2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
36 * from: @(#)vm_pager.c 8.6 (Berkeley) 1/12/94
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
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.
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.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
64 * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $
65 * $DragonFly: src/sys/vm/vm_pager.c,v 1.10 2004/05/13 17:40:19 dillon Exp $
69 * Paging space routine stubs. Emulates a matchmaker-like interface
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/kernel.h>
76 #include <sys/vnode.h>
78 #include <sys/ucred.h>
79 #include <sys/malloc.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>
91 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
93 extern struct pagerops defaultpagerops;
94 extern struct pagerops swappagerops;
95 extern struct pagerops vnodepagerops;
96 extern struct pagerops devicepagerops;
97 extern struct pagerops physpagerops;
99 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */
101 static int dead_pager_getpages (vm_object_t, vm_page_t *, int, int);
102 static vm_object_t dead_pager_alloc (void *, vm_ooffset_t, vm_prot_t,
104 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
105 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t, int *, int *);
106 static void dead_pager_dealloc (vm_object_t);
109 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int req)
111 return VM_PAGER_FAIL;
115 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
122 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
127 for (i = 0; i < count; i++) {
128 rtvals[i] = VM_PAGER_AGAIN;
133 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next)
143 dead_pager_dealloc(vm_object_t object)
148 static struct pagerops deadpagerops = {
158 struct pagerops *pagertab[] = {
159 &defaultpagerops, /* OBJT_DEFAULT */
160 &swappagerops, /* OBJT_SWAP */
161 &vnodepagerops, /* OBJT_VNODE */
162 &devicepagerops, /* OBJT_DEVICE */
163 &physpagerops, /* OBJT_PHYS */
164 &deadpagerops /* OBJT_DEAD */
167 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
170 * Kernel address space for mapping pages.
171 * Used by pagers where KVAs are needed for IO.
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.
177 #define PAGER_MAP_SIZE (8 * 1024 * 1024)
179 int pager_map_size = PAGER_MAP_SIZE;
181 static int bswneeded;
182 static vm_offset_t swapbkva; /* swap buffers kva */
187 struct pagerops **pgops;
190 * Initialize known pagers
192 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
193 if (pgops && ((*pgops)->pgo_init != NULL))
194 (*(*pgops)->pgo_init) ();
198 vm_pager_bufferinit(void)
205 * Now set up swap and physical I/O buffer headers.
207 for (i = 0; i < nswbuf; i++, bp++) {
208 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
210 LIST_INIT(&bp->b_dep);
214 cluster_pbuf_freecnt = nswbuf / 2;
216 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
218 panic("Not enough pager_map VM space for physical buffers");
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).
227 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot,
230 struct pagerops *ops;
232 ops = pagertab[type];
234 return ((*ops->pgo_alloc) (handle, size, prot, off));
239 vm_pager_deallocate(vm_object_t object)
241 (*pagertab[object->type]->pgo_dealloc) (object);
247 * called with no specific spl
248 * Execute strategy routine directly to pager.
252 vm_pager_strategy(vm_object_t object, struct buf *bp)
254 if (pagertab[object->type]->pgo_strategy) {
255 (*pagertab[object->type]->pgo_strategy)(object, bp);
257 bp->b_flags |= B_ERROR;
264 * vm_pager_get_pages() - inline, see vm/vm_pager.h
265 * vm_pager_put_pages() - inline, see vm/vm_pager.h
266 * vm_pager_has_page() - inline, see vm/vm_pager.h
267 * vm_pager_page_inserted() - inline, see vm/vm_pager.h
268 * vm_pager_page_removed() - inline, see vm/vm_pager.h
275 * Called by pageout daemon before going back to sleep.
276 * Gives pagers a chance to clean up any completed async pageing
282 struct pagerops **pgops;
284 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
285 if (pgops && ((*pgops)->pgo_sync != NULL))
286 (*(*pgops)->pgo_sync) ();
292 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
296 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list))
297 if (object->handle == handle)
303 * initialize a physical buffer
307 initpbuf(struct buf *bp)
309 bp->b_qindex = QUEUE_NONE;
310 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
311 bp->b_kvabase = bp->b_data;
312 bp->b_kvasize = MAXPHYS;
316 BUF_LOCK(bp, LK_EXCLUSIVE);
320 * allocate a physical buffer
322 * There are a limited number (nswbuf) of physical buffers. We need
323 * to make sure that no single subsystem is able to hog all of them,
324 * so each subsystem implements a counter which is typically initialized
325 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and
326 * increments it on release, and blocks if the counter hits zero. A
327 * subsystem may initialize the counter to -1 to disable the feature,
328 * but it must still be sure to match up all uses of getpbuf() with
329 * relpbuf() using the same variable.
331 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
332 * relatively soon when the rest of the subsystems get smart about it. XXX
335 getpbuf(int *pfreecnt)
344 while (*pfreecnt == 0) {
345 tsleep(pfreecnt, 0, "wswbuf0", 0);
349 /* get a bp from the swap buffer header pool */
350 if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
354 tsleep(&bswneeded, 0, "wswbuf1", 0);
355 /* loop in case someone else grabbed one */
357 TAILQ_REMOVE(&bswlist, bp, b_freelist);
367 * allocate a physical buffer, if one is available.
369 * Note that there is no NULL hack here - all subsystems using this
370 * call understand how to use pfreecnt.
373 trypbuf(int *pfreecnt)
379 if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
383 TAILQ_REMOVE(&bswlist, bp, b_freelist);
395 * release a physical buffer
397 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
398 * relatively soon when the rest of the subsystems get smart about it. XXX
401 relpbuf(struct buf *bp, int *pfreecnt)
412 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
419 if (++*pfreecnt == 1)
425 /********************************************************
426 * CHAINING FUNCTIONS *
427 ********************************************************
429 * These functions support recursion of I/O operations
430 * on bp's, typically by chaining one or more 'child' bp's
431 * to the parent. Synchronous, asynchronous, and semi-synchronous
432 * chaining is possible.
436 * vm_pager_chain_iodone:
438 * io completion routine for child bp. Currently we fudge a bit
439 * on dealing with b_resid. Since users of these routines may issue
440 * multiple children simultaniously, sequencing of the error can be lost.
444 vm_pager_chain_iodone(struct buf *nbp)
448 if ((bp = nbp->b_chain.parent) != NULL) {
449 if (nbp->b_flags & B_ERROR) {
450 bp->b_flags |= B_ERROR;
451 bp->b_error = nbp->b_error;
452 } else if (nbp->b_resid != 0) {
453 bp->b_flags |= B_ERROR;
454 bp->b_error = EINVAL;
456 bp->b_resid -= nbp->b_bcount;
458 nbp->b_chain.parent = NULL;
460 if (bp->b_flags & B_WANT) {
461 bp->b_flags &= ~B_WANT;
464 if (!bp->b_chain.count && (bp->b_xflags & BX_AUTOCHAINDONE)) {
465 bp->b_xflags &= ~BX_AUTOCHAINDONE;
466 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) {
467 bp->b_flags |= B_ERROR;
468 bp->b_error = EINVAL;
473 nbp->b_flags |= B_DONE;
474 nbp->b_flags &= ~B_ASYNC;
481 * Obtain a physical buffer and chain it to its parent buffer. When
482 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are
483 * automatically propogated to the parent
485 * Since these are brand new buffers, we do not have to clear B_INVAL
486 * and B_ERROR because they are already clear.
490 getchainbuf(struct buf *bp, struct vnode *vp, int flags)
492 struct buf *nbp = getpbuf(NULL);
494 nbp->b_chain.parent = bp;
497 if (bp->b_chain.count > 4)
498 waitchainbuf(bp, 4, 0);
500 nbp->b_flags = B_CALL | (bp->b_flags & B_ORDERED) | flags;
501 nbp->b_iodone = vm_pager_chain_iodone;
509 flushchainbuf(struct buf *nbp)
512 nbp->b_bufsize = nbp->b_bcount;
513 if ((nbp->b_flags & B_READ) == 0)
514 nbp->b_dirtyend = nbp->b_bcount;
516 VOP_STRATEGY(nbp->b_vp, nbp);
523 waitchainbuf(struct buf *bp, int count, int done)
528 while (bp->b_chain.count > count) {
529 bp->b_flags |= B_WANT;
530 tsleep(bp, 0, "bpchain", 0);
533 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) {
534 bp->b_flags |= B_ERROR;
535 bp->b_error = EINVAL;
543 autochaindone(struct buf *bp)
548 if (bp->b_chain.count == 0)
551 bp->b_xflags |= BX_AUTOCHAINDONE;