xref: /freebsd/sys/vm/vm_pager.c (revision 1670a1c2a47d10ecccd001970b859caf93cd3b6e)
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 MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "XXX: VM pager private data");
86 
87 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
88 
89 static int dead_pager_getpages(vm_object_t, vm_page_t *, int, int);
90 static vm_object_t dead_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
91     vm_ooffset_t, struct ucred *);
92 static void dead_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
93 static boolean_t dead_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
94 static void dead_pager_dealloc(vm_object_t);
95 
96 static int
97 dead_pager_getpages(obj, ma, count, req)
98 	vm_object_t obj;
99 	vm_page_t *ma;
100 	int count;
101 	int req;
102 {
103 	return VM_PAGER_FAIL;
104 }
105 
106 static vm_object_t
107 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
108     vm_ooffset_t off, struct ucred *cred)
109 {
110 	return NULL;
111 }
112 
113 static void
114 dead_pager_putpages(object, m, count, flags, rtvals)
115 	vm_object_t object;
116 	vm_page_t *m;
117 	int count;
118 	int flags;
119 	int *rtvals;
120 {
121 	int i;
122 
123 	for (i = 0; i < count; i++) {
124 		rtvals[i] = VM_PAGER_AGAIN;
125 	}
126 }
127 
128 static int
129 dead_pager_haspage(object, pindex, prev, next)
130 	vm_object_t object;
131 	vm_pindex_t pindex;
132 	int *prev;
133 	int *next;
134 {
135 	if (prev)
136 		*prev = 0;
137 	if (next)
138 		*next = 0;
139 	return FALSE;
140 }
141 
142 static void
143 dead_pager_dealloc(object)
144 	vm_object_t object;
145 {
146 	return;
147 }
148 
149 static struct pagerops deadpagerops = {
150 	.pgo_alloc = 	dead_pager_alloc,
151 	.pgo_dealloc =	dead_pager_dealloc,
152 	.pgo_getpages =	dead_pager_getpages,
153 	.pgo_putpages =	dead_pager_putpages,
154 	.pgo_haspage =	dead_pager_haspage,
155 };
156 
157 struct pagerops *pagertab[] = {
158 	&defaultpagerops,	/* OBJT_DEFAULT */
159 	&swappagerops,		/* OBJT_SWAP */
160 	&vnodepagerops,		/* OBJT_VNODE */
161 	&devicepagerops,	/* OBJT_DEVICE */
162 	&physpagerops,		/* OBJT_PHYS */
163 	&deadpagerops,		/* OBJT_DEAD */
164 	&sgpagerops		/* OBJT_SG */
165 };
166 
167 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
168 
169 /*
170  * Kernel address space for mapping pages.
171  * Used by pagers where KVAs are needed for IO.
172  *
173  * XXX needs to be large enough to support the number of pending async
174  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
175  * (MAXPHYS == 64k) if you want to get the most efficiency.
176  */
177 vm_map_t pager_map;
178 static int bswneeded;
179 static vm_offset_t swapbkva;		/* swap buffers kva */
180 struct mtx pbuf_mtx;
181 static TAILQ_HEAD(swqueue, buf) bswlist;
182 
183 void
184 vm_pager_init()
185 {
186 	struct pagerops **pgops;
187 
188 	TAILQ_INIT(&bswlist);
189 	/*
190 	 * Initialize known pagers
191 	 */
192 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
193 		if (pgops && ((*pgops)->pgo_init != NULL))
194 			(*(*pgops)->pgo_init) ();
195 }
196 
197 void
198 vm_pager_bufferinit()
199 {
200 	struct buf *bp;
201 	int i;
202 
203 	mtx_init(&pbuf_mtx, "pbuf mutex", NULL, MTX_DEF);
204 	bp = swbuf;
205 	/*
206 	 * Now set up swap and physical I/O buffer headers.
207 	 */
208 	for (i = 0; i < nswbuf; i++, bp++) {
209 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
210 		BUF_LOCKINIT(bp);
211 		LIST_INIT(&bp->b_dep);
212 		bp->b_rcred = bp->b_wcred = NOCRED;
213 		bp->b_xflags = 0;
214 	}
215 
216 	cluster_pbuf_freecnt = nswbuf / 2;
217 	vnode_pbuf_freecnt = nswbuf / 2 + 1;
218 
219 	swapbkva = kmem_alloc_nofault(pager_map, nswbuf * MAXPHYS);
220 	if (!swapbkva)
221 		panic("Not enough pager_map VM space for physical buffers");
222 }
223 
224 /*
225  * Allocate an instance of a pager of the given type.
226  * Size, protection and offset parameters are passed in for pagers that
227  * need to perform page-level validation (e.g. the device pager).
228  */
229 vm_object_t
230 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
231     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
232 {
233 	vm_object_t ret;
234 	struct pagerops *ops;
235 
236 	ops = pagertab[type];
237 	if (ops)
238 		ret = (*ops->pgo_alloc) (handle, size, prot, off, cred);
239 	else
240 		ret = NULL;
241 	return (ret);
242 }
243 
244 /*
245  *	The object must be locked.
246  */
247 void
248 vm_pager_deallocate(object)
249 	vm_object_t object;
250 {
251 
252 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
253 	(*pagertab[object->type]->pgo_dealloc) (object);
254 }
255 
256 /*
257  * vm_pager_get_pages() - inline, see vm/vm_pager.h
258  * vm_pager_put_pages() - inline, see vm/vm_pager.h
259  * vm_pager_has_page() - inline, see vm/vm_pager.h
260  */
261 
262 /*
263  * Search the specified pager object list for an object with the
264  * specified handle.  If an object with the specified handle is found,
265  * increase its reference count and return it.  Otherwise, return NULL.
266  *
267  * The pager object list must be locked.
268  */
269 vm_object_t
270 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
271 {
272 	vm_object_t object;
273 
274 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
275 		VM_OBJECT_LOCK(object);
276 		if (object->handle == handle &&
277 		    (object->flags & OBJ_DEAD) == 0) {
278 			vm_object_reference_locked(object);
279 			VM_OBJECT_UNLOCK(object);
280 			break;
281 		}
282 		VM_OBJECT_UNLOCK(object);
283 	}
284 	return (object);
285 }
286 
287 /*
288  * initialize a physical buffer
289  */
290 
291 /*
292  * XXX This probably belongs in vfs_bio.c
293  */
294 static void
295 initpbuf(struct buf *bp)
296 {
297 	KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj"));
298 	KASSERT(bp->b_vp == NULL, ("initpbuf with vp"));
299 	bp->b_rcred = NOCRED;
300 	bp->b_wcred = NOCRED;
301 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
302 	bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
303 	bp->b_data = bp->b_saveaddr;
304 	bp->b_kvabase = bp->b_saveaddr;
305 	bp->b_kvasize = MAXPHYS;
306 	bp->b_xflags = 0;
307 	bp->b_flags = 0;
308 	bp->b_ioflags = 0;
309 	bp->b_iodone = NULL;
310 	bp->b_error = 0;
311 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
312 }
313 
314 /*
315  * allocate a physical buffer
316  *
317  *	There are a limited number (nswbuf) of physical buffers.  We need
318  *	to make sure that no single subsystem is able to hog all of them,
319  *	so each subsystem implements a counter which is typically initialized
320  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
321  *	increments it on release, and blocks if the counter hits zero.  A
322  *	subsystem may initialize the counter to -1 to disable the feature,
323  *	but it must still be sure to match up all uses of getpbuf() with
324  *	relpbuf() using the same variable.
325  *
326  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
327  *	relatively soon when the rest of the subsystems get smart about it. XXX
328  */
329 struct buf *
330 getpbuf(int *pfreecnt)
331 {
332 	struct buf *bp;
333 
334 	mtx_lock(&pbuf_mtx);
335 
336 	for (;;) {
337 		if (pfreecnt) {
338 			while (*pfreecnt == 0) {
339 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
340 			}
341 		}
342 
343 		/* get a bp from the swap buffer header pool */
344 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
345 			break;
346 
347 		bswneeded = 1;
348 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
349 		/* loop in case someone else grabbed one */
350 	}
351 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
352 	if (pfreecnt)
353 		--*pfreecnt;
354 	mtx_unlock(&pbuf_mtx);
355 
356 	initpbuf(bp);
357 	return bp;
358 }
359 
360 /*
361  * allocate a physical buffer, if one is available.
362  *
363  *	Note that there is no NULL hack here - all subsystems using this
364  *	call understand how to use pfreecnt.
365  */
366 struct buf *
367 trypbuf(int *pfreecnt)
368 {
369 	struct buf *bp;
370 
371 	mtx_lock(&pbuf_mtx);
372 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
373 		mtx_unlock(&pbuf_mtx);
374 		return NULL;
375 	}
376 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
377 
378 	--*pfreecnt;
379 
380 	mtx_unlock(&pbuf_mtx);
381 
382 	initpbuf(bp);
383 
384 	return bp;
385 }
386 
387 /*
388  * release a physical buffer
389  *
390  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
391  *	relatively soon when the rest of the subsystems get smart about it. XXX
392  */
393 void
394 relpbuf(struct buf *bp, int *pfreecnt)
395 {
396 
397 	if (bp->b_rcred != NOCRED) {
398 		crfree(bp->b_rcred);
399 		bp->b_rcred = NOCRED;
400 	}
401 	if (bp->b_wcred != NOCRED) {
402 		crfree(bp->b_wcred);
403 		bp->b_wcred = NOCRED;
404 	}
405 
406 	KASSERT(bp->b_vp == NULL, ("relpbuf with vp"));
407 	KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj"));
408 
409 	BUF_UNLOCK(bp);
410 
411 	mtx_lock(&pbuf_mtx);
412 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
413 
414 	if (bswneeded) {
415 		bswneeded = 0;
416 		wakeup(&bswneeded);
417 	}
418 	if (pfreecnt) {
419 		if (++*pfreecnt == 1)
420 			wakeup(pfreecnt);
421 	}
422 	mtx_unlock(&pbuf_mtx);
423 }
424 
425 /*
426  * Associate a p-buffer with a vnode.
427  *
428  * Also sets B_PAGING flag to indicate that vnode is not fully associated
429  * with the buffer.  i.e. the bp has not been linked into the vnode or
430  * ref-counted.
431  */
432 void
433 pbgetvp(struct vnode *vp, struct buf *bp)
434 {
435 
436 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
437 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
438 
439 	bp->b_vp = vp;
440 	bp->b_flags |= B_PAGING;
441 	bp->b_bufobj = &vp->v_bufobj;
442 }
443 
444 /*
445  * Associate a p-buffer with a vnode.
446  *
447  * Also sets B_PAGING flag to indicate that vnode is not fully associated
448  * with the buffer.  i.e. the bp has not been linked into the vnode or
449  * ref-counted.
450  */
451 void
452 pbgetbo(struct bufobj *bo, struct buf *bp)
453 {
454 
455 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
456 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
457 
458 	bp->b_flags |= B_PAGING;
459 	bp->b_bufobj = bo;
460 }
461 
462 /*
463  * Disassociate a p-buffer from a vnode.
464  */
465 void
466 pbrelvp(struct buf *bp)
467 {
468 
469 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
470 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
471 
472 	/* XXX REMOVE ME */
473 	BO_LOCK(bp->b_bufobj);
474 	if (TAILQ_NEXT(bp, b_bobufs) != NULL) {
475 		panic(
476 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
477 		    bp,
478 		    (int)bp->b_flags
479 		);
480 	}
481 	BO_UNLOCK(bp->b_bufobj);
482 	bp->b_vp = NULL;
483 	bp->b_bufobj = NULL;
484 	bp->b_flags &= ~B_PAGING;
485 }
486 
487 /*
488  * Disassociate a p-buffer from a bufobj.
489  */
490 void
491 pbrelbo(struct buf *bp)
492 {
493 
494 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
495 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
496 
497 	/* XXX REMOVE ME */
498 	BO_LOCK(bp->b_bufobj);
499 	if (TAILQ_NEXT(bp, b_bobufs) != NULL) {
500 		panic(
501 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
502 		    bp,
503 		    (int)bp->b_flags
504 		);
505 	}
506 	BO_UNLOCK(bp->b_bufobj);
507 	bp->b_bufobj = NULL;
508 	bp->b_flags &= ~B_PAGING;
509 }
510