xref: /netbsd-src/sys/dev/fss.c (revision 5b84b3983f71fd20a534cfa5d1556623a8aaa717)
1 /*	$NetBSD: fss.c,v 1.15 2005/04/17 16:28:26 hannken Exp $	*/
2 
3 /*-
4  * Copyright (c) 2003 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * File system snapshot disk driver.
41  *
42  * Block/character interface to the snapshot of a mounted file system.
43  */
44 
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.15 2005/04/17 16:28:26 hannken Exp $");
47 
48 #include "fss.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/namei.h>
53 #include <sys/proc.h>
54 #include <sys/errno.h>
55 #include <sys/buf.h>
56 #include <sys/malloc.h>
57 #include <sys/ioctl.h>
58 #include <sys/disklabel.h>
59 #include <sys/device.h>
60 #include <sys/disk.h>
61 #include <sys/stat.h>
62 #include <sys/mount.h>
63 #include <sys/vnode.h>
64 #include <sys/file.h>
65 #include <sys/uio.h>
66 #include <sys/conf.h>
67 #include <sys/kthread.h>
68 
69 #include <miscfs/specfs/specdev.h>
70 
71 #include <dev/fssvar.h>
72 
73 #include <machine/stdarg.h>
74 
75 #ifdef DEBUG
76 #define FSS_STATISTICS
77 #endif
78 
79 #ifdef FSS_STATISTICS
80 struct fss_stat {
81 	u_int64_t	cow_calls;
82 	u_int64_t	cow_copied;
83 	u_int64_t	cow_cache_full;
84 	u_int64_t	indir_read;
85 	u_int64_t	indir_write;
86 };
87 
88 static struct fss_stat fss_stat[NFSS];
89 
90 #define FSS_STAT_INC(sc, field)	\
91 			do { \
92 				fss_stat[sc->sc_unit].field++; \
93 			} while (0)
94 #define FSS_STAT_SET(sc, field, value) \
95 			do { \
96 				fss_stat[sc->sc_unit].field = value; \
97 			} while (0)
98 #define FSS_STAT_ADD(sc, field, value) \
99 			do { \
100 				fss_stat[sc->sc_unit].field += value; \
101 			} while (0)
102 #define FSS_STAT_VAL(sc, field) fss_stat[sc->sc_unit].field
103 #define FSS_STAT_CLEAR(sc) \
104 			do { \
105 				memset(&fss_stat[sc->sc_unit], 0, \
106 				    sizeof(struct fss_stat)); \
107 			} while (0)
108 #else /* FSS_STATISTICS */
109 #define FSS_STAT_INC(sc, field)
110 #define FSS_STAT_SET(sc, field, value)
111 #define FSS_STAT_ADD(sc, field, value)
112 #define FSS_STAT_CLEAR(sc)
113 #endif /* FSS_STATISTICS */
114 
115 static struct fss_softc fss_softc[NFSS];
116 
117 void fssattach(int);
118 
119 dev_type_open(fss_open);
120 dev_type_close(fss_close);
121 dev_type_read(fss_read);
122 dev_type_write(fss_write);
123 dev_type_ioctl(fss_ioctl);
124 dev_type_strategy(fss_strategy);
125 dev_type_dump(fss_dump);
126 dev_type_size(fss_size);
127 
128 static int fss_copy_on_write(void *, struct buf *);
129 static inline void fss_error(struct fss_softc *, const char *, ...);
130 static int fss_create_files(struct fss_softc *, struct fss_set *,
131     off_t *, struct proc *);
132 static int fss_create_snapshot(struct fss_softc *, struct fss_set *,
133     struct proc *);
134 static int fss_delete_snapshot(struct fss_softc *, struct proc *);
135 static int fss_softc_alloc(struct fss_softc *);
136 static void fss_softc_free(struct fss_softc *);
137 static void fss_cluster_iodone(struct buf *);
138 static void fss_read_cluster(struct fss_softc *, u_int32_t);
139 static int fss_write_cluster(struct fss_cache *, u_int32_t);
140 static void fss_bs_thread(void *);
141 static int fss_bmap(struct fss_softc *, off_t, int,
142     struct vnode **, daddr_t *, int *);
143 static int fss_bs_io(struct fss_softc *, fss_io_type,
144     u_int32_t, off_t, int, caddr_t);
145 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t);
146 
147 const struct bdevsw fss_bdevsw = {
148 	fss_open, fss_close, fss_strategy, fss_ioctl,
149 	fss_dump, fss_size, D_DISK
150 };
151 
152 const struct cdevsw fss_cdevsw = {
153 	fss_open, fss_close, fss_read, fss_write, fss_ioctl,
154 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK
155 };
156 
157 void
158 fssattach(int num)
159 {
160 	int i;
161 	struct fss_softc *sc;
162 
163 	for (i = 0; i < NFSS; i++) {
164 		sc = &fss_softc[i];
165 		sc->sc_unit = i;
166 		sc->sc_bdev = NODEV;
167 		simple_lock_init(&sc->sc_slock);
168 		bufq_alloc(&sc->sc_bufq, BUFQ_FCFS|BUFQ_SORT_RAWBLOCK);
169 	}
170 }
171 
172 int
173 fss_open(dev_t dev, int flags, int mode, struct proc *p)
174 {
175 	int s, mflag;
176 	struct fss_softc *sc;
177 
178 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
179 
180 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
181 		return ENODEV;
182 
183 	FSS_LOCK(sc, s);
184 
185 	sc->sc_flags |= mflag;
186 
187 	FSS_UNLOCK(sc, s);
188 
189 	return 0;
190 }
191 
192 int
193 fss_close(dev_t dev, int flags, int mode, struct proc *p)
194 {
195 	int s, mflag, error;
196 	struct fss_softc *sc;
197 
198 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
199 
200 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
201 		return ENODEV;
202 
203 	FSS_LOCK(sc, s);
204 
205 	if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag) {
206 		if ((sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0 &&
207 		    (sc->sc_flags & FSS_ACTIVE) != 0) {
208 			FSS_UNLOCK(sc, s);
209 			error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, p);
210 			if (error)
211 				return error;
212 			FSS_LOCK(sc, s);
213 		}
214 		sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE;
215 	}
216 
217 	sc->sc_flags &= ~mflag;
218 
219 	FSS_UNLOCK(sc, s);
220 
221 	return 0;
222 }
223 
224 void
225 fss_strategy(struct buf *bp)
226 {
227 	int s;
228 	struct fss_softc *sc;
229 
230 	sc = FSS_DEV_TO_SOFTC(bp->b_dev);
231 
232 	FSS_LOCK(sc, s);
233 
234 	if ((bp->b_flags & B_READ) != B_READ ||
235 	    sc == NULL || !FSS_ISVALID(sc)) {
236 
237 		FSS_UNLOCK(sc, s);
238 
239 		bp->b_error = (sc == NULL ? ENODEV : EROFS);
240 		bp->b_flags |= B_ERROR;
241 		bp->b_resid = bp->b_bcount;
242 		biodone(bp);
243 		return;
244 	}
245 
246 	bp->b_rawblkno = bp->b_blkno;
247 	BUFQ_PUT(&sc->sc_bufq, bp);
248 	wakeup(&sc->sc_bs_proc);
249 
250 	FSS_UNLOCK(sc, s);
251 }
252 
253 int
254 fss_read(dev_t dev, struct uio *uio, int flags)
255 {
256 	return physio(fss_strategy, NULL, dev, B_READ, minphys, uio);
257 }
258 
259 int
260 fss_write(dev_t dev, struct uio *uio, int flags)
261 {
262 	return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio);
263 }
264 
265 int
266 fss_ioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
267 {
268 	int s, error;
269 	struct fss_softc *sc;
270 	struct fss_set *fss = (struct fss_set *)data;
271 	struct fss_get *fsg = (struct fss_get *)data;
272 
273 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
274 		return ENODEV;
275 
276 	FSS_LOCK(sc, s);
277 	while ((sc->sc_flags & FSS_EXCL) == FSS_EXCL) {
278 		error = ltsleep(sc, PRIBIO|PCATCH, "fsslock", 0, &sc->sc_slock);
279 		if (error) {
280 			FSS_UNLOCK(sc, s);
281 			return error;
282 		}
283 	}
284 	sc->sc_flags |= FSS_EXCL;
285 	FSS_UNLOCK(sc, s);
286 
287 	switch (cmd) {
288 	case FSSIOCSET:
289 		if ((flag & FWRITE) == 0)
290 			error = EPERM;
291 		else if ((sc->sc_flags & FSS_ACTIVE) != 0)
292 			error = EBUSY;
293 		else
294 			error = fss_create_snapshot(sc, fss, p);
295 		break;
296 
297 	case FSSIOCCLR:
298 		if ((flag & FWRITE) == 0)
299 			error = EPERM;
300 		else if ((sc->sc_flags & FSS_ACTIVE) == 0)
301 			error = ENXIO;
302 		else
303 			error = fss_delete_snapshot(sc, p);
304 		break;
305 
306 	case FSSIOCGET:
307 		switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) {
308 		case FSS_ACTIVE:
309 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
310 			fsg->fsg_csize = FSS_CLSIZE(sc);
311 			fsg->fsg_time = sc->sc_time;
312 			fsg->fsg_mount_size = sc->sc_clcount;
313 			fsg->fsg_bs_size = sc->sc_clnext;
314 			error = 0;
315 			break;
316 		case FSS_PERSISTENT | FSS_ACTIVE:
317 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
318 			fsg->fsg_csize = 0;
319 			fsg->fsg_time = sc->sc_time;
320 			fsg->fsg_mount_size = 0;
321 			fsg->fsg_bs_size = 0;
322 			error = 0;
323 			break;
324 		default:
325 			error = ENXIO;
326 			break;
327 		}
328 		break;
329 
330 	case FSSIOFSET:
331 		sc->sc_uflags = *(int *)data;
332 		error = 0;
333 		break;
334 
335 	case FSSIOFGET:
336 		*(int *)data = sc->sc_uflags;
337 		error = 0;
338 		break;
339 
340 	default:
341 		error = EINVAL;
342 		break;
343 	}
344 
345 	FSS_LOCK(sc, s);
346 	sc->sc_flags &= ~FSS_EXCL;
347 	FSS_UNLOCK(sc, s);
348 	wakeup(sc);
349 
350 	return error;
351 }
352 
353 int
354 fss_size(dev_t dev)
355 {
356 	return -1;
357 }
358 
359 int
360 fss_dump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
361 {
362 	return EROFS;
363 }
364 
365 /*
366  * An error occurred reading or writing the snapshot or backing store.
367  * If it is the first error log to console.
368  * The caller holds the simplelock.
369  */
370 static inline void
371 fss_error(struct fss_softc *sc, const char *fmt, ...)
372 {
373 	va_list ap;
374 
375 	if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) {
376 		va_start(ap, fmt);
377 		printf("fss%d: snapshot invalid: ", sc->sc_unit);
378 		vprintf(fmt, ap);
379 		printf("\n");
380 		va_end(ap);
381 	}
382 	if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE)
383 		sc->sc_flags |= FSS_ERROR;
384 }
385 
386 /*
387  * Allocate the variable sized parts of the softc and
388  * fork the kernel thread.
389  *
390  * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size
391  * must be initialized.
392  */
393 static int
394 fss_softc_alloc(struct fss_softc *sc)
395 {
396 	int i, len, error;
397 
398 	len = (sc->sc_clcount+NBBY-1)/NBBY;
399 	sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
400 	if (sc->sc_copied == NULL)
401 		return(ENOMEM);
402 
403 	len = sc->sc_cache_size*sizeof(struct fss_cache);
404 	sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
405 	if (sc->sc_cache == NULL)
406 		return(ENOMEM);
407 
408 	len = FSS_CLSIZE(sc);
409 	for (i = 0; i < sc->sc_cache_size; i++) {
410 		sc->sc_cache[i].fc_type = FSS_CACHE_FREE;
411 		sc->sc_cache[i].fc_softc = sc;
412 		sc->sc_cache[i].fc_xfercount = 0;
413 		sc->sc_cache[i].fc_data = malloc(len, M_TEMP,
414 		    M_WAITOK|M_CANFAIL);
415 		if (sc->sc_cache[i].fc_data == NULL)
416 			return(ENOMEM);
417 	}
418 
419 	len = (sc->sc_indir_size+NBBY-1)/NBBY;
420 	sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
421 	if (sc->sc_indir_valid == NULL)
422 		return(ENOMEM);
423 
424 	len = FSS_CLSIZE(sc);
425 	sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
426 	if (sc->sc_indir_data == NULL)
427 		return(ENOMEM);
428 
429 	if ((error = kthread_create1(fss_bs_thread, sc, &sc->sc_bs_proc,
430 	    "fssbs%d", sc->sc_unit)) != 0)
431 		return error;
432 
433 	sc->sc_flags |= FSS_BS_THREAD;
434 	return 0;
435 }
436 
437 /*
438  * Free the variable sized parts of the softc.
439  */
440 static void
441 fss_softc_free(struct fss_softc *sc)
442 {
443 	int s, i;
444 
445 	if ((sc->sc_flags & FSS_BS_THREAD) != 0) {
446 		FSS_LOCK(sc, s);
447 		sc->sc_flags &= ~FSS_BS_THREAD;
448 		wakeup(&sc->sc_bs_proc);
449 		while (sc->sc_bs_proc != NULL)
450 			ltsleep(&sc->sc_bs_proc, PRIBIO, "fssthread", 0,
451 			    &sc->sc_slock);
452 		FSS_UNLOCK(sc, s);
453 	}
454 
455 	if (sc->sc_copied != NULL)
456 		free(sc->sc_copied, M_TEMP);
457 	sc->sc_copied = NULL;
458 
459 	if (sc->sc_cache != NULL) {
460 		for (i = 0; i < sc->sc_cache_size; i++)
461 			if (sc->sc_cache[i].fc_data != NULL)
462 				free(sc->sc_cache[i].fc_data, M_TEMP);
463 		free(sc->sc_cache, M_TEMP);
464 	}
465 	sc->sc_cache = NULL;
466 
467 	if (sc->sc_indir_valid != NULL)
468 		free(sc->sc_indir_valid, M_TEMP);
469 	sc->sc_indir_valid = NULL;
470 
471 	if (sc->sc_indir_data != NULL)
472 		free(sc->sc_indir_data, M_TEMP);
473 	sc->sc_indir_data = NULL;
474 }
475 
476 /*
477  * Check if an unmount is ok. If forced, set this snapshot into ERROR state.
478  */
479 int
480 fss_umount_hook(struct mount *mp, int forced)
481 {
482 	int i, s;
483 
484 	for (i = 0; i < NFSS; i++) {
485 		FSS_LOCK(&fss_softc[i], s);
486 		if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 &&
487 		    fss_softc[i].sc_mount == mp) {
488 			if (forced)
489 				fss_error(&fss_softc[i], "forced unmount");
490 			else {
491 				FSS_UNLOCK(&fss_softc[i], s);
492 				return EBUSY;
493 			}
494 		}
495 		FSS_UNLOCK(&fss_softc[i], s);
496 	}
497 
498 	return 0;
499 }
500 
501 /*
502  * A buffer is written to the snapshotted block device. Copy to
503  * backing store if needed.
504  */
505 static int
506 fss_copy_on_write(void *v, struct buf *bp)
507 {
508 	int s;
509 	u_int32_t cl, ch, c;
510 	struct fss_softc *sc = v;
511 
512 	FSS_LOCK(sc, s);
513 	if (!FSS_ISVALID(sc)) {
514 		FSS_UNLOCK(sc, s);
515 		return 0;
516 	}
517 
518 	FSS_UNLOCK(sc, s);
519 
520 	FSS_STAT_INC(sc, cow_calls);
521 
522 	cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
523 	ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
524 
525 	for (c = cl; c <= ch; c++)
526 		fss_read_cluster(sc, c);
527 
528 	return 0;
529 }
530 
531 /*
532  * Lookup and open needed files.
533  *
534  * For file system internal snapshot initializes sc_mntname, sc_mount,
535  * sc_bs_vp and sc_time.
536  *
537  * Otherwise returns dev and size of the underlying block device.
538  * Initializes sc_mntname, sc_mount_vp, sc_bdev, sc_bs_vp and sc_mount
539  */
540 static int
541 fss_create_files(struct fss_softc *sc, struct fss_set *fss,
542     off_t *bsize, struct proc *p)
543 {
544 	int error, bits, fsbsize;
545 	struct timespec ts;
546 	struct partinfo dpart;
547 	struct vattr va;
548 	struct nameidata nd;
549 
550 	/*
551 	 * Get the mounted file system.
552 	 */
553 
554 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount, p);
555 	if ((error = namei(&nd)) != 0)
556 		return error;
557 
558 	if ((nd.ni_vp->v_flag & VROOT) != VROOT) {
559 		vrele(nd.ni_vp);
560 		return EINVAL;
561 	}
562 
563 	sc->sc_mount = nd.ni_vp->v_mount;
564 	memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN);
565 
566 	vrele(nd.ni_vp);
567 
568 	/*
569 	 * Check for file system internal snapshot.
570 	 */
571 
572 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
573 	if ((error = namei(&nd)) != 0)
574 		return error;
575 
576 	if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) {
577 		vrele(nd.ni_vp);
578 		sc->sc_flags |= FSS_PERSISTENT;
579 
580 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
581 		if ((error = vn_open(&nd, FREAD, 0)) != 0)
582 			return error;
583 		sc->sc_bs_vp = nd.ni_vp;
584 
585 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
586 		bits = sizeof(sc->sc_bs_bshift)*NBBY;
587 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits;
588 		    sc->sc_bs_bshift++)
589 			if (FSS_FSBSIZE(sc) == fsbsize)
590 				break;
591 		if (sc->sc_bs_bshift >= bits) {
592 			VOP_UNLOCK(sc->sc_bs_vp, 0);
593 			return EINVAL;
594 		}
595 
596 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
597 		sc->sc_clshift = 0;
598 
599 		error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
600 		TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
601 
602 		VOP_UNLOCK(sc->sc_bs_vp, 0);
603 
604 		return error;
605 	}
606 	vrele(nd.ni_vp);
607 
608 	/*
609 	 * Get the block device it is mounted on.
610 	 */
611 
612 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE,
613 	    sc->sc_mount->mnt_stat.f_mntfromname, p);
614 	if ((error = namei(&nd)) != 0)
615 		return error;
616 
617 	if (nd.ni_vp->v_type != VBLK) {
618 		vrele(nd.ni_vp);
619 		return EINVAL;
620 	}
621 
622 	error = VOP_IOCTL(nd.ni_vp, DIOCGPART, &dpart, FREAD, p->p_ucred, p);
623 	if (error) {
624 		vrele(nd.ni_vp);
625 		return error;
626 	}
627 
628 	sc->sc_mount_vp = nd.ni_vp;
629 	sc->sc_bdev = nd.ni_vp->v_rdev;
630 	*bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size;
631 	vrele(nd.ni_vp);
632 
633 	/*
634 	 * Get the backing store
635 	 */
636 
637 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
638 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0)
639 		return error;
640 	VOP_UNLOCK(nd.ni_vp, 0);
641 
642 	sc->sc_bs_vp = nd.ni_vp;
643 
644 	if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR)
645 		return EINVAL;
646 
647 	if (sc->sc_bs_vp->v_type == VREG) {
648 		error = VOP_GETATTR(sc->sc_bs_vp, &va, p->p_ucred, p);
649 		if (error != 0)
650 			return error;
651 		sc->sc_bs_size = va.va_size;
652 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
653 		if (fsbsize & (fsbsize-1))	/* No power of two */
654 			return EINVAL;
655 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
656 		    sc->sc_bs_bshift++)
657 			if (FSS_FSBSIZE(sc) == fsbsize)
658 				break;
659 		if (sc->sc_bs_bshift >= 32)
660 			return EINVAL;
661 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
662 		sc->sc_flags |= FSS_BS_ALLOC;
663 	} else {
664 		sc->sc_bs_bshift = DEV_BSHIFT;
665 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
666 		sc->sc_flags &= ~FSS_BS_ALLOC;
667 	}
668 
669 	/*
670 	 * As all IO to from/to the backing store goes through
671 	 * VOP_STRATEGY() clean the buffer cache to prevent
672 	 * cache incoherencies.
673 	 */
674 	if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, p->p_ucred, p, 0, 0)) != 0)
675 		return error;
676 
677 	return 0;
678 }
679 
680 /*
681  * Create a snapshot.
682  */
683 static int
684 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct proc *p)
685 {
686 	int len, error;
687 	u_int32_t csize;
688 	off_t bsize;
689 
690 	/*
691 	 * Open needed files.
692 	 */
693 	if ((error = fss_create_files(sc, fss, &bsize, p)) != 0)
694 		goto bad;
695 
696 	if (sc->sc_flags & FSS_PERSISTENT) {
697 		fss_softc_alloc(sc);
698 		sc->sc_flags |= FSS_ACTIVE;
699 		return 0;
700 	}
701 
702 	/*
703 	 * Set cluster size. Must be a power of two and
704 	 * a multiple of backing store block size.
705 	 */
706 	if (fss->fss_csize <= 0)
707 		csize = MAXPHYS;
708 	else
709 		csize = fss->fss_csize;
710 	if (bsize/csize > FSS_CLUSTER_MAX)
711 		csize = bsize/FSS_CLUSTER_MAX+1;
712 
713 	for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32;
714 	    sc->sc_clshift++)
715 		if (FSS_CLSIZE(sc) >= csize)
716 			break;
717 	if (sc->sc_clshift >= 32) {
718 		error = EINVAL;
719 		goto bad;
720 	}
721 	sc->sc_clmask = FSS_CLSIZE(sc)-1;
722 
723 	/*
724 	 * Set number of cache slots.
725 	 */
726 	if (FSS_CLSIZE(sc) <= 8192)
727 		sc->sc_cache_size = 32;
728 	else if (FSS_CLSIZE(sc) <= 65536)
729 		sc->sc_cache_size = 8;
730 	else
731 		sc->sc_cache_size = 4;
732 
733 	/*
734 	 * Set number of clusters and size of last cluster.
735 	 */
736 	sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1;
737 	sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1;
738 
739 	/*
740 	 * Set size of indirect table.
741 	 */
742 	len = sc->sc_clcount*sizeof(u_int32_t);
743 	sc->sc_indir_size = FSS_BTOCL(sc, len)+1;
744 	sc->sc_clnext = sc->sc_indir_size;
745 	sc->sc_indir_cur = 0;
746 
747 	if ((error = fss_softc_alloc(sc)) != 0)
748 		goto bad;
749 
750 	/*
751 	 * Activate the snapshot.
752 	 */
753 
754 	if ((error = vfs_write_suspend(sc->sc_mount, PUSER|PCATCH, 0)) != 0)
755 		goto bad;
756 
757 	microtime(&sc->sc_time);
758 
759 	if (error == 0)
760 		error = vn_cow_establish(sc->sc_mount_vp,
761 		    fss_copy_on_write, sc);
762 	if (error == 0)
763 		sc->sc_flags |= FSS_ACTIVE;
764 
765 	vfs_write_resume(sc->sc_mount);
766 
767 	if (error != 0)
768 		goto bad;
769 
770 #ifdef DEBUG
771 	printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname);
772 	printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n",
773 	    sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc),
774 	    sc->sc_cache_size, sc->sc_indir_size);
775 #endif
776 
777 	return 0;
778 
779 bad:
780 	fss_softc_free(sc);
781 	if (sc->sc_bs_vp != NULL) {
782 		if (sc->sc_flags & FSS_PERSISTENT)
783 			vn_close(sc->sc_bs_vp, FREAD, p->p_ucred, p);
784 		else
785 			vn_close(sc->sc_bs_vp, FREAD|FWRITE, p->p_ucred, p);
786 	}
787 	sc->sc_bs_vp = NULL;
788 
789 	return error;
790 }
791 
792 /*
793  * Delete a snapshot.
794  */
795 static int
796 fss_delete_snapshot(struct fss_softc *sc, struct proc *p)
797 {
798 	int s;
799 
800 	if ((sc->sc_flags & FSS_PERSISTENT) == 0)
801 		vn_cow_disestablish(sc->sc_mount_vp, fss_copy_on_write, sc);
802 
803 	FSS_LOCK(sc, s);
804 	sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR);
805 	sc->sc_mount = NULL;
806 	sc->sc_bdev = NODEV;
807 	FSS_UNLOCK(sc, s);
808 
809 	fss_softc_free(sc);
810 	if (sc->sc_flags & FSS_PERSISTENT)
811 		vn_close(sc->sc_bs_vp, FREAD, p->p_ucred, p);
812 	else
813 		vn_close(sc->sc_bs_vp, FREAD|FWRITE, p->p_ucred, p);
814 	sc->sc_bs_vp = NULL;
815 	sc->sc_flags &= ~(FSS_PERSISTENT|FSS_BS_ALLOC);
816 
817 	FSS_STAT_CLEAR(sc);
818 
819 	return 0;
820 }
821 
822 /*
823  * Get the block address and number of contiguous blocks.
824  * If the file contains a hole, try to allocate.
825  * Backing store is locked by caller.
826  */
827 static int
828 fss_bmap(struct fss_softc *sc, off_t start, int len,
829     struct vnode **vpp, daddr_t *bnp, int *runp)
830 {
831 	int l, s, error;
832 	struct buf *bp, **bpp;
833 
834 	if ((sc->sc_bs_vp->v_mount->mnt_flag & MNT_SOFTDEP) != 0)
835 		bpp = &bp;
836 	else
837 		bpp = NULL;
838 
839 	error = VOP_BMAP(sc->sc_bs_vp, FSS_BTOFSB(sc, start), vpp, bnp, runp);
840 	if ((error == 0 && *bnp != (daddr_t)-1) ||
841 	    (sc->sc_flags & FSS_BS_ALLOC) == 0)
842 		goto out;
843 
844 	if (start+len >= sc->sc_bs_size) {
845 		error = ENOSPC;
846 		goto out;
847 	}
848 
849 	for (l = 0; l < len; l += FSS_FSBSIZE(sc)) {
850 		error = VOP_BALLOC(sc->sc_bs_vp, start+l, FSS_FSBSIZE(sc),
851 		    sc->sc_bs_proc->p_ucred, 0, bpp);
852 		if (error)
853 			goto out;
854 
855 		if (bpp == NULL)
856 			continue;
857 
858 		s = splbio();
859 		simple_lock(&bp->b_interlock);
860 
861 		if (LIST_FIRST(&bp->b_dep) != NULL && bioops.io_start)
862 			(*bioops.io_start)(bp);
863 		if (LIST_FIRST(&bp->b_dep) != NULL && bioops.io_complete)
864 			(*bioops.io_complete)(bp);
865 
866 		bp->b_flags |= B_INVAL;
867 		simple_unlock(&bp->b_interlock);
868 		splx(s);
869 
870 		brelse(bp);
871 	}
872 
873 	error = VOP_BMAP(sc->sc_bs_vp, FSS_BTOFSB(sc, start), vpp, bnp, runp);
874 
875 out:
876 
877 	if ((sc->sc_flags & FSS_PERSISTENT) == 0 &&
878 	    error == 0 && *bnp == (daddr_t)-1)
879 		error = ENOSPC;
880 
881 	return error;
882 }
883 
884 /*
885  * A read from the snapshotted block device has completed.
886  */
887 static void
888 fss_cluster_iodone(struct buf *bp)
889 {
890 	int s;
891 	struct fss_cache *scp = bp->b_private;
892 
893 	FSS_LOCK(scp->fc_softc, s);
894 
895 	if (bp->b_flags & B_EINTR)
896 		fss_error(scp->fc_softc, "fs read interrupted");
897 	if (bp->b_flags & B_ERROR)
898 		fss_error(scp->fc_softc, "fs read error %d", bp->b_error);
899 
900 	if (bp->b_vp != NULL)
901 		brelvp(bp);
902 
903 	if (--scp->fc_xfercount == 0)
904 		wakeup(&scp->fc_data);
905 
906 	FSS_UNLOCK(scp->fc_softc, s);
907 
908 	s = splbio();
909 	pool_put(&bufpool, bp);
910 	splx(s);
911 }
912 
913 /*
914  * Read a cluster from the snapshotted block device to the cache.
915  */
916 static void
917 fss_read_cluster(struct fss_softc *sc, u_int32_t cl)
918 {
919 	int s, todo, len;
920 	caddr_t addr;
921 	daddr_t dblk;
922 	struct buf *bp;
923 	struct fss_cache *scp, *scl;
924 
925 	/*
926 	 * Get a free cache slot.
927 	 */
928 	scl = sc->sc_cache+sc->sc_cache_size;
929 
930 	FSS_LOCK(sc, s);
931 
932 restart:
933 	if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) {
934 		FSS_UNLOCK(sc, s);
935 		return;
936 	}
937 
938 	for (scp = sc->sc_cache; scp < scl; scp++)
939 		if (scp->fc_type != FSS_CACHE_FREE &&
940 		    scp->fc_cluster == cl) {
941 			ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0,
942 			    &sc->sc_slock);
943 			goto restart;
944 		}
945 
946 	for (scp = sc->sc_cache; scp < scl; scp++)
947 		if (scp->fc_type == FSS_CACHE_FREE) {
948 			scp->fc_type = FSS_CACHE_BUSY;
949 			scp->fc_cluster = cl;
950 			break;
951 		}
952 	if (scp >= scl) {
953 		FSS_STAT_INC(sc, cow_cache_full);
954 		ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock);
955 		goto restart;
956 	}
957 
958 	FSS_UNLOCK(sc, s);
959 
960 	/*
961 	 * Start the read.
962 	 */
963 	FSS_STAT_INC(sc, cow_copied);
964 
965 	dblk = btodb(FSS_CLTOB(sc, cl));
966 	addr = scp->fc_data;
967 	if (cl == sc->sc_clcount-1) {
968 		todo = sc->sc_clresid;
969 		memset(addr+todo, 0, FSS_CLSIZE(sc)-todo);
970 	} else
971 		todo = FSS_CLSIZE(sc);
972 	while (todo > 0) {
973 		len = todo;
974 		if (len > MAXPHYS)
975 			len = MAXPHYS;
976 
977 		s = splbio();
978 		bp = pool_get(&bufpool, PR_WAITOK);
979 		splx(s);
980 
981 		BUF_INIT(bp);
982 		bp->b_flags = B_READ|B_CALL;
983 		bp->b_bcount = len;
984 		bp->b_bufsize = bp->b_bcount;
985 		bp->b_error = 0;
986 		bp->b_data = addr;
987 		bp->b_blkno = bp->b_rawblkno = dblk;
988 		bp->b_proc = NULL;
989 		bp->b_dev = sc->sc_bdev;
990 		bp->b_vp = NULLVP;
991 		bp->b_private = scp;
992 		bp->b_iodone = fss_cluster_iodone;
993 
994 		DEV_STRATEGY(bp);
995 
996 		FSS_LOCK(sc, s);
997 		scp->fc_xfercount++;
998 		FSS_UNLOCK(sc, s);
999 
1000 		dblk += btodb(len);
1001 		addr += len;
1002 		todo -= len;
1003 	}
1004 
1005 	/*
1006 	 * Wait for all read requests to complete.
1007 	 */
1008 	FSS_LOCK(sc, s);
1009 	while (scp->fc_xfercount > 0)
1010 		ltsleep(&scp->fc_data, PRIBIO, "cowwait", 0, &sc->sc_slock);
1011 
1012 	scp->fc_type = FSS_CACHE_VALID;
1013 	setbit(sc->sc_copied, scp->fc_cluster);
1014 	FSS_UNLOCK(sc, s);
1015 
1016 	wakeup(&sc->sc_bs_proc);
1017 }
1018 
1019 /*
1020  * Write a cluster from the cache to the backing store.
1021  */
1022 static int
1023 fss_write_cluster(struct fss_cache *scp, u_int32_t cl)
1024 {
1025 	int s, error, todo, len, nra;
1026 	daddr_t nbn;
1027 	caddr_t addr;
1028 	off_t pos;
1029 	struct buf *bp;
1030 	struct vnode *vp;
1031 	struct fss_softc *sc;
1032 
1033 	error = 0;
1034 	sc = scp->fc_softc;
1035 
1036 	pos = FSS_CLTOB(sc, cl);
1037 	addr = scp->fc_data;
1038 	todo = FSS_CLSIZE(sc);
1039 
1040 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
1041 	simple_lock(&sc->sc_bs_vp->v_interlock);
1042 	error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(pos),
1043 	    round_page(pos+todo), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
1044 
1045 	while (error == 0 && todo > 0) {
1046 		error = fss_bmap(sc, pos, todo, &vp, &nbn, &nra);
1047 		if (error)
1048 			break;
1049 
1050 		len = FSS_FSBTOB(sc, nra+1)-FSS_FSBOFF(sc, pos);
1051 		if (len > todo)
1052 			len = todo;
1053 
1054 		s = splbio();
1055 		bp = pool_get(&bufpool, PR_WAITOK);
1056 		splx(s);
1057 
1058 		BUF_INIT(bp);
1059 		bp->b_flags = B_CALL;
1060 		bp->b_bcount = len;
1061 		bp->b_bufsize = bp->b_bcount;
1062 		bp->b_error = 0;
1063 		bp->b_data = addr;
1064 		bp->b_blkno = bp->b_rawblkno = nbn+btodb(FSS_FSBOFF(sc, pos));
1065 		bp->b_proc = NULL;
1066 		bp->b_vp = NULLVP;
1067 		bp->b_private = scp;
1068 		bp->b_iodone = fss_cluster_iodone;
1069 		bgetvp(vp, bp);
1070 		bp->b_vp->v_numoutput++;
1071 
1072 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
1073 		VOP_STRATEGY(vp, bp);
1074 
1075 		FSS_LOCK(sc, s);
1076 		scp->fc_xfercount++;
1077 		FSS_UNLOCK(sc, s);
1078 
1079 		pos += len;
1080 		addr += len;
1081 		todo -= len;
1082 	}
1083 
1084 	/*
1085 	 * Wait for all write requests to complete.
1086 	 */
1087 	FSS_LOCK(sc, s);
1088 	while (scp->fc_xfercount > 0)
1089 		ltsleep(&scp->fc_data, PRIBIO, "bswwait", 0, &sc->sc_slock);
1090 	FSS_UNLOCK(sc, s);
1091 
1092 	VOP_UNLOCK(sc->sc_bs_vp, 0);
1093 
1094 	return error;
1095 }
1096 
1097 /*
1098  * Read/write clusters from/to backing store.
1099  * For persistent snapshots must be called with cl == 0. off is the
1100  * offset into the snapshot.
1101  */
1102 static int
1103 fss_bs_io(struct fss_softc *sc, fss_io_type rw,
1104     u_int32_t cl, off_t off, int len, caddr_t data)
1105 {
1106 	int s, error, todo, count, nra;
1107 	off_t pos;
1108 	daddr_t nbn;
1109 	struct buf *bp;
1110 	struct vnode *vp;
1111 
1112 	todo = len;
1113 	pos = FSS_CLTOB(sc, cl)+off;
1114 	error = 0;
1115 
1116 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
1117 	simple_lock(&sc->sc_bs_vp->v_interlock);
1118 	error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(pos),
1119 	    round_page(pos+todo), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
1120 
1121 	while (error == 0 && todo > 0) {
1122 		error = fss_bmap(sc, pos, todo, &vp, &nbn, &nra);
1123 		if (error)
1124 			break;
1125 
1126 		count = FSS_FSBTOB(sc, nra+1)-FSS_FSBOFF(sc, pos);
1127 		if (count > todo)
1128 			count = todo;
1129 
1130 		if (nbn == (daddr_t)-1) {
1131 			bzero(data, count);
1132 			todo -= count;
1133 			data += count;
1134 			pos += count;
1135 			continue;
1136 		}
1137 
1138 		s = splbio();
1139 		bp = pool_get(&bufpool, PR_WAITOK);
1140 		splx(s);
1141 
1142 		BUF_INIT(bp);
1143 		bp->b_flags = (rw == FSS_READ ? B_READ : 0);
1144 		bp->b_bcount = count;
1145 		bp->b_bufsize = bp->b_bcount;
1146 		bp->b_error = 0;
1147 		bp->b_data = data;
1148 		bp->b_blkno = bp->b_rawblkno = nbn+btodb(FSS_FSBOFF(sc, pos));
1149 		bp->b_proc = NULL;
1150 		bp->b_vp = NULLVP;
1151 		bgetvp(vp, bp);
1152 		if ((bp->b_flags & B_READ) == 0)
1153 			bp->b_vp->v_numoutput++;
1154 
1155 		if ((bp->b_flags & B_READ) == 0 || cl < sc->sc_indir_size)
1156 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
1157 		VOP_STRATEGY(vp, bp);
1158 
1159 		error = biowait(bp);
1160 
1161 		if (bp->b_vp != NULL)
1162 			brelvp(bp);
1163 
1164 		s = splbio();
1165 		pool_put(&bufpool, bp);
1166 		splx(s);
1167 
1168 		if (error)
1169 			break;
1170 
1171 		todo -= count;
1172 		data += count;
1173 		pos += count;
1174 	}
1175 
1176 	VOP_UNLOCK(sc->sc_bs_vp, 0);
1177 
1178 	return error;
1179 }
1180 
1181 /*
1182  * Get a pointer to the indirect slot for this cluster.
1183  */
1184 static u_int32_t *
1185 fss_bs_indir(struct fss_softc *sc, u_int32_t cl)
1186 {
1187 	u_int32_t icl;
1188 	int ioff;
1189 
1190 	icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1191 	ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1192 
1193 	if (sc->sc_indir_cur == icl)
1194 		return &sc->sc_indir_data[ioff];
1195 
1196 	if (sc->sc_indir_dirty) {
1197 		FSS_STAT_INC(sc, indir_write);
1198 		if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0,
1199 		    FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0)
1200 			return NULL;
1201 		setbit(sc->sc_indir_valid, sc->sc_indir_cur);
1202 	}
1203 
1204 	sc->sc_indir_dirty = 0;
1205 	sc->sc_indir_cur = icl;
1206 
1207 	if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) {
1208 		FSS_STAT_INC(sc, indir_read);
1209 		if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0,
1210 		    FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0)
1211 			return NULL;
1212 	} else
1213 		memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc));
1214 
1215 	return &sc->sc_indir_data[ioff];
1216 }
1217 
1218 /*
1219  * The kernel thread (one for every active snapshot).
1220  *
1221  * After wakeup it cleans the cache and runs the I/O requests.
1222  */
1223 static void
1224 fss_bs_thread(void *arg)
1225 {
1226 	int error, len, nfreed, nio, s;
1227 	long off;
1228 	caddr_t addr;
1229 	u_int32_t c, cl, ch, *indirp;
1230 	struct buf *bp, *nbp;
1231 	struct fss_softc *sc;
1232 	struct fss_cache *scp, *scl;
1233 
1234 	sc = arg;
1235 
1236 	scl = sc->sc_cache+sc->sc_cache_size;
1237 
1238 	s = splbio();
1239 	nbp = pool_get(&bufpool, PR_WAITOK);
1240 	splx(s);
1241 
1242 	nfreed = nio = 1;		/* Dont sleep the first time */
1243 
1244 	FSS_LOCK(sc, s);
1245 
1246 	for (;;) {
1247 		if (nfreed == 0 && nio == 0)
1248 			ltsleep(&sc->sc_bs_proc, PVM-1, "fssbs", 0,
1249 			    &sc->sc_slock);
1250 
1251 		if ((sc->sc_flags & FSS_BS_THREAD) == 0) {
1252 			sc->sc_bs_proc = NULL;
1253 			wakeup(&sc->sc_bs_proc);
1254 
1255 			FSS_UNLOCK(sc, s);
1256 
1257 			s = splbio();
1258 			pool_put(&bufpool, nbp);
1259 			splx(s);
1260 #ifdef FSS_STATISTICS
1261 			if ((sc->sc_flags & FSS_PERSISTENT) == 0) {
1262 				printf("fss%d: cow called %" PRId64 " times,"
1263 				    " copied %" PRId64 " clusters,"
1264 				    " cache full %" PRId64 " times\n",
1265 				    sc->sc_unit,
1266 				    FSS_STAT_VAL(sc, cow_calls),
1267 				    FSS_STAT_VAL(sc, cow_copied),
1268 				    FSS_STAT_VAL(sc, cow_cache_full));
1269 				printf("fss%d: %" PRId64 " indir reads,"
1270 				    " %" PRId64 " indir writes\n",
1271 				    sc->sc_unit,
1272 				    FSS_STAT_VAL(sc, indir_read),
1273 				    FSS_STAT_VAL(sc, indir_write));
1274 			}
1275 #endif /* FSS_STATISTICS */
1276 			kthread_exit(0);
1277 		}
1278 
1279 		/*
1280 		 * Process I/O requests (persistent)
1281 		 */
1282 
1283 		if (sc->sc_flags & FSS_PERSISTENT) {
1284 			nfreed = nio = 0;
1285 
1286 			if ((bp = BUFQ_GET(&sc->sc_bufq)) == NULL)
1287 				continue;
1288 
1289 			nio++;
1290 
1291 			if (FSS_ISVALID(sc)) {
1292 				FSS_UNLOCK(sc, s);
1293 
1294 				error = fss_bs_io(sc, FSS_READ, 0,
1295 				    dbtob(bp->b_blkno), bp->b_bcount,
1296 				    bp->b_data);
1297 
1298 				FSS_LOCK(sc, s);
1299 			} else
1300 				error = ENXIO;
1301 
1302 			if (error) {
1303 				bp->b_error = error;
1304 				bp->b_flags |= B_ERROR;
1305 				bp->b_resid = bp->b_bcount;
1306 			}
1307 			biodone(bp);
1308 
1309 			continue;
1310 		}
1311 
1312 		/*
1313 		 * Clean the cache
1314 		 */
1315 		nfreed = 0;
1316 		for (scp = sc->sc_cache; scp < scl; scp++) {
1317 			if (scp->fc_type != FSS_CACHE_VALID)
1318 				continue;
1319 
1320 			FSS_UNLOCK(sc, s);
1321 
1322 			indirp = fss_bs_indir(sc, scp->fc_cluster);
1323 			if (indirp != NULL) {
1324 				error = fss_write_cluster(scp, sc->sc_clnext);
1325 			} else
1326 				error = EIO;
1327 
1328 			FSS_LOCK(sc, s);
1329 
1330 			if (error == 0) {
1331 				*indirp = sc->sc_clnext++;
1332 				sc->sc_indir_dirty = 1;
1333 			} else
1334 				fss_error(sc, "write bs error %d", error);
1335 
1336 			scp->fc_type = FSS_CACHE_FREE;
1337 			nfreed++;
1338 			wakeup(&scp->fc_type);
1339 		}
1340 
1341 		if (nfreed)
1342 			wakeup(&sc->sc_cache);
1343 
1344 		/*
1345 		 * Process I/O requests
1346 		 */
1347 		nio = 0;
1348 
1349 		if ((bp = BUFQ_GET(&sc->sc_bufq)) == NULL)
1350 			continue;
1351 
1352 		nio++;
1353 
1354 		if (!FSS_ISVALID(sc)) {
1355 			bp->b_error = ENXIO;
1356 			bp->b_flags |= B_ERROR;
1357 			bp->b_resid = bp->b_bcount;
1358 			biodone(bp);
1359 			continue;
1360 		}
1361 
1362 		/*
1363 		 * First read from the snapshotted block device.
1364 		 * XXX Split to only read those parts that have not
1365 		 * been saved to backing store?
1366 		 */
1367 
1368 		FSS_UNLOCK(sc, s);
1369 
1370 		BUF_INIT(nbp);
1371 		nbp->b_flags = B_READ;
1372 		nbp->b_bcount = bp->b_bcount;
1373 		nbp->b_bufsize = bp->b_bcount;
1374 		nbp->b_error = 0;
1375 		nbp->b_data = bp->b_data;
1376 		nbp->b_blkno = nbp->b_rawblkno = bp->b_blkno;
1377 		nbp->b_proc = bp->b_proc;
1378 		nbp->b_dev = sc->sc_bdev;
1379 		nbp->b_vp = NULLVP;
1380 
1381 		DEV_STRATEGY(nbp);
1382 
1383 		if (biowait(nbp) != 0) {
1384 			bp->b_resid = bp->b_bcount;
1385 			bp->b_error = nbp->b_error;
1386 			bp->b_flags |= B_ERROR;
1387 			biodone(bp);
1388 			continue;
1389 		}
1390 
1391 		cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
1392 		off = FSS_CLOFF(sc, dbtob(bp->b_blkno));
1393 		ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
1394 		bp->b_resid = bp->b_bcount;
1395 		addr = bp->b_data;
1396 
1397 		FSS_LOCK(sc, s);
1398 
1399 		/*
1400 		 * Replace those parts that have been saved to backing store.
1401 		 */
1402 
1403 		for (c = cl; c <= ch;
1404 		    c++, off = 0, bp->b_resid -= len, addr += len) {
1405 			len = FSS_CLSIZE(sc)-off;
1406 			if (len > bp->b_resid)
1407 				len = bp->b_resid;
1408 
1409 			if (isclr(sc->sc_copied, c))
1410 				continue;
1411 
1412 			FSS_UNLOCK(sc, s);
1413 
1414 			indirp = fss_bs_indir(sc, c);
1415 
1416 			FSS_LOCK(sc, s);
1417 
1418 			if (indirp == NULL || *indirp == 0) {
1419 				/*
1420 				 * Not on backing store. Either in cache
1421 				 * or hole in the snapshotted block device.
1422 				 */
1423 				for (scp = sc->sc_cache; scp < scl; scp++)
1424 					if (scp->fc_type == FSS_CACHE_VALID &&
1425 					    scp->fc_cluster == c)
1426 						break;
1427 				if (scp < scl)
1428 					memcpy(addr, scp->fc_data+off, len);
1429 				else
1430 					memset(addr, 0, len);
1431 				continue;
1432 			}
1433 			/*
1434 			 * Read from backing store.
1435 			 */
1436 
1437 			FSS_UNLOCK(sc, s);
1438 
1439 			if ((error = fss_bs_io(sc, FSS_READ, *indirp,
1440 			    off, len, addr)) != 0) {
1441 				bp->b_resid = bp->b_bcount;
1442 				bp->b_error = error;
1443 				bp->b_flags |= B_ERROR;
1444 				break;
1445 			}
1446 
1447 			FSS_LOCK(sc, s);
1448 
1449 		}
1450 
1451 		biodone(bp);
1452 	}
1453 }
1454