xref: /netbsd-src/sys/dev/fss.c (revision 3816d47b2c42fcd6e549e3407f842a5b1a1d23ad)
1 /*	$NetBSD: fss.c,v 1.65 2009/10/17 10:29:29 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  *
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 1.65 2009/10/17 10:29:29 hannken 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/simplelock.h>
61 
62 #include <miscfs/specfs/specdev.h>
63 
64 #include <dev/fssvar.h>
65 
66 #include <uvm/uvm.h>
67 
68 void fssattach(int);
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 *);
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 	fss_open, fss_close, fss_strategy, fss_ioctl,
103 	fss_dump, fss_size, D_DISK | D_MPSAFE
104 };
105 
106 const struct cdevsw fss_cdevsw = {
107 	fss_open, fss_close, fss_read, fss_write, fss_ioctl,
108 	nostop, notty, nopoll, nommap, nokqfilter, D_DISK | D_MPSAFE
109 };
110 
111 static int fss_match(device_t, cfdata_t, void *);
112 static void fss_attach(device_t, device_t, void *);
113 static int fss_detach(device_t, int);
114 
115 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc),
116     fss_match, fss_attach, fss_detach, NULL);
117 extern struct cfdriver fss_cd;
118 
119 void
120 fssattach(int num)
121 {
122 
123 	mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE);
124 	if (config_cfattach_attach(fss_cd.cd_name, &fss_ca))
125 		aprint_error("%s: unable to register\n", fss_cd.cd_name);
126 }
127 
128 static int
129 fss_match(device_t self, cfdata_t cfdata, void *aux)
130 {
131 	return 1;
132 }
133 
134 static void
135 fss_attach(device_t parent, device_t self, void *aux)
136 {
137 	struct fss_softc *sc = device_private(self);
138 
139 	sc->sc_dev = self;
140 	sc->sc_bdev = NODEV;
141 	mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE);
142 	mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE);
143 	cv_init(&sc->sc_work_cv, "fssbs");
144 	cv_init(&sc->sc_cache_cv, "cowwait");
145 	bufq_alloc(&sc->sc_bufq, "fcfs", 0);
146 	sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK);
147 	sc->sc_dkdev->dk_info = NULL;
148 	disk_init(sc->sc_dkdev, device_xname(self), NULL);
149 	if (!pmf_device_register(self, NULL, NULL))
150 		aprint_error_dev(self, "couldn't establish power handler\n");
151 
152 	if (fss_num_attached++ == 0)
153 		vfs_hooks_attach(&fss_vfs_hooks);
154 }
155 
156 static int
157 fss_detach(device_t self, int flags)
158 {
159 	struct fss_softc *sc = device_private(self);
160 
161 	if (sc->sc_flags & FSS_ACTIVE)
162 		return EBUSY;
163 
164 	if (--fss_num_attached == 0)
165 		vfs_hooks_detach(&fss_vfs_hooks);
166 
167 	pmf_device_deregister(self);
168 	mutex_destroy(&sc->sc_slock);
169 	mutex_destroy(&sc->sc_lock);
170 	cv_destroy(&sc->sc_work_cv);
171 	cv_destroy(&sc->sc_cache_cv);
172 	bufq_drain(sc->sc_bufq);
173 	bufq_free(sc->sc_bufq);
174 	disk_destroy(sc->sc_dkdev);
175 	free(sc->sc_dkdev, M_DEVBUF);
176 
177 	return 0;
178 }
179 
180 int
181 fss_open(dev_t dev, int flags, int mode, struct lwp *l)
182 {
183 	int mflag;
184 	cfdata_t cf;
185 	struct fss_softc *sc;
186 
187 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
188 
189 	mutex_enter(&fss_device_lock);
190 
191 	sc = device_lookup_private(&fss_cd, minor(dev));
192 	if (sc == NULL) {
193 		cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK);
194 		cf->cf_name = fss_cd.cd_name;
195 		cf->cf_atname = fss_cd.cd_name;
196 		cf->cf_unit = minor(dev);
197 		cf->cf_fstate = FSTATE_STAR;
198 		sc = device_private(config_attach_pseudo(cf));
199 		if (sc == NULL)
200 			return ENOMEM;
201 	}
202 
203 	mutex_enter(&sc->sc_slock);
204 
205 	sc->sc_flags |= mflag;
206 
207 	mutex_exit(&sc->sc_slock);
208 	mutex_exit(&fss_device_lock);
209 
210 	return 0;
211 }
212 
213 int
214 fss_close(dev_t dev, int flags, int mode, struct lwp *l)
215 {
216 	int mflag, error;
217 	cfdata_t cf;
218 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev));
219 
220 	mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN);
221 	error = 0;
222 
223 restart:
224 	mutex_enter(&sc->sc_slock);
225 	if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) {
226 		sc->sc_flags &= ~mflag;
227 		mutex_exit(&sc->sc_slock);
228 		return 0;
229 	}
230 	if ((sc->sc_flags & FSS_ACTIVE) != 0 &&
231 	    (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) {
232 		sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE;
233 		mutex_exit(&sc->sc_slock);
234 		error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l);
235 		goto restart;
236 	}
237 	if ((sc->sc_flags & FSS_ACTIVE) != 0) {
238 		mutex_exit(&sc->sc_slock);
239 		return error;
240 	}
241 	if (! mutex_tryenter(&fss_device_lock)) {
242 		mutex_exit(&sc->sc_slock);
243 		goto restart;
244 	}
245 
246 	KASSERT((sc->sc_flags & FSS_ACTIVE) == 0);
247 	KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag);
248 	mutex_exit(&sc->sc_slock);
249 	cf = device_cfdata(sc->sc_dev);
250 	error = config_detach(sc->sc_dev, DETACH_QUIET);
251 	if (! error)
252 		free(cf, M_DEVBUF);
253 	mutex_exit(&fss_device_lock);
254 
255 	return error;
256 }
257 
258 void
259 fss_strategy(struct buf *bp)
260 {
261 	const bool write = ((bp->b_flags & B_READ) != B_READ);
262 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev));
263 
264 	mutex_enter(&sc->sc_slock);
265 
266 	if (write || !FSS_ISVALID(sc)) {
267 
268 		mutex_exit(&sc->sc_slock);
269 
270 		bp->b_error = (write ? EROFS : ENXIO);
271 		bp->b_resid = bp->b_bcount;
272 		biodone(bp);
273 		return;
274 	}
275 
276 	bp->b_rawblkno = bp->b_blkno;
277 	bufq_put(sc->sc_bufq, bp);
278 	cv_signal(&sc->sc_work_cv);
279 
280 	mutex_exit(&sc->sc_slock);
281 }
282 
283 int
284 fss_read(dev_t dev, struct uio *uio, int flags)
285 {
286 	return physio(fss_strategy, NULL, dev, B_READ, minphys, uio);
287 }
288 
289 int
290 fss_write(dev_t dev, struct uio *uio, int flags)
291 {
292 	return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio);
293 }
294 
295 int
296 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
297 {
298 	int error;
299 	struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev));
300 	struct fss_set *fss = (struct fss_set *)data;
301 	struct fss_get *fsg = (struct fss_get *)data;
302 
303 	switch (cmd) {
304 	case FSSIOCSET:
305 		mutex_enter(&sc->sc_lock);
306 		if ((flag & FWRITE) == 0)
307 			error = EPERM;
308 		else if ((sc->sc_flags & FSS_ACTIVE) != 0)
309 			error = EBUSY;
310 		else
311 			error = fss_create_snapshot(sc, fss, l);
312 		mutex_exit(&sc->sc_lock);
313 		break;
314 
315 	case FSSIOCCLR:
316 		mutex_enter(&sc->sc_lock);
317 		if ((flag & FWRITE) == 0)
318 			error = EPERM;
319 		else if ((sc->sc_flags & FSS_ACTIVE) == 0)
320 			error = ENXIO;
321 		else
322 			error = fss_delete_snapshot(sc, l);
323 		mutex_exit(&sc->sc_lock);
324 		break;
325 
326 	case FSSIOCGET:
327 		mutex_enter(&sc->sc_lock);
328 		switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) {
329 		case FSS_ACTIVE:
330 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
331 			fsg->fsg_csize = FSS_CLSIZE(sc);
332 			fsg->fsg_time = sc->sc_time;
333 			fsg->fsg_mount_size = sc->sc_clcount;
334 			fsg->fsg_bs_size = sc->sc_clnext;
335 			error = 0;
336 			break;
337 		case FSS_PERSISTENT | FSS_ACTIVE:
338 			memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN);
339 			fsg->fsg_csize = 0;
340 			fsg->fsg_time = sc->sc_time;
341 			fsg->fsg_mount_size = 0;
342 			fsg->fsg_bs_size = 0;
343 			error = 0;
344 			break;
345 		default:
346 			error = ENXIO;
347 			break;
348 		}
349 		mutex_exit(&sc->sc_lock);
350 		break;
351 
352 	case FSSIOFSET:
353 		mutex_enter(&sc->sc_slock);
354 		sc->sc_uflags = *(int *)data;
355 		mutex_exit(&sc->sc_slock);
356 		error = 0;
357 		break;
358 
359 	case FSSIOFGET:
360 		mutex_enter(&sc->sc_slock);
361 		*(int *)data = sc->sc_uflags;
362 		mutex_exit(&sc->sc_slock);
363 		error = 0;
364 		break;
365 
366 	default:
367 		error = EINVAL;
368 		break;
369 	}
370 
371 	return error;
372 }
373 
374 int
375 fss_size(dev_t dev)
376 {
377 	return -1;
378 }
379 
380 int
381 fss_dump(dev_t dev, daddr_t blkno, void *va,
382     size_t size)
383 {
384 	return EROFS;
385 }
386 
387 /*
388  * An error occurred reading or writing the snapshot or backing store.
389  * If it is the first error log to console.
390  * The caller holds the mutex.
391  */
392 static inline void
393 fss_error(struct fss_softc *sc, const char *msg)
394 {
395 
396 	if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE)
397 		aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg);
398 	if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE)
399 		sc->sc_flags |= FSS_ERROR;
400 }
401 
402 /*
403  * Allocate the variable sized parts of the softc and
404  * fork the kernel thread.
405  *
406  * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size
407  * must be initialized.
408  */
409 static int
410 fss_softc_alloc(struct fss_softc *sc)
411 {
412 	int i, error;
413 
414 	if ((sc->sc_flags & FSS_PERSISTENT) == 0) {
415 		sc->sc_copied =
416 		    kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP);
417 		if (sc->sc_copied == NULL)
418 			return(ENOMEM);
419 
420 		sc->sc_cache = kmem_alloc(sc->sc_cache_size *
421 		    sizeof(struct fss_cache), KM_SLEEP);
422 		if (sc->sc_cache == NULL)
423 			return(ENOMEM);
424 
425 		for (i = 0; i < sc->sc_cache_size; i++) {
426 			sc->sc_cache[i].fc_type = FSS_CACHE_FREE;
427 			sc->sc_cache[i].fc_data =
428 			    kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP);
429 			if (sc->sc_cache[i].fc_data == NULL)
430 				return(ENOMEM);
431 			cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1");
432 		}
433 
434 		sc->sc_indir_valid =
435 		    kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP);
436 		if (sc->sc_indir_valid == NULL)
437 			return(ENOMEM);
438 
439 		sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP);
440 		if (sc->sc_indir_data == NULL)
441 			return(ENOMEM);
442 	} else {
443 		sc->sc_copied = NULL;
444 		sc->sc_cache = NULL;
445 		sc->sc_indir_valid = NULL;
446 		sc->sc_indir_data = NULL;
447 	}
448 
449 	sc->sc_flags |= FSS_BS_THREAD;
450 	if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc,
451 	    &sc->sc_bs_lwp, device_xname(sc->sc_dev))) != 0) {
452 		sc->sc_flags &= ~FSS_BS_THREAD;
453 		return error;
454 	}
455 
456 	disk_attach(sc->sc_dkdev);
457 
458 	return 0;
459 }
460 
461 /*
462  * Free the variable sized parts of the softc.
463  */
464 static void
465 fss_softc_free(struct fss_softc *sc)
466 {
467 	int i;
468 
469 	if ((sc->sc_flags & FSS_BS_THREAD) != 0) {
470 		mutex_enter(&sc->sc_slock);
471 		sc->sc_flags &= ~FSS_BS_THREAD;
472 		cv_signal(&sc->sc_work_cv);
473 		while (sc->sc_bs_lwp != NULL)
474 			kpause("fssdetach", false, 1, &sc->sc_slock);
475 		mutex_exit(&sc->sc_slock);
476 	}
477 
478 	disk_detach(sc->sc_dkdev);
479 
480 	if (sc->sc_copied != NULL)
481 		kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY));
482 	sc->sc_copied = NULL;
483 
484 	if (sc->sc_cache != NULL) {
485 		for (i = 0; i < sc->sc_cache_size; i++)
486 			if (sc->sc_cache[i].fc_data != NULL) {
487 				cv_destroy(&sc->sc_cache[i].fc_state_cv);
488 				kmem_free(sc->sc_cache[i].fc_data,
489 				    FSS_CLSIZE(sc));
490 			}
491 		kmem_free(sc->sc_cache,
492 		    sc->sc_cache_size*sizeof(struct fss_cache));
493 	}
494 	sc->sc_cache = NULL;
495 
496 	if (sc->sc_indir_valid != NULL)
497 		kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY));
498 	sc->sc_indir_valid = NULL;
499 
500 	if (sc->sc_indir_data != NULL)
501 		kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc));
502 	sc->sc_indir_data = NULL;
503 }
504 
505 /*
506  * Set all active snapshots on this file system into ERROR state.
507  */
508 static void
509 fss_unmount_hook(struct mount *mp)
510 {
511 	int i;
512 	struct fss_softc *sc;
513 
514 	mutex_enter(&fss_device_lock);
515 	for (i = 0; i < fss_cd.cd_ndevs; i++) {
516 		if ((sc = device_lookup_private(&fss_cd, i)) == NULL)
517 			continue;
518 		mutex_enter(&sc->sc_slock);
519 		if ((sc->sc_flags & FSS_ACTIVE) != 0 &&
520 		    sc->sc_mount == mp)
521 			fss_error(sc, "forced unmount");
522 		mutex_exit(&sc->sc_slock);
523 	}
524 	mutex_exit(&fss_device_lock);
525 }
526 
527 /*
528  * A buffer is written to the snapshotted block device. Copy to
529  * backing store if needed.
530  */
531 static int
532 fss_copy_on_write(void *v, struct buf *bp, bool data_valid)
533 {
534 	int error;
535 	u_int32_t cl, ch, c;
536 	struct fss_softc *sc = v;
537 
538 	mutex_enter(&sc->sc_slock);
539 	if (!FSS_ISVALID(sc)) {
540 		mutex_exit(&sc->sc_slock);
541 		return 0;
542 	}
543 
544 	cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
545 	ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
546 	error = 0;
547 	if (curlwp == uvm.pagedaemon_lwp) {
548 		for (c = cl; c <= ch; c++)
549 			if (isclr(sc->sc_copied, c)) {
550 				error = ENOMEM;
551 				break;
552 			}
553 	}
554 	mutex_exit(&sc->sc_slock);
555 
556 	if (error == 0)
557 		for (c = cl; c <= ch; c++) {
558 			error = fss_read_cluster(sc, c);
559 			if (error)
560 				break;
561 		}
562 
563 	return error;
564 }
565 
566 /*
567  * Lookup and open needed files.
568  *
569  * For file system internal snapshot initializes sc_mntname, sc_mount,
570  * sc_bs_vp and sc_time.
571  *
572  * Otherwise returns dev and size of the underlying block device.
573  * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount
574  */
575 static int
576 fss_create_files(struct fss_softc *sc, struct fss_set *fss,
577     off_t *bsize, struct lwp *l)
578 {
579 	int error, bits, fsbsize;
580 	struct timespec ts;
581 	struct partinfo dpart;
582 	struct vattr va;
583 	/* nd -> nd2 to reduce mistakes while updating only some namei calls */
584 	struct nameidata nd2;
585 	struct vnode *vp;
586 
587 	/*
588 	 * Get the mounted file system.
589 	 */
590 
591 	error = namei_simple_user(fss->fss_mount,
592 				NSM_FOLLOW_NOEMULROOT, &vp);
593 	if (error != 0)
594 		return error;
595 
596 	if ((vp->v_vflag & VV_ROOT) != VV_ROOT) {
597 		vrele(vp);
598 		return EINVAL;
599 	}
600 
601 	sc->sc_mount = vp->v_mount;
602 	memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN);
603 
604 	vrele(vp);
605 
606 	/*
607 	 * Check for file system internal snapshot.
608 	 */
609 
610 	error = namei_simple_user(fss->fss_bstore,
611 				NSM_FOLLOW_NOEMULROOT, &vp);
612 	if (error != 0)
613 		return error;
614 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
615 	if (error != 0) {
616 		vrele(vp);
617 		return error;
618 	}
619 
620 	if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) {
621 		sc->sc_flags |= FSS_PERSISTENT;
622 		sc->sc_bs_vp = vp;
623 
624 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
625 		bits = sizeof(sc->sc_bs_bshift)*NBBY;
626 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits;
627 		    sc->sc_bs_bshift++)
628 			if (FSS_FSBSIZE(sc) == fsbsize)
629 				break;
630 		if (sc->sc_bs_bshift >= bits) {
631 			VOP_UNLOCK(sc->sc_bs_vp, 0);
632 			return EINVAL;
633 		}
634 
635 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
636 		sc->sc_clshift = 0;
637 
638 		error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
639 		TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
640 
641 		VOP_UNLOCK(sc->sc_bs_vp, 0);
642 
643 		return error;
644 	}
645 	vput(vp);
646 
647 	/*
648 	 * Get the block device it is mounted on.
649 	 */
650 
651 	error = namei_simple_kernel(sc->sc_mount->mnt_stat.f_mntfromname,
652 				NSM_FOLLOW_NOEMULROOT, &vp);
653 	if (error != 0)
654 		return error;
655 
656 	if (vp->v_type != VBLK) {
657 		vrele(vp);
658 		return EINVAL;
659 	}
660 
661 	sc->sc_bdev = vp->v_rdev;
662 	vrele(vp);
663 
664 	/*
665 	 * Get the block device size.
666 	 */
667 
668 	error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l);
669 	if (error)
670 		return error;
671 
672 	*bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size;
673 
674 	/*
675 	 * Get the backing store
676 	 */
677 
678 	NDINIT(&nd2, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore);
679 	if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0)
680 		return error;
681 	VOP_UNLOCK(nd2.ni_vp, 0);
682 
683 	sc->sc_bs_vp = nd2.ni_vp;
684 
685 	if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR)
686 		return EINVAL;
687 
688 	if (sc->sc_bs_vp->v_type == VREG) {
689 		error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred);
690 		if (error != 0)
691 			return error;
692 		sc->sc_bs_size = va.va_size;
693 		fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
694 		if (fsbsize & (fsbsize-1))	/* No power of two */
695 			return EINVAL;
696 		for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
697 		    sc->sc_bs_bshift++)
698 			if (FSS_FSBSIZE(sc) == fsbsize)
699 				break;
700 		if (sc->sc_bs_bshift >= 32)
701 			return EINVAL;
702 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
703 	} else {
704 		sc->sc_bs_bshift = DEV_BSHIFT;
705 		sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
706 	}
707 
708 	return 0;
709 }
710 
711 /*
712  * Create a snapshot.
713  */
714 static int
715 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l)
716 {
717 	int len, error;
718 	u_int32_t csize;
719 	off_t bsize;
720 
721 	bsize = 0;	/* XXX gcc */
722 
723 	/*
724 	 * Open needed files.
725 	 */
726 	if ((error = fss_create_files(sc, fss, &bsize, l)) != 0)
727 		goto bad;
728 
729 	if (sc->sc_flags & FSS_PERSISTENT) {
730 		fss_softc_alloc(sc);
731 		sc->sc_flags |= FSS_ACTIVE;
732 		return 0;
733 	}
734 
735 	/*
736 	 * Set cluster size. Must be a power of two and
737 	 * a multiple of backing store block size.
738 	 */
739 	if (fss->fss_csize <= 0)
740 		csize = MAXPHYS;
741 	else
742 		csize = fss->fss_csize;
743 	if (bsize/csize > FSS_CLUSTER_MAX)
744 		csize = bsize/FSS_CLUSTER_MAX+1;
745 
746 	for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32;
747 	    sc->sc_clshift++)
748 		if (FSS_CLSIZE(sc) >= csize)
749 			break;
750 	if (sc->sc_clshift >= 32) {
751 		error = EINVAL;
752 		goto bad;
753 	}
754 	sc->sc_clmask = FSS_CLSIZE(sc)-1;
755 
756 	/*
757 	 * Set number of cache slots.
758 	 */
759 	if (FSS_CLSIZE(sc) <= 8192)
760 		sc->sc_cache_size = 32;
761 	else if (FSS_CLSIZE(sc) <= 65536)
762 		sc->sc_cache_size = 8;
763 	else
764 		sc->sc_cache_size = 4;
765 
766 	/*
767 	 * Set number of clusters and size of last cluster.
768 	 */
769 	sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1;
770 	sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1;
771 
772 	/*
773 	 * Set size of indirect table.
774 	 */
775 	len = sc->sc_clcount*sizeof(u_int32_t);
776 	sc->sc_indir_size = FSS_BTOCL(sc, len)+1;
777 	sc->sc_clnext = sc->sc_indir_size;
778 	sc->sc_indir_cur = 0;
779 
780 	if ((error = fss_softc_alloc(sc)) != 0)
781 		goto bad;
782 
783 	/*
784 	 * Activate the snapshot.
785 	 */
786 
787 	if ((error = vfs_suspend(sc->sc_mount, 0)) != 0)
788 		goto bad;
789 
790 	microtime(&sc->sc_time);
791 
792 	if (error == 0)
793 		error = fscow_establish(sc->sc_mount,
794 		    fss_copy_on_write, sc);
795 	if (error == 0)
796 		sc->sc_flags |= FSS_ACTIVE;
797 
798 	vfs_resume(sc->sc_mount);
799 
800 	if (error != 0)
801 		goto bad;
802 
803 	aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname);
804 	aprint_debug_dev(sc->sc_dev,
805 	    "%u clusters of %u, %u cache slots, %u indir clusters\n",
806 	    sc->sc_clcount, FSS_CLSIZE(sc),
807 	    sc->sc_cache_size, sc->sc_indir_size);
808 
809 	return 0;
810 
811 bad:
812 	fss_softc_free(sc);
813 	if (sc->sc_bs_vp != NULL) {
814 		if (sc->sc_flags & FSS_PERSISTENT)
815 			vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
816 		else
817 			vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
818 	}
819 	sc->sc_bs_vp = NULL;
820 
821 	return error;
822 }
823 
824 /*
825  * Delete a snapshot.
826  */
827 static int
828 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l)
829 {
830 
831 	if ((sc->sc_flags & FSS_PERSISTENT) == 0)
832 		fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc);
833 
834 	mutex_enter(&sc->sc_slock);
835 	sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR);
836 	sc->sc_mount = NULL;
837 	sc->sc_bdev = NODEV;
838 	mutex_exit(&sc->sc_slock);
839 
840 	fss_softc_free(sc);
841 	if (sc->sc_flags & FSS_PERSISTENT)
842 		vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
843 	else
844 		vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
845 	sc->sc_bs_vp = NULL;
846 	sc->sc_flags &= ~FSS_PERSISTENT;
847 
848 	return 0;
849 }
850 
851 /*
852  * Read a cluster from the snapshotted block device to the cache.
853  */
854 static int
855 fss_read_cluster(struct fss_softc *sc, u_int32_t cl)
856 {
857 	int error, todo, offset, len;
858 	daddr_t dblk;
859 	struct buf *bp, *mbp;
860 	struct fss_cache *scp, *scl;
861 
862 	/*
863 	 * Get a free cache slot.
864 	 */
865 	scl = sc->sc_cache+sc->sc_cache_size;
866 
867 	mutex_enter(&sc->sc_slock);
868 
869 restart:
870 	if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) {
871 		mutex_exit(&sc->sc_slock);
872 		return 0;
873 	}
874 
875 	for (scp = sc->sc_cache; scp < scl; scp++)
876 		if (scp->fc_cluster == cl) {
877 			if (scp->fc_type == FSS_CACHE_VALID) {
878 				mutex_exit(&sc->sc_slock);
879 				return 0;
880 			} else if (scp->fc_type == FSS_CACHE_BUSY) {
881 				cv_wait(&scp->fc_state_cv, &sc->sc_slock);
882 				goto restart;
883 			}
884 		}
885 
886 	for (scp = sc->sc_cache; scp < scl; scp++)
887 		if (scp->fc_type == FSS_CACHE_FREE) {
888 			scp->fc_type = FSS_CACHE_BUSY;
889 			scp->fc_cluster = cl;
890 			break;
891 		}
892 	if (scp >= scl) {
893 		cv_wait(&sc->sc_cache_cv, &sc->sc_slock);
894 		goto restart;
895 	}
896 
897 	mutex_exit(&sc->sc_slock);
898 
899 	/*
900 	 * Start the read.
901 	 */
902 	dblk = btodb(FSS_CLTOB(sc, cl));
903 	if (cl == sc->sc_clcount-1) {
904 		todo = sc->sc_clresid;
905 		memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo);
906 	} else
907 		todo = FSS_CLSIZE(sc);
908 	offset = 0;
909 	mbp = getiobuf(NULL, true);
910 	mbp->b_bufsize = todo;
911 	mbp->b_data = scp->fc_data;
912 	mbp->b_resid = mbp->b_bcount = todo;
913 	mbp->b_flags = B_READ;
914 	mbp->b_cflags = BC_BUSY;
915 	mbp->b_dev = sc->sc_bdev;
916 	while (todo > 0) {
917 		len = todo;
918 		if (len > MAXPHYS)
919 			len = MAXPHYS;
920 		if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo)
921 			bp = mbp;
922 		else {
923 			bp = getiobuf(NULL, true);
924 			nestiobuf_setup(mbp, bp, offset, len);
925 		}
926 		bp->b_lblkno = 0;
927 		bp->b_blkno = dblk;
928 		bdev_strategy(bp);
929 		dblk += btodb(len);
930 		offset += len;
931 		todo -= len;
932 	}
933 	error = biowait(mbp);
934 	putiobuf(mbp);
935 
936 	mutex_enter(&sc->sc_slock);
937 	scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID);
938 	cv_broadcast(&scp->fc_state_cv);
939 	if (error == 0) {
940 		setbit(sc->sc_copied, scp->fc_cluster);
941 		cv_signal(&sc->sc_work_cv);
942 	}
943 	mutex_exit(&sc->sc_slock);
944 
945 	return error;
946 }
947 
948 /*
949  * Read/write clusters from/to backing store.
950  * For persistent snapshots must be called with cl == 0. off is the
951  * offset into the snapshot.
952  */
953 static int
954 fss_bs_io(struct fss_softc *sc, fss_io_type rw,
955     u_int32_t cl, off_t off, int len, void *data)
956 {
957 	int error;
958 
959 	off += FSS_CLTOB(sc, cl);
960 
961 	vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
962 
963 	error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp,
964 	    data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED,
965 	    sc->sc_bs_lwp->l_cred, NULL, NULL);
966 	if (error == 0) {
967 		mutex_enter(&sc->sc_bs_vp->v_interlock);
968 		error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off),
969 		    round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
970 	}
971 
972 	VOP_UNLOCK(sc->sc_bs_vp, 0);
973 
974 	return error;
975 }
976 
977 /*
978  * Get a pointer to the indirect slot for this cluster.
979  */
980 static u_int32_t *
981 fss_bs_indir(struct fss_softc *sc, u_int32_t cl)
982 {
983 	u_int32_t icl;
984 	int ioff;
985 
986 	icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t));
987 	ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t));
988 
989 	if (sc->sc_indir_cur == icl)
990 		return &sc->sc_indir_data[ioff];
991 
992 	if (sc->sc_indir_dirty) {
993 		if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0,
994 		    FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0)
995 			return NULL;
996 		setbit(sc->sc_indir_valid, sc->sc_indir_cur);
997 	}
998 
999 	sc->sc_indir_dirty = 0;
1000 	sc->sc_indir_cur = icl;
1001 
1002 	if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) {
1003 		if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0,
1004 		    FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0)
1005 			return NULL;
1006 	} else
1007 		memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc));
1008 
1009 	return &sc->sc_indir_data[ioff];
1010 }
1011 
1012 /*
1013  * The kernel thread (one for every active snapshot).
1014  *
1015  * After wakeup it cleans the cache and runs the I/O requests.
1016  */
1017 static void
1018 fss_bs_thread(void *arg)
1019 {
1020 	bool thread_idle, is_valid;
1021 	int error, i, todo, len, crotor, is_read;
1022 	long off;
1023 	char *addr;
1024 	u_int32_t c, cl, ch, *indirp;
1025 	struct buf *bp, *nbp;
1026 	struct fss_softc *sc;
1027 	struct fss_cache *scp, *scl;
1028 
1029 	sc = arg;
1030 	scl = sc->sc_cache+sc->sc_cache_size;
1031 	crotor = 0;
1032 	thread_idle = false;
1033 
1034 	mutex_enter(&sc->sc_slock);
1035 
1036 	for (;;) {
1037 		if (thread_idle)
1038 			cv_wait(&sc->sc_work_cv, &sc->sc_slock);
1039 		thread_idle = true;
1040 		if ((sc->sc_flags & FSS_BS_THREAD) == 0) {
1041 			sc->sc_bs_lwp = NULL;
1042 			mutex_exit(&sc->sc_slock);
1043 			kthread_exit(0);
1044 		}
1045 
1046 		/*
1047 		 * Process I/O requests (persistent)
1048 		 */
1049 
1050 		if (sc->sc_flags & FSS_PERSISTENT) {
1051 			if ((bp = bufq_get(sc->sc_bufq)) == NULL)
1052 				continue;
1053 			is_valid = FSS_ISVALID(sc);
1054 			is_read = (bp->b_flags & B_READ);
1055 			thread_idle = false;
1056 			mutex_exit(&sc->sc_slock);
1057 
1058 			if (is_valid) {
1059 				disk_busy(sc->sc_dkdev);
1060 				error = fss_bs_io(sc, FSS_READ, 0,
1061 				    dbtob(bp->b_blkno), bp->b_bcount,
1062 				    bp->b_data);
1063 				disk_unbusy(sc->sc_dkdev,
1064 				    (error ? 0 : bp->b_bcount), is_read);
1065 			} else
1066 				error = ENXIO;
1067 
1068 			bp->b_error = error;
1069 			bp->b_resid = (error ? bp->b_bcount : 0);
1070 			biodone(bp);
1071 
1072 			mutex_enter(&sc->sc_slock);
1073 			continue;
1074 		}
1075 
1076 		/*
1077 		 * Clean the cache
1078 		 */
1079 		for (i = 0; i < sc->sc_cache_size; i++) {
1080 			crotor = (crotor + 1) % sc->sc_cache_size;
1081 			scp = sc->sc_cache + crotor;
1082 			if (scp->fc_type != FSS_CACHE_VALID)
1083 				continue;
1084 			mutex_exit(&sc->sc_slock);
1085 
1086 			thread_idle = false;
1087 			indirp = fss_bs_indir(sc, scp->fc_cluster);
1088 			if (indirp != NULL) {
1089 				error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext,
1090 				    0, FSS_CLSIZE(sc), scp->fc_data);
1091 			} else
1092 				error = EIO;
1093 
1094 			mutex_enter(&sc->sc_slock);
1095 			if (error == 0) {
1096 				*indirp = sc->sc_clnext++;
1097 				sc->sc_indir_dirty = 1;
1098 			} else
1099 				fss_error(sc, "write error on backing store");
1100 
1101 			scp->fc_type = FSS_CACHE_FREE;
1102 			cv_signal(&sc->sc_cache_cv);
1103 			break;
1104 		}
1105 
1106 		/*
1107 		 * Process I/O requests
1108 		 */
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 
1115 		if (!is_valid) {
1116 			mutex_exit(&sc->sc_slock);
1117 
1118 			bp->b_error = ENXIO;
1119 			bp->b_resid = bp->b_bcount;
1120 			biodone(bp);
1121 
1122 			mutex_enter(&sc->sc_slock);
1123 			continue;
1124 		}
1125 
1126 		disk_busy(sc->sc_dkdev);
1127 
1128 		/*
1129 		 * First read from the snapshotted block device unless
1130 		 * this request is completely covered by backing store.
1131 		 */
1132 
1133 		cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
1134 		off = FSS_CLOFF(sc, dbtob(bp->b_blkno));
1135 		ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
1136 		error = 0;
1137 		bp->b_resid = 0;
1138 		bp->b_error = 0;
1139 		for (c = cl; c <= ch; c++) {
1140 			if (isset(sc->sc_copied, c))
1141 				continue;
1142 			mutex_exit(&sc->sc_slock);
1143 
1144 			/* Not on backing store, read from device. */
1145 			nbp = getiobuf(NULL, true);
1146 			nbp->b_flags = B_READ;
1147 			nbp->b_resid = nbp->b_bcount = bp->b_bcount;
1148 			nbp->b_bufsize = bp->b_bcount;
1149 			nbp->b_data = bp->b_data;
1150 			nbp->b_blkno = bp->b_blkno;
1151 			nbp->b_lblkno = 0;
1152 			nbp->b_dev = sc->sc_bdev;
1153 			SET(nbp->b_cflags, BC_BUSY);	/* mark buffer busy */
1154 
1155 			bdev_strategy(nbp);
1156 
1157 			error = biowait(nbp);
1158 			if (error != 0) {
1159 				bp->b_resid = bp->b_bcount;
1160 				bp->b_error = nbp->b_error;
1161 				disk_unbusy(sc->sc_dkdev, 0, is_read);
1162 				biodone(bp);
1163 			}
1164 			putiobuf(nbp);
1165 
1166 			mutex_enter(&sc->sc_slock);
1167 			break;
1168 		}
1169 		if (error)
1170 			continue;
1171 
1172 		/*
1173 		 * Replace those parts that have been saved to backing store.
1174 		 */
1175 
1176 		addr = bp->b_data;
1177 		todo = bp->b_bcount;
1178 		for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) {
1179 			len = FSS_CLSIZE(sc)-off;
1180 			if (len > todo)
1181 				len = todo;
1182 			if (isclr(sc->sc_copied, c))
1183 				continue;
1184 			mutex_exit(&sc->sc_slock);
1185 
1186 			indirp = fss_bs_indir(sc, c);
1187 			if (indirp == NULL || *indirp == 0) {
1188 				/*
1189 				 * Not on backing store. Either in cache
1190 				 * or hole in the snapshotted block device.
1191 				 */
1192 
1193 				mutex_enter(&sc->sc_slock);
1194 				for (scp = sc->sc_cache; scp < scl; scp++)
1195 					if (scp->fc_type == FSS_CACHE_VALID &&
1196 					    scp->fc_cluster == c)
1197 						break;
1198 				if (scp < scl)
1199 					memcpy(addr, (char *)scp->fc_data+off,
1200 					    len);
1201 				else
1202 					memset(addr, 0, len);
1203 				continue;
1204 			}
1205 
1206 			/*
1207 			 * Read from backing store.
1208 			 */
1209 			error =
1210 			    fss_bs_io(sc, FSS_READ, *indirp, off, len, addr);
1211 
1212 			mutex_enter(&sc->sc_slock);
1213 			if (error) {
1214 				bp->b_resid = bp->b_bcount;
1215 				bp->b_error = error;
1216 				break;
1217 			}
1218 		}
1219 		mutex_exit(&sc->sc_slock);
1220 
1221 		disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read);
1222 		biodone(bp);
1223 
1224 		mutex_enter(&sc->sc_slock);
1225 	}
1226 }
1227 
1228 #ifdef _MODULE
1229 
1230 #include <sys/module.h>
1231 
1232 MODULE(MODULE_CLASS_DRIVER, fss, NULL);
1233 CFDRIVER_DECL(fss, DV_DISK, NULL);
1234 
1235 static int
1236 fss_modcmd(modcmd_t cmd, void *arg)
1237 {
1238 	int bmajor = -1, cmajor = -1,  error = 0;
1239 
1240 	switch (cmd) {
1241 	case MODULE_CMD_INIT:
1242 		mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE);
1243 		error = config_cfdriver_attach(&fss_cd);
1244 		if (error) {
1245 			mutex_destroy(&fss_device_lock);
1246 			break;
1247 		}
1248 		error = config_cfattach_attach(fss_cd.cd_name, &fss_ca);
1249 		if (error) {
1250 			config_cfdriver_detach(&fss_cd);
1251 			mutex_destroy(&fss_device_lock);
1252 			break;
1253 		}
1254 		error = devsw_attach(fss_cd.cd_name,
1255 		    &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor);
1256 		if (error) {
1257 			config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1258 			config_cfdriver_detach(&fss_cd);
1259 			mutex_destroy(&fss_device_lock);
1260 			break;
1261 		}
1262 		break;
1263 
1264 	case MODULE_CMD_FINI:
1265 		error = config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1266 		if (error)
1267 			break;
1268 		config_cfdriver_detach(&fss_cd);
1269 		devsw_detach(&fss_bdevsw, &fss_cdevsw);
1270 		mutex_destroy(&fss_device_lock);
1271 		break;
1272 
1273 	default:
1274 		error = ENOTTY;
1275 		break;
1276 	}
1277 
1278 	return error;
1279 }
1280 
1281 #endif /* _MODULE */
1282