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
44 * Page to/from files (vnodes).
49 * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will
50 * greatly re-simplify the vnode_pager.
53 #include <sys/cdefs.h>
54 __FBSDID("$FreeBSD$");
56 #include <sys/param.h>
57 #include <sys/systm.h>
59 #include <sys/vnode.h>
60 #include <sys/mount.h>
63 #include <sys/vmmeter.h>
64 #include <sys/limits.h>
66 #include <sys/sf_buf.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_pager.h>
72 #include <vm/vm_map.h>
73 #include <vm/vnode_pager.h>
74 #include <vm/vm_extern.h>
76 static void vnode_pager_init(void);
77 static vm_offset_t vnode_pager_addr(struct vnode *vp, vm_ooffset_t address,
79 static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m);
80 static int vnode_pager_input_old(vm_object_t object, vm_page_t m);
81 static void vnode_pager_dealloc(vm_object_t);
82 static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int);
83 static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *);
84 static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
85 static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t, vm_ooffset_t);
87 struct pagerops vnodepagerops = {
88 .pgo_init = vnode_pager_init,
89 .pgo_alloc = vnode_pager_alloc,
90 .pgo_dealloc = vnode_pager_dealloc,
91 .pgo_getpages = vnode_pager_getpages,
92 .pgo_putpages = vnode_pager_putpages,
93 .pgo_haspage = vnode_pager_haspage,
96 int vnode_pbuf_freecnt;
99 vnode_pager_init(void)
102 vnode_pbuf_freecnt = nswbuf / 2 + 1;
105 /* Create the VM system backing object for this vnode */
107 vnode_create_vobject(struct vnode *vp, size_t isize, struct thread *td)
110 vm_ooffset_t size = isize;
113 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
116 while ((object = vp->v_object) != NULL) {
117 VM_OBJECT_LOCK(object);
118 if (!(object->flags & OBJ_DEAD)) {
119 VM_OBJECT_UNLOCK(object);
122 VOP_UNLOCK(vp, 0, td);
123 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
124 msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vodead", 0);
125 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
129 if (vn_isdisk(vp, NULL)) {
130 size = IDX_TO_OFF(INT_MAX);
132 if (VOP_GETATTR(vp, &va, td->td_ucred, td) != 0)
138 object = vnode_pager_alloc(vp, size, 0, 0);
140 * Dereference the reference we just created. This assumes
141 * that the object is associated with the vp.
143 VM_OBJECT_LOCK(object);
145 VM_OBJECT_UNLOCK(object);
148 KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
154 vnode_destroy_vobject(struct vnode *vp)
156 struct vm_object *obj;
163 if (obj->ref_count == 0) {
165 * vclean() may be called twice. The first time
166 * removes the primary reference to the object,
167 * the second time goes one further and is a
168 * special-case to terminate the object.
170 * don't double-terminate the object
172 if ((obj->flags & OBJ_DEAD) == 0)
173 vm_object_terminate(obj);
175 VM_OBJECT_UNLOCK(obj);
178 * Woe to the process that tries to page now :-).
180 vm_pager_deallocate(obj);
181 VM_OBJECT_UNLOCK(obj);
187 * Allocate (or lookup) pager for a vnode.
188 * Handle is a vnode pointer.
193 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
200 * Pageout to vnode, no can do yet.
205 vp = (struct vnode *) handle;
207 ASSERT_VOP_LOCKED(vp, "vnode_pager_alloc");
210 * Prevent race condition when allocating the object. This
211 * can happen with NFS vnodes since the nfsnode isn't locked.
214 while (vp->v_iflag & VI_OLOCK) {
215 vp->v_iflag |= VI_OWANT;
216 msleep(vp, VI_MTX(vp), PVM, "vnpobj", 0);
218 vp->v_iflag |= VI_OLOCK;
222 * If the object is being terminated, wait for it to
225 while ((object = vp->v_object) != NULL) {
226 VM_OBJECT_LOCK(object);
227 if ((object->flags & OBJ_DEAD) == 0)
229 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
230 msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0);
233 if (vp->v_usecount == 0)
234 panic("vnode_pager_alloc: no vnode reference");
236 if (object == NULL) {
238 * And an object of the appropriate size
240 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
242 object->un_pager.vnp.vnp_size = size;
244 object->handle = handle;
245 vp->v_object = object;
248 VM_OBJECT_UNLOCK(object);
252 vp->v_iflag &= ~VI_OLOCK;
253 if (vp->v_iflag & VI_OWANT) {
254 vp->v_iflag &= ~VI_OWANT;
262 * The object must be locked.
265 vnode_pager_dealloc(object)
268 struct vnode *vp = object->handle;
271 panic("vnode_pager_dealloc: pager already dealloced");
273 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
274 vm_object_pip_wait(object, "vnpdea");
276 object->handle = NULL;
277 object->type = OBJT_DEAD;
278 if (object->flags & OBJ_DISCONNECTWNT) {
279 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
282 ASSERT_VOP_LOCKED(vp, "vnode_pager_dealloc");
284 vp->v_vflag &= ~VV_TEXT;
288 vnode_pager_haspage(object, pindex, before, after)
294 struct vnode *vp = object->handle;
300 int pagesperblock, blocksperpage;
303 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
305 * If no vp or vp is doomed or marked transparent to VM, we do not
312 if (vp->v_iflag & VI_DOOMED) {
318 * If filesystem no longer mounted or offset beyond end of file we do
321 if ((vp->v_mount == NULL) ||
322 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
325 bsize = vp->v_mount->mnt_stat.f_iosize;
326 pagesperblock = bsize / PAGE_SIZE;
328 if (pagesperblock > 0) {
329 reqblock = pindex / pagesperblock;
331 blocksperpage = (PAGE_SIZE / bsize);
332 reqblock = pindex * blocksperpage;
334 VM_OBJECT_UNLOCK(object);
335 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
336 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
337 VFS_UNLOCK_GIANT(vfslocked);
338 VM_OBJECT_LOCK(object);
343 if (pagesperblock > 0) {
344 poff = pindex - (reqblock * pagesperblock);
346 *before *= pagesperblock;
351 *after *= pagesperblock;
352 numafter = pagesperblock - (poff + 1);
353 if (IDX_TO_OFF(pindex + numafter) >
354 object->un_pager.vnp.vnp_size) {
356 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
363 *before /= blocksperpage;
367 *after /= blocksperpage;
374 * Lets the VM system know about a change in size for a file.
375 * We adjust our own internal size and flush any cached pages in
376 * the associated object that are affected by the size change.
378 * Note: this routine may be invoked as a result of a pager put
379 * operation (possibly at object termination time), so we must be careful.
382 vnode_pager_setsize(vp, nsize)
388 vm_pindex_t nobjsize;
390 if ((object = vp->v_object) == NULL)
392 VM_OBJECT_LOCK(object);
393 if (nsize == object->un_pager.vnp.vnp_size) {
395 * Hasn't changed size
397 VM_OBJECT_UNLOCK(object);
400 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
401 if (nsize < object->un_pager.vnp.vnp_size) {
403 * File has shrunk. Toss any cached pages beyond the new EOF.
405 if (nobjsize < object->size)
406 vm_object_page_remove(object, nobjsize, object->size,
409 * this gets rid of garbage at the end of a page that is now
410 * only partially backed by the vnode.
412 * XXX for some reason (I don't know yet), if we take a
413 * completely invalid page and mark it partially valid
414 * it can screw up NFS reads, so we don't allow the case.
416 if ((nsize & PAGE_MASK) &&
417 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
419 int base = (int)nsize & PAGE_MASK;
420 int size = PAGE_SIZE - base;
423 * Clear out partial-page garbage in case
424 * the page has been mapped.
426 pmap_zero_page_area(m, base, size);
429 * XXX work around SMP data integrity race
430 * by unmapping the page from user processes.
431 * The garbage we just cleared may be mapped
432 * to a user process running on another cpu
433 * and this code is not running through normal
434 * I/O channels which handle SMP issues for
435 * us, so unmap page to synchronize all cpus.
437 * XXX should vm_pager_unmap_page() have
440 vm_page_lock_queues();
444 * Clear out partial-page dirty bits. This
445 * has the side effect of setting the valid
446 * bits, but that is ok. There are a bunch
447 * of places in the VM system where we expected
448 * m->dirty == VM_PAGE_BITS_ALL. The file EOF
449 * case is one of them. If the page is still
450 * partially dirty, make it fully dirty.
452 * note that we do not clear out the valid
453 * bits. This would prevent bogus_page
454 * replacement from working properly.
456 vm_page_set_validclean(m, base, size);
458 m->dirty = VM_PAGE_BITS_ALL;
459 vm_page_unlock_queues();
462 object->un_pager.vnp.vnp_size = nsize;
463 object->size = nobjsize;
464 VM_OBJECT_UNLOCK(object);
468 * calculate the linear (byte) disk address of specified virtual
472 vnode_pager_addr(vp, address, run)
474 vm_ooffset_t address;
487 if (vp->v_mount == NULL)
490 bsize = vp->v_mount->mnt_stat.f_iosize;
491 vblock = address / bsize;
492 voffset = address % bsize;
494 err = VOP_BMAP(vp, vblock, NULL, &block, run, NULL);
496 if (err || (block == -1))
499 rtaddress = block + voffset / DEV_BSIZE;
502 *run *= bsize/PAGE_SIZE;
503 *run -= voffset/PAGE_SIZE;
511 * small block filesystem vnode pager input
514 vnode_pager_input_smlfs(object, m)
528 if (vp->v_mount == NULL)
531 bsize = vp->v_mount->mnt_stat.f_iosize;
533 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
535 sf = sf_buf_alloc(m, 0);
537 for (i = 0; i < PAGE_SIZE / bsize; i++) {
538 vm_ooffset_t address;
540 if (vm_page_bits(i * bsize, bsize) & m->valid)
543 address = IDX_TO_OFF(m->pindex) + i * bsize;
544 if (address >= object->un_pager.vnp.vnp_size) {
547 fileaddr = vnode_pager_addr(vp, address, NULL);
549 if (fileaddr != -1) {
550 bp = getpbuf(&vnode_pbuf_freecnt);
552 /* build a minimal buffer header */
553 bp->b_iocmd = BIO_READ;
554 bp->b_iodone = bdone;
555 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
556 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
557 bp->b_rcred = crhold(curthread->td_ucred);
558 bp->b_wcred = crhold(curthread->td_ucred);
559 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
560 bp->b_blkno = fileaddr;
562 bp->b_bcount = bsize;
563 bp->b_bufsize = bsize;
564 bp->b_runningbufspace = bp->b_bufsize;
565 runningbufspace += bp->b_runningbufspace;
568 bp->b_iooffset = dbtob(bp->b_blkno);
571 /* we definitely need to be at splvm here */
573 bwait(bp, PVM, "vnsrd");
575 if ((bp->b_ioflags & BIO_ERROR) != 0)
579 * free the buffer header back to the swap buffer pool
582 relpbuf(bp, &vnode_pbuf_freecnt);
586 VM_OBJECT_LOCK(object);
587 vm_page_lock_queues();
588 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
589 vm_page_unlock_queues();
590 VM_OBJECT_UNLOCK(object);
592 VM_OBJECT_LOCK(object);
593 vm_page_lock_queues();
594 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
595 vm_page_unlock_queues();
596 VM_OBJECT_UNLOCK(object);
597 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
601 vm_page_lock_queues();
602 pmap_clear_modify(m);
603 vm_page_unlock_queues();
605 return VM_PAGER_ERROR;
613 * old style vnode pager input routine
616 vnode_pager_input_old(object, m)
627 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
631 * Return failure if beyond current EOF
633 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
637 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
638 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
640 VM_OBJECT_UNLOCK(object);
643 * Allocate a kernel virtual address and initialize so that
644 * we can use VOP_READ/WRITE routines.
646 sf = sf_buf_alloc(m, 0);
648 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
650 auio.uio_iov = &aiov;
652 auio.uio_offset = IDX_TO_OFF(m->pindex);
653 auio.uio_segflg = UIO_SYSSPACE;
654 auio.uio_rw = UIO_READ;
655 auio.uio_resid = size;
656 auio.uio_td = curthread;
658 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
660 int count = size - auio.uio_resid;
664 else if (count != PAGE_SIZE)
665 bzero((caddr_t)sf_buf_kva(sf) + count,
670 VM_OBJECT_LOCK(object);
672 vm_page_lock_queues();
673 pmap_clear_modify(m);
675 vm_page_unlock_queues();
677 m->valid = VM_PAGE_BITS_ALL;
678 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
682 * generic vnode pager input routine
686 * Local media VFS's that do not implement their own VOP_GETPAGES
687 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
688 * to implement the previous behaviour.
690 * All other FS's should use the bypass to get to the local media
691 * backing vp's VOP_GETPAGES.
694 vnode_pager_getpages(object, m, count, reqpage)
702 int bytes = count * PAGE_SIZE;
706 VM_OBJECT_UNLOCK(object);
707 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
708 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
709 KASSERT(rtval != EOPNOTSUPP,
710 ("vnode_pager: FS getpages not implemented\n"));
711 VFS_UNLOCK_GIANT(vfslocked);
712 VM_OBJECT_LOCK(object);
717 * This is now called from local media FS's to operate against their
718 * own vnodes if they fail to implement VOP_GETPAGES.
721 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
729 off_t foff, tfoff, nextoff;
730 int i, j, size, bsize, first, firstaddr;
738 object = vp->v_object;
739 count = bytecount / PAGE_SIZE;
741 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
742 ("vnode_pager_generic_getpages does not support devices"));
743 if (vp->v_mount == NULL)
746 bsize = vp->v_mount->mnt_stat.f_iosize;
748 /* get the UNDERLYING device for the file with VOP_BMAP() */
751 * originally, we did not check for an error return value -- assuming
752 * an fs always has a bmap entry point -- that assumption is wrong!!!
754 foff = IDX_TO_OFF(m[reqpage]->pindex);
757 * if we can't bmap, use old VOP code
759 if (VOP_BMAP(vp, 0, &bo, 0, NULL, NULL)) {
760 VM_OBJECT_LOCK(object);
761 vm_page_lock_queues();
762 for (i = 0; i < count; i++)
765 vm_page_unlock_queues();
768 error = vnode_pager_input_old(object, m[reqpage]);
769 VM_OBJECT_UNLOCK(object);
773 * if the blocksize is smaller than a page size, then use
774 * special small filesystem code. NFS sometimes has a small
775 * blocksize, but it can handle large reads itself.
777 } else if ((PAGE_SIZE / bsize) > 1 &&
778 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
779 VM_OBJECT_LOCK(object);
780 vm_page_lock_queues();
781 for (i = 0; i < count; i++)
784 vm_page_unlock_queues();
785 VM_OBJECT_UNLOCK(object);
788 return vnode_pager_input_smlfs(object, m[reqpage]);
792 * If we have a completely valid page available to us, we can
793 * clean up and return. Otherwise we have to re-read the
796 VM_OBJECT_LOCK(object);
797 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
798 vm_page_lock_queues();
799 for (i = 0; i < count; i++)
802 vm_page_unlock_queues();
803 VM_OBJECT_UNLOCK(object);
806 m[reqpage]->valid = 0;
807 VM_OBJECT_UNLOCK(object);
810 * here on direct device I/O
815 * calculate the run that includes the required page
817 for (first = 0, i = 0; i < count; i = runend) {
818 firstaddr = vnode_pager_addr(vp,
819 IDX_TO_OFF(m[i]->pindex), &runpg);
820 if (firstaddr == -1) {
821 VM_OBJECT_LOCK(object);
822 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
823 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
824 firstaddr, (uintmax_t)(foff >> 32),
827 (object->un_pager.vnp.vnp_size >> 32),
828 (uintmax_t)object->un_pager.vnp.vnp_size);
830 vm_page_lock_queues();
832 vm_page_unlock_queues();
833 VM_OBJECT_UNLOCK(object);
839 if (runend <= reqpage) {
840 VM_OBJECT_LOCK(object);
841 vm_page_lock_queues();
842 for (j = i; j < runend; j++)
844 vm_page_unlock_queues();
845 VM_OBJECT_UNLOCK(object);
847 if (runpg < (count - first)) {
848 VM_OBJECT_LOCK(object);
849 vm_page_lock_queues();
850 for (i = first + runpg; i < count; i++)
852 vm_page_unlock_queues();
853 VM_OBJECT_UNLOCK(object);
854 count = first + runpg;
862 * the first and last page have been calculated now, move input pages
863 * to be zero based...
866 for (i = first; i < count; i++) {
874 * calculate the file virtual address for the transfer
876 foff = IDX_TO_OFF(m[0]->pindex);
879 * calculate the size of the transfer
881 size = count * PAGE_SIZE;
882 KASSERT(count > 0, ("zero count"));
883 if ((foff + size) > object->un_pager.vnp.vnp_size)
884 size = object->un_pager.vnp.vnp_size - foff;
885 KASSERT(size > 0, ("zero size"));
888 * round up physical size for real devices.
891 int secmask = bo->bo_bsize - 1;
892 KASSERT(secmask < PAGE_SIZE && secmask > 0,
893 ("vnode_pager_generic_getpages: sector size %d too large",
895 size = (size + secmask) & ~secmask;
898 bp = getpbuf(&vnode_pbuf_freecnt);
899 kva = (vm_offset_t) bp->b_data;
902 * and map the pages to be read into the kva
904 pmap_qenter(kva, m, count);
906 /* build a minimal buffer header */
907 bp->b_iocmd = BIO_READ;
908 bp->b_iodone = bdone;
909 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
910 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
911 bp->b_rcred = crhold(curthread->td_ucred);
912 bp->b_wcred = crhold(curthread->td_ucred);
913 bp->b_blkno = firstaddr;
916 bp->b_bufsize = size;
917 bp->b_runningbufspace = bp->b_bufsize;
918 runningbufspace += bp->b_runningbufspace;
921 cnt.v_vnodepgsin += count;
924 bp->b_iooffset = dbtob(bp->b_blkno);
927 bwait(bp, PVM, "vnread");
929 if ((bp->b_ioflags & BIO_ERROR) != 0)
933 if (size != count * PAGE_SIZE)
934 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
936 pmap_qremove(kva, count);
939 * free the buffer header back to the swap buffer pool
942 relpbuf(bp, &vnode_pbuf_freecnt);
944 VM_OBJECT_LOCK(object);
945 vm_page_lock_queues();
946 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
949 nextoff = tfoff + PAGE_SIZE;
952 if (nextoff <= object->un_pager.vnp.vnp_size) {
954 * Read filled up entire page.
956 mt->valid = VM_PAGE_BITS_ALL;
957 vm_page_undirty(mt); /* should be an assert? XXX */
958 pmap_clear_modify(mt);
961 * Read did not fill up entire page. Since this
962 * is getpages, the page may be mapped, so we have
963 * to zero the invalid portions of the page even
964 * though we aren't setting them valid.
966 * Currently we do not set the entire page valid,
967 * we just try to clear the piece that we couldn't
970 vm_page_set_validclean(mt, 0,
971 object->un_pager.vnp.vnp_size - tfoff);
972 /* handled by vm_fault now */
973 /* vm_page_zero_invalid(mt, FALSE); */
979 * whether or not to leave the page activated is up in
980 * the air, but we should put the page on a page queue
981 * somewhere. (it already is in the object). Result:
982 * It appears that empirical results show that
983 * deactivating pages is best.
987 * just in case someone was asking for this page we
988 * now tell them that it is ok to use
991 if (mt->flags & PG_WANTED)
992 vm_page_activate(mt);
994 vm_page_deactivate(mt);
1001 vm_page_unlock_queues();
1002 VM_OBJECT_UNLOCK(object);
1004 printf("vnode_pager_getpages: I/O read error\n");
1006 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
1010 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
1011 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
1012 * vnode_pager_generic_putpages() to implement the previous behaviour.
1014 * All other FS's should use the bypass to get to the local media
1015 * backing vp's VOP_PUTPAGES.
1018 vnode_pager_putpages(object, m, count, sync, rtvals)
1028 int bytes = count * PAGE_SIZE;
1031 * Force synchronous operation if we are extremely low on memory
1032 * to prevent a low-memory deadlock. VOP operations often need to
1033 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1034 * operation ). The swapper handles the case by limiting the amount
1035 * of asynchronous I/O, but that sort of solution doesn't scale well
1036 * for the vnode pager without a lot of work.
1038 * Also, the backing vnode's iodone routine may not wake the pageout
1039 * daemon up. This should be probably be addressed XXX.
1042 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
1046 * Call device-specific putpages function
1048 vp = object->handle;
1049 VM_OBJECT_UNLOCK(object);
1050 if (vp->v_type != VREG)
1052 (void)vn_start_write(vp, &mp, V_WAIT);
1053 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
1054 KASSERT(rtval != EOPNOTSUPP,
1055 ("vnode_pager: stale FS putpages\n"));
1056 vn_finished_write(mp);
1057 VM_OBJECT_LOCK(object);
1062 * This is now called from local media FS's to operate against their
1063 * own vnodes if they fail to implement VOP_PUTPAGES.
1065 * This is typically called indirectly via the pageout daemon and
1066 * clustering has already typically occured, so in general we ask the
1067 * underlying filesystem to write the data out asynchronously rather
1071 vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals)
1082 int maxsize, ncount;
1083 vm_ooffset_t poffset;
1089 object = vp->v_object;
1090 count = bytecount / PAGE_SIZE;
1092 for (i = 0; i < count; i++)
1093 rtvals[i] = VM_PAGER_AGAIN;
1095 if ((int64_t)m[0]->pindex < 0) {
1096 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1097 (long)m[0]->pindex, (u_long)m[0]->dirty);
1098 rtvals[0] = VM_PAGER_BAD;
1099 return VM_PAGER_BAD;
1102 maxsize = count * PAGE_SIZE;
1105 poffset = IDX_TO_OFF(m[0]->pindex);
1108 * If the page-aligned write is larger then the actual file we
1109 * have to invalidate pages occuring beyond the file EOF. However,
1110 * there is an edge case where a file may not be page-aligned where
1111 * the last page is partially invalid. In this case the filesystem
1112 * may not properly clear the dirty bits for the entire page (which
1113 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1114 * With the page locked we are free to fix-up the dirty bits here.
1116 * We do not under any circumstances truncate the valid bits, as
1117 * this will screw up bogus page replacement.
1119 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1120 if (object->un_pager.vnp.vnp_size > poffset) {
1123 maxsize = object->un_pager.vnp.vnp_size - poffset;
1124 ncount = btoc(maxsize);
1125 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1126 vm_page_lock_queues();
1127 vm_page_clear_dirty(m[ncount - 1], pgoff,
1129 vm_page_unlock_queues();
1135 if (ncount < count) {
1136 for (i = ncount; i < count; i++) {
1137 rtvals[i] = VM_PAGER_BAD;
1143 * pageouts are already clustered, use IO_ASYNC t o force a bawrite()
1144 * rather then a bdwrite() to prevent paging I/O from saturating
1145 * the buffer cache. Dummy-up the sequential heuristic to cause
1146 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1147 * the system decides how to cluster.
1150 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1152 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1153 ioflags |= IO_ASYNC;
1154 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1155 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1157 aiov.iov_base = (caddr_t) 0;
1158 aiov.iov_len = maxsize;
1159 auio.uio_iov = &aiov;
1160 auio.uio_iovcnt = 1;
1161 auio.uio_offset = poffset;
1162 auio.uio_segflg = UIO_NOCOPY;
1163 auio.uio_rw = UIO_WRITE;
1164 auio.uio_resid = maxsize;
1165 auio.uio_td = (struct thread *) 0;
1166 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1168 cnt.v_vnodepgsout += ncount;
1171 printf("vnode_pager_putpages: I/O error %d\n", error);
1173 if (auio.uio_resid) {
1174 printf("vnode_pager_putpages: residual I/O %d at %lu\n",
1175 auio.uio_resid, (u_long)m[0]->pindex);
1177 for (i = 0; i < ncount; i++) {
1178 rtvals[i] = VM_PAGER_OK;
1184 vnode_pager_lock(vm_object_t first_object)
1187 vm_object_t backing_object, object;
1189 VM_OBJECT_LOCK_ASSERT(first_object, MA_OWNED);
1190 for (object = first_object; object != NULL; object = backing_object) {
1191 if (object->type != OBJT_VNODE) {
1192 if ((backing_object = object->backing_object) != NULL)
1193 VM_OBJECT_LOCK(backing_object);
1194 if (object != first_object)
1195 VM_OBJECT_UNLOCK(object);
1199 if (object->flags & OBJ_DEAD) {
1200 if (object != first_object)
1201 VM_OBJECT_UNLOCK(object);
1204 vp = object->handle;
1206 VM_OBJECT_UNLOCK(object);
1207 if (first_object != object)
1208 VM_OBJECT_UNLOCK(first_object);
1209 if (vget(vp, LK_CANRECURSE | LK_INTERLOCK | LK_NOPAUSE |
1210 LK_RETRY | LK_SHARED, curthread)) {
1211 VM_OBJECT_LOCK(first_object);
1212 if (object != first_object)
1213 VM_OBJECT_LOCK(object);
1214 if (object->type != OBJT_VNODE) {
1215 if (object != first_object)
1216 VM_OBJECT_UNLOCK(object);
1219 printf("vnode_pager_lock: retrying\n");
1222 VM_OBJECT_LOCK(first_object);