1 /*-
2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
8 * Science Department.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35 *
36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
37 */
38
39 /*
40 * Mapped file (mmap) interface to VM
41 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD: stable/9/sys/vm/vm_mmap.c 266498 2014-05-21 15:17:22Z pho $");
45
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capability.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/procctl.h>
60 #include <sys/racct.h>
61 #include <sys/resource.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 #include <sys/fcntl.h>
66 #include <sys/file.h>
67 #include <sys/mman.h>
68 #include <sys/mount.h>
69 #include <sys/conf.h>
70 #include <sys/stat.h>
71 #include <sys/syscallsubr.h>
72 #include <sys/sysent.h>
73 #include <sys/vmmeter.h>
74
75 #include <security/mac/mac_framework.h>
76
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_pager.h>
84 #include <vm/vm_pageout.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
88
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92
93 int old_mlock = 1;
94 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RW | CTLFLAG_TUN, &old_mlock, 0,
95 "Do not apply RLIMIT_MEMLOCK on mlockall");
96 TUNABLE_INT("vm.old_mlock", &old_mlock);
97
98 #ifndef _SYS_SYSPROTO_H_
99 struct sbrk_args {
100 int incr;
101 };
102 #endif
103
104 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
105 int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *);
106 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
107 int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
108 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
109 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
110
111 /*
112 * MPSAFE
113 */
114 /* ARGSUSED */
115 int
sys_sbrk(td,uap)116 sys_sbrk(td, uap)
117 struct thread *td;
118 struct sbrk_args *uap;
119 {
120 /* Not yet implemented */
121 return (EOPNOTSUPP);
122 }
123
124 #ifndef _SYS_SYSPROTO_H_
125 struct sstk_args {
126 int incr;
127 };
128 #endif
129
130 /*
131 * MPSAFE
132 */
133 /* ARGSUSED */
134 int
sys_sstk(td,uap)135 sys_sstk(td, uap)
136 struct thread *td;
137 struct sstk_args *uap;
138 {
139 /* Not yet implemented */
140 return (EOPNOTSUPP);
141 }
142
143 #if defined(COMPAT_43)
144 #ifndef _SYS_SYSPROTO_H_
145 struct getpagesize_args {
146 int dummy;
147 };
148 #endif
149
150 /* ARGSUSED */
151 int
ogetpagesize(td,uap)152 ogetpagesize(td, uap)
153 struct thread *td;
154 struct getpagesize_args *uap;
155 {
156 /* MP SAFE */
157 td->td_retval[0] = PAGE_SIZE;
158 return (0);
159 }
160 #endif /* COMPAT_43 */
161
162
163 /*
164 * Memory Map (mmap) system call. Note that the file offset
165 * and address are allowed to be NOT page aligned, though if
166 * the MAP_FIXED flag it set, both must have the same remainder
167 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
168 * page-aligned, the actual mapping starts at trunc_page(addr)
169 * and the return value is adjusted up by the page offset.
170 *
171 * Generally speaking, only character devices which are themselves
172 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
173 * there would be no cache coherency between a descriptor and a VM mapping
174 * both to the same character device.
175 */
176 #ifndef _SYS_SYSPROTO_H_
177 struct mmap_args {
178 void *addr;
179 size_t len;
180 int prot;
181 int flags;
182 int fd;
183 long pad;
184 off_t pos;
185 };
186 #endif
187
188 /*
189 * MPSAFE
190 */
191 int
sys_mmap(td,uap)192 sys_mmap(td, uap)
193 struct thread *td;
194 struct mmap_args *uap;
195 {
196 #ifdef HWPMC_HOOKS
197 struct pmckern_map_in pkm;
198 #endif
199 struct file *fp;
200 struct vnode *vp;
201 vm_offset_t addr;
202 vm_size_t size, pageoff;
203 vm_prot_t cap_maxprot, prot, maxprot;
204 void *handle;
205 objtype_t handle_type;
206 int align, error, flags;
207 off_t pos;
208 struct vmspace *vms = td->td_proc->p_vmspace;
209 cap_rights_t rights;
210
211 addr = (vm_offset_t) uap->addr;
212 size = uap->len;
213 prot = uap->prot & VM_PROT_ALL;
214 flags = uap->flags;
215 pos = uap->pos;
216
217 fp = NULL;
218
219 /*
220 * Enforce the constraints.
221 * Mapping of length 0 is only allowed for old binaries.
222 * Anonymous mapping shall specify -1 as filedescriptor and
223 * zero position for new code. Be nice to ancient a.out
224 * binaries and correct pos for anonymous mapping, since old
225 * ld.so sometimes issues anonymous map requests with non-zero
226 * pos.
227 */
228 if (!SV_CURPROC_FLAG(SV_AOUT)) {
229 if ((uap->len == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
230 ((flags & MAP_ANON) != 0 && (uap->fd != -1 || pos != 0)))
231 return (EINVAL);
232 } else {
233 if ((flags & MAP_ANON) != 0)
234 pos = 0;
235 }
236
237 if (flags & MAP_STACK) {
238 if ((uap->fd != -1) ||
239 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
240 return (EINVAL);
241 flags |= MAP_ANON;
242 pos = 0;
243 }
244
245 /*
246 * Align the file position to a page boundary,
247 * and save its page offset component.
248 */
249 pageoff = (pos & PAGE_MASK);
250 pos -= pageoff;
251
252 /* Adjust size for rounding (on both ends). */
253 size += pageoff; /* low end... */
254 size = (vm_size_t) round_page(size); /* hi end */
255
256 /* Ensure alignment is at least a page and fits in a pointer. */
257 align = flags & MAP_ALIGNMENT_MASK;
258 if (align != 0 && align != MAP_ALIGNED_SUPER &&
259 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
260 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
261 return (EINVAL);
262
263 /*
264 * Check for illegal addresses. Watch out for address wrap... Note
265 * that VM_*_ADDRESS are not constants due to casts (argh).
266 */
267 if (flags & MAP_FIXED) {
268 /*
269 * The specified address must have the same remainder
270 * as the file offset taken modulo PAGE_SIZE, so it
271 * should be aligned after adjustment by pageoff.
272 */
273 addr -= pageoff;
274 if (addr & PAGE_MASK)
275 return (EINVAL);
276
277 /* Address range must be all in user VM space. */
278 if (addr < vm_map_min(&vms->vm_map) ||
279 addr + size > vm_map_max(&vms->vm_map))
280 return (EINVAL);
281 if (addr + size < addr)
282 return (EINVAL);
283 } else {
284 /*
285 * XXX for non-fixed mappings where no hint is provided or
286 * the hint would fall in the potential heap space,
287 * place it after the end of the largest possible heap.
288 *
289 * There should really be a pmap call to determine a reasonable
290 * location.
291 */
292 PROC_LOCK(td->td_proc);
293 if (addr == 0 ||
294 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
295 addr < round_page((vm_offset_t)vms->vm_daddr +
296 lim_max(td->td_proc, RLIMIT_DATA))))
297 addr = round_page((vm_offset_t)vms->vm_daddr +
298 lim_max(td->td_proc, RLIMIT_DATA));
299 PROC_UNLOCK(td->td_proc);
300 }
301 if (flags & MAP_ANON) {
302 /*
303 * Mapping blank space is trivial.
304 */
305 handle = NULL;
306 handle_type = OBJT_DEFAULT;
307 maxprot = VM_PROT_ALL;
308 cap_maxprot = VM_PROT_ALL;
309 } else {
310 /*
311 * Mapping file, get fp for validation and don't let the
312 * descriptor disappear on us if we block. Check capability
313 * rights, but also return the maximum rights to be combined
314 * with maxprot later.
315 */
316 rights = CAP_MMAP;
317 if (prot & PROT_READ)
318 rights |= CAP_READ;
319 if ((flags & MAP_SHARED) != 0) {
320 if (prot & PROT_WRITE)
321 rights |= CAP_WRITE;
322 }
323 if (prot & PROT_EXEC)
324 rights |= CAP_MAPEXEC;
325 if ((error = fget_mmap(td, uap->fd, rights, &cap_maxprot,
326 &fp)) != 0)
327 goto done;
328 if (fp->f_type == DTYPE_SHM) {
329 handle = fp->f_data;
330 handle_type = OBJT_SWAP;
331 maxprot = VM_PROT_NONE;
332
333 /* FREAD should always be set. */
334 if (fp->f_flag & FREAD)
335 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
336 if (fp->f_flag & FWRITE)
337 maxprot |= VM_PROT_WRITE;
338 goto map;
339 }
340 if (fp->f_type != DTYPE_VNODE) {
341 error = ENODEV;
342 goto done;
343 }
344 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
345 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
346 /*
347 * POSIX shared-memory objects are defined to have
348 * kernel persistence, and are not defined to support
349 * read(2)/write(2) -- or even open(2). Thus, we can
350 * use MAP_ASYNC to trade on-disk coherence for speed.
351 * The shm_open(3) library routine turns on the FPOSIXSHM
352 * flag to request this behavior.
353 */
354 if (fp->f_flag & FPOSIXSHM)
355 flags |= MAP_NOSYNC;
356 #endif
357 vp = fp->f_vnode;
358 /*
359 * Ensure that file and memory protections are
360 * compatible. Note that we only worry about
361 * writability if mapping is shared; in this case,
362 * current and max prot are dictated by the open file.
363 * XXX use the vnode instead? Problem is: what
364 * credentials do we use for determination? What if
365 * proc does a setuid?
366 */
367 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
368 maxprot = VM_PROT_NONE;
369 else
370 maxprot = VM_PROT_EXECUTE;
371 if (fp->f_flag & FREAD) {
372 maxprot |= VM_PROT_READ;
373 } else if (prot & PROT_READ) {
374 error = EACCES;
375 goto done;
376 }
377 /*
378 * If we are sharing potential changes (either via
379 * MAP_SHARED or via the implicit sharing of character
380 * device mappings), and we are trying to get write
381 * permission although we opened it without asking
382 * for it, bail out.
383 */
384 if ((flags & MAP_SHARED) != 0) {
385 if ((fp->f_flag & FWRITE) != 0) {
386 maxprot |= VM_PROT_WRITE;
387 } else if ((prot & PROT_WRITE) != 0) {
388 error = EACCES;
389 goto done;
390 }
391 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
392 maxprot |= VM_PROT_WRITE;
393 cap_maxprot |= VM_PROT_WRITE;
394 }
395 handle = (void *)vp;
396 handle_type = OBJT_VNODE;
397 }
398 map:
399 td->td_fpop = fp;
400 maxprot &= cap_maxprot;
401 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
402 flags, handle_type, handle, pos);
403 td->td_fpop = NULL;
404 #ifdef HWPMC_HOOKS
405 /* inform hwpmc(4) if an executable is being mapped */
406 if (error == 0 && handle_type == OBJT_VNODE &&
407 (prot & PROT_EXEC)) {
408 pkm.pm_file = handle;
409 pkm.pm_address = (uintptr_t) addr;
410 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
411 }
412 #endif
413 if (error == 0)
414 td->td_retval[0] = (register_t) (addr + pageoff);
415 done:
416 if (fp)
417 fdrop(fp, td);
418
419 return (error);
420 }
421
422 int
freebsd6_mmap(struct thread * td,struct freebsd6_mmap_args * uap)423 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
424 {
425 struct mmap_args oargs;
426
427 oargs.addr = uap->addr;
428 oargs.len = uap->len;
429 oargs.prot = uap->prot;
430 oargs.flags = uap->flags;
431 oargs.fd = uap->fd;
432 oargs.pos = uap->pos;
433 return (sys_mmap(td, &oargs));
434 }
435
436 #ifdef COMPAT_43
437 #ifndef _SYS_SYSPROTO_H_
438 struct ommap_args {
439 caddr_t addr;
440 int len;
441 int prot;
442 int flags;
443 int fd;
444 long pos;
445 };
446 #endif
447 int
ommap(td,uap)448 ommap(td, uap)
449 struct thread *td;
450 struct ommap_args *uap;
451 {
452 struct mmap_args nargs;
453 static const char cvtbsdprot[8] = {
454 0,
455 PROT_EXEC,
456 PROT_WRITE,
457 PROT_EXEC | PROT_WRITE,
458 PROT_READ,
459 PROT_EXEC | PROT_READ,
460 PROT_WRITE | PROT_READ,
461 PROT_EXEC | PROT_WRITE | PROT_READ,
462 };
463
464 #define OMAP_ANON 0x0002
465 #define OMAP_COPY 0x0020
466 #define OMAP_SHARED 0x0010
467 #define OMAP_FIXED 0x0100
468
469 nargs.addr = uap->addr;
470 nargs.len = uap->len;
471 nargs.prot = cvtbsdprot[uap->prot & 0x7];
472 #ifdef COMPAT_FREEBSD32
473 #if defined(__amd64__) || defined(__ia64__)
474 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
475 nargs.prot != 0)
476 nargs.prot |= PROT_EXEC;
477 #endif
478 #endif
479 nargs.flags = 0;
480 if (uap->flags & OMAP_ANON)
481 nargs.flags |= MAP_ANON;
482 if (uap->flags & OMAP_COPY)
483 nargs.flags |= MAP_COPY;
484 if (uap->flags & OMAP_SHARED)
485 nargs.flags |= MAP_SHARED;
486 else
487 nargs.flags |= MAP_PRIVATE;
488 if (uap->flags & OMAP_FIXED)
489 nargs.flags |= MAP_FIXED;
490 nargs.fd = uap->fd;
491 nargs.pos = uap->pos;
492 return (sys_mmap(td, &nargs));
493 }
494 #endif /* COMPAT_43 */
495
496
497 #ifndef _SYS_SYSPROTO_H_
498 struct msync_args {
499 void *addr;
500 size_t len;
501 int flags;
502 };
503 #endif
504 /*
505 * MPSAFE
506 */
507 int
sys_msync(td,uap)508 sys_msync(td, uap)
509 struct thread *td;
510 struct msync_args *uap;
511 {
512 vm_offset_t addr;
513 vm_size_t size, pageoff;
514 int flags;
515 vm_map_t map;
516 int rv;
517
518 addr = (vm_offset_t) uap->addr;
519 size = uap->len;
520 flags = uap->flags;
521
522 pageoff = (addr & PAGE_MASK);
523 addr -= pageoff;
524 size += pageoff;
525 size = (vm_size_t) round_page(size);
526 if (addr + size < addr)
527 return (EINVAL);
528
529 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
530 return (EINVAL);
531
532 map = &td->td_proc->p_vmspace->vm_map;
533
534 /*
535 * Clean the pages and interpret the return value.
536 */
537 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
538 (flags & MS_INVALIDATE) != 0);
539 switch (rv) {
540 case KERN_SUCCESS:
541 return (0);
542 case KERN_INVALID_ADDRESS:
543 return (ENOMEM);
544 case KERN_INVALID_ARGUMENT:
545 return (EBUSY);
546 case KERN_FAILURE:
547 return (EIO);
548 default:
549 return (EINVAL);
550 }
551 }
552
553 #ifndef _SYS_SYSPROTO_H_
554 struct munmap_args {
555 void *addr;
556 size_t len;
557 };
558 #endif
559 /*
560 * MPSAFE
561 */
562 int
sys_munmap(td,uap)563 sys_munmap(td, uap)
564 struct thread *td;
565 struct munmap_args *uap;
566 {
567 #ifdef HWPMC_HOOKS
568 struct pmckern_map_out pkm;
569 vm_map_entry_t entry;
570 #endif
571 vm_offset_t addr;
572 vm_size_t size, pageoff;
573 vm_map_t map;
574
575 addr = (vm_offset_t) uap->addr;
576 size = uap->len;
577 if (size == 0)
578 return (EINVAL);
579
580 pageoff = (addr & PAGE_MASK);
581 addr -= pageoff;
582 size += pageoff;
583 size = (vm_size_t) round_page(size);
584 if (addr + size < addr)
585 return (EINVAL);
586
587 /*
588 * Check for illegal addresses. Watch out for address wrap...
589 */
590 map = &td->td_proc->p_vmspace->vm_map;
591 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
592 return (EINVAL);
593 vm_map_lock(map);
594 #ifdef HWPMC_HOOKS
595 /*
596 * Inform hwpmc if the address range being unmapped contains
597 * an executable region.
598 */
599 pkm.pm_address = (uintptr_t) NULL;
600 if (vm_map_lookup_entry(map, addr, &entry)) {
601 for (;
602 entry != &map->header && entry->start < addr + size;
603 entry = entry->next) {
604 if (vm_map_check_protection(map, entry->start,
605 entry->end, VM_PROT_EXECUTE) == TRUE) {
606 pkm.pm_address = (uintptr_t) addr;
607 pkm.pm_size = (size_t) size;
608 break;
609 }
610 }
611 }
612 #endif
613 vm_map_delete(map, addr, addr + size);
614
615 #ifdef HWPMC_HOOKS
616 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
617 vm_map_lock_downgrade(map);
618 if (pkm.pm_address != (uintptr_t) NULL)
619 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
620 vm_map_unlock_read(map);
621 #else
622 vm_map_unlock(map);
623 #endif
624 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
625 return (0);
626 }
627
628 #ifndef _SYS_SYSPROTO_H_
629 struct mprotect_args {
630 const void *addr;
631 size_t len;
632 int prot;
633 };
634 #endif
635 /*
636 * MPSAFE
637 */
638 int
sys_mprotect(td,uap)639 sys_mprotect(td, uap)
640 struct thread *td;
641 struct mprotect_args *uap;
642 {
643 vm_offset_t addr;
644 vm_size_t size, pageoff;
645 vm_prot_t prot;
646
647 addr = (vm_offset_t) uap->addr;
648 size = uap->len;
649 prot = uap->prot & VM_PROT_ALL;
650
651 pageoff = (addr & PAGE_MASK);
652 addr -= pageoff;
653 size += pageoff;
654 size = (vm_size_t) round_page(size);
655 if (addr + size < addr)
656 return (EINVAL);
657
658 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
659 addr + size, prot, FALSE)) {
660 case KERN_SUCCESS:
661 return (0);
662 case KERN_PROTECTION_FAILURE:
663 return (EACCES);
664 case KERN_RESOURCE_SHORTAGE:
665 return (ENOMEM);
666 }
667 return (EINVAL);
668 }
669
670 #ifndef _SYS_SYSPROTO_H_
671 struct minherit_args {
672 void *addr;
673 size_t len;
674 int inherit;
675 };
676 #endif
677 /*
678 * MPSAFE
679 */
680 int
sys_minherit(td,uap)681 sys_minherit(td, uap)
682 struct thread *td;
683 struct minherit_args *uap;
684 {
685 vm_offset_t addr;
686 vm_size_t size, pageoff;
687 vm_inherit_t inherit;
688
689 addr = (vm_offset_t)uap->addr;
690 size = uap->len;
691 inherit = uap->inherit;
692
693 pageoff = (addr & PAGE_MASK);
694 addr -= pageoff;
695 size += pageoff;
696 size = (vm_size_t) round_page(size);
697 if (addr + size < addr)
698 return (EINVAL);
699
700 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
701 addr + size, inherit)) {
702 case KERN_SUCCESS:
703 return (0);
704 case KERN_PROTECTION_FAILURE:
705 return (EACCES);
706 }
707 return (EINVAL);
708 }
709
710 #ifndef _SYS_SYSPROTO_H_
711 struct madvise_args {
712 void *addr;
713 size_t len;
714 int behav;
715 };
716 #endif
717
718 /*
719 * MPSAFE
720 */
721 /* ARGSUSED */
722 int
sys_madvise(td,uap)723 sys_madvise(td, uap)
724 struct thread *td;
725 struct madvise_args *uap;
726 {
727 vm_offset_t start, end;
728 vm_map_t map;
729 int flags;
730
731 /*
732 * Check for our special case, advising the swap pager we are
733 * "immortal."
734 */
735 if (uap->behav == MADV_PROTECT) {
736 flags = PPROT_SET;
737 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
738 PROC_SPROTECT, &flags));
739 }
740
741 /*
742 * Check for illegal behavior
743 */
744 if (uap->behav < 0 || uap->behav > MADV_CORE)
745 return (EINVAL);
746 /*
747 * Check for illegal addresses. Watch out for address wrap... Note
748 * that VM_*_ADDRESS are not constants due to casts (argh).
749 */
750 map = &td->td_proc->p_vmspace->vm_map;
751 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
752 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
753 return (EINVAL);
754 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
755 return (EINVAL);
756
757 /*
758 * Since this routine is only advisory, we default to conservative
759 * behavior.
760 */
761 start = trunc_page((vm_offset_t) uap->addr);
762 end = round_page((vm_offset_t) uap->addr + uap->len);
763
764 if (vm_map_madvise(map, start, end, uap->behav))
765 return (EINVAL);
766 return (0);
767 }
768
769 #ifndef _SYS_SYSPROTO_H_
770 struct mincore_args {
771 const void *addr;
772 size_t len;
773 char *vec;
774 };
775 #endif
776
777 /*
778 * MPSAFE
779 */
780 /* ARGSUSED */
781 int
sys_mincore(td,uap)782 sys_mincore(td, uap)
783 struct thread *td;
784 struct mincore_args *uap;
785 {
786 vm_offset_t addr, first_addr;
787 vm_offset_t end, cend;
788 pmap_t pmap;
789 vm_map_t map;
790 char *vec;
791 int error = 0;
792 int vecindex, lastvecindex;
793 vm_map_entry_t current;
794 vm_map_entry_t entry;
795 vm_object_t object;
796 vm_paddr_t locked_pa;
797 vm_page_t m;
798 vm_pindex_t pindex;
799 int mincoreinfo;
800 unsigned int timestamp;
801 boolean_t locked;
802
803 /*
804 * Make sure that the addresses presented are valid for user
805 * mode.
806 */
807 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
808 end = addr + (vm_size_t)round_page(uap->len);
809 map = &td->td_proc->p_vmspace->vm_map;
810 if (end > vm_map_max(map) || end < addr)
811 return (ENOMEM);
812
813 /*
814 * Address of byte vector
815 */
816 vec = uap->vec;
817
818 pmap = vmspace_pmap(td->td_proc->p_vmspace);
819
820 vm_map_lock_read(map);
821 RestartScan:
822 timestamp = map->timestamp;
823
824 if (!vm_map_lookup_entry(map, addr, &entry)) {
825 vm_map_unlock_read(map);
826 return (ENOMEM);
827 }
828
829 /*
830 * Do this on a map entry basis so that if the pages are not
831 * in the current processes address space, we can easily look
832 * up the pages elsewhere.
833 */
834 lastvecindex = -1;
835 for (current = entry;
836 (current != &map->header) && (current->start < end);
837 current = current->next) {
838
839 /*
840 * check for contiguity
841 */
842 if (current->end < end &&
843 (entry->next == &map->header ||
844 current->next->start > current->end)) {
845 vm_map_unlock_read(map);
846 return (ENOMEM);
847 }
848
849 /*
850 * ignore submaps (for now) or null objects
851 */
852 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
853 current->object.vm_object == NULL)
854 continue;
855
856 /*
857 * limit this scan to the current map entry and the
858 * limits for the mincore call
859 */
860 if (addr < current->start)
861 addr = current->start;
862 cend = current->end;
863 if (cend > end)
864 cend = end;
865
866 /*
867 * scan this entry one page at a time
868 */
869 while (addr < cend) {
870 /*
871 * Check pmap first, it is likely faster, also
872 * it can provide info as to whether we are the
873 * one referencing or modifying the page.
874 */
875 object = NULL;
876 locked_pa = 0;
877 retry:
878 m = NULL;
879 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
880 if (locked_pa != 0) {
881 /*
882 * The page is mapped by this process but not
883 * both accessed and modified. It is also
884 * managed. Acquire the object lock so that
885 * other mappings might be examined.
886 */
887 m = PHYS_TO_VM_PAGE(locked_pa);
888 if (m->object != object) {
889 if (object != NULL)
890 VM_OBJECT_UNLOCK(object);
891 object = m->object;
892 locked = VM_OBJECT_TRYLOCK(object);
893 vm_page_unlock(m);
894 if (!locked) {
895 VM_OBJECT_LOCK(object);
896 vm_page_lock(m);
897 goto retry;
898 }
899 } else
900 vm_page_unlock(m);
901 KASSERT(m->valid == VM_PAGE_BITS_ALL,
902 ("mincore: page %p is mapped but invalid",
903 m));
904 } else if (mincoreinfo == 0) {
905 /*
906 * The page is not mapped by this process. If
907 * the object implements managed pages, then
908 * determine if the page is resident so that
909 * the mappings might be examined.
910 */
911 if (current->object.vm_object != object) {
912 if (object != NULL)
913 VM_OBJECT_UNLOCK(object);
914 object = current->object.vm_object;
915 VM_OBJECT_LOCK(object);
916 }
917 if (object->type == OBJT_DEFAULT ||
918 object->type == OBJT_SWAP ||
919 object->type == OBJT_VNODE) {
920 pindex = OFF_TO_IDX(current->offset +
921 (addr - current->start));
922 m = vm_page_lookup(object, pindex);
923 if (m == NULL &&
924 vm_page_is_cached(object, pindex))
925 mincoreinfo = MINCORE_INCORE;
926 if (m != NULL && m->valid == 0)
927 m = NULL;
928 if (m != NULL)
929 mincoreinfo = MINCORE_INCORE;
930 }
931 }
932 if (m != NULL) {
933 /* Examine other mappings to the page. */
934 if (m->dirty == 0 && pmap_is_modified(m))
935 vm_page_dirty(m);
936 if (m->dirty != 0)
937 mincoreinfo |= MINCORE_MODIFIED_OTHER;
938 /*
939 * The first test for PGA_REFERENCED is an
940 * optimization. The second test is
941 * required because a concurrent pmap
942 * operation could clear the last reference
943 * and set PGA_REFERENCED before the call to
944 * pmap_is_referenced().
945 */
946 if ((m->aflags & PGA_REFERENCED) != 0 ||
947 pmap_is_referenced(m) ||
948 (m->aflags & PGA_REFERENCED) != 0)
949 mincoreinfo |= MINCORE_REFERENCED_OTHER;
950 }
951 if (object != NULL)
952 VM_OBJECT_UNLOCK(object);
953
954 /*
955 * subyte may page fault. In case it needs to modify
956 * the map, we release the lock.
957 */
958 vm_map_unlock_read(map);
959
960 /*
961 * calculate index into user supplied byte vector
962 */
963 vecindex = OFF_TO_IDX(addr - first_addr);
964
965 /*
966 * If we have skipped map entries, we need to make sure that
967 * the byte vector is zeroed for those skipped entries.
968 */
969 while ((lastvecindex + 1) < vecindex) {
970 ++lastvecindex;
971 error = subyte(vec + lastvecindex, 0);
972 if (error) {
973 error = EFAULT;
974 goto done2;
975 }
976 }
977
978 /*
979 * Pass the page information to the user
980 */
981 error = subyte(vec + vecindex, mincoreinfo);
982 if (error) {
983 error = EFAULT;
984 goto done2;
985 }
986
987 /*
988 * If the map has changed, due to the subyte, the previous
989 * output may be invalid.
990 */
991 vm_map_lock_read(map);
992 if (timestamp != map->timestamp)
993 goto RestartScan;
994
995 lastvecindex = vecindex;
996 addr += PAGE_SIZE;
997 }
998 }
999
1000 /*
1001 * subyte may page fault. In case it needs to modify
1002 * the map, we release the lock.
1003 */
1004 vm_map_unlock_read(map);
1005
1006 /*
1007 * Zero the last entries in the byte vector.
1008 */
1009 vecindex = OFF_TO_IDX(end - first_addr);
1010 while ((lastvecindex + 1) < vecindex) {
1011 ++lastvecindex;
1012 error = subyte(vec + lastvecindex, 0);
1013 if (error) {
1014 error = EFAULT;
1015 goto done2;
1016 }
1017 }
1018
1019 /*
1020 * If the map has changed, due to the subyte, the previous
1021 * output may be invalid.
1022 */
1023 vm_map_lock_read(map);
1024 if (timestamp != map->timestamp)
1025 goto RestartScan;
1026 vm_map_unlock_read(map);
1027 done2:
1028 return (error);
1029 }
1030
1031 #ifndef _SYS_SYSPROTO_H_
1032 struct mlock_args {
1033 const void *addr;
1034 size_t len;
1035 };
1036 #endif
1037 /*
1038 * MPSAFE
1039 */
1040 int
sys_mlock(td,uap)1041 sys_mlock(td, uap)
1042 struct thread *td;
1043 struct mlock_args *uap;
1044 {
1045 struct proc *proc;
1046 vm_offset_t addr, end, last, start;
1047 vm_size_t npages, size;
1048 vm_map_t map;
1049 unsigned long nsize;
1050 int error;
1051
1052 error = priv_check(td, PRIV_VM_MLOCK);
1053 if (error)
1054 return (error);
1055 addr = (vm_offset_t)uap->addr;
1056 size = uap->len;
1057 last = addr + size;
1058 start = trunc_page(addr);
1059 end = round_page(last);
1060 if (last < addr || end < addr)
1061 return (EINVAL);
1062 npages = atop(end - start);
1063 if (npages > vm_page_max_wired)
1064 return (ENOMEM);
1065 proc = td->td_proc;
1066 map = &proc->p_vmspace->vm_map;
1067 PROC_LOCK(proc);
1068 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1069 if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1070 PROC_UNLOCK(proc);
1071 return (ENOMEM);
1072 }
1073 PROC_UNLOCK(proc);
1074 if (npages + cnt.v_wire_count > vm_page_max_wired)
1075 return (EAGAIN);
1076 #ifdef RACCT
1077 PROC_LOCK(proc);
1078 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1079 PROC_UNLOCK(proc);
1080 if (error != 0)
1081 return (ENOMEM);
1082 #endif
1083 error = vm_map_wire(map, start, end,
1084 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1085 #ifdef RACCT
1086 if (error != KERN_SUCCESS) {
1087 PROC_LOCK(proc);
1088 racct_set(proc, RACCT_MEMLOCK,
1089 ptoa(pmap_wired_count(map->pmap)));
1090 PROC_UNLOCK(proc);
1091 }
1092 #endif
1093 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1094 }
1095
1096 #ifndef _SYS_SYSPROTO_H_
1097 struct mlockall_args {
1098 int how;
1099 };
1100 #endif
1101
1102 /*
1103 * MPSAFE
1104 */
1105 int
sys_mlockall(td,uap)1106 sys_mlockall(td, uap)
1107 struct thread *td;
1108 struct mlockall_args *uap;
1109 {
1110 vm_map_t map;
1111 int error;
1112
1113 map = &td->td_proc->p_vmspace->vm_map;
1114 error = priv_check(td, PRIV_VM_MLOCK);
1115 if (error)
1116 return (error);
1117
1118 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1119 return (EINVAL);
1120
1121 /*
1122 * If wiring all pages in the process would cause it to exceed
1123 * a hard resource limit, return ENOMEM.
1124 */
1125 if (!old_mlock && uap->how & MCL_CURRENT) {
1126 PROC_LOCK(td->td_proc);
1127 if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1128 PROC_UNLOCK(td->td_proc);
1129 return (ENOMEM);
1130 }
1131 PROC_UNLOCK(td->td_proc);
1132 }
1133 #ifdef RACCT
1134 PROC_LOCK(td->td_proc);
1135 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1136 PROC_UNLOCK(td->td_proc);
1137 if (error != 0)
1138 return (ENOMEM);
1139 #endif
1140
1141 if (uap->how & MCL_FUTURE) {
1142 vm_map_lock(map);
1143 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1144 vm_map_unlock(map);
1145 error = 0;
1146 }
1147
1148 if (uap->how & MCL_CURRENT) {
1149 /*
1150 * P1003.1-2001 mandates that all currently mapped pages
1151 * will be memory resident and locked (wired) upon return
1152 * from mlockall(). vm_map_wire() will wire pages, by
1153 * calling vm_fault_wire() for each page in the region.
1154 */
1155 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1156 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1157 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1158 }
1159 #ifdef RACCT
1160 if (error != KERN_SUCCESS) {
1161 PROC_LOCK(td->td_proc);
1162 racct_set(td->td_proc, RACCT_MEMLOCK,
1163 ptoa(pmap_wired_count(map->pmap)));
1164 PROC_UNLOCK(td->td_proc);
1165 }
1166 #endif
1167
1168 return (error);
1169 }
1170
1171 #ifndef _SYS_SYSPROTO_H_
1172 struct munlockall_args {
1173 register_t dummy;
1174 };
1175 #endif
1176
1177 /*
1178 * MPSAFE
1179 */
1180 int
sys_munlockall(td,uap)1181 sys_munlockall(td, uap)
1182 struct thread *td;
1183 struct munlockall_args *uap;
1184 {
1185 vm_map_t map;
1186 int error;
1187
1188 map = &td->td_proc->p_vmspace->vm_map;
1189 error = priv_check(td, PRIV_VM_MUNLOCK);
1190 if (error)
1191 return (error);
1192
1193 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1194 vm_map_lock(map);
1195 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1196 vm_map_unlock(map);
1197
1198 /* Forcibly unwire all pages. */
1199 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1200 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1201 #ifdef RACCT
1202 if (error == KERN_SUCCESS) {
1203 PROC_LOCK(td->td_proc);
1204 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1205 PROC_UNLOCK(td->td_proc);
1206 }
1207 #endif
1208
1209 return (error);
1210 }
1211
1212 #ifndef _SYS_SYSPROTO_H_
1213 struct munlock_args {
1214 const void *addr;
1215 size_t len;
1216 };
1217 #endif
1218 /*
1219 * MPSAFE
1220 */
1221 int
sys_munlock(td,uap)1222 sys_munlock(td, uap)
1223 struct thread *td;
1224 struct munlock_args *uap;
1225 {
1226 vm_offset_t addr, end, last, start;
1227 vm_size_t size;
1228 #ifdef RACCT
1229 vm_map_t map;
1230 #endif
1231 int error;
1232
1233 error = priv_check(td, PRIV_VM_MUNLOCK);
1234 if (error)
1235 return (error);
1236 addr = (vm_offset_t)uap->addr;
1237 size = uap->len;
1238 last = addr + size;
1239 start = trunc_page(addr);
1240 end = round_page(last);
1241 if (last < addr || end < addr)
1242 return (EINVAL);
1243 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1244 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1245 #ifdef RACCT
1246 if (error == KERN_SUCCESS) {
1247 PROC_LOCK(td->td_proc);
1248 map = &td->td_proc->p_vmspace->vm_map;
1249 racct_set(td->td_proc, RACCT_MEMLOCK,
1250 ptoa(pmap_wired_count(map->pmap)));
1251 PROC_UNLOCK(td->td_proc);
1252 }
1253 #endif
1254 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1255 }
1256
1257 /*
1258 * vm_mmap_vnode()
1259 *
1260 * Helper function for vm_mmap. Perform sanity check specific for mmap
1261 * operations on vnodes.
1262 *
1263 * For VCHR vnodes, the vnode lock is held over the call to
1264 * vm_mmap_cdev() to keep vp->v_rdev valid.
1265 */
1266 int
vm_mmap_vnode(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct vnode * vp,vm_ooffset_t * foffp,vm_object_t * objp,boolean_t * writecounted)1267 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1268 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1269 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1270 boolean_t *writecounted)
1271 {
1272 struct vattr va;
1273 vm_object_t obj;
1274 vm_offset_t foff;
1275 struct mount *mp;
1276 struct ucred *cred;
1277 int error, flags, locktype, vfslocked;
1278
1279 mp = vp->v_mount;
1280 cred = td->td_ucred;
1281 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1282 locktype = LK_EXCLUSIVE;
1283 else
1284 locktype = LK_SHARED;
1285 vfslocked = VFS_LOCK_GIANT(mp);
1286 if ((error = vget(vp, locktype, td)) != 0) {
1287 VFS_UNLOCK_GIANT(vfslocked);
1288 return (error);
1289 }
1290 foff = *foffp;
1291 flags = *flagsp;
1292 obj = vp->v_object;
1293 if (vp->v_type == VREG) {
1294 /*
1295 * Get the proper underlying object
1296 */
1297 if (obj == NULL) {
1298 error = EINVAL;
1299 goto done;
1300 }
1301 if (obj->handle != vp) {
1302 vput(vp);
1303 vp = (struct vnode *)obj->handle;
1304 /*
1305 * Bypass filesystems obey the mpsafety of the
1306 * underlying fs.
1307 */
1308 error = vget(vp, locktype, td);
1309 if (error != 0) {
1310 VFS_UNLOCK_GIANT(vfslocked);
1311 return (error);
1312 }
1313 }
1314 if (locktype == LK_EXCLUSIVE) {
1315 *writecounted = TRUE;
1316 vnode_pager_update_writecount(obj, 0, objsize);
1317 }
1318 } else if (vp->v_type == VCHR) {
1319 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1320 vp->v_rdev, foffp, objp);
1321 if (error == 0)
1322 goto mark_atime;
1323 goto done;
1324 } else {
1325 error = EINVAL;
1326 goto done;
1327 }
1328 if ((error = VOP_GETATTR(vp, &va, cred)))
1329 goto done;
1330 #ifdef MAC
1331 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1332 if (error != 0)
1333 goto done;
1334 #endif
1335 if ((flags & MAP_SHARED) != 0) {
1336 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1337 if (prot & PROT_WRITE) {
1338 error = EPERM;
1339 goto done;
1340 }
1341 *maxprotp &= ~VM_PROT_WRITE;
1342 }
1343 }
1344 /*
1345 * If it is a regular file without any references
1346 * we do not need to sync it.
1347 * Adjust object size to be the size of actual file.
1348 */
1349 objsize = round_page(va.va_size);
1350 if (va.va_nlink == 0)
1351 flags |= MAP_NOSYNC;
1352 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, cred);
1353 if (obj == NULL) {
1354 error = ENOMEM;
1355 goto done;
1356 }
1357 *objp = obj;
1358 *flagsp = flags;
1359
1360 mark_atime:
1361 vfs_mark_atime(vp, cred);
1362
1363 done:
1364 if (error != 0 && *writecounted) {
1365 *writecounted = FALSE;
1366 vnode_pager_update_writecount(obj, objsize, 0);
1367 }
1368 vput(vp);
1369 VFS_UNLOCK_GIANT(vfslocked);
1370 return (error);
1371 }
1372
1373 /*
1374 * vm_mmap_cdev()
1375 *
1376 * MPSAFE
1377 *
1378 * Helper function for vm_mmap. Perform sanity check specific for mmap
1379 * operations on cdevs.
1380 */
1381 int
vm_mmap_cdev(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct cdev * cdev,vm_ooffset_t * foff,vm_object_t * objp)1382 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1383 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1384 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1385 {
1386 vm_object_t obj;
1387 struct cdevsw *dsw;
1388 int error, flags, ref;
1389
1390 flags = *flagsp;
1391
1392 dsw = dev_refthread(cdev, &ref);
1393 if (dsw == NULL)
1394 return (ENXIO);
1395 if (dsw->d_flags & D_MMAP_ANON) {
1396 dev_relthread(cdev, ref);
1397 *maxprotp = VM_PROT_ALL;
1398 *flagsp |= MAP_ANON;
1399 return (0);
1400 }
1401 /*
1402 * cdevs do not provide private mappings of any kind.
1403 */
1404 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1405 (prot & PROT_WRITE) != 0) {
1406 dev_relthread(cdev, ref);
1407 return (EACCES);
1408 }
1409 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1410 dev_relthread(cdev, ref);
1411 return (EINVAL);
1412 }
1413 /*
1414 * Force device mappings to be shared.
1415 */
1416 flags |= MAP_SHARED;
1417 #ifdef MAC_XXX
1418 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1419 if (error != 0) {
1420 dev_relthread(cdev, ref);
1421 return (error);
1422 }
1423 #endif
1424 /*
1425 * First, try d_mmap_single(). If that is not implemented
1426 * (returns ENODEV), fall back to using the device pager.
1427 * Note that d_mmap_single() must return a reference to the
1428 * object (it needs to bump the reference count of the object
1429 * it returns somehow).
1430 *
1431 * XXX assumes VM_PROT_* == PROT_*
1432 */
1433 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1434 dev_relthread(cdev, ref);
1435 if (error != ENODEV)
1436 return (error);
1437 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1438 td->td_ucred);
1439 if (obj == NULL)
1440 return (EINVAL);
1441 *objp = obj;
1442 *flagsp = flags;
1443 return (0);
1444 }
1445
1446 /*
1447 * vm_mmap_shm()
1448 *
1449 * MPSAFE
1450 *
1451 * Helper function for vm_mmap. Perform sanity check specific for mmap
1452 * operations on shm file descriptors.
1453 */
1454 int
vm_mmap_shm(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct shmfd * shmfd,vm_ooffset_t foff,vm_object_t * objp)1455 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1456 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1457 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1458 {
1459 int error;
1460
1461 if ((*flagsp & MAP_SHARED) != 0 &&
1462 (*maxprotp & VM_PROT_WRITE) == 0 &&
1463 (prot & PROT_WRITE) != 0)
1464 return (EACCES);
1465 #ifdef MAC
1466 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1467 if (error != 0)
1468 return (error);
1469 #endif
1470 error = shm_mmap(shmfd, objsize, foff, objp);
1471 if (error)
1472 return (error);
1473 return (0);
1474 }
1475
1476 /*
1477 * vm_mmap()
1478 *
1479 * MPSAFE
1480 *
1481 * Internal version of mmap. Currently used by mmap, exec, and sys5
1482 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1483 */
1484 int
vm_mmap(vm_map_t map,vm_offset_t * addr,vm_size_t size,vm_prot_t prot,vm_prot_t maxprot,int flags,objtype_t handle_type,void * handle,vm_ooffset_t foff)1485 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1486 vm_prot_t maxprot, int flags,
1487 objtype_t handle_type, void *handle,
1488 vm_ooffset_t foff)
1489 {
1490 boolean_t fitit;
1491 vm_object_t object = NULL;
1492 struct thread *td = curthread;
1493 int docow, error, findspace, rv;
1494 boolean_t writecounted;
1495
1496 if (size == 0)
1497 return (0);
1498
1499 size = round_page(size);
1500
1501 PROC_LOCK(td->td_proc);
1502 if (td->td_proc->p_vmspace->vm_map.size + size >
1503 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1504 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1505 if (ptoa(pmap_wired_count(map->pmap)) + size >
1506 lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1507 racct_set_force(td->td_proc, RACCT_VMEM,
1508 map->size);
1509 PROC_UNLOCK(td->td_proc);
1510 return (ENOMEM);
1511 }
1512 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1513 ptoa(pmap_wired_count(map->pmap)) + size);
1514 if (error != 0) {
1515 racct_set_force(td->td_proc, RACCT_VMEM,
1516 map->size);
1517 PROC_UNLOCK(td->td_proc);
1518 return (error);
1519 }
1520 }
1521 PROC_UNLOCK(td->td_proc);
1522 return (ENOMEM);
1523 }
1524 if (racct_set(td->td_proc, RACCT_VMEM,
1525 td->td_proc->p_vmspace->vm_map.size + size)) {
1526 PROC_UNLOCK(td->td_proc);
1527 return (ENOMEM);
1528 }
1529 PROC_UNLOCK(td->td_proc);
1530
1531 /*
1532 * We currently can only deal with page aligned file offsets.
1533 * The check is here rather than in the syscall because the
1534 * kernel calls this function internally for other mmaping
1535 * operations (such as in exec) and non-aligned offsets will
1536 * cause pmap inconsistencies...so we want to be sure to
1537 * disallow this in all cases.
1538 */
1539 if (foff & PAGE_MASK)
1540 return (EINVAL);
1541
1542 if ((flags & MAP_FIXED) == 0) {
1543 fitit = TRUE;
1544 *addr = round_page(*addr);
1545 } else {
1546 if (*addr != trunc_page(*addr))
1547 return (EINVAL);
1548 fitit = FALSE;
1549 }
1550 writecounted = FALSE;
1551
1552 /*
1553 * Lookup/allocate object.
1554 */
1555 switch (handle_type) {
1556 case OBJT_DEVICE:
1557 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1558 handle, &foff, &object);
1559 break;
1560 case OBJT_VNODE:
1561 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1562 handle, &foff, &object, &writecounted);
1563 break;
1564 case OBJT_SWAP:
1565 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1566 handle, foff, &object);
1567 break;
1568 case OBJT_DEFAULT:
1569 if (handle == NULL) {
1570 error = 0;
1571 break;
1572 }
1573 /* FALLTHROUGH */
1574 default:
1575 error = EINVAL;
1576 break;
1577 }
1578 if (error)
1579 return (error);
1580 if (flags & MAP_ANON) {
1581 object = NULL;
1582 docow = 0;
1583 /*
1584 * Unnamed anonymous regions always start at 0.
1585 */
1586 if (handle == 0)
1587 foff = 0;
1588 } else if (flags & MAP_PREFAULT_READ)
1589 docow = MAP_PREFAULT;
1590 else
1591 docow = MAP_PREFAULT_PARTIAL;
1592
1593 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1594 docow |= MAP_COPY_ON_WRITE;
1595 if (flags & MAP_NOSYNC)
1596 docow |= MAP_DISABLE_SYNCER;
1597 if (flags & MAP_NOCORE)
1598 docow |= MAP_DISABLE_COREDUMP;
1599 /* Shared memory is also shared with children. */
1600 if (flags & MAP_SHARED)
1601 docow |= MAP_INHERIT_SHARE;
1602 if (writecounted)
1603 docow |= MAP_VN_WRITECOUNT;
1604
1605 if (flags & MAP_STACK)
1606 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1607 docow | MAP_STACK_GROWS_DOWN);
1608 else if (fitit) {
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);
1614 else
1615 findspace = VMFS_OPTIMAL_SPACE;
1616 rv = vm_map_find(map, object, foff, addr, size, findspace,
1617 prot, maxprot, docow);
1618 } else
1619 rv = vm_map_fixed(map, object, foff, *addr, size,
1620 prot, maxprot, docow);
1621
1622 if (rv == KERN_SUCCESS) {
1623 /*
1624 * If the process has requested that all future mappings
1625 * be wired, then heed this.
1626 */
1627 if (map->flags & MAP_WIREFUTURE) {
1628 vm_map_wire(map, *addr, *addr + size,
1629 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1630 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1631 }
1632 } else {
1633 /*
1634 * If this mapping was accounted for in the vnode's
1635 * writecount, then undo that now.
1636 */
1637 if (writecounted)
1638 vnode_pager_release_writecount(object, 0, size);
1639 /*
1640 * Lose the object reference. Will destroy the
1641 * object if it's an unnamed anonymous mapping
1642 * or named anonymous without other references.
1643 */
1644 vm_object_deallocate(object);
1645 }
1646 return (vm_mmap_to_errno(rv));
1647 }
1648
1649 /*
1650 * Translate a Mach VM return code to zero on success or the appropriate errno
1651 * on failure.
1652 */
1653 int
vm_mmap_to_errno(int rv)1654 vm_mmap_to_errno(int rv)
1655 {
1656
1657 switch (rv) {
1658 case KERN_SUCCESS:
1659 return (0);
1660 case KERN_INVALID_ADDRESS:
1661 case KERN_NO_SPACE:
1662 return (ENOMEM);
1663 case KERN_PROTECTION_FAILURE:
1664 return (EACCES);
1665 default:
1666 return (EINVAL);
1667 }
1668 }
1669