2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1988 University of Utah.
5 * Copyright (c) 1991, 1993
6 * The Regents of the University of California. All rights reserved.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
38 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
42 * Mapped file (mmap) interface to VM
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include "opt_hwpmc_hooks.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/capsicum.h>
54 #include <sys/kernel.h>
56 #include <sys/mutex.h>
57 #include <sys/sysproto.h>
59 #include <sys/filedesc.h>
62 #include <sys/procctl.h>
63 #include <sys/racct.h>
64 #include <sys/resource.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/fcntl.h>
72 #include <sys/mount.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/vmmeter.h>
78 #if defined(__amd64__) || defined(__i386__) /* for i386_read_exec */
79 #include <machine/md_var.h>
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
86 #include <vm/vm_param.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_pager.h>
92 #include <vm/vm_pageout.h>
93 #include <vm/vm_extern.h>
94 #include <vm/vm_page.h>
95 #include <vm/vnode_pager.h>
98 #include <sys/pmckern.h>
102 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
103 "Do not apply RLIMIT_MEMLOCK on mlockall");
104 static int mincore_mapped = 1;
105 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
106 "mincore reports mappings, not residency");
107 static int imply_prot_max = 0;
108 SYSCTL_INT(_vm, OID_AUTO, imply_prot_max, CTLFLAG_RWTUN, &imply_prot_max, 0,
109 "Imply maximum page permissions in mmap() when none are specified");
112 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
115 #ifndef _SYS_SYSPROTO_H_
122 sys_sbrk(struct thread *td, struct sbrk_args *uap)
124 /* Not yet implemented */
128 #ifndef _SYS_SYSPROTO_H_
135 sys_sstk(struct thread *td, struct sstk_args *uap)
137 /* Not yet implemented */
141 #if defined(COMPAT_43)
143 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
146 td->td_retval[0] = PAGE_SIZE;
149 #endif /* COMPAT_43 */
153 * Memory Map (mmap) system call. Note that the file offset
154 * and address are allowed to be NOT page aligned, though if
155 * the MAP_FIXED flag it set, both must have the same remainder
156 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
157 * page-aligned, the actual mapping starts at trunc_page(addr)
158 * and the return value is adjusted up by the page offset.
160 * Generally speaking, only character devices which are themselves
161 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
162 * there would be no cache coherency between a descriptor and a VM mapping
163 * both to the same character device.
165 #ifndef _SYS_SYSPROTO_H_
178 sys_mmap(struct thread *td, struct mmap_args *uap)
181 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
182 uap->flags, uap->fd, uap->pos));
186 kern_mmap_maxprot(struct proc *p, int prot)
189 if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
190 (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
192 if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
199 kern_mmap(struct thread *td, uintptr_t addr0, size_t len, int prot, int flags,
202 struct mmap_req mr = {
211 return (kern_mmap_req(td, &mr));
215 kern_mmap_req(struct thread *td, const struct mmap_req *mrp)
222 vm_size_t len, pageoff, size;
223 vm_prot_t cap_maxprot;
224 int align, error, fd, flags, max_prot, prot;
226 mmap_check_fp_fn check_fp_fn;
231 flags = mrp->mr_flags;
234 check_fp_fn = mrp->mr_check_fp_fn;
236 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
238 max_prot = PROT_MAX_EXTRACT(prot);
239 prot = PROT_EXTRACT(prot);
240 if (max_prot != 0 && (max_prot & prot) != prot)
246 * Always honor PROT_MAX if set. If not, default to all
247 * permissions unless we're implying maximum permissions.
250 max_prot = kern_mmap_maxprot(p, prot);
257 * Ignore old flags that used to be defined but did not do anything.
259 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
262 * Enforce the constraints.
263 * Mapping of length 0 is only allowed for old binaries.
264 * Anonymous mapping shall specify -1 as filedescriptor and
265 * zero position for new code. Be nice to ancient a.out
266 * binaries and correct pos for anonymous mapping, since old
267 * ld.so sometimes issues anonymous map requests with non-zero
270 if (!SV_CURPROC_FLAG(SV_AOUT)) {
271 if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
272 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
275 if ((flags & MAP_ANON) != 0)
279 if (flags & MAP_STACK) {
281 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
286 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
287 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
288 MAP_PREFAULT_READ | MAP_GUARD |
292 MAP_ALIGNMENT_MASK)) != 0)
294 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
296 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
298 if (prot != PROT_NONE &&
299 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
301 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
302 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
306 MAP_ALIGNMENT_MASK)) != 0))
310 * Align the file position to a page boundary,
311 * and save its page offset component.
313 pageoff = (pos & PAGE_MASK);
316 /* Compute size from len by rounding (on both ends). */
317 size = len + pageoff; /* low end... */
318 size = round_page(size); /* hi end */
319 /* Check for rounding up to zero. */
323 /* Ensure alignment is at least a page and fits in a pointer. */
324 align = flags & MAP_ALIGNMENT_MASK;
325 if (align != 0 && align != MAP_ALIGNED_SUPER &&
326 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
327 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
331 * Check for illegal addresses. Watch out for address wrap... Note
332 * that VM_*_ADDRESS are not constants due to casts (argh).
334 if (flags & MAP_FIXED) {
336 * The specified address must have the same remainder
337 * as the file offset taken modulo PAGE_SIZE, so it
338 * should be aligned after adjustment by pageoff.
341 if (addr & PAGE_MASK)
344 /* Address range must be all in user VM space. */
345 if (addr < vm_map_min(&vms->vm_map) ||
346 addr + size > vm_map_max(&vms->vm_map))
348 if (addr + size < addr)
351 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
353 } else if (flags & MAP_32BIT) {
355 * For MAP_32BIT, override the hint if it is too high and
356 * do not bother moving the mapping past the heap (since
357 * the heap is usually above 2GB).
359 if (addr + size > MAP_32BIT_MAX_ADDR)
364 * XXX for non-fixed mappings where no hint is provided or
365 * the hint would fall in the potential heap space,
366 * place it after the end of the largest possible heap.
368 * There should really be a pmap call to determine a reasonable
372 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
373 addr < round_page((vm_offset_t)vms->vm_daddr +
374 lim_max(td, RLIMIT_DATA))))
375 addr = round_page((vm_offset_t)vms->vm_daddr +
376 lim_max(td, RLIMIT_DATA));
380 * Return success without mapping anything for old
381 * binaries that request a page-aligned mapping of
382 * length 0. For modern binaries, this function
383 * returns an error earlier.
386 } else if ((flags & MAP_GUARD) != 0) {
387 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
388 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
389 } else if ((flags & MAP_ANON) != 0) {
391 * Mapping blank space is trivial.
393 * This relies on VM_PROT_* matching PROT_*.
395 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
396 max_prot, flags, NULL, pos, FALSE, td);
399 * Mapping file, get fp for validation and don't let the
400 * descriptor disappear on us if we block. Check capability
401 * rights, but also return the maximum rights to be combined
402 * with maxprot later.
404 cap_rights_init_one(&rights, CAP_MMAP);
405 if (prot & PROT_READ)
406 cap_rights_set_one(&rights, CAP_MMAP_R);
407 if ((flags & MAP_SHARED) != 0) {
408 if (prot & PROT_WRITE)
409 cap_rights_set_one(&rights, CAP_MMAP_W);
411 if (prot & PROT_EXEC)
412 cap_rights_set_one(&rights, CAP_MMAP_X);
413 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
416 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
417 p->p_osrel >= P_OSREL_MAP_FSTRICT) {
421 if (check_fp_fn != NULL) {
422 error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
427 /* This relies on VM_PROT_* matching PROT_*. */
428 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
429 max_prot & cap_maxprot, flags, pos, td);
433 td->td_retval[0] = (register_t) (addr + pageoff);
441 #if defined(COMPAT_FREEBSD6)
443 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
446 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
447 uap->flags, uap->fd, uap->pos));
452 #ifndef _SYS_SYSPROTO_H_
463 ommap(struct thread *td, struct ommap_args *uap)
465 static const char cvtbsdprot[8] = {
469 PROT_EXEC | PROT_WRITE,
471 PROT_EXEC | PROT_READ,
472 PROT_WRITE | PROT_READ,
473 PROT_EXEC | PROT_WRITE | PROT_READ,
477 #define OMAP_ANON 0x0002
478 #define OMAP_COPY 0x0020
479 #define OMAP_SHARED 0x0010
480 #define OMAP_FIXED 0x0100
482 prot = cvtbsdprot[uap->prot & 0x7];
483 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
484 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
489 if (uap->flags & OMAP_ANON)
491 if (uap->flags & OMAP_COPY)
493 if (uap->flags & OMAP_SHARED)
496 flags |= MAP_PRIVATE;
497 if (uap->flags & OMAP_FIXED)
499 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
502 #endif /* COMPAT_43 */
505 #ifndef _SYS_SYSPROTO_H_
513 sys_msync(struct thread *td, struct msync_args *uap)
516 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
520 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
528 pageoff = (addr & PAGE_MASK);
531 size = (vm_size_t) round_page(size);
532 if (addr + size < addr)
535 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
538 map = &td->td_proc->p_vmspace->vm_map;
541 * Clean the pages and interpret the return value.
543 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
544 (flags & MS_INVALIDATE) != 0);
548 case KERN_INVALID_ADDRESS:
550 case KERN_INVALID_ARGUMENT:
559 #ifndef _SYS_SYSPROTO_H_
566 sys_munmap(struct thread *td, struct munmap_args *uap)
569 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
573 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
576 struct pmckern_map_out pkm;
577 vm_map_entry_t entry;
588 pageoff = (addr & PAGE_MASK);
591 size = (vm_size_t) round_page(size);
592 if (addr + size < addr)
596 * Check for illegal addresses. Watch out for address wrap...
598 map = &td->td_proc->p_vmspace->vm_map;
599 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
604 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
607 * Inform hwpmc if the address range being unmapped contains
608 * an executable region.
610 pkm.pm_address = (uintptr_t) NULL;
611 if (vm_map_lookup_entry(map, addr, &entry)) {
612 for (; entry->start < addr + size;
613 entry = vm_map_entry_succ(entry)) {
614 if (vm_map_check_protection(map, entry->start,
615 entry->end, VM_PROT_EXECUTE) == TRUE) {
616 pkm.pm_address = (uintptr_t) addr;
617 pkm.pm_size = (size_t) size;
624 vm_map_delete(map, addr, addr + size);
627 if (__predict_false(pmc_handled)) {
628 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
629 vm_map_lock_downgrade(map);
630 if (pkm.pm_address != (uintptr_t) NULL)
631 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
632 vm_map_unlock_read(map);
637 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
641 #ifndef _SYS_SYSPROTO_H_
642 struct mprotect_args {
649 sys_mprotect(struct thread *td, struct mprotect_args *uap)
652 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
656 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
660 int vm_error, max_prot;
663 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
665 max_prot = PROT_MAX_EXTRACT(prot);
666 prot = PROT_EXTRACT(prot);
667 pageoff = (addr & PAGE_MASK);
670 size = (vm_size_t) round_page(size);
671 #ifdef COMPAT_FREEBSD32
672 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
673 if (((addr + size) & 0xffffffff) < addr)
677 if (addr + size < addr)
680 vm_error = KERN_SUCCESS;
682 if ((max_prot & prot) != prot)
684 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
685 addr, addr + size, max_prot, TRUE);
687 if (vm_error == KERN_SUCCESS)
688 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
689 addr, addr + size, prot, FALSE);
694 case KERN_PROTECTION_FAILURE:
696 case KERN_RESOURCE_SHORTAGE:
702 #ifndef _SYS_SYSPROTO_H_
703 struct minherit_args {
710 sys_minherit(struct thread *td, struct minherit_args *uap)
713 vm_size_t size, pageoff;
714 vm_inherit_t inherit;
716 addr = (vm_offset_t)uap->addr;
718 inherit = uap->inherit;
720 pageoff = (addr & PAGE_MASK);
723 size = (vm_size_t) round_page(size);
724 if (addr + size < addr)
727 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
728 addr + size, inherit)) {
731 case KERN_PROTECTION_FAILURE:
737 #ifndef _SYS_SYSPROTO_H_
738 struct madvise_args {
746 sys_madvise(struct thread *td, struct madvise_args *uap)
749 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
753 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
756 vm_offset_t addr, end, start;
760 * Check for our special case, advising the swap pager we are
763 if (behav == MADV_PROTECT) {
765 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
766 PROC_SPROTECT, &flags));
770 * Check for illegal addresses. Watch out for address wrap... Note
771 * that VM_*_ADDRESS are not constants due to casts (argh).
773 map = &td->td_proc->p_vmspace->vm_map;
775 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
777 if ((addr + len) < addr)
781 * Since this routine is only advisory, we default to conservative
784 start = trunc_page(addr);
785 end = round_page(addr + len);
788 * vm_map_madvise() checks for illegal values of behav.
790 return (vm_map_madvise(map, start, end, behav));
793 #ifndef _SYS_SYSPROTO_H_
794 struct mincore_args {
802 sys_mincore(struct thread *td, struct mincore_args *uap)
805 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
809 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
813 vm_map_entry_t current, entry;
815 vm_offset_t addr, cend, end, first_addr;
819 int error, lastvecindex, mincoreinfo, vecindex;
820 unsigned int timestamp;
823 * Make sure that the addresses presented are valid for user
826 first_addr = addr = trunc_page(addr0);
827 end = round_page(addr0 + len);
828 map = &td->td_proc->p_vmspace->vm_map;
829 if (end > vm_map_max(map) || end < addr)
832 pmap = vmspace_pmap(td->td_proc->p_vmspace);
834 vm_map_lock_read(map);
836 timestamp = map->timestamp;
838 if (!vm_map_lookup_entry(map, addr, &entry)) {
839 vm_map_unlock_read(map);
844 * Do this on a map entry basis so that if the pages are not
845 * in the current processes address space, we can easily look
846 * up the pages elsewhere.
849 while (entry->start < end) {
852 * check for contiguity
855 entry = vm_map_entry_succ(current);
856 if (current->end < end &&
857 entry->start > current->end) {
858 vm_map_unlock_read(map);
863 * ignore submaps (for now) or null objects
865 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
866 current->object.vm_object == NULL)
870 * limit this scan to the current map entry and the
871 * limits for the mincore call
873 if (addr < current->start)
874 addr = current->start;
879 for (; addr < cend; addr += PAGE_SIZE) {
881 * Check pmap first, it is likely faster, also
882 * it can provide info as to whether we are the
883 * one referencing or modifying the page.
889 mincoreinfo = pmap_mincore(pmap, addr, &pa);
890 if (mincore_mapped) {
892 * We only care about this pmap's
893 * mapping of the page, if any.
896 } else if (pa != 0) {
898 * The page is mapped by this process but not
899 * both accessed and modified. It is also
900 * managed. Acquire the object lock so that
901 * other mappings might be examined. The page's
902 * identity may change at any point before its
903 * object lock is acquired, so re-validate if
906 m = PHYS_TO_VM_PAGE(pa);
907 while (object == NULL || m->object != object) {
909 VM_OBJECT_WUNLOCK(object);
910 object = atomic_load_ptr(&m->object);
913 VM_OBJECT_WLOCK(object);
915 if (pa != pmap_extract(pmap, addr))
917 KASSERT(vm_page_all_valid(m),
918 ("mincore: page %p is mapped but invalid",
920 } else if (mincoreinfo == 0) {
922 * The page is not mapped by this process. If
923 * the object implements managed pages, then
924 * determine if the page is resident so that
925 * the mappings might be examined.
927 if (current->object.vm_object != object) {
929 VM_OBJECT_WUNLOCK(object);
930 object = current->object.vm_object;
931 VM_OBJECT_WLOCK(object);
933 if (object->type == OBJT_DEFAULT ||
934 object->type == OBJT_SWAP ||
935 object->type == OBJT_VNODE) {
936 pindex = OFF_TO_IDX(current->offset +
937 (addr - current->start));
938 m = vm_page_lookup(object, pindex);
939 if (m != NULL && vm_page_none_valid(m))
942 mincoreinfo = MINCORE_INCORE;
946 VM_OBJECT_ASSERT_WLOCKED(m->object);
948 /* Examine other mappings of the page. */
949 if (m->dirty == 0 && pmap_is_modified(m))
952 mincoreinfo |= MINCORE_MODIFIED_OTHER;
955 * The first test for PGA_REFERENCED is an
956 * optimization. The second test is
957 * required because a concurrent pmap
958 * operation could clear the last reference
959 * and set PGA_REFERENCED before the call to
960 * pmap_is_referenced().
962 if ((m->a.flags & PGA_REFERENCED) != 0 ||
963 pmap_is_referenced(m) ||
964 (m->a.flags & PGA_REFERENCED) != 0)
965 mincoreinfo |= MINCORE_REFERENCED_OTHER;
968 VM_OBJECT_WUNLOCK(object);
971 * subyte may page fault. In case it needs to modify
972 * the map, we release the lock.
974 vm_map_unlock_read(map);
977 * calculate index into user supplied byte vector
979 vecindex = atop(addr - first_addr);
982 * If we have skipped map entries, we need to make sure that
983 * the byte vector is zeroed for those skipped entries.
985 while ((lastvecindex + 1) < vecindex) {
987 error = subyte(vec + lastvecindex, 0);
995 * Pass the page information to the user
997 error = subyte(vec + vecindex, mincoreinfo);
1004 * If the map has changed, due to the subyte, the previous
1005 * output may be invalid.
1007 vm_map_lock_read(map);
1008 if (timestamp != map->timestamp)
1011 lastvecindex = vecindex;
1016 * subyte may page fault. In case it needs to modify
1017 * the map, we release the lock.
1019 vm_map_unlock_read(map);
1022 * Zero the last entries in the byte vector.
1024 vecindex = atop(end - first_addr);
1025 while ((lastvecindex + 1) < vecindex) {
1027 error = subyte(vec + lastvecindex, 0);
1035 * If the map has changed, due to the subyte, the previous
1036 * output may be invalid.
1038 vm_map_lock_read(map);
1039 if (timestamp != map->timestamp)
1041 vm_map_unlock_read(map);
1046 #ifndef _SYS_SYSPROTO_H_
1053 sys_mlock(struct thread *td, struct mlock_args *uap)
1056 return (kern_mlock(td->td_proc, td->td_ucred,
1057 __DECONST(uintptr_t, uap->addr), uap->len));
1061 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1063 vm_offset_t addr, end, last, start;
1064 vm_size_t npages, size;
1066 unsigned long nsize;
1069 error = priv_check_cred(cred, PRIV_VM_MLOCK);
1075 start = trunc_page(addr);
1076 end = round_page(last);
1077 if (last < addr || end < addr)
1079 npages = atop(end - start);
1080 if (npages > vm_page_max_user_wired)
1082 map = &proc->p_vmspace->vm_map;
1084 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1085 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1093 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1099 error = vm_map_wire(map, start, end,
1100 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1102 if (racct_enable && error != KERN_SUCCESS) {
1104 racct_set(proc, RACCT_MEMLOCK,
1105 ptoa(pmap_wired_count(map->pmap)));
1109 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1112 #ifndef _SYS_SYSPROTO_H_
1113 struct mlockall_args {
1119 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1124 map = &td->td_proc->p_vmspace->vm_map;
1125 error = priv_check(td, PRIV_VM_MLOCK);
1129 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1133 * If wiring all pages in the process would cause it to exceed
1134 * a hard resource limit, return ENOMEM.
1136 if (!old_mlock && uap->how & MCL_CURRENT) {
1137 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1142 PROC_LOCK(td->td_proc);
1143 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1144 PROC_UNLOCK(td->td_proc);
1150 if (uap->how & MCL_FUTURE) {
1152 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1157 if (uap->how & MCL_CURRENT) {
1159 * P1003.1-2001 mandates that all currently mapped pages
1160 * will be memory resident and locked (wired) upon return
1161 * from mlockall(). vm_map_wire() will wire pages, by
1162 * calling vm_fault_wire() for each page in the region.
1164 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1165 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1166 if (error == KERN_SUCCESS)
1168 else if (error == KERN_RESOURCE_SHORTAGE)
1174 if (racct_enable && error != KERN_SUCCESS) {
1175 PROC_LOCK(td->td_proc);
1176 racct_set(td->td_proc, RACCT_MEMLOCK,
1177 ptoa(pmap_wired_count(map->pmap)));
1178 PROC_UNLOCK(td->td_proc);
1185 #ifndef _SYS_SYSPROTO_H_
1186 struct munlockall_args {
1192 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1197 map = &td->td_proc->p_vmspace->vm_map;
1198 error = priv_check(td, PRIV_VM_MUNLOCK);
1202 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1204 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1207 /* Forcibly unwire all pages. */
1208 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1209 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1211 if (racct_enable && error == KERN_SUCCESS) {
1212 PROC_LOCK(td->td_proc);
1213 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1214 PROC_UNLOCK(td->td_proc);
1221 #ifndef _SYS_SYSPROTO_H_
1222 struct munlock_args {
1228 sys_munlock(struct thread *td, struct munlock_args *uap)
1231 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1235 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1237 vm_offset_t addr, end, last, start;
1243 error = priv_check(td, PRIV_VM_MUNLOCK);
1248 start = trunc_page(addr);
1249 end = round_page(last);
1250 if (last < addr || end < addr)
1252 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1253 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1255 if (racct_enable && error == KERN_SUCCESS) {
1256 PROC_LOCK(td->td_proc);
1257 map = &td->td_proc->p_vmspace->vm_map;
1258 racct_set(td->td_proc, RACCT_MEMLOCK,
1259 ptoa(pmap_wired_count(map->pmap)));
1260 PROC_UNLOCK(td->td_proc);
1263 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1269 * Helper function for vm_mmap. Perform sanity check specific for mmap
1270 * operations on vnodes.
1273 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1274 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1275 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1276 boolean_t *writecounted)
1285 cred = td->td_ucred;
1286 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1287 (*flagsp & MAP_SHARED) != 0;
1288 if ((error = vget(vp, LK_SHARED, td)) != 0)
1290 AUDIT_ARG_VNODE1(vp);
1294 if (vp->v_type == VREG) {
1296 * Get the proper underlying object
1302 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1304 vp = (struct vnode *)obj->handle;
1306 * Bypass filesystems obey the mpsafety of the
1307 * underlying fs. Tmpfs never bypasses.
1309 error = vget(vp, LK_SHARED, td);
1314 *writecounted = TRUE;
1315 vm_pager_update_writecount(obj, 0, objsize);
1321 if ((error = VOP_GETATTR(vp, &va, cred)))
1324 /* This relies on VM_PROT_* matching PROT_*. */
1325 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1329 if ((flags & MAP_SHARED) != 0) {
1330 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1331 if (prot & VM_PROT_WRITE) {
1335 *maxprotp &= ~VM_PROT_WRITE;
1339 * If it is a regular file without any references
1340 * we do not need to sync it.
1341 * Adjust object size to be the size of actual file.
1343 objsize = round_page(va.va_size);
1344 if (va.va_nlink == 0)
1345 flags |= MAP_NOSYNC;
1346 if (obj->type == OBJT_VNODE) {
1347 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1354 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1355 ("wrong object type"));
1356 vm_object_reference(obj);
1357 #if VM_NRESERVLEVEL > 0
1358 if ((obj->flags & OBJ_COLORED) == 0) {
1359 VM_OBJECT_WLOCK(obj);
1360 vm_object_color(obj, 0);
1361 VM_OBJECT_WUNLOCK(obj);
1371 if (error != 0 && *writecounted) {
1372 *writecounted = FALSE;
1373 vm_pager_update_writecount(obj, objsize, 0);
1382 * Helper function for vm_mmap. Perform sanity check specific for mmap
1383 * operations on cdevs.
1386 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1387 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1388 vm_ooffset_t *foff, vm_object_t *objp)
1395 if (dsw->d_flags & D_MMAP_ANON) {
1398 *maxprotp = VM_PROT_ALL;
1399 *flagsp |= MAP_ANON;
1403 * cdevs do not provide private mappings of any kind.
1405 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1406 (prot & VM_PROT_WRITE) != 0)
1408 if (flags & (MAP_PRIVATE|MAP_COPY))
1411 * Force device mappings to be shared.
1413 flags |= MAP_SHARED;
1415 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1420 * First, try d_mmap_single(). If that is not implemented
1421 * (returns ENODEV), fall back to using the device pager.
1422 * Note that d_mmap_single() must return a reference to the
1423 * object (it needs to bump the reference count of the object
1424 * it returns somehow).
1426 * XXX assumes VM_PROT_* == PROT_*
1428 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1429 if (error != ENODEV)
1431 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1443 * Internal version of mmap used by exec, sys5 shared memory, and
1444 * various device drivers. Handle is either a vnode pointer, a
1445 * character device, or NULL for MAP_ANON.
1448 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1449 vm_prot_t maxprot, int flags,
1450 objtype_t handle_type, void *handle,
1454 struct thread *td = curthread;
1456 boolean_t writecounted;
1461 size = round_page(size);
1463 writecounted = FALSE;
1466 * Lookup/allocate object.
1468 switch (handle_type) {
1475 dsw = dev_refthread(cdev, &ref);
1478 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1479 dsw, &foff, &object);
1480 dev_relthread(cdev, ref);
1484 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1485 handle, &foff, &object, &writecounted);
1488 if (handle == NULL) {
1500 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1501 foff, writecounted, td);
1502 if (error != 0 && object != NULL) {
1504 * If this mapping was accounted for in the vnode's
1505 * writecount, then undo that now.
1508 vm_pager_release_writecount(object, 0, size);
1509 vm_object_deallocate(object);
1515 * Internal version of mmap that maps a specific VM object into an
1516 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1519 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1520 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1521 boolean_t writecounted, struct thread *td)
1523 boolean_t curmap, fitit;
1524 vm_offset_t max_addr;
1525 int docow, error, findspace, rv;
1527 curmap = map == &td->td_proc->p_vmspace->vm_map;
1529 RACCT_PROC_LOCK(td->td_proc);
1530 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1531 RACCT_PROC_UNLOCK(td->td_proc);
1534 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1535 RACCT_PROC_UNLOCK(td->td_proc);
1538 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1539 if (ptoa(pmap_wired_count(map->pmap)) + size >
1540 lim_cur(td, RLIMIT_MEMLOCK)) {
1541 racct_set_force(td->td_proc, RACCT_VMEM,
1543 RACCT_PROC_UNLOCK(td->td_proc);
1546 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1547 ptoa(pmap_wired_count(map->pmap)) + size);
1549 racct_set_force(td->td_proc, RACCT_VMEM,
1551 RACCT_PROC_UNLOCK(td->td_proc);
1555 RACCT_PROC_UNLOCK(td->td_proc);
1559 * We currently can only deal with page aligned file offsets.
1560 * The mmap() system call already enforces this by subtracting
1561 * the page offset from the file offset, but checking here
1562 * catches errors in device drivers (e.g. d_single_mmap()
1563 * callbacks) and other internal mapping requests (such as in
1566 if (foff & PAGE_MASK)
1569 if ((flags & MAP_FIXED) == 0) {
1571 *addr = round_page(*addr);
1573 if (*addr != trunc_page(*addr))
1578 if (flags & MAP_ANON) {
1579 if (object != NULL || foff != 0)
1582 } else if (flags & MAP_PREFAULT_READ)
1583 docow = MAP_PREFAULT;
1585 docow = MAP_PREFAULT_PARTIAL;
1587 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1588 docow |= MAP_COPY_ON_WRITE;
1589 if (flags & MAP_NOSYNC)
1590 docow |= MAP_DISABLE_SYNCER;
1591 if (flags & MAP_NOCORE)
1592 docow |= MAP_DISABLE_COREDUMP;
1593 /* Shared memory is also shared with children. */
1594 if (flags & MAP_SHARED)
1595 docow |= MAP_INHERIT_SHARE;
1597 docow |= MAP_WRITECOUNT;
1598 if (flags & MAP_STACK) {
1601 docow |= MAP_STACK_GROWS_DOWN;
1603 if ((flags & MAP_EXCL) != 0)
1604 docow |= MAP_CHECK_EXCL;
1605 if ((flags & MAP_GUARD) != 0)
1606 docow |= MAP_CREATE_GUARD;
1609 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1610 findspace = VMFS_SUPER_SPACE;
1611 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1612 findspace = VMFS_ALIGNED_SPACE(flags >>
1613 MAP_ALIGNMENT_SHIFT);
1615 findspace = VMFS_OPTIMAL_SPACE;
1618 if ((flags & MAP_32BIT) != 0)
1619 max_addr = MAP_32BIT_MAX_ADDR;
1622 rv = vm_map_find_min(map, object, foff, addr, size,
1623 round_page((vm_offset_t)td->td_proc->p_vmspace->
1624 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1625 findspace, prot, maxprot, docow);
1627 rv = vm_map_find(map, object, foff, addr, size,
1628 max_addr, findspace, prot, maxprot, docow);
1631 rv = vm_map_fixed(map, object, foff, *addr, size,
1632 prot, maxprot, docow);
1635 if (rv == KERN_SUCCESS) {
1637 * If the process has requested that all future mappings
1638 * be wired, then heed this.
1640 if ((map->flags & MAP_WIREFUTURE) != 0) {
1642 if ((map->flags & MAP_WIREFUTURE) != 0)
1643 (void)vm_map_wire_locked(map, *addr,
1644 *addr + size, VM_MAP_WIRE_USER |
1645 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1646 VM_MAP_WIRE_NOHOLES));
1650 return (vm_mmap_to_errno(rv));
1654 * Translate a Mach VM return code to zero on success or the appropriate errno
1658 vm_mmap_to_errno(int rv)
1664 case KERN_INVALID_ADDRESS:
1667 case KERN_PROTECTION_FAILURE: