1 /*
2 * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved.
3 *
4 * Further information about snapshots can be obtained from:
5 *
6 * Marshall Kirk McKusick http://www.mckusick.com/softdep/
7 * 1614 Oxford Street mckusick@mckusick.com
8 * Berkeley, CA 94709-1608 +1-510-843-9542
9 * USA
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 *
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 *
21 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
22 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
23 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
25 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00
34 */
35
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: releng/5.2/sys/ufs/ffs/ffs_snapshot.c 122596 2003-11-13 03:56:32Z alc $");
38
39 #include <sys/param.h>
40 #include <sys/kernel.h>
41 #include <sys/systm.h>
42 #include <sys/conf.h>
43 #include <sys/bio.h>
44 #include <sys/buf.h>
45 #include <sys/proc.h>
46 #include <sys/namei.h>
47 #include <sys/sched.h>
48 #include <sys/stat.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/resource.h>
52 #include <sys/resourcevar.h>
53 #include <sys/vnode.h>
54
55 #include <ufs/ufs/extattr.h>
56 #include <ufs/ufs/quota.h>
57 #include <ufs/ufs/ufsmount.h>
58 #include <ufs/ufs/inode.h>
59 #include <ufs/ufs/ufs_extern.h>
60
61 #include <ufs/ffs/fs.h>
62 #include <ufs/ffs/ffs_extern.h>
63
64 #define KERNCRED thread0.td_ucred
65 #define DEBUG 1
66
67 static int cgaccount(int, struct vnode *, struct buf *, int);
68 static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
69 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
70 ufs_lbn_t, int), int);
71 static int indiracct_ufs1(struct vnode *, struct vnode *, int,
72 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
73 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
74 ufs_lbn_t, int), int);
75 static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
76 struct fs *, ufs_lbn_t, int);
77 static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
78 struct fs *, ufs_lbn_t, int);
79 static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
80 struct fs *, ufs_lbn_t, int);
81 static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
82 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
83 ufs_lbn_t, int), int);
84 static int indiracct_ufs2(struct vnode *, struct vnode *, int,
85 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
86 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
87 ufs_lbn_t, int), int);
88 static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
89 struct fs *, ufs_lbn_t, int);
90 static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
91 struct fs *, ufs_lbn_t, int);
92 static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
93 struct fs *, ufs_lbn_t, int);
94 static int ffs_copyonwrite(struct vnode *, struct buf *);
95 static int readblock(struct buf *, ufs2_daddr_t);
96
97 /*
98 * To ensure the consistency of snapshots across crashes, we must
99 * synchronously write out copied blocks before allowing the
100 * originals to be modified. Because of the rather severe speed
101 * penalty that this imposes, the following flag allows this
102 * crash persistence to be disabled.
103 */
104 int dopersistence = 0;
105
106 #ifdef DEBUG
107 #include <sys/sysctl.h>
108 SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, "");
109 static int snapdebug = 0;
110 SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
111 int collectsnapstats = 0;
112 SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats,
113 0, "");
114 #endif /* DEBUG */
115
116 /*
117 * Create a snapshot file and initialize it for the filesystem.
118 */
119 int
120 ffs_snapshot(mp, snapfile)
121 struct mount *mp;
122 char *snapfile;
123 {
124 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist;
125 int error, cg, snaploc;
126 int i, size, len, loc;
127 int flag = mp->mnt_flag;
128 struct timespec starttime = {0, 0}, endtime;
129 char saved_nice = 0;
130 long redo = 0, snaplistsize = 0;
131 int32_t *lp;
132 void *space;
133 struct fs *copy_fs = NULL, *fs = VFSTOUFS(mp)->um_fs;
134 struct snaphead *snaphead;
135 struct thread *td = curthread;
136 struct inode *ip, *xp;
137 struct buf *bp, *nbp, *ibp, *sbp = NULL;
138 struct nameidata nd;
139 struct mount *wrtmp;
140 struct vattr vat;
141 struct vnode *vp, *xvp, *nvp, *devvp;
142 struct uio auio;
143 struct iovec aiov;
144
145 /*
146 * Need to serialize access to snapshot code per filesystem.
147 */
148 /*
149 * Assign a snapshot slot in the superblock.
150 */
151 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
152 if (fs->fs_snapinum[snaploc] == 0)
153 break;
154 if (snaploc == FSMAXSNAP)
155 return (ENOSPC);
156 /*
157 * Create the snapshot file.
158 */
159 restart:
160 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_USERSPACE, snapfile, td);
161 if ((error = namei(&nd)) != 0)
162 return (error);
163 if (nd.ni_vp != NULL) {
164 vput(nd.ni_vp);
165 error = EEXIST;
166 }
167 if (nd.ni_dvp->v_mount != mp)
168 error = EXDEV;
169 if (error) {
170 NDFREE(&nd, NDF_ONLY_PNBUF);
171 if (nd.ni_dvp == nd.ni_vp)
172 vrele(nd.ni_dvp);
173 else
174 vput(nd.ni_dvp);
175 return (error);
176 }
177 VATTR_NULL(&vat);
178 vat.va_type = VREG;
179 vat.va_mode = S_IRUSR;
180 vat.va_vaflags |= VA_EXCLUSIVE;
181 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
182 wrtmp = NULL;
183 if (wrtmp != mp)
184 panic("ffs_snapshot: mount mismatch");
185 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
186 NDFREE(&nd, NDF_ONLY_PNBUF);
187 vput(nd.ni_dvp);
188 if ((error = vn_start_write(NULL, &wrtmp,
189 V_XSLEEP | PCATCH)) != 0)
190 return (error);
191 goto restart;
192 }
193 VOP_LEASE(nd.ni_dvp, td, KERNCRED, LEASE_WRITE);
194 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
195 vput(nd.ni_dvp);
196 if (error) {
197 NDFREE(&nd, NDF_ONLY_PNBUF);
198 vn_finished_write(wrtmp);
199 return (error);
200 }
201 vp = nd.ni_vp;
202 ip = VTOI(vp);
203 devvp = ip->i_devvp;
204 /*
205 * Allocate and copy the last block contents so as to be able
206 * to set size to that of the filesystem.
207 */
208 numblks = howmany(fs->fs_size, fs->fs_frag);
209 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
210 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
211 if (error)
212 goto out;
213 ip->i_size = lblktosize(fs, (off_t)numblks);
214 DIP(ip, i_size) = ip->i_size;
215 ip->i_flag |= IN_CHANGE | IN_UPDATE;
216 if ((error = readblock(bp, numblks - 1)) != 0)
217 goto out;
218 bawrite(bp);
219 /*
220 * Preallocate critical data structures so that we can copy
221 * them in without further allocation after we suspend all
222 * operations on the filesystem. We would like to just release
223 * the allocated buffers without writing them since they will
224 * be filled in below once we are ready to go, but this upsets
225 * the soft update code, so we go ahead and write the new buffers.
226 *
227 * Allocate all indirect blocks and mark all of them as not
228 * needing to be copied.
229 */
230 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
231 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
232 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp);
233 if (error)
234 goto out;
235 bawrite(ibp);
236 }
237 /*
238 * Allocate copies for the superblock and its summary information.
239 */
240 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
241 0, &nbp);
242 if (error)
243 goto out;
244 bawrite(nbp);
245 blkno = fragstoblks(fs, fs->fs_csaddr);
246 len = howmany(fs->fs_cssize, fs->fs_bsize);
247 for (loc = 0; loc < len; loc++) {
248 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
249 fs->fs_bsize, KERNCRED, 0, &nbp);
250 if (error)
251 goto out;
252 bawrite(nbp);
253 }
254 /*
255 * Allocate all cylinder group blocks.
256 */
257 for (cg = 0; cg < fs->fs_ncg; cg++) {
258 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
259 fs->fs_bsize, KERNCRED, 0, &nbp);
260 if (error)
261 goto out;
262 bawrite(nbp);
263 }
264 /*
265 * Copy all the cylinder group maps. Although the
266 * filesystem is still active, we hope that only a few
267 * cylinder groups will change between now and when we
268 * suspend operations. Thus, we will be able to quickly
269 * touch up the few cylinder groups that changed during
270 * the suspension period.
271 */
272 len = howmany(fs->fs_ncg, NBBY);
273 MALLOC(fs->fs_active, int *, len, M_DEVBUF, M_WAITOK);
274 bzero(fs->fs_active, len);
275 for (cg = 0; cg < fs->fs_ncg; cg++) {
276 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
277 fs->fs_bsize, KERNCRED, 0, &nbp);
278 if (error)
279 goto out;
280 error = cgaccount(cg, vp, nbp, 1);
281 bawrite(nbp);
282 if (error)
283 goto out;
284 }
285 /*
286 * Change inode to snapshot type file.
287 */
288 ip->i_flags |= SF_SNAPSHOT;
289 DIP(ip, i_flags) = ip->i_flags;
290 ip->i_flag |= IN_CHANGE | IN_UPDATE;
291 /*
292 * Ensure that the snapshot is completely on disk.
293 * Since we have marked it as a snapshot it is safe to
294 * unlock it as no process will be allowed to write to it.
295 */
296 if ((error = VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td)) != 0)
297 goto out;
298 VOP_UNLOCK(vp, 0, td);
299 /*
300 * All allocations are done, so we can now snapshot the system.
301 *
302 * Recind nice scheduling while running with the filesystem suspended.
303 */
304 if (td->td_ksegrp->kg_nice > 0) {
305 PROC_LOCK(td->td_proc);
306 mtx_lock_spin(&sched_lock);
307 saved_nice = td->td_ksegrp->kg_nice;
308 sched_nice(td->td_ksegrp, 0);
309 mtx_unlock_spin(&sched_lock);
310 PROC_UNLOCK(td->td_proc);
311 }
312 /*
313 * Suspend operation on filesystem.
314 */
315 for (;;) {
316 vn_finished_write(wrtmp);
317 if ((error = vfs_write_suspend(vp->v_mount)) != 0) {
318 vn_start_write(NULL, &wrtmp, V_WAIT);
319 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
320 goto out;
321 }
322 if (mp->mnt_kern_flag & MNTK_SUSPENDED)
323 break;
324 vn_start_write(NULL, &wrtmp, V_WAIT);
325 }
326 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
327 if (collectsnapstats)
328 nanotime(&starttime);
329 /*
330 * First, copy all the cylinder group maps that have changed.
331 */
332 for (cg = 0; cg < fs->fs_ncg; cg++) {
333 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0)
334 continue;
335 redo++;
336 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
337 fs->fs_bsize, KERNCRED, 0, &nbp);
338 if (error)
339 goto out1;
340 error = cgaccount(cg, vp, nbp, 2);
341 bawrite(nbp);
342 if (error)
343 goto out1;
344 }
345 /*
346 * Grab a copy of the superblock and its summary information.
347 * We delay writing it until the suspension is released below.
348 */
349 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
350 KERNCRED, &sbp);
351 if (error) {
352 brelse(sbp);
353 sbp = NULL;
354 goto out1;
355 }
356 loc = blkoff(fs, fs->fs_sblockloc);
357 copy_fs = (struct fs *)(sbp->b_data + loc);
358 bcopy(fs, copy_fs, fs->fs_sbsize);
359 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
360 copy_fs->fs_clean = 1;
361 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
362 if (fs->fs_sbsize < size)
363 bzero(&sbp->b_data[loc + fs->fs_sbsize], size - fs->fs_sbsize);
364 size = blkroundup(fs, fs->fs_cssize);
365 if (fs->fs_contigsumsize > 0)
366 size += fs->fs_ncg * sizeof(int32_t);
367 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
368 copy_fs->fs_csp = space;
369 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
370 (char *)space += fs->fs_cssize;
371 loc = howmany(fs->fs_cssize, fs->fs_fsize);
372 i = fs->fs_frag - loc % fs->fs_frag;
373 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
374 if (len > 0) {
375 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
376 len, KERNCRED, &bp)) != 0) {
377 brelse(bp);
378 free(copy_fs->fs_csp, M_UFSMNT);
379 bawrite(sbp);
380 sbp = NULL;
381 goto out1;
382 }
383 bcopy(bp->b_data, space, (u_int)len);
384 (char *)space += len;
385 bp->b_flags |= B_INVAL | B_NOCACHE;
386 brelse(bp);
387 }
388 if (fs->fs_contigsumsize > 0) {
389 copy_fs->fs_maxcluster = lp = space;
390 for (i = 0; i < fs->fs_ncg; i++)
391 *lp++ = fs->fs_contigsumsize;
392 }
393 /*
394 * We must check for active files that have been unlinked
395 * (e.g., with a zero link count). We have to expunge all
396 * trace of these files from the snapshot so that they are
397 * not reclaimed prematurely by fsck or unnecessarily dumped.
398 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
399 * spec_strategy about writing on a suspended filesystem.
400 * Note that we skip unlinked snapshot files as they will
401 * be handled separately below.
402 *
403 * We also calculate the needed size for the snapshot list.
404 */
405 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
406 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
407 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
408 MNT_ILOCK(mp);
409 loop:
410 for (xvp = TAILQ_FIRST(&mp->mnt_nvnodelist); xvp; xvp = nvp) {
411 /*
412 * Make sure this vnode wasn't reclaimed in getnewvnode().
413 * Start over if it has (it won't be on the list anymore).
414 */
415 if (xvp->v_mount != mp)
416 goto loop;
417 nvp = TAILQ_NEXT(xvp, v_nmntvnodes);
418 VI_LOCK(xvp);
419 MNT_IUNLOCK(mp);
420 if ((xvp->v_iflag & VI_XLOCK) ||
421 xvp->v_usecount == 0 || xvp->v_type == VNON ||
422 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
423 VI_UNLOCK(xvp);
424 MNT_ILOCK(mp);
425 continue;
426 }
427 if (snapdebug)
428 vprint("ffs_snapshot: busy vnode", xvp);
429 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK, td) != 0) {
430 MNT_ILOCK(mp);
431 goto loop;
432 }
433 if (VOP_GETATTR(xvp, &vat, td->td_ucred, td) == 0 &&
434 vat.va_nlink > 0) {
435 VOP_UNLOCK(xvp, 0, td);
436 MNT_ILOCK(mp);
437 continue;
438 }
439 xp = VTOI(xvp);
440 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
441 VOP_UNLOCK(xvp, 0, td);
442 MNT_ILOCK(mp);
443 continue;
444 }
445 /*
446 * If there is a fragment, clear it here.
447 */
448 blkno = 0;
449 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
450 if (loc < NDADDR) {
451 len = fragroundup(fs, blkoff(fs, xp->i_size));
452 if (len < fs->fs_bsize) {
453 ffs_blkfree(copy_fs, vp, DIP(xp, i_db[loc]),
454 len, xp->i_number);
455 blkno = DIP(xp, i_db[loc]);
456 DIP(xp, i_db[loc]) = 0;
457 }
458 }
459 snaplistsize += 1;
460 if (xp->i_ump->um_fstype == UFS1)
461 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
462 BLK_NOCOPY);
463 else
464 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
465 BLK_NOCOPY);
466 if (blkno)
467 DIP(xp, i_db[loc]) = blkno;
468 if (!error)
469 error = ffs_freefile(copy_fs, vp, xp->i_number,
470 xp->i_mode);
471 VOP_UNLOCK(xvp, 0, td);
472 if (error) {
473 free(copy_fs->fs_csp, M_UFSMNT);
474 bawrite(sbp);
475 sbp = NULL;
476 goto out1;
477 }
478 MNT_ILOCK(mp);
479 }
480 MNT_IUNLOCK(mp);
481 /*
482 * If there already exist snapshots on this filesystem, grab a
483 * reference to their shared lock. If this is the first snapshot
484 * on this filesystem, we need to allocate a lock for the snapshots
485 * to share. In either case, acquire the snapshot lock and give
486 * up our original private lock.
487 */
488 VI_LOCK(devvp);
489 snaphead = &devvp->v_rdev->si_snapshots;
490 if ((xp = TAILQ_FIRST(snaphead)) != NULL) {
491 VI_LOCK(vp);
492 vp->v_vnlock = ITOV(xp)->v_vnlock;
493 VI_UNLOCK(devvp);
494 } else {
495 struct lock *lkp;
496
497 VI_UNLOCK(devvp);
498 MALLOC(lkp, struct lock *, sizeof(struct lock), M_UFSMNT,
499 M_WAITOK);
500 lockinit(lkp, PVFS, "snaplk", VLKTIMEOUT,
501 LK_CANRECURSE | LK_NOPAUSE);
502 VI_LOCK(vp);
503 vp->v_vnlock = lkp;
504 }
505 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
506 transferlockers(&vp->v_lock, vp->v_vnlock);
507 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
508 /*
509 * If this is the first snapshot on this filesystem, then we need
510 * to allocate the space for the list of preallocated snapshot blocks.
511 * This list will be refined below, but this preliminary one will
512 * keep us out of deadlock until the full one is ready.
513 */
514 if (xp == NULL) {
515 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
516 M_UFSMNT, M_WAITOK);
517 blkp = &snapblklist[1];
518 *blkp++ = lblkno(fs, fs->fs_sblockloc);
519 blkno = fragstoblks(fs, fs->fs_csaddr);
520 for (cg = 0; cg < fs->fs_ncg; cg++) {
521 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
522 break;
523 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
524 }
525 len = howmany(fs->fs_cssize, fs->fs_bsize);
526 for (loc = 0; loc < len; loc++)
527 *blkp++ = blkno + loc;
528 for (; cg < fs->fs_ncg; cg++)
529 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
530 snapblklist[0] = blkp - snapblklist;
531 VI_LOCK(devvp);
532 if (devvp->v_rdev->si_snapblklist != NULL)
533 panic("ffs_snapshot: non-empty list");
534 devvp->v_rdev->si_snapblklist = snapblklist;
535 devvp->v_rdev->si_snaplistsize = blkp - snapblklist;
536 VI_UNLOCK(devvp);
537 }
538 /*
539 * Record snapshot inode. Since this is the newest snapshot,
540 * it must be placed at the end of the list.
541 */
542 VI_LOCK(devvp);
543 fs->fs_snapinum[snaploc] = ip->i_number;
544 if (ip->i_nextsnap.tqe_prev != 0)
545 panic("ffs_snapshot: %d already on list", ip->i_number);
546 TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap);
547 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
548 devvp->v_vflag |= VV_COPYONWRITE;
549 VI_UNLOCK(devvp);
550 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
551 vp->v_vflag |= VV_SYSTEM;
552 out1:
553 /*
554 * Resume operation on filesystem.
555 */
556 vfs_write_resume(vp->v_mount);
557 vn_start_write(NULL, &wrtmp, V_WAIT);
558 if (collectsnapstats && starttime.tv_sec > 0) {
559 nanotime(&endtime);
560 timespecsub(&endtime, &starttime);
561 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
562 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
563 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
564 }
565 if (sbp == NULL)
566 goto out;
567 /*
568 * Copy allocation information from all the snapshots in
569 * this snapshot and then expunge them from its view.
570 */
571 snaphead = &devvp->v_rdev->si_snapshots;
572 TAILQ_FOREACH(xp, snaphead, i_nextsnap) {
573 if (xp == ip)
574 break;
575 if (xp->i_ump->um_fstype == UFS1)
576 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
577 BLK_SNAP);
578 else
579 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
580 BLK_SNAP);
581 if (error) {
582 fs->fs_snapinum[snaploc] = 0;
583 goto done;
584 }
585 }
586 /*
587 * Allocate space for the full list of preallocated snapshot blocks.
588 */
589 MALLOC(snapblklist, daddr_t *, snaplistsize * sizeof(daddr_t),
590 M_UFSMNT, M_WAITOK);
591 ip->i_snapblklist = &snapblklist[1];
592 /*
593 * Expunge the blocks used by the snapshots from the set of
594 * blocks marked as used in the snapshot bitmaps. Also, collect
595 * the list of allocated blocks in i_snapblklist.
596 */
597 if (ip->i_ump->um_fstype == UFS1)
598 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP);
599 else
600 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP);
601 if (error) {
602 fs->fs_snapinum[snaploc] = 0;
603 FREE(snapblklist, M_UFSMNT);
604 goto done;
605 }
606 if (snaplistsize < ip->i_snapblklist - snapblklist)
607 panic("ffs_snapshot: list too small");
608 snaplistsize = ip->i_snapblklist - snapblklist;
609 snapblklist[0] = snaplistsize;
610 ip->i_snapblklist = 0;
611 /*
612 * Write out the list of allocated blocks to the end of the snapshot.
613 */
614 auio.uio_iov = &aiov;
615 auio.uio_iovcnt = 1;
616 aiov.iov_base = (void *)snapblklist;
617 aiov.iov_len = snaplistsize * sizeof(daddr_t);
618 auio.uio_resid = aiov.iov_len;;
619 auio.uio_offset = ip->i_size;
620 auio.uio_segflg = UIO_SYSSPACE;
621 auio.uio_rw = UIO_WRITE;
622 auio.uio_td = td;
623 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
624 fs->fs_snapinum[snaploc] = 0;
625 FREE(snapblklist, M_UFSMNT);
626 goto done;
627 }
628 /*
629 * Write the superblock and its summary information
630 * to the snapshot.
631 */
632 blkno = fragstoblks(fs, fs->fs_csaddr);
633 len = howmany(fs->fs_cssize, fs->fs_bsize);
634 space = copy_fs->fs_csp;
635 for (loc = 0; loc < len; loc++) {
636 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
637 if (error) {
638 brelse(nbp);
639 fs->fs_snapinum[snaploc] = 0;
640 FREE(snapblklist, M_UFSMNT);
641 goto done;
642 }
643 bcopy(space, nbp->b_data, fs->fs_bsize);
644 space = (char *)space + fs->fs_bsize;
645 bawrite(nbp);
646 }
647 /*
648 * As this is the newest list, it is the most inclusive, so
649 * should replace the previous list.
650 */
651 VI_LOCK(devvp);
652 space = devvp->v_rdev->si_snapblklist;
653 devvp->v_rdev->si_snapblklist = snapblklist;
654 devvp->v_rdev->si_snaplistsize = snaplistsize;
655 VI_UNLOCK(devvp);
656 if (space != NULL)
657 FREE(space, M_UFSMNT);
658 done:
659 free(copy_fs->fs_csp, M_UFSMNT);
660 bawrite(sbp);
661 out:
662 if (saved_nice > 0) {
663 PROC_LOCK(td->td_proc);
664 mtx_lock_spin(&sched_lock);
665 sched_nice(td->td_ksegrp, saved_nice);
666 mtx_unlock_spin(&sched_lock);
667 PROC_UNLOCK(td->td_proc);
668 }
669 if (fs->fs_active != 0) {
670 FREE(fs->fs_active, M_DEVBUF);
671 fs->fs_active = 0;
672 }
673 mp->mnt_flag = flag;
674 if (error)
675 (void) UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
676 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
677 if (error)
678 vput(vp);
679 else
680 VOP_UNLOCK(vp, 0, td);
681 vn_finished_write(wrtmp);
682 return (error);
683 }
684
685 /*
686 * Copy a cylinder group map. All the unallocated blocks are marked
687 * BLK_NOCOPY so that the snapshot knows that it need not copy them
688 * if they are later written. If passno is one, then this is a first
689 * pass, so only setting needs to be done. If passno is 2, then this
690 * is a revision to a previous pass which must be undone as the
691 * replacement pass is done.
692 */
693 static int
694 cgaccount(cg, vp, nbp, passno)
695 int cg;
696 struct vnode *vp;
697 struct buf *nbp;
698 int passno;
699 {
700 struct buf *bp, *ibp;
701 struct inode *ip;
702 struct cg *cgp;
703 struct fs *fs;
704 ufs2_daddr_t base, numblks;
705 int error, len, loc, indiroff;
706
707 ip = VTOI(vp);
708 fs = ip->i_fs;
709 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
710 (int)fs->fs_cgsize, KERNCRED, &bp);
711 if (error) {
712 brelse(bp);
713 return (error);
714 }
715 cgp = (struct cg *)bp->b_data;
716 if (!cg_chkmagic(cgp)) {
717 brelse(bp);
718 return (EIO);
719 }
720 atomic_set_int(&ACTIVECGNUM(fs, cg), ACTIVECGOFF(cg));
721 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
722 if (fs->fs_cgsize < fs->fs_bsize)
723 bzero(&nbp->b_data[fs->fs_cgsize],
724 fs->fs_bsize - fs->fs_cgsize);
725 if (passno == 2)
726 nbp->b_flags |= B_VALIDSUSPWRT;
727 numblks = howmany(fs->fs_size, fs->fs_frag);
728 len = howmany(fs->fs_fpg, fs->fs_frag);
729 base = cg * fs->fs_fpg / fs->fs_frag;
730 if (base + len >= numblks)
731 len = numblks - base - 1;
732 loc = 0;
733 if (base < NDADDR) {
734 for ( ; loc < NDADDR; loc++) {
735 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
736 DIP(ip, i_db[loc]) = BLK_NOCOPY;
737 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
738 DIP(ip, i_db[loc]) = 0;
739 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
740 panic("ffs_snapshot: lost direct block");
741 }
742 }
743 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
744 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
745 if (error) {
746 brelse(bp);
747 return (error);
748 }
749 indiroff = (base + loc - NDADDR) % NINDIR(fs);
750 for ( ; loc < len; loc++, indiroff++) {
751 if (indiroff >= NINDIR(fs)) {
752 if (passno == 2)
753 ibp->b_flags |= B_VALIDSUSPWRT;
754 bawrite(ibp);
755 error = UFS_BALLOC(vp,
756 lblktosize(fs, (off_t)(base + loc)),
757 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
758 if (error) {
759 brelse(bp);
760 return (error);
761 }
762 indiroff = 0;
763 }
764 if (ip->i_ump->um_fstype == UFS1) {
765 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
766 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
767 BLK_NOCOPY;
768 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
769 [indiroff] == BLK_NOCOPY)
770 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
771 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
772 [indiroff] == BLK_NOCOPY)
773 panic("ffs_snapshot: lost indirect block");
774 continue;
775 }
776 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
777 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
778 else if (passno == 2 &&
779 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
780 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
781 else if (passno == 1 &&
782 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
783 panic("ffs_snapshot: lost indirect block");
784 }
785 bqrelse(bp);
786 if (passno == 2)
787 ibp->b_flags |= B_VALIDSUSPWRT;
788 bdwrite(ibp);
789 return (0);
790 }
791
792 /*
793 * Before expunging a snapshot inode, note all the
794 * blocks that it claims with BLK_SNAP so that fsck will
795 * be able to account for those blocks properly and so
796 * that this snapshot knows that it need not copy them
797 * if the other snapshot holding them is freed. This code
798 * is reproduced once each for UFS1 and UFS2.
799 */
800 static int
801 expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype)
802 struct vnode *snapvp;
803 struct inode *cancelip;
804 struct fs *fs;
805 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
806 struct fs *, ufs_lbn_t, int);
807 int expungetype;
808 {
809 int i, error, indiroff;
810 ufs_lbn_t lbn, rlbn;
811 ufs2_daddr_t len, blkno, numblks, blksperindir;
812 struct ufs1_dinode *dip;
813 struct thread *td = curthread;
814 struct buf *bp;
815
816 /*
817 * Prepare to expunge the inode. If its inode block has not
818 * yet been copied, then allocate and fill the copy.
819 */
820 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
821 blkno = 0;
822 if (lbn < NDADDR) {
823 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
824 } else {
825 td->td_pflags |= TDP_COWINPROGRESS;
826 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
827 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
828 td->td_pflags &= ~TDP_COWINPROGRESS;
829 if (error)
830 return (error);
831 indiroff = (lbn - NDADDR) % NINDIR(fs);
832 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
833 bqrelse(bp);
834 }
835 if (blkno != 0) {
836 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
837 return (error);
838 } else {
839 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
840 fs->fs_bsize, KERNCRED, 0, &bp);
841 if (error)
842 return (error);
843 if ((error = readblock(bp, lbn)) != 0)
844 return (error);
845 }
846 /*
847 * Set a snapshot inode to be a zero length file, regular files
848 * to be completely unallocated.
849 */
850 dip = (struct ufs1_dinode *)bp->b_data +
851 ino_to_fsbo(fs, cancelip->i_number);
852 if (expungetype == BLK_NOCOPY)
853 dip->di_mode = 0;
854 dip->di_size = 0;
855 dip->di_blocks = 0;
856 dip->di_flags &= ~SF_SNAPSHOT;
857 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
858 bdwrite(bp);
859 /*
860 * Now go through and expunge all the blocks in the file
861 * using the function requested.
862 */
863 numblks = howmany(cancelip->i_size, fs->fs_bsize);
864 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
865 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
866 return (error);
867 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
868 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
869 return (error);
870 blksperindir = 1;
871 lbn = -NDADDR;
872 len = numblks - NDADDR;
873 rlbn = NDADDR;
874 for (i = 0; len > 0 && i < NIADDR; i++) {
875 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
876 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
877 blksperindir, fs, acctfunc, expungetype);
878 if (error)
879 return (error);
880 blksperindir *= NINDIR(fs);
881 lbn -= blksperindir + 1;
882 len -= blksperindir;
883 rlbn += blksperindir;
884 }
885 return (0);
886 }
887
888 /*
889 * Descend an indirect block chain for vnode cancelvp accounting for all
890 * its indirect blocks in snapvp.
891 */
892 static int
893 indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
894 blksperindir, fs, acctfunc, expungetype)
895 struct vnode *snapvp;
896 struct vnode *cancelvp;
897 int level;
898 ufs1_daddr_t blkno;
899 ufs_lbn_t lbn;
900 ufs_lbn_t rlbn;
901 ufs_lbn_t remblks;
902 ufs_lbn_t blksperindir;
903 struct fs *fs;
904 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
905 struct fs *, ufs_lbn_t, int);
906 int expungetype;
907 {
908 int error, num, i;
909 ufs_lbn_t subblksperindir;
910 struct indir indirs[NIADDR + 2];
911 ufs1_daddr_t last, *bap;
912 struct buf *bp;
913
914 if (blkno == 0) {
915 if (expungetype == BLK_NOCOPY)
916 return (0);
917 panic("indiracct_ufs1: missing indir");
918 }
919 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
920 return (error);
921 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
922 panic("indiracct_ufs1: botched params");
923 /*
924 * We have to expand bread here since it will deadlock looking
925 * up the block number for any blocks that are not in the cache.
926 */
927 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
928 bp->b_blkno = fsbtodb(fs, blkno);
929 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
930 (error = readblock(bp, fragstoblks(fs, blkno)))) {
931 brelse(bp);
932 return (error);
933 }
934 /*
935 * Account for the block pointers in this indirect block.
936 */
937 last = howmany(remblks, blksperindir);
938 if (last > NINDIR(fs))
939 last = NINDIR(fs);
940 MALLOC(bap, ufs1_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
941 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
942 bqrelse(bp);
943 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
944 level == 0 ? rlbn : -1, expungetype);
945 if (error || level == 0)
946 goto out;
947 /*
948 * Account for the block pointers in each of the indirect blocks
949 * in the levels below us.
950 */
951 subblksperindir = blksperindir / NINDIR(fs);
952 for (lbn++, level--, i = 0; i < last; i++) {
953 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
954 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
955 if (error)
956 goto out;
957 rlbn += blksperindir;
958 lbn -= blksperindir;
959 remblks -= blksperindir;
960 }
961 out:
962 FREE(bap, M_DEVBUF);
963 return (error);
964 }
965
966 /*
967 * Do both snap accounting and map accounting.
968 */
969 static int
970 fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
971 struct vnode *vp;
972 ufs1_daddr_t *oldblkp, *lastblkp;
973 struct fs *fs;
974 ufs_lbn_t lblkno;
975 int exptype; /* BLK_SNAP or BLK_NOCOPY */
976 {
977 int error;
978
979 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
980 return (error);
981 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
982 }
983
984 /*
985 * Identify a set of blocks allocated in a snapshot inode.
986 */
987 static int
988 snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
989 struct vnode *vp;
990 ufs1_daddr_t *oldblkp, *lastblkp;
991 struct fs *fs;
992 ufs_lbn_t lblkno;
993 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
994 {
995 struct inode *ip = VTOI(vp);
996 ufs1_daddr_t blkno, *blkp;
997 ufs_lbn_t lbn;
998 struct buf *ibp;
999 int error;
1000
1001 for ( ; oldblkp < lastblkp; oldblkp++) {
1002 blkno = *oldblkp;
1003 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1004 continue;
1005 lbn = fragstoblks(fs, blkno);
1006 if (lbn < NDADDR) {
1007 blkp = &ip->i_din1->di_db[lbn];
1008 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1009 } else {
1010 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1011 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1012 if (error)
1013 return (error);
1014 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1015 [(lbn - NDADDR) % NINDIR(fs)];
1016 }
1017 /*
1018 * If we are expunging a snapshot vnode and we
1019 * find a block marked BLK_NOCOPY, then it is
1020 * one that has been allocated to this snapshot after
1021 * we took our current snapshot and can be ignored.
1022 */
1023 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1024 if (lbn >= NDADDR)
1025 brelse(ibp);
1026 } else {
1027 if (*blkp != 0)
1028 panic("snapacct_ufs1: bad block");
1029 *blkp = expungetype;
1030 if (lbn >= NDADDR)
1031 bdwrite(ibp);
1032 }
1033 }
1034 return (0);
1035 }
1036
1037 /*
1038 * Account for a set of blocks allocated in a snapshot inode.
1039 */
1040 static int
1041 mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1042 struct vnode *vp;
1043 ufs1_daddr_t *oldblkp, *lastblkp;
1044 struct fs *fs;
1045 ufs_lbn_t lblkno;
1046 int expungetype;
1047 {
1048 ufs1_daddr_t blkno;
1049 struct inode *ip;
1050 ino_t inum;
1051 int acctit;
1052
1053 ip = VTOI(vp);
1054 inum = ip->i_number;
1055 if (lblkno == -1)
1056 acctit = 0;
1057 else
1058 acctit = 1;
1059 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1060 blkno = *oldblkp;
1061 if (blkno == 0 || blkno == BLK_NOCOPY)
1062 continue;
1063 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1064 *ip->i_snapblklist++ = lblkno;
1065 if (blkno == BLK_SNAP)
1066 blkno = blkstofrags(fs, lblkno);
1067 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1068 }
1069 return (0);
1070 }
1071
1072 /*
1073 * Before expunging a snapshot inode, note all the
1074 * blocks that it claims with BLK_SNAP so that fsck will
1075 * be able to account for those blocks properly and so
1076 * that this snapshot knows that it need not copy them
1077 * if the other snapshot holding them is freed. This code
1078 * is reproduced once each for UFS1 and UFS2.
1079 */
1080 static int
1081 expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype)
1082 struct vnode *snapvp;
1083 struct inode *cancelip;
1084 struct fs *fs;
1085 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1086 struct fs *, ufs_lbn_t, int);
1087 int expungetype;
1088 {
1089 int i, error, indiroff;
1090 ufs_lbn_t lbn, rlbn;
1091 ufs2_daddr_t len, blkno, numblks, blksperindir;
1092 struct ufs2_dinode *dip;
1093 struct thread *td = curthread;
1094 struct buf *bp;
1095
1096 /*
1097 * Prepare to expunge the inode. If its inode block has not
1098 * yet been copied, then allocate and fill the copy.
1099 */
1100 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1101 blkno = 0;
1102 if (lbn < NDADDR) {
1103 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1104 } else {
1105 td->td_pflags |= TDP_COWINPROGRESS;
1106 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1107 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1108 td->td_pflags &= ~TDP_COWINPROGRESS;
1109 if (error)
1110 return (error);
1111 indiroff = (lbn - NDADDR) % NINDIR(fs);
1112 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1113 bqrelse(bp);
1114 }
1115 if (blkno != 0) {
1116 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1117 return (error);
1118 } else {
1119 error = UFS_BALLOC(snapvp, lblktosize(fs, (off_t)lbn),
1120 fs->fs_bsize, KERNCRED, 0, &bp);
1121 if (error)
1122 return (error);
1123 if ((error = readblock(bp, lbn)) != 0)
1124 return (error);
1125 }
1126 /*
1127 * Set a snapshot inode to be a zero length file, regular files
1128 * to be completely unallocated.
1129 */
1130 dip = (struct ufs2_dinode *)bp->b_data +
1131 ino_to_fsbo(fs, cancelip->i_number);
1132 if (expungetype == BLK_NOCOPY)
1133 dip->di_mode = 0;
1134 dip->di_size = 0;
1135 dip->di_blocks = 0;
1136 dip->di_flags &= ~SF_SNAPSHOT;
1137 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1138 bdwrite(bp);
1139 /*
1140 * Now go through and expunge all the blocks in the file
1141 * using the function requested.
1142 */
1143 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1144 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1145 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1146 return (error);
1147 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1148 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1149 return (error);
1150 blksperindir = 1;
1151 lbn = -NDADDR;
1152 len = numblks - NDADDR;
1153 rlbn = NDADDR;
1154 for (i = 0; len > 0 && i < NIADDR; i++) {
1155 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1156 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1157 blksperindir, fs, acctfunc, expungetype);
1158 if (error)
1159 return (error);
1160 blksperindir *= NINDIR(fs);
1161 lbn -= blksperindir + 1;
1162 len -= blksperindir;
1163 rlbn += blksperindir;
1164 }
1165 return (0);
1166 }
1167
1168 /*
1169 * Descend an indirect block chain for vnode cancelvp accounting for all
1170 * its indirect blocks in snapvp.
1171 */
1172 static int
1173 indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1174 blksperindir, fs, acctfunc, expungetype)
1175 struct vnode *snapvp;
1176 struct vnode *cancelvp;
1177 int level;
1178 ufs2_daddr_t blkno;
1179 ufs_lbn_t lbn;
1180 ufs_lbn_t rlbn;
1181 ufs_lbn_t remblks;
1182 ufs_lbn_t blksperindir;
1183 struct fs *fs;
1184 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1185 struct fs *, ufs_lbn_t, int);
1186 int expungetype;
1187 {
1188 int error, num, i;
1189 ufs_lbn_t subblksperindir;
1190 struct indir indirs[NIADDR + 2];
1191 ufs2_daddr_t last, *bap;
1192 struct buf *bp;
1193
1194 if (blkno == 0) {
1195 if (expungetype == BLK_NOCOPY)
1196 return (0);
1197 panic("indiracct_ufs2: missing indir");
1198 }
1199 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1200 return (error);
1201 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1202 panic("indiracct_ufs2: botched params");
1203 /*
1204 * We have to expand bread here since it will deadlock looking
1205 * up the block number for any blocks that are not in the cache.
1206 */
1207 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1208 bp->b_blkno = fsbtodb(fs, blkno);
1209 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1210 (error = readblock(bp, fragstoblks(fs, blkno)))) {
1211 brelse(bp);
1212 return (error);
1213 }
1214 /*
1215 * Account for the block pointers in this indirect block.
1216 */
1217 last = howmany(remblks, blksperindir);
1218 if (last > NINDIR(fs))
1219 last = NINDIR(fs);
1220 MALLOC(bap, ufs2_daddr_t *, fs->fs_bsize, M_DEVBUF, M_WAITOK);
1221 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1222 bqrelse(bp);
1223 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1224 level == 0 ? rlbn : -1, expungetype);
1225 if (error || level == 0)
1226 goto out;
1227 /*
1228 * Account for the block pointers in each of the indirect blocks
1229 * in the levels below us.
1230 */
1231 subblksperindir = blksperindir / NINDIR(fs);
1232 for (lbn++, level--, i = 0; i < last; i++) {
1233 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1234 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1235 if (error)
1236 goto out;
1237 rlbn += blksperindir;
1238 lbn -= blksperindir;
1239 remblks -= blksperindir;
1240 }
1241 out:
1242 FREE(bap, M_DEVBUF);
1243 return (error);
1244 }
1245
1246 /*
1247 * Do both snap accounting and map accounting.
1248 */
1249 static int
1250 fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1251 struct vnode *vp;
1252 ufs2_daddr_t *oldblkp, *lastblkp;
1253 struct fs *fs;
1254 ufs_lbn_t lblkno;
1255 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1256 {
1257 int error;
1258
1259 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1260 return (error);
1261 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1262 }
1263
1264 /*
1265 * Identify a set of blocks allocated in a snapshot inode.
1266 */
1267 static int
1268 snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1269 struct vnode *vp;
1270 ufs2_daddr_t *oldblkp, *lastblkp;
1271 struct fs *fs;
1272 ufs_lbn_t lblkno;
1273 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1274 {
1275 struct inode *ip = VTOI(vp);
1276 ufs2_daddr_t blkno, *blkp;
1277 ufs_lbn_t lbn;
1278 struct buf *ibp;
1279 int error;
1280
1281 for ( ; oldblkp < lastblkp; oldblkp++) {
1282 blkno = *oldblkp;
1283 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1284 continue;
1285 lbn = fragstoblks(fs, blkno);
1286 if (lbn < NDADDR) {
1287 blkp = &ip->i_din2->di_db[lbn];
1288 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1289 } else {
1290 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1291 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1292 if (error)
1293 return (error);
1294 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1295 [(lbn - NDADDR) % NINDIR(fs)];
1296 }
1297 /*
1298 * If we are expunging a snapshot vnode and we
1299 * find a block marked BLK_NOCOPY, then it is
1300 * one that has been allocated to this snapshot after
1301 * we took our current snapshot and can be ignored.
1302 */
1303 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1304 if (lbn >= NDADDR)
1305 brelse(ibp);
1306 } else {
1307 if (*blkp != 0)
1308 panic("snapacct_ufs2: bad block");
1309 *blkp = expungetype;
1310 if (lbn >= NDADDR)
1311 bdwrite(ibp);
1312 }
1313 }
1314 return (0);
1315 }
1316
1317 /*
1318 * Account for a set of blocks allocated in a snapshot inode.
1319 */
1320 static int
1321 mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1322 struct vnode *vp;
1323 ufs2_daddr_t *oldblkp, *lastblkp;
1324 struct fs *fs;
1325 ufs_lbn_t lblkno;
1326 int expungetype;
1327 {
1328 ufs2_daddr_t blkno;
1329 struct inode *ip;
1330 ino_t inum;
1331 int acctit;
1332
1333 ip = VTOI(vp);
1334 inum = ip->i_number;
1335 if (lblkno == -1)
1336 acctit = 0;
1337 else
1338 acctit = 1;
1339 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1340 blkno = *oldblkp;
1341 if (blkno == 0 || blkno == BLK_NOCOPY)
1342 continue;
1343 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1344 *ip->i_snapblklist++ = lblkno;
1345 if (blkno == BLK_SNAP)
1346 blkno = blkstofrags(fs, lblkno);
1347 ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
1348 }
1349 return (0);
1350 }
1351
1352 /*
1353 * Decrement extra reference on snapshot when last name is removed.
1354 * It will not be freed until the last open reference goes away.
1355 */
1356 void
1357 ffs_snapgone(ip)
1358 struct inode *ip;
1359 {
1360 struct inode *xp;
1361 struct fs *fs;
1362 int snaploc;
1363
1364 /*
1365 * Find snapshot in incore list.
1366 */
1367 TAILQ_FOREACH(xp, &ip->i_devvp->v_rdev->si_snapshots, i_nextsnap)
1368 if (xp == ip)
1369 break;
1370 if (xp != NULL)
1371 vrele(ITOV(ip));
1372 else if (snapdebug)
1373 printf("ffs_snapgone: lost snapshot vnode %d\n",
1374 ip->i_number);
1375 /*
1376 * Delete snapshot inode from superblock. Keep list dense.
1377 */
1378 fs = ip->i_fs;
1379 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1380 if (fs->fs_snapinum[snaploc] == ip->i_number)
1381 break;
1382 if (snaploc < FSMAXSNAP) {
1383 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1384 if (fs->fs_snapinum[snaploc] == 0)
1385 break;
1386 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1387 }
1388 fs->fs_snapinum[snaploc - 1] = 0;
1389 }
1390 }
1391
1392 /*
1393 * Prepare a snapshot file for being removed.
1394 */
1395 void
1396 ffs_snapremove(vp)
1397 struct vnode *vp;
1398 {
1399 struct inode *ip;
1400 struct vnode *devvp;
1401 struct lock *lkp;
1402 struct buf *ibp;
1403 struct fs *fs;
1404 struct thread *td = curthread;
1405 ufs2_daddr_t numblks, blkno, dblk, *snapblklist;
1406 int error, loc, last;
1407
1408 ip = VTOI(vp);
1409 fs = ip->i_fs;
1410 devvp = ip->i_devvp;
1411 /*
1412 * If active, delete from incore list (this snapshot may
1413 * already have been in the process of being deleted, so
1414 * would not have been active).
1415 *
1416 * Clear copy-on-write flag if last snapshot.
1417 */
1418 if (ip->i_nextsnap.tqe_prev != 0) {
1419 VI_LOCK(devvp);
1420 lockmgr(&vp->v_lock, LK_INTERLOCK | LK_EXCLUSIVE,
1421 VI_MTX(devvp), td);
1422 VI_LOCK(devvp);
1423 TAILQ_REMOVE(&devvp->v_rdev->si_snapshots, ip, i_nextsnap);
1424 ip->i_nextsnap.tqe_prev = 0;
1425 lkp = vp->v_vnlock;
1426 vp->v_vnlock = &vp->v_lock;
1427 lockmgr(lkp, LK_RELEASE, NULL, td);
1428 if (TAILQ_FIRST(&devvp->v_rdev->si_snapshots) != 0) {
1429 VI_UNLOCK(devvp);
1430 } else {
1431 snapblklist = devvp->v_rdev->si_snapblklist;
1432 devvp->v_rdev->si_snapblklist = 0;
1433 devvp->v_rdev->si_snaplistsize = 0;
1434 devvp->v_rdev->si_copyonwrite = 0;
1435 devvp->v_vflag &= ~VV_COPYONWRITE;
1436 lockmgr(lkp, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp), td);
1437 lockmgr(lkp, LK_RELEASE, NULL, td);
1438 lockdestroy(lkp);
1439 FREE(lkp, M_UFSMNT);
1440 FREE(snapblklist, M_UFSMNT);
1441 }
1442 }
1443 /*
1444 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1445 * snapshots that want them (see ffs_snapblkfree below).
1446 */
1447 for (blkno = 1; blkno < NDADDR; blkno++) {
1448 dblk = DIP(ip, i_db[blkno]);
1449 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1450 DIP(ip, i_db[blkno]) = 0;
1451 else if ((dblk == blkstofrags(fs, blkno) &&
1452 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1453 ip->i_number))) {
1454 DIP(ip, i_blocks) -= btodb(fs->fs_bsize);
1455 DIP(ip, i_db[blkno]) = 0;
1456 }
1457 }
1458 numblks = howmany(ip->i_size, fs->fs_bsize);
1459 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1460 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1461 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1462 if (error)
1463 continue;
1464 if (fs->fs_size - blkno > NINDIR(fs))
1465 last = NINDIR(fs);
1466 else
1467 last = fs->fs_size - blkno;
1468 for (loc = 0; loc < last; loc++) {
1469 if (ip->i_ump->um_fstype == UFS1) {
1470 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1471 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1472 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1473 else if ((dblk == blkstofrags(fs, blkno) &&
1474 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1475 fs->fs_bsize, ip->i_number))) {
1476 ip->i_din1->di_blocks -=
1477 btodb(fs->fs_bsize);
1478 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1479 }
1480 continue;
1481 }
1482 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1483 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1484 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1485 else if ((dblk == blkstofrags(fs, blkno) &&
1486 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1487 fs->fs_bsize, ip->i_number))) {
1488 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1489 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1490 }
1491 }
1492 bawrite(ibp);
1493 }
1494 /*
1495 * Clear snapshot flag and drop reference.
1496 */
1497 ip->i_flags &= ~SF_SNAPSHOT;
1498 DIP(ip, i_flags) = ip->i_flags;
1499 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1500 }
1501
1502 /*
1503 * Notification that a block is being freed. Return zero if the free
1504 * should be allowed to proceed. Return non-zero if the snapshot file
1505 * wants to claim the block. The block will be claimed if it is an
1506 * uncopied part of one of the snapshots. It will be freed if it is
1507 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1508 * If a fragment is being freed, then all snapshots that care about
1509 * it must make a copy since a snapshot file can only claim full sized
1510 * blocks. Note that if more than one snapshot file maps the block,
1511 * we can pick one at random to claim it. Since none of the snapshots
1512 * can change, we are assurred that they will all see the same unmodified
1513 * image. When deleting a snapshot file (see ffs_snapremove above), we
1514 * must push any of these claimed blocks to one of the other snapshots
1515 * that maps it. These claimed blocks are easily identified as they will
1516 * have a block number equal to their logical block number within the
1517 * snapshot. A copied block can never have this property because they
1518 * must always have been allocated from a BLK_NOCOPY location.
1519 */
1520 int
1521 ffs_snapblkfree(fs, devvp, bno, size, inum)
1522 struct fs *fs;
1523 struct vnode *devvp;
1524 ufs2_daddr_t bno;
1525 long size;
1526 ino_t inum;
1527 {
1528 struct buf *ibp, *cbp, *savedcbp = 0;
1529 struct thread *td = curthread;
1530 struct inode *ip;
1531 struct vnode *vp = NULL;
1532 ufs_lbn_t lbn;
1533 ufs2_daddr_t blkno;
1534 int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
1535 struct snaphead *snaphead;
1536
1537 lbn = fragstoblks(fs, bno);
1538 retry:
1539 VI_LOCK(devvp);
1540 snaphead = &devvp->v_rdev->si_snapshots;
1541 TAILQ_FOREACH(ip, snaphead, i_nextsnap) {
1542 vp = ITOV(ip);
1543 /*
1544 * Lookup block being written.
1545 */
1546 if (lbn < NDADDR) {
1547 blkno = DIP(ip, i_db[lbn]);
1548 } else {
1549 if (snapshot_locked == 0 &&
1550 lockmgr(vp->v_vnlock,
1551 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1552 VI_MTX(devvp), td) != 0)
1553 goto retry;
1554 snapshot_locked = 1;
1555 td->td_pflags |= TDP_COWINPROGRESS;
1556 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1557 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1558 td->td_pflags &= ~TDP_COWINPROGRESS;
1559 if (error)
1560 break;
1561 indiroff = (lbn - NDADDR) % NINDIR(fs);
1562 if (ip->i_ump->um_fstype == UFS1)
1563 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1564 else
1565 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1566 }
1567 /*
1568 * Check to see if block needs to be copied.
1569 */
1570 if (blkno == 0) {
1571 /*
1572 * A block that we map is being freed. If it has not
1573 * been claimed yet, we will claim or copy it (below).
1574 */
1575 claimedblk = 1;
1576 } else if (blkno == BLK_SNAP) {
1577 /*
1578 * No previous snapshot claimed the block,
1579 * so it will be freed and become a BLK_NOCOPY
1580 * (don't care) for us.
1581 */
1582 if (claimedblk)
1583 panic("snapblkfree: inconsistent block type");
1584 if (snapshot_locked == 0 &&
1585 lockmgr(vp->v_vnlock,
1586 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1587 VI_MTX(devvp), td) != 0) {
1588 if (lbn >= NDADDR)
1589 bqrelse(ibp);
1590 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1591 goto retry;
1592 }
1593 snapshot_locked = 1;
1594 if (lbn < NDADDR) {
1595 DIP(ip, i_db[lbn]) = BLK_NOCOPY;
1596 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1597 } else if (ip->i_ump->um_fstype == UFS1) {
1598 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1599 BLK_NOCOPY;
1600 bdwrite(ibp);
1601 } else {
1602 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1603 BLK_NOCOPY;
1604 bdwrite(ibp);
1605 }
1606 continue;
1607 } else /* BLK_NOCOPY or default */ {
1608 /*
1609 * If the snapshot has already copied the block
1610 * (default), or does not care about the block,
1611 * it is not needed.
1612 */
1613 if (lbn >= NDADDR)
1614 bqrelse(ibp);
1615 continue;
1616 }
1617 /*
1618 * If this is a full size block, we will just grab it
1619 * and assign it to the snapshot inode. Otherwise we
1620 * will proceed to copy it. See explanation for this
1621 * routine as to why only a single snapshot needs to
1622 * claim this block.
1623 */
1624 if (snapshot_locked == 0 &&
1625 lockmgr(vp->v_vnlock,
1626 LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
1627 VI_MTX(devvp), td) != 0) {
1628 if (lbn >= NDADDR)
1629 bqrelse(ibp);
1630 vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL, td);
1631 goto retry;
1632 }
1633 snapshot_locked = 1;
1634 if (size == fs->fs_bsize) {
1635 #ifdef DEBUG
1636 if (snapdebug)
1637 printf("%s %d lbn %jd from inum %d\n",
1638 "Grabonremove: snapino", ip->i_number,
1639 (intmax_t)lbn, inum);
1640 #endif
1641 if (lbn < NDADDR) {
1642 DIP(ip, i_db[lbn]) = bno;
1643 } else if (ip->i_ump->um_fstype == UFS1) {
1644 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1645 bdwrite(ibp);
1646 } else {
1647 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1648 bdwrite(ibp);
1649 }
1650 DIP(ip, i_blocks) += btodb(size);
1651 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1652 VOP_UNLOCK(vp, 0, td);
1653 return (1);
1654 }
1655 if (lbn >= NDADDR)
1656 bqrelse(ibp);
1657 /*
1658 * Allocate the block into which to do the copy. Note that this
1659 * allocation will never require any additional allocations for
1660 * the snapshot inode.
1661 */
1662 td->td_pflags |= TDP_COWINPROGRESS;
1663 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1664 fs->fs_bsize, KERNCRED, 0, &cbp);
1665 td->td_pflags &= ~TDP_COWINPROGRESS;
1666 if (error)
1667 break;
1668 #ifdef DEBUG
1669 if (snapdebug)
1670 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n",
1671 "Copyonremove: snapino ", ip->i_number,
1672 (intmax_t)lbn, "for inum", inum, size,
1673 (intmax_t)cbp->b_blkno);
1674 #endif
1675 /*
1676 * If we have already read the old block contents, then
1677 * simply copy them to the new block. Note that we need
1678 * to synchronously write snapshots that have not been
1679 * unlinked, and hence will be visible after a crash,
1680 * to ensure their integrity.
1681 */
1682 if (savedcbp != 0) {
1683 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1684 bawrite(cbp);
1685 if (dopersistence && ip->i_effnlink > 0)
1686 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1687 continue;
1688 }
1689 /*
1690 * Otherwise, read the old block contents into the buffer.
1691 */
1692 if ((error = readblock(cbp, lbn)) != 0) {
1693 bzero(cbp->b_data, fs->fs_bsize);
1694 bawrite(cbp);
1695 if (dopersistence && ip->i_effnlink > 0)
1696 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1697 break;
1698 }
1699 savedcbp = cbp;
1700 }
1701 /*
1702 * Note that we need to synchronously write snapshots that
1703 * have not been unlinked, and hence will be visible after
1704 * a crash, to ensure their integrity.
1705 */
1706 if (savedcbp) {
1707 vp = savedcbp->b_vp;
1708 bawrite(savedcbp);
1709 if (dopersistence && VTOI(vp)->i_effnlink > 0)
1710 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1711 }
1712 /*
1713 * If we have been unable to allocate a block in which to do
1714 * the copy, then return non-zero so that the fragment will
1715 * not be freed. Although space will be lost, the snapshot
1716 * will stay consistent.
1717 */
1718 if (snapshot_locked)
1719 VOP_UNLOCK(vp, 0, td);
1720 else
1721 VI_UNLOCK(devvp);
1722 return (error);
1723 }
1724
1725 /*
1726 * Associate snapshot files when mounting.
1727 */
1728 void
1729 ffs_snapshot_mount(mp)
1730 struct mount *mp;
1731 {
1732 struct ufsmount *ump = VFSTOUFS(mp);
1733 struct vnode *devvp = ump->um_devvp;
1734 struct fs *fs = ump->um_fs;
1735 struct thread *td = curthread;
1736 struct snaphead *snaphead;
1737 struct vnode *vp;
1738 struct inode *ip, *xp;
1739 struct uio auio;
1740 struct iovec aiov;
1741 void *snapblklist;
1742 char *reason;
1743 daddr_t snaplistsize;
1744 int error, snaploc, loc;
1745
1746 /*
1747 * XXX The following needs to be set before UFS_TRUNCATE or
1748 * VOP_READ can be called.
1749 */
1750 mp->mnt_stat.f_iosize = fs->fs_bsize;
1751 /*
1752 * Process each snapshot listed in the superblock.
1753 */
1754 vp = NULL;
1755 snaphead = &devvp->v_rdev->si_snapshots;
1756 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1757 if (fs->fs_snapinum[snaploc] == 0)
1758 break;
1759 if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc],
1760 LK_EXCLUSIVE, &vp)) != 0){
1761 printf("ffs_snapshot_mount: vget failed %d\n", error);
1762 continue;
1763 }
1764 ip = VTOI(vp);
1765 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size ==
1766 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1767 if ((ip->i_flags & SF_SNAPSHOT) == 0) {
1768 reason = "non-snapshot";
1769 } else {
1770 reason = "old format snapshot";
1771 (void)UFS_TRUNCATE(vp, (off_t)0, 0, NOCRED, td);
1772 (void)VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
1773 }
1774 printf("ffs_snapshot_mount: %s inode %d\n",
1775 reason, fs->fs_snapinum[snaploc]);
1776 vput(vp);
1777 vp = NULL;
1778 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
1779 if (fs->fs_snapinum[loc] == 0)
1780 break;
1781 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
1782 }
1783 fs->fs_snapinum[loc - 1] = 0;
1784 snaploc--;
1785 continue;
1786 }
1787 /*
1788 * If there already exist snapshots on this filesystem, grab a
1789 * reference to their shared lock. If this is the first snapshot
1790 * on this filesystem, we need to allocate a lock for the
1791 * snapshots to share. In either case, acquire the snapshot
1792 * lock and give up our original private lock.
1793 */
1794 VI_LOCK(devvp);
1795 if ((xp = TAILQ_FIRST(snaphead)) != NULL) {
1796 VI_LOCK(vp);
1797 vp->v_vnlock = ITOV(xp)->v_vnlock;
1798 VI_UNLOCK(devvp);
1799 } else {
1800 struct lock *lkp;
1801
1802 VI_UNLOCK(devvp);
1803 MALLOC(lkp, struct lock *, sizeof(struct lock),
1804 M_UFSMNT, M_WAITOK);
1805 lockinit(lkp, PVFS, "snaplk", VLKTIMEOUT,
1806 LK_CANRECURSE | LK_NOPAUSE);
1807 VI_LOCK(vp);
1808 vp->v_vnlock = lkp;
1809 }
1810 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
1811 transferlockers(&vp->v_lock, vp->v_vnlock);
1812 lockmgr(&vp->v_lock, LK_RELEASE, NULL, td);
1813 /*
1814 * Link it onto the active snapshot list.
1815 */
1816 VI_LOCK(devvp);
1817 if (ip->i_nextsnap.tqe_prev != 0)
1818 panic("ffs_snapshot_mount: %d already on list",
1819 ip->i_number);
1820 else
1821 TAILQ_INSERT_TAIL(snaphead, ip, i_nextsnap);
1822 vp->v_vflag |= VV_SYSTEM;
1823 VI_UNLOCK(devvp);
1824 VOP_UNLOCK(vp, 0, td);
1825 }
1826 /*
1827 * No usable snapshots found.
1828 */
1829 if (vp == NULL)
1830 return;
1831 /*
1832 * Allocate the space for the block hints list. We always want to
1833 * use the list from the newest snapshot.
1834 */
1835 auio.uio_iov = &aiov;
1836 auio.uio_iovcnt = 1;
1837 aiov.iov_base = (void *)&snaplistsize;
1838 aiov.iov_len = sizeof(snaplistsize);
1839 auio.uio_resid = aiov.iov_len;
1840 auio.uio_offset =
1841 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
1842 auio.uio_segflg = UIO_SYSSPACE;
1843 auio.uio_rw = UIO_READ;
1844 auio.uio_td = td;
1845 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1846 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1847 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
1848 VOP_UNLOCK(vp, 0, td);
1849 return;
1850 }
1851 MALLOC(snapblklist, void *, snaplistsize * sizeof(daddr_t),
1852 M_UFSMNT, M_WAITOK);
1853 auio.uio_iovcnt = 1;
1854 aiov.iov_base = snapblklist;
1855 aiov.iov_len = snaplistsize * sizeof (daddr_t);
1856 auio.uio_resid = aiov.iov_len;
1857 auio.uio_offset -= sizeof(snaplistsize);
1858 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
1859 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
1860 VOP_UNLOCK(vp, 0, td);
1861 FREE(snapblklist, M_UFSMNT);
1862 return;
1863 }
1864 VOP_UNLOCK(vp, 0, td);
1865 VI_LOCK(devvp);
1866 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
1867 devvp->v_rdev->si_snaplistsize = snaplistsize;
1868 devvp->v_rdev->si_snapblklist = (daddr_t *)snapblklist;
1869 devvp->v_rdev->si_copyonwrite = ffs_copyonwrite;
1870 devvp->v_vflag |= VV_COPYONWRITE;
1871 VI_UNLOCK(devvp);
1872 }
1873
1874 /*
1875 * Disassociate snapshot files when unmounting.
1876 */
1877 void
1878 ffs_snapshot_unmount(mp)
1879 struct mount *mp;
1880 {
1881 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
1882 struct snaphead *snaphead = &devvp->v_rdev->si_snapshots;
1883 struct lock *lkp = NULL;
1884 struct inode *xp;
1885 struct vnode *vp;
1886
1887 VI_LOCK(devvp);
1888 while ((xp = TAILQ_FIRST(snaphead)) != 0) {
1889 vp = ITOV(xp);
1890 lkp = vp->v_vnlock;
1891 vp->v_vnlock = &vp->v_lock;
1892 TAILQ_REMOVE(snaphead, xp, i_nextsnap);
1893 xp->i_nextsnap.tqe_prev = 0;
1894 if (xp->i_effnlink > 0) {
1895 VI_UNLOCK(devvp);
1896 vrele(vp);
1897 VI_LOCK(devvp);
1898 }
1899 }
1900 if (devvp->v_rdev->si_snapblklist != NULL) {
1901 FREE(devvp->v_rdev->si_snapblklist, M_UFSMNT);
1902 devvp->v_rdev->si_snapblklist = NULL;
1903 devvp->v_rdev->si_snaplistsize = 0;
1904 }
1905 if (lkp != NULL) {
1906 lockdestroy(lkp);
1907 FREE(lkp, M_UFSMNT);
1908 }
1909 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
1910 devvp->v_rdev->si_copyonwrite = 0;
1911 devvp->v_vflag &= ~VV_COPYONWRITE;
1912 VI_UNLOCK(devvp);
1913 }
1914
1915 /*
1916 * Check for need to copy block that is about to be written,
1917 * copying the block if necessary.
1918 */
1919 static int
1920 ffs_copyonwrite(devvp, bp)
1921 struct vnode *devvp;
1922 struct buf *bp;
1923 {
1924 struct snaphead *snaphead;
1925 struct buf *ibp, *cbp, *savedcbp = 0;
1926 struct thread *td = curthread;
1927 struct fs *fs;
1928 struct inode *ip;
1929 struct vnode *vp = 0;
1930 ufs2_daddr_t lbn, blkno, *snapblklist;
1931 int lower, upper, mid, indiroff, snapshot_locked = 0, error = 0;
1932
1933 if (td->td_pflags & TDP_COWINPROGRESS)
1934 panic("ffs_copyonwrite: recursive call");
1935 /*
1936 * First check to see if it is in the preallocated list.
1937 * By doing this check we avoid several potential deadlocks.
1938 */
1939 VI_LOCK(devvp);
1940 snaphead = &devvp->v_rdev->si_snapshots;
1941 ip = TAILQ_FIRST(snaphead);
1942 fs = ip->i_fs;
1943 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
1944 snapblklist = devvp->v_rdev->si_snapblklist;
1945 upper = devvp->v_rdev->si_snaplistsize - 1;
1946 lower = 1;
1947 while (lower <= upper) {
1948 mid = (lower + upper) / 2;
1949 if (snapblklist[mid] == lbn)
1950 break;
1951 if (snapblklist[mid] < lbn)
1952 lower = mid + 1;
1953 else
1954 upper = mid - 1;
1955 }
1956 if (lower <= upper) {
1957 VI_UNLOCK(devvp);
1958 return (0);
1959 }
1960 /*
1961 * Not in the precomputed list, so check the snapshots.
1962 */
1963 retry:
1964 TAILQ_FOREACH(ip, snaphead, i_nextsnap) {
1965 vp = ITOV(ip);
1966 /*
1967 * We ensure that everything of our own that needs to be
1968 * copied will be done at the time that ffs_snapshot is
1969 * called. Thus we can skip the check here which can
1970 * deadlock in doing the lookup in UFS_BALLOC.
1971 */
1972 if (bp->b_vp == vp)
1973 continue;
1974 /*
1975 * Check to see if block needs to be copied. We do not have
1976 * to hold the snapshot lock while doing this lookup as it
1977 * will never require any additional allocations for the
1978 * snapshot inode.
1979 */
1980 if (lbn < NDADDR) {
1981 blkno = DIP(ip, i_db[lbn]);
1982 } else {
1983 if (snapshot_locked == 0 &&
1984 lockmgr(vp->v_vnlock,
1985 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1986 VI_MTX(devvp), td) != 0) {
1987 VI_LOCK(devvp);
1988 goto retry;
1989 }
1990 snapshot_locked = 1;
1991 td->td_pflags |= TDP_COWINPROGRESS;
1992 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1993 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1994 td->td_pflags &= ~TDP_COWINPROGRESS;
1995 if (error)
1996 break;
1997 indiroff = (lbn - NDADDR) % NINDIR(fs);
1998 if (ip->i_ump->um_fstype == UFS1)
1999 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2000 else
2001 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2002 bqrelse(ibp);
2003 }
2004 #ifdef DIAGNOSTIC
2005 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2006 panic("ffs_copyonwrite: bad copy block");
2007 #endif
2008 if (blkno != 0)
2009 continue;
2010 /*
2011 * Allocate the block into which to do the copy. Since
2012 * multiple processes may all try to copy the same block,
2013 * we have to recheck our need to do a copy if we sleep
2014 * waiting for the lock.
2015 *
2016 * Because all snapshots on a filesystem share a single
2017 * lock, we ensure that we will never be in competition
2018 * with another process to allocate a block.
2019 */
2020 if (snapshot_locked == 0 &&
2021 lockmgr(vp->v_vnlock,
2022 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2023 VI_MTX(devvp), td) != 0) {
2024 VI_LOCK(devvp);
2025 goto retry;
2026 }
2027 snapshot_locked = 1;
2028 td->td_pflags |= TDP_COWINPROGRESS;
2029 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2030 fs->fs_bsize, KERNCRED, 0, &cbp);
2031 td->td_pflags &= ~TDP_COWINPROGRESS;
2032 if (error)
2033 break;
2034 #ifdef DEBUG
2035 if (snapdebug) {
2036 printf("Copyonwrite: snapino %d lbn %jd for ",
2037 ip->i_number, (intmax_t)lbn);
2038 if (bp->b_vp == devvp)
2039 printf("fs metadata");
2040 else
2041 printf("inum %d", VTOI(bp->b_vp)->i_number);
2042 printf(" lblkno %jd to blkno %jd\n",
2043 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2044 }
2045 #endif
2046 /*
2047 * If we have already read the old block contents, then
2048 * simply copy them to the new block. Note that we need
2049 * to synchronously write snapshots that have not been
2050 * unlinked, and hence will be visible after a crash,
2051 * to ensure their integrity.
2052 */
2053 if (savedcbp != 0) {
2054 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2055 bawrite(cbp);
2056 if (dopersistence && ip->i_effnlink > 0)
2057 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2058 continue;
2059 }
2060 /*
2061 * Otherwise, read the old block contents into the buffer.
2062 */
2063 if ((error = readblock(cbp, lbn)) != 0) {
2064 bzero(cbp->b_data, fs->fs_bsize);
2065 bawrite(cbp);
2066 if (dopersistence && ip->i_effnlink > 0)
2067 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2068 break;
2069 }
2070 savedcbp = cbp;
2071 }
2072 /*
2073 * Note that we need to synchronously write snapshots that
2074 * have not been unlinked, and hence will be visible after
2075 * a crash, to ensure their integrity.
2076 */
2077 if (savedcbp) {
2078 vp = savedcbp->b_vp;
2079 bawrite(savedcbp);
2080 if (dopersistence && VTOI(vp)->i_effnlink > 0)
2081 (void) VOP_FSYNC(vp, KERNCRED, MNT_WAIT, td);
2082 }
2083 if (snapshot_locked)
2084 VOP_UNLOCK(vp, 0, td);
2085 else
2086 VI_UNLOCK(devvp);
2087 return (error);
2088 }
2089
2090 /*
2091 * Read the specified block into the given buffer.
2092 * Much of this boiler-plate comes from bwrite().
2093 */
2094 static int
2095 readblock(bp, lbn)
2096 struct buf *bp;
2097 ufs2_daddr_t lbn;
2098 {
2099 struct uio auio;
2100 struct iovec aiov;
2101 struct thread *td = curthread;
2102 struct inode *ip = VTOI(bp->b_vp);
2103
2104 aiov.iov_base = bp->b_data;
2105 aiov.iov_len = bp->b_bcount;
2106 auio.uio_iov = &aiov;
2107 auio.uio_iovcnt = 1;
2108 auio.uio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2109 auio.uio_resid = bp->b_bcount;
2110 auio.uio_rw = UIO_READ;
2111 auio.uio_segflg = UIO_SYSSPACE;
2112 auio.uio_td = td;
2113 return (physio(ip->i_devvp->v_rdev, &auio, 0));
2114 }
Cache object: b331614e97d93af43a3a9e84e335c991
|