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_WLOCK(object);
231 KASSERT(object->ref_count == 1,
232 ("leaked ref %p %d", object, object->ref_count));
233 object->type = OBJT_DEAD;
234 object->ref_count = 0;
235 VM_OBJECT_WUNLOCK(object);
236 vm_object_destroy(object);
239 vp->v_object = object;
243 VM_OBJECT_WUNLOCK(object);
250 * The object must be locked.
253 vnode_pager_dealloc(object)
261 panic("vnode_pager_dealloc: pager already dealloced");
263 VM_OBJECT_ASSERT_WLOCKED(object);
264 vm_object_pip_wait(object, "vnpdea");
265 refs = object->ref_count;
267 object->handle = NULL;
268 object->type = OBJT_DEAD;
269 if (object->flags & OBJ_DISCONNECTWNT) {
270 vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
273 ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
274 if (object->un_pager.vnp.writemappings > 0) {
275 object->un_pager.vnp.writemappings = 0;
276 VOP_ADD_WRITECOUNT(vp, -1);
277 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
278 __func__, vp, vp->v_writecount);
282 VM_OBJECT_WUNLOCK(object);
285 VM_OBJECT_WLOCK(object);
289 vnode_pager_haspage(object, pindex, before, after)
295 struct vnode *vp = object->handle;
301 int pagesperblock, blocksperpage;
303 VM_OBJECT_ASSERT_WLOCKED(object);
305 * If no vp or vp is doomed or marked transparent to VM, we do not
308 if (vp == NULL || vp->v_iflag & VI_DOOMED)
311 * If the offset is beyond end of file we do
314 if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size)
317 bsize = vp->v_mount->mnt_stat.f_iosize;
318 pagesperblock = bsize / PAGE_SIZE;
320 if (pagesperblock > 0) {
321 reqblock = pindex / pagesperblock;
323 blocksperpage = (PAGE_SIZE / bsize);
324 reqblock = pindex * blocksperpage;
326 VM_OBJECT_WUNLOCK(object);
327 err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before);
328 VM_OBJECT_WLOCK(object);
333 if (pagesperblock > 0) {
334 poff = pindex - (reqblock * pagesperblock);
336 *before *= pagesperblock;
341 *after *= pagesperblock;
342 numafter = pagesperblock - (poff + 1);
343 if (IDX_TO_OFF(pindex + numafter) >
344 object->un_pager.vnp.vnp_size) {
346 OFF_TO_IDX(object->un_pager.vnp.vnp_size) -
353 *before /= blocksperpage;
357 *after /= blocksperpage;
364 * Lets the VM system know about a change in size for a file.
365 * We adjust our own internal size and flush any cached pages in
366 * the associated object that are affected by the size change.
368 * Note: this routine may be invoked as a result of a pager put
369 * operation (possibly at object termination time), so we must be careful.
372 vnode_pager_setsize(vp, nsize)
378 vm_pindex_t nobjsize;
380 if ((object = vp->v_object) == NULL)
382 /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
383 VM_OBJECT_WLOCK(object);
384 if (object->type == OBJT_DEAD) {
385 VM_OBJECT_WUNLOCK(object);
388 KASSERT(object->type == OBJT_VNODE,
389 ("not vnode-backed object %p", object));
390 if (nsize == object->un_pager.vnp.vnp_size) {
392 * Hasn't changed size
394 VM_OBJECT_WUNLOCK(object);
397 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
398 if (nsize < object->un_pager.vnp.vnp_size) {
400 * File has shrunk. Toss any cached pages beyond the new EOF.
402 if (nobjsize < object->size)
403 vm_object_page_remove(object, nobjsize, object->size,
406 * this gets rid of garbage at the end of a page that is now
407 * only partially backed by the vnode.
409 * XXX for some reason (I don't know yet), if we take a
410 * completely invalid page and mark it partially valid
411 * it can screw up NFS reads, so we don't allow the case.
413 if ((nsize & PAGE_MASK) &&
414 (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
416 int base = (int)nsize & PAGE_MASK;
417 int size = PAGE_SIZE - base;
420 * Clear out partial-page garbage in case
421 * the page has been mapped.
423 pmap_zero_page_area(m, base, size);
426 * Update the valid bits to reflect the blocks that
427 * have been zeroed. Some of these valid bits may
428 * have already been set.
430 vm_page_set_valid_range(m, base, size);
433 * Round "base" to the next block boundary so that the
434 * dirty bit for a partially zeroed block is not
437 base = roundup2(base, DEV_BSIZE);
440 * Clear out partial-page dirty bits.
442 * note that we do not clear out the valid
443 * bits. This would prevent bogus_page
444 * replacement from working properly.
446 vm_page_clear_dirty(m, base, PAGE_SIZE - base);
447 } else if ((nsize & PAGE_MASK) &&
448 vm_page_is_cached(object, OFF_TO_IDX(nsize))) {
449 vm_page_cache_free(object, OFF_TO_IDX(nsize),
453 object->un_pager.vnp.vnp_size = nsize;
454 object->size = nobjsize;
455 VM_OBJECT_WUNLOCK(object);
459 * calculate the linear (byte) disk address of specified virtual
463 vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress,
474 if (vp->v_iflag & VI_DOOMED)
477 bsize = vp->v_mount->mnt_stat.f_iosize;
478 vblock = address / bsize;
479 voffset = address % bsize;
481 err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL);
483 if (*rtaddress != -1)
484 *rtaddress += voffset / DEV_BSIZE;
487 *run *= bsize/PAGE_SIZE;
488 *run -= voffset/PAGE_SIZE;
496 * small block filesystem vnode pager input
499 vnode_pager_input_smlfs(object, m)
514 if (vp->v_iflag & VI_DOOMED)
517 bsize = vp->v_mount->mnt_stat.f_iosize;
519 VOP_BMAP(vp, 0, &bo, 0, NULL, NULL);
521 sf = sf_buf_alloc(m, 0);
523 for (i = 0; i < PAGE_SIZE / bsize; i++) {
524 vm_ooffset_t address;
526 bits = vm_page_bits(i * bsize, bsize);
530 address = IDX_TO_OFF(m->pindex) + i * bsize;
531 if (address >= object->un_pager.vnp.vnp_size) {
534 error = vnode_pager_addr(vp, address, &fileaddr, NULL);
538 if (fileaddr != -1) {
539 bp = getpbuf(&vnode_pbuf_freecnt);
541 /* build a minimal buffer header */
542 bp->b_iocmd = BIO_READ;
543 bp->b_iodone = bdone;
544 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
545 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
546 bp->b_rcred = crhold(curthread->td_ucred);
547 bp->b_wcred = crhold(curthread->td_ucred);
548 bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize;
549 bp->b_blkno = fileaddr;
552 bp->b_bcount = bsize;
553 bp->b_bufsize = bsize;
554 bp->b_runningbufspace = bp->b_bufsize;
555 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
558 bp->b_iooffset = dbtob(bp->b_blkno);
561 bwait(bp, PVM, "vnsrd");
563 if ((bp->b_ioflags & BIO_ERROR) != 0)
567 * free the buffer header back to the swap buffer pool
571 relpbuf(bp, &vnode_pbuf_freecnt);
575 bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize);
576 KASSERT((m->dirty & bits) == 0,
577 ("vnode_pager_input_smlfs: page %p is dirty", m));
578 VM_OBJECT_WLOCK(object);
580 VM_OBJECT_WUNLOCK(object);
584 return VM_PAGER_ERROR;
590 * old style vnode pager input routine
593 vnode_pager_input_old(object, m)
604 VM_OBJECT_ASSERT_WLOCKED(object);
608 * Return failure if beyond current EOF
610 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
614 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
615 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
617 VM_OBJECT_WUNLOCK(object);
620 * Allocate a kernel virtual address and initialize so that
621 * we can use VOP_READ/WRITE routines.
623 sf = sf_buf_alloc(m, 0);
625 aiov.iov_base = (caddr_t)sf_buf_kva(sf);
627 auio.uio_iov = &aiov;
629 auio.uio_offset = IDX_TO_OFF(m->pindex);
630 auio.uio_segflg = UIO_SYSSPACE;
631 auio.uio_rw = UIO_READ;
632 auio.uio_resid = size;
633 auio.uio_td = curthread;
635 error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
637 int count = size - auio.uio_resid;
641 else if (count != PAGE_SIZE)
642 bzero((caddr_t)sf_buf_kva(sf) + count,
647 VM_OBJECT_WLOCK(object);
649 KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
651 m->valid = VM_PAGE_BITS_ALL;
652 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
656 * generic vnode pager input routine
660 * Local media VFS's that do not implement their own VOP_GETPAGES
661 * should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
662 * to implement the previous behaviour.
664 * All other FS's should use the bypass to get to the local media
665 * backing vp's VOP_GETPAGES.
668 vnode_pager_getpages(object, m, count, reqpage)
676 int bytes = count * PAGE_SIZE;
679 VM_OBJECT_WUNLOCK(object);
680 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
681 KASSERT(rtval != EOPNOTSUPP,
682 ("vnode_pager: FS getpages not implemented\n"));
683 VM_OBJECT_WLOCK(object);
688 * This is now called from local media FS's to operate against their
689 * own vnodes if they fail to implement VOP_GETPAGES.
692 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
700 off_t foff, tfoff, nextoff;
701 int i, j, size, bsize, first;
702 daddr_t firstaddr, reqblock;
711 object = vp->v_object;
712 count = bytecount / PAGE_SIZE;
714 KASSERT(vp->v_type != VCHR && vp->v_type != VBLK,
715 ("vnode_pager_generic_getpages does not support devices"));
716 if (vp->v_iflag & VI_DOOMED)
719 bsize = vp->v_mount->mnt_stat.f_iosize;
721 /* get the UNDERLYING device for the file with VOP_BMAP() */
724 * originally, we did not check for an error return value -- assuming
725 * an fs always has a bmap entry point -- that assumption is wrong!!!
727 foff = IDX_TO_OFF(m[reqpage]->pindex);
730 * if we can't bmap, use old VOP code
732 error = VOP_BMAP(vp, foff / bsize, &bo, &reqblock, NULL, NULL);
733 if (error == EOPNOTSUPP) {
734 VM_OBJECT_WLOCK(object);
736 for (i = 0; i < count; i++)
740 vm_page_unlock(m[i]);
742 PCPU_INC(cnt.v_vnodein);
743 PCPU_INC(cnt.v_vnodepgsin);
744 error = vnode_pager_input_old(object, m[reqpage]);
745 VM_OBJECT_WUNLOCK(object);
747 } else if (error != 0) {
748 VM_OBJECT_WLOCK(object);
749 for (i = 0; i < count; i++)
753 vm_page_unlock(m[i]);
755 VM_OBJECT_WUNLOCK(object);
756 return (VM_PAGER_ERROR);
759 * if the blocksize is smaller than a page size, then use
760 * special small filesystem code. NFS sometimes has a small
761 * blocksize, but it can handle large reads itself.
763 } else if ((PAGE_SIZE / bsize) > 1 &&
764 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
765 VM_OBJECT_WLOCK(object);
766 for (i = 0; i < count; i++)
770 vm_page_unlock(m[i]);
772 VM_OBJECT_WUNLOCK(object);
773 PCPU_INC(cnt.v_vnodein);
774 PCPU_INC(cnt.v_vnodepgsin);
775 return vnode_pager_input_smlfs(object, m[reqpage]);
779 * If we have a completely valid page available to us, we can
780 * clean up and return. Otherwise we have to re-read the
783 VM_OBJECT_WLOCK(object);
784 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
785 for (i = 0; i < count; i++)
789 vm_page_unlock(m[i]);
791 VM_OBJECT_WUNLOCK(object);
793 } else if (reqblock == -1) {
794 pmap_zero_page(m[reqpage]);
795 KASSERT(m[reqpage]->dirty == 0,
796 ("vnode_pager_generic_getpages: page %p is dirty", m));
797 m[reqpage]->valid = VM_PAGE_BITS_ALL;
798 for (i = 0; i < count; i++)
802 vm_page_unlock(m[i]);
804 VM_OBJECT_WUNLOCK(object);
805 return (VM_PAGER_OK);
807 m[reqpage]->valid = 0;
808 VM_OBJECT_WUNLOCK(object);
811 * here on direct device I/O
816 * calculate the run that includes the required page
818 for (first = 0, i = 0; i < count; i = runend) {
819 if (vnode_pager_addr(vp, IDX_TO_OFF(m[i]->pindex), &firstaddr,
821 VM_OBJECT_WLOCK(object);
822 for (; i < count; i++)
826 vm_page_unlock(m[i]);
828 VM_OBJECT_WUNLOCK(object);
829 return (VM_PAGER_ERROR);
831 if (firstaddr == -1) {
832 VM_OBJECT_WLOCK(object);
833 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
834 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %jd, foff: 0x%jx%08jx, vnp_size: 0x%jx%08jx",
835 (intmax_t)firstaddr, (uintmax_t)(foff >> 32),
838 (object->un_pager.vnp.vnp_size >> 32),
839 (uintmax_t)object->un_pager.vnp.vnp_size);
843 vm_page_unlock(m[i]);
844 VM_OBJECT_WUNLOCK(object);
850 if (runend <= reqpage) {
851 VM_OBJECT_WLOCK(object);
852 for (j = i; j < runend; j++) {
855 vm_page_unlock(m[j]);
857 VM_OBJECT_WUNLOCK(object);
859 if (runpg < (count - first)) {
860 VM_OBJECT_WLOCK(object);
861 for (i = first + runpg; i < count; i++) {
864 vm_page_unlock(m[i]);
866 VM_OBJECT_WUNLOCK(object);
867 count = first + runpg;
875 * the first and last page have been calculated now, move input pages
876 * to be zero based...
885 * calculate the file virtual address for the transfer
887 foff = IDX_TO_OFF(m[0]->pindex);
890 * calculate the size of the transfer
892 size = count * PAGE_SIZE;
893 KASSERT(count > 0, ("zero count"));
894 if ((foff + size) > object->un_pager.vnp.vnp_size)
895 size = object->un_pager.vnp.vnp_size - foff;
896 KASSERT(size > 0, ("zero size"));
899 * round up physical size for real devices.
902 int secmask = bo->bo_bsize - 1;
903 KASSERT(secmask < PAGE_SIZE && secmask > 0,
904 ("vnode_pager_generic_getpages: sector size %d too large",
906 size = (size + secmask) & ~secmask;
909 bp = getpbuf(&vnode_pbuf_freecnt);
910 kva = (vm_offset_t)bp->b_data;
913 * and map the pages to be read into the kva, if the filesystem
914 * requires mapped buffers.
917 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 &&
918 unmapped_buf_allowed) {
919 bp->b_data = unmapped_buf;
920 bp->b_kvabase = unmapped_buf;
922 bp->b_flags |= B_UNMAPPED;
923 bp->b_npages = count;
924 for (i = 0; i < count; i++)
925 bp->b_pages[i] = m[i];
927 pmap_qenter(kva, m, count);
929 /* build a minimal buffer header */
930 bp->b_iocmd = BIO_READ;
931 bp->b_iodone = bdone;
932 KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred"));
933 KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred"));
934 bp->b_rcred = crhold(curthread->td_ucred);
935 bp->b_wcred = crhold(curthread->td_ucred);
936 bp->b_blkno = firstaddr;
940 bp->b_bufsize = size;
941 bp->b_runningbufspace = bp->b_bufsize;
942 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
944 PCPU_INC(cnt.v_vnodein);
945 PCPU_ADD(cnt.v_vnodepgsin, count);
948 bp->b_iooffset = dbtob(bp->b_blkno);
951 bwait(bp, PVM, "vnread");
953 if ((bp->b_ioflags & BIO_ERROR) != 0)
956 if (error == 0 && size != count * PAGE_SIZE) {
957 if ((bp->b_flags & B_UNMAPPED) != 0) {
958 bp->b_flags &= ~B_UNMAPPED;
959 pmap_qenter(kva, m, count);
961 bzero((caddr_t)kva + size, PAGE_SIZE * count - size);
963 if ((bp->b_flags & B_UNMAPPED) == 0)
964 pmap_qremove(kva, count);
965 if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) {
966 bp->b_data = (caddr_t)kva;
967 bp->b_kvabase = (caddr_t)kva;
968 bp->b_flags &= ~B_UNMAPPED;
969 for (i = 0; i < count; i++)
970 bp->b_pages[i] = NULL;
974 * free the buffer header back to the swap buffer pool
978 relpbuf(bp, &vnode_pbuf_freecnt);
980 VM_OBJECT_WLOCK(object);
981 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
984 nextoff = tfoff + PAGE_SIZE;
987 if (nextoff <= object->un_pager.vnp.vnp_size) {
989 * Read filled up entire page.
991 mt->valid = VM_PAGE_BITS_ALL;
992 KASSERT(mt->dirty == 0,
993 ("vnode_pager_generic_getpages: page %p is dirty",
995 KASSERT(!pmap_page_is_mapped(mt),
996 ("vnode_pager_generic_getpages: page %p is mapped",
1000 * Read did not fill up entire page.
1002 * Currently we do not set the entire page valid,
1003 * we just try to clear the piece that we couldn't
1006 vm_page_set_valid_range(mt, 0,
1007 object->un_pager.vnp.vnp_size - tfoff);
1008 KASSERT((mt->dirty & vm_page_bits(0,
1009 object->un_pager.vnp.vnp_size - tfoff)) == 0,
1010 ("vnode_pager_generic_getpages: page %p is dirty",
1015 vm_page_readahead_finish(mt);
1017 VM_OBJECT_WUNLOCK(object);
1019 printf("vnode_pager_getpages: I/O read error\n");
1021 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
1025 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
1026 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
1027 * vnode_pager_generic_putpages() to implement the previous behaviour.
1029 * All other FS's should use the bypass to get to the local media
1030 * backing vp's VOP_PUTPAGES.
1033 vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
1034 int flags, int *rtvals)
1038 int bytes = count * PAGE_SIZE;
1041 * Force synchronous operation if we are extremely low on memory
1042 * to prevent a low-memory deadlock. VOP operations often need to
1043 * allocate more memory to initiate the I/O ( i.e. do a BMAP
1044 * operation ). The swapper handles the case by limiting the amount
1045 * of asynchronous I/O, but that sort of solution doesn't scale well
1046 * for the vnode pager without a lot of work.
1048 * Also, the backing vnode's iodone routine may not wake the pageout
1049 * daemon up. This should be probably be addressed XXX.
1052 if (cnt.v_free_count + cnt.v_cache_count < cnt.v_pageout_free_min)
1053 flags |= VM_PAGER_PUT_SYNC;
1056 * Call device-specific putpages function
1058 vp = object->handle;
1059 VM_OBJECT_WUNLOCK(object);
1060 rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals, 0);
1061 KASSERT(rtval != EOPNOTSUPP,
1062 ("vnode_pager: stale FS putpages\n"));
1063 VM_OBJECT_WLOCK(object);
1068 * This is now called from local media FS's to operate against their
1069 * own vnodes if they fail to implement VOP_PUTPAGES.
1071 * This is typically called indirectly via the pageout daemon and
1072 * clustering has already typically occured, so in general we ask the
1073 * underlying filesystem to write the data out asynchronously rather
1077 vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
1078 int flags, int *rtvals)
1085 int maxsize, ncount;
1086 vm_ooffset_t poffset;
1092 static struct timeval lastfail;
1095 object = vp->v_object;
1096 count = bytecount / PAGE_SIZE;
1098 for (i = 0; i < count; i++)
1099 rtvals[i] = VM_PAGER_ERROR;
1101 if ((int64_t)ma[0]->pindex < 0) {
1102 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
1103 (long)ma[0]->pindex, (u_long)ma[0]->dirty);
1104 rtvals[0] = VM_PAGER_BAD;
1105 return VM_PAGER_BAD;
1108 maxsize = count * PAGE_SIZE;
1111 poffset = IDX_TO_OFF(ma[0]->pindex);
1114 * If the page-aligned write is larger then the actual file we
1115 * have to invalidate pages occuring beyond the file EOF. However,
1116 * there is an edge case where a file may not be page-aligned where
1117 * the last page is partially invalid. In this case the filesystem
1118 * may not properly clear the dirty bits for the entire page (which
1119 * could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
1120 * With the page locked we are free to fix-up the dirty bits here.
1122 * We do not under any circumstances truncate the valid bits, as
1123 * this will screw up bogus page replacement.
1125 VM_OBJECT_WLOCK(object);
1126 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
1127 if (object->un_pager.vnp.vnp_size > poffset) {
1130 maxsize = object->un_pager.vnp.vnp_size - poffset;
1131 ncount = btoc(maxsize);
1132 if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
1134 * If the object is locked and the following
1135 * conditions hold, then the page's dirty
1136 * field cannot be concurrently changed by a
1140 vm_page_assert_sbusied(m);
1141 KASSERT(!pmap_page_is_write_mapped(m),
1142 ("vnode_pager_generic_putpages: page %p is not read-only", m));
1143 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
1150 if (ncount < count) {
1151 for (i = ncount; i < count; i++) {
1152 rtvals[i] = VM_PAGER_BAD;
1156 VM_OBJECT_WUNLOCK(object);
1159 * pageouts are already clustered, use IO_ASYNC to force a bawrite()
1160 * rather then a bdwrite() to prevent paging I/O from saturating
1161 * the buffer cache. Dummy-up the sequential heuristic to cause
1162 * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
1163 * the system decides how to cluster.
1166 if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
1168 else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
1169 ioflags |= IO_ASYNC;
1170 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
1171 ioflags |= IO_SEQMAX << IO_SEQSHIFT;
1173 aiov.iov_base = (caddr_t) 0;
1174 aiov.iov_len = maxsize;
1175 auio.uio_iov = &aiov;
1176 auio.uio_iovcnt = 1;
1177 auio.uio_offset = poffset;
1178 auio.uio_segflg = UIO_NOCOPY;
1179 auio.uio_rw = UIO_WRITE;
1180 auio.uio_resid = maxsize;
1181 auio.uio_td = (struct thread *) 0;
1182 error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
1183 PCPU_INC(cnt.v_vnodeout);
1184 PCPU_ADD(cnt.v_vnodepgsout, ncount);
1187 if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
1188 printf("vnode_pager_putpages: I/O error %d\n", error);
1190 if (auio.uio_resid) {
1191 if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
1192 printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
1193 auio.uio_resid, (u_long)ma[0]->pindex);
1195 for (i = 0; i < ncount; i++) {
1196 rtvals[i] = VM_PAGER_OK;
1202 vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written)
1209 obj = ma[0]->object;
1210 VM_OBJECT_WLOCK(obj);
1211 for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) {
1212 if (pos < trunc_page(written)) {
1213 rtvals[i] = VM_PAGER_OK;
1214 vm_page_undirty(ma[i]);
1216 /* Partially written page. */
1217 rtvals[i] = VM_PAGER_AGAIN;
1218 vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK);
1221 VM_OBJECT_WUNLOCK(obj);
1225 vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
1229 vm_ooffset_t old_wm;
1231 VM_OBJECT_WLOCK(object);
1232 if (object->type != OBJT_VNODE) {
1233 VM_OBJECT_WUNLOCK(object);
1236 old_wm = object->un_pager.vnp.writemappings;
1237 object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
1238 vp = object->handle;
1239 if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
1240 ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
1241 VOP_ADD_WRITECOUNT(vp, 1);
1242 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1243 __func__, vp, vp->v_writecount);
1244 } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
1245 ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
1246 VOP_ADD_WRITECOUNT(vp, -1);
1247 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1248 __func__, vp, vp->v_writecount);
1250 VM_OBJECT_WUNLOCK(object);
1254 vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
1261 VM_OBJECT_WLOCK(object);
1264 * First, recheck the object type to account for the race when
1265 * the vnode is reclaimed.
1267 if (object->type != OBJT_VNODE) {
1268 VM_OBJECT_WUNLOCK(object);
1273 * Optimize for the case when writemappings is not going to
1277 if (object->un_pager.vnp.writemappings != inc) {
1278 object->un_pager.vnp.writemappings -= inc;
1279 VM_OBJECT_WUNLOCK(object);
1283 vp = object->handle;
1285 VM_OBJECT_WUNLOCK(object);
1287 vn_start_write(vp, &mp, V_WAIT);
1288 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1291 * Decrement the object's writemappings, by swapping the start
1292 * and end arguments for vnode_pager_update_writecount(). If
1293 * there was not a race with vnode reclaimation, then the
1294 * vnode's v_writecount is decremented.
1296 vnode_pager_update_writecount(object, end, start);
1300 vn_finished_write(mp);