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");
102 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
105 #ifndef _SYS_SYSPROTO_H_
112 sys_sbrk(struct thread *td, struct sbrk_args *uap)
114 /* Not yet implemented */
118 #ifndef _SYS_SYSPROTO_H_
125 sys_sstk(struct thread *td, struct sstk_args *uap)
127 /* Not yet implemented */
131 #if defined(COMPAT_43)
132 #ifndef _SYS_SYSPROTO_H_
133 struct getpagesize_args {
139 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
142 td->td_retval[0] = PAGE_SIZE;
145 #endif /* COMPAT_43 */
149 * Memory Map (mmap) system call. Note that the file offset
150 * and address are allowed to be NOT page aligned, though if
151 * the MAP_FIXED flag it set, both must have the same remainder
152 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
153 * page-aligned, the actual mapping starts at trunc_page(addr)
154 * and the return value is adjusted up by the page offset.
156 * Generally speaking, only character devices which are themselves
157 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
158 * there would be no cache coherency between a descriptor and a VM mapping
159 * both to the same character device.
161 #ifndef _SYS_SYSPROTO_H_
174 sys_mmap(struct thread *td, struct mmap_args *uap)
177 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
178 uap->flags, uap->fd, uap->pos));
182 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
189 vm_prot_t cap_maxprot;
193 vms = td->td_proc->p_vmspace;
199 * Ignore old flags that used to be defined but did not do anything.
201 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
204 * Enforce the constraints.
205 * Mapping of length 0 is only allowed for old binaries.
206 * Anonymous mapping shall specify -1 as filedescriptor and
207 * zero position for new code. Be nice to ancient a.out
208 * binaries and correct pos for anonymous mapping, since old
209 * ld.so sometimes issues anonymous map requests with non-zero
212 if (!SV_CURPROC_FLAG(SV_AOUT)) {
213 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
214 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
217 if ((flags & MAP_ANON) != 0)
221 if (flags & MAP_STACK) {
223 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
228 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
229 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
230 MAP_PREFAULT_READ | MAP_GUARD |
234 MAP_ALIGNMENT_MASK)) != 0)
236 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
238 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
240 if (prot != PROT_NONE &&
241 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
243 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
244 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
248 MAP_ALIGNMENT_MASK)) != 0))
252 * Align the file position to a page boundary,
253 * and save its page offset component.
255 pageoff = (pos & PAGE_MASK);
258 /* Adjust size for rounding (on both ends). */
259 size += pageoff; /* low end... */
260 size = (vm_size_t) round_page(size); /* hi end */
262 /* Ensure alignment is at least a page and fits in a pointer. */
263 align = flags & MAP_ALIGNMENT_MASK;
264 if (align != 0 && align != MAP_ALIGNED_SUPER &&
265 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
266 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
270 * Check for illegal addresses. Watch out for address wrap... Note
271 * that VM_*_ADDRESS are not constants due to casts (argh).
273 if (flags & MAP_FIXED) {
275 * The specified address must have the same remainder
276 * as the file offset taken modulo PAGE_SIZE, so it
277 * should be aligned after adjustment by pageoff.
280 if (addr & PAGE_MASK)
283 /* Address range must be all in user VM space. */
284 if (addr < vm_map_min(&vms->vm_map) ||
285 addr + size > vm_map_max(&vms->vm_map))
287 if (addr + size < addr)
290 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
292 } else if (flags & MAP_32BIT) {
294 * For MAP_32BIT, override the hint if it is too high and
295 * do not bother moving the mapping past the heap (since
296 * the heap is usually above 2GB).
298 if (addr + size > MAP_32BIT_MAX_ADDR)
303 * XXX for non-fixed mappings where no hint is provided or
304 * the hint would fall in the potential heap space,
305 * place it after the end of the largest possible heap.
307 * There should really be a pmap call to determine a reasonable
311 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
312 addr < round_page((vm_offset_t)vms->vm_daddr +
313 lim_max(td, RLIMIT_DATA))))
314 addr = round_page((vm_offset_t)vms->vm_daddr +
315 lim_max(td, RLIMIT_DATA));
319 * Return success without mapping anything for old
320 * binaries that request a page-aligned mapping of
321 * length 0. For modern binaries, this function
322 * returns an error earlier.
325 } else if ((flags & MAP_GUARD) != 0) {
326 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
327 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
328 } else if ((flags & MAP_ANON) != 0) {
330 * Mapping blank space is trivial.
332 * This relies on VM_PROT_* matching PROT_*.
334 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
335 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
338 * Mapping file, get fp for validation and don't let the
339 * descriptor disappear on us if we block. Check capability
340 * rights, but also return the maximum rights to be combined
341 * with maxprot later.
343 cap_rights_init(&rights, CAP_MMAP);
344 if (prot & PROT_READ)
345 cap_rights_set(&rights, CAP_MMAP_R);
346 if ((flags & MAP_SHARED) != 0) {
347 if (prot & PROT_WRITE)
348 cap_rights_set(&rights, CAP_MMAP_W);
350 if (prot & PROT_EXEC)
351 cap_rights_set(&rights, CAP_MMAP_X);
352 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
355 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
356 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
361 /* This relies on VM_PROT_* matching PROT_*. */
362 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
363 cap_maxprot, flags, pos, td);
367 td->td_retval[0] = (register_t) (addr + pageoff);
375 #if defined(COMPAT_FREEBSD6)
377 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
380 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
381 uap->flags, uap->fd, uap->pos));
386 #ifndef _SYS_SYSPROTO_H_
397 ommap(struct thread *td, struct ommap_args *uap)
399 static const char cvtbsdprot[8] = {
403 PROT_EXEC | PROT_WRITE,
405 PROT_EXEC | PROT_READ,
406 PROT_WRITE | PROT_READ,
407 PROT_EXEC | PROT_WRITE | PROT_READ,
411 #define OMAP_ANON 0x0002
412 #define OMAP_COPY 0x0020
413 #define OMAP_SHARED 0x0010
414 #define OMAP_FIXED 0x0100
416 prot = cvtbsdprot[uap->prot & 0x7];
417 #ifdef COMPAT_FREEBSD32
418 #if defined(__amd64__)
419 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
425 if (uap->flags & OMAP_ANON)
427 if (uap->flags & OMAP_COPY)
429 if (uap->flags & OMAP_SHARED)
432 flags |= MAP_PRIVATE;
433 if (uap->flags & OMAP_FIXED)
435 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
438 #endif /* COMPAT_43 */
441 #ifndef _SYS_SYSPROTO_H_
449 sys_msync(struct thread *td, struct msync_args *uap)
452 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
456 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
464 pageoff = (addr & PAGE_MASK);
467 size = (vm_size_t) round_page(size);
468 if (addr + size < addr)
471 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
474 map = &td->td_proc->p_vmspace->vm_map;
477 * Clean the pages and interpret the return value.
479 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
480 (flags & MS_INVALIDATE) != 0);
484 case KERN_INVALID_ADDRESS:
486 case KERN_INVALID_ARGUMENT:
495 #ifndef _SYS_SYSPROTO_H_
502 sys_munmap(struct thread *td, struct munmap_args *uap)
505 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
509 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
512 struct pmckern_map_out pkm;
513 vm_map_entry_t entry;
524 pageoff = (addr & PAGE_MASK);
527 size = (vm_size_t) round_page(size);
528 if (addr + size < addr)
532 * Check for illegal addresses. Watch out for address wrap...
534 map = &td->td_proc->p_vmspace->vm_map;
535 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
540 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
543 * Inform hwpmc if the address range being unmapped contains
544 * an executable region.
546 pkm.pm_address = (uintptr_t) NULL;
547 if (vm_map_lookup_entry(map, addr, &entry)) {
548 for (; entry->start < addr + size;
549 entry = entry->next) {
550 if (vm_map_check_protection(map, entry->start,
551 entry->end, VM_PROT_EXECUTE) == TRUE) {
552 pkm.pm_address = (uintptr_t) addr;
553 pkm.pm_size = (size_t) size;
560 vm_map_delete(map, addr, addr + size);
563 if (__predict_false(pmc_handled)) {
564 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
565 vm_map_lock_downgrade(map);
566 if (pkm.pm_address != (uintptr_t) NULL)
567 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
568 vm_map_unlock_read(map);
573 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
577 #ifndef _SYS_SYSPROTO_H_
578 struct mprotect_args {
585 sys_mprotect(struct thread *td, struct mprotect_args *uap)
588 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
592 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
598 prot = (prot & VM_PROT_ALL);
599 pageoff = (addr & PAGE_MASK);
602 size = (vm_size_t) round_page(size);
603 #ifdef COMPAT_FREEBSD32
604 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
605 if (((addr + size) & 0xffffffff) < addr)
609 if (addr + size < addr)
612 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
613 addr + size, prot, FALSE)) {
616 case KERN_PROTECTION_FAILURE:
618 case KERN_RESOURCE_SHORTAGE:
624 #ifndef _SYS_SYSPROTO_H_
625 struct minherit_args {
632 sys_minherit(struct thread *td, struct minherit_args *uap)
635 vm_size_t size, pageoff;
636 vm_inherit_t inherit;
638 addr = (vm_offset_t)uap->addr;
640 inherit = uap->inherit;
642 pageoff = (addr & PAGE_MASK);
645 size = (vm_size_t) round_page(size);
646 if (addr + size < addr)
649 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
650 addr + size, inherit)) {
653 case KERN_PROTECTION_FAILURE:
659 #ifndef _SYS_SYSPROTO_H_
660 struct madvise_args {
668 sys_madvise(struct thread *td, struct madvise_args *uap)
671 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
675 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
678 vm_offset_t addr, end, start;
682 * Check for our special case, advising the swap pager we are
685 if (behav == MADV_PROTECT) {
687 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
688 PROC_SPROTECT, &flags));
692 * Check for illegal addresses. Watch out for address wrap... Note
693 * that VM_*_ADDRESS are not constants due to casts (argh).
695 map = &td->td_proc->p_vmspace->vm_map;
697 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
699 if ((addr + len) < addr)
703 * Since this routine is only advisory, we default to conservative
706 start = trunc_page(addr);
707 end = round_page(addr + len);
710 * vm_map_madvise() checks for illegal values of behav.
712 return (vm_map_madvise(map, start, end, behav));
715 #ifndef _SYS_SYSPROTO_H_
716 struct mincore_args {
724 sys_mincore(struct thread *td, struct mincore_args *uap)
727 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
731 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
733 vm_offset_t addr, first_addr;
734 vm_offset_t end, cend;
738 int vecindex, lastvecindex;
739 vm_map_entry_t current;
740 vm_map_entry_t entry;
742 vm_paddr_t locked_pa;
746 unsigned int timestamp;
750 * Make sure that the addresses presented are valid for user
753 first_addr = addr = trunc_page(addr0);
754 end = addr + (vm_size_t)round_page(len);
755 map = &td->td_proc->p_vmspace->vm_map;
756 if (end > vm_map_max(map) || end < addr)
759 pmap = vmspace_pmap(td->td_proc->p_vmspace);
761 vm_map_lock_read(map);
763 timestamp = map->timestamp;
765 if (!vm_map_lookup_entry(map, addr, &entry)) {
766 vm_map_unlock_read(map);
771 * Do this on a map entry basis so that if the pages are not
772 * in the current processes address space, we can easily look
773 * up the pages elsewhere.
776 for (current = entry; current->start < end; current = current->next) {
779 * check for contiguity
781 if (current->end < end && current->next->start > current->end) {
782 vm_map_unlock_read(map);
787 * ignore submaps (for now) or null objects
789 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
790 current->object.vm_object == NULL)
794 * limit this scan to the current map entry and the
795 * limits for the mincore call
797 if (addr < current->start)
798 addr = current->start;
804 * scan this entry one page at a time
806 while (addr < cend) {
808 * Check pmap first, it is likely faster, also
809 * it can provide info as to whether we are the
810 * one referencing or modifying the page.
816 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
817 if (locked_pa != 0) {
819 * The page is mapped by this process but not
820 * both accessed and modified. It is also
821 * managed. Acquire the object lock so that
822 * other mappings might be examined.
824 m = PHYS_TO_VM_PAGE(locked_pa);
825 if (m->object != object) {
827 VM_OBJECT_WUNLOCK(object);
829 locked = VM_OBJECT_TRYWLOCK(object);
832 VM_OBJECT_WLOCK(object);
838 KASSERT(m->valid == VM_PAGE_BITS_ALL,
839 ("mincore: page %p is mapped but invalid",
841 } else if (mincoreinfo == 0) {
843 * The page is not mapped by this process. If
844 * the object implements managed pages, then
845 * determine if the page is resident so that
846 * the mappings might be examined.
848 if (current->object.vm_object != object) {
850 VM_OBJECT_WUNLOCK(object);
851 object = current->object.vm_object;
852 VM_OBJECT_WLOCK(object);
854 if (object->type == OBJT_DEFAULT ||
855 object->type == OBJT_SWAP ||
856 object->type == OBJT_VNODE) {
857 pindex = OFF_TO_IDX(current->offset +
858 (addr - current->start));
859 m = vm_page_lookup(object, pindex);
860 if (m != NULL && m->valid == 0)
863 mincoreinfo = MINCORE_INCORE;
867 /* Examine other mappings to the page. */
868 if (m->dirty == 0 && pmap_is_modified(m))
871 mincoreinfo |= MINCORE_MODIFIED_OTHER;
873 * The first test for PGA_REFERENCED is an
874 * optimization. The second test is
875 * required because a concurrent pmap
876 * operation could clear the last reference
877 * and set PGA_REFERENCED before the call to
878 * pmap_is_referenced().
880 if ((m->aflags & PGA_REFERENCED) != 0 ||
881 pmap_is_referenced(m) ||
882 (m->aflags & PGA_REFERENCED) != 0)
883 mincoreinfo |= MINCORE_REFERENCED_OTHER;
886 VM_OBJECT_WUNLOCK(object);
889 * subyte may page fault. In case it needs to modify
890 * the map, we release the lock.
892 vm_map_unlock_read(map);
895 * calculate index into user supplied byte vector
897 vecindex = atop(addr - first_addr);
900 * If we have skipped map entries, we need to make sure that
901 * the byte vector is zeroed for those skipped entries.
903 while ((lastvecindex + 1) < vecindex) {
905 error = subyte(vec + lastvecindex, 0);
913 * Pass the page information to the user
915 error = subyte(vec + vecindex, mincoreinfo);
922 * If the map has changed, due to the subyte, the previous
923 * output may be invalid.
925 vm_map_lock_read(map);
926 if (timestamp != map->timestamp)
929 lastvecindex = vecindex;
935 * subyte may page fault. In case it needs to modify
936 * the map, we release the lock.
938 vm_map_unlock_read(map);
941 * Zero the last entries in the byte vector.
943 vecindex = atop(end - first_addr);
944 while ((lastvecindex + 1) < vecindex) {
946 error = subyte(vec + lastvecindex, 0);
954 * If the map has changed, due to the subyte, the previous
955 * output may be invalid.
957 vm_map_lock_read(map);
958 if (timestamp != map->timestamp)
960 vm_map_unlock_read(map);
965 #ifndef _SYS_SYSPROTO_H_
972 sys_mlock(struct thread *td, struct mlock_args *uap)
975 return (kern_mlock(td->td_proc, td->td_ucred,
976 __DECONST(uintptr_t, uap->addr), uap->len));
980 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
982 vm_offset_t addr, end, last, start;
983 vm_size_t npages, size;
988 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
994 start = trunc_page(addr);
995 end = round_page(last);
996 if (last < addr || end < addr)
998 npages = atop(end - start);
999 if (npages > vm_page_max_wired)
1001 map = &proc->p_vmspace->vm_map;
1003 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1004 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1009 if (npages + vm_wire_count() > vm_page_max_wired)
1014 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1020 error = vm_map_wire(map, start, end,
1021 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1023 if (racct_enable && error != KERN_SUCCESS) {
1025 racct_set(proc, RACCT_MEMLOCK,
1026 ptoa(pmap_wired_count(map->pmap)));
1030 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1033 #ifndef _SYS_SYSPROTO_H_
1034 struct mlockall_args {
1040 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1045 map = &td->td_proc->p_vmspace->vm_map;
1046 error = priv_check(td, PRIV_VM_MLOCK);
1050 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1054 * If wiring all pages in the process would cause it to exceed
1055 * a hard resource limit, return ENOMEM.
1057 if (!old_mlock && uap->how & MCL_CURRENT) {
1058 PROC_LOCK(td->td_proc);
1059 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1060 PROC_UNLOCK(td->td_proc);
1063 PROC_UNLOCK(td->td_proc);
1067 PROC_LOCK(td->td_proc);
1068 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1069 PROC_UNLOCK(td->td_proc);
1075 if (uap->how & MCL_FUTURE) {
1077 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1082 if (uap->how & MCL_CURRENT) {
1084 * P1003.1-2001 mandates that all currently mapped pages
1085 * will be memory resident and locked (wired) upon return
1086 * from mlockall(). vm_map_wire() will wire pages, by
1087 * calling vm_fault_wire() for each page in the region.
1089 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1090 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1091 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1094 if (racct_enable && error != KERN_SUCCESS) {
1095 PROC_LOCK(td->td_proc);
1096 racct_set(td->td_proc, RACCT_MEMLOCK,
1097 ptoa(pmap_wired_count(map->pmap)));
1098 PROC_UNLOCK(td->td_proc);
1105 #ifndef _SYS_SYSPROTO_H_
1106 struct munlockall_args {
1112 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1117 map = &td->td_proc->p_vmspace->vm_map;
1118 error = priv_check(td, PRIV_VM_MUNLOCK);
1122 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1124 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1127 /* Forcibly unwire all pages. */
1128 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1129 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1131 if (racct_enable && error == KERN_SUCCESS) {
1132 PROC_LOCK(td->td_proc);
1133 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1134 PROC_UNLOCK(td->td_proc);
1141 #ifndef _SYS_SYSPROTO_H_
1142 struct munlock_args {
1148 sys_munlock(struct thread *td, struct munlock_args *uap)
1151 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1155 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1157 vm_offset_t addr, end, last, start;
1163 error = priv_check(td, PRIV_VM_MUNLOCK);
1168 start = trunc_page(addr);
1169 end = round_page(last);
1170 if (last < addr || end < addr)
1172 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1173 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1175 if (racct_enable && error == KERN_SUCCESS) {
1176 PROC_LOCK(td->td_proc);
1177 map = &td->td_proc->p_vmspace->vm_map;
1178 racct_set(td->td_proc, RACCT_MEMLOCK,
1179 ptoa(pmap_wired_count(map->pmap)));
1180 PROC_UNLOCK(td->td_proc);
1183 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1189 * Helper function for vm_mmap. Perform sanity check specific for mmap
1190 * operations on vnodes.
1193 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1194 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1195 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1196 boolean_t *writecounted)
1202 int error, flags, locktype;
1204 cred = td->td_ucred;
1205 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1206 locktype = LK_EXCLUSIVE;
1208 locktype = LK_SHARED;
1209 if ((error = vget(vp, locktype, td)) != 0)
1211 AUDIT_ARG_VNODE1(vp);
1215 if (vp->v_type == VREG) {
1217 * Get the proper underlying object
1223 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1225 vp = (struct vnode *)obj->handle;
1227 * Bypass filesystems obey the mpsafety of the
1228 * underlying fs. Tmpfs never bypasses.
1230 error = vget(vp, locktype, td);
1234 if (locktype == LK_EXCLUSIVE) {
1235 *writecounted = TRUE;
1236 vnode_pager_update_writecount(obj, 0, objsize);
1242 if ((error = VOP_GETATTR(vp, &va, cred)))
1245 /* This relies on VM_PROT_* matching PROT_*. */
1246 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1250 if ((flags & MAP_SHARED) != 0) {
1251 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1252 if (prot & VM_PROT_WRITE) {
1256 *maxprotp &= ~VM_PROT_WRITE;
1260 * If it is a regular file without any references
1261 * we do not need to sync it.
1262 * Adjust object size to be the size of actual file.
1264 objsize = round_page(va.va_size);
1265 if (va.va_nlink == 0)
1266 flags |= MAP_NOSYNC;
1267 if (obj->type == OBJT_VNODE) {
1268 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1275 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1276 ("wrong object type"));
1277 VM_OBJECT_WLOCK(obj);
1278 vm_object_reference_locked(obj);
1279 #if VM_NRESERVLEVEL > 0
1280 vm_object_color(obj, 0);
1282 VM_OBJECT_WUNLOCK(obj);
1287 vfs_mark_atime(vp, cred);
1290 if (error != 0 && *writecounted) {
1291 *writecounted = FALSE;
1292 vnode_pager_update_writecount(obj, objsize, 0);
1301 * Helper function for vm_mmap. Perform sanity check specific for mmap
1302 * operations on cdevs.
1305 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1306 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1307 vm_ooffset_t *foff, vm_object_t *objp)
1314 if (dsw->d_flags & D_MMAP_ANON) {
1317 *maxprotp = VM_PROT_ALL;
1318 *flagsp |= MAP_ANON;
1322 * cdevs do not provide private mappings of any kind.
1324 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1325 (prot & VM_PROT_WRITE) != 0)
1327 if (flags & (MAP_PRIVATE|MAP_COPY))
1330 * Force device mappings to be shared.
1332 flags |= MAP_SHARED;
1334 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1339 * First, try d_mmap_single(). If that is not implemented
1340 * (returns ENODEV), fall back to using the device pager.
1341 * Note that d_mmap_single() must return a reference to the
1342 * object (it needs to bump the reference count of the object
1343 * it returns somehow).
1345 * XXX assumes VM_PROT_* == PROT_*
1347 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1348 if (error != ENODEV)
1350 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1362 * Internal version of mmap used by exec, sys5 shared memory, and
1363 * various device drivers. Handle is either a vnode pointer, a
1364 * character device, or NULL for MAP_ANON.
1367 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1368 vm_prot_t maxprot, int flags,
1369 objtype_t handle_type, void *handle,
1373 struct thread *td = curthread;
1375 boolean_t writecounted;
1380 size = round_page(size);
1382 writecounted = FALSE;
1385 * Lookup/allocate object.
1387 switch (handle_type) {
1394 dsw = dev_refthread(cdev, &ref);
1397 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1398 dsw, &foff, &object);
1399 dev_relthread(cdev, ref);
1403 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1404 handle, &foff, &object, &writecounted);
1407 if (handle == NULL) {
1419 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1420 foff, writecounted, td);
1421 if (error != 0 && object != NULL) {
1423 * If this mapping was accounted for in the vnode's
1424 * writecount, then undo that now.
1427 vnode_pager_release_writecount(object, 0, size);
1428 vm_object_deallocate(object);
1434 * Internal version of mmap that maps a specific VM object into an
1435 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1438 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1439 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1440 boolean_t writecounted, struct thread *td)
1442 boolean_t curmap, fitit;
1443 vm_offset_t max_addr;
1444 int docow, error, findspace, rv;
1446 curmap = map == &td->td_proc->p_vmspace->vm_map;
1448 PROC_LOCK(td->td_proc);
1449 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1450 PROC_UNLOCK(td->td_proc);
1453 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1454 PROC_UNLOCK(td->td_proc);
1457 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1458 if (ptoa(pmap_wired_count(map->pmap)) + size >
1459 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1460 racct_set_force(td->td_proc, RACCT_VMEM,
1462 PROC_UNLOCK(td->td_proc);
1465 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1466 ptoa(pmap_wired_count(map->pmap)) + size);
1468 racct_set_force(td->td_proc, RACCT_VMEM,
1470 PROC_UNLOCK(td->td_proc);
1474 PROC_UNLOCK(td->td_proc);
1478 * We currently can only deal with page aligned file offsets.
1479 * The mmap() system call already enforces this by subtracting
1480 * the page offset from the file offset, but checking here
1481 * catches errors in device drivers (e.g. d_single_mmap()
1482 * callbacks) and other internal mapping requests (such as in
1485 if (foff & PAGE_MASK)
1488 if ((flags & MAP_FIXED) == 0) {
1490 *addr = round_page(*addr);
1492 if (*addr != trunc_page(*addr))
1497 if (flags & MAP_ANON) {
1498 if (object != NULL || foff != 0)
1501 } else if (flags & MAP_PREFAULT_READ)
1502 docow = MAP_PREFAULT;
1504 docow = MAP_PREFAULT_PARTIAL;
1506 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1507 docow |= MAP_COPY_ON_WRITE;
1508 if (flags & MAP_NOSYNC)
1509 docow |= MAP_DISABLE_SYNCER;
1510 if (flags & MAP_NOCORE)
1511 docow |= MAP_DISABLE_COREDUMP;
1512 /* Shared memory is also shared with children. */
1513 if (flags & MAP_SHARED)
1514 docow |= MAP_INHERIT_SHARE;
1516 docow |= MAP_VN_WRITECOUNT;
1517 if (flags & MAP_STACK) {
1520 docow |= MAP_STACK_GROWS_DOWN;
1522 if ((flags & MAP_EXCL) != 0)
1523 docow |= MAP_CHECK_EXCL;
1524 if ((flags & MAP_GUARD) != 0)
1525 docow |= MAP_CREATE_GUARD;
1528 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1529 findspace = VMFS_SUPER_SPACE;
1530 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1531 findspace = VMFS_ALIGNED_SPACE(flags >>
1532 MAP_ALIGNMENT_SHIFT);
1534 findspace = VMFS_OPTIMAL_SPACE;
1537 if ((flags & MAP_32BIT) != 0)
1538 max_addr = MAP_32BIT_MAX_ADDR;
1541 rv = vm_map_find_min(map, object, foff, addr, size,
1542 round_page((vm_offset_t)td->td_proc->p_vmspace->
1543 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1544 findspace, prot, maxprot, docow);
1546 rv = vm_map_find(map, object, foff, addr, size,
1547 max_addr, findspace, prot, maxprot, docow);
1550 rv = vm_map_fixed(map, object, foff, *addr, size,
1551 prot, maxprot, docow);
1554 if (rv == KERN_SUCCESS) {
1556 * If the process has requested that all future mappings
1557 * be wired, then heed this.
1559 if (map->flags & MAP_WIREFUTURE) {
1560 vm_map_wire(map, *addr, *addr + size,
1561 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1562 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1565 return (vm_mmap_to_errno(rv));
1569 * Translate a Mach VM return code to zero on success or the appropriate errno
1573 vm_mmap_to_errno(int rv)
1579 case KERN_INVALID_ADDRESS:
1582 case KERN_PROTECTION_FAILURE: