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 protections 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 (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
348 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
350 } else if (flags & MAP_32BIT) {
352 * For MAP_32BIT, override the hint if it is too high and
353 * do not bother moving the mapping past the heap (since
354 * the heap is usually above 2GB).
356 if (addr + size > MAP_32BIT_MAX_ADDR)
361 * XXX for non-fixed mappings where no hint is provided or
362 * the hint would fall in the potential heap space,
363 * place it after the end of the largest possible heap.
365 * There should really be a pmap call to determine a reasonable
369 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
370 addr < round_page((vm_offset_t)vms->vm_daddr +
371 lim_max(td, RLIMIT_DATA))))
372 addr = round_page((vm_offset_t)vms->vm_daddr +
373 lim_max(td, RLIMIT_DATA));
377 * Return success without mapping anything for old
378 * binaries that request a page-aligned mapping of
379 * length 0. For modern binaries, this function
380 * returns an error earlier.
383 } else if ((flags & MAP_GUARD) != 0) {
384 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
385 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
386 } else if ((flags & MAP_ANON) != 0) {
388 * Mapping blank space is trivial.
390 * This relies on VM_PROT_* matching PROT_*.
392 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
393 max_prot, flags, NULL, pos, FALSE, td);
396 * Mapping file, get fp for validation and don't let the
397 * descriptor disappear on us if we block. Check capability
398 * rights, but also return the maximum rights to be combined
399 * with maxprot later.
401 cap_rights_init_one(&rights, CAP_MMAP);
402 if (prot & PROT_READ)
403 cap_rights_set_one(&rights, CAP_MMAP_R);
404 if ((flags & MAP_SHARED) != 0) {
405 if (prot & PROT_WRITE)
406 cap_rights_set_one(&rights, CAP_MMAP_W);
408 if (prot & PROT_EXEC)
409 cap_rights_set_one(&rights, CAP_MMAP_X);
410 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
413 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
414 p->p_osrel >= P_OSREL_MAP_FSTRICT) {
418 if (check_fp_fn != NULL) {
419 error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
424 /* This relies on VM_PROT_* matching PROT_*. */
425 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
426 max_prot & cap_maxprot, flags, pos, td);
430 td->td_retval[0] = (register_t) (addr + pageoff);
438 #if defined(COMPAT_FREEBSD6)
440 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
443 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
444 uap->flags, uap->fd, uap->pos));
449 #ifndef _SYS_SYSPROTO_H_
460 ommap(struct thread *td, struct ommap_args *uap)
462 static const char cvtbsdprot[8] = {
466 PROT_EXEC | PROT_WRITE,
468 PROT_EXEC | PROT_READ,
469 PROT_WRITE | PROT_READ,
470 PROT_EXEC | PROT_WRITE | PROT_READ,
474 #define OMAP_ANON 0x0002
475 #define OMAP_COPY 0x0020
476 #define OMAP_SHARED 0x0010
477 #define OMAP_FIXED 0x0100
479 prot = cvtbsdprot[uap->prot & 0x7];
480 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
481 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
486 if (uap->flags & OMAP_ANON)
488 if (uap->flags & OMAP_COPY)
490 if (uap->flags & OMAP_SHARED)
493 flags |= MAP_PRIVATE;
494 if (uap->flags & OMAP_FIXED)
496 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
499 #endif /* COMPAT_43 */
502 #ifndef _SYS_SYSPROTO_H_
510 sys_msync(struct thread *td, struct msync_args *uap)
513 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
517 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
525 pageoff = (addr & PAGE_MASK);
528 size = (vm_size_t) round_page(size);
529 if (addr + size < addr)
532 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
535 map = &td->td_proc->p_vmspace->vm_map;
538 * Clean the pages and interpret the return value.
540 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
541 (flags & MS_INVALIDATE) != 0);
545 case KERN_INVALID_ADDRESS:
547 case KERN_INVALID_ARGUMENT:
556 #ifndef _SYS_SYSPROTO_H_
563 sys_munmap(struct thread *td, struct munmap_args *uap)
566 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
570 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
573 struct pmckern_map_out pkm;
574 vm_map_entry_t entry;
577 vm_offset_t addr, end;
585 pageoff = (addr & PAGE_MASK);
588 size = (vm_size_t) round_page(size);
590 map = &td->td_proc->p_vmspace->vm_map;
591 if (!vm_map_range_valid(map, addr, end))
597 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
600 * Inform hwpmc if the address range being unmapped contains
601 * an executable region.
603 pkm.pm_address = (uintptr_t) NULL;
604 if (vm_map_lookup_entry(map, addr, &entry)) {
605 for (; entry->start < end;
606 entry = vm_map_entry_succ(entry)) {
607 if (vm_map_check_protection(map, entry->start,
608 entry->end, VM_PROT_EXECUTE) == TRUE) {
609 pkm.pm_address = (uintptr_t) addr;
610 pkm.pm_size = (size_t) size;
617 vm_map_delete(map, addr, end);
620 if (__predict_false(pmc_handled)) {
621 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
622 vm_map_lock_downgrade(map);
623 if (pkm.pm_address != (uintptr_t) NULL)
624 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
625 vm_map_unlock_read(map);
630 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
634 #ifndef _SYS_SYSPROTO_H_
635 struct mprotect_args {
642 sys_mprotect(struct thread *td, struct mprotect_args *uap)
645 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
649 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
653 int vm_error, max_prot;
656 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
658 max_prot = PROT_MAX_EXTRACT(prot);
659 prot = PROT_EXTRACT(prot);
660 pageoff = (addr & PAGE_MASK);
663 size = (vm_size_t) round_page(size);
664 #ifdef COMPAT_FREEBSD32
665 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
666 if (((addr + size) & 0xffffffff) < addr)
670 if (addr + size < addr)
673 vm_error = KERN_SUCCESS;
675 if ((max_prot & prot) != prot)
677 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
678 addr, addr + size, max_prot, TRUE);
680 if (vm_error == KERN_SUCCESS)
681 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
682 addr, addr + size, prot, FALSE);
687 case KERN_PROTECTION_FAILURE:
689 case KERN_RESOURCE_SHORTAGE:
695 #ifndef _SYS_SYSPROTO_H_
696 struct minherit_args {
703 sys_minherit(struct thread *td, struct minherit_args *uap)
706 return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
711 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
714 vm_size_t size, pageoff;
715 vm_inherit_t inherit;
717 addr = (vm_offset_t)addr0;
721 pageoff = (addr & PAGE_MASK);
724 size = (vm_size_t) round_page(size);
725 if (addr + size < addr)
728 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
729 addr + size, inherit)) {
732 case KERN_PROTECTION_FAILURE:
738 #ifndef _SYS_SYSPROTO_H_
739 struct madvise_args {
747 sys_madvise(struct thread *td, struct madvise_args *uap)
750 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
754 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
757 vm_offset_t addr, end, start;
761 * Check for our special case, advising the swap pager we are
764 if (behav == MADV_PROTECT) {
766 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
767 PROC_SPROTECT, &flags));
771 * Check for illegal addresses. Watch out for address wrap... Note
772 * that VM_*_ADDRESS are not constants due to casts (argh).
774 map = &td->td_proc->p_vmspace->vm_map;
776 if (!vm_map_range_valid(map, addr, addr + len))
780 * Since this routine is only advisory, we default to conservative
783 start = trunc_page(addr);
784 end = round_page(addr + len);
787 * vm_map_madvise() checks for illegal values of behav.
789 return (vm_map_madvise(map, start, end, behav));
792 #ifndef _SYS_SYSPROTO_H_
793 struct mincore_args {
801 sys_mincore(struct thread *td, struct mincore_args *uap)
804 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
808 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
812 vm_map_entry_t current, entry;
814 vm_offset_t addr, cend, end, first_addr;
818 int error, lastvecindex, mincoreinfo, vecindex;
819 unsigned int timestamp;
822 * Make sure that the addresses presented are valid for user
825 first_addr = addr = trunc_page(addr0);
826 end = round_page(addr0 + len);
827 map = &td->td_proc->p_vmspace->vm_map;
828 if (end > vm_map_max(map) || end < addr)
831 pmap = vmspace_pmap(td->td_proc->p_vmspace);
833 vm_map_lock_read(map);
835 timestamp = map->timestamp;
837 if (!vm_map_lookup_entry(map, addr, &entry)) {
838 vm_map_unlock_read(map);
843 * Do this on a map entry basis so that if the pages are not
844 * in the current processes address space, we can easily look
845 * up the pages elsewhere.
848 while (entry->start < end) {
851 * check for contiguity
854 entry = vm_map_entry_succ(current);
855 if (current->end < end &&
856 entry->start > current->end) {
857 vm_map_unlock_read(map);
862 * ignore submaps (for now) or null objects
864 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
865 current->object.vm_object == NULL)
869 * limit this scan to the current map entry and the
870 * limits for the mincore call
872 if (addr < current->start)
873 addr = current->start;
878 for (; addr < cend; addr += PAGE_SIZE) {
880 * Check pmap first, it is likely faster, also
881 * it can provide info as to whether we are the
882 * one referencing or modifying the page.
888 mincoreinfo = pmap_mincore(pmap, addr, &pa);
889 if (mincore_mapped) {
891 * We only care about this pmap's
892 * mapping of the page, if any.
895 } else if (pa != 0) {
897 * The page is mapped by this process but not
898 * both accessed and modified. It is also
899 * managed. Acquire the object lock so that
900 * other mappings might be examined. The page's
901 * identity may change at any point before its
902 * object lock is acquired, so re-validate if
905 m = PHYS_TO_VM_PAGE(pa);
906 while (object == NULL || m->object != object) {
908 VM_OBJECT_WUNLOCK(object);
909 object = atomic_load_ptr(&m->object);
912 VM_OBJECT_WLOCK(object);
914 if (pa != pmap_extract(pmap, addr))
916 KASSERT(vm_page_all_valid(m),
917 ("mincore: page %p is mapped but invalid",
919 } else if (mincoreinfo == 0) {
921 * The page is not mapped by this process. If
922 * the object implements managed pages, then
923 * determine if the page is resident so that
924 * the mappings might be examined.
926 if (current->object.vm_object != object) {
928 VM_OBJECT_WUNLOCK(object);
929 object = current->object.vm_object;
930 VM_OBJECT_WLOCK(object);
932 if (object->type == OBJT_DEFAULT ||
933 object->type == OBJT_SWAP ||
934 object->type == OBJT_VNODE) {
935 pindex = OFF_TO_IDX(current->offset +
936 (addr - current->start));
937 m = vm_page_lookup(object, pindex);
938 if (m != NULL && vm_page_none_valid(m))
941 mincoreinfo = MINCORE_INCORE;
945 VM_OBJECT_ASSERT_WLOCKED(m->object);
947 /* Examine other mappings of the page. */
948 if (m->dirty == 0 && pmap_is_modified(m))
951 mincoreinfo |= MINCORE_MODIFIED_OTHER;
954 * The first test for PGA_REFERENCED is an
955 * optimization. The second test is
956 * required because a concurrent pmap
957 * operation could clear the last reference
958 * and set PGA_REFERENCED before the call to
959 * pmap_is_referenced().
961 if ((m->a.flags & PGA_REFERENCED) != 0 ||
962 pmap_is_referenced(m) ||
963 (m->a.flags & PGA_REFERENCED) != 0)
964 mincoreinfo |= MINCORE_REFERENCED_OTHER;
967 VM_OBJECT_WUNLOCK(object);
970 * subyte may page fault. In case it needs to modify
971 * the map, we release the lock.
973 vm_map_unlock_read(map);
976 * calculate index into user supplied byte vector
978 vecindex = atop(addr - first_addr);
981 * If we have skipped map entries, we need to make sure that
982 * the byte vector is zeroed for those skipped entries.
984 while ((lastvecindex + 1) < vecindex) {
986 error = subyte(vec + lastvecindex, 0);
994 * Pass the page information to the user
996 error = subyte(vec + vecindex, mincoreinfo);
1003 * If the map has changed, due to the subyte, the previous
1004 * output may be invalid.
1006 vm_map_lock_read(map);
1007 if (timestamp != map->timestamp)
1010 lastvecindex = vecindex;
1015 * subyte may page fault. In case it needs to modify
1016 * the map, we release the lock.
1018 vm_map_unlock_read(map);
1021 * Zero the last entries in the byte vector.
1023 vecindex = atop(end - first_addr);
1024 while ((lastvecindex + 1) < vecindex) {
1026 error = subyte(vec + lastvecindex, 0);
1034 * If the map has changed, due to the subyte, the previous
1035 * output may be invalid.
1037 vm_map_lock_read(map);
1038 if (timestamp != map->timestamp)
1040 vm_map_unlock_read(map);
1045 #ifndef _SYS_SYSPROTO_H_
1052 sys_mlock(struct thread *td, struct mlock_args *uap)
1055 return (kern_mlock(td->td_proc, td->td_ucred,
1056 __DECONST(uintptr_t, uap->addr), uap->len));
1060 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1062 vm_offset_t addr, end, last, start;
1063 vm_size_t npages, size;
1065 unsigned long nsize;
1068 error = priv_check_cred(cred, PRIV_VM_MLOCK);
1074 start = trunc_page(addr);
1075 end = round_page(last);
1076 if (last < addr || end < addr)
1078 npages = atop(end - start);
1079 if (npages > vm_page_max_user_wired)
1081 map = &proc->p_vmspace->vm_map;
1083 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1084 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1092 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1098 error = vm_map_wire(map, start, end,
1099 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1101 if (racct_enable && error != KERN_SUCCESS) {
1103 racct_set(proc, RACCT_MEMLOCK,
1104 ptoa(pmap_wired_count(map->pmap)));
1108 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1111 #ifndef _SYS_SYSPROTO_H_
1112 struct mlockall_args {
1118 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1123 map = &td->td_proc->p_vmspace->vm_map;
1124 error = priv_check(td, PRIV_VM_MLOCK);
1128 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1132 * If wiring all pages in the process would cause it to exceed
1133 * a hard resource limit, return ENOMEM.
1135 if (!old_mlock && uap->how & MCL_CURRENT) {
1136 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1141 PROC_LOCK(td->td_proc);
1142 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1143 PROC_UNLOCK(td->td_proc);
1149 if (uap->how & MCL_FUTURE) {
1151 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1156 if (uap->how & MCL_CURRENT) {
1158 * P1003.1-2001 mandates that all currently mapped pages
1159 * will be memory resident and locked (wired) upon return
1160 * from mlockall(). vm_map_wire() will wire pages, by
1161 * calling vm_fault_wire() for each page in the region.
1163 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1164 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1165 if (error == KERN_SUCCESS)
1167 else if (error == KERN_RESOURCE_SHORTAGE)
1173 if (racct_enable && error != KERN_SUCCESS) {
1174 PROC_LOCK(td->td_proc);
1175 racct_set(td->td_proc, RACCT_MEMLOCK,
1176 ptoa(pmap_wired_count(map->pmap)));
1177 PROC_UNLOCK(td->td_proc);
1184 #ifndef _SYS_SYSPROTO_H_
1185 struct munlockall_args {
1191 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1196 map = &td->td_proc->p_vmspace->vm_map;
1197 error = priv_check(td, PRIV_VM_MUNLOCK);
1201 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1203 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1206 /* Forcibly unwire all pages. */
1207 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1208 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1210 if (racct_enable && error == KERN_SUCCESS) {
1211 PROC_LOCK(td->td_proc);
1212 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1213 PROC_UNLOCK(td->td_proc);
1220 #ifndef _SYS_SYSPROTO_H_
1221 struct munlock_args {
1227 sys_munlock(struct thread *td, struct munlock_args *uap)
1230 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1234 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1236 vm_offset_t addr, end, last, start;
1242 error = priv_check(td, PRIV_VM_MUNLOCK);
1247 start = trunc_page(addr);
1248 end = round_page(last);
1249 if (last < addr || end < addr)
1251 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1252 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1254 if (racct_enable && error == KERN_SUCCESS) {
1255 PROC_LOCK(td->td_proc);
1256 map = &td->td_proc->p_vmspace->vm_map;
1257 racct_set(td->td_proc, RACCT_MEMLOCK,
1258 ptoa(pmap_wired_count(map->pmap)));
1259 PROC_UNLOCK(td->td_proc);
1262 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1268 * Helper function for vm_mmap. Perform sanity check specific for mmap
1269 * operations on vnodes.
1272 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1273 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1274 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1275 boolean_t *writecounted)
1284 cred = td->td_ucred;
1285 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1286 (*flagsp & MAP_SHARED) != 0;
1287 if ((error = vget(vp, LK_SHARED)) != 0)
1289 AUDIT_ARG_VNODE1(vp);
1293 if (vp->v_type == VREG) {
1295 * Get the proper underlying object
1301 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1303 vp = (struct vnode *)obj->handle;
1305 * Bypass filesystems obey the mpsafety of the
1306 * underlying fs. Tmpfs never bypasses.
1308 error = vget(vp, LK_SHARED);
1313 *writecounted = TRUE;
1314 vm_pager_update_writecount(obj, 0, objsize);
1320 if ((error = VOP_GETATTR(vp, &va, cred)))
1323 /* This relies on VM_PROT_* matching PROT_*. */
1324 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1328 if ((flags & MAP_SHARED) != 0) {
1329 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1330 if (prot & VM_PROT_WRITE) {
1334 *maxprotp &= ~VM_PROT_WRITE;
1338 * If it is a regular file without any references
1339 * we do not need to sync it.
1340 * Adjust object size to be the size of actual file.
1342 objsize = round_page(va.va_size);
1343 if (va.va_nlink == 0)
1344 flags |= MAP_NOSYNC;
1345 if (obj->type == OBJT_VNODE) {
1346 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1353 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1354 ("wrong object type"));
1355 vm_object_reference(obj);
1356 #if VM_NRESERVLEVEL > 0
1357 if ((obj->flags & OBJ_COLORED) == 0) {
1358 VM_OBJECT_WLOCK(obj);
1359 vm_object_color(obj, 0);
1360 VM_OBJECT_WUNLOCK(obj);
1370 if (error != 0 && *writecounted) {
1371 *writecounted = FALSE;
1372 vm_pager_update_writecount(obj, objsize, 0);
1381 * Helper function for vm_mmap. Perform sanity check specific for mmap
1382 * operations on cdevs.
1385 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1386 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1387 vm_ooffset_t *foff, vm_object_t *objp)
1394 if (dsw->d_flags & D_MMAP_ANON) {
1397 *maxprotp = VM_PROT_ALL;
1398 *flagsp |= MAP_ANON;
1402 * cdevs do not provide private mappings of any kind.
1404 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1405 (prot & VM_PROT_WRITE) != 0)
1407 if (flags & (MAP_PRIVATE|MAP_COPY))
1410 * Force device mappings to be shared.
1412 flags |= MAP_SHARED;
1414 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1419 * First, try d_mmap_single(). If that is not implemented
1420 * (returns ENODEV), fall back to using the device pager.
1421 * Note that d_mmap_single() must return a reference to the
1422 * object (it needs to bump the reference count of the object
1423 * it returns somehow).
1425 * XXX assumes VM_PROT_* == PROT_*
1427 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1428 if (error != ENODEV)
1430 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1442 * Internal version of mmap used by exec, sys5 shared memory, and
1443 * various device drivers. Handle is either a vnode pointer, a
1444 * character device, or NULL for MAP_ANON.
1447 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1448 vm_prot_t maxprot, int flags,
1449 objtype_t handle_type, void *handle,
1453 struct thread *td = curthread;
1455 boolean_t writecounted;
1460 size = round_page(size);
1462 writecounted = FALSE;
1465 * Lookup/allocate object.
1467 switch (handle_type) {
1474 dsw = dev_refthread(cdev, &ref);
1477 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1478 dsw, &foff, &object);
1479 dev_relthread(cdev, ref);
1483 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1484 handle, &foff, &object, &writecounted);
1487 if (handle == NULL) {
1499 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1500 foff, writecounted, td);
1501 if (error != 0 && object != NULL) {
1503 * If this mapping was accounted for in the vnode's
1504 * writecount, then undo that now.
1507 vm_pager_release_writecount(object, 0, size);
1508 vm_object_deallocate(object);
1514 * Internal version of mmap that maps a specific VM object into an
1515 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1518 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1519 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1520 boolean_t writecounted, struct thread *td)
1522 boolean_t curmap, fitit;
1523 vm_offset_t max_addr;
1524 int docow, error, findspace, rv;
1526 curmap = map == &td->td_proc->p_vmspace->vm_map;
1528 RACCT_PROC_LOCK(td->td_proc);
1529 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1530 RACCT_PROC_UNLOCK(td->td_proc);
1533 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1534 RACCT_PROC_UNLOCK(td->td_proc);
1537 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1538 if (ptoa(pmap_wired_count(map->pmap)) + size >
1539 lim_cur(td, RLIMIT_MEMLOCK)) {
1540 racct_set_force(td->td_proc, RACCT_VMEM,
1542 RACCT_PROC_UNLOCK(td->td_proc);
1545 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1546 ptoa(pmap_wired_count(map->pmap)) + size);
1548 racct_set_force(td->td_proc, RACCT_VMEM,
1550 RACCT_PROC_UNLOCK(td->td_proc);
1554 RACCT_PROC_UNLOCK(td->td_proc);
1558 * We currently can only deal with page aligned file offsets.
1559 * The mmap() system call already enforces this by subtracting
1560 * the page offset from the file offset, but checking here
1561 * catches errors in device drivers (e.g. d_single_mmap()
1562 * callbacks) and other internal mapping requests (such as in
1565 if (foff & PAGE_MASK)
1568 if ((flags & MAP_FIXED) == 0) {
1570 *addr = round_page(*addr);
1572 if (*addr != trunc_page(*addr))
1577 if (flags & MAP_ANON) {
1578 if (object != NULL || foff != 0)
1581 } else if (flags & MAP_PREFAULT_READ)
1582 docow = MAP_PREFAULT;
1584 docow = MAP_PREFAULT_PARTIAL;
1586 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1587 docow |= MAP_COPY_ON_WRITE;
1588 if (flags & MAP_NOSYNC)
1589 docow |= MAP_DISABLE_SYNCER;
1590 if (flags & MAP_NOCORE)
1591 docow |= MAP_DISABLE_COREDUMP;
1592 /* Shared memory is also shared with children. */
1593 if (flags & MAP_SHARED)
1594 docow |= MAP_INHERIT_SHARE;
1596 docow |= MAP_WRITECOUNT;
1597 if (flags & MAP_STACK) {
1600 docow |= MAP_STACK_GROWS_DOWN;
1602 if ((flags & MAP_EXCL) != 0)
1603 docow |= MAP_CHECK_EXCL;
1604 if ((flags & MAP_GUARD) != 0)
1605 docow |= MAP_CREATE_GUARD;
1608 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1609 findspace = VMFS_SUPER_SPACE;
1610 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1611 findspace = VMFS_ALIGNED_SPACE(flags >>
1612 MAP_ALIGNMENT_SHIFT);
1614 findspace = VMFS_OPTIMAL_SPACE;
1617 if ((flags & MAP_32BIT) != 0)
1618 max_addr = MAP_32BIT_MAX_ADDR;
1621 rv = vm_map_find_min(map, object, foff, addr, size,
1622 round_page((vm_offset_t)td->td_proc->p_vmspace->
1623 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1624 findspace, prot, maxprot, docow);
1626 rv = vm_map_find(map, object, foff, addr, size,
1627 max_addr, findspace, prot, maxprot, docow);
1630 rv = vm_map_fixed(map, object, foff, *addr, size,
1631 prot, maxprot, docow);
1634 if (rv == KERN_SUCCESS) {
1636 * If the process has requested that all future mappings
1637 * be wired, then heed this.
1639 if ((map->flags & MAP_WIREFUTURE) != 0) {
1641 if ((map->flags & MAP_WIREFUTURE) != 0)
1642 (void)vm_map_wire_locked(map, *addr,
1643 *addr + size, VM_MAP_WIRE_USER |
1644 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1645 VM_MAP_WIRE_NOHOLES));
1649 return (vm_mmap_to_errno(rv));
1653 * Translate a Mach VM return code to zero on success or the appropriate errno
1657 vm_mmap_to_errno(int rv)
1663 case KERN_INVALID_ADDRESS:
1666 case KERN_PROTECTION_FAILURE: