xref: /freebsd/sys/vm/vm_pager.c (revision 1b6c76a2fe091c74f08427e6c870851025a9cf67)
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 
81 #include <vm/vm.h>
82 #include <vm/vm_param.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 
88 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
89 
90 extern struct pagerops defaultpagerops;
91 extern struct pagerops swappagerops;
92 extern struct pagerops vnodepagerops;
93 extern struct pagerops devicepagerops;
94 extern struct pagerops physpagerops;
95 
96 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
97 
98 static int dead_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
99 static vm_object_t dead_pager_alloc __P((void *, vm_ooffset_t, vm_prot_t,
100 	vm_ooffset_t));
101 static void dead_pager_putpages __P((vm_object_t, vm_page_t *, int, int, int *));
102 static boolean_t dead_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
103 static void dead_pager_dealloc __P((vm_object_t));
104 
105 static int
106 dead_pager_getpages(obj, ma, count, req)
107 	vm_object_t obj;
108 	vm_page_t *ma;
109 	int count;
110 	int req;
111 {
112 	return VM_PAGER_FAIL;
113 }
114 
115 static vm_object_t
116 dead_pager_alloc(handle, size, prot, off)
117 	void *handle;
118 	vm_ooffset_t size;
119 	vm_prot_t prot;
120 	vm_ooffset_t off;
121 {
122 	return NULL;
123 }
124 
125 static void
126 dead_pager_putpages(object, m, count, flags, rtvals)
127 	vm_object_t object;
128 	vm_page_t *m;
129 	int count;
130 	int flags;
131 	int *rtvals;
132 {
133 	int i;
134 
135 	for (i = 0; i < count; i++) {
136 		rtvals[i] = VM_PAGER_AGAIN;
137 	}
138 }
139 
140 static int
141 dead_pager_haspage(object, pindex, prev, next)
142 	vm_object_t object;
143 	vm_pindex_t pindex;
144 	int *prev;
145 	int *next;
146 {
147 	if (prev)
148 		*prev = 0;
149 	if (next)
150 		*next = 0;
151 	return FALSE;
152 }
153 
154 static void
155 dead_pager_dealloc(object)
156 	vm_object_t object;
157 {
158 	return;
159 }
160 
161 static struct pagerops deadpagerops = {
162 	NULL,
163 	dead_pager_alloc,
164 	dead_pager_dealloc,
165 	dead_pager_getpages,
166 	dead_pager_putpages,
167 	dead_pager_haspage,
168 	NULL
169 };
170 
171 struct pagerops *pagertab[] = {
172 	&defaultpagerops,	/* OBJT_DEFAULT */
173 	&swappagerops,		/* OBJT_SWAP */
174 	&vnodepagerops,		/* OBJT_VNODE */
175 	&devicepagerops,	/* OBJT_DEVICE */
176 	&physpagerops,		/* OBJT_PHYS */
177 	&deadpagerops		/* OBJT_DEAD */
178 };
179 
180 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
181 
182 /*
183  * Kernel address space for mapping pages.
184  * Used by pagers where KVAs are needed for IO.
185  *
186  * XXX needs to be large enough to support the number of pending async
187  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
188  * (MAXPHYS == 64k) if you want to get the most efficiency.
189  */
190 #define PAGER_MAP_SIZE	(8 * 1024 * 1024)
191 
192 int pager_map_size = PAGER_MAP_SIZE;
193 vm_map_t pager_map;
194 static int bswneeded;
195 static vm_offset_t swapbkva;		/* swap buffers kva */
196 struct mtx pbuf_mtx;
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 	mtx_init(&pbuf_mtx, "pbuf mutex", MTX_DEF);
218 	bp = swbuf;
219 	/*
220 	 * Now set up swap and physical I/O buffer headers.
221 	 */
222 	for (i = 0; i < nswbuf; i++, bp++) {
223 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
224 		BUF_LOCKINIT(bp);
225 		LIST_INIT(&bp->b_dep);
226 		bp->b_rcred = bp->b_wcred = NOCRED;
227 		bp->b_xflags = 0;
228 	}
229 
230 	cluster_pbuf_freecnt = nswbuf / 2;
231 
232 	swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
233 	if (!swapbkva)
234 		panic("Not enough pager_map VM space for physical buffers");
235 }
236 
237 /*
238  * Allocate an instance of a pager of the given type.
239  * Size, protection and offset parameters are passed in for pagers that
240  * need to perform page-level validation (e.g. the device pager).
241  */
242 vm_object_t
243 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
244 		  vm_prot_t prot, vm_ooffset_t off)
245 {
246 	vm_object_t ret;
247 	struct pagerops *ops;
248 	int hadvmlock;
249 
250 	hadvmlock = mtx_owned(&vm_mtx);
251 	if (!hadvmlock)
252 		mtx_lock(&vm_mtx);
253 	ops = pagertab[type];
254 	if (ops)
255 		ret = (*ops->pgo_alloc) (handle, size, prot, off);
256 	else
257 		ret = NULL;
258 	if (!hadvmlock)
259 		mtx_unlock(&vm_mtx);
260 	return (ret);
261 }
262 
263 void
264 vm_pager_deallocate(object)
265 	vm_object_t object;
266 {
267 
268 	mtx_assert(&vm_mtx, MA_OWNED);
269 	(*pagertab[object->type]->pgo_dealloc) (object);
270 }
271 
272 /*
273  *      vm_pager_strategy:
274  *
275  *      called with no specific spl
276  *      Execute strategy routine directly to pager.
277  */
278 
279 void
280 vm_pager_strategy(vm_object_t object, struct bio *bp)
281 {
282 	if (pagertab[object->type]->pgo_strategy) {
283 	    (*pagertab[object->type]->pgo_strategy)(object, bp);
284 	} else {
285 		bp->bio_flags |= BIO_ERROR;
286 		bp->bio_error = ENXIO;
287 		biodone(bp);
288 	}
289 }
290 
291 /*
292  * vm_pager_get_pages() - inline, see vm/vm_pager.h
293  * vm_pager_put_pages() - inline, see vm/vm_pager.h
294  * vm_pager_has_page() - inline, see vm/vm_pager.h
295  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
296  * vm_pager_page_removed() - inline, see vm/vm_pager.h
297  */
298 
299 #if 0
300 /*
301  *	vm_pager_sync:
302  *
303  *	Called by pageout daemon before going back to sleep.
304  *	Gives pagers a chance to clean up any completed async pageing
305  *	operations.
306  */
307 void
308 vm_pager_sync()
309 {
310 	struct pagerops **pgops;
311 
312 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
313 		if (pgops && ((*pgops)->pgo_sync != NULL))
314 			(*(*pgops)->pgo_sync) ();
315 }
316 
317 #endif
318 
319 vm_offset_t
320 vm_pager_map_page(m)
321 	vm_page_t m;
322 {
323 	vm_offset_t kva;
324 
325 	kva = kmem_alloc_wait(pager_map, PAGE_SIZE);
326 	pmap_kenter(kva, VM_PAGE_TO_PHYS(m));
327 	return (kva);
328 }
329 
330 void
331 vm_pager_unmap_page(kva)
332 	vm_offset_t kva;
333 {
334 	pmap_kremove(kva);
335 	kmem_free_wakeup(pager_map, kva, PAGE_SIZE);
336 }
337 
338 vm_object_t
339 vm_pager_object_lookup(pg_list, handle)
340 	register struct pagerlst *pg_list;
341 	void *handle;
342 {
343 	register vm_object_t object;
344 
345 	TAILQ_FOREACH(object, pg_list, pager_object_list)
346 		if (object->handle == handle)
347 			return (object);
348 	return (NULL);
349 }
350 
351 /*
352  * initialize a physical buffer
353  */
354 
355 static void
356 initpbuf(struct buf *bp)
357 {
358 	bp->b_rcred = NOCRED;
359 	bp->b_wcred = NOCRED;
360 	bp->b_qindex = QUEUE_NONE;
361 	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
362 	bp->b_kvabase = bp->b_data;
363 	bp->b_kvasize = MAXPHYS;
364 	bp->b_xflags = 0;
365 	bp->b_flags = 0;
366 	bp->b_ioflags = 0;
367 	bp->b_iodone = NULL;
368 	bp->b_error = 0;
369 	bp->b_magic = B_MAGIC_BIO;
370 	bp->b_op = &buf_ops_bio;
371 	BUF_LOCK(bp, LK_EXCLUSIVE);
372 }
373 
374 /*
375  * allocate a physical buffer
376  *
377  *	There are a limited number (nswbuf) of physical buffers.  We need
378  *	to make sure that no single subsystem is able to hog all of them,
379  *	so each subsystem implements a counter which is typically initialized
380  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
381  *	increments it on release, and blocks if the counter hits zero.  A
382  *	subsystem may initialize the counter to -1 to disable the feature,
383  *	but it must still be sure to match up all uses of getpbuf() with
384  *	relpbuf() using the same variable.
385  *
386  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
387  *	relatively soon when the rest of the subsystems get smart about it. XXX
388  *
389  *	vm_mtx can be held or unheld
390  */
391 struct buf *
392 getpbuf(pfreecnt)
393 	int *pfreecnt;
394 {
395 	int s;
396 	struct buf *bp;
397 	int hadvmlock;
398 
399 	s = splvm();
400 	hadvmlock = mtx_owned(&vm_mtx);
401 	if (hadvmlock)
402 		mtx_unlock(&vm_mtx);
403 	mtx_lock(&pbuf_mtx);
404 
405 	for (;;) {
406 		if (pfreecnt) {
407 			while (*pfreecnt == 0) {
408 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
409 			}
410 		}
411 
412 		/* get a bp from the swap buffer header pool */
413 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
414 			break;
415 
416 		bswneeded = 1;
417 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
418 		/* loop in case someone else grabbed one */
419 	}
420 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
421 	if (pfreecnt)
422 		--*pfreecnt;
423 	mtx_unlock(&pbuf_mtx);
424 	splx(s);
425 
426 	initpbuf(bp);
427 	if (hadvmlock)
428 		mtx_lock(&vm_mtx);
429 	return bp;
430 }
431 
432 /*
433  * allocate a physical buffer, if one is available.
434  *
435  *	Note that there is no NULL hack here - all subsystems using this
436  *	call understand how to use pfreecnt.
437  */
438 struct buf *
439 trypbuf(pfreecnt)
440 	int *pfreecnt;
441 {
442 	int s;
443 	struct buf *bp;
444 
445 	s = splvm();
446 	mtx_lock(&pbuf_mtx);
447 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
448 		mtx_unlock(&pbuf_mtx);
449 		splx(s);
450 		return NULL;
451 	}
452 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
453 
454 	--*pfreecnt;
455 
456 	mtx_unlock(&pbuf_mtx);
457 	splx(s);
458 
459 	initpbuf(bp);
460 
461 	return bp;
462 }
463 
464 /*
465  * release a physical buffer
466  *
467  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
468  *	relatively soon when the rest of the subsystems get smart about it. XXX
469  */
470 void
471 relpbuf(bp, pfreecnt)
472 	struct buf *bp;
473 	int *pfreecnt;
474 {
475 	int s;
476 
477 	s = splvm();
478 	mtx_lock(&pbuf_mtx);
479 
480 	if (bp->b_rcred != NOCRED) {
481 		crfree(bp->b_rcred);
482 		bp->b_rcred = NOCRED;
483 	}
484 	if (bp->b_wcred != NOCRED) {
485 		crfree(bp->b_wcred);
486 		bp->b_wcred = NOCRED;
487 	}
488 
489 	if (bp->b_vp)
490 		pbrelvp(bp);
491 
492 	BUF_UNLOCK(bp);
493 
494 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
495 
496 	if (bswneeded) {
497 		bswneeded = 0;
498 		wakeup(&bswneeded);
499 	}
500 	if (pfreecnt) {
501 		if (++*pfreecnt == 1)
502 			wakeup(pfreecnt);
503 	}
504 	mtx_unlock(&pbuf_mtx);
505 	splx(s);
506 }
507