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 * 4. 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_page.h>
76 #include <vm/vm_kern.h>
78 #include <sys/file2.h>
79 #include <sys/thread.h>
80 #include <sys/thread2.h>
82 static int max_proc_mmap;
83 SYSCTL_INT(_vm, OID_AUTO, max_proc_mmap, CTLFLAG_RW, &max_proc_mmap, 0, "");
85 SYSCTL_INT(_vm, OID_AUTO, vkernel_enable, CTLFLAG_RW, &vkernel_enable, 0, "");
88 * Set the maximum number of vm_map_entry structures per process. Roughly
89 * speaking vm_map_entry structures are tiny, so allowing them to eat 1/100
90 * of our KVM malloc space still results in generous limits. We want a
91 * default that is good enough to prevent the kernel running out of resources
92 * if attacked from compromised user account but generous enough such that
93 * multi-threaded processes are not unduly inconvenienced.
96 static void vmmapentry_rsrc_init (void *);
97 SYSINIT(vmmersrc, SI_BOOT1_POST, SI_ORDER_ANY, vmmapentry_rsrc_init, NULL)
100 vmmapentry_rsrc_init(void *dummy)
102 max_proc_mmap = KvaSize / sizeof(struct vm_map_entry);
103 max_proc_mmap /= 100;
110 sys_sbrk(struct sbrk_args *uap)
112 /* Not yet implemented */
117 * sstk_args(int incr)
122 sys_sstk(struct sstk_args *uap)
124 /* Not yet implemented */
129 * mmap_args(void *addr, size_t len, int prot, int flags, int fd,
130 * long pad, off_t pos)
132 * Memory Map (mmap) system call. Note that the file offset
133 * and address are allowed to be NOT page aligned, though if
134 * the MAP_FIXED flag it set, both must have the same remainder
135 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
136 * page-aligned, the actual mapping starts at trunc_page(addr)
137 * and the return value is adjusted up by the page offset.
139 * Generally speaking, only character devices which are themselves
140 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
141 * there would be no cache coherency between a descriptor and a VM mapping
142 * both to the same character device.
144 * Block devices can be mmap'd no matter what they represent. Cache coherency
145 * is maintained as long as you do not write directly to the underlying
148 * No requirements; sys_mmap path holds the vm_token
151 kern_mmap(struct vmspace *vms, caddr_t uaddr, size_t ulen,
152 int uprot, int uflags, int fd, off_t upos, void **res)
154 struct thread *td = curthread;
155 struct proc *p = td->td_proc;
156 struct file *fp = NULL;
160 vm_size_t size, pageoff;
161 vm_prot_t prot, maxprot;
169 addr = (vm_offset_t) uaddr;
171 prot = uprot & VM_PROT_ALL;
176 * Make sure mapping fits into numeric range etc.
178 * NOTE: We support the full unsigned range for size now.
180 if (((flags & MAP_ANON) && (fd != -1 || pos != 0)))
183 if (flags & MAP_STACK) {
185 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
192 * Virtual page tables cannot be used with MAP_STACK. Apart from
193 * it not making any sense, the aux union is used by both
196 * Because the virtual page table is stored in the backing object
197 * and might be updated by the kernel, the mapping must be R+W.
199 if (flags & MAP_VPAGETABLE) {
200 if (vkernel_enable == 0)
202 if (flags & MAP_STACK)
204 if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
209 * Align the file position to a page boundary,
210 * and save its page offset component.
212 pageoff = (pos & PAGE_MASK);
215 /* Adjust size for rounding (on both ends). */
216 size += pageoff; /* low end... */
217 size = (vm_size_t) round_page(size); /* hi end */
218 if (size < ulen) /* wrap */
222 * Check for illegal addresses. Watch out for address wrap... Note
223 * that VM_*_ADDRESS are not constants due to casts (argh).
225 if (flags & (MAP_FIXED | MAP_TRYFIXED)) {
227 * The specified address must have the same remainder
228 * as the file offset taken modulo PAGE_SIZE, so it
229 * should be aligned after adjustment by pageoff.
232 if (addr & PAGE_MASK)
236 * Address range must be all in user VM space and not wrap.
238 tmpaddr = addr + size;
241 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
243 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
247 * Get a hint of where to map. It also provides mmap offset
248 * randomization if enabled.
250 addr = vm_map_hint(p, addr, prot);
253 if (flags & MAP_ANON) {
255 * Mapping blank space is trivial.
258 maxprot = VM_PROT_ALL;
261 * Mapping file, get fp for validation. Obtain vnode and make
262 * sure it is of appropriate type.
264 fp = holdfp(p->p_fd, fd, -1);
267 if (fp->f_type != DTYPE_VNODE) {
272 * POSIX shared-memory objects are defined to have
273 * kernel persistence, and are not defined to support
274 * read(2)/write(2) -- or even open(2). Thus, we can
275 * use MAP_ASYNC to trade on-disk coherence for speed.
276 * The shm_open(3) library routine turns on the FPOSIXSHM
277 * flag to request this behavior.
279 if (fp->f_flag & FPOSIXSHM)
281 vp = (struct vnode *) fp->f_data;
284 * Validate the vnode for the operation.
289 * Get the proper underlying object
291 if ((obj = vp->v_object) == NULL) {
295 KKASSERT((struct vnode *)obj->handle == vp);
299 * Make sure a device has not been revoked.
300 * Mappability is handled by the device layer.
302 if (vp->v_rdev == NULL) {
309 * Nothing else is mappable.
316 * XXX hack to handle use of /dev/zero to map anon memory (ala
319 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
321 maxprot = VM_PROT_ALL;
326 * cdevs does not provide private mappings of any kind.
328 if (vp->v_type == VCHR &&
329 (flags & (MAP_PRIVATE|MAP_COPY))) {
334 * Ensure that file and memory protections are
335 * compatible. Note that we only worry about
336 * writability if mapping is shared; in this case,
337 * current and max prot are dictated by the open file.
338 * XXX use the vnode instead? Problem is: what
339 * credentials do we use for determination? What if
340 * proc does a setuid?
342 maxprot = VM_PROT_EXECUTE; /* ??? */
343 if (fp->f_flag & FREAD) {
344 maxprot |= VM_PROT_READ;
345 } else if (prot & PROT_READ) {
350 * If we are sharing potential changes (either via
351 * MAP_SHARED or via the implicit sharing of character
352 * device mappings), and we are trying to get write
353 * permission although we opened it without asking
354 * for it, bail out. Check for superuser, only if
355 * we're at securelevel < 1, to allow the XIG X server
356 * to continue to work.
358 if ((flags & MAP_SHARED) != 0 || vp->v_type == VCHR) {
359 if ((fp->f_flag & FWRITE) != 0) {
361 if ((error = VOP_GETATTR(vp, &va))) {
365 (IMMUTABLE|APPEND)) == 0) {
366 maxprot |= VM_PROT_WRITE;
367 } else if (prot & PROT_WRITE) {
371 } else if ((prot & PROT_WRITE) != 0) {
376 maxprot |= VM_PROT_WRITE;
382 /* Token serializes access to vm_map.nentries against vm_mmap */
383 lwkt_gettoken(&vm_token);
386 * Do not allow more then a certain number of vm_map_entry structures
387 * per process. Scale with the number of rforks sharing the map
388 * to make the limit reasonable for threads.
391 vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
393 lwkt_reltoken(&vm_token);
397 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
400 *res = (void *)(addr + pageoff);
402 lwkt_reltoken(&vm_token);
411 * mmap system call handler
416 sys_mmap(struct mmap_args *uap)
420 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
421 uap->prot, uap->flags,
422 uap->fd, uap->pos, &uap->sysmsg_resultp);
428 * msync system call handler
430 * msync_args(void *addr, size_t len, int flags)
435 sys_msync(struct msync_args *uap)
437 struct proc *p = curproc;
440 vm_size_t size, pageoff;
445 addr = (vm_offset_t) uap->addr;
449 pageoff = (addr & PAGE_MASK);
452 size = (vm_size_t) round_page(size);
453 if (size < uap->len) /* wrap */
455 tmpaddr = addr + size; /* workaround gcc4 opt */
456 if (tmpaddr < addr) /* wrap */
459 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
462 map = &p->p_vmspace->vm_map;
465 * vm_token serializes extracting the address range for size == 0
466 * msyncs with the vm_map_clean call; if the token were not held
467 * across the two calls, an intervening munmap/mmap pair, for example,
468 * could cause msync to occur on a wrong region.
470 lwkt_gettoken(&vm_token);
473 * XXX Gak! If size is zero we are supposed to sync "all modified
474 * pages with the region containing addr". Unfortunately, we don't
475 * really keep track of individual mmaps so we approximate by flushing
476 * the range of the map entry containing addr. This can be incorrect
477 * if the region splits or is coalesced with a neighbor.
480 vm_map_entry_t entry;
482 vm_map_lock_read(map);
483 rv = vm_map_lookup_entry(map, addr, &entry);
485 vm_map_unlock_read(map);
486 rv = KERN_INVALID_ADDRESS;
490 size = entry->end - entry->start;
491 vm_map_unlock_read(map);
495 * Clean the pages and interpret the return value.
497 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
498 (flags & MS_INVALIDATE) != 0);
500 lwkt_reltoken(&vm_token);
505 case KERN_INVALID_ADDRESS:
506 return (EINVAL); /* Sun returns ENOMEM? */
517 * munmap system call handler
519 * munmap_args(void *addr, size_t len)
524 sys_munmap(struct munmap_args *uap)
526 struct proc *p = curproc;
529 vm_size_t size, pageoff;
532 addr = (vm_offset_t) uap->addr;
535 pageoff = (addr & PAGE_MASK);
538 size = (vm_size_t) round_page(size);
539 if (size < uap->len) /* wrap */
541 tmpaddr = addr + size; /* workaround gcc4 opt */
542 if (tmpaddr < addr) /* wrap */
549 * Check for illegal addresses. Watch out for address wrap... Note
550 * that VM_*_ADDRESS are not constants due to casts (argh).
552 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
554 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
557 map = &p->p_vmspace->vm_map;
559 /* vm_token serializes between the map check and the actual unmap */
560 lwkt_gettoken(&vm_token);
563 * Make sure entire range is allocated.
565 if (!vm_map_check_protection(map, addr, addr + size,
566 VM_PROT_NONE, FALSE)) {
567 lwkt_reltoken(&vm_token);
570 /* returns nothing but KERN_SUCCESS anyway */
571 vm_map_remove(map, addr, addr + size);
572 lwkt_reltoken(&vm_token);
577 * mprotect_args(const void *addr, size_t len, int prot)
582 sys_mprotect(struct mprotect_args *uap)
584 struct proc *p = curproc;
587 vm_size_t size, pageoff;
591 addr = (vm_offset_t) uap->addr;
593 prot = uap->prot & VM_PROT_ALL;
594 #if defined(VM_PROT_READ_IS_EXEC)
595 if (prot & VM_PROT_READ)
596 prot |= VM_PROT_EXECUTE;
599 pageoff = (addr & PAGE_MASK);
602 size = (vm_size_t) round_page(size);
603 if (size < uap->len) /* wrap */
605 tmpaddr = addr + size; /* workaround gcc4 opt */
606 if (tmpaddr < addr) /* wrap */
609 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size,
614 case KERN_PROTECTION_FAILURE:
625 * minherit system call handler
627 * minherit_args(void *addr, size_t len, int inherit)
632 sys_minherit(struct minherit_args *uap)
634 struct proc *p = curproc;
637 vm_size_t size, pageoff;
638 vm_inherit_t inherit;
641 addr = (vm_offset_t)uap->addr;
643 inherit = uap->inherit;
645 pageoff = (addr & PAGE_MASK);
648 size = (vm_size_t) round_page(size);
649 if (size < uap->len) /* wrap */
651 tmpaddr = addr + size; /* workaround gcc4 opt */
652 if (tmpaddr < addr) /* wrap */
655 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr,
656 addr + size, inherit)) {
660 case KERN_PROTECTION_FAILURE:
671 * madvise system call handler
673 * madvise_args(void *addr, size_t len, int behav)
678 sys_madvise(struct madvise_args *uap)
680 struct proc *p = curproc;
681 vm_offset_t start, end;
682 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
686 * Check for illegal behavior
688 if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
691 * Check for illegal addresses. Watch out for address wrap... Note
692 * that VM_*_ADDRESS are not constants due to casts (argh).
694 if (tmpaddr < (vm_offset_t)uap->addr)
696 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
698 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
702 * Since this routine is only advisory, we default to conservative
705 start = trunc_page((vm_offset_t)uap->addr);
706 end = round_page(tmpaddr);
708 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
714 * mcontrol system call handler
716 * mcontrol_args(void *addr, size_t len, int behav, off_t value)
721 sys_mcontrol(struct mcontrol_args *uap)
723 struct proc *p = curproc;
724 vm_offset_t start, end;
725 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
729 * Check for illegal behavior
731 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
734 * Check for illegal addresses. Watch out for address wrap... Note
735 * that VM_*_ADDRESS are not constants due to casts (argh).
737 if (tmpaddr < (vm_offset_t) uap->addr)
739 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
741 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
745 * Since this routine is only advisory, we default to conservative
748 start = trunc_page((vm_offset_t)uap->addr);
749 end = round_page(tmpaddr);
751 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
752 uap->behav, uap->value);
758 * mincore system call handler
760 * mincore_args(const void *addr, size_t len, char *vec)
765 sys_mincore(struct mincore_args *uap)
767 struct proc *p = curproc;
768 vm_offset_t addr, first_addr;
769 vm_offset_t end, cend;
774 int vecindex, lastvecindex;
775 vm_map_entry_t current;
776 vm_map_entry_t entry;
778 unsigned int timestamp;
781 * Make sure that the addresses presented are valid for user
784 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
785 end = addr + (vm_size_t)round_page(uap->len);
788 if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
792 * Address of byte vector
796 map = &p->p_vmspace->vm_map;
797 pmap = vmspace_pmap(p->p_vmspace);
799 lwkt_gettoken(&vm_token);
800 vm_map_lock_read(map);
802 timestamp = map->timestamp;
804 if (!vm_map_lookup_entry(map, addr, &entry))
808 * Do this on a map entry basis so that if the pages are not
809 * in the current processes address space, we can easily look
810 * up the pages elsewhere.
814 (current != &map->header) && (current->start < end);
815 current = current->next) {
818 * ignore submaps (for now) or null objects
820 if (current->maptype != VM_MAPTYPE_NORMAL &&
821 current->maptype != VM_MAPTYPE_VPAGETABLE) {
824 if (current->object.vm_object == NULL)
828 * limit this scan to the current map entry and the
829 * limits for the mincore call
831 if (addr < current->start)
832 addr = current->start;
838 * scan this entry one page at a time
840 while (addr < cend) {
842 * Check pmap first, it is likely faster, also
843 * it can provide info as to whether we are the
844 * one referencing or modifying the page.
846 * If we have to check the VM object, only mess
847 * around with normal maps. Do not mess around
848 * with virtual page tables (XXX).
850 mincoreinfo = pmap_mincore(pmap, addr);
851 if (mincoreinfo == 0 &&
852 current->maptype == VM_MAPTYPE_NORMAL) {
858 * calculate the page index into the object
860 offset = current->offset + (addr - current->start);
861 pindex = OFF_TO_IDX(offset);
864 * if the page is resident, then gather
865 * information about it. spl protection is
866 * required to maintain the object
867 * association. And XXX what if the page is
868 * busy? What's the deal with that?
871 m = vm_page_lookup(current->object.vm_object,
874 mincoreinfo = MINCORE_INCORE;
877 mincoreinfo |= MINCORE_MODIFIED_OTHER;
878 if ((m->flags & PG_REFERENCED) ||
879 pmap_ts_referenced(m)) {
880 vm_page_flag_set(m, PG_REFERENCED);
881 mincoreinfo |= MINCORE_REFERENCED_OTHER;
888 * subyte may page fault. In case it needs to modify
889 * the map, we release the lock.
891 vm_map_unlock_read(map);
894 * calculate index into user supplied byte vector
896 vecindex = OFF_TO_IDX(addr - first_addr);
899 * If we have skipped map entries, we need to make sure that
900 * the byte vector is zeroed for those skipped entries.
902 while((lastvecindex + 1) < vecindex) {
903 error = subyte( vec + lastvecindex, 0);
912 * Pass the page information to the user
914 error = subyte( vec + vecindex, mincoreinfo);
921 * If the map has changed, due to the subyte, the previous
922 * output may be invalid.
924 vm_map_lock_read(map);
925 if (timestamp != map->timestamp)
928 lastvecindex = vecindex;
934 * subyte may page fault. In case it needs to modify
935 * the map, we release the lock.
937 vm_map_unlock_read(map);
940 * Zero the last entries in the byte vector.
942 vecindex = OFF_TO_IDX(end - first_addr);
943 while((lastvecindex + 1) < vecindex) {
944 error = subyte( vec + lastvecindex, 0);
953 * If the map has changed, due to the subyte, the previous
954 * output may be invalid.
956 vm_map_lock_read(map);
957 if (timestamp != map->timestamp)
959 vm_map_unlock_read(map);
963 lwkt_reltoken(&vm_token);
968 * mlock system call handler
970 * mlock_args(const void *addr, size_t len)
975 sys_mlock(struct mlock_args *uap)
979 vm_size_t size, pageoff;
980 struct thread *td = curthread;
981 struct proc *p = td->td_proc;
984 addr = (vm_offset_t) uap->addr;
987 pageoff = (addr & PAGE_MASK);
990 size = (vm_size_t) round_page(size);
991 if (size < uap->len) /* wrap */
993 tmpaddr = addr + size; /* workaround gcc4 opt */
994 if (tmpaddr < addr) /* wrap */
997 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
1001 * We do not need to synchronize against other threads updating ucred;
1002 * they update p->ucred, which is synchronized into td_ucred ourselves.
1004 #ifdef pmap_wired_count
1005 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
1006 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) {
1010 error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1015 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
1016 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1025 sys_mlockall(struct mlockall_args *uap)
1027 #ifdef _P1003_1B_VISIBLE
1028 struct thread *td = curthread;
1029 struct proc *p = td->td_proc;
1030 vm_map_t map = &p->p_vmspace->vm_map;
1031 vm_map_entry_t entry;
1033 int rc = KERN_SUCCESS;
1035 if (((how & MCL_CURRENT) == 0) && ((how & MCL_FUTURE) == 0))
1038 rc = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1044 if (how & MCL_CURRENT) {
1045 for(entry = map->header.next;
1046 entry != &map->header;
1047 entry = entry->next);
1053 if (how & MCL_FUTURE)
1054 map->flags |= MAP_WIREFUTURE;
1059 #else /* !_P1003_1B_VISIBLE */
1061 #endif /* _P1003_1B_VISIBLE */
1067 * Unwire all user-wired map entries, cancel MCL_FUTURE.
1072 sys_munlockall(struct munlockall_args *uap)
1074 struct thread *td = curthread;
1075 struct proc *p = td->td_proc;
1076 vm_map_t map = &p->p_vmspace->vm_map;
1077 vm_map_entry_t entry;
1078 int rc = KERN_SUCCESS;
1082 /* Clear MAP_WIREFUTURE to cancel mlockall(MCL_FUTURE) */
1083 map->flags &= ~MAP_WIREFUTURE;
1086 for (entry = map->header.next;
1087 entry != &map->header;
1088 entry = entry->next) {
1089 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
1093 * If we encounter an in-transition entry, we release the
1094 * map lock and retry the scan; we do not decrement any
1095 * wired_count more than once because we do not touch
1096 * any entries with MAP_ENTRY_USER_WIRED not set.
1098 * There is a potential interleaving with concurrent
1099 * mlockall()s here -- if we abort a scan, an mlockall()
1100 * could start, wire a number of entries before our
1101 * current position in, and then stall itself on this
1102 * or any other in-transition entry. If that occurs, when
1103 * we resume, we will unwire those entries.
1105 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1106 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1107 ++mycpu->gd_cnt.v_intrans_coll;
1108 ++mycpu->gd_cnt.v_intrans_wait;
1109 vm_map_transition_wait(map);
1113 KASSERT(entry->wired_count > 0,
1114 ("wired_count was 0 with USER_WIRED set! %p", entry));
1116 /* Drop wired count, if it hits zero, unwire the entry */
1117 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1118 entry->wired_count--;
1119 if (entry->wired_count == 0)
1120 vm_fault_unwire(map, entry);
1130 * munlock system call handler
1132 * munlock_args(const void *addr, size_t len)
1137 sys_munlock(struct munlock_args *uap)
1139 struct thread *td = curthread;
1140 struct proc *p = td->td_proc;
1142 vm_offset_t tmpaddr;
1143 vm_size_t size, pageoff;
1146 addr = (vm_offset_t) uap->addr;
1149 pageoff = (addr & PAGE_MASK);
1152 size = (vm_size_t) round_page(size);
1154 tmpaddr = addr + size;
1155 if (tmpaddr < addr) /* wrap */
1158 #ifndef pmap_wired_count
1159 error = priv_check(td, PRIV_ROOT);
1164 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1165 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1169 * Internal version of mmap.
1170 * Currently used by mmap, exec, and sys5 shared memory.
1171 * Handle is either a vnode pointer or NULL for MAP_ANON.
1173 * No requirements; kern_mmap path holds the vm_token
1176 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1177 vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
1184 struct thread *td = curthread;
1186 int rv = KERN_SUCCESS;
1193 objsize = round_page(size);
1198 lwkt_gettoken(&vm_token);
1201 * XXX messy code, fixme
1203 * NOTE: Overflow checks require discrete statements or GCC4
1204 * will optimize it out.
1206 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1207 esize = map->size + size; /* workaround gcc4 opt */
1208 if (esize < map->size ||
1209 esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1210 lwkt_reltoken(&vm_token);
1216 * We currently can only deal with page aligned file offsets.
1217 * The check is here rather than in the syscall because the
1218 * kernel calls this function internally for other mmaping
1219 * operations (such as in exec) and non-aligned offsets will
1220 * cause pmap inconsistencies...so we want to be sure to
1221 * disallow this in all cases.
1223 * NOTE: Overflow checks require discrete statements or GCC4
1224 * will optimize it out.
1226 if (foff & PAGE_MASK) {
1227 lwkt_reltoken(&vm_token);
1231 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1233 *addr = round_page(*addr);
1235 if (*addr != trunc_page(*addr)) {
1236 lwkt_reltoken(&vm_token);
1239 eaddr = *addr + size;
1240 if (eaddr < *addr) {
1241 lwkt_reltoken(&vm_token);
1245 if ((flags & MAP_TRYFIXED) == 0)
1246 vm_map_remove(map, *addr, *addr + size);
1250 * Lookup/allocate object.
1252 if (flags & MAP_ANON) {
1254 * Unnamed anonymous regions always start at 0.
1258 * Default memory object
1260 object = default_pager_alloc(handle, objsize,
1262 if (object == NULL) {
1263 lwkt_reltoken(&vm_token);
1266 docow = MAP_PREFAULT_PARTIAL;
1269 * Implicit single instance of a default memory
1270 * object, so we don't need a VM object yet.
1278 vp = (struct vnode *)handle;
1279 if (vp->v_type == VCHR) {
1281 * Device mappings (device size unknown?).
1282 * Force them to be shared.
1284 handle = (void *)(intptr_t)vp->v_rdev;
1285 object = dev_pager_alloc(handle, objsize, prot, foff);
1286 if (object == NULL) {
1287 lwkt_reltoken(&vm_token);
1290 docow = MAP_PREFAULT_PARTIAL;
1291 flags &= ~(MAP_PRIVATE|MAP_COPY);
1292 flags |= MAP_SHARED;
1295 * Regular file mapping (typically). The attribute
1296 * check is for the link count test only. Mmapble
1297 * vnodes must already have a VM object assigned.
1302 error = VOP_GETATTR(vp, &vat);
1304 lwkt_reltoken(&vm_token);
1307 docow = MAP_PREFAULT_PARTIAL;
1308 object = vnode_pager_reference(vp);
1309 if (object == NULL && vp->v_type == VREG) {
1310 lwkt_reltoken(&vm_token);
1311 kprintf("Warning: cannot mmap vnode %p, no "
1317 * If it is a regular file without any references
1318 * we do not need to sync it.
1320 if (vp->v_type == VREG && vat.va_nlink == 0) {
1321 flags |= MAP_NOSYNC;
1327 * Deal with the adjusted flags
1329 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1330 docow |= MAP_COPY_ON_WRITE;
1331 if (flags & MAP_NOSYNC)
1332 docow |= MAP_DISABLE_SYNCER;
1333 if (flags & MAP_NOCORE)
1334 docow |= MAP_DISABLE_COREDUMP;
1336 #if defined(VM_PROT_READ_IS_EXEC)
1337 if (prot & VM_PROT_READ)
1338 prot |= VM_PROT_EXECUTE;
1340 if (maxprot & VM_PROT_READ)
1341 maxprot |= VM_PROT_EXECUTE;
1345 * This may place the area in its own page directory if (size) is
1346 * large enough, otherwise it typically returns its argument.
1349 *addr = pmap_addr_hint(object, *addr, size);
1353 * Stack mappings need special attention.
1355 * Mappings that use virtual page tables will default to storing
1356 * the page table at offset 0.
1358 if (flags & MAP_STACK) {
1359 rv = vm_map_stack(map, *addr, size, flags,
1360 prot, maxprot, docow);
1361 } else if (flags & MAP_VPAGETABLE) {
1362 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1363 fitit, VM_MAPTYPE_VPAGETABLE,
1364 prot, maxprot, docow);
1366 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1367 fitit, VM_MAPTYPE_NORMAL,
1368 prot, maxprot, docow);
1371 if (rv != KERN_SUCCESS) {
1373 * Lose the object reference. Will destroy the
1374 * object if it's an unnamed anonymous mapping
1375 * or named anonymous without other references.
1377 vm_object_deallocate(object);
1382 * Shared memory is also shared with children.
1384 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1385 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1386 if (rv != KERN_SUCCESS) {
1387 vm_map_remove(map, *addr, *addr + size);
1392 /* If a process has marked all future mappings for wiring, do so */
1393 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1394 vm_map_unwire(map, *addr, *addr + size, FALSE);
1397 * Set the access time on the vnode
1400 vn_mark_atime(vp, td);
1402 lwkt_reltoken(&vm_token);
1407 case KERN_INVALID_ADDRESS:
1410 case KERN_PROTECTION_FAILURE: