FreeBSD/Linux Kernel Cross Reference
sys/uvm/uvm_vnode.c
1 /* $NetBSD: uvm_vnode.c,v 1.77.2.1 2007/02/17 23:27:53 tron 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 1.77.2.1 2007/02/17 23:27:53 tron 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_attach
109 *
110 * attach a vnode structure to a VM object. if the vnode is already
111 * attached, then just bump the reference count by one and return the
112 * VM object. if not already attached, attach and return the new VM obj.
113 * the "accessprot" tells the max access the attaching thread wants to
114 * our pages.
115 *
116 * => caller must _not_ already be holding the lock on the uvm_object.
117 * => in fact, nothing should be locked so that we can sleep here.
118 * => note that uvm_object is first thing in vnode structure, so their
119 * pointers are equiv.
120 */
121
122 struct uvm_object *
123 uvn_attach(void *arg, vm_prot_t accessprot)
124 {
125 struct vnode *vp = arg;
126 struct uvm_object *uobj = &vp->v_uobj;
127 struct vattr vattr;
128 const struct bdevsw *bdev;
129 int result;
130 struct partinfo pi;
131 voff_t used_vnode_size;
132 UVMHIST_FUNC("uvn_attach"); UVMHIST_CALLED(maphist);
133
134 UVMHIST_LOG(maphist, "(vn=0x%x)", arg,0,0,0);
135 used_vnode_size = (voff_t)0;
136
137 /*
138 * first get a lock on the uobj.
139 */
140
141 simple_lock(&uobj->vmobjlock);
142 while (vp->v_flag & VXLOCK) {
143 vp->v_flag |= VXWANT;
144 UVMHIST_LOG(maphist, " SLEEPING on blocked vn",0,0,0,0);
145 UVM_UNLOCK_AND_WAIT(uobj, &uobj->vmobjlock, FALSE,
146 "uvn_attach", 0);
147 simple_lock(&uobj->vmobjlock);
148 UVMHIST_LOG(maphist," WOKE UP",0,0,0,0);
149 }
150
151 /*
152 * if we're mapping a BLK device, make sure it is a disk.
153 */
154 if (vp->v_type == VBLK) {
155 bdev = bdevsw_lookup(vp->v_rdev);
156 if (bdev == NULL || bdev->d_type != D_DISK) {
157 simple_unlock(&uobj->vmobjlock);
158 UVMHIST_LOG(maphist,"<- done (VBLK not D_DISK!)",
159 0,0,0,0);
160 return(NULL);
161 }
162 }
163 KASSERT(vp->v_type == VREG || vp->v_type == VBLK);
164
165 /*
166 * set up our idea of the size
167 * if this hasn't been done already.
168 */
169 if (vp->v_size == VSIZENOTSET) {
170
171
172 vp->v_flag |= VXLOCK;
173 simple_unlock(&uobj->vmobjlock); /* drop lock in case we sleep */
174 /* XXX: curproc? */
175 if (vp->v_type == VBLK) {
176 /*
177 * We could implement this as a specfs getattr call, but:
178 *
179 * (1) VOP_GETATTR() would get the file system
180 * vnode operation, not the specfs operation.
181 *
182 * (2) All we want is the size, anyhow.
183 */
184 bdev = bdevsw_lookup(vp->v_rdev);
185 if (bdev != NULL) {
186 result = (*bdev->d_ioctl)(vp->v_rdev, DIOCGPART,
187 (caddr_t)&pi, FREAD, curlwp);
188 } else {
189 result = ENXIO;
190 }
191 if (result == 0) {
192 /* XXX should remember blocksize */
193 used_vnode_size = (voff_t)pi.disklab->d_secsize *
194 (voff_t)pi.part->p_size;
195 }
196 } else {
197 result = VOP_GETATTR(vp, &vattr, curlwp->l_cred, curlwp);
198 if (result == 0)
199 used_vnode_size = vattr.va_size;
200 }
201
202 /* relock object */
203 simple_lock(&uobj->vmobjlock);
204
205 if (vp->v_flag & VXWANT) {
206 wakeup(vp);
207 }
208 vp->v_flag &= ~(VXLOCK|VXWANT);
209
210 if (result != 0) {
211 simple_unlock(&uobj->vmobjlock);
212 UVMHIST_LOG(maphist,"<- done (VOP_GETATTR FAILED!)", 0,0,0,0);
213 return(NULL);
214 }
215 vp->v_size = used_vnode_size;
216
217 }
218
219 simple_unlock(&uobj->vmobjlock);
220 UVMHIST_LOG(maphist,"<- done, refcnt=%d", vp->v_usecount,
221 0, 0, 0);
222 return uobj;
223 }
224
225
226 /*
227 * uvn_reference
228 *
229 * duplicate a reference to a VM object. Note that the reference
230 * count must already be at least one (the passed in reference) so
231 * there is no chance of the uvn being killed or locked out here.
232 *
233 * => caller must call with object unlocked.
234 * => caller must be using the same accessprot as was used at attach time
235 */
236
237 static void
238 uvn_reference(struct uvm_object *uobj)
239 {
240 VREF((struct vnode *)uobj);
241 }
242
243
244 /*
245 * uvn_detach
246 *
247 * remove a reference to a VM object.
248 *
249 * => caller must call with object unlocked and map locked.
250 */
251
252 static void
253 uvn_detach(struct uvm_object *uobj)
254 {
255 vrele((struct vnode *)uobj);
256 }
257
258 /*
259 * uvn_put: flush page data to backing store.
260 *
261 * => object must be locked on entry! VOP_PUTPAGES must unlock it.
262 * => flags: PGO_SYNCIO -- use sync. I/O
263 * => note: caller must set PG_CLEAN and pmap_clear_modify (if needed)
264 */
265
266 static int
267 uvn_put(struct uvm_object *uobj, voff_t offlo, voff_t offhi, int flags)
268 {
269 struct vnode *vp = (struct vnode *)uobj;
270 int error;
271
272 LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
273 error = VOP_PUTPAGES(vp, offlo, offhi, flags);
274 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
275 return error;
276 }
277
278
279 /*
280 * uvn_get: get pages (synchronously) from backing store
281 *
282 * => prefer map unlocked (not required)
283 * => object must be locked! we will _unlock_ it before starting any I/O.
284 * => flags: PGO_ALLPAGES: get all of the pages
285 * PGO_LOCKED: fault data structures are locked
286 * => NOTE: offset is the offset of pps[0], _NOT_ pps[centeridx]
287 * => NOTE: caller must check for released pages!!
288 */
289
290 static int
291 uvn_get(struct uvm_object *uobj, voff_t offset,
292 struct vm_page **pps /* IN/OUT */,
293 int *npagesp /* IN (OUT if PGO_LOCKED)*/,
294 int centeridx, vm_prot_t access_type, int advice, int flags)
295 {
296 struct vnode *vp = (struct vnode *)uobj;
297 int error;
298
299 UVMHIST_FUNC("uvn_get"); UVMHIST_CALLED(ubchist);
300
301 UVMHIST_LOG(ubchist, "vp %p off 0x%x", vp, (int)offset, 0,0);
302
303 if ((access_type & VM_PROT_WRITE) == 0 && (flags & PGO_LOCKED) == 0) {
304 simple_unlock(&vp->v_interlock);
305 vn_ra_allocctx(vp);
306 uvm_ra_request(vp->v_ractx, advice, uobj, offset,
307 *npagesp << PAGE_SHIFT);
308 simple_lock(&vp->v_interlock);
309 }
310
311 error = VOP_GETPAGES(vp, offset, pps, npagesp, centeridx,
312 access_type, advice, flags);
313
314 LOCK_ASSERT(((flags & PGO_LOCKED) != 0 &&
315 simple_lock_held(&vp->v_interlock)) ||
316 ((flags & PGO_LOCKED) == 0 &&
317 !simple_lock_held(&vp->v_interlock)));
318 return error;
319 }
320
321
322 /*
323 * uvn_findpages:
324 * return the page for the uobj and offset requested, allocating if needed.
325 * => uobj must be locked.
326 * => returned pages will be BUSY.
327 */
328
329 int
330 uvn_findpages(struct uvm_object *uobj, voff_t offset, int *npagesp,
331 struct vm_page **pgs, int flags)
332 {
333 int i, count, found, npages, rv;
334
335 count = found = 0;
336 npages = *npagesp;
337 if (flags & UFP_BACKWARD) {
338 for (i = npages - 1; i >= 0; i--, offset -= PAGE_SIZE) {
339 rv = uvn_findpage(uobj, offset, &pgs[i], flags);
340 if (rv == 0) {
341 if (flags & UFP_DIRTYONLY)
342 break;
343 } else
344 found++;
345 count++;
346 }
347 } else {
348 for (i = 0; i < npages; i++, offset += PAGE_SIZE) {
349 rv = uvn_findpage(uobj, offset, &pgs[i], flags);
350 if (rv == 0) {
351 if (flags & UFP_DIRTYONLY)
352 break;
353 } else
354 found++;
355 count++;
356 }
357 }
358 *npagesp = count;
359 return (found);
360 }
361
362 static int
363 uvn_findpage(struct uvm_object *uobj, voff_t offset, struct vm_page **pgp,
364 int flags)
365 {
366 struct vm_page *pg;
367 boolean_t dirty;
368 UVMHIST_FUNC("uvn_findpage"); UVMHIST_CALLED(ubchist);
369 UVMHIST_LOG(ubchist, "vp %p off 0x%lx", uobj, offset,0,0);
370
371 if (*pgp != NULL) {
372 UVMHIST_LOG(ubchist, "dontcare", 0,0,0,0);
373 return 0;
374 }
375 for (;;) {
376 /* look for an existing page */
377 pg = uvm_pagelookup(uobj, offset);
378
379 /* nope? allocate one now */
380 if (pg == NULL) {
381 if (flags & UFP_NOALLOC) {
382 UVMHIST_LOG(ubchist, "noalloc", 0,0,0,0);
383 return 0;
384 }
385 pg = uvm_pagealloc(uobj, offset, NULL, 0);
386 if (pg == NULL) {
387 if (flags & UFP_NOWAIT) {
388 UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
389 return 0;
390 }
391 simple_unlock(&uobj->vmobjlock);
392 uvm_wait("uvn_fp1");
393 simple_lock(&uobj->vmobjlock);
394 continue;
395 }
396 UVMHIST_LOG(ubchist, "alloced %p", pg,0,0,0);
397 break;
398 } else if (flags & UFP_NOCACHE) {
399 UVMHIST_LOG(ubchist, "nocache",0,0,0,0);
400 return 0;
401 }
402
403 /* page is there, see if we need to wait on it */
404 if ((pg->flags & PG_BUSY) != 0) {
405 if (flags & UFP_NOWAIT) {
406 UVMHIST_LOG(ubchist, "nowait",0,0,0,0);
407 return 0;
408 }
409 pg->flags |= PG_WANTED;
410 UVMHIST_LOG(ubchist, "wait %p", pg,0,0,0);
411 UVM_UNLOCK_AND_WAIT(pg, &uobj->vmobjlock, 0,
412 "uvn_fp2", 0);
413 simple_lock(&uobj->vmobjlock);
414 continue;
415 }
416
417 /* skip PG_RDONLY pages if requested */
418 if ((flags & UFP_NORDONLY) && (pg->flags & PG_RDONLY)) {
419 UVMHIST_LOG(ubchist, "nordonly",0,0,0,0);
420 return 0;
421 }
422
423 /* stop on clean pages if requested */
424 if (flags & UFP_DIRTYONLY) {
425 dirty = pmap_clear_modify(pg) ||
426 (pg->flags & PG_CLEAN) == 0;
427 pg->flags |= PG_CLEAN;
428 if (!dirty) {
429 UVMHIST_LOG(ubchist, "dirtonly", 0,0,0,0);
430 return 0;
431 }
432 }
433
434 /* mark the page BUSY and we're done. */
435 pg->flags |= PG_BUSY;
436 UVM_PAGE_OWN(pg, "uvn_findpage");
437 UVMHIST_LOG(ubchist, "found %p", pg,0,0,0);
438 break;
439 }
440 *pgp = pg;
441 return 1;
442 }
443
444 /*
445 * uvm_vnp_setsize: grow or shrink a vnode uobj
446 *
447 * grow => just update size value
448 * shrink => toss un-needed pages
449 *
450 * => we assume that the caller has a reference of some sort to the
451 * vnode in question so that it will not be yanked out from under
452 * us.
453 */
454
455 void
456 uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
457 {
458 struct uvm_object *uobj = &vp->v_uobj;
459 voff_t pgend = round_page(newsize);
460 voff_t oldsize;
461 UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
462
463 simple_lock(&uobj->vmobjlock);
464 UVMHIST_LOG(ubchist, "vp %p old 0x%x new 0x%x",
465 vp, vp->v_size, newsize, 0);
466
467 /*
468 * now check if the size has changed: if we shrink we had better
469 * toss some pages...
470 */
471
472 oldsize = vp->v_size;
473 if (oldsize > pgend && oldsize != VSIZENOTSET) {
474 (void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
475 simple_lock(&uobj->vmobjlock);
476 }
477 vp->v_size = newsize;
478 simple_unlock(&uobj->vmobjlock);
479 }
480
481 /*
482 * uvm_vnp_zerorange: set a range of bytes in a file to zero.
483 */
484
485 void
486 uvm_vnp_zerorange(struct vnode *vp, off_t off, size_t len)
487 {
488 void *win;
489 int flags;
490
491 /*
492 * XXXUBC invent kzero() and use it
493 */
494
495 while (len) {
496 vsize_t bytelen = len;
497
498 win = ubc_alloc(&vp->v_uobj, off, &bytelen, UVM_ADV_NORMAL,
499 UBC_WRITE);
500 memset(win, 0, bytelen);
501 flags = UBC_WANT_UNMAP(vp) ? UBC_UNMAP : 0;
502 ubc_release(win, flags);
503
504 off += bytelen;
505 len -= bytelen;
506 }
507 }
508
509 boolean_t
510 uvn_text_p(struct uvm_object *uobj)
511 {
512 struct vnode *vp = (struct vnode *)uobj;
513
514 return (vp->v_flag & VEXECMAP) != 0;
515 }
516
517 boolean_t
518 uvn_clean_p(struct uvm_object *uobj)
519 {
520 struct vnode *vp = (struct vnode *)uobj;
521
522 return (vp->v_flag & VONWORKLST) == 0;
523 }
524
525 boolean_t
526 uvn_needs_writefault_p(struct uvm_object *uobj)
527 {
528 struct vnode *vp = (struct vnode *)uobj;
529
530 return uvn_clean_p(uobj) ||
531 (vp->v_flag & (VWRITEMAP|VWRITEMAPDIRTY)) == VWRITEMAP;
532 }
Cache object: f36f83ca355c766ac4197ad84e02bfd5
|