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"
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capsicum.h>
52 #include <sys/kernel.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
59 #include <sys/procctl.h>
60 #include <sys/racct.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/rwlock.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
66 #include <sys/fcntl.h>
69 #include <sys/mount.h>
72 #include <sys/syscallsubr.h>
73 #include <sys/sysent.h>
74 #include <sys/vmmeter.h>
76 #include <security/mac/mac_framework.h>
79 #include <vm/vm_param.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/vm_pageout.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_page.h>
88 #include <vm/vnode_pager.h>
91 #include <sys/pmckern.h>
95 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
96 "Do not apply RLIMIT_MEMLOCK on mlockall");
99 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
102 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
103 int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
104 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
105 int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
106 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
107 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
109 #ifndef _SYS_SYSPROTO_H_
122 struct sbrk_args *uap;
124 /* Not yet implemented */
128 #ifndef _SYS_SYSPROTO_H_
141 struct sstk_args *uap;
143 /* Not yet implemented */
147 #if defined(COMPAT_43)
148 #ifndef _SYS_SYSPROTO_H_
149 struct getpagesize_args {
155 ogetpagesize(td, uap)
157 struct getpagesize_args *uap;
160 td->td_retval[0] = PAGE_SIZE;
163 #endif /* COMPAT_43 */
167 * Memory Map (mmap) system call. Note that the file offset
168 * and address are allowed to be NOT page aligned, though if
169 * the MAP_FIXED flag it set, both must have the same remainder
170 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
171 * page-aligned, the actual mapping starts at trunc_page(addr)
172 * and the return value is adjusted up by the page offset.
174 * Generally speaking, only character devices which are themselves
175 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
176 * there would be no cache coherency between a descriptor and a VM mapping
177 * both to the same character device.
179 #ifndef _SYS_SYSPROTO_H_
197 struct mmap_args *uap;
200 struct pmckern_map_in pkm;
205 vm_size_t size, pageoff;
206 vm_prot_t cap_maxprot, maxprot;
208 objtype_t handle_type;
209 int align, error, flags, prot;
211 struct vmspace *vms = td->td_proc->p_vmspace;
214 addr = (vm_offset_t) uap->addr;
223 * Ignore old flags that used to be defined but did not do anything.
225 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
228 * Enforce the constraints.
229 * Mapping of length 0 is only allowed for old binaries.
230 * Anonymous mapping shall specify -1 as filedescriptor and
231 * zero position for new code. Be nice to ancient a.out
232 * binaries and correct pos for anonymous mapping, since old
233 * ld.so sometimes issues anonymous map requests with non-zero
236 if (!SV_CURPROC_FLAG(SV_AOUT)) {
237 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
238 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
241 if ((flags & MAP_ANON) != 0)
245 if (flags & MAP_STACK) {
246 if ((uap->fd != -1) ||
247 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
252 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
253 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
258 MAP_ALIGNMENT_MASK)) != 0)
260 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
262 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
264 if (prot != PROT_NONE &&
265 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
269 * Align the file position to a page boundary,
270 * and save its page offset component.
272 pageoff = (pos & PAGE_MASK);
275 /* Adjust size for rounding (on both ends). */
276 size += pageoff; /* low end... */
277 size = (vm_size_t) round_page(size); /* hi end */
279 /* Ensure alignment is at least a page and fits in a pointer. */
280 align = flags & MAP_ALIGNMENT_MASK;
281 if (align != 0 && align != MAP_ALIGNED_SUPER &&
282 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
283 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
287 * Check for illegal addresses. Watch out for address wrap... Note
288 * that VM_*_ADDRESS are not constants due to casts (argh).
290 if (flags & MAP_FIXED) {
292 * The specified address must have the same remainder
293 * as the file offset taken modulo PAGE_SIZE, so it
294 * should be aligned after adjustment by pageoff.
297 if (addr & PAGE_MASK)
300 /* Address range must be all in user VM space. */
301 if (addr < vm_map_min(&vms->vm_map) ||
302 addr + size > vm_map_max(&vms->vm_map))
304 if (addr + size < addr)
307 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
309 } else if (flags & MAP_32BIT) {
311 * For MAP_32BIT, override the hint if it is too high and
312 * do not bother moving the mapping past the heap (since
313 * the heap is usually above 2GB).
315 if (addr + size > MAP_32BIT_MAX_ADDR)
320 * XXX for non-fixed mappings where no hint is provided or
321 * the hint would fall in the potential heap space,
322 * place it after the end of the largest possible heap.
324 * There should really be a pmap call to determine a reasonable
327 PROC_LOCK(td->td_proc);
329 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
330 addr < round_page((vm_offset_t)vms->vm_daddr +
331 lim_max(td->td_proc, RLIMIT_DATA))))
332 addr = round_page((vm_offset_t)vms->vm_daddr +
333 lim_max(td->td_proc, RLIMIT_DATA));
334 PROC_UNLOCK(td->td_proc);
336 if (flags & MAP_ANON) {
338 * Mapping blank space is trivial.
341 handle_type = OBJT_DEFAULT;
342 maxprot = VM_PROT_ALL;
343 cap_maxprot = VM_PROT_ALL;
346 * Mapping file, get fp for validation and don't let the
347 * descriptor disappear on us if we block. Check capability
348 * rights, but also return the maximum rights to be combined
349 * with maxprot later.
351 cap_rights_init(&rights, CAP_MMAP);
352 if (prot & PROT_READ)
353 cap_rights_set(&rights, CAP_MMAP_R);
354 if ((flags & MAP_SHARED) != 0) {
355 if (prot & PROT_WRITE)
356 cap_rights_set(&rights, CAP_MMAP_W);
358 if (prot & PROT_EXEC)
359 cap_rights_set(&rights, CAP_MMAP_X);
360 error = fget_mmap(td, uap->fd, &rights, &cap_maxprot, &fp);
363 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
364 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
368 if (fp->f_type == DTYPE_SHM) {
370 handle_type = OBJT_SWAP;
371 maxprot = VM_PROT_NONE;
373 /* FREAD should always be set. */
374 if (fp->f_flag & FREAD)
375 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
376 if (fp->f_flag & FWRITE)
377 maxprot |= VM_PROT_WRITE;
380 if (fp->f_type != DTYPE_VNODE) {
384 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
385 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
387 * POSIX shared-memory objects are defined to have
388 * kernel persistence, and are not defined to support
389 * read(2)/write(2) -- or even open(2). Thus, we can
390 * use MAP_ASYNC to trade on-disk coherence for speed.
391 * The shm_open(3) library routine turns on the FPOSIXSHM
392 * flag to request this behavior.
394 if (fp->f_flag & FPOSIXSHM)
399 * Ensure that file and memory protections are
400 * compatible. Note that we only worry about
401 * writability if mapping is shared; in this case,
402 * current and max prot are dictated by the open file.
403 * XXX use the vnode instead? Problem is: what
404 * credentials do we use for determination? What if
405 * proc does a setuid?
407 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
408 maxprot = VM_PROT_NONE;
410 maxprot = VM_PROT_EXECUTE;
411 if (fp->f_flag & FREAD) {
412 maxprot |= VM_PROT_READ;
413 } else if (prot & PROT_READ) {
418 * If we are sharing potential changes (either via
419 * MAP_SHARED or via the implicit sharing of character
420 * device mappings), and we are trying to get write
421 * permission although we opened it without asking
424 if ((flags & MAP_SHARED) != 0) {
425 if ((fp->f_flag & FWRITE) != 0) {
426 maxprot |= VM_PROT_WRITE;
427 } else if ((prot & PROT_WRITE) != 0) {
431 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
432 maxprot |= VM_PROT_WRITE;
433 cap_maxprot |= VM_PROT_WRITE;
436 handle_type = OBJT_VNODE;
440 maxprot &= cap_maxprot;
442 /* This relies on VM_PROT_* matching PROT_*. */
443 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
444 flags, handle_type, handle, pos);
447 /* inform hwpmc(4) if an executable is being mapped */
448 if (error == 0 && handle_type == OBJT_VNODE &&
449 (prot & PROT_EXEC)) {
450 pkm.pm_file = handle;
451 pkm.pm_address = (uintptr_t) addr;
452 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
456 td->td_retval[0] = (register_t) (addr + pageoff);
465 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
467 struct mmap_args oargs;
469 oargs.addr = uap->addr;
470 oargs.len = uap->len;
471 oargs.prot = uap->prot;
472 oargs.flags = uap->flags;
474 oargs.pos = uap->pos;
475 return (sys_mmap(td, &oargs));
479 #ifndef _SYS_SYSPROTO_H_
492 struct ommap_args *uap;
494 struct mmap_args nargs;
495 static const char cvtbsdprot[8] = {
499 PROT_EXEC | PROT_WRITE,
501 PROT_EXEC | PROT_READ,
502 PROT_WRITE | PROT_READ,
503 PROT_EXEC | PROT_WRITE | PROT_READ,
506 #define OMAP_ANON 0x0002
507 #define OMAP_COPY 0x0020
508 #define OMAP_SHARED 0x0010
509 #define OMAP_FIXED 0x0100
511 nargs.addr = uap->addr;
512 nargs.len = uap->len;
513 nargs.prot = cvtbsdprot[uap->prot & 0x7];
514 #ifdef COMPAT_FREEBSD32
515 #if defined(__amd64__)
516 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
518 nargs.prot |= PROT_EXEC;
522 if (uap->flags & OMAP_ANON)
523 nargs.flags |= MAP_ANON;
524 if (uap->flags & OMAP_COPY)
525 nargs.flags |= MAP_COPY;
526 if (uap->flags & OMAP_SHARED)
527 nargs.flags |= MAP_SHARED;
529 nargs.flags |= MAP_PRIVATE;
530 if (uap->flags & OMAP_FIXED)
531 nargs.flags |= MAP_FIXED;
533 nargs.pos = uap->pos;
534 return (sys_mmap(td, &nargs));
536 #endif /* COMPAT_43 */
539 #ifndef _SYS_SYSPROTO_H_
552 struct msync_args *uap;
555 vm_size_t size, pageoff;
560 addr = (vm_offset_t) uap->addr;
564 pageoff = (addr & PAGE_MASK);
567 size = (vm_size_t) round_page(size);
568 if (addr + size < addr)
571 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
574 map = &td->td_proc->p_vmspace->vm_map;
577 * Clean the pages and interpret the return value.
579 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
580 (flags & MS_INVALIDATE) != 0);
584 case KERN_INVALID_ADDRESS:
586 case KERN_INVALID_ARGUMENT:
595 #ifndef _SYS_SYSPROTO_H_
607 struct munmap_args *uap;
610 struct pmckern_map_out pkm;
611 vm_map_entry_t entry;
614 vm_size_t size, pageoff;
617 addr = (vm_offset_t) uap->addr;
622 pageoff = (addr & PAGE_MASK);
625 size = (vm_size_t) round_page(size);
626 if (addr + size < addr)
630 * Check for illegal addresses. Watch out for address wrap...
632 map = &td->td_proc->p_vmspace->vm_map;
633 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
638 * Inform hwpmc if the address range being unmapped contains
639 * an executable region.
641 pkm.pm_address = (uintptr_t) NULL;
642 if (vm_map_lookup_entry(map, addr, &entry)) {
644 entry != &map->header && entry->start < addr + size;
645 entry = entry->next) {
646 if (vm_map_check_protection(map, entry->start,
647 entry->end, VM_PROT_EXECUTE) == TRUE) {
648 pkm.pm_address = (uintptr_t) addr;
649 pkm.pm_size = (size_t) size;
655 vm_map_delete(map, addr, addr + size);
658 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
659 vm_map_lock_downgrade(map);
660 if (pkm.pm_address != (uintptr_t) NULL)
661 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
662 vm_map_unlock_read(map);
666 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
670 #ifndef _SYS_SYSPROTO_H_
671 struct mprotect_args {
681 sys_mprotect(td, uap)
683 struct mprotect_args *uap;
686 vm_size_t size, pageoff;
689 addr = (vm_offset_t) uap->addr;
691 prot = uap->prot & VM_PROT_ALL;
693 pageoff = (addr & PAGE_MASK);
696 size = (vm_size_t) round_page(size);
697 if (addr + size < addr)
700 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
701 addr + size, prot, FALSE)) {
704 case KERN_PROTECTION_FAILURE:
706 case KERN_RESOURCE_SHORTAGE:
712 #ifndef _SYS_SYSPROTO_H_
713 struct minherit_args {
723 sys_minherit(td, uap)
725 struct minherit_args *uap;
728 vm_size_t size, pageoff;
729 vm_inherit_t inherit;
731 addr = (vm_offset_t)uap->addr;
733 inherit = uap->inherit;
735 pageoff = (addr & PAGE_MASK);
738 size = (vm_size_t) round_page(size);
739 if (addr + size < addr)
742 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
743 addr + size, inherit)) {
746 case KERN_PROTECTION_FAILURE:
752 #ifndef _SYS_SYSPROTO_H_
753 struct madvise_args {
766 struct madvise_args *uap;
768 vm_offset_t start, end;
773 * Check for our special case, advising the swap pager we are
776 if (uap->behav == MADV_PROTECT) {
778 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
779 PROC_SPROTECT, &flags));
783 * Check for illegal behavior
785 if (uap->behav < 0 || uap->behav > MADV_CORE)
788 * Check for illegal addresses. Watch out for address wrap... Note
789 * that VM_*_ADDRESS are not constants due to casts (argh).
791 map = &td->td_proc->p_vmspace->vm_map;
792 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
793 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
795 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
799 * Since this routine is only advisory, we default to conservative
802 start = trunc_page((vm_offset_t) uap->addr);
803 end = round_page((vm_offset_t) uap->addr + uap->len);
805 if (vm_map_madvise(map, start, end, uap->behav))
810 #ifndef _SYS_SYSPROTO_H_
811 struct mincore_args {
824 struct mincore_args *uap;
826 vm_offset_t addr, first_addr;
827 vm_offset_t end, cend;
832 int vecindex, lastvecindex;
833 vm_map_entry_t current;
834 vm_map_entry_t entry;
836 vm_paddr_t locked_pa;
840 unsigned int timestamp;
844 * Make sure that the addresses presented are valid for user
847 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
848 end = addr + (vm_size_t)round_page(uap->len);
849 map = &td->td_proc->p_vmspace->vm_map;
850 if (end > vm_map_max(map) || end < addr)
854 * Address of byte vector
858 pmap = vmspace_pmap(td->td_proc->p_vmspace);
860 vm_map_lock_read(map);
862 timestamp = map->timestamp;
864 if (!vm_map_lookup_entry(map, addr, &entry)) {
865 vm_map_unlock_read(map);
870 * Do this on a map entry basis so that if the pages are not
871 * in the current processes address space, we can easily look
872 * up the pages elsewhere.
875 for (current = entry;
876 (current != &map->header) && (current->start < end);
877 current = current->next) {
880 * check for contiguity
882 if (current->end < end &&
883 (entry->next == &map->header ||
884 current->next->start > current->end)) {
885 vm_map_unlock_read(map);
890 * ignore submaps (for now) or null objects
892 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
893 current->object.vm_object == NULL)
897 * limit this scan to the current map entry and the
898 * limits for the mincore call
900 if (addr < current->start)
901 addr = current->start;
907 * scan this entry one page at a time
909 while (addr < cend) {
911 * Check pmap first, it is likely faster, also
912 * it can provide info as to whether we are the
913 * one referencing or modifying the page.
919 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
920 if (locked_pa != 0) {
922 * The page is mapped by this process but not
923 * both accessed and modified. It is also
924 * managed. Acquire the object lock so that
925 * other mappings might be examined.
927 m = PHYS_TO_VM_PAGE(locked_pa);
928 if (m->object != object) {
930 VM_OBJECT_WUNLOCK(object);
932 locked = VM_OBJECT_TRYWLOCK(object);
935 VM_OBJECT_WLOCK(object);
941 KASSERT(m->valid == VM_PAGE_BITS_ALL,
942 ("mincore: page %p is mapped but invalid",
944 } else if (mincoreinfo == 0) {
946 * The page is not mapped by this process. If
947 * the object implements managed pages, then
948 * determine if the page is resident so that
949 * the mappings might be examined.
951 if (current->object.vm_object != object) {
953 VM_OBJECT_WUNLOCK(object);
954 object = current->object.vm_object;
955 VM_OBJECT_WLOCK(object);
957 if (object->type == OBJT_DEFAULT ||
958 object->type == OBJT_SWAP ||
959 object->type == OBJT_VNODE) {
960 pindex = OFF_TO_IDX(current->offset +
961 (addr - current->start));
962 m = vm_page_lookup(object, pindex);
964 vm_page_is_cached(object, pindex))
965 mincoreinfo = MINCORE_INCORE;
966 if (m != NULL && m->valid == 0)
969 mincoreinfo = MINCORE_INCORE;
973 /* Examine other mappings to the page. */
974 if (m->dirty == 0 && pmap_is_modified(m))
977 mincoreinfo |= MINCORE_MODIFIED_OTHER;
979 * The first test for PGA_REFERENCED is an
980 * optimization. The second test is
981 * required because a concurrent pmap
982 * operation could clear the last reference
983 * and set PGA_REFERENCED before the call to
984 * pmap_is_referenced().
986 if ((m->aflags & PGA_REFERENCED) != 0 ||
987 pmap_is_referenced(m) ||
988 (m->aflags & PGA_REFERENCED) != 0)
989 mincoreinfo |= MINCORE_REFERENCED_OTHER;
992 VM_OBJECT_WUNLOCK(object);
995 * subyte may page fault. In case it needs to modify
996 * the map, we release the lock.
998 vm_map_unlock_read(map);
1001 * calculate index into user supplied byte vector
1003 vecindex = OFF_TO_IDX(addr - first_addr);
1006 * If we have skipped map entries, we need to make sure that
1007 * the byte vector is zeroed for those skipped entries.
1009 while ((lastvecindex + 1) < vecindex) {
1011 error = subyte(vec + lastvecindex, 0);
1019 * Pass the page information to the user
1021 error = subyte(vec + vecindex, mincoreinfo);
1028 * If the map has changed, due to the subyte, the previous
1029 * output may be invalid.
1031 vm_map_lock_read(map);
1032 if (timestamp != map->timestamp)
1035 lastvecindex = vecindex;
1041 * subyte may page fault. In case it needs to modify
1042 * the map, we release the lock.
1044 vm_map_unlock_read(map);
1047 * Zero the last entries in the byte vector.
1049 vecindex = OFF_TO_IDX(end - first_addr);
1050 while ((lastvecindex + 1) < vecindex) {
1052 error = subyte(vec + lastvecindex, 0);
1060 * If the map has changed, due to the subyte, the previous
1061 * output may be invalid.
1063 vm_map_lock_read(map);
1064 if (timestamp != map->timestamp)
1066 vm_map_unlock_read(map);
1071 #ifndef _SYS_SYSPROTO_H_
1083 struct mlock_args *uap;
1086 return (vm_mlock(td->td_proc, td->td_ucred, uap->addr, uap->len));
1090 vm_mlock(struct proc *proc, struct ucred *cred, const void *addr0, size_t len)
1092 vm_offset_t addr, end, last, start;
1093 vm_size_t npages, size;
1095 unsigned long nsize;
1098 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1101 addr = (vm_offset_t)addr0;
1104 start = trunc_page(addr);
1105 end = round_page(last);
1106 if (last < addr || end < addr)
1108 npages = atop(end - start);
1109 if (npages > vm_page_max_wired)
1111 map = &proc->p_vmspace->vm_map;
1113 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1114 if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1119 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1123 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1128 error = vm_map_wire(map, start, end,
1129 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1131 if (error != KERN_SUCCESS) {
1133 racct_set(proc, RACCT_MEMLOCK,
1134 ptoa(pmap_wired_count(map->pmap)));
1138 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1141 #ifndef _SYS_SYSPROTO_H_
1142 struct mlockall_args {
1151 sys_mlockall(td, uap)
1153 struct mlockall_args *uap;
1158 map = &td->td_proc->p_vmspace->vm_map;
1159 error = priv_check(td, PRIV_VM_MLOCK);
1163 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1167 * If wiring all pages in the process would cause it to exceed
1168 * a hard resource limit, return ENOMEM.
1170 if (!old_mlock && uap->how & MCL_CURRENT) {
1171 PROC_LOCK(td->td_proc);
1172 if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1173 PROC_UNLOCK(td->td_proc);
1176 PROC_UNLOCK(td->td_proc);
1179 PROC_LOCK(td->td_proc);
1180 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1181 PROC_UNLOCK(td->td_proc);
1186 if (uap->how & MCL_FUTURE) {
1188 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1193 if (uap->how & MCL_CURRENT) {
1195 * P1003.1-2001 mandates that all currently mapped pages
1196 * will be memory resident and locked (wired) upon return
1197 * from mlockall(). vm_map_wire() will wire pages, by
1198 * calling vm_fault_wire() for each page in the region.
1200 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1201 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1202 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1205 if (error != KERN_SUCCESS) {
1206 PROC_LOCK(td->td_proc);
1207 racct_set(td->td_proc, RACCT_MEMLOCK,
1208 ptoa(pmap_wired_count(map->pmap)));
1209 PROC_UNLOCK(td->td_proc);
1216 #ifndef _SYS_SYSPROTO_H_
1217 struct munlockall_args {
1226 sys_munlockall(td, uap)
1228 struct munlockall_args *uap;
1233 map = &td->td_proc->p_vmspace->vm_map;
1234 error = priv_check(td, PRIV_VM_MUNLOCK);
1238 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1240 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1243 /* Forcibly unwire all pages. */
1244 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1245 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1247 if (error == KERN_SUCCESS) {
1248 PROC_LOCK(td->td_proc);
1249 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1250 PROC_UNLOCK(td->td_proc);
1257 #ifndef _SYS_SYSPROTO_H_
1258 struct munlock_args {
1267 sys_munlock(td, uap)
1269 struct munlock_args *uap;
1271 vm_offset_t addr, end, last, start;
1278 error = priv_check(td, PRIV_VM_MUNLOCK);
1281 addr = (vm_offset_t)uap->addr;
1284 start = trunc_page(addr);
1285 end = round_page(last);
1286 if (last < addr || end < addr)
1288 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1289 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1291 if (error == KERN_SUCCESS) {
1292 PROC_LOCK(td->td_proc);
1293 map = &td->td_proc->p_vmspace->vm_map;
1294 racct_set(td->td_proc, RACCT_MEMLOCK,
1295 ptoa(pmap_wired_count(map->pmap)));
1296 PROC_UNLOCK(td->td_proc);
1299 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1305 * Helper function for vm_mmap. Perform sanity check specific for mmap
1306 * operations on vnodes.
1308 * For VCHR vnodes, the vnode lock is held over the call to
1309 * vm_mmap_cdev() to keep vp->v_rdev valid.
1312 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1313 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1314 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1315 boolean_t *writecounted)
1322 int error, flags, locktype;
1325 cred = td->td_ucred;
1326 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1327 locktype = LK_EXCLUSIVE;
1329 locktype = LK_SHARED;
1330 if ((error = vget(vp, locktype, td)) != 0)
1335 if (vp->v_type == VREG) {
1337 * Get the proper underlying object
1343 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1345 vp = (struct vnode *)obj->handle;
1347 * Bypass filesystems obey the mpsafety of the
1348 * underlying fs. Tmpfs never bypasses.
1350 error = vget(vp, locktype, td);
1354 if (locktype == LK_EXCLUSIVE) {
1355 *writecounted = TRUE;
1356 vnode_pager_update_writecount(obj, 0, objsize);
1358 } else if (vp->v_type == VCHR) {
1359 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1360 vp->v_rdev, foffp, objp);
1368 if ((error = VOP_GETATTR(vp, &va, cred)))
1371 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1375 if ((flags & MAP_SHARED) != 0) {
1376 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1377 if (prot & PROT_WRITE) {
1381 *maxprotp &= ~VM_PROT_WRITE;
1385 * If it is a regular file without any references
1386 * we do not need to sync it.
1387 * Adjust object size to be the size of actual file.
1389 objsize = round_page(va.va_size);
1390 if (va.va_nlink == 0)
1391 flags |= MAP_NOSYNC;
1392 if (obj->type == OBJT_VNODE)
1393 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1396 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1397 ("wrong object type"));
1398 vm_object_reference(obj);
1408 vfs_mark_atime(vp, cred);
1411 if (error != 0 && *writecounted) {
1412 *writecounted = FALSE;
1413 vnode_pager_update_writecount(obj, objsize, 0);
1424 * Helper function for vm_mmap. Perform sanity check specific for mmap
1425 * operations on cdevs.
1428 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1429 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1430 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1434 int error, flags, ref;
1438 dsw = dev_refthread(cdev, &ref);
1441 if (dsw->d_flags & D_MMAP_ANON) {
1442 dev_relthread(cdev, ref);
1443 *maxprotp = VM_PROT_ALL;
1444 *flagsp |= MAP_ANON;
1448 * cdevs do not provide private mappings of any kind.
1450 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1451 (prot & PROT_WRITE) != 0) {
1452 dev_relthread(cdev, ref);
1455 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1456 dev_relthread(cdev, ref);
1460 * Force device mappings to be shared.
1462 flags |= MAP_SHARED;
1464 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1466 dev_relthread(cdev, ref);
1471 * First, try d_mmap_single(). If that is not implemented
1472 * (returns ENODEV), fall back to using the device pager.
1473 * Note that d_mmap_single() must return a reference to the
1474 * object (it needs to bump the reference count of the object
1475 * it returns somehow).
1477 * XXX assumes VM_PROT_* == PROT_*
1479 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1480 dev_relthread(cdev, ref);
1481 if (error != ENODEV)
1483 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1497 * Helper function for vm_mmap. Perform sanity check specific for mmap
1498 * operations on shm file descriptors.
1501 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1502 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1503 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1507 if ((*flagsp & MAP_SHARED) != 0 &&
1508 (*maxprotp & VM_PROT_WRITE) == 0 &&
1509 (prot & PROT_WRITE) != 0)
1512 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1516 error = shm_mmap(shmfd, objsize, foff, objp);
1527 * Internal version of mmap. Currently used by mmap, exec, and sys5
1528 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1531 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1532 vm_prot_t maxprot, int flags,
1533 objtype_t handle_type, void *handle,
1537 vm_object_t object = NULL;
1538 struct thread *td = curthread;
1539 int docow, error, findspace, rv;
1540 boolean_t writecounted;
1545 size = round_page(size);
1547 if (map == &td->td_proc->p_vmspace->vm_map) {
1548 PROC_LOCK(td->td_proc);
1549 if (map->size + size > lim_cur(td->td_proc, RLIMIT_VMEM)) {
1550 PROC_UNLOCK(td->td_proc);
1553 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1554 PROC_UNLOCK(td->td_proc);
1557 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1558 if (ptoa(pmap_wired_count(map->pmap)) + size >
1559 lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1560 racct_set_force(td->td_proc, RACCT_VMEM,
1562 PROC_UNLOCK(td->td_proc);
1565 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1566 ptoa(pmap_wired_count(map->pmap)) + size);
1568 racct_set_force(td->td_proc, RACCT_VMEM,
1570 PROC_UNLOCK(td->td_proc);
1574 PROC_UNLOCK(td->td_proc);
1578 * We currently can only deal with page aligned file offsets.
1579 * The check is here rather than in the syscall because the
1580 * kernel calls this function internally for other mmaping
1581 * operations (such as in exec) and non-aligned offsets will
1582 * cause pmap inconsistencies...so we want to be sure to
1583 * disallow this in all cases.
1585 if (foff & PAGE_MASK)
1588 if ((flags & MAP_FIXED) == 0) {
1590 *addr = round_page(*addr);
1592 if (*addr != trunc_page(*addr))
1596 writecounted = FALSE;
1599 * Lookup/allocate object.
1601 switch (handle_type) {
1603 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1604 handle, &foff, &object);
1607 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1608 handle, &foff, &object, &writecounted);
1611 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1612 handle, foff, &object);
1615 if (handle == NULL) {
1626 if (flags & MAP_ANON) {
1630 * Unnamed anonymous regions always start at 0.
1634 } else if (flags & MAP_PREFAULT_READ)
1635 docow = MAP_PREFAULT;
1637 docow = MAP_PREFAULT_PARTIAL;
1639 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1640 docow |= MAP_COPY_ON_WRITE;
1641 if (flags & MAP_NOSYNC)
1642 docow |= MAP_DISABLE_SYNCER;
1643 if (flags & MAP_NOCORE)
1644 docow |= MAP_DISABLE_COREDUMP;
1645 /* Shared memory is also shared with children. */
1646 if (flags & MAP_SHARED)
1647 docow |= MAP_INHERIT_SHARE;
1649 docow |= MAP_VN_WRITECOUNT;
1650 if (flags & MAP_STACK) {
1653 docow |= MAP_STACK_GROWS_DOWN;
1655 if ((flags & MAP_EXCL) != 0)
1656 docow |= MAP_CHECK_EXCL;
1659 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1660 findspace = VMFS_SUPER_SPACE;
1661 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1662 findspace = VMFS_ALIGNED_SPACE(flags >>
1663 MAP_ALIGNMENT_SHIFT);
1665 findspace = VMFS_OPTIMAL_SPACE;
1666 rv = vm_map_find(map, object, foff, addr, size,
1668 flags & MAP_32BIT ? MAP_32BIT_MAX_ADDR :
1670 0, findspace, prot, maxprot, docow);
1672 rv = vm_map_fixed(map, object, foff, *addr, size,
1673 prot, maxprot, docow);
1676 if (rv == KERN_SUCCESS) {
1678 * If the process has requested that all future mappings
1679 * be wired, then heed this.
1681 if (map->flags & MAP_WIREFUTURE) {
1682 vm_map_wire(map, *addr, *addr + size,
1683 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1684 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1688 * If this mapping was accounted for in the vnode's
1689 * writecount, then undo that now.
1692 vnode_pager_release_writecount(object, 0, size);
1694 * Lose the object reference. Will destroy the
1695 * object if it's an unnamed anonymous mapping
1696 * or named anonymous without other references.
1698 vm_object_deallocate(object);
1700 return (vm_mmap_to_errno(rv));
1704 * Translate a Mach VM return code to zero on success or the appropriate errno
1708 vm_mmap_to_errno(int rv)
1714 case KERN_INVALID_ADDRESS:
1717 case KERN_PROTECTION_FAILURE: