Home | History | Annotate | Line # | Download | only in kern
vfs_trans.c revision 1.34.2.1
      1 /*	$NetBSD: vfs_trans.c,v 1.34.2.1 2017/03/20 06:57:48 pgoyette 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 1.34.2.1 2017/03/20 06:57:48 pgoyette 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 #define _FSTRANS_API_PRIVATE
     52 #include <sys/fstrans.h>
     53 #include <sys/proc.h>
     54 
     55 #include <miscfs/specfs/specdev.h>
     56 
     57 struct fscow_handler {
     58 	LIST_ENTRY(fscow_handler) ch_list;
     59 	int (*ch_func)(void *, struct buf *, bool);
     60 	void *ch_arg;
     61 };
     62 struct fstrans_lwp_info {
     63 	struct fstrans_lwp_info *fli_succ;
     64 	struct lwp *fli_self;
     65 	struct mount *fli_mount;
     66 	int fli_trans_cnt;
     67 	int fli_cow_cnt;
     68 	enum fstrans_lock_type fli_lock_type;
     69 	LIST_ENTRY(fstrans_lwp_info) fli_list;
     70 };
     71 struct fstrans_mount_info {
     72 	enum fstrans_state fmi_state;
     73 	unsigned int fmi_ref_cnt;
     74 	bool fmi_cow_change;
     75 	LIST_HEAD(, fscow_handler) fmi_cow_handler;
     76 };
     77 
     78 static specificdata_key_t lwp_data_key;	/* Our specific data key. */
     79 static kmutex_t vfs_suspend_lock;	/* Serialize suspensions. */
     80 static kmutex_t fstrans_lock;		/* Fstrans big lock. */
     81 static kmutex_t fstrans_mount_lock;	/* Fstrans mount big lock. */
     82 static kcondvar_t fstrans_state_cv;	/* Fstrans or cow state changed. */
     83 static kcondvar_t fstrans_count_cv;	/* Fstrans or cow count changed. */
     84 static pserialize_t fstrans_psz;	/* Pserialize state. */
     85 static LIST_HEAD(fstrans_lwp_head, fstrans_lwp_info) fstrans_fli_head;
     86 					/* List of all fstrans_lwp_info. */
     87 
     88 static inline struct mount *fstrans_normalize_mount(struct mount *);
     89 static void fstrans_lwp_dtor(void *);
     90 static void fstrans_mount_dtor(struct mount *);
     91 static struct fstrans_lwp_info *fstrans_get_lwp_info(struct mount *, bool);
     92 static bool grant_lock(const enum fstrans_state, const enum fstrans_lock_type);
     93 static bool state_change_done(const struct mount *);
     94 static bool cow_state_change_done(const struct mount *);
     95 static void cow_change_enter(const struct mount *);
     96 static void cow_change_done(const struct mount *);
     97 
     98 /*
     99  * Initialize.
    100  */
    101 void
    102 fstrans_init(void)
    103 {
    104 	int error __diagused;
    105 
    106 	error = lwp_specific_key_create(&lwp_data_key, fstrans_lwp_dtor);
    107 	KASSERT(error == 0);
    108 
    109 	mutex_init(&vfs_suspend_lock, MUTEX_DEFAULT, IPL_NONE);
    110 	mutex_init(&fstrans_lock, MUTEX_DEFAULT, IPL_NONE);
    111 	mutex_init(&fstrans_mount_lock, MUTEX_DEFAULT, IPL_NONE);
    112 	cv_init(&fstrans_state_cv, "fstchg");
    113 	cv_init(&fstrans_count_cv, "fstcnt");
    114 	fstrans_psz = pserialize_create();
    115 	LIST_INIT(&fstrans_fli_head);
    116 }
    117 
    118 /*
    119  * Normalize mount.
    120  * Return mount if file system supports fstrans, NULL otherwise.
    121  */
    122 static inline struct mount *
    123 fstrans_normalize_mount(struct mount *mp)
    124 {
    125 
    126 	while (mp && mp->mnt_lower)
    127 		mp = mp->mnt_lower;
    128 	if (mp == NULL)
    129 		return NULL;
    130 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    131 		return NULL;
    132 	return mp;
    133 }
    134 
    135 /*
    136  * Deallocate lwp state.
    137  */
    138 static void
    139 fstrans_lwp_dtor(void *arg)
    140 {
    141 	struct fstrans_lwp_info *fli, *fli_next;
    142 
    143 	for (fli = arg; fli; fli = fli_next) {
    144 		KASSERT(fli->fli_trans_cnt == 0);
    145 		KASSERT(fli->fli_cow_cnt == 0);
    146 		if (fli->fli_mount != NULL)
    147 			fstrans_mount_dtor(fli->fli_mount);
    148 		fli_next = fli->fli_succ;
    149 		fli->fli_mount = NULL;
    150 		membar_sync();
    151 		fli->fli_self = NULL;
    152 	}
    153 }
    154 
    155 /*
    156  * Dereference mount state.
    157  */
    158 static void
    159 fstrans_mount_dtor(struct mount *mp)
    160 {
    161 	struct fstrans_mount_info *fmi;
    162 
    163 	mutex_enter(&fstrans_mount_lock);
    164 
    165 	fmi = mp->mnt_transinfo;
    166 	KASSERT(fmi != NULL);
    167 	fmi->fmi_ref_cnt -= 1;
    168 	if (fmi->fmi_ref_cnt > 0) {
    169 		mutex_exit(&fstrans_mount_lock);
    170 		return;
    171 	}
    172 
    173 	KASSERT(fmi->fmi_state == FSTRANS_NORMAL);
    174 	KASSERT(LIST_FIRST(&fmi->fmi_cow_handler) == NULL);
    175 
    176 	mp->mnt_iflag &= ~IMNT_HAS_TRANS;
    177 	mp->mnt_transinfo = NULL;
    178 
    179 	mutex_exit(&fstrans_mount_lock);
    180 
    181 	kmem_free(fmi, sizeof(*fmi));
    182 	vfs_destroy(mp);
    183 }
    184 
    185 /*
    186  * Allocate mount state.
    187  */
    188 int
    189 fstrans_mount(struct mount *mp)
    190 {
    191 	int error;
    192 	struct fstrans_mount_info *newfmi;
    193 
    194 	error = vfs_busy(mp, NULL);
    195 	if (error)
    196 		return error;
    197 	newfmi = kmem_alloc(sizeof(*newfmi), KM_SLEEP);
    198 	newfmi->fmi_state = FSTRANS_NORMAL;
    199 	newfmi->fmi_ref_cnt = 1;
    200 	LIST_INIT(&newfmi->fmi_cow_handler);
    201 	newfmi->fmi_cow_change = false;
    202 
    203 	mutex_enter(&fstrans_mount_lock);
    204 	mp->mnt_transinfo = newfmi;
    205 	mp->mnt_iflag |= IMNT_HAS_TRANS;
    206 	mutex_exit(&fstrans_mount_lock);
    207 
    208 	vfs_unbusy(mp, true, NULL);
    209 
    210 	return 0;
    211 }
    212 
    213 /*
    214  * Deallocate mount state.
    215  */
    216 void
    217 fstrans_unmount(struct mount *mp)
    218 {
    219 
    220 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    221 		return;
    222 
    223 	KASSERT(mp->mnt_transinfo != NULL);
    224 
    225 	fstrans_mount_dtor(mp);
    226 }
    227 
    228 /*
    229  * Retrieve the per lwp info for this mount allocating if necessary.
    230  */
    231 static struct fstrans_lwp_info *
    232 fstrans_get_lwp_info(struct mount *mp, bool do_alloc)
    233 {
    234 	struct fstrans_lwp_info *fli, *res;
    235 	struct fstrans_mount_info *fmi;
    236 
    237 	/*
    238 	 * Scan our list for a match clearing entries whose mount is gone.
    239 	 */
    240 	res = NULL;
    241 	for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
    242 		if (fli->fli_mount == mp) {
    243 			KASSERT(res == NULL);
    244 			res = fli;
    245 		} else if (fli->fli_mount != NULL &&
    246 		    (fli->fli_mount->mnt_iflag & IMNT_GONE) != 0 &&
    247 		    fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
    248 			fstrans_mount_dtor(fli->fli_mount);
    249 			fli->fli_mount = NULL;
    250 		}
    251 	}
    252 	if (__predict_true(res != NULL))
    253 		return res;
    254 
    255 	if (! do_alloc)
    256 		return NULL;
    257 
    258 	/*
    259 	 * Try to reuse a cleared entry or allocate a new one.
    260 	 */
    261 	for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
    262 		if (fli->fli_mount == NULL) {
    263 			KASSERT(fli->fli_trans_cnt == 0);
    264 			KASSERT(fli->fli_cow_cnt == 0);
    265 			break;
    266 		}
    267 	}
    268 	if (fli == NULL) {
    269 		mutex_enter(&fstrans_lock);
    270 		LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    271 			if (fli->fli_self == NULL) {
    272 				KASSERT(fli->fli_mount == NULL);
    273 				KASSERT(fli->fli_trans_cnt == 0);
    274 				KASSERT(fli->fli_cow_cnt == 0);
    275 				fli->fli_self = curlwp;
    276 				fli->fli_succ = lwp_getspecific(lwp_data_key);
    277 				lwp_setspecific(lwp_data_key, fli);
    278 				break;
    279 			}
    280 		}
    281 		mutex_exit(&fstrans_lock);
    282 	}
    283 	if (fli == NULL) {
    284 		fli = kmem_alloc(sizeof(*fli), KM_SLEEP);
    285 		mutex_enter(&fstrans_lock);
    286 		memset(fli, 0, sizeof(*fli));
    287 		fli->fli_self = curlwp;
    288 		LIST_INSERT_HEAD(&fstrans_fli_head, fli, fli_list);
    289 		mutex_exit(&fstrans_lock);
    290 		fli->fli_succ = lwp_getspecific(lwp_data_key);
    291 		lwp_setspecific(lwp_data_key, fli);
    292 	}
    293 
    294 	/*
    295 	 * Attach the entry to the mount.
    296 	 */
    297 	mutex_enter(&fstrans_mount_lock);
    298 	fmi = mp->mnt_transinfo;
    299 	KASSERT(fmi != NULL);
    300 	fli->fli_mount = mp;
    301 	fmi->fmi_ref_cnt += 1;
    302 	mutex_exit(&fstrans_mount_lock);
    303 
    304 	return fli;
    305 }
    306 
    307 /*
    308  * Check if this lock type is granted at this state.
    309  */
    310 static bool
    311 grant_lock(const enum fstrans_state state, const enum fstrans_lock_type type)
    312 {
    313 
    314 	if (__predict_true(state == FSTRANS_NORMAL))
    315 		return true;
    316 	if (type == FSTRANS_EXCL)
    317 		return true;
    318 	if  (state == FSTRANS_SUSPENDING && type == FSTRANS_LAZY)
    319 		return true;
    320 
    321 	return false;
    322 }
    323 
    324 /*
    325  * Start a transaction.  If this thread already has a transaction on this
    326  * file system increment the reference counter.
    327  */
    328 int
    329 _fstrans_start(struct mount *mp, enum fstrans_lock_type lock_type, int wait)
    330 {
    331 	int s;
    332 	struct fstrans_lwp_info *fli;
    333 	struct fstrans_mount_info *fmi;
    334 
    335 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    336 		return 0;
    337 
    338 	ASSERT_SLEEPABLE();
    339 
    340 	if ((fli = fstrans_get_lwp_info(mp, true)) == NULL)
    341 		return 0;
    342 
    343 	if (fli->fli_trans_cnt > 0) {
    344 		KASSERT(lock_type != FSTRANS_EXCL);
    345 		fli->fli_trans_cnt += 1;
    346 
    347 		return 0;
    348 	}
    349 
    350 	s = pserialize_read_enter();
    351 	fmi = mp->mnt_transinfo;
    352 	if (__predict_true(grant_lock(fmi->fmi_state, lock_type))) {
    353 		fli->fli_trans_cnt = 1;
    354 		fli->fli_lock_type = lock_type;
    355 		pserialize_read_exit(s);
    356 
    357 		return 0;
    358 	}
    359 	pserialize_read_exit(s);
    360 
    361 	if (! wait)
    362 		return EBUSY;
    363 
    364 	mutex_enter(&fstrans_lock);
    365 	while (! grant_lock(fmi->fmi_state, lock_type))
    366 		cv_wait(&fstrans_state_cv, &fstrans_lock);
    367 	fli->fli_trans_cnt = 1;
    368 	fli->fli_lock_type = lock_type;
    369 	mutex_exit(&fstrans_lock);
    370 
    371 	return 0;
    372 }
    373 
    374 /*
    375  * Finish a transaction.
    376  */
    377 void
    378 fstrans_done(struct mount *mp)
    379 {
    380 	int s;
    381 	struct fstrans_lwp_info *fli;
    382 	struct fstrans_mount_info *fmi;
    383 
    384 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    385 		return;
    386 	if ((fli = fstrans_get_lwp_info(mp, true)) == NULL)
    387 		return;
    388 
    389 	KASSERT(fli->fli_trans_cnt > 0);
    390 
    391 	if (fli->fli_trans_cnt > 1) {
    392 		fli->fli_trans_cnt -= 1;
    393 
    394 		return;
    395 	}
    396 
    397 	s = pserialize_read_enter();
    398 	fmi = mp->mnt_transinfo;
    399 	if (__predict_true(fmi->fmi_state == FSTRANS_NORMAL)) {
    400 		fli->fli_trans_cnt = 0;
    401 		pserialize_read_exit(s);
    402 
    403 		return;
    404 	}
    405 	pserialize_read_exit(s);
    406 
    407 	mutex_enter(&fstrans_lock);
    408 	fli->fli_trans_cnt = 0;
    409 	cv_signal(&fstrans_count_cv);
    410 	mutex_exit(&fstrans_lock);
    411 }
    412 
    413 /*
    414  * Check if this thread has an exclusive lock.
    415  */
    416 int
    417 fstrans_is_owner(struct mount *mp)
    418 {
    419 	struct fstrans_lwp_info *fli;
    420 
    421 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    422 		return 0;
    423 	if ((fli = fstrans_get_lwp_info(mp, false)) == NULL)
    424 		return 0;
    425 
    426 	if (fli->fli_trans_cnt == 0)
    427 		return 0;
    428 
    429 	KASSERT(fli->fli_mount == mp);
    430 	KASSERT(fli->fli_trans_cnt > 0);
    431 
    432 	return (fli->fli_lock_type == FSTRANS_EXCL);
    433 }
    434 
    435 /*
    436  * True, if no thread is in a transaction not granted at the current state.
    437  */
    438 static bool
    439 state_change_done(const struct mount *mp)
    440 {
    441 	struct fstrans_lwp_info *fli;
    442 	struct fstrans_mount_info *fmi;
    443 
    444 	KASSERT(mutex_owned(&fstrans_lock));
    445 
    446 	fmi = mp->mnt_transinfo;
    447 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    448 		if (fli->fli_mount != mp)
    449 			continue;
    450 		if (fli->fli_trans_cnt == 0)
    451 			continue;
    452 		if (grant_lock(fmi->fmi_state, fli->fli_lock_type))
    453 			continue;
    454 
    455 		return false;
    456 	}
    457 
    458 	return true;
    459 }
    460 
    461 /*
    462  * Set new file system state.
    463  */
    464 int
    465 fstrans_setstate(struct mount *mp, enum fstrans_state new_state)
    466 {
    467 	int error;
    468 	enum fstrans_state old_state;
    469 	struct fstrans_mount_info *fmi;
    470 
    471 	fmi = mp->mnt_transinfo;
    472 	old_state = fmi->fmi_state;
    473 	if (old_state == new_state)
    474 		return 0;
    475 
    476 	mutex_enter(&fstrans_lock);
    477 	fmi->fmi_state = new_state;
    478 	pserialize_perform(fstrans_psz);
    479 
    480 	/*
    481 	 * All threads see the new state now.
    482 	 * Wait for transactions invalid at this state to leave.
    483 	 */
    484 	error = 0;
    485 	while (! state_change_done(mp)) {
    486 		error = cv_wait_sig(&fstrans_count_cv, &fstrans_lock);
    487 		if (error) {
    488 			new_state = fmi->fmi_state = FSTRANS_NORMAL;
    489 			break;
    490 		}
    491 	}
    492 	cv_broadcast(&fstrans_state_cv);
    493 	mutex_exit(&fstrans_lock);
    494 
    495 	if (old_state != new_state) {
    496 		if (old_state == FSTRANS_NORMAL)
    497 			fstrans_start(mp, FSTRANS_EXCL);
    498 		if (new_state == FSTRANS_NORMAL)
    499 			fstrans_done(mp);
    500 	}
    501 
    502 	return error;
    503 }
    504 
    505 /*
    506  * Get current file system state.
    507  */
    508 enum fstrans_state
    509 fstrans_getstate(struct mount *mp)
    510 {
    511 	struct fstrans_mount_info *fmi;
    512 
    513 	fmi = mp->mnt_transinfo;
    514 	KASSERT(fmi != NULL);
    515 
    516 	return fmi->fmi_state;
    517 }
    518 
    519 /*
    520  * Request a filesystem to suspend all operations.
    521  */
    522 int
    523 vfs_suspend(struct mount *mp, int nowait)
    524 {
    525 	int error;
    526 
    527 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    528 		return EOPNOTSUPP;
    529 	if (nowait) {
    530 		if (!mutex_tryenter(&vfs_suspend_lock))
    531 			return EWOULDBLOCK;
    532 	} else
    533 		mutex_enter(&vfs_suspend_lock);
    534 
    535 	if ((error = VFS_SUSPENDCTL(mp, SUSPEND_SUSPEND)) != 0)
    536 		mutex_exit(&vfs_suspend_lock);
    537 
    538 	return error;
    539 }
    540 
    541 /*
    542  * Request a filesystem to resume all operations.
    543  */
    544 void
    545 vfs_resume(struct mount *mp)
    546 {
    547 
    548 	mp = fstrans_normalize_mount(mp);
    549 	KASSERT(mp != NULL);
    550 
    551 	VFS_SUSPENDCTL(mp, SUSPEND_RESUME);
    552 	mutex_exit(&vfs_suspend_lock);
    553 }
    554 
    555 
    556 /*
    557  * True, if no thread is running a cow handler.
    558  */
    559 static bool
    560 cow_state_change_done(const struct mount *mp)
    561 {
    562 	struct fstrans_lwp_info *fli;
    563 	struct fstrans_mount_info *fmi __diagused;
    564 
    565 	fmi = mp->mnt_transinfo;
    566 
    567 	KASSERT(mutex_owned(&fstrans_lock));
    568 	KASSERT(fmi->fmi_cow_change);
    569 
    570 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    571 		if (fli->fli_mount != mp)
    572 			continue;
    573 		if (fli->fli_cow_cnt == 0)
    574 			continue;
    575 
    576 		return false;
    577 	}
    578 
    579 	return true;
    580 }
    581 
    582 /*
    583  * Prepare for changing this mounts cow list.
    584  * Returns with fstrans_lock locked.
    585  */
    586 static void
    587 cow_change_enter(const struct mount *mp)
    588 {
    589 	struct fstrans_mount_info *fmi;
    590 
    591 	fmi = mp->mnt_transinfo;
    592 
    593 	mutex_enter(&fstrans_lock);
    594 
    595 	/*
    596 	 * Wait for other threads changing the list.
    597 	 */
    598 	while (fmi->fmi_cow_change)
    599 		cv_wait(&fstrans_state_cv, &fstrans_lock);
    600 
    601 	/*
    602 	 * Wait until all threads are aware of a state change.
    603 	 */
    604 	fmi->fmi_cow_change = true;
    605 	pserialize_perform(fstrans_psz);
    606 
    607 	while (! cow_state_change_done(mp))
    608 		cv_wait(&fstrans_count_cv, &fstrans_lock);
    609 }
    610 
    611 /*
    612  * Done changing this mounts cow list.
    613  */
    614 static void
    615 cow_change_done(const struct mount *mp)
    616 {
    617 	struct fstrans_mount_info *fmi;
    618 
    619 	KASSERT(mutex_owned(&fstrans_lock));
    620 
    621 	fmi = mp->mnt_transinfo;
    622 
    623 	fmi->fmi_cow_change = false;
    624 	pserialize_perform(fstrans_psz);
    625 
    626 	cv_broadcast(&fstrans_state_cv);
    627 
    628 	mutex_exit(&fstrans_lock);
    629 }
    630 
    631 /*
    632  * Add a handler to this mount.
    633  */
    634 int
    635 fscow_establish(struct mount *mp, int (*func)(void *, struct buf *, bool),
    636     void *arg)
    637 {
    638 	struct fstrans_mount_info *fmi;
    639 	struct fscow_handler *newch;
    640 
    641 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    642 		return EINVAL;
    643 
    644 	fmi = mp->mnt_transinfo;
    645 	KASSERT(fmi != NULL);
    646 
    647 	newch = kmem_alloc(sizeof(*newch), KM_SLEEP);
    648 	newch->ch_func = func;
    649 	newch->ch_arg = arg;
    650 
    651 	cow_change_enter(mp);
    652 	LIST_INSERT_HEAD(&fmi->fmi_cow_handler, newch, ch_list);
    653 	cow_change_done(mp);
    654 
    655 	return 0;
    656 }
    657 
    658 /*
    659  * Remove a handler from this mount.
    660  */
    661 int
    662 fscow_disestablish(struct mount *mp, int (*func)(void *, struct buf *, bool),
    663     void *arg)
    664 {
    665 	struct fstrans_mount_info *fmi;
    666 	struct fscow_handler *hp = NULL;
    667 
    668 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    669 		return EINVAL;
    670 
    671 	fmi = mp->mnt_transinfo;
    672 	KASSERT(fmi != NULL);
    673 
    674 	cow_change_enter(mp);
    675 	LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
    676 		if (hp->ch_func == func && hp->ch_arg == arg)
    677 			break;
    678 	if (hp != NULL) {
    679 		LIST_REMOVE(hp, ch_list);
    680 		kmem_free(hp, sizeof(*hp));
    681 	}
    682 	cow_change_done(mp);
    683 
    684 	return hp ? 0 : EINVAL;
    685 }
    686 
    687 /*
    688  * Check for need to copy block that is about to be written.
    689  */
    690 int
    691 fscow_run(struct buf *bp, bool data_valid)
    692 {
    693 	int error, s;
    694 	struct mount *mp;
    695 	struct fstrans_lwp_info *fli;
    696 	struct fstrans_mount_info *fmi;
    697 	struct fscow_handler *hp;
    698 
    699 	/*
    700 	 * First check if we need run the copy-on-write handler.
    701 	 */
    702 	if ((bp->b_flags & B_COWDONE))
    703 		return 0;
    704 	if (bp->b_vp == NULL) {
    705 		bp->b_flags |= B_COWDONE;
    706 		return 0;
    707 	}
    708 	if (bp->b_vp->v_type == VBLK)
    709 		mp = spec_node_getmountedfs(bp->b_vp);
    710 	else
    711 		mp = bp->b_vp->v_mount;
    712 	if (mp == NULL || (mp->mnt_iflag & IMNT_HAS_TRANS) == 0) {
    713 		bp->b_flags |= B_COWDONE;
    714 		return 0;
    715 	}
    716 
    717 	fli = fstrans_get_lwp_info(mp, true);
    718 	fmi = mp->mnt_transinfo;
    719 
    720 	/*
    721 	 * On non-recursed run check if other threads
    722 	 * want to change the list.
    723 	 */
    724 	if (fli->fli_cow_cnt == 0) {
    725 		s = pserialize_read_enter();
    726 		if (__predict_false(fmi->fmi_cow_change)) {
    727 			pserialize_read_exit(s);
    728 			mutex_enter(&fstrans_lock);
    729 			while (fmi->fmi_cow_change)
    730 				cv_wait(&fstrans_state_cv, &fstrans_lock);
    731 			fli->fli_cow_cnt = 1;
    732 			mutex_exit(&fstrans_lock);
    733 		} else {
    734 			fli->fli_cow_cnt = 1;
    735 			pserialize_read_exit(s);
    736 		}
    737 	} else
    738 		fli->fli_cow_cnt += 1;
    739 
    740 	/*
    741 	 * Run all copy-on-write handlers, stop on error.
    742 	 */
    743 	error = 0;
    744 	LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
    745 		if ((error = (*hp->ch_func)(hp->ch_arg, bp, data_valid)) != 0)
    746 			break;
    747  	if (error == 0)
    748  		bp->b_flags |= B_COWDONE;
    749 
    750 	/*
    751 	 * Check if other threads want to change the list.
    752 	 */
    753 	if (fli->fli_cow_cnt > 1) {
    754 		fli->fli_cow_cnt -= 1;
    755 	} else {
    756 		s = pserialize_read_enter();
    757 		if (__predict_false(fmi->fmi_cow_change)) {
    758 			pserialize_read_exit(s);
    759 			mutex_enter(&fstrans_lock);
    760 			fli->fli_cow_cnt = 0;
    761 			cv_signal(&fstrans_count_cv);
    762 			mutex_exit(&fstrans_lock);
    763 		} else {
    764 			fli->fli_cow_cnt = 0;
    765 			pserialize_read_exit(s);
    766 		}
    767 	}
    768 
    769 	return error;
    770 }
    771 
    772 #if defined(DDB)
    773 void fstrans_dump(int);
    774 
    775 static void
    776 fstrans_print_lwp(struct proc *p, struct lwp *l, int verbose)
    777 {
    778 	char prefix[9];
    779 	struct fstrans_lwp_info *fli;
    780 
    781 	snprintf(prefix, sizeof(prefix), "%d.%d", p->p_pid, l->l_lid);
    782 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    783 		if (fli->fli_self != l)
    784 			continue;
    785 		if (fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
    786 			if (! verbose)
    787 				continue;
    788 		}
    789 		printf("%-8s", prefix);
    790 		if (verbose)
    791 			printf(" @%p", fli);
    792 		if (fli->fli_mount != NULL)
    793 			printf(" (%s)", fli->fli_mount->mnt_stat.f_mntonname);
    794 		else
    795 			printf(" NULL");
    796 		if (fli->fli_trans_cnt == 0) {
    797 			printf(" -");
    798 		} else {
    799 			switch (fli->fli_lock_type) {
    800 			case FSTRANS_LAZY:
    801 				printf(" lazy");
    802 				break;
    803 			case FSTRANS_SHARED:
    804 				printf(" shared");
    805 				break;
    806 			case FSTRANS_EXCL:
    807 				printf(" excl");
    808 				break;
    809 			default:
    810 				printf(" %#x", fli->fli_lock_type);
    811 				break;
    812 			}
    813 		}
    814 		printf(" %d cow %d\n", fli->fli_trans_cnt, fli->fli_cow_cnt);
    815 		prefix[0] = '\0';
    816 	}
    817 }
    818 
    819 static void
    820 fstrans_print_mount(struct mount *mp, int verbose)
    821 {
    822 	struct fstrans_mount_info *fmi;
    823 
    824 	fmi = mp->mnt_transinfo;
    825 	if (!verbose && (fmi == NULL || fmi->fmi_state == FSTRANS_NORMAL))
    826 		return;
    827 
    828 	printf("%-16s ", mp->mnt_stat.f_mntonname);
    829 	if (fmi == NULL) {
    830 		printf("(null)\n");
    831 		return;
    832 	}
    833 	switch (fmi->fmi_state) {
    834 	case FSTRANS_NORMAL:
    835 		printf("state normal\n");
    836 		break;
    837 	case FSTRANS_SUSPENDING:
    838 		printf("state suspending\n");
    839 		break;
    840 	case FSTRANS_SUSPENDED:
    841 		printf("state suspended\n");
    842 		break;
    843 	default:
    844 		printf("state %#x\n", fmi->fmi_state);
    845 		break;
    846 	}
    847 }
    848 
    849 void
    850 fstrans_dump(int full)
    851 {
    852 	const struct proclist_desc *pd;
    853 	struct proc *p;
    854 	struct lwp *l;
    855 	struct mount *mp;
    856 
    857 	printf("Fstrans locks by lwp:\n");
    858 	for (pd = proclists; pd->pd_list != NULL; pd++)
    859 		PROCLIST_FOREACH(p, pd->pd_list)
    860 			LIST_FOREACH(l, &p->p_lwps, l_sibling)
    861 				fstrans_print_lwp(p, l, full == 1);
    862 
    863 	printf("Fstrans state by mount:\n");
    864 	TAILQ_FOREACH(mp, &mountlist, mnt_list)
    865 		fstrans_print_mount(mp, full == 1);
    866 }
    867 #endif /* defined(DDB) */
    868