Home | History | Annotate | Line # | Download | only in kern
vfs_trans.c revision 1.34.2.2
      1 /*	$NetBSD: vfs_trans.c,v 1.34.2.2 2017/04/26 02:53:27 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 1.34.2.2 2017/04/26 02:53:27 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_rele(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);
    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_ref(mp);
    209 	vfs_unbusy(mp);
    210 
    211 	return 0;
    212 }
    213 
    214 /*
    215  * Deallocate mount state.
    216  */
    217 void
    218 fstrans_unmount(struct mount *mp)
    219 {
    220 
    221 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    222 		return;
    223 
    224 	KASSERT(mp->mnt_transinfo != NULL);
    225 
    226 	fstrans_mount_dtor(mp);
    227 }
    228 
    229 /*
    230  * Retrieve the per lwp info for this mount allocating if necessary.
    231  */
    232 static struct fstrans_lwp_info *
    233 fstrans_get_lwp_info(struct mount *mp, bool do_alloc)
    234 {
    235 	struct fstrans_lwp_info *fli, *res;
    236 	struct fstrans_mount_info *fmi;
    237 
    238 	/*
    239 	 * Scan our list for a match clearing entries whose mount is gone.
    240 	 */
    241 	res = NULL;
    242 	for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
    243 		if (fli->fli_mount == mp) {
    244 			KASSERT(res == NULL);
    245 			res = fli;
    246 		} else if (fli->fli_mount != NULL &&
    247 		    (fli->fli_mount->mnt_iflag & IMNT_GONE) != 0 &&
    248 		    fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
    249 			fstrans_mount_dtor(fli->fli_mount);
    250 			fli->fli_mount = NULL;
    251 		}
    252 	}
    253 	if (__predict_true(res != NULL))
    254 		return res;
    255 
    256 	if (! do_alloc)
    257 		return NULL;
    258 
    259 	/*
    260 	 * Try to reuse a cleared entry or allocate a new one.
    261 	 */
    262 	for (fli = lwp_getspecific(lwp_data_key); fli; fli = fli->fli_succ) {
    263 		if (fli->fli_mount == NULL) {
    264 			KASSERT(fli->fli_trans_cnt == 0);
    265 			KASSERT(fli->fli_cow_cnt == 0);
    266 			break;
    267 		}
    268 	}
    269 	if (fli == NULL) {
    270 		mutex_enter(&fstrans_lock);
    271 		LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    272 			if (fli->fli_self == NULL) {
    273 				KASSERT(fli->fli_mount == NULL);
    274 				KASSERT(fli->fli_trans_cnt == 0);
    275 				KASSERT(fli->fli_cow_cnt == 0);
    276 				fli->fli_self = curlwp;
    277 				fli->fli_succ = lwp_getspecific(lwp_data_key);
    278 				lwp_setspecific(lwp_data_key, fli);
    279 				break;
    280 			}
    281 		}
    282 		mutex_exit(&fstrans_lock);
    283 	}
    284 	if (fli == NULL) {
    285 		fli = kmem_alloc(sizeof(*fli), KM_SLEEP);
    286 		mutex_enter(&fstrans_lock);
    287 		memset(fli, 0, sizeof(*fli));
    288 		fli->fli_self = curlwp;
    289 		LIST_INSERT_HEAD(&fstrans_fli_head, fli, fli_list);
    290 		mutex_exit(&fstrans_lock);
    291 		fli->fli_succ = lwp_getspecific(lwp_data_key);
    292 		lwp_setspecific(lwp_data_key, fli);
    293 	}
    294 
    295 	/*
    296 	 * Attach the entry to the mount.
    297 	 */
    298 	mutex_enter(&fstrans_mount_lock);
    299 	fmi = mp->mnt_transinfo;
    300 	KASSERT(fmi != NULL);
    301 	fli->fli_mount = mp;
    302 	fmi->fmi_ref_cnt += 1;
    303 	mutex_exit(&fstrans_mount_lock);
    304 
    305 	return fli;
    306 }
    307 
    308 /*
    309  * Check if this lock type is granted at this state.
    310  */
    311 static bool
    312 grant_lock(const enum fstrans_state state, const enum fstrans_lock_type type)
    313 {
    314 
    315 	if (__predict_true(state == FSTRANS_NORMAL))
    316 		return true;
    317 	if (type == FSTRANS_EXCL)
    318 		return true;
    319 	if  (state == FSTRANS_SUSPENDING && type == FSTRANS_LAZY)
    320 		return true;
    321 
    322 	return false;
    323 }
    324 
    325 /*
    326  * Start a transaction.  If this thread already has a transaction on this
    327  * file system increment the reference counter.
    328  */
    329 int
    330 _fstrans_start(struct mount *mp, enum fstrans_lock_type lock_type, int wait)
    331 {
    332 	int s;
    333 	struct mount *lmp;
    334 	struct fstrans_lwp_info *fli;
    335 	struct fstrans_mount_info *fmi;
    336 
    337 	if ((lmp = fstrans_normalize_mount(mp)) == NULL)
    338 		return 0;
    339 
    340 	ASSERT_SLEEPABLE();
    341 
    342 	/*
    343 	 * Allocate per lwp info for layered file systems to
    344 	 * get a reference to the mount.  No need to increment
    345 	 * the reference counter here.
    346 	 */
    347 	for (lmp = mp; lmp->mnt_lower; lmp = lmp->mnt_lower) {
    348 		fli = fstrans_get_lwp_info(lmp, true);
    349 		KASSERT(fli != NULL);
    350 	}
    351 
    352 	if ((fli = fstrans_get_lwp_info(lmp, true)) == NULL)
    353 		return 0;
    354 
    355 	if (fli->fli_trans_cnt > 0) {
    356 		KASSERT(lock_type != FSTRANS_EXCL);
    357 		fli->fli_trans_cnt += 1;
    358 
    359 		return 0;
    360 	}
    361 
    362 	s = pserialize_read_enter();
    363 	fmi = lmp->mnt_transinfo;
    364 	if (__predict_true(grant_lock(fmi->fmi_state, lock_type))) {
    365 		fli->fli_trans_cnt = 1;
    366 		fli->fli_lock_type = lock_type;
    367 		pserialize_read_exit(s);
    368 
    369 		return 0;
    370 	}
    371 	pserialize_read_exit(s);
    372 
    373 	if (! wait)
    374 		return EBUSY;
    375 
    376 	mutex_enter(&fstrans_lock);
    377 	while (! grant_lock(fmi->fmi_state, lock_type))
    378 		cv_wait(&fstrans_state_cv, &fstrans_lock);
    379 	fli->fli_trans_cnt = 1;
    380 	fli->fli_lock_type = lock_type;
    381 	mutex_exit(&fstrans_lock);
    382 
    383 	return 0;
    384 }
    385 
    386 /*
    387  * Finish a transaction.
    388  */
    389 void
    390 fstrans_done(struct mount *mp)
    391 {
    392 	int s;
    393 	struct fstrans_lwp_info *fli;
    394 	struct fstrans_mount_info *fmi;
    395 
    396 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    397 		return;
    398 	fli = fstrans_get_lwp_info(mp, false);
    399 	KASSERT(fli != NULL);
    400 	KASSERT(fli->fli_trans_cnt > 0);
    401 
    402 	if (fli->fli_trans_cnt > 1) {
    403 		fli->fli_trans_cnt -= 1;
    404 
    405 		return;
    406 	}
    407 
    408 	s = pserialize_read_enter();
    409 	fmi = mp->mnt_transinfo;
    410 	if (__predict_true(fmi->fmi_state == FSTRANS_NORMAL)) {
    411 		fli->fli_trans_cnt = 0;
    412 		pserialize_read_exit(s);
    413 
    414 		return;
    415 	}
    416 	pserialize_read_exit(s);
    417 
    418 	mutex_enter(&fstrans_lock);
    419 	fli->fli_trans_cnt = 0;
    420 	cv_signal(&fstrans_count_cv);
    421 	mutex_exit(&fstrans_lock);
    422 }
    423 
    424 /*
    425  * Check if this thread has an exclusive lock.
    426  */
    427 int
    428 fstrans_is_owner(struct mount *mp)
    429 {
    430 	struct fstrans_lwp_info *fli;
    431 
    432 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    433 		return 0;
    434 	if ((fli = fstrans_get_lwp_info(mp, false)) == NULL)
    435 		return 0;
    436 
    437 	if (fli->fli_trans_cnt == 0)
    438 		return 0;
    439 
    440 	KASSERT(fli->fli_mount == mp);
    441 	KASSERT(fli->fli_trans_cnt > 0);
    442 
    443 	return (fli->fli_lock_type == FSTRANS_EXCL);
    444 }
    445 
    446 /*
    447  * True, if no thread is in a transaction not granted at the current state.
    448  */
    449 static bool
    450 state_change_done(const struct mount *mp)
    451 {
    452 	struct fstrans_lwp_info *fli;
    453 	struct fstrans_mount_info *fmi;
    454 
    455 	KASSERT(mutex_owned(&fstrans_lock));
    456 
    457 	fmi = mp->mnt_transinfo;
    458 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    459 		if (fli->fli_mount != mp)
    460 			continue;
    461 		if (fli->fli_trans_cnt == 0)
    462 			continue;
    463 		if (grant_lock(fmi->fmi_state, fli->fli_lock_type))
    464 			continue;
    465 
    466 		return false;
    467 	}
    468 
    469 	return true;
    470 }
    471 
    472 /*
    473  * Set new file system state.
    474  */
    475 int
    476 fstrans_setstate(struct mount *mp, enum fstrans_state new_state)
    477 {
    478 	int error;
    479 	enum fstrans_state old_state;
    480 	struct fstrans_mount_info *fmi;
    481 
    482 	fmi = mp->mnt_transinfo;
    483 	old_state = fmi->fmi_state;
    484 	if (old_state == new_state)
    485 		return 0;
    486 
    487 	mutex_enter(&fstrans_lock);
    488 	fmi->fmi_state = new_state;
    489 	pserialize_perform(fstrans_psz);
    490 
    491 	/*
    492 	 * All threads see the new state now.
    493 	 * Wait for transactions invalid at this state to leave.
    494 	 */
    495 	error = 0;
    496 	while (! state_change_done(mp)) {
    497 		error = cv_wait_sig(&fstrans_count_cv, &fstrans_lock);
    498 		if (error) {
    499 			new_state = fmi->fmi_state = FSTRANS_NORMAL;
    500 			break;
    501 		}
    502 	}
    503 	cv_broadcast(&fstrans_state_cv);
    504 	mutex_exit(&fstrans_lock);
    505 
    506 	if (old_state != new_state) {
    507 		if (old_state == FSTRANS_NORMAL)
    508 			fstrans_start(mp, FSTRANS_EXCL);
    509 		if (new_state == FSTRANS_NORMAL)
    510 			fstrans_done(mp);
    511 	}
    512 
    513 	return error;
    514 }
    515 
    516 /*
    517  * Get current file system state.
    518  */
    519 enum fstrans_state
    520 fstrans_getstate(struct mount *mp)
    521 {
    522 	struct fstrans_mount_info *fmi;
    523 
    524 	fmi = mp->mnt_transinfo;
    525 	KASSERT(fmi != NULL);
    526 
    527 	return fmi->fmi_state;
    528 }
    529 
    530 /*
    531  * Request a filesystem to suspend all operations.
    532  */
    533 int
    534 vfs_suspend(struct mount *mp, int nowait)
    535 {
    536 	int error;
    537 
    538 	if ((mp = fstrans_normalize_mount(mp)) == NULL)
    539 		return EOPNOTSUPP;
    540 	if (nowait) {
    541 		if (!mutex_tryenter(&vfs_suspend_lock))
    542 			return EWOULDBLOCK;
    543 	} else
    544 		mutex_enter(&vfs_suspend_lock);
    545 
    546 	if ((error = VFS_SUSPENDCTL(mp, SUSPEND_SUSPEND)) != 0)
    547 		mutex_exit(&vfs_suspend_lock);
    548 
    549 	return error;
    550 }
    551 
    552 /*
    553  * Request a filesystem to resume all operations.
    554  */
    555 void
    556 vfs_resume(struct mount *mp)
    557 {
    558 
    559 	mp = fstrans_normalize_mount(mp);
    560 	KASSERT(mp != NULL);
    561 
    562 	VFS_SUSPENDCTL(mp, SUSPEND_RESUME);
    563 	mutex_exit(&vfs_suspend_lock);
    564 }
    565 
    566 
    567 /*
    568  * True, if no thread is running a cow handler.
    569  */
    570 static bool
    571 cow_state_change_done(const struct mount *mp)
    572 {
    573 	struct fstrans_lwp_info *fli;
    574 	struct fstrans_mount_info *fmi __diagused;
    575 
    576 	fmi = mp->mnt_transinfo;
    577 
    578 	KASSERT(mutex_owned(&fstrans_lock));
    579 	KASSERT(fmi->fmi_cow_change);
    580 
    581 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    582 		if (fli->fli_mount != mp)
    583 			continue;
    584 		if (fli->fli_cow_cnt == 0)
    585 			continue;
    586 
    587 		return false;
    588 	}
    589 
    590 	return true;
    591 }
    592 
    593 /*
    594  * Prepare for changing this mounts cow list.
    595  * Returns with fstrans_lock locked.
    596  */
    597 static void
    598 cow_change_enter(const struct mount *mp)
    599 {
    600 	struct fstrans_mount_info *fmi;
    601 
    602 	fmi = mp->mnt_transinfo;
    603 
    604 	mutex_enter(&fstrans_lock);
    605 
    606 	/*
    607 	 * Wait for other threads changing the list.
    608 	 */
    609 	while (fmi->fmi_cow_change)
    610 		cv_wait(&fstrans_state_cv, &fstrans_lock);
    611 
    612 	/*
    613 	 * Wait until all threads are aware of a state change.
    614 	 */
    615 	fmi->fmi_cow_change = true;
    616 	pserialize_perform(fstrans_psz);
    617 
    618 	while (! cow_state_change_done(mp))
    619 		cv_wait(&fstrans_count_cv, &fstrans_lock);
    620 }
    621 
    622 /*
    623  * Done changing this mounts cow list.
    624  */
    625 static void
    626 cow_change_done(const struct mount *mp)
    627 {
    628 	struct fstrans_mount_info *fmi;
    629 
    630 	KASSERT(mutex_owned(&fstrans_lock));
    631 
    632 	fmi = mp->mnt_transinfo;
    633 
    634 	fmi->fmi_cow_change = false;
    635 	pserialize_perform(fstrans_psz);
    636 
    637 	cv_broadcast(&fstrans_state_cv);
    638 
    639 	mutex_exit(&fstrans_lock);
    640 }
    641 
    642 /*
    643  * Add a handler to this mount.
    644  */
    645 int
    646 fscow_establish(struct mount *mp, int (*func)(void *, struct buf *, bool),
    647     void *arg)
    648 {
    649 	struct fstrans_mount_info *fmi;
    650 	struct fscow_handler *newch;
    651 
    652 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    653 		return EINVAL;
    654 
    655 	fmi = mp->mnt_transinfo;
    656 	KASSERT(fmi != NULL);
    657 
    658 	newch = kmem_alloc(sizeof(*newch), KM_SLEEP);
    659 	newch->ch_func = func;
    660 	newch->ch_arg = arg;
    661 
    662 	cow_change_enter(mp);
    663 	LIST_INSERT_HEAD(&fmi->fmi_cow_handler, newch, ch_list);
    664 	cow_change_done(mp);
    665 
    666 	return 0;
    667 }
    668 
    669 /*
    670  * Remove a handler from this mount.
    671  */
    672 int
    673 fscow_disestablish(struct mount *mp, int (*func)(void *, struct buf *, bool),
    674     void *arg)
    675 {
    676 	struct fstrans_mount_info *fmi;
    677 	struct fscow_handler *hp = NULL;
    678 
    679 	if ((mp->mnt_iflag & IMNT_HAS_TRANS) == 0)
    680 		return EINVAL;
    681 
    682 	fmi = mp->mnt_transinfo;
    683 	KASSERT(fmi != NULL);
    684 
    685 	cow_change_enter(mp);
    686 	LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
    687 		if (hp->ch_func == func && hp->ch_arg == arg)
    688 			break;
    689 	if (hp != NULL) {
    690 		LIST_REMOVE(hp, ch_list);
    691 		kmem_free(hp, sizeof(*hp));
    692 	}
    693 	cow_change_done(mp);
    694 
    695 	return hp ? 0 : EINVAL;
    696 }
    697 
    698 /*
    699  * Check for need to copy block that is about to be written.
    700  */
    701 int
    702 fscow_run(struct buf *bp, bool data_valid)
    703 {
    704 	int error, s;
    705 	struct mount *mp;
    706 	struct fstrans_lwp_info *fli;
    707 	struct fstrans_mount_info *fmi;
    708 	struct fscow_handler *hp;
    709 
    710 	/*
    711 	 * First check if we need run the copy-on-write handler.
    712 	 */
    713 	if ((bp->b_flags & B_COWDONE))
    714 		return 0;
    715 	if (bp->b_vp == NULL) {
    716 		bp->b_flags |= B_COWDONE;
    717 		return 0;
    718 	}
    719 	if (bp->b_vp->v_type == VBLK)
    720 		mp = spec_node_getmountedfs(bp->b_vp);
    721 	else
    722 		mp = bp->b_vp->v_mount;
    723 	if (mp == NULL || (mp->mnt_iflag & IMNT_HAS_TRANS) == 0) {
    724 		bp->b_flags |= B_COWDONE;
    725 		return 0;
    726 	}
    727 
    728 	fli = fstrans_get_lwp_info(mp, true);
    729 	fmi = mp->mnt_transinfo;
    730 
    731 	/*
    732 	 * On non-recursed run check if other threads
    733 	 * want to change the list.
    734 	 */
    735 	if (fli->fli_cow_cnt == 0) {
    736 		s = pserialize_read_enter();
    737 		if (__predict_false(fmi->fmi_cow_change)) {
    738 			pserialize_read_exit(s);
    739 			mutex_enter(&fstrans_lock);
    740 			while (fmi->fmi_cow_change)
    741 				cv_wait(&fstrans_state_cv, &fstrans_lock);
    742 			fli->fli_cow_cnt = 1;
    743 			mutex_exit(&fstrans_lock);
    744 		} else {
    745 			fli->fli_cow_cnt = 1;
    746 			pserialize_read_exit(s);
    747 		}
    748 	} else
    749 		fli->fli_cow_cnt += 1;
    750 
    751 	/*
    752 	 * Run all copy-on-write handlers, stop on error.
    753 	 */
    754 	error = 0;
    755 	LIST_FOREACH(hp, &fmi->fmi_cow_handler, ch_list)
    756 		if ((error = (*hp->ch_func)(hp->ch_arg, bp, data_valid)) != 0)
    757 			break;
    758  	if (error == 0)
    759  		bp->b_flags |= B_COWDONE;
    760 
    761 	/*
    762 	 * Check if other threads want to change the list.
    763 	 */
    764 	if (fli->fli_cow_cnt > 1) {
    765 		fli->fli_cow_cnt -= 1;
    766 	} else {
    767 		s = pserialize_read_enter();
    768 		if (__predict_false(fmi->fmi_cow_change)) {
    769 			pserialize_read_exit(s);
    770 			mutex_enter(&fstrans_lock);
    771 			fli->fli_cow_cnt = 0;
    772 			cv_signal(&fstrans_count_cv);
    773 			mutex_exit(&fstrans_lock);
    774 		} else {
    775 			fli->fli_cow_cnt = 0;
    776 			pserialize_read_exit(s);
    777 		}
    778 	}
    779 
    780 	return error;
    781 }
    782 
    783 #if defined(DDB)
    784 void fstrans_dump(int);
    785 
    786 static void
    787 fstrans_print_lwp(struct proc *p, struct lwp *l, int verbose)
    788 {
    789 	char prefix[9];
    790 	struct fstrans_lwp_info *fli;
    791 
    792 	snprintf(prefix, sizeof(prefix), "%d.%d", p->p_pid, l->l_lid);
    793 	LIST_FOREACH(fli, &fstrans_fli_head, fli_list) {
    794 		if (fli->fli_self != l)
    795 			continue;
    796 		if (fli->fli_trans_cnt == 0 && fli->fli_cow_cnt == 0) {
    797 			if (! verbose)
    798 				continue;
    799 		}
    800 		printf("%-8s", prefix);
    801 		if (verbose)
    802 			printf(" @%p", fli);
    803 		if (fli->fli_mount != NULL)
    804 			printf(" (%s)", fli->fli_mount->mnt_stat.f_mntonname);
    805 		else
    806 			printf(" NULL");
    807 		if (fli->fli_trans_cnt == 0) {
    808 			printf(" -");
    809 		} else {
    810 			switch (fli->fli_lock_type) {
    811 			case FSTRANS_LAZY:
    812 				printf(" lazy");
    813 				break;
    814 			case FSTRANS_SHARED:
    815 				printf(" shared");
    816 				break;
    817 			case FSTRANS_EXCL:
    818 				printf(" excl");
    819 				break;
    820 			default:
    821 				printf(" %#x", fli->fli_lock_type);
    822 				break;
    823 			}
    824 		}
    825 		printf(" %d cow %d\n", fli->fli_trans_cnt, fli->fli_cow_cnt);
    826 		prefix[0] = '\0';
    827 	}
    828 }
    829 
    830 static void
    831 fstrans_print_mount(struct mount *mp, int verbose)
    832 {
    833 	struct fstrans_mount_info *fmi;
    834 
    835 	fmi = mp->mnt_transinfo;
    836 	if (!verbose && (fmi == NULL || fmi->fmi_state == FSTRANS_NORMAL))
    837 		return;
    838 
    839 	printf("%-16s ", mp->mnt_stat.f_mntonname);
    840 	if (fmi == NULL) {
    841 		printf("(null)\n");
    842 		return;
    843 	}
    844 	switch (fmi->fmi_state) {
    845 	case FSTRANS_NORMAL:
    846 		printf("state normal\n");
    847 		break;
    848 	case FSTRANS_SUSPENDING:
    849 		printf("state suspending\n");
    850 		break;
    851 	case FSTRANS_SUSPENDED:
    852 		printf("state suspended\n");
    853 		break;
    854 	default:
    855 		printf("state %#x\n", fmi->fmi_state);
    856 		break;
    857 	}
    858 }
    859 
    860 void
    861 fstrans_dump(int full)
    862 {
    863 	const struct proclist_desc *pd;
    864 	struct proc *p;
    865 	struct lwp *l;
    866 	struct mount *mp;
    867 
    868 	printf("Fstrans locks by lwp:\n");
    869 	for (pd = proclists; pd->pd_list != NULL; pd++)
    870 		PROCLIST_FOREACH(p, pd->pd_list)
    871 			LIST_FOREACH(l, &p->p_lwps, l_sibling)
    872 				fstrans_print_lwp(p, l, full == 1);
    873 
    874 	printf("Fstrans state by mount:\n");
    875 	for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp))
    876 		fstrans_print_mount(mp, full == 1);
    877 }
    878 #endif /* defined(DDB) */
    879