xref: /netbsd-src/sys/dev/fss.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: fss.c,v 1.104 2018/01/23 22:42:29 pgoyette 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  *
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 /*
33  * File system snapshot disk driver.
34  *
35  * Block/character interface to the snapshot of a mounted file system.
36  */
37 
38 #include <sys/cdefs.h>
39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.104 2018/01/23 22:42:29 pgoyette Exp $");
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/namei.h>
44 #include <sys/proc.h>
45 #include <sys/errno.h>
46 #include <sys/malloc.h>
47 #include <sys/buf.h>
48 #include <sys/ioctl.h>
49 #include <sys/disklabel.h>
50 #include <sys/device.h>
51 #include <sys/disk.h>
52 #include <sys/stat.h>
53 #include <sys/mount.h>
54 #include <sys/vnode.h>
55 #include <sys/file.h>
56 #include <sys/uio.h>
57 #include <sys/conf.h>
58 #include <sys/kthread.h>
59 #include <sys/fstrans.h>
60 #include <sys/vfs_syscalls.h>		/* For do_sys_unlink(). */
61 
62 #include <miscfs/specfs/specdev.h>
63 
64 #include <dev/fssvar.h>
65 
66 #include <uvm/uvm.h>
67 
68 #include "ioconf.h"
69 
70 dev_type_open(fss_open);
71 dev_type_close(fss_close);
72 dev_type_read(fss_read);
73 dev_type_write(fss_write);
74 dev_type_ioctl(fss_ioctl);
75 dev_type_strategy(fss_strategy);
76 dev_type_dump(fss_dump);
77 dev_type_size(fss_size);
78 
79 static void fss_unmount_hook(struct mount *);
80 static int fss_copy_on_write(void *, struct buf *, bool);
81 static inline void fss_error(struct fss_softc *, const char *);
82 static int fss_create_files(struct fss_softc *, struct fss_set *,
83     off_t *, struct lwp *);
84 static int fss_create_snapshot(struct fss_softc *, struct fss_set *,
85     struct lwp *);
86 static int fss_delete_snapshot(struct fss_softc *, struct lwp *);
87 static int fss_softc_alloc(struct fss_softc *);
88 static void fss_softc_free(struct fss_softc *);
89 static int fss_read_cluster(struct fss_softc *, u_int32_t);
90 static void fss_bs_thread(void *);
91 static int fss_bs_io(struct fss_softc *, fss_io_type,
92     u_int32_t, off_t, int, void *, size_t *);
93 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t);
94 
95 static kmutex_t fss_device_lock;	/* Protect all units. */
96 static int fss_num_attached = 0;	/* Number of attached devices. */
97 static struct vfs_hooks fss_vfs_hooks = {
98 	.vh_unmount = fss_unmount_hook
99 };
100 
101 const struct bdevsw fss_bdevsw = {
102 	.d_open = fss_open,
103 	.d_close = fss_close,
104 	.d_strategy = fss_strategy,
105 	.d_ioctl = fss_ioctl,
106 	.d_dump = fss_dump,
107 	.d_psize = fss_size,
108 	.d_discard = nodiscard,
109 	.d_flag = D_DISK | D_MPSAFE
110 };
111 
112 const struct cdevsw fss_cdevsw = {
113 	.d_open = fss_open,
114 	.d_close = fss_close,
115 	.d_read = fss_read,
116 	.d_write = fss_write,
117 	.d_ioctl = fss_ioctl,
118 	.d_stop = nostop,
119 	.d_tty = notty,
120 	.d_poll = nopoll,
121 	.d_mmap = nommap,
122 	.d_kqfilter = nokqfilter,
123 	.d_discard = nodiscard,
124 	.d_flag = D_DISK | D_MPSAFE
125 };
126 
127 static int fss_match(device_t, cfdata_t, void *);
128 static void fss_attach(device_t, device_t, void *);
129 static int fss_detach(device_t, int);
130 
131 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc),
132     fss_match, fss_attach, fss_detach, NULL);
133 
134 void
135 fssattach(int num)
136 {
137 
138 	mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE);
139 	if (config_cfattach_attach(fss_cd.cd_name, &fss_ca))
140 		aprint_error("%s: unable to register\n", fss_cd.cd_name);
141 }
142 
143 static int
144 fss_match(device_t self, cfdata_t cfdata, void *aux)
145 {
146 	return 1;
147 }
148 
149 static void
150 fss_attach(device_t parent, device_t self, void *aux)
151 {
152 	struct fss_softc *sc = device_private(self);
153 
154 	sc->sc_dev = self;
155 	sc->sc_bdev = NODEV;
156 	mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE);
157 	mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE);
158 	cv_init(&sc->sc_work_cv, "fssbs");
159 	cv_init(&sc->sc_cache_cv, "cowwait");
160 	bufq_alloc(&sc->sc_bufq, "fcfs", 0);
161 	sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK);
162 	sc->sc_dkdev->dk_info = NULL;
163 	disk_init(sc->sc_dkdev, device_xname(self), NULL);
164 	if (!pmf_device_register(self, NULL, NULL))
165 		aprint_error_dev(self, "couldn't establish power handler\n");
166 
167 	if (fss_num_attached++ == 0)
168 		vfs_hooks_attach(&fss_vfs_hooks);
169 }
170 
171 static int
172 fss_detach(device_t self, int flags)
173 {
174 	struct fss_softc *sc = device_private(self);
175 
176 	if (sc->sc_flags & FSS_ACTIVE)
177 		return EBUSY;
178 
179 	if (--fss_num_attached == 0)
180 		vfs_hooks_detach(&fss_vfs_hooks);
181 
182 	pmf_device_deregister(self);
183 	mutex_destroy(&sc->sc_slock);
184 	mutex_destroy(&sc->sc_lock);
185 	cv_destroy(&sc->sc_work_cv);
186 	cv_destroy(&sc->sc_cache_cv);
187 	bufq_drain(sc->sc_bufq);
188 	bufq_free(sc->sc_bufq);
189 	disk_destroy(sc->sc_dkdev);
190 	free(sc->sc_dkdev, M_DEVBUF);
191 
192 	return 0;
193 }
194 
195 int
196 fss_open(dev_t dev, int flags, int mode, struct lwp *l)
197 {
198 	int mflag;
199 	cfdata_t cf;
200 	struct fss_softc *sc;
201 
202 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
203 
204 	mutex_enter(&fss_device_lock);
205 
206 	sc = device_lookup_private(&fss_cd, minor(dev));
207 	if (sc == NULL) {
208 		cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
209 		cf->cf_name = fss_cd.cd_name;
210 		cf->cf_atname = fss_cd.cd_name;
211 		cf->cf_unit = minor(dev);
212 		cf->cf_fstate = FSTATE_STAR;
213 		sc = device_private(config_attach_pseudo(cf));
214 		if (sc == NULL) {
215 			mutex_exit(&fss_device_lock);
216 			return ENOMEM;
217 		}
218 	}
219 
220 	mutex_enter(&sc->sc_slock);
221 
222 	sc->sc_flags |= mflag;
223 
224 	mutex_exit(&sc->sc_slock);
225 	mutex_exit(&fss_device_lock);
226 
227 	return 0;
228 }
229 
230 int
231 fss_close(dev_t dev, int flags, int mode, struct lwp *l)
232 {
233 	int mflag, error;
234 	cfdata_t cf;
235 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev));
236 
237 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
238 	error = 0;
239 
240 	mutex_enter(&fss_device_lock);
241 restart:
242 	mutex_enter(&sc->sc_slock);
243 	if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) {
244 		sc->sc_flags &= ~mflag;
245 		mutex_exit(&sc->sc_slock);
246 		mutex_exit(&fss_device_lock);
247 		return 0;
248 	}
249 	if ((sc->sc_flags & FSS_ACTIVE) != 0 &&
250 	    (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) {
251 		sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE;
252 		mutex_exit(&sc->sc_slock);
253 		error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l);
254 		goto restart;
255 	}
256 	if ((sc->sc_flags & FSS_ACTIVE) != 0) {
257 		mutex_exit(&sc->sc_slock);
258 		mutex_exit(&fss_device_lock);
259 		return error;
260 	}
261 
262 	KASSERT((sc->sc_flags & FSS_ACTIVE) == 0);
263 	KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag);
264 	mutex_exit(&sc->sc_slock);
265 	cf = device_cfdata(sc->sc_dev);
266 	error = config_detach(sc->sc_dev, DETACH_QUIET);
267 	if (! error)
268 		free(cf, M_DEVBUF);
269 	mutex_exit(&fss_device_lock);
270 
271 	return error;
272 }
273 
274 void
275 fss_strategy(struct buf *bp)
276 {
277 	const bool write = ((bp->b_flags & B_READ) != B_READ);
278 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev));
279 
280 	mutex_enter(&sc->sc_slock);
281 
282 	if (write || !FSS_ISVALID(sc)) {
283 		bp->b_error = (write ? EROFS : ENXIO);
284 		goto done;
285 	}
286 	/* Check bounds for non-persistent snapshots. */
287 	if ((sc->sc_flags & FSS_PERSISTENT) == 0 &&
288 	    bounds_check_with_mediasize(bp, DEV_BSIZE,
289 	    btodb(FSS_CLTOB(sc, sc->sc_clcount - 1) + sc->sc_clresid)) <= 0)
290 		goto done;
291 
292 	bp->b_rawblkno = bp->b_blkno;
293 	bufq_put(sc->sc_bufq, bp);
294 	cv_signal(&sc->sc_work_cv);
295 
296 	mutex_exit(&sc->sc_slock);
297 	return;
298 
299 done:
300 	mutex_exit(&sc->sc_slock);
301 	bp->b_resid = bp->b_bcount;
302 	biodone(bp);
303 }
304 
305 int
306 fss_read(dev_t dev, struct uio *uio, int flags)
307 {
308 	return physio(fss_strategy, NULL, dev, B_READ, minphys, uio);
309 }
310 
311 int
312 fss_write(dev_t dev, struct uio *uio, int flags)
313 {
314 	return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio);
315 }
316 
317 int
318 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
319 {
320 	int error;
321 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev));
322 	struct fss_set _fss;
323 	struct fss_set *fss = (struct fss_set *)data;
324 	struct fss_set50 *fss50 = (struct fss_set50 *)data;
325 	struct fss_get *fsg = (struct fss_get *)data;
326 #ifndef _LP64
327 	struct fss_get50 *fsg50 = (struct fss_get50 *)data;
328 #endif
329 
330 	switch (cmd) {
331 	case FSSIOCSET50:
332 		fss = &_fss;
333 		fss->fss_mount = fss50->fss_mount;
334 		fss->fss_bstore = fss50->fss_bstore;
335 		fss->fss_csize = fss50->fss_csize;
336 		fss->fss_flags = 0;
337 		/* Fall through */
338 	case FSSIOCSET:
339 		mutex_enter(&sc->sc_lock);
340 		if ((flag & FWRITE) == 0)
341 			error = EPERM;
342 		else if ((sc->sc_flags & FSS_ACTIVE) != 0)
343 			error = EBUSY;
344 		else
345 			error = fss_create_snapshot(sc, fss, l);
346 		if (error == 0)
347 			sc->sc_uflags = fss->fss_flags;
348 		mutex_exit(&sc->sc_lock);
349 		break;
350 
351 	case FSSIOCCLR:
352 		mutex_enter(&sc->sc_lock);
353 		if ((flag & FWRITE) == 0)
354 			error = EPERM;
355 		else if ((sc->sc_flags & FSS_ACTIVE) == 0)
356 			error = ENXIO;
357 		else
358 			error = fss_delete_snapshot(sc, l);
359 		mutex_exit(&sc->sc_lock);
360 		break;
361 
362 #ifndef _LP64
363 	case FSSIOCGET50:
364 		mutex_enter(&sc->sc_lock);
365 		switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) {
366 		case FSS_ACTIVE:
367 			memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN);
368 			fsg50->fsg_csize = FSS_CLSIZE(sc);
369 			timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time);
370 			fsg50->fsg_mount_size = sc->sc_clcount;
371 			fsg50->fsg_bs_size = sc->sc_clnext;
372 			error = 0;
373 			break;
374 		case FSS_PERSISTENT | FSS_ACTIVE:
375 			memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN);
376 			fsg50->fsg_csize = 0;
377 			timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time);
378 			fsg50->fsg_mount_size = 0;
379 			fsg50->fsg_bs_size = 0;
380 			error = 0;
381 			break;
382 		default:
383 			error = ENXIO;
384 			break;
385 		}
386 		mutex_exit(&sc->sc_lock);
387 		break;
388 #endif /* _LP64 */
389 
390 	case FSSIOCGET:
391 		mutex_enter(&sc->sc_lock);
392 		switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) {
393 		case FSS_ACTIVE:
394 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
395 			fsg->fsg_csize = FSS_CLSIZE(sc);
396 			fsg->fsg_time = sc->sc_time;
397 			fsg->fsg_mount_size = sc->sc_clcount;
398 			fsg->fsg_bs_size = sc->sc_clnext;
399 			error = 0;
400 			break;
401 		case FSS_PERSISTENT | FSS_ACTIVE:
402 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
403 			fsg->fsg_csize = 0;
404 			fsg->fsg_time = sc->sc_time;
405 			fsg->fsg_mount_size = 0;
406 			fsg->fsg_bs_size = 0;
407 			error = 0;
408 			break;
409 		default:
410 			error = ENXIO;
411 			break;
412 		}
413 		mutex_exit(&sc->sc_lock);
414 		break;
415 
416 	case FSSIOFSET:
417 		mutex_enter(&sc->sc_slock);
418 		sc->sc_uflags = *(int *)data;
419 		mutex_exit(&sc->sc_slock);
420 		error = 0;
421 		break;
422 
423 	case FSSIOFGET:
424 		mutex_enter(&sc->sc_slock);
425 		*(int *)data = sc->sc_uflags;
426 		mutex_exit(&sc->sc_slock);
427 		error = 0;
428 		break;
429 
430 	default:
431 		error = EINVAL;
432 		break;
433 	}
434 
435 	return error;
436 }
437 
438 int
439 fss_size(dev_t dev)
440 {
441 	return -1;
442 }
443 
444 int
445 fss_dump(dev_t dev, daddr_t blkno, void *va,
446     size_t size)
447 {
448 	return EROFS;
449 }
450 
451 /*
452  * An error occurred reading or writing the snapshot or backing store.
453  * If it is the first error log to console and disestablish cow handler.
454  * The caller holds the mutex.
455  */
456 static inline void
457 fss_error(struct fss_softc *sc, const char *msg)
458 {
459 
460 	if ((sc->sc_flags & (FSS_ACTIVE | FSS_ERROR)) != FSS_ACTIVE)
461 		return;
462 
463 	aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg);
464 	if ((sc->sc_flags & FSS_PERSISTENT) == 0)
465 		fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc);
466 	sc->sc_flags |= FSS_ERROR;
467 }
468 
469 /*
470  * Allocate the variable sized parts of the softc and
471  * fork the kernel thread.
472  *
473  * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size
474  * must be initialized.
475  */
476 static int
477 fss_softc_alloc(struct fss_softc *sc)
478 {
479 	int i, error;
480 
481 	if ((sc->sc_flags & FSS_PERSISTENT) == 0) {
482 		sc->sc_copied =
483 		    kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP);
484 		sc->sc_cache = kmem_alloc(sc->sc_cache_size *
485 		    sizeof(struct fss_cache), KM_SLEEP);
486 		for (i = 0; i < sc->sc_cache_size; i++) {
487 			sc->sc_cache[i].fc_type = FSS_CACHE_FREE;
488 			sc->sc_cache[i].fc_data =
489 			    kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP);
490 			cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1");
491 		}
492 
493 		sc->sc_indir_valid =
494 		    kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP);
495 		sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP);
496 	} else {
497 		sc->sc_copied = NULL;
498 		sc->sc_cache = NULL;
499 		sc->sc_indir_valid = NULL;
500 		sc->sc_indir_data = NULL;
501 	}
502 
503 	sc->sc_flags |= FSS_BS_THREAD;
504 	if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL,
505 	    fss_bs_thread, sc, &sc->sc_bs_lwp,
506 	    "%s", device_xname(sc->sc_dev))) != 0) {
507 		sc->sc_flags &= ~FSS_BS_THREAD;
508 		return error;
509 	}
510 
511 	disk_attach(sc->sc_dkdev);
512 
513 	return 0;
514 }
515 
516 /*
517  * Free the variable sized parts of the softc.
518  */
519 static void
520 fss_softc_free(struct fss_softc *sc)
521 {
522 	int i;
523 
524 	if ((sc->sc_flags & FSS_BS_THREAD) != 0) {
525 		mutex_enter(&sc->sc_slock);
526 		sc->sc_flags &= ~FSS_BS_THREAD;
527 		cv_signal(&sc->sc_work_cv);
528 		mutex_exit(&sc->sc_slock);
529 		kthread_join(sc->sc_bs_lwp);
530 
531 		disk_detach(sc->sc_dkdev);
532 	}
533 
534 	if (sc->sc_copied != NULL)
535 		kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY));
536 	sc->sc_copied = NULL;
537 
538 	if (sc->sc_cache != NULL) {
539 		for (i = 0; i < sc->sc_cache_size; i++)
540 			if (sc->sc_cache[i].fc_data != NULL) {
541 				cv_destroy(&sc->sc_cache[i].fc_state_cv);
542 				kmem_free(sc->sc_cache[i].fc_data,
543 				    FSS_CLSIZE(sc));
544 			}
545 		kmem_free(sc->sc_cache,
546 		    sc->sc_cache_size*sizeof(struct fss_cache));
547 	}
548 	sc->sc_cache = NULL;
549 
550 	if (sc->sc_indir_valid != NULL)
551 		kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY));
552 	sc->sc_indir_valid = NULL;
553 
554 	if (sc->sc_indir_data != NULL)
555 		kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc));
556 	sc->sc_indir_data = NULL;
557 }
558 
559 /*
560  * Set all active snapshots on this file system into ERROR state.
561  */
562 static void
563 fss_unmount_hook(struct mount *mp)
564 {
565 	int i;
566 	struct fss_softc *sc;
567 
568 	mutex_enter(&fss_device_lock);
569 	for (i = 0; i < fss_cd.cd_ndevs; i++) {
570 		if ((sc = device_lookup_private(&fss_cd, i)) == NULL)
571 			continue;
572 		mutex_enter(&sc->sc_slock);
573 		if ((sc->sc_flags & FSS_ACTIVE) != 0 && sc->sc_mount == mp)
574 			fss_error(sc, "forced by unmount");
575 		mutex_exit(&sc->sc_slock);
576 	}
577 	mutex_exit(&fss_device_lock);
578 }
579 
580 /*
581  * A buffer is written to the snapshotted block device. Copy to
582  * backing store if needed.
583  */
584 static int
585 fss_copy_on_write(void *v, struct buf *bp, bool data_valid)
586 {
587 	int error;
588 	u_int32_t cl, ch, c;
589 	struct fss_softc *sc = v;
590 
591 	mutex_enter(&sc->sc_slock);
592 	if (!FSS_ISVALID(sc)) {
593 		mutex_exit(&sc->sc_slock);
594 		return 0;
595 	}
596 
597 	cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
598 	ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
599 	error = 0;
600 	if (curlwp == uvm.pagedaemon_lwp) {
601 		for (c = cl; c <= ch; c++)
602 			if (isclr(sc->sc_copied, c)) {
603 				error = ENOMEM;
604 				break;
605 			}
606 	}
607 	mutex_exit(&sc->sc_slock);
608 
609 	if (error == 0)
610 		for (c = cl; c <= ch; c++) {
611 			error = fss_read_cluster(sc, c);
612 			if (error)
613 				break;
614 		}
615 
616 	return error;
617 }
618 
619 /*
620  * Lookup and open needed files.
621  *
622  * For file system internal snapshot initializes sc_mntname, sc_mount,
623  * sc_bs_vp and sc_time.
624  *
625  * Otherwise returns dev and size of the underlying block device.
626  * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount
627  */
628 static int
629 fss_create_files(struct fss_softc *sc, struct fss_set *fss,
630     off_t *bsize, struct lwp *l)
631 {
632 	int error, bits, fsbsize;
633 	uint64_t numsec;
634 	unsigned int secsize;
635 	struct timespec ts;
636 	/* nd -> nd2 to reduce mistakes while updating only some namei calls */
637 	struct pathbuf *pb2;
638 	struct nameidata nd2;
639 	struct vnode *vp;
640 
641 	/*
642 	 * Get the mounted file system.
643 	 */
644 
645 	error = namei_simple_user(fss->fss_mount,
646 				NSM_FOLLOW_NOEMULROOT, &vp);
647 	if (error != 0)
648 		return error;
649 
650 	if ((vp->v_vflag & VV_ROOT) != VV_ROOT) {
651 		vrele(vp);
652 		return EINVAL;
653 	}
654 
655 	sc->sc_mount = vp->v_mount;
656 	memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN);
657 
658 	vrele(vp);
659 
660 	/*
661 	 * Check for file system internal snapshot.
662 	 */
663 
664 	error = namei_simple_user(fss->fss_bstore,
665 				NSM_FOLLOW_NOEMULROOT, &vp);
666 	if (error != 0)
667 		return error;
668 
669 	if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) {
670 		sc->sc_flags |= FSS_PERSISTENT;
671 		sc->sc_bs_vp = vp;
672 
673 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
674 		bits = sizeof(sc->sc_bs_bshift)*NBBY;
675 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits;
676 		    sc->sc_bs_bshift++)
677 			if (FSS_FSBSIZE(sc) == fsbsize)
678 				break;
679 		if (sc->sc_bs_bshift >= bits)
680 			return EINVAL;
681 
682 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
683 		sc->sc_clshift = 0;
684 
685 		if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) {
686 			error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE);
687 			if (error)
688 				return error;
689 		}
690 		error = vn_lock(vp, LK_EXCLUSIVE);
691 		if (error != 0)
692 			return error;
693 		error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
694 		TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
695 
696 		VOP_UNLOCK(sc->sc_bs_vp);
697 
698 		return error;
699 	}
700 	vrele(vp);
701 
702 	/*
703 	 * Get the block device it is mounted on and its size.
704 	 */
705 
706 	error = spec_node_lookup_by_mount(sc->sc_mount, &vp);
707 	if (error)
708 		return error;
709 	sc->sc_bdev = vp->v_rdev;
710 
711 	error = getdisksize(vp, &numsec, &secsize);
712 	vrele(vp);
713 	if (error)
714 		return error;
715 
716 	*bsize = (off_t)numsec*secsize;
717 
718 	/*
719 	 * Get the backing store
720 	 */
721 
722 	error = pathbuf_copyin(fss->fss_bstore, &pb2);
723 	if (error) {
724  		return error;
725 	}
726 	NDINIT(&nd2, LOOKUP, FOLLOW, pb2);
727 	if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) {
728 		pathbuf_destroy(pb2);
729 		return error;
730 	}
731 	VOP_UNLOCK(nd2.ni_vp);
732 
733 	sc->sc_bs_vp = nd2.ni_vp;
734 
735 	if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) {
736 		pathbuf_destroy(pb2);
737 		return EINVAL;
738 	}
739 	pathbuf_destroy(pb2);
740 
741 	if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) {
742 		error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE);
743 		if (error)
744 			return error;
745 	}
746 	if (sc->sc_bs_vp->v_type == VREG) {
747 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
748 		if (fsbsize & (fsbsize-1))	/* No power of two */
749 			return EINVAL;
750 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
751 		    sc->sc_bs_bshift++)
752 			if (FSS_FSBSIZE(sc) == fsbsize)
753 				break;
754 		if (sc->sc_bs_bshift >= 32)
755 			return EINVAL;
756 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
757 	} else {
758 		sc->sc_bs_bshift = DEV_BSHIFT;
759 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
760 	}
761 
762 	return 0;
763 }
764 
765 /*
766  * Create a snapshot.
767  */
768 static int
769 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l)
770 {
771 	int len, error;
772 	u_int32_t csize;
773 	off_t bsize;
774 
775 	bsize = 0;	/* XXX gcc */
776 
777 	/*
778 	 * Open needed files.
779 	 */
780 	if ((error = fss_create_files(sc, fss, &bsize, l)) != 0)
781 		goto bad;
782 
783 	if (sc->sc_flags & FSS_PERSISTENT) {
784 		fss_softc_alloc(sc);
785 		sc->sc_flags |= FSS_ACTIVE;
786 		return 0;
787 	}
788 
789 	/*
790 	 * Set cluster size. Must be a power of two and
791 	 * a multiple of backing store block size.
792 	 */
793 	if (fss->fss_csize <= 0)
794 		csize = MAXPHYS;
795 	else
796 		csize = fss->fss_csize;
797 	if (bsize/csize > FSS_CLUSTER_MAX)
798 		csize = bsize/FSS_CLUSTER_MAX+1;
799 
800 	for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32;
801 	    sc->sc_clshift++)
802 		if (FSS_CLSIZE(sc) >= csize)
803 			break;
804 	if (sc->sc_clshift >= 32) {
805 		error = EINVAL;
806 		goto bad;
807 	}
808 	sc->sc_clmask = FSS_CLSIZE(sc)-1;
809 
810 	/*
811 	 * Set number of cache slots.
812 	 */
813 	if (FSS_CLSIZE(sc) <= 8192)
814 		sc->sc_cache_size = 32;
815 	else if (FSS_CLSIZE(sc) <= 65536)
816 		sc->sc_cache_size = 8;
817 	else
818 		sc->sc_cache_size = 4;
819 
820 	/*
821 	 * Set number of clusters and size of last cluster.
822 	 */
823 	sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1;
824 	sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1;
825 
826 	/*
827 	 * Set size of indirect table.
828 	 */
829 	len = sc->sc_clcount*sizeof(u_int32_t);
830 	sc->sc_indir_size = FSS_BTOCL(sc, len)+1;
831 	sc->sc_clnext = sc->sc_indir_size;
832 	sc->sc_indir_cur = 0;
833 
834 	if ((error = fss_softc_alloc(sc)) != 0)
835 		goto bad;
836 
837 	/*
838 	 * Activate the snapshot.
839 	 */
840 
841 	if ((error = vfs_suspend(sc->sc_mount, 0)) != 0)
842 		goto bad;
843 
844 	microtime(&sc->sc_time);
845 
846 	vrele_flush(sc->sc_mount);
847 	error = VFS_SYNC(sc->sc_mount, MNT_WAIT, curlwp->l_cred);
848 	if (error == 0)
849 		error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc);
850 	if (error == 0)
851 		sc->sc_flags |= FSS_ACTIVE;
852 
853 	vfs_resume(sc->sc_mount);
854 
855 	if (error != 0)
856 		goto bad;
857 
858 	aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname);
859 	aprint_debug_dev(sc->sc_dev,
860 	    "%u clusters of %u, %u cache slots, %u indir clusters\n",
861 	    sc->sc_clcount, FSS_CLSIZE(sc),
862 	    sc->sc_cache_size, sc->sc_indir_size);
863 
864 	return 0;
865 
866 bad:
867 	fss_softc_free(sc);
868 	if (sc->sc_bs_vp != NULL) {
869 		if (sc->sc_flags & FSS_PERSISTENT)
870 			vrele(sc->sc_bs_vp);
871 		else
872 			vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
873 	}
874 	sc->sc_bs_vp = NULL;
875 
876 	return error;
877 }
878 
879 /*
880  * Delete a snapshot.
881  */
882 static int
883 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l)
884 {
885 
886 	if ((sc->sc_flags & (FSS_PERSISTENT | FSS_ERROR)) == 0)
887 		fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc);
888 
889 	mutex_enter(&sc->sc_slock);
890 	sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR);
891 	sc->sc_mount = NULL;
892 	sc->sc_bdev = NODEV;
893 	mutex_exit(&sc->sc_slock);
894 
895 	fss_softc_free(sc);
896 	if (sc->sc_flags & FSS_PERSISTENT)
897 		vrele(sc->sc_bs_vp);
898 	else
899 		vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
900 	sc->sc_bs_vp = NULL;
901 	sc->sc_flags &= ~FSS_PERSISTENT;
902 
903 	return 0;
904 }
905 
906 /*
907  * Read a cluster from the snapshotted block device to the cache.
908  */
909 static int
910 fss_read_cluster(struct fss_softc *sc, u_int32_t cl)
911 {
912 	int error, todo, offset, len;
913 	daddr_t dblk;
914 	struct buf *bp, *mbp;
915 	struct fss_cache *scp, *scl;
916 
917 	/*
918 	 * Get a free cache slot.
919 	 */
920 	scl = sc->sc_cache+sc->sc_cache_size;
921 
922 	mutex_enter(&sc->sc_slock);
923 
924 restart:
925 	if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) {
926 		mutex_exit(&sc->sc_slock);
927 		return 0;
928 	}
929 
930 	for (scp = sc->sc_cache; scp < scl; scp++)
931 		if (scp->fc_cluster == cl) {
932 			if (scp->fc_type == FSS_CACHE_VALID) {
933 				mutex_exit(&sc->sc_slock);
934 				return 0;
935 			} else if (scp->fc_type == FSS_CACHE_BUSY) {
936 				cv_wait(&scp->fc_state_cv, &sc->sc_slock);
937 				goto restart;
938 			}
939 		}
940 
941 	for (scp = sc->sc_cache; scp < scl; scp++)
942 		if (scp->fc_type == FSS_CACHE_FREE) {
943 			scp->fc_type = FSS_CACHE_BUSY;
944 			scp->fc_cluster = cl;
945 			break;
946 		}
947 	if (scp >= scl) {
948 		cv_wait(&sc->sc_cache_cv, &sc->sc_slock);
949 		goto restart;
950 	}
951 
952 	mutex_exit(&sc->sc_slock);
953 
954 	/*
955 	 * Start the read.
956 	 */
957 	dblk = btodb(FSS_CLTOB(sc, cl));
958 	if (cl == sc->sc_clcount-1) {
959 		todo = sc->sc_clresid;
960 		memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo);
961 	} else
962 		todo = FSS_CLSIZE(sc);
963 	offset = 0;
964 	mbp = getiobuf(NULL, true);
965 	mbp->b_bufsize = todo;
966 	mbp->b_data = scp->fc_data;
967 	mbp->b_resid = mbp->b_bcount = todo;
968 	mbp->b_flags = B_READ;
969 	mbp->b_cflags = BC_BUSY;
970 	mbp->b_dev = sc->sc_bdev;
971 	while (todo > 0) {
972 		len = todo;
973 		if (len > MAXPHYS)
974 			len = MAXPHYS;
975 		if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo)
976 			bp = mbp;
977 		else {
978 			bp = getiobuf(NULL, true);
979 			nestiobuf_setup(mbp, bp, offset, len);
980 		}
981 		bp->b_lblkno = 0;
982 		bp->b_blkno = dblk;
983 		bdev_strategy(bp);
984 		dblk += btodb(len);
985 		offset += len;
986 		todo -= len;
987 	}
988 	error = biowait(mbp);
989 	if (error == 0 && mbp->b_resid != 0)
990 		error = EIO;
991 	putiobuf(mbp);
992 
993 	mutex_enter(&sc->sc_slock);
994 	scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID);
995 	cv_broadcast(&scp->fc_state_cv);
996 	if (error == 0) {
997 		setbit(sc->sc_copied, scp->fc_cluster);
998 		cv_signal(&sc->sc_work_cv);
999 	}
1000 	mutex_exit(&sc->sc_slock);
1001 
1002 	return error;
1003 }
1004 
1005 /*
1006  * Read/write clusters from/to backing store.
1007  * For persistent snapshots must be called with cl == 0. off is the
1008  * offset into the snapshot.
1009  */
1010 static int
1011 fss_bs_io(struct fss_softc *sc, fss_io_type rw,
1012     u_int32_t cl, off_t off, int len, void *data, size_t *resid)
1013 {
1014 	int error;
1015 
1016 	off += FSS_CLTOB(sc, cl);
1017 
1018 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
1019 
1020 	error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp,
1021 	    data, len, off, UIO_SYSSPACE,
1022 	    IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED,
1023 	    sc->sc_bs_lwp->l_cred, resid, NULL);
1024 	if (error == 0) {
1025 		mutex_enter(sc->sc_bs_vp->v_interlock);
1026 		error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off),
1027 		    round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO);
1028 	}
1029 
1030 	VOP_UNLOCK(sc->sc_bs_vp);
1031 
1032 	return error;
1033 }
1034 
1035 /*
1036  * Get a pointer to the indirect slot for this cluster.
1037  */
1038 static u_int32_t *
1039 fss_bs_indir(struct fss_softc *sc, u_int32_t cl)
1040 {
1041 	u_int32_t icl;
1042 	int ioff;
1043 
1044 	icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1045 	ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t));
1046 
1047 	if (sc->sc_indir_cur == icl)
1048 		return &sc->sc_indir_data[ioff];
1049 
1050 	if (sc->sc_indir_dirty) {
1051 		if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0,
1052 		    FSS_CLSIZE(sc), (void *)sc->sc_indir_data, NULL) != 0)
1053 			return NULL;
1054 		setbit(sc->sc_indir_valid, sc->sc_indir_cur);
1055 	}
1056 
1057 	sc->sc_indir_dirty = 0;
1058 	sc->sc_indir_cur = icl;
1059 
1060 	if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) {
1061 		if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0,
1062 		    FSS_CLSIZE(sc), (void *)sc->sc_indir_data, NULL) != 0)
1063 			return NULL;
1064 	} else
1065 		memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc));
1066 
1067 	return &sc->sc_indir_data[ioff];
1068 }
1069 
1070 /*
1071  * The kernel thread (one for every active snapshot).
1072  *
1073  * After wakeup it cleans the cache and runs the I/O requests.
1074  */
1075 static void
1076 fss_bs_thread(void *arg)
1077 {
1078 	bool thread_idle, is_valid;
1079 	int error, i, todo, len, crotor, is_read;
1080 	long off;
1081 	char *addr;
1082 	u_int32_t c, cl, ch, *indirp;
1083 	size_t resid;
1084 	struct buf *bp, *nbp;
1085 	struct fss_softc *sc;
1086 	struct fss_cache *scp, *scl;
1087 
1088 	sc = arg;
1089 	scl = sc->sc_cache+sc->sc_cache_size;
1090 	crotor = 0;
1091 	thread_idle = false;
1092 
1093 	mutex_enter(&sc->sc_slock);
1094 
1095 	for (;;) {
1096 		if (thread_idle)
1097 			cv_wait(&sc->sc_work_cv, &sc->sc_slock);
1098 		thread_idle = true;
1099 		if ((sc->sc_flags & FSS_BS_THREAD) == 0) {
1100 			mutex_exit(&sc->sc_slock);
1101 			kthread_exit(0);
1102 		}
1103 
1104 		/*
1105 		 * Process I/O requests (persistent)
1106 		 */
1107 
1108 		if (sc->sc_flags & FSS_PERSISTENT) {
1109 			if ((bp = bufq_get(sc->sc_bufq)) == NULL)
1110 				continue;
1111 			is_valid = FSS_ISVALID(sc);
1112 			is_read = (bp->b_flags & B_READ);
1113 			thread_idle = false;
1114 			mutex_exit(&sc->sc_slock);
1115 
1116 			if (is_valid) {
1117 				disk_busy(sc->sc_dkdev);
1118 				error = fss_bs_io(sc, FSS_READ, 0,
1119 				    dbtob(bp->b_blkno), bp->b_bcount,
1120 				    bp->b_data, &resid);
1121 				if (error)
1122 					resid = bp->b_bcount;
1123 				disk_unbusy(sc->sc_dkdev,
1124 				    (error ? 0 : bp->b_bcount), is_read);
1125 			} else {
1126 				error = ENXIO;
1127 				resid = bp->b_bcount;
1128 			}
1129 
1130 			bp->b_error = error;
1131 			bp->b_resid = resid;
1132 			biodone(bp);
1133 
1134 			mutex_enter(&sc->sc_slock);
1135 			continue;
1136 		}
1137 
1138 		/*
1139 		 * Clean the cache
1140 		 */
1141 		for (i = 0; i < sc->sc_cache_size; i++) {
1142 			crotor = (crotor + 1) % sc->sc_cache_size;
1143 			scp = sc->sc_cache + crotor;
1144 			if (scp->fc_type != FSS_CACHE_VALID)
1145 				continue;
1146 			mutex_exit(&sc->sc_slock);
1147 
1148 			thread_idle = false;
1149 			indirp = fss_bs_indir(sc, scp->fc_cluster);
1150 			if (indirp != NULL) {
1151 				error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext,
1152 				    0, FSS_CLSIZE(sc), scp->fc_data, NULL);
1153 			} else
1154 				error = EIO;
1155 
1156 			mutex_enter(&sc->sc_slock);
1157 			if (error == 0) {
1158 				*indirp = sc->sc_clnext++;
1159 				sc->sc_indir_dirty = 1;
1160 			} else
1161 				fss_error(sc, "write error on backing store");
1162 
1163 			scp->fc_type = FSS_CACHE_FREE;
1164 			cv_broadcast(&sc->sc_cache_cv);
1165 			break;
1166 		}
1167 
1168 		/*
1169 		 * Process I/O requests
1170 		 */
1171 		if ((bp = bufq_get(sc->sc_bufq)) == NULL)
1172 			continue;
1173 		is_valid = FSS_ISVALID(sc);
1174 		is_read = (bp->b_flags & B_READ);
1175 		thread_idle = false;
1176 
1177 		if (!is_valid) {
1178 			mutex_exit(&sc->sc_slock);
1179 
1180 			bp->b_error = ENXIO;
1181 			bp->b_resid = bp->b_bcount;
1182 			biodone(bp);
1183 
1184 			mutex_enter(&sc->sc_slock);
1185 			continue;
1186 		}
1187 
1188 		disk_busy(sc->sc_dkdev);
1189 
1190 		/*
1191 		 * First read from the snapshotted block device unless
1192 		 * this request is completely covered by backing store.
1193 		 */
1194 
1195 		cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
1196 		off = FSS_CLOFF(sc, dbtob(bp->b_blkno));
1197 		ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
1198 		error = 0;
1199 		bp->b_resid = 0;
1200 		bp->b_error = 0;
1201 		for (c = cl; c <= ch; c++) {
1202 			if (isset(sc->sc_copied, c))
1203 				continue;
1204 			mutex_exit(&sc->sc_slock);
1205 
1206 			/* Not on backing store, read from device. */
1207 			nbp = getiobuf(NULL, true);
1208 			nbp->b_flags = B_READ;
1209 			nbp->b_resid = nbp->b_bcount = bp->b_bcount;
1210 			nbp->b_bufsize = bp->b_bcount;
1211 			nbp->b_data = bp->b_data;
1212 			nbp->b_blkno = bp->b_blkno;
1213 			nbp->b_lblkno = 0;
1214 			nbp->b_dev = sc->sc_bdev;
1215 			SET(nbp->b_cflags, BC_BUSY);	/* mark buffer busy */
1216 
1217 			bdev_strategy(nbp);
1218 
1219 			error = biowait(nbp);
1220 			if (error == 0 && nbp->b_resid != 0)
1221 				error = EIO;
1222 			if (error != 0) {
1223 				bp->b_resid = bp->b_bcount;
1224 				bp->b_error = nbp->b_error;
1225 				disk_unbusy(sc->sc_dkdev, 0, is_read);
1226 				biodone(bp);
1227 			}
1228 			putiobuf(nbp);
1229 
1230 			mutex_enter(&sc->sc_slock);
1231 			break;
1232 		}
1233 		if (error)
1234 			continue;
1235 
1236 		/*
1237 		 * Replace those parts that have been saved to backing store.
1238 		 */
1239 
1240 		addr = bp->b_data;
1241 		todo = bp->b_bcount;
1242 		for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) {
1243 			len = FSS_CLSIZE(sc)-off;
1244 			if (len > todo)
1245 				len = todo;
1246 			if (isclr(sc->sc_copied, c))
1247 				continue;
1248 			mutex_exit(&sc->sc_slock);
1249 
1250 			indirp = fss_bs_indir(sc, c);
1251 			if (indirp == NULL || *indirp == 0) {
1252 				/*
1253 				 * Not on backing store. Either in cache
1254 				 * or hole in the snapshotted block device.
1255 				 */
1256 
1257 				mutex_enter(&sc->sc_slock);
1258 				for (scp = sc->sc_cache; scp < scl; scp++)
1259 					if (scp->fc_type == FSS_CACHE_VALID &&
1260 					    scp->fc_cluster == c)
1261 						break;
1262 				if (scp < scl)
1263 					memcpy(addr, (char *)scp->fc_data+off,
1264 					    len);
1265 				else
1266 					memset(addr, 0, len);
1267 				continue;
1268 			}
1269 
1270 			/*
1271 			 * Read from backing store.
1272 			 */
1273 			error = fss_bs_io(sc, FSS_READ,
1274 			    *indirp, off, len, addr, NULL);
1275 
1276 			mutex_enter(&sc->sc_slock);
1277 			if (error) {
1278 				bp->b_resid = bp->b_bcount;
1279 				bp->b_error = error;
1280 				break;
1281 			}
1282 		}
1283 		mutex_exit(&sc->sc_slock);
1284 
1285 		disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read);
1286 		biodone(bp);
1287 
1288 		mutex_enter(&sc->sc_slock);
1289 	}
1290 }
1291 
1292 #ifdef _MODULE
1293 
1294 #include <sys/module.h>
1295 
1296 MODULE(MODULE_CLASS_DRIVER, fss, "bufq_fcfs");
1297 CFDRIVER_DECL(fss, DV_DISK, NULL);
1298 
1299 devmajor_t fss_bmajor = -1, fss_cmajor = -1;
1300 
1301 static int
1302 fss_modcmd(modcmd_t cmd, void *arg)
1303 {
1304 	int error = 0;
1305 
1306 	switch (cmd) {
1307 	case MODULE_CMD_INIT:
1308 		mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE);
1309 		error = config_cfdriver_attach(&fss_cd);
1310 		if (error) {
1311 			mutex_destroy(&fss_device_lock);
1312 			break;
1313 		}
1314 		error = config_cfattach_attach(fss_cd.cd_name, &fss_ca);
1315 		if (error) {
1316 			config_cfdriver_detach(&fss_cd);
1317 			mutex_destroy(&fss_device_lock);
1318 			break;
1319 		}
1320 		error = devsw_attach(fss_cd.cd_name,
1321 		    &fss_bdevsw, &fss_bmajor, &fss_cdevsw, &fss_cmajor);
1322 
1323 		if (error) {
1324 			config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1325 			config_cfdriver_detach(&fss_cd);
1326 			mutex_destroy(&fss_device_lock);
1327 			break;
1328 		}
1329 		break;
1330 
1331 	case MODULE_CMD_FINI:
1332 		devsw_detach(&fss_bdevsw, &fss_cdevsw);
1333 		error = config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1334 		if (error) {
1335 			devsw_attach(fss_cd.cd_name, &fss_bdevsw, &fss_bmajor,
1336 			    &fss_cdevsw, &fss_cmajor);
1337 			break;
1338 		}
1339 		error = config_cfdriver_detach(&fss_cd);
1340 		if (error) {
1341 			devsw_attach(fss_cd.cd_name,
1342 			    &fss_bdevsw, &fss_bmajor, &fss_cdevsw, &fss_cmajor);
1343 			devsw_attach(fss_cd.cd_name, &fss_bdevsw, &fss_bmajor,
1344 			    &fss_cdevsw, &fss_cmajor);
1345 			break;
1346 		}
1347 		mutex_destroy(&fss_device_lock);
1348 		break;
1349 
1350 	default:
1351 		error = ENOTTY;
1352 		break;
1353 	}
1354 
1355 	return error;
1356 }
1357 
1358 #endif /* _MODULE */
1359