xref: /freebsd/sys/vm/vm_pager.c (revision a79b71281cd63ad7a6cc43a6d5673a2510b51630)
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  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_pager.c	8.6 (Berkeley) 1/12/94
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD$
65  */
66 
67 /*
68  *	Paging space routine stubs.  Emulates a matchmaker-like interface
69  *	for builtin pagers.
70  */
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/kernel.h>
75 #include <sys/vnode.h>
76 #include <sys/bio.h>
77 #include <sys/buf.h>
78 #include <sys/ucred.h>
79 #include <sys/malloc.h>
80 #include <sys/proc.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_pager.h>
87 #include <vm/vm_extern.h>
88 
89 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
90 
91 extern struct pagerops defaultpagerops;
92 extern struct pagerops swappagerops;
93 extern struct pagerops vnodepagerops;
94 extern struct pagerops devicepagerops;
95 extern struct pagerops physpagerops;
96 
97 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
98 
99 static int dead_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
100 static vm_object_t dead_pager_alloc __P((void *, vm_ooffset_t, vm_prot_t,
101 	vm_ooffset_t));
102 static void dead_pager_putpages __P((vm_object_t, vm_page_t *, int, int, int *));
103 static boolean_t dead_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
104 static void dead_pager_dealloc __P((vm_object_t));
105 
106 static int
107 dead_pager_getpages(obj, ma, count, req)
108 	vm_object_t obj;
109 	vm_page_t *ma;
110 	int count;
111 	int req;
112 {
113 	return VM_PAGER_FAIL;
114 }
115 
116 static vm_object_t
117 dead_pager_alloc(handle, size, prot, off)
118 	void *handle;
119 	vm_ooffset_t size;
120 	vm_prot_t prot;
121 	vm_ooffset_t off;
122 {
123 	return NULL;
124 }
125 
126 static void
127 dead_pager_putpages(object, m, count, flags, rtvals)
128 	vm_object_t object;
129 	vm_page_t *m;
130 	int count;
131 	int flags;
132 	int *rtvals;
133 {
134 	int i;
135 
136 	for (i = 0; i < count; i++) {
137 		rtvals[i] = VM_PAGER_AGAIN;
138 	}
139 }
140 
141 static int
142 dead_pager_haspage(object, pindex, prev, next)
143 	vm_object_t object;
144 	vm_pindex_t pindex;
145 	int *prev;
146 	int *next;
147 {
148 	if (prev)
149 		*prev = 0;
150 	if (next)
151 		*next = 0;
152 	return FALSE;
153 }
154 
155 static void
156 dead_pager_dealloc(object)
157 	vm_object_t object;
158 {
159 	return;
160 }
161 
162 static struct pagerops deadpagerops = {
163 	NULL,
164 	dead_pager_alloc,
165 	dead_pager_dealloc,
166 	dead_pager_getpages,
167 	dead_pager_putpages,
168 	dead_pager_haspage,
169 	NULL
170 };
171 
172 struct pagerops *pagertab[] = {
173 	&defaultpagerops,	/* OBJT_DEFAULT */
174 	&swappagerops,		/* OBJT_SWAP */
175 	&vnodepagerops,		/* OBJT_VNODE */
176 	&devicepagerops,	/* OBJT_DEVICE */
177 	&physpagerops,		/* OBJT_PHYS */
178 	&deadpagerops		/* OBJT_DEAD */
179 };
180 
181 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
182 
183 /*
184  * Kernel address space for mapping pages.
185  * Used by pagers where KVAs are needed for IO.
186  *
187  * XXX needs to be large enough to support the number of pending async
188  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
189  * (MAXPHYS == 64k) if you want to get the most efficiency.
190  */
191 #define PAGER_MAP_SIZE	(8 * 1024 * 1024)
192 
193 int pager_map_size = PAGER_MAP_SIZE;
194 vm_map_t pager_map;
195 static int bswneeded;
196 static vm_offset_t swapbkva;		/* swap buffers kva */
197 
198 void
199 vm_pager_init()
200 {
201 	struct pagerops **pgops;
202 
203 	/*
204 	 * Initialize known pagers
205 	 */
206 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
207 		if (pgops && ((*pgops)->pgo_init != NULL))
208 			(*(*pgops)->pgo_init) ();
209 }
210 
211 void
212 vm_pager_bufferinit()
213 {
214 	struct buf *bp;
215 	int i;
216 
217 	bp = swbuf;
218 	/*
219 	 * Now set up swap and physical I/O buffer headers.
220 	 */
221 	for (i = 0; i < nswbuf; i++, bp++) {
222 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
223 		BUF_LOCKINIT(bp);
224 		LIST_INIT(&bp->b_dep);
225 		bp->b_rcred = bp->b_wcred = NOCRED;
226 		bp->b_xflags = 0;
227 	}
228 
229 	cluster_pbuf_freecnt = nswbuf / 2;
230 
231 	swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
232 	if (!swapbkva)
233 		panic("Not enough pager_map VM space for physical buffers");
234 }
235 
236 /*
237  * Allocate an instance of a pager of the given type.
238  * Size, protection and offset parameters are passed in for pagers that
239  * need to perform page-level validation (e.g. the device pager).
240  */
241 vm_object_t
242 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot,
243 		  vm_ooffset_t off)
244 {
245 	struct pagerops *ops;
246 
247 	ops = pagertab[type];
248 	if (ops)
249 		return ((*ops->pgo_alloc) (handle, size, prot, off));
250 	return (NULL);
251 }
252 
253 void
254 vm_pager_deallocate(object)
255 	vm_object_t object;
256 {
257 	(*pagertab[object->type]->pgo_dealloc) (object);
258 }
259 
260 /*
261  *      vm_pager_strategy:
262  *
263  *      called with no specific spl
264  *      Execute strategy routine directly to pager.
265  */
266 
267 void
268 vm_pager_strategy(vm_object_t object, struct bio *bp)
269 {
270 	if (pagertab[object->type]->pgo_strategy) {
271 	    (*pagertab[object->type]->pgo_strategy)(object, bp);
272 	} else {
273 		bp->bio_flags |= BIO_ERROR;
274 		bp->bio_error = ENXIO;
275 		biodone(bp);
276 	}
277 }
278 
279 /*
280  * vm_pager_get_pages() - inline, see vm/vm_pager.h
281  * vm_pager_put_pages() - inline, see vm/vm_pager.h
282  * vm_pager_has_page() - inline, see vm/vm_pager.h
283  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
284  * vm_pager_page_removed() - inline, see vm/vm_pager.h
285  */
286 
287 #if 0
288 /*
289  *	vm_pager_sync:
290  *
291  *	Called by pageout daemon before going back to sleep.
292  *	Gives pagers a chance to clean up any completed async pageing
293  *	operations.
294  */
295 void
296 vm_pager_sync()
297 {
298 	struct pagerops **pgops;
299 
300 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
301 		if (pgops && ((*pgops)->pgo_sync != NULL))
302 			(*(*pgops)->pgo_sync) ();
303 }
304 
305 #endif
306 
307 vm_offset_t
308 vm_pager_map_page(m)
309 	vm_page_t m;
310 {
311 	vm_offset_t kva;
312 
313 	kva = kmem_alloc_wait(pager_map, PAGE_SIZE);
314 	pmap_kenter(kva, VM_PAGE_TO_PHYS(m));
315 	return (kva);
316 }
317 
318 void
319 vm_pager_unmap_page(kva)
320 	vm_offset_t kva;
321 {
322 	pmap_kremove(kva);
323 	kmem_free_wakeup(pager_map, kva, PAGE_SIZE);
324 }
325 
326 vm_object_t
327 vm_pager_object_lookup(pg_list, handle)
328 	register struct pagerlst *pg_list;
329 	void *handle;
330 {
331 	register vm_object_t object;
332 
333 	for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list))
334 		if (object->handle == handle)
335 			return (object);
336 	return (NULL);
337 }
338 
339 /*
340  * initialize a physical buffer
341  */
342 
343 static void
344 initpbuf(struct buf *bp)
345 {
346 	bp->b_rcred = NOCRED;
347 	bp->b_wcred = NOCRED;
348 	bp->b_qindex = QUEUE_NONE;
349 	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
350 	bp->b_kvabase = bp->b_data;
351 	bp->b_kvasize = MAXPHYS;
352 	bp->b_xflags = 0;
353 	bp->b_flags = 0;
354 	bp->b_ioflags = 0;
355 	bp->b_iodone = NULL;
356 	bp->b_error = 0;
357 	BUF_LOCK(bp, LK_EXCLUSIVE);
358 }
359 
360 /*
361  * allocate a physical buffer
362  *
363  *	There are a limited number (nswbuf) of physical buffers.  We need
364  *	to make sure that no single subsystem is able to hog all of them,
365  *	so each subsystem implements a counter which is typically initialized
366  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
367  *	increments it on release, and blocks if the counter hits zero.  A
368  *	subsystem may initialize the counter to -1 to disable the feature,
369  *	but it must still be sure to match up all uses of getpbuf() with
370  *	relpbuf() using the same variable.
371  *
372  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
373  *	relatively soon when the rest of the subsystems get smart about it. XXX
374  */
375 struct buf *
376 getpbuf(pfreecnt)
377 	int *pfreecnt;
378 {
379 	int s;
380 	struct buf *bp;
381 
382 	s = splvm();
383 
384 	for (;;) {
385 		if (pfreecnt) {
386 			while (*pfreecnt == 0) {
387 				tsleep(pfreecnt, PVM, "wswbuf0", 0);
388 			}
389 		}
390 
391 		/* get a bp from the swap buffer header pool */
392 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
393 			break;
394 
395 		bswneeded = 1;
396 		tsleep(&bswneeded, PVM, "wswbuf1", 0);
397 		/* loop in case someone else grabbed one */
398 	}
399 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
400 	if (pfreecnt)
401 		--*pfreecnt;
402 	splx(s);
403 
404 	initpbuf(bp);
405 	return bp;
406 }
407 
408 /*
409  * allocate a physical buffer, if one is available.
410  *
411  *	Note that there is no NULL hack here - all subsystems using this
412  *	call understand how to use pfreecnt.
413  */
414 struct buf *
415 trypbuf(pfreecnt)
416 	int *pfreecnt;
417 {
418 	int s;
419 	struct buf *bp;
420 
421 	s = splvm();
422 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
423 		splx(s);
424 		return NULL;
425 	}
426 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
427 
428 	--*pfreecnt;
429 
430 	splx(s);
431 
432 	initpbuf(bp);
433 
434 	return bp;
435 }
436 
437 /*
438  * release a physical buffer
439  *
440  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
441  *	relatively soon when the rest of the subsystems get smart about it. XXX
442  */
443 void
444 relpbuf(bp, pfreecnt)
445 	struct buf *bp;
446 	int *pfreecnt;
447 {
448 	int s;
449 
450 	s = splvm();
451 
452 	if (bp->b_rcred != NOCRED) {
453 		crfree(bp->b_rcred);
454 		bp->b_rcred = NOCRED;
455 	}
456 	if (bp->b_wcred != NOCRED) {
457 		crfree(bp->b_wcred);
458 		bp->b_wcred = NOCRED;
459 	}
460 
461 	if (bp->b_vp)
462 		pbrelvp(bp);
463 
464 	BUF_UNLOCK(bp);
465 
466 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
467 
468 	if (bswneeded) {
469 		bswneeded = 0;
470 		wakeup(&bswneeded);
471 	}
472 	if (pfreecnt) {
473 		if (++*pfreecnt == 1)
474 			wakeup(pfreecnt);
475 	}
476 	splx(s);
477 }
478