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