xref: /freebsd/sys/vm/vm_pager.c (revision b52f49a9a0f22207ad5130ad8faba08de3ed23d8)
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 
65 /*
66  *	Paging space routine stubs.  Emulates a matchmaker-like interface
67  *	for builtin pagers.
68  */
69 
70 #include <sys/cdefs.h>
71 __FBSDID("$FreeBSD$");
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/kernel.h>
76 #include <sys/vnode.h>
77 #include <sys/bio.h>
78 #include <sys/buf.h>
79 #include <sys/ucred.h>
80 #include <sys/malloc.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(vm_object_t, vm_page_t *, int, int);
100 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
101 	vm_ooffset_t);
102 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
103 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
104 static void dead_pager_dealloc(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 struct mtx pbuf_mtx;
198 static TAILQ_HEAD(swqueue, buf) bswlist;
199 
200 void
201 vm_pager_init()
202 {
203 	struct pagerops **pgops;
204 
205 	TAILQ_INIT(&bswlist);
206 	/*
207 	 * Initialize known pagers
208 	 */
209 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
210 		if (pgops && ((*pgops)->pgo_init != NULL))
211 			(*(*pgops)->pgo_init) ();
212 }
213 
214 void
215 vm_pager_bufferinit()
216 {
217 	struct buf *bp;
218 	int i;
219 
220 	mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF);
221 	bp = swbuf;
222 	/*
223 	 * Now set up swap and physical I/O buffer headers.
224 	 */
225 	for (i = 0; i < nswbuf; i++, bp++) {
226 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
227 		BUF_LOCKINIT(bp);
228 		LIST_INIT(&bp->b_dep);
229 		bp->b_rcred = bp->b_wcred = NOCRED;
230 		bp->b_xflags = 0;
231 	}
232 
233 	cluster_pbuf_freecnt = nswbuf / 2;
234 
235 	swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
236 	if (!swapbkva)
237 		panic("Not enough pager_map VM space for physical buffers");
238 }
239 
240 /*
241  * Allocate an instance of a pager of the given type.
242  * Size, protection and offset parameters are passed in for pagers that
243  * need to perform page-level validation (e.g. the device pager).
244  */
245 vm_object_t
246 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
247 		  vm_prot_t prot, vm_ooffset_t off)
248 {
249 	vm_object_t ret;
250 	struct pagerops *ops;
251 
252 	ops = pagertab[type];
253 	if (ops)
254 		ret = (*ops->pgo_alloc) (handle, size, prot, off);
255 	else
256 		ret = NULL;
257 	return (ret);
258 }
259 
260 /*
261  *	The object must be locked.
262  */
263 void
264 vm_pager_deallocate(object)
265 	vm_object_t object;
266 {
267 
268 	VM_OBJECT_LOCK_ASSERT(object, 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 void
279 vm_pager_strategy(vm_object_t object, struct bio *bp)
280 {
281 	if (pagertab[object->type]->pgo_strategy) {
282 		(*pagertab[object->type]->pgo_strategy)(object, bp);
283 	} else {
284 		bp->bio_flags |= BIO_ERROR;
285 		bp->bio_error = ENXIO;
286 		biodone(bp);
287 	}
288 }
289 
290 /*
291  * vm_pager_get_pages() - inline, see vm/vm_pager.h
292  * vm_pager_put_pages() - inline, see vm/vm_pager.h
293  * vm_pager_has_page() - inline, see vm/vm_pager.h
294  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
295  * vm_pager_page_removed() - inline, see vm/vm_pager.h
296  */
297 
298 vm_offset_t
299 vm_pager_map_page(m)
300 	vm_page_t m;
301 {
302 	vm_offset_t kva;
303 
304 	kva = kmem_alloc_wait(pager_map, PAGE_SIZE);
305 	pmap_qenter(kva, &m, 1);
306 	return (kva);
307 }
308 
309 void
310 vm_pager_unmap_page(kva)
311 	vm_offset_t kva;
312 {
313 
314 	pmap_qremove(kva, 1);
315 	kmem_free_wakeup(pager_map, kva, PAGE_SIZE);
316 }
317 
318 vm_object_t
319 vm_pager_object_lookup(pg_list, handle)
320 	struct pagerlst *pg_list;
321 	void *handle;
322 {
323 	vm_object_t object;
324 
325 	TAILQ_FOREACH(object, pg_list, pager_object_list)
326 		if (object->handle == handle)
327 			return (object);
328 	return (NULL);
329 }
330 
331 /*
332  * initialize a physical buffer
333  */
334 
335 /*
336  * XXX This probably belongs in vfs_bio.c
337  */
338 static void
339 initpbuf(struct buf *bp)
340 {
341 	bp->b_rcred = NOCRED;
342 	bp->b_wcred = NOCRED;
343 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
344 	bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
345 	bp->b_data = bp->b_saveaddr;
346 	bp->b_kvabase = bp->b_saveaddr;
347 	bp->b_kvasize = MAXPHYS;
348 	bp->b_xflags = 0;
349 	bp->b_flags = 0;
350 	bp->b_ioflags = 0;
351 	bp->b_iodone = NULL;
352 	bp->b_error = 0;
353 	bp->b_magic = B_MAGIC_BIO;
354 	bp->b_op = &buf_ops_bio;
355 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
356 }
357 
358 /*
359  * allocate a physical buffer
360  *
361  *	There are a limited number (nswbuf) of physical buffers.  We need
362  *	to make sure that no single subsystem is able to hog all of them,
363  *	so each subsystem implements a counter which is typically initialized
364  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
365  *	increments it on release, and blocks if the counter hits zero.  A
366  *	subsystem may initialize the counter to -1 to disable the feature,
367  *	but it must still be sure to match up all uses of getpbuf() with
368  *	relpbuf() using the same variable.
369  *
370  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
371  *	relatively soon when the rest of the subsystems get smart about it. XXX
372  */
373 struct buf *
374 getpbuf(pfreecnt)
375 	int *pfreecnt;
376 {
377 	int s;
378 	struct buf *bp;
379 
380 	s = splvm();
381 	mtx_lock(&pbuf_mtx);
382 
383 	for (;;) {
384 		if (pfreecnt) {
385 			while (*pfreecnt == 0) {
386 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
387 			}
388 		}
389 
390 		/* get a bp from the swap buffer header pool */
391 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
392 			break;
393 
394 		bswneeded = 1;
395 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
396 		/* loop in case someone else grabbed one */
397 	}
398 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
399 	if (pfreecnt)
400 		--*pfreecnt;
401 	mtx_unlock(&pbuf_mtx);
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 	mtx_lock(&pbuf_mtx);
423 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
424 		mtx_unlock(&pbuf_mtx);
425 		splx(s);
426 		return NULL;
427 	}
428 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
429 
430 	--*pfreecnt;
431 
432 	mtx_unlock(&pbuf_mtx);
433 	splx(s);
434 
435 	initpbuf(bp);
436 
437 	return bp;
438 }
439 
440 /*
441  * release a physical buffer
442  *
443  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
444  *	relatively soon when the rest of the subsystems get smart about it. XXX
445  */
446 void
447 relpbuf(bp, pfreecnt)
448 	struct buf *bp;
449 	int *pfreecnt;
450 {
451 	int s;
452 
453 	s = splvm();
454 
455 	if (bp->b_rcred != NOCRED) {
456 		crfree(bp->b_rcred);
457 		bp->b_rcred = NOCRED;
458 	}
459 	if (bp->b_wcred != NOCRED) {
460 		crfree(bp->b_wcred);
461 		bp->b_wcred = NOCRED;
462 	}
463 
464 	if (bp->b_vp)
465 		pbrelvp(bp);
466 
467 	BUF_UNLOCK(bp);
468 
469 	mtx_lock(&pbuf_mtx);
470 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
471 
472 	if (bswneeded) {
473 		bswneeded = 0;
474 		wakeup(&bswneeded);
475 	}
476 	if (pfreecnt) {
477 		if (++*pfreecnt == 1)
478 			wakeup(pfreecnt);
479 	}
480 	mtx_unlock(&pbuf_mtx);
481 	splx(s);
482 }
483