xref: /freebsd/sys/vm/vm_pager.c (revision 6404f697a01c9b274232db3c1bba19c75163df6e)
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 #include <sys/rwlock.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_pager.h>
85 #include <vm/vm_extern.h>
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 	&mgtdevicepagerops,	/* OBJT_MGTDEVICE */
166 };
167 
168 static const int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
169 
170 /*
171  * Kernel address space for mapping pages.
172  * Used by pagers where KVAs are needed for IO.
173  *
174  * XXX needs to be large enough to support the number of pending async
175  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
176  * (MAXPHYS == 64k) if you want to get the most efficiency.
177  */
178 struct mtx_padalign pbuf_mtx;
179 static TAILQ_HEAD(swqueue, buf) bswlist;
180 static int bswneeded;
181 vm_offset_t swapbkva;		/* swap buffers kva */
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)->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 	vnode_async_pbuf_freecnt = nswbuf / 2;
219 }
220 
221 /*
222  * Allocate an instance of a pager of the given type.
223  * Size, protection and offset parameters are passed in for pagers that
224  * need to perform page-level validation (e.g. the device pager).
225  */
226 vm_object_t
227 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size,
228     vm_prot_t prot, vm_ooffset_t off, struct ucred *cred)
229 {
230 	vm_object_t ret;
231 	struct pagerops *ops;
232 
233 	ops = pagertab[type];
234 	if (ops)
235 		ret = (*ops->pgo_alloc) (handle, size, prot, off, cred);
236 	else
237 		ret = NULL;
238 	return (ret);
239 }
240 
241 /*
242  *	The object must be locked.
243  */
244 void
245 vm_pager_deallocate(object)
246 	vm_object_t object;
247 {
248 
249 	VM_OBJECT_ASSERT_WLOCKED(object);
250 	(*pagertab[object->type]->pgo_dealloc) (object);
251 }
252 
253 /*
254  * vm_pager_get_pages() - inline, see vm/vm_pager.h
255  * vm_pager_put_pages() - inline, see vm/vm_pager.h
256  * vm_pager_has_page() - inline, see vm/vm_pager.h
257  */
258 
259 /*
260  * Search the specified pager object list for an object with the
261  * specified handle.  If an object with the specified handle is found,
262  * increase its reference count and return it.  Otherwise, return NULL.
263  *
264  * The pager object list must be locked.
265  */
266 vm_object_t
267 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
268 {
269 	vm_object_t object;
270 
271 	TAILQ_FOREACH(object, pg_list, pager_object_list) {
272 		if (object->handle == handle) {
273 			VM_OBJECT_WLOCK(object);
274 			if ((object->flags & OBJ_DEAD) == 0) {
275 				vm_object_reference_locked(object);
276 				VM_OBJECT_WUNLOCK(object);
277 				break;
278 			}
279 			VM_OBJECT_WUNLOCK(object);
280 		}
281 	}
282 	return (object);
283 }
284 
285 /*
286  * Free the non-requested pages from the given array.
287  */
288 void
289 vm_pager_free_nonreq(vm_object_t object, vm_page_t ma[], int reqpage,
290     int npages)
291 {
292 	int i;
293 	boolean_t object_locked;
294 
295 	VM_OBJECT_ASSERT_UNLOCKED(object);
296 	object_locked = FALSE;
297 	for (i = 0; i < npages; ++i) {
298 		if (i != reqpage) {
299 			if (!object_locked) {
300 				VM_OBJECT_WLOCK(object);
301 				object_locked = TRUE;
302 			}
303 			vm_page_lock(ma[i]);
304 			vm_page_free(ma[i]);
305 			vm_page_unlock(ma[i]);
306 		}
307 	}
308 	if (object_locked)
309 		VM_OBJECT_WUNLOCK(object);
310 }
311 
312 /*
313  * initialize a physical buffer
314  */
315 
316 /*
317  * XXX This probably belongs in vfs_bio.c
318  */
319 static void
320 initpbuf(struct buf *bp)
321 {
322 	KASSERT(bp->b_bufobj == NULL, ("initpbuf with bufobj"));
323 	KASSERT(bp->b_vp == NULL, ("initpbuf with vp"));
324 	bp->b_rcred = NOCRED;
325 	bp->b_wcred = NOCRED;
326 	bp->b_qindex = 0;	/* On no queue (QUEUE_NONE) */
327 	bp->b_saveaddr = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
328 	bp->b_data = bp->b_saveaddr;
329 	bp->b_kvabase = bp->b_saveaddr;
330 	bp->b_kvasize = MAXPHYS;
331 	bp->b_xflags = 0;
332 	bp->b_flags = 0;
333 	bp->b_ioflags = 0;
334 	bp->b_iodone = NULL;
335 	bp->b_error = 0;
336 	BUF_LOCK(bp, LK_EXCLUSIVE, NULL);
337 }
338 
339 /*
340  * allocate a physical buffer
341  *
342  *	There are a limited number (nswbuf) of physical buffers.  We need
343  *	to make sure that no single subsystem is able to hog all of them,
344  *	so each subsystem implements a counter which is typically initialized
345  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
346  *	increments it on release, and blocks if the counter hits zero.  A
347  *	subsystem may initialize the counter to -1 to disable the feature,
348  *	but it must still be sure to match up all uses of getpbuf() with
349  *	relpbuf() using the same variable.
350  *
351  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
352  *	relatively soon when the rest of the subsystems get smart about it. XXX
353  */
354 struct buf *
355 getpbuf(int *pfreecnt)
356 {
357 	struct buf *bp;
358 
359 	mtx_lock(&pbuf_mtx);
360 
361 	for (;;) {
362 		if (pfreecnt) {
363 			while (*pfreecnt == 0) {
364 				msleep(pfreecnt, &pbuf_mtx, PVM, "wswbuf0", 0);
365 			}
366 		}
367 
368 		/* get a bp from the swap buffer header pool */
369 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
370 			break;
371 
372 		bswneeded = 1;
373 		msleep(&bswneeded, &pbuf_mtx, PVM, "wswbuf1", 0);
374 		/* loop in case someone else grabbed one */
375 	}
376 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
377 	if (pfreecnt)
378 		--*pfreecnt;
379 	mtx_unlock(&pbuf_mtx);
380 
381 	initpbuf(bp);
382 	return bp;
383 }
384 
385 /*
386  * allocate a physical buffer, if one is available.
387  *
388  *	Note that there is no NULL hack here - all subsystems using this
389  *	call understand how to use pfreecnt.
390  */
391 struct buf *
392 trypbuf(int *pfreecnt)
393 {
394 	struct buf *bp;
395 
396 	mtx_lock(&pbuf_mtx);
397 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
398 		mtx_unlock(&pbuf_mtx);
399 		return NULL;
400 	}
401 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
402 
403 	--*pfreecnt;
404 
405 	mtx_unlock(&pbuf_mtx);
406 
407 	initpbuf(bp);
408 
409 	return bp;
410 }
411 
412 /*
413  * release a physical buffer
414  *
415  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
416  *	relatively soon when the rest of the subsystems get smart about it. XXX
417  */
418 void
419 relpbuf(struct buf *bp, int *pfreecnt)
420 {
421 
422 	if (bp->b_rcred != NOCRED) {
423 		crfree(bp->b_rcred);
424 		bp->b_rcred = NOCRED;
425 	}
426 	if (bp->b_wcred != NOCRED) {
427 		crfree(bp->b_wcred);
428 		bp->b_wcred = NOCRED;
429 	}
430 
431 	KASSERT(bp->b_vp == NULL, ("relpbuf with vp"));
432 	KASSERT(bp->b_bufobj == NULL, ("relpbuf with bufobj"));
433 
434 	BUF_UNLOCK(bp);
435 
436 	mtx_lock(&pbuf_mtx);
437 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
438 
439 	if (bswneeded) {
440 		bswneeded = 0;
441 		wakeup(&bswneeded);
442 	}
443 	if (pfreecnt) {
444 		if (++*pfreecnt == 1)
445 			wakeup(pfreecnt);
446 	}
447 	mtx_unlock(&pbuf_mtx);
448 }
449 
450 /*
451  * Associate a p-buffer with a vnode.
452  *
453  * Also sets B_PAGING flag to indicate that vnode is not fully associated
454  * with the buffer.  i.e. the bp has not been linked into the vnode or
455  * ref-counted.
456  */
457 void
458 pbgetvp(struct vnode *vp, struct buf *bp)
459 {
460 
461 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
462 	KASSERT(bp->b_bufobj == NULL, ("pbgetvp: not free (bufobj)"));
463 
464 	bp->b_vp = vp;
465 	bp->b_flags |= B_PAGING;
466 	bp->b_bufobj = &vp->v_bufobj;
467 }
468 
469 /*
470  * Associate a p-buffer with a vnode.
471  *
472  * Also sets B_PAGING flag to indicate that vnode is not fully associated
473  * with the buffer.  i.e. the bp has not been linked into the vnode or
474  * ref-counted.
475  */
476 void
477 pbgetbo(struct bufobj *bo, struct buf *bp)
478 {
479 
480 	KASSERT(bp->b_vp == NULL, ("pbgetbo: not free (vnode)"));
481 	KASSERT(bp->b_bufobj == NULL, ("pbgetbo: not free (bufobj)"));
482 
483 	bp->b_flags |= B_PAGING;
484 	bp->b_bufobj = bo;
485 }
486 
487 /*
488  * Disassociate a p-buffer from a vnode.
489  */
490 void
491 pbrelvp(struct buf *bp)
492 {
493 
494 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
495 	KASSERT(bp->b_bufobj != NULL, ("pbrelvp: NULL bufobj"));
496 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
497 	    ("pbrelvp: pager buf on vnode list."));
498 
499 	bp->b_vp = NULL;
500 	bp->b_bufobj = NULL;
501 	bp->b_flags &= ~B_PAGING;
502 }
503 
504 /*
505  * Disassociate a p-buffer from a bufobj.
506  */
507 void
508 pbrelbo(struct buf *bp)
509 {
510 
511 	KASSERT(bp->b_vp == NULL, ("pbrelbo: vnode"));
512 	KASSERT(bp->b_bufobj != NULL, ("pbrelbo: NULL bufobj"));
513 	KASSERT((bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) == 0,
514 	    ("pbrelbo: pager buf on vnode list."));
515 
516 	bp->b_bufobj = NULL;
517 	bp->b_flags &= ~B_PAGING;
518 }
519