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))
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;
304 * Mapping file, get fp for validation and
305 * don't let the descriptor disappear on us if we block.
307 if ((error = fget(td, uap->fd, &fp)) != 0)
309 if (fp->f_type == DTYPE_SHM) {
311 handle_type = OBJT_SWAP;
312 maxprot = VM_PROT_NONE;
314 /* FREAD should always be set. */
315 if (fp->f_flag & FREAD)
316 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
317 if (fp->f_flag & FWRITE)
318 maxprot |= VM_PROT_WRITE;
321 if (fp->f_type != DTYPE_VNODE) {
325 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
326 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
328 * POSIX shared-memory objects are defined to have
329 * kernel persistence, and are not defined to support
330 * read(2)/write(2) -- or even open(2). Thus, we can
331 * use MAP_ASYNC to trade on-disk coherence for speed.
332 * The shm_open(3) library routine turns on the FPOSIXSHM
333 * flag to request this behavior.
335 if (fp->f_flag & FPOSIXSHM)
340 * Ensure that file and memory protections are
341 * compatible. Note that we only worry about
342 * writability if mapping is shared; in this case,
343 * current and max prot are dictated by the open file.
344 * XXX use the vnode instead? Problem is: what
345 * credentials do we use for determination? What if
346 * proc does a setuid?
348 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
349 maxprot = VM_PROT_NONE;
351 maxprot = VM_PROT_EXECUTE;
352 if (fp->f_flag & FREAD) {
353 maxprot |= VM_PROT_READ;
354 } else if (prot & PROT_READ) {
359 * If we are sharing potential changes (either via
360 * MAP_SHARED or via the implicit sharing of character
361 * device mappings), and we are trying to get write
362 * permission although we opened it without asking
365 if ((flags & MAP_SHARED) != 0) {
366 if ((fp->f_flag & FWRITE) != 0) {
367 maxprot |= VM_PROT_WRITE;
368 } else if ((prot & PROT_WRITE) != 0) {
372 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
373 maxprot |= VM_PROT_WRITE;
376 handle_type = OBJT_VNODE;
381 * Do not allow more then a certain number of vm_map_entry structures
382 * per process. Scale with the number of rforks sharing the map
383 * to make the limit reasonable for threads.
386 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
392 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
393 flags, handle_type, handle, pos);
396 /* inform hwpmc(4) if an executable is being mapped */
397 if (error == 0 && handle_type == OBJT_VNODE &&
398 (prot & PROT_EXEC)) {
399 pkm.pm_file = handle;
400 pkm.pm_address = (uintptr_t) addr;
401 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
405 td->td_retval[0] = (register_t) (addr + pageoff);
414 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
416 struct mmap_args oargs;
418 oargs.addr = uap->addr;
419 oargs.len = uap->len;
420 oargs.prot = uap->prot;
421 oargs.flags = uap->flags;
423 oargs.pos = uap->pos;
424 return (mmap(td, &oargs));
428 #ifndef _SYS_SYSPROTO_H_
441 struct ommap_args *uap;
443 struct mmap_args nargs;
444 static const char cvtbsdprot[8] = {
448 PROT_EXEC | PROT_WRITE,
450 PROT_EXEC | PROT_READ,
451 PROT_WRITE | PROT_READ,
452 PROT_EXEC | PROT_WRITE | PROT_READ,
455 #define OMAP_ANON 0x0002
456 #define OMAP_COPY 0x0020
457 #define OMAP_SHARED 0x0010
458 #define OMAP_FIXED 0x0100
460 nargs.addr = uap->addr;
461 nargs.len = uap->len;
462 nargs.prot = cvtbsdprot[uap->prot & 0x7];
464 if (uap->flags & OMAP_ANON)
465 nargs.flags |= MAP_ANON;
466 if (uap->flags & OMAP_COPY)
467 nargs.flags |= MAP_COPY;
468 if (uap->flags & OMAP_SHARED)
469 nargs.flags |= MAP_SHARED;
471 nargs.flags |= MAP_PRIVATE;
472 if (uap->flags & OMAP_FIXED)
473 nargs.flags |= MAP_FIXED;
475 nargs.pos = uap->pos;
476 return (mmap(td, &nargs));
478 #endif /* COMPAT_43 */
481 #ifndef _SYS_SYSPROTO_H_
494 struct msync_args *uap;
497 vm_size_t size, pageoff;
502 addr = (vm_offset_t) uap->addr;
506 pageoff = (addr & PAGE_MASK);
509 size = (vm_size_t) round_page(size);
510 if (addr + size < addr)
513 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
516 map = &td->td_proc->p_vmspace->vm_map;
519 * Clean the pages and interpret the return value.
521 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
522 (flags & MS_INVALIDATE) != 0);
526 case KERN_INVALID_ADDRESS:
527 return (EINVAL); /* Sun returns ENOMEM? */
528 case KERN_INVALID_ARGUMENT:
535 #ifndef _SYS_SYSPROTO_H_
547 struct munmap_args *uap;
550 struct pmckern_map_out pkm;
551 vm_map_entry_t entry;
554 vm_size_t size, pageoff;
557 addr = (vm_offset_t) uap->addr;
562 pageoff = (addr & PAGE_MASK);
565 size = (vm_size_t) round_page(size);
566 if (addr + size < addr)
570 * Check for illegal addresses. Watch out for address wrap...
572 map = &td->td_proc->p_vmspace->vm_map;
573 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
578 * Inform hwpmc if the address range being unmapped contains
579 * an executable region.
581 if (vm_map_lookup_entry(map, addr, &entry)) {
583 entry != &map->header && entry->start < addr + size;
584 entry = entry->next) {
585 if (vm_map_check_protection(map, entry->start,
586 entry->end, VM_PROT_EXECUTE) == TRUE) {
587 pkm.pm_address = (uintptr_t) addr;
588 pkm.pm_size = (size_t) size;
589 PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
596 /* returns nothing but KERN_SUCCESS anyway */
597 vm_map_delete(map, addr, addr + size);
602 #ifndef _SYS_SYSPROTO_H_
603 struct mprotect_args {
615 struct mprotect_args *uap;
618 vm_size_t size, pageoff;
621 addr = (vm_offset_t) uap->addr;
623 prot = uap->prot & VM_PROT_ALL;
625 pageoff = (addr & PAGE_MASK);
628 size = (vm_size_t) round_page(size);
629 if (addr + size < addr)
632 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
633 addr + size, prot, FALSE)) {
636 case KERN_PROTECTION_FAILURE:
638 case KERN_RESOURCE_SHORTAGE:
644 #ifndef _SYS_SYSPROTO_H_
645 struct minherit_args {
657 struct minherit_args *uap;
660 vm_size_t size, pageoff;
661 vm_inherit_t inherit;
663 addr = (vm_offset_t)uap->addr;
665 inherit = uap->inherit;
667 pageoff = (addr & PAGE_MASK);
670 size = (vm_size_t) round_page(size);
671 if (addr + size < addr)
674 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
675 addr + size, inherit)) {
678 case KERN_PROTECTION_FAILURE:
684 #ifndef _SYS_SYSPROTO_H_
685 struct madvise_args {
699 struct madvise_args *uap;
701 vm_offset_t start, end;
707 * Check for our special case, advising the swap pager we are
710 if (uap->behav == MADV_PROTECT) {
711 error = priv_check(td, PRIV_VM_MADV_PROTECT);
715 p->p_flag |= P_PROTECTED;
721 * Check for illegal behavior
723 if (uap->behav < 0 || uap->behav > MADV_CORE)
726 * Check for illegal addresses. Watch out for address wrap... Note
727 * that VM_*_ADDRESS are not constants due to casts (argh).
729 map = &td->td_proc->p_vmspace->vm_map;
730 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
731 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
733 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
737 * Since this routine is only advisory, we default to conservative
740 start = trunc_page((vm_offset_t) uap->addr);
741 end = round_page((vm_offset_t) uap->addr + uap->len);
743 if (vm_map_madvise(map, start, end, uap->behav))
748 #ifndef _SYS_SYSPROTO_H_
749 struct mincore_args {
763 struct mincore_args *uap;
765 vm_offset_t addr, first_addr;
766 vm_offset_t end, cend;
771 int vecindex, lastvecindex;
772 vm_map_entry_t current;
773 vm_map_entry_t entry;
775 unsigned int timestamp;
778 * Make sure that the addresses presented are valid for user
781 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
782 end = addr + (vm_size_t)round_page(uap->len);
783 map = &td->td_proc->p_vmspace->vm_map;
784 if (end > vm_map_max(map) || end < addr)
788 * Address of byte vector
792 pmap = vmspace_pmap(td->td_proc->p_vmspace);
794 vm_map_lock_read(map);
796 timestamp = map->timestamp;
798 if (!vm_map_lookup_entry(map, addr, &entry)) {
799 vm_map_unlock_read(map);
804 * Do this on a map entry basis so that if the pages are not
805 * in the current processes address space, we can easily look
806 * up the pages elsewhere.
809 for (current = entry;
810 (current != &map->header) && (current->start < end);
811 current = current->next) {
814 * check for contiguity
816 if (current->end < end &&
817 (entry->next == &map->header ||
818 current->next->start > current->end)) {
819 vm_map_unlock_read(map);
824 * ignore submaps (for now) or null objects
826 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
827 current->object.vm_object == NULL)
831 * limit this scan to the current map entry and the
832 * limits for the mincore call
834 if (addr < current->start)
835 addr = current->start;
841 * scan this entry one page at a time
843 while (addr < cend) {
845 * Check pmap first, it is likely faster, also
846 * it can provide info as to whether we are the
847 * one referencing or modifying the page.
849 mincoreinfo = pmap_mincore(pmap, addr);
855 * calculate the page index into the object
857 offset = current->offset + (addr - current->start);
858 pindex = OFF_TO_IDX(offset);
859 VM_OBJECT_LOCK(current->object.vm_object);
860 m = vm_page_lookup(current->object.vm_object,
863 * if the page is resident, then gather information about
866 if (m != NULL && m->valid != 0) {
867 mincoreinfo = MINCORE_INCORE;
868 vm_page_lock_queues();
871 mincoreinfo |= MINCORE_MODIFIED_OTHER;
872 if ((m->flags & PG_REFERENCED) ||
873 pmap_ts_referenced(m)) {
874 vm_page_flag_set(m, PG_REFERENCED);
875 mincoreinfo |= MINCORE_REFERENCED_OTHER;
877 vm_page_unlock_queues();
879 VM_OBJECT_UNLOCK(current->object.vm_object);
883 * subyte may page fault. In case it needs to modify
884 * the map, we release the lock.
886 vm_map_unlock_read(map);
889 * calculate index into user supplied byte vector
891 vecindex = OFF_TO_IDX(addr - first_addr);
894 * If we have skipped map entries, we need to make sure that
895 * the byte vector is zeroed for those skipped entries.
897 while ((lastvecindex + 1) < vecindex) {
898 error = subyte(vec + lastvecindex, 0);
907 * Pass the page information to the user
909 error = subyte(vec + vecindex, mincoreinfo);
916 * If the map has changed, due to the subyte, the previous
917 * output may be invalid.
919 vm_map_lock_read(map);
920 if (timestamp != map->timestamp)
923 lastvecindex = vecindex;
929 * subyte may page fault. In case it needs to modify
930 * the map, we release the lock.
932 vm_map_unlock_read(map);
935 * Zero the last entries in the byte vector.
937 vecindex = OFF_TO_IDX(end - first_addr);
938 while ((lastvecindex + 1) < vecindex) {
939 error = subyte(vec + lastvecindex, 0);
948 * If the map has changed, due to the subyte, the previous
949 * output may be invalid.
951 vm_map_lock_read(map);
952 if (timestamp != map->timestamp)
954 vm_map_unlock_read(map);
959 #ifndef _SYS_SYSPROTO_H_
971 struct mlock_args *uap;
974 vm_offset_t addr, end, last, start;
975 vm_size_t npages, size;
978 error = priv_check(td, PRIV_VM_MLOCK);
981 addr = (vm_offset_t)uap->addr;
984 start = trunc_page(addr);
985 end = round_page(last);
986 if (last < addr || end < addr)
988 npages = atop(end - start);
989 if (npages > vm_page_max_wired)
994 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
995 lim_cur(proc, RLIMIT_MEMLOCK)) {
1000 if (npages + cnt.v_wire_count > vm_page_max_wired)
1002 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1003 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1004 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1007 #ifndef _SYS_SYSPROTO_H_
1008 struct mlockall_args {
1019 struct mlockall_args *uap;
1024 map = &td->td_proc->p_vmspace->vm_map;
1027 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1032 * If wiring all pages in the process would cause it to exceed
1033 * a hard resource limit, return ENOMEM.
1035 PROC_LOCK(td->td_proc);
1036 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1037 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1038 PROC_UNLOCK(td->td_proc);
1041 PROC_UNLOCK(td->td_proc);
1043 error = priv_check(td, PRIV_VM_MLOCK);
1048 if (uap->how & MCL_FUTURE) {
1050 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1055 if (uap->how & MCL_CURRENT) {
1057 * P1003.1-2001 mandates that all currently mapped pages
1058 * will be memory resident and locked (wired) upon return
1059 * from mlockall(). vm_map_wire() will wire pages, by
1060 * calling vm_fault_wire() for each page in the region.
1062 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1063 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1064 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1070 #ifndef _SYS_SYSPROTO_H_
1071 struct munlockall_args {
1082 struct munlockall_args *uap;
1087 map = &td->td_proc->p_vmspace->vm_map;
1088 error = priv_check(td, PRIV_VM_MUNLOCK);
1092 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1094 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1097 /* Forcibly unwire all pages. */
1098 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1099 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1104 #ifndef _SYS_SYSPROTO_H_
1105 struct munlock_args {
1116 struct munlock_args *uap;
1118 vm_offset_t addr, end, last, start;
1122 error = priv_check(td, PRIV_VM_MUNLOCK);
1125 addr = (vm_offset_t)uap->addr;
1128 start = trunc_page(addr);
1129 end = round_page(last);
1130 if (last < addr || end < addr)
1132 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1133 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1134 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1142 * Helper function for vm_mmap. Perform sanity check specific for mmap
1143 * operations on vnodes.
1146 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1147 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1148 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp)
1159 cred = td->td_ucred;
1160 vfslocked = VFS_LOCK_GIANT(mp);
1161 if ((error = vget(vp, LK_SHARED, td)) != 0) {
1162 VFS_UNLOCK_GIANT(vfslocked);
1168 if (vp->v_type == VREG) {
1170 * Get the proper underlying object
1176 if (obj->handle != vp) {
1178 vp = (struct vnode*)obj->handle;
1179 vget(vp, LK_SHARED, td);
1181 } else if (vp->v_type == VCHR) {
1182 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1183 vp->v_rdev, foffp, objp);
1191 if ((error = VOP_GETATTR(vp, &va, cred)))
1194 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1198 if ((flags & MAP_SHARED) != 0) {
1199 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1200 if (prot & PROT_WRITE) {
1204 *maxprotp &= ~VM_PROT_WRITE;
1208 * If it is a regular file without any references
1209 * we do not need to sync it.
1210 * Adjust object size to be the size of actual file.
1212 objsize = round_page(va.va_size);
1213 if (va.va_nlink == 0)
1214 flags |= MAP_NOSYNC;
1215 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, td->td_ucred);
1224 vfs_mark_atime(vp, cred);
1228 VFS_UNLOCK_GIANT(vfslocked);
1237 * Helper function for vm_mmap. Perform sanity check specific for mmap
1238 * operations on cdevs.
1241 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1242 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1243 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1251 dsw = dev_refthread(cdev);
1254 if (dsw->d_flags & D_MMAP_ANON) {
1255 dev_relthread(cdev);
1256 *maxprotp = VM_PROT_ALL;
1257 *flagsp |= MAP_ANON;
1261 * cdevs do not provide private mappings of any kind.
1263 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1264 (prot & PROT_WRITE) != 0) {
1265 dev_relthread(cdev);
1268 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1269 dev_relthread(cdev);
1273 * Force device mappings to be shared.
1275 flags |= MAP_SHARED;
1277 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1279 dev_relthread(cdev);
1284 * First, try d_mmap_single(). If that is not implemented
1285 * (returns ENODEV), fall back to using the device pager.
1286 * Note that d_mmap_single() must return a reference to the
1287 * object (it needs to bump the reference count of the object
1288 * it returns somehow).
1290 * XXX assumes VM_PROT_* == PROT_*
1292 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1293 dev_relthread(cdev);
1294 if (error != ENODEV)
1296 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1310 * Helper function for vm_mmap. Perform sanity check specific for mmap
1311 * operations on shm file descriptors.
1314 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1315 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1316 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1320 if ((*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: