xref: /netbsd-src/sys/uvm/uvm_vnode.c (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*	$NetBSD: uvm_vnode.c,v 1.94 2011/02/02 20:07:25 chuck Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Charles D. Cranor and Washington University.
5  * Copyright (c) 1991, 1993
6  *      The Regents of the University of California.
7  * Copyright (c) 1990 University of Utah.
8  *
9  * All rights reserved.
10  *
11  * This code is derived from software contributed to Berkeley by
12  * the Systems Programming Group of the University of Utah Computer
13  * Science Department.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  *
39  *      @(#)vnode_pager.c       8.8 (Berkeley) 2/13/94
40  * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp
41  */
42 
43 /*
44  * uvm_vnode.c: the vnode pager.
45  */
46 
47 #include <sys/cdefs.h>
48 __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.94 2011/02/02 20:07:25 chuck Exp $");
49 
50 #include "opt_uvmhist.h"
51 
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kernel.h>
55 #include <sys/proc.h>
56 #include <sys/malloc.h>
57 #include <sys/vnode.h>
58 #include <sys/disklabel.h>
59 #include <sys/ioctl.h>
60 #include <sys/fcntl.h>
61 #include <sys/conf.h>
62 #include <sys/pool.h>
63 #include <sys/mount.h>
64 
65 #include <miscfs/specfs/specdev.h>
66 
67 #include <uvm/uvm.h>
68 #include <uvm/uvm_readahead.h>
69 
70 /*
71  * functions
72  */
73 
74 static void	uvn_detach(struct uvm_object *);
75 static int	uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
76 			int, vm_prot_t, int, int);
77 static int	uvn_put(struct uvm_object *, voff_t, voff_t, int);
78 static void	uvn_reference(struct uvm_object *);
79 
80 static int	uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
81 			     int);
82 
83 /*
84  * master pager structure
85  */
86 
87 const struct uvm_pagerops uvm_vnodeops = {
88 	.pgo_reference = uvn_reference,
89 	.pgo_detach = uvn_detach,
90 	.pgo_get = uvn_get,
91 	.pgo_put = uvn_put,
92 };
93 
94 /*
95  * the ops!
96  */
97 
98 /*
99  * uvn_reference
100  *
101  * duplicate a reference to a VM object.  Note that the reference
102  * count must already be at least one (the passed in reference) so
103  * there is no chance of the uvn being killed or locked out here.
104  *
105  * => caller must call with object unlocked.
106  * => caller must be using the same accessprot as was used at attach time
107  */
108 
109 static void
110 uvn_reference(struct uvm_object *uobj)
111 {
112 	vref((struct vnode *)uobj);
113 }
114 
115 
116 /*
117  * uvn_detach
118  *
119  * remove a reference to a VM object.
120  *
121  * => caller must call with object unlocked and map locked.
122  */
123 
124 static void
125 uvn_detach(struct uvm_object *uobj)
126 {
127 	vrele((struct vnode *)uobj);
128 }
129 
130 /*
131  * uvn_put: flush page data to backing store.
132  *
133  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
134  * => flags: PGO_SYNCIO -- use sync. I/O
135  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
136  */
137 
138 static int
139 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
140 {
141 	struct vnode *vp = (struct vnode *)uobj;
142 	int error;
143 
144 	KASSERT(mutex_owned(&vp->v_interlock));
145 	error = VOP_PUTPAGES(vp, offlo, offhi, flags);
146 
147 	return error;
148 }
149 
150 
151 /*
152  * uvn_get: get pages (synchronously) from backing store
153  *
154  * => prefer map unlocked (not required)
155  * => object must be locked!  we will _unlock_ it before starting any I/O.
156  * => flags: PGO_ALLPAGES: get all of the pages
157  *           PGO_LOCKED: fault data structures are locked
158  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
159  * => NOTE: caller must check for released pages!!
160  */
161 
162 static int
163 uvn_get(struct uvm_object *uobj, voff_t offset,
164     struct vm_page **pps /* IN/OUT */,
165     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
166     int centeridx, vm_prot_t access_type, int advice, int flags)
167 {
168 	struct vnode *vp = (struct vnode *)uobj;
169 	int error;
170 
171 	UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
172 
173 	UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
174 
175 	if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) {
176 		vn_ra_allocctx(vp);
177 		uvm_ra_request(vp->v_ractx, advice, uobj, offset,
178 		    *npagesp << PAGE_SHIFT);
179 	}
180 
181 	error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
182 			     access_type, advice, flags);
183 
184 	KASSERT(((flags & PGO_LOCKED) != 0 && mutex_owned(&vp->v_interlock)) ||
185 	    (flags & PGO_LOCKED) == 0);
186 	return error;
187 }
188 
189 
190 /*
191  * uvn_findpages:
192  * return the page for the uobj and offset requested, allocating if needed.
193  * => uobj must be locked.
194  * => returned pages will be BUSY.
195  */
196 
197 int
198 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
199     struct vm_page **pgs, int flags)
200 {
201 	int i, count, found, npages, rv;
202 
203 	count = found = 0;
204 	npages = *npagesp;
205 	if (flags & UFP_BACKWARD) {
206 		for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
207 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
208 			if (rv == 0) {
209 				if (flags & UFP_DIRTYONLY)
210 					break;
211 			} else
212 				found++;
213 			count++;
214 		}
215 	} else {
216 		for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
217 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
218 			if (rv == 0) {
219 				if (flags & UFP_DIRTYONLY)
220 					break;
221 			} else
222 				found++;
223 			count++;
224 		}
225 	}
226 	*npagesp = count;
227 	return (found);
228 }
229 
230 static int
231 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
232     int flags)
233 {
234 	struct vm_page *pg;
235 	bool dirty;
236 	UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
237 	UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
238 
239 	if (*pgp != NULL) {
240 		UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
241 		return 0;
242 	}
243 	for (;;) {
244 		/* look for an existing page */
245 		pg = uvm_pagelookup(uobj, offset);
246 
247 		/* nope?  allocate one now */
248 		if (pg == NULL) {
249 			if (flags & UFP_NOALLOC) {
250 				UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
251 				return 0;
252 			}
253 			pg = uvm_pagealloc(uobj, offset, NULL, 0);
254 			if (pg == NULL) {
255 				if (flags & UFP_NOWAIT) {
256 					UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
257 					return 0;
258 				}
259 				mutex_exit(&uobj->vmobjlock);
260 				uvm_wait("uvn_fp1");
261 				mutex_enter(&uobj->vmobjlock);
262 				continue;
263 			}
264 			UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
265 			break;
266 		} else if (flags & UFP_NOCACHE) {
267 			UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
268 			return 0;
269 		}
270 
271 		/* page is there, see if we need to wait on it */
272 		if ((pg->flags & PG_BUSY) != 0) {
273 			if (flags & UFP_NOWAIT) {
274 				UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
275 				return 0;
276 			}
277 			pg->flags |= PG_WANTED;
278 			UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
279 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
280 					    "uvn_fp2", 0);
281 			mutex_enter(&uobj->vmobjlock);
282 			continue;
283 		}
284 
285 		/* skip PG_RDONLY pages if requested */
286 		if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
287 			UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
288 			return 0;
289 		}
290 
291 		/* stop on clean pages if requested */
292 		if (flags & UFP_DIRTYONLY) {
293 			dirty = pmap_clear_modify(pg) ||
294 				(pg->flags & PG_CLEAN) == 0;
295 			pg->flags |= PG_CLEAN;
296 			if (!dirty) {
297 				UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
298 				return 0;
299 			}
300 		}
301 
302 		/* mark the page BUSY and we're done. */
303 		pg->flags |= PG_BUSY;
304 		UVM_PAGE_OWN(pg, "uvn_findpage");
305 		UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
306 		break;
307 	}
308 	*pgp = pg;
309 	return 1;
310 }
311 
312 /*
313  * uvm_vnp_setsize: grow or shrink a vnode uobj
314  *
315  * grow   => just update size value
316  * shrink => toss un-needed pages
317  *
318  * => we assume that the caller has a reference of some sort to the
319  *	vnode in question so that it will not be yanked out from under
320  *	us.
321  */
322 
323 void
324 uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
325 {
326 	struct uvm_object *uobj = &vp->v_uobj;
327 	voff_t pgend = round_page(newsize);
328 	voff_t oldsize;
329 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
330 
331 	mutex_enter(&uobj->vmobjlock);
332 	UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
333 	    vp, vp->v_size, newsize, 0);
334 
335 	/*
336 	 * now check if the size has changed: if we shrink we had better
337 	 * toss some pages...
338 	 */
339 
340 	KASSERT(newsize != VSIZENOTSET);
341 	KASSERT(vp->v_size <= vp->v_writesize);
342 	KASSERT(vp->v_size == vp->v_writesize ||
343 	    newsize == vp->v_writesize || newsize <= vp->v_size);
344 
345 	oldsize = vp->v_writesize;
346 	KASSERT(oldsize != VSIZENOTSET || pgend > oldsize);
347 
348 	if (oldsize > pgend) {
349 		(void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
350 		mutex_enter(&uobj->vmobjlock);
351 	}
352 	vp->v_size = vp->v_writesize = newsize;
353 	mutex_exit(&uobj->vmobjlock);
354 }
355 
356 void
357 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize)
358 {
359 
360 	mutex_enter(&vp->v_interlock);
361 	KASSERT(newsize != VSIZENOTSET);
362 	KASSERT(vp->v_size != VSIZENOTSET);
363 	KASSERT(vp->v_writesize != VSIZENOTSET);
364 	KASSERT(vp->v_size <= vp->v_writesize);
365 	KASSERT(vp->v_size <= newsize);
366 	vp->v_writesize = newsize;
367 	mutex_exit(&vp->v_interlock);
368 }
369 
370 bool
371 uvn_text_p(struct uvm_object *uobj)
372 {
373 	struct vnode *vp = (struct vnode *)uobj;
374 
375 	return (vp->v_iflag & VI_EXECMAP) != 0;
376 }
377 
378 bool
379 uvn_clean_p(struct uvm_object *uobj)
380 {
381 	struct vnode *vp = (struct vnode *)uobj;
382 
383 	return (vp->v_iflag & VI_ONWORKLST) == 0;
384 }
385 
386 bool
387 uvn_needs_writefault_p(struct uvm_object *uobj)
388 {
389 	struct vnode *vp = (struct vnode *)uobj;
390 
391 	return uvn_clean_p(uobj) ||
392 	    (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP;
393 }
394