xref: /freebsd/sys/vm/vm_pager.c (revision ebbd4fa8c8427d3dd847ba33c45c996e0500e6ff)
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(vm_object_t, vm_page_t *, int, int);
99 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
100 	vm_ooffset_t);
101 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
102 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
103 static void dead_pager_dealloc(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 static TAILQ_HEAD(swqueue, buf) bswlist;
198 
199 void
200 vm_pager_init()
201 {
202 	struct pagerops **pgops;
203 
204 	TAILQ_INIT(&bswlist);
205 	/*
206 	 * Initialize known pagers
207 	 */
208 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
209 		if (pgops && ((*pgops)->pgo_init != NULL))
210 			(*(*pgops)->pgo_init) ();
211 }
212 
213 void
214 vm_pager_bufferinit()
215 {
216 	struct buf *bp;
217 	int i;
218 
219 	mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF);
220 	bp = swbuf;
221 	/*
222 	 * Now set up swap and physical I/O buffer headers.
223 	 */
224 	for (i = 0; i < nswbuf; i++, bp++) {
225 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
226 		BUF_LOCKINIT(bp);
227 		LIST_INIT(&bp->b_dep);
228 		bp->b_rcred = bp->b_wcred = NOCRED;
229 		bp->b_xflags = 0;
230 	}
231 
232 	cluster_pbuf_freecnt = nswbuf / 2;
233 
234 	swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
235 	if (!swapbkva)
236 		panic("Not enough pager_map VM space for physical buffers");
237 }
238 
239 /*
240  * Allocate an instance of a pager of the given type.
241  * Size, protection and offset parameters are passed in for pagers that
242  * need to perform page-level validation (e.g. the device pager).
243  */
244 vm_object_t
245 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
246 		  vm_prot_t prot, vm_ooffset_t off)
247 {
248 	vm_object_t ret;
249 	struct pagerops *ops;
250 
251 	ops = pagertab[type];
252 	if (ops)
253 		ret = (*ops->pgo_alloc) (handle, size, prot, off);
254 	else
255 		ret = NULL;
256 	return (ret);
257 }
258 
259 void
260 vm_pager_deallocate(object)
261 	vm_object_t object;
262 {
263 	(*pagertab[object->type]->pgo_dealloc) (object);
264 }
265 
266 /*
267  *      vm_pager_strategy:
268  *
269  *      called with no specific spl
270  *      Execute strategy routine directly to pager.
271  */
272 void
273 vm_pager_strategy(vm_object_t object, struct bio *bp)
274 {
275 	if (pagertab[object->type]->pgo_strategy) {
276 		(*pagertab[object->type]->pgo_strategy)(object, bp);
277 	} else {
278 		bp->bio_flags |= BIO_ERROR;
279 		bp->bio_error = ENXIO;
280 		biodone(bp);
281 	}
282 }
283 
284 /*
285  * vm_pager_get_pages() - inline, see vm/vm_pager.h
286  * vm_pager_put_pages() - inline, see vm/vm_pager.h
287  * vm_pager_has_page() - inline, see vm/vm_pager.h
288  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
289  * vm_pager_page_removed() - inline, see vm/vm_pager.h
290  */
291 
292 vm_offset_t
293 vm_pager_map_page(m)
294 	vm_page_t m;
295 {
296 	vm_offset_t kva;
297 
298 	kva = kmem_alloc_wait(pager_map, PAGE_SIZE);
299 	pmap_qenter(kva, &m, 1);
300 	return (kva);
301 }
302 
303 void
304 vm_pager_unmap_page(kva)
305 	vm_offset_t kva;
306 {
307 
308 	pmap_qremove(kva, 1);
309 	kmem_free_wakeup(pager_map, kva, PAGE_SIZE);
310 }
311 
312 vm_object_t
313 vm_pager_object_lookup(pg_list, handle)
314 	struct pagerlst *pg_list;
315 	void *handle;
316 {
317 	vm_object_t object;
318 
319 	TAILQ_FOREACH(object, pg_list, pager_object_list)
320 		if (object->handle == handle)
321 			return (object);
322 	return (NULL);
323 }
324 
325 /*
326  * initialize a physical buffer
327  */
328 
329 /*
330  * XXX This probably belongs in vfs_bio.c
331  */
332 static void
333 initpbuf(struct buf *bp)
334 {
335 	bp->b_rcred = NOCRED;
336 	bp->b_wcred = NOCRED;
337 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
338 	bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
339 	bp->b_kvabase = bp->b_data;
340 	bp->b_kvasize = MAXPHYS;
341 	bp->b_xflags = 0;
342 	bp->b_flags = 0;
343 	bp->b_ioflags = 0;
344 	bp->b_iodone = NULL;
345 	bp->b_error = 0;
346 	bp->b_magic = B_MAGIC_BIO;
347 	bp->b_op = &buf_ops_bio;
348 	BUF_LOCK(bp, LK_EXCLUSIVE);
349 }
350 
351 /*
352  * allocate a physical buffer
353  *
354  *	There are a limited number (nswbuf) of physical buffers.  We need
355  *	to make sure that no single subsystem is able to hog all of them,
356  *	so each subsystem implements a counter which is typically initialized
357  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
358  *	increments it on release, and blocks if the counter hits zero.  A
359  *	subsystem may initialize the counter to -1 to disable the feature,
360  *	but it must still be sure to match up all uses of getpbuf() with
361  *	relpbuf() using the same variable.
362  *
363  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
364  *	relatively soon when the rest of the subsystems get smart about it. XXX
365  */
366 struct buf *
367 getpbuf(pfreecnt)
368 	int *pfreecnt;
369 {
370 	int s;
371 	struct buf *bp;
372 
373 	s = splvm();
374 	GIANT_REQUIRED;
375 	mtx_lock(&pbuf_mtx);
376 
377 	for (;;) {
378 		if (pfreecnt) {
379 			while (*pfreecnt == 0) {
380 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
381 			}
382 		}
383 
384 		/* get a bp from the swap buffer header pool */
385 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
386 			break;
387 
388 		bswneeded = 1;
389 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
390 		/* loop in case someone else grabbed one */
391 	}
392 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
393 	if (pfreecnt)
394 		--*pfreecnt;
395 	mtx_unlock(&pbuf_mtx);
396 	splx(s);
397 
398 	initpbuf(bp);
399 	return bp;
400 }
401 
402 /*
403  * allocate a physical buffer, if one is available.
404  *
405  *	Note that there is no NULL hack here - all subsystems using this
406  *	call understand how to use pfreecnt.
407  */
408 struct buf *
409 trypbuf(pfreecnt)
410 	int *pfreecnt;
411 {
412 	int s;
413 	struct buf *bp;
414 
415 	s = splvm();
416 	mtx_lock(&pbuf_mtx);
417 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
418 		mtx_unlock(&pbuf_mtx);
419 		splx(s);
420 		return NULL;
421 	}
422 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
423 
424 	--*pfreecnt;
425 
426 	mtx_unlock(&pbuf_mtx);
427 	splx(s);
428 
429 	initpbuf(bp);
430 
431 	return bp;
432 }
433 
434 /*
435  * release a physical buffer
436  *
437  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
438  *	relatively soon when the rest of the subsystems get smart about it. XXX
439  */
440 void
441 relpbuf(bp, pfreecnt)
442 	struct buf *bp;
443 	int *pfreecnt;
444 {
445 	int s;
446 
447 	s = splvm();
448 	mtx_lock(&pbuf_mtx);
449 
450 	if (bp->b_rcred != NOCRED) {
451 		crfree(bp->b_rcred);
452 		bp->b_rcred = NOCRED;
453 	}
454 	if (bp->b_wcred != NOCRED) {
455 		crfree(bp->b_wcred);
456 		bp->b_wcred = NOCRED;
457 	}
458 
459 	if (bp->b_vp)
460 		pbrelvp(bp);
461 
462 	BUF_UNLOCK(bp);
463 
464 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
465 
466 	if (bswneeded) {
467 		bswneeded = 0;
468 		wakeup(&bswneeded);
469 	}
470 	if (pfreecnt) {
471 		if (++*pfreecnt == 1)
472 			wakeup(pfreecnt);
473 	}
474 	mtx_unlock(&pbuf_mtx);
475 	splx(s);
476 }
477