]> git.saurik.com Git - apple/xnu.git/blame - bsd/kern/ubc_subr.c
xnu-344.2.tar.gz
[apple/xnu.git] / bsd / kern / ubc_subr.c
CommitLineData
1c79356b 1/*
0b4e3aa0 2 * Copyright (c) 1999-2001 Apple Computer, Inc. All rights reserved.
1c79356b
A
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
11 *
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
18 * under the License.
19 *
20 * @APPLE_LICENSE_HEADER_END@
21 */
22/*
23 * File: ubc_subr.c
24 * Author: Umesh Vaishampayan [umeshv@apple.com]
25 * 05-Aug-1999 umeshv Created.
26 *
27 * Functions related to Unified Buffer cache.
28 *
0b4e3aa0
A
29 * Caller of UBC functions MUST have a valid reference on the vnode.
30 *
1c79356b
A
31 */
32
0b4e3aa0 33#undef DIAGNOSTIC
1c79356b
A
34#define DIAGNOSTIC 1
35
36#include <sys/types.h>
37#include <sys/param.h>
38#include <sys/systm.h>
39#include <sys/lock.h>
40#include <sys/ubc.h>
41#include <sys/mount.h>
42#include <sys/vnode.h>
43#include <sys/ubc.h>
44#include <sys/ucred.h>
45#include <sys/proc.h>
46#include <sys/buf.h>
47
48#include <mach/mach_types.h>
49#include <mach/memory_object_types.h>
50
51#include <kern/zalloc.h>
52
53#if DIAGNOSTIC
54#if defined(assert)
55#undef assert()
56#endif
57#define assert(cond) \
9bccf70c 58 ((void) ((cond) ? 0 : panic("%s:%d (%s)", __FILE__, __LINE__, # cond)))
1c79356b
A
59#else
60#include <kern/assert.h>
61#endif /* DIAGNOSTIC */
62
63struct zone *ubc_info_zone;
64
0b4e3aa0
A
65/* lock for changes to struct UBC */
66static __inline__ void
67ubc_lock(struct vnode *vp)
1c79356b 68{
0b4e3aa0
A
69 /* For now, just use the v_interlock */
70 simple_lock(&vp->v_interlock);
71}
1c79356b 72
0b4e3aa0
A
73/* unlock */
74static __inline__ void
75ubc_unlock(struct vnode *vp)
76{
77 /* For now, just use the v_interlock */
78 simple_unlock(&vp->v_interlock);
1c79356b
A
79}
80
81/*
82 * Initialization of the zone for Unified Buffer Cache.
83 */
0b4e3aa0 84__private_extern__ void
1c79356b
A
85ubc_init()
86{
87 int i;
88
89 i = (vm_size_t) sizeof (struct ubc_info);
90 /* XXX the number of elements should be tied in to maxvnodes */
91 ubc_info_zone = zinit (i, 10000*i, 8192, "ubc_info zone");
92 return;
93}
94
95/*
96 * Initialize a ubc_info structure for a vnode.
97 */
98int
99ubc_info_init(struct vnode *vp)
100{
101 register struct ubc_info *uip;
102 void * pager;
103 struct vattr vattr;
104 struct proc *p = current_proc();
105 int error = 0;
106 kern_return_t kret;
0b4e3aa0 107 memory_object_control_t control;
1c79356b 108
0b4e3aa0
A
109 if (!UBCISVALID(vp))
110 return (EINVAL);
1c79356b
A
111
112 ubc_lock(vp);
113 if (ISSET(vp->v_flag, VUINIT)) {
114 /*
115 * other thread is already doing this
116 * wait till done
117 */
118 while (ISSET(vp->v_flag, VUINIT)) {
119 SET(vp->v_flag, VUWANT); /* XXX overloaded! */
120 ubc_unlock(vp);
121 (void) tsleep((caddr_t)vp, PINOD, "ubcinfo", 0);
122 ubc_lock(vp);
123 }
124 ubc_unlock(vp);
125 return (0);
126 } else {
127 SET(vp->v_flag, VUINIT);
128 }
129
130 uip = vp->v_ubcinfo;
131 if ((uip == UBC_INFO_NULL) || (uip == UBC_NOINFO)) {
132 ubc_unlock(vp);
133 uip = (struct ubc_info *) zalloc(ubc_info_zone);
0b4e3aa0
A
134 uip->ui_pager = MEMORY_OBJECT_NULL;
135 uip->ui_control = MEMORY_OBJECT_CONTROL_NULL;
136 uip->ui_flags = UI_INITED;
1c79356b
A
137 uip->ui_vnode = vp;
138 uip->ui_ucred = NOCRED;
0b4e3aa0
A
139 uip->ui_refcount = 1;
140 uip->ui_size = 0;
141 uip->ui_mapped = 0;
142 ubc_lock(vp);
1c79356b 143 }
0b4e3aa0
A
144#if DIAGNOSTIC
145 else
146 Debugger("ubc_info_init: already");
147#endif /* DIAGNOSTIC */
1c79356b
A
148
149 assert(uip->ui_flags != UI_NONE);
150 assert(uip->ui_vnode == vp);
151
152#if 0
153 if(ISSET(uip->ui_flags, UI_HASPAGER))
154 goto done;
155#endif /* 0 */
156
157 /* now set this ubc_info in the vnode */
158 vp->v_ubcinfo = uip;
159 SET(uip->ui_flags, UI_HASPAGER);
160 ubc_unlock(vp);
161 pager = (void *)vnode_pager_setup(vp, uip->ui_pager);
162 assert(pager);
0b4e3aa0 163 ubc_setpager(vp, pager);
1c79356b
A
164
165 /*
0b4e3aa0
A
166 * Note: We can not use VOP_GETATTR() to get accurate
167 * value of ui_size. Thanks to NFS.
1c79356b
A
168 * nfs_getattr() can call vinvalbuf() and in this case
169 * ubc_info is not set up to deal with that.
170 * So use bogus size.
171 */
172
1c79356b 173 /*
0b4e3aa0
A
174 * create a vnode - vm_object association
175 * memory_object_create_named() creates a "named" reference on the
176 * memory object we hold this reference as long as the vnode is
177 * "alive." Since memory_object_create_named() took its own reference
178 * on the vnode pager we passed it, we can drop the reference
179 * vnode_pager_setup() returned here.
1c79356b 180 */
0b4e3aa0
A
181 kret = memory_object_create_named(pager,
182 (memory_object_size_t)uip->ui_size, &control);
183 vnode_pager_deallocate(pager);
184 if (kret != KERN_SUCCESS)
185 panic("ubc_info_init: memory_object_create_named returned %d", kret);
1c79356b 186
0b4e3aa0
A
187 assert(control);
188 uip->ui_control = control; /* cache the value of the mo control */
189 SET(uip->ui_flags, UI_HASOBJREF); /* with a named reference */
1c79356b 190 /* create a pager reference on the vnode */
0b4e3aa0 191 error = vnode_pager_vget(vp);
1c79356b 192 if (error)
0b4e3aa0 193 panic("ubc_info_init: vnode_pager_vget error = %d", error);
1c79356b
A
194
195 /* initialize the size */
196 error = VOP_GETATTR(vp, &vattr, p->p_ucred, p);
197
198 ubc_lock(vp);
199 uip->ui_size = (error ? 0: vattr.va_size);
200
201done:
202 CLR(vp->v_flag, VUINIT);
203 if (ISSET(vp->v_flag, VUWANT)) {
204 CLR(vp->v_flag, VUWANT);
205 ubc_unlock(vp);
206 wakeup((caddr_t)vp);
207 } else
208 ubc_unlock(vp);
209
0b4e3aa0 210 return (error);
1c79356b
A
211}
212
213/* Free the ubc_info */
0b4e3aa0
A
214static void
215ubc_info_free(struct ubc_info *uip)
1c79356b 216{
1c79356b
A
217 struct ucred *credp;
218
1c79356b
A
219 credp = uip->ui_ucred;
220 if (credp != NOCRED) {
221 uip->ui_ucred = NOCRED;
222 crfree(credp);
223 }
0b4e3aa0
A
224
225 if (uip->ui_control != MEMORY_OBJECT_CONTROL_NULL)
226 memory_object_control_deallocate(uip->ui_control);
227
1c79356b
A
228 zfree(ubc_info_zone, (vm_offset_t)uip);
229 return;
230}
231
0b4e3aa0
A
232void
233ubc_info_deallocate(struct ubc_info *uip)
234{
235 assert(uip->ui_refcount > 0);
236
237 if (uip->ui_refcount-- == 1)
238 ubc_info_free(uip);
239}
240
1c79356b
A
241/*
242 * Communicate with VM the size change of the file
243 * returns 1 on success, 0 on failure
244 */
245int
246ubc_setsize(struct vnode *vp, off_t nsize)
247{
248 off_t osize; /* ui_size before change */
249 off_t lastpg, olastpgend, lastoff;
250 struct ubc_info *uip;
0b4e3aa0 251 memory_object_control_t control;
1c79356b 252 kern_return_t kret;
1c79356b 253
1c79356b 254 assert(nsize >= (off_t)0);
1c79356b
A
255
256 if (UBCINVALID(vp))
0b4e3aa0 257 return (0);
1c79356b
A
258
259 if (!UBCINFOEXISTS(vp))
0b4e3aa0 260 return (0);
1c79356b
A
261
262 uip = vp->v_ubcinfo;
263 osize = uip->ui_size; /* call ubc_getsize() ??? */
264 /* Update the size before flushing the VM */
265 uip->ui_size = nsize;
266
267 if (nsize >= osize) /* Nothing more to do */
0b4e3aa0 268 return (1); /* return success */
1c79356b
A
269
270 /*
271 * When the file shrinks, invalidate the pages beyond the
272 * new size. Also get rid of garbage beyond nsize on the
273 * last page. The ui_size already has the nsize. This
274 * insures that the pageout would not write beyond the new
275 * end of the file.
276 */
277
1c79356b
A
278 lastpg = trunc_page_64(nsize);
279 olastpgend = round_page_64(osize);
0b4e3aa0
A
280 control = uip->ui_control;
281 assert(control);
1c79356b
A
282 lastoff = (nsize & PAGE_MASK_64);
283
284 /*
285 * If length is multiple of page size, we should not flush
286 * invalidating is sufficient
287 */
288 if (!lastoff) {
1c79356b 289 /* invalidate last page and old contents beyond nsize */
0b4e3aa0
A
290 kret = memory_object_lock_request(control,
291 (memory_object_offset_t)lastpg,
1c79356b 292 (memory_object_size_t)(olastpgend - lastpg),
0b4e3aa0
A
293 MEMORY_OBJECT_RETURN_NONE, MEMORY_OBJECT_DATA_FLUSH,
294 VM_PROT_NO_CHANGE);
1c79356b
A
295 if (kret != KERN_SUCCESS)
296 printf("ubc_setsize: invalidate failed (error = %d)\n", kret);
297
1c79356b
A
298 return ((kret == KERN_SUCCESS) ? 1 : 0);
299 }
300
1c79356b 301 /* flush the last page */
0b4e3aa0
A
302 kret = memory_object_lock_request(control,
303 (memory_object_offset_t)lastpg,
1c79356b 304 PAGE_SIZE_64,
0b4e3aa0
A
305 MEMORY_OBJECT_RETURN_DIRTY, FALSE,
306 VM_PROT_NO_CHANGE);
1c79356b
A
307
308 if (kret == KERN_SUCCESS) {
1c79356b 309 /* invalidate last page and old contents beyond nsize */
0b4e3aa0
A
310 kret = memory_object_lock_request(control,
311 (memory_object_offset_t)lastpg,
1c79356b 312 (memory_object_size_t)(olastpgend - lastpg),
0b4e3aa0
A
313 MEMORY_OBJECT_RETURN_NONE, MEMORY_OBJECT_DATA_FLUSH,
314 VM_PROT_NO_CHANGE);
1c79356b
A
315 if (kret != KERN_SUCCESS)
316 printf("ubc_setsize: invalidate failed (error = %d)\n", kret);
317 } else
318 printf("ubc_setsize: flush failed (error = %d)\n", kret);
319
1c79356b
A
320 return ((kret == KERN_SUCCESS) ? 1 : 0);
321}
322
323/*
324 * Get the size of the file
1c79356b
A
325 */
326off_t
327ubc_getsize(struct vnode *vp)
328{
1c79356b
A
329 return (vp->v_ubcinfo->ui_size);
330}
331
1c79356b
A
332/*
333 * Caller indicate that the object corresponding to the vnode
334 * can not be cached in object cache. Make it so.
335 * returns 1 on success, 0 on failure
1c79356b
A
336 */
337int
338ubc_uncache(struct vnode *vp)
339{
1c79356b
A
340 kern_return_t kret;
341 struct ubc_info *uip;
0b4e3aa0 342 memory_object_control_t control;
1c79356b 343 memory_object_perf_info_data_t perf;
1c79356b
A
344
345 if (!UBCINFOEXISTS(vp))
346 return (0);
347
348 uip = vp->v_ubcinfo;
349
350 assert(uip != UBC_INFO_NULL);
351
352 /*
353 * AGE it so that vfree() can make sure that it
354 * would get recycled soon after the last reference is gone
355 * This will insure that .nfs turds would not linger
356 */
357 vagevp(vp);
358
359 /* set the "do not cache" bit */
360 SET(uip->ui_flags, UI_DONTCACHE);
361
0b4e3aa0
A
362 control = uip->ui_control;
363 assert(control);
1c79356b
A
364
365 perf.cluster_size = PAGE_SIZE; /* XXX use real cluster_size. */
366 perf.may_cache = FALSE;
0b4e3aa0 367 kret = memory_object_change_attributes(control,
1c79356b
A
368 MEMORY_OBJECT_PERFORMANCE_INFO,
369 (memory_object_info_t) &perf,
0b4e3aa0 370 MEMORY_OBJECT_PERF_INFO_COUNT);
1c79356b
A
371
372 if (kret != KERN_SUCCESS) {
0b4e3aa0 373 printf("ubc_uncache: memory_object_change_attributes_named "
1c79356b 374 "kret = %d", kret);
1c79356b
A
375 return (0);
376 }
377
0b4e3aa0
A
378 ubc_release_named(vp);
379
1c79356b
A
380 return (1);
381}
382
1c79356b
A
383/*
384 * call ubc_clean() and ubc_uncache() on all the vnodes
385 * for this mount point.
386 * returns 1 on success, 0 on failure
387 */
0b4e3aa0 388__private_extern__ int
1c79356b
A
389ubc_umount(struct mount *mp)
390{
391 struct proc *p = current_proc();
392 struct vnode *vp, *nvp;
393 int ret = 1;
394
395loop:
396 simple_lock(&mntvnode_slock);
397 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
398 if (vp->v_mount != mp) {
399 simple_unlock(&mntvnode_slock);
400 goto loop;
401 }
402 nvp = vp->v_mntvnodes.le_next;
403 simple_unlock(&mntvnode_slock);
404 if (UBCINFOEXISTS(vp)) {
0b4e3aa0
A
405
406 /*
407 * Must get a valid reference on the vnode
408 * before callig UBC functions
409 */
410 if (vget(vp, 0, p)) {
411 ret = 0;
412 simple_lock(&mntvnode_slock);
413 continue; /* move on to the next vnode */
414 }
1c79356b
A
415 ret &= ubc_clean(vp, 0); /* do not invalidate */
416 ret &= ubc_uncache(vp);
0b4e3aa0 417 vrele(vp);
1c79356b
A
418 }
419 simple_lock(&mntvnode_slock);
420 }
421 simple_unlock(&mntvnode_slock);
422 return (ret);
423}
424
425/*
426 * Call ubc_unmount() for all filesystems.
427 * The list is traversed in reverse order
428 * of mounting to avoid dependencies.
429 */
0b4e3aa0 430__private_extern__ void
1c79356b
A
431ubc_unmountall()
432{
433 struct mount *mp, *nmp;
434
435 /*
436 * Since this only runs when rebooting, it is not interlocked.
437 */
438 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
439 nmp = mp->mnt_list.cqe_prev;
440 (void) ubc_umount(mp);
441 }
442}
443
444/* Get the credentials */
445struct ucred *
446ubc_getcred(struct vnode *vp)
447{
448 struct ubc_info *uip;
449
1c79356b
A
450 uip = vp->v_ubcinfo;
451
0b4e3aa0 452 if (UBCINVALID(vp))
1c79356b 453 return (NOCRED);
1c79356b
A
454
455 return (uip->ui_ucred);
456}
457
458/*
459 * Set the credentials
460 * existing credentials are not changed
461 * returns 1 on success and 0 on failure
462 */
1c79356b
A
463int
464ubc_setcred(struct vnode *vp, struct proc *p)
465{
466 struct ubc_info *uip;
467 struct ucred *credp;
468
1c79356b
A
469 uip = vp->v_ubcinfo;
470
0b4e3aa0 471 if (UBCINVALID(vp))
1c79356b 472 return (0);
1c79356b
A
473
474 credp = uip->ui_ucred;
475 if (credp == NOCRED) {
476 crhold(p->p_ucred);
477 uip->ui_ucred = p->p_ucred;
478 }
479
480 return (1);
481}
482
483/* Get the pager */
0b4e3aa0 484__private_extern__ memory_object_t
1c79356b
A
485ubc_getpager(struct vnode *vp)
486{
487 struct ubc_info *uip;
488
1c79356b
A
489 uip = vp->v_ubcinfo;
490
0b4e3aa0 491 if (UBCINVALID(vp))
1c79356b 492 return (0);
1c79356b
A
493
494 return (uip->ui_pager);
495}
496
497/*
498 * Get the memory object associated with this vnode
499 * If the vnode was reactivated, memory object would not exist.
500 * Unless "do not rectivate" was specified, look it up using the pager.
1c79356b
A
501 * If hold was requested create an object reference of one does not
502 * exist already.
503 */
504
0b4e3aa0
A
505memory_object_control_t
506ubc_getobject(struct vnode *vp, int flags)
1c79356b
A
507{
508 struct ubc_info *uip;
0b4e3aa0 509 memory_object_control_t control;
1c79356b
A
510
511 uip = vp->v_ubcinfo;
1c79356b 512
0b4e3aa0
A
513 if (UBCINVALID(vp))
514 return (0);
1c79356b 515
0b4e3aa0 516 ubc_lock(vp);
1c79356b 517
0b4e3aa0 518 control = uip->ui_control;
1c79356b 519
0b4e3aa0 520 if ((flags & UBC_HOLDOBJECT) && (!ISSET(uip->ui_flags, UI_HASOBJREF))) {
1c79356b 521
0b4e3aa0
A
522 /*
523 * Take a temporary reference on the ubc info so that it won't go
524 * away during our recovery attempt.
525 */
526 uip->ui_refcount++;
527 ubc_unlock(vp);
528 if (memory_object_recover_named(control, TRUE) == KERN_SUCCESS) {
529 ubc_lock(vp);
1c79356b 530 SET(uip->ui_flags, UI_HASOBJREF);
0b4e3aa0
A
531 ubc_unlock(vp);
532 } else {
533 control = MEMORY_OBJECT_CONTROL_NULL;
1c79356b 534 }
0b4e3aa0 535 ubc_info_deallocate(uip);
1c79356b 536
0b4e3aa0
A
537 } else {
538 ubc_unlock(vp);
1c79356b
A
539 }
540
0b4e3aa0 541 return (control);
1c79356b
A
542}
543
544/* Set the pager */
545int
0b4e3aa0 546ubc_setpager(struct vnode *vp, memory_object_t pager)
1c79356b
A
547{
548 struct ubc_info *uip;
549
1c79356b
A
550 uip = vp->v_ubcinfo;
551
0b4e3aa0 552 if (UBCINVALID(vp))
1c79356b 553 return (0);
1c79356b
A
554
555 uip->ui_pager = pager;
556 return (1);
557}
558
559int
560ubc_setflags(struct vnode * vp, int flags)
561{
562 struct ubc_info *uip;
563
0b4e3aa0
A
564 if (UBCINVALID(vp))
565 return (0);
1c79356b
A
566
567 uip = vp->v_ubcinfo;
568
1c79356b
A
569 SET(uip->ui_flags, flags);
570
0b4e3aa0 571 return (1);
1c79356b
A
572}
573
574int
575ubc_clearflags(struct vnode * vp, int flags)
576{
577 struct ubc_info *uip;
578
0b4e3aa0
A
579 if (UBCINVALID(vp))
580 return (0);
1c79356b
A
581
582 uip = vp->v_ubcinfo;
583
1c79356b
A
584 CLR(uip->ui_flags, flags);
585
0b4e3aa0 586 return (1);
1c79356b
A
587}
588
589
590int
591ubc_issetflags(struct vnode * vp, int flags)
592{
593 struct ubc_info *uip;
594
0b4e3aa0
A
595 if (UBCINVALID(vp))
596 return (0);
1c79356b
A
597
598 uip = vp->v_ubcinfo;
599
0b4e3aa0 600 return (ISSET(uip->ui_flags, flags));
1c79356b
A
601}
602
603off_t
604ubc_blktooff(struct vnode *vp, daddr_t blkno)
605{
606 off_t file_offset;
607 int error;
608
0b4e3aa0 609 if (UBCINVALID(vp))
1c79356b 610 return ((off_t)-1);
1c79356b
A
611
612 error = VOP_BLKTOOFF(vp, blkno, &file_offset);
613 if (error)
614 file_offset = -1;
615
616 return (file_offset);
617}
0b4e3aa0 618
1c79356b
A
619daddr_t
620ubc_offtoblk(struct vnode *vp, off_t offset)
621{
622 daddr_t blkno;
0b4e3aa0 623 int error = 0;
1c79356b 624
1c79356b
A
625 if (UBCINVALID(vp)) {
626 return ((daddr_t)-1);
627 }
628
629 error = VOP_OFFTOBLK(vp, offset, &blkno);
630 if (error)
631 blkno = -1;
632
633 return (blkno);
634}
635
636/*
637 * Cause the file data in VM to be pushed out to the storage
638 * it also causes all currently valid pages to be released
639 * returns 1 on success, 0 on failure
640 */
641int
642ubc_clean(struct vnode *vp, int invalidate)
643{
644 off_t size;
645 struct ubc_info *uip;
0b4e3aa0 646 memory_object_control_t control;
1c79356b
A
647 kern_return_t kret;
648 int flags = 0;
1c79356b
A
649
650 if (UBCINVALID(vp))
0b4e3aa0 651 return (0);
1c79356b
A
652
653 if (!UBCINFOEXISTS(vp))
0b4e3aa0 654 return (0);
1c79356b
A
655
656 /*
657 * if invalidate was requested, write dirty data and then discard
658 * the resident pages
659 */
660 if (invalidate)
661 flags = (MEMORY_OBJECT_DATA_FLUSH | MEMORY_OBJECT_DATA_NO_CHANGE);
662
1c79356b
A
663 uip = vp->v_ubcinfo;
664 size = uip->ui_size; /* call ubc_getsize() ??? */
665
0b4e3aa0
A
666 control = uip->ui_control;
667 assert(control);
1c79356b
A
668
669 vp->v_flag &= ~VHASDIRTY;
670 vp->v_clen = 0;
671
672 /* Write the dirty data in the file and discard cached pages */
0b4e3aa0
A
673 kret = memory_object_lock_request(control,
674 (memory_object_offset_t)0,
1c79356b
A
675 (memory_object_size_t)round_page_64(size),
676 MEMORY_OBJECT_RETURN_ALL, flags,
0b4e3aa0 677 VM_PROT_NO_CHANGE);
1c79356b 678
0b4e3aa0 679 if (kret != KERN_SUCCESS)
1c79356b 680 printf("ubc_clean: clean failed (error = %d)\n", kret);
1c79356b
A
681
682 return ((kret == KERN_SUCCESS) ? 1 : 0);
683}
684
685/*
686 * Cause the file data in VM to be pushed out to the storage
687 * currently valid pages are NOT invalidated
688 * returns 1 on success, 0 on failure
689 */
690int
691ubc_pushdirty(struct vnode *vp)
692{
693 off_t size;
694 struct ubc_info *uip;
0b4e3aa0 695 memory_object_control_t control;
1c79356b 696 kern_return_t kret;
1c79356b
A
697
698 if (UBCINVALID(vp))
0b4e3aa0 699 return (0);
1c79356b
A
700
701 if (!UBCINFOEXISTS(vp))
0b4e3aa0 702 return (0);
1c79356b 703
1c79356b
A
704 uip = vp->v_ubcinfo;
705 size = uip->ui_size; /* call ubc_getsize() ??? */
706
0b4e3aa0
A
707 control = uip->ui_control;
708 assert(control);
1c79356b
A
709
710 vp->v_flag &= ~VHASDIRTY;
711 vp->v_clen = 0;
712
713 /* Write the dirty data in the file and discard cached pages */
0b4e3aa0
A
714 kret = memory_object_lock_request(control,
715 (memory_object_offset_t)0,
1c79356b 716 (memory_object_size_t)round_page_64(size),
0b4e3aa0
A
717 MEMORY_OBJECT_RETURN_DIRTY, FALSE,
718 VM_PROT_NO_CHANGE);
1c79356b 719
0b4e3aa0 720 if (kret != KERN_SUCCESS)
1c79356b 721 printf("ubc_pushdirty: flush failed (error = %d)\n", kret);
1c79356b 722
0b4e3aa0
A
723 return ((kret == KERN_SUCCESS) ? 1 : 0);
724}
725
726/*
727 * Cause the file data in VM to be pushed out to the storage
728 * currently valid pages are NOT invalidated
729 * returns 1 on success, 0 on failure
730 */
731int
732ubc_pushdirty_range(struct vnode *vp, off_t offset, off_t size)
733{
734 struct ubc_info *uip;
735 memory_object_control_t control;
736 kern_return_t kret;
737
738 if (UBCINVALID(vp))
739 return (0);
740
741 if (!UBCINFOEXISTS(vp))
742 return (0);
743
744 uip = vp->v_ubcinfo;
745
746 control = uip->ui_control;
747 assert(control);
748
749 /* Write any dirty pages in the requested range of the file: */
750 kret = memory_object_lock_request(control,
751 (memory_object_offset_t)offset,
752 (memory_object_size_t)round_page_64(size),
753 MEMORY_OBJECT_RETURN_DIRTY, FALSE,
754 VM_PROT_NO_CHANGE);
755
756 if (kret != KERN_SUCCESS)
757 printf("ubc_pushdirty_range: flush failed (error = %d)\n", kret);
1c79356b
A
758
759 return ((kret == KERN_SUCCESS) ? 1 : 0);
760}
761
762/*
763 * Make sure the vm object does not vanish
764 * returns 1 if the hold count was incremented
765 * returns 0 if the hold count was not incremented
766 * This return value should be used to balance
767 * ubc_hold() and ubc_rele().
768 */
769int
770ubc_hold(struct vnode *vp)
771{
772 struct ubc_info *uip;
0b4e3aa0 773 memory_object_control_t object;
1c79356b
A
774
775 if (UBCINVALID(vp))
776 return (0);
777
778 if (!UBCINFOEXISTS(vp)) {
0b4e3aa0
A
779 /* must be invalid or dying vnode */
780 assert(UBCINVALID(vp) ||
781 ((vp->v_flag & VXLOCK) || (vp->v_flag & VTERMINATE)));
782 return (0);
1c79356b 783 }
0b4e3aa0 784
1c79356b 785 uip = vp->v_ubcinfo;
0b4e3aa0 786 assert(uip->ui_control != MEMORY_OBJECT_CONTROL_NULL);
1c79356b 787
0b4e3aa0
A
788 ubc_lock(vp);
789 uip->ui_refcount++;
1c79356b 790
0b4e3aa0
A
791 if (!ISSET(uip->ui_flags, UI_HASOBJREF)) {
792 ubc_unlock(vp);
793 if (memory_object_recover_named(uip->ui_control, TRUE) != KERN_SUCCESS) {
794 ubc_info_deallocate(uip);
795 return (0);
796 }
797 ubc_lock(vp);
798 SET(uip->ui_flags, UI_HASOBJREF);
799 ubc_unlock(vp);
800 } else {
801 ubc_unlock(vp);
802 }
803
804 assert(uip->ui_refcount > 0);
1c79356b
A
805 return (1);
806}
807
0b4e3aa0
A
808/*
809 * Drop the holdcount.
810 * release the reference on the vm object if the this is "uncached"
811 * ubc_info.
812 */
1c79356b
A
813void
814ubc_rele(struct vnode *vp)
815{
816 struct ubc_info *uip;
1c79356b
A
817
818 if (UBCINVALID(vp))
819 return;
820
821 if (!UBCINFOEXISTS(vp)) {
822 /* nothing more to do for a dying vnode */
823 if ((vp->v_flag & VXLOCK) || (vp->v_flag & VTERMINATE))
824 return;
825 panic("ubc_rele: can not");
826 }
827
828 uip = vp->v_ubcinfo;
829
0b4e3aa0
A
830 if (uip->ui_refcount == 1)
831 panic("ubc_rele: ui_refcount");
1c79356b 832
0b4e3aa0 833 --uip->ui_refcount;
1c79356b 834
0b4e3aa0
A
835 if ((uip->ui_refcount == 1)
836 && ISSET(uip->ui_flags, UI_DONTCACHE))
837 (void) ubc_release_named(vp);
1c79356b
A
838
839 return;
840}
841
842/*
0b4e3aa0 843 * The vnode is mapped explicitly, mark it so.
1c79356b 844 */
0b4e3aa0 845__private_extern__ void
1c79356b
A
846ubc_map(struct vnode *vp)
847{
848 struct ubc_info *uip;
1c79356b 849
0b4e3aa0 850 if (UBCINVALID(vp))
1c79356b 851 return;
1c79356b
A
852
853 if (!UBCINFOEXISTS(vp))
0b4e3aa0 854 return;
1c79356b 855
0b4e3aa0 856 ubc_lock(vp);
1c79356b
A
857 uip = vp->v_ubcinfo;
858
859 SET(uip->ui_flags, UI_WASMAPPED);
860 uip->ui_mapped = 1;
861 ubc_unlock(vp);
862
1c79356b 863 return;
1c79356b
A
864}
865
866/*
867 * Release the memory object reference on the vnode
868 * only if it is not in use
869 * Return 1 if the reference was released, 0 otherwise.
870 */
871int
0b4e3aa0 872ubc_release_named(struct vnode *vp)
1c79356b
A
873{
874 struct ubc_info *uip;
0b4e3aa0
A
875 memory_object_control_t control;
876 kern_return_t kret;
1c79356b
A
877
878 if (UBCINVALID(vp))
879 return (0);
880
881 if (!UBCINFOEXISTS(vp))
0b4e3aa0 882 return (0);
1c79356b
A
883
884 uip = vp->v_ubcinfo;
885
0b4e3aa0
A
886 /* can not release held or mapped vnodes */
887 if (ISSET(uip->ui_flags, UI_HASOBJREF) &&
888 (uip->ui_refcount == 1) && !uip->ui_mapped) {
889 control = uip->ui_control;
890 assert(control);
891 CLR(uip->ui_flags, UI_HASOBJREF);
892 kret = memory_object_release_name(control,
893 MEMORY_OBJECT_RESPECT_CACHE);
894 return ((kret != KERN_SUCCESS) ? 0 : 1);
895 } else
1c79356b 896 return (0);
0b4e3aa0 897}
1c79356b 898
0b4e3aa0
A
899/*
900 * This function used to called by extensions directly. Some may
901 * still exist with this behavior. In those cases, we will do the
902 * release as part of reclaiming or cleaning the vnode. We don't
903 * need anything explicit - so just stub this out until those callers
904 * get cleaned up.
905 */
906int
907ubc_release(
908 struct vnode *vp)
909{
910 return 0;
911}
912
913/*
914 * destroy the named reference for a given vnode
915 */
916__private_extern__ int
917ubc_destroy_named(
918 struct vnode *vp)
919{
920 memory_object_control_t control;
921 struct proc *p;
922 struct ubc_info *uip;
923 kern_return_t kret;
924
925 /*
926 * We may already have had the object terminated
927 * and the ubcinfo released as a side effect of
928 * some earlier processing. If so, pretend we did
929 * it, because it probably was a result of our
930 * efforts.
931 */
932 if (!UBCINFOEXISTS(vp))
1c79356b 933 return (1);
0b4e3aa0
A
934
935 uip = vp->v_ubcinfo;
936
937 /* can not destroy held vnodes */
938 if (uip->ui_refcount > 1)
1c79356b 939 return (0);
0b4e3aa0
A
940
941 /*
942 * Terminate the memory object.
943 * memory_object_destroy() will result in
944 * vnode_pager_no_senders().
945 * That will release the pager reference
946 * and the vnode will move to the free list.
947 */
948 control = ubc_getobject(vp, UBC_HOLDOBJECT);
949 if (control != MEMORY_OBJECT_CONTROL_NULL) {
950
951 if (ISSET(vp->v_flag, VTERMINATE))
952 panic("ubc_destroy_named: already teminating");
953 SET(vp->v_flag, VTERMINATE);
954
955 kret = memory_object_destroy(control, 0);
956 if (kret != KERN_SUCCESS)
957 return (0);
958
959 /*
960 * memory_object_destroy() is asynchronous
961 * with respect to vnode_pager_no_senders().
962 * wait for vnode_pager_no_senders() to clear
963 * VTERMINATE
964 */
965 while (ISSET(vp->v_flag, VTERMINATE)) {
966 SET(vp->v_flag, VTERMWANT);
967 (void)tsleep((caddr_t)&vp->v_ubcinfo,
968 PINOD, "ubc_destroy_named", 0);
969 }
970 }
971 return (1);
1c79356b
A
972}
973
0b4e3aa0 974
1c79356b
A
975/*
976 * Invalidate a range in the memory object that backs this
977 * vnode. The offset is truncated to the page boundary and the
978 * size is adjusted to include the last page in the range.
979 */
980int
981ubc_invalidate(struct vnode *vp, off_t offset, size_t size)
982{
983 struct ubc_info *uip;
0b4e3aa0 984 memory_object_control_t control;
1c79356b
A
985 kern_return_t kret;
986 off_t toff;
987 size_t tsize;
1c79356b
A
988
989 if (UBCINVALID(vp))
0b4e3aa0 990 return (0);
1c79356b
A
991
992 if (!UBCINFOEXISTS(vp))
0b4e3aa0 993 return (0);
1c79356b 994
1c79356b
A
995 toff = trunc_page_64(offset);
996 tsize = (size_t)(round_page_64(offset+size) - toff);
997 uip = vp->v_ubcinfo;
0b4e3aa0
A
998 control = uip->ui_control;
999 assert(control);
1c79356b
A
1000
1001 /* invalidate pages in the range requested */
0b4e3aa0
A
1002 kret = memory_object_lock_request(control,
1003 (memory_object_offset_t)toff,
1c79356b
A
1004 (memory_object_size_t)tsize,
1005 MEMORY_OBJECT_RETURN_NONE,
1006 (MEMORY_OBJECT_DATA_NO_CHANGE| MEMORY_OBJECT_DATA_FLUSH),
0b4e3aa0 1007 VM_PROT_NO_CHANGE);
1c79356b
A
1008 if (kret != KERN_SUCCESS)
1009 printf("ubc_invalidate: invalidate failed (error = %d)\n", kret);
1010
1c79356b
A
1011 return ((kret == KERN_SUCCESS) ? 1 : 0);
1012}
1013
1014/*
1015 * Find out whether a vnode is in use by UBC
1016 * Returns 1 if file is in use by UBC, 0 if not
1017 */
1018int
1019ubc_isinuse(struct vnode *vp, int tookref)
1020{
1021 int busycount = tookref ? 2 : 1;
1022
1023 if (!UBCINFOEXISTS(vp))
0b4e3aa0 1024 return (0);
1c79356b 1025
9bccf70c
A
1026 if (tookref == 0) {
1027 printf("ubc_isinuse: called without a valid reference"
1028 ": v_tag = %d\v", vp->v_tag);
1029 vprint("ubc_isinuse", vp);
1030 return (0);
1031 }
1032
1c79356b
A
1033 if (vp->v_usecount > busycount)
1034 return (1);
1035
1036 if ((vp->v_usecount == busycount)
1037 && (vp->v_ubcinfo->ui_mapped == 1))
0b4e3aa0 1038 return (1);
1c79356b 1039 else
0b4e3aa0 1040 return (0);
1c79356b
A
1041}
1042
1c79356b 1043/*
0b4e3aa0 1044 * The backdoor routine to clear the ui_mapped.
1c79356b
A
1045 * MUST only be called by the VM
1046 *
0b4e3aa0
A
1047 * Note that this routine is not called under funnel. There are numerous
1048 * things about the calling sequence that make this work on SMP.
1c79356b
A
1049 * Any code change in those paths can break this.
1050 *
1c79356b 1051 */
0b4e3aa0 1052__private_extern__ void
1c79356b
A
1053ubc_unmap(struct vnode *vp)
1054{
1055 struct ubc_info *uip;
0b4e3aa0 1056 boolean_t funnel_state;
1c79356b 1057
0b4e3aa0 1058 if (UBCINVALID(vp))
1c79356b 1059 return;
1c79356b
A
1060
1061 if (!UBCINFOEXISTS(vp))
0b4e3aa0 1062 return;
1c79356b
A
1063
1064 ubc_lock(vp);
1065 uip = vp->v_ubcinfo;
1c79356b 1066 uip->ui_mapped = 0;
0b4e3aa0
A
1067 if ((uip->ui_refcount > 1) || !ISSET(uip->ui_flags, UI_DONTCACHE)) {
1068 ubc_unlock(vp);
1069 return;
1070 }
1c79356b
A
1071 ubc_unlock(vp);
1072
0b4e3aa0
A
1073 funnel_state = thread_funnel_set(kernel_flock, TRUE);
1074 (void) ubc_release_named(vp);
1075 (void) thread_funnel_set(kernel_flock, funnel_state);
1076}
1077
1078kern_return_t
1079ubc_page_op(
1080 struct vnode *vp,
1081 off_t f_offset,
1082 int ops,
1083 vm_offset_t *phys_entryp,
1084 int *flagsp)
1085{
1086 memory_object_control_t control;
1087
1088 control = ubc_getobject(vp, UBC_FLAGS_NONE);
1089 if (control == MEMORY_OBJECT_CONTROL_NULL)
1090 return KERN_INVALID_ARGUMENT;
1091
1092 return (memory_object_page_op(control,
1093 (memory_object_offset_t)f_offset,
1094 ops,
1095 phys_entryp,
1096 flagsp));
1097}
1098
1099kern_return_t
1100ubc_create_upl(
1101 struct vnode *vp,
1102 off_t f_offset,
1103 long bufsize,
1104 upl_t *uplp,
1105 upl_page_info_t **plp,
1106 int uplflags)
1107{
1108 memory_object_control_t control;
1109 int count;
1110 off_t file_offset;
1111 kern_return_t kr;
1112
1113 if (bufsize & 0xfff)
1114 return KERN_INVALID_ARGUMENT;
1115
1116 control = ubc_getobject(vp, UBC_FLAGS_NONE);
1117 if (control == MEMORY_OBJECT_CONTROL_NULL)
1118 return KERN_INVALID_ARGUMENT;
1119
1120 uplflags |= (UPL_NO_SYNC|UPL_CLEAN_IN_PLACE|UPL_SET_INTERNAL);
1121 count = 0;
1122 kr = memory_object_upl_request(control, f_offset, bufsize,
1123 uplp, NULL, &count, uplflags);
1124 if (plp != NULL)
1125 *plp = UPL_GET_INTERNAL_PAGE_LIST(*uplp);
1126 return kr;
1127}
1128
1129
1130kern_return_t
1131ubc_upl_map(
1132 upl_t upl,
1133 vm_offset_t *dst_addr)
1134{
1135 return (vm_upl_map(kernel_map, upl, dst_addr));
1136}
1137
1138
1139kern_return_t
1140ubc_upl_unmap(
1141 upl_t upl)
1142{
1143 return(vm_upl_unmap(kernel_map, upl));
1144}
1145
1146kern_return_t
1147ubc_upl_commit(
1148 upl_t upl)
1149{
1150 upl_page_info_t *pl;
1151 kern_return_t kr;
1152
1153 pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
1154 kr = upl_commit(upl, pl, MAX_UPL_TRANSFER);
1155 upl_deallocate(upl);
1156 return kr;
1c79356b
A
1157}
1158
0b4e3aa0
A
1159
1160kern_return_t
1161ubc_upl_commit_range(
1162 upl_t upl,
1163 vm_offset_t offset,
1164 vm_size_t size,
1165 int flags)
1166{
1167 upl_page_info_t *pl;
1168 boolean_t empty;
1169 kern_return_t kr;
1170
1171 if (flags & UPL_COMMIT_FREE_ON_EMPTY)
1172 flags |= UPL_COMMIT_NOTIFY_EMPTY;
1173
1174 pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
1175
1176 kr = upl_commit_range(upl, offset, size, flags,
1177 pl, MAX_UPL_TRANSFER, &empty);
1178
1179 if((flags & UPL_COMMIT_FREE_ON_EMPTY) && empty)
1180 upl_deallocate(upl);
1181
1182 return kr;
1183}
1184
1185kern_return_t
1186ubc_upl_abort_range(
1187 upl_t upl,
1188 vm_offset_t offset,
1189 vm_size_t size,
1190 int abort_flags)
1191{
1192 kern_return_t kr;
1193 boolean_t empty = FALSE;
1194
1195 if (abort_flags & UPL_ABORT_FREE_ON_EMPTY)
1196 abort_flags |= UPL_ABORT_NOTIFY_EMPTY;
1197
1198 kr = upl_abort_range(upl, offset, size, abort_flags, &empty);
1199
1200 if((abort_flags & UPL_ABORT_FREE_ON_EMPTY) && empty)
1201 upl_deallocate(upl);
1202
1203 return kr;
1204}
1205
1206kern_return_t
1207ubc_upl_abort(
1208 upl_t upl,
1209 int abort_type)
1210{
1211 kern_return_t kr;
1212
1213 kr = upl_abort(upl, abort_type);
1214 upl_deallocate(upl);
1215 return kr;
1216}
1217
1218upl_page_info_t *
1219ubc_upl_pageinfo(
1220 upl_t upl)
1221{
1222 return (UPL_GET_INTERNAL_PAGE_LIST(upl));
1223}