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>
68 #include <machine/atomic.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_pager.h>
75 #include <vm/vm_map.h>
76 #include <vm/vnode_pager.h>
77 #include <vm/vm_extern.h>
79 static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address,
80 daddr_t *rtaddress, int *run);
81 static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m);
82 static int vnode_pager_input_old(vm_object_t object, vm_page_t m);
83 static void vnode_pager_dealloc(vm_object_t);
84 static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int);
85 static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *);
86 static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
87 static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
88 vm_ooffset_t, struct ucred *cred);
90 struct pagerops vnodepagerops = {
91 .pgo_alloc = vnode_pager_alloc,
92 .pgo_dealloc = vnode_pager_dealloc,
93 .pgo_getpages = vnode_pager_getpages,
94 .pgo_putpages = vnode_pager_putpages,
95 .pgo_haspage = vnode_pager_haspage,
98 int vnode_pbuf_freecnt;
100 /* Create the VM system backing object for this vnode */
102 vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td)
105 vm_ooffset_t size = isize;
108 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
111 while ((object = vp->v_object) != NULL) {
112 VM_OBJECT_LOCK(object);
113 if (!(object->flags & OBJ_DEAD)) {
114 VM_OBJECT_UNLOCK(object);
118 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
119 VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vodead", 0);
120 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
124 if (vn_isdisk(vp, NULL)) {
125 size = IDX_TO_OFF(INT_MAX);
127 if (VOP_GETATTR(vp, &va, td->td_ucred))
133 object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred);
135 * Dereference the reference we just created. This assumes
136 * that the object is associated with the vp.
138 VM_OBJECT_LOCK(object);
140 VM_OBJECT_UNLOCK(object);
143 KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
149 vnode_destroy_vobject(struct vnode *vp)
151 struct vm_object *obj;
156 ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
158 if (obj->ref_count == 0) {
160 * vclean() may be called twice. The first time
161 * removes the primary reference to the object,
162 * the second time goes one further and is a
163 * special-case to terminate the object.
165 * don't double-terminate the object
167 if ((obj->flags & OBJ_DEAD) == 0)
168 vm_object_terminate(obj);
170 VM_OBJECT_UNLOCK(obj);
173 * Woe to the process that tries to page now :-).
175 vm_pager_deallocate(obj);
176 VM_OBJECT_UNLOCK(obj);
183 * Allocate (or lookup) pager for a vnode.
184 * Handle is a vnode pointer.
189 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
190 vm_ooffset_t offset, struct ucred *cred)
196 * Pageout to vnode, no can do yet.
201 vp = (struct vnode *) handle;
204 * If the object is being terminated, wait for it to
208 while ((object = vp->v_object) != NULL) {
209 VM_OBJECT_LOCK(object);
210 if ((object->flags & OBJ_DEAD) == 0)
212 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
213 VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0);
216 if (vp->v_usecount == 0)
217 panic("vnode_pager_alloc: no vnode reference");
219 if (object == NULL) {
221 * Add an object of the appropriate size
223 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
225 object->un_pager.vnp.vnp_size = size;
226 object->un_pager.vnp.writemappings = 0;
228 object->handle = handle;
230 if (vp->v_object != NULL) {
232 * Object has been created while we were sleeping
235 vm_object_destroy(object);
238 vp->v_object = object;
242 VM_OBJECT_UNLOCK(object);
249 * The object must be locked.
252 vnode_pager_dealloc(object)
260 panic("vnode_pager_dealloc: pager already dealloced");
262 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
263 vm_object_pip_wait(object, "vnpdea");
264 refs = object->ref_count;
266 object->handle = NULL;
267 object->type = OBJT_DEAD;
268 if (object->flags & OBJ_DISCONNECTWNT) {
269 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
272 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
273 if (object->un_pager.vnp.writemappings > 0) {
274 object->un_pager.vnp.writemappings = 0;
275 VOP_ADD_WRITECOUNT(vp, -1);
276 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
277 __func__, vp, vp->v_writecount);
281 VM_OBJECT_UNLOCK(object);
284 VM_OBJECT_LOCK(object);
288 vnode_pager_haspage(object, pindex, before, after)
294 struct vnode *vp = object->handle;
300 int pagesperblock, blocksperpage;
302 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
304 * If no vp or vp is doomed or marked transparent to VM, we do not
307 if (vp == NULL || vp->v_iflag & VI_DOOMED)
310 * If the offset is beyond end of file we do
313 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
316 bsize = vp->v_mount->mnt_stat.f_iosize;
317 pagesperblock = bsize / PAGE_SIZE;
319 if (pagesperblock > 0) {
320 reqblock = pindex / pagesperblock;
322 blocksperpage = (PAGE_SIZE / bsize);
323 reqblock = pindex * blocksperpage;
325 VM_OBJECT_UNLOCK(object);
326 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
327 VM_OBJECT_LOCK(object);
332 if (pagesperblock > 0) {
333 poff = pindex - (reqblock * pagesperblock);
335 *before *= pagesperblock;
340 *after *= pagesperblock;
341 numafter = pagesperblock - (poff + 1);
342 if (IDX_TO_OFF(pindex + numafter) >
343 object->un_pager.vnp.vnp_size) {
345 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
352 *before /= blocksperpage;
356 *after /= blocksperpage;
363 * Lets the VM system know about a change in size for a file.
364 * We adjust our own internal size and flush any cached pages in
365 * the associated object that are affected by the size change.
367 * Note: this routine may be invoked as a result of a pager put
368 * operation (possibly at object termination time), so we must be careful.
371 vnode_pager_setsize(vp, nsize)
377 vm_pindex_t nobjsize;
379 if ((object = vp->v_object) == NULL)
381 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
382 VM_OBJECT_LOCK(object);
383 if (nsize == object->un_pager.vnp.vnp_size) {
385 * Hasn't changed size
387 VM_OBJECT_UNLOCK(object);
390 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
391 if (nsize < object->un_pager.vnp.vnp_size) {
393 * File has shrunk. Toss any cached pages beyond the new EOF.
395 if (nobjsize < object->size)
396 vm_object_page_remove(object, nobjsize, object->size,
399 * this gets rid of garbage at the end of a page that is now
400 * only partially backed by the vnode.
402 * XXX for some reason (I don't know yet), if we take a
403 * completely invalid page and mark it partially valid
404 * it can screw up NFS reads, so we don't allow the case.
406 if ((nsize & PAGE_MASK) &&
407 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
409 int base = (int)nsize & PAGE_MASK;
410 int size = PAGE_SIZE - base;
413 * Clear out partial-page garbage in case
414 * the page has been mapped.
416 pmap_zero_page_area(m, base, size);
419 * Update the valid bits to reflect the blocks that
420 * have been zeroed. Some of these valid bits may
421 * have already been set.
423 vm_page_set_valid_range(m, base, size);
426 * Round "base" to the next block boundary so that the
427 * dirty bit for a partially zeroed block is not
430 base = roundup2(base, DEV_BSIZE);
433 * Clear out partial-page dirty bits.
435 * note that we do not clear out the valid
436 * bits. This would prevent bogus_page
437 * replacement from working properly.
439 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
440 } else if ((nsize & PAGE_MASK) &&
441 vm_page_is_cached(object, OFF_TO_IDX(nsize))) {
442 vm_page_cache_free(object, OFF_TO_IDX(nsize),
446 object->un_pager.vnp.vnp_size = nsize;
447 object->size = nobjsize;
448 VM_OBJECT_UNLOCK(object);
452 * calculate the linear (byte) disk address of specified virtual
456 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
467 if (vp->v_iflag & VI_DOOMED)
470 bsize = vp->v_mount->mnt_stat.f_iosize;
471 vblock = address / bsize;
472 voffset = address % bsize;
474 err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
476 if (*rtaddress != -1)
477 *rtaddress += voffset / DEV_BSIZE;
480 *run *= bsize/PAGE_SIZE;
481 *run -= voffset/PAGE_SIZE;
489 * small block filesystem vnode pager input
492 vnode_pager_input_smlfs(object, m)
507 if (vp->v_iflag & VI_DOOMED)
510 bsize = vp->v_mount->mnt_stat.f_iosize;
512 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
514 sf = sf_buf_alloc(m, 0);
516 for (i = 0; i < PAGE_SIZE / bsize; i++) {
517 vm_ooffset_t address;
519 bits = vm_page_bits(i * bsize, bsize);
523 address = IDX_TO_OFF(m->pindex) + i * bsize;
524 if (address >= object->un_pager.vnp.vnp_size) {
527 error = vnode_pager_addr(vp, address, &fileaddr, NULL);
531 if (fileaddr != -1) {
532 bp = getpbuf(&vnode_pbuf_freecnt);
534 /* build a minimal buffer header */
535 bp->b_iocmd = BIO_READ;
536 bp->b_iodone = bdone;
537 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
538 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
539 bp->b_rcred = crhold(curthread->td_ucred);
540 bp->b_wcred = crhold(curthread->td_ucred);
541 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
542 bp->b_blkno = fileaddr;
545 bp->b_bcount = bsize;
546 bp->b_bufsize = bsize;
547 bp->b_runningbufspace = bp->b_bufsize;
548 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
551 bp->b_iooffset = dbtob(bp->b_blkno);
554 bwait(bp, PVM, "vnsrd");
556 if ((bp->b_ioflags & BIO_ERROR) != 0)
560 * free the buffer header back to the swap buffer pool
564 relpbuf(bp, &vnode_pbuf_freecnt);
568 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
569 KASSERT((m->dirty & bits) == 0,
570 ("vnode_pager_input_smlfs: page %p is dirty", m));
571 VM_OBJECT_LOCK(object);
573 VM_OBJECT_UNLOCK(object);
577 return VM_PAGER_ERROR;
583 * old style vnode pager input routine
586 vnode_pager_input_old(object, m)
597 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
601 * Return failure if beyond current EOF
603 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
607 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
608 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
610 VM_OBJECT_UNLOCK(object);
613 * Allocate a kernel virtual address and initialize so that
614 * we can use VOP_READ/WRITE routines.
616 sf = sf_buf_alloc(m, 0);
618 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
620 auio.uio_iov = &aiov;
622 auio.uio_offset = IDX_TO_OFF(m->pindex);
623 auio.uio_segflg = UIO_SYSSPACE;
624 auio.uio_rw = UIO_READ;
625 auio.uio_resid = size;
626 auio.uio_td = curthread;
628 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
630 int count = size - auio.uio_resid;
634 else if (count != PAGE_SIZE)
635 bzero((caddr_t)sf_buf_kva(sf) + count,
640 VM_OBJECT_LOCK(object);
642 KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
644 m->valid = VM_PAGE_BITS_ALL;
645 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
649 * generic vnode pager input routine
653 * Local media VFS's that do not implement their own VOP_GETPAGES
654 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
655 * to implement the previous behaviour.
657 * All other FS's should use the bypass to get to the local media
658 * backing vp's VOP_GETPAGES.
661 vnode_pager_getpages(object, m, count, reqpage)
669 int bytes = count * PAGE_SIZE;
672 VM_OBJECT_UNLOCK(object);
673 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
674 KASSERT(rtval != EOPNOTSUPP,
675 ("vnode_pager: FS getpages not implemented\n"));
676 VM_OBJECT_LOCK(object);
681 * This is now called from local media FS's to operate against their
682 * own vnodes if they fail to implement VOP_GETPAGES.
685 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
693 off_t foff, tfoff, nextoff;
694 int i, j, size, bsize, first;
695 daddr_t firstaddr, reqblock;
703 object = vp->v_object;
704 count = bytecount / PAGE_SIZE;
706 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
707 ("vnode_pager_generic_getpages does not support devices"));
708 if (vp->v_iflag & VI_DOOMED)
711 bsize = vp->v_mount->mnt_stat.f_iosize;
713 /* get the UNDERLYING device for the file with VOP_BMAP() */
716 * originally, we did not check for an error return value -- assuming
717 * an fs always has a bmap entry point -- that assumption is wrong!!!
719 foff = IDX_TO_OFF(m[reqpage]->pindex);
722 * if we can't bmap, use old VOP code
724 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
725 if (error == EOPNOTSUPP) {
726 VM_OBJECT_LOCK(object);
728 for (i = 0; i < count; i++)
732 vm_page_unlock(m[i]);
734 PCPU_INC(cnt.v_vnodein);
735 PCPU_INC(cnt.v_vnodepgsin);
736 error = vnode_pager_input_old(object, m[reqpage]);
737 VM_OBJECT_UNLOCK(object);
739 } else if (error != 0) {
740 VM_OBJECT_LOCK(object);
741 for (i = 0; i < count; i++)
745 vm_page_unlock(m[i]);
747 VM_OBJECT_UNLOCK(object);
748 return (VM_PAGER_ERROR);
751 * if the blocksize is smaller than a page size, then use
752 * special small filesystem code. NFS sometimes has a small
753 * blocksize, but it can handle large reads itself.
755 } else if ((PAGE_SIZE / bsize) > 1 &&
756 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
757 VM_OBJECT_LOCK(object);
758 for (i = 0; i < count; i++)
762 vm_page_unlock(m[i]);
764 VM_OBJECT_UNLOCK(object);
765 PCPU_INC(cnt.v_vnodein);
766 PCPU_INC(cnt.v_vnodepgsin);
767 return vnode_pager_input_smlfs(object, m[reqpage]);
771 * If we have a completely valid page available to us, we can
772 * clean up and return. Otherwise we have to re-read the
775 VM_OBJECT_LOCK(object);
776 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
777 for (i = 0; i < count; i++)
781 vm_page_unlock(m[i]);
783 VM_OBJECT_UNLOCK(object);
785 } else if (reqblock == -1) {
786 pmap_zero_page(m[reqpage]);
787 KASSERT(m[reqpage]->dirty == 0,
788 ("vnode_pager_generic_getpages: page %p is dirty", m));
789 m[reqpage]->valid = VM_PAGE_BITS_ALL;
790 for (i = 0; i < count; i++)
794 vm_page_unlock(m[i]);
796 VM_OBJECT_UNLOCK(object);
797 return (VM_PAGER_OK);
799 m[reqpage]->valid = 0;
800 VM_OBJECT_UNLOCK(object);
803 * here on direct device I/O
808 * calculate the run that includes the required page
810 for (first = 0, i = 0; i < count; i = runend) {
811 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
813 VM_OBJECT_LOCK(object);
814 for (; i < count; i++)
818 vm_page_unlock(m[i]);
820 VM_OBJECT_UNLOCK(object);
821 return (VM_PAGER_ERROR);
823 if (firstaddr == -1) {
824 VM_OBJECT_LOCK(object);
825 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
826 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
827 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
830 (object->un_pager.vnp.vnp_size >> 32),
831 (uintmax_t)object->un_pager.vnp.vnp_size);
835 vm_page_unlock(m[i]);
836 VM_OBJECT_UNLOCK(object);
842 if (runend <= reqpage) {
843 VM_OBJECT_LOCK(object);
844 for (j = i; j < runend; j++) {
847 vm_page_unlock(m[j]);
849 VM_OBJECT_UNLOCK(object);
851 if (runpg < (count - first)) {
852 VM_OBJECT_LOCK(object);
853 for (i = first + runpg; i < count; i++) {
856 vm_page_unlock(m[i]);
858 VM_OBJECT_UNLOCK(object);
859 count = first + runpg;
867 * the first and last page have been calculated now, move input pages
868 * to be zero based...
877 * calculate the file virtual address for the transfer
879 foff = IDX_TO_OFF(m[0]->pindex);
882 * calculate the size of the transfer
884 size = count * PAGE_SIZE;
885 KASSERT(count > 0, ("zero count"));
886 if ((foff + size) > object->un_pager.vnp.vnp_size)
887 size = object->un_pager.vnp.vnp_size - foff;
888 KASSERT(size > 0, ("zero size"));
891 * round up physical size for real devices.
894 int secmask = bo->bo_bsize - 1;
895 KASSERT(secmask < PAGE_SIZE && secmask > 0,
896 ("vnode_pager_generic_getpages: sector size %d too large",
898 size = (size + secmask) & ~secmask;
901 bp = getpbuf(&vnode_pbuf_freecnt);
902 kva = (vm_offset_t) bp->b_data;
905 * and map the pages to be read into the kva
907 pmap_qenter(kva, m, count);
909 /* build a minimal buffer header */
910 bp->b_iocmd = BIO_READ;
911 bp->b_iodone = bdone;
912 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
913 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
914 bp->b_rcred = crhold(curthread->td_ucred);
915 bp->b_wcred = crhold(curthread->td_ucred);
916 bp->b_blkno = firstaddr;
920 bp->b_bufsize = size;
921 bp->b_runningbufspace = bp->b_bufsize;
922 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
924 PCPU_INC(cnt.v_vnodein);
925 PCPU_ADD(cnt.v_vnodepgsin, count);
928 bp->b_iooffset = dbtob(bp->b_blkno);
931 bwait(bp, PVM, "vnread");
933 if ((bp->b_ioflags & BIO_ERROR) != 0)
937 if (size != count * PAGE_SIZE)
938 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
940 pmap_qremove(kva, count);
943 * free the buffer header back to the swap buffer pool
947 relpbuf(bp, &vnode_pbuf_freecnt);
949 VM_OBJECT_LOCK(object);
950 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
953 nextoff = tfoff + PAGE_SIZE;
956 if (nextoff <= object->un_pager.vnp.vnp_size) {
958 * Read filled up entire page.
960 mt->valid = VM_PAGE_BITS_ALL;
961 KASSERT(mt->dirty == 0,
962 ("vnode_pager_generic_getpages: page %p is dirty",
964 KASSERT(!pmap_page_is_mapped(mt),
965 ("vnode_pager_generic_getpages: page %p is mapped",
969 * Read did not fill up entire page.
971 * Currently we do not set the entire page valid,
972 * we just try to clear the piece that we couldn't
975 vm_page_set_valid_range(mt, 0,
976 object->un_pager.vnp.vnp_size - tfoff);
977 KASSERT((mt->dirty & vm_page_bits(0,
978 object->un_pager.vnp.vnp_size - tfoff)) == 0,
979 ("vnode_pager_generic_getpages: page %p is dirty",
984 vm_page_readahead_finish(mt);
986 VM_OBJECT_UNLOCK(object);
988 printf("vnode_pager_getpages: I/O read error\n");
990 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
994 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
995 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
996 * vnode_pager_generic_putpages() to implement the previous behaviour.
998 * All other FS's should use the bypass to get to the local media
999 * backing vp's VOP_PUTPAGES.
1002 vnode_pager_putpages(object, m, count, sync, rtvals)
1011 int bytes = count * PAGE_SIZE;
1014 * Force synchronous operation if we are extremely low on memory
1015 * to prevent a low-memory deadlock. VOP operations often need to
1016 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1017 * operation ). The swapper handles the case by limiting the amount
1018 * of asynchronous I/O, but that sort of solution doesn't scale well
1019 * for the vnode pager without a lot of work.
1021 * Also, the backing vnode's iodone routine may not wake the pageout
1022 * daemon up. This should be probably be addressed XXX.
1025 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
1029 * Call device-specific putpages function
1031 vp = object->handle;
1032 VM_OBJECT_UNLOCK(object);
1033 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
1034 KASSERT(rtval != EOPNOTSUPP,
1035 ("vnode_pager: stale FS putpages\n"));
1036 VM_OBJECT_LOCK(object);
1041 * This is now called from local media FS's to operate against their
1042 * own vnodes if they fail to implement VOP_PUTPAGES.
1044 * This is typically called indirectly via the pageout daemon and
1045 * clustering has already typically occured, so in general we ask the
1046 * underlying filesystem to write the data out asynchronously rather
1050 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1051 int flags, int *rtvals)
1058 int maxsize, ncount;
1059 vm_ooffset_t poffset;
1065 static struct timeval lastfail;
1068 object = vp->v_object;
1069 count = bytecount / PAGE_SIZE;
1071 for (i = 0; i < count; i++)
1072 rtvals[i] = VM_PAGER_ERROR;
1074 if ((int64_t)ma[0]->pindex < 0) {
1075 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1076 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1077 rtvals[0] = VM_PAGER_BAD;
1078 return VM_PAGER_BAD;
1081 maxsize = count * PAGE_SIZE;
1084 poffset = IDX_TO_OFF(ma[0]->pindex);
1087 * If the page-aligned write is larger then the actual file we
1088 * have to invalidate pages occuring beyond the file EOF. However,
1089 * there is an edge case where a file may not be page-aligned where
1090 * the last page is partially invalid. In this case the filesystem
1091 * may not properly clear the dirty bits for the entire page (which
1092 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1093 * With the page locked we are free to fix-up the dirty bits here.
1095 * We do not under any circumstances truncate the valid bits, as
1096 * this will screw up bogus page replacement.
1098 VM_OBJECT_LOCK(object);
1099 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1100 if (object->un_pager.vnp.vnp_size > poffset) {
1103 maxsize = object->un_pager.vnp.vnp_size - poffset;
1104 ncount = btoc(maxsize);
1105 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1107 * If the object is locked and the following
1108 * conditions hold, then the page's dirty
1109 * field cannot be concurrently changed by a
1113 KASSERT(m->busy > 0,
1114 ("vnode_pager_generic_putpages: page %p is not busy", m));
1115 KASSERT(!pmap_page_is_write_mapped(m),
1116 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1117 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1124 if (ncount < count) {
1125 for (i = ncount; i < count; i++) {
1126 rtvals[i] = VM_PAGER_BAD;
1130 VM_OBJECT_UNLOCK(object);
1133 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
1134 * rather then a bdwrite() to prevent paging I/O from saturating
1135 * the buffer cache. Dummy-up the sequential heuristic to cause
1136 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1137 * the system decides how to cluster.
1140 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1142 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1143 ioflags |= IO_ASYNC;
1144 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1145 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1147 aiov.iov_base = (caddr_t) 0;
1148 aiov.iov_len = maxsize;
1149 auio.uio_iov = &aiov;
1150 auio.uio_iovcnt = 1;
1151 auio.uio_offset = poffset;
1152 auio.uio_segflg = UIO_NOCOPY;
1153 auio.uio_rw = UIO_WRITE;
1154 auio.uio_resid = maxsize;
1155 auio.uio_td = (struct thread *) 0;
1156 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1157 PCPU_INC(cnt.v_vnodeout);
1158 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1161 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1162 printf("vnode_pager_putpages: I/O error %d\n", error);
1164 if (auio.uio_resid) {
1165 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1166 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1167 auio.uio_resid, (u_long)ma[0]->pindex);
1169 for (i = 0; i < ncount; i++) {
1170 rtvals[i] = VM_PAGER_OK;
1176 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1183 obj = ma[0]->object;
1184 VM_OBJECT_LOCK(obj);
1185 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1186 if (pos < trunc_page(written)) {
1187 rtvals[i] = VM_PAGER_OK;
1188 vm_page_undirty(ma[i]);
1190 /* Partially written page. */
1191 rtvals[i] = VM_PAGER_AGAIN;
1192 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1195 VM_OBJECT_UNLOCK(obj);
1199 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1203 vm_ooffset_t old_wm;
1205 VM_OBJECT_LOCK(object);
1206 if (object->type != OBJT_VNODE) {
1207 VM_OBJECT_UNLOCK(object);
1210 old_wm = object->un_pager.vnp.writemappings;
1211 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1212 vp = object->handle;
1213 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1214 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1215 VOP_ADD_WRITECOUNT(vp, 1);
1216 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1217 __func__, vp, vp->v_writecount);
1218 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1219 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1220 VOP_ADD_WRITECOUNT(vp, -1);
1221 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1222 __func__, vp, vp->v_writecount);
1224 VM_OBJECT_UNLOCK(object);
1228 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1235 VM_OBJECT_LOCK(object);
1238 * First, recheck the object type to account for the race when
1239 * the vnode is reclaimed.
1241 if (object->type != OBJT_VNODE) {
1242 VM_OBJECT_UNLOCK(object);
1247 * Optimize for the case when writemappings is not going to
1251 if (object->un_pager.vnp.writemappings != inc) {
1252 object->un_pager.vnp.writemappings -= inc;
1253 VM_OBJECT_UNLOCK(object);
1257 vp = object->handle;
1259 VM_OBJECT_UNLOCK(object);
1261 vn_start_write(vp, &mp, V_WAIT);
1262 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1265 * Decrement the object's writemappings, by swapping the start
1266 * and end arguments for vnode_pager_update_writecount(). If
1267 * there was not a race with vnode reclaimation, then the
1268 * vnode's v_writecount is decremented.
1270 vnode_pager_update_writecount(object, end, start);
1274 vn_finished_write(mp);