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 msleep(object, VM_OBJECT_MTX(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 msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0);
216 KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference"));
218 if (object == NULL) {
220 * Add an object of the appropriate size
222 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
224 object->un_pager.vnp.vnp_size = size;
225 object->un_pager.vnp.writemappings = 0;
227 object->handle = handle;
229 if (vp->v_object != NULL) {
231 * Object has been created while we were sleeping
234 VM_OBJECT_LOCK(object);
235 KASSERT(object->ref_count == 1,
236 ("leaked ref %p %d", object, object->ref_count));
237 object->type = OBJT_DEAD;
238 object->ref_count = 0;
239 VM_OBJECT_UNLOCK(object);
240 vm_object_destroy(object);
243 vp->v_object = object;
247 VM_OBJECT_UNLOCK(object);
254 * The object must be locked.
257 vnode_pager_dealloc(object)
265 panic("vnode_pager_dealloc: pager already dealloced");
267 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
268 vm_object_pip_wait(object, "vnpdea");
269 refs = object->ref_count;
271 object->handle = NULL;
272 object->type = OBJT_DEAD;
273 if (object->flags & OBJ_DISCONNECTWNT) {
274 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
277 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
278 if (object->un_pager.vnp.writemappings > 0) {
279 object->un_pager.vnp.writemappings = 0;
280 VOP_ADD_WRITECOUNT(vp, -1);
284 VM_OBJECT_UNLOCK(object);
287 VM_OBJECT_LOCK(object);
291 vnode_pager_haspage(object, pindex, before, after)
297 struct vnode *vp = object->handle;
303 int pagesperblock, blocksperpage;
306 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
308 * If no vp or vp is doomed or marked transparent to VM, we do not
311 if (vp == NULL || vp->v_iflag & VI_DOOMED)
314 * If the offset is beyond end of file we do
317 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
320 bsize = vp->v_mount->mnt_stat.f_iosize;
321 pagesperblock = bsize / PAGE_SIZE;
323 if (pagesperblock > 0) {
324 reqblock = pindex / pagesperblock;
326 blocksperpage = (PAGE_SIZE / bsize);
327 reqblock = pindex * blocksperpage;
329 VM_OBJECT_UNLOCK(object);
330 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
331 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
332 VFS_UNLOCK_GIANT(vfslocked);
333 VM_OBJECT_LOCK(object);
338 if (pagesperblock > 0) {
339 poff = pindex - (reqblock * pagesperblock);
341 *before *= pagesperblock;
346 *after *= pagesperblock;
347 numafter = pagesperblock - (poff + 1);
348 if (IDX_TO_OFF(pindex + numafter) >
349 object->un_pager.vnp.vnp_size) {
351 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
358 *before /= blocksperpage;
362 *after /= blocksperpage;
369 * Lets the VM system know about a change in size for a file.
370 * We adjust our own internal size and flush any cached pages in
371 * the associated object that are affected by the size change.
373 * Note: this routine may be invoked as a result of a pager put
374 * operation (possibly at object termination time), so we must be careful.
377 vnode_pager_setsize(vp, nsize)
383 vm_pindex_t nobjsize;
385 if ((object = vp->v_object) == NULL)
387 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
388 VM_OBJECT_LOCK(object);
389 if (object->type == OBJT_DEAD) {
390 VM_OBJECT_UNLOCK(object);
393 KASSERT(object->type == OBJT_VNODE,
394 ("not vnode-backed object %p", object));
395 if (nsize == object->un_pager.vnp.vnp_size) {
397 * Hasn't changed size
399 VM_OBJECT_UNLOCK(object);
402 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
403 if (nsize < object->un_pager.vnp.vnp_size) {
405 * File has shrunk. Toss any cached pages beyond the new EOF.
407 if (nobjsize < object->size)
408 vm_object_page_remove(object, nobjsize, object->size,
411 * this gets rid of garbage at the end of a page that is now
412 * only partially backed by the vnode.
414 * XXX for some reason (I don't know yet), if we take a
415 * completely invalid page and mark it partially valid
416 * it can screw up NFS reads, so we don't allow the case.
418 if ((nsize & PAGE_MASK) &&
419 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
421 int base = (int)nsize & PAGE_MASK;
422 int size = PAGE_SIZE - base;
425 * Clear out partial-page garbage in case
426 * the page has been mapped.
428 pmap_zero_page_area(m, base, size);
431 * Update the valid bits to reflect the blocks that
432 * have been zeroed. Some of these valid bits may
433 * have already been set.
435 vm_page_set_valid(m, base, size);
438 * Round "base" to the next block boundary so that the
439 * dirty bit for a partially zeroed block is not
442 base = roundup2(base, DEV_BSIZE);
445 * Clear out partial-page dirty bits.
447 * note that we do not clear out the valid
448 * bits. This would prevent bogus_page
449 * replacement from working properly.
451 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
452 } else if ((nsize & PAGE_MASK) &&
453 __predict_false(object->cache != NULL)) {
454 vm_page_cache_free(object, OFF_TO_IDX(nsize),
458 object->un_pager.vnp.vnp_size = nsize;
459 object->size = nobjsize;
460 VM_OBJECT_UNLOCK(object);
464 * calculate the linear (byte) disk address of specified virtual
468 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
479 if (vp->v_iflag & VI_DOOMED)
482 bsize = vp->v_mount->mnt_stat.f_iosize;
483 vblock = address / bsize;
484 voffset = address % bsize;
486 err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
488 if (*rtaddress != -1)
489 *rtaddress += voffset / DEV_BSIZE;
492 *run *= bsize/PAGE_SIZE;
493 *run -= voffset/PAGE_SIZE;
501 * small block filesystem vnode pager input
504 vnode_pager_input_smlfs(object, m)
519 if (vp->v_iflag & VI_DOOMED)
522 bsize = vp->v_mount->mnt_stat.f_iosize;
524 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
526 sf = sf_buf_alloc(m, 0);
528 for (i = 0; i < PAGE_SIZE / bsize; i++) {
529 vm_ooffset_t address;
531 bits = vm_page_bits(i * bsize, bsize);
535 address = IDX_TO_OFF(m->pindex) + i * bsize;
536 if (address >= object->un_pager.vnp.vnp_size) {
539 error = vnode_pager_addr(vp, address, &fileaddr, NULL);
543 if (fileaddr != -1) {
544 bp = getpbuf(&vnode_pbuf_freecnt);
546 /* build a minimal buffer header */
547 bp->b_iocmd = BIO_READ;
548 bp->b_iodone = bdone;
549 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
550 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
551 bp->b_rcred = crhold(curthread->td_ucred);
552 bp->b_wcred = crhold(curthread->td_ucred);
553 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
554 bp->b_blkno = fileaddr;
557 bp->b_bcount = bsize;
558 bp->b_bufsize = bsize;
559 bp->b_runningbufspace = bp->b_bufsize;
560 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
563 bp->b_iooffset = dbtob(bp->b_blkno);
566 bwait(bp, PVM, "vnsrd");
568 if ((bp->b_ioflags & BIO_ERROR) != 0)
572 * free the buffer header back to the swap buffer pool
576 relpbuf(bp, &vnode_pbuf_freecnt);
580 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
581 KASSERT((m->dirty & bits) == 0,
582 ("vnode_pager_input_smlfs: page %p is dirty", m));
583 VM_OBJECT_LOCK(object);
585 VM_OBJECT_UNLOCK(object);
589 return VM_PAGER_ERROR;
595 * old style vnode pager input routine
598 vnode_pager_input_old(object, m)
609 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
613 * Return failure if beyond current EOF
615 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
619 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
620 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
622 VM_OBJECT_UNLOCK(object);
625 * Allocate a kernel virtual address and initialize so that
626 * we can use VOP_READ/WRITE routines.
628 sf = sf_buf_alloc(m, 0);
630 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
632 auio.uio_iov = &aiov;
634 auio.uio_offset = IDX_TO_OFF(m->pindex);
635 auio.uio_segflg = UIO_SYSSPACE;
636 auio.uio_rw = UIO_READ;
637 auio.uio_resid = size;
638 auio.uio_td = curthread;
640 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
642 int count = size - auio.uio_resid;
646 else if (count != PAGE_SIZE)
647 bzero((caddr_t)sf_buf_kva(sf) + count,
652 VM_OBJECT_LOCK(object);
654 KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
656 m->valid = VM_PAGE_BITS_ALL;
657 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
661 * generic vnode pager input routine
665 * Local media VFS's that do not implement their own VOP_GETPAGES
666 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
667 * to implement the previous behaviour.
669 * All other FS's should use the bypass to get to the local media
670 * backing vp's VOP_GETPAGES.
673 vnode_pager_getpages(object, m, count, reqpage)
681 int bytes = count * PAGE_SIZE;
685 VM_OBJECT_UNLOCK(object);
686 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
687 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
688 KASSERT(rtval != EOPNOTSUPP,
689 ("vnode_pager: FS getpages not implemented\n"));
690 VFS_UNLOCK_GIANT(vfslocked);
691 VM_OBJECT_LOCK(object);
696 * This is now called from local media FS's to operate against their
697 * own vnodes if they fail to implement VOP_GETPAGES.
700 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
708 off_t foff, tfoff, nextoff;
709 int i, j, size, bsize, first;
710 daddr_t firstaddr, reqblock;
719 object = vp->v_object;
720 count = bytecount / PAGE_SIZE;
722 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
723 ("vnode_pager_generic_getpages does not support devices"));
724 if (vp->v_iflag & VI_DOOMED)
727 bsize = vp->v_mount->mnt_stat.f_iosize;
729 /* get the UNDERLYING device for the file with VOP_BMAP() */
732 * originally, we did not check for an error return value -- assuming
733 * an fs always has a bmap entry point -- that assumption is wrong!!!
735 foff = IDX_TO_OFF(m[reqpage]->pindex);
738 * if we can't bmap, use old VOP code
740 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
741 if (error == EOPNOTSUPP) {
742 VM_OBJECT_LOCK(object);
744 for (i = 0; i < count; i++)
748 vm_page_unlock(m[i]);
750 PCPU_INC(cnt.v_vnodein);
751 PCPU_INC(cnt.v_vnodepgsin);
752 error = vnode_pager_input_old(object, m[reqpage]);
753 VM_OBJECT_UNLOCK(object);
755 } else if (error != 0) {
756 VM_OBJECT_LOCK(object);
757 for (i = 0; i < count; i++)
761 vm_page_unlock(m[i]);
763 VM_OBJECT_UNLOCK(object);
764 return (VM_PAGER_ERROR);
767 * if the blocksize is smaller than a page size, then use
768 * special small filesystem code. NFS sometimes has a small
769 * blocksize, but it can handle large reads itself.
771 } else if ((PAGE_SIZE / bsize) > 1 &&
772 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
773 VM_OBJECT_LOCK(object);
774 for (i = 0; i < count; i++)
778 vm_page_unlock(m[i]);
780 VM_OBJECT_UNLOCK(object);
781 PCPU_INC(cnt.v_vnodein);
782 PCPU_INC(cnt.v_vnodepgsin);
783 return vnode_pager_input_smlfs(object, m[reqpage]);
787 * If we have a completely valid page available to us, we can
788 * clean up and return. Otherwise we have to re-read the
791 VM_OBJECT_LOCK(object);
792 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
793 for (i = 0; i < count; i++)
797 vm_page_unlock(m[i]);
799 VM_OBJECT_UNLOCK(object);
801 } else if (reqblock == -1) {
802 pmap_zero_page(m[reqpage]);
803 KASSERT(m[reqpage]->dirty == 0,
804 ("vnode_pager_generic_getpages: page %p is dirty", m));
805 m[reqpage]->valid = VM_PAGE_BITS_ALL;
806 for (i = 0; i < count; i++)
810 vm_page_unlock(m[i]);
812 VM_OBJECT_UNLOCK(object);
813 return (VM_PAGER_OK);
815 m[reqpage]->valid = 0;
816 VM_OBJECT_UNLOCK(object);
819 * here on direct device I/O
824 * calculate the run that includes the required page
826 for (first = 0, i = 0; i < count; i = runend) {
827 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
829 VM_OBJECT_LOCK(object);
830 for (; i < count; i++)
834 vm_page_unlock(m[i]);
836 VM_OBJECT_UNLOCK(object);
837 return (VM_PAGER_ERROR);
839 if (firstaddr == -1) {
840 VM_OBJECT_LOCK(object);
841 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
842 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
843 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
846 (object->un_pager.vnp.vnp_size >> 32),
847 (uintmax_t)object->un_pager.vnp.vnp_size);
851 vm_page_unlock(m[i]);
852 VM_OBJECT_UNLOCK(object);
858 if (runend <= reqpage) {
859 VM_OBJECT_LOCK(object);
860 for (j = i; j < runend; j++) {
863 vm_page_unlock(m[j]);
865 VM_OBJECT_UNLOCK(object);
867 if (runpg < (count - first)) {
868 VM_OBJECT_LOCK(object);
869 for (i = first + runpg; i < count; i++) {
872 vm_page_unlock(m[i]);
874 VM_OBJECT_UNLOCK(object);
875 count = first + runpg;
883 * the first and last page have been calculated now, move input pages
884 * to be zero based...
893 * calculate the file virtual address for the transfer
895 foff = IDX_TO_OFF(m[0]->pindex);
898 * calculate the size of the transfer
900 size = count * PAGE_SIZE;
901 KASSERT(count > 0, ("zero count"));
902 if ((foff + size) > object->un_pager.vnp.vnp_size)
903 size = object->un_pager.vnp.vnp_size - foff;
904 KASSERT(size > 0, ("zero size"));
907 * round up physical size for real devices.
910 int secmask = bo->bo_bsize - 1;
911 KASSERT(secmask < PAGE_SIZE && secmask > 0,
912 ("vnode_pager_generic_getpages: sector size %d too large",
914 size = (size + secmask) & ~secmask;
917 bp = getpbuf(&vnode_pbuf_freecnt);
918 kva = (vm_offset_t)bp->b_data;
921 * and map the pages to be read into the kva, if the filesystem
922 * requires mapped buffers.
925 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 &&
926 unmapped_buf_allowed) {
927 bp->b_data = unmapped_buf;
928 bp->b_kvabase = unmapped_buf;
930 bp->b_flags |= B_UNMAPPED;
931 bp->b_npages = count;
932 for (i = 0; i < count; i++)
933 bp->b_pages[i] = m[i];
935 pmap_qenter(kva, m, count);
937 /* build a minimal buffer header */
938 bp->b_iocmd = BIO_READ;
939 bp->b_iodone = bdone;
940 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
941 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
942 bp->b_rcred = crhold(curthread->td_ucred);
943 bp->b_wcred = crhold(curthread->td_ucred);
944 bp->b_blkno = firstaddr;
948 bp->b_bufsize = size;
949 bp->b_runningbufspace = bp->b_bufsize;
950 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
952 PCPU_INC(cnt.v_vnodein);
953 PCPU_ADD(cnt.v_vnodepgsin, count);
956 bp->b_iooffset = dbtob(bp->b_blkno);
959 bwait(bp, PVM, "vnread");
961 if ((bp->b_ioflags & BIO_ERROR) != 0)
964 if (error == 0 && size != count * PAGE_SIZE) {
965 if ((bp->b_flags & B_UNMAPPED) != 0) {
966 bp->b_flags &= ~B_UNMAPPED;
967 pmap_qenter(kva, m, count);
969 bzero((caddr_t)kva + size, PAGE_SIZE * count - size);
971 if ((bp->b_flags & B_UNMAPPED) == 0)
972 pmap_qremove(kva, count);
973 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) {
974 bp->b_data = (caddr_t)kva;
975 bp->b_kvabase = (caddr_t)kva;
976 bp->b_flags &= ~B_UNMAPPED;
977 for (i = 0; i < count; i++)
978 bp->b_pages[i] = NULL;
982 * free the buffer header back to the swap buffer pool
986 relpbuf(bp, &vnode_pbuf_freecnt);
988 VM_OBJECT_LOCK(object);
989 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
992 nextoff = tfoff + PAGE_SIZE;
995 if (nextoff <= object->un_pager.vnp.vnp_size) {
997 * Read filled up entire page.
999 mt->valid = VM_PAGE_BITS_ALL;
1000 KASSERT(mt->dirty == 0,
1001 ("vnode_pager_generic_getpages: page %p is dirty",
1003 KASSERT(!pmap_page_is_mapped(mt),
1004 ("vnode_pager_generic_getpages: page %p is mapped",
1008 * Read did not fill up entire page.
1010 * Currently we do not set the entire page valid,
1011 * we just try to clear the piece that we couldn't
1014 vm_page_set_valid(mt, 0,
1015 object->un_pager.vnp.vnp_size - tfoff);
1016 KASSERT((mt->dirty & vm_page_bits(0,
1017 object->un_pager.vnp.vnp_size - tfoff)) == 0,
1018 ("vnode_pager_generic_getpages: page %p is dirty",
1023 vm_page_readahead_finish(mt);
1025 VM_OBJECT_UNLOCK(object);
1027 printf("vnode_pager_getpages: I/O read error\n");
1029 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
1033 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
1034 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
1035 * vnode_pager_generic_putpages() to implement the previous behaviour.
1037 * All other FS's should use the bypass to get to the local media
1038 * backing vp's VOP_PUTPAGES.
1041 vnode_pager_putpages(object, m, count, sync, rtvals)
1050 int bytes = count * PAGE_SIZE;
1053 * Force synchronous operation if we are extremely low on memory
1054 * to prevent a low-memory deadlock. VOP operations often need to
1055 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1056 * operation ). The swapper handles the case by limiting the amount
1057 * of asynchronous I/O, but that sort of solution doesn't scale well
1058 * for the vnode pager without a lot of work.
1060 * Also, the backing vnode's iodone routine may not wake the pageout
1061 * daemon up. This should be probably be addressed XXX.
1064 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
1068 * Call device-specific putpages function
1070 vp = object->handle;
1071 VM_OBJECT_UNLOCK(object);
1072 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
1073 KASSERT(rtval != EOPNOTSUPP,
1074 ("vnode_pager: stale FS putpages\n"));
1075 VM_OBJECT_LOCK(object);
1080 * This is now called from local media FS's to operate against their
1081 * own vnodes if they fail to implement VOP_PUTPAGES.
1083 * This is typically called indirectly via the pageout daemon and
1084 * clustering has already typically occured, so in general we ask the
1085 * underlying filesystem to write the data out asynchronously rather
1089 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1090 int flags, int *rtvals)
1097 int maxsize, ncount;
1098 vm_ooffset_t poffset;
1104 static struct timeval lastfail;
1107 object = vp->v_object;
1108 count = bytecount / PAGE_SIZE;
1110 for (i = 0; i < count; i++)
1111 rtvals[i] = VM_PAGER_ERROR;
1113 if ((int64_t)ma[0]->pindex < 0) {
1114 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1115 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1116 rtvals[0] = VM_PAGER_BAD;
1117 return VM_PAGER_BAD;
1120 maxsize = count * PAGE_SIZE;
1123 poffset = IDX_TO_OFF(ma[0]->pindex);
1126 * If the page-aligned write is larger then the actual file we
1127 * have to invalidate pages occuring beyond the file EOF. However,
1128 * there is an edge case where a file may not be page-aligned where
1129 * the last page is partially invalid. In this case the filesystem
1130 * may not properly clear the dirty bits for the entire page (which
1131 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1132 * With the page locked we are free to fix-up the dirty bits here.
1134 * We do not under any circumstances truncate the valid bits, as
1135 * this will screw up bogus page replacement.
1137 VM_OBJECT_LOCK(object);
1138 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1139 if (object->un_pager.vnp.vnp_size > poffset) {
1142 maxsize = object->un_pager.vnp.vnp_size - poffset;
1143 ncount = btoc(maxsize);
1144 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1146 * If the object is locked and the following
1147 * conditions hold, then the page's dirty
1148 * field cannot be concurrently changed by a
1152 KASSERT(m->busy > 0,
1153 ("vnode_pager_generic_putpages: page %p is not busy", m));
1154 KASSERT(!pmap_page_is_write_mapped(m),
1155 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1156 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1163 if (ncount < count) {
1164 for (i = ncount; i < count; i++) {
1165 rtvals[i] = VM_PAGER_BAD;
1169 VM_OBJECT_UNLOCK(object);
1172 * pageouts are already clustered, use IO_ASYNC t o force a bawrite()
1173 * rather then a bdwrite() to prevent paging I/O from saturating
1174 * the buffer cache. Dummy-up the sequential heuristic to cause
1175 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1176 * the system decides how to cluster.
1179 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1181 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1182 ioflags |= IO_ASYNC;
1183 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1184 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1186 aiov.iov_base = (caddr_t) 0;
1187 aiov.iov_len = maxsize;
1188 auio.uio_iov = &aiov;
1189 auio.uio_iovcnt = 1;
1190 auio.uio_offset = poffset;
1191 auio.uio_segflg = UIO_NOCOPY;
1192 auio.uio_rw = UIO_WRITE;
1193 auio.uio_resid = maxsize;
1194 auio.uio_td = (struct thread *) 0;
1195 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1196 PCPU_INC(cnt.v_vnodeout);
1197 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1200 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1201 printf("vnode_pager_putpages: I/O error %d\n", error);
1203 if (auio.uio_resid) {
1204 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1205 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1206 auio.uio_resid, (u_long)ma[0]->pindex);
1208 for (i = 0; i < ncount; i++) {
1209 rtvals[i] = VM_PAGER_OK;
1215 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1222 obj = ma[0]->object;
1223 VM_OBJECT_LOCK(obj);
1224 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1225 if (pos < trunc_page(written)) {
1226 rtvals[i] = VM_PAGER_OK;
1227 vm_page_undirty(ma[i]);
1229 /* Partially written page. */
1230 rtvals[i] = VM_PAGER_AGAIN;
1231 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1234 VM_OBJECT_UNLOCK(obj);
1238 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1242 vm_ooffset_t old_wm;
1244 VM_OBJECT_LOCK(object);
1245 if (object->type != OBJT_VNODE) {
1246 VM_OBJECT_UNLOCK(object);
1249 old_wm = object->un_pager.vnp.writemappings;
1250 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1251 vp = object->handle;
1252 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1253 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1254 VOP_ADD_WRITECOUNT(vp, 1);
1255 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1256 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1257 VOP_ADD_WRITECOUNT(vp, -1);
1259 VM_OBJECT_UNLOCK(object);
1263 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1271 VM_OBJECT_LOCK(object);
1274 * First, recheck the object type to account for the race when
1275 * the vnode is reclaimed.
1277 if (object->type != OBJT_VNODE) {
1278 VM_OBJECT_UNLOCK(object);
1283 * Optimize for the case when writemappings is not going to
1287 if (object->un_pager.vnp.writemappings != inc) {
1288 object->un_pager.vnp.writemappings -= inc;
1289 VM_OBJECT_UNLOCK(object);
1293 vp = object->handle;
1295 VM_OBJECT_UNLOCK(object);
1296 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1298 vn_start_write(vp, &mp, V_WAIT);
1299 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1302 * Decrement the object's writemappings, by swapping the start
1303 * and end arguments for vnode_pager_update_writecount(). If
1304 * there was not a race with vnode reclaimation, then the
1305 * vnode's v_writecount is decremented.
1307 vnode_pager_update_writecount(object, end, start);
1311 vn_finished_write(mp);
1312 VFS_UNLOCK_GIANT(vfslocked);