xref: /freebsd/sys/vm/vnode_pager.c (revision 2ad872c5794e4c26fdf6ed219ad3f09ca0d5304a)
1 /*
2  * Copyright (c) 1990 University of Utah.
3  * Copyright (c) 1991 The Regents of the University of California.
4  * All rights reserved.
5  * Copyright (c) 1993, 1994 John S. Dyson
6  * Copyright (c) 1995, David Greenman
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
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.
27  *
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
38  * SUCH DAMAGE.
39  *
40  *	from: @(#)vnode_pager.c	7.5 (Berkeley) 4/20/91
41  *	$Id: vnode_pager.c,v 1.100 1998/10/13 08:24:44 dg Exp $
42  */
43 
44 /*
45  * Page to/from files (vnodes).
46  */
47 
48 /*
49  * TODO:
50  *	Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will
51  *	greatly re-simplify the vnode_pager.
52  */
53 
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/proc.h>
57 #include <sys/vnode.h>
58 #include <sys/mount.h>
59 #include <sys/buf.h>
60 #include <sys/vmmeter.h>
61 
62 #include <vm/vm.h>
63 #include <vm/vm_prot.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_pager.h>
67 #include <vm/vm_map.h>
68 #include <vm/vnode_pager.h>
69 #include <vm/vm_extern.h>
70 
71 static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address,
72 					 int *run));
73 static void vnode_pager_iodone __P((struct buf *bp));
74 static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m));
75 static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m));
76 static void vnode_pager_dealloc __P((vm_object_t));
77 static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
78 static int vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *));
79 static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
80 
81 struct pagerops vnodepagerops = {
82 	NULL,
83 	vnode_pager_alloc,
84 	vnode_pager_dealloc,
85 	vnode_pager_getpages,
86 	vnode_pager_putpages,
87 	vnode_pager_haspage,
88 	NULL
89 };
90 
91 
92 /*
93  * Allocate (or lookup) pager for a vnode.
94  * Handle is a vnode pointer.
95  */
96 vm_object_t
97 vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
98 		  vm_ooffset_t offset)
99 {
100 	vm_object_t object;
101 	struct vnode *vp;
102 
103 	/*
104 	 * Pageout to vnode, no can do yet.
105 	 */
106 	if (handle == NULL)
107 		return (NULL);
108 
109 	vp = (struct vnode *) handle;
110 
111 	/*
112 	 * Prevent race condition when allocating the object. This
113 	 * can happen with NFS vnodes since the nfsnode isn't locked.
114 	 */
115 	while (vp->v_flag & VOLOCK) {
116 		vp->v_flag |= VOWANT;
117 		tsleep(vp, PVM, "vnpobj", 0);
118 	}
119 	vp->v_flag |= VOLOCK;
120 
121 	/*
122 	 * If the object is being terminated, wait for it to
123 	 * go away.
124 	 */
125 	while (((object = vp->v_object) != NULL) &&
126 		(object->flags & OBJ_DEAD)) {
127 		tsleep(object, PVM, "vadead", 0);
128 	}
129 
130 	if (vp->v_usecount == 0)
131 		panic("vnode_pager_alloc: no vnode reference");
132 
133 	if (object == NULL) {
134 		/*
135 		 * And an object of the appropriate size
136 		 */
137 		object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
138 		object->flags = 0;
139 
140 		object->un_pager.vnp.vnp_size = size;
141 
142 		object->handle = handle;
143 		vp->v_object = object;
144 		vp->v_usecount++;
145 	} else {
146 		object->ref_count++;
147 		vp->v_usecount++;
148 	}
149 
150 	vp->v_flag &= ~VOLOCK;
151 	if (vp->v_flag & VOWANT) {
152 		vp->v_flag &= ~VOWANT;
153 		wakeup(vp);
154 	}
155 	return (object);
156 }
157 
158 static void
159 vnode_pager_dealloc(object)
160 	vm_object_t object;
161 {
162 	register struct vnode *vp = object->handle;
163 
164 	if (vp == NULL)
165 		panic("vnode_pager_dealloc: pager already dealloced");
166 
167 	vm_object_pip_wait(object, "vnpdea");
168 
169 	object->handle = NULL;
170 	object->type = OBJT_DEAD;
171 	vp->v_object = NULL;
172 	vp->v_flag &= ~(VTEXT | VOBJBUF);
173 }
174 
175 static boolean_t
176 vnode_pager_haspage(object, pindex, before, after)
177 	vm_object_t object;
178 	vm_pindex_t pindex;
179 	int *before;
180 	int *after;
181 {
182 	struct vnode *vp = object->handle;
183 	daddr_t bn;
184 	int err;
185 	daddr_t reqblock;
186 	int poff;
187 	int bsize;
188 	int pagesperblock, blocksperpage;
189 
190 	if ((vp == NULL) || (vp->v_flag & VDOOMED))
191 		return FALSE;
192 
193 	/*
194 	 * If filesystem no longer mounted or offset beyond end of file we do
195 	 * not have the page.
196 	 */
197 	if ((vp->v_mount == NULL) ||
198 		(IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
199 		return FALSE;
200 
201 	bsize = vp->v_mount->mnt_stat.f_iosize;
202 	pagesperblock = bsize / PAGE_SIZE;
203 	blocksperpage = 0;
204 	if (pagesperblock > 0) {
205 		reqblock = pindex / pagesperblock;
206 	} else {
207 		blocksperpage = (PAGE_SIZE / bsize);
208 		reqblock = pindex * blocksperpage;
209 	}
210 	err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn,
211 		after, before);
212 	if (err)
213 		return TRUE;
214 	if ( bn == -1)
215 		return FALSE;
216 	if (pagesperblock > 0) {
217 		poff = pindex - (reqblock * pagesperblock);
218 		if (before) {
219 			*before *= pagesperblock;
220 			*before += poff;
221 		}
222 		if (after) {
223 			int numafter;
224 			*after *= pagesperblock;
225 			numafter = pagesperblock - (poff + 1);
226 			if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) {
227 				numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex)));
228 			}
229 			*after += numafter;
230 		}
231 	} else {
232 		if (before) {
233 			*before /= blocksperpage;
234 		}
235 
236 		if (after) {
237 			*after /= blocksperpage;
238 		}
239 	}
240 	return TRUE;
241 }
242 
243 /*
244  * Lets the VM system know about a change in size for a file.
245  * We adjust our own internal size and flush any cached pages in
246  * the associated object that are affected by the size change.
247  *
248  * Note: this routine may be invoked as a result of a pager put
249  * operation (possibly at object termination time), so we must be careful.
250  */
251 void
252 vnode_pager_setsize(vp, nsize)
253 	struct vnode *vp;
254 	vm_ooffset_t nsize;
255 {
256 	vm_pindex_t nobjsize;
257 	vm_object_t object = vp->v_object;
258 
259 	if (object == NULL)
260 		return;
261 
262 	/*
263 	 * Hasn't changed size
264 	 */
265 	if (nsize == object->un_pager.vnp.vnp_size)
266 		return;
267 
268 	nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
269 
270 	/*
271 	 * File has shrunk. Toss any cached pages beyond the new EOF.
272 	 */
273 	if (nsize < object->un_pager.vnp.vnp_size) {
274 		vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size);
275 		if (nobjsize < object->size) {
276 			vm_object_page_remove(object, nobjsize, object->size,
277 				FALSE);
278 		}
279 		/*
280 		 * this gets rid of garbage at the end of a page that is now
281 		 * only partially backed by the vnode...
282 		 */
283 		if (nsize & PAGE_MASK) {
284 			vm_offset_t kva;
285 			vm_page_t m;
286 
287 			m = vm_page_lookup(object, OFF_TO_IDX(nsize));
288 			if (m) {
289 				kva = vm_pager_map_page(m);
290 				bzero((caddr_t) kva + (nsize & PAGE_MASK),
291 				    (int) (round_page(nsize) - nsize));
292 				vm_pager_unmap_page(kva);
293 			}
294 		}
295 	}
296 	object->un_pager.vnp.vnp_size = nsize;
297 	object->size = nobjsize;
298 }
299 
300 void
301 vnode_pager_freepage(m)
302 	vm_page_t m;
303 {
304 	vm_page_free(m);
305 }
306 
307 /*
308  * calculate the linear (byte) disk address of specified virtual
309  * file address
310  */
311 static vm_offset_t
312 vnode_pager_addr(vp, address, run)
313 	struct vnode *vp;
314 	vm_ooffset_t address;
315 	int *run;
316 {
317 	int rtaddress;
318 	int bsize;
319 	daddr_t block;
320 	struct vnode *rtvp;
321 	int err;
322 	daddr_t vblock;
323 	int voffset;
324 
325 	if ((int) address < 0)
326 		return -1;
327 
328 	if (vp->v_mount == NULL)
329 		return -1;
330 
331 	bsize = vp->v_mount->mnt_stat.f_iosize;
332 	vblock = address / bsize;
333 	voffset = address % bsize;
334 
335 	err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL);
336 
337 	if (err || (block == -1))
338 		rtaddress = -1;
339 	else {
340 		rtaddress = block + voffset / DEV_BSIZE;
341 		if( run) {
342 			*run += 1;
343 			*run *= bsize/PAGE_SIZE;
344 			*run -= voffset/PAGE_SIZE;
345 		}
346 	}
347 
348 	return rtaddress;
349 }
350 
351 /*
352  * interrupt routine for I/O completion
353  */
354 static void
355 vnode_pager_iodone(bp)
356 	struct buf *bp;
357 {
358 	bp->b_flags |= B_DONE;
359 	wakeup(bp);
360 }
361 
362 /*
363  * small block file system vnode pager input
364  */
365 static int
366 vnode_pager_input_smlfs(object, m)
367 	vm_object_t object;
368 	vm_page_t m;
369 {
370 	int i;
371 	int s;
372 	struct vnode *dp, *vp;
373 	struct buf *bp;
374 	vm_offset_t kva;
375 	int fileaddr;
376 	vm_offset_t bsize;
377 	int error = 0;
378 
379 	vp = object->handle;
380 	if (vp->v_mount == NULL)
381 		return VM_PAGER_BAD;
382 
383 	bsize = vp->v_mount->mnt_stat.f_iosize;
384 
385 
386 	VOP_BMAP(vp, 0, &dp, 0, NULL, NULL);
387 
388 	kva = vm_pager_map_page(m);
389 
390 	for (i = 0; i < PAGE_SIZE / bsize; i++) {
391 
392 		if ((vm_page_bits(IDX_TO_OFF(m->pindex) + i * bsize, bsize) & m->valid))
393 			continue;
394 
395 		fileaddr = vnode_pager_addr(vp,
396 			IDX_TO_OFF(m->pindex) + i * bsize, (int *)0);
397 		if (fileaddr != -1) {
398 			bp = getpbuf();
399 
400 			/* build a minimal buffer header */
401 			bp->b_flags = B_BUSY | B_READ | B_CALL;
402 			bp->b_iodone = vnode_pager_iodone;
403 			bp->b_proc = curproc;
404 			bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred;
405 			if (bp->b_rcred != NOCRED)
406 				crhold(bp->b_rcred);
407 			if (bp->b_wcred != NOCRED)
408 				crhold(bp->b_wcred);
409 			bp->b_data = (caddr_t) kva + i * bsize;
410 			bp->b_blkno = fileaddr;
411 			pbgetvp(dp, bp);
412 			bp->b_bcount = bsize;
413 			bp->b_bufsize = bsize;
414 
415 			/* do the input */
416 			VOP_STRATEGY(bp->b_vp, bp);
417 
418 			/* we definitely need to be at splvm here */
419 
420 			s = splvm();
421 			while ((bp->b_flags & B_DONE) == 0) {
422 				tsleep(bp, PVM, "vnsrd", 0);
423 			}
424 			splx(s);
425 			if ((bp->b_flags & B_ERROR) != 0)
426 				error = EIO;
427 
428 			/*
429 			 * free the buffer header back to the swap buffer pool
430 			 */
431 			relpbuf(bp);
432 			if (error)
433 				break;
434 
435 			vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
436 		} else {
437 			vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
438 			bzero((caddr_t) kva + i * bsize, bsize);
439 		}
440 	}
441 	vm_pager_unmap_page(kva);
442 	pmap_clear_modify(VM_PAGE_TO_PHYS(m));
443 	vm_page_flag_clear(m, PG_ZERO);
444 	if (error) {
445 		return VM_PAGER_ERROR;
446 	}
447 	return VM_PAGER_OK;
448 
449 }
450 
451 
452 /*
453  * old style vnode pager output routine
454  */
455 static int
456 vnode_pager_input_old(object, m)
457 	vm_object_t object;
458 	vm_page_t m;
459 {
460 	struct uio auio;
461 	struct iovec aiov;
462 	int error;
463 	int size;
464 	vm_offset_t kva;
465 
466 	error = 0;
467 
468 	/*
469 	 * Return failure if beyond current EOF
470 	 */
471 	if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
472 		return VM_PAGER_BAD;
473 	} else {
474 		size = PAGE_SIZE;
475 		if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
476 			size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
477 
478 		/*
479 		 * Allocate a kernel virtual address and initialize so that
480 		 * we can use VOP_READ/WRITE routines.
481 		 */
482 		kva = vm_pager_map_page(m);
483 
484 		aiov.iov_base = (caddr_t) kva;
485 		aiov.iov_len = size;
486 		auio.uio_iov = &aiov;
487 		auio.uio_iovcnt = 1;
488 		auio.uio_offset = IDX_TO_OFF(m->pindex);
489 		auio.uio_segflg = UIO_SYSSPACE;
490 		auio.uio_rw = UIO_READ;
491 		auio.uio_resid = size;
492 		auio.uio_procp = curproc;
493 
494 		error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred);
495 		if (!error) {
496 			register int count = size - auio.uio_resid;
497 
498 			if (count == 0)
499 				error = EINVAL;
500 			else if (count != PAGE_SIZE)
501 				bzero((caddr_t) kva + count, PAGE_SIZE - count);
502 		}
503 		vm_pager_unmap_page(kva);
504 	}
505 	pmap_clear_modify(VM_PAGE_TO_PHYS(m));
506 	m->dirty = 0;
507 	vm_page_flag_clear(m, PG_ZERO);
508 	if (!error)
509 		m->valid = VM_PAGE_BITS_ALL;
510 	return error ? VM_PAGER_ERROR : VM_PAGER_OK;
511 }
512 
513 /*
514  * generic vnode pager input routine
515  */
516 
517 /*
518  * EOPNOTSUPP is no longer legal.  For local media VFS's that do not
519  * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
520  * vnode_pager_generic_getpages() to implement the previous behaviour.
521  *
522  * All other FS's should use the bypass to get to the local media
523  * backing vp's VOP_GETPAGES.
524  */
525 static int
526 vnode_pager_getpages(object, m, count, reqpage)
527 	vm_object_t object;
528 	vm_page_t *m;
529 	int count;
530 	int reqpage;
531 {
532 	int rtval;
533 	struct vnode *vp;
534 	int bytes = count * PAGE_SIZE;
535 
536 	vp = object->handle;
537 	/*
538 	 * XXX temporary diagnostic message to help track stale FS code,
539 	 * Returning EOPNOTSUPP from here may make things unhappy.
540 	 */
541 	rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
542 	if (rtval == EOPNOTSUPP) {
543 	    printf("vnode_pager: *** WARNING *** stale FS getpages\n");
544 	    rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage);
545 	}
546 	return rtval;
547 }
548 
549 
550 /*
551  * This is now called from local media FS's to operate against their
552  * own vnodes if they fail to implement VOP_GETPAGES.
553  */
554 int
555 vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
556 	struct vnode *vp;
557 	vm_page_t *m;
558 	int bytecount;
559 	int reqpage;
560 {
561 	vm_object_t object;
562 	vm_offset_t kva;
563 	off_t foff, tfoff, nextoff;
564 	int i, size, bsize, first, firstaddr;
565 	struct vnode *dp;
566 	int runpg;
567 	int runend;
568 	struct buf *bp;
569 	int s;
570 	int count;
571 	int error = 0;
572 
573 	object = vp->v_object;
574 	count = bytecount / PAGE_SIZE;
575 
576 	if (vp->v_mount == NULL)
577 		return VM_PAGER_BAD;
578 
579 	bsize = vp->v_mount->mnt_stat.f_iosize;
580 
581 	/* get the UNDERLYING device for the file with VOP_BMAP() */
582 
583 	/*
584 	 * originally, we did not check for an error return value -- assuming
585 	 * an fs always has a bmap entry point -- that assumption is wrong!!!
586 	 */
587 	foff = IDX_TO_OFF(m[reqpage]->pindex);
588 
589 	/*
590 	 * if we can't bmap, use old VOP code
591 	 */
592 	if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) {
593 		for (i = 0; i < count; i++) {
594 			if (i != reqpage) {
595 				vnode_pager_freepage(m[i]);
596 			}
597 		}
598 		cnt.v_vnodein++;
599 		cnt.v_vnodepgsin++;
600 		return vnode_pager_input_old(object, m[reqpage]);
601 
602 		/*
603 		 * if the blocksize is smaller than a page size, then use
604 		 * special small filesystem code.  NFS sometimes has a small
605 		 * blocksize, but it can handle large reads itself.
606 		 */
607 	} else if ((PAGE_SIZE / bsize) > 1 &&
608 	    (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
609 		for (i = 0; i < count; i++) {
610 			if (i != reqpage) {
611 				vnode_pager_freepage(m[i]);
612 			}
613 		}
614 		cnt.v_vnodein++;
615 		cnt.v_vnodepgsin++;
616 		return vnode_pager_input_smlfs(object, m[reqpage]);
617 	}
618 	/*
619 	 * if ANY DEV_BSIZE blocks are valid on a large filesystem block
620 	 * then, the entire page is valid --
621 	 * XXX no it isn't
622 	 */
623 
624 	if (m[reqpage]->valid != VM_PAGE_BITS_ALL)
625 	    m[reqpage]->valid = 0;
626 
627 	if (m[reqpage]->valid) {
628 		m[reqpage]->valid = VM_PAGE_BITS_ALL;
629 		for (i = 0; i < count; i++) {
630 			if (i != reqpage)
631 				vnode_pager_freepage(m[i]);
632 		}
633 		return VM_PAGER_OK;
634 	}
635 
636 	/*
637 	 * here on direct device I/O
638 	 */
639 
640 	firstaddr = -1;
641 	/*
642 	 * calculate the run that includes the required page
643 	 */
644 	for(first = 0, i = 0; i < count; i = runend) {
645 		firstaddr = vnode_pager_addr(vp,
646 			IDX_TO_OFF(m[i]->pindex), &runpg);
647 		if (firstaddr == -1) {
648 			if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
649 				/* XXX no %qd in kernel. */
650 				panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx",
651 			   	 firstaddr, (u_long)(foff >> 32),
652 			   	 (u_long)(u_int32_t)foff,
653 				 (u_long)(u_int32_t)
654 				 (object->un_pager.vnp.vnp_size >> 32),
655 				 (u_long)(u_int32_t)
656 				 object->un_pager.vnp.vnp_size);
657 			}
658 			vnode_pager_freepage(m[i]);
659 			runend = i + 1;
660 			first = runend;
661 			continue;
662 		}
663 		runend = i + runpg;
664 		if (runend <= reqpage) {
665 			int j;
666 			for (j = i; j < runend; j++) {
667 				vnode_pager_freepage(m[j]);
668 			}
669 		} else {
670 			if (runpg < (count - first)) {
671 				for (i = first + runpg; i < count; i++)
672 					vnode_pager_freepage(m[i]);
673 				count = first + runpg;
674 			}
675 			break;
676 		}
677 		first = runend;
678 	}
679 
680 	/*
681 	 * the first and last page have been calculated now, move input pages
682 	 * to be zero based...
683 	 */
684 	if (first != 0) {
685 		for (i = first; i < count; i++) {
686 			m[i - first] = m[i];
687 		}
688 		count -= first;
689 		reqpage -= first;
690 	}
691 
692 	/*
693 	 * calculate the file virtual address for the transfer
694 	 */
695 	foff = IDX_TO_OFF(m[0]->pindex);
696 
697 	/*
698 	 * calculate the size of the transfer
699 	 */
700 	size = count * PAGE_SIZE;
701 	if ((foff + size) > object->un_pager.vnp.vnp_size)
702 		size = object->un_pager.vnp.vnp_size - foff;
703 
704 	/*
705 	 * round up physical size for real devices
706 	 */
707 	if (dp->v_type == VBLK || dp->v_type == VCHR)
708 		size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1);
709 
710 	bp = getpbuf();
711 	kva = (vm_offset_t) bp->b_data;
712 
713 	/*
714 	 * and map the pages to be read into the kva
715 	 */
716 	pmap_qenter(kva, m, count);
717 
718 	/* build a minimal buffer header */
719 	bp->b_flags = B_BUSY | B_READ | B_CALL;
720 	bp->b_iodone = vnode_pager_iodone;
721 	/* B_PHYS is not set, but it is nice to fill this in */
722 	bp->b_proc = curproc;
723 	bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred;
724 	if (bp->b_rcred != NOCRED)
725 		crhold(bp->b_rcred);
726 	if (bp->b_wcred != NOCRED)
727 		crhold(bp->b_wcred);
728 	bp->b_blkno = firstaddr;
729 	pbgetvp(dp, bp);
730 	bp->b_bcount = size;
731 	bp->b_bufsize = size;
732 
733 	cnt.v_vnodein++;
734 	cnt.v_vnodepgsin += count;
735 
736 	/* do the input */
737 	VOP_STRATEGY(bp->b_vp, bp);
738 
739 	s = splvm();
740 	/* we definitely need to be at splvm here */
741 
742 	while ((bp->b_flags & B_DONE) == 0) {
743 		tsleep(bp, PVM, "vnread", 0);
744 	}
745 	splx(s);
746 	if ((bp->b_flags & B_ERROR) != 0)
747 		error = EIO;
748 
749 	if (!error) {
750 		if (size != count * PAGE_SIZE)
751 			bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
752 	}
753 	pmap_qremove(kva, count);
754 
755 	/*
756 	 * free the buffer header back to the swap buffer pool
757 	 */
758 	relpbuf(bp);
759 
760 	for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
761 		vm_page_t mt;
762 
763 		nextoff = tfoff + PAGE_SIZE;
764 		mt = m[i];
765 
766 		if (nextoff <= size) {
767 			mt->valid = VM_PAGE_BITS_ALL;
768 			mt->dirty = 0;
769 			pmap_clear_modify(VM_PAGE_TO_PHYS(mt));
770 		} else {
771 			int nvalid = ((size + DEV_BSIZE - 1) - tfoff) & ~(DEV_BSIZE - 1);
772 			vm_page_set_validclean(mt, 0, nvalid);
773 		}
774 
775 		vm_page_flag_clear(mt, PG_ZERO);
776 		if (i != reqpage) {
777 
778 			/*
779 			 * whether or not to leave the page activated is up in
780 			 * the air, but we should put the page on a page queue
781 			 * somewhere. (it already is in the object). Result:
782 			 * It appears that emperical results show that
783 			 * deactivating pages is best.
784 			 */
785 
786 			/*
787 			 * just in case someone was asking for this page we
788 			 * now tell them that it is ok to use
789 			 */
790 			if (!error) {
791 				if (mt->flags & PG_WANTED)
792 					vm_page_activate(mt);
793 				else
794 					vm_page_deactivate(mt);
795 				vm_page_wakeup(mt);
796 			} else {
797 				vnode_pager_freepage(mt);
798 			}
799 		}
800 	}
801 	if (error) {
802 		printf("vnode_pager_getpages: I/O read error\n");
803 	}
804 	return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
805 }
806 
807 /*
808  * EOPNOTSUPP is no longer legal.  For local media VFS's that do not
809  * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
810  * vnode_pager_generic_putpages() to implement the previous behaviour.
811  *
812  * All other FS's should use the bypass to get to the local media
813  * backing vp's VOP_PUTPAGES.
814  */
815 static int
816 vnode_pager_putpages(object, m, count, sync, rtvals)
817 	vm_object_t object;
818 	vm_page_t *m;
819 	int count;
820 	boolean_t sync;
821 	int *rtvals;
822 {
823 	int rtval;
824 	struct vnode *vp;
825 	int bytes = count * PAGE_SIZE;
826 
827 	vp = object->handle;
828 	rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
829 	if (rtval == EOPNOTSUPP) {
830 	    printf("vnode_pager: *** WARNING *** stale FS putpages\n");
831 	    rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
832 	}
833 	return rtval;
834 }
835 
836 
837 /*
838  * This is now called from local media FS's to operate against their
839  * own vnodes if they fail to implement VOP_GETPAGES.
840  */
841 int
842 vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals)
843 	struct vnode *vp;
844 	vm_page_t *m;
845 	int bytecount;
846 	int flags;
847 	int *rtvals;
848 {
849 	int i;
850 	vm_object_t object;
851 	int count;
852 
853 	int maxsize, ncount;
854 	vm_ooffset_t poffset;
855 	struct uio auio;
856 	struct iovec aiov;
857 	int error;
858 	int ioflags;
859 
860 	object = vp->v_object;
861 	count = bytecount / PAGE_SIZE;
862 
863 	for (i = 0; i < count; i++)
864 		rtvals[i] = VM_PAGER_AGAIN;
865 
866 	if ((int) m[0]->pindex < 0) {
867 		printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n",
868 			m[0]->pindex, m[0]->dirty);
869 		rtvals[0] = VM_PAGER_BAD;
870 		return VM_PAGER_BAD;
871 	}
872 
873 	maxsize = count * PAGE_SIZE;
874 	ncount = count;
875 
876 	poffset = IDX_TO_OFF(m[0]->pindex);
877 	if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
878 		if (object->un_pager.vnp.vnp_size > poffset)
879 			maxsize = object->un_pager.vnp.vnp_size - poffset;
880 		else
881 			maxsize = 0;
882 		ncount = btoc(maxsize);
883 		if (ncount < count) {
884 			for (i = ncount; i < count; i++) {
885 				rtvals[i] = VM_PAGER_BAD;
886 			}
887 		}
888 	}
889 
890 	ioflags = IO_VMIO;
891 	ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: 0;
892 	ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
893 
894 	aiov.iov_base = (caddr_t) 0;
895 	aiov.iov_len = maxsize;
896 	auio.uio_iov = &aiov;
897 	auio.uio_iovcnt = 1;
898 	auio.uio_offset = poffset;
899 	auio.uio_segflg = UIO_NOCOPY;
900 	auio.uio_rw = UIO_WRITE;
901 	auio.uio_resid = maxsize;
902 	auio.uio_procp = (struct proc *) 0;
903 	error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred);
904 	cnt.v_vnodeout++;
905 	cnt.v_vnodepgsout += ncount;
906 
907 	if (error) {
908 		printf("vnode_pager_putpages: I/O error %d\n", error);
909 	}
910 	if (auio.uio_resid) {
911 		printf("vnode_pager_putpages: residual I/O %d at %lu\n",
912 		    auio.uio_resid, (u_long)m[0]->pindex);
913 	}
914 	for (i = 0; i < ncount; i++) {
915 		rtvals[i] = VM_PAGER_OK;
916 	}
917 	return rtvals[0];
918 }
919 
920 struct vnode *
921 vnode_pager_lock(object)
922 	vm_object_t object;
923 {
924 	struct proc *p = curproc;	/* XXX */
925 
926 	for (; object != NULL; object = object->backing_object) {
927 		if (object->type != OBJT_VNODE)
928 			continue;
929 		if (object->flags & OBJ_DEAD)
930 			return NULL;
931 
932 		while (vget(object->handle,
933 			LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) {
934 			if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE))
935 				return NULL;
936 			printf("vnode_pager_lock: retrying\n");
937 		}
938 		return object->handle;
939 	}
940 	return NULL;
941 }
942