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 _Static_assert(MAXPAGESIZES <= 4, "MINCORE_SUPER too narrow");
117 #ifndef _SYS_SYSPROTO_H_
124 sys_sbrk(struct thread *td, struct sbrk_args *uap)
126 /* Not yet implemented */
130 #ifndef _SYS_SYSPROTO_H_
137 sys_sstk(struct thread *td, struct sstk_args *uap)
139 /* Not yet implemented */
143 #if defined(COMPAT_43)
145 ogetpagesize(struct thread *td, struct ogetpagesize_args *uap)
148 td->td_retval[0] = PAGE_SIZE;
151 #endif /* COMPAT_43 */
154 * Memory Map (mmap) system call. Note that the file offset
155 * and address are allowed to be NOT page aligned, though if
156 * the MAP_FIXED flag it set, both must have the same remainder
157 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
158 * page-aligned, the actual mapping starts at trunc_page(addr)
159 * and the return value is adjusted up by the page offset.
161 * Generally speaking, only character devices which are themselves
162 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
163 * there would be no cache coherency between a descriptor and a VM mapping
164 * both to the same character device.
166 #ifndef _SYS_SYSPROTO_H_
179 sys_mmap(struct thread *td, struct mmap_args *uap)
182 return (kern_mmap(td, &(struct mmap_req){
183 .mr_hint = (uintptr_t)uap->addr,
185 .mr_prot = uap->prot,
186 .mr_flags = uap->flags,
193 kern_mmap_maxprot(struct proc *p, int prot)
196 if ((p->p_flag2 & P2_PROTMAX_DISABLE) != 0 ||
197 (p->p_fctl0 & NT_FREEBSD_FCTL_PROTMAX_DISABLE) != 0)
199 if (((p->p_flag2 & P2_PROTMAX_ENABLE) != 0 || imply_prot_max) &&
206 kern_mmap(struct thread *td, const struct mmap_req *mrp)
212 vm_offset_t addr, orig_addr;
213 vm_size_t len, pageoff, size;
214 vm_prot_t cap_maxprot;
215 int align, error, fd, flags, max_prot, prot;
217 mmap_check_fp_fn check_fp_fn;
219 orig_addr = addr = mrp->mr_hint;
222 flags = mrp->mr_flags;
225 check_fp_fn = mrp->mr_check_fp_fn;
227 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
229 max_prot = PROT_MAX_EXTRACT(prot);
230 prot = PROT_EXTRACT(prot);
231 if (max_prot != 0 && (max_prot & prot) != prot)
237 * Always honor PROT_MAX if set. If not, default to all
238 * permissions unless we're implying maximum permissions.
241 max_prot = kern_mmap_maxprot(p, prot);
248 * Ignore old flags that used to be defined but did not do anything.
250 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
253 * Enforce the constraints.
254 * Mapping of length 0 is only allowed for old binaries.
255 * Anonymous mapping shall specify -1 as filedescriptor and
256 * zero position for new code. Be nice to ancient a.out
257 * binaries and correct pos for anonymous mapping, since old
258 * ld.so sometimes issues anonymous map requests with non-zero
261 if (!SV_CURPROC_FLAG(SV_AOUT)) {
262 if ((len == 0 && p->p_osrel >= P_OSREL_MAP_ANON) ||
263 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
266 if ((flags & MAP_ANON) != 0)
270 if (flags & MAP_STACK) {
272 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
277 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
278 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
279 MAP_PREFAULT_READ | MAP_GUARD |
283 MAP_ALIGNMENT_MASK)) != 0)
285 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
287 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
289 if (prot != PROT_NONE &&
290 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
292 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
293 pos != 0 || (flags & ~(MAP_FIXED | MAP_GUARD | MAP_EXCL |
297 MAP_ALIGNMENT_MASK)) != 0))
301 * Align the file position to a page boundary,
302 * and save its page offset component.
304 pageoff = (pos & PAGE_MASK);
307 /* Compute size from len by rounding (on both ends). */
308 size = len + pageoff; /* low end... */
309 size = round_page(size); /* hi end */
310 /* Check for rounding up to zero. */
314 /* Ensure alignment is at least a page and fits in a pointer. */
315 align = flags & MAP_ALIGNMENT_MASK;
316 if (align != 0 && align != MAP_ALIGNED_SUPER &&
317 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
318 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
322 * Check for illegal addresses. Watch out for address wrap... Note
323 * that VM_*_ADDRESS are not constants due to casts (argh).
325 if (flags & MAP_FIXED) {
327 * The specified address must have the same remainder
328 * as the file offset taken modulo PAGE_SIZE, so it
329 * should be aligned after adjustment by pageoff.
332 if (addr & PAGE_MASK)
335 /* Address range must be all in user VM space. */
336 if (!vm_map_range_valid(&vms->vm_map, addr, addr + size))
339 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
341 } else if (flags & MAP_32BIT) {
343 * For MAP_32BIT, override the hint if it is too high and
344 * do not bother moving the mapping past the heap (since
345 * the heap is usually above 2GB).
347 if (addr + size > MAP_32BIT_MAX_ADDR)
352 * XXX for non-fixed mappings where no hint is provided or
353 * the hint would fall in the potential heap space,
354 * place it after the end of the largest possible heap.
356 * There should really be a pmap call to determine a reasonable
360 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
361 addr < round_page((vm_offset_t)vms->vm_daddr +
362 lim_max(td, RLIMIT_DATA))))
363 addr = round_page((vm_offset_t)vms->vm_daddr +
364 lim_max(td, RLIMIT_DATA));
368 * Return success without mapping anything for old
369 * binaries that request a page-aligned mapping of
370 * length 0. For modern binaries, this function
371 * returns an error earlier.
374 } else if ((flags & MAP_GUARD) != 0) {
375 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
376 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
377 } else if ((flags & MAP_ANON) != 0) {
379 * Mapping blank space is trivial.
381 * This relies on VM_PROT_* matching PROT_*.
383 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
384 max_prot, flags, NULL, pos, FALSE, td);
387 * Mapping file, get fp for validation and don't let the
388 * descriptor disappear on us if we block. Check capability
389 * rights, but also return the maximum rights to be combined
390 * with maxprot later.
392 cap_rights_init_one(&rights, CAP_MMAP);
393 if (prot & PROT_READ)
394 cap_rights_set_one(&rights, CAP_MMAP_R);
395 if ((flags & MAP_SHARED) != 0) {
396 if (prot & PROT_WRITE)
397 cap_rights_set_one(&rights, CAP_MMAP_W);
399 if (prot & PROT_EXEC)
400 cap_rights_set_one(&rights, CAP_MMAP_X);
401 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
404 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
405 p->p_osrel >= P_OSREL_MAP_FSTRICT) {
409 if (check_fp_fn != NULL) {
410 error = check_fp_fn(fp, prot, max_prot & cap_maxprot,
415 if (fp->f_ops == &shm_ops && shm_largepage(fp->f_data))
417 /* This relies on VM_PROT_* matching PROT_*. */
418 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
419 max_prot & cap_maxprot, flags, pos, td);
423 td->td_retval[0] = addr + pageoff;
431 #if defined(COMPAT_FREEBSD6)
433 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
435 return (kern_mmap(td, &(struct mmap_req){
436 .mr_hint = (uintptr_t)uap->addr,
438 .mr_prot = uap->prot,
439 .mr_flags = uap->flags,
447 #ifndef _SYS_SYSPROTO_H_
458 ommap(struct thread *td, struct ommap_args *uap)
460 return (kern_ommap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
461 uap->flags, uap->fd, uap->pos));
465 kern_ommap(struct thread *td, uintptr_t hint, int len, int oprot,
466 int oflags, int fd, long pos)
468 static const char cvtbsdprot[8] = {
472 PROT_EXEC | PROT_WRITE,
474 PROT_EXEC | PROT_READ,
475 PROT_WRITE | PROT_READ,
476 PROT_EXEC | PROT_WRITE | PROT_READ,
483 #define OMAP_ANON 0x0002
484 #define OMAP_COPY 0x0020
485 #define OMAP_SHARED 0x0010
486 #define OMAP_FIXED 0x0100
488 prot = cvtbsdprot[oprot & 0x7];
489 #if (defined(COMPAT_FREEBSD32) && defined(__amd64__)) || defined(__i386__)
490 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
495 if (oflags & OMAP_ANON)
497 if (oflags & OMAP_COPY)
499 if (oflags & OMAP_SHARED)
502 flags |= MAP_PRIVATE;
503 if (oflags & OMAP_FIXED)
505 return (kern_mmap(td, &(struct mmap_req){
514 #endif /* COMPAT_43 */
516 #ifndef _SYS_SYSPROTO_H_
524 sys_msync(struct thread *td, struct msync_args *uap)
527 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
531 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
539 pageoff = (addr & PAGE_MASK);
542 size = (vm_size_t) round_page(size);
543 if (addr + size < addr)
546 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
549 map = &td->td_proc->p_vmspace->vm_map;
552 * Clean the pages and interpret the return value.
554 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
555 (flags & MS_INVALIDATE) != 0);
559 case KERN_INVALID_ADDRESS:
561 case KERN_INVALID_ARGUMENT:
570 #ifndef _SYS_SYSPROTO_H_
577 sys_munmap(struct thread *td, struct munmap_args *uap)
580 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
584 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
587 struct pmckern_map_out pkm;
588 vm_map_entry_t entry;
591 vm_offset_t addr, end;
600 pageoff = (addr & PAGE_MASK);
603 size = (vm_size_t) round_page(size);
605 map = &td->td_proc->p_vmspace->vm_map;
606 if (!vm_map_range_valid(map, addr, end))
612 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
615 * Inform hwpmc if the address range being unmapped contains
616 * an executable region.
618 pkm.pm_address = (uintptr_t) NULL;
619 if (vm_map_lookup_entry(map, addr, &entry)) {
620 for (; entry->start < end;
621 entry = vm_map_entry_succ(entry)) {
622 if (vm_map_check_protection(map, entry->start,
623 entry->end, VM_PROT_EXECUTE) == TRUE) {
624 pkm.pm_address = (uintptr_t) addr;
625 pkm.pm_size = (size_t) size;
632 rv = vm_map_delete(map, addr, end);
635 if (rv == KERN_SUCCESS && __predict_false(pmc_handled)) {
636 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
637 vm_map_lock_downgrade(map);
638 if (pkm.pm_address != (uintptr_t) NULL)
639 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
640 vm_map_unlock_read(map);
645 return (vm_mmap_to_errno(rv));
648 #ifndef _SYS_SYSPROTO_H_
649 struct mprotect_args {
656 sys_mprotect(struct thread *td, struct mprotect_args *uap)
659 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
663 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
667 int vm_error, max_prot;
671 if ((prot & ~(_PROT_ALL | PROT_MAX(_PROT_ALL))) != 0)
673 max_prot = PROT_MAX_EXTRACT(prot);
674 prot = PROT_EXTRACT(prot);
675 pageoff = (addr & PAGE_MASK);
678 size = (vm_size_t) round_page(size);
679 #ifdef COMPAT_FREEBSD32
680 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
681 if (((addr + size) & 0xffffffff) < addr)
685 if (addr + size < addr)
688 flags = VM_MAP_PROTECT_SET_PROT;
690 flags |= VM_MAP_PROTECT_SET_MAXPROT;
691 vm_error = vm_map_protect(&td->td_proc->p_vmspace->vm_map,
692 addr, addr + size, prot, max_prot, flags);
697 case KERN_PROTECTION_FAILURE:
699 case KERN_RESOURCE_SHORTAGE:
701 case KERN_OUT_OF_BOUNDS:
707 #ifndef _SYS_SYSPROTO_H_
708 struct minherit_args {
715 sys_minherit(struct thread *td, struct minherit_args *uap)
718 return (kern_minherit(td, (uintptr_t)uap->addr, uap->len,
723 kern_minherit(struct thread *td, uintptr_t addr0, size_t len, int inherit0)
726 vm_size_t size, pageoff;
727 vm_inherit_t inherit;
729 addr = (vm_offset_t)addr0;
733 pageoff = (addr & PAGE_MASK);
736 size = (vm_size_t) round_page(size);
737 if (addr + size < addr)
740 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
741 addr + size, inherit)) {
744 case KERN_PROTECTION_FAILURE:
750 #ifndef _SYS_SYSPROTO_H_
751 struct madvise_args {
759 sys_madvise(struct thread *td, struct madvise_args *uap)
762 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
766 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
769 vm_offset_t addr, end, start;
773 * Check for our special case, advising the swap pager we are
776 if (behav == MADV_PROTECT) {
778 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
779 PROC_SPROTECT, &flags));
783 * Check for illegal addresses. Watch out for address wrap... Note
784 * that VM_*_ADDRESS are not constants due to casts (argh).
786 map = &td->td_proc->p_vmspace->vm_map;
788 if (!vm_map_range_valid(map, addr, addr + len))
792 * Since this routine is only advisory, we default to conservative
795 start = trunc_page(addr);
796 end = round_page(addr + len);
799 * vm_map_madvise() checks for illegal values of behav.
801 return (vm_map_madvise(map, start, end, behav));
804 #ifndef _SYS_SYSPROTO_H_
805 struct mincore_args {
813 sys_mincore(struct thread *td, struct mincore_args *uap)
816 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
820 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
824 vm_map_entry_t current, entry;
826 vm_offset_t addr, cend, end, first_addr;
830 int error, lastvecindex, mincoreinfo, vecindex;
831 unsigned int timestamp;
834 * Make sure that the addresses presented are valid for user
837 first_addr = addr = trunc_page(addr0);
838 end = round_page(addr0 + len);
839 map = &td->td_proc->p_vmspace->vm_map;
840 if (end > vm_map_max(map) || end < addr)
843 pmap = vmspace_pmap(td->td_proc->p_vmspace);
845 vm_map_lock_read(map);
847 timestamp = map->timestamp;
849 if (!vm_map_lookup_entry(map, addr, &entry)) {
850 vm_map_unlock_read(map);
855 * Do this on a map entry basis so that if the pages are not
856 * in the current processes address space, we can easily look
857 * up the pages elsewhere.
860 while (entry->start < end) {
862 * check for contiguity
865 entry = vm_map_entry_succ(current);
866 if (current->end < end &&
867 entry->start > current->end) {
868 vm_map_unlock_read(map);
873 * ignore submaps (for now) or null objects
875 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
876 current->object.vm_object == NULL)
880 * limit this scan to the current map entry and the
881 * limits for the mincore call
883 if (addr < current->start)
884 addr = current->start;
889 for (; addr < cend; addr += PAGE_SIZE) {
891 * Check pmap first, it is likely faster, also
892 * it can provide info as to whether we are the
893 * one referencing or modifying the page.
899 mincoreinfo = pmap_mincore(pmap, addr, &pa);
900 if (mincore_mapped) {
902 * We only care about this pmap's
903 * mapping of the page, if any.
906 } else if (pa != 0) {
908 * The page is mapped by this process but not
909 * both accessed and modified. It is also
910 * managed. Acquire the object lock so that
911 * other mappings might be examined. The page's
912 * identity may change at any point before its
913 * object lock is acquired, so re-validate if
916 m = PHYS_TO_VM_PAGE(pa);
917 while (object == NULL || m->object != object) {
919 VM_OBJECT_WUNLOCK(object);
920 object = atomic_load_ptr(&m->object);
923 VM_OBJECT_WLOCK(object);
925 if (pa != pmap_extract(pmap, addr))
927 KASSERT(vm_page_all_valid(m),
928 ("mincore: page %p is mapped but invalid",
930 } else if (mincoreinfo == 0) {
932 * The page is not mapped by this process. If
933 * the object implements managed pages, then
934 * determine if the page is resident so that
935 * the mappings might be examined.
937 if (current->object.vm_object != object) {
939 VM_OBJECT_WUNLOCK(object);
940 object = current->object.vm_object;
941 VM_OBJECT_WLOCK(object);
943 if ((object->flags & OBJ_SWAP) != 0 ||
944 object->type == OBJT_VNODE) {
945 pindex = OFF_TO_IDX(current->offset +
946 (addr - current->start));
947 m = vm_page_lookup(object, pindex);
948 if (m != NULL && vm_page_none_valid(m))
951 mincoreinfo = MINCORE_INCORE;
955 VM_OBJECT_ASSERT_WLOCKED(m->object);
957 /* Examine other mappings of the page. */
958 if (m->dirty == 0 && pmap_is_modified(m))
961 mincoreinfo |= MINCORE_MODIFIED_OTHER;
964 * The first test for PGA_REFERENCED is an
965 * optimization. The second test is
966 * required because a concurrent pmap
967 * operation could clear the last reference
968 * and set PGA_REFERENCED before the call to
969 * pmap_is_referenced().
971 if ((m->a.flags & PGA_REFERENCED) != 0 ||
972 pmap_is_referenced(m) ||
973 (m->a.flags & PGA_REFERENCED) != 0)
974 mincoreinfo |= MINCORE_REFERENCED_OTHER;
977 VM_OBJECT_WUNLOCK(object);
980 * subyte may page fault. In case it needs to modify
981 * the map, we release the lock.
983 vm_map_unlock_read(map);
986 * calculate index into user supplied byte vector
988 vecindex = atop(addr - first_addr);
991 * If we have skipped map entries, we need to make sure that
992 * the byte vector is zeroed for those skipped entries.
994 while ((lastvecindex + 1) < vecindex) {
996 error = subyte(vec + lastvecindex, 0);
1004 * Pass the page information to the user
1006 error = subyte(vec + vecindex, mincoreinfo);
1013 * If the map has changed, due to the subyte, the previous
1014 * output may be invalid.
1016 vm_map_lock_read(map);
1017 if (timestamp != map->timestamp)
1020 lastvecindex = vecindex;
1025 * subyte may page fault. In case it needs to modify
1026 * the map, we release the lock.
1028 vm_map_unlock_read(map);
1031 * Zero the last entries in the byte vector.
1033 vecindex = atop(end - first_addr);
1034 while ((lastvecindex + 1) < vecindex) {
1036 error = subyte(vec + lastvecindex, 0);
1044 * If the map has changed, due to the subyte, the previous
1045 * output may be invalid.
1047 vm_map_lock_read(map);
1048 if (timestamp != map->timestamp)
1050 vm_map_unlock_read(map);
1055 #ifndef _SYS_SYSPROTO_H_
1062 sys_mlock(struct thread *td, struct mlock_args *uap)
1065 return (kern_mlock(td->td_proc, td->td_ucred,
1066 __DECONST(uintptr_t, uap->addr), uap->len));
1070 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1072 vm_offset_t addr, end, last, start;
1073 vm_size_t npages, size;
1075 unsigned long nsize;
1078 error = priv_check_cred(cred, PRIV_VM_MLOCK);
1084 start = trunc_page(addr);
1085 end = round_page(last);
1086 if (last < addr || end < addr)
1088 npages = atop(end - start);
1089 if (npages > vm_page_max_user_wired)
1091 map = &proc->p_vmspace->vm_map;
1093 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1094 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1102 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1108 error = vm_map_wire(map, start, end,
1109 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1111 if (racct_enable && error != KERN_SUCCESS) {
1113 racct_set(proc, RACCT_MEMLOCK,
1114 ptoa(pmap_wired_count(map->pmap)));
1121 case KERN_INVALID_ARGUMENT:
1128 #ifndef _SYS_SYSPROTO_H_
1129 struct mlockall_args {
1135 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1140 map = &td->td_proc->p_vmspace->vm_map;
1141 error = priv_check(td, PRIV_VM_MLOCK);
1145 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1149 * If wiring all pages in the process would cause it to exceed
1150 * a hard resource limit, return ENOMEM.
1152 if (!old_mlock && uap->how & MCL_CURRENT) {
1153 if (map->size > lim_cur(td, RLIMIT_MEMLOCK))
1158 PROC_LOCK(td->td_proc);
1159 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1160 PROC_UNLOCK(td->td_proc);
1166 if (uap->how & MCL_FUTURE) {
1168 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1173 if (uap->how & MCL_CURRENT) {
1175 * P1003.1-2001 mandates that all currently mapped pages
1176 * will be memory resident and locked (wired) upon return
1177 * from mlockall(). vm_map_wire() will wire pages, by
1178 * calling vm_fault_wire() for each page in the region.
1180 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1181 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1182 if (error == KERN_SUCCESS)
1184 else if (error == KERN_RESOURCE_SHORTAGE)
1190 if (racct_enable && error != KERN_SUCCESS) {
1191 PROC_LOCK(td->td_proc);
1192 racct_set(td->td_proc, RACCT_MEMLOCK,
1193 ptoa(pmap_wired_count(map->pmap)));
1194 PROC_UNLOCK(td->td_proc);
1201 #ifndef _SYS_SYSPROTO_H_
1202 struct munlockall_args {
1208 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1213 map = &td->td_proc->p_vmspace->vm_map;
1214 error = priv_check(td, PRIV_VM_MUNLOCK);
1218 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1220 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1223 /* Forcibly unwire all pages. */
1224 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1225 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1227 if (racct_enable && error == KERN_SUCCESS) {
1228 PROC_LOCK(td->td_proc);
1229 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1230 PROC_UNLOCK(td->td_proc);
1237 #ifndef _SYS_SYSPROTO_H_
1238 struct munlock_args {
1244 sys_munlock(struct thread *td, struct munlock_args *uap)
1247 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1251 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1253 vm_offset_t addr, end, last, start;
1259 error = priv_check(td, PRIV_VM_MUNLOCK);
1264 start = trunc_page(addr);
1265 end = round_page(last);
1266 if (last < addr || end < addr)
1268 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1269 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1271 if (racct_enable && error == KERN_SUCCESS) {
1272 PROC_LOCK(td->td_proc);
1273 map = &td->td_proc->p_vmspace->vm_map;
1274 racct_set(td->td_proc, RACCT_MEMLOCK,
1275 ptoa(pmap_wired_count(map->pmap)));
1276 PROC_UNLOCK(td->td_proc);
1279 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1285 * Helper function for vm_mmap. Perform sanity check specific for mmap
1286 * operations on vnodes.
1289 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1290 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1291 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1292 boolean_t *writecounted)
1301 cred = td->td_ucred;
1302 writex = (*maxprotp & VM_PROT_WRITE) != 0 &&
1303 (*flagsp & MAP_SHARED) != 0;
1304 if ((error = vget(vp, LK_SHARED)) != 0)
1306 AUDIT_ARG_VNODE1(vp);
1310 if (vp->v_type == VREG) {
1312 * Get the proper underlying object
1318 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1320 vp = (struct vnode *)obj->handle;
1322 * Bypass filesystems obey the mpsafety of the
1323 * underlying fs. Tmpfs never bypasses.
1325 error = vget(vp, LK_SHARED);
1330 *writecounted = TRUE;
1331 vm_pager_update_writecount(obj, 0, objsize);
1337 if ((error = VOP_GETATTR(vp, &va, cred)))
1340 /* This relies on VM_PROT_* matching PROT_*. */
1341 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1345 if ((flags & MAP_SHARED) != 0) {
1346 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1347 if (prot & VM_PROT_WRITE) {
1351 *maxprotp &= ~VM_PROT_WRITE;
1355 * If it is a regular file without any references
1356 * we do not need to sync it.
1357 * Adjust object size to be the size of actual file.
1359 objsize = round_page(va.va_size);
1360 if (va.va_nlink == 0)
1361 flags |= MAP_NOSYNC;
1362 if (obj->type == OBJT_VNODE) {
1363 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1370 KASSERT((obj->flags & OBJ_SWAP) != 0, ("wrong object type"));
1371 vm_object_reference(obj);
1372 #if VM_NRESERVLEVEL > 0
1373 if ((obj->flags & OBJ_COLORED) == 0) {
1374 VM_OBJECT_WLOCK(obj);
1375 vm_object_color(obj, 0);
1376 VM_OBJECT_WUNLOCK(obj);
1386 if (error != 0 && *writecounted) {
1387 *writecounted = FALSE;
1388 vm_pager_update_writecount(obj, objsize, 0);
1397 * Helper function for vm_mmap. Perform sanity check specific for mmap
1398 * operations on cdevs.
1401 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1402 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1403 vm_ooffset_t *foff, vm_object_t *objp)
1410 if (dsw->d_flags & D_MMAP_ANON) {
1413 *maxprotp = VM_PROT_ALL;
1414 *flagsp |= MAP_ANON;
1418 * cdevs do not provide private mappings of any kind.
1420 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1421 (prot & VM_PROT_WRITE) != 0)
1423 if (flags & (MAP_PRIVATE|MAP_COPY))
1426 * Force device mappings to be shared.
1428 flags |= MAP_SHARED;
1430 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1435 * First, try d_mmap_single(). If that is not implemented
1436 * (returns ENODEV), fall back to using the device pager.
1437 * Note that d_mmap_single() must return a reference to the
1438 * object (it needs to bump the reference count of the object
1439 * it returns somehow).
1441 * XXX assumes VM_PROT_* == PROT_*
1443 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1444 if (error != ENODEV)
1446 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1456 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1457 vm_prot_t maxprot, int flags,
1458 objtype_t handle_type, void *handle,
1462 struct thread *td = curthread;
1464 boolean_t writecounted;
1469 size = round_page(size);
1471 writecounted = FALSE;
1473 switch (handle_type) {
1480 dsw = dev_refthread(cdev, &ref);
1483 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1484 dsw, &foff, &object);
1485 dev_relthread(cdev, ref);
1489 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1490 handle, &foff, &object, &writecounted);
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 kern_mmap_racct_check(struct thread *td, vm_map_t map, vm_size_t size)
1518 RACCT_PROC_LOCK(td->td_proc);
1519 if (map->size + size > lim_cur(td, RLIMIT_VMEM)) {
1520 RACCT_PROC_UNLOCK(td->td_proc);
1523 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1524 RACCT_PROC_UNLOCK(td->td_proc);
1527 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1528 if (ptoa(pmap_wired_count(map->pmap)) + size >
1529 lim_cur(td, RLIMIT_MEMLOCK)) {
1530 racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1531 RACCT_PROC_UNLOCK(td->td_proc);
1534 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1535 ptoa(pmap_wired_count(map->pmap)) + size);
1537 racct_set_force(td->td_proc, RACCT_VMEM, map->size);
1538 RACCT_PROC_UNLOCK(td->td_proc);
1542 RACCT_PROC_UNLOCK(td->td_proc);
1547 * Internal version of mmap that maps a specific VM object into an
1548 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1551 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1552 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1553 boolean_t writecounted, struct thread *td)
1555 vm_offset_t max_addr;
1556 int docow, error, findspace, rv;
1559 curmap = map == &td->td_proc->p_vmspace->vm_map;
1561 error = kern_mmap_racct_check(td, map, size);
1567 * We currently can only deal with page aligned file offsets.
1568 * The mmap() system call already enforces this by subtracting
1569 * the page offset from the file offset, but checking here
1570 * catches errors in device drivers (e.g. d_single_mmap()
1571 * callbacks) and other internal mapping requests (such as in
1574 if (foff & PAGE_MASK)
1577 if ((flags & MAP_FIXED) == 0) {
1579 *addr = round_page(*addr);
1581 if (*addr != trunc_page(*addr))
1586 if (flags & MAP_ANON) {
1587 if (object != NULL || foff != 0)
1590 } else if (flags & MAP_PREFAULT_READ)
1591 docow = MAP_PREFAULT;
1593 docow = MAP_PREFAULT_PARTIAL;
1595 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1596 docow |= MAP_COPY_ON_WRITE;
1597 if (flags & MAP_NOSYNC)
1598 docow |= MAP_DISABLE_SYNCER;
1599 if (flags & MAP_NOCORE)
1600 docow |= MAP_DISABLE_COREDUMP;
1601 /* Shared memory is also shared with children. */
1602 if (flags & MAP_SHARED)
1603 docow |= MAP_INHERIT_SHARE;
1605 docow |= MAP_WRITECOUNT;
1606 if (flags & MAP_STACK) {
1609 docow |= MAP_STACK_GROWS_DOWN;
1611 if ((flags & MAP_EXCL) != 0)
1612 docow |= MAP_CHECK_EXCL;
1613 if ((flags & MAP_GUARD) != 0)
1614 docow |= MAP_CREATE_GUARD;
1617 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1618 findspace = VMFS_SUPER_SPACE;
1619 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1620 findspace = VMFS_ALIGNED_SPACE(flags >>
1621 MAP_ALIGNMENT_SHIFT);
1623 findspace = VMFS_OPTIMAL_SPACE;
1626 if ((flags & MAP_32BIT) != 0)
1627 max_addr = MAP_32BIT_MAX_ADDR;
1630 rv = vm_map_find_min(map, object, foff, addr, size,
1631 round_page((vm_offset_t)td->td_proc->p_vmspace->
1632 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1633 findspace, prot, maxprot, docow);
1635 rv = vm_map_find(map, object, foff, addr, size,
1636 max_addr, findspace, prot, maxprot, docow);
1639 rv = vm_map_fixed(map, object, foff, *addr, size,
1640 prot, maxprot, docow);
1643 if (rv == KERN_SUCCESS) {
1645 * If the process has requested that all future mappings
1646 * be wired, then heed this.
1648 if ((map->flags & MAP_WIREFUTURE) != 0) {
1650 if ((map->flags & MAP_WIREFUTURE) != 0)
1651 (void)vm_map_wire_locked(map, *addr,
1652 *addr + size, VM_MAP_WIRE_USER |
1653 ((flags & MAP_STACK) ? VM_MAP_WIRE_HOLESOK :
1654 VM_MAP_WIRE_NOHOLES));
1658 return (vm_mmap_to_errno(rv));
1662 * Translate a Mach VM return code to zero on success or the appropriate errno
1666 vm_mmap_to_errno(int rv)
1672 case KERN_INVALID_ADDRESS:
1675 case KERN_PROTECTION_FAILURE: