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 1.12 2005/06/02 20:57:21 swildner 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>
81 #include <sys/thread2.h>
84 #include <vm/vm_param.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pager.h>
88 #include <vm/vm_extern.h>
92 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
94 extern struct pagerops defaultpagerops;
95 extern struct pagerops swappagerops;
96 extern struct pagerops vnodepagerops;
97 extern struct pagerops devicepagerops;
98 extern struct pagerops physpagerops;
100 int cluster_pbuf_freecnt = -1; /* unlimited to begin with */
102 static int dead_pager_getpages (vm_object_t, vm_page_t *, int, int);
103 static vm_object_t dead_pager_alloc (void *, vm_ooffset_t, vm_prot_t,
105 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
106 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t, int *, int *);
107 static void dead_pager_dealloc (vm_object_t);
110 dead_pager_getpages(vm_object_t obj, vm_page_t *ma, int count, int req)
112 return VM_PAGER_FAIL;
116 dead_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
123 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
128 for (i = 0; i < count; i++) {
129 rtvals[i] = VM_PAGER_AGAIN;
134 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *prev, int *next)
144 dead_pager_dealloc(vm_object_t object)
149 static struct pagerops deadpagerops = {
159 struct pagerops *pagertab[] = {
160 &defaultpagerops, /* OBJT_DEFAULT */
161 &swappagerops, /* OBJT_SWAP */
162 &vnodepagerops, /* OBJT_VNODE */
163 &devicepagerops, /* OBJT_DEVICE */
164 &physpagerops, /* OBJT_PHYS */
165 &deadpagerops /* OBJT_DEAD */
168 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
171 * Kernel address space for mapping pages.
172 * Used by pagers where KVAs are needed for IO.
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.
178 #define PAGER_MAP_SIZE (8 * 1024 * 1024)
180 int pager_map_size = PAGER_MAP_SIZE;
182 static int bswneeded;
183 static vm_offset_t swapbkva; /* swap buffers kva */
188 struct pagerops **pgops;
191 * Initialize known pagers
193 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
194 if (pgops && ((*pgops)->pgo_init != NULL))
195 (*(*pgops)->pgo_init) ();
199 vm_pager_bufferinit(void)
206 * Now set up swap and physical I/O buffer headers.
208 for (i = 0; i < nswbuf; i++, bp++) {
209 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
211 LIST_INIT(&bp->b_dep);
215 cluster_pbuf_freecnt = nswbuf / 2;
217 swapbkva = kmem_alloc_pageable(pager_map, nswbuf * MAXPHYS);
219 panic("Not enough pager_map VM space for physical buffers");
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).
228 vm_pager_allocate(objtype_t type, void *handle, vm_ooffset_t size, vm_prot_t prot,
231 struct pagerops *ops;
233 ops = pagertab[type];
235 return ((*ops->pgo_alloc) (handle, size, prot, off));
240 vm_pager_deallocate(vm_object_t object)
242 (*pagertab[object->type]->pgo_dealloc) (object);
248 * called with no specific spl
249 * Execute strategy routine directly to pager.
253 vm_pager_strategy(vm_object_t object, struct buf *bp)
255 if (pagertab[object->type]->pgo_strategy) {
256 (*pagertab[object->type]->pgo_strategy)(object, bp);
258 bp->b_flags |= B_ERROR;
265 * vm_pager_get_pages() - inline, see vm/vm_pager.h
266 * vm_pager_put_pages() - inline, see vm/vm_pager.h
267 * vm_pager_has_page() - inline, see vm/vm_pager.h
268 * vm_pager_page_inserted() - inline, see vm/vm_pager.h
269 * vm_pager_page_removed() - inline, see vm/vm_pager.h
276 * Called by pageout daemon before going back to sleep.
277 * Gives pagers a chance to clean up any completed async pageing
283 struct pagerops **pgops;
285 for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
286 if (pgops && ((*pgops)->pgo_sync != NULL))
287 (*(*pgops)->pgo_sync) ();
293 vm_pager_object_lookup(struct pagerlst *pg_list, void *handle)
297 for (object = TAILQ_FIRST(pg_list); object != NULL; object = TAILQ_NEXT(object,pager_object_list))
298 if (object->handle == handle)
304 * initialize a physical buffer
308 initpbuf(struct buf *bp)
310 bp->b_qindex = QUEUE_NONE;
311 bp->b_data = (caddr_t) (MAXPHYS * (bp - swbuf)) + swapbkva;
312 bp->b_kvabase = bp->b_data;
313 bp->b_kvasize = MAXPHYS;
317 xio_init(&bp->b_xio);
318 BUF_LOCK(bp, LK_EXCLUSIVE);
322 * allocate a physical buffer
324 * There are a limited number (nswbuf) of physical buffers. We need
325 * to make sure that no single subsystem is able to hog all of them,
326 * so each subsystem implements a counter which is typically initialized
327 * to 1/2 nswbuf. getpbuf() decrements this counter in allocation and
328 * increments it on release, and blocks if the counter hits zero. A
329 * subsystem may initialize the counter to -1 to disable the feature,
330 * but it must still be sure to match up all uses of getpbuf() with
331 * relpbuf() using the same variable.
333 * NOTE: pfreecnt can be NULL, but this 'feature' will be removed
334 * relatively soon when the rest of the subsystems get smart about it. XXX
337 getpbuf(int *pfreecnt)
345 while (*pfreecnt == 0) {
346 tsleep(pfreecnt, 0, "wswbuf0", 0);
350 /* get a bp from the swap buffer header pool */
351 if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
355 tsleep(&bswneeded, 0, "wswbuf1", 0);
356 /* loop in case someone else grabbed one */
358 TAILQ_REMOVE(&bswlist, bp, b_freelist);
368 * allocate a physical buffer, if one is available.
370 * Note that there is no NULL hack here - all subsystems using this
371 * call understand how to use pfreecnt.
374 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)
410 TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
417 if (++*pfreecnt == 1)
423 /********************************************************
424 * CHAINING FUNCTIONS *
425 ********************************************************
427 * These functions support recursion of I/O operations
428 * on bp's, typically by chaining one or more 'child' bp's
429 * to the parent. Synchronous, asynchronous, and semi-synchronous
430 * chaining is possible.
434 * vm_pager_chain_iodone:
436 * io completion routine for child bp. Currently we fudge a bit
437 * on dealing with b_resid. Since users of these routines may issue
438 * multiple children simultaniously, sequencing of the error can be lost.
442 vm_pager_chain_iodone(struct buf *nbp)
446 if ((bp = nbp->b_chain.parent) != NULL) {
447 if (nbp->b_flags & B_ERROR) {
448 bp->b_flags |= B_ERROR;
449 bp->b_error = nbp->b_error;
450 } else if (nbp->b_resid != 0) {
451 bp->b_flags |= B_ERROR;
452 bp->b_error = EINVAL;
454 bp->b_resid -= nbp->b_bcount;
456 nbp->b_chain.parent = NULL;
458 if (bp->b_flags & B_WANT) {
459 bp->b_flags &= ~B_WANT;
462 if (!bp->b_chain.count && (bp->b_xflags & BX_AUTOCHAINDONE)) {
463 bp->b_xflags &= ~BX_AUTOCHAINDONE;
464 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) {
465 bp->b_flags |= B_ERROR;
466 bp->b_error = EINVAL;
471 nbp->b_flags |= B_DONE;
472 nbp->b_flags &= ~B_ASYNC;
479 * Obtain a physical buffer and chain it to its parent buffer. When
480 * I/O completes, the parent buffer will be B_SIGNAL'd. Errors are
481 * automatically propogated to the parent
483 * Since these are brand new buffers, we do not have to clear B_INVAL
484 * and B_ERROR because they are already clear.
488 getchainbuf(struct buf *bp, struct vnode *vp, int flags)
490 struct buf *nbp = getpbuf(NULL);
492 nbp->b_chain.parent = bp;
495 if (bp->b_chain.count > 4)
496 waitchainbuf(bp, 4, 0);
498 nbp->b_flags = B_CALL | (bp->b_flags & B_ORDERED) | flags;
499 nbp->b_iodone = vm_pager_chain_iodone;
507 flushchainbuf(struct buf *nbp)
510 nbp->b_bufsize = nbp->b_bcount;
511 if ((nbp->b_flags & B_READ) == 0)
512 nbp->b_dirtyend = nbp->b_bcount;
514 VOP_STRATEGY(nbp->b_vp, nbp);
521 waitchainbuf(struct buf *bp, int count, int done)
524 while (bp->b_chain.count > count) {
525 bp->b_flags |= B_WANT;
526 tsleep(bp, 0, "bpchain", 0);
529 if (bp->b_resid != 0 && !(bp->b_flags & B_ERROR)) {
530 bp->b_flags |= B_ERROR;
531 bp->b_error = EINVAL;
539 autochaindone(struct buf *bp)
542 if (bp->b_chain.count == 0)
545 bp->b_xflags |= BX_AUTOCHAINDONE;