xref: /freebsd/sys/vm/vm_pager.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
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  * 3. 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: @(#)vm_pager.c	8.6 (Berkeley) 1/12/94
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *	Paging space routine stubs.  Emulates a matchmaker-like interface
65  *	for builtin pagers.
66  */
67 
68 #include <sys/cdefs.h>
69 #include "opt_param.h"
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/vnode.h>
75 #include <sys/bio.h>
76 #include <sys/buf.h>
77 #include <sys/ucred.h>
78 #include <sys/malloc.h>
79 #include <sys/rwlock.h>
80 #include <sys/user.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_extern.h>
89 #include <vm/uma.h>
90 
91 uma_zone_t pbuf_zone;
92 static int	pbuf_init(void *, int, int);
93 static int	pbuf_ctor(void *, int, void *, int);
94 static void	pbuf_dtor(void *, int, void *);
95 
96 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
97 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
98     vm_ooffset_t, struct ucred *);
99 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
100 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
101 static void dead_pager_dealloc(vm_object_t);
102 static void dead_pager_getvp(vm_object_t, struct vnode **, bool *);
103 
104 static int
105 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind,
106     int *rahead)
107 {
108 
109 	return (VM_PAGER_FAIL);
110 }
111 
112 static vm_object_t
113 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
114     vm_ooffset_t off, struct ucred *cred)
115 {
116 
117 	return (NULL);
118 }
119 
120 static void
121 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count,
122     int flags, int *rtvals)
123 {
124 	int i;
125 
126 	for (i = 0; i < count; i++)
127 		rtvals[i] = VM_PAGER_AGAIN;
128 }
129 
130 static boolean_t
131 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next)
132 {
133 
134 	if (prev != NULL)
135 		*prev = 0;
136 	if (next != NULL)
137 		*next = 0;
138 	return (FALSE);
139 }
140 
141 static void
142 dead_pager_dealloc(vm_object_t object)
143 {
144 
145 }
146 
147 static void
148 dead_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp)
149 {
150 	/*
151 	 * For OBJT_DEAD objects, v_writecount was handled in
152 	 * vnode_pager_dealloc().
153 	 */
154 }
155 
156 static const struct pagerops deadpagerops = {
157 	.pgo_kvme_type = KVME_TYPE_DEAD,
158 	.pgo_alloc = 	dead_pager_alloc,
159 	.pgo_dealloc =	dead_pager_dealloc,
160 	.pgo_getpages =	dead_pager_getpages,
161 	.pgo_putpages =	dead_pager_putpages,
162 	.pgo_haspage =	dead_pager_haspage,
163 	.pgo_getvp =	dead_pager_getvp,
164 };
165 
166 const struct pagerops *pagertab[16] __read_mostly = {
167 	[OBJT_SWAP] =		&swappagerops,
168 	[OBJT_VNODE] =		&vnodepagerops,
169 	[OBJT_DEVICE] =		&devicepagerops,
170 	[OBJT_PHYS] =		&physpagerops,
171 	[OBJT_DEAD] =		&deadpagerops,
172 	[OBJT_SG] = 		&sgpagerops,
173 	[OBJT_MGTDEVICE] = 	&mgtdevicepagerops,
174 };
175 static struct mtx pagertab_lock;
176 
177 void
178 vm_pager_init(void)
179 {
180 	const struct pagerops **pgops;
181 	int i;
182 
183 	mtx_init(&pagertab_lock, "dynpag", NULL, MTX_DEF);
184 
185 	/*
186 	 * Initialize known pagers
187 	 */
188 	for (i = 0; i < OBJT_FIRST_DYN; i++) {
189 		pgops = &pagertab[i];
190 		if (*pgops != NULL && (*pgops)->pgo_init != NULL)
191 			(*(*pgops)->pgo_init)();
192 	}
193 }
194 
195 static int nswbuf_max;
196 
197 void
198 vm_pager_bufferinit(void)
199 {
200 
201 	/* Main zone for paging bufs. */
202 	pbuf_zone = uma_zcreate("pbuf",
203 	    sizeof(struct buf) + PBUF_PAGES * sizeof(vm_page_t),
204 	    pbuf_ctor, pbuf_dtor, pbuf_init, NULL, UMA_ALIGN_CACHE,
205 	    UMA_ZONE_NOFREE);
206 	/* Few systems may still use this zone directly, so it needs a limit. */
207 	nswbuf_max += uma_zone_set_max(pbuf_zone, NSWBUF_MIN);
208 }
209 
210 uma_zone_t
211 pbuf_zsecond_create(const char *name, int max)
212 {
213 	uma_zone_t zone;
214 
215 	zone = uma_zsecond_create(name, pbuf_ctor, pbuf_dtor, NULL, NULL,
216 	    pbuf_zone);
217 
218 #ifdef KMSAN
219 	/*
220 	 * Shrink the size of the pbuf pools if KMSAN is enabled, otherwise the
221 	 * shadows of the large KVA allocations eat up too much memory.
222 	 */
223 	max /= 3;
224 #endif
225 
226 	/*
227 	 * uma_prealloc() rounds up to items per slab. If we would prealloc
228 	 * immediately on every pbuf_zsecond_create(), we may accumulate too
229 	 * much of difference between hard limit and prealloced items, which
230 	 * means wasted memory.
231 	 */
232 	if (nswbuf_max > 0)
233 		nswbuf_max += uma_zone_set_max(zone, max);
234 	else
235 		uma_prealloc(pbuf_zone, uma_zone_set_max(zone, max));
236 
237 	return (zone);
238 }
239 
240 static void
241 pbuf_prealloc(void *arg __unused)
242 {
243 
244 	uma_prealloc(pbuf_zone, nswbuf_max);
245 	nswbuf_max = -1;
246 }
247 
248 SYSINIT(pbuf, SI_SUB_KTHREAD_BUF, SI_ORDER_ANY, pbuf_prealloc, NULL);
249 
250 /*
251  * Allocate an instance of a pager of the given type.
252  * Size, protection and offset parameters are passed in for pagers that
253  * need to perform page-level validation (e.g. the device pager).
254  */
255 vm_object_t
256 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
257     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
258 {
259 	vm_object_t object;
260 
261 	MPASS(type < nitems(pagertab));
262 
263 	object = (*pagertab[type]->pgo_alloc)(handle, size, prot, off, cred);
264 	if (object != NULL)
265 		object->type = type;
266 	return (object);
267 }
268 
269 /*
270  *	The object must be locked.
271  */
272 void
273 vm_pager_deallocate(vm_object_t object)
274 {
275 
276 	VM_OBJECT_ASSERT_WLOCKED(object);
277 	MPASS(object->type < nitems(pagertab));
278 	(*pagertab[object->type]->pgo_dealloc) (object);
279 }
280 
281 static void
282 vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count)
283 {
284 #ifdef INVARIANTS
285 
286 	/*
287 	 * All pages must be consecutive, busied, not mapped, not fully valid,
288 	 * not dirty and belong to the proper object.  Some pages may be the
289 	 * bogus page, but the first and last pages must be a real ones.
290 	 */
291 
292 	VM_OBJECT_ASSERT_UNLOCKED(object);
293 	VM_OBJECT_ASSERT_PAGING(object);
294 	KASSERT(count > 0, ("%s: 0 count", __func__));
295 	for (int i = 0 ; i < count; i++) {
296 		if (m[i] == bogus_page) {
297 			KASSERT(i != 0 && i != count - 1,
298 			    ("%s: page %d is the bogus page", __func__, i));
299 			continue;
300 		}
301 		vm_page_assert_xbusied(m[i]);
302 		KASSERT(!pmap_page_is_mapped(m[i]),
303 		    ("%s: page %p is mapped", __func__, m[i]));
304 		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
305 		    ("%s: request for a valid page %p", __func__, m[i]));
306 		KASSERT(m[i]->dirty == 0,
307 		    ("%s: page %p is dirty", __func__, m[i]));
308 		KASSERT(m[i]->object == object,
309 		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
310 		KASSERT(m[i]->pindex == m[0]->pindex + i,
311 		    ("%s: page %p isn't consecutive", __func__, m[i]));
312 	}
313 #endif
314 }
315 
316 /*
317  * Page in the pages for the object using its associated pager.
318  * The requested page must be fully valid on successful return.
319  */
320 int
321 vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
322     int *rahead)
323 {
324 #ifdef INVARIANTS
325 	vm_pindex_t pindex = m[0]->pindex;
326 #endif
327 	int r;
328 
329 	MPASS(object->type < nitems(pagertab));
330 	vm_pager_assert_in(object, m, count);
331 
332 	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
333 	    rahead);
334 	if (r != VM_PAGER_OK)
335 		return (r);
336 
337 	for (int i = 0; i < count; i++) {
338 		/*
339 		 * If pager has replaced a page, assert that it had
340 		 * updated the array.
341 		 */
342 #ifdef INVARIANTS
343 		KASSERT(m[i] == vm_page_relookup(object, pindex++),
344 		    ("%s: mismatch page %p pindex %ju", __func__,
345 		    m[i], (uintmax_t )pindex - 1));
346 #endif
347 
348 		/*
349 		 * Zero out partially filled data.
350 		 */
351 		if (m[i]->valid != VM_PAGE_BITS_ALL)
352 			vm_page_zero_invalid(m[i], TRUE);
353 	}
354 	return (VM_PAGER_OK);
355 }
356 
357 int
358 vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
359     int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
360 {
361 
362 	MPASS(object->type < nitems(pagertab));
363 	vm_pager_assert_in(object, m, count);
364 
365 	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
366 	    count, rbehind, rahead, iodone, arg));
367 }
368 
369 /*
370  * vm_pager_put_pages() - inline, see vm/vm_pager.h
371  * vm_pager_has_page() - inline, see vm/vm_pager.h
372  */
373 
374 /*
375  * Search the specified pager object list for an object with the
376  * specified handle.  If an object with the specified handle is found,
377  * increase its reference count and return it.  Otherwise, return NULL.
378  *
379  * The pager object list must be locked.
380  */
381 vm_object_t
382 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
383 {
384 	vm_object_t object;
385 
386 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
387 		if (object->handle == handle) {
388 			VM_OBJECT_WLOCK(object);
389 			if ((object->flags & OBJ_DEAD) == 0) {
390 				vm_object_reference_locked(object);
391 				VM_OBJECT_WUNLOCK(object);
392 				break;
393 			}
394 			VM_OBJECT_WUNLOCK(object);
395 		}
396 	}
397 	return (object);
398 }
399 
400 int
401 vm_pager_alloc_dyn_type(struct pagerops *ops, int base_type)
402 {
403 	int res;
404 
405 	mtx_lock(&pagertab_lock);
406 	MPASS(base_type == -1 ||
407 	    (base_type >= OBJT_SWAP && base_type < nitems(pagertab)));
408 	for (res = OBJT_FIRST_DYN; res < nitems(pagertab); res++) {
409 		if (pagertab[res] == NULL)
410 			break;
411 	}
412 	if (res == nitems(pagertab)) {
413 		mtx_unlock(&pagertab_lock);
414 		return (-1);
415 	}
416 	if (base_type != -1) {
417 		MPASS(pagertab[base_type] != NULL);
418 #define	FIX(n)								\
419 		if (ops->pgo_##n == NULL)				\
420 			ops->pgo_##n = pagertab[base_type]->pgo_##n
421 		FIX(init);
422 		FIX(alloc);
423 		FIX(dealloc);
424 		FIX(getpages);
425 		FIX(getpages_async);
426 		FIX(putpages);
427 		FIX(haspage);
428 		FIX(populate);
429 		FIX(pageunswapped);
430 		FIX(update_writecount);
431 		FIX(release_writecount);
432 		FIX(set_writeable_dirty);
433 		FIX(mightbedirty);
434 		FIX(getvp);
435 		FIX(freespace);
436 		FIX(page_inserted);
437 		FIX(page_removed);
438 		FIX(can_alloc_page);
439 #undef FIX
440 	}
441 	pagertab[res] = ops;	/* XXXKIB should be rel, but acq is too much */
442 	mtx_unlock(&pagertab_lock);
443 	return (res);
444 }
445 
446 void
447 vm_pager_free_dyn_type(objtype_t type)
448 {
449 	MPASS(type >= OBJT_FIRST_DYN && type < nitems(pagertab));
450 
451 	mtx_lock(&pagertab_lock);
452 	MPASS(pagertab[type] != NULL);
453 	pagertab[type] = NULL;
454 	mtx_unlock(&pagertab_lock);
455 }
456 
457 static int
458 pbuf_ctor(void *mem, int size, void *arg, int flags)
459 {
460 	struct buf *bp = mem;
461 
462 	bp->b_vp = NULL;
463 	bp->b_bufobj = NULL;
464 
465 	/* copied from initpbuf() */
466 	bp->b_rcred = NOCRED;
467 	bp->b_wcred = NOCRED;
468 	bp->b_qindex = 0;       /* On no queue (QUEUE_NONE) */
469 	bp->b_data = bp->b_kvabase;
470 	bp->b_xflags = 0;
471 	bp->b_flags = B_MAXPHYS;
472 	bp->b_ioflags = 0;
473 	bp->b_iodone = NULL;
474 	bp->b_error = 0;
475 	BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL);
476 
477 	return (0);
478 }
479 
480 static void
481 pbuf_dtor(void *mem, int size, void *arg)
482 {
483 	struct buf *bp = mem;
484 
485 	if (bp->b_rcred != NOCRED) {
486 		crfree(bp->b_rcred);
487 		bp->b_rcred = NOCRED;
488 	}
489 	if (bp->b_wcred != NOCRED) {
490 		crfree(bp->b_wcred);
491 		bp->b_wcred = NOCRED;
492 	}
493 
494 	BUF_UNLOCK(bp);
495 }
496 
497 static const char pbuf_wmesg[] = "pbufwait";
498 
499 static int
500 pbuf_init(void *mem, int size, int flags)
501 {
502 	struct buf *bp = mem;
503 
504 	TSENTER();
505 
506 	bp->b_kvabase = (void *)kva_alloc(ptoa(PBUF_PAGES));
507 	if (bp->b_kvabase == NULL)
508 		return (ENOMEM);
509 	bp->b_kvasize = ptoa(PBUF_PAGES);
510 	BUF_LOCKINIT(bp, pbuf_wmesg);
511 	LIST_INIT(&bp->b_dep);
512 	bp->b_rcred = bp->b_wcred = NOCRED;
513 	bp->b_xflags = 0;
514 
515 	TSEXIT();
516 
517 	return (0);
518 }
519 
520 /*
521  * Associate a p-buffer with a vnode.
522  *
523  * Also sets B_PAGING flag to indicate that vnode is not fully associated
524  * with the buffer.  i.e. the bp has not been linked into the vnode or
525  * ref-counted.
526  */
527 void
528 pbgetvp(struct vnode *vp, struct buf *bp)
529 {
530 
531 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
532 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
533 
534 	bp->b_vp = vp;
535 	bp->b_flags |= B_PAGING;
536 	bp->b_bufobj = &vp->v_bufobj;
537 }
538 
539 /*
540  * Associate a p-buffer with a vnode.
541  *
542  * Also sets B_PAGING flag to indicate that vnode is not fully associated
543  * with the buffer.  i.e. the bp has not been linked into the vnode or
544  * ref-counted.
545  */
546 void
547 pbgetbo(struct bufobj *bo, struct buf *bp)
548 {
549 
550 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
551 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
552 
553 	bp->b_flags |= B_PAGING;
554 	bp->b_bufobj = bo;
555 }
556 
557 /*
558  * Disassociate a p-buffer from a vnode.
559  */
560 void
561 pbrelvp(struct buf *bp)
562 {
563 
564 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
565 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
566 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
567 	    ("pbrelvp: pager buf on vnode list."));
568 
569 	bp->b_vp = NULL;
570 	bp->b_bufobj = NULL;
571 	bp->b_flags &= ~B_PAGING;
572 }
573 
574 /*
575  * Disassociate a p-buffer from a bufobj.
576  */
577 void
578 pbrelbo(struct buf *bp)
579 {
580 
581 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
582 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
583 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
584 	    ("pbrelbo: pager buf on vnode list."));
585 
586 	bp->b_bufobj = NULL;
587 	bp->b_flags &= ~B_PAGING;
588 }
589 
590 void
591 vm_object_set_writeable_dirty(vm_object_t object)
592 {
593 	pgo_set_writeable_dirty_t *method;
594 
595 	MPASS(object->type < nitems(pagertab));
596 
597 	method = pagertab[object->type]->pgo_set_writeable_dirty;
598 	if (method != NULL)
599 		method(object);
600 }
601 
602 bool
603 vm_object_mightbedirty(vm_object_t object)
604 {
605 	pgo_mightbedirty_t *method;
606 
607 	MPASS(object->type < nitems(pagertab));
608 
609 	method = pagertab[object->type]->pgo_mightbedirty;
610 	if (method == NULL)
611 		return (false);
612 	return (method(object));
613 }
614 
615 /*
616  * Return the kvme type of the given object.
617  * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL.
618  */
619 int
620 vm_object_kvme_type(vm_object_t object, struct vnode **vpp)
621 {
622 	VM_OBJECT_ASSERT_LOCKED(object);
623 	MPASS(object->type < nitems(pagertab));
624 
625 	if (vpp != NULL)
626 		*vpp = vm_object_vnode(object);
627 	return (pagertab[object->type]->pgo_kvme_type);
628 }
629