xref: /netbsd-src/sys/uvm/uvm_vnode.c (revision df0caa2637da0538ecdf6b878c4d08e684b43d8f)
1 /*	$NetBSD: uvm_vnode.c,v 1.66 2005/06/27 02:29:32 thorpej 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 1.66 2005/06/27 02:29:32 thorpej 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 
76 /*
77  * functions
78  */
79 
80 static void	uvn_detach(struct uvm_object *);
81 static int	uvn_get(struct uvm_object *, voff_t, struct vm_page **, int *,
82 			int, vm_prot_t, int, int);
83 static int	uvn_put(struct uvm_object *, voff_t, voff_t, int);
84 static void	uvn_reference(struct uvm_object *);
85 
86 static int	uvn_findpage(struct uvm_object *, voff_t, struct vm_page **,
87 			     int);
88 
89 /*
90  * master pager structure
91  */
92 
93 struct uvm_pagerops uvm_vnodeops = {
94 	NULL,
95 	uvn_reference,
96 	uvn_detach,
97 	NULL,
98 	uvn_get,
99 	uvn_put,
100 };
101 
102 /*
103  * the ops!
104  */
105 
106 /*
107  * uvn_attach
108  *
109  * attach a vnode structure to a VM object.  if the vnode is already
110  * attached, then just bump the reference count by one and return the
111  * VM object.   if not already attached, attach and return the new VM obj.
112  * the "accessprot" tells the max access the attaching thread wants to
113  * our pages.
114  *
115  * => caller must _not_ already be holding the lock on the uvm_object.
116  * => in fact, nothing should be locked so that we can sleep here.
117  * => note that uvm_object is first thing in vnode structure, so their
118  *    pointers are equiv.
119  */
120 
121 struct uvm_object *
122 uvn_attach(void *arg, vm_prot_t accessprot)
123 {
124 	struct vnode *vp = arg;
125 	struct uvm_object *uobj = &vp->v_uobj;
126 	struct vattr vattr;
127 	const struct bdevsw *bdev;
128 	int result;
129 	struct partinfo pi;
130 	voff_t used_vnode_size;
131 	UVMHIST_FUNC("uvn_attach"); UVMHIST_CALLED(maphist);
132 
133 	UVMHIST_LOG(maphist, "(vn=0x%x)", arg,0,0,0);
134 	used_vnode_size = (voff_t)0;
135 
136 	/*
137 	 * first get a lock on the uobj.
138 	 */
139 
140 	simple_lock(&uobj->vmobjlock);
141 	while (vp->v_flag & VXLOCK) {
142 		vp->v_flag |= VXWANT;
143 		UVMHIST_LOG(maphist, "  SLEEPING on blocked vn",0,0,0,0);
144 		UVM_UNLOCK_AND_WAIT(uobj, &uobj->vmobjlock, FALSE,
145 		    "uvn_attach", 0);
146 		simple_lock(&uobj->vmobjlock);
147 		UVMHIST_LOG(maphist,"  WOKE UP",0,0,0,0);
148 	}
149 
150 	/*
151 	 * if we're mapping a BLK device, make sure it is a disk.
152 	 */
153 	if (vp->v_type == VBLK) {
154 		bdev = bdevsw_lookup(vp->v_rdev);
155 		if (bdev == NULL || bdev->d_type != D_DISK) {
156 			simple_unlock(&uobj->vmobjlock);
157 			UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)",
158 				    0,0,0,0);
159 			return(NULL);
160 		}
161 	}
162 	KASSERT(vp->v_type == VREG || vp->v_type == VBLK);
163 
164 	/*
165 	 * set up our idea of the size
166 	 * if this hasn't been done already.
167 	 */
168 	if (vp->v_size == VSIZENOTSET) {
169 
170 
171 	vp->v_flag |= VXLOCK;
172 	simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */
173 		/* XXX: curproc? */
174 	if (vp->v_type == VBLK) {
175 		/*
176 		 * We could implement this as a specfs getattr call, but:
177 		 *
178 		 *	(1) VOP_GETATTR() would get the file system
179 		 *	    vnode operation, not the specfs operation.
180 		 *
181 		 *	(2) All we want is the size, anyhow.
182 		 */
183 		bdev = bdevsw_lookup(vp->v_rdev);
184 		if (bdev != NULL) {
185 			result = (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART,
186 						  (caddr_t)&pi, FREAD, curproc);
187 		} else {
188 			result = ENXIO;
189 		}
190 		if (result == 0) {
191 			/* XXX should remember blocksize */
192 			used_vnode_size = (voff_t)pi.disklab->d_secsize *
193 			    (voff_t)pi.part->p_size;
194 		}
195 	} else {
196 		result = VOP_GETATTR(vp, &vattr, curproc->p_ucred, curproc);
197 		if (result == 0)
198 			used_vnode_size = vattr.va_size;
199 	}
200 
201 	/* relock object */
202 	simple_lock(&uobj->vmobjlock);
203 
204 	if (vp->v_flag & VXWANT) {
205 		wakeup(vp);
206 	}
207 	vp->v_flag &= ~(VXLOCK|VXWANT);
208 
209 	if (result != 0) {
210 		simple_unlock(&uobj->vmobjlock);
211 		UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0);
212 		return(NULL);
213 	}
214 	vp->v_size = used_vnode_size;
215 
216 	}
217 
218 	simple_unlock(&uobj->vmobjlock);
219 	UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount,
220 	    0, 0, 0);
221 	return uobj;
222 }
223 
224 
225 /*
226  * uvn_reference
227  *
228  * duplicate a reference to a VM object.  Note that the reference
229  * count must already be at least one (the passed in reference) so
230  * there is no chance of the uvn being killed or locked out here.
231  *
232  * => caller must call with object unlocked.
233  * => caller must be using the same accessprot as was used at attach time
234  */
235 
236 static void
237 uvn_reference(struct uvm_object *uobj)
238 {
239 	VREF((struct vnode *)uobj);
240 }
241 
242 
243 /*
244  * uvn_detach
245  *
246  * remove a reference to a VM object.
247  *
248  * => caller must call with object unlocked and map locked.
249  */
250 
251 static void
252 uvn_detach(struct uvm_object *uobj)
253 {
254 	vrele((struct vnode *)uobj);
255 }
256 
257 /*
258  * uvn_put: flush page data to backing store.
259  *
260  * => object must be locked on entry!   VOP_PUTPAGES must unlock it.
261  * => flags: PGO_SYNCIO -- use sync. I/O
262  * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
263  */
264 
265 static int
266 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
267 {
268 	struct vnode *vp = (struct vnode *)uobj;
269 	int error;
270 
271 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
272 	error = VOP_PUTPAGES(vp, offlo, offhi, flags);
273 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
274 	return error;
275 }
276 
277 
278 /*
279  * uvn_get: get pages (synchronously) from backing store
280  *
281  * => prefer map unlocked (not required)
282  * => object must be locked!  we will _unlock_ it before starting any I/O.
283  * => flags: PGO_ALLPAGES: get all of the pages
284  *           PGO_LOCKED: fault data structures are locked
285  * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
286  * => NOTE: caller must check for released pages!!
287  */
288 
289 static int
290 uvn_get(struct uvm_object *uobj, voff_t offset,
291     struct vm_page **pps /* IN/OUT */,
292     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
293     int centeridx, vm_prot_t access_type, int advice, int flags)
294 {
295 	struct vnode *vp = (struct vnode *)uobj;
296 	int error;
297 	UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
298 
299 	UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
300 	error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
301 			     access_type, advice, flags);
302 	return error;
303 }
304 
305 
306 /*
307  * uvn_findpages:
308  * return the page for the uobj and offset requested, allocating if needed.
309  * => uobj must be locked.
310  * => returned pages will be BUSY.
311  */
312 
313 int
314 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
315     struct vm_page **pgs, int flags)
316 {
317 	int i, count, found, npages, rv;
318 
319 	count = found = 0;
320 	npages = *npagesp;
321 	if (flags & UFP_BACKWARD) {
322 		for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
323 			rv = uvn_findpage(uobj, offset, &pgs[i], flags);
324 			if (rv == 0) {
325 				if (flags & UFP_DIRTYONLY)
326 					break;
327 			} else
328 				found++;
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 (rv == 0) {
335 				if (flags & UFP_DIRTYONLY)
336 					break;
337 			} else
338 				found++;
339 			count++;
340 		}
341 	}
342 	*npagesp = count;
343 	return (found);
344 }
345 
346 static int
347 uvn_findpage(struct uvm_object *uobj, voff_t offset, 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 			UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
381 			break;
382 		} else if (flags & UFP_NOCACHE) {
383 			UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
384 			return 0;
385 		}
386 
387 		/* page is there, see if we need to wait on it */
388 		if ((pg->flags & PG_BUSY) != 0) {
389 			if (flags & UFP_NOWAIT) {
390 				UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
391 				return 0;
392 			}
393 			pg->flags |= PG_WANTED;
394 			UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
395 			UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
396 					    "uvn_fp2", 0);
397 			simple_lock(&uobj->vmobjlock);
398 			continue;
399 		}
400 
401 		/* skip PG_RDONLY pages if requested */
402 		if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
403 			UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
404 			return 0;
405 		}
406 
407 		/* stop on clean pages if requested */
408 		if (flags & UFP_DIRTYONLY) {
409 			dirty = pmap_clear_modify(pg) ||
410 				(pg->flags & PG_CLEAN) == 0;
411 			pg->flags |= PG_CLEAN;
412 			if (!dirty) {
413 				UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
414 				return 0;
415 			}
416 		}
417 
418 		/* mark the page BUSY and we're done. */
419 		pg->flags |= PG_BUSY;
420 		UVM_PAGE_OWN(pg, "uvn_findpage");
421 		UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
422 		break;
423 	}
424 	*pgp = pg;
425 	return 1;
426 }
427 
428 /*
429  * uvm_vnp_setsize: grow or shrink a vnode uobj
430  *
431  * grow   => just update size value
432  * shrink => toss un-needed pages
433  *
434  * => we assume that the caller has a reference of some sort to the
435  *	vnode in question so that it will not be yanked out from under
436  *	us.
437  */
438 
439 void
440 uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
441 {
442 	struct uvm_object *uobj = &vp->v_uobj;
443 	voff_t pgend = round_page(newsize);
444 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
445 
446 	simple_lock(&uobj->vmobjlock);
447 	UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
448 	    vp, vp->v_size, newsize, 0);
449 
450 	/*
451 	 * now check if the size has changed: if we shrink we had better
452 	 * toss some pages...
453 	 */
454 
455 	if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) {
456 		(void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
457 	} else {
458 		simple_unlock(&uobj->vmobjlock);
459 	}
460 	vp->v_size = newsize;
461 }
462 
463 /*
464  * uvm_vnp_zerorange:  set a range of bytes in a file to zero.
465  */
466 
467 void
468 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len)
469 {
470 	void *win;
471 	int flags;
472 
473 	/*
474 	 * XXXUBC invent kzero() and use it
475 	 */
476 
477 	while (len) {
478 		vsize_t bytelen = len;
479 
480 		win = ubc_alloc(&vp->v_uobj, off, &bytelen, UBC_WRITE);
481 		memset(win, 0, bytelen);
482 		flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
483 		ubc_release(win, flags);
484 
485 		off += bytelen;
486 		len -= bytelen;
487 	}
488 }
489