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 * don't double-terminate the object
163 if ((obj->flags & OBJ_DEAD) == 0)
164 vm_object_terminate(obj);
166 VM_OBJECT_WUNLOCK(obj);
169 * Woe to the process that tries to page now :-).
171 vm_pager_deallocate(obj);
172 VM_OBJECT_WUNLOCK(obj);
179 * Allocate (or lookup) pager for a vnode.
180 * Handle is a vnode pointer.
185 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
186 vm_ooffset_t offset, struct ucred *cred)
192 * Pageout to vnode, no can do yet.
197 vp = (struct vnode *) handle;
200 * If the object is being terminated, wait for it to
204 while ((object = vp->v_object) != NULL) {
205 VM_OBJECT_WLOCK(object);
206 if ((object->flags & OBJ_DEAD) == 0)
208 vm_object_set_flag(object, OBJ_DISCONNECTWNT);
209 VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0);
212 KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference"));
214 if (object == NULL) {
216 * Add an object of the appropriate size
218 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
220 object->un_pager.vnp.vnp_size = size;
221 object->un_pager.vnp.writemappings = 0;
223 object->handle = handle;
225 if (vp->v_object != NULL) {
227 * Object has been created while we were sleeping
230 vm_object_destroy(object);
233 vp->v_object = object;
237 VM_OBJECT_WUNLOCK(object);
244 * The object must be locked.
247 vnode_pager_dealloc(object)
255 panic("vnode_pager_dealloc: pager already dealloced");
257 VM_OBJECT_ASSERT_WLOCKED(object);
258 vm_object_pip_wait(object, "vnpdea");
259 refs = object->ref_count;
261 object->handle = NULL;
262 object->type = OBJT_DEAD;
263 if (object->flags & OBJ_DISCONNECTWNT) {
264 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
267 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
268 if (object->un_pager.vnp.writemappings > 0) {
269 object->un_pager.vnp.writemappings = 0;
270 VOP_ADD_WRITECOUNT(vp, -1);
271 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
272 __func__, vp, vp->v_writecount);
276 VM_OBJECT_WUNLOCK(object);
279 VM_OBJECT_WLOCK(object);
283 vnode_pager_haspage(object, pindex, before, after)
289 struct vnode *vp = object->handle;
295 int pagesperblock, blocksperpage;
297 VM_OBJECT_ASSERT_WLOCKED(object);
299 * If no vp or vp is doomed or marked transparent to VM, we do not
302 if (vp == NULL || vp->v_iflag & VI_DOOMED)
305 * If the offset is beyond end of file we do
308 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
311 bsize = vp->v_mount->mnt_stat.f_iosize;
312 pagesperblock = bsize / PAGE_SIZE;
314 if (pagesperblock > 0) {
315 reqblock = pindex / pagesperblock;
317 blocksperpage = (PAGE_SIZE / bsize);
318 reqblock = pindex * blocksperpage;
320 VM_OBJECT_WUNLOCK(object);
321 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
322 VM_OBJECT_WLOCK(object);
327 if (pagesperblock > 0) {
328 poff = pindex - (reqblock * pagesperblock);
330 *before *= pagesperblock;
335 *after *= pagesperblock;
336 numafter = pagesperblock - (poff + 1);
337 if (IDX_TO_OFF(pindex + numafter) >
338 object->un_pager.vnp.vnp_size) {
340 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
347 *before /= blocksperpage;
351 *after /= blocksperpage;
358 * Lets the VM system know about a change in size for a file.
359 * We adjust our own internal size and flush any cached pages in
360 * the associated object that are affected by the size change.
362 * Note: this routine may be invoked as a result of a pager put
363 * operation (possibly at object termination time), so we must be careful.
366 vnode_pager_setsize(vp, nsize)
372 vm_pindex_t nobjsize;
374 if ((object = vp->v_object) == NULL)
376 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
377 VM_OBJECT_WLOCK(object);
378 if (object->type == OBJT_DEAD) {
379 VM_OBJECT_WUNLOCK(object);
382 KASSERT(object->type == OBJT_VNODE,
383 ("not vnode-backed object %p", 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;
705 object = vp->v_object;
706 count = bytecount / PAGE_SIZE;
708 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
709 ("vnode_pager_generic_getpages does not support devices"));
710 if (vp->v_iflag & VI_DOOMED)
713 bsize = vp->v_mount->mnt_stat.f_iosize;
715 /* get the UNDERLYING device for the file with VOP_BMAP() */
718 * originally, we did not check for an error return value -- assuming
719 * an fs always has a bmap entry point -- that assumption is wrong!!!
721 foff = IDX_TO_OFF(m[reqpage]->pindex);
724 * if we can't bmap, use old VOP code
726 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
727 if (error == EOPNOTSUPP) {
728 VM_OBJECT_WLOCK(object);
730 for (i = 0; i < count; i++)
734 vm_page_unlock(m[i]);
736 PCPU_INC(cnt.v_vnodein);
737 PCPU_INC(cnt.v_vnodepgsin);
738 error = vnode_pager_input_old(object, m[reqpage]);
739 VM_OBJECT_WUNLOCK(object);
741 } else if (error != 0) {
742 VM_OBJECT_WLOCK(object);
743 for (i = 0; i < count; i++)
747 vm_page_unlock(m[i]);
749 VM_OBJECT_WUNLOCK(object);
750 return (VM_PAGER_ERROR);
753 * if the blocksize is smaller than a page size, then use
754 * special small filesystem code. NFS sometimes has a small
755 * blocksize, but it can handle large reads itself.
757 } else if ((PAGE_SIZE / bsize) > 1 &&
758 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
759 VM_OBJECT_WLOCK(object);
760 for (i = 0; i < count; i++)
764 vm_page_unlock(m[i]);
766 VM_OBJECT_WUNLOCK(object);
767 PCPU_INC(cnt.v_vnodein);
768 PCPU_INC(cnt.v_vnodepgsin);
769 return vnode_pager_input_smlfs(object, m[reqpage]);
773 * If we have a completely valid page available to us, we can
774 * clean up and return. Otherwise we have to re-read the
777 VM_OBJECT_WLOCK(object);
778 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
779 for (i = 0; i < count; i++)
783 vm_page_unlock(m[i]);
785 VM_OBJECT_WUNLOCK(object);
787 } else if (reqblock == -1) {
788 pmap_zero_page(m[reqpage]);
789 KASSERT(m[reqpage]->dirty == 0,
790 ("vnode_pager_generic_getpages: page %p is dirty", m));
791 m[reqpage]->valid = VM_PAGE_BITS_ALL;
792 for (i = 0; i < count; i++)
796 vm_page_unlock(m[i]);
798 VM_OBJECT_WUNLOCK(object);
799 return (VM_PAGER_OK);
801 m[reqpage]->valid = 0;
802 VM_OBJECT_WUNLOCK(object);
805 * here on direct device I/O
810 * calculate the run that includes the required page
812 for (first = 0, i = 0; i < count; i = runend) {
813 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
815 VM_OBJECT_WLOCK(object);
816 for (; i < count; i++)
820 vm_page_unlock(m[i]);
822 VM_OBJECT_WUNLOCK(object);
823 return (VM_PAGER_ERROR);
825 if (firstaddr == -1) {
826 VM_OBJECT_WLOCK(object);
827 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
828 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
829 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
832 (object->un_pager.vnp.vnp_size >> 32),
833 (uintmax_t)object->un_pager.vnp.vnp_size);
837 vm_page_unlock(m[i]);
838 VM_OBJECT_WUNLOCK(object);
844 if (runend <= reqpage) {
845 VM_OBJECT_WLOCK(object);
846 for (j = i; j < runend; j++) {
849 vm_page_unlock(m[j]);
851 VM_OBJECT_WUNLOCK(object);
853 if (runpg < (count - first)) {
854 VM_OBJECT_WLOCK(object);
855 for (i = first + runpg; i < count; i++) {
858 vm_page_unlock(m[i]);
860 VM_OBJECT_WUNLOCK(object);
861 count = first + runpg;
869 * the first and last page have been calculated now, move input pages
870 * to be zero based...
879 * calculate the file virtual address for the transfer
881 foff = IDX_TO_OFF(m[0]->pindex);
884 * calculate the size of the transfer
886 size = count * PAGE_SIZE;
887 KASSERT(count > 0, ("zero count"));
888 if ((foff + size) > object->un_pager.vnp.vnp_size)
889 size = object->un_pager.vnp.vnp_size - foff;
890 KASSERT(size > 0, ("zero size"));
893 * round up physical size for real devices.
896 int secmask = bo->bo_bsize - 1;
897 KASSERT(secmask < PAGE_SIZE && secmask > 0,
898 ("vnode_pager_generic_getpages: sector size %d too large",
900 size = (size + secmask) & ~secmask;
903 bp = getpbuf(&vnode_pbuf_freecnt);
904 kva = (vm_offset_t)bp->b_data;
907 * and map the pages to be read into the kva, if the filesystem
908 * requires mapped buffers.
911 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 &&
912 unmapped_buf_allowed) {
913 bp->b_data = unmapped_buf;
914 bp->b_kvabase = unmapped_buf;
916 bp->b_flags |= B_UNMAPPED;
917 bp->b_npages = count;
918 for (i = 0; i < count; i++)
919 bp->b_pages[i] = m[i];
921 pmap_qenter(kva, m, count);
923 /* build a minimal buffer header */
924 bp->b_iocmd = BIO_READ;
925 bp->b_iodone = bdone;
926 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
927 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
928 bp->b_rcred = crhold(curthread->td_ucred);
929 bp->b_wcred = crhold(curthread->td_ucred);
930 bp->b_blkno = firstaddr;
934 bp->b_bufsize = size;
935 bp->b_runningbufspace = bp->b_bufsize;
936 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
938 PCPU_INC(cnt.v_vnodein);
939 PCPU_ADD(cnt.v_vnodepgsin, count);
942 bp->b_iooffset = dbtob(bp->b_blkno);
945 bwait(bp, PVM, "vnread");
947 if ((bp->b_ioflags & BIO_ERROR) != 0)
950 if (error == 0 && size != count * PAGE_SIZE) {
951 if ((bp->b_flags & B_UNMAPPED) != 0) {
952 bp->b_flags &= ~B_UNMAPPED;
953 pmap_qenter(kva, m, count);
955 bzero((caddr_t)kva + size, PAGE_SIZE * count - size);
957 if ((bp->b_flags & B_UNMAPPED) == 0)
958 pmap_qremove(kva, count);
959 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) {
960 bp->b_data = (caddr_t)kva;
961 bp->b_kvabase = (caddr_t)kva;
962 bp->b_flags &= ~B_UNMAPPED;
963 for (i = 0; i < count; i++)
964 bp->b_pages[i] = NULL;
968 * free the buffer header back to the swap buffer pool
972 relpbuf(bp, &vnode_pbuf_freecnt);
974 VM_OBJECT_WLOCK(object);
975 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
978 nextoff = tfoff + PAGE_SIZE;
981 if (nextoff <= object->un_pager.vnp.vnp_size) {
983 * Read filled up entire page.
985 mt->valid = VM_PAGE_BITS_ALL;
986 KASSERT(mt->dirty == 0,
987 ("vnode_pager_generic_getpages: page %p is dirty",
989 KASSERT(!pmap_page_is_mapped(mt),
990 ("vnode_pager_generic_getpages: page %p is mapped",
994 * Read did not fill up entire page.
996 * Currently we do not set the entire page valid,
997 * we just try to clear the piece that we couldn't
1000 vm_page_set_valid_range(mt, 0,
1001 object->un_pager.vnp.vnp_size - tfoff);
1002 KASSERT((mt->dirty & vm_page_bits(0,
1003 object->un_pager.vnp.vnp_size - tfoff)) == 0,
1004 ("vnode_pager_generic_getpages: page %p is dirty",
1009 vm_page_readahead_finish(mt);
1011 VM_OBJECT_WUNLOCK(object);
1013 printf("vnode_pager_getpages: I/O read error\n");
1015 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
1019 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
1020 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
1021 * vnode_pager_generic_putpages() to implement the previous behaviour.
1023 * All other FS's should use the bypass to get to the local media
1024 * backing vp's VOP_PUTPAGES.
1027 vnode_pager_putpages(object, m, count, sync, rtvals)
1036 int bytes = count * PAGE_SIZE;
1039 * Force synchronous operation if we are extremely low on memory
1040 * to prevent a low-memory deadlock. VOP operations often need to
1041 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1042 * operation ). The swapper handles the case by limiting the amount
1043 * of asynchronous I/O, but that sort of solution doesn't scale well
1044 * for the vnode pager without a lot of work.
1046 * Also, the backing vnode's iodone routine may not wake the pageout
1047 * daemon up. This should be probably be addressed XXX.
1050 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
1054 * Call device-specific putpages function
1056 vp = object->handle;
1057 VM_OBJECT_WUNLOCK(object);
1058 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
1059 KASSERT(rtval != EOPNOTSUPP,
1060 ("vnode_pager: stale FS putpages\n"));
1061 VM_OBJECT_WLOCK(object);
1066 * This is now called from local media FS's to operate against their
1067 * own vnodes if they fail to implement VOP_PUTPAGES.
1069 * This is typically called indirectly via the pageout daemon and
1070 * clustering has already typically occured, so in general we ask the
1071 * underlying filesystem to write the data out asynchronously rather
1075 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1076 int flags, int *rtvals)
1083 int maxsize, ncount;
1084 vm_ooffset_t poffset;
1090 static struct timeval lastfail;
1093 object = vp->v_object;
1094 count = bytecount / PAGE_SIZE;
1096 for (i = 0; i < count; i++)
1097 rtvals[i] = VM_PAGER_ERROR;
1099 if ((int64_t)ma[0]->pindex < 0) {
1100 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1101 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1102 rtvals[0] = VM_PAGER_BAD;
1103 return VM_PAGER_BAD;
1106 maxsize = count * PAGE_SIZE;
1109 poffset = IDX_TO_OFF(ma[0]->pindex);
1112 * If the page-aligned write is larger then the actual file we
1113 * have to invalidate pages occuring beyond the file EOF. However,
1114 * there is an edge case where a file may not be page-aligned where
1115 * the last page is partially invalid. In this case the filesystem
1116 * may not properly clear the dirty bits for the entire page (which
1117 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1118 * With the page locked we are free to fix-up the dirty bits here.
1120 * We do not under any circumstances truncate the valid bits, as
1121 * this will screw up bogus page replacement.
1123 VM_OBJECT_WLOCK(object);
1124 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1125 if (object->un_pager.vnp.vnp_size > poffset) {
1128 maxsize = object->un_pager.vnp.vnp_size - poffset;
1129 ncount = btoc(maxsize);
1130 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1132 * If the object is locked and the following
1133 * conditions hold, then the page's dirty
1134 * field cannot be concurrently changed by a
1138 vm_page_assert_sbusied(m);
1139 KASSERT(!pmap_page_is_write_mapped(m),
1140 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1141 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1148 if (ncount < count) {
1149 for (i = ncount; i < count; i++) {
1150 rtvals[i] = VM_PAGER_BAD;
1154 VM_OBJECT_WUNLOCK(object);
1157 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
1158 * rather then a bdwrite() to prevent paging I/O from saturating
1159 * the buffer cache. Dummy-up the sequential heuristic to cause
1160 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1161 * the system decides how to cluster.
1164 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1166 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1167 ioflags |= IO_ASYNC;
1168 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1169 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1171 aiov.iov_base = (caddr_t) 0;
1172 aiov.iov_len = maxsize;
1173 auio.uio_iov = &aiov;
1174 auio.uio_iovcnt = 1;
1175 auio.uio_offset = poffset;
1176 auio.uio_segflg = UIO_NOCOPY;
1177 auio.uio_rw = UIO_WRITE;
1178 auio.uio_resid = maxsize;
1179 auio.uio_td = (struct thread *) 0;
1180 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1181 PCPU_INC(cnt.v_vnodeout);
1182 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1185 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1186 printf("vnode_pager_putpages: I/O error %d\n", error);
1188 if (auio.uio_resid) {
1189 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1190 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1191 auio.uio_resid, (u_long)ma[0]->pindex);
1193 for (i = 0; i < ncount; i++) {
1194 rtvals[i] = VM_PAGER_OK;
1200 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1207 obj = ma[0]->object;
1208 VM_OBJECT_WLOCK(obj);
1209 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1210 if (pos < trunc_page(written)) {
1211 rtvals[i] = VM_PAGER_OK;
1212 vm_page_undirty(ma[i]);
1214 /* Partially written page. */
1215 rtvals[i] = VM_PAGER_AGAIN;
1216 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1219 VM_OBJECT_WUNLOCK(obj);
1223 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1227 vm_ooffset_t old_wm;
1229 VM_OBJECT_WLOCK(object);
1230 if (object->type != OBJT_VNODE) {
1231 VM_OBJECT_WUNLOCK(object);
1234 old_wm = object->un_pager.vnp.writemappings;
1235 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1236 vp = object->handle;
1237 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1238 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1239 VOP_ADD_WRITECOUNT(vp, 1);
1240 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1241 __func__, vp, vp->v_writecount);
1242 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1243 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1244 VOP_ADD_WRITECOUNT(vp, -1);
1245 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1246 __func__, vp, vp->v_writecount);
1248 VM_OBJECT_WUNLOCK(object);
1252 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1259 VM_OBJECT_WLOCK(object);
1262 * First, recheck the object type to account for the race when
1263 * the vnode is reclaimed.
1265 if (object->type != OBJT_VNODE) {
1266 VM_OBJECT_WUNLOCK(object);
1271 * Optimize for the case when writemappings is not going to
1275 if (object->un_pager.vnp.writemappings != inc) {
1276 object->un_pager.vnp.writemappings -= inc;
1277 VM_OBJECT_WUNLOCK(object);
1281 vp = object->handle;
1283 VM_OBJECT_WUNLOCK(object);
1285 vn_start_write(vp, &mp, V_WAIT);
1286 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1289 * Decrement the object's writemappings, by swapping the start
1290 * and end arguments for vnode_pager_update_writecount(). If
1291 * there was not a race with vnode reclaimation, then the
1292 * vnode's v_writecount is decremented.
1294 vnode_pager_update_writecount(object, end, start);
1298 vn_finished_write(mp);