uvm_vnode.c revision 1.94 1 1.94 chuck /* $NetBSD: uvm_vnode.c,v 1.94 2011/02/02 20:07:25 chuck Exp $ */
2 1.1 mrg
3 1.1 mrg /*
4 1.1 mrg * Copyright (c) 1997 Charles D. Cranor and Washington University.
5 1.1 mrg * Copyright (c) 1991, 1993
6 1.49 chs * The Regents of the University of California.
7 1.1 mrg * Copyright (c) 1990 University of Utah.
8 1.1 mrg *
9 1.1 mrg * All rights reserved.
10 1.1 mrg *
11 1.1 mrg * This code is derived from software contributed to Berkeley by
12 1.1 mrg * the Systems Programming Group of the University of Utah Computer
13 1.1 mrg * Science Department.
14 1.1 mrg *
15 1.1 mrg * Redistribution and use in source and binary forms, with or without
16 1.1 mrg * modification, are permitted provided that the following conditions
17 1.1 mrg * are met:
18 1.1 mrg * 1. Redistributions of source code must retain the above copyright
19 1.1 mrg * notice, this list of conditions and the following disclaimer.
20 1.1 mrg * 2. Redistributions in binary form must reproduce the above copyright
21 1.1 mrg * notice, this list of conditions and the following disclaimer in the
22 1.1 mrg * documentation and/or other materials provided with the distribution.
23 1.94 chuck * 3. Neither the name of the University nor the names of its contributors
24 1.1 mrg * may be used to endorse or promote products derived from this software
25 1.1 mrg * without specific prior written permission.
26 1.1 mrg *
27 1.1 mrg * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 1.1 mrg * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 1.1 mrg * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 1.1 mrg * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 1.1 mrg * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 1.1 mrg * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 1.1 mrg * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 1.1 mrg * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 1.1 mrg * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 1.1 mrg * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 1.1 mrg * SUCH DAMAGE.
38 1.1 mrg *
39 1.1 mrg * @(#)vnode_pager.c 8.8 (Berkeley) 2/13/94
40 1.3 mrg * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp
41 1.1 mrg */
42 1.1 mrg
43 1.55 lukem /*
44 1.55 lukem * uvm_vnode.c: the vnode pager.
45 1.55 lukem */
46 1.55 lukem
47 1.55 lukem #include <sys/cdefs.h>
48 1.94 chuck __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.94 2011/02/02 20:07:25 chuck Exp $");
49 1.55 lukem
50 1.4 mrg #include "opt_uvmhist.h"
51 1.1 mrg
52 1.1 mrg #include <sys/param.h>
53 1.1 mrg #include <sys/systm.h>
54 1.37 chs #include <sys/kernel.h>
55 1.1 mrg #include <sys/proc.h>
56 1.1 mrg #include <sys/malloc.h>
57 1.1 mrg #include <sys/vnode.h>
58 1.13 thorpej #include <sys/disklabel.h>
59 1.13 thorpej #include <sys/ioctl.h>
60 1.13 thorpej #include <sys/fcntl.h>
61 1.13 thorpej #include <sys/conf.h>
62 1.37 chs #include <sys/pool.h>
63 1.37 chs #include <sys/mount.h>
64 1.13 thorpej
65 1.13 thorpej #include <miscfs/specfs/specdev.h>
66 1.1 mrg
67 1.1 mrg #include <uvm/uvm.h>
68 1.68 yamt #include <uvm/uvm_readahead.h>
69 1.1 mrg
70 1.1 mrg /*
71 1.1 mrg * functions
72 1.1 mrg */
73 1.1 mrg
74 1.66 thorpej static void uvn_detach(struct uvm_object *);
75 1.66 thorpej static int uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
76 1.66 thorpej int, vm_prot_t, int, int);
77 1.66 thorpej static int uvn_put(struct uvm_object *, voff_t, voff_t, int);
78 1.66 thorpej static void uvn_reference(struct uvm_object *);
79 1.52 chs
80 1.66 thorpej static int uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
81 1.66 thorpej int);
82 1.1 mrg
83 1.1 mrg /*
84 1.1 mrg * master pager structure
85 1.1 mrg */
86 1.1 mrg
87 1.89 yamt const struct uvm_pagerops uvm_vnodeops = {
88 1.88 yamt .pgo_reference = uvn_reference,
89 1.88 yamt .pgo_detach = uvn_detach,
90 1.88 yamt .pgo_get = uvn_get,
91 1.88 yamt .pgo_put = uvn_put,
92 1.1 mrg };
93 1.1 mrg
94 1.1 mrg /*
95 1.1 mrg * the ops!
96 1.1 mrg */
97 1.1 mrg
98 1.1 mrg /*
99 1.1 mrg * uvn_reference
100 1.1 mrg *
101 1.1 mrg * duplicate a reference to a VM object. Note that the reference
102 1.49 chs * count must already be at least one (the passed in reference) so
103 1.1 mrg * there is no chance of the uvn being killed or locked out here.
104 1.1 mrg *
105 1.49 chs * => caller must call with object unlocked.
106 1.1 mrg * => caller must be using the same accessprot as was used at attach time
107 1.1 mrg */
108 1.1 mrg
109 1.66 thorpej static void
110 1.65 thorpej uvn_reference(struct uvm_object *uobj)
111 1.1 mrg {
112 1.93 pooka vref((struct vnode *)uobj);
113 1.1 mrg }
114 1.1 mrg
115 1.52 chs
116 1.1 mrg /*
117 1.1 mrg * uvn_detach
118 1.1 mrg *
119 1.1 mrg * remove a reference to a VM object.
120 1.1 mrg *
121 1.1 mrg * => caller must call with object unlocked and map locked.
122 1.1 mrg */
123 1.52 chs
124 1.66 thorpej static void
125 1.65 thorpej uvn_detach(struct uvm_object *uobj)
126 1.8 mrg {
127 1.37 chs vrele((struct vnode *)uobj);
128 1.1 mrg }
129 1.1 mrg
130 1.1 mrg /*
131 1.1 mrg * uvn_put: flush page data to backing store.
132 1.1 mrg *
133 1.53 sommerfe * => object must be locked on entry! VOP_PUTPAGES must unlock it.
134 1.1 mrg * => flags: PGO_SYNCIO -- use sync. I/O
135 1.1 mrg * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
136 1.1 mrg */
137 1.1 mrg
138 1.66 thorpej static int
139 1.65 thorpej uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
140 1.1 mrg {
141 1.37 chs struct vnode *vp = (struct vnode *)uobj;
142 1.37 chs int error;
143 1.1 mrg
144 1.90 ad KASSERT(mutex_owned(&vp->v_interlock));
145 1.54 chs error = VOP_PUTPAGES(vp, offlo, offhi, flags);
146 1.90 ad
147 1.48 chs return error;
148 1.1 mrg }
149 1.1 mrg
150 1.1 mrg
151 1.1 mrg /*
152 1.1 mrg * uvn_get: get pages (synchronously) from backing store
153 1.1 mrg *
154 1.1 mrg * => prefer map unlocked (not required)
155 1.1 mrg * => object must be locked! we will _unlock_ it before starting any I/O.
156 1.1 mrg * => flags: PGO_ALLPAGES: get all of the pages
157 1.1 mrg * PGO_LOCKED: fault data structures are locked
158 1.1 mrg * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
159 1.1 mrg * => NOTE: caller must check for released pages!!
160 1.1 mrg */
161 1.49 chs
162 1.66 thorpej static int
163 1.65 thorpej uvn_get(struct uvm_object *uobj, voff_t offset,
164 1.65 thorpej struct vm_page **pps /* IN/OUT */,
165 1.65 thorpej int *npagesp /* IN (OUT if PGO_LOCKED)*/,
166 1.65 thorpej int centeridx, vm_prot_t access_type, int advice, int flags)
167 1.8 mrg {
168 1.37 chs struct vnode *vp = (struct vnode *)uobj;
169 1.37 chs int error;
170 1.67 yamt
171 1.37 chs UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
172 1.37 chs
173 1.37 chs UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
174 1.68 yamt
175 1.68 yamt if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) {
176 1.68 yamt vn_ra_allocctx(vp);
177 1.68 yamt uvm_ra_request(vp->v_ractx, advice, uobj, offset,
178 1.68 yamt *npagesp << PAGE_SHIFT);
179 1.68 yamt }
180 1.68 yamt
181 1.37 chs error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
182 1.37 chs access_type, advice, flags);
183 1.67 yamt
184 1.90 ad KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(&vp->v_interlock)) ||
185 1.90 ad (flags & PGO_LOCKED) == 0);
186 1.48 chs return error;
187 1.37 chs }
188 1.8 mrg
189 1.8 mrg
190 1.37 chs /*
191 1.37 chs * uvn_findpages:
192 1.37 chs * return the page for the uobj and offset requested, allocating if needed.
193 1.37 chs * => uobj must be locked.
194 1.52 chs * => returned pages will be BUSY.
195 1.37 chs */
196 1.1 mrg
197 1.58 enami int
198 1.65 thorpej uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
199 1.65 thorpej struct vm_page **pgs, int flags)
200 1.37 chs {
201 1.58 enami int i, count, found, npages, rv;
202 1.8 mrg
203 1.58 enami count = found = 0;
204 1.37 chs npages = *npagesp;
205 1.52 chs if (flags & UFP_BACKWARD) {
206 1.52 chs for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
207 1.52 chs rv = uvn_findpage(uobj, offset, &pgs[i], flags);
208 1.58 enami if (rv == 0) {
209 1.58 enami if (flags & UFP_DIRTYONLY)
210 1.58 enami break;
211 1.58 enami } else
212 1.58 enami found++;
213 1.52 chs count++;
214 1.52 chs }
215 1.52 chs } else {
216 1.52 chs for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
217 1.52 chs rv = uvn_findpage(uobj, offset, &pgs[i], flags);
218 1.58 enami if (rv == 0) {
219 1.58 enami if (flags & UFP_DIRTYONLY)
220 1.58 enami break;
221 1.58 enami } else
222 1.58 enami found++;
223 1.52 chs count++;
224 1.52 chs }
225 1.37 chs }
226 1.52 chs *npagesp = count;
227 1.58 enami return (found);
228 1.37 chs }
229 1.8 mrg
230 1.66 thorpej static int
231 1.65 thorpej uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
232 1.65 thorpej int flags)
233 1.37 chs {
234 1.37 chs struct vm_page *pg;
235 1.79 thorpej bool dirty;
236 1.37 chs UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
237 1.37 chs UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
238 1.8 mrg
239 1.37 chs if (*pgp != NULL) {
240 1.37 chs UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
241 1.37 chs return 0;
242 1.37 chs }
243 1.37 chs for (;;) {
244 1.37 chs /* look for an existing page */
245 1.37 chs pg = uvm_pagelookup(uobj, offset);
246 1.37 chs
247 1.52 chs /* nope? allocate one now */
248 1.37 chs if (pg == NULL) {
249 1.37 chs if (flags & UFP_NOALLOC) {
250 1.37 chs UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
251 1.37 chs return 0;
252 1.37 chs }
253 1.47 chs pg = uvm_pagealloc(uobj, offset, NULL, 0);
254 1.37 chs if (pg == NULL) {
255 1.37 chs if (flags & UFP_NOWAIT) {
256 1.37 chs UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
257 1.37 chs return 0;
258 1.8 mrg }
259 1.90 ad mutex_exit(&uobj->vmobjlock);
260 1.37 chs uvm_wait("uvn_fp1");
261 1.90 ad mutex_enter(&uobj->vmobjlock);
262 1.37 chs continue;
263 1.47 chs }
264 1.52 chs UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
265 1.37 chs break;
266 1.37 chs } else if (flags & UFP_NOCACHE) {
267 1.37 chs UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
268 1.37 chs return 0;
269 1.8 mrg }
270 1.8 mrg
271 1.37 chs /* page is there, see if we need to wait on it */
272 1.52 chs if ((pg->flags & PG_BUSY) != 0) {
273 1.37 chs if (flags & UFP_NOWAIT) {
274 1.37 chs UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
275 1.37 chs return 0;
276 1.37 chs }
277 1.37 chs pg->flags |= PG_WANTED;
278 1.58 enami UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
279 1.37 chs UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
280 1.37 chs "uvn_fp2", 0);
281 1.90 ad mutex_enter(&uobj->vmobjlock);
282 1.37 chs continue;
283 1.8 mrg }
284 1.49 chs
285 1.37 chs /* skip PG_RDONLY pages if requested */
286 1.37 chs if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
287 1.37 chs UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
288 1.37 chs return 0;
289 1.8 mrg }
290 1.8 mrg
291 1.52 chs /* stop on clean pages if requested */
292 1.52 chs if (flags & UFP_DIRTYONLY) {
293 1.52 chs dirty = pmap_clear_modify(pg) ||
294 1.52 chs (pg->flags & PG_CLEAN) == 0;
295 1.52 chs pg->flags |= PG_CLEAN;
296 1.52 chs if (!dirty) {
297 1.58 enami UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
298 1.52 chs return 0;
299 1.52 chs }
300 1.52 chs }
301 1.52 chs
302 1.37 chs /* mark the page BUSY and we're done. */
303 1.37 chs pg->flags |= PG_BUSY;
304 1.37 chs UVM_PAGE_OWN(pg, "uvn_findpage");
305 1.52 chs UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
306 1.37 chs break;
307 1.8 mrg }
308 1.37 chs *pgp = pg;
309 1.37 chs return 1;
310 1.1 mrg }
311 1.1 mrg
312 1.1 mrg /*
313 1.52 chs * uvm_vnp_setsize: grow or shrink a vnode uobj
314 1.1 mrg *
315 1.1 mrg * grow => just update size value
316 1.1 mrg * shrink => toss un-needed pages
317 1.1 mrg *
318 1.49 chs * => we assume that the caller has a reference of some sort to the
319 1.1 mrg * vnode in question so that it will not be yanked out from under
320 1.1 mrg * us.
321 1.1 mrg */
322 1.1 mrg
323 1.8 mrg void
324 1.65 thorpej uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
325 1.8 mrg {
326 1.52 chs struct uvm_object *uobj = &vp->v_uobj;
327 1.46 enami voff_t pgend = round_page(newsize);
328 1.72 yamt voff_t oldsize;
329 1.37 chs UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
330 1.37 chs
331 1.90 ad mutex_enter(&uobj->vmobjlock);
332 1.52 chs UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
333 1.52 chs vp, vp->v_size, newsize, 0);
334 1.1 mrg
335 1.8 mrg /*
336 1.37 chs * now check if the size has changed: if we shrink we had better
337 1.37 chs * toss some pages...
338 1.8 mrg */
339 1.1 mrg
340 1.85 pooka KASSERT(newsize != VSIZENOTSET);
341 1.85 pooka KASSERT(vp->v_size <= vp->v_writesize);
342 1.85 pooka KASSERT(vp->v_size == vp->v_writesize ||
343 1.85 pooka newsize == vp->v_writesize || newsize <= vp->v_size);
344 1.85 pooka
345 1.85 pooka oldsize = vp->v_writesize;
346 1.85 pooka KASSERT(oldsize != VSIZENOTSET || pgend > oldsize);
347 1.85 pooka
348 1.85 pooka if (oldsize > pgend) {
349 1.57 chs (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
350 1.90 ad mutex_enter(&uobj->vmobjlock);
351 1.8 mrg }
352 1.82 yamt vp->v_size = vp->v_writesize = newsize;
353 1.90 ad mutex_exit(&uobj->vmobjlock);
354 1.1 mrg }
355 1.1 mrg
356 1.82 yamt void
357 1.82 yamt uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize)
358 1.82 yamt {
359 1.82 yamt
360 1.90 ad mutex_enter(&vp->v_interlock);
361 1.85 pooka KASSERT(newsize != VSIZENOTSET);
362 1.82 yamt KASSERT(vp->v_size != VSIZENOTSET);
363 1.82 yamt KASSERT(vp->v_writesize != VSIZENOTSET);
364 1.82 yamt KASSERT(vp->v_size <= vp->v_writesize);
365 1.82 yamt KASSERT(vp->v_size <= newsize);
366 1.82 yamt vp->v_writesize = newsize;
367 1.90 ad mutex_exit(&vp->v_interlock);
368 1.82 yamt }
369 1.82 yamt
370 1.79 thorpej bool
371 1.75 yamt uvn_text_p(struct uvm_object *uobj)
372 1.75 yamt {
373 1.75 yamt struct vnode *vp = (struct vnode *)uobj;
374 1.75 yamt
375 1.86 ad return (vp->v_iflag & VI_EXECMAP) != 0;
376 1.75 yamt }
377 1.75 yamt
378 1.79 thorpej bool
379 1.75 yamt uvn_clean_p(struct uvm_object *uobj)
380 1.75 yamt {
381 1.75 yamt struct vnode *vp = (struct vnode *)uobj;
382 1.75 yamt
383 1.86 ad return (vp->v_iflag & VI_ONWORKLST) == 0;
384 1.75 yamt }
385 1.75 yamt
386 1.79 thorpej bool
387 1.75 yamt uvn_needs_writefault_p(struct uvm_object *uobj)
388 1.75 yamt {
389 1.75 yamt struct vnode *vp = (struct vnode *)uobj;
390 1.75 yamt
391 1.75 yamt return uvn_clean_p(uobj) ||
392 1.86 ad (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP;
393 1.75 yamt }
394