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
| 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 1.107 1999/04/10 20:52:11 dt Exp $
| 41 * $Id: vnode_pager.c,v 1.108 1999/05/02 23:57:16 alc 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
71static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address,
72 int *run));
73static void vnode_pager_iodone __P((struct buf *bp));
74static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m));
75static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m));
76static void vnode_pager_dealloc __P((vm_object_t));
77static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
78static void vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *));
79static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
80
81struct 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
91int vnode_pbuf_freecnt = -1; /* start out unlimited */
92
93
94/*
95 * Allocate (or lookup) pager for a vnode.
96 * Handle is a vnode pointer.
97 */
98vm_object_t
99vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
100 vm_ooffset_t offset)
101{
102 vm_object_t object;
103 struct vnode *vp;
104
105 /*
106 * Pageout to vnode, no can do yet.
107 */
108 if (handle == NULL)
109 return (NULL);
110
111 /*
112 * XXX hack - This initialization should be put somewhere else.
113 */
114 if (vnode_pbuf_freecnt < 0) {
115 vnode_pbuf_freecnt = nswbuf / 2 + 1;
116 }
117
118 vp = (struct vnode *) handle;
119
120 /*
121 * Prevent race condition when allocating the object. This
122 * can happen with NFS vnodes since the nfsnode isn't locked.
123 */
124 while (vp->v_flag & VOLOCK) {
125 vp->v_flag |= VOWANT;
126 tsleep(vp, PVM, "vnpobj", 0);
127 }
128 vp->v_flag |= VOLOCK;
129
130 /*
131 * If the object is being terminated, wait for it to
132 * go away.
133 */
134 while (((object = vp->v_object) != NULL) &&
135 (object->flags & OBJ_DEAD)) {
136 tsleep(object, PVM, "vadead", 0);
137 }
138
139 if (vp->v_usecount == 0)
140 panic("vnode_pager_alloc: no vnode reference");
141
142 if (object == NULL) {
143 /*
144 * And an object of the appropriate size
145 */
146 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
147 object->flags = 0;
148
149 object->un_pager.vnp.vnp_size = size;
150
151 object->handle = handle;
152 vp->v_object = object;
153 vp->v_usecount++;
154 } else {
155 object->ref_count++;
156 vp->v_usecount++;
157 }
158
159 vp->v_flag &= ~VOLOCK;
160 if (vp->v_flag & VOWANT) {
161 vp->v_flag &= ~VOWANT;
162 wakeup(vp);
163 }
164 return (object);
165}
166
167static void
168vnode_pager_dealloc(object)
169 vm_object_t object;
170{
171 register struct vnode *vp = object->handle;
172
173 if (vp == NULL)
174 panic("vnode_pager_dealloc: pager already dealloced");
175
176 vm_object_pip_wait(object, "vnpdea");
177
178 object->handle = NULL;
179 object->type = OBJT_DEAD;
180 vp->v_object = NULL;
181 vp->v_flag &= ~(VTEXT | VOBJBUF);
182}
183
184static boolean_t
185vnode_pager_haspage(object, pindex, before, after)
186 vm_object_t object;
187 vm_pindex_t pindex;
188 int *before;
189 int *after;
190{
191 struct vnode *vp = object->handle;
192 daddr_t bn;
193 int err;
194 daddr_t reqblock;
195 int poff;
196 int bsize;
197 int pagesperblock, blocksperpage;
198
199 if ((vp == NULL) || (vp->v_flag & VDOOMED))
200 return FALSE;
201
202 /*
203 * If filesystem no longer mounted or offset beyond end of file we do
204 * not have the page.
205 */
206 if ((vp->v_mount == NULL) ||
207 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
208 return FALSE;
209
210 bsize = vp->v_mount->mnt_stat.f_iosize;
211 pagesperblock = bsize / PAGE_SIZE;
212 blocksperpage = 0;
213 if (pagesperblock > 0) {
214 reqblock = pindex / pagesperblock;
215 } else {
216 blocksperpage = (PAGE_SIZE / bsize);
217 reqblock = pindex * blocksperpage;
218 }
219 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn,
220 after, before);
221 if (err)
222 return TRUE;
223 if ( bn == -1)
224 return FALSE;
225 if (pagesperblock > 0) {
226 poff = pindex - (reqblock * pagesperblock);
227 if (before) {
228 *before *= pagesperblock;
229 *before += poff;
230 }
231 if (after) {
232 int numafter;
233 *after *= pagesperblock;
234 numafter = pagesperblock - (poff + 1);
235 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) {
236 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex)));
237 }
238 *after += numafter;
239 }
240 } else {
241 if (before) {
242 *before /= blocksperpage;
243 }
244
245 if (after) {
246 *after /= blocksperpage;
247 }
248 }
249 return TRUE;
250}
251
252/*
253 * Lets the VM system know about a change in size for a file.
254 * We adjust our own internal size and flush any cached pages in
255 * the associated object that are affected by the size change.
256 *
257 * Note: this routine may be invoked as a result of a pager put
258 * operation (possibly at object termination time), so we must be careful.
259 */
260void
261vnode_pager_setsize(vp, nsize)
262 struct vnode *vp;
263 vm_ooffset_t nsize;
264{
265 vm_pindex_t nobjsize;
266 vm_object_t object = vp->v_object;
267
268 if (object == NULL)
269 return;
270
271 /*
272 * Hasn't changed size
273 */
274 if (nsize == object->un_pager.vnp.vnp_size)
275 return;
276
277 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
278
279 /*
280 * File has shrunk. Toss any cached pages beyond the new EOF.
281 */
282 if (nsize < object->un_pager.vnp.vnp_size) {
283 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size);
284 if (nobjsize < object->size) {
285 vm_object_page_remove(object, nobjsize, object->size,
286 FALSE);
287 }
288 /*
289 * this gets rid of garbage at the end of a page that is now
290 * only partially backed by the vnode...
291 */
292 if (nsize & PAGE_MASK) {
293 vm_offset_t kva;
294 vm_page_t m;
295
296 m = vm_page_lookup(object, OFF_TO_IDX(nsize));
297 if (m) {
298 kva = vm_pager_map_page(m);
299 bzero((caddr_t) kva + (nsize & PAGE_MASK),
300 (int) (round_page(nsize) - nsize));
301 vm_pager_unmap_page(kva);
302 }
303 }
304 }
305 object->un_pager.vnp.vnp_size = nsize;
306 object->size = nobjsize;
307}
308
309void
310vnode_pager_freepage(m)
311 vm_page_t m;
312{
313 vm_page_free(m);
314}
315
316/*
317 * calculate the linear (byte) disk address of specified virtual
318 * file address
319 */
320static vm_offset_t
321vnode_pager_addr(vp, address, run)
322 struct vnode *vp;
323 vm_ooffset_t address;
324 int *run;
325{
326 int rtaddress;
327 int bsize;
328 daddr_t block;
329 struct vnode *rtvp;
330 int err;
331 daddr_t vblock;
332 int voffset;
333
334 if ((int) address < 0)
335 return -1;
336
337 if (vp->v_mount == NULL)
338 return -1;
339
340 bsize = vp->v_mount->mnt_stat.f_iosize;
341 vblock = address / bsize;
342 voffset = address % bsize;
343
344 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL);
345
346 if (err || (block == -1))
347 rtaddress = -1;
348 else {
349 rtaddress = block + voffset / DEV_BSIZE;
350 if( run) {
351 *run += 1;
352 *run *= bsize/PAGE_SIZE;
353 *run -= voffset/PAGE_SIZE;
354 }
355 }
356
357 return rtaddress;
358}
359
360/*
361 * interrupt routine for I/O completion
362 */
363static void
364vnode_pager_iodone(bp)
365 struct buf *bp;
366{
367 bp->b_flags |= B_DONE;
368 wakeup(bp);
369}
370
371/*
372 * small block file system vnode pager input
373 */
374static int
375vnode_pager_input_smlfs(object, m)
376 vm_object_t object;
377 vm_page_t m;
378{
379 int i;
380 int s;
381 struct vnode *dp, *vp;
382 struct buf *bp;
383 vm_offset_t kva;
384 int fileaddr;
385 vm_offset_t bsize;
386 int error = 0;
387
388 vp = object->handle;
389 if (vp->v_mount == NULL)
390 return VM_PAGER_BAD;
391
392 bsize = vp->v_mount->mnt_stat.f_iosize;
393
394
395 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL);
396
397 kva = vm_pager_map_page(m);
398
399 for (i = 0; i < PAGE_SIZE / bsize; i++) {
400
401 if (vm_page_bits(i * bsize, bsize) & m->valid)
402 continue;
403
404 fileaddr = vnode_pager_addr(vp,
405 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0);
406 if (fileaddr != -1) {
407 bp = getpbuf(&vnode_pbuf_freecnt);
408
409 /* build a minimal buffer header */
410 bp->b_flags = B_BUSY | B_READ | B_CALL;
411 bp->b_iodone = vnode_pager_iodone;
| 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
71static vm_offset_t vnode_pager_addr __P((struct vnode *vp, vm_ooffset_t address,
72 int *run));
73static void vnode_pager_iodone __P((struct buf *bp));
74static int vnode_pager_input_smlfs __P((vm_object_t object, vm_page_t m));
75static int vnode_pager_input_old __P((vm_object_t object, vm_page_t m));
76static void vnode_pager_dealloc __P((vm_object_t));
77static int vnode_pager_getpages __P((vm_object_t, vm_page_t *, int, int));
78static void vnode_pager_putpages __P((vm_object_t, vm_page_t *, int, boolean_t, int *));
79static boolean_t vnode_pager_haspage __P((vm_object_t, vm_pindex_t, int *, int *));
80
81struct 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
91int vnode_pbuf_freecnt = -1; /* start out unlimited */
92
93
94/*
95 * Allocate (or lookup) pager for a vnode.
96 * Handle is a vnode pointer.
97 */
98vm_object_t
99vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
100 vm_ooffset_t offset)
101{
102 vm_object_t object;
103 struct vnode *vp;
104
105 /*
106 * Pageout to vnode, no can do yet.
107 */
108 if (handle == NULL)
109 return (NULL);
110
111 /*
112 * XXX hack - This initialization should be put somewhere else.
113 */
114 if (vnode_pbuf_freecnt < 0) {
115 vnode_pbuf_freecnt = nswbuf / 2 + 1;
116 }
117
118 vp = (struct vnode *) handle;
119
120 /*
121 * Prevent race condition when allocating the object. This
122 * can happen with NFS vnodes since the nfsnode isn't locked.
123 */
124 while (vp->v_flag & VOLOCK) {
125 vp->v_flag |= VOWANT;
126 tsleep(vp, PVM, "vnpobj", 0);
127 }
128 vp->v_flag |= VOLOCK;
129
130 /*
131 * If the object is being terminated, wait for it to
132 * go away.
133 */
134 while (((object = vp->v_object) != NULL) &&
135 (object->flags & OBJ_DEAD)) {
136 tsleep(object, PVM, "vadead", 0);
137 }
138
139 if (vp->v_usecount == 0)
140 panic("vnode_pager_alloc: no vnode reference");
141
142 if (object == NULL) {
143 /*
144 * And an object of the appropriate size
145 */
146 object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
147 object->flags = 0;
148
149 object->un_pager.vnp.vnp_size = size;
150
151 object->handle = handle;
152 vp->v_object = object;
153 vp->v_usecount++;
154 } else {
155 object->ref_count++;
156 vp->v_usecount++;
157 }
158
159 vp->v_flag &= ~VOLOCK;
160 if (vp->v_flag & VOWANT) {
161 vp->v_flag &= ~VOWANT;
162 wakeup(vp);
163 }
164 return (object);
165}
166
167static void
168vnode_pager_dealloc(object)
169 vm_object_t object;
170{
171 register struct vnode *vp = object->handle;
172
173 if (vp == NULL)
174 panic("vnode_pager_dealloc: pager already dealloced");
175
176 vm_object_pip_wait(object, "vnpdea");
177
178 object->handle = NULL;
179 object->type = OBJT_DEAD;
180 vp->v_object = NULL;
181 vp->v_flag &= ~(VTEXT | VOBJBUF);
182}
183
184static boolean_t
185vnode_pager_haspage(object, pindex, before, after)
186 vm_object_t object;
187 vm_pindex_t pindex;
188 int *before;
189 int *after;
190{
191 struct vnode *vp = object->handle;
192 daddr_t bn;
193 int err;
194 daddr_t reqblock;
195 int poff;
196 int bsize;
197 int pagesperblock, blocksperpage;
198
199 if ((vp == NULL) || (vp->v_flag & VDOOMED))
200 return FALSE;
201
202 /*
203 * If filesystem no longer mounted or offset beyond end of file we do
204 * not have the page.
205 */
206 if ((vp->v_mount == NULL) ||
207 (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size))
208 return FALSE;
209
210 bsize = vp->v_mount->mnt_stat.f_iosize;
211 pagesperblock = bsize / PAGE_SIZE;
212 blocksperpage = 0;
213 if (pagesperblock > 0) {
214 reqblock = pindex / pagesperblock;
215 } else {
216 blocksperpage = (PAGE_SIZE / bsize);
217 reqblock = pindex * blocksperpage;
218 }
219 err = VOP_BMAP(vp, reqblock, (struct vnode **) 0, &bn,
220 after, before);
221 if (err)
222 return TRUE;
223 if ( bn == -1)
224 return FALSE;
225 if (pagesperblock > 0) {
226 poff = pindex - (reqblock * pagesperblock);
227 if (before) {
228 *before *= pagesperblock;
229 *before += poff;
230 }
231 if (after) {
232 int numafter;
233 *after *= pagesperblock;
234 numafter = pagesperblock - (poff + 1);
235 if (IDX_TO_OFF(pindex + numafter) > object->un_pager.vnp.vnp_size) {
236 numafter = OFF_TO_IDX((object->un_pager.vnp.vnp_size - IDX_TO_OFF(pindex)));
237 }
238 *after += numafter;
239 }
240 } else {
241 if (before) {
242 *before /= blocksperpage;
243 }
244
245 if (after) {
246 *after /= blocksperpage;
247 }
248 }
249 return TRUE;
250}
251
252/*
253 * Lets the VM system know about a change in size for a file.
254 * We adjust our own internal size and flush any cached pages in
255 * the associated object that are affected by the size change.
256 *
257 * Note: this routine may be invoked as a result of a pager put
258 * operation (possibly at object termination time), so we must be careful.
259 */
260void
261vnode_pager_setsize(vp, nsize)
262 struct vnode *vp;
263 vm_ooffset_t nsize;
264{
265 vm_pindex_t nobjsize;
266 vm_object_t object = vp->v_object;
267
268 if (object == NULL)
269 return;
270
271 /*
272 * Hasn't changed size
273 */
274 if (nsize == object->un_pager.vnp.vnp_size)
275 return;
276
277 nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
278
279 /*
280 * File has shrunk. Toss any cached pages beyond the new EOF.
281 */
282 if (nsize < object->un_pager.vnp.vnp_size) {
283 vm_freeze_copyopts(object, OFF_TO_IDX(nsize), object->size);
284 if (nobjsize < object->size) {
285 vm_object_page_remove(object, nobjsize, object->size,
286 FALSE);
287 }
288 /*
289 * this gets rid of garbage at the end of a page that is now
290 * only partially backed by the vnode...
291 */
292 if (nsize & PAGE_MASK) {
293 vm_offset_t kva;
294 vm_page_t m;
295
296 m = vm_page_lookup(object, OFF_TO_IDX(nsize));
297 if (m) {
298 kva = vm_pager_map_page(m);
299 bzero((caddr_t) kva + (nsize & PAGE_MASK),
300 (int) (round_page(nsize) - nsize));
301 vm_pager_unmap_page(kva);
302 }
303 }
304 }
305 object->un_pager.vnp.vnp_size = nsize;
306 object->size = nobjsize;
307}
308
309void
310vnode_pager_freepage(m)
311 vm_page_t m;
312{
313 vm_page_free(m);
314}
315
316/*
317 * calculate the linear (byte) disk address of specified virtual
318 * file address
319 */
320static vm_offset_t
321vnode_pager_addr(vp, address, run)
322 struct vnode *vp;
323 vm_ooffset_t address;
324 int *run;
325{
326 int rtaddress;
327 int bsize;
328 daddr_t block;
329 struct vnode *rtvp;
330 int err;
331 daddr_t vblock;
332 int voffset;
333
334 if ((int) address < 0)
335 return -1;
336
337 if (vp->v_mount == NULL)
338 return -1;
339
340 bsize = vp->v_mount->mnt_stat.f_iosize;
341 vblock = address / bsize;
342 voffset = address % bsize;
343
344 err = VOP_BMAP(vp, vblock, &rtvp, &block, run, NULL);
345
346 if (err || (block == -1))
347 rtaddress = -1;
348 else {
349 rtaddress = block + voffset / DEV_BSIZE;
350 if( run) {
351 *run += 1;
352 *run *= bsize/PAGE_SIZE;
353 *run -= voffset/PAGE_SIZE;
354 }
355 }
356
357 return rtaddress;
358}
359
360/*
361 * interrupt routine for I/O completion
362 */
363static void
364vnode_pager_iodone(bp)
365 struct buf *bp;
366{
367 bp->b_flags |= B_DONE;
368 wakeup(bp);
369}
370
371/*
372 * small block file system vnode pager input
373 */
374static int
375vnode_pager_input_smlfs(object, m)
376 vm_object_t object;
377 vm_page_t m;
378{
379 int i;
380 int s;
381 struct vnode *dp, *vp;
382 struct buf *bp;
383 vm_offset_t kva;
384 int fileaddr;
385 vm_offset_t bsize;
386 int error = 0;
387
388 vp = object->handle;
389 if (vp->v_mount == NULL)
390 return VM_PAGER_BAD;
391
392 bsize = vp->v_mount->mnt_stat.f_iosize;
393
394
395 VOP_BMAP(vp, 0, &dp, 0, NULL, NULL);
396
397 kva = vm_pager_map_page(m);
398
399 for (i = 0; i < PAGE_SIZE / bsize; i++) {
400
401 if (vm_page_bits(i * bsize, bsize) & m->valid)
402 continue;
403
404 fileaddr = vnode_pager_addr(vp,
405 IDX_TO_OFF(m->pindex) + i * bsize, (int *)0);
406 if (fileaddr != -1) {
407 bp = getpbuf(&vnode_pbuf_freecnt);
408
409 /* build a minimal buffer header */
410 bp->b_flags = B_BUSY | B_READ | B_CALL;
411 bp->b_iodone = vnode_pager_iodone;
|
412 bp->b_proc = curproc;
413 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred;
| 412 bp->b_rcred = bp->b_wcred = curproc->p_ucred;
|
414 if (bp->b_rcred != NOCRED)
415 crhold(bp->b_rcred);
416 if (bp->b_wcred != NOCRED)
417 crhold(bp->b_wcred);
418 bp->b_data = (caddr_t) kva + i * bsize;
419 bp->b_blkno = fileaddr;
420 pbgetvp(dp, bp);
421 bp->b_bcount = bsize;
422 bp->b_bufsize = bsize;
423
424 /* do the input */
425 VOP_STRATEGY(bp->b_vp, bp);
426
427 /* we definitely need to be at splvm here */
428
429 s = splvm();
430 while ((bp->b_flags & B_DONE) == 0) {
431 tsleep(bp, PVM, "vnsrd", 0);
432 }
433 splx(s);
434 if ((bp->b_flags & B_ERROR) != 0)
435 error = EIO;
436
437 /*
438 * free the buffer header back to the swap buffer pool
439 */
440 relpbuf(bp, &vnode_pbuf_freecnt);
441 if (error)
442 break;
443
444 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
445 } else {
446 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
447 bzero((caddr_t) kva + i * bsize, bsize);
448 }
449 }
450 vm_pager_unmap_page(kva);
451 pmap_clear_modify(VM_PAGE_TO_PHYS(m));
452 vm_page_flag_clear(m, PG_ZERO);
453 if (error) {
454 return VM_PAGER_ERROR;
455 }
456 return VM_PAGER_OK;
457
458}
459
460
461/*
462 * old style vnode pager output routine
463 */
464static int
465vnode_pager_input_old(object, m)
466 vm_object_t object;
467 vm_page_t m;
468{
469 struct uio auio;
470 struct iovec aiov;
471 int error;
472 int size;
473 vm_offset_t kva;
474
475 error = 0;
476
477 /*
478 * Return failure if beyond current EOF
479 */
480 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
481 return VM_PAGER_BAD;
482 } else {
483 size = PAGE_SIZE;
484 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
485 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
486
487 /*
488 * Allocate a kernel virtual address and initialize so that
489 * we can use VOP_READ/WRITE routines.
490 */
491 kva = vm_pager_map_page(m);
492
493 aiov.iov_base = (caddr_t) kva;
494 aiov.iov_len = size;
495 auio.uio_iov = &aiov;
496 auio.uio_iovcnt = 1;
497 auio.uio_offset = IDX_TO_OFF(m->pindex);
498 auio.uio_segflg = UIO_SYSSPACE;
499 auio.uio_rw = UIO_READ;
500 auio.uio_resid = size;
501 auio.uio_procp = curproc;
502
503 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred);
504 if (!error) {
505 register int count = size - auio.uio_resid;
506
507 if (count == 0)
508 error = EINVAL;
509 else if (count != PAGE_SIZE)
510 bzero((caddr_t) kva + count, PAGE_SIZE - count);
511 }
512 vm_pager_unmap_page(kva);
513 }
514 pmap_clear_modify(VM_PAGE_TO_PHYS(m));
515 m->dirty = 0;
516 vm_page_flag_clear(m, PG_ZERO);
517 if (!error)
518 m->valid = VM_PAGE_BITS_ALL;
519 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
520}
521
522/*
523 * generic vnode pager input routine
524 */
525
526/*
527 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
528 * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
529 * vnode_pager_generic_getpages() to implement the previous behaviour.
530 *
531 * All other FS's should use the bypass to get to the local media
532 * backing vp's VOP_GETPAGES.
533 */
534static int
535vnode_pager_getpages(object, m, count, reqpage)
536 vm_object_t object;
537 vm_page_t *m;
538 int count;
539 int reqpage;
540{
541 int rtval;
542 struct vnode *vp;
543 int bytes = count * PAGE_SIZE;
544
545 vp = object->handle;
546 /*
547 * XXX temporary diagnostic message to help track stale FS code,
548 * Returning EOPNOTSUPP from here may make things unhappy.
549 */
550 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
551 if (rtval == EOPNOTSUPP) {
552 printf("vnode_pager: *** WARNING *** stale FS getpages\n");
553 rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage);
554 }
555 return rtval;
556}
557
558
559/*
560 * This is now called from local media FS's to operate against their
561 * own vnodes if they fail to implement VOP_GETPAGES.
562 */
563int
564vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
565 struct vnode *vp;
566 vm_page_t *m;
567 int bytecount;
568 int reqpage;
569{
570 vm_object_t object;
571 vm_offset_t kva;
572 off_t foff, tfoff, nextoff;
573 int i, size, bsize, first, firstaddr;
574 struct vnode *dp;
575 int runpg;
576 int runend;
577 struct buf *bp;
578 int s;
579 int count;
580 int error = 0;
581
582 object = vp->v_object;
583 count = bytecount / PAGE_SIZE;
584
585 if (vp->v_mount == NULL)
586 return VM_PAGER_BAD;
587
588 bsize = vp->v_mount->mnt_stat.f_iosize;
589
590 /* get the UNDERLYING device for the file with VOP_BMAP() */
591
592 /*
593 * originally, we did not check for an error return value -- assuming
594 * an fs always has a bmap entry point -- that assumption is wrong!!!
595 */
596 foff = IDX_TO_OFF(m[reqpage]->pindex);
597
598 /*
599 * if we can't bmap, use old VOP code
600 */
601 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) {
602 for (i = 0; i < count; i++) {
603 if (i != reqpage) {
604 vnode_pager_freepage(m[i]);
605 }
606 }
607 cnt.v_vnodein++;
608 cnt.v_vnodepgsin++;
609 return vnode_pager_input_old(object, m[reqpage]);
610
611 /*
612 * if the blocksize is smaller than a page size, then use
613 * special small filesystem code. NFS sometimes has a small
614 * blocksize, but it can handle large reads itself.
615 */
616 } else if ((PAGE_SIZE / bsize) > 1 &&
617 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
618 for (i = 0; i < count; i++) {
619 if (i != reqpage) {
620 vnode_pager_freepage(m[i]);
621 }
622 }
623 cnt.v_vnodein++;
624 cnt.v_vnodepgsin++;
625 return vnode_pager_input_smlfs(object, m[reqpage]);
626 }
627
628 /*
629 * If we have a completely valid page available to us, we can
630 * clean up and return. Otherwise we have to re-read the
631 * media.
632 */
633
634 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
635 for (i = 0; i < count; i++) {
636 if (i != reqpage)
637 vnode_pager_freepage(m[i]);
638 }
639 return VM_PAGER_OK;
640 }
641 m[reqpage]->valid = 0;
642
643 /*
644 * here on direct device I/O
645 */
646
647 firstaddr = -1;
648 /*
649 * calculate the run that includes the required page
650 */
651 for(first = 0, i = 0; i < count; i = runend) {
652 firstaddr = vnode_pager_addr(vp,
653 IDX_TO_OFF(m[i]->pindex), &runpg);
654 if (firstaddr == -1) {
655 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
656 /* XXX no %qd in kernel. */
657 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx",
658 firstaddr, (u_long)(foff >> 32),
659 (u_long)(u_int32_t)foff,
660 (u_long)(u_int32_t)
661 (object->un_pager.vnp.vnp_size >> 32),
662 (u_long)(u_int32_t)
663 object->un_pager.vnp.vnp_size);
664 }
665 vnode_pager_freepage(m[i]);
666 runend = i + 1;
667 first = runend;
668 continue;
669 }
670 runend = i + runpg;
671 if (runend <= reqpage) {
672 int j;
673 for (j = i; j < runend; j++) {
674 vnode_pager_freepage(m[j]);
675 }
676 } else {
677 if (runpg < (count - first)) {
678 for (i = first + runpg; i < count; i++)
679 vnode_pager_freepage(m[i]);
680 count = first + runpg;
681 }
682 break;
683 }
684 first = runend;
685 }
686
687 /*
688 * the first and last page have been calculated now, move input pages
689 * to be zero based...
690 */
691 if (first != 0) {
692 for (i = first; i < count; i++) {
693 m[i - first] = m[i];
694 }
695 count -= first;
696 reqpage -= first;
697 }
698
699 /*
700 * calculate the file virtual address for the transfer
701 */
702 foff = IDX_TO_OFF(m[0]->pindex);
703
704 /*
705 * calculate the size of the transfer
706 */
707 size = count * PAGE_SIZE;
708 if ((foff + size) > object->un_pager.vnp.vnp_size)
709 size = object->un_pager.vnp.vnp_size - foff;
710
711 /*
712 * round up physical size for real devices
713 */
714 if (dp->v_type == VBLK || dp->v_type == VCHR)
715 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1);
716
717 bp = getpbuf(&vnode_pbuf_freecnt);
718 kva = (vm_offset_t) bp->b_data;
719
720 /*
721 * and map the pages to be read into the kva
722 */
723 pmap_qenter(kva, m, count);
724
725 /* build a minimal buffer header */
726 bp->b_flags = B_BUSY | B_READ | B_CALL;
727 bp->b_iodone = vnode_pager_iodone;
728 /* B_PHYS is not set, but it is nice to fill this in */
| 413 if (bp->b_rcred != NOCRED)
414 crhold(bp->b_rcred);
415 if (bp->b_wcred != NOCRED)
416 crhold(bp->b_wcred);
417 bp->b_data = (caddr_t) kva + i * bsize;
418 bp->b_blkno = fileaddr;
419 pbgetvp(dp, bp);
420 bp->b_bcount = bsize;
421 bp->b_bufsize = bsize;
422
423 /* do the input */
424 VOP_STRATEGY(bp->b_vp, bp);
425
426 /* we definitely need to be at splvm here */
427
428 s = splvm();
429 while ((bp->b_flags & B_DONE) == 0) {
430 tsleep(bp, PVM, "vnsrd", 0);
431 }
432 splx(s);
433 if ((bp->b_flags & B_ERROR) != 0)
434 error = EIO;
435
436 /*
437 * free the buffer header back to the swap buffer pool
438 */
439 relpbuf(bp, &vnode_pbuf_freecnt);
440 if (error)
441 break;
442
443 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
444 } else {
445 vm_page_set_validclean(m, (i * bsize) & PAGE_MASK, bsize);
446 bzero((caddr_t) kva + i * bsize, bsize);
447 }
448 }
449 vm_pager_unmap_page(kva);
450 pmap_clear_modify(VM_PAGE_TO_PHYS(m));
451 vm_page_flag_clear(m, PG_ZERO);
452 if (error) {
453 return VM_PAGER_ERROR;
454 }
455 return VM_PAGER_OK;
456
457}
458
459
460/*
461 * old style vnode pager output routine
462 */
463static int
464vnode_pager_input_old(object, m)
465 vm_object_t object;
466 vm_page_t m;
467{
468 struct uio auio;
469 struct iovec aiov;
470 int error;
471 int size;
472 vm_offset_t kva;
473
474 error = 0;
475
476 /*
477 * Return failure if beyond current EOF
478 */
479 if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
480 return VM_PAGER_BAD;
481 } else {
482 size = PAGE_SIZE;
483 if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
484 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
485
486 /*
487 * Allocate a kernel virtual address and initialize so that
488 * we can use VOP_READ/WRITE routines.
489 */
490 kva = vm_pager_map_page(m);
491
492 aiov.iov_base = (caddr_t) kva;
493 aiov.iov_len = size;
494 auio.uio_iov = &aiov;
495 auio.uio_iovcnt = 1;
496 auio.uio_offset = IDX_TO_OFF(m->pindex);
497 auio.uio_segflg = UIO_SYSSPACE;
498 auio.uio_rw = UIO_READ;
499 auio.uio_resid = size;
500 auio.uio_procp = curproc;
501
502 error = VOP_READ(object->handle, &auio, 0, curproc->p_ucred);
503 if (!error) {
504 register int count = size - auio.uio_resid;
505
506 if (count == 0)
507 error = EINVAL;
508 else if (count != PAGE_SIZE)
509 bzero((caddr_t) kva + count, PAGE_SIZE - count);
510 }
511 vm_pager_unmap_page(kva);
512 }
513 pmap_clear_modify(VM_PAGE_TO_PHYS(m));
514 m->dirty = 0;
515 vm_page_flag_clear(m, PG_ZERO);
516 if (!error)
517 m->valid = VM_PAGE_BITS_ALL;
518 return error ? VM_PAGER_ERROR : VM_PAGER_OK;
519}
520
521/*
522 * generic vnode pager input routine
523 */
524
525/*
526 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
527 * implement their own VOP_GETPAGES, their VOP_GETPAGES should call to
528 * vnode_pager_generic_getpages() to implement the previous behaviour.
529 *
530 * All other FS's should use the bypass to get to the local media
531 * backing vp's VOP_GETPAGES.
532 */
533static int
534vnode_pager_getpages(object, m, count, reqpage)
535 vm_object_t object;
536 vm_page_t *m;
537 int count;
538 int reqpage;
539{
540 int rtval;
541 struct vnode *vp;
542 int bytes = count * PAGE_SIZE;
543
544 vp = object->handle;
545 /*
546 * XXX temporary diagnostic message to help track stale FS code,
547 * Returning EOPNOTSUPP from here may make things unhappy.
548 */
549 rtval = VOP_GETPAGES(vp, m, bytes, reqpage, 0);
550 if (rtval == EOPNOTSUPP) {
551 printf("vnode_pager: *** WARNING *** stale FS getpages\n");
552 rtval = vnode_pager_generic_getpages( vp, m, bytes, reqpage);
553 }
554 return rtval;
555}
556
557
558/*
559 * This is now called from local media FS's to operate against their
560 * own vnodes if they fail to implement VOP_GETPAGES.
561 */
562int
563vnode_pager_generic_getpages(vp, m, bytecount, reqpage)
564 struct vnode *vp;
565 vm_page_t *m;
566 int bytecount;
567 int reqpage;
568{
569 vm_object_t object;
570 vm_offset_t kva;
571 off_t foff, tfoff, nextoff;
572 int i, size, bsize, first, firstaddr;
573 struct vnode *dp;
574 int runpg;
575 int runend;
576 struct buf *bp;
577 int s;
578 int count;
579 int error = 0;
580
581 object = vp->v_object;
582 count = bytecount / PAGE_SIZE;
583
584 if (vp->v_mount == NULL)
585 return VM_PAGER_BAD;
586
587 bsize = vp->v_mount->mnt_stat.f_iosize;
588
589 /* get the UNDERLYING device for the file with VOP_BMAP() */
590
591 /*
592 * originally, we did not check for an error return value -- assuming
593 * an fs always has a bmap entry point -- that assumption is wrong!!!
594 */
595 foff = IDX_TO_OFF(m[reqpage]->pindex);
596
597 /*
598 * if we can't bmap, use old VOP code
599 */
600 if (VOP_BMAP(vp, 0, &dp, 0, NULL, NULL)) {
601 for (i = 0; i < count; i++) {
602 if (i != reqpage) {
603 vnode_pager_freepage(m[i]);
604 }
605 }
606 cnt.v_vnodein++;
607 cnt.v_vnodepgsin++;
608 return vnode_pager_input_old(object, m[reqpage]);
609
610 /*
611 * if the blocksize is smaller than a page size, then use
612 * special small filesystem code. NFS sometimes has a small
613 * blocksize, but it can handle large reads itself.
614 */
615 } else if ((PAGE_SIZE / bsize) > 1 &&
616 (vp->v_mount->mnt_stat.f_type != nfs_mount_type)) {
617 for (i = 0; i < count; i++) {
618 if (i != reqpage) {
619 vnode_pager_freepage(m[i]);
620 }
621 }
622 cnt.v_vnodein++;
623 cnt.v_vnodepgsin++;
624 return vnode_pager_input_smlfs(object, m[reqpage]);
625 }
626
627 /*
628 * If we have a completely valid page available to us, we can
629 * clean up and return. Otherwise we have to re-read the
630 * media.
631 */
632
633 if (m[reqpage]->valid == VM_PAGE_BITS_ALL) {
634 for (i = 0; i < count; i++) {
635 if (i != reqpage)
636 vnode_pager_freepage(m[i]);
637 }
638 return VM_PAGER_OK;
639 }
640 m[reqpage]->valid = 0;
641
642 /*
643 * here on direct device I/O
644 */
645
646 firstaddr = -1;
647 /*
648 * calculate the run that includes the required page
649 */
650 for(first = 0, i = 0; i < count; i = runend) {
651 firstaddr = vnode_pager_addr(vp,
652 IDX_TO_OFF(m[i]->pindex), &runpg);
653 if (firstaddr == -1) {
654 if (i == reqpage && foff < object->un_pager.vnp.vnp_size) {
655 /* XXX no %qd in kernel. */
656 panic("vnode_pager_getpages: unexpected missing page: firstaddr: %d, foff: 0x%lx%08lx, vnp_size: 0x%lx%08lx",
657 firstaddr, (u_long)(foff >> 32),
658 (u_long)(u_int32_t)foff,
659 (u_long)(u_int32_t)
660 (object->un_pager.vnp.vnp_size >> 32),
661 (u_long)(u_int32_t)
662 object->un_pager.vnp.vnp_size);
663 }
664 vnode_pager_freepage(m[i]);
665 runend = i + 1;
666 first = runend;
667 continue;
668 }
669 runend = i + runpg;
670 if (runend <= reqpage) {
671 int j;
672 for (j = i; j < runend; j++) {
673 vnode_pager_freepage(m[j]);
674 }
675 } else {
676 if (runpg < (count - first)) {
677 for (i = first + runpg; i < count; i++)
678 vnode_pager_freepage(m[i]);
679 count = first + runpg;
680 }
681 break;
682 }
683 first = runend;
684 }
685
686 /*
687 * the first and last page have been calculated now, move input pages
688 * to be zero based...
689 */
690 if (first != 0) {
691 for (i = first; i < count; i++) {
692 m[i - first] = m[i];
693 }
694 count -= first;
695 reqpage -= first;
696 }
697
698 /*
699 * calculate the file virtual address for the transfer
700 */
701 foff = IDX_TO_OFF(m[0]->pindex);
702
703 /*
704 * calculate the size of the transfer
705 */
706 size = count * PAGE_SIZE;
707 if ((foff + size) > object->un_pager.vnp.vnp_size)
708 size = object->un_pager.vnp.vnp_size - foff;
709
710 /*
711 * round up physical size for real devices
712 */
713 if (dp->v_type == VBLK || dp->v_type == VCHR)
714 size = (size + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1);
715
716 bp = getpbuf(&vnode_pbuf_freecnt);
717 kva = (vm_offset_t) bp->b_data;
718
719 /*
720 * and map the pages to be read into the kva
721 */
722 pmap_qenter(kva, m, count);
723
724 /* build a minimal buffer header */
725 bp->b_flags = B_BUSY | B_READ | B_CALL;
726 bp->b_iodone = vnode_pager_iodone;
727 /* B_PHYS is not set, but it is nice to fill this in */
|
729 bp->b_proc = curproc;
730 bp->b_rcred = bp->b_wcred = bp->b_proc->p_ucred;
| 728 bp->b_rcred = bp->b_wcred = curproc->p_ucred;
|
731 if (bp->b_rcred != NOCRED)
732 crhold(bp->b_rcred);
733 if (bp->b_wcred != NOCRED)
734 crhold(bp->b_wcred);
735 bp->b_blkno = firstaddr;
736 pbgetvp(dp, bp);
737 bp->b_bcount = size;
738 bp->b_bufsize = size;
739
740 cnt.v_vnodein++;
741 cnt.v_vnodepgsin += count;
742
743 /* do the input */
744 VOP_STRATEGY(bp->b_vp, bp);
745
746 s = splvm();
747 /* we definitely need to be at splvm here */
748
749 while ((bp->b_flags & B_DONE) == 0) {
750 tsleep(bp, PVM, "vnread", 0);
751 }
752 splx(s);
753 if ((bp->b_flags & B_ERROR) != 0)
754 error = EIO;
755
756 if (!error) {
757 if (size != count * PAGE_SIZE)
758 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
759 }
760 pmap_qremove(kva, count);
761
762 /*
763 * free the buffer header back to the swap buffer pool
764 */
765 relpbuf(bp, &vnode_pbuf_freecnt);
766
767 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
768 vm_page_t mt;
769
770 nextoff = tfoff + PAGE_SIZE;
771 mt = m[i];
772
773 if (nextoff <= size) {
774 /*
775 * Read filled up entire page.
776 */
777 mt->valid = VM_PAGE_BITS_ALL;
778 mt->dirty = 0;
779 pmap_clear_modify(VM_PAGE_TO_PHYS(mt));
780 } else {
781 /*
782 * Read did not fill up entire page. Since this
783 * is getpages, the page may be mapped, so we have
784 * to zero the invalid portions of the page even
785 * though we aren't setting them valid.
786 *
787 * Currently we do not set the entire page valid,
788 * we just try to clear the piece that we couldn't
789 * read.
790 */
791 vm_page_set_validclean(mt, 0, size - tfoff);
792 /* handled by vm_fault now */
793 /* vm_page_zero_invalid(mt, FALSE); */
794 }
795
796 vm_page_flag_clear(mt, PG_ZERO);
797 if (i != reqpage) {
798
799 /*
800 * whether or not to leave the page activated is up in
801 * the air, but we should put the page on a page queue
802 * somewhere. (it already is in the object). Result:
803 * It appears that emperical results show that
804 * deactivating pages is best.
805 */
806
807 /*
808 * just in case someone was asking for this page we
809 * now tell them that it is ok to use
810 */
811 if (!error) {
812 if (mt->flags & PG_WANTED)
813 vm_page_activate(mt);
814 else
815 vm_page_deactivate(mt);
816 vm_page_wakeup(mt);
817 } else {
818 vnode_pager_freepage(mt);
819 }
820 }
821 }
822 if (error) {
823 printf("vnode_pager_getpages: I/O read error\n");
824 }
825 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
826}
827
828/*
829 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
830 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
831 * vnode_pager_generic_putpages() to implement the previous behaviour.
832 *
833 * All other FS's should use the bypass to get to the local media
834 * backing vp's VOP_PUTPAGES.
835 */
836static void
837vnode_pager_putpages(object, m, count, sync, rtvals)
838 vm_object_t object;
839 vm_page_t *m;
840 int count;
841 boolean_t sync;
842 int *rtvals;
843{
844 int rtval;
845 struct vnode *vp;
846 int bytes = count * PAGE_SIZE;
847
848 /*
849 * Force synchronous operation if we are extremely low on memory
850 * to prevent a low-memory deadlock. VOP operations often need to
851 * allocate more memory to initiate the I/O ( i.e. do a BMAP
852 * operation ). The swapper handles the case by limiting the amount
853 * of asynchronous I/O, but that sort of solution doesn't scale well
854 * for the vnode pager without a lot of work.
855 *
856 * Also, the backing vnode's iodone routine may not wake the pageout
857 * daemon up. This should be probably be addressed XXX.
858 */
859
860 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
861 sync |= OBJPC_SYNC;
862
863 /*
864 * Call device-specific putpages function
865 */
866
867 vp = object->handle;
868 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
869 if (rtval == EOPNOTSUPP) {
870 printf("vnode_pager: *** WARNING *** stale FS putpages\n");
871 rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
872 }
873}
874
875
876/*
877 * This is now called from local media FS's to operate against their
878 * own vnodes if they fail to implement VOP_PUTPAGES.
879 */
880int
881vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals)
882 struct vnode *vp;
883 vm_page_t *m;
884 int bytecount;
885 int flags;
886 int *rtvals;
887{
888 int i;
889 vm_object_t object;
890 int count;
891
892 int maxsize, ncount;
893 vm_ooffset_t poffset;
894 struct uio auio;
895 struct iovec aiov;
896 int error;
897 int ioflags;
898
899 object = vp->v_object;
900 count = bytecount / PAGE_SIZE;
901
902 for (i = 0; i < count; i++)
903 rtvals[i] = VM_PAGER_AGAIN;
904
905 if ((int) m[0]->pindex < 0) {
906 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n",
907 m[0]->pindex, m[0]->dirty);
908 rtvals[0] = VM_PAGER_BAD;
909 return VM_PAGER_BAD;
910 }
911
912 maxsize = count * PAGE_SIZE;
913 ncount = count;
914
915 poffset = IDX_TO_OFF(m[0]->pindex);
916 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
917 if (object->un_pager.vnp.vnp_size > poffset)
918 maxsize = object->un_pager.vnp.vnp_size - poffset;
919 else
920 maxsize = 0;
921 ncount = btoc(maxsize);
922 if (ncount < count) {
923 for (i = ncount; i < count; i++) {
924 rtvals[i] = VM_PAGER_BAD;
925 }
926 }
927 }
928
929 ioflags = IO_VMIO;
930 ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: 0;
931 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
932
933 aiov.iov_base = (caddr_t) 0;
934 aiov.iov_len = maxsize;
935 auio.uio_iov = &aiov;
936 auio.uio_iovcnt = 1;
937 auio.uio_offset = poffset;
938 auio.uio_segflg = UIO_NOCOPY;
939 auio.uio_rw = UIO_WRITE;
940 auio.uio_resid = maxsize;
941 auio.uio_procp = (struct proc *) 0;
942 error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred);
943 cnt.v_vnodeout++;
944 cnt.v_vnodepgsout += ncount;
945
946 if (error) {
947 printf("vnode_pager_putpages: I/O error %d\n", error);
948 }
949 if (auio.uio_resid) {
950 printf("vnode_pager_putpages: residual I/O %d at %lu\n",
951 auio.uio_resid, (u_long)m[0]->pindex);
952 }
953 for (i = 0; i < ncount; i++) {
954 rtvals[i] = VM_PAGER_OK;
955 }
956 return rtvals[0];
957}
958
959struct vnode *
960vnode_pager_lock(object)
961 vm_object_t object;
962{
963 struct proc *p = curproc; /* XXX */
964
965 for (; object != NULL; object = object->backing_object) {
966 if (object->type != OBJT_VNODE)
967 continue;
968 if (object->flags & OBJ_DEAD)
969 return NULL;
970
971 while (vget(object->handle,
972 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) {
973 if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE))
974 return NULL;
975 printf("vnode_pager_lock: retrying\n");
976 }
977 return object->handle;
978 }
979 return NULL;
980}
| 729 if (bp->b_rcred != NOCRED)
730 crhold(bp->b_rcred);
731 if (bp->b_wcred != NOCRED)
732 crhold(bp->b_wcred);
733 bp->b_blkno = firstaddr;
734 pbgetvp(dp, bp);
735 bp->b_bcount = size;
736 bp->b_bufsize = size;
737
738 cnt.v_vnodein++;
739 cnt.v_vnodepgsin += count;
740
741 /* do the input */
742 VOP_STRATEGY(bp->b_vp, bp);
743
744 s = splvm();
745 /* we definitely need to be at splvm here */
746
747 while ((bp->b_flags & B_DONE) == 0) {
748 tsleep(bp, PVM, "vnread", 0);
749 }
750 splx(s);
751 if ((bp->b_flags & B_ERROR) != 0)
752 error = EIO;
753
754 if (!error) {
755 if (size != count * PAGE_SIZE)
756 bzero((caddr_t) kva + size, PAGE_SIZE * count - size);
757 }
758 pmap_qremove(kva, count);
759
760 /*
761 * free the buffer header back to the swap buffer pool
762 */
763 relpbuf(bp, &vnode_pbuf_freecnt);
764
765 for (i = 0, tfoff = foff; i < count; i++, tfoff = nextoff) {
766 vm_page_t mt;
767
768 nextoff = tfoff + PAGE_SIZE;
769 mt = m[i];
770
771 if (nextoff <= size) {
772 /*
773 * Read filled up entire page.
774 */
775 mt->valid = VM_PAGE_BITS_ALL;
776 mt->dirty = 0;
777 pmap_clear_modify(VM_PAGE_TO_PHYS(mt));
778 } else {
779 /*
780 * Read did not fill up entire page. Since this
781 * is getpages, the page may be mapped, so we have
782 * to zero the invalid portions of the page even
783 * though we aren't setting them valid.
784 *
785 * Currently we do not set the entire page valid,
786 * we just try to clear the piece that we couldn't
787 * read.
788 */
789 vm_page_set_validclean(mt, 0, size - tfoff);
790 /* handled by vm_fault now */
791 /* vm_page_zero_invalid(mt, FALSE); */
792 }
793
794 vm_page_flag_clear(mt, PG_ZERO);
795 if (i != reqpage) {
796
797 /*
798 * whether or not to leave the page activated is up in
799 * the air, but we should put the page on a page queue
800 * somewhere. (it already is in the object). Result:
801 * It appears that emperical results show that
802 * deactivating pages is best.
803 */
804
805 /*
806 * just in case someone was asking for this page we
807 * now tell them that it is ok to use
808 */
809 if (!error) {
810 if (mt->flags & PG_WANTED)
811 vm_page_activate(mt);
812 else
813 vm_page_deactivate(mt);
814 vm_page_wakeup(mt);
815 } else {
816 vnode_pager_freepage(mt);
817 }
818 }
819 }
820 if (error) {
821 printf("vnode_pager_getpages: I/O read error\n");
822 }
823 return (error ? VM_PAGER_ERROR : VM_PAGER_OK);
824}
825
826/*
827 * EOPNOTSUPP is no longer legal. For local media VFS's that do not
828 * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
829 * vnode_pager_generic_putpages() to implement the previous behaviour.
830 *
831 * All other FS's should use the bypass to get to the local media
832 * backing vp's VOP_PUTPAGES.
833 */
834static void
835vnode_pager_putpages(object, m, count, sync, rtvals)
836 vm_object_t object;
837 vm_page_t *m;
838 int count;
839 boolean_t sync;
840 int *rtvals;
841{
842 int rtval;
843 struct vnode *vp;
844 int bytes = count * PAGE_SIZE;
845
846 /*
847 * Force synchronous operation if we are extremely low on memory
848 * to prevent a low-memory deadlock. VOP operations often need to
849 * allocate more memory to initiate the I/O ( i.e. do a BMAP
850 * operation ). The swapper handles the case by limiting the amount
851 * of asynchronous I/O, but that sort of solution doesn't scale well
852 * for the vnode pager without a lot of work.
853 *
854 * Also, the backing vnode's iodone routine may not wake the pageout
855 * daemon up. This should be probably be addressed XXX.
856 */
857
858 if ((cnt.v_free_count + cnt.v_cache_count) < cnt.v_pageout_free_min)
859 sync |= OBJPC_SYNC;
860
861 /*
862 * Call device-specific putpages function
863 */
864
865 vp = object->handle;
866 rtval = VOP_PUTPAGES(vp, m, bytes, sync, rtvals, 0);
867 if (rtval == EOPNOTSUPP) {
868 printf("vnode_pager: *** WARNING *** stale FS putpages\n");
869 rtval = vnode_pager_generic_putpages( vp, m, bytes, sync, rtvals);
870 }
871}
872
873
874/*
875 * This is now called from local media FS's to operate against their
876 * own vnodes if they fail to implement VOP_PUTPAGES.
877 */
878int
879vnode_pager_generic_putpages(vp, m, bytecount, flags, rtvals)
880 struct vnode *vp;
881 vm_page_t *m;
882 int bytecount;
883 int flags;
884 int *rtvals;
885{
886 int i;
887 vm_object_t object;
888 int count;
889
890 int maxsize, ncount;
891 vm_ooffset_t poffset;
892 struct uio auio;
893 struct iovec aiov;
894 int error;
895 int ioflags;
896
897 object = vp->v_object;
898 count = bytecount / PAGE_SIZE;
899
900 for (i = 0; i < count; i++)
901 rtvals[i] = VM_PAGER_AGAIN;
902
903 if ((int) m[0]->pindex < 0) {
904 printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%x(%x)\n",
905 m[0]->pindex, m[0]->dirty);
906 rtvals[0] = VM_PAGER_BAD;
907 return VM_PAGER_BAD;
908 }
909
910 maxsize = count * PAGE_SIZE;
911 ncount = count;
912
913 poffset = IDX_TO_OFF(m[0]->pindex);
914 if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
915 if (object->un_pager.vnp.vnp_size > poffset)
916 maxsize = object->un_pager.vnp.vnp_size - poffset;
917 else
918 maxsize = 0;
919 ncount = btoc(maxsize);
920 if (ncount < count) {
921 for (i = ncount; i < count; i++) {
922 rtvals[i] = VM_PAGER_BAD;
923 }
924 }
925 }
926
927 ioflags = IO_VMIO;
928 ioflags |= (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ? IO_SYNC: 0;
929 ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
930
931 aiov.iov_base = (caddr_t) 0;
932 aiov.iov_len = maxsize;
933 auio.uio_iov = &aiov;
934 auio.uio_iovcnt = 1;
935 auio.uio_offset = poffset;
936 auio.uio_segflg = UIO_NOCOPY;
937 auio.uio_rw = UIO_WRITE;
938 auio.uio_resid = maxsize;
939 auio.uio_procp = (struct proc *) 0;
940 error = VOP_WRITE(vp, &auio, ioflags, curproc->p_ucred);
941 cnt.v_vnodeout++;
942 cnt.v_vnodepgsout += ncount;
943
944 if (error) {
945 printf("vnode_pager_putpages: I/O error %d\n", error);
946 }
947 if (auio.uio_resid) {
948 printf("vnode_pager_putpages: residual I/O %d at %lu\n",
949 auio.uio_resid, (u_long)m[0]->pindex);
950 }
951 for (i = 0; i < ncount; i++) {
952 rtvals[i] = VM_PAGER_OK;
953 }
954 return rtvals[0];
955}
956
957struct vnode *
958vnode_pager_lock(object)
959 vm_object_t object;
960{
961 struct proc *p = curproc; /* XXX */
962
963 for (; object != NULL; object = object->backing_object) {
964 if (object->type != OBJT_VNODE)
965 continue;
966 if (object->flags & OBJ_DEAD)
967 return NULL;
968
969 while (vget(object->handle,
970 LK_NOPAUSE | LK_SHARED | LK_RETRY | LK_CANRECURSE, p)) {
971 if ((object->flags & OBJ_DEAD) || (object->type != OBJT_VNODE))
972 return NULL;
973 printf("vnode_pager_lock: retrying\n");
974 }
975 return object->handle;
976 }
977 return NULL;
978}
|