2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
40 * Mapped file (mmap) interface to VM
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/kernel.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
59 #include <sys/resource.h>
60 #include <sys/resourcevar.h>
61 #include <sys/vnode.h>
62 #include <sys/fcntl.h>
65 #include <sys/mount.h>
68 #include <sys/vmmeter.h>
69 #include <sys/sysctl.h>
71 #include <security/mac/mac_framework.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_pager.h>
80 #include <vm/vm_pageout.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_kern.h>
86 #include <sys/pmckern.h>
89 #ifndef _SYS_SYSPROTO_H_
95 static int max_proc_mmap;
96 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0,
97 "Maximum number of memory-mapped files per process");
100 * Set the maximum number of vm_map_entry structures per process. Roughly
101 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
102 * of our KVM malloc space still results in generous limits. We want a
103 * default that is good enough to prevent the kernel running out of resources
104 * if attacked from compromised user account but generous enough such that
105 * multi-threaded processes are not unduly inconvenienced.
107 static void vmmapentry_rsrc_init(void *);
108 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init,
112 vmmapentry_rsrc_init(dummy)
115 max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
116 max_proc_mmap /= 100;
119 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
120 int *, struct vnode *, vm_ooffset_t, vm_object_t *);
121 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
122 int *, struct cdev *, vm_ooffset_t, vm_object_t *);
123 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
124 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
133 struct sbrk_args *uap;
135 /* Not yet implemented */
139 #ifndef _SYS_SYSPROTO_H_
152 struct sstk_args *uap;
154 /* Not yet implemented */
158 #if defined(COMPAT_43)
159 #ifndef _SYS_SYSPROTO_H_
160 struct getpagesize_args {
167 ogetpagesize(td, uap)
169 struct getpagesize_args *uap;
172 td->td_retval[0] = PAGE_SIZE;
175 #endif /* COMPAT_43 */
179 * Memory Map (mmap) system call. Note that the file offset
180 * and address are allowed to be NOT page aligned, though if
181 * the MAP_FIXED flag it set, both must have the same remainder
182 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
183 * page-aligned, the actual mapping starts at trunc_page(addr)
184 * and the return value is adjusted up by the page offset.
186 * Generally speaking, only character devices which are themselves
187 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
188 * there would be no cache coherency between a descriptor and a VM mapping
189 * both to the same character device.
191 * Block devices can be mmap'd no matter what they represent. Cache coherency
192 * is maintained as long as you do not write directly to the underlying
195 #ifndef _SYS_SYSPROTO_H_
213 struct mmap_args *uap;
216 struct pmckern_map_in pkm;
221 vm_size_t size, pageoff;
222 vm_prot_t prot, maxprot;
224 objtype_t handle_type;
227 struct vmspace *vms = td->td_proc->p_vmspace;
229 addr = (vm_offset_t) uap->addr;
231 prot = uap->prot & VM_PROT_ALL;
236 /* make sure mapping fits into numeric range etc */
237 if ((ssize_t) uap->len < 0 ||
238 ((flags & MAP_ANON) && uap->fd != -1))
241 if (flags & MAP_STACK) {
242 if ((uap->fd != -1) ||
243 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
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 */
261 * Check for illegal addresses. Watch out for address wrap... Note
262 * that VM_*_ADDRESS are not constants due to casts (argh).
264 if (flags & MAP_FIXED) {
266 * The specified address must have the same remainder
267 * as the file offset taken modulo PAGE_SIZE, so it
268 * should be aligned after adjustment by pageoff.
271 if (addr & PAGE_MASK)
273 /* Address range must be all in user VM space. */
274 if (addr < vm_map_min(&vms->vm_map) ||
275 addr + size > vm_map_max(&vms->vm_map))
277 if (addr + size < addr)
281 * XXX for non-fixed mappings where no hint is provided or
282 * the hint would fall in the potential heap space,
283 * place it after the end of the largest possible heap.
285 * There should really be a pmap call to determine a reasonable
288 PROC_LOCK(td->td_proc);
290 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
291 addr < round_page((vm_offset_t)vms->vm_daddr +
292 lim_max(td->td_proc, RLIMIT_DATA))))
293 addr = round_page((vm_offset_t)vms->vm_daddr +
294 lim_max(td->td_proc, RLIMIT_DATA));
295 PROC_UNLOCK(td->td_proc);
297 if (flags & MAP_ANON) {
299 * Mapping blank space is trivial.
302 handle_type = OBJT_DEFAULT;
303 maxprot = VM_PROT_ALL;
307 * Mapping file, get fp for validation and
308 * don't let the descriptor disappear on us if we block.
310 if ((error = fget(td, uap->fd, &fp)) != 0)
312 if (fp->f_type == DTYPE_SHM) {
314 handle_type = OBJT_SWAP;
315 maxprot = VM_PROT_NONE;
317 /* FREAD should always be set. */
318 if (fp->f_flag & FREAD)
319 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
320 if (fp->f_flag & FWRITE)
321 maxprot |= VM_PROT_WRITE;
324 if (fp->f_type != DTYPE_VNODE) {
328 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
329 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
331 * POSIX shared-memory objects are defined to have
332 * kernel persistence, and are not defined to support
333 * read(2)/write(2) -- or even open(2). Thus, we can
334 * use MAP_ASYNC to trade on-disk coherence for speed.
335 * The shm_open(3) library routine turns on the FPOSIXSHM
336 * flag to request this behavior.
338 if (fp->f_flag & FPOSIXSHM)
343 * Ensure that file and memory protections are
344 * compatible. Note that we only worry about
345 * writability if mapping is shared; in this case,
346 * current and max prot are dictated by the open file.
347 * XXX use the vnode instead? Problem is: what
348 * credentials do we use for determination? What if
349 * proc does a setuid?
351 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
352 maxprot = VM_PROT_NONE;
354 maxprot = VM_PROT_EXECUTE;
355 if (fp->f_flag & FREAD) {
356 maxprot |= VM_PROT_READ;
357 } else if (prot & PROT_READ) {
362 * If we are sharing potential changes (either via
363 * MAP_SHARED or via the implicit sharing of character
364 * device mappings), and we are trying to get write
365 * permission although we opened it without asking
368 if ((flags & MAP_SHARED) != 0) {
369 if ((fp->f_flag & FWRITE) != 0) {
370 maxprot |= VM_PROT_WRITE;
371 } else if ((prot & PROT_WRITE) != 0) {
375 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
376 maxprot |= VM_PROT_WRITE;
379 handle_type = OBJT_VNODE;
384 * Do not allow more then a certain number of vm_map_entry structures
385 * per process. Scale with the number of rforks sharing the map
386 * to make the limit reasonable for threads.
389 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
394 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
395 flags, handle_type, handle, pos);
397 /* inform hwpmc(4) if an executable is being mapped */
398 if (error == 0 && handle_type == OBJT_VNODE &&
399 (prot & PROT_EXEC)) {
400 pkm.pm_file = handle;
401 pkm.pm_address = (uintptr_t) addr;
402 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
406 td->td_retval[0] = (register_t) (addr + pageoff);
415 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
417 struct mmap_args oargs;
419 oargs.addr = uap->addr;
420 oargs.len = uap->len;
421 oargs.prot = uap->prot;
422 oargs.flags = uap->flags;
424 oargs.pos = uap->pos;
425 return (mmap(td, &oargs));
429 #ifndef _SYS_SYSPROTO_H_
442 struct ommap_args *uap;
444 struct mmap_args nargs;
445 static const char cvtbsdprot[8] = {
449 PROT_EXEC | PROT_WRITE,
451 PROT_EXEC | PROT_READ,
452 PROT_WRITE | PROT_READ,
453 PROT_EXEC | PROT_WRITE | PROT_READ,
456 #define OMAP_ANON 0x0002
457 #define OMAP_COPY 0x0020
458 #define OMAP_SHARED 0x0010
459 #define OMAP_FIXED 0x0100
461 nargs.addr = uap->addr;
462 nargs.len = uap->len;
463 nargs.prot = cvtbsdprot[uap->prot & 0x7];
465 if (uap->flags & OMAP_ANON)
466 nargs.flags |= MAP_ANON;
467 if (uap->flags & OMAP_COPY)
468 nargs.flags |= MAP_COPY;
469 if (uap->flags & OMAP_SHARED)
470 nargs.flags |= MAP_SHARED;
472 nargs.flags |= MAP_PRIVATE;
473 if (uap->flags & OMAP_FIXED)
474 nargs.flags |= MAP_FIXED;
476 nargs.pos = uap->pos;
477 return (mmap(td, &nargs));
479 #endif /* COMPAT_43 */
482 #ifndef _SYS_SYSPROTO_H_
495 struct msync_args *uap;
498 vm_size_t size, pageoff;
503 addr = (vm_offset_t) uap->addr;
507 pageoff = (addr & PAGE_MASK);
510 size = (vm_size_t) round_page(size);
511 if (addr + size < addr)
514 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
517 map = &td->td_proc->p_vmspace->vm_map;
520 * Clean the pages and interpret the return value.
522 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
523 (flags & MS_INVALIDATE) != 0);
527 case KERN_INVALID_ADDRESS:
528 return (EINVAL); /* Sun returns ENOMEM? */
529 case KERN_INVALID_ARGUMENT:
536 #ifndef _SYS_SYSPROTO_H_
548 struct munmap_args *uap;
551 struct pmckern_map_out pkm;
552 vm_map_entry_t entry;
555 vm_size_t size, pageoff;
558 addr = (vm_offset_t) uap->addr;
563 pageoff = (addr & PAGE_MASK);
566 size = (vm_size_t) round_page(size);
567 if (addr + size < addr)
571 * Check for illegal addresses. Watch out for address wrap...
573 map = &td->td_proc->p_vmspace->vm_map;
574 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
579 * Inform hwpmc if the address range being unmapped contains
580 * an executable region.
582 if (vm_map_lookup_entry(map, addr, &entry)) {
584 entry != &map->header && entry->start < addr + size;
585 entry = entry->next) {
586 if (vm_map_check_protection(map, entry->start,
587 entry->end, VM_PROT_EXECUTE) == TRUE) {
588 pkm.pm_address = (uintptr_t) addr;
589 pkm.pm_size = (size_t) size;
590 PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
597 /* returns nothing but KERN_SUCCESS anyway */
598 vm_map_delete(map, addr, addr + size);
603 #ifndef _SYS_SYSPROTO_H_
604 struct mprotect_args {
616 struct mprotect_args *uap;
619 vm_size_t size, pageoff;
622 addr = (vm_offset_t) uap->addr;
624 prot = uap->prot & VM_PROT_ALL;
625 #if defined(VM_PROT_READ_IS_EXEC)
626 if (prot & VM_PROT_READ)
627 prot |= VM_PROT_EXECUTE;
630 pageoff = (addr & PAGE_MASK);
633 size = (vm_size_t) round_page(size);
634 if (addr + size < addr)
637 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
638 addr + size, prot, FALSE)) {
641 case KERN_PROTECTION_FAILURE:
647 #ifndef _SYS_SYSPROTO_H_
648 struct minherit_args {
660 struct minherit_args *uap;
663 vm_size_t size, pageoff;
664 vm_inherit_t inherit;
666 addr = (vm_offset_t)uap->addr;
668 inherit = uap->inherit;
670 pageoff = (addr & PAGE_MASK);
673 size = (vm_size_t) round_page(size);
674 if (addr + size < addr)
677 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
678 addr + size, inherit)) {
681 case KERN_PROTECTION_FAILURE:
687 #ifndef _SYS_SYSPROTO_H_
688 struct madvise_args {
702 struct madvise_args *uap;
704 vm_offset_t start, end;
710 * Check for our special case, advising the swap pager we are
713 if (uap->behav == MADV_PROTECT) {
714 error = priv_check(td, PRIV_VM_MADV_PROTECT);
718 p->p_flag |= P_PROTECTED;
724 * Check for illegal behavior
726 if (uap->behav < 0 || uap->behav > MADV_CORE)
729 * Check for illegal addresses. Watch out for address wrap... Note
730 * that VM_*_ADDRESS are not constants due to casts (argh).
732 map = &td->td_proc->p_vmspace->vm_map;
733 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
734 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
736 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
740 * Since this routine is only advisory, we default to conservative
743 start = trunc_page((vm_offset_t) uap->addr);
744 end = round_page((vm_offset_t) uap->addr + uap->len);
746 if (vm_map_madvise(map, start, end, uap->behav))
751 #ifndef _SYS_SYSPROTO_H_
752 struct mincore_args {
766 struct mincore_args *uap;
768 vm_offset_t addr, first_addr;
769 vm_offset_t end, cend;
774 int vecindex, lastvecindex;
775 vm_map_entry_t current;
776 vm_map_entry_t entry;
778 unsigned int timestamp;
781 * Make sure that the addresses presented are valid for user
784 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
785 end = addr + (vm_size_t)round_page(uap->len);
786 map = &td->td_proc->p_vmspace->vm_map;
787 if (end > vm_map_max(map) || end < addr)
791 * Address of byte vector
795 pmap = vmspace_pmap(td->td_proc->p_vmspace);
797 vm_map_lock_read(map);
799 timestamp = map->timestamp;
801 if (!vm_map_lookup_entry(map, addr, &entry)) {
802 vm_map_unlock_read(map);
807 * Do this on a map entry basis so that if the pages are not
808 * in the current processes address space, we can easily look
809 * up the pages elsewhere.
812 for (current = entry;
813 (current != &map->header) && (current->start < end);
814 current = current->next) {
817 * check for contiguity
819 if (current->end < end &&
820 (entry->next == &map->header ||
821 current->next->start > current->end)) {
822 vm_map_unlock_read(map);
827 * ignore submaps (for now) or null objects
829 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
830 current->object.vm_object == NULL)
834 * limit this scan to the current map entry and the
835 * limits for the mincore call
837 if (addr < current->start)
838 addr = current->start;
844 * scan this entry one page at a time
846 while (addr < cend) {
848 * Check pmap first, it is likely faster, also
849 * it can provide info as to whether we are the
850 * one referencing or modifying the page.
852 mincoreinfo = pmap_mincore(pmap, addr);
858 * calculate the page index into the object
860 offset = current->offset + (addr - current->start);
861 pindex = OFF_TO_IDX(offset);
862 VM_OBJECT_LOCK(current->object.vm_object);
863 m = vm_page_lookup(current->object.vm_object,
866 * if the page is resident, then gather information about
869 if (m != NULL && m->valid != 0) {
870 mincoreinfo = MINCORE_INCORE;
871 vm_page_lock_queues();
874 mincoreinfo |= MINCORE_MODIFIED_OTHER;
875 if ((m->flags & PG_REFERENCED) ||
876 pmap_ts_referenced(m)) {
877 vm_page_flag_set(m, PG_REFERENCED);
878 mincoreinfo |= MINCORE_REFERENCED_OTHER;
880 vm_page_unlock_queues();
882 VM_OBJECT_UNLOCK(current->object.vm_object);
886 * subyte may page fault. In case it needs to modify
887 * the map, we release the lock.
889 vm_map_unlock_read(map);
892 * calculate index into user supplied byte vector
894 vecindex = OFF_TO_IDX(addr - first_addr);
897 * If we have skipped map entries, we need to make sure that
898 * the byte vector is zeroed for those skipped entries.
900 while ((lastvecindex + 1) < vecindex) {
901 error = subyte(vec + lastvecindex, 0);
910 * Pass the page information to the user
912 error = subyte(vec + vecindex, mincoreinfo);
919 * If the map has changed, due to the subyte, the previous
920 * output may be invalid.
922 vm_map_lock_read(map);
923 if (timestamp != map->timestamp)
926 lastvecindex = vecindex;
932 * subyte may page fault. In case it needs to modify
933 * the map, we release the lock.
935 vm_map_unlock_read(map);
938 * Zero the last entries in the byte vector.
940 vecindex = OFF_TO_IDX(end - first_addr);
941 while ((lastvecindex + 1) < vecindex) {
942 error = subyte(vec + lastvecindex, 0);
951 * If the map has changed, due to the subyte, the previous
952 * output may be invalid.
954 vm_map_lock_read(map);
955 if (timestamp != map->timestamp)
957 vm_map_unlock_read(map);
962 #ifndef _SYS_SYSPROTO_H_
974 struct mlock_args *uap;
977 vm_offset_t addr, end, last, start;
978 vm_size_t npages, size;
981 error = priv_check(td, PRIV_VM_MLOCK);
984 addr = (vm_offset_t)uap->addr;
987 start = trunc_page(addr);
988 end = round_page(last);
989 if (last < addr || end < addr)
991 npages = atop(end - start);
992 if (npages > vm_page_max_wired)
997 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
998 lim_cur(proc, RLIMIT_MEMLOCK)) {
1003 if (npages + cnt.v_wire_count > vm_page_max_wired)
1005 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1006 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1007 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1010 #ifndef _SYS_SYSPROTO_H_
1011 struct mlockall_args {
1022 struct mlockall_args *uap;
1027 map = &td->td_proc->p_vmspace->vm_map;
1030 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1035 * If wiring all pages in the process would cause it to exceed
1036 * a hard resource limit, return ENOMEM.
1038 PROC_LOCK(td->td_proc);
1039 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1040 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1041 PROC_UNLOCK(td->td_proc);
1044 PROC_UNLOCK(td->td_proc);
1046 error = priv_check(td, PRIV_VM_MLOCK);
1051 if (uap->how & MCL_FUTURE) {
1053 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1058 if (uap->how & MCL_CURRENT) {
1060 * P1003.1-2001 mandates that all currently mapped pages
1061 * will be memory resident and locked (wired) upon return
1062 * from mlockall(). vm_map_wire() will wire pages, by
1063 * calling vm_fault_wire() for each page in the region.
1065 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1066 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1067 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1073 #ifndef _SYS_SYSPROTO_H_
1074 struct munlockall_args {
1085 struct munlockall_args *uap;
1090 map = &td->td_proc->p_vmspace->vm_map;
1091 error = priv_check(td, PRIV_VM_MUNLOCK);
1095 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1097 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1100 /* Forcibly unwire all pages. */
1101 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1102 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1107 #ifndef _SYS_SYSPROTO_H_
1108 struct munlock_args {
1119 struct munlock_args *uap;
1121 vm_offset_t addr, end, last, start;
1125 error = priv_check(td, PRIV_VM_MUNLOCK);
1128 addr = (vm_offset_t)uap->addr;
1131 start = trunc_page(addr);
1132 end = round_page(last);
1133 if (last < addr || end < addr)
1135 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1136 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1137 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1145 * Helper function for vm_mmap. Perform sanity check specific for mmap
1146 * operations on vnodes.
1149 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1150 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1151 struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1158 int error, flags, type;
1162 vfslocked = VFS_LOCK_GIANT(mp);
1163 if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1164 VFS_UNLOCK_GIANT(vfslocked);
1169 if (vp->v_type == VREG) {
1171 * Get the proper underlying object
1177 if (obj->handle != vp) {
1179 vp = (struct vnode*)obj->handle;
1180 vget(vp, LK_EXCLUSIVE, td);
1184 } else if (vp->v_type == VCHR) {
1186 handle = vp->v_rdev;
1188 dsw = dev_refthread(handle);
1193 if (dsw->d_flags & D_MMAP_ANON) {
1194 dev_relthread(handle);
1195 *maxprotp = VM_PROT_ALL;
1196 *flagsp |= MAP_ANON;
1200 dev_relthread(handle);
1202 * cdevs does not provide private mappings of any kind.
1204 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1205 (prot & PROT_WRITE) != 0) {
1209 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1214 * Force device mappings to be shared.
1216 flags |= MAP_SHARED;
1221 if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1225 error = mac_vnode_check_mmap(td->td_ucred, vp, prot, flags);
1229 if ((flags & MAP_SHARED) != 0) {
1230 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1231 if (prot & PROT_WRITE) {
1235 *maxprotp &= ~VM_PROT_WRITE;
1239 * If it is a regular file without any references
1240 * we do not need to sync it.
1241 * Adjust object size to be the size of actual file.
1243 if (vp->v_type == VREG) {
1244 objsize = round_page(va.va_size);
1245 if (va.va_nlink == 0)
1246 flags |= MAP_NOSYNC;
1248 obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1250 error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1255 vfs_mark_atime(vp, td);
1259 VFS_UNLOCK_GIANT(vfslocked);
1268 * Helper function for vm_mmap. Perform sanity check specific for mmap
1269 * operations on cdevs.
1272 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1273 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1274 struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1282 dsw = dev_refthread(cdev);
1285 if (dsw->d_flags & D_MMAP_ANON) {
1286 dev_relthread(cdev);
1287 *maxprotp = VM_PROT_ALL;
1288 *flagsp |= MAP_ANON;
1291 dev_relthread(cdev);
1293 * cdevs does not provide private mappings of any kind.
1295 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1296 (prot & PROT_WRITE) != 0)
1298 if (flags & (MAP_PRIVATE|MAP_COPY))
1301 * Force device mappings to be shared.
1303 flags |= MAP_SHARED;
1305 error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1309 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1322 * Helper function for vm_mmap. Perform sanity check specific for mmap
1323 * operations on shm file descriptors.
1326 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1327 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1328 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1332 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1333 (prot & PROT_WRITE) != 0)
1336 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1340 error = shm_mmap(shmfd, objsize, foff, objp);
1351 * Internal version of mmap. Currently used by mmap, exec, and sys5
1352 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1355 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1356 vm_prot_t maxprot, int flags,
1357 objtype_t handle_type, void *handle,
1361 vm_object_t object = NULL;
1362 int rv = KERN_SUCCESS;
1364 struct thread *td = curthread;
1369 size = round_page(size);
1371 PROC_LOCK(td->td_proc);
1372 if (td->td_proc->p_vmspace->vm_map.size + size >
1373 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1374 PROC_UNLOCK(td->td_proc);
1377 PROC_UNLOCK(td->td_proc);
1380 * We currently can only deal with page aligned file offsets.
1381 * The check is here rather than in the syscall because the
1382 * kernel calls this function internally for other mmaping
1383 * operations (such as in exec) and non-aligned offsets will
1384 * cause pmap inconsistencies...so we want to be sure to
1385 * disallow this in all cases.
1387 if (foff & PAGE_MASK)
1390 if ((flags & MAP_FIXED) == 0) {
1392 *addr = round_page(*addr);
1394 if (*addr != trunc_page(*addr))
1399 * Lookup/allocate object.
1401 switch (handle_type) {
1403 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1404 handle, foff, &object);
1407 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1408 handle, foff, &object);
1411 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1412 handle, foff, &object);
1415 if (handle == NULL) {
1426 if (flags & MAP_ANON) {
1430 * Unnamed anonymous regions always start at 0.
1435 docow = MAP_PREFAULT_PARTIAL;
1438 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1439 docow |= MAP_COPY_ON_WRITE;
1440 if (flags & MAP_NOSYNC)
1441 docow |= MAP_DISABLE_SYNCER;
1442 if (flags & MAP_NOCORE)
1443 docow |= MAP_DISABLE_COREDUMP;
1445 #if defined(VM_PROT_READ_IS_EXEC)
1446 if (prot & VM_PROT_READ)
1447 prot |= VM_PROT_EXECUTE;
1449 if (maxprot & VM_PROT_READ)
1450 maxprot |= VM_PROT_EXECUTE;
1453 if (flags & MAP_STACK)
1454 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1455 docow | MAP_STACK_GROWS_DOWN);
1457 rv = vm_map_find(map, object, foff, addr, size,
1458 object != NULL && object->type == OBJT_DEVICE ?
1459 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1461 rv = vm_map_fixed(map, object, foff, *addr, size,
1462 prot, maxprot, docow);
1464 if (rv != KERN_SUCCESS) {
1466 * Lose the object reference. Will destroy the
1467 * object if it's an unnamed anonymous mapping
1468 * or named anonymous without other references.
1470 vm_object_deallocate(object);
1471 } else if (flags & MAP_SHARED) {
1473 * Shared memory is also shared with children.
1475 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1476 if (rv != KERN_SUCCESS)
1477 (void) vm_map_remove(map, *addr, *addr + size);
1481 * If the process has requested that all future mappings
1482 * be wired, then heed this.
1484 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1485 vm_map_wire(map, *addr, *addr + size,
1486 VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1491 case KERN_INVALID_ADDRESS:
1494 case KERN_PROTECTION_FAILURE: