vfs_trans.c revision 1.45.2.2 1 /* $NetBSD: vfs_trans.c,v 1.45.2.2 2017/06/21 18:24:26 snj Exp $ */
2
3 /*-
4 * Copyright (c) 2007 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Juergen Hannken-Illjes.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: vfs_trans.c,v 1.45.2.2 2017/06/21 18:24:26 snj Exp $");
34
35 /*
36 * File system transaction operations.
37 */
38
39 #ifdef _KERNEL_OPT
40 #include "opt_ddb.h"
41 #endif
42
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/atomic.h>
46 #include <sys/buf.h>
47 #include <sys/kmem.h>
48 #include <sys/mount.h>
49 #include <sys/pserialize.h>
50 #include <sys/vnode.h>
51 #include <sys/fstrans.h>
52 #include <sys/proc.h>
53
54 #include <miscfs/specfs/specdev.h>
55
56 enum fstrans_lock_type {
57 FSTRANS_SHARED, /* Granted while not suspending */
58 FSTRANS_EXCL /* Internal: exclusive lock */
59 };
60
61 struct fscow_handler {
62 LIST_ENTRY(fscow_handler) ch_list;
63 int (*ch_func)(void *, struct buf *, bool);
64 void *ch_arg;
65 };
66 struct fstrans_lwp_info {
67 struct fstrans_lwp_info *fli_succ;
68 struct lwp *fli_self;
69 struct mount *fli_mount;
70 int fli_trans_cnt;
71 int fli_cow_cnt;
72 enum fstrans_lock_type fli_lock_type;
73 LIST_ENTRY(fstrans_lwp_info) fli_list;
74 };
75 struct fstrans_mount_info {
76 enum fstrans_state fmi_state;
77 unsigned int fmi_ref_cnt;
78 bool fmi_cow_change;
79 LIST_HEAD(, fscow_handler) fmi_cow_handler;
80 };
81
82 static specificdata_key_t lwp_data_key; /* Our specific data key. */
83 static kmutex_t vfs_suspend_lock; /* Serialize suspensions. */
84 static kmutex_t fstrans_lock; /* Fstrans big lock. */
85 static kmutex_t fstrans_mount_lock; /* Fstrans mount big lock. */
86 static kcondvar_t fstrans_state_cv; /* Fstrans or cow state changed. */
87 static kcondvar_t fstrans_count_cv; /* Fstrans or cow count changed. */
88 static pserialize_t fstrans_psz; /* Pserialize state. */
89 static LIST_HEAD(fstrans_lwp_head, fstrans_lwp_info) fstrans_fli_head;
90 /* List of all fstrans_lwp_info. */
91
92 static inline struct mount *fstrans_normalize_mount(struct mount *);
93 static void fstrans_lwp_dtor(void *);
94 static void fstrans_mount_dtor(struct mount *);
95 static void fstrans_clear_lwp_info(void);
96 static inline struct fstrans_lwp_info *
97 fstrans_get_lwp_info(struct mount *, bool);
98 static struct fstrans_lwp_info *fstrans_alloc_lwp_info(struct mount *);
99 static inline int _fstrans_start(struct mount *, enum fstrans_lock_type, int);
100 static bool grant_lock(const enum fstrans_state, const enum fstrans_lock_type);
101 static bool state_change_done(const struct mount *);
102 static bool cow_state_change_done(const struct mount *);
103 static void cow_change_enter(const struct mount *);
104 static void cow_change_done(const struct mount *);
105
106 /*
107 * Initialize.
108 */
109 void
110 fstrans_init(void)
111 {
112 int error __diagused;
113
114 error = lwp_specific_key_create(&lwp_data_key, fstrans_lwp_dtor);
115 KASSERT(error == 0);
116
117 mutex_init(&vfs_suspend_lock, MUTEX_DEFAULT, IPL_NONE);
118 mutex_init(&fstrans_lock, MUTEX_DEFAULT, IPL_NONE);
119 mutex_init(&fstrans_mount_lock, MUTEX_DEFAULT, IPL_NONE);
120 cv_init(&fstrans_state_cv, "fstchg");
121 cv_init(&fstrans_count_cv, "fstcnt");
122 fstrans_psz = pserialize_create();
123 LIST_INIT(&fstrans_fli_head);
124 }
125
126 /*
127 * Normalize mount.
128 * Return mount if file system supports fstrans, NULL otherwise.
129 */
130 static inline struct mount *
131 fstrans_normalize_mount(struct mount *mp)
132 {
133
134 while (mp && mp->mnt_lower)
135 mp = mp->mnt_lower;
136 if (mp == NULL)
137 return NULL;
138 if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
139 return NULL;
140 return mp;
141 }
142
143 /*
144 * Deallocate lwp state.
145 */
146 static void
147 fstrans_lwp_dtor(void *arg)
148 {
149 struct fstrans_lwp_info *fli, *fli_next;
150
151 for (fli = arg; fli; fli = fli_next) {
152 KASSERT(fli->fli_trans_cnt == 0);
153 KASSERT(fli->fli_cow_cnt == 0);
154 if (fli->fli_mount != NULL)
155 fstrans_mount_dtor(fli->fli_mount);
156 fli_next = fli->fli_succ;
157 fli->fli_mount = NULL;
158 membar_sync();
159 fli->fli_self = NULL;
160 }
161 }
162
163 /*
164 * Dereference mount state.
165 */
166 static void
167 fstrans_mount_dtor(struct mount *mp)
168 {
169 struct fstrans_mount_info *fmi;
170
171 mutex_enter(&fstrans_mount_lock);
172
173 fmi = mp->mnt_transinfo;
174 KASSERT(fmi != NULL);
175 fmi->fmi_ref_cnt -= 1;
176 if (fmi->fmi_ref_cnt > 0) {
177 mutex_exit(&fstrans_mount_lock);
178 return;
179 }
180
181 KASSERT(fmi->fmi_state == FSTRANS_NORMAL);
182 KASSERT(LIST_FIRST(&fmi->fmi_cow_handler) == NULL);
183
184 mp->mnt_iflag &= ~IMNT_HAS_TRANS;
185 mp->mnt_transinfo = NULL;
186
187 mutex_exit(&fstrans_mount_lock);
188
189 kmem_free(fmi, sizeof(*fmi));
190 vfs_rele(mp);
191 }
192
193 /*
194 * Allocate mount state.
195 */
196 int
197 fstrans_mount(struct mount *mp)
198 {
199 struct fstrans_mount_info *newfmi;
200
201 newfmi = kmem_alloc(sizeof(*newfmi), KM_SLEEP);
202 newfmi->fmi_state = FSTRANS_NORMAL;
203 newfmi->fmi_ref_cnt = 1;
204 LIST_INIT(&newfmi->fmi_cow_handler);
205 newfmi->fmi_cow_change = false;
206
207 mutex_enter(&fstrans_mount_lock);
208 mp->mnt_transinfo = newfmi;
209 mp->mnt_iflag |= IMNT_HAS_TRANS;
210 mutex_exit(&fstrans_mount_lock);
211
212 vfs_ref(mp);
213
214 return 0;
215 }
216
217 /*
218 * Deallocate mount state.
219 */
220 void
221 fstrans_unmount(struct mount *mp)
222 {
223
224 if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
225 return;
226
227 KASSERT(mp->mnt_transinfo != NULL);
228
229 fstrans_mount_dtor(mp);
230 }
231
232 /*
233 * Clear mount entries whose mount is gone.
234 */
235 static void
236 fstrans_clear_lwp_info(void)
237 {
238 struct fstrans_lwp_info *fli;
239
240 /*
241 * Scan our list clearing entries whose mount is gone.
242 */
243 for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
244 if (fli->fli_mount != NULL &&
245 (fli->fli_mount->mnt_iflag & IMNT_GONE) != 0 &&
246 fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
247 fstrans_mount_dtor(fli->fli_mount);
248 fli->fli_mount = NULL;
249 }
250 }
251 }
252
253 /*
254 * Allocate and return per lwp info for this mount.
255 */
256 static struct fstrans_lwp_info *
257 fstrans_alloc_lwp_info(struct mount *mp)
258 {
259 struct fstrans_lwp_info *fli;
260 struct fstrans_mount_info *fmi;
261
262 /*
263 * Try to reuse a cleared entry or allocate a new one.
264 */
265 for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
266 KASSERT(fli->fli_mount != mp);
267 if (fli->fli_mount == NULL) {
268 KASSERT(fli->fli_trans_cnt == 0);
269 KASSERT(fli->fli_cow_cnt == 0);
270 break;
271 }
272 }
273 if (fli == NULL) {
274 mutex_enter(&fstrans_lock);
275 LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
276 if (fli->fli_self == NULL) {
277 KASSERT(fli->fli_mount == NULL);
278 KASSERT(fli->fli_trans_cnt == 0);
279 KASSERT(fli->fli_cow_cnt == 0);
280 fli->fli_self = curlwp;
281 fli->fli_succ = lwp_getspecific(lwp_data_key);
282 lwp_setspecific(lwp_data_key, fli);
283 break;
284 }
285 }
286 mutex_exit(&fstrans_lock);
287 }
288 if (fli == NULL) {
289 fli = kmem_alloc(sizeof(*fli), KM_SLEEP);
290 mutex_enter(&fstrans_lock);
291 memset(fli, 0, sizeof(*fli));
292 fli->fli_self = curlwp;
293 LIST_INSERT_HEAD(&fstrans_fli_head, fli, fli_list);
294 mutex_exit(&fstrans_lock);
295 fli->fli_succ = lwp_getspecific(lwp_data_key);
296 lwp_setspecific(lwp_data_key, fli);
297 }
298
299 /*
300 * Attach the entry to the mount if its mnt_transinfo is valid.
301 */
302 mutex_enter(&fstrans_mount_lock);
303 fmi = mp->mnt_transinfo;
304 if (__predict_true(fmi != NULL)) {
305 fli->fli_mount = mp;
306 fmi->fmi_ref_cnt += 1;
307 } else {
308 fli = NULL;
309 }
310 mutex_exit(&fstrans_mount_lock);
311
312 return fli;
313 }
314
315 /*
316 * Retrieve the per lwp info for this mount allocating if necessary.
317 */
318 static inline struct fstrans_lwp_info *
319 fstrans_get_lwp_info(struct mount *mp, bool do_alloc)
320 {
321 struct fstrans_lwp_info *fli;
322
323 /*
324 * Scan our list for a match.
325 */
326 for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
327 if (fli->fli_mount == mp)
328 return fli;
329 }
330
331 return (do_alloc ? fstrans_alloc_lwp_info(mp) : NULL);
332 }
333
334 /*
335 * Check if this lock type is granted at this state.
336 */
337 static bool
338 grant_lock(const enum fstrans_state state, const enum fstrans_lock_type type)
339 {
340
341 if (__predict_true(state == FSTRANS_NORMAL))
342 return true;
343 if (type == FSTRANS_EXCL)
344 return true;
345
346 return false;
347 }
348
349 /*
350 * Start a transaction. If this thread already has a transaction on this
351 * file system increment the reference counter.
352 */
353 static inline int
354 _fstrans_start(struct mount *mp, enum fstrans_lock_type lock_type, int wait)
355 {
356 int s;
357 struct mount *lmp;
358 struct fstrans_lwp_info *fli;
359 struct fstrans_mount_info *fmi;
360
361 if ((lmp = fstrans_normalize_mount(mp)) == NULL)
362 return 0;
363
364 ASSERT_SLEEPABLE();
365
366 /*
367 * Allocate per lwp info for layered file systems to
368 * get a reference to the mount. No need to increment
369 * the reference counter here.
370 */
371 for (lmp = mp; lmp->mnt_lower; lmp = lmp->mnt_lower) {
372 fli = fstrans_get_lwp_info(lmp, true);
373 }
374
375 if ((fli = fstrans_get_lwp_info(lmp, true)) == NULL)
376 return 0;
377
378 if (fli->fli_trans_cnt > 0) {
379 KASSERT(lock_type != FSTRANS_EXCL);
380 fli->fli_trans_cnt += 1;
381
382 return 0;
383 }
384
385 s = pserialize_read_enter();
386 fmi = lmp->mnt_transinfo;
387 if (__predict_true(grant_lock(fmi->fmi_state, lock_type))) {
388 fli->fli_trans_cnt = 1;
389 fli->fli_lock_type = lock_type;
390 pserialize_read_exit(s);
391
392 return 0;
393 }
394 pserialize_read_exit(s);
395
396 if (! wait)
397 return EBUSY;
398
399 mutex_enter(&fstrans_lock);
400 while (! grant_lock(fmi->fmi_state, lock_type))
401 cv_wait(&fstrans_state_cv, &fstrans_lock);
402 fli->fli_trans_cnt = 1;
403 fli->fli_lock_type = lock_type;
404 mutex_exit(&fstrans_lock);
405
406 return 0;
407 }
408
409 void
410 fstrans_start(struct mount *mp)
411 {
412 int error __diagused;
413
414 error = _fstrans_start(mp, FSTRANS_SHARED, 1);
415 KASSERT(error == 0);
416 }
417
418 int
419 fstrans_start_nowait(struct mount *mp)
420 {
421
422 return _fstrans_start(mp, FSTRANS_SHARED, 0);
423 }
424
425 /*
426 * Finish a transaction.
427 */
428 void
429 fstrans_done(struct mount *mp)
430 {
431 int s;
432 struct fstrans_lwp_info *fli;
433 struct fstrans_mount_info *fmi;
434
435 if ((mp = fstrans_normalize_mount(mp)) == NULL)
436 return;
437 if ((fli = fstrans_get_lwp_info(mp, false)) == NULL)
438 return;
439 KASSERT(fli->fli_trans_cnt > 0);
440
441 if (fli->fli_trans_cnt > 1) {
442 fli->fli_trans_cnt -= 1;
443
444 return;
445 }
446
447 fstrans_clear_lwp_info();
448
449 s = pserialize_read_enter();
450 fmi = mp->mnt_transinfo;
451 if (__predict_true(fmi->fmi_state == FSTRANS_NORMAL)) {
452 fli->fli_trans_cnt = 0;
453 pserialize_read_exit(s);
454
455 return;
456 }
457 pserialize_read_exit(s);
458
459 mutex_enter(&fstrans_lock);
460 fli->fli_trans_cnt = 0;
461 cv_signal(&fstrans_count_cv);
462 mutex_exit(&fstrans_lock);
463 }
464
465 /*
466 * Check if this thread has an exclusive lock.
467 */
468 int
469 fstrans_is_owner(struct mount *mp)
470 {
471 struct fstrans_lwp_info *fli;
472
473 if ((mp = fstrans_normalize_mount(mp)) == NULL)
474 return 0;
475 if ((fli = fstrans_get_lwp_info(mp, false)) == NULL)
476 return 0;
477
478 if (fli->fli_trans_cnt == 0)
479 return 0;
480
481 KASSERT(fli->fli_mount == mp);
482 KASSERT(fli->fli_trans_cnt > 0);
483
484 return (fli->fli_lock_type == FSTRANS_EXCL);
485 }
486
487 /*
488 * True, if no thread is in a transaction not granted at the current state.
489 */
490 static bool
491 state_change_done(const struct mount *mp)
492 {
493 struct fstrans_lwp_info *fli;
494 struct fstrans_mount_info *fmi;
495
496 KASSERT(mutex_owned(&fstrans_lock));
497
498 fmi = mp->mnt_transinfo;
499 LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
500 if (fli->fli_mount != mp)
501 continue;
502 if (fli->fli_trans_cnt == 0)
503 continue;
504 if (grant_lock(fmi->fmi_state, fli->fli_lock_type))
505 continue;
506
507 return false;
508 }
509
510 return true;
511 }
512
513 /*
514 * Set new file system state.
515 */
516 int
517 fstrans_setstate(struct mount *mp, enum fstrans_state new_state)
518 {
519 int error;
520 enum fstrans_state old_state;
521 struct fstrans_mount_info *fmi;
522
523 fmi = mp->mnt_transinfo;
524 old_state = fmi->fmi_state;
525 if (old_state == new_state)
526 return 0;
527
528 mutex_enter(&fstrans_lock);
529 fmi->fmi_state = new_state;
530 pserialize_perform(fstrans_psz);
531
532 /*
533 * All threads see the new state now.
534 * Wait for transactions invalid at this state to leave.
535 */
536 error = 0;
537 while (! state_change_done(mp)) {
538 error = cv_wait_sig(&fstrans_count_cv, &fstrans_lock);
539 if (error) {
540 new_state = fmi->fmi_state = FSTRANS_NORMAL;
541 break;
542 }
543 }
544 cv_broadcast(&fstrans_state_cv);
545 mutex_exit(&fstrans_lock);
546
547 if (old_state != new_state) {
548 if (old_state == FSTRANS_NORMAL)
549 _fstrans_start(mp, FSTRANS_EXCL, 1);
550 if (new_state == FSTRANS_NORMAL)
551 fstrans_done(mp);
552 }
553
554 return error;
555 }
556
557 /*
558 * Get current file system state.
559 */
560 enum fstrans_state
561 fstrans_getstate(struct mount *mp)
562 {
563 struct fstrans_mount_info *fmi;
564
565 fmi = mp->mnt_transinfo;
566 KASSERT(fmi != NULL);
567
568 return fmi->fmi_state;
569 }
570
571 /*
572 * Request a filesystem to suspend all operations.
573 */
574 int
575 vfs_suspend(struct mount *mp, int nowait)
576 {
577 int error;
578
579 if ((mp = fstrans_normalize_mount(mp)) == NULL)
580 return EOPNOTSUPP;
581 if (nowait) {
582 if (!mutex_tryenter(&vfs_suspend_lock))
583 return EWOULDBLOCK;
584 } else
585 mutex_enter(&vfs_suspend_lock);
586
587 if ((error = VFS_SUSPENDCTL(mp, SUSPEND_SUSPEND)) != 0)
588 mutex_exit(&vfs_suspend_lock);
589
590 return error;
591 }
592
593 /*
594 * Request a filesystem to resume all operations.
595 */
596 void
597 vfs_resume(struct mount *mp)
598 {
599
600 mp = fstrans_normalize_mount(mp);
601 KASSERT(mp != NULL);
602
603 VFS_SUSPENDCTL(mp, SUSPEND_RESUME);
604 mutex_exit(&vfs_suspend_lock);
605 }
606
607
608 /*
609 * True, if no thread is running a cow handler.
610 */
611 static bool
612 cow_state_change_done(const struct mount *mp)
613 {
614 struct fstrans_lwp_info *fli;
615 struct fstrans_mount_info *fmi __diagused;
616
617 fmi = mp->mnt_transinfo;
618
619 KASSERT(mutex_owned(&fstrans_lock));
620 KASSERT(fmi->fmi_cow_change);
621
622 LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
623 if (fli->fli_mount != mp)
624 continue;
625 if (fli->fli_cow_cnt == 0)
626 continue;
627
628 return false;
629 }
630
631 return true;
632 }
633
634 /*
635 * Prepare for changing this mounts cow list.
636 * Returns with fstrans_lock locked.
637 */
638 static void
639 cow_change_enter(const struct mount *mp)
640 {
641 struct fstrans_mount_info *fmi;
642
643 fmi = mp->mnt_transinfo;
644
645 mutex_enter(&fstrans_lock);
646
647 /*
648 * Wait for other threads changing the list.
649 */
650 while (fmi->fmi_cow_change)
651 cv_wait(&fstrans_state_cv, &fstrans_lock);
652
653 /*
654 * Wait until all threads are aware of a state change.
655 */
656 fmi->fmi_cow_change = true;
657 pserialize_perform(fstrans_psz);
658
659 while (! cow_state_change_done(mp))
660 cv_wait(&fstrans_count_cv, &fstrans_lock);
661 }
662
663 /*
664 * Done changing this mounts cow list.
665 */
666 static void
667 cow_change_done(const struct mount *mp)
668 {
669 struct fstrans_mount_info *fmi;
670
671 KASSERT(mutex_owned(&fstrans_lock));
672
673 fmi = mp->mnt_transinfo;
674
675 fmi->fmi_cow_change = false;
676 pserialize_perform(fstrans_psz);
677
678 cv_broadcast(&fstrans_state_cv);
679
680 mutex_exit(&fstrans_lock);
681 }
682
683 /*
684 * Add a handler to this mount.
685 */
686 int
687 fscow_establish(struct mount *mp, int (*func)(void *, struct buf *, bool),
688 void *arg)
689 {
690 struct fstrans_mount_info *fmi;
691 struct fscow_handler *newch;
692
693 if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
694 return EINVAL;
695
696 fmi = mp->mnt_transinfo;
697 KASSERT(fmi != NULL);
698
699 newch = kmem_alloc(sizeof(*newch), KM_SLEEP);
700 newch->ch_func = func;
701 newch->ch_arg = arg;
702
703 cow_change_enter(mp);
704 LIST_INSERT_HEAD(&fmi->fmi_cow_handler, newch, ch_list);
705 cow_change_done(mp);
706
707 return 0;
708 }
709
710 /*
711 * Remove a handler from this mount.
712 */
713 int
714 fscow_disestablish(struct mount *mp, int (*func)(void *, struct buf *, bool),
715 void *arg)
716 {
717 struct fstrans_mount_info *fmi;
718 struct fscow_handler *hp = NULL;
719
720 if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
721 return EINVAL;
722
723 fmi = mp->mnt_transinfo;
724 KASSERT(fmi != NULL);
725
726 cow_change_enter(mp);
727 LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
728 if (hp->ch_func == func && hp->ch_arg == arg)
729 break;
730 if (hp != NULL) {
731 LIST_REMOVE(hp, ch_list);
732 kmem_free(hp, sizeof(*hp));
733 }
734 cow_change_done(mp);
735
736 return hp ? 0 : EINVAL;
737 }
738
739 /*
740 * Check for need to copy block that is about to be written.
741 */
742 int
743 fscow_run(struct buf *bp, bool data_valid)
744 {
745 int error, s;
746 struct mount *mp;
747 struct fstrans_lwp_info *fli;
748 struct fstrans_mount_info *fmi;
749 struct fscow_handler *hp;
750
751 /*
752 * First check if we need run the copy-on-write handler.
753 */
754 if ((bp->b_flags & B_COWDONE))
755 return 0;
756 if (bp->b_vp == NULL) {
757 bp->b_flags |= B_COWDONE;
758 return 0;
759 }
760 if (bp->b_vp->v_type == VBLK)
761 mp = spec_node_getmountedfs(bp->b_vp);
762 else
763 mp = bp->b_vp->v_mount;
764 if (mp == NULL || (mp->mnt_iflag & IMNT_HAS_TRANS) == 0) {
765 bp->b_flags |= B_COWDONE;
766 return 0;
767 }
768
769 fli = fstrans_get_lwp_info(mp, true);
770 fmi = mp->mnt_transinfo;
771
772 /*
773 * On non-recursed run check if other threads
774 * want to change the list.
775 */
776 if (fli->fli_cow_cnt == 0) {
777 s = pserialize_read_enter();
778 if (__predict_false(fmi->fmi_cow_change)) {
779 pserialize_read_exit(s);
780 mutex_enter(&fstrans_lock);
781 while (fmi->fmi_cow_change)
782 cv_wait(&fstrans_state_cv, &fstrans_lock);
783 fli->fli_cow_cnt = 1;
784 mutex_exit(&fstrans_lock);
785 } else {
786 fli->fli_cow_cnt = 1;
787 pserialize_read_exit(s);
788 }
789 } else
790 fli->fli_cow_cnt += 1;
791
792 /*
793 * Run all copy-on-write handlers, stop on error.
794 */
795 error = 0;
796 LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
797 if ((error = (*hp->ch_func)(hp->ch_arg, bp, data_valid)) != 0)
798 break;
799 if (error == 0)
800 bp->b_flags |= B_COWDONE;
801
802 /*
803 * Check if other threads want to change the list.
804 */
805 if (fli->fli_cow_cnt > 1) {
806 fli->fli_cow_cnt -= 1;
807 } else {
808 s = pserialize_read_enter();
809 if (__predict_false(fmi->fmi_cow_change)) {
810 pserialize_read_exit(s);
811 mutex_enter(&fstrans_lock);
812 fli->fli_cow_cnt = 0;
813 cv_signal(&fstrans_count_cv);
814 mutex_exit(&fstrans_lock);
815 } else {
816 fli->fli_cow_cnt = 0;
817 pserialize_read_exit(s);
818 }
819 }
820
821 return error;
822 }
823
824 #if defined(DDB)
825 void fstrans_dump(int);
826
827 static void
828 fstrans_print_lwp(struct proc *p, struct lwp *l, int verbose)
829 {
830 char prefix[9];
831 struct fstrans_lwp_info *fli;
832
833 snprintf(prefix, sizeof(prefix), "%d.%d", p->p_pid, l->l_lid);
834 LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
835 if (fli->fli_self != l)
836 continue;
837 if (fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
838 if (! verbose)
839 continue;
840 }
841 printf("%-8s", prefix);
842 if (verbose)
843 printf(" @%p", fli);
844 if (fli->fli_mount != NULL)
845 printf(" (%s)", fli->fli_mount->mnt_stat.f_mntonname);
846 else
847 printf(" NULL");
848 if (fli->fli_trans_cnt == 0) {
849 printf(" -");
850 } else {
851 switch (fli->fli_lock_type) {
852 case FSTRANS_SHARED:
853 printf(" shared");
854 break;
855 case FSTRANS_EXCL:
856 printf(" excl");
857 break;
858 default:
859 printf(" %#x", fli->fli_lock_type);
860 break;
861 }
862 }
863 printf(" %d cow %d\n", fli->fli_trans_cnt, fli->fli_cow_cnt);
864 prefix[0] = '\0';
865 }
866 }
867
868 static void
869 fstrans_print_mount(struct mount *mp, int verbose)
870 {
871 struct fstrans_mount_info *fmi;
872
873 fmi = mp->mnt_transinfo;
874 if (!verbose && (fmi == NULL || fmi->fmi_state == FSTRANS_NORMAL))
875 return;
876
877 printf("%-16s ", mp->mnt_stat.f_mntonname);
878 if (fmi == NULL) {
879 printf("(null)\n");
880 return;
881 }
882 switch (fmi->fmi_state) {
883 case FSTRANS_NORMAL:
884 printf("state normal\n");
885 break;
886 case FSTRANS_SUSPENDED:
887 printf("state suspended\n");
888 break;
889 default:
890 printf("state %#x\n", fmi->fmi_state);
891 break;
892 }
893 }
894
895 void
896 fstrans_dump(int full)
897 {
898 const struct proclist_desc *pd;
899 struct proc *p;
900 struct lwp *l;
901 struct mount *mp;
902
903 printf("Fstrans locks by lwp:\n");
904 for (pd = proclists; pd->pd_list != NULL; pd++)
905 PROCLIST_FOREACH(p, pd->pd_list)
906 LIST_FOREACH(l, &p->p_lwps, l_sibling)
907 fstrans_print_lwp(p, l, full == 1);
908
909 printf("Fstrans state by mount:\n");
910 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp))
911 fstrans_print_mount(mp, full == 1);
912 }
913 #endif /* defined(DDB) */
914