xref: /netbsd-src/sys/uvm/uvm_vnode.c (revision 8a8f936f250a330d54f8a24ed0e92aadf9743a7b)
1 /*	$NetBSD: uvm_vnode.c,v 1.54 2001/09/26 07:23:51 chs 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 #include "fs_nfs.h"
49 #include "opt_uvmhist.h"
50 #include "opt_ddb.h"
51 
52 /*
53  * uvm_vnode.c: the vnode pager.
54  */
55 
56 #include <sys/param.h>
57 #include <sys/systm.h>
58 #include <sys/kernel.h>
59 #include <sys/proc.h>
60 #include <sys/malloc.h>
61 #include <sys/vnode.h>
62 #include <sys/disklabel.h>
63 #include <sys/ioctl.h>
64 #include <sys/fcntl.h>
65 #include <sys/conf.h>
66 #include <sys/pool.h>
67 #include <sys/mount.h>
68 
69 #include <miscfs/specfs/specdev.h>
70 
71 #include <uvm/uvm.h>
72 
73 /*
74  * functions
75  */
76 
77 void	uvn_detach __P((struct uvm_object *));
78 int	uvn_get __P((struct uvm_object *, voff_t, struct vm_page **, int *, int,
79 	    vm_prot_t, int, int));
80 int	uvn_put __P((struct uvm_object *, voff_t, voff_t, int));
81 void	uvn_reference __P((struct uvm_object *));
82 
83 int	uvn_findpage __P((struct uvm_object *, voff_t, struct vm_page **, int));
84 
85 /*
86  * master pager structure
87  */
88 
89 struct uvm_pagerops uvm_vnodeops = {
90 	NULL,
91 	uvn_reference,
92 	uvn_detach,
93 	NULL,
94 	uvn_get,
95 	uvn_put,
96 };
97 
98 /*
99  * the ops!
100  */
101 
102 /*
103  * uvn_attach
104  *
105  * attach a vnode structure to a VM object.  if the vnode is already
106  * attached, then just bump the reference count by one and return the
107  * VM object.   if not already attached, attach and return the new VM obj.
108  * the "accessprot" tells the max access the attaching thread wants to
109  * our pages.
110  *
111  * => caller must _not_ already be holding the lock on the uvm_object.
112  * => in fact, nothing should be locked so that we can sleep here.
113  * => note that uvm_object is first thing in vnode structure, so their
114  *    pointers are equiv.
115  */
116 
117 struct uvm_object *
118 uvn_attach(arg, accessprot)
119 	void *arg;
120 	vm_prot_t accessprot;
121 {
122 	struct vnode *vp = arg;
123 	struct uvm_object *uobj = &vp->v_uobj;
124 	struct vattr vattr;
125 	int result;
126 	struct partinfo pi;
127 	voff_t used_vnode_size;
128 	UVMHIST_FUNC("uvn_attach"); UVMHIST_CALLED(maphist);
129 
130 	UVMHIST_LOG(maphist, "(vn=0x%x)", arg,0,0,0);
131 	used_vnode_size = (voff_t)0;
132 
133 	/*
134 	 * first get a lock on the uobj.
135 	 */
136 
137 	simple_lock(&uobj->vmobjlock);
138 	while (vp->v_flag & VXLOCK) {
139 		vp->v_flag |= VXWANT;
140 		UVMHIST_LOG(maphist, "  SLEEPING on blocked vn",0,0,0,0);
141 		UVM_UNLOCK_AND_WAIT(uobj, &uobj->vmobjlock, FALSE,
142 		    "uvn_attach", 0);
143 		simple_lock(&uobj->vmobjlock);
144 		UVMHIST_LOG(maphist,"  WOKE UP",0,0,0,0);
145 	}
146 
147 	/*
148 	 * if we're mapping a BLK device, make sure it is a disk.
149 	 */
150 	if (vp->v_type == VBLK && bdevsw[major(vp->v_rdev)].d_type != D_DISK) {
151 		simple_unlock(&uobj->vmobjlock);
152 		UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)", 0,0,0,0);
153 		return(NULL);
154 	}
155 	KASSERT(vp->v_type == VREG || vp->v_type == VBLK);
156 
157 	/*
158 	 * set up our idea of the size
159 	 * if this hasn't been done already.
160 	 */
161 	if (vp->v_size == VSIZENOTSET) {
162 
163 
164 	vp->v_flag |= VXLOCK;
165 	simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */
166 		/* XXX: curproc? */
167 	if (vp->v_type == VBLK) {
168 		/*
169 		 * We could implement this as a specfs getattr call, but:
170 		 *
171 		 *	(1) VOP_GETATTR() would get the file system
172 		 *	    vnode operation, not the specfs operation.
173 		 *
174 		 *	(2) All we want is the size, anyhow.
175 		 */
176 		result = (*bdevsw[major(vp->v_rdev)].d_ioctl)(vp->v_rdev,
177 		    DIOCGPART, (caddr_t)&pi, FREAD, curproc);
178 		if (result == 0) {
179 			/* XXX should remember blocksize */
180 			used_vnode_size = (voff_t)pi.disklab->d_secsize *
181 			    (voff_t)pi.part->p_size;
182 		}
183 	} else {
184 		result = VOP_GETATTR(vp, &vattr, curproc->p_ucred, curproc);
185 		if (result == 0)
186 			used_vnode_size = vattr.va_size;
187 	}
188 
189 	/* relock object */
190 	simple_lock(&uobj->vmobjlock);
191 
192 	if (vp->v_flag & VXWANT) {
193 		wakeup(vp);
194 	}
195 	vp->v_flag &= ~(VXLOCK|VXWANT);
196 
197 	if (result != 0) {
198 		simple_unlock(&uobj->vmobjlock);
199 		UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0);
200 		return(NULL);
201 	}
202 	vp->v_size = used_vnode_size;
203 
204 	}
205 
206 	simple_unlock(&uobj->vmobjlock);
207 	UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount,
208 	    0, 0, 0);
209 	return uobj;
210 }
211 
212 
213 /*
214  * uvn_reference
215  *
216  * duplicate a reference to a VM object.  Note that the reference
217  * count must already be at least one (the passed in reference) so
218  * there is no chance of the uvn being killed or locked out here.
219  *
220  * => caller must call with object unlocked.
221  * => caller must be using the same accessprot as was used at attach time
222  */
223 
224 void
225 uvn_reference(uobj)
226 	struct uvm_object *uobj;
227 {
228 	VREF((struct vnode *)uobj);
229 }
230 
231 
232 /*
233  * uvn_detach
234  *
235  * remove a reference to a VM object.
236  *
237  * => caller must call with object unlocked and map locked.
238  */
239 
240 void
241 uvn_detach(uobj)
242 	struct uvm_object *uobj;
243 {
244 	vrele((struct vnode *)uobj);
245 }
246 
247 /*
248  * uvn_put: flush page data to backing store.
249  *
250  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
251  * => flags: PGO_SYNCIO -- use sync. I/O
252  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
253  */
254 
255 int
256 uvn_put(uobj, offlo, offhi, flags)
257 	struct uvm_object *uobj;
258 	voff_t offlo;
259 	voff_t offhi;
260 	int flags;
261 {
262 	struct vnode *vp = (struct vnode *)uobj;
263 	int error;
264 
265 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
266 	error = VOP_PUTPAGES(vp, offlo, offhi, flags);
267 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
268 	return error;
269 }
270 
271 
272 /*
273  * uvn_get: get pages (synchronously) from backing store
274  *
275  * => prefer map unlocked (not required)
276  * => object must be locked!  we will _unlock_ it before starting any I/O.
277  * => flags: PGO_ALLPAGES: get all of the pages
278  *           PGO_LOCKED: fault data structures are locked
279  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
280  * => NOTE: caller must check for released pages!!
281  */
282 
283 int
284 uvn_get(uobj, offset, pps, npagesp, centeridx, access_type, advice, flags)
285 	struct uvm_object *uobj;
286 	voff_t offset;
287 	struct vm_page **pps;		/* IN/OUT */
288 	int *npagesp;			/* IN (OUT if PGO_LOCKED) */
289 	int centeridx;
290 	vm_prot_t access_type;
291 	int advice, flags;
292 {
293 	struct vnode *vp = (struct vnode *)uobj;
294 	int error;
295 	UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
296 
297 	UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
298 	error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
299 			     access_type, advice, flags);
300 	return error;
301 }
302 
303 
304 /*
305  * uvn_findpages:
306  * return the page for the uobj and offset requested, allocating if needed.
307  * => uobj must be locked.
308  * => returned pages will be BUSY.
309  */
310 
311 void
312 uvn_findpages(uobj, offset, npagesp, pgs, flags)
313 	struct uvm_object *uobj;
314 	voff_t offset;
315 	int *npagesp;
316 	struct vm_page **pgs;
317 	int flags;
318 {
319 	int i, count, npages, rv;
320 
321 	count = 0;
322 	npages = *npagesp;
323 	if (flags & UFP_BACKWARD) {
324 		for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
325 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
326 			if (flags & UFP_DIRTYONLY && rv == 0) {
327 				break;
328 			}
329 			count++;
330 		}
331 	} else {
332 		for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
333 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
334 			if (flags & UFP_DIRTYONLY && rv == 0) {
335 				break;
336 			}
337 			count++;
338 		}
339 	}
340 	*npagesp = count;
341 }
342 
343 int
344 uvn_findpage(uobj, offset, pgp, flags)
345 	struct uvm_object *uobj;
346 	voff_t offset;
347 	struct vm_page **pgp;
348 	int flags;
349 {
350 	struct vm_page *pg;
351 	boolean_t dirty;
352 	UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
353 	UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
354 
355 	if (*pgp != NULL) {
356 		UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
357 		return 0;
358 	}
359 	for (;;) {
360 		/* look for an existing page */
361 		pg = uvm_pagelookup(uobj, offset);
362 
363 		/* nope?  allocate one now */
364 		if (pg == NULL) {
365 			if (flags & UFP_NOALLOC) {
366 				UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
367 				return 0;
368 			}
369 			pg = uvm_pagealloc(uobj, offset, NULL, 0);
370 			if (pg == NULL) {
371 				if (flags & UFP_NOWAIT) {
372 					UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
373 					return 0;
374 				}
375 				simple_unlock(&uobj->vmobjlock);
376 				uvm_wait("uvn_fp1");
377 				simple_lock(&uobj->vmobjlock);
378 				continue;
379 			}
380 			if (UVM_OBJ_IS_VTEXT(uobj)) {
381 				uvmexp.vtextpages++;
382 			} else {
383 				uvmexp.vnodepages++;
384 			}
385 			UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
386 			break;
387 		} else if (flags & UFP_NOCACHE) {
388 			UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
389 			return 0;
390 		}
391 
392 		/* page is there, see if we need to wait on it */
393 		if ((pg->flags & PG_BUSY) != 0) {
394 			if (flags & UFP_NOWAIT) {
395 				UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
396 				return 0;
397 			}
398 			pg->flags |= PG_WANTED;
399 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
400 					    "uvn_fp2", 0);
401 			simple_lock(&uobj->vmobjlock);
402 			continue;
403 		}
404 
405 		/* skip PG_RDONLY pages if requested */
406 		if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
407 			UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
408 			return 0;
409 		}
410 
411 		/* stop on clean pages if requested */
412 		if (flags & UFP_DIRTYONLY) {
413 			dirty = pmap_clear_modify(pg) ||
414 				(pg->flags & PG_CLEAN) == 0;
415 			pg->flags |= PG_CLEAN;
416 			if (!dirty) {
417 				return 0;
418 			}
419 		}
420 
421 		/* mark the page BUSY and we're done. */
422 		pg->flags |= PG_BUSY;
423 		UVM_PAGE_OWN(pg, "uvn_findpage");
424 		UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
425 		break;
426 	}
427 	*pgp = pg;
428 	return 1;
429 }
430 
431 /*
432  * uvm_vnp_setsize: grow or shrink a vnode uobj
433  *
434  * grow   => just update size value
435  * shrink => toss un-needed pages
436  *
437  * => we assume that the caller has a reference of some sort to the
438  *	vnode in question so that it will not be yanked out from under
439  *	us.
440  */
441 
442 void
443 uvm_vnp_setsize(vp, newsize)
444 	struct vnode *vp;
445 	voff_t newsize;
446 {
447 	struct uvm_object *uobj = &vp->v_uobj;
448 	voff_t pgend = round_page(newsize);
449 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
450 
451 	simple_lock(&uobj->vmobjlock);
452 	UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
453 	    vp, vp->v_size, newsize, 0);
454 
455 	/*
456 	 * now check if the size has changed: if we shrink we had better
457 	 * toss some pages...
458 	 */
459 
460 	if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) {
461 		(void) uvn_put(uobj, pgend, 0, PGO_FREE);
462 	} else {
463 		simple_unlock(&uobj->vmobjlock);
464 	}
465 	vp->v_size = newsize;
466 }
467 
468 /*
469  * uvm_vnp_zerorange:  set a range of bytes in a file to zero.
470  */
471 
472 void
473 uvm_vnp_zerorange(vp, off, len)
474 	struct vnode *vp;
475 	off_t off;
476 	size_t len;
477 {
478         void *win;
479 
480         /*
481          * XXXUBC invent kzero() and use it
482          */
483 
484         while (len) {
485                 vsize_t bytelen = len;
486 
487                 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UBC_WRITE);
488                 memset(win, 0, bytelen);
489                 ubc_release(win, 0);
490 
491                 off += bytelen;
492                 len -= bytelen;
493         }
494 }
495