FreeBSD/Linux Kernel Cross Reference
sys/dev/fss.c
1 /* $NetBSD: fss.c,v 1.60.4.3 2010/03/28 17:26:26 snj 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 1.60.4.3 2010/03/28 17:26:26 snj 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 struct nameidata nd;
584
585 /*
586 * Get the mounted file system.
587 */
588
589 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_mount);
590 if ((error = namei(&nd)) != 0)
591 return error;
592
593 if ((nd.ni_vp->v_vflag & VV_ROOT) != VV_ROOT) {
594 vrele(nd.ni_vp);
595 return EINVAL;
596 }
597
598 sc->sc_mount = nd.ni_vp->v_mount;
599 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN);
600
601 vrele(nd.ni_vp);
602
603 /*
604 * Check for file system internal snapshot.
605 */
606
607 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore);
608 if ((error = namei(&nd)) != 0)
609 return error;
610
611 if (nd.ni_vp->v_type == VREG && nd.ni_vp->v_mount == sc->sc_mount) {
612 sc->sc_flags |= FSS_PERSISTENT;
613 sc->sc_bs_vp = nd.ni_vp;
614
615 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
616 bits = sizeof(sc->sc_bs_bshift)*NBBY;
617 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits;
618 sc->sc_bs_bshift++)
619 if (FSS_FSBSIZE(sc) == fsbsize)
620 break;
621 if (sc->sc_bs_bshift >= bits) {
622 VOP_UNLOCK(sc->sc_bs_vp, 0);
623 return EINVAL;
624 }
625
626 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
627 sc->sc_clshift = 0;
628
629 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
630 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
631
632 VOP_UNLOCK(sc->sc_bs_vp, 0);
633
634 return error;
635 }
636 vput(nd.ni_vp);
637
638 /*
639 * Get the block device it is mounted on.
640 */
641
642 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE,
643 sc->sc_mount->mnt_stat.f_mntfromname);
644 if ((error = namei(&nd)) != 0)
645 return error;
646
647 if (nd.ni_vp->v_type != VBLK) {
648 vrele(nd.ni_vp);
649 return EINVAL;
650 }
651
652 sc->sc_bdev = nd.ni_vp->v_rdev;
653 vrele(nd.ni_vp);
654
655 /*
656 * Get the block device size.
657 */
658
659 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l);
660 if (error)
661 return error;
662
663 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size;
664
665 /*
666 * Get the backing store
667 */
668
669 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore);
670 if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0)
671 return error;
672 VOP_UNLOCK(nd.ni_vp, 0);
673
674 sc->sc_bs_vp = nd.ni_vp;
675
676 if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR)
677 return EINVAL;
678
679 if (sc->sc_bs_vp->v_type == VREG) {
680 error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred);
681 if (error != 0)
682 return error;
683 sc->sc_bs_size = va.va_size;
684 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize;
685 if (fsbsize & (fsbsize-1)) /* No power of two */
686 return EINVAL;
687 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32;
688 sc->sc_bs_bshift++)
689 if (FSS_FSBSIZE(sc) == fsbsize)
690 break;
691 if (sc->sc_bs_bshift >= 32)
692 return EINVAL;
693 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
694 } else {
695 sc->sc_bs_bshift = DEV_BSHIFT;
696 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
697 }
698
699 return 0;
700 }
701
702 /*
703 * Create a snapshot.
704 */
705 static int
706 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l)
707 {
708 int len, error;
709 u_int32_t csize;
710 off_t bsize;
711
712 bsize = 0; /* XXX gcc */
713
714 /*
715 * Open needed files.
716 */
717 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0)
718 goto bad;
719
720 if (sc->sc_flags & FSS_PERSISTENT) {
721 fss_softc_alloc(sc);
722 sc->sc_flags |= FSS_ACTIVE;
723 return 0;
724 }
725
726 /*
727 * Set cluster size. Must be a power of two and
728 * a multiple of backing store block size.
729 */
730 if (fss->fss_csize <= 0)
731 csize = MAXPHYS;
732 else
733 csize = fss->fss_csize;
734 if (bsize/csize > FSS_CLUSTER_MAX)
735 csize = bsize/FSS_CLUSTER_MAX+1;
736
737 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32;
738 sc->sc_clshift++)
739 if (FSS_CLSIZE(sc) >= csize)
740 break;
741 if (sc->sc_clshift >= 32) {
742 error = EINVAL;
743 goto bad;
744 }
745 sc->sc_clmask = FSS_CLSIZE(sc)-1;
746
747 /*
748 * Set number of cache slots.
749 */
750 if (FSS_CLSIZE(sc) <= 8192)
751 sc->sc_cache_size = 32;
752 else if (FSS_CLSIZE(sc) <= 65536)
753 sc->sc_cache_size = 8;
754 else
755 sc->sc_cache_size = 4;
756
757 /*
758 * Set number of clusters and size of last cluster.
759 */
760 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1;
761 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1;
762
763 /*
764 * Set size of indirect table.
765 */
766 len = sc->sc_clcount*sizeof(u_int32_t);
767 sc->sc_indir_size = FSS_BTOCL(sc, len)+1;
768 sc->sc_clnext = sc->sc_indir_size;
769 sc->sc_indir_cur = 0;
770
771 if ((error = fss_softc_alloc(sc)) != 0)
772 goto bad;
773
774 /*
775 * Activate the snapshot.
776 */
777
778 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0)
779 goto bad;
780
781 microtime(&sc->sc_time);
782
783 if (error == 0)
784 error = fscow_establish(sc->sc_mount,
785 fss_copy_on_write, sc);
786 if (error == 0)
787 sc->sc_flags |= FSS_ACTIVE;
788
789 vfs_resume(sc->sc_mount);
790
791 if (error != 0)
792 goto bad;
793
794 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname);
795 aprint_debug_dev(sc->sc_dev,
796 "%u clusters of %u, %u cache slots, %u indir clusters\n",
797 sc->sc_clcount, FSS_CLSIZE(sc),
798 sc->sc_cache_size, sc->sc_indir_size);
799
800 return 0;
801
802 bad:
803 fss_softc_free(sc);
804 if (sc->sc_bs_vp != NULL) {
805 if (sc->sc_flags & FSS_PERSISTENT)
806 vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
807 else
808 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
809 }
810 sc->sc_bs_vp = NULL;
811
812 return error;
813 }
814
815 /*
816 * Delete a snapshot.
817 */
818 static int
819 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l)
820 {
821
822 if ((sc->sc_flags & FSS_PERSISTENT) == 0)
823 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc);
824
825 mutex_enter(&sc->sc_slock);
826 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR);
827 sc->sc_mount = NULL;
828 sc->sc_bdev = NODEV;
829 mutex_exit(&sc->sc_slock);
830
831 fss_softc_free(sc);
832 if (sc->sc_flags & FSS_PERSISTENT)
833 vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
834 else
835 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
836 sc->sc_bs_vp = NULL;
837 sc->sc_flags &= ~FSS_PERSISTENT;
838
839 return 0;
840 }
841
842 /*
843 * Read a cluster from the snapshotted block device to the cache.
844 */
845 static int
846 fss_read_cluster(struct fss_softc *sc, u_int32_t cl)
847 {
848 int error, todo, offset, len;
849 daddr_t dblk;
850 struct buf *bp, *mbp;
851 struct fss_cache *scp, *scl;
852
853 /*
854 * Get a free cache slot.
855 */
856 scl = sc->sc_cache+sc->sc_cache_size;
857
858 mutex_enter(&sc->sc_slock);
859
860 restart:
861 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) {
862 mutex_exit(&sc->sc_slock);
863 return 0;
864 }
865
866 for (scp = sc->sc_cache; scp < scl; scp++)
867 if (scp->fc_cluster == cl) {
868 if (scp->fc_type == FSS_CACHE_VALID) {
869 mutex_exit(&sc->sc_slock);
870 return 0;
871 } else if (scp->fc_type == FSS_CACHE_BUSY) {
872 cv_wait(&scp->fc_state_cv, &sc->sc_slock);
873 goto restart;
874 }
875 }
876
877 for (scp = sc->sc_cache; scp < scl; scp++)
878 if (scp->fc_type == FSS_CACHE_FREE) {
879 scp->fc_type = FSS_CACHE_BUSY;
880 scp->fc_cluster = cl;
881 break;
882 }
883 if (scp >= scl) {
884 cv_wait(&sc->sc_cache_cv, &sc->sc_slock);
885 goto restart;
886 }
887
888 mutex_exit(&sc->sc_slock);
889
890 /*
891 * Start the read.
892 */
893 dblk = btodb(FSS_CLTOB(sc, cl));
894 if (cl == sc->sc_clcount-1) {
895 todo = sc->sc_clresid;
896 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo);
897 } else
898 todo = FSS_CLSIZE(sc);
899 offset = 0;
900 mbp = getiobuf(NULL, true);
901 mbp->b_bufsize = todo;
902 mbp->b_data = scp->fc_data;
903 mbp->b_resid = mbp->b_bcount = todo;
904 mbp->b_flags = B_READ;
905 mbp->b_cflags = BC_BUSY;
906 mbp->b_dev = sc->sc_bdev;
907 while (todo > 0) {
908 len = todo;
909 if (len > MAXPHYS)
910 len = MAXPHYS;
911 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo)
912 bp = mbp;
913 else {
914 bp = getiobuf(NULL, true);
915 nestiobuf_setup(mbp, bp, offset, len);
916 }
917 bp->b_lblkno = 0;
918 bp->b_blkno = dblk;
919 bdev_strategy(bp);
920 dblk += btodb(len);
921 offset += len;
922 todo -= len;
923 }
924 error = biowait(mbp);
925 putiobuf(mbp);
926
927 mutex_enter(&sc->sc_slock);
928 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID);
929 cv_broadcast(&scp->fc_state_cv);
930 if (error == 0) {
931 setbit(sc->sc_copied, scp->fc_cluster);
932 cv_signal(&sc->sc_work_cv);
933 }
934 mutex_exit(&sc->sc_slock);
935
936 return error;
937 }
938
939 /*
940 * Read/write clusters from/to backing store.
941 * For persistent snapshots must be called with cl == 0. off is the
942 * offset into the snapshot.
943 */
944 static int
945 fss_bs_io(struct fss_softc *sc, fss_io_type rw,
946 u_int32_t cl, off_t off, int len, void *data)
947 {
948 int error;
949
950 off += FSS_CLTOB(sc, cl);
951
952 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY);
953
954 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp,
955 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED,
956 sc->sc_bs_lwp->l_cred, NULL, NULL);
957 if (error == 0) {
958 mutex_enter(&sc->sc_bs_vp->v_interlock);
959 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off),
960 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
961 }
962
963 VOP_UNLOCK(sc->sc_bs_vp, 0);
964
965 return error;
966 }
967
968 /*
969 * Get a pointer to the indirect slot for this cluster.
970 */
971 static u_int32_t *
972 fss_bs_indir(struct fss_softc *sc, u_int32_t cl)
973 {
974 u_int32_t icl;
975 int ioff;
976
977 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t));
978 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t));
979
980 if (sc->sc_indir_cur == icl)
981 return &sc->sc_indir_data[ioff];
982
983 if (sc->sc_indir_dirty) {
984 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0,
985 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0)
986 return NULL;
987 setbit(sc->sc_indir_valid, sc->sc_indir_cur);
988 }
989
990 sc->sc_indir_dirty = 0;
991 sc->sc_indir_cur = icl;
992
993 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) {
994 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0,
995 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0)
996 return NULL;
997 } else
998 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc));
999
1000 return &sc->sc_indir_data[ioff];
1001 }
1002
1003 /*
1004 * The kernel thread (one for every active snapshot).
1005 *
1006 * After wakeup it cleans the cache and runs the I/O requests.
1007 */
1008 static void
1009 fss_bs_thread(void *arg)
1010 {
1011 bool thread_idle, is_valid;
1012 int error, i, todo, len, crotor, is_read;
1013 long off;
1014 char *addr;
1015 u_int32_t c, cl, ch, *indirp;
1016 struct buf *bp, *nbp;
1017 struct fss_softc *sc;
1018 struct fss_cache *scp, *scl;
1019
1020 sc = arg;
1021 scl = sc->sc_cache+sc->sc_cache_size;
1022 crotor = 0;
1023 thread_idle = false;
1024
1025 mutex_enter(&sc->sc_slock);
1026
1027 for (;;) {
1028 if (thread_idle)
1029 cv_wait(&sc->sc_work_cv, &sc->sc_slock);
1030 thread_idle = true;
1031 if ((sc->sc_flags & FSS_BS_THREAD) == 0) {
1032 sc->sc_bs_lwp = NULL;
1033 mutex_exit(&sc->sc_slock);
1034 kthread_exit(0);
1035 }
1036
1037 /*
1038 * Process I/O requests (persistent)
1039 */
1040
1041 if (sc->sc_flags & FSS_PERSISTENT) {
1042 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL)
1043 continue;
1044 is_valid = FSS_ISVALID(sc);
1045 is_read = (bp->b_flags & B_READ);
1046 thread_idle = false;
1047 mutex_exit(&sc->sc_slock);
1048
1049 if (is_valid) {
1050 disk_busy(sc->sc_dkdev);
1051 error = fss_bs_io(sc, FSS_READ, 0,
1052 dbtob(bp->b_blkno), bp->b_bcount,
1053 bp->b_data);
1054 disk_unbusy(sc->sc_dkdev,
1055 (error ? 0 : bp->b_bcount), is_read);
1056 } else
1057 error = ENXIO;
1058
1059 bp->b_error = error;
1060 bp->b_resid = (error ? bp->b_bcount : 0);
1061 biodone(bp);
1062
1063 mutex_enter(&sc->sc_slock);
1064 continue;
1065 }
1066
1067 /*
1068 * Clean the cache
1069 */
1070 for (i = 0; i < sc->sc_cache_size; i++) {
1071 crotor = (crotor + 1) % sc->sc_cache_size;
1072 scp = sc->sc_cache + crotor;
1073 if (scp->fc_type != FSS_CACHE_VALID)
1074 continue;
1075 mutex_exit(&sc->sc_slock);
1076
1077 thread_idle = false;
1078 indirp = fss_bs_indir(sc, scp->fc_cluster);
1079 if (indirp != NULL) {
1080 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext,
1081 0, FSS_CLSIZE(sc), scp->fc_data);
1082 } else
1083 error = EIO;
1084
1085 mutex_enter(&sc->sc_slock);
1086 if (error == 0) {
1087 *indirp = sc->sc_clnext++;
1088 sc->sc_indir_dirty = 1;
1089 } else
1090 fss_error(sc, "write error on backing store");
1091
1092 scp->fc_type = FSS_CACHE_FREE;
1093 cv_signal(&sc->sc_cache_cv);
1094 break;
1095 }
1096
1097 /*
1098 * Process I/O requests
1099 */
1100 if ((bp = BUFQ_GET(sc->sc_bufq)) == NULL)
1101 continue;
1102 is_valid = FSS_ISVALID(sc);
1103 is_read = (bp->b_flags & B_READ);
1104 thread_idle = false;
1105
1106 if (!is_valid) {
1107 mutex_exit(&sc->sc_slock);
1108
1109 bp->b_error = ENXIO;
1110 bp->b_resid = bp->b_bcount;
1111 biodone(bp);
1112
1113 mutex_enter(&sc->sc_slock);
1114 continue;
1115 }
1116
1117 disk_busy(sc->sc_dkdev);
1118
1119 /*
1120 * First read from the snapshotted block device unless
1121 * this request is completely covered by backing store.
1122 */
1123
1124 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno));
1125 off = FSS_CLOFF(sc, dbtob(bp->b_blkno));
1126 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1);
1127 error = 0;
1128 bp->b_resid = 0;
1129 bp->b_error = 0;
1130 for (c = cl; c <= ch; c++) {
1131 if (isset(sc->sc_copied, c))
1132 continue;
1133 mutex_exit(&sc->sc_slock);
1134
1135 /* Not on backing store, read from device. */
1136 nbp = getiobuf(NULL, true);
1137 nbp->b_flags = B_READ;
1138 nbp->b_resid = nbp->b_bcount = bp->b_bcount;
1139 nbp->b_bufsize = bp->b_bcount;
1140 nbp->b_data = bp->b_data;
1141 nbp->b_blkno = bp->b_blkno;
1142 nbp->b_lblkno = 0;
1143 nbp->b_dev = sc->sc_bdev;
1144 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */
1145
1146 bdev_strategy(nbp);
1147
1148 error = biowait(nbp);
1149 if (error != 0) {
1150 bp->b_resid = bp->b_bcount;
1151 bp->b_error = nbp->b_error;
1152 disk_unbusy(sc->sc_dkdev, 0, is_read);
1153 biodone(bp);
1154 }
1155 putiobuf(nbp);
1156
1157 mutex_enter(&sc->sc_slock);
1158 break;
1159 }
1160 if (error)
1161 continue;
1162
1163 /*
1164 * Replace those parts that have been saved to backing store.
1165 */
1166
1167 addr = bp->b_data;
1168 todo = bp->b_bcount;
1169 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) {
1170 len = FSS_CLSIZE(sc)-off;
1171 if (len > todo)
1172 len = todo;
1173 if (isclr(sc->sc_copied, c))
1174 continue;
1175 mutex_exit(&sc->sc_slock);
1176
1177 indirp = fss_bs_indir(sc, c);
1178 if (indirp == NULL || *indirp == 0) {
1179 /*
1180 * Not on backing store. Either in cache
1181 * or hole in the snapshotted block device.
1182 */
1183
1184 mutex_enter(&sc->sc_slock);
1185 for (scp = sc->sc_cache; scp < scl; scp++)
1186 if (scp->fc_type == FSS_CACHE_VALID &&
1187 scp->fc_cluster == c)
1188 break;
1189 if (scp < scl)
1190 memcpy(addr, (char *)scp->fc_data+off,
1191 len);
1192 else
1193 memset(addr, 0, len);
1194 continue;
1195 }
1196
1197 /*
1198 * Read from backing store.
1199 */
1200 error =
1201 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr);
1202
1203 mutex_enter(&sc->sc_slock);
1204 if (error) {
1205 bp->b_resid = bp->b_bcount;
1206 bp->b_error = error;
1207 break;
1208 }
1209 }
1210 mutex_exit(&sc->sc_slock);
1211
1212 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read);
1213 biodone(bp);
1214
1215 mutex_enter(&sc->sc_slock);
1216 }
1217 }
1218
1219 #ifdef _MODULE
1220
1221 #include <sys/module.h>
1222
1223 MODULE(MODULE_CLASS_DRIVER, fss, NULL);
1224 CFDRIVER_DECL(fss, DV_DISK, NULL);
1225
1226 static int
1227 fss_modcmd(modcmd_t cmd, void *arg)
1228 {
1229 int bmajor = -1, cmajor = -1, error = 0;
1230
1231 switch (cmd) {
1232 case MODULE_CMD_INIT:
1233 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE);
1234 error = config_cfdriver_attach(&fss_cd);
1235 if (error) {
1236 mutex_destroy(&fss_device_lock);
1237 break;
1238 }
1239 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca);
1240 if (error) {
1241 config_cfdriver_detach(&fss_cd);
1242 mutex_destroy(&fss_device_lock);
1243 break;
1244 }
1245 error = devsw_attach(fss_cd.cd_name,
1246 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor);
1247 if (error) {
1248 config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1249 config_cfdriver_detach(&fss_cd);
1250 mutex_destroy(&fss_device_lock);
1251 break;
1252 }
1253 break;
1254
1255 case MODULE_CMD_FINI:
1256 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca);
1257 if (error)
1258 break;
1259 config_cfdriver_detach(&fss_cd);
1260 devsw_detach(&fss_bdevsw, &fss_cdevsw);
1261 mutex_destroy(&fss_device_lock);
1262 break;
1263
1264 default:
1265 error = ENOTTY;
1266 break;
1267 }
1268
1269 return error;
1270 }
1271
1272 #endif /* _MODULE */
Cache object: 619d79cc849032bf0dc0e943b7f343d6
|