xref: /netbsd-src/sys/dev/fss.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: fss.c,v 1.11 2004/10/29 15:39:38 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 1.11 2004/10/29 15:39:38 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 	struct fss_softc *sc;
176 
177 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
178 		return ENODEV;
179 
180 	return 0;
181 }
182 
183 int
184 fss_close(dev_t dev, int flags, int mode, struct proc *p)
185 {
186 	struct fss_softc *sc;
187 
188 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
189 		return ENODEV;
190 
191 	return 0;
192 }
193 
194 void
195 fss_strategy(struct buf *bp)
196 {
197 	int s;
198 	struct fss_softc *sc;
199 
200 	sc = FSS_DEV_TO_SOFTC(bp->b_dev);
201 
202 	FSS_LOCK(sc, s);
203 
204 	if ((bp->b_flags & B_READ) != B_READ ||
205 	    sc == NULL || !FSS_ISVALID(sc)) {
206 
207 		FSS_UNLOCK(sc, s);
208 
209 		bp->b_error = (sc == NULL ? ENODEV : EROFS);
210 		bp->b_flags |= B_ERROR;
211 		bp->b_resid = bp->b_bcount;
212 		biodone(bp);
213 		return;
214 	}
215 
216 	bp->b_rawblkno = bp->b_blkno;
217 	BUFQ_PUT(&sc->sc_bufq, bp);
218 	wakeup(&sc->sc_bs_proc);
219 
220 	FSS_UNLOCK(sc, s);
221 }
222 
223 int
224 fss_read(dev_t dev, struct uio *uio, int flags)
225 {
226 	return physio(fss_strategy, NULL, dev, B_READ, minphys, uio);
227 }
228 
229 int
230 fss_write(dev_t dev, struct uio *uio, int flags)
231 {
232 	return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio);
233 }
234 
235 int
236 fss_ioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct proc *p)
237 {
238 	int s, error;
239 	struct fss_softc *sc;
240 	struct fss_set *fss = (struct fss_set *)data;
241 	struct fss_get *fsg = (struct fss_get *)data;
242 
243 	if ((sc = FSS_DEV_TO_SOFTC(dev)) == NULL)
244 		return ENODEV;
245 
246 	FSS_LOCK(sc, s);
247 	while ((sc->sc_flags & FSS_EXCL) == FSS_EXCL) {
248 		error = ltsleep(sc, PRIBIO|PCATCH, "fsslock", 0, &sc->sc_slock);
249 		if (error) {
250 			FSS_UNLOCK(sc, s);
251 			return error;
252 		}
253 	}
254 	sc->sc_flags |= FSS_EXCL;
255 	FSS_UNLOCK(sc, s);
256 
257 	error = EINVAL;
258 
259 	switch (cmd) {
260 	case FSSIOCSET:
261 		if ((flag & FWRITE) == 0)
262 			error = EPERM;
263 		else if ((sc->sc_flags & FSS_ACTIVE) != 0)
264 			error = EBUSY;
265 		else
266 			error = fss_create_snapshot(sc, fss, p);
267 		break;
268 
269 	case FSSIOCCLR:
270 		if ((flag & FWRITE) == 0)
271 			error = EPERM;
272 		else if ((sc->sc_flags & FSS_ACTIVE) == 0)
273 			error = ENXIO;
274 		else
275 			error = fss_delete_snapshot(sc, p);
276 		break;
277 
278 	case FSSIOCGET:
279 		switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) {
280 		case FSS_ACTIVE:
281 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
282 			fsg->fsg_csize = FSS_CLSIZE(sc);
283 			fsg->fsg_time = sc->sc_time;
284 			fsg->fsg_mount_size = sc->sc_clcount;
285 			fsg->fsg_bs_size = sc->sc_clnext;
286 			error = 0;
287 			break;
288 		case FSS_PERSISTENT | FSS_ACTIVE:
289 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
290 			fsg->fsg_csize = 0;
291 			fsg->fsg_time = sc->sc_time;
292 			fsg->fsg_mount_size = 0;
293 			fsg->fsg_bs_size = 0;
294 			error = 0;
295 			break;
296 		default:
297 			error = ENXIO;
298 			break;
299 		}
300 		break;
301 	}
302 
303 	FSS_LOCK(sc, s);
304 	sc->sc_flags &= ~FSS_EXCL;
305 	FSS_UNLOCK(sc, s);
306 	wakeup(sc);
307 
308 	return error;
309 }
310 
311 int
312 fss_size(dev_t dev)
313 {
314 	return -1;
315 }
316 
317 int
318 fss_dump(dev_t dev, daddr_t blkno, caddr_t va, size_t size)
319 {
320 	return EROFS;
321 }
322 
323 /*
324  * An error occurred reading or writing the snapshot or backing store.
325  * If it is the first error log to console.
326  * The caller holds the simplelock.
327  */
328 static inline void
329 fss_error(struct fss_softc *sc, const char *fmt, ...)
330 {
331 	va_list ap;
332 
333 	if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) {
334 		va_start(ap, fmt);
335 		printf("fss%d: snapshot invalid: ", sc->sc_unit);
336 		vprintf(fmt, ap);
337 		printf("\n");
338 		va_end(ap);
339 	}
340 	if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE)
341 		sc->sc_flags |= FSS_ERROR;
342 }
343 
344 /*
345  * Allocate the variable sized parts of the softc and
346  * fork the kernel thread.
347  *
348  * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size
349  * must be initialized.
350  */
351 static int
352 fss_softc_alloc(struct fss_softc *sc)
353 {
354 	int i, len, error;
355 
356 	len = (sc->sc_clcount+NBBY-1)/NBBY;
357 	sc->sc_copied = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
358 	if (sc->sc_copied == NULL)
359 		return(ENOMEM);
360 
361 	len = sc->sc_cache_size*sizeof(struct fss_cache);
362 	sc->sc_cache = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
363 	if (sc->sc_cache == NULL)
364 		return(ENOMEM);
365 
366 	len = FSS_CLSIZE(sc);
367 	for (i = 0; i < sc->sc_cache_size; i++) {
368 		sc->sc_cache[i].fc_type = FSS_CACHE_FREE;
369 		sc->sc_cache[i].fc_softc = sc;
370 		sc->sc_cache[i].fc_xfercount = 0;
371 		sc->sc_cache[i].fc_data = malloc(len, M_TEMP,
372 		    M_WAITOK|M_CANFAIL);
373 		if (sc->sc_cache[i].fc_data == NULL)
374 			return(ENOMEM);
375 	}
376 
377 	len = (sc->sc_indir_size+NBBY-1)/NBBY;
378 	sc->sc_indir_valid = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
379 	if (sc->sc_indir_valid == NULL)
380 		return(ENOMEM);
381 
382 	len = FSS_CLSIZE(sc);
383 	sc->sc_indir_data = malloc(len, M_TEMP, M_ZERO|M_WAITOK|M_CANFAIL);
384 	if (sc->sc_indir_data == NULL)
385 		return(ENOMEM);
386 
387 	if ((error = kthread_create1(fss_bs_thread, sc, &sc->sc_bs_proc,
388 	    "fssbs%d", sc->sc_unit)) != 0)
389 		return error;
390 
391 	sc->sc_flags |= FSS_BS_THREAD;
392 	return 0;
393 }
394 
395 /*
396  * Free the variable sized parts of the softc.
397  */
398 static void
399 fss_softc_free(struct fss_softc *sc)
400 {
401 	int s, i;
402 
403 	if ((sc->sc_flags & FSS_BS_THREAD) != 0) {
404 		FSS_LOCK(sc, s);
405 		sc->sc_flags &= ~FSS_BS_THREAD;
406 		wakeup(&sc->sc_bs_proc);
407 		while (sc->sc_bs_proc != NULL)
408 			ltsleep(&sc->sc_bs_proc, PRIBIO, "fssthread", 0,
409 			    &sc->sc_slock);
410 		FSS_UNLOCK(sc, s);
411 	}
412 
413 	if (sc->sc_copied != NULL)
414 		free(sc->sc_copied, M_TEMP);
415 	sc->sc_copied = NULL;
416 
417 	if (sc->sc_cache != NULL) {
418 		for (i = 0; i < sc->sc_cache_size; i++)
419 			if (sc->sc_cache[i].fc_data != NULL)
420 				free(sc->sc_cache[i].fc_data, M_TEMP);
421 		free(sc->sc_cache, M_TEMP);
422 	}
423 	sc->sc_cache = NULL;
424 
425 	if (sc->sc_indir_valid != NULL)
426 		free(sc->sc_indir_valid, M_TEMP);
427 	sc->sc_indir_valid = NULL;
428 
429 	if (sc->sc_indir_data != NULL)
430 		free(sc->sc_indir_data, M_TEMP);
431 	sc->sc_indir_data = NULL;
432 }
433 
434 /*
435  * Check if an unmount is ok. If forced, set this snapshot into ERROR state.
436  */
437 int
438 fss_umount_hook(struct mount *mp, int forced)
439 {
440 	int i, s;
441 
442 	for (i = 0; i < NFSS; i++) {
443 		FSS_LOCK(&fss_softc[i], s);
444 		if ((fss_softc[i].sc_flags & FSS_ACTIVE) != 0 &&
445 		    fss_softc[i].sc_mount == mp) {
446 			if (forced)
447 				fss_error(&fss_softc[i], "forced unmount");
448 			else {
449 				FSS_UNLOCK(&fss_softc[i], s);
450 				return EBUSY;
451 			}
452 		}
453 		FSS_UNLOCK(&fss_softc[i], s);
454 	}
455 
456 	return 0;
457 }
458 
459 /*
460  * A buffer is written to the snapshotted block device. Copy to
461  * backing store if needed.
462  */
463 static int
464 fss_copy_on_write(void *v, struct buf *bp)
465 {
466 	int s;
467 	u_int32_t cl, ch, c;
468 	struct fss_softc *sc = v;
469 
470 	FSS_LOCK(sc, s);
471 	if (!FSS_ISVALID(sc)) {
472 		FSS_UNLOCK(sc, s);
473 		return 0;
474 	}
475 
476 	FSS_UNLOCK(sc, s);
477 
478 	FSS_STAT_INC(sc, cow_calls);
479 
480 	cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
481 	ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
482 
483 	for (c = cl; c <= ch; c++)
484 		fss_read_cluster(sc, c);
485 
486 	return 0;
487 }
488 
489 /*
490  * Lookup and open needed files.
491  *
492  * For file system internal snapshot initializes sc_mntname, sc_mount,
493  * sc_bs_vp and sc_time.
494  *
495  * Otherwise returns dev and size of the underlying block device.
496  * Initializes sc_mntname, sc_mount_vp, sc_bdev, sc_bs_vp and sc_mount
497  */
498 static int
499 fss_create_files(struct fss_softc *sc, struct fss_set *fss,
500     off_t *bsize, struct proc *p)
501 {
502 	int error, fsbsize;
503 	struct timespec ts;
504 	struct partinfo dpart;
505 	struct vattr va;
506 	struct nameidata nd;
507 
508 	/*
509 	 * Get the mounted file system.
510 	 */
511 
512 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount, p);
513 	if ((error = namei(&nd)) != 0)
514 		return error;
515 
516 	if ((nd.ni_vp->v_flag & VROOT) != VROOT) {
517 		vrele(nd.ni_vp);
518 		return EINVAL;
519 	}
520 
521 	sc->sc_mount = nd.ni_vp->v_mount;
522 	memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN);
523 
524 	vrele(nd.ni_vp);
525 
526 	/*
527 	 * Check for file system internal snapshot.
528 	 */
529 
530 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
531 	if ((error = namei(&nd)) != 0)
532 		return error;
533 
534 	if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) {
535 		vrele(nd.ni_vp);
536 		sc->sc_flags |= FSS_PERSISTENT;
537 
538 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
539 		if ((error = vn_open(&nd, FREAD, 0)) != 0)
540 			return error;
541 		sc->sc_bs_vp = nd.ni_vp;
542 
543 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
544 		if (fsbsize & (fsbsize-1))	/* No power of two */
545 			return EINVAL;
546 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
547 		    sc->sc_bs_bshift++)
548 			if (FSS_FSBSIZE(sc) == fsbsize)
549 				break;
550 		if (sc->sc_bs_bshift >= 32)
551 			return EINVAL;
552 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
553 
554 		sc->sc_clshift = 0;
555 
556 		error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
557 		TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
558 
559 		VOP_UNLOCK(sc->sc_bs_vp, 0);
560 
561 		return error;
562 	}
563 	vrele(nd.ni_vp);
564 
565 	/*
566 	 * Get the block device it is mounted on.
567 	 */
568 
569 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE,
570 	    sc->sc_mount->mnt_stat.f_mntfromname, p);
571 	if ((error = namei(&nd)) != 0)
572 		return error;
573 
574 	if (nd.ni_vp->v_type != VBLK) {
575 		vrele(nd.ni_vp);
576 		return EINVAL;
577 	}
578 
579 	error = VOP_IOCTL(nd.ni_vp, DIOCGPART, &dpart, FREAD, p->p_ucred, p);
580 	if (error) {
581 		vrele(nd.ni_vp);
582 		return error;
583 	}
584 
585 	sc->sc_mount_vp = nd.ni_vp;
586 	sc->sc_bdev = nd.ni_vp->v_rdev;
587 	*bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size;
588 	vrele(nd.ni_vp);
589 
590 	/*
591 	 * Get the backing store
592 	 */
593 
594 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore, p);
595 	if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0)
596 		return error;
597 	VOP_UNLOCK(nd.ni_vp, 0);
598 
599 	sc->sc_bs_vp = nd.ni_vp;
600 
601 	if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR)
602 		return EINVAL;
603 
604 	if (sc->sc_bs_vp->v_type == VREG) {
605 		error = VOP_GETATTR(sc->sc_bs_vp, &va, p->p_ucred, p);
606 		if (error != 0)
607 			return error;
608 		sc->sc_bs_size = va.va_size;
609 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
610 		if (fsbsize & (fsbsize-1))	/* No power of two */
611 			return EINVAL;
612 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
613 		    sc->sc_bs_bshift++)
614 			if (FSS_FSBSIZE(sc) == fsbsize)
615 				break;
616 		if (sc->sc_bs_bshift >= 32)
617 			return EINVAL;
618 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
619 		sc->sc_flags |= FSS_BS_ALLOC;
620 	} else {
621 		sc->sc_bs_bshift = DEV_BSHIFT;
622 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
623 		sc->sc_flags &= ~FSS_BS_ALLOC;
624 	}
625 
626 	/*
627 	 * As all IO to from/to the backing store goes through
628 	 * VOP_STRATEGY() clean the buffer cache to prevent
629 	 * cache incoherencies.
630 	 */
631 	if ((error = vinvalbuf(sc->sc_bs_vp, V_SAVE, p->p_ucred, p, 0, 0)) != 0)
632 		return error;
633 
634 	return 0;
635 }
636 
637 /*
638  * Create a snapshot.
639  */
640 static int
641 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct proc *p)
642 {
643 	int len, error;
644 	u_int32_t csize;
645 	off_t bsize;
646 
647 	/*
648 	 * Open needed files.
649 	 */
650 	if ((error = fss_create_files(sc, fss, &bsize, p)) != 0)
651 		goto bad;
652 
653 	if (sc->sc_flags & FSS_PERSISTENT) {
654 		fss_softc_alloc(sc);
655 		sc->sc_flags |= FSS_ACTIVE;
656 		return 0;
657 	}
658 
659 	/*
660 	 * Set cluster size. Must be a power of two and
661 	 * a multiple of backing store block size.
662 	 */
663 	if (fss->fss_csize <= 0)
664 		csize = MAXPHYS;
665 	else
666 		csize = fss->fss_csize;
667 	if (bsize/csize > FSS_CLUSTER_MAX)
668 		csize = bsize/FSS_CLUSTER_MAX+1;
669 
670 	for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32;
671 	    sc->sc_clshift++)
672 		if (FSS_CLSIZE(sc) >= csize)
673 			break;
674 	if (sc->sc_clshift >= 32) {
675 		error = EINVAL;
676 		goto bad;
677 	}
678 	sc->sc_clmask = FSS_CLSIZE(sc)-1;
679 
680 	/*
681 	 * Set number of cache slots.
682 	 */
683 	if (FSS_CLSIZE(sc) <= 8192)
684 		sc->sc_cache_size = 32;
685 	else if (FSS_CLSIZE(sc) <= 65536)
686 		sc->sc_cache_size = 8;
687 	else
688 		sc->sc_cache_size = 4;
689 
690 	/*
691 	 * Set number of clusters and size of last cluster.
692 	 */
693 	sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1;
694 	sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1;
695 
696 	/*
697 	 * Set size of indirect table.
698 	 */
699 	len = sc->sc_clcount*sizeof(u_int32_t);
700 	sc->sc_indir_size = FSS_BTOCL(sc, len)+1;
701 	sc->sc_clnext = sc->sc_indir_size;
702 	sc->sc_indir_cur = 0;
703 
704 	if ((error = fss_softc_alloc(sc)) != 0)
705 		goto bad;
706 
707 	/*
708 	 * Activate the snapshot.
709 	 */
710 
711 	if ((error = vfs_write_suspend(sc->sc_mount, PUSER|PCATCH, 0)) != 0)
712 		goto bad;
713 
714 	microtime(&sc->sc_time);
715 
716 	if (error == 0)
717 		error = vn_cow_establish(sc->sc_mount_vp,
718 		    fss_copy_on_write, sc);
719 	if (error == 0)
720 		sc->sc_flags |= FSS_ACTIVE;
721 
722 	vfs_write_resume(sc->sc_mount);
723 
724 	if (error != 0)
725 		goto bad;
726 
727 #ifdef DEBUG
728 	printf("fss%d: %s snapshot active\n", sc->sc_unit, sc->sc_mntname);
729 	printf("fss%d: %u clusters of %u, %u cache slots, %u indir clusters\n",
730 	    sc->sc_unit, sc->sc_clcount, FSS_CLSIZE(sc),
731 	    sc->sc_cache_size, sc->sc_indir_size);
732 #endif
733 
734 	return 0;
735 
736 bad:
737 	fss_softc_free(sc);
738 	if (sc->sc_bs_vp != NULL) {
739 		if (sc->sc_flags & FSS_PERSISTENT)
740 			vn_close(sc->sc_bs_vp, FREAD, p->p_ucred, p);
741 		else
742 			vn_close(sc->sc_bs_vp, FREAD|FWRITE, p->p_ucred, p);
743 	}
744 	sc->sc_bs_vp = NULL;
745 
746 	return error;
747 }
748 
749 /*
750  * Delete a snapshot.
751  */
752 static int
753 fss_delete_snapshot(struct fss_softc *sc, struct proc *p)
754 {
755 	int s;
756 
757 	if ((sc->sc_flags & FSS_PERSISTENT) == 0)
758 		vn_cow_disestablish(sc->sc_mount_vp, fss_copy_on_write, sc);
759 
760 	FSS_LOCK(sc, s);
761 	sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR);
762 	sc->sc_mount = NULL;
763 	sc->sc_bdev = NODEV;
764 	FSS_UNLOCK(sc, s);
765 
766 	fss_softc_free(sc);
767 	if (sc->sc_flags & FSS_PERSISTENT)
768 		vn_close(sc->sc_bs_vp, FREAD, p->p_ucred, p);
769 	else
770 		vn_close(sc->sc_bs_vp, FREAD|FWRITE, p->p_ucred, p);
771 	sc->sc_bs_vp = NULL;
772 	sc->sc_flags &= ~FSS_PERSISTENT;
773 
774 	FSS_STAT_CLEAR(sc);
775 
776 	return 0;
777 }
778 
779 /*
780  * Get the block address and number of contiguous blocks.
781  * If the file contains a hole, try to allocate.
782  * Backing store is locked by caller.
783  */
784 static int
785 fss_bmap(struct fss_softc *sc, off_t start, int len,
786     struct vnode **vpp, daddr_t *bnp, int *runp)
787 {
788 	int l, s, error;
789 	struct buf *bp, **bpp;
790 
791 	if ((sc->sc_bs_vp->v_mount->mnt_flag & MNT_SOFTDEP) != 0)
792 		bpp = &bp;
793 	else
794 		bpp = NULL;
795 
796 	error = VOP_BMAP(sc->sc_bs_vp, FSS_BTOFSB(sc, start), vpp, bnp, runp);
797 	if ((error == 0 && *bnp != (daddr_t)-1) ||
798 	    (sc->sc_flags & FSS_BS_ALLOC) == 0)
799 		goto out;
800 
801 	if (start+len >= sc->sc_bs_size) {
802 		error = ENOSPC;
803 		goto out;
804 	}
805 
806 	for (l = 0; l < len; l += FSS_FSBSIZE(sc)) {
807 		error = VOP_BALLOC(sc->sc_bs_vp, start+l, FSS_FSBSIZE(sc),
808 		    sc->sc_bs_proc->p_ucred, 0, bpp);
809 		if (error)
810 			goto out;
811 
812 		if (bpp == NULL)
813 			continue;
814 
815 		s = splbio();
816 		simple_lock(&bp->b_interlock);
817 
818 		if (LIST_FIRST(&bp->b_dep) != NULL && bioops.io_start)
819 			(*bioops.io_start)(bp);
820 		if (LIST_FIRST(&bp->b_dep) != NULL && bioops.io_complete)
821 			(*bioops.io_complete)(bp);
822 
823 		bp->b_flags |= B_INVAL;
824 		simple_unlock(&bp->b_interlock);
825 		splx(s);
826 
827 		brelse(bp);
828 	}
829 
830 	error = VOP_BMAP(sc->sc_bs_vp, FSS_BTOFSB(sc, start), vpp, bnp, runp);
831 
832 out:
833 
834 	if ((sc->sc_flags & FSS_PERSISTENT) == 0 &&
835 	    error == 0 && *bnp == (daddr_t)-1)
836 		error = ENOSPC;
837 
838 	return error;
839 }
840 
841 /*
842  * A read from the snapshotted block device has completed.
843  */
844 static void
845 fss_cluster_iodone(struct buf *bp)
846 {
847 	int s;
848 	struct fss_cache *scp = bp->b_private;
849 
850 	FSS_LOCK(scp->fc_softc, s);
851 
852 	if (bp->b_flags & B_EINTR)
853 		fss_error(scp->fc_softc, "fs read interrupted");
854 	if (bp->b_flags & B_ERROR)
855 		fss_error(scp->fc_softc, "fs read error %d", bp->b_error);
856 
857 	if (bp->b_vp != NULL)
858 		brelvp(bp);
859 
860 	if (--scp->fc_xfercount == 0)
861 		wakeup(&scp->fc_data);
862 
863 	FSS_UNLOCK(scp->fc_softc, s);
864 
865 	s = splbio();
866 	pool_put(&bufpool, bp);
867 	splx(s);
868 }
869 
870 /*
871  * Read a cluster from the snapshotted block device to the cache.
872  */
873 static void
874 fss_read_cluster(struct fss_softc *sc, u_int32_t cl)
875 {
876 	int s, todo, len;
877 	caddr_t addr;
878 	daddr_t dblk;
879 	struct buf *bp;
880 	struct fss_cache *scp, *scl;
881 
882 	/*
883 	 * Get a free cache slot.
884 	 */
885 	scl = sc->sc_cache+sc->sc_cache_size;
886 
887 	FSS_LOCK(sc, s);
888 
889 restart:
890 	if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) {
891 		FSS_UNLOCK(sc, s);
892 		return;
893 	}
894 
895 	for (scp = sc->sc_cache; scp < scl; scp++)
896 		if (scp->fc_type != FSS_CACHE_FREE &&
897 		    scp->fc_cluster == cl) {
898 			ltsleep(&scp->fc_type, PRIBIO, "cowwait2", 0,
899 			    &sc->sc_slock);
900 			goto restart;
901 		}
902 
903 	for (scp = sc->sc_cache; scp < scl; scp++)
904 		if (scp->fc_type == FSS_CACHE_FREE) {
905 			scp->fc_type = FSS_CACHE_BUSY;
906 			scp->fc_cluster = cl;
907 			break;
908 		}
909 	if (scp >= scl) {
910 		FSS_STAT_INC(sc, cow_cache_full);
911 		ltsleep(&sc->sc_cache, PRIBIO, "cowwait3", 0, &sc->sc_slock);
912 		goto restart;
913 	}
914 
915 	FSS_UNLOCK(sc, s);
916 
917 	/*
918 	 * Start the read.
919 	 */
920 	FSS_STAT_INC(sc, cow_copied);
921 
922 	dblk = btodb(FSS_CLTOB(sc, cl));
923 	addr = scp->fc_data;
924 	if (cl == sc->sc_clcount-1) {
925 		todo = sc->sc_clresid;
926 		memset(addr+todo, 0, FSS_CLSIZE(sc)-todo);
927 	} else
928 		todo = FSS_CLSIZE(sc);
929 	while (todo > 0) {
930 		len = todo;
931 		if (len > MAXPHYS)
932 			len = MAXPHYS;
933 
934 		s = splbio();
935 		bp = pool_get(&bufpool, PR_WAITOK);
936 		splx(s);
937 
938 		BUF_INIT(bp);
939 		bp->b_flags = B_READ|B_CALL;
940 		bp->b_bcount = len;
941 		bp->b_bufsize = bp->b_bcount;
942 		bp->b_error = 0;
943 		bp->b_data = addr;
944 		bp->b_blkno = bp->b_rawblkno = dblk;
945 		bp->b_proc = NULL;
946 		bp->b_dev = sc->sc_bdev;
947 		bp->b_vp = NULLVP;
948 		bp->b_private = scp;
949 		bp->b_iodone = fss_cluster_iodone;
950 
951 		DEV_STRATEGY(bp);
952 
953 		FSS_LOCK(sc, s);
954 		scp->fc_xfercount++;
955 		FSS_UNLOCK(sc, s);
956 
957 		dblk += btodb(len);
958 		addr += len;
959 		todo -= len;
960 	}
961 
962 	/*
963 	 * Wait for all read requests to complete.
964 	 */
965 	FSS_LOCK(sc, s);
966 	while (scp->fc_xfercount > 0)
967 		ltsleep(&scp->fc_data, PRIBIO, "cowwait", 0, &sc->sc_slock);
968 
969 	scp->fc_type = FSS_CACHE_VALID;
970 	setbit(sc->sc_copied, scp->fc_cluster);
971 	FSS_UNLOCK(sc, s);
972 
973 	wakeup(&sc->sc_bs_proc);
974 }
975 
976 /*
977  * Write a cluster from the cache to the backing store.
978  */
979 static int
980 fss_write_cluster(struct fss_cache *scp, u_int32_t cl)
981 {
982 	int s, error, todo, len, nra;
983 	daddr_t nbn;
984 	caddr_t addr;
985 	off_t pos;
986 	struct buf *bp;
987 	struct vnode *vp;
988 	struct fss_softc *sc;
989 
990 	error = 0;
991 	sc = scp->fc_softc;
992 
993 	pos = FSS_CLTOB(sc, cl);
994 	addr = scp->fc_data;
995 	todo = FSS_CLSIZE(sc);
996 
997 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
998 	simple_lock(&sc->sc_bs_vp->v_interlock);
999 	error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(pos),
1000 	    round_page(pos+todo), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
1001 
1002 	while (error == 0 && todo > 0) {
1003 		error = fss_bmap(sc, pos, todo, &vp, &nbn, &nra);
1004 		if (error)
1005 			break;
1006 
1007 		len = FSS_FSBTOB(sc, nra+1)-FSS_FSBOFF(sc, pos);
1008 		if (len > todo)
1009 			len = todo;
1010 
1011 		s = splbio();
1012 		bp = pool_get(&bufpool, PR_WAITOK);
1013 		splx(s);
1014 
1015 		BUF_INIT(bp);
1016 		bp->b_flags = B_CALL;
1017 		bp->b_bcount = len;
1018 		bp->b_bufsize = bp->b_bcount;
1019 		bp->b_error = 0;
1020 		bp->b_data = addr;
1021 		bp->b_blkno = bp->b_rawblkno = nbn+btodb(FSS_FSBOFF(sc, pos));
1022 		bp->b_proc = NULL;
1023 		bp->b_vp = NULLVP;
1024 		bp->b_private = scp;
1025 		bp->b_iodone = fss_cluster_iodone;
1026 		bgetvp(vp, bp);
1027 		bp->b_vp->v_numoutput++;
1028 
1029 		BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
1030 		VOP_STRATEGY(vp, bp);
1031 
1032 		FSS_LOCK(sc, s);
1033 		scp->fc_xfercount++;
1034 		FSS_UNLOCK(sc, s);
1035 
1036 		pos += len;
1037 		addr += len;
1038 		todo -= len;
1039 	}
1040 
1041 	/*
1042 	 * Wait for all write requests to complete.
1043 	 */
1044 	FSS_LOCK(sc, s);
1045 	while (scp->fc_xfercount > 0)
1046 		ltsleep(&scp->fc_data, PRIBIO, "bswwait", 0, &sc->sc_slock);
1047 	FSS_UNLOCK(sc, s);
1048 
1049 	VOP_UNLOCK(sc->sc_bs_vp, 0);
1050 
1051 	return error;
1052 }
1053 
1054 /*
1055  * Read/write clusters from/to backing store.
1056  * For persistent snapshots must be called with cl == 0. off is the
1057  * offset into the snapshot.
1058  */
1059 static int
1060 fss_bs_io(struct fss_softc *sc, fss_io_type rw,
1061     u_int32_t cl, off_t off, int len, caddr_t data)
1062 {
1063 	int s, error, todo, count, nra;
1064 	off_t pos;
1065 	daddr_t nbn;
1066 	struct buf *bp;
1067 	struct vnode *vp;
1068 
1069 	todo = len;
1070 	pos = FSS_CLTOB(sc, cl)+off;
1071 	error = 0;
1072 
1073 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
1074 	simple_lock(&sc->sc_bs_vp->v_interlock);
1075 	error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(pos),
1076 	    round_page(pos+todo), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
1077 
1078 	while (error == 0 && todo > 0) {
1079 		error = fss_bmap(sc, pos, todo, &vp, &nbn, &nra);
1080 		if (error)
1081 			break;
1082 
1083 		count = FSS_FSBTOB(sc, nra+1)-FSS_FSBOFF(sc, pos);
1084 		if (count > todo)
1085 			count = todo;
1086 
1087 		if (nbn == (daddr_t)-1) {
1088 			bzero(data, count);
1089 			todo -= count;
1090 			data += count;
1091 			pos += count;
1092 			continue;
1093 		}
1094 
1095 		s = splbio();
1096 		bp = pool_get(&bufpool, PR_WAITOK);
1097 		splx(s);
1098 
1099 		BUF_INIT(bp);
1100 		bp->b_flags = (rw == FSS_READ ? B_READ : 0);
1101 		bp->b_bcount = count;
1102 		bp->b_bufsize = bp->b_bcount;
1103 		bp->b_error = 0;
1104 		bp->b_data = data;
1105 		bp->b_blkno = bp->b_rawblkno = nbn+btodb(FSS_FSBOFF(sc, pos));
1106 		bp->b_proc = NULL;
1107 		bp->b_vp = NULLVP;
1108 		bgetvp(vp, bp);
1109 		if ((bp->b_flags & B_READ) == 0)
1110 			bp->b_vp->v_numoutput++;
1111 
1112 		if ((bp->b_flags & B_READ) == 0 || cl < sc->sc_indir_size)
1113 			BIO_SETPRIO(bp, BPRIO_TIMECRITICAL);
1114 		VOP_STRATEGY(vp, bp);
1115 
1116 		error = biowait(bp);
1117 
1118 		if (bp->b_vp != NULL)
1119 			brelvp(bp);
1120 
1121 		s = splbio();
1122 		pool_put(&bufpool, bp);
1123 		splx(s);
1124 
1125 		if (error)
1126 			break;
1127 
1128 		todo -= count;
1129 		data += count;
1130 		pos += count;
1131 	}
1132 
1133 	VOP_UNLOCK(sc->sc_bs_vp, 0);
1134 
1135 	return error;
1136 }
1137 
1138 /*
1139  * Get a pointer to the indirect slot for this cluster.
1140  */
1141 static u_int32_t *
1142 fss_bs_indir(struct fss_softc *sc, u_int32_t cl)
1143 {
1144 	u_int32_t icl;
1145 	int ioff;
1146 
1147 	icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1148 	ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1149 
1150 	if (sc->sc_indir_cur == icl)
1151 		return &sc->sc_indir_data[ioff];
1152 
1153 	if (sc->sc_indir_dirty) {
1154 		FSS_STAT_INC(sc, indir_write);
1155 		if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0,
1156 		    FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0)
1157 			return NULL;
1158 		setbit(sc->sc_indir_valid, sc->sc_indir_cur);
1159 	}
1160 
1161 	sc->sc_indir_dirty = 0;
1162 	sc->sc_indir_cur = icl;
1163 
1164 	if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) {
1165 		FSS_STAT_INC(sc, indir_read);
1166 		if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0,
1167 		    FSS_CLSIZE(sc), (caddr_t)sc->sc_indir_data) != 0)
1168 			return NULL;
1169 	} else
1170 		memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc));
1171 
1172 	return &sc->sc_indir_data[ioff];
1173 }
1174 
1175 /*
1176  * The kernel thread (one for every active snapshot).
1177  *
1178  * After wakeup it cleans the cache and runs the I/O requests.
1179  */
1180 static void
1181 fss_bs_thread(void *arg)
1182 {
1183 	int error, len, nfreed, nio, s;
1184 	long off;
1185 	caddr_t addr;
1186 	u_int32_t c, cl, ch, *indirp;
1187 	struct buf *bp, *nbp;
1188 	struct fss_softc *sc;
1189 	struct fss_cache *scp, *scl;
1190 
1191 	sc = arg;
1192 
1193 	scl = sc->sc_cache+sc->sc_cache_size;
1194 
1195 	s = splbio();
1196 	nbp = pool_get(&bufpool, PR_WAITOK);
1197 	splx(s);
1198 
1199 	nfreed = nio = 1;		/* Dont sleep the first time */
1200 
1201 	FSS_LOCK(sc, s);
1202 
1203 	for (;;) {
1204 		if (nfreed == 0 && nio == 0)
1205 			ltsleep(&sc->sc_bs_proc, PVM-1, "fssbs", 0,
1206 			    &sc->sc_slock);
1207 
1208 		if ((sc->sc_flags & FSS_BS_THREAD) == 0) {
1209 			sc->sc_bs_proc = NULL;
1210 			wakeup(&sc->sc_bs_proc);
1211 
1212 			FSS_UNLOCK(sc, s);
1213 
1214 			s = splbio();
1215 			pool_put(&bufpool, nbp);
1216 			splx(s);
1217 #ifdef FSS_STATISTICS
1218 			if ((sc->sc_flags & FSS_PERSISTENT) == 0) {
1219 				printf("fss%d: cow called %" PRId64 " times,"
1220 				    " copied %" PRId64 " clusters,"
1221 				    " cache full %" PRId64 " times\n",
1222 				    sc->sc_unit,
1223 				    FSS_STAT_VAL(sc, cow_calls),
1224 				    FSS_STAT_VAL(sc, cow_copied),
1225 				    FSS_STAT_VAL(sc, cow_cache_full));
1226 				printf("fss%d: %" PRId64 " indir reads,"
1227 				    " %" PRId64 " indir writes\n",
1228 				    sc->sc_unit,
1229 				    FSS_STAT_VAL(sc, indir_read),
1230 				    FSS_STAT_VAL(sc, indir_write));
1231 			}
1232 #endif /* FSS_STATISTICS */
1233 			kthread_exit(0);
1234 		}
1235 
1236 		/*
1237 		 * Process I/O requests (persistent)
1238 		 */
1239 
1240 		if (sc->sc_flags & FSS_PERSISTENT) {
1241 			nfreed = nio = 0;
1242 
1243 			if ((bp = BUFQ_GET(&sc->sc_bufq)) == NULL)
1244 				continue;
1245 
1246 			nio++;
1247 
1248 			if (FSS_ISVALID(sc)) {
1249 				FSS_UNLOCK(sc, s);
1250 
1251 				error = fss_bs_io(sc, FSS_READ, 0,
1252 				    dbtob(bp->b_blkno), bp->b_bcount,
1253 				    bp->b_data);
1254 
1255 				FSS_LOCK(sc, s);
1256 			} else
1257 				error = ENXIO;
1258 
1259 			if (error) {
1260 				bp->b_error = error;
1261 				bp->b_flags |= B_ERROR;
1262 				bp->b_resid = bp->b_bcount;
1263 			}
1264 			biodone(bp);
1265 
1266 			continue;
1267 		}
1268 
1269 		/*
1270 		 * Clean the cache
1271 		 */
1272 		nfreed = 0;
1273 		for (scp = sc->sc_cache; scp < scl; scp++) {
1274 			if (scp->fc_type != FSS_CACHE_VALID)
1275 				continue;
1276 
1277 			FSS_UNLOCK(sc, s);
1278 
1279 			indirp = fss_bs_indir(sc, scp->fc_cluster);
1280 			if (indirp != NULL) {
1281 				error = fss_write_cluster(scp, sc->sc_clnext);
1282 			} else
1283 				error = EIO;
1284 
1285 			FSS_LOCK(sc, s);
1286 
1287 			if (error == 0) {
1288 				*indirp = sc->sc_clnext++;
1289 				sc->sc_indir_dirty = 1;
1290 			} else
1291 				fss_error(sc, "write bs error %d", error);
1292 
1293 			scp->fc_type = FSS_CACHE_FREE;
1294 			nfreed++;
1295 			wakeup(&scp->fc_type);
1296 		}
1297 
1298 		if (nfreed)
1299 			wakeup(&sc->sc_cache);
1300 
1301 		/*
1302 		 * Process I/O requests
1303 		 */
1304 		nio = 0;
1305 
1306 		if ((bp = BUFQ_GET(&sc->sc_bufq)) == NULL)
1307 			continue;
1308 
1309 		nio++;
1310 
1311 		if (!FSS_ISVALID(sc)) {
1312 			bp->b_error = ENXIO;
1313 			bp->b_flags |= B_ERROR;
1314 			bp->b_resid = bp->b_bcount;
1315 			biodone(bp);
1316 			continue;
1317 		}
1318 
1319 		/*
1320 		 * First read from the snapshotted block device.
1321 		 * XXX Split to only read those parts that have not
1322 		 * been saved to backing store?
1323 		 */
1324 
1325 		FSS_UNLOCK(sc, s);
1326 
1327 		BUF_INIT(nbp);
1328 		nbp->b_flags = B_READ;
1329 		nbp->b_bcount = bp->b_bcount;
1330 		nbp->b_bufsize = bp->b_bcount;
1331 		nbp->b_error = 0;
1332 		nbp->b_data = bp->b_data;
1333 		nbp->b_blkno = nbp->b_rawblkno = bp->b_blkno;
1334 		nbp->b_proc = bp->b_proc;
1335 		nbp->b_dev = sc->sc_bdev;
1336 		nbp->b_vp = NULLVP;
1337 
1338 		DEV_STRATEGY(nbp);
1339 
1340 		if (biowait(nbp) != 0) {
1341 			bp->b_resid = bp->b_bcount;
1342 			bp->b_error = nbp->b_error;
1343 			bp->b_flags |= B_ERROR;
1344 			biodone(bp);
1345 			continue;
1346 		}
1347 
1348 		cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
1349 		off = FSS_CLOFF(sc, dbtob(bp->b_blkno));
1350 		ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
1351 		bp->b_resid = bp->b_bcount;
1352 		addr = bp->b_data;
1353 
1354 		FSS_LOCK(sc, s);
1355 
1356 		/*
1357 		 * Replace those parts that have been saved to backing store.
1358 		 */
1359 
1360 		for (c = cl; c <= ch;
1361 		    c++, off = 0, bp->b_resid -= len, addr += len) {
1362 			len = FSS_CLSIZE(sc)-off;
1363 			if (len > bp->b_resid)
1364 				len = bp->b_resid;
1365 
1366 			if (isclr(sc->sc_copied, c))
1367 				continue;
1368 
1369 			FSS_UNLOCK(sc, s);
1370 
1371 			indirp = fss_bs_indir(sc, c);
1372 
1373 			FSS_LOCK(sc, s);
1374 
1375 			if (indirp == NULL || *indirp == 0) {
1376 				/*
1377 				 * Not on backing store. Either in cache
1378 				 * or hole in the snapshotted block device.
1379 				 */
1380 				for (scp = sc->sc_cache; scp < scl; scp++)
1381 					if (scp->fc_type == FSS_CACHE_VALID &&
1382 					    scp->fc_cluster == c)
1383 						break;
1384 				if (scp < scl)
1385 					memcpy(addr, scp->fc_data+off, len);
1386 				else
1387 					memset(addr, 0, len);
1388 				continue;
1389 			}
1390 			/*
1391 			 * Read from backing store.
1392 			 */
1393 
1394 			FSS_UNLOCK(sc, s);
1395 
1396 			if ((error = fss_bs_io(sc, FSS_READ, *indirp,
1397 			    off, len, addr)) != 0) {
1398 				bp->b_resid = bp->b_bcount;
1399 				bp->b_error = error;
1400 				bp->b_flags |= B_ERROR;
1401 				break;
1402 			}
1403 
1404 			FSS_LOCK(sc, s);
1405 
1406 		}
1407 
1408 		biodone(bp);
1409 	}
1410 }
1411