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_SHARED | MAP_PRIVATE | MAP_PREFAULT |
245 MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
249 * Align the file position to a page boundary,
250 * and save its page offset component.
252 pageoff = (pos & PAGE_MASK);
255 /* Adjust size for rounding (on both ends). */
256 size += pageoff; /* low end... */
257 size = (vm_size_t) round_page(size); /* hi end */
259 /* Ensure alignment is at least a page and fits in a pointer. */
260 align = flags & MAP_ALIGNMENT_MASK;
261 if (align != 0 && align != MAP_ALIGNED_SUPER &&
262 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
263 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
267 * Check for illegal addresses. Watch out for address wrap... Note
268 * that VM_*_ADDRESS are not constants due to casts (argh).
270 if (flags & MAP_FIXED) {
272 * The specified address must have the same remainder
273 * as the file offset taken modulo PAGE_SIZE, so it
274 * should be aligned after adjustment by pageoff.
277 if (addr & PAGE_MASK)
280 /* Address range must be all in user VM space. */
281 if (addr < vm_map_min(&vms->vm_map) ||
282 addr + size > vm_map_max(&vms->vm_map))
284 if (addr + size < addr)
287 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
289 } else if (flags & MAP_32BIT) {
291 * For MAP_32BIT, override the hint if it is too high and
292 * do not bother moving the mapping past the heap (since
293 * the heap is usually above 2GB).
295 if (addr + size > MAP_32BIT_MAX_ADDR)
300 * XXX for non-fixed mappings where no hint is provided or
301 * the hint would fall in the potential heap space,
302 * place it after the end of the largest possible heap.
304 * There should really be a pmap call to determine a reasonable
308 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
309 addr < round_page((vm_offset_t)vms->vm_daddr +
310 lim_max(td, RLIMIT_DATA))))
311 addr = round_page((vm_offset_t)vms->vm_daddr +
312 lim_max(td, RLIMIT_DATA));
316 * Return success without mapping anything for old
317 * binaries that request a page-aligned mapping of
318 * length 0. For modern binaries, this function
319 * returns an error earlier.
322 } else if ((flags & MAP_GUARD) != 0) {
323 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
324 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
325 } else if ((flags & MAP_ANON) != 0) {
327 * Mapping blank space is trivial.
329 * This relies on VM_PROT_* matching PROT_*.
331 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
332 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
335 * Mapping file, get fp for validation and don't let the
336 * descriptor disappear on us if we block. Check capability
337 * rights, but also return the maximum rights to be combined
338 * with maxprot later.
340 cap_rights_init(&rights, CAP_MMAP);
341 if (prot & PROT_READ)
342 cap_rights_set(&rights, CAP_MMAP_R);
343 if ((flags & MAP_SHARED) != 0) {
344 if (prot & PROT_WRITE)
345 cap_rights_set(&rights, CAP_MMAP_W);
347 if (prot & PROT_EXEC)
348 cap_rights_set(&rights, CAP_MMAP_X);
349 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
352 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
353 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
358 /* This relies on VM_PROT_* matching PROT_*. */
359 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
360 cap_maxprot, flags, pos, td);
364 td->td_retval[0] = (register_t) (addr + pageoff);
372 #if defined(COMPAT_FREEBSD6)
374 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
377 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
378 uap->flags, uap->fd, uap->pos));
383 #ifndef _SYS_SYSPROTO_H_
394 ommap(struct thread *td, struct ommap_args *uap)
396 static const char cvtbsdprot[8] = {
400 PROT_EXEC | PROT_WRITE,
402 PROT_EXEC | PROT_READ,
403 PROT_WRITE | PROT_READ,
404 PROT_EXEC | PROT_WRITE | PROT_READ,
408 #define OMAP_ANON 0x0002
409 #define OMAP_COPY 0x0020
410 #define OMAP_SHARED 0x0010
411 #define OMAP_FIXED 0x0100
413 prot = cvtbsdprot[uap->prot & 0x7];
414 #ifdef COMPAT_FREEBSD32
415 #if defined(__amd64__)
416 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
422 if (uap->flags & OMAP_ANON)
424 if (uap->flags & OMAP_COPY)
426 if (uap->flags & OMAP_SHARED)
429 flags |= MAP_PRIVATE;
430 if (uap->flags & OMAP_FIXED)
432 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
435 #endif /* COMPAT_43 */
438 #ifndef _SYS_SYSPROTO_H_
446 sys_msync(struct thread *td, struct msync_args *uap)
449 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
453 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
461 pageoff = (addr & PAGE_MASK);
464 size = (vm_size_t) round_page(size);
465 if (addr + size < addr)
468 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
471 map = &td->td_proc->p_vmspace->vm_map;
474 * Clean the pages and interpret the return value.
476 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
477 (flags & MS_INVALIDATE) != 0);
481 case KERN_INVALID_ADDRESS:
483 case KERN_INVALID_ARGUMENT:
492 #ifndef _SYS_SYSPROTO_H_
499 sys_munmap(struct thread *td, struct munmap_args *uap)
502 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
506 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
509 struct pmckern_map_out pkm;
510 vm_map_entry_t entry;
521 pageoff = (addr & PAGE_MASK);
524 size = (vm_size_t) round_page(size);
525 if (addr + size < addr)
529 * Check for illegal addresses. Watch out for address wrap...
531 map = &td->td_proc->p_vmspace->vm_map;
532 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
537 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
540 * Inform hwpmc if the address range being unmapped contains
541 * an executable region.
543 pkm.pm_address = (uintptr_t) NULL;
544 if (vm_map_lookup_entry(map, addr, &entry)) {
545 for (; entry->start < addr + size;
546 entry = entry->next) {
547 if (vm_map_check_protection(map, entry->start,
548 entry->end, VM_PROT_EXECUTE) == TRUE) {
549 pkm.pm_address = (uintptr_t) addr;
550 pkm.pm_size = (size_t) size;
557 vm_map_delete(map, addr, addr + size);
560 if (__predict_false(pmc_handled)) {
561 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
562 vm_map_lock_downgrade(map);
563 if (pkm.pm_address != (uintptr_t) NULL)
564 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
565 vm_map_unlock_read(map);
570 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
574 #ifndef _SYS_SYSPROTO_H_
575 struct mprotect_args {
582 sys_mprotect(struct thread *td, struct mprotect_args *uap)
585 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
589 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
595 prot = (prot & VM_PROT_ALL);
596 pageoff = (addr & PAGE_MASK);
599 size = (vm_size_t) round_page(size);
600 if (addr + size < addr)
603 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
604 addr + size, prot, FALSE)) {
607 case KERN_PROTECTION_FAILURE:
609 case KERN_RESOURCE_SHORTAGE:
615 #ifndef _SYS_SYSPROTO_H_
616 struct minherit_args {
623 sys_minherit(struct thread *td, struct minherit_args *uap)
626 vm_size_t size, pageoff;
627 vm_inherit_t inherit;
629 addr = (vm_offset_t)uap->addr;
631 inherit = uap->inherit;
633 pageoff = (addr & PAGE_MASK);
636 size = (vm_size_t) round_page(size);
637 if (addr + size < addr)
640 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
641 addr + size, inherit)) {
644 case KERN_PROTECTION_FAILURE:
650 #ifndef _SYS_SYSPROTO_H_
651 struct madvise_args {
659 sys_madvise(struct thread *td, struct madvise_args *uap)
662 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
666 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
669 vm_offset_t addr, end, start;
673 * Check for our special case, advising the swap pager we are
676 if (behav == MADV_PROTECT) {
678 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
679 PROC_SPROTECT, &flags));
683 * Check for illegal behavior
685 if (behav < 0 || behav > MADV_CORE)
688 * Check for illegal addresses. Watch out for address wrap... Note
689 * that VM_*_ADDRESS are not constants due to casts (argh).
691 map = &td->td_proc->p_vmspace->vm_map;
693 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
695 if ((addr + len) < addr)
699 * Since this routine is only advisory, we default to conservative
702 start = trunc_page(addr);
703 end = round_page(addr + len);
705 if (vm_map_madvise(map, start, end, behav))
710 #ifndef _SYS_SYSPROTO_H_
711 struct mincore_args {
719 sys_mincore(struct thread *td, struct mincore_args *uap)
722 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
726 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
728 vm_offset_t addr, first_addr;
729 vm_offset_t end, cend;
733 int vecindex, lastvecindex;
734 vm_map_entry_t current;
735 vm_map_entry_t entry;
737 vm_paddr_t locked_pa;
741 unsigned int timestamp;
745 * Make sure that the addresses presented are valid for user
748 first_addr = addr = trunc_page(addr0);
749 end = addr + (vm_size_t)round_page(len);
750 map = &td->td_proc->p_vmspace->vm_map;
751 if (end > vm_map_max(map) || end < addr)
754 pmap = vmspace_pmap(td->td_proc->p_vmspace);
756 vm_map_lock_read(map);
758 timestamp = map->timestamp;
760 if (!vm_map_lookup_entry(map, addr, &entry)) {
761 vm_map_unlock_read(map);
766 * Do this on a map entry basis so that if the pages are not
767 * in the current processes address space, we can easily look
768 * up the pages elsewhere.
771 for (current = entry; current->start < end; current = current->next) {
774 * check for contiguity
776 if (current->end < end && current->next->start > current->end) {
777 vm_map_unlock_read(map);
782 * ignore submaps (for now) or null objects
784 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
785 current->object.vm_object == NULL)
789 * limit this scan to the current map entry and the
790 * limits for the mincore call
792 if (addr < current->start)
793 addr = current->start;
799 * scan this entry one page at a time
801 while (addr < cend) {
803 * Check pmap first, it is likely faster, also
804 * it can provide info as to whether we are the
805 * one referencing or modifying the page.
811 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
812 if (locked_pa != 0) {
814 * The page is mapped by this process but not
815 * both accessed and modified. It is also
816 * managed. Acquire the object lock so that
817 * other mappings might be examined.
819 m = PHYS_TO_VM_PAGE(locked_pa);
820 if (m->object != object) {
822 VM_OBJECT_WUNLOCK(object);
824 locked = VM_OBJECT_TRYWLOCK(object);
827 VM_OBJECT_WLOCK(object);
833 KASSERT(m->valid == VM_PAGE_BITS_ALL,
834 ("mincore: page %p is mapped but invalid",
836 } else if (mincoreinfo == 0) {
838 * The page is not mapped by this process. If
839 * the object implements managed pages, then
840 * determine if the page is resident so that
841 * the mappings might be examined.
843 if (current->object.vm_object != object) {
845 VM_OBJECT_WUNLOCK(object);
846 object = current->object.vm_object;
847 VM_OBJECT_WLOCK(object);
849 if (object->type == OBJT_DEFAULT ||
850 object->type == OBJT_SWAP ||
851 object->type == OBJT_VNODE) {
852 pindex = OFF_TO_IDX(current->offset +
853 (addr - current->start));
854 m = vm_page_lookup(object, pindex);
855 if (m != NULL && m->valid == 0)
858 mincoreinfo = MINCORE_INCORE;
862 /* Examine other mappings to the page. */
863 if (m->dirty == 0 && pmap_is_modified(m))
866 mincoreinfo |= MINCORE_MODIFIED_OTHER;
868 * The first test for PGA_REFERENCED is an
869 * optimization. The second test is
870 * required because a concurrent pmap
871 * operation could clear the last reference
872 * and set PGA_REFERENCED before the call to
873 * pmap_is_referenced().
875 if ((m->aflags & PGA_REFERENCED) != 0 ||
876 pmap_is_referenced(m) ||
877 (m->aflags & PGA_REFERENCED) != 0)
878 mincoreinfo |= MINCORE_REFERENCED_OTHER;
881 VM_OBJECT_WUNLOCK(object);
884 * subyte may page fault. In case it needs to modify
885 * the map, we release the lock.
887 vm_map_unlock_read(map);
890 * calculate index into user supplied byte vector
892 vecindex = atop(addr - first_addr);
895 * If we have skipped map entries, we need to make sure that
896 * the byte vector is zeroed for those skipped entries.
898 while ((lastvecindex + 1) < vecindex) {
900 error = subyte(vec + lastvecindex, 0);
908 * Pass the page information to the user
910 error = subyte(vec + vecindex, mincoreinfo);
917 * If the map has changed, due to the subyte, the previous
918 * output may be invalid.
920 vm_map_lock_read(map);
921 if (timestamp != map->timestamp)
924 lastvecindex = vecindex;
930 * subyte may page fault. In case it needs to modify
931 * the map, we release the lock.
933 vm_map_unlock_read(map);
936 * Zero the last entries in the byte vector.
938 vecindex = atop(end - first_addr);
939 while ((lastvecindex + 1) < vecindex) {
941 error = subyte(vec + lastvecindex, 0);
949 * If the map has changed, due to the subyte, the previous
950 * output may be invalid.
952 vm_map_lock_read(map);
953 if (timestamp != map->timestamp)
955 vm_map_unlock_read(map);
960 #ifndef _SYS_SYSPROTO_H_
967 sys_mlock(struct thread *td, struct mlock_args *uap)
970 return (kern_mlock(td->td_proc, td->td_ucred,
971 __DECONST(uintptr_t, uap->addr), uap->len));
975 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
977 vm_offset_t addr, end, last, start;
978 vm_size_t npages, size;
983 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
989 start = trunc_page(addr);
990 end = round_page(last);
991 if (last < addr || end < addr)
993 npages = atop(end - start);
994 if (npages > vm_page_max_wired)
996 map = &proc->p_vmspace->vm_map;
998 nsize = ptoa(npages + pmap_wired_count(map->pmap));
999 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1004 if (npages + vm_wire_count() > vm_page_max_wired)
1009 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1015 error = vm_map_wire(map, start, end,
1016 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1018 if (racct_enable && error != KERN_SUCCESS) {
1020 racct_set(proc, RACCT_MEMLOCK,
1021 ptoa(pmap_wired_count(map->pmap)));
1025 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1028 #ifndef _SYS_SYSPROTO_H_
1029 struct mlockall_args {
1035 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1040 map = &td->td_proc->p_vmspace->vm_map;
1041 error = priv_check(td, PRIV_VM_MLOCK);
1045 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1049 * If wiring all pages in the process would cause it to exceed
1050 * a hard resource limit, return ENOMEM.
1052 if (!old_mlock && uap->how & MCL_CURRENT) {
1053 PROC_LOCK(td->td_proc);
1054 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1055 PROC_UNLOCK(td->td_proc);
1058 PROC_UNLOCK(td->td_proc);
1062 PROC_LOCK(td->td_proc);
1063 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1064 PROC_UNLOCK(td->td_proc);
1070 if (uap->how & MCL_FUTURE) {
1072 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1077 if (uap->how & MCL_CURRENT) {
1079 * P1003.1-2001 mandates that all currently mapped pages
1080 * will be memory resident and locked (wired) upon return
1081 * from mlockall(). vm_map_wire() will wire pages, by
1082 * calling vm_fault_wire() for each page in the region.
1084 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1085 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1086 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1089 if (racct_enable && error != KERN_SUCCESS) {
1090 PROC_LOCK(td->td_proc);
1091 racct_set(td->td_proc, RACCT_MEMLOCK,
1092 ptoa(pmap_wired_count(map->pmap)));
1093 PROC_UNLOCK(td->td_proc);
1100 #ifndef _SYS_SYSPROTO_H_
1101 struct munlockall_args {
1107 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1112 map = &td->td_proc->p_vmspace->vm_map;
1113 error = priv_check(td, PRIV_VM_MUNLOCK);
1117 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1119 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1122 /* Forcibly unwire all pages. */
1123 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1124 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1126 if (racct_enable && error == KERN_SUCCESS) {
1127 PROC_LOCK(td->td_proc);
1128 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1129 PROC_UNLOCK(td->td_proc);
1136 #ifndef _SYS_SYSPROTO_H_
1137 struct munlock_args {
1143 sys_munlock(struct thread *td, struct munlock_args *uap)
1146 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1150 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1152 vm_offset_t addr, end, last, start;
1158 error = priv_check(td, PRIV_VM_MUNLOCK);
1163 start = trunc_page(addr);
1164 end = round_page(last);
1165 if (last < addr || end < addr)
1167 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1168 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1170 if (racct_enable && error == KERN_SUCCESS) {
1171 PROC_LOCK(td->td_proc);
1172 map = &td->td_proc->p_vmspace->vm_map;
1173 racct_set(td->td_proc, RACCT_MEMLOCK,
1174 ptoa(pmap_wired_count(map->pmap)));
1175 PROC_UNLOCK(td->td_proc);
1178 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1184 * Helper function for vm_mmap. Perform sanity check specific for mmap
1185 * operations on vnodes.
1188 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1189 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1190 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1191 boolean_t *writecounted)
1197 int error, flags, locktype;
1199 cred = td->td_ucred;
1200 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1201 locktype = LK_EXCLUSIVE;
1203 locktype = LK_SHARED;
1204 if ((error = vget(vp, locktype, td)) != 0)
1206 AUDIT_ARG_VNODE1(vp);
1210 if (vp->v_type == VREG) {
1212 * Get the proper underlying object
1218 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1220 vp = (struct vnode *)obj->handle;
1222 * Bypass filesystems obey the mpsafety of the
1223 * underlying fs. Tmpfs never bypasses.
1225 error = vget(vp, locktype, td);
1229 if (locktype == LK_EXCLUSIVE) {
1230 *writecounted = TRUE;
1231 vnode_pager_update_writecount(obj, 0, objsize);
1237 if ((error = VOP_GETATTR(vp, &va, cred)))
1240 /* This relies on VM_PROT_* matching PROT_*. */
1241 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1245 if ((flags & MAP_SHARED) != 0) {
1246 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1247 if (prot & VM_PROT_WRITE) {
1251 *maxprotp &= ~VM_PROT_WRITE;
1255 * If it is a regular file without any references
1256 * we do not need to sync it.
1257 * Adjust object size to be the size of actual file.
1259 objsize = round_page(va.va_size);
1260 if (va.va_nlink == 0)
1261 flags |= MAP_NOSYNC;
1262 if (obj->type == OBJT_VNODE) {
1263 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1270 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1271 ("wrong object type"));
1272 VM_OBJECT_WLOCK(obj);
1273 vm_object_reference_locked(obj);
1274 #if VM_NRESERVLEVEL > 0
1275 vm_object_color(obj, 0);
1277 VM_OBJECT_WUNLOCK(obj);
1282 vfs_mark_atime(vp, cred);
1285 if (error != 0 && *writecounted) {
1286 *writecounted = FALSE;
1287 vnode_pager_update_writecount(obj, objsize, 0);
1296 * Helper function for vm_mmap. Perform sanity check specific for mmap
1297 * operations on cdevs.
1300 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1301 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1302 vm_ooffset_t *foff, vm_object_t *objp)
1309 if (dsw->d_flags & D_MMAP_ANON) {
1312 *maxprotp = VM_PROT_ALL;
1313 *flagsp |= MAP_ANON;
1317 * cdevs do not provide private mappings of any kind.
1319 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1320 (prot & VM_PROT_WRITE) != 0)
1322 if (flags & (MAP_PRIVATE|MAP_COPY))
1325 * Force device mappings to be shared.
1327 flags |= MAP_SHARED;
1329 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1334 * First, try d_mmap_single(). If that is not implemented
1335 * (returns ENODEV), fall back to using the device pager.
1336 * Note that d_mmap_single() must return a reference to the
1337 * object (it needs to bump the reference count of the object
1338 * it returns somehow).
1340 * XXX assumes VM_PROT_* == PROT_*
1342 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1343 if (error != ENODEV)
1345 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1357 * Internal version of mmap used by exec, sys5 shared memory, and
1358 * various device drivers. Handle is either a vnode pointer, a
1359 * character device, or NULL for MAP_ANON.
1362 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1363 vm_prot_t maxprot, int flags,
1364 objtype_t handle_type, void *handle,
1368 struct thread *td = curthread;
1370 boolean_t writecounted;
1375 size = round_page(size);
1377 writecounted = FALSE;
1380 * Lookup/allocate object.
1382 switch (handle_type) {
1389 dsw = dev_refthread(cdev, &ref);
1392 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1393 dsw, &foff, &object);
1394 dev_relthread(cdev, ref);
1398 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1399 handle, &foff, &object, &writecounted);
1402 if (handle == NULL) {
1414 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1415 foff, writecounted, td);
1416 if (error != 0 && object != NULL) {
1418 * If this mapping was accounted for in the vnode's
1419 * writecount, then undo that now.
1422 vnode_pager_release_writecount(object, 0, size);
1423 vm_object_deallocate(object);
1429 * Internal version of mmap that maps a specific VM object into an
1430 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1433 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1434 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1435 boolean_t writecounted, struct thread *td)
1437 boolean_t curmap, fitit;
1438 vm_offset_t max_addr;
1439 int docow, error, findspace, rv;
1441 curmap = map == &td->td_proc->p_vmspace->vm_map;
1443 PROC_LOCK(td->td_proc);
1444 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1445 PROC_UNLOCK(td->td_proc);
1448 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1449 PROC_UNLOCK(td->td_proc);
1452 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1453 if (ptoa(pmap_wired_count(map->pmap)) + size >
1454 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1455 racct_set_force(td->td_proc, RACCT_VMEM,
1457 PROC_UNLOCK(td->td_proc);
1460 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1461 ptoa(pmap_wired_count(map->pmap)) + size);
1463 racct_set_force(td->td_proc, RACCT_VMEM,
1465 PROC_UNLOCK(td->td_proc);
1469 PROC_UNLOCK(td->td_proc);
1473 * We currently can only deal with page aligned file offsets.
1474 * The mmap() system call already enforces this by subtracting
1475 * the page offset from the file offset, but checking here
1476 * catches errors in device drivers (e.g. d_single_mmap()
1477 * callbacks) and other internal mapping requests (such as in
1480 if (foff & PAGE_MASK)
1483 if ((flags & MAP_FIXED) == 0) {
1485 *addr = round_page(*addr);
1487 if (*addr != trunc_page(*addr))
1492 if (flags & MAP_ANON) {
1493 if (object != NULL || foff != 0)
1496 } else if (flags & MAP_PREFAULT_READ)
1497 docow = MAP_PREFAULT;
1499 docow = MAP_PREFAULT_PARTIAL;
1501 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1502 docow |= MAP_COPY_ON_WRITE;
1503 if (flags & MAP_NOSYNC)
1504 docow |= MAP_DISABLE_SYNCER;
1505 if (flags & MAP_NOCORE)
1506 docow |= MAP_DISABLE_COREDUMP;
1507 /* Shared memory is also shared with children. */
1508 if (flags & MAP_SHARED)
1509 docow |= MAP_INHERIT_SHARE;
1511 docow |= MAP_VN_WRITECOUNT;
1512 if (flags & MAP_STACK) {
1515 docow |= MAP_STACK_GROWS_DOWN;
1517 if ((flags & MAP_EXCL) != 0)
1518 docow |= MAP_CHECK_EXCL;
1519 if ((flags & MAP_GUARD) != 0)
1520 docow |= MAP_CREATE_GUARD;
1523 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1524 findspace = VMFS_SUPER_SPACE;
1525 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1526 findspace = VMFS_ALIGNED_SPACE(flags >>
1527 MAP_ALIGNMENT_SHIFT);
1529 findspace = VMFS_OPTIMAL_SPACE;
1532 if ((flags & MAP_32BIT) != 0)
1533 max_addr = MAP_32BIT_MAX_ADDR;
1536 rv = vm_map_find_min(map, object, foff, addr, size,
1537 round_page((vm_offset_t)td->td_proc->p_vmspace->
1538 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1539 findspace, prot, maxprot, docow);
1541 rv = vm_map_find(map, object, foff, addr, size,
1542 max_addr, findspace, prot, maxprot, docow);
1545 rv = vm_map_fixed(map, object, foff, *addr, size,
1546 prot, maxprot, docow);
1549 if (rv == KERN_SUCCESS) {
1551 * If the process has requested that all future mappings
1552 * be wired, then heed this.
1554 if (map->flags & MAP_WIREFUTURE) {
1555 vm_map_wire(map, *addr, *addr + size,
1556 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1557 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1560 return (vm_mmap_to_errno(rv));
1564 * Translate a Mach VM return code to zero on success or the appropriate errno
1568 vm_mmap_to_errno(int rv)
1574 case KERN_INVALID_ADDRESS:
1577 case KERN_PROTECTION_FAILURE: