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"
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/kernel.h>
53 #include <sys/mutex.h>
54 #include <sys/sysproto.h>
55 #include <sys/filedesc.h>
58 #include <sys/resource.h>
59 #include <sys/resourcevar.h>
60 #include <sys/vnode.h>
61 #include <sys/fcntl.h>
64 #include <sys/mount.h>
67 #include <sys/sysent.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 #ifndef _SYS_SYSPROTO_H_
209 struct mmap_args *uap;
212 struct pmckern_map_in pkm;
217 vm_size_t size, pageoff;
218 vm_prot_t prot, maxprot;
220 objtype_t handle_type;
223 struct vmspace *vms = td->td_proc->p_vmspace;
225 addr = (vm_offset_t) uap->addr;
227 prot = uap->prot & VM_PROT_ALL;
232 /* make sure mapping fits into numeric range etc */
233 if ((uap->len == 0 && !SV_CURPROC_FLAG(SV_AOUT) &&
234 curproc->p_osrel >= 800104) ||
235 ((flags & MAP_ANON) && (uap->fd != -1 || pos != 0)))
238 if (flags & MAP_STACK) {
239 if ((uap->fd != -1) ||
240 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
247 * Align the file position to a page boundary,
248 * and save its page offset component.
250 pageoff = (pos & PAGE_MASK);
253 /* Adjust size for rounding (on both ends). */
254 size += pageoff; /* low end... */
255 size = (vm_size_t) round_page(size); /* hi end */
258 * Check for illegal addresses. Watch out for address wrap... Note
259 * that VM_*_ADDRESS are not constants due to casts (argh).
261 if (flags & MAP_FIXED) {
263 * The specified address must have the same remainder
264 * as the file offset taken modulo PAGE_SIZE, so it
265 * should be aligned after adjustment by pageoff.
268 if (addr & PAGE_MASK)
270 /* Address range must be all in user VM space. */
271 if (addr < vm_map_min(&vms->vm_map) ||
272 addr + size > vm_map_max(&vms->vm_map))
274 if (addr + size < addr)
278 * XXX for non-fixed mappings where no hint is provided or
279 * the hint would fall in the potential heap space,
280 * place it after the end of the largest possible heap.
282 * There should really be a pmap call to determine a reasonable
285 PROC_LOCK(td->td_proc);
287 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
288 addr < round_page((vm_offset_t)vms->vm_daddr +
289 lim_max(td->td_proc, RLIMIT_DATA))))
290 addr = round_page((vm_offset_t)vms->vm_daddr +
291 lim_max(td->td_proc, RLIMIT_DATA));
292 PROC_UNLOCK(td->td_proc);
294 if (flags & MAP_ANON) {
296 * Mapping blank space is trivial.
299 handle_type = OBJT_DEFAULT;
300 maxprot = VM_PROT_ALL;
303 * Mapping file, get fp for validation and
304 * don't let the descriptor disappear on us if we block.
306 if ((error = fget(td, uap->fd, &fp)) != 0)
308 if (fp->f_type == DTYPE_SHM) {
310 handle_type = OBJT_SWAP;
311 maxprot = VM_PROT_NONE;
313 /* FREAD should always be set. */
314 if (fp->f_flag & FREAD)
315 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
316 if (fp->f_flag & FWRITE)
317 maxprot |= VM_PROT_WRITE;
320 if (fp->f_type != DTYPE_VNODE) {
324 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
325 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
327 * POSIX shared-memory objects are defined to have
328 * kernel persistence, and are not defined to support
329 * read(2)/write(2) -- or even open(2). Thus, we can
330 * use MAP_ASYNC to trade on-disk coherence for speed.
331 * The shm_open(3) library routine turns on the FPOSIXSHM
332 * flag to request this behavior.
334 if (fp->f_flag & FPOSIXSHM)
339 * Ensure that file and memory protections are
340 * compatible. Note that we only worry about
341 * writability if mapping is shared; in this case,
342 * current and max prot are dictated by the open file.
343 * XXX use the vnode instead? Problem is: what
344 * credentials do we use for determination? What if
345 * proc does a setuid?
347 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
348 maxprot = VM_PROT_NONE;
350 maxprot = VM_PROT_EXECUTE;
351 if (fp->f_flag & FREAD) {
352 maxprot |= VM_PROT_READ;
353 } else if (prot & PROT_READ) {
358 * If we are sharing potential changes (either via
359 * MAP_SHARED or via the implicit sharing of character
360 * device mappings), and we are trying to get write
361 * permission although we opened it without asking
364 if ((flags & MAP_SHARED) != 0) {
365 if ((fp->f_flag & FWRITE) != 0) {
366 maxprot |= VM_PROT_WRITE;
367 } else if ((prot & PROT_WRITE) != 0) {
371 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
372 maxprot |= VM_PROT_WRITE;
375 handle_type = OBJT_VNODE;
380 * Do not allow more then a certain number of vm_map_entry structures
381 * per process. Scale with the number of rforks sharing the map
382 * to make the limit reasonable for threads.
385 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
391 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
392 flags, handle_type, handle, pos);
395 /* inform hwpmc(4) if an executable is being mapped */
396 if (error == 0 && handle_type == OBJT_VNODE &&
397 (prot & PROT_EXEC)) {
398 pkm.pm_file = handle;
399 pkm.pm_address = (uintptr_t) addr;
400 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
404 td->td_retval[0] = (register_t) (addr + pageoff);
413 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
415 struct mmap_args oargs;
417 oargs.addr = uap->addr;
418 oargs.len = uap->len;
419 oargs.prot = uap->prot;
420 oargs.flags = uap->flags;
422 oargs.pos = uap->pos;
423 return (mmap(td, &oargs));
427 #ifndef _SYS_SYSPROTO_H_
440 struct ommap_args *uap;
442 struct mmap_args nargs;
443 static const char cvtbsdprot[8] = {
447 PROT_EXEC | PROT_WRITE,
449 PROT_EXEC | PROT_READ,
450 PROT_WRITE | PROT_READ,
451 PROT_EXEC | PROT_WRITE | PROT_READ,
454 #define OMAP_ANON 0x0002
455 #define OMAP_COPY 0x0020
456 #define OMAP_SHARED 0x0010
457 #define OMAP_FIXED 0x0100
459 nargs.addr = uap->addr;
460 nargs.len = uap->len;
461 nargs.prot = cvtbsdprot[uap->prot & 0x7];
463 if (uap->flags & OMAP_ANON)
464 nargs.flags |= MAP_ANON;
465 if (uap->flags & OMAP_COPY)
466 nargs.flags |= MAP_COPY;
467 if (uap->flags & OMAP_SHARED)
468 nargs.flags |= MAP_SHARED;
470 nargs.flags |= MAP_PRIVATE;
471 if (uap->flags & OMAP_FIXED)
472 nargs.flags |= MAP_FIXED;
474 nargs.pos = uap->pos;
475 return (mmap(td, &nargs));
477 #endif /* COMPAT_43 */
480 #ifndef _SYS_SYSPROTO_H_
493 struct msync_args *uap;
496 vm_size_t size, pageoff;
501 addr = (vm_offset_t) uap->addr;
505 pageoff = (addr & PAGE_MASK);
508 size = (vm_size_t) round_page(size);
509 if (addr + size < addr)
512 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
515 map = &td->td_proc->p_vmspace->vm_map;
518 * Clean the pages and interpret the return value.
520 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
521 (flags & MS_INVALIDATE) != 0);
525 case KERN_INVALID_ADDRESS:
526 return (EINVAL); /* Sun returns ENOMEM? */
527 case KERN_INVALID_ARGUMENT:
534 #ifndef _SYS_SYSPROTO_H_
546 struct munmap_args *uap;
549 struct pmckern_map_out pkm;
550 vm_map_entry_t entry;
553 vm_size_t size, pageoff;
556 addr = (vm_offset_t) uap->addr;
561 pageoff = (addr & PAGE_MASK);
564 size = (vm_size_t) round_page(size);
565 if (addr + size < addr)
569 * Check for illegal addresses. Watch out for address wrap...
571 map = &td->td_proc->p_vmspace->vm_map;
572 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
577 * Inform hwpmc if the address range being unmapped contains
578 * an executable region.
580 if (vm_map_lookup_entry(map, addr, &entry)) {
582 entry != &map->header && entry->start < addr + size;
583 entry = entry->next) {
584 if (vm_map_check_protection(map, entry->start,
585 entry->end, VM_PROT_EXECUTE) == TRUE) {
586 pkm.pm_address = (uintptr_t) addr;
587 pkm.pm_size = (size_t) size;
588 PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
595 /* returns nothing but KERN_SUCCESS anyway */
596 vm_map_delete(map, addr, addr + size);
601 #ifndef _SYS_SYSPROTO_H_
602 struct mprotect_args {
614 struct mprotect_args *uap;
617 vm_size_t size, pageoff;
620 addr = (vm_offset_t) uap->addr;
622 prot = uap->prot & VM_PROT_ALL;
624 pageoff = (addr & PAGE_MASK);
627 size = (vm_size_t) round_page(size);
628 if (addr + size < addr)
631 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
632 addr + size, prot, FALSE)) {
635 case KERN_PROTECTION_FAILURE:
637 case KERN_RESOURCE_SHORTAGE:
643 #ifndef _SYS_SYSPROTO_H_
644 struct minherit_args {
656 struct minherit_args *uap;
659 vm_size_t size, pageoff;
660 vm_inherit_t inherit;
662 addr = (vm_offset_t)uap->addr;
664 inherit = uap->inherit;
666 pageoff = (addr & PAGE_MASK);
669 size = (vm_size_t) round_page(size);
670 if (addr + size < addr)
673 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
674 addr + size, inherit)) {
677 case KERN_PROTECTION_FAILURE:
683 #ifndef _SYS_SYSPROTO_H_
684 struct madvise_args {
698 struct madvise_args *uap;
700 vm_offset_t start, end;
706 * Check for our special case, advising the swap pager we are
709 if (uap->behav == MADV_PROTECT) {
710 error = priv_check(td, PRIV_VM_MADV_PROTECT);
714 p->p_flag |= P_PROTECTED;
720 * Check for illegal behavior
722 if (uap->behav < 0 || uap->behav > MADV_CORE)
725 * Check for illegal addresses. Watch out for address wrap... Note
726 * that VM_*_ADDRESS are not constants due to casts (argh).
728 map = &td->td_proc->p_vmspace->vm_map;
729 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
730 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
732 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
736 * Since this routine is only advisory, we default to conservative
739 start = trunc_page((vm_offset_t) uap->addr);
740 end = round_page((vm_offset_t) uap->addr + uap->len);
742 if (vm_map_madvise(map, start, end, uap->behav))
747 #ifndef _SYS_SYSPROTO_H_
748 struct mincore_args {
762 struct mincore_args *uap;
764 vm_offset_t addr, first_addr;
765 vm_offset_t end, cend;
770 int vecindex, lastvecindex;
771 vm_map_entry_t current;
772 vm_map_entry_t entry;
774 unsigned int timestamp;
777 * Make sure that the addresses presented are valid for user
780 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
781 end = addr + (vm_size_t)round_page(uap->len);
782 map = &td->td_proc->p_vmspace->vm_map;
783 if (end > vm_map_max(map) || end < addr)
787 * Address of byte vector
791 pmap = vmspace_pmap(td->td_proc->p_vmspace);
793 vm_map_lock_read(map);
795 timestamp = map->timestamp;
797 if (!vm_map_lookup_entry(map, addr, &entry)) {
798 vm_map_unlock_read(map);
803 * Do this on a map entry basis so that if the pages are not
804 * in the current processes address space, we can easily look
805 * up the pages elsewhere.
808 for (current = entry;
809 (current != &map->header) && (current->start < end);
810 current = current->next) {
813 * check for contiguity
815 if (current->end < end &&
816 (entry->next == &map->header ||
817 current->next->start > current->end)) {
818 vm_map_unlock_read(map);
823 * ignore submaps (for now) or null objects
825 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
826 current->object.vm_object == NULL)
830 * limit this scan to the current map entry and the
831 * limits for the mincore call
833 if (addr < current->start)
834 addr = current->start;
840 * scan this entry one page at a time
842 while (addr < cend) {
844 * Check pmap first, it is likely faster, also
845 * it can provide info as to whether we are the
846 * one referencing or modifying the page.
848 mincoreinfo = pmap_mincore(pmap, addr);
854 * calculate the page index into the object
856 offset = current->offset + (addr - current->start);
857 pindex = OFF_TO_IDX(offset);
858 VM_OBJECT_LOCK(current->object.vm_object);
859 m = vm_page_lookup(current->object.vm_object,
862 * if the page is resident, then gather information about
865 if (m != NULL && m->valid != 0) {
866 mincoreinfo = MINCORE_INCORE;
867 vm_page_lock_queues();
870 mincoreinfo |= MINCORE_MODIFIED_OTHER;
871 if ((m->flags & PG_REFERENCED) ||
872 pmap_ts_referenced(m)) {
873 vm_page_flag_set(m, PG_REFERENCED);
874 mincoreinfo |= MINCORE_REFERENCED_OTHER;
876 vm_page_unlock_queues();
878 VM_OBJECT_UNLOCK(current->object.vm_object);
882 * subyte may page fault. In case it needs to modify
883 * the map, we release the lock.
885 vm_map_unlock_read(map);
888 * calculate index into user supplied byte vector
890 vecindex = OFF_TO_IDX(addr - first_addr);
893 * If we have skipped map entries, we need to make sure that
894 * the byte vector is zeroed for those skipped entries.
896 while ((lastvecindex + 1) < vecindex) {
897 error = subyte(vec + lastvecindex, 0);
906 * Pass the page information to the user
908 error = subyte(vec + vecindex, mincoreinfo);
915 * If the map has changed, due to the subyte, the previous
916 * output may be invalid.
918 vm_map_lock_read(map);
919 if (timestamp != map->timestamp)
922 lastvecindex = vecindex;
928 * subyte may page fault. In case it needs to modify
929 * the map, we release the lock.
931 vm_map_unlock_read(map);
934 * Zero the last entries in the byte vector.
936 vecindex = OFF_TO_IDX(end - first_addr);
937 while ((lastvecindex + 1) < vecindex) {
938 error = subyte(vec + lastvecindex, 0);
947 * If the map has changed, due to the subyte, the previous
948 * output may be invalid.
950 vm_map_lock_read(map);
951 if (timestamp != map->timestamp)
953 vm_map_unlock_read(map);
958 #ifndef _SYS_SYSPROTO_H_
970 struct mlock_args *uap;
973 vm_offset_t addr, end, last, start;
974 vm_size_t npages, size;
977 error = priv_check(td, PRIV_VM_MLOCK);
980 addr = (vm_offset_t)uap->addr;
983 start = trunc_page(addr);
984 end = round_page(last);
985 if (last < addr || end < addr)
987 npages = atop(end - start);
988 if (npages > vm_page_max_wired)
993 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
994 lim_cur(proc, RLIMIT_MEMLOCK)) {
999 if (npages + cnt.v_wire_count > vm_page_max_wired)
1001 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1002 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1003 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1006 #ifndef _SYS_SYSPROTO_H_
1007 struct mlockall_args {
1018 struct mlockall_args *uap;
1023 map = &td->td_proc->p_vmspace->vm_map;
1026 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1031 * If wiring all pages in the process would cause it to exceed
1032 * a hard resource limit, return ENOMEM.
1034 PROC_LOCK(td->td_proc);
1035 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1036 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1037 PROC_UNLOCK(td->td_proc);
1040 PROC_UNLOCK(td->td_proc);
1042 error = priv_check(td, PRIV_VM_MLOCK);
1047 if (uap->how & MCL_FUTURE) {
1049 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1054 if (uap->how & MCL_CURRENT) {
1056 * P1003.1-2001 mandates that all currently mapped pages
1057 * will be memory resident and locked (wired) upon return
1058 * from mlockall(). vm_map_wire() will wire pages, by
1059 * calling vm_fault_wire() for each page in the region.
1061 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1062 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1063 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1069 #ifndef _SYS_SYSPROTO_H_
1070 struct munlockall_args {
1081 struct munlockall_args *uap;
1086 map = &td->td_proc->p_vmspace->vm_map;
1087 error = priv_check(td, PRIV_VM_MUNLOCK);
1091 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1093 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1096 /* Forcibly unwire all pages. */
1097 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1098 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1103 #ifndef _SYS_SYSPROTO_H_
1104 struct munlock_args {
1115 struct munlock_args *uap;
1117 vm_offset_t addr, end, last, start;
1121 error = priv_check(td, PRIV_VM_MUNLOCK);
1124 addr = (vm_offset_t)uap->addr;
1127 start = trunc_page(addr);
1128 end = round_page(last);
1129 if (last < addr || end < addr)
1131 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1132 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1133 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1141 * Helper function for vm_mmap. Perform sanity check specific for mmap
1142 * operations on vnodes.
1145 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1146 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1147 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp)
1158 cred = td->td_ucred;
1159 vfslocked = VFS_LOCK_GIANT(mp);
1160 if ((error = vget(vp, LK_SHARED, td)) != 0) {
1161 VFS_UNLOCK_GIANT(vfslocked);
1167 if (vp->v_type == VREG) {
1169 * Get the proper underlying object
1175 if (obj->handle != vp) {
1177 vp = (struct vnode*)obj->handle;
1178 vget(vp, LK_SHARED, td);
1180 } else if (vp->v_type == VCHR) {
1181 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1182 vp->v_rdev, foffp, objp);
1190 if ((error = VOP_GETATTR(vp, &va, cred)))
1193 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1197 if ((flags & MAP_SHARED) != 0) {
1198 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1199 if (prot & PROT_WRITE) {
1203 *maxprotp &= ~VM_PROT_WRITE;
1207 * If it is a regular file without any references
1208 * we do not need to sync it.
1209 * Adjust object size to be the size of actual file.
1211 objsize = round_page(va.va_size);
1212 if (va.va_nlink == 0)
1213 flags |= MAP_NOSYNC;
1214 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, td->td_ucred);
1223 vfs_mark_atime(vp, cred);
1227 VFS_UNLOCK_GIANT(vfslocked);
1236 * Helper function for vm_mmap. Perform sanity check specific for mmap
1237 * operations on cdevs.
1240 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1241 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1242 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1246 int error, flags, ref;
1250 dsw = dev_refthread(cdev, &ref);
1253 if (dsw->d_flags & D_MMAP_ANON) {
1254 dev_relthread(cdev, ref);
1255 *maxprotp = VM_PROT_ALL;
1256 *flagsp |= MAP_ANON;
1260 * cdevs do not provide private mappings of any kind.
1262 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1263 (prot & PROT_WRITE) != 0) {
1264 dev_relthread(cdev, ref);
1267 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1268 dev_relthread(cdev, ref);
1272 * Force device mappings to be shared.
1274 flags |= MAP_SHARED;
1276 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1278 dev_relthread(cdev, ref);
1283 * First, try d_mmap_single(). If that is not implemented
1284 * (returns ENODEV), fall back to using the device pager.
1285 * Note that d_mmap_single() must return a reference to the
1286 * object (it needs to bump the reference count of the object
1287 * it returns somehow).
1289 * XXX assumes VM_PROT_* == PROT_*
1291 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1292 dev_relthread(cdev, ref);
1293 if (error != ENODEV)
1295 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1309 * Helper function for vm_mmap. Perform sanity check specific for mmap
1310 * operations on shm file descriptors.
1313 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1314 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1315 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1319 if ((*flagsp & MAP_SHARED) != 0 &&
1320 (*maxprotp & VM_PROT_WRITE) == 0 &&
1321 (prot & PROT_WRITE) != 0)
1324 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1328 error = shm_mmap(shmfd, objsize, foff, objp);
1339 * Internal version of mmap. Currently used by mmap, exec, and sys5
1340 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1343 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1344 vm_prot_t maxprot, int flags,
1345 objtype_t handle_type, void *handle,
1349 vm_object_t object = NULL;
1350 int rv = KERN_SUCCESS;
1352 struct thread *td = curthread;
1357 size = round_page(size);
1359 PROC_LOCK(td->td_proc);
1360 if (td->td_proc->p_vmspace->vm_map.size + size >
1361 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1362 PROC_UNLOCK(td->td_proc);
1365 PROC_UNLOCK(td->td_proc);
1368 * We currently can only deal with page aligned file offsets.
1369 * The check is here rather than in the syscall because the
1370 * kernel calls this function internally for other mmaping
1371 * operations (such as in exec) and non-aligned offsets will
1372 * cause pmap inconsistencies...so we want to be sure to
1373 * disallow this in all cases.
1375 if (foff & PAGE_MASK)
1378 if ((flags & MAP_FIXED) == 0) {
1380 *addr = round_page(*addr);
1382 if (*addr != trunc_page(*addr))
1387 * Lookup/allocate object.
1389 switch (handle_type) {
1391 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1392 handle, &foff, &object);
1395 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1396 handle, &foff, &object);
1399 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1400 handle, foff, &object);
1403 if (handle == NULL) {
1414 if (flags & MAP_ANON) {
1418 * Unnamed anonymous regions always start at 0.
1423 docow = MAP_PREFAULT_PARTIAL;
1426 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1427 docow |= MAP_COPY_ON_WRITE;
1428 if (flags & MAP_NOSYNC)
1429 docow |= MAP_DISABLE_SYNCER;
1430 if (flags & MAP_NOCORE)
1431 docow |= MAP_DISABLE_COREDUMP;
1433 if (flags & MAP_STACK)
1434 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1435 docow | MAP_STACK_GROWS_DOWN);
1437 rv = vm_map_find(map, object, foff, addr, size,
1438 object != NULL && object->type == OBJT_DEVICE ?
1439 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1441 rv = vm_map_fixed(map, object, foff, *addr, size,
1442 prot, maxprot, docow);
1444 if (rv != KERN_SUCCESS) {
1446 * Lose the object reference. Will destroy the
1447 * object if it's an unnamed anonymous mapping
1448 * or named anonymous without other references.
1450 vm_object_deallocate(object);
1451 } else if (flags & MAP_SHARED) {
1453 * Shared memory is also shared with children.
1455 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1456 if (rv != KERN_SUCCESS)
1457 (void) vm_map_remove(map, *addr, *addr + size);
1461 * If the process has requested that all future mappings
1462 * be wired, then heed this.
1464 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1465 vm_map_wire(map, *addr, *addr + size,
1466 VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1471 case KERN_INVALID_ADDRESS:
1474 case KERN_PROTECTION_FAILURE: