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 * 3. 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/audit/audit.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_param.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
93 #include <sys/pmckern.h>
97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
98 "Do not apply RLIMIT_MEMLOCK on mlockall");
101 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
104 #ifndef _SYS_SYSPROTO_H_
111 sys_sbrk(struct thread *td, struct sbrk_args *uap)
113 /* Not yet implemented */
117 #ifndef _SYS_SYSPROTO_H_
124 sys_sstk(struct thread *td, struct sstk_args *uap)
126 /* Not yet implemented */
130 #if defined(COMPAT_43)
131 #ifndef _SYS_SYSPROTO_H_
132 struct getpagesize_args {
138 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
141 td->td_retval[0] = PAGE_SIZE;
144 #endif /* COMPAT_43 */
148 * Memory Map (mmap) system call. Note that the file offset
149 * and address are allowed to be NOT page aligned, though if
150 * the MAP_FIXED flag it set, both must have the same remainder
151 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
152 * page-aligned, the actual mapping starts at trunc_page(addr)
153 * and the return value is adjusted up by the page offset.
155 * Generally speaking, only character devices which are themselves
156 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
157 * there would be no cache coherency between a descriptor and a VM mapping
158 * both to the same character device.
160 #ifndef _SYS_SYSPROTO_H_
173 sys_mmap(struct thread *td, struct mmap_args *uap)
176 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
177 uap->flags, uap->fd, uap->pos));
181 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
188 vm_prot_t cap_maxprot;
192 vms = td->td_proc->p_vmspace;
198 * Ignore old flags that used to be defined but did not do anything.
200 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
203 * Enforce the constraints.
204 * Mapping of length 0 is only allowed for old binaries.
205 * Anonymous mapping shall specify -1 as filedescriptor and
206 * zero position for new code. Be nice to ancient a.out
207 * binaries and correct pos for anonymous mapping, since old
208 * ld.so sometimes issues anonymous map requests with non-zero
211 if (!SV_CURPROC_FLAG(SV_AOUT)) {
212 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
213 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
216 if ((flags & MAP_ANON) != 0)
220 if (flags & MAP_STACK) {
222 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
227 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
228 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
229 MAP_PREFAULT_READ | MAP_GUARD |
233 MAP_ALIGNMENT_MASK)) != 0)
235 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
237 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
239 if (prot != PROT_NONE &&
240 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
242 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
243 pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
244 MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
248 * Align the file position to a page boundary,
249 * and save its page offset component.
251 pageoff = (pos & PAGE_MASK);
254 /* Adjust size for rounding (on both ends). */
255 size += pageoff; /* low end... */
256 size = (vm_size_t) round_page(size); /* hi end */
258 /* Ensure alignment is at least a page and fits in a pointer. */
259 align = flags & MAP_ALIGNMENT_MASK;
260 if (align != 0 && align != MAP_ALIGNED_SUPER &&
261 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
262 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
266 * Check for illegal addresses. Watch out for address wrap... Note
267 * that VM_*_ADDRESS are not constants due to casts (argh).
269 if (flags & MAP_FIXED) {
271 * The specified address must have the same remainder
272 * as the file offset taken modulo PAGE_SIZE, so it
273 * should be aligned after adjustment by pageoff.
276 if (addr & PAGE_MASK)
279 /* Address range must be all in user VM space. */
280 if (addr < vm_map_min(&vms->vm_map) ||
281 addr + size > vm_map_max(&vms->vm_map))
283 if (addr + size < addr)
286 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
288 } else if (flags & MAP_32BIT) {
290 * For MAP_32BIT, override the hint if it is too high and
291 * do not bother moving the mapping past the heap (since
292 * the heap is usually above 2GB).
294 if (addr + size > MAP_32BIT_MAX_ADDR)
299 * XXX for non-fixed mappings where no hint is provided or
300 * the hint would fall in the potential heap space,
301 * place it after the end of the largest possible heap.
303 * There should really be a pmap call to determine a reasonable
307 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
308 addr < round_page((vm_offset_t)vms->vm_daddr +
309 lim_max(td, RLIMIT_DATA))))
310 addr = round_page((vm_offset_t)vms->vm_daddr +
311 lim_max(td, RLIMIT_DATA));
315 * Return success without mapping anything for old
316 * binaries that request a page-aligned mapping of
317 * length 0. For modern binaries, this function
318 * returns an error earlier.
321 } else if ((flags & MAP_GUARD) != 0) {
322 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
323 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
324 } else if ((flags & MAP_ANON) != 0) {
326 * Mapping blank space is trivial.
328 * This relies on VM_PROT_* matching PROT_*.
330 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
331 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
334 * Mapping file, get fp for validation and don't let the
335 * descriptor disappear on us if we block. Check capability
336 * rights, but also return the maximum rights to be combined
337 * with maxprot later.
339 cap_rights_init(&rights, CAP_MMAP);
340 if (prot & PROT_READ)
341 cap_rights_set(&rights, CAP_MMAP_R);
342 if ((flags & MAP_SHARED) != 0) {
343 if (prot & PROT_WRITE)
344 cap_rights_set(&rights, CAP_MMAP_W);
346 if (prot & PROT_EXEC)
347 cap_rights_set(&rights, CAP_MMAP_X);
348 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
351 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
352 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
357 /* This relies on VM_PROT_* matching PROT_*. */
358 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
359 cap_maxprot, flags, pos, td);
363 td->td_retval[0] = (register_t) (addr + pageoff);
371 #if defined(COMPAT_FREEBSD6)
373 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
376 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
377 uap->flags, uap->fd, uap->pos));
382 #ifndef _SYS_SYSPROTO_H_
393 ommap(struct thread *td, struct ommap_args *uap)
395 static const char cvtbsdprot[8] = {
399 PROT_EXEC | PROT_WRITE,
401 PROT_EXEC | PROT_READ,
402 PROT_WRITE | PROT_READ,
403 PROT_EXEC | PROT_WRITE | PROT_READ,
407 #define OMAP_ANON 0x0002
408 #define OMAP_COPY 0x0020
409 #define OMAP_SHARED 0x0010
410 #define OMAP_FIXED 0x0100
412 prot = cvtbsdprot[uap->prot & 0x7];
413 #ifdef COMPAT_FREEBSD32
414 #if defined(__amd64__)
415 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
421 if (uap->flags & OMAP_ANON)
423 if (uap->flags & OMAP_COPY)
425 if (uap->flags & OMAP_SHARED)
428 flags |= MAP_PRIVATE;
429 if (uap->flags & OMAP_FIXED)
431 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
434 #endif /* COMPAT_43 */
437 #ifndef _SYS_SYSPROTO_H_
445 sys_msync(struct thread *td, struct msync_args *uap)
448 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
452 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
460 pageoff = (addr & PAGE_MASK);
463 size = (vm_size_t) round_page(size);
464 if (addr + size < addr)
467 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
470 map = &td->td_proc->p_vmspace->vm_map;
473 * Clean the pages and interpret the return value.
475 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
476 (flags & MS_INVALIDATE) != 0);
480 case KERN_INVALID_ADDRESS:
482 case KERN_INVALID_ARGUMENT:
491 #ifndef _SYS_SYSPROTO_H_
498 sys_munmap(struct thread *td, struct munmap_args *uap)
501 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
505 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
508 struct pmckern_map_out pkm;
509 vm_map_entry_t entry;
520 pageoff = (addr & PAGE_MASK);
523 size = (vm_size_t) round_page(size);
524 if (addr + size < addr)
528 * Check for illegal addresses. Watch out for address wrap...
530 map = &td->td_proc->p_vmspace->vm_map;
531 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
536 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
539 * Inform hwpmc if the address range being unmapped contains
540 * an executable region.
542 pkm.pm_address = (uintptr_t) NULL;
543 if (vm_map_lookup_entry(map, addr, &entry)) {
545 entry != &map->header && entry->start < addr + size;
546 entry = entry->next) {
547 if (vm_map_check_protection(map, entry->start,
548 entry->end, VM_PROT_EXECUTE) == TRUE) {
549 pkm.pm_address = (uintptr_t) addr;
550 pkm.pm_size = (size_t) size;
557 vm_map_delete(map, addr, addr + size);
560 if (__predict_false(pmc_handled)) {
561 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
562 vm_map_lock_downgrade(map);
563 if (pkm.pm_address != (uintptr_t) NULL)
564 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
565 vm_map_unlock_read(map);
570 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
574 #ifndef _SYS_SYSPROTO_H_
575 struct mprotect_args {
582 sys_mprotect(struct thread *td, struct mprotect_args *uap)
585 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
589 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
595 prot = (prot & VM_PROT_ALL);
596 pageoff = (addr & PAGE_MASK);
599 size = (vm_size_t) round_page(size);
600 if (addr + size < addr)
603 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
604 addr + size, prot, FALSE)) {
607 case KERN_PROTECTION_FAILURE:
609 case KERN_RESOURCE_SHORTAGE:
615 #ifndef _SYS_SYSPROTO_H_
616 struct minherit_args {
623 sys_minherit(struct thread *td, struct minherit_args *uap)
626 vm_size_t size, pageoff;
627 vm_inherit_t inherit;
629 addr = (vm_offset_t)uap->addr;
631 inherit = uap->inherit;
633 pageoff = (addr & PAGE_MASK);
636 size = (vm_size_t) round_page(size);
637 if (addr + size < addr)
640 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
641 addr + size, inherit)) {
644 case KERN_PROTECTION_FAILURE:
650 #ifndef _SYS_SYSPROTO_H_
651 struct madvise_args {
659 sys_madvise(struct thread *td, struct madvise_args *uap)
662 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
666 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
669 vm_offset_t addr, end, start;
673 * Check for our special case, advising the swap pager we are
676 if (behav == MADV_PROTECT) {
678 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
679 PROC_SPROTECT, &flags));
683 * Check for illegal behavior
685 if (behav < 0 || behav > MADV_CORE)
688 * Check for illegal addresses. Watch out for address wrap... Note
689 * that VM_*_ADDRESS are not constants due to casts (argh).
691 map = &td->td_proc->p_vmspace->vm_map;
693 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
695 if ((addr + len) < addr)
699 * Since this routine is only advisory, we default to conservative
702 start = trunc_page(addr);
703 end = round_page(addr + len);
705 if (vm_map_madvise(map, start, end, behav))
710 #ifndef _SYS_SYSPROTO_H_
711 struct mincore_args {
719 sys_mincore(struct thread *td, struct mincore_args *uap)
722 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
726 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
728 vm_offset_t addr, first_addr;
729 vm_offset_t end, cend;
733 int vecindex, lastvecindex;
734 vm_map_entry_t current;
735 vm_map_entry_t entry;
737 vm_paddr_t locked_pa;
741 unsigned int timestamp;
745 * Make sure that the addresses presented are valid for user
748 first_addr = addr = trunc_page(addr0);
749 end = addr + (vm_size_t)round_page(len);
750 map = &td->td_proc->p_vmspace->vm_map;
751 if (end > vm_map_max(map) || end < addr)
754 pmap = vmspace_pmap(td->td_proc->p_vmspace);
756 vm_map_lock_read(map);
758 timestamp = map->timestamp;
760 if (!vm_map_lookup_entry(map, addr, &entry)) {
761 vm_map_unlock_read(map);
766 * Do this on a map entry basis so that if the pages are not
767 * in the current processes address space, we can easily look
768 * up the pages elsewhere.
771 for (current = entry;
772 (current != &map->header) && (current->start < end);
773 current = current->next) {
776 * check for contiguity
778 if (current->end < end &&
779 (entry->next == &map->header ||
780 current->next->start > current->end)) {
781 vm_map_unlock_read(map);
786 * ignore submaps (for now) or null objects
788 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
789 current->object.vm_object == NULL)
793 * limit this scan to the current map entry and the
794 * limits for the mincore call
796 if (addr < current->start)
797 addr = current->start;
803 * scan this entry one page at a time
805 while (addr < cend) {
807 * Check pmap first, it is likely faster, also
808 * it can provide info as to whether we are the
809 * one referencing or modifying the page.
815 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
816 if (locked_pa != 0) {
818 * The page is mapped by this process but not
819 * both accessed and modified. It is also
820 * managed. Acquire the object lock so that
821 * other mappings might be examined.
823 m = PHYS_TO_VM_PAGE(locked_pa);
824 if (m->object != object) {
826 VM_OBJECT_WUNLOCK(object);
828 locked = VM_OBJECT_TRYWLOCK(object);
831 VM_OBJECT_WLOCK(object);
837 KASSERT(m->valid == VM_PAGE_BITS_ALL,
838 ("mincore: page %p is mapped but invalid",
840 } else if (mincoreinfo == 0) {
842 * The page is not mapped by this process. If
843 * the object implements managed pages, then
844 * determine if the page is resident so that
845 * the mappings might be examined.
847 if (current->object.vm_object != object) {
849 VM_OBJECT_WUNLOCK(object);
850 object = current->object.vm_object;
851 VM_OBJECT_WLOCK(object);
853 if (object->type == OBJT_DEFAULT ||
854 object->type == OBJT_SWAP ||
855 object->type == OBJT_VNODE) {
856 pindex = OFF_TO_IDX(current->offset +
857 (addr - current->start));
858 m = vm_page_lookup(object, pindex);
859 if (m != NULL && m->valid == 0)
862 mincoreinfo = MINCORE_INCORE;
866 /* Examine other mappings to the page. */
867 if (m->dirty == 0 && pmap_is_modified(m))
870 mincoreinfo |= MINCORE_MODIFIED_OTHER;
872 * The first test for PGA_REFERENCED is an
873 * optimization. The second test is
874 * required because a concurrent pmap
875 * operation could clear the last reference
876 * and set PGA_REFERENCED before the call to
877 * pmap_is_referenced().
879 if ((m->aflags & PGA_REFERENCED) != 0 ||
880 pmap_is_referenced(m) ||
881 (m->aflags & PGA_REFERENCED) != 0)
882 mincoreinfo |= MINCORE_REFERENCED_OTHER;
885 VM_OBJECT_WUNLOCK(object);
888 * subyte may page fault. In case it needs to modify
889 * the map, we release the lock.
891 vm_map_unlock_read(map);
894 * calculate index into user supplied byte vector
896 vecindex = atop(addr - first_addr);
899 * If we have skipped map entries, we need to make sure that
900 * the byte vector is zeroed for those skipped entries.
902 while ((lastvecindex + 1) < vecindex) {
904 error = subyte(vec + lastvecindex, 0);
912 * Pass the page information to the user
914 error = subyte(vec + vecindex, mincoreinfo);
921 * If the map has changed, due to the subyte, the previous
922 * output may be invalid.
924 vm_map_lock_read(map);
925 if (timestamp != map->timestamp)
928 lastvecindex = vecindex;
934 * subyte may page fault. In case it needs to modify
935 * the map, we release the lock.
937 vm_map_unlock_read(map);
940 * Zero the last entries in the byte vector.
942 vecindex = atop(end - first_addr);
943 while ((lastvecindex + 1) < vecindex) {
945 error = subyte(vec + lastvecindex, 0);
953 * If the map has changed, due to the subyte, the previous
954 * output may be invalid.
956 vm_map_lock_read(map);
957 if (timestamp != map->timestamp)
959 vm_map_unlock_read(map);
964 #ifndef _SYS_SYSPROTO_H_
971 sys_mlock(struct thread *td, struct mlock_args *uap)
974 return (kern_mlock(td->td_proc, td->td_ucred,
975 __DECONST(uintptr_t, uap->addr), uap->len));
979 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
981 vm_offset_t addr, end, last, start;
982 vm_size_t npages, size;
987 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
993 start = trunc_page(addr);
994 end = round_page(last);
995 if (last < addr || end < addr)
997 npages = atop(end - start);
998 if (npages > vm_page_max_wired)
1000 map = &proc->p_vmspace->vm_map;
1002 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1003 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1008 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1013 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1019 error = vm_map_wire(map, start, end,
1020 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1022 if (racct_enable && error != KERN_SUCCESS) {
1024 racct_set(proc, RACCT_MEMLOCK,
1025 ptoa(pmap_wired_count(map->pmap)));
1029 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1032 #ifndef _SYS_SYSPROTO_H_
1033 struct mlockall_args {
1039 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1044 map = &td->td_proc->p_vmspace->vm_map;
1045 error = priv_check(td, PRIV_VM_MLOCK);
1049 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1053 * If wiring all pages in the process would cause it to exceed
1054 * a hard resource limit, return ENOMEM.
1056 if (!old_mlock && uap->how & MCL_CURRENT) {
1057 PROC_LOCK(td->td_proc);
1058 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1059 PROC_UNLOCK(td->td_proc);
1062 PROC_UNLOCK(td->td_proc);
1066 PROC_LOCK(td->td_proc);
1067 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1068 PROC_UNLOCK(td->td_proc);
1074 if (uap->how & MCL_FUTURE) {
1076 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1081 if (uap->how & MCL_CURRENT) {
1083 * P1003.1-2001 mandates that all currently mapped pages
1084 * will be memory resident and locked (wired) upon return
1085 * from mlockall(). vm_map_wire() will wire pages, by
1086 * calling vm_fault_wire() for each page in the region.
1088 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1089 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1090 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1093 if (racct_enable && error != KERN_SUCCESS) {
1094 PROC_LOCK(td->td_proc);
1095 racct_set(td->td_proc, RACCT_MEMLOCK,
1096 ptoa(pmap_wired_count(map->pmap)));
1097 PROC_UNLOCK(td->td_proc);
1104 #ifndef _SYS_SYSPROTO_H_
1105 struct munlockall_args {
1111 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1116 map = &td->td_proc->p_vmspace->vm_map;
1117 error = priv_check(td, PRIV_VM_MUNLOCK);
1121 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1123 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1126 /* Forcibly unwire all pages. */
1127 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1128 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1130 if (racct_enable && error == KERN_SUCCESS) {
1131 PROC_LOCK(td->td_proc);
1132 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1133 PROC_UNLOCK(td->td_proc);
1140 #ifndef _SYS_SYSPROTO_H_
1141 struct munlock_args {
1147 sys_munlock(struct thread *td, struct munlock_args *uap)
1150 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1154 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1156 vm_offset_t addr, end, last, start;
1162 error = priv_check(td, PRIV_VM_MUNLOCK);
1167 start = trunc_page(addr);
1168 end = round_page(last);
1169 if (last < addr || end < addr)
1171 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1172 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1174 if (racct_enable && error == KERN_SUCCESS) {
1175 PROC_LOCK(td->td_proc);
1176 map = &td->td_proc->p_vmspace->vm_map;
1177 racct_set(td->td_proc, RACCT_MEMLOCK,
1178 ptoa(pmap_wired_count(map->pmap)));
1179 PROC_UNLOCK(td->td_proc);
1182 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1188 * Helper function for vm_mmap. Perform sanity check specific for mmap
1189 * operations on vnodes.
1192 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1193 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1194 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1195 boolean_t *writecounted)
1201 int error, flags, locktype;
1203 cred = td->td_ucred;
1204 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1205 locktype = LK_EXCLUSIVE;
1207 locktype = LK_SHARED;
1208 if ((error = vget(vp, locktype, td)) != 0)
1210 AUDIT_ARG_VNODE1(vp);
1214 if (vp->v_type == VREG) {
1216 * Get the proper underlying object
1222 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1224 vp = (struct vnode *)obj->handle;
1226 * Bypass filesystems obey the mpsafety of the
1227 * underlying fs. Tmpfs never bypasses.
1229 error = vget(vp, locktype, td);
1233 if (locktype == LK_EXCLUSIVE) {
1234 *writecounted = TRUE;
1235 vnode_pager_update_writecount(obj, 0, objsize);
1241 if ((error = VOP_GETATTR(vp, &va, cred)))
1244 /* This relies on VM_PROT_* matching PROT_*. */
1245 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1249 if ((flags & MAP_SHARED) != 0) {
1250 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1251 if (prot & VM_PROT_WRITE) {
1255 *maxprotp &= ~VM_PROT_WRITE;
1259 * If it is a regular file without any references
1260 * we do not need to sync it.
1261 * Adjust object size to be the size of actual file.
1263 objsize = round_page(va.va_size);
1264 if (va.va_nlink == 0)
1265 flags |= MAP_NOSYNC;
1266 if (obj->type == OBJT_VNODE) {
1267 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1274 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1275 ("wrong object type"));
1276 VM_OBJECT_WLOCK(obj);
1277 vm_object_reference_locked(obj);
1278 #if VM_NRESERVLEVEL > 0
1279 vm_object_color(obj, 0);
1281 VM_OBJECT_WUNLOCK(obj);
1286 vfs_mark_atime(vp, cred);
1289 if (error != 0 && *writecounted) {
1290 *writecounted = FALSE;
1291 vnode_pager_update_writecount(obj, objsize, 0);
1300 * Helper function for vm_mmap. Perform sanity check specific for mmap
1301 * operations on cdevs.
1304 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1305 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1306 vm_ooffset_t *foff, vm_object_t *objp)
1313 if (dsw->d_flags & D_MMAP_ANON) {
1316 *maxprotp = VM_PROT_ALL;
1317 *flagsp |= MAP_ANON;
1321 * cdevs do not provide private mappings of any kind.
1323 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1324 (prot & VM_PROT_WRITE) != 0)
1326 if (flags & (MAP_PRIVATE|MAP_COPY))
1329 * Force device mappings to be shared.
1331 flags |= MAP_SHARED;
1333 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1338 * First, try d_mmap_single(). If that is not implemented
1339 * (returns ENODEV), fall back to using the device pager.
1340 * Note that d_mmap_single() must return a reference to the
1341 * object (it needs to bump the reference count of the object
1342 * it returns somehow).
1344 * XXX assumes VM_PROT_* == PROT_*
1346 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1347 if (error != ENODEV)
1349 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1361 * Internal version of mmap used by exec, sys5 shared memory, and
1362 * various device drivers. Handle is either a vnode pointer, a
1363 * character device, or NULL for MAP_ANON.
1366 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1367 vm_prot_t maxprot, int flags,
1368 objtype_t handle_type, void *handle,
1372 struct thread *td = curthread;
1374 boolean_t writecounted;
1379 size = round_page(size);
1381 writecounted = FALSE;
1384 * Lookup/allocate object.
1386 switch (handle_type) {
1393 dsw = dev_refthread(cdev, &ref);
1396 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1397 dsw, &foff, &object);
1398 dev_relthread(cdev, ref);
1402 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1403 handle, &foff, &object, &writecounted);
1406 if (handle == NULL) {
1418 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1419 foff, writecounted, td);
1420 if (error != 0 && object != NULL) {
1422 * If this mapping was accounted for in the vnode's
1423 * writecount, then undo that now.
1426 vnode_pager_release_writecount(object, 0, size);
1427 vm_object_deallocate(object);
1433 * Internal version of mmap that maps a specific VM object into an
1434 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1437 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1438 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1439 boolean_t writecounted, struct thread *td)
1441 boolean_t curmap, fitit;
1442 vm_offset_t max_addr;
1443 int docow, error, findspace, rv;
1445 curmap = map == &td->td_proc->p_vmspace->vm_map;
1447 PROC_LOCK(td->td_proc);
1448 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1449 PROC_UNLOCK(td->td_proc);
1452 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1453 PROC_UNLOCK(td->td_proc);
1456 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1457 if (ptoa(pmap_wired_count(map->pmap)) + size >
1458 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1459 racct_set_force(td->td_proc, RACCT_VMEM,
1461 PROC_UNLOCK(td->td_proc);
1464 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1465 ptoa(pmap_wired_count(map->pmap)) + size);
1467 racct_set_force(td->td_proc, RACCT_VMEM,
1469 PROC_UNLOCK(td->td_proc);
1473 PROC_UNLOCK(td->td_proc);
1477 * We currently can only deal with page aligned file offsets.
1478 * The mmap() system call already enforces this by subtracting
1479 * the page offset from the file offset, but checking here
1480 * catches errors in device drivers (e.g. d_single_mmap()
1481 * callbacks) and other internal mapping requests (such as in
1484 if (foff & PAGE_MASK)
1487 if ((flags & MAP_FIXED) == 0) {
1489 *addr = round_page(*addr);
1491 if (*addr != trunc_page(*addr))
1496 if (flags & MAP_ANON) {
1497 if (object != NULL || foff != 0)
1500 } else if (flags & MAP_PREFAULT_READ)
1501 docow = MAP_PREFAULT;
1503 docow = MAP_PREFAULT_PARTIAL;
1505 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1506 docow |= MAP_COPY_ON_WRITE;
1507 if (flags & MAP_NOSYNC)
1508 docow |= MAP_DISABLE_SYNCER;
1509 if (flags & MAP_NOCORE)
1510 docow |= MAP_DISABLE_COREDUMP;
1511 /* Shared memory is also shared with children. */
1512 if (flags & MAP_SHARED)
1513 docow |= MAP_INHERIT_SHARE;
1515 docow |= MAP_VN_WRITECOUNT;
1516 if (flags & MAP_STACK) {
1519 docow |= MAP_STACK_GROWS_DOWN;
1521 if ((flags & MAP_EXCL) != 0)
1522 docow |= MAP_CHECK_EXCL;
1523 if ((flags & MAP_GUARD) != 0)
1524 docow |= MAP_CREATE_GUARD;
1527 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1528 findspace = VMFS_SUPER_SPACE;
1529 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1530 findspace = VMFS_ALIGNED_SPACE(flags >>
1531 MAP_ALIGNMENT_SHIFT);
1533 findspace = VMFS_OPTIMAL_SPACE;
1536 if ((flags & MAP_32BIT) != 0)
1537 max_addr = MAP_32BIT_MAX_ADDR;
1540 rv = vm_map_find_min(map, object, foff, addr, size,
1541 round_page((vm_offset_t)td->td_proc->p_vmspace->
1542 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1543 findspace, prot, maxprot, docow);
1545 rv = vm_map_find(map, object, foff, addr, size,
1546 max_addr, findspace, prot, maxprot, docow);
1549 rv = vm_map_fixed(map, object, foff, *addr, size,
1550 prot, maxprot, docow);
1553 if (rv == KERN_SUCCESS) {
1555 * If the process has requested that all future mappings
1556 * be wired, then heed this.
1558 if (map->flags & MAP_WIREFUTURE) {
1559 vm_map_wire(map, *addr, *addr + size,
1560 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1561 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1564 return (vm_mmap_to_errno(rv));
1568 * Translate a Mach VM return code to zero on success or the appropriate errno
1572 vm_mmap_to_errno(int rv)
1578 case KERN_INVALID_ADDRESS:
1581 case KERN_PROTECTION_FAILURE: