xref: /freebsd/sys/vm/vm_pager.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vm_extern.h>
84 
85 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
86 
87 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int);
88 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
89     vm_ooffset_t, struct ucred *);
90 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
91 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
92 static void dead_pager_dealloc(vm_object_t);
93 
94 static int
95 dead_pager_getpages(obj, ma, count, req)
96 	vm_object_t obj;
97 	vm_page_t *ma;
98 	int count;
99 	int req;
100 {
101 	return VM_PAGER_FAIL;
102 }
103 
104 static vm_object_t
105 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
106     vm_ooffset_t off, struct ucred *cred)
107 {
108 	return NULL;
109 }
110 
111 static void
112 dead_pager_putpages(object, m, count, flags, rtvals)
113 	vm_object_t object;
114 	vm_page_t *m;
115 	int count;
116 	int flags;
117 	int *rtvals;
118 {
119 	int i;
120 
121 	for (i = 0; i < count; i++) {
122 		rtvals[i] = VM_PAGER_AGAIN;
123 	}
124 }
125 
126 static int
127 dead_pager_haspage(object, pindex, prev, next)
128 	vm_object_t object;
129 	vm_pindex_t pindex;
130 	int *prev;
131 	int *next;
132 {
133 	if (prev)
134 		*prev = 0;
135 	if (next)
136 		*next = 0;
137 	return FALSE;
138 }
139 
140 static void
141 dead_pager_dealloc(object)
142 	vm_object_t object;
143 {
144 	return;
145 }
146 
147 static struct pagerops deadpagerops = {
148 	.pgo_alloc = 	dead_pager_alloc,
149 	.pgo_dealloc =	dead_pager_dealloc,
150 	.pgo_getpages =	dead_pager_getpages,
151 	.pgo_putpages =	dead_pager_putpages,
152 	.pgo_haspage =	dead_pager_haspage,
153 };
154 
155 struct pagerops *pagertab[] = {
156 	&defaultpagerops,	/* OBJT_DEFAULT */
157 	&swappagerops,		/* OBJT_SWAP */
158 	&vnodepagerops,		/* OBJT_VNODE */
159 	&devicepagerops,	/* OBJT_DEVICE */
160 	&physpagerops,		/* OBJT_PHYS */
161 	&deadpagerops,		/* OBJT_DEAD */
162 	&sgpagerops		/* OBJT_SG */
163 };
164 
165 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
166 
167 /*
168  * Kernel address space for mapping pages.
169  * Used by pagers where KVAs are needed for IO.
170  *
171  * XXX needs to be large enough to support the number of pending async
172  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
173  * (MAXPHYS == 64k) if you want to get the most efficiency.
174  */
175 vm_map_t pager_map;
176 static int bswneeded;
177 static vm_offset_t swapbkva;		/* swap buffers kva */
178 struct mtx pbuf_mtx;
179 static TAILQ_HEAD(swqueue, buf) bswlist;
180 
181 void
182 vm_pager_init()
183 {
184 	struct pagerops **pgops;
185 
186 	TAILQ_INIT(&bswlist);
187 	/*
188 	 * Initialize known pagers
189 	 */
190 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
191 		if ((*pgops)->pgo_init != NULL)
192 			(*(*pgops)->pgo_init) ();
193 }
194 
195 void
196 vm_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 
217 	swapbkva = kmem_alloc_nofault(pager_map, nswbuf * MAXPHYS);
218 	if (!swapbkva)
219 		panic("Not enough pager_map VM space for physical buffers");
220 }
221 
222 /*
223  * Allocate an instance of a pager of the given type.
224  * Size, protection and offset parameters are passed in for pagers that
225  * need to perform page-level validation (e.g. the device pager).
226  */
227 vm_object_t
228 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
229     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
230 {
231 	vm_object_t ret;
232 	struct pagerops *ops;
233 
234 	ops = pagertab[type];
235 	if (ops)
236 		ret = (*ops->pgo_alloc) (handle, size, prot, off, cred);
237 	else
238 		ret = NULL;
239 	return (ret);
240 }
241 
242 /*
243  *	The object must be locked.
244  */
245 void
246 vm_pager_deallocate(object)
247 	vm_object_t object;
248 {
249 
250 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
251 	(*pagertab[object->type]->pgo_dealloc) (object);
252 }
253 
254 /*
255  * vm_pager_get_pages() - inline, see vm/vm_pager.h
256  * vm_pager_put_pages() - inline, see vm/vm_pager.h
257  * vm_pager_has_page() - inline, see vm/vm_pager.h
258  */
259 
260 /*
261  * Search the specified pager object list for an object with the
262  * specified handle.  If an object with the specified handle is found,
263  * increase its reference count and return it.  Otherwise, return NULL.
264  *
265  * The pager object list must be locked.
266  */
267 vm_object_t
268 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
269 {
270 	vm_object_t object;
271 
272 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
273 		VM_OBJECT_LOCK(object);
274 		if (object->handle == handle &&
275 		    (object->flags & OBJ_DEAD) == 0) {
276 			vm_object_reference_locked(object);
277 			VM_OBJECT_UNLOCK(object);
278 			break;
279 		}
280 		VM_OBJECT_UNLOCK(object);
281 	}
282 	return (object);
283 }
284 
285 /*
286  * initialize a physical buffer
287  */
288 
289 /*
290  * XXX This probably belongs in vfs_bio.c
291  */
292 static void
293 initpbuf(struct buf *bp)
294 {
295 	KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj"));
296 	KASSERT(bp->b_vp == NULL, ("initpbuf with vp"));
297 	bp->b_rcred = NOCRED;
298 	bp->b_wcred = NOCRED;
299 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
300 	bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
301 	bp->b_data = bp->b_saveaddr;
302 	bp->b_kvabase = bp->b_saveaddr;
303 	bp->b_kvasize = MAXPHYS;
304 	bp->b_xflags = 0;
305 	bp->b_flags = 0;
306 	bp->b_ioflags = 0;
307 	bp->b_iodone = NULL;
308 	bp->b_error = 0;
309 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
310 }
311 
312 /*
313  * allocate a physical buffer
314  *
315  *	There are a limited number (nswbuf) of physical buffers.  We need
316  *	to make sure that no single subsystem is able to hog all of them,
317  *	so each subsystem implements a counter which is typically initialized
318  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
319  *	increments it on release, and blocks if the counter hits zero.  A
320  *	subsystem may initialize the counter to -1 to disable the feature,
321  *	but it must still be sure to match up all uses of getpbuf() with
322  *	relpbuf() using the same variable.
323  *
324  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
325  *	relatively soon when the rest of the subsystems get smart about it. XXX
326  */
327 struct buf *
328 getpbuf(int *pfreecnt)
329 {
330 	struct buf *bp;
331 
332 	mtx_lock(&pbuf_mtx);
333 
334 	for (;;) {
335 		if (pfreecnt) {
336 			while (*pfreecnt == 0) {
337 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
338 			}
339 		}
340 
341 		/* get a bp from the swap buffer header pool */
342 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
343 			break;
344 
345 		bswneeded = 1;
346 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
347 		/* loop in case someone else grabbed one */
348 	}
349 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
350 	if (pfreecnt)
351 		--*pfreecnt;
352 	mtx_unlock(&pbuf_mtx);
353 
354 	initpbuf(bp);
355 	return bp;
356 }
357 
358 /*
359  * allocate a physical buffer, if one is available.
360  *
361  *	Note that there is no NULL hack here - all subsystems using this
362  *	call understand how to use pfreecnt.
363  */
364 struct buf *
365 trypbuf(int *pfreecnt)
366 {
367 	struct buf *bp;
368 
369 	mtx_lock(&pbuf_mtx);
370 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
371 		mtx_unlock(&pbuf_mtx);
372 		return NULL;
373 	}
374 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
375 
376 	--*pfreecnt;
377 
378 	mtx_unlock(&pbuf_mtx);
379 
380 	initpbuf(bp);
381 
382 	return bp;
383 }
384 
385 /*
386  * release a physical buffer
387  *
388  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
389  *	relatively soon when the rest of the subsystems get smart about it. XXX
390  */
391 void
392 relpbuf(struct buf *bp, int *pfreecnt)
393 {
394 
395 	if (bp->b_rcred != NOCRED) {
396 		crfree(bp->b_rcred);
397 		bp->b_rcred = NOCRED;
398 	}
399 	if (bp->b_wcred != NOCRED) {
400 		crfree(bp->b_wcred);
401 		bp->b_wcred = NOCRED;
402 	}
403 
404 	KASSERT(bp->b_vp == NULL, ("relpbuf with vp"));
405 	KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj"));
406 
407 	BUF_UNLOCK(bp);
408 
409 	mtx_lock(&pbuf_mtx);
410 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
411 
412 	if (bswneeded) {
413 		bswneeded = 0;
414 		wakeup(&bswneeded);
415 	}
416 	if (pfreecnt) {
417 		if (++*pfreecnt == 1)
418 			wakeup(pfreecnt);
419 	}
420 	mtx_unlock(&pbuf_mtx);
421 }
422 
423 /*
424  * Associate a p-buffer with a vnode.
425  *
426  * Also sets B_PAGING flag to indicate that vnode is not fully associated
427  * with the buffer.  i.e. the bp has not been linked into the vnode or
428  * ref-counted.
429  */
430 void
431 pbgetvp(struct vnode *vp, struct buf *bp)
432 {
433 
434 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
435 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
436 
437 	bp->b_vp = vp;
438 	bp->b_flags |= B_PAGING;
439 	bp->b_bufobj = &vp->v_bufobj;
440 }
441 
442 /*
443  * Associate a p-buffer with a vnode.
444  *
445  * Also sets B_PAGING flag to indicate that vnode is not fully associated
446  * with the buffer.  i.e. the bp has not been linked into the vnode or
447  * ref-counted.
448  */
449 void
450 pbgetbo(struct bufobj *bo, struct buf *bp)
451 {
452 
453 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
454 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
455 
456 	bp->b_flags |= B_PAGING;
457 	bp->b_bufobj = bo;
458 }
459 
460 /*
461  * Disassociate a p-buffer from a vnode.
462  */
463 void
464 pbrelvp(struct buf *bp)
465 {
466 
467 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
468 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
469 
470 	/* XXX REMOVE ME */
471 	BO_LOCK(bp->b_bufobj);
472 	if (TAILQ_NEXT(bp, b_bobufs) != NULL) {
473 		panic(
474 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
475 		    bp,
476 		    (int)bp->b_flags
477 		);
478 	}
479 	BO_UNLOCK(bp->b_bufobj);
480 	bp->b_vp = NULL;
481 	bp->b_bufobj = NULL;
482 	bp->b_flags &= ~B_PAGING;
483 }
484 
485 /*
486  * Disassociate a p-buffer from a bufobj.
487  */
488 void
489 pbrelbo(struct buf *bp)
490 {
491 
492 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
493 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
494 
495 	/* XXX REMOVE ME */
496 	BO_LOCK(bp->b_bufobj);
497 	if (TAILQ_NEXT(bp, b_bobufs) != NULL) {
498 		panic(
499 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
500 		    bp,
501 		    (int)bp->b_flags
502 		);
503 	}
504 	BO_UNLOCK(bp->b_bufobj);
505 	bp->b_bufobj = NULL;
506 	bp->b_flags &= ~B_PAGING;
507 }
508