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/rwlock.h>
67 #include <sys/sf_buf.h>
69 #include <machine/atomic.h>
72 #include <vm/vm_param.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pager.h>
76 #include <vm/vm_map.h>
77 #include <vm/vnode_pager.h>
78 #include <vm/vm_extern.h>
80 static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address,
81 daddr_t *rtaddress, int *run);
82 static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m);
83 static int vnode_pager_input_old(vm_object_t object, vm_page_t m);
84 static void vnode_pager_dealloc(vm_object_t);
85 static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int);
86 static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *);
87 static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
88 static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
89 vm_ooffset_t, struct ucred *cred);
91 struct pagerops vnodepagerops = {
92 .pgo_alloc = vnode_pager_alloc,
93 .pgo_dealloc = vnode_pager_dealloc,
94 .pgo_getpages = vnode_pager_getpages,
95 .pgo_putpages = vnode_pager_putpages,
96 .pgo_haspage = vnode_pager_haspage,
99 int vnode_pbuf_freecnt;
101 /* Create the VM system backing object for this vnode */
103 vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td)
106 vm_ooffset_t size = isize;
109 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
112 while ((object = vp->v_object) != NULL) {
113 VM_OBJECT_WLOCK(object);
114 if (!(object->flags & OBJ_DEAD)) {
115 VM_OBJECT_WUNLOCK(object);
119 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
120 VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vodead", 0);
121 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
125 if (vn_isdisk(vp, NULL)) {
126 size = IDX_TO_OFF(INT_MAX);
128 if (VOP_GETATTR(vp, &va, td->td_ucred))
134 object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred);
136 * Dereference the reference we just created. This assumes
137 * that the object is associated with the vp.
139 VM_OBJECT_WLOCK(object);
141 VM_OBJECT_WUNLOCK(object);
144 KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
150 vnode_destroy_vobject(struct vnode *vp)
152 struct vm_object *obj;
157 ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
158 VM_OBJECT_WLOCK(obj);
159 if (obj->ref_count == 0) {
161 * vclean() may be called twice. The first time
162 * removes the primary reference to the object,
163 * the second time goes one further and is a
164 * special-case to terminate the object.
166 * don't double-terminate the object
168 if ((obj->flags & OBJ_DEAD) == 0)
169 vm_object_terminate(obj);
171 VM_OBJECT_WUNLOCK(obj);
174 * Woe to the process that tries to page now :-).
176 vm_pager_deallocate(obj);
177 VM_OBJECT_WUNLOCK(obj);
184 * Allocate (or lookup) pager for a vnode.
185 * Handle is a vnode pointer.
190 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
191 vm_ooffset_t offset, struct ucred *cred)
197 * Pageout to vnode, no can do yet.
202 vp = (struct vnode *) handle;
205 * If the object is being terminated, wait for it to
209 while ((object = vp->v_object) != NULL) {
210 VM_OBJECT_WLOCK(object);
211 if ((object->flags & OBJ_DEAD) == 0)
213 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
214 VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0);
217 if (vp->v_usecount == 0)
218 panic("vnode_pager_alloc: no vnode reference");
220 if (object == NULL) {
222 * Add an object of the appropriate size
224 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
226 object->un_pager.vnp.vnp_size = size;
227 object->un_pager.vnp.writemappings = 0;
229 object->handle = handle;
231 if (vp->v_object != NULL) {
233 * Object has been created while we were sleeping
236 vm_object_destroy(object);
239 vp->v_object = object;
243 VM_OBJECT_WUNLOCK(object);
250 * The object must be locked.
253 vnode_pager_dealloc(object)
261 panic("vnode_pager_dealloc: pager already dealloced");
263 VM_OBJECT_ASSERT_WLOCKED(object);
264 vm_object_pip_wait(object, "vnpdea");
265 refs = object->ref_count;
267 object->handle = NULL;
268 object->type = OBJT_DEAD;
269 if (object->flags & OBJ_DISCONNECTWNT) {
270 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
273 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
274 if (object->un_pager.vnp.writemappings > 0) {
275 object->un_pager.vnp.writemappings = 0;
276 VOP_ADD_WRITECOUNT(vp, -1);
277 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
278 __func__, vp, vp->v_writecount);
282 VM_OBJECT_WUNLOCK(object);
285 VM_OBJECT_WLOCK(object);
289 vnode_pager_haspage(object, pindex, before, after)
295 struct vnode *vp = object->handle;
301 int pagesperblock, blocksperpage;
303 VM_OBJECT_ASSERT_WLOCKED(object);
305 * If no vp or vp is doomed or marked transparent to VM, we do not
308 if (vp == NULL || vp->v_iflag & VI_DOOMED)
311 * If the offset is beyond end of file we do
314 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
317 bsize = vp->v_mount->mnt_stat.f_iosize;
318 pagesperblock = bsize / PAGE_SIZE;
320 if (pagesperblock > 0) {
321 reqblock = pindex / pagesperblock;
323 blocksperpage = (PAGE_SIZE / bsize);
324 reqblock = pindex * blocksperpage;
326 VM_OBJECT_WUNLOCK(object);
327 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
328 VM_OBJECT_WLOCK(object);
333 if (pagesperblock > 0) {
334 poff = pindex - (reqblock * pagesperblock);
336 *before *= pagesperblock;
341 *after *= pagesperblock;
342 numafter = pagesperblock - (poff + 1);
343 if (IDX_TO_OFF(pindex + numafter) >
344 object->un_pager.vnp.vnp_size) {
346 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
353 *before /= blocksperpage;
357 *after /= blocksperpage;
364 * Lets the VM system know about a change in size for a file.
365 * We adjust our own internal size and flush any cached pages in
366 * the associated object that are affected by the size change.
368 * Note: this routine may be invoked as a result of a pager put
369 * operation (possibly at object termination time), so we must be careful.
372 vnode_pager_setsize(vp, nsize)
378 vm_pindex_t nobjsize;
380 if ((object = vp->v_object) == NULL)
382 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
383 VM_OBJECT_WLOCK(object);
384 if (nsize == object->un_pager.vnp.vnp_size) {
386 * Hasn't changed size
388 VM_OBJECT_WUNLOCK(object);
391 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
392 if (nsize < object->un_pager.vnp.vnp_size) {
394 * File has shrunk. Toss any cached pages beyond the new EOF.
396 if (nobjsize < object->size)
397 vm_object_page_remove(object, nobjsize, object->size,
400 * this gets rid of garbage at the end of a page that is now
401 * only partially backed by the vnode.
403 * XXX for some reason (I don't know yet), if we take a
404 * completely invalid page and mark it partially valid
405 * it can screw up NFS reads, so we don't allow the case.
407 if ((nsize & PAGE_MASK) &&
408 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
410 int base = (int)nsize & PAGE_MASK;
411 int size = PAGE_SIZE - base;
414 * Clear out partial-page garbage in case
415 * the page has been mapped.
417 pmap_zero_page_area(m, base, size);
420 * Update the valid bits to reflect the blocks that
421 * have been zeroed. Some of these valid bits may
422 * have already been set.
424 vm_page_set_valid_range(m, base, size);
427 * Round "base" to the next block boundary so that the
428 * dirty bit for a partially zeroed block is not
431 base = roundup2(base, DEV_BSIZE);
434 * Clear out partial-page dirty bits.
436 * note that we do not clear out the valid
437 * bits. This would prevent bogus_page
438 * replacement from working properly.
440 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
441 } else if ((nsize & PAGE_MASK) &&
442 vm_page_is_cached(object, OFF_TO_IDX(nsize))) {
443 vm_page_cache_free(object, OFF_TO_IDX(nsize),
447 object->un_pager.vnp.vnp_size = nsize;
448 object->size = nobjsize;
449 VM_OBJECT_WUNLOCK(object);
453 * calculate the linear (byte) disk address of specified virtual
457 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
468 if (vp->v_iflag & VI_DOOMED)
471 bsize = vp->v_mount->mnt_stat.f_iosize;
472 vblock = address / bsize;
473 voffset = address % bsize;
475 err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
477 if (*rtaddress != -1)
478 *rtaddress += voffset / DEV_BSIZE;
481 *run *= bsize/PAGE_SIZE;
482 *run -= voffset/PAGE_SIZE;
490 * small block filesystem vnode pager input
493 vnode_pager_input_smlfs(object, m)
508 if (vp->v_iflag & VI_DOOMED)
511 bsize = vp->v_mount->mnt_stat.f_iosize;
513 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
515 sf = sf_buf_alloc(m, 0);
517 for (i = 0; i < PAGE_SIZE / bsize; i++) {
518 vm_ooffset_t address;
520 bits = vm_page_bits(i * bsize, bsize);
524 address = IDX_TO_OFF(m->pindex) + i * bsize;
525 if (address >= object->un_pager.vnp.vnp_size) {
528 error = vnode_pager_addr(vp, address, &fileaddr, NULL);
532 if (fileaddr != -1) {
533 bp = getpbuf(&vnode_pbuf_freecnt);
535 /* build a minimal buffer header */
536 bp->b_iocmd = BIO_READ;
537 bp->b_iodone = bdone;
538 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
539 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
540 bp->b_rcred = crhold(curthread->td_ucred);
541 bp->b_wcred = crhold(curthread->td_ucred);
542 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
543 bp->b_blkno = fileaddr;
546 bp->b_bcount = bsize;
547 bp->b_bufsize = bsize;
548 bp->b_runningbufspace = bp->b_bufsize;
549 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
552 bp->b_iooffset = dbtob(bp->b_blkno);
555 bwait(bp, PVM, "vnsrd");
557 if ((bp->b_ioflags & BIO_ERROR) != 0)
561 * free the buffer header back to the swap buffer pool
565 relpbuf(bp, &vnode_pbuf_freecnt);
569 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
570 KASSERT((m->dirty & bits) == 0,
571 ("vnode_pager_input_smlfs: page %p is dirty", m));
572 VM_OBJECT_WLOCK(object);
574 VM_OBJECT_WUNLOCK(object);
578 return VM_PAGER_ERROR;
584 * old style vnode pager input routine
587 vnode_pager_input_old(object, m)
598 VM_OBJECT_ASSERT_WLOCKED(object);
602 * Return failure if beyond current EOF
604 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
608 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
609 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
611 VM_OBJECT_WUNLOCK(object);
614 * Allocate a kernel virtual address and initialize so that
615 * we can use VOP_READ/WRITE routines.
617 sf = sf_buf_alloc(m, 0);
619 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
621 auio.uio_iov = &aiov;
623 auio.uio_offset = IDX_TO_OFF(m->pindex);
624 auio.uio_segflg = UIO_SYSSPACE;
625 auio.uio_rw = UIO_READ;
626 auio.uio_resid = size;
627 auio.uio_td = curthread;
629 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
631 int count = size - auio.uio_resid;
635 else if (count != PAGE_SIZE)
636 bzero((caddr_t)sf_buf_kva(sf) + count,
641 VM_OBJECT_WLOCK(object);
643 KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
645 m->valid = VM_PAGE_BITS_ALL;
646 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
650 * generic vnode pager input routine
654 * Local media VFS's that do not implement their own VOP_GETPAGES
655 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
656 * to implement the previous behaviour.
658 * All other FS's should use the bypass to get to the local media
659 * backing vp's VOP_GETPAGES.
662 vnode_pager_getpages(object, m, count, reqpage)
670 int bytes = count * PAGE_SIZE;
673 VM_OBJECT_WUNLOCK(object);
674 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
675 KASSERT(rtval != EOPNOTSUPP,
676 ("vnode_pager: FS getpages not implemented\n"));
677 VM_OBJECT_WLOCK(object);
682 * This is now called from local media FS's to operate against their
683 * own vnodes if they fail to implement VOP_GETPAGES.
686 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
694 off_t foff, tfoff, nextoff;
695 int i, j, size, bsize, first;
696 daddr_t firstaddr, reqblock;
704 object = vp->v_object;
705 count = bytecount / PAGE_SIZE;
707 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
708 ("vnode_pager_generic_getpages does not support devices"));
709 if (vp->v_iflag & VI_DOOMED)
712 bsize = vp->v_mount->mnt_stat.f_iosize;
714 /* get the UNDERLYING device for the file with VOP_BMAP() */
717 * originally, we did not check for an error return value -- assuming
718 * an fs always has a bmap entry point -- that assumption is wrong!!!
720 foff = IDX_TO_OFF(m[reqpage]->pindex);
723 * if we can't bmap, use old VOP code
725 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
726 if (error == EOPNOTSUPP) {
727 VM_OBJECT_WLOCK(object);
729 for (i = 0; i < count; i++)
733 vm_page_unlock(m[i]);
735 PCPU_INC(cnt.v_vnodein);
736 PCPU_INC(cnt.v_vnodepgsin);
737 error = vnode_pager_input_old(object, m[reqpage]);
738 VM_OBJECT_WUNLOCK(object);
740 } else if (error != 0) {
741 VM_OBJECT_WLOCK(object);
742 for (i = 0; i < count; i++)
746 vm_page_unlock(m[i]);
748 VM_OBJECT_WUNLOCK(object);
749 return (VM_PAGER_ERROR);
752 * if the blocksize is smaller than a page size, then use
753 * special small filesystem code. NFS sometimes has a small
754 * blocksize, but it can handle large reads itself.
756 } else if ((PAGE_SIZE / bsize) > 1 &&
757 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
758 VM_OBJECT_WLOCK(object);
759 for (i = 0; i < count; i++)
763 vm_page_unlock(m[i]);
765 VM_OBJECT_WUNLOCK(object);
766 PCPU_INC(cnt.v_vnodein);
767 PCPU_INC(cnt.v_vnodepgsin);
768 return vnode_pager_input_smlfs(object, m[reqpage]);
772 * If we have a completely valid page available to us, we can
773 * clean up and return. Otherwise we have to re-read the
776 VM_OBJECT_WLOCK(object);
777 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
778 for (i = 0; i < count; i++)
782 vm_page_unlock(m[i]);
784 VM_OBJECT_WUNLOCK(object);
786 } else if (reqblock == -1) {
787 pmap_zero_page(m[reqpage]);
788 KASSERT(m[reqpage]->dirty == 0,
789 ("vnode_pager_generic_getpages: page %p is dirty", m));
790 m[reqpage]->valid = VM_PAGE_BITS_ALL;
791 for (i = 0; i < count; i++)
795 vm_page_unlock(m[i]);
797 VM_OBJECT_WUNLOCK(object);
798 return (VM_PAGER_OK);
800 m[reqpage]->valid = 0;
801 VM_OBJECT_WUNLOCK(object);
804 * here on direct device I/O
809 * calculate the run that includes the required page
811 for (first = 0, i = 0; i < count; i = runend) {
812 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
814 VM_OBJECT_WLOCK(object);
815 for (; i < count; i++)
819 vm_page_unlock(m[i]);
821 VM_OBJECT_WUNLOCK(object);
822 return (VM_PAGER_ERROR);
824 if (firstaddr == -1) {
825 VM_OBJECT_WLOCK(object);
826 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
827 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
828 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
831 (object->un_pager.vnp.vnp_size >> 32),
832 (uintmax_t)object->un_pager.vnp.vnp_size);
836 vm_page_unlock(m[i]);
837 VM_OBJECT_WUNLOCK(object);
843 if (runend <= reqpage) {
844 VM_OBJECT_WLOCK(object);
845 for (j = i; j < runend; j++) {
848 vm_page_unlock(m[j]);
850 VM_OBJECT_WUNLOCK(object);
852 if (runpg < (count - first)) {
853 VM_OBJECT_WLOCK(object);
854 for (i = first + runpg; i < count; i++) {
857 vm_page_unlock(m[i]);
859 VM_OBJECT_WUNLOCK(object);
860 count = first + runpg;
868 * the first and last page have been calculated now, move input pages
869 * to be zero based...
878 * calculate the file virtual address for the transfer
880 foff = IDX_TO_OFF(m[0]->pindex);
883 * calculate the size of the transfer
885 size = count * PAGE_SIZE;
886 KASSERT(count > 0, ("zero count"));
887 if ((foff + size) > object->un_pager.vnp.vnp_size)
888 size = object->un_pager.vnp.vnp_size - foff;
889 KASSERT(size > 0, ("zero size"));
892 * round up physical size for real devices.
895 int secmask = bo->bo_bsize - 1;
896 KASSERT(secmask < PAGE_SIZE && secmask > 0,
897 ("vnode_pager_generic_getpages: sector size %d too large",
899 size = (size + secmask) & ~secmask;
902 bp = getpbuf(&vnode_pbuf_freecnt);
903 kva = (vm_offset_t)bp->b_data;
906 * and map the pages to be read into the kva
908 pmap_qenter(kva, m, count);
910 /* build a minimal buffer header */
911 bp->b_iocmd = BIO_READ;
912 bp->b_iodone = bdone;
913 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
914 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
915 bp->b_rcred = crhold(curthread->td_ucred);
916 bp->b_wcred = crhold(curthread->td_ucred);
917 bp->b_blkno = firstaddr;
921 bp->b_bufsize = size;
922 bp->b_runningbufspace = bp->b_bufsize;
923 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
925 PCPU_INC(cnt.v_vnodein);
926 PCPU_ADD(cnt.v_vnodepgsin, count);
929 bp->b_iooffset = dbtob(bp->b_blkno);
932 bwait(bp, PVM, "vnread");
934 if ((bp->b_ioflags & BIO_ERROR) != 0)
938 if (size != count * PAGE_SIZE)
939 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
941 pmap_qremove(kva, count);
944 * free the buffer header back to the swap buffer pool
948 relpbuf(bp, &vnode_pbuf_freecnt);
950 VM_OBJECT_WLOCK(object);
951 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
954 nextoff = tfoff + PAGE_SIZE;
957 if (nextoff <= object->un_pager.vnp.vnp_size) {
959 * Read filled up entire page.
961 mt->valid = VM_PAGE_BITS_ALL;
962 KASSERT(mt->dirty == 0,
963 ("vnode_pager_generic_getpages: page %p is dirty",
965 KASSERT(!pmap_page_is_mapped(mt),
966 ("vnode_pager_generic_getpages: page %p is mapped",
970 * Read did not fill up entire page.
972 * Currently we do not set the entire page valid,
973 * we just try to clear the piece that we couldn't
976 vm_page_set_valid_range(mt, 0,
977 object->un_pager.vnp.vnp_size - tfoff);
978 KASSERT((mt->dirty & vm_page_bits(0,
979 object->un_pager.vnp.vnp_size - tfoff)) == 0,
980 ("vnode_pager_generic_getpages: page %p is dirty",
985 vm_page_readahead_finish(mt);
987 VM_OBJECT_WUNLOCK(object);
989 printf("vnode_pager_getpages: I/O read error\n");
991 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
995 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
996 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
997 * vnode_pager_generic_putpages() to implement the previous behaviour.
999 * All other FS's should use the bypass to get to the local media
1000 * backing vp's VOP_PUTPAGES.
1003 vnode_pager_putpages(object, m, count, sync, rtvals)
1012 int bytes = count * PAGE_SIZE;
1015 * Force synchronous operation if we are extremely low on memory
1016 * to prevent a low-memory deadlock. VOP operations often need to
1017 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1018 * operation ). The swapper handles the case by limiting the amount
1019 * of asynchronous I/O, but that sort of solution doesn't scale well
1020 * for the vnode pager without a lot of work.
1022 * Also, the backing vnode's iodone routine may not wake the pageout
1023 * daemon up. This should be probably be addressed XXX.
1026 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
1030 * Call device-specific putpages function
1032 vp = object->handle;
1033 VM_OBJECT_WUNLOCK(object);
1034 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
1035 KASSERT(rtval != EOPNOTSUPP,
1036 ("vnode_pager: stale FS putpages\n"));
1037 VM_OBJECT_WLOCK(object);
1042 * This is now called from local media FS's to operate against their
1043 * own vnodes if they fail to implement VOP_PUTPAGES.
1045 * This is typically called indirectly via the pageout daemon and
1046 * clustering has already typically occured, so in general we ask the
1047 * underlying filesystem to write the data out asynchronously rather
1051 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1052 int flags, int *rtvals)
1059 int maxsize, ncount;
1060 vm_ooffset_t poffset;
1066 static struct timeval lastfail;
1069 object = vp->v_object;
1070 count = bytecount / PAGE_SIZE;
1072 for (i = 0; i < count; i++)
1073 rtvals[i] = VM_PAGER_ERROR;
1075 if ((int64_t)ma[0]->pindex < 0) {
1076 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1077 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1078 rtvals[0] = VM_PAGER_BAD;
1079 return VM_PAGER_BAD;
1082 maxsize = count * PAGE_SIZE;
1085 poffset = IDX_TO_OFF(ma[0]->pindex);
1088 * If the page-aligned write is larger then the actual file we
1089 * have to invalidate pages occuring beyond the file EOF. However,
1090 * there is an edge case where a file may not be page-aligned where
1091 * the last page is partially invalid. In this case the filesystem
1092 * may not properly clear the dirty bits for the entire page (which
1093 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1094 * With the page locked we are free to fix-up the dirty bits here.
1096 * We do not under any circumstances truncate the valid bits, as
1097 * this will screw up bogus page replacement.
1099 VM_OBJECT_WLOCK(object);
1100 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1101 if (object->un_pager.vnp.vnp_size > poffset) {
1104 maxsize = object->un_pager.vnp.vnp_size - poffset;
1105 ncount = btoc(maxsize);
1106 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1108 * If the object is locked and the following
1109 * conditions hold, then the page's dirty
1110 * field cannot be concurrently changed by a
1114 KASSERT(m->busy > 0,
1115 ("vnode_pager_generic_putpages: page %p is not busy", m));
1116 KASSERT(!pmap_page_is_write_mapped(m),
1117 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1118 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1125 if (ncount < count) {
1126 for (i = ncount; i < count; i++) {
1127 rtvals[i] = VM_PAGER_BAD;
1131 VM_OBJECT_WUNLOCK(object);
1134 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
1135 * rather then a bdwrite() to prevent paging I/O from saturating
1136 * the buffer cache. Dummy-up the sequential heuristic to cause
1137 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1138 * the system decides how to cluster.
1141 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1143 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1144 ioflags |= IO_ASYNC;
1145 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1146 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1148 aiov.iov_base = (caddr_t) 0;
1149 aiov.iov_len = maxsize;
1150 auio.uio_iov = &aiov;
1151 auio.uio_iovcnt = 1;
1152 auio.uio_offset = poffset;
1153 auio.uio_segflg = UIO_NOCOPY;
1154 auio.uio_rw = UIO_WRITE;
1155 auio.uio_resid = maxsize;
1156 auio.uio_td = (struct thread *) 0;
1157 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1158 PCPU_INC(cnt.v_vnodeout);
1159 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1162 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1163 printf("vnode_pager_putpages: I/O error %d\n", error);
1165 if (auio.uio_resid) {
1166 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1167 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1168 auio.uio_resid, (u_long)ma[0]->pindex);
1170 for (i = 0; i < ncount; i++) {
1171 rtvals[i] = VM_PAGER_OK;
1177 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1184 obj = ma[0]->object;
1185 VM_OBJECT_WLOCK(obj);
1186 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1187 if (pos < trunc_page(written)) {
1188 rtvals[i] = VM_PAGER_OK;
1189 vm_page_undirty(ma[i]);
1191 /* Partially written page. */
1192 rtvals[i] = VM_PAGER_AGAIN;
1193 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1196 VM_OBJECT_WUNLOCK(obj);
1200 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1204 vm_ooffset_t old_wm;
1206 VM_OBJECT_WLOCK(object);
1207 if (object->type != OBJT_VNODE) {
1208 VM_OBJECT_WUNLOCK(object);
1211 old_wm = object->un_pager.vnp.writemappings;
1212 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1213 vp = object->handle;
1214 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1215 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1216 VOP_ADD_WRITECOUNT(vp, 1);
1217 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1218 __func__, vp, vp->v_writecount);
1219 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1220 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1221 VOP_ADD_WRITECOUNT(vp, -1);
1222 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1223 __func__, vp, vp->v_writecount);
1225 VM_OBJECT_WUNLOCK(object);
1229 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1236 VM_OBJECT_WLOCK(object);
1239 * First, recheck the object type to account for the race when
1240 * the vnode is reclaimed.
1242 if (object->type != OBJT_VNODE) {
1243 VM_OBJECT_WUNLOCK(object);
1248 * Optimize for the case when writemappings is not going to
1252 if (object->un_pager.vnp.writemappings != inc) {
1253 object->un_pager.vnp.writemappings -= inc;
1254 VM_OBJECT_WUNLOCK(object);
1258 vp = object->handle;
1260 VM_OBJECT_WUNLOCK(object);
1262 vn_start_write(vp, &mp, V_WAIT);
1263 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1266 * Decrement the object's writemappings, by swapping the start
1267 * and end arguments for vnode_pager_update_writecount(). If
1268 * there was not a race with vnode reclaimation, then the
1269 * vnode's v_writecount is decremented.
1271 vnode_pager_update_writecount(object, end, start);
1275 vn_finished_write(mp);