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/capsicum.h>
53 #include <sys/kernel.h>
55 #include <sys/mutex.h>
56 #include <sys/sysproto.h>
57 #include <sys/filedesc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
70 #include <sys/mount.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_param.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_pager.h>
86 #include <vm/vm_pageout.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_page.h>
89 #include <vm/vnode_pager.h>
92 #include <sys/pmckern.h>
96 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
97 "Do not apply RLIMIT_MEMLOCK on mlockall");
100 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
103 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
104 int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
105 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
106 int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
107 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
108 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
110 #ifndef _SYS_SYSPROTO_H_
123 struct sbrk_args *uap;
125 /* Not yet implemented */
129 #ifndef _SYS_SYSPROTO_H_
142 struct sstk_args *uap;
144 /* Not yet implemented */
148 #if defined(COMPAT_43)
149 #ifndef _SYS_SYSPROTO_H_
150 struct getpagesize_args {
156 ogetpagesize(td, uap)
158 struct getpagesize_args *uap;
161 td->td_retval[0] = PAGE_SIZE;
164 #endif /* COMPAT_43 */
168 * Memory Map (mmap) system call. Note that the file offset
169 * and address are allowed to be NOT page aligned, though if
170 * the MAP_FIXED flag it set, both must have the same remainder
171 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
172 * page-aligned, the actual mapping starts at trunc_page(addr)
173 * and the return value is adjusted up by the page offset.
175 * Generally speaking, only character devices which are themselves
176 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
177 * there would be no cache coherency between a descriptor and a VM mapping
178 * both to the same character device.
180 #ifndef _SYS_SYSPROTO_H_
198 struct mmap_args *uap;
201 struct pmckern_map_in pkm;
206 vm_size_t size, pageoff;
207 vm_prot_t cap_maxprot, maxprot;
209 objtype_t handle_type;
210 int align, error, flags, prot;
212 struct vmspace *vms = td->td_proc->p_vmspace;
215 addr = (vm_offset_t) uap->addr;
224 * Ignore old flags that used to be defined but did not do anything.
226 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
229 * Enforce the constraints.
230 * Mapping of length 0 is only allowed for old binaries.
231 * Anonymous mapping shall specify -1 as filedescriptor and
232 * zero position for new code. Be nice to ancient a.out
233 * binaries and correct pos for anonymous mapping, since old
234 * ld.so sometimes issues anonymous map requests with non-zero
237 if (!SV_CURPROC_FLAG(SV_AOUT)) {
238 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
239 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
242 if ((flags & MAP_ANON) != 0)
246 if (flags & MAP_STACK) {
247 if ((uap->fd != -1) ||
248 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
253 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
254 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
259 MAP_ALIGNMENT_MASK)) != 0)
261 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
263 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
265 if (prot != PROT_NONE &&
266 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
270 * Align the file position to a page boundary,
271 * and save its page offset component.
273 pageoff = (pos & PAGE_MASK);
276 /* Adjust size for rounding (on both ends). */
277 size += pageoff; /* low end... */
278 size = (vm_size_t) round_page(size); /* hi end */
280 /* Ensure alignment is at least a page and fits in a pointer. */
281 align = flags & MAP_ALIGNMENT_MASK;
282 if (align != 0 && align != MAP_ALIGNED_SUPER &&
283 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
284 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
288 * Check for illegal addresses. Watch out for address wrap... Note
289 * that VM_*_ADDRESS are not constants due to casts (argh).
291 if (flags & MAP_FIXED) {
293 * The specified address must have the same remainder
294 * as the file offset taken modulo PAGE_SIZE, so it
295 * should be aligned after adjustment by pageoff.
298 if (addr & PAGE_MASK)
301 /* Address range must be all in user VM space. */
302 if (addr < vm_map_min(&vms->vm_map) ||
303 addr + size > vm_map_max(&vms->vm_map))
305 if (addr + size < addr)
308 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
310 } else if (flags & MAP_32BIT) {
312 * For MAP_32BIT, override the hint if it is too high and
313 * do not bother moving the mapping past the heap (since
314 * the heap is usually above 2GB).
316 if (addr + size > MAP_32BIT_MAX_ADDR)
321 * XXX for non-fixed mappings where no hint is provided or
322 * the hint would fall in the potential heap space,
323 * place it after the end of the largest possible heap.
325 * There should really be a pmap call to determine a reasonable
328 PROC_LOCK(td->td_proc);
330 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
331 addr < round_page((vm_offset_t)vms->vm_daddr +
332 lim_max(td->td_proc, RLIMIT_DATA))))
333 addr = round_page((vm_offset_t)vms->vm_daddr +
334 lim_max(td->td_proc, RLIMIT_DATA));
335 PROC_UNLOCK(td->td_proc);
337 if (flags & MAP_ANON) {
339 * Mapping blank space is trivial.
342 handle_type = OBJT_DEFAULT;
343 maxprot = VM_PROT_ALL;
344 cap_maxprot = VM_PROT_ALL;
347 * Mapping file, get fp for validation and don't let the
348 * descriptor disappear on us if we block. Check capability
349 * rights, but also return the maximum rights to be combined
350 * with maxprot later.
352 cap_rights_init(&rights, CAP_MMAP);
353 if (prot & PROT_READ)
354 cap_rights_set(&rights, CAP_MMAP_R);
355 if ((flags & MAP_SHARED) != 0) {
356 if (prot & PROT_WRITE)
357 cap_rights_set(&rights, CAP_MMAP_W);
359 if (prot & PROT_EXEC)
360 cap_rights_set(&rights, CAP_MMAP_X);
361 error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
364 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
365 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
369 if (fp->f_type == DTYPE_SHM) {
371 handle_type = OBJT_SWAP;
372 maxprot = VM_PROT_NONE;
374 /* FREAD should always be set. */
375 if (fp->f_flag & FREAD)
376 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
377 if (fp->f_flag & FWRITE)
378 maxprot |= VM_PROT_WRITE;
381 if (fp->f_type != DTYPE_VNODE) {
385 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
386 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
388 * POSIX shared-memory objects are defined to have
389 * kernel persistence, and are not defined to support
390 * read(2)/write(2) -- or even open(2). Thus, we can
391 * use MAP_ASYNC to trade on-disk coherence for speed.
392 * The shm_open(3) library routine turns on the FPOSIXSHM
393 * flag to request this behavior.
395 if (fp->f_flag & FPOSIXSHM)
400 * Ensure that file and memory protections are
401 * compatible. Note that we only worry about
402 * writability if mapping is shared; in this case,
403 * current and max prot are dictated by the open file.
404 * XXX use the vnode instead? Problem is: what
405 * credentials do we use for determination? What if
406 * proc does a setuid?
408 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
409 maxprot = VM_PROT_NONE;
411 maxprot = VM_PROT_EXECUTE;
412 if (fp->f_flag & FREAD) {
413 maxprot |= VM_PROT_READ;
414 } else if (prot & PROT_READ) {
419 * If we are sharing potential changes (either via
420 * MAP_SHARED or via the implicit sharing of character
421 * device mappings), and we are trying to get write
422 * permission although we opened it without asking
425 if ((flags & MAP_SHARED) != 0) {
426 if ((fp->f_flag & FWRITE) != 0) {
427 maxprot |= VM_PROT_WRITE;
428 } else if ((prot & PROT_WRITE) != 0) {
432 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
433 maxprot |= VM_PROT_WRITE;
434 cap_maxprot |= VM_PROT_WRITE;
437 handle_type = OBJT_VNODE;
441 maxprot &= cap_maxprot;
443 /* This relies on VM_PROT_* matching PROT_*. */
444 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
445 flags, handle_type, handle, pos);
448 /* inform hwpmc(4) if an executable is being mapped */
449 if (error == 0 && handle_type == OBJT_VNODE &&
450 (prot & PROT_EXEC)) {
451 pkm.pm_file = handle;
452 pkm.pm_address = (uintptr_t) addr;
453 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
457 td->td_retval[0] = (register_t) (addr + pageoff);
465 #if defined(COMPAT_FREEBSD6)
467 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
469 struct mmap_args oargs;
471 oargs.addr = uap->addr;
472 oargs.len = uap->len;
473 oargs.prot = uap->prot;
474 oargs.flags = uap->flags;
476 oargs.pos = uap->pos;
477 return (sys_mmap(td, &oargs));
482 #ifndef _SYS_SYSPROTO_H_
495 struct ommap_args *uap;
497 struct mmap_args nargs;
498 static const char cvtbsdprot[8] = {
502 PROT_EXEC | PROT_WRITE,
504 PROT_EXEC | PROT_READ,
505 PROT_WRITE | PROT_READ,
506 PROT_EXEC | PROT_WRITE | PROT_READ,
509 #define OMAP_ANON 0x0002
510 #define OMAP_COPY 0x0020
511 #define OMAP_SHARED 0x0010
512 #define OMAP_FIXED 0x0100
514 nargs.addr = uap->addr;
515 nargs.len = uap->len;
516 nargs.prot = cvtbsdprot[uap->prot & 0x7];
517 #ifdef COMPAT_FREEBSD32
518 #if defined(__amd64__)
519 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
521 nargs.prot |= PROT_EXEC;
525 if (uap->flags & OMAP_ANON)
526 nargs.flags |= MAP_ANON;
527 if (uap->flags & OMAP_COPY)
528 nargs.flags |= MAP_COPY;
529 if (uap->flags & OMAP_SHARED)
530 nargs.flags |= MAP_SHARED;
532 nargs.flags |= MAP_PRIVATE;
533 if (uap->flags & OMAP_FIXED)
534 nargs.flags |= MAP_FIXED;
536 nargs.pos = uap->pos;
537 return (sys_mmap(td, &nargs));
539 #endif /* COMPAT_43 */
542 #ifndef _SYS_SYSPROTO_H_
555 struct msync_args *uap;
558 vm_size_t size, pageoff;
563 addr = (vm_offset_t) uap->addr;
567 pageoff = (addr & PAGE_MASK);
570 size = (vm_size_t) round_page(size);
571 if (addr + size < addr)
574 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
577 map = &td->td_proc->p_vmspace->vm_map;
580 * Clean the pages and interpret the return value.
582 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
583 (flags & MS_INVALIDATE) != 0);
587 case KERN_INVALID_ADDRESS:
589 case KERN_INVALID_ARGUMENT:
598 #ifndef _SYS_SYSPROTO_H_
610 struct munmap_args *uap;
613 struct pmckern_map_out pkm;
614 vm_map_entry_t entry;
617 vm_size_t size, pageoff;
620 addr = (vm_offset_t) uap->addr;
625 pageoff = (addr & PAGE_MASK);
628 size = (vm_size_t) round_page(size);
629 if (addr + size < addr)
633 * Check for illegal addresses. Watch out for address wrap...
635 map = &td->td_proc->p_vmspace->vm_map;
636 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
641 * Inform hwpmc if the address range being unmapped contains
642 * an executable region.
644 pkm.pm_address = (uintptr_t) NULL;
645 if (vm_map_lookup_entry(map, addr, &entry)) {
647 entry != &map->header && entry->start < addr + size;
648 entry = entry->next) {
649 if (vm_map_check_protection(map, entry->start,
650 entry->end, VM_PROT_EXECUTE) == TRUE) {
651 pkm.pm_address = (uintptr_t) addr;
652 pkm.pm_size = (size_t) size;
658 vm_map_delete(map, addr, addr + size);
661 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
662 vm_map_lock_downgrade(map);
663 if (pkm.pm_address != (uintptr_t) NULL)
664 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
665 vm_map_unlock_read(map);
669 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
673 #ifndef _SYS_SYSPROTO_H_
674 struct mprotect_args {
684 sys_mprotect(td, uap)
686 struct mprotect_args *uap;
689 vm_size_t size, pageoff;
692 addr = (vm_offset_t) uap->addr;
694 prot = uap->prot & VM_PROT_ALL;
696 pageoff = (addr & PAGE_MASK);
699 size = (vm_size_t) round_page(size);
700 if (addr + size < addr)
703 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
704 addr + size, prot, FALSE)) {
707 case KERN_PROTECTION_FAILURE:
709 case KERN_RESOURCE_SHORTAGE:
715 #ifndef _SYS_SYSPROTO_H_
716 struct minherit_args {
726 sys_minherit(td, uap)
728 struct minherit_args *uap;
731 vm_size_t size, pageoff;
732 vm_inherit_t inherit;
734 addr = (vm_offset_t)uap->addr;
736 inherit = uap->inherit;
738 pageoff = (addr & PAGE_MASK);
741 size = (vm_size_t) round_page(size);
742 if (addr + size < addr)
745 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
746 addr + size, inherit)) {
749 case KERN_PROTECTION_FAILURE:
755 #ifndef _SYS_SYSPROTO_H_
756 struct madvise_args {
769 struct madvise_args *uap;
771 vm_offset_t start, end;
776 * Check for our special case, advising the swap pager we are
779 if (uap->behav == MADV_PROTECT) {
781 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
782 PROC_SPROTECT, &flags));
786 * Check for illegal behavior
788 if (uap->behav < 0 || uap->behav > MADV_CORE)
791 * Check for illegal addresses. Watch out for address wrap... Note
792 * that VM_*_ADDRESS are not constants due to casts (argh).
794 map = &td->td_proc->p_vmspace->vm_map;
795 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
796 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
798 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
802 * Since this routine is only advisory, we default to conservative
805 start = trunc_page((vm_offset_t) uap->addr);
806 end = round_page((vm_offset_t) uap->addr + uap->len);
808 if (vm_map_madvise(map, start, end, uap->behav))
813 #ifndef _SYS_SYSPROTO_H_
814 struct mincore_args {
827 struct mincore_args *uap;
829 vm_offset_t addr, first_addr;
830 vm_offset_t end, cend;
835 int vecindex, lastvecindex;
836 vm_map_entry_t current;
837 vm_map_entry_t entry;
839 vm_paddr_t locked_pa;
843 unsigned int timestamp;
847 * Make sure that the addresses presented are valid for user
850 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
851 end = addr + (vm_size_t)round_page(uap->len);
852 map = &td->td_proc->p_vmspace->vm_map;
853 if (end > vm_map_max(map) || end < addr)
857 * Address of byte vector
861 pmap = vmspace_pmap(td->td_proc->p_vmspace);
863 vm_map_lock_read(map);
865 timestamp = map->timestamp;
867 if (!vm_map_lookup_entry(map, addr, &entry)) {
868 vm_map_unlock_read(map);
873 * Do this on a map entry basis so that if the pages are not
874 * in the current processes address space, we can easily look
875 * up the pages elsewhere.
878 for (current = entry;
879 (current != &map->header) && (current->start < end);
880 current = current->next) {
883 * check for contiguity
885 if (current->end < end &&
886 (entry->next == &map->header ||
887 current->next->start > current->end)) {
888 vm_map_unlock_read(map);
893 * ignore submaps (for now) or null objects
895 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
896 current->object.vm_object == NULL)
900 * limit this scan to the current map entry and the
901 * limits for the mincore call
903 if (addr < current->start)
904 addr = current->start;
910 * scan this entry one page at a time
912 while (addr < cend) {
914 * Check pmap first, it is likely faster, also
915 * it can provide info as to whether we are the
916 * one referencing or modifying the page.
922 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
923 if (locked_pa != 0) {
925 * The page is mapped by this process but not
926 * both accessed and modified. It is also
927 * managed. Acquire the object lock so that
928 * other mappings might be examined.
930 m = PHYS_TO_VM_PAGE(locked_pa);
931 if (m->object != object) {
933 VM_OBJECT_WUNLOCK(object);
935 locked = VM_OBJECT_TRYWLOCK(object);
938 VM_OBJECT_WLOCK(object);
944 KASSERT(m->valid == VM_PAGE_BITS_ALL,
945 ("mincore: page %p is mapped but invalid",
947 } else if (mincoreinfo == 0) {
949 * The page is not mapped by this process. If
950 * the object implements managed pages, then
951 * determine if the page is resident so that
952 * the mappings might be examined.
954 if (current->object.vm_object != object) {
956 VM_OBJECT_WUNLOCK(object);
957 object = current->object.vm_object;
958 VM_OBJECT_WLOCK(object);
960 if (object->type == OBJT_DEFAULT ||
961 object->type == OBJT_SWAP ||
962 object->type == OBJT_VNODE) {
963 pindex = OFF_TO_IDX(current->offset +
964 (addr - current->start));
965 m = vm_page_lookup(object, pindex);
967 vm_page_is_cached(object, pindex))
968 mincoreinfo = MINCORE_INCORE;
969 if (m != NULL && m->valid == 0)
972 mincoreinfo = MINCORE_INCORE;
976 /* Examine other mappings to the page. */
977 if (m->dirty == 0 && pmap_is_modified(m))
980 mincoreinfo |= MINCORE_MODIFIED_OTHER;
982 * The first test for PGA_REFERENCED is an
983 * optimization. The second test is
984 * required because a concurrent pmap
985 * operation could clear the last reference
986 * and set PGA_REFERENCED before the call to
987 * pmap_is_referenced().
989 if ((m->aflags & PGA_REFERENCED) != 0 ||
990 pmap_is_referenced(m) ||
991 (m->aflags & PGA_REFERENCED) != 0)
992 mincoreinfo |= MINCORE_REFERENCED_OTHER;
995 VM_OBJECT_WUNLOCK(object);
998 * subyte may page fault. In case it needs to modify
999 * the map, we release the lock.
1001 vm_map_unlock_read(map);
1004 * calculate index into user supplied byte vector
1006 vecindex = OFF_TO_IDX(addr - first_addr);
1009 * If we have skipped map entries, we need to make sure that
1010 * the byte vector is zeroed for those skipped entries.
1012 while ((lastvecindex + 1) < vecindex) {
1014 error = subyte(vec + lastvecindex, 0);
1022 * Pass the page information to the user
1024 error = subyte(vec + vecindex, mincoreinfo);
1031 * If the map has changed, due to the subyte, the previous
1032 * output may be invalid.
1034 vm_map_lock_read(map);
1035 if (timestamp != map->timestamp)
1038 lastvecindex = vecindex;
1044 * subyte may page fault. In case it needs to modify
1045 * the map, we release the lock.
1047 vm_map_unlock_read(map);
1050 * Zero the last entries in the byte vector.
1052 vecindex = OFF_TO_IDX(end - first_addr);
1053 while ((lastvecindex + 1) < vecindex) {
1055 error = subyte(vec + lastvecindex, 0);
1063 * If the map has changed, due to the subyte, the previous
1064 * output may be invalid.
1066 vm_map_lock_read(map);
1067 if (timestamp != map->timestamp)
1069 vm_map_unlock_read(map);
1074 #ifndef _SYS_SYSPROTO_H_
1086 struct mlock_args *uap;
1089 return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1093 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1095 vm_offset_t addr, end, last, start;
1096 vm_size_t npages, size;
1098 unsigned long nsize;
1101 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1104 addr = (vm_offset_t)addr0;
1107 start = trunc_page(addr);
1108 end = round_page(last);
1109 if (last < addr || end < addr)
1111 npages = atop(end - start);
1112 if (npages > vm_page_max_wired)
1114 map = &proc->p_vmspace->vm_map;
1116 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1117 if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1122 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1126 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1131 error = vm_map_wire(map, start, end,
1132 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1134 if (error != KERN_SUCCESS) {
1136 racct_set(proc, RACCT_MEMLOCK,
1137 ptoa(pmap_wired_count(map->pmap)));
1141 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1144 #ifndef _SYS_SYSPROTO_H_
1145 struct mlockall_args {
1154 sys_mlockall(td, uap)
1156 struct mlockall_args *uap;
1161 map = &td->td_proc->p_vmspace->vm_map;
1162 error = priv_check(td, PRIV_VM_MLOCK);
1166 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1170 * If wiring all pages in the process would cause it to exceed
1171 * a hard resource limit, return ENOMEM.
1173 if (!old_mlock && uap->how & MCL_CURRENT) {
1174 PROC_LOCK(td->td_proc);
1175 if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1176 PROC_UNLOCK(td->td_proc);
1179 PROC_UNLOCK(td->td_proc);
1182 PROC_LOCK(td->td_proc);
1183 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1184 PROC_UNLOCK(td->td_proc);
1189 if (uap->how & MCL_FUTURE) {
1191 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1196 if (uap->how & MCL_CURRENT) {
1198 * P1003.1-2001 mandates that all currently mapped pages
1199 * will be memory resident and locked (wired) upon return
1200 * from mlockall(). vm_map_wire() will wire pages, by
1201 * calling vm_fault_wire() for each page in the region.
1203 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1204 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1205 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1208 if (error != KERN_SUCCESS) {
1209 PROC_LOCK(td->td_proc);
1210 racct_set(td->td_proc, RACCT_MEMLOCK,
1211 ptoa(pmap_wired_count(map->pmap)));
1212 PROC_UNLOCK(td->td_proc);
1219 #ifndef _SYS_SYSPROTO_H_
1220 struct munlockall_args {
1229 sys_munlockall(td, uap)
1231 struct munlockall_args *uap;
1236 map = &td->td_proc->p_vmspace->vm_map;
1237 error = priv_check(td, PRIV_VM_MUNLOCK);
1241 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1243 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1246 /* Forcibly unwire all pages. */
1247 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1248 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1250 if (error == KERN_SUCCESS) {
1251 PROC_LOCK(td->td_proc);
1252 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1253 PROC_UNLOCK(td->td_proc);
1260 #ifndef _SYS_SYSPROTO_H_
1261 struct munlock_args {
1270 sys_munlock(td, uap)
1272 struct munlock_args *uap;
1274 vm_offset_t addr, end, last, start;
1281 error = priv_check(td, PRIV_VM_MUNLOCK);
1284 addr = (vm_offset_t)uap->addr;
1287 start = trunc_page(addr);
1288 end = round_page(last);
1289 if (last < addr || end < addr)
1291 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1292 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1294 if (error == KERN_SUCCESS) {
1295 PROC_LOCK(td->td_proc);
1296 map = &td->td_proc->p_vmspace->vm_map;
1297 racct_set(td->td_proc, RACCT_MEMLOCK,
1298 ptoa(pmap_wired_count(map->pmap)));
1299 PROC_UNLOCK(td->td_proc);
1302 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1308 * Helper function for vm_mmap. Perform sanity check specific for mmap
1309 * operations on vnodes.
1311 * For VCHR vnodes, the vnode lock is held over the call to
1312 * vm_mmap_cdev() to keep vp->v_rdev valid.
1315 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1316 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1317 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1318 boolean_t *writecounted)
1324 int error, flags, locktype;
1326 cred = td->td_ucred;
1327 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1328 locktype = LK_EXCLUSIVE;
1330 locktype = LK_SHARED;
1331 if ((error = vget(vp, locktype, td)) != 0)
1336 if (vp->v_type == VREG) {
1338 * Get the proper underlying object
1344 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1346 vp = (struct vnode *)obj->handle;
1348 * Bypass filesystems obey the mpsafety of the
1349 * underlying fs. Tmpfs never bypasses.
1351 error = vget(vp, locktype, td);
1355 if (locktype == LK_EXCLUSIVE) {
1356 *writecounted = TRUE;
1357 vnode_pager_update_writecount(obj, 0, objsize);
1359 } else if (vp->v_type == VCHR) {
1360 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1361 vp->v_rdev, foffp, objp);
1369 if ((error = VOP_GETATTR(vp, &va, cred)))
1372 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1376 if ((flags & MAP_SHARED) != 0) {
1377 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1378 if (prot & PROT_WRITE) {
1382 *maxprotp &= ~VM_PROT_WRITE;
1386 * If it is a regular file without any references
1387 * we do not need to sync it.
1388 * Adjust object size to be the size of actual file.
1390 objsize = round_page(va.va_size);
1391 if (va.va_nlink == 0)
1392 flags |= MAP_NOSYNC;
1393 if (obj->type == OBJT_VNODE) {
1394 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1401 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1402 ("wrong object type"));
1403 VM_OBJECT_WLOCK(obj);
1404 vm_object_reference_locked(obj);
1405 #if VM_NRESERVLEVEL > 0
1406 vm_object_color(obj, 0);
1408 VM_OBJECT_WUNLOCK(obj);
1414 vfs_mark_atime(vp, cred);
1417 if (error != 0 && *writecounted) {
1418 *writecounted = FALSE;
1419 vnode_pager_update_writecount(obj, objsize, 0);
1430 * Helper function for vm_mmap. Perform sanity check specific for mmap
1431 * operations on cdevs.
1434 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1435 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1436 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1440 int error, flags, ref;
1444 dsw = dev_refthread(cdev, &ref);
1447 if (dsw->d_flags & D_MMAP_ANON) {
1448 dev_relthread(cdev, ref);
1449 *maxprotp = VM_PROT_ALL;
1450 *flagsp |= MAP_ANON;
1454 * cdevs do not provide private mappings of any kind.
1456 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1457 (prot & PROT_WRITE) != 0) {
1458 dev_relthread(cdev, ref);
1461 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1462 dev_relthread(cdev, ref);
1466 * Force device mappings to be shared.
1468 flags |= MAP_SHARED;
1470 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1472 dev_relthread(cdev, ref);
1477 * First, try d_mmap_single(). If that is not implemented
1478 * (returns ENODEV), fall back to using the device pager.
1479 * Note that d_mmap_single() must return a reference to the
1480 * object (it needs to bump the reference count of the object
1481 * it returns somehow).
1483 * XXX assumes VM_PROT_* == PROT_*
1485 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1486 dev_relthread(cdev, ref);
1487 if (error != ENODEV)
1489 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1503 * Helper function for vm_mmap. Perform sanity check specific for mmap
1504 * operations on shm file descriptors.
1507 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1508 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1509 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1513 if ((*flagsp & MAP_SHARED) != 0 &&
1514 (*maxprotp & VM_PROT_WRITE) == 0 &&
1515 (prot & PROT_WRITE) != 0)
1518 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1522 error = shm_mmap(shmfd, objsize, foff, objp);
1533 * Internal version of mmap. Currently used by mmap, exec, and sys5
1534 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1537 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1538 vm_prot_t maxprot, int flags,
1539 objtype_t handle_type, void *handle,
1543 vm_object_t object = NULL;
1544 struct thread *td = curthread;
1545 int docow, error, findspace, rv;
1546 boolean_t writecounted;
1551 size = round_page(size);
1553 if (map == &td->td_proc->p_vmspace->vm_map) {
1554 PROC_LOCK(td->td_proc);
1555 if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1556 PROC_UNLOCK(td->td_proc);
1559 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1560 PROC_UNLOCK(td->td_proc);
1563 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1564 if (ptoa(pmap_wired_count(map->pmap)) + size >
1565 lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1566 racct_set_force(td->td_proc, RACCT_VMEM,
1568 PROC_UNLOCK(td->td_proc);
1571 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1572 ptoa(pmap_wired_count(map->pmap)) + size);
1574 racct_set_force(td->td_proc, RACCT_VMEM,
1576 PROC_UNLOCK(td->td_proc);
1580 PROC_UNLOCK(td->td_proc);
1584 * We currently can only deal with page aligned file offsets.
1585 * The check is here rather than in the syscall because the
1586 * kernel calls this function internally for other mmaping
1587 * operations (such as in exec) and non-aligned offsets will
1588 * cause pmap inconsistencies...so we want to be sure to
1589 * disallow this in all cases.
1591 if (foff & PAGE_MASK)
1594 if ((flags & MAP_FIXED) == 0) {
1596 *addr = round_page(*addr);
1598 if (*addr != trunc_page(*addr))
1602 writecounted = FALSE;
1605 * Lookup/allocate object.
1607 switch (handle_type) {
1609 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1610 handle, &foff, &object);
1613 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1614 handle, &foff, &object, &writecounted);
1617 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1618 handle, foff, &object);
1621 if (handle == NULL) {
1632 if (flags & MAP_ANON) {
1636 * Unnamed anonymous regions always start at 0.
1640 } else if (flags & MAP_PREFAULT_READ)
1641 docow = MAP_PREFAULT;
1643 docow = MAP_PREFAULT_PARTIAL;
1645 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1646 docow |= MAP_COPY_ON_WRITE;
1647 if (flags & MAP_NOSYNC)
1648 docow |= MAP_DISABLE_SYNCER;
1649 if (flags & MAP_NOCORE)
1650 docow |= MAP_DISABLE_COREDUMP;
1651 /* Shared memory is also shared with children. */
1652 if (flags & MAP_SHARED)
1653 docow |= MAP_INHERIT_SHARE;
1655 docow |= MAP_VN_WRITECOUNT;
1656 if (flags & MAP_STACK) {
1659 docow |= MAP_STACK_GROWS_DOWN;
1661 if ((flags & MAP_EXCL) != 0)
1662 docow |= MAP_CHECK_EXCL;
1665 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1666 findspace = VMFS_SUPER_SPACE;
1667 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1668 findspace = VMFS_ALIGNED_SPACE(flags >>
1669 MAP_ALIGNMENT_SHIFT);
1671 findspace = VMFS_OPTIMAL_SPACE;
1672 rv = vm_map_find(map, object, foff, addr, size,
1674 flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1676 0, findspace, prot, maxprot, docow);
1678 rv = vm_map_fixed(map, object, foff, *addr, size,
1679 prot, maxprot, docow);
1682 if (rv == KERN_SUCCESS) {
1684 * If the process has requested that all future mappings
1685 * be wired, then heed this.
1687 if (map->flags & MAP_WIREFUTURE) {
1688 vm_map_wire(map, *addr, *addr + size,
1689 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1690 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1694 * If this mapping was accounted for in the vnode's
1695 * writecount, then undo that now.
1698 vnode_pager_release_writecount(object, 0, size);
1700 * Lose the object reference. Will destroy the
1701 * object if it's an unnamed anonymous mapping
1702 * or named anonymous without other references.
1704 vm_object_deallocate(object);
1706 return (vm_mmap_to_errno(rv));
1710 * Translate a Mach VM return code to zero on success or the appropriate errno
1714 vm_mmap_to_errno(int rv)
1720 case KERN_INVALID_ADDRESS:
1723 case KERN_PROTECTION_FAILURE: