xref: /netbsd-src/sys/uvm/uvm_vnode.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: uvm_vnode.c,v 1.87 2007/10/11 19:53:43 ad 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. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *      This product includes software developed by Charles D. Cranor,
26  *	Washington University, the University of California, Berkeley and
27  *	its contributors.
28  * 4. Neither the name of the University nor the names of its contributors
29  *    may be used to endorse or promote products derived from this software
30  *    without specific prior written permission.
31  *
32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42  * SUCH DAMAGE.
43  *
44  *      @(#)vnode_pager.c       8.8 (Berkeley) 2/13/94
45  * from: Id: uvm_vnode.c,v 1.1.2.26 1998/02/02 20:38:07 chuck Exp
46  */
47 
48 /*
49  * uvm_vnode.c: the vnode pager.
50  */
51 
52 #include <sys/cdefs.h>
53 __KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.87 2007/10/11 19:53:43 ad Exp $");
54 
55 #include "fs_nfs.h"
56 #include "opt_uvmhist.h"
57 #include "opt_ddb.h"
58 
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/proc.h>
63 #include <sys/malloc.h>
64 #include <sys/vnode.h>
65 #include <sys/disklabel.h>
66 #include <sys/ioctl.h>
67 #include <sys/fcntl.h>
68 #include <sys/conf.h>
69 #include <sys/pool.h>
70 #include <sys/mount.h>
71 
72 #include <miscfs/specfs/specdev.h>
73 
74 #include <uvm/uvm.h>
75 #include <uvm/uvm_readahead.h>
76 
77 /*
78  * functions
79  */
80 
81 static void	uvn_detach(struct uvm_object *);
82 static int	uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
83 			int, vm_prot_t, int, int);
84 static int	uvn_put(struct uvm_object *, voff_t, voff_t, int);
85 static void	uvn_reference(struct uvm_object *);
86 
87 static int	uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
88 			     int);
89 
90 /*
91  * master pager structure
92  */
93 
94 struct uvm_pagerops uvm_vnodeops = {
95 	NULL,
96 	uvn_reference,
97 	uvn_detach,
98 	NULL,
99 	uvn_get,
100 	uvn_put,
101 };
102 
103 /*
104  * the ops!
105  */
106 
107 /*
108  * uvn_reference
109  *
110  * duplicate a reference to a VM object.  Note that the reference
111  * count must already be at least one (the passed in reference) so
112  * there is no chance of the uvn being killed or locked out here.
113  *
114  * => caller must call with object unlocked.
115  * => caller must be using the same accessprot as was used at attach time
116  */
117 
118 static void
119 uvn_reference(struct uvm_object *uobj)
120 {
121 	VREF((struct vnode *)uobj);
122 }
123 
124 
125 /*
126  * uvn_detach
127  *
128  * remove a reference to a VM object.
129  *
130  * => caller must call with object unlocked and map locked.
131  */
132 
133 static void
134 uvn_detach(struct uvm_object *uobj)
135 {
136 	vrele((struct vnode *)uobj);
137 }
138 
139 /*
140  * uvn_put: flush page data to backing store.
141  *
142  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
143  * => flags: PGO_SYNCIO -- use sync. I/O
144  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
145  */
146 
147 static int
148 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
149 {
150 	struct vnode *vp = (struct vnode *)uobj;
151 	int error;
152 
153 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
154 	error = VOP_PUTPAGES(vp, offlo, offhi, flags);
155 	return error;
156 }
157 
158 
159 /*
160  * uvn_get: get pages (synchronously) from backing store
161  *
162  * => prefer map unlocked (not required)
163  * => object must be locked!  we will _unlock_ it before starting any I/O.
164  * => flags: PGO_ALLPAGES: get all of the pages
165  *           PGO_LOCKED: fault data structures are locked
166  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
167  * => NOTE: caller must check for released pages!!
168  */
169 
170 static int
171 uvn_get(struct uvm_object *uobj, voff_t offset,
172     struct vm_page **pps /* IN/OUT */,
173     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
174     int centeridx, vm_prot_t access_type, int advice, int flags)
175 {
176 	struct vnode *vp = (struct vnode *)uobj;
177 	int error;
178 
179 	UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
180 
181 	UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
182 
183 	if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) {
184 		simple_unlock(&vp->v_interlock);
185 		vn_ra_allocctx(vp);
186 		uvm_ra_request(vp->v_ractx, advice, uobj, offset,
187 		    *npagesp << PAGE_SHIFT);
188 		simple_lock(&vp->v_interlock);
189 	}
190 
191 	error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
192 			     access_type, advice, flags);
193 
194 	LOCK_ASSERT(((flags & PGO_LOCKED) != 0 &&
195 		     simple_lock_held(&vp->v_interlock)) ||
196 		    (flags & PGO_LOCKED) == 0);
197 	return error;
198 }
199 
200 
201 /*
202  * uvn_findpages:
203  * return the page for the uobj and offset requested, allocating if needed.
204  * => uobj must be locked.
205  * => returned pages will be BUSY.
206  */
207 
208 int
209 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
210     struct vm_page **pgs, int flags)
211 {
212 	int i, count, found, npages, rv;
213 
214 	count = found = 0;
215 	npages = *npagesp;
216 	if (flags & UFP_BACKWARD) {
217 		for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
218 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
219 			if (rv == 0) {
220 				if (flags & UFP_DIRTYONLY)
221 					break;
222 			} else
223 				found++;
224 			count++;
225 		}
226 	} else {
227 		for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
228 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
229 			if (rv == 0) {
230 				if (flags & UFP_DIRTYONLY)
231 					break;
232 			} else
233 				found++;
234 			count++;
235 		}
236 	}
237 	*npagesp = count;
238 	return (found);
239 }
240 
241 static int
242 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
243     int flags)
244 {
245 	struct vm_page *pg;
246 	bool dirty;
247 	UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
248 	UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
249 
250 	if (*pgp != NULL) {
251 		UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
252 		return 0;
253 	}
254 	for (;;) {
255 		/* look for an existing page */
256 		pg = uvm_pagelookup(uobj, offset);
257 
258 		/* nope?  allocate one now */
259 		if (pg == NULL) {
260 			if (flags & UFP_NOALLOC) {
261 				UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
262 				return 0;
263 			}
264 			pg = uvm_pagealloc(uobj, offset, NULL, 0);
265 			if (pg == NULL) {
266 				if (flags & UFP_NOWAIT) {
267 					UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
268 					return 0;
269 				}
270 				simple_unlock(&uobj->vmobjlock);
271 				uvm_wait("uvn_fp1");
272 				simple_lock(&uobj->vmobjlock);
273 				continue;
274 			}
275 			UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
276 			break;
277 		} else if (flags & UFP_NOCACHE) {
278 			UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
279 			return 0;
280 		}
281 
282 		/* page is there, see if we need to wait on it */
283 		if ((pg->flags & PG_BUSY) != 0) {
284 			if (flags & UFP_NOWAIT) {
285 				UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
286 				return 0;
287 			}
288 			pg->flags |= PG_WANTED;
289 			UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
290 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
291 					    "uvn_fp2", 0);
292 			simple_lock(&uobj->vmobjlock);
293 			continue;
294 		}
295 
296 		/* skip PG_RDONLY pages if requested */
297 		if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
298 			UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
299 			return 0;
300 		}
301 
302 		/* stop on clean pages if requested */
303 		if (flags & UFP_DIRTYONLY) {
304 			dirty = pmap_clear_modify(pg) ||
305 				(pg->flags & PG_CLEAN) == 0;
306 			pg->flags |= PG_CLEAN;
307 			if (!dirty) {
308 				UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
309 				return 0;
310 			}
311 		}
312 
313 		/* mark the page BUSY and we're done. */
314 		pg->flags |= PG_BUSY;
315 		UVM_PAGE_OWN(pg, "uvn_findpage");
316 		UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
317 		break;
318 	}
319 	*pgp = pg;
320 	return 1;
321 }
322 
323 /*
324  * uvm_vnp_setsize: grow or shrink a vnode uobj
325  *
326  * grow   => just update size value
327  * shrink => toss un-needed pages
328  *
329  * => we assume that the caller has a reference of some sort to the
330  *	vnode in question so that it will not be yanked out from under
331  *	us.
332  */
333 
334 void
335 uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
336 {
337 	struct uvm_object *uobj = &vp->v_uobj;
338 	voff_t pgend = round_page(newsize);
339 	voff_t oldsize;
340 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
341 
342 	simple_lock(&uobj->vmobjlock);
343 	UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
344 	    vp, vp->v_size, newsize, 0);
345 
346 	/*
347 	 * now check if the size has changed: if we shrink we had better
348 	 * toss some pages...
349 	 */
350 
351 	KASSERT(newsize != VSIZENOTSET);
352 	KASSERT(vp->v_size <= vp->v_writesize);
353 	KASSERT(vp->v_size == vp->v_writesize ||
354 	    newsize == vp->v_writesize || newsize <= vp->v_size);
355 
356 	oldsize = vp->v_writesize;
357 	KASSERT(oldsize != VSIZENOTSET || pgend > oldsize);
358 
359 	if (oldsize > pgend) {
360 		(void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
361 		simple_lock(&uobj->vmobjlock);
362 	}
363 	vp->v_size = vp->v_writesize = newsize;
364 	simple_unlock(&uobj->vmobjlock);
365 }
366 
367 void
368 uvm_vnp_setwritesize(struct vnode *vp, voff_t newsize)
369 {
370 
371 	simple_lock(&vp->v_interlock);
372 	KASSERT(newsize != VSIZENOTSET);
373 	KASSERT(vp->v_size != VSIZENOTSET);
374 	KASSERT(vp->v_writesize != VSIZENOTSET);
375 	KASSERT(vp->v_size <= vp->v_writesize);
376 	KASSERT(vp->v_size <= newsize);
377 	vp->v_writesize = newsize;
378 	simple_unlock(&vp->v_interlock);
379 }
380 
381 /*
382  * uvm_vnp_zerorange:  set a range of bytes in a file to zero.
383  */
384 
385 void
386 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len)
387 {
388 	void *win;
389 	int flags;
390 
391 	/*
392 	 * XXXUBC invent kzero() and use it
393 	 */
394 
395 	while (len) {
396 		vsize_t bytelen = len;
397 
398 		win = ubc_alloc(&vp->v_uobj, off, &bytelen, UVM_ADV_NORMAL,
399 		    UBC_WRITE);
400 		memset(win, 0, bytelen);
401 		flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
402 		ubc_release(win, flags);
403 
404 		off += bytelen;
405 		len -= bytelen;
406 	}
407 }
408 
409 bool
410 uvn_text_p(struct uvm_object *uobj)
411 {
412 	struct vnode *vp = (struct vnode *)uobj;
413 
414 	return (vp->v_iflag & VI_EXECMAP) != 0;
415 }
416 
417 bool
418 uvn_clean_p(struct uvm_object *uobj)
419 {
420 	struct vnode *vp = (struct vnode *)uobj;
421 
422 	return (vp->v_iflag & VI_ONWORKLST) == 0;
423 }
424 
425 bool
426 uvn_needs_writefault_p(struct uvm_object *uobj)
427 {
428 	struct vnode *vp = (struct vnode *)uobj;
429 
430 	return uvn_clean_p(uobj) ||
431 	    (vp->v_iflag & (VI_WRMAP|VI_WRMAPDIRTY)) == VI_WRMAP;
432 }
433