2 * Copyright (c) 1990 University of Utah.
3 * Copyright (c) 1991 The Regents of the University of California.
5 * Copyright (c) 1993, 1994 John S. Dyson
6 * Copyright (c) 1995, David Greenman
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91
41 * $FreeBSD: src/sys/vm/vnode_pager.c,v 1.116.2.7 2002/12/31 09:34:51 dillon Exp $
42 * $DragonFly: src/sys/vm/vnode_pager.c,v 1.35 2007/07/30 21:41:30 dillon Exp $
46 * Page to/from files (vnodes).
51 * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will
52 * greatly re-simplify the vnode_pager.
55 #include <sys/param.h>
56 #include <sys/systm.h>
57 #include <sys/kernel.h>
59 #include <sys/vnode.h>
60 #include <sys/mount.h>
62 #include <sys/vmmeter.h>
64 #include <sys/sfbuf.h>
65 #include <sys/thread2.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_pager.h>
71 #include <vm/vm_map.h>
72 #include <vm/vnode_pager.h>
73 #include <vm/vm_extern.h>
75 static off_t
vnode_pager_addr (struct vnode
*vp
, off_t loffset
, int *run
);
76 static void vnode_pager_iodone (struct bio
*bio
);
77 static int vnode_pager_input_smlfs (vm_object_t object
, vm_page_t m
);
78 static int vnode_pager_input_old (vm_object_t object
, vm_page_t m
);
79 static void vnode_pager_dealloc (vm_object_t
);
80 static int vnode_pager_getpages (vm_object_t
, vm_page_t
*, int, int);
81 static void vnode_pager_putpages (vm_object_t
, vm_page_t
*, int, boolean_t
, int *);
82 static boolean_t
vnode_pager_haspage (vm_object_t
, vm_pindex_t
, int *, int *);
84 struct pagerops vnodepagerops
= {
94 static struct krate vbadrate
= { 1 };
95 static struct krate vresrate
= { 1 };
97 int vnode_pbuf_freecnt
= -1; /* start out unlimited */
100 * Allocate (or lookup) pager for a vnode.
101 * Handle is a vnode pointer.
104 vnode_pager_alloc(void *handle
, off_t size
, vm_prot_t prot
, off_t offset
)
110 * Pageout to vnode, no can do yet.
116 * XXX hack - This initialization should be put somewhere else.
118 if (vnode_pbuf_freecnt
< 0) {
119 vnode_pbuf_freecnt
= nswbuf
/ 2 + 1;
122 vp
= (struct vnode
*) handle
;
125 * Prevent race condition when allocating the object. This
126 * can happen with NFS vnodes since the nfsnode isn't locked.
128 while (vp
->v_flag
& VOLOCK
) {
129 vp
->v_flag
|= VOWANT
;
130 tsleep(vp
, 0, "vnpobj", 0);
132 vp
->v_flag
|= VOLOCK
;
135 * If the object is being terminated, wait for it to
138 while (((object
= vp
->v_object
) != NULL
) &&
139 (object
->flags
& OBJ_DEAD
)) {
140 vm_object_dead_sleep(object
, "vadead");
143 if (vp
->v_sysref
.refcnt
<= 0)
144 panic("vnode_pager_alloc: no vnode reference");
146 if (object
== NULL
) {
148 * And an object of the appropriate size
150 object
= vm_object_allocate(OBJT_VNODE
, OFF_TO_IDX(round_page(size
)));
152 object
->handle
= handle
;
153 vp
->v_object
= object
;
154 vp
->v_filesize
= size
;
157 if (vp
->v_filesize
!= size
)
158 kprintf("vnode_pager_alloc: Warning, filesize mismatch %lld/%lld\n", vp
->v_filesize
, size
);
162 vp
->v_flag
&= ~VOLOCK
;
163 if (vp
->v_flag
& VOWANT
) {
164 vp
->v_flag
&= ~VOWANT
;
171 vnode_pager_dealloc(vm_object_t object
)
173 struct vnode
*vp
= object
->handle
;
176 panic("vnode_pager_dealloc: pager already dealloced");
178 vm_object_pip_wait(object
, "vnpdea");
180 object
->handle
= NULL
;
181 object
->type
= OBJT_DEAD
;
183 vp
->v_filesize
= NOOFFSET
;
184 vp
->v_flag
&= ~(VTEXT
| VOBJBUF
);
188 * Return whether the vnode pager has the requested page. Return the
189 * number of disk-contiguous pages before and after the requested page,
190 * not including the requested page.
193 vnode_pager_haspage(vm_object_t object
, vm_pindex_t pindex
, int *before
,
196 struct vnode
*vp
= object
->handle
;
204 * If no vp or vp is doomed or marked transparent to VM, we do not
207 if ((vp
== NULL
) || (vp
->v_flag
& VRECLAIMED
))
211 * If filesystem no longer mounted or offset beyond end of file we do
214 loffset
= IDX_TO_OFF(pindex
);
216 if (vp
->v_mount
== NULL
|| loffset
>= vp
->v_filesize
)
219 bsize
= vp
->v_mount
->mnt_stat
.f_iosize
;
220 voff
= loffset
% bsize
;
222 error
= VOP_BMAP(vp
, loffset
- voff
, NULL
, &doffset
, after
, before
);
225 if (doffset
== NOOFFSET
)
229 *before
= (*before
+ voff
) >> PAGE_SHIFT
;
233 if (loffset
+ *after
> vp
->v_filesize
)
234 *after
= vp
->v_filesize
- loffset
;
235 *after
>>= PAGE_SHIFT
;
243 * Lets the VM system know about a change in size for a file.
244 * We adjust our own internal size and flush any cached pages in
245 * the associated object that are affected by the size change.
247 * NOTE: This routine may be invoked as a result of a pager put
248 * operation (possibly at object termination time), so we must be careful.
250 * NOTE: vp->v_filesize is initialized to NOOFFSET (-1), be sure that
251 * we do not blow up on the case. nsize will always be >= 0, however.
254 vnode_pager_setsize(struct vnode
*vp
, vm_ooffset_t nsize
)
256 vm_pindex_t nobjsize
;
257 vm_pindex_t oobjsize
;
258 vm_object_t object
= vp
->v_object
;
264 * Hasn't changed size
266 if (nsize
== vp
->v_filesize
)
270 * Has changed size. Adjust the VM object's size and v_filesize
271 * before we start scanning pages to prevent new pages from being
272 * allocated during the scan.
274 nobjsize
= OFF_TO_IDX(nsize
+ PAGE_MASK
);
275 oobjsize
= object
->size
;
276 object
->size
= nobjsize
;
279 * File has shrunk. Toss any cached pages beyond the new EOF.
281 if (nsize
< vp
->v_filesize
) {
282 vp
->v_filesize
= nsize
;
283 if (nobjsize
< oobjsize
) {
284 vm_object_page_remove(object
, nobjsize
, oobjsize
,
288 * This gets rid of garbage at the end of a page that is now
289 * only partially backed by the vnode. Since we are setting
290 * the entire page valid & clean after we are done we have
291 * to be sure that the portion of the page within the file
292 * bounds is already valid. If it isn't then making it
293 * valid would create a corrupt block.
295 if (nsize
& PAGE_MASK
) {
299 m
= vm_page_lookup(object
, OFF_TO_IDX(nsize
));
301 int base
= (int)nsize
& PAGE_MASK
;
302 int size
= PAGE_SIZE
- base
;
306 * Clear out partial-page garbage in case
307 * the page has been mapped.
309 sf
= sf_buf_alloc(m
, SFB_CPUPRIVATE
);
310 kva
= sf_buf_kva(sf
);
311 bzero((caddr_t
)kva
+ base
, size
);
315 * XXX work around SMP data integrity race
316 * by unmapping the page from user processes.
317 * The garbage we just cleared may be mapped
318 * to a user process running on another cpu
319 * and this code is not running through normal
320 * I/O channels which handle SMP issues for
321 * us, so unmap page to synchronize all cpus.
323 * XXX should vm_pager_unmap_page() have
326 vm_page_protect(m
, VM_PROT_NONE
);
329 * Clear out partial-page dirty bits. This
330 * has the side effect of setting the valid
331 * bits, but that is ok. There are a bunch
332 * of places in the VM system where we expected
333 * m->dirty == VM_PAGE_BITS_ALL. The file EOF
334 * case is one of them. If the page is still
335 * partially dirty, make it fully dirty.
337 * note that we do not clear out the valid
338 * bits. This would prevent bogus_page
339 * replacement from working properly.
341 vm_page_set_validclean(m
, base
, size
);
343 m
->dirty
= VM_PAGE_BITS_ALL
;
347 vp
->v_filesize
= nsize
;
352 vnode_pager_freepage(vm_page_t m
)
358 * calculate the disk byte address of specified logical byte offset. The
359 * logical offset will be block-aligned. Return the number of contiguous
360 * pages that may be read from the underlying block device in *run. If
361 * *run is non-NULL, it will be set to a value of at least 1.
364 vnode_pager_addr(struct vnode
*vp
, off_t loffset
, int *run
)
375 if (vp
->v_mount
== NULL
)
379 * Align loffset to a block boundary for the BMAP, then adjust the
380 * returned disk address appropriately.
382 bsize
= vp
->v_mount
->mnt_stat
.f_iosize
;
383 voff
= loffset
% bsize
;
386 * Map the block, adjust the disk offset so it represents the
387 * passed loffset rather then the block containing loffset.
389 error
= VOP_BMAP(vp
, loffset
- voff
, &rtvp
, &doffset
, run
, NULL
);
390 if (error
|| doffset
== NOOFFSET
) {
396 * When calculating *run, which is the number of pages
397 * worth of data which can be read linearly from disk,
398 * the minimum return value is 1 page.
401 *run
= (*run
- voff
) >> PAGE_SHIFT
;
411 * interrupt routine for I/O completion
414 vnode_pager_iodone(struct bio
*bio
)
416 struct buf
*bp
= bio
->bio_buf
;
418 bp
->b_cmd
= BUF_CMD_DONE
;
423 * small block file system vnode pager input
426 vnode_pager_input_smlfs(vm_object_t object
, vm_page_t m
)
429 struct vnode
*dp
, *vp
;
438 if (vp
->v_mount
== NULL
)
441 bsize
= vp
->v_mount
->mnt_stat
.f_iosize
;
444 VOP_BMAP(vp
, (off_t
)0, &dp
, NULL
, NULL
, NULL
);
446 sf
= sf_buf_alloc(m
, 0);
447 kva
= sf_buf_kva(sf
);
449 for (i
= 0; i
< PAGE_SIZE
/ bsize
; i
++) {
452 if (vm_page_bits(i
* bsize
, bsize
) & m
->valid
)
455 loffset
= IDX_TO_OFF(m
->pindex
) + i
* bsize
;
456 if (loffset
>= vp
->v_filesize
) {
459 doffset
= vnode_pager_addr(vp
, loffset
, NULL
);
461 if (doffset
!= NOOFFSET
) {
462 bp
= getpbuf(&vnode_pbuf_freecnt
);
464 /* build a minimal buffer header */
465 bp
->b_data
= (caddr_t
) kva
+ i
* bsize
;
466 bp
->b_bio1
.bio_done
= vnode_pager_iodone
;
467 bp
->b_bio1
.bio_offset
= doffset
;
468 bp
->b_bcount
= bsize
;
469 bp
->b_runningbufspace
= bsize
;
470 runningbufspace
+= bp
->b_runningbufspace
;
471 bp
->b_cmd
= BUF_CMD_READ
;
474 vn_strategy(dp
, &bp
->b_bio1
);
476 /* we definitely need to be at splvm here */
479 while (bp
->b_cmd
!= BUF_CMD_DONE
)
480 tsleep(bp
, 0, "vnsrd", 0);
482 if ((bp
->b_flags
& B_ERROR
) != 0)
486 * free the buffer header back to the swap buffer pool
488 relpbuf(bp
, &vnode_pbuf_freecnt
);
492 vm_page_set_validclean(m
, (i
* bsize
) & PAGE_MASK
, bsize
);
494 vm_page_set_validclean(m
, (i
* bsize
) & PAGE_MASK
, bsize
);
495 bzero((caddr_t
) kva
+ i
* bsize
, bsize
);
499 pmap_clear_modify(m
);
500 vm_page_flag_clear(m
, PG_ZERO
);
502 return VM_PAGER_ERROR
;
510 * old style vnode pager output routine
513 vnode_pager_input_old(vm_object_t object
, vm_page_t m
)
527 * Return failure if beyond current EOF
529 if (IDX_TO_OFF(m
->pindex
) >= vp
->v_filesize
) {
533 if (IDX_TO_OFF(m
->pindex
) + size
> vp
->v_filesize
)
534 size
= vp
->v_filesize
- IDX_TO_OFF(m
->pindex
);
537 * Allocate a kernel virtual address and initialize so that
538 * we can use VOP_READ/WRITE routines.
540 sf
= sf_buf_alloc(m
, 0);
541 kva
= sf_buf_kva(sf
);
543 aiov
.iov_base
= (caddr_t
) kva
;
545 auio
.uio_iov
= &aiov
;
547 auio
.uio_offset
= IDX_TO_OFF(m
->pindex
);
548 auio
.uio_segflg
= UIO_SYSSPACE
;
549 auio
.uio_rw
= UIO_READ
;
550 auio
.uio_resid
= size
;
551 auio
.uio_td
= curthread
;
553 error
= VOP_READ(((struct vnode
*)object
->handle
),
554 &auio
, 0, proc0
.p_ucred
);
556 int count
= size
- auio
.uio_resid
;
560 else if (count
!= PAGE_SIZE
)
561 bzero((caddr_t
) kva
+ count
, PAGE_SIZE
- count
);
565 pmap_clear_modify(m
);
567 vm_page_flag_clear(m
, PG_ZERO
);
569 m
->valid
= VM_PAGE_BITS_ALL
;
570 return error
? VM_PAGER_ERROR
: VM_PAGER_OK
;
574 * generic vnode pager input routine
578 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
579 * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
580 * vnode_pager_generic_getpages() to implement the previous behaviour.
582 * All other FS's should use the bypass to get to the local media
583 * backing vp's VOP_GETPAGES.
586 vnode_pager_getpages(vm_object_t object
, vm_page_t
*m
, int count
, int reqpage
)
590 int bytes
= count
* PAGE_SIZE
;
594 * XXX temporary diagnostic message to help track stale FS code,
595 * Returning EOPNOTSUPP from here may make things unhappy.
597 rtval
= VOP_GETPAGES(vp
, m
, bytes
, reqpage
, 0);
598 if (rtval
== EOPNOTSUPP
) {
599 kprintf("vnode_pager: *** WARNING *** stale FS getpages\n");
600 rtval
= vnode_pager_generic_getpages( vp
, m
, bytes
, reqpage
);
607 * This is now called from local media FS's to operate against their
608 * own vnodes if they fail to implement VOP_GETPAGES.
611 vnode_pager_generic_getpages(struct vnode
*vp
, vm_page_t
*m
, int bytecount
,
616 off_t foff
, tfoff
, nextoff
;
617 int i
, size
, bsize
, first
;
626 object
= vp
->v_object
;
627 count
= bytecount
/ PAGE_SIZE
;
629 if (vp
->v_mount
== NULL
)
632 bsize
= vp
->v_mount
->mnt_stat
.f_iosize
;
634 /* get the UNDERLYING device for the file with VOP_BMAP() */
637 * originally, we did not check for an error return value -- assuming
638 * an fs always has a bmap entry point -- that assumption is wrong!!!
640 foff
= IDX_TO_OFF(m
[reqpage
]->pindex
);
643 * if we can't bmap, use old VOP code
645 if (VOP_BMAP(vp
, (off_t
)0, &dp
, NULL
, NULL
, NULL
)) {
646 for (i
= 0; i
< count
; i
++) {
648 vnode_pager_freepage(m
[i
]);
651 mycpu
->gd_cnt
.v_vnodein
++;
652 mycpu
->gd_cnt
.v_vnodepgsin
++;
653 return vnode_pager_input_old(object
, m
[reqpage
]);
656 * if the blocksize is smaller than a page size, then use
657 * special small filesystem code. NFS sometimes has a small
658 * blocksize, but it can handle large reads itself.
660 } else if ((PAGE_SIZE
/ bsize
) > 1 &&
661 (vp
->v_mount
->mnt_stat
.f_type
!= nfs_mount_type
)) {
662 for (i
= 0; i
< count
; i
++) {
664 vnode_pager_freepage(m
[i
]);
667 mycpu
->gd_cnt
.v_vnodein
++;
668 mycpu
->gd_cnt
.v_vnodepgsin
++;
669 return vnode_pager_input_smlfs(object
, m
[reqpage
]);
673 * If we have a completely valid page available to us, we can
674 * clean up and return. Otherwise we have to re-read the
677 * Note that this does not work with NFS, so NFS has its own
678 * getpages routine. The problem is that NFS can have partially
679 * valid pages associated with the buffer cache due to the piecemeal
680 * write support. If we were to fall through and re-read the media
681 * as we do here, dirty data could be lost.
684 if (m
[reqpage
]->valid
== VM_PAGE_BITS_ALL
) {
685 for (i
= 0; i
< count
; i
++) {
687 vnode_pager_freepage(m
[i
]);
691 m
[reqpage
]->valid
= 0;
694 * here on direct device I/O
699 * calculate the run that includes the required page
701 for(first
= 0, i
= 0; i
< count
; i
= runend
) {
702 firstaddr
= vnode_pager_addr(vp
, IDX_TO_OFF(m
[i
]->pindex
),
704 if (firstaddr
== -1) {
705 if (i
== reqpage
&& foff
< vp
->v_filesize
) {
706 /* XXX no %qd in kernel. */
707 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %012llx, foff: 0x%012llx, v_filesize: 0x%012llx",
708 firstaddr
, foff
, vp
->v_filesize
);
710 vnode_pager_freepage(m
[i
]);
716 if (runend
<= reqpage
) {
718 for (j
= i
; j
< runend
; j
++) {
719 vnode_pager_freepage(m
[j
]);
722 if (runpg
< (count
- first
)) {
723 for (i
= first
+ runpg
; i
< count
; i
++)
724 vnode_pager_freepage(m
[i
]);
725 count
= first
+ runpg
;
733 * the first and last page have been calculated now, move input pages
734 * to be zero based...
737 for (i
= first
; i
< count
; i
++) {
745 * calculate the file virtual address for the transfer
747 foff
= IDX_TO_OFF(m
[0]->pindex
);
750 * calculate the size of the transfer
752 size
= count
* PAGE_SIZE
;
753 if ((foff
+ size
) > vp
->v_filesize
)
754 size
= vp
->v_filesize
- foff
;
757 * round up physical size for real devices.
759 if (dp
->v_type
== VBLK
|| dp
->v_type
== VCHR
) {
760 int secmask
= dp
->v_rdev
->si_bsize_phys
- 1;
761 KASSERT(secmask
< PAGE_SIZE
, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask
+ 1));
762 size
= (size
+ secmask
) & ~secmask
;
765 bp
= getpbuf(&vnode_pbuf_freecnt
);
766 kva
= (vm_offset_t
) bp
->b_data
;
769 * and map the pages to be read into the kva
771 pmap_qenter(kva
, m
, count
);
773 /* build a minimal buffer header */
774 bp
->b_bio1
.bio_done
= vnode_pager_iodone
;
775 bp
->b_bio1
.bio_offset
= firstaddr
;
777 bp
->b_runningbufspace
= size
;
778 runningbufspace
+= bp
->b_runningbufspace
;
779 bp
->b_cmd
= BUF_CMD_READ
;
781 mycpu
->gd_cnt
.v_vnodein
++;
782 mycpu
->gd_cnt
.v_vnodepgsin
+= count
;
785 vn_strategy(dp
, &bp
->b_bio1
);
788 /* we definitely need to be at splvm here */
790 while (bp
->b_cmd
!= BUF_CMD_DONE
)
791 tsleep(bp
, 0, "vnread", 0);
793 if ((bp
->b_flags
& B_ERROR
) != 0)
797 if (size
!= count
* PAGE_SIZE
)
798 bzero((caddr_t
) kva
+ size
, PAGE_SIZE
* count
- size
);
800 pmap_qremove(kva
, count
);
803 * free the buffer header back to the swap buffer pool
805 relpbuf(bp
, &vnode_pbuf_freecnt
);
807 for (i
= 0, tfoff
= foff
; i
< count
; i
++, tfoff
= nextoff
) {
810 nextoff
= tfoff
+ PAGE_SIZE
;
813 if (nextoff
<= vp
->v_filesize
) {
815 * Read filled up entire page.
817 mt
->valid
= VM_PAGE_BITS_ALL
;
818 vm_page_undirty(mt
); /* should be an assert? XXX */
819 pmap_clear_modify(mt
);
822 * Read did not fill up entire page. Since this
823 * is getpages, the page may be mapped, so we have
824 * to zero the invalid portions of the page even
825 * though we aren't setting them valid.
827 * Currently we do not set the entire page valid,
828 * we just try to clear the piece that we couldn't
831 vm_page_set_validclean(mt
, 0, vp
->v_filesize
- tfoff
);
832 /* handled by vm_fault now */
833 /* vm_page_zero_invalid(mt, FALSE); */
836 vm_page_flag_clear(mt
, PG_ZERO
);
840 * whether or not to leave the page activated is up in
841 * the air, but we should put the page on a page queue
842 * somewhere. (it already is in the object). Result:
843 * It appears that empirical results show that
844 * deactivating pages is best.
848 * just in case someone was asking for this page we
849 * now tell them that it is ok to use
852 if (mt
->flags
& PG_WANTED
)
853 vm_page_activate(mt
);
855 vm_page_deactivate(mt
);
858 vnode_pager_freepage(mt
);
863 kprintf("vnode_pager_getpages: I/O read error\n");
865 return (error
? VM_PAGER_ERROR
: VM_PAGER_OK
);
869 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
870 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
871 * vnode_pager_generic_putpages() to implement the previous behaviour.
873 * All other FS's should use the bypass to get to the local media
874 * backing vp's VOP_PUTPAGES.
877 vnode_pager_putpages(vm_object_t object
, vm_page_t
*m
, int count
,
878 boolean_t sync
, int *rtvals
)
882 int bytes
= count
* PAGE_SIZE
;
885 * Force synchronous operation if we are extremely low on memory
886 * to prevent a low-memory deadlock. VOP operations often need to
887 * allocate more memory to initiate the I/O ( i.e. do a BMAP
888 * operation ). The swapper handles the case by limiting the amount
889 * of asynchronous I/O, but that sort of solution doesn't scale well
890 * for the vnode pager without a lot of work.
892 * Also, the backing vnode's iodone routine may not wake the pageout
893 * daemon up. This should be probably be addressed XXX.
896 if ((vmstats
.v_free_count
+ vmstats
.v_cache_count
) < vmstats
.v_pageout_free_min
)
900 * Call device-specific putpages function
904 rtval
= VOP_PUTPAGES(vp
, m
, bytes
, sync
, rtvals
, 0);
905 if (rtval
== EOPNOTSUPP
) {
906 kprintf("vnode_pager: *** WARNING *** stale FS putpages\n");
907 rtval
= vnode_pager_generic_putpages( vp
, m
, bytes
, sync
, rtvals
);
913 * This is now called from local media FS's to operate against their
914 * own vnodes if they fail to implement VOP_PUTPAGES.
916 * This is typically called indirectly via the pageout daemon and
917 * clustering has already typically occured, so in general we ask the
918 * underlying filesystem to write the data out asynchronously rather
922 vnode_pager_generic_putpages(struct vnode
*vp
, vm_page_t
*m
, int bytecount
,
923 int flags
, int *rtvals
)
930 vm_ooffset_t poffset
;
936 object
= vp
->v_object
;
937 count
= bytecount
/ PAGE_SIZE
;
939 for (i
= 0; i
< count
; i
++)
940 rtvals
[i
] = VM_PAGER_AGAIN
;
942 if ((int) m
[0]->pindex
< 0) {
943 kprintf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n",
944 (long)m
[0]->pindex
, m
[0]->dirty
);
945 rtvals
[0] = VM_PAGER_BAD
;
949 maxsize
= count
* PAGE_SIZE
;
952 poffset
= IDX_TO_OFF(m
[0]->pindex
);
955 * If the page-aligned write is larger then the actual file we
956 * have to invalidate pages occuring beyond the file EOF. However,
957 * there is an edge case where a file may not be page-aligned where
958 * the last page is partially invalid. In this case the filesystem
959 * may not properly clear the dirty bits for the entire page (which
960 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
961 * With the page locked we are free to fix-up the dirty bits here.
963 * We do not under any circumstances truncate the valid bits, as
964 * this will screw up bogus page replacement.
966 if (maxsize
+ poffset
> vp
->v_filesize
) {
967 if (vp
->v_filesize
> poffset
) {
970 maxsize
= vp
->v_filesize
- poffset
;
971 ncount
= btoc(maxsize
);
972 if ((pgoff
= (int)maxsize
& PAGE_MASK
) != 0) {
973 vm_page_clear_dirty(m
[ncount
- 1], pgoff
,
980 if (ncount
< count
) {
981 for (i
= ncount
; i
< count
; i
++) {
982 rtvals
[i
] = VM_PAGER_BAD
;
988 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
989 * rather then a bdwrite() to prevent paging I/O from saturating
990 * the buffer cache. Dummy-up the sequential heuristic to cause
991 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
992 * the system decides how to cluster.
995 if (flags
& (VM_PAGER_PUT_SYNC
| VM_PAGER_PUT_INVAL
))
997 else if ((flags
& VM_PAGER_CLUSTER_OK
) == 0)
999 ioflags
|= (flags
& VM_PAGER_PUT_INVAL
) ? IO_INVAL
: 0;
1000 ioflags
|= IO_SEQMAX
<< IO_SEQSHIFT
;
1002 aiov
.iov_base
= (caddr_t
) 0;
1003 aiov
.iov_len
= maxsize
;
1004 auio
.uio_iov
= &aiov
;
1005 auio
.uio_iovcnt
= 1;
1006 auio
.uio_offset
= poffset
;
1007 auio
.uio_segflg
= UIO_NOCOPY
;
1008 auio
.uio_rw
= UIO_WRITE
;
1009 auio
.uio_resid
= maxsize
;
1011 error
= VOP_WRITE(vp
, &auio
, ioflags
, proc0
.p_ucred
);
1012 mycpu
->gd_cnt
.v_vnodeout
++;
1013 mycpu
->gd_cnt
.v_vnodepgsout
+= ncount
;
1016 krateprintf(&vbadrate
,
1017 "vnode_pager_putpages: I/O error %d\n", error
);
1019 if (auio
.uio_resid
) {
1020 krateprintf(&vresrate
,
1021 "vnode_pager_putpages: residual I/O %d at %lu\n",
1022 auio
.uio_resid
, (u_long
)m
[0]->pindex
);
1024 for (i
= 0; i
< ncount
; i
++) {
1025 rtvals
[i
] = VM_PAGER_OK
;
1031 vnode_pager_lock(vm_object_t object
)
1033 struct thread
*td
= curthread
; /* XXX */
1036 for (; object
!= NULL
; object
= object
->backing_object
) {
1037 if (object
->type
!= OBJT_VNODE
)
1039 if (object
->flags
& OBJ_DEAD
)
1043 struct vnode
*vp
= object
->handle
;
1044 error
= vget(vp
, LK_SHARED
| LK_RETRY
| LK_CANRECURSE
);
1046 if (object
->handle
!= vp
) {
1052 if ((object
->flags
& OBJ_DEAD
) ||
1053 (object
->type
!= OBJT_VNODE
)) {
1056 kprintf("vnode_pager_lock: vp %p error %d lockstatus %d, retrying\n", vp
, error
, lockstatus(&vp
->v_lock
, td
));
1057 tsleep(object
->handle
, 0, "vnpgrl", hz
);