size;
548 entry = entry->next) {
549 if (vm_map_check_protection(map, entry->start,
550 entry->end, VM_PROT_EXECUTE) == TRUE) {
551 pkm.pm_address = (uintptr_t) addr;
552 pkm.pm_size = (size_t) size;
559 vm_map_delete(map, addr, addr + size);
562 if (__predict_false(pmc_handled)) {
563 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
564 vm_map_lock_downgrade(map);
565 if (pkm.pm_address != (uintptr_t) NULL)
566 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
567 vm_map_unlock_read(map);
572 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
576 #ifndef _SYS_SYSPROTO_H_
577 struct mprotect_args {
584 sys_mprotect(struct thread *td, struct mprotect_args *uap)
587 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
591 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
597 prot = (prot & VM_PROT_ALL);
598 pageoff = (addr & PAGE_MASK);
601 size = (vm_size_t) round_page(size);
602 if (addr + size < addr)
605 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
606 addr + size, prot, FALSE)) {
609 case KERN_PROTECTION_FAILURE:
611 case KERN_RESOURCE_SHORTAGE:
617 #ifndef _SYS_SYSPROTO_H_
618 struct minherit_args {
625 sys_minherit(struct thread *td, struct minherit_args *uap)
628 vm_size_t size, pageoff;
629 vm_inherit_t inherit;
631 addr = (vm_offset_t)uap->addr;
633 inherit = uap->inherit;
635 pageoff = (addr & PAGE_MASK);
638 size = (vm_size_t) round_page(size);
639 if (addr + size < addr)
642 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
643 addr + size, inherit)) {
646 case KERN_PROTECTION_FAILURE:
652 #ifndef _SYS_SYSPROTO_H_
653 struct madvise_args {
661 sys_madvise(struct thread *td, struct madvise_args *uap)
664 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
668 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
671 vm_offset_t addr, end, start;
675 * Check for our special case, advising the swap pager we are
678 if (behav == MADV_PROTECT) {
680 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
681 PROC_SPROTECT, &flags));
685 * Check for illegal behavior
687 if (behav < 0 || behav > MADV_CORE)
690 * Check for illegal addresses. Watch out for address wrap... Note
691 * that VM_*_ADDRESS are not constants due to casts (argh).
693 map = &td->td_proc->p_vmspace->vm_map;
695 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
697 if ((addr + len) < addr)
701 * Since this routine is only advisory, we default to conservative
704 start = trunc_page(addr);
705 end = round_page(addr + len);
707 if (vm_map_madvise(map, start, end, behav))
712 #ifndef _SYS_SYSPROTO_H_
713 struct mincore_args {
721 sys_mincore(struct thread *td, struct mincore_args *uap)
724 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
728 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
730 vm_offset_t addr, first_addr;
731 vm_offset_t end, cend;
735 int vecindex, lastvecindex;
736 vm_map_entry_t current;
737 vm_map_entry_t entry;
739 vm_paddr_t locked_pa;
743 unsigned int timestamp;
747 * Make sure that the addresses presented are valid for user
750 first_addr = addr = trunc_page(addr0);
751 end = addr + (vm_size_t)round_page(len);
752 map = &td->td_proc->p_vmspace->vm_map;
753 if (end > vm_map_max(map) || end < addr)
756 pmap = vmspace_pmap(td->td_proc->p_vmspace);
758 vm_map_lock_read(map);
760 timestamp = map->timestamp;
762 if (!vm_map_lookup_entry(map, addr, &entry)) {
763 vm_map_unlock_read(map);
768 * Do this on a map entry basis so that if the pages are not
769 * in the current processes address space, we can easily look
770 * up the pages elsewhere.
773 for (current = entry;
774 (current != &map->header) && (current->start < end);
775 current = current->next) {
778 * check for contiguity
780 if (current->end < end &&
781 (entry->next == &map->header ||
782 current->next->start > current->end)) {
783 vm_map_unlock_read(map);
788 * ignore submaps (for now) or null objects
790 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
791 current->object.vm_object == NULL)
795 * limit this scan to the current map entry and the
796 * limits for the mincore call
798 if (addr < current->start)
799 addr = current->start;
805 * scan this entry one page at a time
807 while (addr < cend) {
809 * Check pmap first, it is likely faster, also
810 * it can provide info as to whether we are the
811 * one referencing or modifying the page.
817 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
818 if (locked_pa != 0) {
820 * The page is mapped by this process but not
821 * both accessed and modified. It is also
822 * managed. Acquire the object lock so that
823 * other mappings might be examined.
825 m = PHYS_TO_VM_PAGE(locked_pa);
826 if (m->object != object) {
828 VM_OBJECT_WUNLOCK(object);
830 locked = VM_OBJECT_TRYWLOCK(object);
833 VM_OBJECT_WLOCK(object);
839 KASSERT(m->valid == VM_PAGE_BITS_ALL,
840 ("mincore: page %p is mapped but invalid",
842 } else if (mincoreinfo == 0) {
844 * The page is not mapped by this process. If
845 * the object implements managed pages, then
846 * determine if the page is resident so that
847 * the mappings might be examined.
849 if (current->object.vm_object != object) {
851 VM_OBJECT_WUNLOCK(object);
852 object = current->object.vm_object;
853 VM_OBJECT_WLOCK(object);
855 if (object->type == OBJT_DEFAULT ||
856 object->type == OBJT_SWAP ||
857 object->type == OBJT_VNODE) {
858 pindex = OFF_TO_IDX(current->offset +
859 (addr - current->start));
860 m = vm_page_lookup(object, pindex);
861 if (m != NULL && m->valid == 0)
864 mincoreinfo = MINCORE_INCORE;
868 /* Examine other mappings to the page. */
869 if (m->dirty == 0 && pmap_is_modified(m))
872 mincoreinfo |= MINCORE_MODIFIED_OTHER;
874 * The first test for PGA_REFERENCED is an
875 * optimization. The second test is
876 * required because a concurrent pmap
877 * operation could clear the last reference
878 * and set PGA_REFERENCED before the call to
879 * pmap_is_referenced().
881 if ((m->aflags & PGA_REFERENCED) != 0 ||
882 pmap_is_referenced(m) ||
883 (m->aflags & PGA_REFERENCED) != 0)
884 mincoreinfo |= MINCORE_REFERENCED_OTHER;
887 VM_OBJECT_WUNLOCK(object);
890 * subyte may page fault. In case it needs to modify
891 * the map, we release the lock.
893 vm_map_unlock_read(map);
896 * calculate index into user supplied byte vector
898 vecindex = atop(addr - first_addr);
901 * If we have skipped map entries, we need to make sure that
902 * the byte vector is zeroed for those skipped entries.
904 while ((lastvecindex + 1) < vecindex) {
906 error = subyte(vec + lastvecindex, 0);
914 * Pass the page information to the user
916 error = subyte(vec + vecindex, mincoreinfo);
923 * If the map has changed, due to the subyte, the previous
924 * output may be invalid.
926 vm_map_lock_read(map);
927 if (timestamp != map->timestamp)
930 lastvecindex = vecindex;
936 * subyte may page fault. In case it needs to modify
937 * the map, we release the lock.
939 vm_map_unlock_read(map);
942 * Zero the last entries in the byte vector.
944 vecindex = atop(end - first_addr);
945 while ((lastvecindex + 1) < vecindex) {
947 error = subyte(vec + lastvecindex, 0);
955 * If the map has changed, due to the subyte, the previous
956 * output may be invalid.
958 vm_map_lock_read(map);
959 if (timestamp != map->timestamp)
961 vm_map_unlock_read(map);
966 #ifndef _SYS_SYSPROTO_H_
973 sys_mlock(struct thread *td, struct mlock_args *uap)
976 return (kern_mlock(td->td_proc, td->td_ucred,
977 __DECONST(uintptr_t, uap->addr), uap->len));
981 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
983 vm_offset_t addr, end, last, start;
984 vm_size_t npages, size;
989 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
995 start = trunc_page(addr);
996 end = round_page(last);
997 if (last < addr || end < addr)
999 npages = atop(end - start);
1000 if (npages > vm_page_max_wired)
1002 map = &proc->p_vmspace->vm_map;
1004 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1005 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1010 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1015 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1021 error = vm_map_wire(map, start, end,
1022 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1024 if (racct_enable && error != KERN_SUCCESS) {
1026 racct_set(proc, RACCT_MEMLOCK,
1027 ptoa(pmap_wired_count(map->pmap)));
1031 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1034 #ifndef _SYS_SYSPROTO_H_
1035 struct mlockall_args {
1041 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1046 map = &td->td_proc->p_vmspace->vm_map;
1047 error = priv_check(td, PRIV_VM_MLOCK);
1051 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1055 * If wiring all pages in the process would cause it to exceed
1056 * a hard resource limit, return ENOMEM.
1058 if (!old_mlock && uap->how & MCL_CURRENT) {
1059 PROC_LOCK(td->td_proc);
1060 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1061 PROC_UNLOCK(td->td_proc);
1064 PROC_UNLOCK(td->td_proc);
1068 PROC_LOCK(td->td_proc);
1069 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1070 PROC_UNLOCK(td->td_proc);
1076 if (uap->how & MCL_FUTURE) {
1078 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1083 if (uap->how & MCL_CURRENT) {
1085 * P1003.1-2001 mandates that all currently mapped pages
1086 * will be memory resident and locked (wired) upon return
1087 * from mlockall(). vm_map_wire() will wire pages, by
1088 * calling vm_fault_wire() for each page in the region.
1090 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1091 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1092 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1095 if (racct_enable && error != KERN_SUCCESS) {
1096 PROC_LOCK(td->td_proc);
1097 racct_set(td->td_proc, RACCT_MEMLOCK,
1098 ptoa(pmap_wired_count(map->pmap)));
1099 PROC_UNLOCK(td->td_proc);
1106 #ifndef _SYS_SYSPROTO_H_
1107 struct munlockall_args {
1113 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1118 map = &td->td_proc->p_vmspace->vm_map;
1119 error = priv_check(td, PRIV_VM_MUNLOCK);
1123 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1125 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1128 /* Forcibly unwire all pages. */
1129 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1130 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1132 if (racct_enable && error == KERN_SUCCESS) {
1133 PROC_LOCK(td->td_proc);
1134 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1135 PROC_UNLOCK(td->td_proc);
1142 #ifndef _SYS_SYSPROTO_H_
1143 struct munlock_args {
1149 sys_munlock(struct thread *td, struct munlock_args *uap)
1152 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1156 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1158 vm_offset_t addr, end, last, start;
1164 error = priv_check(td, PRIV_VM_MUNLOCK);
1169 start = trunc_page(addr);
1170 end = round_page(last);
1171 if (last < addr || end < addr)
1173 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1174 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1176 if (racct_enable && error == KERN_SUCCESS) {
1177 PROC_LOCK(td->td_proc);
1178 map = &td->td_proc->p_vmspace->vm_map;
1179 racct_set(td->td_proc, RACCT_MEMLOCK,
1180 ptoa(pmap_wired_count(map->pmap)));
1181 PROC_UNLOCK(td->td_proc);
1184 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1190 * Helper function for vm_mmap. Perform sanity check specific for mmap
1191 * operations on vnodes.
1194 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1195 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1196 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1197 boolean_t *writecounted)
1203 int error, flags, locktype;
1205 cred = td->td_ucred;
1206 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1207 locktype = LK_EXCLUSIVE;
1209 locktype = LK_SHARED;
1210 if ((error = vget(vp, locktype, td)) != 0)
1212 AUDIT_ARG_VNODE1(vp);
1216 if (vp->v_type == VREG) {
1218 * Get the proper underlying object
1224 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1226 vp = (struct vnode *)obj->handle;
1228 * Bypass filesystems obey the mpsafety of the
1229 * underlying fs. Tmpfs never bypasses.
1231 error = vget(vp, locktype, td);
1235 if (locktype == LK_EXCLUSIVE) {
1236 *writecounted = TRUE;
1237 vnode_pager_update_writecount(obj, 0, objsize);
1243 if ((error = VOP_GETATTR(vp, &va, cred)))
1246 /* This relies on VM_PROT_* matching PROT_*. */
1247 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1251 if ((flags & MAP_SHARED) != 0) {
1252 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1253 if (prot & VM_PROT_WRITE) {
1257 *maxprotp &= ~VM_PROT_WRITE;
1261 * If it is a regular file without any references
1262 * we do not need to sync it.
1263 * Adjust object size to be the size of actual file.
1265 objsize = round_page(va.va_size);
1266 if (va.va_nlink == 0)
1267 flags |= MAP_NOSYNC;
1268 if (obj->type == OBJT_VNODE) {
1269 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1276 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1277 ("wrong object type"));
1278 VM_OBJECT_WLOCK(obj);
1279 vm_object_reference_locked(obj);
1280 #if VM_NRESERVLEVEL > 0
1281 vm_object_color(obj, 0);
1283 VM_OBJECT_WUNLOCK(obj);
1288 vfs_mark_atime(vp, cred);
1291 if (error != 0 && *writecounted) {
1292 *writecounted = FALSE;
1293 vnode_pager_update_writecount(obj, objsize, 0);
1302 * Helper function for vm_mmap. Perform sanity check specific for mmap
1303 * operations on cdevs.
1306 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1307 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1308 vm_ooffset_t *foff, vm_object_t *objp)
1315 if (dsw->d_flags & D_MMAP_ANON) {
1318 *maxprotp = VM_PROT_ALL;
1319 *flagsp |= MAP_ANON;
1323 * cdevs do not provide private mappings of any kind.
1325 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1326 (prot & VM_PROT_WRITE) != 0)
1328 if (flags & (MAP_PRIVATE|MAP_COPY))
1331 * Force device mappings to be shared.
1333 flags |= MAP_SHARED;
1335 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1340 * First, try d_mmap_single(). If that is not implemented
1341 * (returns ENODEV), fall back to using the device pager.
1342 * Note that d_mmap_single() must return a reference to the
1343 * object (it needs to bump the reference count of the object
1344 * it returns somehow).
1346 * XXX assumes VM_PROT_* == PROT_*
1348 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1349 if (error != ENODEV)
1351 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1363 * Internal version of mmap used by exec, sys5 shared memory, and
1364 * various device drivers. Handle is either a vnode pointer, a
1365 * character device, or NULL for MAP_ANON.
1368 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1369 vm_prot_t maxprot, int flags,
1370 objtype_t handle_type, void *handle,
1374 struct thread *td = curthread;
1376 boolean_t writecounted;
1381 size = round_page(size);
1383 writecounted = FALSE;
1386 * Lookup/allocate object.
1388 switch (handle_type) {
1395 dsw = dev_refthread(cdev, &ref);
1398 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1399 dsw, &foff, &object);
1400 dev_relthread(cdev, ref);
1404 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1405 handle, &foff, &object, &writecounted);
1408 if (handle == NULL) {
1420 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1421 foff, writecounted, td);
1422 if (error != 0 && object != NULL) {
1424 * If this mapping was accounted for in the vnode's
1425 * writecount, then undo that now.
1428 vnode_pager_release_writecount(object, 0, size);
1429 vm_object_deallocate(object);
1435 * Internal version of mmap that maps a specific VM object into an
1436 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1439 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1440 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1441 boolean_t writecounted, struct thread *td)
1443 boolean_t curmap, fitit;
1444 vm_offset_t max_addr;
1445 int docow, error, findspace, rv;
1447 curmap = map == &td->td_proc->p_vmspace->vm_map;
1449 PROC_LOCK(td->td_proc);
1450 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1451 PROC_UNLOCK(td->td_proc);
1454 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1455 PROC_UNLOCK(td->td_proc);
1458 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1459 if (ptoa(pmap_wired_count(map->pmap)) + size >
1460 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1461 racct_set_force(td->td_proc, RACCT_VMEM,
1463 PROC_UNLOCK(td->td_proc);
1466 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1467 ptoa(pmap_wired_count(map->pmap)) + size);
1469 racct_set_force(td->td_proc, RACCT_VMEM,
1471 PROC_UNLOCK(td->td_proc);
1475 PROC_UNLOCK(td->td_proc);
1479 * We currently can only deal with page aligned file offsets.
1480 * The mmap() system call already enforces this by subtracting
1481 * the page offset from the file offset, but checking here
1482 * catches errors in device drivers (e.g. d_single_mmap()
1483 * callbacks) and other internal mapping requests (such as in
1486 if (foff & PAGE_MASK)
1489 if ((flags & MAP_FIXED) == 0) {
1491 *addr = round_page(*addr);
1493 if (*addr != trunc_page(*addr))
1498 if (flags & MAP_ANON) {
1499 if (object != NULL || foff != 0)
1502 } else if (flags & MAP_PREFAULT_READ)
1503 docow = MAP_PREFAULT;
1505 docow = MAP_PREFAULT_PARTIAL;
1507 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1508 docow |= MAP_COPY_ON_WRITE;
1509 if (flags & MAP_NOSYNC)
1510 docow |= MAP_DISABLE_SYNCER;
1511 if (flags & MAP_NOCORE)
1512 docow |= MAP_DISABLE_COREDUMP;
1513 /* Shared memory is also shared with children. */
1514 if (flags & MAP_SHARED)
1515 docow |= MAP_INHERIT_SHARE;
1517 docow |= MAP_VN_WRITECOUNT;
1518 if (flags & MAP_STACK) {
1521 docow |= MAP_STACK_GROWS_DOWN;
1523 if ((flags & MAP_EXCL) != 0)
1524 docow |= MAP_CHECK_EXCL;
1525 if ((flags & MAP_GUARD) != 0)
1526 docow |= MAP_CREATE_GUARD;
1529 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1530 findspace = VMFS_SUPER_SPACE;
1531 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1532 findspace = VMFS_ALIGNED_SPACE(flags >>
1533 MAP_ALIGNMENT_SHIFT);
1535 findspace = VMFS_OPTIMAL_SPACE;
1538 if ((flags & MAP_32BIT) != 0)
1539 max_addr = MAP_32BIT_MAX_ADDR;
1542 rv = vm_map_find_min(map, object, foff, addr, size,
1543 round_page((vm_offset_t)td->td_proc->p_vmspace->
1544 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1545 findspace, prot, maxprot, docow);
1547 rv = vm_map_find(map, object, foff, addr, size,
1548 max_addr, findspace, prot, maxprot, docow);
1551 rv = vm_map_fixed(map, object, foff, *addr, size,
1552 prot, maxprot, docow);
1555 if (rv == KERN_SUCCESS) {
1557 * If the process has requested that all future mappings
1558 * be wired, then heed this.
1560 if (map->flags & MAP_WIREFUTURE) {
1561 vm_map_wire(map, *addr, *addr + size,
1562 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1563 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1566 return (vm_mmap_to_errno(rv));
1570 * Translate a Mach VM return code to zero on success or the appropriate errno
1574 vm_mmap_to_errno(int rv)
1580 case KERN_INVALID_ADDRESS:
1583 case KERN_PROTECTION_FAILURE: