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)))
186 if (flags & MAP_STACK) {
188 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
195 * Virtual page tables cannot be used with MAP_STACK. Apart from
196 * it not making any sense, the aux union is used by both
199 * Because the virtual page table is stored in the backing object
200 * and might be updated by the kernel, the mapping must be R+W.
202 if (flags & MAP_VPAGETABLE) {
203 if (vkernel_enable == 0)
205 if (flags & MAP_STACK)
207 if ((prot & (PROT_READ|PROT_WRITE)) != (PROT_READ|PROT_WRITE))
212 * Align the file position to a page boundary,
213 * and save its page offset component.
215 pageoff = (pos & PAGE_MASK);
218 /* Adjust size for rounding (on both ends). */
219 size += pageoff; /* low end... */
220 size = (vm_size_t) round_page(size); /* hi end */
221 if (size < ulen) /* wrap */
225 * Check for illegal addresses. Watch out for address wrap... Note
226 * that VM_*_ADDRESS are not constants due to casts (argh).
228 if (flags & (MAP_FIXED | MAP_TRYFIXED)) {
230 * The specified address must have the same remainder
231 * as the file offset taken modulo PAGE_SIZE, so it
232 * should be aligned after adjustment by pageoff.
235 if (addr & PAGE_MASK)
239 * Address range must be all in user VM space and not wrap.
241 tmpaddr = addr + size;
244 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
246 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
250 * Get a hint of where to map. It also provides mmap offset
251 * randomization if enabled.
253 addr = vm_map_hint(p, addr, prot);
256 if (flags & MAP_ANON) {
258 * Mapping blank space is trivial.
261 maxprot = VM_PROT_ALL;
264 * Mapping file, get fp for validation. Obtain vnode and make
265 * sure it is of appropriate type.
267 fp = holdfp(p->p_fd, fd, -1);
270 if (fp->f_type != DTYPE_VNODE) {
275 * POSIX shared-memory objects are defined to have
276 * kernel persistence, and are not defined to support
277 * read(2)/write(2) -- or even open(2). Thus, we can
278 * use MAP_ASYNC to trade on-disk coherence for speed.
279 * The shm_open(3) library routine turns on the FPOSIXSHM
280 * flag to request this behavior.
282 if (fp->f_flag & FPOSIXSHM)
284 vp = (struct vnode *) fp->f_data;
287 * Validate the vnode for the operation.
292 * Get the proper underlying object
294 if ((obj = vp->v_object) == NULL) {
298 KKASSERT((struct vnode *)obj->handle == vp);
302 * Make sure a device has not been revoked.
303 * Mappability is handled by the device layer.
305 if (vp->v_rdev == NULL) {
312 * Nothing else is mappable.
319 * XXX hack to handle use of /dev/zero to map anon memory (ala
322 if (vp->v_type == VCHR && iszerodev(vp->v_rdev)) {
324 maxprot = VM_PROT_ALL;
329 * cdevs does not provide private mappings of any kind.
331 if (vp->v_type == VCHR &&
332 (flags & (MAP_PRIVATE|MAP_COPY))) {
337 * Ensure that file and memory protections are
338 * compatible. Note that we only worry about
339 * writability if mapping is shared; in this case,
340 * current and max prot are dictated by the open file.
341 * XXX use the vnode instead? Problem is: what
342 * credentials do we use for determination? What if
343 * proc does a setuid?
345 maxprot = VM_PROT_EXECUTE; /* ??? */
346 if (fp->f_flag & FREAD) {
347 maxprot |= VM_PROT_READ;
348 } else if (prot & PROT_READ) {
353 * If we are sharing potential changes (either via
354 * MAP_SHARED or via the implicit sharing of character
355 * device mappings), and we are trying to get write
356 * permission although we opened it without asking
357 * for it, bail out. Check for superuser, only if
358 * we're at securelevel < 1, to allow the XIG X server
359 * to continue to work.
361 if ((flags & MAP_SHARED) != 0 || vp->v_type == VCHR) {
362 if ((fp->f_flag & FWRITE) != 0) {
364 if ((error = VOP_GETATTR(vp, &va))) {
368 (IMMUTABLE|APPEND)) == 0) {
369 maxprot |= VM_PROT_WRITE;
370 } else if (prot & PROT_WRITE) {
374 } else if ((prot & PROT_WRITE) != 0) {
379 maxprot |= VM_PROT_WRITE;
385 /* Token serializes access to vm_map.nentries against vm_mmap */
386 lwkt_gettoken(&vm_token);
389 * Do not allow more then a certain number of vm_map_entry structures
390 * per process. Scale with the number of rforks sharing the map
391 * to make the limit reasonable for threads.
394 vms->vm_map.nentries >= max_proc_mmap * vms->vm_sysref.refcnt) {
396 lwkt_reltoken(&vm_token);
400 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
403 *res = (void *)(addr + pageoff);
405 lwkt_reltoken(&vm_token);
414 * mmap system call handler
419 sys_mmap(struct mmap_args *uap)
423 error = kern_mmap(curproc->p_vmspace, uap->addr, uap->len,
424 uap->prot, uap->flags,
425 uap->fd, uap->pos, &uap->sysmsg_resultp);
431 * msync system call handler
433 * msync_args(void *addr, size_t len, int flags)
438 sys_msync(struct msync_args *uap)
440 struct proc *p = curproc;
443 vm_size_t size, pageoff;
448 addr = (vm_offset_t) uap->addr;
452 pageoff = (addr & PAGE_MASK);
455 size = (vm_size_t) round_page(size);
456 if (size < uap->len) /* wrap */
458 tmpaddr = addr + size; /* workaround gcc4 opt */
459 if (tmpaddr < addr) /* wrap */
462 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
465 map = &p->p_vmspace->vm_map;
468 * vm_token serializes extracting the address range for size == 0
469 * msyncs with the vm_map_clean call; if the token were not held
470 * across the two calls, an intervening munmap/mmap pair, for example,
471 * could cause msync to occur on a wrong region.
473 lwkt_gettoken(&vm_token);
476 * XXX Gak! If size is zero we are supposed to sync "all modified
477 * pages with the region containing addr". Unfortunately, we don't
478 * really keep track of individual mmaps so we approximate by flushing
479 * the range of the map entry containing addr. This can be incorrect
480 * if the region splits or is coalesced with a neighbor.
483 vm_map_entry_t entry;
485 vm_map_lock_read(map);
486 rv = vm_map_lookup_entry(map, addr, &entry);
488 vm_map_unlock_read(map);
489 rv = KERN_INVALID_ADDRESS;
493 size = entry->end - entry->start;
494 vm_map_unlock_read(map);
498 * Clean the pages and interpret the return value.
500 rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
501 (flags & MS_INVALIDATE) != 0);
503 lwkt_reltoken(&vm_token);
508 case KERN_INVALID_ADDRESS:
509 return (EINVAL); /* Sun returns ENOMEM? */
520 * munmap system call handler
522 * munmap_args(void *addr, size_t len)
527 sys_munmap(struct munmap_args *uap)
529 struct proc *p = curproc;
532 vm_size_t size, pageoff;
535 addr = (vm_offset_t) uap->addr;
538 pageoff = (addr & PAGE_MASK);
541 size = (vm_size_t) round_page(size);
542 if (size < uap->len) /* wrap */
544 tmpaddr = addr + size; /* workaround gcc4 opt */
545 if (tmpaddr < addr) /* wrap */
552 * Check for illegal addresses. Watch out for address wrap... Note
553 * that VM_*_ADDRESS are not constants due to casts (argh).
555 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
557 if (VM_MIN_USER_ADDRESS > 0 && addr < VM_MIN_USER_ADDRESS)
560 map = &p->p_vmspace->vm_map;
562 /* vm_token serializes between the map check and the actual unmap */
563 lwkt_gettoken(&vm_token);
566 * Make sure entire range is allocated.
568 if (!vm_map_check_protection(map, addr, addr + size,
569 VM_PROT_NONE, FALSE)) {
570 lwkt_reltoken(&vm_token);
573 /* returns nothing but KERN_SUCCESS anyway */
574 vm_map_remove(map, addr, addr + size);
575 lwkt_reltoken(&vm_token);
580 * mprotect_args(const void *addr, size_t len, int prot)
585 sys_mprotect(struct mprotect_args *uap)
587 struct proc *p = curproc;
590 vm_size_t size, pageoff;
594 addr = (vm_offset_t) uap->addr;
596 prot = uap->prot & VM_PROT_ALL;
597 #if defined(VM_PROT_READ_IS_EXEC)
598 if (prot & VM_PROT_READ)
599 prot |= VM_PROT_EXECUTE;
602 pageoff = (addr & PAGE_MASK);
605 size = (vm_size_t) round_page(size);
606 if (size < uap->len) /* wrap */
608 tmpaddr = addr + size; /* workaround gcc4 opt */
609 if (tmpaddr < addr) /* wrap */
612 switch (vm_map_protect(&p->p_vmspace->vm_map, addr, addr + size,
617 case KERN_PROTECTION_FAILURE:
628 * minherit system call handler
630 * minherit_args(void *addr, size_t len, int inherit)
635 sys_minherit(struct minherit_args *uap)
637 struct proc *p = curproc;
640 vm_size_t size, pageoff;
641 vm_inherit_t inherit;
644 addr = (vm_offset_t)uap->addr;
646 inherit = uap->inherit;
648 pageoff = (addr & PAGE_MASK);
651 size = (vm_size_t) round_page(size);
652 if (size < uap->len) /* wrap */
654 tmpaddr = addr + size; /* workaround gcc4 opt */
655 if (tmpaddr < addr) /* wrap */
658 switch (vm_map_inherit(&p->p_vmspace->vm_map, addr,
659 addr + size, inherit)) {
663 case KERN_PROTECTION_FAILURE:
674 * madvise system call handler
676 * madvise_args(void *addr, size_t len, int behav)
681 sys_madvise(struct madvise_args *uap)
683 struct proc *p = curproc;
684 vm_offset_t start, end;
685 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
689 * Check for illegal behavior
691 if (uap->behav < 0 || uap->behav >= MADV_CONTROL_END)
694 * Check for illegal addresses. Watch out for address wrap... Note
695 * that VM_*_ADDRESS are not constants due to casts (argh).
697 if (tmpaddr < (vm_offset_t)uap->addr)
699 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
701 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
705 * Since this routine is only advisory, we default to conservative
708 start = trunc_page((vm_offset_t)uap->addr);
709 end = round_page(tmpaddr);
711 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
717 * mcontrol system call handler
719 * mcontrol_args(void *addr, size_t len, int behav, off_t value)
724 sys_mcontrol(struct mcontrol_args *uap)
726 struct proc *p = curproc;
727 vm_offset_t start, end;
728 vm_offset_t tmpaddr = (vm_offset_t)uap->addr + uap->len;
732 * Check for illegal behavior
734 if (uap->behav < 0 || uap->behav > MADV_CONTROL_END)
737 * Check for illegal addresses. Watch out for address wrap... Note
738 * that VM_*_ADDRESS are not constants due to casts (argh).
740 if (tmpaddr < (vm_offset_t) uap->addr)
742 if (VM_MAX_USER_ADDRESS > 0 && tmpaddr > VM_MAX_USER_ADDRESS)
744 if (VM_MIN_USER_ADDRESS > 0 && uap->addr < VM_MIN_USER_ADDRESS)
748 * Since this routine is only advisory, we default to conservative
751 start = trunc_page((vm_offset_t)uap->addr);
752 end = round_page(tmpaddr);
754 error = vm_map_madvise(&p->p_vmspace->vm_map, start, end,
755 uap->behav, uap->value);
761 * mincore system call handler
763 * mincore_args(const void *addr, size_t len, char *vec)
768 sys_mincore(struct mincore_args *uap)
770 struct proc *p = curproc;
771 vm_offset_t addr, first_addr;
772 vm_offset_t end, cend;
777 int vecindex, lastvecindex;
778 vm_map_entry_t current;
779 vm_map_entry_t entry;
781 unsigned int timestamp;
784 * Make sure that the addresses presented are valid for user
787 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
788 end = addr + (vm_size_t)round_page(uap->len);
791 if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
795 * Address of byte vector
799 map = &p->p_vmspace->vm_map;
800 pmap = vmspace_pmap(p->p_vmspace);
802 lwkt_gettoken(&vm_token);
803 vm_map_lock_read(map);
805 timestamp = map->timestamp;
807 if (!vm_map_lookup_entry(map, addr, &entry))
811 * Do this on a map entry basis so that if the pages are not
812 * in the current processes address space, we can easily look
813 * up the pages elsewhere.
817 (current != &map->header) && (current->start < end);
818 current = current->next) {
821 * ignore submaps (for now) or null objects
823 if (current->maptype != VM_MAPTYPE_NORMAL &&
824 current->maptype != VM_MAPTYPE_VPAGETABLE) {
827 if (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 * If we have to check the VM object, only mess
850 * around with normal maps. Do not mess around
851 * with virtual page tables (XXX).
853 mincoreinfo = pmap_mincore(pmap, addr);
854 if (mincoreinfo == 0 &&
855 current->maptype == VM_MAPTYPE_NORMAL) {
861 * calculate the page index into the object
863 offset = current->offset + (addr - current->start);
864 pindex = OFF_TO_IDX(offset);
867 * if the page is resident, then gather
868 * information about it. spl protection is
869 * required to maintain the object
870 * association. And XXX what if the page is
871 * busy? What's the deal with that?
874 m = vm_page_lookup(current->object.vm_object,
877 mincoreinfo = MINCORE_INCORE;
880 mincoreinfo |= MINCORE_MODIFIED_OTHER;
881 if ((m->flags & PG_REFERENCED) ||
882 pmap_ts_referenced(m)) {
883 vm_page_flag_set(m, PG_REFERENCED);
884 mincoreinfo |= MINCORE_REFERENCED_OTHER;
891 * subyte may page fault. In case it needs to modify
892 * the map, we release the lock.
894 vm_map_unlock_read(map);
897 * calculate index into user supplied byte vector
899 vecindex = OFF_TO_IDX(addr - first_addr);
902 * If we have skipped map entries, we need to make sure that
903 * the byte vector is zeroed for those skipped entries.
905 while((lastvecindex + 1) < vecindex) {
906 error = subyte( vec + lastvecindex, 0);
915 * Pass the page information to the user
917 error = subyte( vec + vecindex, mincoreinfo);
924 * If the map has changed, due to the subyte, the previous
925 * output may be invalid.
927 vm_map_lock_read(map);
928 if (timestamp != map->timestamp)
931 lastvecindex = vecindex;
937 * subyte may page fault. In case it needs to modify
938 * the map, we release the lock.
940 vm_map_unlock_read(map);
943 * Zero the last entries in the byte vector.
945 vecindex = OFF_TO_IDX(end - first_addr);
946 while((lastvecindex + 1) < vecindex) {
947 error = subyte( vec + lastvecindex, 0);
956 * If the map has changed, due to the subyte, the previous
957 * output may be invalid.
959 vm_map_lock_read(map);
960 if (timestamp != map->timestamp)
962 vm_map_unlock_read(map);
966 lwkt_reltoken(&vm_token);
971 * mlock system call handler
973 * mlock_args(const void *addr, size_t len)
978 sys_mlock(struct mlock_args *uap)
982 vm_size_t size, pageoff;
983 struct thread *td = curthread;
984 struct proc *p = td->td_proc;
987 addr = (vm_offset_t) uap->addr;
990 pageoff = (addr & PAGE_MASK);
993 size = (vm_size_t) round_page(size);
994 if (size < uap->len) /* wrap */
996 tmpaddr = addr + size; /* workaround gcc4 opt */
997 if (tmpaddr < addr) /* wrap */
1000 if (atop(size) + vmstats.v_wire_count > vm_page_max_wired)
1004 * We do not need to synchronize against other threads updating ucred;
1005 * they update p->ucred, which is synchronized into td_ucred ourselves.
1007 #ifdef pmap_wired_count
1008 if (size + ptoa(pmap_wired_count(vm_map_pmap(&p->p_vmspace->vm_map))) >
1009 p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur) {
1013 error = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1018 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, FALSE);
1019 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1028 sys_mlockall(struct mlockall_args *uap)
1030 #ifdef _P1003_1B_VISIBLE
1031 struct thread *td = curthread;
1032 struct proc *p = td->td_proc;
1033 vm_map_t map = &p->p_vmspace->vm_map;
1034 vm_map_entry_t entry;
1036 int rc = KERN_SUCCESS;
1038 if (((how & MCL_CURRENT) == 0) && ((how & MCL_FUTURE) == 0))
1041 rc = priv_check_cred(td->td_ucred, PRIV_ROOT, 0);
1047 if (how & MCL_CURRENT) {
1048 for(entry = map->header.next;
1049 entry != &map->header;
1050 entry = entry->next);
1056 if (how & MCL_FUTURE)
1057 map->flags |= MAP_WIREFUTURE;
1062 #else /* !_P1003_1B_VISIBLE */
1064 #endif /* _P1003_1B_VISIBLE */
1070 * Unwire all user-wired map entries, cancel MCL_FUTURE.
1075 sys_munlockall(struct munlockall_args *uap)
1077 struct thread *td = curthread;
1078 struct proc *p = td->td_proc;
1079 vm_map_t map = &p->p_vmspace->vm_map;
1080 vm_map_entry_t entry;
1081 int rc = KERN_SUCCESS;
1085 /* Clear MAP_WIREFUTURE to cancel mlockall(MCL_FUTURE) */
1086 map->flags &= ~MAP_WIREFUTURE;
1089 for (entry = map->header.next;
1090 entry != &map->header;
1091 entry = entry->next) {
1092 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
1096 * If we encounter an in-transition entry, we release the
1097 * map lock and retry the scan; we do not decrement any
1098 * wired_count more than once because we do not touch
1099 * any entries with MAP_ENTRY_USER_WIRED not set.
1101 * There is a potential interleaving with concurrent
1102 * mlockall()s here -- if we abort a scan, an mlockall()
1103 * could start, wire a number of entries before our
1104 * current position in, and then stall itself on this
1105 * or any other in-transition entry. If that occurs, when
1106 * we resume, we will unwire those entries.
1108 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1109 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1110 ++mycpu->gd_cnt.v_intrans_coll;
1111 ++mycpu->gd_cnt.v_intrans_wait;
1112 vm_map_transition_wait(map);
1116 KASSERT(entry->wired_count > 0,
1117 ("wired_count was 0 with USER_WIRED set! %p", entry));
1119 /* Drop wired count, if it hits zero, unwire the entry */
1120 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1121 entry->wired_count--;
1122 if (entry->wired_count == 0)
1123 vm_fault_unwire(map, entry);
1133 * munlock system call handler
1135 * munlock_args(const void *addr, size_t len)
1140 sys_munlock(struct munlock_args *uap)
1142 struct thread *td = curthread;
1143 struct proc *p = td->td_proc;
1145 vm_offset_t tmpaddr;
1146 vm_size_t size, pageoff;
1149 addr = (vm_offset_t) uap->addr;
1152 pageoff = (addr & PAGE_MASK);
1155 size = (vm_size_t) round_page(size);
1157 tmpaddr = addr + size;
1158 if (tmpaddr < addr) /* wrap */
1161 #ifndef pmap_wired_count
1162 error = priv_check(td, PRIV_ROOT);
1167 error = vm_map_unwire(&p->p_vmspace->vm_map, addr, addr + size, TRUE);
1168 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1172 * Internal version of mmap.
1173 * Currently used by mmap, exec, and sys5 shared memory.
1174 * Handle is either a vnode pointer or NULL for MAP_ANON.
1176 * No requirements; kern_mmap path holds the vm_token
1179 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1180 vm_prot_t maxprot, int flags, void *handle, vm_ooffset_t foff)
1187 struct thread *td = curthread;
1189 int rv = KERN_SUCCESS;
1196 objsize = round_page(size);
1201 lwkt_gettoken(&vm_token);
1204 * XXX messy code, fixme
1206 * NOTE: Overflow checks require discrete statements or GCC4
1207 * will optimize it out.
1209 if ((p = curproc) != NULL && map == &p->p_vmspace->vm_map) {
1210 esize = map->size + size; /* workaround gcc4 opt */
1211 if (esize < map->size ||
1212 esize > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
1213 lwkt_reltoken(&vm_token);
1219 * We currently can only deal with page aligned file offsets.
1220 * The check is here rather than in the syscall because the
1221 * kernel calls this function internally for other mmaping
1222 * operations (such as in exec) and non-aligned offsets will
1223 * cause pmap inconsistencies...so we want to be sure to
1224 * disallow this in all cases.
1226 * NOTE: Overflow checks require discrete statements or GCC4
1227 * will optimize it out.
1229 if (foff & PAGE_MASK) {
1230 lwkt_reltoken(&vm_token);
1234 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
1236 *addr = round_page(*addr);
1238 if (*addr != trunc_page(*addr)) {
1239 lwkt_reltoken(&vm_token);
1242 eaddr = *addr + size;
1243 if (eaddr < *addr) {
1244 lwkt_reltoken(&vm_token);
1248 if ((flags & MAP_TRYFIXED) == 0)
1249 vm_map_remove(map, *addr, *addr + size);
1253 * Lookup/allocate object.
1255 if (flags & MAP_ANON) {
1257 * Unnamed anonymous regions always start at 0.
1261 * Default memory object
1263 object = default_pager_alloc(handle, objsize,
1265 if (object == NULL) {
1266 lwkt_reltoken(&vm_token);
1269 docow = MAP_PREFAULT_PARTIAL;
1272 * Implicit single instance of a default memory
1273 * object, so we don't need a VM object yet.
1281 vp = (struct vnode *)handle;
1282 if (vp->v_type == VCHR) {
1284 * Device mappings (device size unknown?).
1285 * Force them to be shared.
1287 handle = (void *)(intptr_t)vp->v_rdev;
1288 object = dev_pager_alloc(handle, objsize, prot, foff);
1289 if (object == NULL) {
1290 lwkt_reltoken(&vm_token);
1293 docow = MAP_PREFAULT_PARTIAL;
1294 flags &= ~(MAP_PRIVATE|MAP_COPY);
1295 flags |= MAP_SHARED;
1298 * Regular file mapping (typically). The attribute
1299 * check is for the link count test only. Mmapble
1300 * vnodes must already have a VM object assigned.
1305 error = VOP_GETATTR(vp, &vat);
1307 lwkt_reltoken(&vm_token);
1310 docow = MAP_PREFAULT_PARTIAL;
1311 object = vnode_pager_reference(vp);
1312 if (object == NULL && vp->v_type == VREG) {
1313 lwkt_reltoken(&vm_token);
1314 kprintf("Warning: cannot mmap vnode %p, no "
1320 * If it is a regular file without any references
1321 * we do not need to sync it.
1323 if (vp->v_type == VREG && vat.va_nlink == 0) {
1324 flags |= MAP_NOSYNC;
1330 * Deal with the adjusted flags
1332 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1333 docow |= MAP_COPY_ON_WRITE;
1334 if (flags & MAP_NOSYNC)
1335 docow |= MAP_DISABLE_SYNCER;
1336 if (flags & MAP_NOCORE)
1337 docow |= MAP_DISABLE_COREDUMP;
1339 #if defined(VM_PROT_READ_IS_EXEC)
1340 if (prot & VM_PROT_READ)
1341 prot |= VM_PROT_EXECUTE;
1343 if (maxprot & VM_PROT_READ)
1344 maxprot |= VM_PROT_EXECUTE;
1348 * This may place the area in its own page directory if (size) is
1349 * large enough, otherwise it typically returns its argument.
1352 *addr = pmap_addr_hint(object, *addr, size);
1356 * Stack mappings need special attention.
1358 * Mappings that use virtual page tables will default to storing
1359 * the page table at offset 0.
1361 if (flags & MAP_STACK) {
1362 rv = vm_map_stack(map, *addr, size, flags,
1363 prot, maxprot, docow);
1364 } else if (flags & MAP_VPAGETABLE) {
1365 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1366 fitit, VM_MAPTYPE_VPAGETABLE,
1367 prot, maxprot, docow);
1369 rv = vm_map_find(map, object, foff, addr, size, PAGE_SIZE,
1370 fitit, VM_MAPTYPE_NORMAL,
1371 prot, maxprot, docow);
1374 if (rv != KERN_SUCCESS) {
1376 * Lose the object reference. Will destroy the
1377 * object if it's an unnamed anonymous mapping
1378 * or named anonymous without other references.
1380 vm_object_deallocate(object);
1385 * Shared memory is also shared with children.
1387 if (flags & (MAP_SHARED|MAP_INHERIT)) {
1388 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1389 if (rv != KERN_SUCCESS) {
1390 vm_map_remove(map, *addr, *addr + size);
1395 /* If a process has marked all future mappings for wiring, do so */
1396 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1397 vm_map_unwire(map, *addr, *addr + size, FALSE);
1400 * Set the access time on the vnode
1403 vn_mark_atime(vp, td);
1405 lwkt_reltoken(&vm_token);
1410 case KERN_INVALID_ADDRESS:
1413 case KERN_PROTECTION_FAILURE: