1/*-
2 * Copyright (c) 1991, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	from: @(#)vm_pager.c	8.6 (Berkeley) 1/12/94
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__FBSDID("$FreeBSD$");
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
79#include <vm/vm.h>
80#include <vm/vm_param.h>
81#include <vm/vm_kern.h>
82#include <vm/vm_object.h>
83#include <vm/vm_page.h>
84#include <vm/vm_pager.h>
85#include <vm/vm_extern.h>
86
87int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
88
89struct buf *swbuf;
90
91static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
92static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
93    vm_ooffset_t, struct ucred *);
94static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
95static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
96static void dead_pager_dealloc(vm_object_t);
97
98static int
99dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int *rbehind,
100    int *rahead)
101{
102
103	return (VM_PAGER_FAIL);
104}
105
106static vm_object_t
107dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
108    vm_ooffset_t off, struct ucred *cred)
109{
110	return NULL;
111}
112
113static void
114dead_pager_putpages(object, m, count, flags, rtvals)
115	vm_object_t object;
116	vm_page_t *m;
117	int count;
118	int flags;
119	int *rtvals;
120{
121	int i;
122
123	for (i = 0; i < count; i++) {
124		rtvals[i] = VM_PAGER_AGAIN;
125	}
126}
127
128static int
129dead_pager_haspage(object, pindex, prev, next)
130	vm_object_t object;
131	vm_pindex_t pindex;
132	int *prev;
133	int *next;
134{
135	if (prev)
136		*prev = 0;
137	if (next)
138		*next = 0;
139	return FALSE;
140}
141
142static void
143dead_pager_dealloc(object)
144	vm_object_t object;
145{
146	return;
147}
148
149static struct pagerops deadpagerops = {
150	.pgo_alloc = 	dead_pager_alloc,
151	.pgo_dealloc =	dead_pager_dealloc,
152	.pgo_getpages =	dead_pager_getpages,
153	.pgo_putpages =	dead_pager_putpages,
154	.pgo_haspage =	dead_pager_haspage,
155};
156
157struct pagerops *pagertab[] = {
158	&defaultpagerops,	/* OBJT_DEFAULT */
159	&swappagerops,		/* OBJT_SWAP */
160	&vnodepagerops,		/* OBJT_VNODE */
161	&devicepagerops,	/* OBJT_DEVICE */
162	&physpagerops,		/* OBJT_PHYS */
163	&deadpagerops,		/* OBJT_DEAD */
164	&sgpagerops,		/* OBJT_SG */
165	&mgtdevicepagerops,	/* OBJT_MGTDEVICE */
166};
167
168/*
169 * Kernel address space for mapping pages.
170 * Used by pagers where KVAs are needed for IO.
171 *
172 * XXX needs to be large enough to support the number of pending async
173 * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
174 * (MAXPHYS == 64k) if you want to get the most efficiency.
175 */
176struct mtx_padalign pbuf_mtx;
177static TAILQ_HEAD(swqueue, buf) bswlist;
178static int bswneeded;
179vm_offset_t swapbkva;		/* swap buffers kva */
180
181void
182vm_pager_init()
183{
184	struct pagerops **pgops;
185
186	TAILQ_INIT(&bswlist);
187	/*
188	 * Initialize known pagers
189	 */
190	for (pgops = pagertab; pgops < &pagertab[nitems(pagertab)]; pgops++)
191		if ((*pgops)->pgo_init != NULL)
192			(*(*pgops)->pgo_init) ();
193}
194
195void
196vm_pager_bufferinit()
197{
198	struct buf *bp;
199	int i;
200
201	mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF);
202	bp = swbuf;
203	/*
204	 * Now set up swap and physical I/O buffer headers.
205	 */
206	for (i = 0; i < nswbuf; i++, bp++) {
207		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
208		BUF_LOCKINIT(bp);
209		LIST_INIT(&bp->b_dep);
210		bp->b_rcred = bp->b_wcred = NOCRED;
211		bp->b_xflags = 0;
212	}
213
214	cluster_pbuf_freecnt = nswbuf / 2;
215	vnode_pbuf_freecnt = nswbuf / 2 + 1;
216	vnode_async_pbuf_freecnt = nswbuf / 2;
217}
218
219/*
220 * Allocate an instance of a pager of the given type.
221 * Size, protection and offset parameters are passed in for pagers that
222 * need to perform page-level validation (e.g. the device pager).
223 */
224vm_object_t
225vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
226    vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
227{
228	vm_object_t ret;
229	struct pagerops *ops;
230
231	ops = pagertab[type];
232	if (ops)
233		ret = (*ops->pgo_alloc) (handle, size, prot, off, cred);
234	else
235		ret = NULL;
236	return (ret);
237}
238
239/*
240 *	The object must be locked.
241 */
242void
243vm_pager_deallocate(object)
244	vm_object_t object;
245{
246
247	VM_OBJECT_ASSERT_WLOCKED(object);
248	(*pagertab[object->type]->pgo_dealloc) (object);
249}
250
251static void
252vm_pager_assert_in(vm_object_t object, vm_page_t *m, int count)
253{
254#ifdef INVARIANTS
255
256	VM_OBJECT_ASSERT_WLOCKED(object);
257	KASSERT(count > 0, ("%s: 0 count", __func__));
258	/*
259	 * All pages must be busied, not mapped, not fully valid,
260	 * not dirty and belong to the proper object.
261	 */
262	for (int i = 0 ; i < count; i++) {
263		vm_page_assert_xbusied(m[i]);
264		KASSERT(!pmap_page_is_mapped(m[i]),
265		    ("%s: page %p is mapped", __func__, m[i]));
266		KASSERT(m[i]->valid != VM_PAGE_BITS_ALL,
267		    ("%s: request for a valid page %p", __func__, m[i]));
268		KASSERT(m[i]->dirty == 0,
269		    ("%s: page %p is dirty", __func__, m[i]));
270		KASSERT(m[i]->object == object,
271		    ("%s: wrong object %p/%p", __func__, object, m[i]->object));
272	}
273#endif
274}
275
276/*
277 * Page in the pages for the object using its associated pager.
278 * The requested page must be fully valid on successful return.
279 */
280int
281vm_pager_get_pages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
282    int *rahead)
283{
284#ifdef INVARIANTS
285	vm_pindex_t pindex = m[0]->pindex;
286#endif
287	int r;
288
289	vm_pager_assert_in(object, m, count);
290
291	r = (*pagertab[object->type]->pgo_getpages)(object, m, count, rbehind,
292	    rahead);
293	if (r != VM_PAGER_OK)
294		return (r);
295
296	for (int i = 0; i < count; i++) {
297		/*
298		 * If pager has replaced a page, assert that it had
299		 * updated the array.
300		 */
301		KASSERT(m[i] == vm_page_lookup(object, pindex++),
302		    ("%s: mismatch page %p pindex %ju", __func__,
303		    m[i], (uintmax_t )pindex - 1));
304		/*
305		 * Zero out partially filled data.
306		 */
307		if (m[i]->valid != VM_PAGE_BITS_ALL)
308			vm_page_zero_invalid(m[i], TRUE);
309	}
310	return (VM_PAGER_OK);
311}
312
313int
314vm_pager_get_pages_async(vm_object_t object, vm_page_t *m, int count,
315    int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg)
316{
317
318	vm_pager_assert_in(object, m, count);
319
320	return ((*pagertab[object->type]->pgo_getpages_async)(object, m,
321	    count, rbehind, rahead, iodone, arg));
322}
323
324/*
325 * vm_pager_put_pages() - inline, see vm/vm_pager.h
326 * vm_pager_has_page() - inline, see vm/vm_pager.h
327 */
328
329/*
330 * Search the specified pager object list for an object with the
331 * specified handle.  If an object with the specified handle is found,
332 * increase its reference count and return it.  Otherwise, return NULL.
333 *
334 * The pager object list must be locked.
335 */
336vm_object_t
337vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
338{
339	vm_object_t object;
340
341	TAILQ_FOREACH(object, pg_list, pager_object_list) {
342		if (object->handle == handle) {
343			VM_OBJECT_WLOCK(object);
344			if ((object->flags & OBJ_DEAD) == 0) {
345				vm_object_reference_locked(object);
346				VM_OBJECT_WUNLOCK(object);
347				break;
348			}
349			VM_OBJECT_WUNLOCK(object);
350		}
351	}
352	return (object);
353}
354
355/*
356 * initialize a physical buffer
357 */
358
359/*
360 * XXX This probably belongs in vfs_bio.c
361 */
362static void
363initpbuf(struct buf *bp)
364{
365	KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj"));
366	KASSERT(bp->b_vp == NULL, ("initpbuf with vp"));
367	bp->b_rcred = NOCRED;
368	bp->b_wcred = NOCRED;
369	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
370	bp->b_kvabase = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
371	bp->b_data = bp->b_kvabase;
372	bp->b_kvasize = MAXPHYS;
373	bp->b_flags = 0;
374	bp->b_xflags = 0;
375	bp->b_ioflags = 0;
376	bp->b_iodone = NULL;
377	bp->b_error = 0;
378	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
379}
380
381/*
382 * allocate a physical buffer
383 *
384 *	There are a limited number (nswbuf) of physical buffers.  We need
385 *	to make sure that no single subsystem is able to hog all of them,
386 *	so each subsystem implements a counter which is typically initialized
387 *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
388 *	increments it on release, and blocks if the counter hits zero.  A
389 *	subsystem may initialize the counter to -1 to disable the feature,
390 *	but it must still be sure to match up all uses of getpbuf() with
391 *	relpbuf() using the same variable.
392 *
393 *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
394 *	relatively soon when the rest of the subsystems get smart about it. XXX
395 */
396struct buf *
397getpbuf(int *pfreecnt)
398{
399	struct buf *bp;
400
401	mtx_lock(&pbuf_mtx);
402
403	for (;;) {
404		if (pfreecnt) {
405			while (*pfreecnt == 0) {
406				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
407			}
408		}
409
410		/* get a bp from the swap buffer header pool */
411		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
412			break;
413
414		bswneeded = 1;
415		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
416		/* loop in case someone else grabbed one */
417	}
418	TAILQ_REMOVE(&bswlist, bp, b_freelist);
419	if (pfreecnt)
420		--*pfreecnt;
421	mtx_unlock(&pbuf_mtx);
422
423	initpbuf(bp);
424	return bp;
425}
426
427/*
428 * allocate a physical buffer, if one is available.
429 *
430 *	Note that there is no NULL hack here - all subsystems using this
431 *	call understand how to use pfreecnt.
432 */
433struct buf *
434trypbuf(int *pfreecnt)
435{
436	struct buf *bp;
437
438	mtx_lock(&pbuf_mtx);
439	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
440		mtx_unlock(&pbuf_mtx);
441		return NULL;
442	}
443	TAILQ_REMOVE(&bswlist, bp, b_freelist);
444
445	--*pfreecnt;
446
447	mtx_unlock(&pbuf_mtx);
448
449	initpbuf(bp);
450
451	return bp;
452}
453
454/*
455 * release a physical buffer
456 *
457 *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
458 *	relatively soon when the rest of the subsystems get smart about it. XXX
459 */
460void
461relpbuf(struct buf *bp, int *pfreecnt)
462{
463
464	if (bp->b_rcred != NOCRED) {
465		crfree(bp->b_rcred);
466		bp->b_rcred = NOCRED;
467	}
468	if (bp->b_wcred != NOCRED) {
469		crfree(bp->b_wcred);
470		bp->b_wcred = NOCRED;
471	}
472
473	KASSERT(bp->b_vp == NULL, ("relpbuf with vp"));
474	KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj"));
475
476	BUF_UNLOCK(bp);
477
478	mtx_lock(&pbuf_mtx);
479	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
480
481	if (bswneeded) {
482		bswneeded = 0;
483		wakeup(&bswneeded);
484	}
485	if (pfreecnt) {
486		if (++*pfreecnt == 1)
487			wakeup(pfreecnt);
488	}
489	mtx_unlock(&pbuf_mtx);
490}
491
492/*
493 * Associate a p-buffer with a vnode.
494 *
495 * Also sets B_PAGING flag to indicate that vnode is not fully associated
496 * with the buffer.  i.e. the bp has not been linked into the vnode or
497 * ref-counted.
498 */
499void
500pbgetvp(struct vnode *vp, struct buf *bp)
501{
502
503	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
504	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
505
506	bp->b_vp = vp;
507	bp->b_flags |= B_PAGING;
508	bp->b_bufobj = &vp->v_bufobj;
509}
510
511/*
512 * Associate a p-buffer with a vnode.
513 *
514 * Also sets B_PAGING flag to indicate that vnode is not fully associated
515 * with the buffer.  i.e. the bp has not been linked into the vnode or
516 * ref-counted.
517 */
518void
519pbgetbo(struct bufobj *bo, struct buf *bp)
520{
521
522	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
523	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
524
525	bp->b_flags |= B_PAGING;
526	bp->b_bufobj = bo;
527}
528
529/*
530 * Disassociate a p-buffer from a vnode.
531 */
532void
533pbrelvp(struct buf *bp)
534{
535
536	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
537	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
538	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
539	    ("pbrelvp: pager buf on vnode list."));
540
541	bp->b_vp = NULL;
542	bp->b_bufobj = NULL;
543	bp->b_flags &= ~B_PAGING;
544}
545
546/*
547 * Disassociate a p-buffer from a bufobj.
548 */
549void
550pbrelbo(struct buf *bp)
551{
552
553	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
554	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
555	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
556	    ("pbrelbo: pager buf on vnode list."));
557
558	bp->b_bufobj = NULL;
559	bp->b_flags &= ~B_PAGING;
560}
561