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, "");
99 * Set the maximum number of vm_map_entry structures per process. Roughly
100 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
101 * of our KVM malloc space still results in generous limits. We want a
102 * default that is good enough to prevent the kernel running out of resources
103 * if attacked from compromised user account but generous enough such that
104 * multi-threaded processes are not unduly inconvenienced.
106 static void vmmapentry_rsrc_init(void *);
107 SYSINIT(vmmersrc, SI_SUB_KVM_RSRC, SI_ORDER_FIRST, vmmapentry_rsrc_init, NULL)
110 vmmapentry_rsrc_init(dummy)
113 max_proc_mmap = vm_kmem_size / sizeof(struct vm_map_entry);
114 max_proc_mmap /= 100;
117 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
118 int *, struct vnode *, vm_ooffset_t, vm_object_t *);
119 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
120 int *, struct cdev *, vm_ooffset_t, vm_object_t *);
129 struct sbrk_args *uap;
131 /* Not yet implemented */
135 #ifndef _SYS_SYSPROTO_H_
148 struct sstk_args *uap;
150 /* Not yet implemented */
154 #if defined(COMPAT_43)
155 #ifndef _SYS_SYSPROTO_H_
156 struct getpagesize_args {
163 ogetpagesize(td, uap)
165 struct getpagesize_args *uap;
168 td->td_retval[0] = PAGE_SIZE;
171 #endif /* COMPAT_43 */
175 * Memory Map (mmap) system call. Note that the file offset
176 * and address are allowed to be NOT page aligned, though if
177 * the MAP_FIXED flag it set, both must have the same remainder
178 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
179 * page-aligned, the actual mapping starts at trunc_page(addr)
180 * and the return value is adjusted up by the page offset.
182 * Generally speaking, only character devices which are themselves
183 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
184 * there would be no cache coherency between a descriptor and a VM mapping
185 * both to the same character device.
187 * Block devices can be mmap'd no matter what they represent. Cache coherency
188 * is maintained as long as you do not write directly to the underlying
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 ((ssize_t) uap->len < 0 ||
234 ((flags & MAP_ANON) && uap->fd != -1))
237 if (flags & MAP_STACK) {
238 if ((uap->fd != -1) ||
239 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
246 * Align the file position to a page boundary,
247 * and save its page offset component.
249 pageoff = (pos & PAGE_MASK);
252 /* Adjust size for rounding (on both ends). */
253 size += pageoff; /* low end... */
254 size = (vm_size_t) round_page(size); /* hi end */
257 * Check for illegal addresses. Watch out for address wrap... Note
258 * that VM_*_ADDRESS are not constants due to casts (argh).
260 if (flags & MAP_FIXED) {
262 * The specified address must have the same remainder
263 * as the file offset taken modulo PAGE_SIZE, so it
264 * should be aligned after adjustment by pageoff.
267 if (addr & PAGE_MASK)
269 /* Address range must be all in user VM space. */
270 if (addr < vm_map_min(&vms->vm_map) ||
271 addr + size > vm_map_max(&vms->vm_map))
273 if (addr + size < addr)
277 * XXX for non-fixed mappings where no hint is provided or
278 * the hint would fall in the potential heap space,
279 * place it after the end of the largest possible heap.
281 * There should really be a pmap call to determine a reasonable
284 PROC_LOCK(td->td_proc);
286 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
287 addr < round_page((vm_offset_t)vms->vm_daddr +
288 lim_max(td->td_proc, RLIMIT_DATA))))
289 addr = round_page((vm_offset_t)vms->vm_daddr +
290 lim_max(td->td_proc, RLIMIT_DATA));
291 PROC_UNLOCK(td->td_proc);
293 if (flags & MAP_ANON) {
295 * Mapping blank space is trivial.
298 handle_type = OBJT_DEFAULT;
299 maxprot = VM_PROT_ALL;
303 * Mapping file, get fp for validation. Obtain vnode and make
304 * sure it is of appropriate type.
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_VNODE) {
314 * POSIX shared-memory objects are defined to have
315 * kernel persistence, and are not defined to support
316 * read(2)/write(2) -- or even open(2). Thus, we can
317 * use MAP_ASYNC to trade on-disk coherence for speed.
318 * The shm_open(3) library routine turns on the FPOSIXSHM
319 * flag to request this behavior.
321 if (fp->f_flag & FPOSIXSHM)
325 * Ensure that file and memory protections are
326 * compatible. Note that we only worry about
327 * writability if mapping is shared; in this case,
328 * current and max prot are dictated by the open file.
329 * XXX use the vnode instead? Problem is: what
330 * credentials do we use for determination? What if
331 * proc does a setuid?
333 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
334 maxprot = VM_PROT_NONE;
336 maxprot = VM_PROT_EXECUTE;
337 if (fp->f_flag & FREAD) {
338 maxprot |= VM_PROT_READ;
339 } else if (prot & PROT_READ) {
344 * If we are sharing potential changes (either via
345 * MAP_SHARED or via the implicit sharing of character
346 * device mappings), and we are trying to get write
347 * permission although we opened it without asking
350 if ((flags & MAP_SHARED) != 0) {
351 if ((fp->f_flag & FWRITE) != 0) {
352 maxprot |= VM_PROT_WRITE;
353 } else if ((prot & PROT_WRITE) != 0) {
357 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
358 maxprot |= VM_PROT_WRITE;
361 handle_type = OBJT_VNODE;
365 * Do not allow more then a certain number of vm_map_entry structures
366 * per process. Scale with the number of rforks sharing the map
367 * to make the limit reasonable for threads.
370 vms->vm_map.nentries >= max_proc_mmap * vms->vm_refcnt) {
375 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
376 flags, handle_type, handle, pos);
378 /* inform hwpmc(4) if an executable is being mapped */
379 if (error == 0 && handle_type == OBJT_VNODE &&
380 (prot & PROT_EXEC)) {
381 pkm.pm_file = handle;
382 pkm.pm_address = (uintptr_t) addr;
383 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
387 td->td_retval[0] = (register_t) (addr + pageoff);
396 #ifndef _SYS_SYSPROTO_H_
409 struct ommap_args *uap;
411 struct mmap_args nargs;
412 static const char cvtbsdprot[8] = {
416 PROT_EXEC | PROT_WRITE,
418 PROT_EXEC | PROT_READ,
419 PROT_WRITE | PROT_READ,
420 PROT_EXEC | PROT_WRITE | PROT_READ,
423 #define OMAP_ANON 0x0002
424 #define OMAP_COPY 0x0020
425 #define OMAP_SHARED 0x0010
426 #define OMAP_FIXED 0x0100
428 nargs.addr = uap->addr;
429 nargs.len = uap->len;
430 nargs.prot = cvtbsdprot[uap->prot & 0x7];
432 if (uap->flags & OMAP_ANON)
433 nargs.flags |= MAP_ANON;
434 if (uap->flags & OMAP_COPY)
435 nargs.flags |= MAP_COPY;
436 if (uap->flags & OMAP_SHARED)
437 nargs.flags |= MAP_SHARED;
439 nargs.flags |= MAP_PRIVATE;
440 if (uap->flags & OMAP_FIXED)
441 nargs.flags |= MAP_FIXED;
443 nargs.pos = uap->pos;
444 return (mmap(td, &nargs));
446 #endif /* COMPAT_43 */
449 #ifndef _SYS_SYSPROTO_H_
462 struct msync_args *uap;
465 vm_size_t size, pageoff;
470 addr = (vm_offset_t) uap->addr;
474 pageoff = (addr & PAGE_MASK);
477 size = (vm_size_t) round_page(size);
478 if (addr + size < addr)
481 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
484 map = &td->td_proc->p_vmspace->vm_map;
487 * Clean the pages and interpret the return value.
489 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
490 (flags & MS_INVALIDATE) != 0);
494 case KERN_INVALID_ADDRESS:
495 return (EINVAL); /* Sun returns ENOMEM? */
496 case KERN_INVALID_ARGUMENT:
503 #ifndef _SYS_SYSPROTO_H_
515 struct munmap_args *uap;
518 struct pmckern_map_out pkm;
519 vm_map_entry_t entry;
522 vm_size_t size, pageoff;
525 addr = (vm_offset_t) uap->addr;
530 pageoff = (addr & PAGE_MASK);
533 size = (vm_size_t) round_page(size);
534 if (addr + size < addr)
538 * Check for illegal addresses. Watch out for address wrap...
540 map = &td->td_proc->p_vmspace->vm_map;
541 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
545 * Make sure entire range is allocated.
547 if (!vm_map_check_protection(map, addr, addr + size, VM_PROT_NONE)) {
553 * Inform hwpmc if the address range being unmapped contains
554 * an executable region.
556 if (vm_map_lookup_entry(map, addr, &entry)) {
558 entry != &map->header && entry->start < addr + size;
559 entry = entry->next) {
560 if (vm_map_check_protection(map, entry->start,
561 entry->end, VM_PROT_EXECUTE) == TRUE) {
562 pkm.pm_address = (uintptr_t) addr;
563 pkm.pm_size = (size_t) size;
564 PMC_CALL_HOOK(td, PMC_FN_MUNMAP,
571 /* returns nothing but KERN_SUCCESS anyway */
572 vm_map_delete(map, addr, addr + size);
577 #ifndef _SYS_SYSPROTO_H_
578 struct mprotect_args {
590 struct mprotect_args *uap;
593 vm_size_t size, pageoff;
596 addr = (vm_offset_t) uap->addr;
598 prot = uap->prot & VM_PROT_ALL;
599 #if defined(VM_PROT_READ_IS_EXEC)
600 if (prot & VM_PROT_READ)
601 prot |= VM_PROT_EXECUTE;
604 pageoff = (addr & PAGE_MASK);
607 size = (vm_size_t) round_page(size);
608 if (addr + size < addr)
611 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
612 addr + size, prot, FALSE)) {
615 case KERN_PROTECTION_FAILURE:
621 #ifndef _SYS_SYSPROTO_H_
622 struct minherit_args {
634 struct minherit_args *uap;
637 vm_size_t size, pageoff;
638 vm_inherit_t inherit;
640 addr = (vm_offset_t)uap->addr;
642 inherit = uap->inherit;
644 pageoff = (addr & PAGE_MASK);
647 size = (vm_size_t) round_page(size);
648 if (addr + size < addr)
651 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
652 addr + size, inherit)) {
655 case KERN_PROTECTION_FAILURE:
661 #ifndef _SYS_SYSPROTO_H_
662 struct madvise_args {
676 struct madvise_args *uap;
678 vm_offset_t start, end;
684 * Check for our special case, advising the swap pager we are
687 if (uap->behav == MADV_PROTECT) {
688 error = priv_check(td, PRIV_VM_MADV_PROTECT);
692 p->p_flag |= P_PROTECTED;
698 * Check for illegal behavior
700 if (uap->behav < 0 || uap->behav > MADV_CORE)
703 * Check for illegal addresses. Watch out for address wrap... Note
704 * that VM_*_ADDRESS are not constants due to casts (argh).
706 map = &td->td_proc->p_vmspace->vm_map;
707 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
708 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
710 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
714 * Since this routine is only advisory, we default to conservative
717 start = trunc_page((vm_offset_t) uap->addr);
718 end = round_page((vm_offset_t) uap->addr + uap->len);
720 if (vm_map_madvise(map, start, end, uap->behav))
725 #ifndef _SYS_SYSPROTO_H_
726 struct mincore_args {
740 struct mincore_args *uap;
742 vm_offset_t addr, first_addr;
743 vm_offset_t end, cend;
748 int vecindex, lastvecindex;
749 vm_map_entry_t current;
750 vm_map_entry_t entry;
752 unsigned int timestamp;
755 * Make sure that the addresses presented are valid for user
758 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
759 end = addr + (vm_size_t)round_page(uap->len);
760 map = &td->td_proc->p_vmspace->vm_map;
761 if (end > vm_map_max(map) || end < addr)
765 * Address of byte vector
769 pmap = vmspace_pmap(td->td_proc->p_vmspace);
771 vm_map_lock_read(map);
773 timestamp = map->timestamp;
775 if (!vm_map_lookup_entry(map, addr, &entry)) {
776 vm_map_unlock_read(map);
781 * Do this on a map entry basis so that if the pages are not
782 * in the current processes address space, we can easily look
783 * up the pages elsewhere.
786 for (current = entry;
787 (current != &map->header) && (current->start < end);
788 current = current->next) {
791 * check for contiguity
793 if (current->end < end &&
794 (entry->next == &map->header ||
795 current->next->start > current->end)) {
796 vm_map_unlock_read(map);
801 * ignore submaps (for now) or null objects
803 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
804 current->object.vm_object == NULL)
808 * limit this scan to the current map entry and the
809 * limits for the mincore call
811 if (addr < current->start)
812 addr = current->start;
818 * scan this entry one page at a time
820 while (addr < cend) {
822 * Check pmap first, it is likely faster, also
823 * it can provide info as to whether we are the
824 * one referencing or modifying the page.
826 mincoreinfo = pmap_mincore(pmap, addr);
832 * calculate the page index into the object
834 offset = current->offset + (addr - current->start);
835 pindex = OFF_TO_IDX(offset);
836 VM_OBJECT_LOCK(current->object.vm_object);
837 m = vm_page_lookup(current->object.vm_object,
840 * if the page is resident, then gather information about
843 if (m != NULL && m->valid != 0) {
844 mincoreinfo = MINCORE_INCORE;
845 vm_page_lock_queues();
848 mincoreinfo |= MINCORE_MODIFIED_OTHER;
849 if ((m->flags & PG_REFERENCED) ||
850 pmap_ts_referenced(m)) {
851 vm_page_flag_set(m, PG_REFERENCED);
852 mincoreinfo |= MINCORE_REFERENCED_OTHER;
854 vm_page_unlock_queues();
856 VM_OBJECT_UNLOCK(current->object.vm_object);
860 * subyte may page fault. In case it needs to modify
861 * the map, we release the lock.
863 vm_map_unlock_read(map);
866 * calculate index into user supplied byte vector
868 vecindex = OFF_TO_IDX(addr - first_addr);
871 * If we have skipped map entries, we need to make sure that
872 * the byte vector is zeroed for those skipped entries.
874 while ((lastvecindex + 1) < vecindex) {
875 error = subyte(vec + lastvecindex, 0);
884 * Pass the page information to the user
886 error = subyte(vec + vecindex, mincoreinfo);
893 * If the map has changed, due to the subyte, the previous
894 * output may be invalid.
896 vm_map_lock_read(map);
897 if (timestamp != map->timestamp)
900 lastvecindex = vecindex;
906 * subyte may page fault. In case it needs to modify
907 * the map, we release the lock.
909 vm_map_unlock_read(map);
912 * Zero the last entries in the byte vector.
914 vecindex = OFF_TO_IDX(end - first_addr);
915 while ((lastvecindex + 1) < vecindex) {
916 error = subyte(vec + lastvecindex, 0);
925 * If the map has changed, due to the subyte, the previous
926 * output may be invalid.
928 vm_map_lock_read(map);
929 if (timestamp != map->timestamp)
931 vm_map_unlock_read(map);
936 #ifndef _SYS_SYSPROTO_H_
948 struct mlock_args *uap;
951 vm_offset_t addr, end, last, start;
952 vm_size_t npages, size;
955 error = priv_check(td, PRIV_VM_MLOCK);
958 addr = (vm_offset_t)uap->addr;
961 start = trunc_page(addr);
962 end = round_page(last);
963 if (last < addr || end < addr)
965 npages = atop(end - start);
966 if (npages > vm_page_max_wired)
971 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))) >
972 lim_cur(proc, RLIMIT_MEMLOCK)) {
977 if (npages + VMCNT_GET(wire_count) > vm_page_max_wired)
979 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
980 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
981 return (error == KERN_SUCCESS ? 0 : ENOMEM);
984 #ifndef _SYS_SYSPROTO_H_
985 struct mlockall_args {
996 struct mlockall_args *uap;
1001 map = &td->td_proc->p_vmspace->vm_map;
1004 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1009 * If wiring all pages in the process would cause it to exceed
1010 * a hard resource limit, return ENOMEM.
1012 PROC_LOCK(td->td_proc);
1013 if (map->size - ptoa(pmap_wired_count(vm_map_pmap(map)) >
1014 lim_cur(td->td_proc, RLIMIT_MEMLOCK))) {
1015 PROC_UNLOCK(td->td_proc);
1018 PROC_UNLOCK(td->td_proc);
1020 error = priv_check(td, PRIV_VM_MLOCK);
1025 if (uap->how & MCL_FUTURE) {
1027 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1032 if (uap->how & MCL_CURRENT) {
1034 * P1003.1-2001 mandates that all currently mapped pages
1035 * will be memory resident and locked (wired) upon return
1036 * from mlockall(). vm_map_wire() will wire pages, by
1037 * calling vm_fault_wire() for each page in the region.
1039 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1040 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1041 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1047 #ifndef _SYS_SYSPROTO_H_
1048 struct munlockall_args {
1059 struct munlockall_args *uap;
1064 map = &td->td_proc->p_vmspace->vm_map;
1065 error = priv_check(td, PRIV_VM_MUNLOCK);
1069 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1071 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1074 /* Forcibly unwire all pages. */
1075 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1076 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1081 #ifndef _SYS_SYSPROTO_H_
1082 struct munlock_args {
1093 struct munlock_args *uap;
1095 vm_offset_t addr, end, last, start;
1099 error = priv_check(td, PRIV_VM_MUNLOCK);
1102 addr = (vm_offset_t)uap->addr;
1105 start = trunc_page(addr);
1106 end = round_page(last);
1107 if (last < addr || end < addr)
1109 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1110 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1111 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1119 * Helper function for vm_mmap. Perform sanity check specific for mmap
1120 * operations on vnodes.
1123 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1124 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1125 struct vnode *vp, vm_ooffset_t foff, vm_object_t *objp)
1131 int error, flags, type;
1135 vfslocked = VFS_LOCK_GIANT(mp);
1136 if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
1137 VFS_UNLOCK_GIANT(vfslocked);
1142 if (vp->v_type == VREG) {
1144 * Get the proper underlying object
1150 if (obj->handle != vp) {
1152 vp = (struct vnode*)obj->handle;
1153 vget(vp, LK_EXCLUSIVE, td);
1157 } else if (vp->v_type == VCHR) {
1159 handle = vp->v_rdev;
1161 /* XXX: lack thredref on device */
1162 if(vp->v_rdev->si_devsw->d_flags & D_MMAP_ANON) {
1163 *maxprotp = VM_PROT_ALL;
1164 *flagsp |= MAP_ANON;
1169 * cdevs does not provide private mappings of any kind.
1171 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1172 (prot & PROT_WRITE) != 0) {
1176 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1181 * Force device mappings to be shared.
1183 flags |= MAP_SHARED;
1188 if ((error = VOP_GETATTR(vp, &va, td->td_ucred, td))) {
1192 error = mac_check_vnode_mmap(td->td_ucred, vp, prot, flags);
1196 if ((flags & MAP_SHARED) != 0) {
1197 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1198 if (prot & PROT_WRITE) {
1202 *maxprotp &= ~VM_PROT_WRITE;
1206 * If it is a regular file without any references
1207 * we do not need to sync it.
1208 * Adjust object size to be the size of actual file.
1210 if (vp->v_type == VREG) {
1211 objsize = round_page(va.va_size);
1212 if (va.va_nlink == 0)
1213 flags |= MAP_NOSYNC;
1215 obj = vm_pager_allocate(type, handle, objsize, prot, foff);
1217 error = (type == OBJT_DEVICE ? EINVAL : ENOMEM);
1222 vfs_mark_atime(vp, td);
1226 VFS_UNLOCK_GIANT(vfslocked);
1235 * Helper function for vm_mmap. Perform sanity check specific for mmap
1236 * operations on cdevs.
1239 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1240 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1241 struct cdev *cdev, vm_ooffset_t foff, vm_object_t *objp)
1248 /* XXX: lack thredref on device */
1249 if (cdev->si_devsw->d_flags & D_MMAP_ANON) {
1250 *maxprotp = VM_PROT_ALL;
1251 *flagsp |= MAP_ANON;
1255 * cdevs does not provide private mappings of any kind.
1257 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1258 (prot & PROT_WRITE) != 0)
1260 if (flags & (MAP_PRIVATE|MAP_COPY))
1263 * Force device mappings to be shared.
1265 flags |= MAP_SHARED;
1267 error = mac_check_cdev_mmap(td->td_ucred, cdev, prot);
1271 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, foff);
1284 * Internal version of mmap. Currently used by mmap, exec, and sys5
1285 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1288 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1289 vm_prot_t maxprot, int flags,
1290 objtype_t handle_type, void *handle,
1295 int rv = KERN_SUCCESS;
1297 struct thread *td = curthread;
1302 size = round_page(size);
1304 PROC_LOCK(td->td_proc);
1305 if (td->td_proc->p_vmspace->vm_map.size + size >
1306 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1307 PROC_UNLOCK(td->td_proc);
1310 PROC_UNLOCK(td->td_proc);
1313 * We currently can only deal with page aligned file offsets.
1314 * The check is here rather than in the syscall because the
1315 * kernel calls this function internally for other mmaping
1316 * operations (such as in exec) and non-aligned offsets will
1317 * cause pmap inconsistencies...so we want to be sure to
1318 * disallow this in all cases.
1320 if (foff & PAGE_MASK)
1323 if ((flags & MAP_FIXED) == 0) {
1325 *addr = round_page(*addr);
1327 if (*addr != trunc_page(*addr))
1330 (void) vm_map_remove(map, *addr, *addr + size);
1333 * Lookup/allocate object.
1335 switch (handle_type) {
1337 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1338 handle, foff, &object);
1341 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1342 handle, foff, &object);
1345 if (handle == NULL) {
1355 if (flags & MAP_ANON) {
1359 * Unnamed anonymous regions always start at 0.
1364 docow = MAP_PREFAULT_PARTIAL;
1367 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1368 docow |= MAP_COPY_ON_WRITE;
1369 if (flags & MAP_NOSYNC)
1370 docow |= MAP_DISABLE_SYNCER;
1371 if (flags & MAP_NOCORE)
1372 docow |= MAP_DISABLE_COREDUMP;
1374 #if defined(VM_PROT_READ_IS_EXEC)
1375 if (prot & VM_PROT_READ)
1376 prot |= VM_PROT_EXECUTE;
1378 if (maxprot & VM_PROT_READ)
1379 maxprot |= VM_PROT_EXECUTE;
1383 *addr = pmap_addr_hint(object, *addr, size);
1385 if (flags & MAP_STACK)
1386 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1387 docow | MAP_STACK_GROWS_DOWN);
1389 rv = vm_map_find(map, object, foff, addr, size, fitit,
1390 prot, maxprot, docow);
1392 if (rv != KERN_SUCCESS) {
1394 * Lose the object reference. Will destroy the
1395 * object if it's an unnamed anonymous mapping
1396 * or named anonymous without other references.
1398 vm_object_deallocate(object);
1399 } else if (flags & MAP_SHARED) {
1401 * Shared memory is also shared with children.
1403 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1404 if (rv != KERN_SUCCESS)
1405 (void) vm_map_remove(map, *addr, *addr + size);
1409 * If the process has requested that all future mappings
1410 * be wired, then heed this.
1412 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1413 vm_map_wire(map, *addr, *addr + size,
1414 VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1419 case KERN_INVALID_ADDRESS:
1422 case KERN_PROTECTION_FAILURE: