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 1.13 2004/05/08 04:11:45 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>
58 #include <sys/vnode.h>
59 #include <sys/mount.h>
61 #include <sys/vmmeter.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_pager.h>
68 #include <vm/vm_map.h>
69 #include <vm/vnode_pager.h>
70 #include <vm/vm_extern.h>
72 static vm_offset_t vnode_pager_addr (struct vnode *vp, vm_ooffset_t address,
74 static void vnode_pager_iodone (struct buf *bp);
75 static int vnode_pager_input_smlfs (vm_object_t object, vm_page_t m);
76 static int vnode_pager_input_old (vm_object_t object, vm_page_t m);
77 static void vnode_pager_dealloc (vm_object_t);
78 static int vnode_pager_getpages (vm_object_t, vm_page_t *, int, int);
79 static void vnode_pager_putpages (vm_object_t, vm_page_t *, int, boolean_t, int *);
80 static boolean_t vnode_pager_haspage (vm_object_t, vm_pindex_t, int *, int *);
82 struct pagerops vnodepagerops = {
92 int vnode_pbuf_freecnt = -1; /* start out unlimited */
95 * Allocate (or lookup) pager for a vnode.
96 * Handle is a vnode pointer.
99 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
106 * Pageout to vnode, no can do yet.
112 * XXX hack - This initialization should be put somewhere else.
114 if (vnode_pbuf_freecnt < 0) {
115 vnode_pbuf_freecnt = nswbuf / 2 + 1;
118 vp = (struct vnode *) handle;
121 * Prevent race condition when allocating the object. This
122 * can happen with NFS vnodes since the nfsnode isn't locked.
124 while (vp->v_flag & VOLOCK) {
125 vp->v_flag |= VOWANT;
126 tsleep(vp, 0, "vnpobj", 0);
128 vp->v_flag |= VOLOCK;
131 * If the object is being terminated, wait for it to
134 while (((object = vp->v_object) != NULL) &&
135 (object->flags & OBJ_DEAD)) {
136 tsleep(object, 0, "vadead", 0);
139 if (vp->v_usecount == 0)
140 panic("vnode_pager_alloc: no vnode reference");
142 if (object == NULL) {
144 * And an object of the appropriate size
146 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
149 object->un_pager.vnp.vnp_size = size;
151 object->handle = handle;
152 vp->v_object = object;
159 vp->v_flag &= ~VOLOCK;
160 if (vp->v_flag & VOWANT) {
161 vp->v_flag &= ~VOWANT;
168 vnode_pager_dealloc(vm_object_t object)
170 struct vnode *vp = object->handle;
173 panic("vnode_pager_dealloc: pager already dealloced");
175 vm_object_pip_wait(object, "vnpdea");
177 object->handle = NULL;
178 object->type = OBJT_DEAD;
180 vp->v_flag &= ~(VTEXT | VOBJBUF);
184 vnode_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
187 struct vnode *vp = object->handle;
193 int pagesperblock, blocksperpage;
196 * If no vp or vp is doomed or marked transparent to VM, we do not
199 if ((vp == NULL) || (vp->v_flag & VDOOMED))
203 * If filesystem no longer mounted or offset beyond end of file we do
206 if ((vp->v_mount == NULL) ||
207 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
210 bsize = vp->v_mount->mnt_stat.f_iosize;
211 pagesperblock = bsize / PAGE_SIZE;
213 if (pagesperblock > 0) {
214 reqblock = pindex / pagesperblock;
216 blocksperpage = (PAGE_SIZE / bsize);
217 reqblock = pindex * blocksperpage;
219 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn,
225 if (pagesperblock > 0) {
226 poff = pindex - (reqblock * pagesperblock);
228 *before *= pagesperblock;
233 *after *= pagesperblock;
234 numafter = pagesperblock - (poff + 1);
235 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) {
236 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex)));
242 *before /= blocksperpage;
246 *after /= blocksperpage;
253 * Lets the VM system know about a change in size for a file.
254 * We adjust our own internal size and flush any cached pages in
255 * the associated object that are affected by the size change.
257 * Note: this routine may be invoked as a result of a pager put
258 * operation (possibly at object termination time), so we must be careful.
261 vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize)
263 vm_pindex_t nobjsize;
264 vm_object_t object = vp->v_object;
270 * Hasn't changed size
272 if (nsize == object->un_pager.vnp.vnp_size)
275 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
278 * File has shrunk. Toss any cached pages beyond the new EOF.
280 if (nsize < object->un_pager.vnp.vnp_size) {
281 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size);
282 if (nobjsize < object->size) {
283 vm_object_page_remove(object, nobjsize, object->size,
287 * This gets rid of garbage at the end of a page that is now
288 * only partially backed by the vnode. Since we are setting
289 * the entire page valid & clean after we are done we have
290 * to be sure that the portion of the page within the file
291 * bounds is already valid. If it isn't then making it
292 * valid would create a corrupt block.
294 if (nsize & PAGE_MASK) {
298 m = vm_page_lookup(object, OFF_TO_IDX(nsize));
300 int base = (int)nsize & PAGE_MASK;
301 int size = PAGE_SIZE - base;
304 * Clear out partial-page garbage in case
305 * the page has been mapped.
307 kva = vm_pager_map_page(m);
308 bzero((caddr_t)kva + base, size);
309 vm_pager_unmap_page(kva);
312 * XXX work around SMP data integrity race
313 * by unmapping the page from user processes.
314 * The garbage we just cleared may be mapped
315 * to a user process running on another cpu
316 * and this code is not running through normal
317 * I/O channels which handle SMP issues for
318 * us, so unmap page to synchronize all cpus.
320 * XXX should vm_pager_unmap_page() have
323 vm_page_protect(m, VM_PROT_NONE);
326 * Clear out partial-page dirty bits. This
327 * has the side effect of setting the valid
328 * bits, but that is ok. There are a bunch
329 * of places in the VM system where we expected
330 * m->dirty == VM_PAGE_BITS_ALL. The file EOF
331 * case is one of them. If the page is still
332 * partially dirty, make it fully dirty.
334 * note that we do not clear out the valid
335 * bits. This would prevent bogus_page
336 * replacement from working properly.
338 vm_page_set_validclean(m, base, size);
340 m->dirty = VM_PAGE_BITS_ALL;
344 object->un_pager.vnp.vnp_size = nsize;
345 object->size = nobjsize;
349 vnode_pager_freepage(vm_page_t m)
355 * calculate the linear (byte) disk address of specified virtual
359 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, int *run)
369 if ((int) address < 0)
372 if (vp->v_mount == NULL)
375 bsize = vp->v_mount->mnt_stat.f_iosize;
376 vblock = address / bsize;
377 voffset = address % bsize;
379 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL);
381 if (err || (block == -1))
384 rtaddress = block + voffset / DEV_BSIZE;
387 *run *= bsize/PAGE_SIZE;
388 *run -= voffset/PAGE_SIZE;
396 * interrupt routine for I/O completion
399 vnode_pager_iodone(struct buf *bp)
401 bp->b_flags |= B_DONE;
406 * small block file system vnode pager input
409 vnode_pager_input_smlfs(vm_object_t object, vm_page_t m)
413 struct vnode *dp, *vp;
421 if (vp->v_mount == NULL)
424 bsize = vp->v_mount->mnt_stat.f_iosize;
427 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL);
429 kva = vm_pager_map_page(m);
431 for (i = 0; i < PAGE_SIZE / bsize; i++) {
432 vm_ooffset_t address;
434 if (vm_page_bits(i * bsize, bsize) & m->valid)
437 address = IDX_TO_OFF(m->pindex) + i * bsize;
438 if (address >= object->un_pager.vnp.vnp_size) {
441 fileaddr = vnode_pager_addr(vp, address, NULL);
443 if (fileaddr != -1) {
444 bp = getpbuf(&vnode_pbuf_freecnt);
446 /* build a minimal buffer header */
447 bp->b_flags = B_READ | B_CALL;
448 bp->b_iodone = vnode_pager_iodone;
449 bp->b_data = (caddr_t) kva + i * bsize;
450 bp->b_blkno = fileaddr;
452 bp->b_bcount = bsize;
453 bp->b_bufsize = bsize;
454 bp->b_runningbufspace = bp->b_bufsize;
455 runningbufspace += bp->b_runningbufspace;
458 VOP_STRATEGY(bp->b_vp, bp);
460 /* we definitely need to be at splvm here */
463 while ((bp->b_flags & B_DONE) == 0) {
464 tsleep(bp, 0, "vnsrd", 0);
467 if ((bp->b_flags & B_ERROR) != 0)
471 * free the buffer header back to the swap buffer pool
473 relpbuf(bp, &vnode_pbuf_freecnt);
477 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
479 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
480 bzero((caddr_t) kva + i * bsize, bsize);
483 vm_pager_unmap_page(kva);
484 pmap_clear_modify(m);
485 vm_page_flag_clear(m, PG_ZERO);
487 return VM_PAGER_ERROR;
495 * old style vnode pager output routine
498 vnode_pager_input_old(vm_object_t object, vm_page_t m)
509 * Return failure if beyond current EOF
511 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
515 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
516 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
519 * Allocate a kernel virtual address and initialize so that
520 * we can use VOP_READ/WRITE routines.
522 kva = vm_pager_map_page(m);
524 aiov.iov_base = (caddr_t) kva;
526 auio.uio_iov = &aiov;
528 auio.uio_offset = IDX_TO_OFF(m->pindex);
529 auio.uio_segflg = UIO_SYSSPACE;
530 auio.uio_rw = UIO_READ;
531 auio.uio_resid = size;
532 auio.uio_td = curthread;
534 error = VOP_READ(object->handle, &auio, 0, proc0.p_ucred);
536 int count = size - auio.uio_resid;
540 else if (count != PAGE_SIZE)
541 bzero((caddr_t) kva + count, PAGE_SIZE - count);
543 vm_pager_unmap_page(kva);
545 pmap_clear_modify(m);
547 vm_page_flag_clear(m, PG_ZERO);
549 m->valid = VM_PAGE_BITS_ALL;
550 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
554 * generic vnode pager input routine
558 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
559 * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
560 * vnode_pager_generic_getpages() to implement the previous behaviour.
562 * All other FS's should use the bypass to get to the local media
563 * backing vp's VOP_GETPAGES.
566 vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
570 int bytes = count * PAGE_SIZE;
574 * XXX temporary diagnostic message to help track stale FS code,
575 * Returning EOPNOTSUPP from here may make things unhappy.
577 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
578 if (rtval == EOPNOTSUPP) {
579 printf("vnode_pager: *** WARNING *** stale FS getpages\n");
580 rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage);
587 * This is now called from local media FS's to operate against their
588 * own vnodes if they fail to implement VOP_GETPAGES.
591 vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int bytecount,
596 off_t foff, tfoff, nextoff;
597 int i, size, bsize, first, firstaddr;
606 object = vp->v_object;
607 count = bytecount / PAGE_SIZE;
609 if (vp->v_mount == NULL)
612 bsize = vp->v_mount->mnt_stat.f_iosize;
614 /* get the UNDERLYING device for the file with VOP_BMAP() */
617 * originally, we did not check for an error return value -- assuming
618 * an fs always has a bmap entry point -- that assumption is wrong!!!
620 foff = IDX_TO_OFF(m[reqpage]->pindex);
623 * if we can't bmap, use old VOP code
625 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) {
626 for (i = 0; i < count; i++) {
628 vnode_pager_freepage(m[i]);
631 mycpu->gd_cnt.v_vnodein++;
632 mycpu->gd_cnt.v_vnodepgsin++;
633 return vnode_pager_input_old(object, m[reqpage]);
636 * if the blocksize is smaller than a page size, then use
637 * special small filesystem code. NFS sometimes has a small
638 * blocksize, but it can handle large reads itself.
640 } else if ((PAGE_SIZE / bsize) > 1 &&
641 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
642 for (i = 0; i < count; i++) {
644 vnode_pager_freepage(m[i]);
647 mycpu->gd_cnt.v_vnodein++;
648 mycpu->gd_cnt.v_vnodepgsin++;
649 return vnode_pager_input_smlfs(object, m[reqpage]);
653 * If we have a completely valid page available to us, we can
654 * clean up and return. Otherwise we have to re-read the
657 * Note that this does not work with NFS, so NFS has its own
658 * getpages routine. The problem is that NFS can have partially
659 * valid pages associated with the buffer cache due to the piecemeal
660 * write support. If we were to fall through and re-read the media
661 * as we do here, dirty data could be lost.
664 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
665 for (i = 0; i < count; i++) {
667 vnode_pager_freepage(m[i]);
671 m[reqpage]->valid = 0;
674 * here on direct device I/O
679 * calculate the run that includes the required page
681 for(first = 0, i = 0; i < count; i = runend) {
682 firstaddr = vnode_pager_addr(vp,
683 IDX_TO_OFF(m[i]->pindex), &runpg);
684 if (firstaddr == -1) {
685 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
686 /* XXX no %qd in kernel. */
687 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx",
688 firstaddr, (u_long)(foff >> 32),
689 (u_long)(u_int32_t)foff,
691 (object->un_pager.vnp.vnp_size >> 32),
693 object->un_pager.vnp.vnp_size);
695 vnode_pager_freepage(m[i]);
701 if (runend <= reqpage) {
703 for (j = i; j < runend; j++) {
704 vnode_pager_freepage(m[j]);
707 if (runpg < (count - first)) {
708 for (i = first + runpg; i < count; i++)
709 vnode_pager_freepage(m[i]);
710 count = first + runpg;
718 * the first and last page have been calculated now, move input pages
719 * to be zero based...
722 for (i = first; i < count; i++) {
730 * calculate the file virtual address for the transfer
732 foff = IDX_TO_OFF(m[0]->pindex);
735 * calculate the size of the transfer
737 size = count * PAGE_SIZE;
738 if ((foff + size) > object->un_pager.vnp.vnp_size)
739 size = object->un_pager.vnp.vnp_size - foff;
742 * round up physical size for real devices.
744 if (dp->v_type == VBLK || dp->v_type == VCHR) {
745 int secmask = dp->v_rdev->si_bsize_phys - 1;
746 KASSERT(secmask < PAGE_SIZE, ("vnode_pager_generic_getpages: sector size %d too large\n", secmask + 1));
747 size = (size + secmask) & ~secmask;
750 bp = getpbuf(&vnode_pbuf_freecnt);
751 kva = (vm_offset_t) bp->b_data;
754 * and map the pages to be read into the kva
756 pmap_qenter(kva, m, count);
758 /* build a minimal buffer header */
759 bp->b_flags = B_READ | B_CALL;
760 bp->b_iodone = vnode_pager_iodone;
761 /* B_PHYS is not set, but it is nice to fill this in */
762 bp->b_blkno = firstaddr;
765 bp->b_bufsize = size;
766 bp->b_runningbufspace = bp->b_bufsize;
767 runningbufspace += bp->b_runningbufspace;
769 mycpu->gd_cnt.v_vnodein++;
770 mycpu->gd_cnt.v_vnodepgsin += count;
773 VOP_STRATEGY(bp->b_vp, bp);
776 /* we definitely need to be at splvm here */
778 while ((bp->b_flags & B_DONE) == 0) {
779 tsleep(bp, 0, "vnread", 0);
782 if ((bp->b_flags & B_ERROR) != 0)
786 if (size != count * PAGE_SIZE)
787 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
789 pmap_qremove(kva, count);
792 * free the buffer header back to the swap buffer pool
794 relpbuf(bp, &vnode_pbuf_freecnt);
796 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
799 nextoff = tfoff + PAGE_SIZE;
802 if (nextoff <= object->un_pager.vnp.vnp_size) {
804 * Read filled up entire page.
806 mt->valid = VM_PAGE_BITS_ALL;
807 vm_page_undirty(mt); /* should be an assert? XXX */
808 pmap_clear_modify(mt);
811 * Read did not fill up entire page. Since this
812 * is getpages, the page may be mapped, so we have
813 * to zero the invalid portions of the page even
814 * though we aren't setting them valid.
816 * Currently we do not set the entire page valid,
817 * we just try to clear the piece that we couldn't
820 vm_page_set_validclean(mt, 0,
821 object->un_pager.vnp.vnp_size - tfoff);
822 /* handled by vm_fault now */
823 /* vm_page_zero_invalid(mt, FALSE); */
826 vm_page_flag_clear(mt, PG_ZERO);
830 * whether or not to leave the page activated is up in
831 * the air, but we should put the page on a page queue
832 * somewhere. (it already is in the object). Result:
833 * It appears that empirical results show that
834 * deactivating pages is best.
838 * just in case someone was asking for this page we
839 * now tell them that it is ok to use
842 if (mt->flags & PG_WANTED)
843 vm_page_activate(mt);
845 vm_page_deactivate(mt);
848 vnode_pager_freepage(mt);
853 printf("vnode_pager_getpages: I/O read error\n");
855 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
859 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
860 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
861 * vnode_pager_generic_putpages() to implement the previous behaviour.
863 * All other FS's should use the bypass to get to the local media
864 * backing vp's VOP_PUTPAGES.
867 vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
868 boolean_t sync, int *rtvals)
872 int bytes = count * PAGE_SIZE;
875 * Force synchronous operation if we are extremely low on memory
876 * to prevent a low-memory deadlock. VOP operations often need to
877 * allocate more memory to initiate the I/O ( i.e. do a BMAP
878 * operation ). The swapper handles the case by limiting the amount
879 * of asynchronous I/O, but that sort of solution doesn't scale well
880 * for the vnode pager without a lot of work.
882 * Also, the backing vnode's iodone routine may not wake the pageout
883 * daemon up. This should be probably be addressed XXX.
886 if ((vmstats.v_free_count + vmstats.v_cache_count) < vmstats.v_pageout_free_min)
890 * Call device-specific putpages function
894 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
895 if (rtval == EOPNOTSUPP) {
896 printf("vnode_pager: *** WARNING *** stale FS putpages\n");
897 rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
903 * This is now called from local media FS's to operate against their
904 * own vnodes if they fail to implement VOP_PUTPAGES.
906 * This is typically called indirectly via the pageout daemon and
907 * clustering has already typically occured, so in general we ask the
908 * underlying filesystem to write the data out asynchronously rather
912 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *m, int bytecount,
913 int flags, int *rtvals)
920 vm_ooffset_t poffset;
926 object = vp->v_object;
927 count = bytecount / PAGE_SIZE;
929 for (i = 0; i < count; i++)
930 rtvals[i] = VM_PAGER_AGAIN;
932 if ((int) m[0]->pindex < 0) {
933 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%x)\n",
934 (long)m[0]->pindex, m[0]->dirty);
935 rtvals[0] = VM_PAGER_BAD;
939 maxsize = count * PAGE_SIZE;
942 poffset = IDX_TO_OFF(m[0]->pindex);
945 * If the page-aligned write is larger then the actual file we
946 * have to invalidate pages occuring beyond the file EOF. However,
947 * there is an edge case where a file may not be page-aligned where
948 * the last page is partially invalid. In this case the filesystem
949 * may not properly clear the dirty bits for the entire page (which
950 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
951 * With the page locked we are free to fix-up the dirty bits here.
953 * We do not under any circumstances truncate the valid bits, as
954 * this will screw up bogus page replacement.
956 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
957 if (object->un_pager.vnp.vnp_size > poffset) {
960 maxsize = object->un_pager.vnp.vnp_size - poffset;
961 ncount = btoc(maxsize);
962 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
963 vm_page_clear_dirty(m[ncount - 1], pgoff,
970 if (ncount < count) {
971 for (i = ncount; i < count; i++) {
972 rtvals[i] = VM_PAGER_BAD;
978 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
979 * rather then a bdwrite() to prevent paging I/O from saturating
980 * the buffer cache. Dummy-up the sequential heuristic to cause
981 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
982 * the system decides how to cluster.
985 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
987 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
989 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
990 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
992 aiov.iov_base = (caddr_t) 0;
993 aiov.iov_len = maxsize;
994 auio.uio_iov = &aiov;
996 auio.uio_offset = poffset;
997 auio.uio_segflg = UIO_NOCOPY;
998 auio.uio_rw = UIO_WRITE;
999 auio.uio_resid = maxsize;
1001 error = VOP_WRITE(vp, &auio, ioflags, proc0.p_ucred);
1002 mycpu->gd_cnt.v_vnodeout++;
1003 mycpu->gd_cnt.v_vnodepgsout += ncount;
1006 printf("vnode_pager_putpages: I/O error %d\n", error);
1008 if (auio.uio_resid) {
1009 printf("vnode_pager_putpages: residual I/O %d at %lu\n",
1010 auio.uio_resid, (u_long)m[0]->pindex);
1012 for (i = 0; i < ncount; i++) {
1013 rtvals[i] = VM_PAGER_OK;
1019 vnode_pager_lock(vm_object_t object)
1021 struct thread *td = curthread; /* XXX */
1023 for (; object != NULL; object = object->backing_object) {
1024 if (object->type != OBJT_VNODE)
1026 if (object->flags & OBJ_DEAD)
1029 while (vget(object->handle, NULL,
1030 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, td)) {
1031 if ((object->flags & OBJ_DEAD) ||
1032 (object->type != OBJT_VNODE)) {
1035 printf("vnode_pager_lock: retrying\n");
1037 return object->handle;