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("$FreeBSD$");
45
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48 #include "opt_vm.h"
49
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/capsicum.h>
53 #include <sys/kernel.h>
54 #include <sys/lock.h>
55 #include <sys/mutex.h>
56 #include <sys/sysproto.h>
57 #include <sys/filedesc.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/procctl.h>
61 #include <sys/racct.h>
62 #include <sys/resource.h>
63 #include <sys/resourcevar.h>
64 #include <sys/rwlock.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/fcntl.h>
68 #include <sys/file.h>
69 #include <sys/mman.h>
70 #include <sys/mount.h>
71 #include <sys/conf.h>
72 #include <sys/stat.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
76
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
79
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_extern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vnode_pager.h>
91
92 #ifdef HWPMC_HOOKS
93 #include <sys/pmckern.h>
94 #endif
95
96 int old_mlock = 0;
97 SYSCTL_INT(_vm, OID_AUTO, old_mlock, CTLFLAG_RWTUN, &old_mlock, 0,
98 "Do not apply RLIMIT_MEMLOCK on mlockall");
99 static int mincore_mapped = 1;
100 SYSCTL_INT(_vm, OID_AUTO, mincore_mapped, CTLFLAG_RWTUN, &mincore_mapped, 0,
101 "mincore reports mappings, not residency");
102
103 #ifdef MAP_32BIT
104 #define MAP_32BIT_MAX_ADDR ((vm_offset_t)1 << 31)
105 #endif
106
107 #ifndef _SYS_SYSPROTO_H_
108 struct sbrk_args {
109 int incr;
110 };
111 #endif
112
113 int
sys_sbrk(struct thread * td,struct sbrk_args * uap)114 sys_sbrk(struct thread *td, struct sbrk_args *uap)
115 {
116 /* Not yet implemented */
117 return (EOPNOTSUPP);
118 }
119
120 #ifndef _SYS_SYSPROTO_H_
121 struct sstk_args {
122 int incr;
123 };
124 #endif
125
126 int
sys_sstk(struct thread * td,struct sstk_args * uap)127 sys_sstk(struct thread *td, struct sstk_args *uap)
128 {
129 /* Not yet implemented */
130 return (EOPNOTSUPP);
131 }
132
133 #if defined(COMPAT_43)
134 #ifndef _SYS_SYSPROTO_H_
135 struct getpagesize_args {
136 int dummy;
137 };
138 #endif
139
140 int
ogetpagesize(struct thread * td,struct getpagesize_args * uap)141 ogetpagesize(struct thread *td, struct getpagesize_args *uap)
142 {
143
144 td->td_retval[0] = PAGE_SIZE;
145 return (0);
146 }
147 #endif /* COMPAT_43 */
148
149
150 /*
151 * Memory Map (mmap) system call. Note that the file offset
152 * and address are allowed to be NOT page aligned, though if
153 * the MAP_FIXED flag it set, both must have the same remainder
154 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
155 * page-aligned, the actual mapping starts at trunc_page(addr)
156 * and the return value is adjusted up by the page offset.
157 *
158 * Generally speaking, only character devices which are themselves
159 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
160 * there would be no cache coherency between a descriptor and a VM mapping
161 * both to the same character device.
162 */
163 #ifndef _SYS_SYSPROTO_H_
164 struct mmap_args {
165 void *addr;
166 size_t len;
167 int prot;
168 int flags;
169 int fd;
170 long pad;
171 off_t pos;
172 };
173 #endif
174
175 int
sys_mmap(struct thread * td,struct mmap_args * uap)176 sys_mmap(struct thread *td, struct mmap_args *uap)
177 {
178
179 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
180 uap->flags, uap->fd, uap->pos));
181 }
182
183 int
kern_mmap(struct thread * td,uintptr_t addr0,size_t size,int prot,int flags,int fd,off_t pos)184 kern_mmap(struct thread *td, uintptr_t addr0, size_t size, int prot, int flags,
185 int fd, off_t pos)
186 {
187
188 return (kern_mmap_fpcheck(td, addr0, size, prot, flags, fd, pos, NULL));
189 }
190
191 /*
192 * When mmap'ing a file, check_fp_fn may be used for the caller to do any
193 * last-minute validation based on the referenced file in a non-racy way.
194 */
195 int
kern_mmap_fpcheck(struct thread * td,uintptr_t addr0,size_t size,int prot,int flags,int fd,off_t pos,mmap_check_fp_fn check_fp_fn)196 kern_mmap_fpcheck(struct thread *td, uintptr_t addr0, size_t size, int prot,
197 int flags, int fd, off_t pos, mmap_check_fp_fn check_fp_fn)
198 {
199 struct vmspace *vms;
200 struct file *fp;
201 vm_offset_t addr;
202 vm_size_t pageoff;
203 vm_prot_t cap_maxprot;
204 int align, error;
205 cap_rights_t rights;
206
207 vms = td->td_proc->p_vmspace;
208 fp = NULL;
209 AUDIT_ARG_FD(fd);
210 addr = addr0;
211
212 /*
213 * Ignore old flags that used to be defined but did not do anything.
214 */
215 flags &= ~(MAP_RESERVED0020 | MAP_RESERVED0040);
216
217 /*
218 * Enforce the constraints.
219 * Mapping of length 0 is only allowed for old binaries.
220 * Anonymous mapping shall specify -1 as filedescriptor and
221 * zero position for new code. Be nice to ancient a.out
222 * binaries and correct pos for anonymous mapping, since old
223 * ld.so sometimes issues anonymous map requests with non-zero
224 * pos.
225 */
226 if (!SV_CURPROC_FLAG(SV_AOUT)) {
227 if ((size == 0 && curproc->p_osrel >= P_OSREL_MAP_ANON) ||
228 ((flags & MAP_ANON) != 0 && (fd != -1 || pos != 0)))
229 return (EINVAL);
230 } else {
231 if ((flags & MAP_ANON) != 0)
232 pos = 0;
233 }
234
235 if (flags & MAP_STACK) {
236 if ((fd != -1) ||
237 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
238 return (EINVAL);
239 flags |= MAP_ANON;
240 pos = 0;
241 }
242 if ((flags & ~(MAP_SHARED | MAP_PRIVATE | MAP_FIXED | MAP_HASSEMAPHORE |
243 MAP_STACK | MAP_NOSYNC | MAP_ANON | MAP_EXCL | MAP_NOCORE |
244 MAP_PREFAULT_READ | MAP_GUARD |
245 #ifdef MAP_32BIT
246 MAP_32BIT |
247 #endif
248 MAP_ALIGNMENT_MASK)) != 0)
249 return (EINVAL);
250 if ((flags & (MAP_EXCL | MAP_FIXED)) == MAP_EXCL)
251 return (EINVAL);
252 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == (MAP_SHARED | MAP_PRIVATE))
253 return (EINVAL);
254 if (prot != PROT_NONE &&
255 (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC)) != 0)
256 return (EINVAL);
257 if ((flags & MAP_GUARD) != 0 && (prot != PROT_NONE || fd != -1 ||
258 pos != 0 || (flags & (MAP_SHARED | MAP_PRIVATE | MAP_PREFAULT |
259 MAP_PREFAULT_READ | MAP_ANON | MAP_STACK)) != 0))
260 return (EINVAL);
261
262 /*
263 * Align the file position to a page boundary,
264 * and save its page offset component.
265 */
266 pageoff = (pos & PAGE_MASK);
267 pos -= pageoff;
268
269 /* Adjust size for rounding (on both ends). */
270 size += pageoff; /* low end... */
271 size = (vm_size_t) round_page(size); /* hi end */
272
273 /* Ensure alignment is at least a page and fits in a pointer. */
274 align = flags & MAP_ALIGNMENT_MASK;
275 if (align != 0 && align != MAP_ALIGNED_SUPER &&
276 (align >> MAP_ALIGNMENT_SHIFT >= sizeof(void *) * NBBY ||
277 align >> MAP_ALIGNMENT_SHIFT < PAGE_SHIFT))
278 return (EINVAL);
279
280 /*
281 * Check for illegal addresses. Watch out for address wrap... Note
282 * that VM_*_ADDRESS are not constants due to casts (argh).
283 */
284 if (flags & MAP_FIXED) {
285 /*
286 * The specified address must have the same remainder
287 * as the file offset taken modulo PAGE_SIZE, so it
288 * should be aligned after adjustment by pageoff.
289 */
290 addr -= pageoff;
291 if (addr & PAGE_MASK)
292 return (EINVAL);
293
294 /* Address range must be all in user VM space. */
295 if (addr < vm_map_min(&vms->vm_map) ||
296 addr + size > vm_map_max(&vms->vm_map))
297 return (EINVAL);
298 if (addr + size < addr)
299 return (EINVAL);
300 #ifdef MAP_32BIT
301 if (flags & MAP_32BIT && addr + size > MAP_32BIT_MAX_ADDR)
302 return (EINVAL);
303 } else if (flags & MAP_32BIT) {
304 /*
305 * For MAP_32BIT, override the hint if it is too high and
306 * do not bother moving the mapping past the heap (since
307 * the heap is usually above 2GB).
308 */
309 if (addr + size > MAP_32BIT_MAX_ADDR)
310 addr = 0;
311 #endif
312 } else {
313 /*
314 * XXX for non-fixed mappings where no hint is provided or
315 * the hint would fall in the potential heap space,
316 * place it after the end of the largest possible heap.
317 *
318 * There should really be a pmap call to determine a reasonable
319 * location.
320 */
321 if (addr == 0 ||
322 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
323 addr < round_page((vm_offset_t)vms->vm_daddr +
324 lim_max(td, RLIMIT_DATA))))
325 addr = round_page((vm_offset_t)vms->vm_daddr +
326 lim_max(td, RLIMIT_DATA));
327 }
328 if (size == 0) {
329 /*
330 * Return success without mapping anything for old
331 * binaries that request a page-aligned mapping of
332 * length 0. For modern binaries, this function
333 * returns an error earlier.
334 */
335 error = 0;
336 } else if ((flags & MAP_GUARD) != 0) {
337 error = vm_mmap_object(&vms->vm_map, &addr, size, VM_PROT_NONE,
338 VM_PROT_NONE, flags, NULL, pos, FALSE, td);
339 } else if ((flags & MAP_ANON) != 0) {
340 /*
341 * Mapping blank space is trivial.
342 *
343 * This relies on VM_PROT_* matching PROT_*.
344 */
345 error = vm_mmap_object(&vms->vm_map, &addr, size, prot,
346 VM_PROT_ALL, flags, NULL, pos, FALSE, td);
347 } else {
348 /*
349 * Mapping file, get fp for validation and don't let the
350 * descriptor disappear on us if we block. Check capability
351 * rights, but also return the maximum rights to be combined
352 * with maxprot later.
353 */
354 cap_rights_init(&rights, CAP_MMAP);
355 if (prot & PROT_READ)
356 cap_rights_set(&rights, CAP_MMAP_R);
357 if ((flags & MAP_SHARED) != 0) {
358 if (prot & PROT_WRITE)
359 cap_rights_set(&rights, CAP_MMAP_W);
360 }
361 if (prot & PROT_EXEC)
362 cap_rights_set(&rights, CAP_MMAP_X);
363 error = fget_mmap(td, fd, &rights, &cap_maxprot, &fp);
364 if (error != 0)
365 goto done;
366 if ((flags & (MAP_SHARED | MAP_PRIVATE)) == 0 &&
367 td->td_proc->p_osrel >= P_OSREL_MAP_FSTRICT) {
368 error = EINVAL;
369 goto done;
370 }
371 if (check_fp_fn != NULL) {
372 error = check_fp_fn(fp, prot, cap_maxprot, flags);
373 if (error != 0)
374 goto done;
375 }
376 /* This relies on VM_PROT_* matching PROT_*. */
377 error = fo_mmap(fp, &vms->vm_map, &addr, size, prot,
378 cap_maxprot, flags, pos, td);
379 }
380
381 if (error == 0)
382 td->td_retval[0] = (register_t) (addr + pageoff);
383 done:
384 if (fp)
385 fdrop(fp, td);
386
387 return (error);
388 }
389
390 #if defined(COMPAT_FREEBSD6)
391 int
freebsd6_mmap(struct thread * td,struct freebsd6_mmap_args * uap)392 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
393 {
394
395 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, uap->prot,
396 uap->flags, uap->fd, uap->pos));
397 }
398 #endif
399
400 #ifdef COMPAT_43
401 #ifndef _SYS_SYSPROTO_H_
402 struct ommap_args {
403 caddr_t addr;
404 int len;
405 int prot;
406 int flags;
407 int fd;
408 long pos;
409 };
410 #endif
411 int
ommap(struct thread * td,struct ommap_args * uap)412 ommap(struct thread *td, struct ommap_args *uap)
413 {
414 static const char cvtbsdprot[8] = {
415 0,
416 PROT_EXEC,
417 PROT_WRITE,
418 PROT_EXEC | PROT_WRITE,
419 PROT_READ,
420 PROT_EXEC | PROT_READ,
421 PROT_WRITE | PROT_READ,
422 PROT_EXEC | PROT_WRITE | PROT_READ,
423 };
424 int flags, prot;
425
426 #define OMAP_ANON 0x0002
427 #define OMAP_COPY 0x0020
428 #define OMAP_SHARED 0x0010
429 #define OMAP_FIXED 0x0100
430
431 prot = cvtbsdprot[uap->prot & 0x7];
432 #ifdef COMPAT_FREEBSD32
433 #if defined(__amd64__)
434 if (i386_read_exec && SV_PROC_FLAG(td->td_proc, SV_ILP32) &&
435 prot != 0)
436 prot |= PROT_EXEC;
437 #endif
438 #endif
439 flags = 0;
440 if (uap->flags & OMAP_ANON)
441 flags |= MAP_ANON;
442 if (uap->flags & OMAP_COPY)
443 flags |= MAP_COPY;
444 if (uap->flags & OMAP_SHARED)
445 flags |= MAP_SHARED;
446 else
447 flags |= MAP_PRIVATE;
448 if (uap->flags & OMAP_FIXED)
449 flags |= MAP_FIXED;
450 return (kern_mmap(td, (uintptr_t)uap->addr, uap->len, prot, flags,
451 uap->fd, uap->pos));
452 }
453 #endif /* COMPAT_43 */
454
455
456 #ifndef _SYS_SYSPROTO_H_
457 struct msync_args {
458 void *addr;
459 size_t len;
460 int flags;
461 };
462 #endif
463 int
sys_msync(struct thread * td,struct msync_args * uap)464 sys_msync(struct thread *td, struct msync_args *uap)
465 {
466
467 return (kern_msync(td, (uintptr_t)uap->addr, uap->len, uap->flags));
468 }
469
470 int
kern_msync(struct thread * td,uintptr_t addr0,size_t size,int flags)471 kern_msync(struct thread *td, uintptr_t addr0, size_t size, int flags)
472 {
473 vm_offset_t addr;
474 vm_size_t pageoff;
475 vm_map_t map;
476 int rv;
477
478 addr = addr0;
479 pageoff = (addr & PAGE_MASK);
480 addr -= pageoff;
481 size += pageoff;
482 size = (vm_size_t) round_page(size);
483 if (addr + size < addr)
484 return (EINVAL);
485
486 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
487 return (EINVAL);
488
489 map = &td->td_proc->p_vmspace->vm_map;
490
491 /*
492 * Clean the pages and interpret the return value.
493 */
494 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
495 (flags & MS_INVALIDATE) != 0);
496 switch (rv) {
497 case KERN_SUCCESS:
498 return (0);
499 case KERN_INVALID_ADDRESS:
500 return (ENOMEM);
501 case KERN_INVALID_ARGUMENT:
502 return (EBUSY);
503 case KERN_FAILURE:
504 return (EIO);
505 default:
506 return (EINVAL);
507 }
508 }
509
510 #ifndef _SYS_SYSPROTO_H_
511 struct munmap_args {
512 void *addr;
513 size_t len;
514 };
515 #endif
516 int
sys_munmap(struct thread * td,struct munmap_args * uap)517 sys_munmap(struct thread *td, struct munmap_args *uap)
518 {
519
520 return (kern_munmap(td, (uintptr_t)uap->addr, uap->len));
521 }
522
523 int
kern_munmap(struct thread * td,uintptr_t addr0,size_t size)524 kern_munmap(struct thread *td, uintptr_t addr0, size_t size)
525 {
526 #ifdef HWPMC_HOOKS
527 struct pmckern_map_out pkm;
528 vm_map_entry_t entry;
529 bool pmc_handled;
530 #endif
531 vm_offset_t addr;
532 vm_size_t pageoff;
533 vm_map_t map;
534
535 if (size == 0)
536 return (EINVAL);
537
538 addr = addr0;
539 pageoff = (addr & PAGE_MASK);
540 addr -= pageoff;
541 size += pageoff;
542 size = (vm_size_t) round_page(size);
543 if (addr + size < addr)
544 return (EINVAL);
545
546 /*
547 * Check for illegal addresses. Watch out for address wrap...
548 */
549 map = &td->td_proc->p_vmspace->vm_map;
550 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
551 return (EINVAL);
552 vm_map_lock(map);
553 #ifdef HWPMC_HOOKS
554 pmc_handled = false;
555 if (PMC_HOOK_INSTALLED(PMC_FN_MUNMAP)) {
556 pmc_handled = true;
557 /*
558 * Inform hwpmc if the address range being unmapped contains
559 * an executable region.
560 */
561 pkm.pm_address = (uintptr_t) NULL;
562 if (vm_map_lookup_entry(map, addr, &entry)) {
563 for (; entry->start < addr + size;
564 entry = entry->next) {
565 if (vm_map_check_protection(map, entry->start,
566 entry->end, VM_PROT_EXECUTE) == TRUE) {
567 pkm.pm_address = (uintptr_t) addr;
568 pkm.pm_size = (size_t) size;
569 break;
570 }
571 }
572 }
573 }
574 #endif
575 vm_map_delete(map, addr, addr + size);
576
577 #ifdef HWPMC_HOOKS
578 if (__predict_false(pmc_handled)) {
579 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
580 vm_map_lock_downgrade(map);
581 if (pkm.pm_address != (uintptr_t) NULL)
582 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
583 vm_map_unlock_read(map);
584 } else
585 #endif
586 vm_map_unlock(map);
587
588 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
589 return (0);
590 }
591
592 #ifndef _SYS_SYSPROTO_H_
593 struct mprotect_args {
594 const void *addr;
595 size_t len;
596 int prot;
597 };
598 #endif
599 int
sys_mprotect(struct thread * td,struct mprotect_args * uap)600 sys_mprotect(struct thread *td, struct mprotect_args *uap)
601 {
602
603 return (kern_mprotect(td, (uintptr_t)uap->addr, uap->len, uap->prot));
604 }
605
606 int
kern_mprotect(struct thread * td,uintptr_t addr0,size_t size,int prot)607 kern_mprotect(struct thread *td, uintptr_t addr0, size_t size, int prot)
608 {
609 vm_offset_t addr;
610 vm_size_t pageoff;
611
612 addr = addr0;
613 prot = (prot & VM_PROT_ALL);
614 pageoff = (addr & PAGE_MASK);
615 addr -= pageoff;
616 size += pageoff;
617 size = (vm_size_t) round_page(size);
618 #ifdef COMPAT_FREEBSD32
619 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
620 if (((addr + size) & 0xffffffff) < addr)
621 return (EINVAL);
622 } else
623 #endif
624 if (addr + size < addr)
625 return (EINVAL);
626
627 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
628 addr + size, prot, FALSE)) {
629 case KERN_SUCCESS:
630 return (0);
631 case KERN_PROTECTION_FAILURE:
632 return (EACCES);
633 case KERN_RESOURCE_SHORTAGE:
634 return (ENOMEM);
635 }
636 return (EINVAL);
637 }
638
639 #ifndef _SYS_SYSPROTO_H_
640 struct minherit_args {
641 void *addr;
642 size_t len;
643 int inherit;
644 };
645 #endif
646 int
sys_minherit(struct thread * td,struct minherit_args * uap)647 sys_minherit(struct thread *td, struct minherit_args *uap)
648 {
649 vm_offset_t addr;
650 vm_size_t size, pageoff;
651 vm_inherit_t inherit;
652
653 addr = (vm_offset_t)uap->addr;
654 size = uap->len;
655 inherit = uap->inherit;
656
657 pageoff = (addr & PAGE_MASK);
658 addr -= pageoff;
659 size += pageoff;
660 size = (vm_size_t) round_page(size);
661 if (addr + size < addr)
662 return (EINVAL);
663
664 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
665 addr + size, inherit)) {
666 case KERN_SUCCESS:
667 return (0);
668 case KERN_PROTECTION_FAILURE:
669 return (EACCES);
670 }
671 return (EINVAL);
672 }
673
674 #ifndef _SYS_SYSPROTO_H_
675 struct madvise_args {
676 void *addr;
677 size_t len;
678 int behav;
679 };
680 #endif
681
682 int
sys_madvise(struct thread * td,struct madvise_args * uap)683 sys_madvise(struct thread *td, struct madvise_args *uap)
684 {
685
686 return (kern_madvise(td, (uintptr_t)uap->addr, uap->len, uap->behav));
687 }
688
689 int
kern_madvise(struct thread * td,uintptr_t addr0,size_t len,int behav)690 kern_madvise(struct thread *td, uintptr_t addr0, size_t len, int behav)
691 {
692 vm_map_t map;
693 vm_offset_t addr, end, start;
694 int flags;
695
696 /*
697 * Check for our special case, advising the swap pager we are
698 * "immortal."
699 */
700 if (behav == MADV_PROTECT) {
701 flags = PPROT_SET;
702 return (kern_procctl(td, P_PID, td->td_proc->p_pid,
703 PROC_SPROTECT, &flags));
704 }
705
706 /*
707 * Check for illegal behavior
708 */
709 if (behav < 0 || behav > MADV_CORE)
710 return (EINVAL);
711 /*
712 * Check for illegal addresses. Watch out for address wrap... Note
713 * that VM_*_ADDRESS are not constants due to casts (argh).
714 */
715 map = &td->td_proc->p_vmspace->vm_map;
716 addr = addr0;
717 if (addr < vm_map_min(map) || addr + len > vm_map_max(map))
718 return (EINVAL);
719 if ((addr + len) < addr)
720 return (EINVAL);
721
722 /*
723 * Since this routine is only advisory, we default to conservative
724 * behavior.
725 */
726 start = trunc_page(addr);
727 end = round_page(addr + len);
728
729 if (vm_map_madvise(map, start, end, behav))
730 return (EINVAL);
731 return (0);
732 }
733
734 #ifndef _SYS_SYSPROTO_H_
735 struct mincore_args {
736 const void *addr;
737 size_t len;
738 char *vec;
739 };
740 #endif
741
742 int
sys_mincore(struct thread * td,struct mincore_args * uap)743 sys_mincore(struct thread *td, struct mincore_args *uap)
744 {
745
746 return (kern_mincore(td, (uintptr_t)uap->addr, uap->len, uap->vec));
747 }
748
749 int
kern_mincore(struct thread * td,uintptr_t addr0,size_t len,char * vec)750 kern_mincore(struct thread *td, uintptr_t addr0, size_t len, char *vec)
751 {
752 vm_offset_t addr, first_addr;
753 vm_offset_t end, cend;
754 pmap_t pmap;
755 vm_map_t map;
756 int error = 0;
757 int vecindex, lastvecindex;
758 vm_map_entry_t current;
759 vm_map_entry_t entry;
760 vm_object_t object;
761 vm_paddr_t locked_pa;
762 vm_page_t m;
763 vm_pindex_t pindex;
764 int mincoreinfo;
765 unsigned int timestamp;
766 boolean_t locked;
767
768 /*
769 * Make sure that the addresses presented are valid for user
770 * mode.
771 */
772 first_addr = addr = trunc_page(addr0);
773 end = addr + (vm_size_t)round_page(len);
774 map = &td->td_proc->p_vmspace->vm_map;
775 if (end > vm_map_max(map) || end < addr)
776 return (ENOMEM);
777
778 pmap = vmspace_pmap(td->td_proc->p_vmspace);
779
780 vm_map_lock_read(map);
781 RestartScan:
782 timestamp = map->timestamp;
783
784 if (!vm_map_lookup_entry(map, addr, &entry)) {
785 vm_map_unlock_read(map);
786 return (ENOMEM);
787 }
788
789 /*
790 * Do this on a map entry basis so that if the pages are not
791 * in the current processes address space, we can easily look
792 * up the pages elsewhere.
793 */
794 lastvecindex = -1;
795 for (current = entry; current->start < end; current = current->next) {
796
797 /*
798 * check for contiguity
799 */
800 if (current->end < end && current->next->start > current->end) {
801 vm_map_unlock_read(map);
802 return (ENOMEM);
803 }
804
805 /*
806 * ignore submaps (for now) or null objects
807 */
808 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
809 current->object.vm_object == NULL)
810 continue;
811
812 /*
813 * limit this scan to the current map entry and the
814 * limits for the mincore call
815 */
816 if (addr < current->start)
817 addr = current->start;
818 cend = current->end;
819 if (cend > end)
820 cend = end;
821
822 /*
823 * scan this entry one page at a time
824 */
825 while (addr < cend) {
826 /*
827 * Check pmap first, it is likely faster, also
828 * it can provide info as to whether we are the
829 * one referencing or modifying the page.
830 */
831 object = NULL;
832 locked_pa = 0;
833 retry:
834 m = NULL;
835 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
836 if (mincore_mapped) {
837 /*
838 * We only care about this pmap's
839 * mapping of the page, if any.
840 */
841 if (locked_pa != 0) {
842 vm_page_unlock(PHYS_TO_VM_PAGE(
843 locked_pa));
844 }
845 } else if (locked_pa != 0) {
846 /*
847 * The page is mapped by this process but not
848 * both accessed and modified. It is also
849 * managed. Acquire the object lock so that
850 * other mappings might be examined.
851 */
852 m = PHYS_TO_VM_PAGE(locked_pa);
853 if (m->object != object) {
854 if (object != NULL)
855 VM_OBJECT_WUNLOCK(object);
856 object = m->object;
857 locked = VM_OBJECT_TRYWLOCK(object);
858 vm_page_unlock(m);
859 if (!locked) {
860 VM_OBJECT_WLOCK(object);
861 vm_page_lock(m);
862 goto retry;
863 }
864 } else
865 vm_page_unlock(m);
866 KASSERT(m->valid == VM_PAGE_BITS_ALL,
867 ("mincore: page %p is mapped but invalid",
868 m));
869 } else if (mincoreinfo == 0) {
870 /*
871 * The page is not mapped by this process. If
872 * the object implements managed pages, then
873 * determine if the page is resident so that
874 * the mappings might be examined.
875 */
876 if (current->object.vm_object != object) {
877 if (object != NULL)
878 VM_OBJECT_WUNLOCK(object);
879 object = current->object.vm_object;
880 VM_OBJECT_WLOCK(object);
881 }
882 if (object->type == OBJT_DEFAULT ||
883 object->type == OBJT_SWAP ||
884 object->type == OBJT_VNODE) {
885 pindex = OFF_TO_IDX(current->offset +
886 (addr - current->start));
887 m = vm_page_lookup(object, pindex);
888 if (m != NULL && m->valid == 0)
889 m = NULL;
890 if (m != NULL)
891 mincoreinfo = MINCORE_INCORE;
892 }
893 }
894 if (m != NULL) {
895 /* Examine other mappings to the page. */
896 if (m->dirty == 0 && pmap_is_modified(m))
897 vm_page_dirty(m);
898 if (m->dirty != 0)
899 mincoreinfo |= MINCORE_MODIFIED_OTHER;
900 /*
901 * The first test for PGA_REFERENCED is an
902 * optimization. The second test is
903 * required because a concurrent pmap
904 * operation could clear the last reference
905 * and set PGA_REFERENCED before the call to
906 * pmap_is_referenced().
907 */
908 if ((m->aflags & PGA_REFERENCED) != 0 ||
909 pmap_is_referenced(m) ||
910 (m->aflags & PGA_REFERENCED) != 0)
911 mincoreinfo |= MINCORE_REFERENCED_OTHER;
912 }
913 if (object != NULL)
914 VM_OBJECT_WUNLOCK(object);
915
916 /*
917 * subyte may page fault. In case it needs to modify
918 * the map, we release the lock.
919 */
920 vm_map_unlock_read(map);
921
922 /*
923 * calculate index into user supplied byte vector
924 */
925 vecindex = atop(addr - first_addr);
926
927 /*
928 * If we have skipped map entries, we need to make sure that
929 * the byte vector is zeroed for those skipped entries.
930 */
931 while ((lastvecindex + 1) < vecindex) {
932 ++lastvecindex;
933 error = subyte(vec + lastvecindex, 0);
934 if (error) {
935 error = EFAULT;
936 goto done2;
937 }
938 }
939
940 /*
941 * Pass the page information to the user
942 */
943 error = subyte(vec + vecindex, mincoreinfo);
944 if (error) {
945 error = EFAULT;
946 goto done2;
947 }
948
949 /*
950 * If the map has changed, due to the subyte, the previous
951 * output may be invalid.
952 */
953 vm_map_lock_read(map);
954 if (timestamp != map->timestamp)
955 goto RestartScan;
956
957 lastvecindex = vecindex;
958 addr += PAGE_SIZE;
959 }
960 }
961
962 /*
963 * subyte may page fault. In case it needs to modify
964 * the map, we release the lock.
965 */
966 vm_map_unlock_read(map);
967
968 /*
969 * Zero the last entries in the byte vector.
970 */
971 vecindex = atop(end - first_addr);
972 while ((lastvecindex + 1) < vecindex) {
973 ++lastvecindex;
974 error = subyte(vec + lastvecindex, 0);
975 if (error) {
976 error = EFAULT;
977 goto done2;
978 }
979 }
980
981 /*
982 * If the map has changed, due to the subyte, the previous
983 * output may be invalid.
984 */
985 vm_map_lock_read(map);
986 if (timestamp != map->timestamp)
987 goto RestartScan;
988 vm_map_unlock_read(map);
989 done2:
990 return (error);
991 }
992
993 #ifndef _SYS_SYSPROTO_H_
994 struct mlock_args {
995 const void *addr;
996 size_t len;
997 };
998 #endif
999 int
sys_mlock(struct thread * td,struct mlock_args * uap)1000 sys_mlock(struct thread *td, struct mlock_args *uap)
1001 {
1002
1003 return (kern_mlock(td->td_proc, td->td_ucred,
1004 __DECONST(uintptr_t, uap->addr), uap->len));
1005 }
1006
1007 int
kern_mlock(struct proc * proc,struct ucred * cred,uintptr_t addr0,size_t len)1008 kern_mlock(struct proc *proc, struct ucred *cred, uintptr_t addr0, size_t len)
1009 {
1010 vm_offset_t addr, end, last, start;
1011 vm_size_t npages, size;
1012 vm_map_t map;
1013 unsigned long nsize;
1014 int error;
1015
1016 error = priv_check_cred(cred, PRIV_VM_MLOCK, 0);
1017 if (error)
1018 return (error);
1019 addr = addr0;
1020 size = len;
1021 last = addr + size;
1022 start = trunc_page(addr);
1023 end = round_page(last);
1024 if (last < addr || end < addr)
1025 return (EINVAL);
1026 npages = atop(end - start);
1027 if (npages > vm_page_max_wired)
1028 return (ENOMEM);
1029 map = &proc->p_vmspace->vm_map;
1030 PROC_LOCK(proc);
1031 nsize = ptoa(npages + pmap_wired_count(map->pmap));
1032 if (nsize > lim_cur_proc(proc, RLIMIT_MEMLOCK)) {
1033 PROC_UNLOCK(proc);
1034 return (ENOMEM);
1035 }
1036 PROC_UNLOCK(proc);
1037 if (npages + vm_cnt.v_wire_count > vm_page_max_wired)
1038 return (EAGAIN);
1039 #ifdef RACCT
1040 if (racct_enable) {
1041 PROC_LOCK(proc);
1042 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1043 PROC_UNLOCK(proc);
1044 if (error != 0)
1045 return (ENOMEM);
1046 }
1047 #endif
1048 error = vm_map_wire(map, start, end,
1049 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1050 #ifdef RACCT
1051 if (racct_enable && error != KERN_SUCCESS) {
1052 PROC_LOCK(proc);
1053 racct_set(proc, RACCT_MEMLOCK,
1054 ptoa(pmap_wired_count(map->pmap)));
1055 PROC_UNLOCK(proc);
1056 }
1057 #endif
1058 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1059 }
1060
1061 #ifndef _SYS_SYSPROTO_H_
1062 struct mlockall_args {
1063 int how;
1064 };
1065 #endif
1066
1067 int
sys_mlockall(struct thread * td,struct mlockall_args * uap)1068 sys_mlockall(struct thread *td, struct mlockall_args *uap)
1069 {
1070 vm_map_t map;
1071 int error;
1072
1073 map = &td->td_proc->p_vmspace->vm_map;
1074 error = priv_check(td, PRIV_VM_MLOCK);
1075 if (error)
1076 return (error);
1077
1078 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1079 return (EINVAL);
1080
1081 /*
1082 * If wiring all pages in the process would cause it to exceed
1083 * a hard resource limit, return ENOMEM.
1084 */
1085 if (!old_mlock && uap->how & MCL_CURRENT) {
1086 PROC_LOCK(td->td_proc);
1087 if (map->size > lim_cur(td, RLIMIT_MEMLOCK)) {
1088 PROC_UNLOCK(td->td_proc);
1089 return (ENOMEM);
1090 }
1091 PROC_UNLOCK(td->td_proc);
1092 }
1093 #ifdef RACCT
1094 if (racct_enable) {
1095 PROC_LOCK(td->td_proc);
1096 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1097 PROC_UNLOCK(td->td_proc);
1098 if (error != 0)
1099 return (ENOMEM);
1100 }
1101 #endif
1102
1103 if (uap->how & MCL_FUTURE) {
1104 vm_map_lock(map);
1105 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1106 vm_map_unlock(map);
1107 error = 0;
1108 }
1109
1110 if (uap->how & MCL_CURRENT) {
1111 /*
1112 * P1003.1-2001 mandates that all currently mapped pages
1113 * will be memory resident and locked (wired) upon return
1114 * from mlockall(). vm_map_wire() will wire pages, by
1115 * calling vm_fault_wire() for each page in the region.
1116 */
1117 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1118 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1119 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1120 }
1121 #ifdef RACCT
1122 if (racct_enable && error != KERN_SUCCESS) {
1123 PROC_LOCK(td->td_proc);
1124 racct_set(td->td_proc, RACCT_MEMLOCK,
1125 ptoa(pmap_wired_count(map->pmap)));
1126 PROC_UNLOCK(td->td_proc);
1127 }
1128 #endif
1129
1130 return (error);
1131 }
1132
1133 #ifndef _SYS_SYSPROTO_H_
1134 struct munlockall_args {
1135 register_t dummy;
1136 };
1137 #endif
1138
1139 int
sys_munlockall(struct thread * td,struct munlockall_args * uap)1140 sys_munlockall(struct thread *td, struct munlockall_args *uap)
1141 {
1142 vm_map_t map;
1143 int error;
1144
1145 map = &td->td_proc->p_vmspace->vm_map;
1146 error = priv_check(td, PRIV_VM_MUNLOCK);
1147 if (error)
1148 return (error);
1149
1150 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1151 vm_map_lock(map);
1152 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1153 vm_map_unlock(map);
1154
1155 /* Forcibly unwire all pages. */
1156 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1157 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1158 #ifdef RACCT
1159 if (racct_enable && error == KERN_SUCCESS) {
1160 PROC_LOCK(td->td_proc);
1161 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1162 PROC_UNLOCK(td->td_proc);
1163 }
1164 #endif
1165
1166 return (error);
1167 }
1168
1169 #ifndef _SYS_SYSPROTO_H_
1170 struct munlock_args {
1171 const void *addr;
1172 size_t len;
1173 };
1174 #endif
1175 int
sys_munlock(struct thread * td,struct munlock_args * uap)1176 sys_munlock(struct thread *td, struct munlock_args *uap)
1177 {
1178
1179 return (kern_munlock(td, (uintptr_t)uap->addr, uap->len));
1180 }
1181
1182 int
kern_munlock(struct thread * td,uintptr_t addr0,size_t size)1183 kern_munlock(struct thread *td, uintptr_t addr0, size_t size)
1184 {
1185 vm_offset_t addr, end, last, start;
1186 #ifdef RACCT
1187 vm_map_t map;
1188 #endif
1189 int error;
1190
1191 error = priv_check(td, PRIV_VM_MUNLOCK);
1192 if (error)
1193 return (error);
1194 addr = addr0;
1195 last = addr + size;
1196 start = trunc_page(addr);
1197 end = round_page(last);
1198 if (last < addr || end < addr)
1199 return (EINVAL);
1200 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1201 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1202 #ifdef RACCT
1203 if (racct_enable && error == KERN_SUCCESS) {
1204 PROC_LOCK(td->td_proc);
1205 map = &td->td_proc->p_vmspace->vm_map;
1206 racct_set(td->td_proc, RACCT_MEMLOCK,
1207 ptoa(pmap_wired_count(map->pmap)));
1208 PROC_UNLOCK(td->td_proc);
1209 }
1210 #endif
1211 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1212 }
1213
1214 /*
1215 * vm_mmap_vnode()
1216 *
1217 * Helper function for vm_mmap. Perform sanity check specific for mmap
1218 * operations on vnodes.
1219 */
1220 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)1221 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1222 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1223 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp,
1224 boolean_t *writecounted)
1225 {
1226 struct vattr va;
1227 vm_object_t obj;
1228 vm_ooffset_t foff;
1229 struct ucred *cred;
1230 int error, flags, locktype;
1231
1232 cred = td->td_ucred;
1233 if ((*maxprotp & VM_PROT_WRITE) && (*flagsp & MAP_SHARED))
1234 locktype = LK_EXCLUSIVE;
1235 else
1236 locktype = LK_SHARED;
1237 if ((error = vget(vp, locktype, td)) != 0)
1238 return (error);
1239 AUDIT_ARG_VNODE1(vp);
1240 foff = *foffp;
1241 flags = *flagsp;
1242 obj = vp->v_object;
1243 if (vp->v_type == VREG) {
1244 /*
1245 * Get the proper underlying object
1246 */
1247 if (obj == NULL) {
1248 error = EINVAL;
1249 goto done;
1250 }
1251 if (obj->type == OBJT_VNODE && obj->handle != vp) {
1252 vput(vp);
1253 vp = (struct vnode *)obj->handle;
1254 /*
1255 * Bypass filesystems obey the mpsafety of the
1256 * underlying fs. Tmpfs never bypasses.
1257 */
1258 error = vget(vp, locktype, td);
1259 if (error != 0)
1260 return (error);
1261 }
1262 if (locktype == LK_EXCLUSIVE) {
1263 *writecounted = TRUE;
1264 vnode_pager_update_writecount(obj, 0, objsize);
1265 }
1266 } else {
1267 error = EINVAL;
1268 goto done;
1269 }
1270 if ((error = VOP_GETATTR(vp, &va, cred)))
1271 goto done;
1272 #ifdef MAC
1273 /* This relies on VM_PROT_* matching PROT_*. */
1274 error = mac_vnode_check_mmap(cred, vp, (int)prot, flags);
1275 if (error != 0)
1276 goto done;
1277 #endif
1278 if ((flags & MAP_SHARED) != 0) {
1279 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1280 if (prot & VM_PROT_WRITE) {
1281 error = EPERM;
1282 goto done;
1283 }
1284 *maxprotp &= ~VM_PROT_WRITE;
1285 }
1286 }
1287 /*
1288 * If it is a regular file without any references
1289 * we do not need to sync it.
1290 * Adjust object size to be the size of actual file.
1291 */
1292 objsize = round_page(va.va_size);
1293 if (va.va_nlink == 0)
1294 flags |= MAP_NOSYNC;
1295 if (obj->type == OBJT_VNODE) {
1296 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff,
1297 cred);
1298 if (obj == NULL) {
1299 error = ENOMEM;
1300 goto done;
1301 }
1302 } else {
1303 KASSERT(obj->type == OBJT_DEFAULT || obj->type == OBJT_SWAP,
1304 ("wrong object type"));
1305 VM_OBJECT_WLOCK(obj);
1306 vm_object_reference_locked(obj);
1307 #if VM_NRESERVLEVEL > 0
1308 vm_object_color(obj, 0);
1309 #endif
1310 VM_OBJECT_WUNLOCK(obj);
1311 }
1312 *objp = obj;
1313 *flagsp = flags;
1314
1315 vfs_mark_atime(vp, cred);
1316
1317 done:
1318 if (error != 0 && *writecounted) {
1319 *writecounted = FALSE;
1320 vnode_pager_update_writecount(obj, objsize, 0);
1321 }
1322 vput(vp);
1323 return (error);
1324 }
1325
1326 /*
1327 * vm_mmap_cdev()
1328 *
1329 * Helper function for vm_mmap. Perform sanity check specific for mmap
1330 * operations on cdevs.
1331 */
1332 int
vm_mmap_cdev(struct thread * td,vm_size_t objsize,vm_prot_t prot,vm_prot_t * maxprotp,int * flagsp,struct cdev * cdev,struct cdevsw * dsw,vm_ooffset_t * foff,vm_object_t * objp)1333 vm_mmap_cdev(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1334 vm_prot_t *maxprotp, int *flagsp, struct cdev *cdev, struct cdevsw *dsw,
1335 vm_ooffset_t *foff, vm_object_t *objp)
1336 {
1337 vm_object_t obj;
1338 int error, flags;
1339
1340 flags = *flagsp;
1341
1342 if (dsw->d_flags & D_MMAP_ANON) {
1343 *objp = NULL;
1344 *foff = 0;
1345 *maxprotp = VM_PROT_ALL;
1346 *flagsp |= MAP_ANON;
1347 return (0);
1348 }
1349 /*
1350 * cdevs do not provide private mappings of any kind.
1351 */
1352 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1353 (prot & VM_PROT_WRITE) != 0)
1354 return (EACCES);
1355 if (flags & (MAP_PRIVATE|MAP_COPY))
1356 return (EINVAL);
1357 /*
1358 * Force device mappings to be shared.
1359 */
1360 flags |= MAP_SHARED;
1361 #ifdef MAC_XXX
1362 error = mac_cdev_check_mmap(td->td_ucred, cdev, (int)prot);
1363 if (error != 0)
1364 return (error);
1365 #endif
1366 /*
1367 * First, try d_mmap_single(). If that is not implemented
1368 * (returns ENODEV), fall back to using the device pager.
1369 * Note that d_mmap_single() must return a reference to the
1370 * object (it needs to bump the reference count of the object
1371 * it returns somehow).
1372 *
1373 * XXX assumes VM_PROT_* == PROT_*
1374 */
1375 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1376 if (error != ENODEV)
1377 return (error);
1378 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1379 td->td_ucred);
1380 if (obj == NULL)
1381 return (EINVAL);
1382 *objp = obj;
1383 *flagsp = flags;
1384 return (0);
1385 }
1386
1387 /*
1388 * vm_mmap()
1389 *
1390 * Internal version of mmap used by exec, sys5 shared memory, and
1391 * various device drivers. Handle is either a vnode pointer, a
1392 * character device, or NULL for MAP_ANON.
1393 */
1394 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)1395 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1396 vm_prot_t maxprot, int flags,
1397 objtype_t handle_type, void *handle,
1398 vm_ooffset_t foff)
1399 {
1400 vm_object_t object;
1401 struct thread *td = curthread;
1402 int error;
1403 boolean_t writecounted;
1404
1405 if (size == 0)
1406 return (EINVAL);
1407
1408 size = round_page(size);
1409 object = NULL;
1410 writecounted = FALSE;
1411
1412 /*
1413 * Lookup/allocate object.
1414 */
1415 switch (handle_type) {
1416 case OBJT_DEVICE: {
1417 struct cdevsw *dsw;
1418 struct cdev *cdev;
1419 int ref;
1420
1421 cdev = handle;
1422 dsw = dev_refthread(cdev, &ref);
1423 if (dsw == NULL)
1424 return (ENXIO);
1425 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, cdev,
1426 dsw, &foff, &object);
1427 dev_relthread(cdev, ref);
1428 break;
1429 }
1430 case OBJT_VNODE:
1431 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1432 handle, &foff, &object, &writecounted);
1433 break;
1434 case OBJT_DEFAULT:
1435 if (handle == NULL) {
1436 error = 0;
1437 break;
1438 }
1439 /* FALLTHROUGH */
1440 default:
1441 error = EINVAL;
1442 break;
1443 }
1444 if (error)
1445 return (error);
1446
1447 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1448 foff, writecounted, td);
1449 if (error != 0 && object != NULL) {
1450 /*
1451 * If this mapping was accounted for in the vnode's
1452 * writecount, then undo that now.
1453 */
1454 if (writecounted)
1455 vnode_pager_release_writecount(object, 0, size);
1456 vm_object_deallocate(object);
1457 }
1458 return (error);
1459 }
1460
1461 /*
1462 * Internal version of mmap that maps a specific VM object into an
1463 * map. Called by mmap for MAP_ANON, vm_mmap, shm_mmap, and vn_mmap.
1464 */
1465 int
vm_mmap_object(vm_map_t map,vm_offset_t * addr,vm_size_t size,vm_prot_t prot,vm_prot_t maxprot,int flags,vm_object_t object,vm_ooffset_t foff,boolean_t writecounted,struct thread * td)1466 vm_mmap_object(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1467 vm_prot_t maxprot, int flags, vm_object_t object, vm_ooffset_t foff,
1468 boolean_t writecounted, struct thread *td)
1469 {
1470 boolean_t curmap, fitit;
1471 vm_offset_t max_addr;
1472 int docow, error, findspace, rv;
1473
1474 curmap = map == &td->td_proc->p_vmspace->vm_map;
1475 if (curmap) {
1476 PROC_LOCK(td->td_proc);
1477 if (map->size + size > lim_cur_proc(td->td_proc, RLIMIT_VMEM)) {
1478 PROC_UNLOCK(td->td_proc);
1479 return (ENOMEM);
1480 }
1481 if (racct_set(td->td_proc, RACCT_VMEM, map->size + size)) {
1482 PROC_UNLOCK(td->td_proc);
1483 return (ENOMEM);
1484 }
1485 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
1486 if (ptoa(pmap_wired_count(map->pmap)) + size >
1487 lim_cur_proc(td->td_proc, RLIMIT_MEMLOCK)) {
1488 racct_set_force(td->td_proc, RACCT_VMEM,
1489 map->size);
1490 PROC_UNLOCK(td->td_proc);
1491 return (ENOMEM);
1492 }
1493 error = racct_set(td->td_proc, RACCT_MEMLOCK,
1494 ptoa(pmap_wired_count(map->pmap)) + size);
1495 if (error != 0) {
1496 racct_set_force(td->td_proc, RACCT_VMEM,
1497 map->size);
1498 PROC_UNLOCK(td->td_proc);
1499 return (error);
1500 }
1501 }
1502 PROC_UNLOCK(td->td_proc);
1503 }
1504
1505 /*
1506 * We currently can only deal with page aligned file offsets.
1507 * The mmap() system call already enforces this by subtracting
1508 * the page offset from the file offset, but checking here
1509 * catches errors in device drivers (e.g. d_single_mmap()
1510 * callbacks) and other internal mapping requests (such as in
1511 * exec).
1512 */
1513 if (foff & PAGE_MASK)
1514 return (EINVAL);
1515
1516 if ((flags & MAP_FIXED) == 0) {
1517 fitit = TRUE;
1518 *addr = round_page(*addr);
1519 } else {
1520 if (*addr != trunc_page(*addr))
1521 return (EINVAL);
1522 fitit = FALSE;
1523 }
1524
1525 if (flags & MAP_ANON) {
1526 if (object != NULL || foff != 0)
1527 return (EINVAL);
1528 docow = 0;
1529 } else if (flags & MAP_PREFAULT_READ)
1530 docow = MAP_PREFAULT;
1531 else
1532 docow = MAP_PREFAULT_PARTIAL;
1533
1534 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1535 docow |= MAP_COPY_ON_WRITE;
1536 if (flags & MAP_NOSYNC)
1537 docow |= MAP_DISABLE_SYNCER;
1538 if (flags & MAP_NOCORE)
1539 docow |= MAP_DISABLE_COREDUMP;
1540 /* Shared memory is also shared with children. */
1541 if (flags & MAP_SHARED)
1542 docow |= MAP_INHERIT_SHARE;
1543 if (writecounted)
1544 docow |= MAP_VN_WRITECOUNT;
1545 if (flags & MAP_STACK) {
1546 if (object != NULL)
1547 return (EINVAL);
1548 docow |= MAP_STACK_GROWS_DOWN;
1549 }
1550 if ((flags & MAP_EXCL) != 0)
1551 docow |= MAP_CHECK_EXCL;
1552 if ((flags & MAP_GUARD) != 0)
1553 docow |= MAP_CREATE_GUARD;
1554
1555 if (fitit) {
1556 if ((flags & MAP_ALIGNMENT_MASK) == MAP_ALIGNED_SUPER)
1557 findspace = VMFS_SUPER_SPACE;
1558 else if ((flags & MAP_ALIGNMENT_MASK) != 0)
1559 findspace = VMFS_ALIGNED_SPACE(flags >>
1560 MAP_ALIGNMENT_SHIFT);
1561 else
1562 findspace = VMFS_OPTIMAL_SPACE;
1563 max_addr = 0;
1564 #ifdef MAP_32BIT
1565 if ((flags & MAP_32BIT) != 0)
1566 max_addr = MAP_32BIT_MAX_ADDR;
1567 #endif
1568 if (curmap) {
1569 rv = vm_map_find_min(map, object, foff, addr, size,
1570 round_page((vm_offset_t)td->td_proc->p_vmspace->
1571 vm_daddr + lim_max(td, RLIMIT_DATA)), max_addr,
1572 findspace, prot, maxprot, docow);
1573 } else {
1574 rv = vm_map_find(map, object, foff, addr, size,
1575 max_addr, findspace, prot, maxprot, docow);
1576 }
1577 } else {
1578 rv = vm_map_fixed(map, object, foff, *addr, size,
1579 prot, maxprot, docow);
1580 }
1581
1582 if (rv == KERN_SUCCESS) {
1583 /*
1584 * If the process has requested that all future mappings
1585 * be wired, then heed this.
1586 */
1587 if (map->flags & MAP_WIREFUTURE) {
1588 vm_map_wire(map, *addr, *addr + size,
1589 VM_MAP_WIRE_USER | ((flags & MAP_STACK) ?
1590 VM_MAP_WIRE_HOLESOK : VM_MAP_WIRE_NOHOLES));
1591 }
1592 }
1593 return (vm_mmap_to_errno(rv));
1594 }
1595
1596 /*
1597 * Translate a Mach VM return code to zero on success or the appropriate errno
1598 * on failure.
1599 */
1600 int
vm_mmap_to_errno(int rv)1601 vm_mmap_to_errno(int rv)
1602 {
1603
1604 switch (rv) {
1605 case KERN_SUCCESS:
1606 return (0);
1607 case KERN_INVALID_ADDRESS:
1608 case KERN_NO_SPACE:
1609 return (ENOMEM);
1610 case KERN_PROTECTION_FAILURE:
1611 return (EACCES);
1612 default:
1613 return (EINVAL);
1614 }
1615 }
1616