2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
40 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
41 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
42 * $DragonFly: src/sys/vm/vm_mmap.c,v 1.34 2006/09/17 21:09:40 dillon Exp $
46 * Mapped file (mmap) interface to VM
49 #include <sys/param.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/sysproto.h>
53 #include <sys/filedesc.h>
54 #include <sys/kern_syscall.h>
56 #include <sys/resource.h>
57 #include <sys/resourcevar.h>
58 #include <sys/vnode.h>
59 #include <sys/fcntl.h>
64 #include <sys/vmmeter.h>
65 #include <sys/sysctl.h>
68 #include <vm/vm_param.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vm_pageout.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_kern.h>
80 #include <sys/file2.h>
81 #include <sys/thread2.h>
83 static int max_proc_mmap;
84 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
86 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
89 * Set the maximum number of vm_map_entry structures per process. Roughly
90 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
91 * of our KVM malloc space still results in generous limits. We want a
92 * default that is good enough to prevent the kernel running out of resources
93 * if attacked from compromised user account but generous enough such that
94 * multi-threaded processes are not unduly inconvenienced.
97 static void vmmapentry_rsrc_init (void *);
98 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
101 vmmapentry_rsrc_init(void *dummy)
103 max_proc_mmap = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) /
104 sizeof(struct vm_map_entry);
105 max_proc_mmap /= 100;
110 sys_sbrk(struct sbrk_args *uap)
112 /* Not yet implemented */
117 * sstk_args(int incr)
121 sys_sstk(struct sstk_args *uap)
123 /* Not yet implemented */
128 * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
129 * long pad, off_t pos)
131 * Memory Map (mmap) system call. Note that the file offset
132 * and address are allowed to be NOT page aligned, though if
133 * the MAP_FIXED flag it set, both must have the same remainder
134 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
135 * page-aligned, the actual mapping starts at trunc_page(addr)
136 * and the return value is adjusted up by the page offset.
138 * Generally speaking, only character devices which are themselves
139 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
140 * there would be no cache coherency between a descriptor and a VM mapping
141 * both to the same character device.
143 * Block devices can be mmap'd no matter what they represent. Cache coherency
144 * is maintained as long as you do not write directly to the underlying
149 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
150 int uprot, int uflags, int fd, off_t upos, void **res)
152 struct thread *td = curthread;
153 struct proc *p = td->td_proc;
154 struct file *fp = NULL;
157 vm_size_t size, pageoff;
158 vm_prot_t prot, maxprot;
161 int disablexworkaround;
167 addr = (vm_offset_t) uaddr;
169 prot = uprot & VM_PROT_ALL;
173 /* make sure mapping fits into numeric range etc */
174 if ((ssize_t) ulen < 0 || ((flags & MAP_ANON) && fd != -1))
177 if (flags & MAP_STACK) {
179 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
186 * Virtual page tables cannot be used with MAP_STACK. Apart from
187 * it not making any sense, the aux union is used by both
190 * Because the virtual page table is stored in the backing object
191 * and might be updated by the kernel, the mapping must be R+W.
193 if (flags & MAP_VPAGETABLE) {
194 if (vkernel_enable == 0)
196 if (flags & MAP_STACK)
198 if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
203 * Align the file position to a page boundary,
204 * and save its page offset component.
206 pageoff = (pos & PAGE_MASK);
209 /* Adjust size for rounding (on both ends). */
210 size += pageoff; /* low end... */
211 size = (vm_size_t) round_page(size); /* hi end */
214 * Check for illegal addresses. Watch out for address wrap... Note
215 * that VM_*_ADDRESS are not constants due to casts (argh).
217 if (flags & MAP_FIXED) {
219 * The specified address must have the same remainder
220 * as the file offset taken modulo PAGE_SIZE, so it
221 * should be aligned after adjustment by pageoff.
224 if (addr & PAGE_MASK)
226 /* Address range must be all in user VM space. */
227 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
230 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
233 if (addr + size < addr)
237 * XXX for non-fixed mappings where no hint is provided or
238 * the hint would fall in the potential heap space,
239 * place it after the end of the largest possible heap.
241 * There should really be a pmap call to determine a reasonable
244 else if (addr == 0 ||
245 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
246 addr < round_page((vm_offset_t)vms->vm_daddr + maxdsiz)))
247 addr = round_page((vm_offset_t)vms->vm_daddr + maxdsiz);
249 if (flags & MAP_ANON) {
251 * Mapping blank space is trivial.
254 maxprot = VM_PROT_ALL;
258 * Mapping file, get fp for validation. Obtain vnode and make
259 * sure it is of appropriate type.
261 fp = holdfp(p->p_fd, fd, -1);
264 if (fp->f_type != DTYPE_VNODE) {
269 * POSIX shared-memory objects are defined to have
270 * kernel persistence, and are not defined to support
271 * read(2)/write(2) -- or even open(2). Thus, we can
272 * use MAP_ASYNC to trade on-disk coherence for speed.
273 * The shm_open(3) library routine turns on the FPOSIXSHM
274 * flag to request this behavior.
276 if (fp->f_flag & FPOSIXSHM)
278 vp = (struct vnode *) fp->f_data;
279 if (vp->v_type != VREG && vp->v_type != VCHR) {
283 if (vp->v_type == VREG) {
285 * Get the proper underlying object
287 if ((obj = vp->v_object) == NULL) {
291 KKASSERT(vp == (struct vnode *)obj->handle);
295 * XXX hack to handle use of /dev/zero to map anon memory (ala
298 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
300 maxprot = VM_PROT_ALL;
305 * cdevs does not provide private mappings of any kind.
308 * However, for XIG X server to continue to work,
309 * we should allow the superuser to do it anyway.
310 * We only allow it at securelevel < 1.
311 * (Because the XIG X server writes directly to video
312 * memory via /dev/mem, it should never work at any
314 * XXX this will have to go
316 if (securelevel >= 1)
317 disablexworkaround = 1;
319 disablexworkaround = suser(td);
320 if (vp->v_type == VCHR && disablexworkaround &&
321 (flags & (MAP_PRIVATE|MAP_COPY))) {
326 * Ensure that file and memory protections are
327 * compatible. Note that we only worry about
328 * writability if mapping is shared; in this case,
329 * current and max prot are dictated by the open file.
330 * XXX use the vnode instead? Problem is: what
331 * credentials do we use for determination? What if
332 * proc does a setuid?
334 maxprot = VM_PROT_EXECUTE; /* ??? */
335 if (fp->f_flag & FREAD) {
336 maxprot |= VM_PROT_READ;
337 } else if (prot & PROT_READ) {
342 * If we are sharing potential changes (either via
343 * MAP_SHARED or via the implicit sharing of character
344 * device mappings), and we are trying to get write
345 * permission although we opened it without asking
346 * for it, bail out. Check for superuser, only if
347 * we're at securelevel < 1, to allow the XIG X server
348 * to continue to work.
351 if ((flags & MAP_SHARED) != 0 ||
352 (vp->v_type == VCHR && disablexworkaround)) {
353 if ((fp->f_flag & FWRITE) != 0) {
355 if ((error = VOP_GETATTR(vp, &va))) {
359 (IMMUTABLE|APPEND)) == 0) {
360 maxprot |= VM_PROT_WRITE;
361 } else if (prot & PROT_WRITE) {
365 } else if ((prot & PROT_WRITE) != 0) {
370 maxprot |= VM_PROT_WRITE;
377 * Do not allow more then a certain number of vm_map_entry structures
378 * per process. Scale with the number of rforks sharing the map
379 * to make the limit reasonable for threads.
382 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
387 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
390 *res = (void *)(addr + pageoff);
398 sys_mmap(struct mmap_args *uap)
402 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
403 uap->prot, uap->flags,
404 uap->fd, uap->pos, &uap->sysmsg_resultp);
410 * msync_args(void *addr, int len, int flags)
413 sys_msync(struct msync_args *uap)
415 struct proc *p = curproc;
417 vm_size_t size, pageoff;
422 addr = (vm_offset_t) uap->addr;
426 pageoff = (addr & PAGE_MASK);
429 size = (vm_size_t) round_page(size);
430 if (addr + size < addr)
433 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
436 map = &p->p_vmspace->vm_map;
439 * XXX Gak! If size is zero we are supposed to sync "all modified
440 * pages with the region containing addr". Unfortunately, we don't
441 * really keep track of individual mmaps so we approximate by flushing
442 * the range of the map entry containing addr. This can be incorrect
443 * if the region splits or is coalesced with a neighbor.
446 vm_map_entry_t entry;
448 vm_map_lock_read(map);
449 rv = vm_map_lookup_entry(map, addr, &entry);
450 vm_map_unlock_read(map);
454 size = entry->end - entry->start;
458 * Clean the pages and interpret the return value.
460 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
461 (flags & MS_INVALIDATE) != 0);
466 case KERN_INVALID_ADDRESS:
467 return (EINVAL); /* Sun returns ENOMEM? */
478 * munmap_args(void *addr, size_t len)
481 sys_munmap(struct munmap_args *uap)
483 struct proc *p = curproc;
485 vm_size_t size, pageoff;
488 addr = (vm_offset_t) uap->addr;
491 pageoff = (addr & PAGE_MASK);
494 size = (vm_size_t) round_page(size);
495 if (addr + size < addr)
502 * Check for illegal addresses. Watch out for address wrap... Note
503 * that VM_*_ADDRESS are not constants due to casts (argh).
505 if (VM_MAXUSER_ADDRESS > 0 && addr + size > VM_MAXUSER_ADDRESS)
508 if (VM_MIN_ADDRESS > 0 && addr < VM_MIN_ADDRESS)
511 map = &p->p_vmspace->vm_map;
513 * Make sure entire range is allocated.
515 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE))
517 /* returns nothing but KERN_SUCCESS anyway */
518 vm_map_remove(map, addr, addr + size);
523 * mprotect_args(const void *addr, size_t len, int prot)
526 sys_mprotect(struct mprotect_args *uap)
528 struct proc *p = curproc;
530 vm_size_t size, pageoff;
533 addr = (vm_offset_t) uap->addr;
535 prot = uap->prot & VM_PROT_ALL;
536 #if defined(VM_PROT_READ_IS_EXEC)
537 if (prot & VM_PROT_READ)
538 prot |= VM_PROT_EXECUTE;
541 pageoff = (addr & PAGE_MASK);
544 size = (vm_size_t) round_page(size);
545 if (addr + size < addr)
548 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size, prot,
552 case KERN_PROTECTION_FAILURE:
559 * minherit_args(void *addr, size_t len, int inherit)
562 sys_minherit(struct minherit_args *uap)
564 struct proc *p = curproc;
566 vm_size_t size, pageoff;
567 vm_inherit_t inherit;
569 addr = (vm_offset_t)uap->addr;
571 inherit = uap->inherit;
573 pageoff = (addr & PAGE_MASK);
576 size = (vm_size_t) round_page(size);
577 if (addr + size < addr)
580 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr, addr+size,
584 case KERN_PROTECTION_FAILURE:
591 * madvise_args(void *addr, size_t len, int behav)
595 sys_madvise(struct madvise_args *uap)
597 struct proc *p = curproc;
598 vm_offset_t start, end;
601 * Check for illegal behavior
603 if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
606 * Check for illegal addresses. Watch out for address wrap... Note
607 * that VM_*_ADDRESS are not constants due to casts (argh).
609 if (VM_MAXUSER_ADDRESS > 0 &&
610 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
613 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
616 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
620 * Since this routine is only advisory, we default to conservative
623 start = trunc_page((vm_offset_t) uap->addr);
624 end = round_page((vm_offset_t) uap->addr + uap->len);
626 return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
631 * mcontrol_args(void *addr, size_t len, int behav, off_t value)
635 sys_mcontrol(struct mcontrol_args *uap)
637 struct proc *p = curproc;
638 vm_offset_t start, end;
641 * Check for illegal behavior
643 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
646 * Check for illegal addresses. Watch out for address wrap... Note
647 * that VM_*_ADDRESS are not constants due to casts (argh).
649 if (VM_MAXUSER_ADDRESS > 0 &&
650 ((vm_offset_t) uap->addr + uap->len) > VM_MAXUSER_ADDRESS)
653 if (VM_MIN_ADDRESS > 0 && uap->addr < VM_MIN_ADDRESS)
656 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
660 * Since this routine is only advisory, we default to conservative
663 start = trunc_page((vm_offset_t) uap->addr);
664 end = round_page((vm_offset_t) uap->addr + uap->len);
666 return (vm_map_madvise(&p->p_vmspace->vm_map, start, end,
667 uap->behav, uap->value));
672 * mincore_args(const void *addr, size_t len, char *vec)
676 sys_mincore(struct mincore_args *uap)
678 struct proc *p = curproc;
679 vm_offset_t addr, first_addr;
680 vm_offset_t end, cend;
685 int vecindex, lastvecindex;
686 vm_map_entry_t current;
687 vm_map_entry_t entry;
689 unsigned int timestamp;
692 * Make sure that the addresses presented are valid for user
695 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
696 end = addr + (vm_size_t)round_page(uap->len);
697 if (VM_MAXUSER_ADDRESS > 0 && end > VM_MAXUSER_ADDRESS)
703 * Address of byte vector
707 map = &p->p_vmspace->vm_map;
708 pmap = vmspace_pmap(p->p_vmspace);
710 vm_map_lock_read(map);
712 timestamp = map->timestamp;
714 if (!vm_map_lookup_entry(map, addr, &entry))
718 * Do this on a map entry basis so that if the pages are not
719 * in the current processes address space, we can easily look
720 * up the pages elsewhere.
724 (current != &map->header) && (current->start < end);
725 current = current->next) {
728 * ignore submaps (for now) or null objects
730 if (current->maptype != VM_MAPTYPE_NORMAL &&
731 current->maptype != VM_MAPTYPE_VPAGETABLE) {
734 if (current->object.vm_object == NULL)
738 * limit this scan to the current map entry and the
739 * limits for the mincore call
741 if (addr < current->start)
742 addr = current->start;
748 * scan this entry one page at a time
750 while (addr < cend) {
752 * Check pmap first, it is likely faster, also
753 * it can provide info as to whether we are the
754 * one referencing or modifying the page.
756 * If we have to check the VM object, only mess
757 * around with normal maps. Do not mess around
758 * with virtual page tables (XXX).
760 mincoreinfo = pmap_mincore(pmap, addr);
761 if (mincoreinfo == 0 &&
762 current->maptype == VM_MAPTYPE_NORMAL) {
768 * calculate the page index into the object
770 offset = current->offset + (addr - current->start);
771 pindex = OFF_TO_IDX(offset);
774 * if the page is resident, then gather
775 * information about it. spl protection is
776 * required to maintain the object
777 * association. And XXX what if the page is
778 * busy? What's the deal with that?
781 m = vm_page_lookup(current->object.vm_object,
784 mincoreinfo = MINCORE_INCORE;
787 mincoreinfo |= MINCORE_MODIFIED_OTHER;
788 if ((m->flags & PG_REFERENCED) ||
789 pmap_ts_referenced(m)) {
790 vm_page_flag_set(m, PG_REFERENCED);
791 mincoreinfo |= MINCORE_REFERENCED_OTHER;
798 * subyte may page fault. In case it needs to modify
799 * the map, we release the lock.
801 vm_map_unlock_read(map);
804 * calculate index into user supplied byte vector
806 vecindex = OFF_TO_IDX(addr - first_addr);
809 * If we have skipped map entries, we need to make sure that
810 * the byte vector is zeroed for those skipped entries.
812 while((lastvecindex + 1) < vecindex) {
813 error = subyte( vec + lastvecindex, 0);
821 * Pass the page information to the user
823 error = subyte( vec + vecindex, mincoreinfo);
829 * If the map has changed, due to the subyte, the previous
830 * output may be invalid.
832 vm_map_lock_read(map);
833 if (timestamp != map->timestamp)
836 lastvecindex = vecindex;
842 * subyte may page fault. In case it needs to modify
843 * the map, we release the lock.
845 vm_map_unlock_read(map);
848 * Zero the last entries in the byte vector.
850 vecindex = OFF_TO_IDX(end - first_addr);
851 while((lastvecindex + 1) < vecindex) {
852 error = subyte( vec + lastvecindex, 0);
860 * If the map has changed, due to the subyte, the previous
861 * output may be invalid.
863 vm_map_lock_read(map);
864 if (timestamp != map->timestamp)
866 vm_map_unlock_read(map);
872 * mlock_args(const void *addr, size_t len)
875 sys_mlock(struct mlock_args *uap)
878 vm_size_t size, pageoff;
880 struct proc *p = curproc;
882 addr = (vm_offset_t) uap->addr;
885 pageoff = (addr & PAGE_MASK);
888 size = (vm_size_t) round_page(size);
890 /* disable wrap around */
891 if (addr + size < addr)
894 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
897 #ifdef pmap_wired_count
898 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
899 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur)
902 error = suser_cred(p->p_ucred, 0);
907 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
908 return (error == KERN_SUCCESS ? 0 : ENOMEM);
912 * mlockall_args(int how)
915 sys_mlockall(struct mlockall_args *uap)
921 * munlockall_args(void)
924 sys_munlockall(struct munlockall_args *uap)
930 * munlock_args(const void *addr, size_t len)
933 sys_munlock(struct munlock_args *uap)
935 struct thread *td = curthread;
936 struct proc *p = td->td_proc;
938 vm_size_t size, pageoff;
941 addr = (vm_offset_t) uap->addr;
944 pageoff = (addr & PAGE_MASK);
947 size = (vm_size_t) round_page(size);
949 /* disable wrap around */
950 if (addr + size < addr)
953 #ifndef pmap_wired_count
959 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
960 return (error == KERN_SUCCESS ? 0 : ENOMEM);
964 * Internal version of mmap.
965 * Currently used by mmap, exec, and sys5 shared memory.
966 * Handle is either a vnode pointer or NULL for MAP_ANON.
969 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
970 vm_prot_t maxprot, int flags,
976 struct vnode *vp = NULL;
979 int rv = KERN_SUCCESS;
986 objsize = size = round_page(size);
989 * XXX messy code, fixme
991 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
992 if (map->size + size > p->p_rlimit[RLIMIT_VMEM].rlim_cur)
997 * We currently can only deal with page aligned file offsets.
998 * The check is here rather than in the syscall because the
999 * kernel calls this function internally for other mmaping
1000 * operations (such as in exec) and non-aligned offsets will
1001 * cause pmap inconsistencies...so we want to be sure to
1002 * disallow this in all cases.
1004 if (foff & PAGE_MASK)
1007 if ((flags & MAP_FIXED) == 0) {
1009 *addr = round_page(*addr);
1011 if (*addr != trunc_page(*addr))
1014 vm_map_remove(map, *addr, *addr + size);
1018 * Lookup/allocate object.
1020 if (flags & MAP_ANON) {
1021 type = OBJT_DEFAULT;
1023 * Unnamed anonymous regions always start at 0.
1028 vp = (struct vnode *) handle;
1029 if (vp->v_type == VCHR) {
1031 handle = (void *)(intptr_t)vp->v_rdev;
1036 error = VOP_GETATTR(vp, &vat);
1039 objsize = vat.va_size;
1042 * if it is a regular file without any references
1043 * we do not need to sync it.
1045 if (vp->v_type == VREG && vat.va_nlink == 0) {
1046 flags |= MAP_NOSYNC;
1051 if (handle == NULL) {
1055 object = vm_pager_allocate(type, handle, objsize, prot, foff);
1057 return (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1058 docow = MAP_PREFAULT_PARTIAL;
1062 * Force device mappings to be shared.
1064 if (type == OBJT_DEVICE || type == OBJT_PHYS) {
1065 flags &= ~(MAP_PRIVATE|MAP_COPY);
1066 flags |= MAP_SHARED;
1069 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1070 docow |= MAP_COPY_ON_WRITE;
1071 if (flags & MAP_NOSYNC)
1072 docow |= MAP_DISABLE_SYNCER;
1073 if (flags & MAP_NOCORE)
1074 docow |= MAP_DISABLE_COREDUMP;
1076 #if defined(VM_PROT_READ_IS_EXEC)
1077 if (prot & VM_PROT_READ)
1078 prot |= VM_PROT_EXECUTE;
1080 if (maxprot & VM_PROT_READ)
1081 maxprot |= VM_PROT_EXECUTE;
1085 *addr = pmap_addr_hint(object, *addr, size);
1089 * Stack mappings need special attention. Mappings that use virtual
1090 * page tables will default to storing the page table at offset 0.
1092 if (flags & MAP_STACK) {
1093 rv = vm_map_stack (map, *addr, size, prot, maxprot, docow);
1094 } else if (flags & MAP_VPAGETABLE) {
1095 rv = vm_map_find(map, object, foff, addr, size, fitit,
1096 VM_MAPTYPE_VPAGETABLE, prot, maxprot, docow);
1098 rv = vm_map_find(map, object, foff, addr, size, fitit,
1099 VM_MAPTYPE_NORMAL, prot, maxprot, docow);
1102 if (rv != KERN_SUCCESS) {
1104 * Lose the object reference. Will destroy the
1105 * object if it's an unnamed anonymous mapping
1106 * or named anonymous without other references.
1108 vm_object_deallocate(object);
1113 * Shared memory is also shared with children.
1115 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1116 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1117 if (rv != KERN_SUCCESS) {
1118 vm_map_remove(map, *addr, *addr + size);
1126 case KERN_INVALID_ADDRESS:
1129 case KERN_PROTECTION_FAILURE: