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, int, 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(vm_object_t object)
254 panic("vnode_pager_dealloc: pager already dealloced");
256 VM_OBJECT_ASSERT_WLOCKED(object);
257 vm_object_pip_wait(object, "vnpdea");
258 refs = object->ref_count;
260 object->handle = NULL;
261 object->type = OBJT_DEAD;
262 if (object->flags & OBJ_DISCONNECTWNT) {
263 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
266 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
267 if (object->un_pager.vnp.writemappings > 0) {
268 object->un_pager.vnp.writemappings = 0;
269 VOP_ADD_WRITECOUNT(vp, -1);
270 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
271 __func__, vp, vp->v_writecount);
275 VM_OBJECT_WUNLOCK(object);
278 VM_OBJECT_WLOCK(object);
282 vnode_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
285 struct vnode *vp = object->handle;
291 int pagesperblock, blocksperpage;
293 VM_OBJECT_ASSERT_WLOCKED(object);
295 * If no vp or vp is doomed or marked transparent to VM, we do not
298 if (vp == NULL || vp->v_iflag & VI_DOOMED)
301 * If the offset is beyond end of file we do
304 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
307 bsize = vp->v_mount->mnt_stat.f_iosize;
308 pagesperblock = bsize / PAGE_SIZE;
310 if (pagesperblock > 0) {
311 reqblock = pindex / pagesperblock;
313 blocksperpage = (PAGE_SIZE / bsize);
314 reqblock = pindex * blocksperpage;
316 VM_OBJECT_WUNLOCK(object);
317 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
318 VM_OBJECT_WLOCK(object);
323 if (pagesperblock > 0) {
324 poff = pindex - (reqblock * pagesperblock);
326 *before *= pagesperblock;
331 *after *= pagesperblock;
332 numafter = pagesperblock - (poff + 1);
333 if (IDX_TO_OFF(pindex + numafter) >
334 object->un_pager.vnp.vnp_size) {
336 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
343 *before /= blocksperpage;
347 *after /= blocksperpage;
354 * Lets the VM system know about a change in size for a file.
355 * We adjust our own internal size and flush any cached pages in
356 * the associated object that are affected by the size change.
358 * Note: this routine may be invoked as a result of a pager put
359 * operation (possibly at object termination time), so we must be careful.
362 vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize)
366 vm_pindex_t nobjsize;
368 if ((object = vp->v_object) == NULL)
370 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
371 VM_OBJECT_WLOCK(object);
372 if (object->type == OBJT_DEAD) {
373 VM_OBJECT_WUNLOCK(object);
376 KASSERT(object->type == OBJT_VNODE,
377 ("not vnode-backed object %p", object));
378 if (nsize == object->un_pager.vnp.vnp_size) {
380 * Hasn't changed size
382 VM_OBJECT_WUNLOCK(object);
385 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
386 if (nsize < object->un_pager.vnp.vnp_size) {
388 * File has shrunk. Toss any cached pages beyond the new EOF.
390 if (nobjsize < object->size)
391 vm_object_page_remove(object, nobjsize, object->size,
394 * this gets rid of garbage at the end of a page that is now
395 * only partially backed by the vnode.
397 * XXX for some reason (I don't know yet), if we take a
398 * completely invalid page and mark it partially valid
399 * it can screw up NFS reads, so we don't allow the case.
401 if ((nsize & PAGE_MASK) &&
402 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
404 int base = (int)nsize & PAGE_MASK;
405 int size = PAGE_SIZE - base;
408 * Clear out partial-page garbage in case
409 * the page has been mapped.
411 pmap_zero_page_area(m, base, size);
414 * Update the valid bits to reflect the blocks that
415 * have been zeroed. Some of these valid bits may
416 * have already been set.
418 vm_page_set_valid_range(m, base, size);
421 * Round "base" to the next block boundary so that the
422 * dirty bit for a partially zeroed block is not
425 base = roundup2(base, DEV_BSIZE);
428 * Clear out partial-page dirty bits.
430 * note that we do not clear out the valid
431 * bits. This would prevent bogus_page
432 * replacement from working properly.
434 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
435 } else if ((nsize & PAGE_MASK) &&
436 vm_page_is_cached(object, OFF_TO_IDX(nsize))) {
437 vm_page_cache_free(object, OFF_TO_IDX(nsize),
441 object->un_pager.vnp.vnp_size = nsize;
442 object->size = nobjsize;
443 VM_OBJECT_WUNLOCK(object);
447 * calculate the linear (byte) disk address of specified virtual
451 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
462 if (vp->v_iflag & VI_DOOMED)
465 bsize = vp->v_mount->mnt_stat.f_iosize;
466 vblock = address / bsize;
467 voffset = address % bsize;
469 err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
471 if (*rtaddress != -1)
472 *rtaddress += voffset / DEV_BSIZE;
475 *run *= bsize/PAGE_SIZE;
476 *run -= voffset/PAGE_SIZE;
484 * small block filesystem vnode pager input
487 vnode_pager_input_smlfs(vm_object_t object, vm_page_t m)
500 if (vp->v_iflag & VI_DOOMED)
503 bsize = vp->v_mount->mnt_stat.f_iosize;
505 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
507 sf = sf_buf_alloc(m, 0);
509 for (i = 0; i < PAGE_SIZE / bsize; i++) {
510 vm_ooffset_t address;
512 bits = vm_page_bits(i * bsize, bsize);
516 address = IDX_TO_OFF(m->pindex) + i * bsize;
517 if (address >= object->un_pager.vnp.vnp_size) {
520 error = vnode_pager_addr(vp, address, &fileaddr, NULL);
524 if (fileaddr != -1) {
525 bp = getpbuf(&vnode_pbuf_freecnt);
527 /* build a minimal buffer header */
528 bp->b_iocmd = BIO_READ;
529 bp->b_iodone = bdone;
530 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
531 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
532 bp->b_rcred = crhold(curthread->td_ucred);
533 bp->b_wcred = crhold(curthread->td_ucred);
534 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
535 bp->b_blkno = fileaddr;
538 bp->b_bcount = bsize;
539 bp->b_bufsize = bsize;
540 bp->b_runningbufspace = bp->b_bufsize;
541 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
544 bp->b_iooffset = dbtob(bp->b_blkno);
547 bwait(bp, PVM, "vnsrd");
549 if ((bp->b_ioflags & BIO_ERROR) != 0)
553 * free the buffer header back to the swap buffer pool
557 relpbuf(bp, &vnode_pbuf_freecnt);
561 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
562 KASSERT((m->dirty & bits) == 0,
563 ("vnode_pager_input_smlfs: page %p is dirty", m));
564 VM_OBJECT_WLOCK(object);
566 VM_OBJECT_WUNLOCK(object);
570 return VM_PAGER_ERROR;
576 * old style vnode pager input routine
579 vnode_pager_input_old(vm_object_t object, vm_page_t m)
588 VM_OBJECT_ASSERT_WLOCKED(object);
592 * Return failure if beyond current EOF
594 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
598 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
599 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
601 VM_OBJECT_WUNLOCK(object);
604 * Allocate a kernel virtual address and initialize so that
605 * we can use VOP_READ/WRITE routines.
607 sf = sf_buf_alloc(m, 0);
609 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
611 auio.uio_iov = &aiov;
613 auio.uio_offset = IDX_TO_OFF(m->pindex);
614 auio.uio_segflg = UIO_SYSSPACE;
615 auio.uio_rw = UIO_READ;
616 auio.uio_resid = size;
617 auio.uio_td = curthread;
619 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
621 int count = size - auio.uio_resid;
625 else if (count != PAGE_SIZE)
626 bzero((caddr_t)sf_buf_kva(sf) + count,
631 VM_OBJECT_WLOCK(object);
633 KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
635 m->valid = VM_PAGE_BITS_ALL;
636 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
640 * generic vnode pager input routine
644 * Local media VFS's that do not implement their own VOP_GETPAGES
645 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
646 * to implement the previous behaviour.
648 * All other FS's should use the bypass to get to the local media
649 * backing vp's VOP_GETPAGES.
652 vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
656 int bytes = count * PAGE_SIZE;
659 VM_OBJECT_WUNLOCK(object);
660 rtval = VOP_GETPAGES(vp, m, bytes, reqpage);
661 KASSERT(rtval != EOPNOTSUPP,
662 ("vnode_pager: FS getpages not implemented\n"));
663 VM_OBJECT_WLOCK(object);
668 * The implementation of VOP_GETPAGES() for local filesystems, where
669 * partially valid pages can only occur at the end of file.
672 vnode_pager_local_getpages(struct vop_getpages_args *ap)
676 mreq = ap->a_m[ap->a_reqpage];
679 * Since the caller has busied the requested page, that page's valid
680 * field will not be changed by other threads.
682 vm_page_assert_xbusied(mreq);
685 * The requested page has valid blocks. Invalid part can only
686 * exist at the end of file, and the page is made fully valid
687 * by zeroing in vm_pager_getpages(). Free non-requested
688 * pages, since no i/o is done to read its content.
690 if (mreq->valid != 0) {
691 vm_pager_free_nonreq(mreq->object, ap->a_m, ap->a_reqpage,
692 round_page(ap->a_count) / PAGE_SIZE);
693 return (VM_PAGER_OK);
696 return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
697 ap->a_count, ap->a_reqpage));
701 * This is now called from local media FS's to operate against their
702 * own vnodes if they fail to implement VOP_GETPAGES.
705 vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int bytecount,
710 off_t foff, tfoff, nextoff;
711 int i, j, size, bsize, first;
712 daddr_t firstaddr, reqblock;
721 object = vp->v_object;
722 count = bytecount / PAGE_SIZE;
724 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
725 ("vnode_pager_generic_getpages does not support devices"));
726 if (vp->v_iflag & VI_DOOMED)
729 bsize = vp->v_mount->mnt_stat.f_iosize;
731 /* get the UNDERLYING device for the file with VOP_BMAP() */
734 * originally, we did not check for an error return value -- assuming
735 * an fs always has a bmap entry point -- that assumption is wrong!!!
737 foff = IDX_TO_OFF(m[reqpage]->pindex);
740 * if we can't bmap, use old VOP code
742 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
743 if (error == EOPNOTSUPP) {
744 VM_OBJECT_WLOCK(object);
746 for (i = 0; i < count; i++)
750 vm_page_unlock(m[i]);
752 PCPU_INC(cnt.v_vnodein);
753 PCPU_INC(cnt.v_vnodepgsin);
754 error = vnode_pager_input_old(object, m[reqpage]);
755 VM_OBJECT_WUNLOCK(object);
757 } else if (error != 0) {
758 vm_pager_free_nonreq(object, m, reqpage, count);
759 return (VM_PAGER_ERROR);
762 * if the blocksize is smaller than a page size, then use
763 * special small filesystem code. NFS sometimes has a small
764 * blocksize, but it can handle large reads itself.
766 } else if ((PAGE_SIZE / bsize) > 1 &&
767 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
768 vm_pager_free_nonreq(object, m, reqpage, count);
769 PCPU_INC(cnt.v_vnodein);
770 PCPU_INC(cnt.v_vnodepgsin);
771 return vnode_pager_input_smlfs(object, m[reqpage]);
775 * Since the caller has busied the requested page, that page's valid
776 * field will not be changed by other threads.
778 vm_page_assert_xbusied(m[reqpage]);
781 * If we have a completely valid page available to us, we can
782 * clean up and return. Otherwise we have to re-read the
785 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
786 vm_pager_free_nonreq(object, m, reqpage, count);
787 return (VM_PAGER_OK);
788 } else if (reqblock == -1) {
789 pmap_zero_page(m[reqpage]);
790 KASSERT(m[reqpage]->dirty == 0,
791 ("vnode_pager_generic_getpages: page %p is dirty", m));
792 VM_OBJECT_WLOCK(object);
793 m[reqpage]->valid = VM_PAGE_BITS_ALL;
794 for (i = 0; i < count; i++)
798 vm_page_unlock(m[i]);
800 VM_OBJECT_WUNLOCK(object);
801 return (VM_PAGER_OK);
802 } else if (m[reqpage]->valid != 0) {
803 VM_OBJECT_WLOCK(object);
804 m[reqpage]->valid = 0;
805 VM_OBJECT_WUNLOCK(object);
809 * here on direct device I/O
814 * calculate the run that includes the required page
816 for (first = 0, i = 0; i < count; i = runend) {
817 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
819 VM_OBJECT_WLOCK(object);
820 for (; i < count; i++)
824 vm_page_unlock(m[i]);
826 VM_OBJECT_WUNLOCK(object);
827 return (VM_PAGER_ERROR);
829 if (firstaddr == -1) {
830 VM_OBJECT_WLOCK(object);
831 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
832 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
833 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
836 (object->un_pager.vnp.vnp_size >> 32),
837 (uintmax_t)object->un_pager.vnp.vnp_size);
841 vm_page_unlock(m[i]);
842 VM_OBJECT_WUNLOCK(object);
848 if (runend <= reqpage) {
849 VM_OBJECT_WLOCK(object);
850 for (j = i; j < runend; j++) {
853 vm_page_unlock(m[j]);
855 VM_OBJECT_WUNLOCK(object);
857 if (runpg < (count - first)) {
858 VM_OBJECT_WLOCK(object);
859 for (i = first + runpg; i < count; i++) {
862 vm_page_unlock(m[i]);
864 VM_OBJECT_WUNLOCK(object);
865 count = first + runpg;
873 * the first and last page have been calculated now, move input pages
874 * to be zero based...
883 * calculate the file virtual address for the transfer
885 foff = IDX_TO_OFF(m[0]->pindex);
888 * calculate the size of the transfer
890 size = count * PAGE_SIZE;
891 KASSERT(count > 0, ("zero count"));
892 if ((foff + size) > object->un_pager.vnp.vnp_size)
893 size = object->un_pager.vnp.vnp_size - foff;
894 KASSERT(size > 0, ("zero size"));
897 * round up physical size for real devices.
900 int secmask = bo->bo_bsize - 1;
901 KASSERT(secmask < PAGE_SIZE && secmask > 0,
902 ("vnode_pager_generic_getpages: sector size %d too large",
904 size = (size + secmask) & ~secmask;
907 bp = getpbuf(&vnode_pbuf_freecnt);
908 kva = (vm_offset_t)bp->b_data;
911 * and map the pages to be read into the kva, if the filesystem
912 * requires mapped buffers.
915 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 &&
916 unmapped_buf_allowed) {
917 bp->b_data = unmapped_buf;
918 bp->b_kvabase = unmapped_buf;
920 bp->b_flags |= B_UNMAPPED;
921 bp->b_npages = count;
922 for (i = 0; i < count; i++)
923 bp->b_pages[i] = m[i];
925 pmap_qenter(kva, m, count);
927 /* build a minimal buffer header */
928 bp->b_iocmd = BIO_READ;
929 bp->b_iodone = bdone;
930 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
931 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
932 bp->b_rcred = crhold(curthread->td_ucred);
933 bp->b_wcred = crhold(curthread->td_ucred);
934 bp->b_blkno = firstaddr;
938 bp->b_bufsize = size;
939 bp->b_runningbufspace = bp->b_bufsize;
940 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
942 PCPU_INC(cnt.v_vnodein);
943 PCPU_ADD(cnt.v_vnodepgsin, count);
946 bp->b_iooffset = dbtob(bp->b_blkno);
949 bwait(bp, PVM, "vnread");
951 if ((bp->b_ioflags & BIO_ERROR) != 0)
954 if (error == 0 && size != count * PAGE_SIZE) {
955 if ((bp->b_flags & B_UNMAPPED) != 0) {
956 bp->b_flags &= ~B_UNMAPPED;
957 pmap_qenter(kva, m, count);
959 bzero((caddr_t)kva + size, PAGE_SIZE * count - size);
961 if ((bp->b_flags & B_UNMAPPED) == 0)
962 pmap_qremove(kva, count);
963 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) {
964 bp->b_data = (caddr_t)kva;
965 bp->b_kvabase = (caddr_t)kva;
966 bp->b_flags &= ~B_UNMAPPED;
967 for (i = 0; i < count; i++)
968 bp->b_pages[i] = NULL;
972 * free the buffer header back to the swap buffer pool
976 relpbuf(bp, &vnode_pbuf_freecnt);
978 VM_OBJECT_WLOCK(object);
979 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
982 nextoff = tfoff + PAGE_SIZE;
985 if (nextoff <= object->un_pager.vnp.vnp_size) {
987 * Read filled up entire page.
989 mt->valid = VM_PAGE_BITS_ALL;
990 KASSERT(mt->dirty == 0,
991 ("vnode_pager_generic_getpages: page %p is dirty",
993 KASSERT(!pmap_page_is_mapped(mt),
994 ("vnode_pager_generic_getpages: page %p is mapped",
998 * Read did not fill up entire page.
1000 * Currently we do not set the entire page valid,
1001 * we just try to clear the piece that we couldn't
1004 vm_page_set_valid_range(mt, 0,
1005 object->un_pager.vnp.vnp_size - tfoff);
1006 KASSERT((mt->dirty & vm_page_bits(0,
1007 object->un_pager.vnp.vnp_size - tfoff)) == 0,
1008 ("vnode_pager_generic_getpages: page %p is dirty",
1013 vm_page_readahead_finish(mt);
1015 VM_OBJECT_WUNLOCK(object);
1017 printf("vnode_pager_getpages: I/O read error\n");
1019 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
1023 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
1024 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
1025 * vnode_pager_generic_putpages() to implement the previous behaviour.
1027 * All other FS's should use the bypass to get to the local media
1028 * backing vp's VOP_PUTPAGES.
1031 vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
1032 int flags, int *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 (vm_cnt.v_free_count + vm_cnt.v_cache_count <
1051 vm_cnt.v_pageout_free_min)
1052 flags |= VM_PAGER_PUT_SYNC;
1055 * Call device-specific putpages function
1057 vp = object->handle;
1058 VM_OBJECT_WUNLOCK(object);
1059 rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals);
1060 KASSERT(rtval != EOPNOTSUPP,
1061 ("vnode_pager: stale FS putpages\n"));
1062 VM_OBJECT_WLOCK(object);
1067 * This is now called from local media FS's to operate against their
1068 * own vnodes if they fail to implement VOP_PUTPAGES.
1070 * This is typically called indirectly via the pageout daemon and
1071 * clustering has already typically occured, so in general we ask the
1072 * underlying filesystem to write the data out asynchronously rather
1076 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1077 int flags, int *rtvals)
1084 int maxsize, ncount;
1085 vm_ooffset_t poffset;
1091 static struct timeval lastfail;
1094 object = vp->v_object;
1095 count = bytecount / PAGE_SIZE;
1097 for (i = 0; i < count; i++)
1098 rtvals[i] = VM_PAGER_ERROR;
1100 if ((int64_t)ma[0]->pindex < 0) {
1101 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1102 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1103 rtvals[0] = VM_PAGER_BAD;
1104 return VM_PAGER_BAD;
1107 maxsize = count * PAGE_SIZE;
1110 poffset = IDX_TO_OFF(ma[0]->pindex);
1113 * If the page-aligned write is larger then the actual file we
1114 * have to invalidate pages occuring beyond the file EOF. However,
1115 * there is an edge case where a file may not be page-aligned where
1116 * the last page is partially invalid. In this case the filesystem
1117 * may not properly clear the dirty bits for the entire page (which
1118 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1119 * With the page locked we are free to fix-up the dirty bits here.
1121 * We do not under any circumstances truncate the valid bits, as
1122 * this will screw up bogus page replacement.
1124 VM_OBJECT_WLOCK(object);
1125 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1126 if (object->un_pager.vnp.vnp_size > poffset) {
1129 maxsize = object->un_pager.vnp.vnp_size - poffset;
1130 ncount = btoc(maxsize);
1131 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1133 * If the object is locked and the following
1134 * conditions hold, then the page's dirty
1135 * field cannot be concurrently changed by a
1139 vm_page_assert_sbusied(m);
1140 KASSERT(!pmap_page_is_write_mapped(m),
1141 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1142 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1149 if (ncount < count) {
1150 for (i = ncount; i < count; i++) {
1151 rtvals[i] = VM_PAGER_BAD;
1155 VM_OBJECT_WUNLOCK(object);
1158 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
1159 * rather then a bdwrite() to prevent paging I/O from saturating
1160 * the buffer cache. Dummy-up the sequential heuristic to cause
1161 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1162 * the system decides how to cluster.
1165 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1167 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1168 ioflags |= IO_ASYNC;
1169 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1170 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1172 aiov.iov_base = (caddr_t) 0;
1173 aiov.iov_len = maxsize;
1174 auio.uio_iov = &aiov;
1175 auio.uio_iovcnt = 1;
1176 auio.uio_offset = poffset;
1177 auio.uio_segflg = UIO_NOCOPY;
1178 auio.uio_rw = UIO_WRITE;
1179 auio.uio_resid = maxsize;
1180 auio.uio_td = (struct thread *) 0;
1181 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1182 PCPU_INC(cnt.v_vnodeout);
1183 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1186 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1187 printf("vnode_pager_putpages: I/O error %d\n", error);
1189 if (auio.uio_resid) {
1190 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1191 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1192 auio.uio_resid, (u_long)ma[0]->pindex);
1194 for (i = 0; i < ncount; i++) {
1195 rtvals[i] = VM_PAGER_OK;
1201 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1208 obj = ma[0]->object;
1209 VM_OBJECT_WLOCK(obj);
1210 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1211 if (pos < trunc_page(written)) {
1212 rtvals[i] = VM_PAGER_OK;
1213 vm_page_undirty(ma[i]);
1215 /* Partially written page. */
1216 rtvals[i] = VM_PAGER_AGAIN;
1217 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1220 VM_OBJECT_WUNLOCK(obj);
1224 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1228 vm_ooffset_t old_wm;
1230 VM_OBJECT_WLOCK(object);
1231 if (object->type != OBJT_VNODE) {
1232 VM_OBJECT_WUNLOCK(object);
1235 old_wm = object->un_pager.vnp.writemappings;
1236 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1237 vp = object->handle;
1238 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1239 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1240 VOP_ADD_WRITECOUNT(vp, 1);
1241 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1242 __func__, vp, vp->v_writecount);
1243 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1244 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1245 VOP_ADD_WRITECOUNT(vp, -1);
1246 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1247 __func__, vp, vp->v_writecount);
1249 VM_OBJECT_WUNLOCK(object);
1253 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1260 VM_OBJECT_WLOCK(object);
1263 * First, recheck the object type to account for the race when
1264 * the vnode is reclaimed.
1266 if (object->type != OBJT_VNODE) {
1267 VM_OBJECT_WUNLOCK(object);
1272 * Optimize for the case when writemappings is not going to
1276 if (object->un_pager.vnp.writemappings != inc) {
1277 object->un_pager.vnp.writemappings -= inc;
1278 VM_OBJECT_WUNLOCK(object);
1282 vp = object->handle;
1284 VM_OBJECT_WUNLOCK(object);
1286 vn_start_write(vp, &mp, V_WAIT);
1287 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1290 * Decrement the object's writemappings, by swapping the start
1291 * and end arguments for vnode_pager_update_writecount(). If
1292 * there was not a race with vnode reclaimation, then the
1293 * vnode's v_writecount is decremented.
1295 vnode_pager_update_writecount(object, end, start);
1299 vn_finished_write(mp);