4 * Copyright (c) 1988 University of Utah.
5 * Copyright (c) 1991, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
39 * $FreeBSD: src/sys/vm/vm_mmap.c,v 1.108.2.6 2002/07/02 20:06:19 dillon Exp $
43 * Mapped file (mmap) interface to VM
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/sysproto.h>
50 #include <sys/filedesc.h>
51 #include <sys/kern_syscall.h>
54 #include <sys/resource.h>
55 #include <sys/resourcevar.h>
56 #include <sys/vnode.h>
57 #include <sys/fcntl.h>
62 #include <sys/vmmeter.h>
63 #include <sys/sysctl.h>
66 #include <vm/vm_param.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pager.h>
73 #include <vm/vm_pageout.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_kern.h>
77 #include <sys/file2.h>
78 #include <sys/thread.h>
79 #include <sys/thread2.h>
81 static int max_proc_mmap;
82 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
84 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
87 * Set the maximum number of vm_map_entry structures per process. Roughly
88 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
89 * of our KVM malloc space still results in generous limits. We want a
90 * default that is good enough to prevent the kernel running out of resources
91 * if attacked from compromised user account but generous enough such that
92 * multi-threaded processes are not unduly inconvenienced.
95 static void vmmapentry_rsrc_init (void *);
96 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL)
99 vmmapentry_rsrc_init(void *dummy)
101 max_proc_mmap = KvaSize / sizeof(struct vm_map_entry);
102 max_proc_mmap /= 100;
109 sys_sbrk(struct sbrk_args *uap)
111 /* Not yet implemented */
116 * 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
150 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
151 int uprot, int uflags, int fd, off_t upos, void **res)
153 struct thread *td = curthread;
154 struct proc *p = td->td_proc;
155 struct file *fp = NULL;
159 vm_size_t size, pageoff;
160 vm_prot_t prot, maxprot;
168 addr = (vm_offset_t) uaddr;
170 prot = uprot & VM_PROT_ALL;
175 * Make sure mapping fits into numeric range etc.
177 * NOTE: We support the full unsigned range for size now.
179 if (((flags & MAP_ANON) && (fd != -1 || pos != 0)))
185 if (flags & MAP_STACK) {
187 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
194 * Virtual page tables cannot be used with MAP_STACK. Apart from
195 * it not making any sense, the aux union is used by both
198 * Because the virtual page table is stored in the backing object
199 * and might be updated by the kernel, the mapping must be R+W.
201 if (flags & MAP_VPAGETABLE) {
202 if (vkernel_enable == 0)
204 if (flags & MAP_STACK)
206 if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
211 * Align the file position to a page boundary,
212 * and save its page offset component.
214 pageoff = (pos & PAGE_MASK);
217 /* Adjust size for rounding (on both ends). */
218 size += pageoff; /* low end... */
219 size = (vm_size_t) round_page(size); /* hi end */
220 if (size < ulen) /* wrap */
224 * Check for illegal addresses. Watch out for address wrap... Note
225 * that VM_*_ADDRESS are not constants due to casts (argh).
227 if (flags & (MAP_FIXED | MAP_TRYFIXED)) {
229 * The specified address must have the same remainder
230 * as the file offset taken modulo PAGE_SIZE, so it
231 * should be aligned after adjustment by pageoff.
234 if (addr & PAGE_MASK)
238 * Address range must be all in user VM space and not wrap.
240 tmpaddr = addr + size;
243 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
245 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
249 * Get a hint of where to map. It also provides mmap offset
250 * randomization if enabled.
252 addr = vm_map_hint(p, addr, prot);
255 if (flags & MAP_ANON) {
257 * Mapping blank space is trivial.
260 maxprot = VM_PROT_ALL;
263 * Mapping file, get fp for validation. Obtain vnode and make
264 * sure it is of appropriate type.
266 fp = holdfp(p->p_fd, fd, -1);
269 if (fp->f_type != DTYPE_VNODE) {
274 * POSIX shared-memory objects are defined to have
275 * kernel persistence, and are not defined to support
276 * read(2)/write(2) -- or even open(2). Thus, we can
277 * use MAP_ASYNC to trade on-disk coherence for speed.
278 * The shm_open(3) library routine turns on the FPOSIXSHM
279 * flag to request this behavior.
281 if (fp->f_flag & FPOSIXSHM)
283 vp = (struct vnode *) fp->f_data;
286 * Validate the vnode for the operation.
291 * Get the proper underlying object
293 if ((obj = vp->v_object) == NULL) {
297 KKASSERT((struct vnode *)obj->handle == vp);
301 * Make sure a device has not been revoked.
302 * Mappability is handled by the device layer.
304 if (vp->v_rdev == NULL) {
311 * Nothing else is mappable.
318 * XXX hack to handle use of /dev/zero to map anon memory (ala
321 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
323 maxprot = VM_PROT_ALL;
328 * cdevs does not provide private mappings of any kind.
330 if (vp->v_type == VCHR &&
331 (flags & (MAP_PRIVATE|MAP_COPY))) {
336 * Ensure that file and memory protections are
337 * compatible. Note that we only worry about
338 * writability if mapping is shared; in this case,
339 * current and max prot are dictated by the open file.
340 * XXX use the vnode instead? Problem is: what
341 * credentials do we use for determination? What if
342 * proc does a setuid?
344 maxprot = VM_PROT_EXECUTE; /* ??? */
345 if (fp->f_flag & FREAD) {
346 maxprot |= VM_PROT_READ;
347 } else if (prot & PROT_READ) {
352 * If we are sharing potential changes (either via
353 * MAP_SHARED or via the implicit sharing of character
354 * device mappings), and we are trying to get write
355 * permission although we opened it without asking
356 * for it, bail out. Check for superuser, only if
357 * we're at securelevel < 1, to allow the XIG X server
358 * to continue to work.
360 if ((flags & MAP_SHARED) != 0 || vp->v_type == VCHR) {
361 if ((fp->f_flag & FWRITE) != 0) {
363 if ((error = VOP_GETATTR(vp, &va))) {
367 (IMMUTABLE|APPEND)) == 0) {
368 maxprot |= VM_PROT_WRITE;
369 } else if (prot & PROT_WRITE) {
373 } else if ((prot & PROT_WRITE) != 0) {
378 maxprot |= VM_PROT_WRITE;
384 lwkt_gettoken(&vms->vm_map.token);
387 * Do not allow more then a certain number of vm_map_entry structures
388 * per process. Scale with the number of rforks sharing the map
389 * to make the limit reasonable for threads.
392 vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
394 lwkt_reltoken(&vms->vm_map.token);
398 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
401 *res = (void *)(addr + pageoff);
403 lwkt_reltoken(&vms->vm_map.token);
412 * mmap system call handler
417 sys_mmap(struct mmap_args *uap)
421 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
422 uap->prot, uap->flags,
423 uap->fd, uap->pos, &uap->sysmsg_resultp);
429 * msync system call handler
431 * msync_args(void *addr, size_t len, int flags)
436 sys_msync(struct msync_args *uap)
438 struct proc *p = curproc;
441 vm_size_t size, pageoff;
446 addr = (vm_offset_t) uap->addr;
450 pageoff = (addr & PAGE_MASK);
453 size = (vm_size_t) round_page(size);
454 if (size < uap->len) /* wrap */
456 tmpaddr = addr + size; /* workaround gcc4 opt */
457 if (tmpaddr < addr) /* wrap */
460 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
463 map = &p->p_vmspace->vm_map;
466 * map->token serializes extracting the address range for size == 0
467 * msyncs with the vm_map_clean call; if the token were not held
468 * across the two calls, an intervening munmap/mmap pair, for example,
469 * could cause msync to occur on a wrong region.
471 lwkt_gettoken(&map->token);
474 * XXX Gak! If size is zero we are supposed to sync "all modified
475 * pages with the region containing addr". Unfortunately, we don't
476 * really keep track of individual mmaps so we approximate by flushing
477 * the range of the map entry containing addr. This can be incorrect
478 * if the region splits or is coalesced with a neighbor.
481 vm_map_entry_t entry;
483 vm_map_lock_read(map);
484 rv = vm_map_lookup_entry(map, addr, &entry);
486 vm_map_unlock_read(map);
487 rv = KERN_INVALID_ADDRESS;
491 size = entry->end - entry->start;
492 vm_map_unlock_read(map);
496 * Clean the pages and interpret the return value.
498 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
499 (flags & MS_INVALIDATE) != 0);
501 lwkt_reltoken(&map->token);
506 case KERN_INVALID_ADDRESS:
507 return (EINVAL); /* Sun returns ENOMEM? */
518 * munmap system call handler
520 * munmap_args(void *addr, size_t len)
525 sys_munmap(struct munmap_args *uap)
527 struct proc *p = curproc;
530 vm_size_t size, pageoff;
533 addr = (vm_offset_t) uap->addr;
536 pageoff = (addr & PAGE_MASK);
539 size = (vm_size_t) round_page(size);
540 if (size < uap->len) /* wrap */
542 tmpaddr = addr + size; /* workaround gcc4 opt */
543 if (tmpaddr < addr) /* wrap */
550 * Check for illegal addresses. Watch out for address wrap... Note
551 * that VM_*_ADDRESS are not constants due to casts (argh).
553 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
555 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
558 map = &p->p_vmspace->vm_map;
560 /* map->token serializes between the map check and the actual unmap */
561 lwkt_gettoken(&map->token);
564 * Make sure entire range is allocated.
566 if (!vm_map_check_protection(map, addr, addr + size,
567 VM_PROT_NONE, FALSE)) {
568 lwkt_reltoken(&map->token);
571 /* returns nothing but KERN_SUCCESS anyway */
572 vm_map_remove(map, addr, addr + size);
573 lwkt_reltoken(&map->token);
578 * mprotect_args(const void *addr, size_t len, int prot)
583 sys_mprotect(struct mprotect_args *uap)
585 struct proc *p = curproc;
588 vm_size_t size, pageoff;
592 addr = (vm_offset_t) uap->addr;
594 prot = uap->prot & VM_PROT_ALL;
595 #if defined(VM_PROT_READ_IS_EXEC)
596 if (prot & VM_PROT_READ)
597 prot |= VM_PROT_EXECUTE;
600 pageoff = (addr & PAGE_MASK);
603 size = (vm_size_t) round_page(size);
604 if (size < uap->len) /* wrap */
606 tmpaddr = addr + size; /* workaround gcc4 opt */
607 if (tmpaddr < addr) /* wrap */
610 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size,
615 case KERN_PROTECTION_FAILURE:
626 * minherit system call handler
628 * minherit_args(void *addr, size_t len, int inherit)
633 sys_minherit(struct minherit_args *uap)
635 struct proc *p = curproc;
638 vm_size_t size, pageoff;
639 vm_inherit_t inherit;
642 addr = (vm_offset_t)uap->addr;
644 inherit = uap->inherit;
646 pageoff = (addr & PAGE_MASK);
649 size = (vm_size_t) round_page(size);
650 if (size < uap->len) /* wrap */
652 tmpaddr = addr + size; /* workaround gcc4 opt */
653 if (tmpaddr < addr) /* wrap */
656 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr,
657 addr + size, inherit)) {
661 case KERN_PROTECTION_FAILURE:
672 * madvise system call handler
674 * madvise_args(void *addr, size_t len, int behav)
679 sys_madvise(struct madvise_args *uap)
681 struct proc *p = curproc;
682 vm_offset_t start, end;
683 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
687 * Check for illegal behavior
689 if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
692 * Check for illegal addresses. Watch out for address wrap... Note
693 * that VM_*_ADDRESS are not constants due to casts (argh).
695 if (tmpaddr < (vm_offset_t)uap->addr)
697 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
699 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
703 * Since this routine is only advisory, we default to conservative
706 start = trunc_page((vm_offset_t)uap->addr);
707 end = round_page(tmpaddr);
709 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
715 * mcontrol system call handler
717 * mcontrol_args(void *addr, size_t len, int behav, off_t value)
722 sys_mcontrol(struct mcontrol_args *uap)
724 struct proc *p = curproc;
725 vm_offset_t start, end;
726 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
730 * Check for illegal behavior
732 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
735 * Check for illegal addresses. Watch out for address wrap... Note
736 * that VM_*_ADDRESS are not constants due to casts (argh).
738 if (tmpaddr < (vm_offset_t) uap->addr)
740 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
742 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
746 * Since this routine is only advisory, we default to conservative
749 start = trunc_page((vm_offset_t)uap->addr);
750 end = round_page(tmpaddr);
752 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
753 uap->behav, uap->value);
759 * mincore system call handler
761 * mincore_args(const void *addr, size_t len, char *vec)
766 sys_mincore(struct mincore_args *uap)
768 struct proc *p = curproc;
769 vm_offset_t addr, first_addr;
770 vm_offset_t end, cend;
775 int vecindex, lastvecindex;
776 vm_map_entry_t current;
777 vm_map_entry_t entry;
779 unsigned int timestamp;
782 * Make sure that the addresses presented are valid for user
785 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
786 end = addr + (vm_size_t)round_page(uap->len);
789 if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
793 * Address of byte vector
797 map = &p->p_vmspace->vm_map;
798 pmap = vmspace_pmap(p->p_vmspace);
800 lwkt_gettoken(&map->token);
801 vm_map_lock_read(map);
803 timestamp = map->timestamp;
805 if (!vm_map_lookup_entry(map, addr, &entry))
809 * Do this on a map entry basis so that if the pages are not
810 * in the current processes address space, we can easily look
811 * up the pages elsewhere.
815 (current != &map->header) && (current->start < end);
816 current = current->next) {
819 * ignore submaps (for now) or null objects
821 if (current->maptype != VM_MAPTYPE_NORMAL &&
822 current->maptype != VM_MAPTYPE_VPAGETABLE) {
825 if (current->object.vm_object == NULL)
829 * limit this scan to the current map entry and the
830 * limits for the mincore call
832 if (addr < current->start)
833 addr = current->start;
839 * scan this entry one page at a time
841 while (addr < cend) {
843 * Check pmap first, it is likely faster, also
844 * it can provide info as to whether we are the
845 * one referencing or modifying the page.
847 * If we have to check the VM object, only mess
848 * around with normal maps. Do not mess around
849 * with virtual page tables (XXX).
851 mincoreinfo = pmap_mincore(pmap, addr);
852 if (mincoreinfo == 0 &&
853 current->maptype == VM_MAPTYPE_NORMAL) {
859 * calculate the page index into the object
861 offset = current->offset + (addr - current->start);
862 pindex = OFF_TO_IDX(offset);
865 * if the page is resident, then gather
866 * information about it. spl protection is
867 * required to maintain the object
868 * association. And XXX what if the page is
869 * busy? What's the deal with that?
871 * XXX vm_token - legacy for pmap_ts_referenced
872 * in i386 and vkernel pmap code.
874 lwkt_gettoken(&vm_token);
875 vm_object_hold(current->object.vm_object);
876 m = vm_page_lookup(current->object.vm_object,
879 mincoreinfo = MINCORE_INCORE;
882 mincoreinfo |= MINCORE_MODIFIED_OTHER;
883 if ((m->flags & PG_REFERENCED) ||
884 pmap_ts_referenced(m)) {
885 vm_page_flag_set(m, PG_REFERENCED);
886 mincoreinfo |= MINCORE_REFERENCED_OTHER;
889 vm_object_drop(current->object.vm_object);
890 lwkt_reltoken(&vm_token);
894 * subyte may page fault. In case it needs to modify
895 * the map, we release the lock.
897 vm_map_unlock_read(map);
900 * calculate index into user supplied byte vector
902 vecindex = OFF_TO_IDX(addr - first_addr);
905 * If we have skipped map entries, we need to make sure that
906 * the byte vector is zeroed for those skipped entries.
908 while((lastvecindex + 1) < vecindex) {
909 error = subyte( vec + lastvecindex, 0);
918 * Pass the page information to the user
920 error = subyte( vec + vecindex, mincoreinfo);
927 * If the map has changed, due to the subyte, the previous
928 * output may be invalid.
930 vm_map_lock_read(map);
931 if (timestamp != map->timestamp)
934 lastvecindex = vecindex;
940 * subyte may page fault. In case it needs to modify
941 * the map, we release the lock.
943 vm_map_unlock_read(map);
946 * Zero the last entries in the byte vector.
948 vecindex = OFF_TO_IDX(end - first_addr);
949 while((lastvecindex + 1) < vecindex) {
950 error = subyte( vec + lastvecindex, 0);
959 * If the map has changed, due to the subyte, the previous
960 * output may be invalid.
962 vm_map_lock_read(map);
963 if (timestamp != map->timestamp)
965 vm_map_unlock_read(map);
969 lwkt_reltoken(&map->token);
974 * mlock system call handler
976 * mlock_args(const void *addr, size_t len)
981 sys_mlock(struct mlock_args *uap)
985 vm_size_t size, pageoff;
986 struct thread *td = curthread;
987 struct proc *p = td->td_proc;
990 addr = (vm_offset_t) uap->addr;
993 pageoff = (addr & PAGE_MASK);
996 size = (vm_size_t) round_page(size);
997 if (size < uap->len) /* wrap */
999 tmpaddr = addr + size; /* workaround gcc4 opt */
1000 if (tmpaddr < addr) /* wrap */
1003 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
1007 * We do not need to synchronize against other threads updating ucred;
1008 * they update p->ucred, which is synchronized into td_ucred ourselves.
1010 #ifdef pmap_wired_count
1011 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
1012 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) {
1016 error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1021 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
1022 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1031 sys_mlockall(struct mlockall_args *uap)
1033 #ifdef _P1003_1B_VISIBLE
1034 struct thread *td = curthread;
1035 struct proc *p = td->td_proc;
1036 vm_map_t map = &p->p_vmspace->vm_map;
1037 vm_map_entry_t entry;
1039 int rc = KERN_SUCCESS;
1041 if (((how & MCL_CURRENT) == 0) && ((how & MCL_FUTURE) == 0))
1044 rc = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1050 if (how & MCL_CURRENT) {
1051 for(entry = map->header.next;
1052 entry != &map->header;
1053 entry = entry->next);
1059 if (how & MCL_FUTURE)
1060 map->flags |= MAP_WIREFUTURE;
1065 #else /* !_P1003_1B_VISIBLE */
1067 #endif /* _P1003_1B_VISIBLE */
1073 * Unwire all user-wired map entries, cancel MCL_FUTURE.
1078 sys_munlockall(struct munlockall_args *uap)
1080 struct thread *td = curthread;
1081 struct proc *p = td->td_proc;
1082 vm_map_t map = &p->p_vmspace->vm_map;
1083 vm_map_entry_t entry;
1084 int rc = KERN_SUCCESS;
1088 /* Clear MAP_WIREFUTURE to cancel mlockall(MCL_FUTURE) */
1089 map->flags &= ~MAP_WIREFUTURE;
1092 for (entry = map->header.next;
1093 entry != &map->header;
1094 entry = entry->next) {
1095 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
1099 * If we encounter an in-transition entry, we release the
1100 * map lock and retry the scan; we do not decrement any
1101 * wired_count more than once because we do not touch
1102 * any entries with MAP_ENTRY_USER_WIRED not set.
1104 * There is a potential interleaving with concurrent
1105 * mlockall()s here -- if we abort a scan, an mlockall()
1106 * could start, wire a number of entries before our
1107 * current position in, and then stall itself on this
1108 * or any other in-transition entry. If that occurs, when
1109 * we resume, we will unwire those entries.
1111 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1112 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1113 ++mycpu->gd_cnt.v_intrans_coll;
1114 ++mycpu->gd_cnt.v_intrans_wait;
1115 vm_map_transition_wait(map);
1119 KASSERT(entry->wired_count > 0,
1120 ("wired_count was 0 with USER_WIRED set! %p", entry));
1122 /* Drop wired count, if it hits zero, unwire the entry */
1123 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1124 entry->wired_count--;
1125 if (entry->wired_count == 0)
1126 vm_fault_unwire(map, entry);
1136 * munlock system call handler
1138 * munlock_args(const void *addr, size_t len)
1143 sys_munlock(struct munlock_args *uap)
1145 struct thread *td = curthread;
1146 struct proc *p = td->td_proc;
1148 vm_offset_t tmpaddr;
1149 vm_size_t size, pageoff;
1152 addr = (vm_offset_t) uap->addr;
1155 pageoff = (addr & PAGE_MASK);
1158 size = (vm_size_t) round_page(size);
1160 tmpaddr = addr + size;
1161 if (tmpaddr < addr) /* wrap */
1164 #ifndef pmap_wired_count
1165 error = priv_check(td, PRIV_ROOT);
1170 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1171 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1175 * Internal version of mmap.
1176 * Currently used by mmap, exec, and sys5 shared memory.
1177 * Handle is either a vnode pointer or NULL for MAP_ANON.
1182 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1183 vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
1191 struct thread *td = curthread;
1193 int rv = KERN_SUCCESS;
1201 objsize = round_page(size);
1206 lwkt_gettoken(&map->token);
1209 * XXX messy code, fixme
1211 * NOTE: Overflow checks require discrete statements or GCC4
1212 * will optimize it out.
1214 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1215 esize = map->size + size; /* workaround gcc4 opt */
1216 if (esize < map->size ||
1217 esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1218 lwkt_reltoken(&map->token);
1224 * We currently can only deal with page aligned file offsets.
1225 * The check is here rather than in the syscall because the
1226 * kernel calls this function internally for other mmaping
1227 * operations (such as in exec) and non-aligned offsets will
1228 * cause pmap inconsistencies...so we want to be sure to
1229 * disallow this in all cases.
1231 * NOTE: Overflow checks require discrete statements or GCC4
1232 * will optimize it out.
1234 if (foff & PAGE_MASK) {
1235 lwkt_reltoken(&map->token);
1240 * Handle alignment. For large memory maps it is possible
1241 * that the MMU can optimize the page table so align anything
1242 * that is a multiple of SEG_SIZE to SEG_SIZE.
1244 * Also align any large mapping (bigger than 16x SG_SIZE) to a
1245 * SEG_SIZE address boundary.
1247 if (flags & MAP_SIZEALIGN) {
1249 if ((align ^ (align - 1)) != (align << 1) - 1) {
1250 lwkt_reltoken(&map->token);
1253 } else if ((flags & MAP_FIXED) == 0 &&
1254 ((size & SEG_MASK) == 0 || size > SEG_SIZE * 16)) {
1260 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1262 *addr = round_page(*addr);
1264 if (*addr != trunc_page(*addr)) {
1265 lwkt_reltoken(&map->token);
1268 eaddr = *addr + size;
1269 if (eaddr < *addr) {
1270 lwkt_reltoken(&map->token);
1274 if ((flags & MAP_TRYFIXED) == 0)
1275 vm_map_remove(map, *addr, *addr + size);
1279 * Lookup/allocate object.
1281 if (flags & MAP_ANON) {
1283 * Unnamed anonymous regions always start at 0.
1287 * Default memory object
1289 object = default_pager_alloc(handle, objsize,
1291 if (object == NULL) {
1292 lwkt_reltoken(&map->token);
1295 docow = MAP_PREFAULT_PARTIAL;
1298 * Implicit single instance of a default memory
1299 * object, so we don't need a VM object yet.
1307 vp = (struct vnode *)handle;
1308 if (vp->v_type == VCHR) {
1310 * Device mappings (device size unknown?).
1311 * Force them to be shared.
1313 error = dev_dmmap_single(vp->v_rdev, &foff, objsize,
1316 if (error == ENODEV) {
1317 handle = (void *)(intptr_t)vp->v_rdev;
1318 object = dev_pager_alloc(handle, objsize, prot, foff);
1319 if (object == NULL) {
1320 lwkt_reltoken(&map->token);
1324 docow = MAP_PREFAULT_PARTIAL;
1325 flags &= ~(MAP_PRIVATE|MAP_COPY);
1326 flags |= MAP_SHARED;
1329 * Regular file mapping (typically). The attribute
1330 * check is for the link count test only. Mmapble
1331 * vnodes must already have a VM object assigned.
1336 error = VOP_GETATTR(vp, &vat);
1338 lwkt_reltoken(&map->token);
1341 docow = MAP_PREFAULT_PARTIAL;
1342 object = vnode_pager_reference(vp);
1343 if (object == NULL && vp->v_type == VREG) {
1344 lwkt_reltoken(&map->token);
1345 kprintf("Warning: cannot mmap vnode %p, no "
1351 * If it is a regular file without any references
1352 * we do not need to sync it.
1354 if (vp->v_type == VREG && vat.va_nlink == 0) {
1355 flags |= MAP_NOSYNC;
1361 * Deal with the adjusted flags
1363 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1364 docow |= MAP_COPY_ON_WRITE;
1365 if (flags & MAP_NOSYNC)
1366 docow |= MAP_DISABLE_SYNCER;
1367 if (flags & MAP_NOCORE)
1368 docow |= MAP_DISABLE_COREDUMP;
1370 #if defined(VM_PROT_READ_IS_EXEC)
1371 if (prot & VM_PROT_READ)
1372 prot |= VM_PROT_EXECUTE;
1374 if (maxprot & VM_PROT_READ)
1375 maxprot |= VM_PROT_EXECUTE;
1379 * This may place the area in its own page directory if (size) is
1380 * large enough, otherwise it typically returns its argument.
1383 *addr = pmap_addr_hint(object, *addr, size);
1387 * Stack mappings need special attention.
1389 * Mappings that use virtual page tables will default to storing
1390 * the page table at offset 0.
1392 if (flags & MAP_STACK) {
1393 rv = vm_map_stack(map, *addr, size, flags,
1394 prot, maxprot, docow);
1395 } else if (flags & MAP_VPAGETABLE) {
1396 rv = vm_map_find(map, object, foff, addr, size, align,
1397 fitit, VM_MAPTYPE_VPAGETABLE,
1398 prot, maxprot, docow);
1400 rv = vm_map_find(map, object, foff, addr, size, align,
1401 fitit, VM_MAPTYPE_NORMAL,
1402 prot, maxprot, docow);
1405 if (rv != KERN_SUCCESS) {
1407 * Lose the object reference. Will destroy the
1408 * object if it's an unnamed anonymous mapping
1409 * or named anonymous without other references.
1411 vm_object_deallocate(object);
1416 * Shared memory is also shared with children.
1418 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1419 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1420 if (rv != KERN_SUCCESS) {
1421 vm_map_remove(map, *addr, *addr + size);
1426 /* If a process has marked all future mappings for wiring, do so */
1427 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1428 vm_map_unwire(map, *addr, *addr + size, FALSE);
1431 * Set the access time on the vnode
1434 vn_mark_atime(vp, td);
1436 lwkt_reltoken(&map->token);
1441 case KERN_INVALID_ADDRESS:
1444 case KERN_PROTECTION_FAILURE:
1452 * Translate a Mach VM return code to zero on success or the appropriate errno
1456 vm_mmap_to_errno(int rv)
1462 case KERN_INVALID_ADDRESS:
1465 case KERN_PROTECTION_FAILURE: