2 * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved.
4 * Further information about snapshots can be obtained from:
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
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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.
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
33 * @(#)ffs_snapshot.c 8.11 (McKusick) 7/23/00
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include "opt_quota.h"
41 #include <sys/param.h>
42 #include <sys/kernel.h>
43 #include <sys/systm.h>
47 #include <sys/fcntl.h>
49 #include <sys/namei.h>
50 #include <sys/sched.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/resource.h>
55 #include <sys/resourcevar.h>
56 #include <sys/vnode.h>
58 #include <geom/geom.h>
60 #include <ufs/ufs/extattr.h>
61 #include <ufs/ufs/quota.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/inode.h>
64 #include <ufs/ufs/ufs_extern.h>
66 #include <ufs/ffs/fs.h>
67 #include <ufs/ffs/ffs_extern.h>
69 #define KERNCRED thread0.td_ucred
74 #ifdef NO_FFS_SNAPSHOT
76 ffs_snapshot(mp, snapfile)
84 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
91 struct workhead *wkhd;
103 ffs_snapshot_mount(mp)
109 ffs_snapshot_unmount(mp)
121 ffs_copyonwrite(devvp, bp)
129 ffs_sync_snap(mp, waitfor)
136 FEATURE(ffs_snapshot, "FFS snapshot support");
138 LIST_HEAD(, snapdata) snapfree;
139 static struct mtx snapfree_lock;
140 MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF);
142 static int cgaccount(int, struct vnode *, struct buf *, int);
143 static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
144 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
145 ufs_lbn_t, int), int, int);
146 static int indiracct_ufs1(struct vnode *, struct vnode *, int,
147 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
148 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
149 ufs_lbn_t, int), int);
150 static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
151 struct fs *, ufs_lbn_t, int);
152 static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
153 struct fs *, ufs_lbn_t, int);
154 static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
155 struct fs *, ufs_lbn_t, int);
156 static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
157 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
158 ufs_lbn_t, int), int, int);
159 static int indiracct_ufs2(struct vnode *, struct vnode *, int,
160 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
161 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
162 ufs_lbn_t, int), int);
163 static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
164 struct fs *, ufs_lbn_t, int);
165 static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
166 struct fs *, ufs_lbn_t, int);
167 static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
168 struct fs *, ufs_lbn_t, int);
169 static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t);
170 static void try_free_snapdata(struct vnode *devvp);
171 static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp);
172 static int ffs_bp_snapblk(struct vnode *, struct buf *);
175 * To ensure the consistency of snapshots across crashes, we must
176 * synchronously write out copied blocks before allowing the
177 * originals to be modified. Because of the rather severe speed
178 * penalty that this imposes, the code normally only ensures
179 * persistence for the filesystem metadata contained within a
180 * snapshot. Setting the following flag allows this crash
181 * persistence to be enabled for file contents.
183 int dopersistence = 0;
186 #include <sys/sysctl.h>
187 SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, "");
188 static int snapdebug = 0;
189 SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
190 int collectsnapstats = 0;
191 SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats,
196 * Create a snapshot file and initialize it for the filesystem.
199 ffs_snapshot(mp, snapfile)
203 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist;
204 int error, cg, snaploc;
205 int i, size, len, loc;
207 struct timespec starttime = {0, 0}, endtime;
209 long redo = 0, snaplistsize = 0;
212 struct fs *copy_fs = NULL, *fs;
213 struct thread *td = curthread;
214 struct inode *ip, *xp;
215 struct buf *bp, *nbp, *ibp;
219 struct vnode *vp, *xvp, *mvp, *devvp;
223 struct ufsmount *ump;
233 * Need to serialize access to snapshot code per filesystem.
236 * Assign a snapshot slot in the superblock.
239 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
240 if (fs->fs_snapinum[snaploc] == 0)
243 if (snaploc == FSMAXSNAP)
246 * Create the snapshot file.
249 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF, UIO_SYSSPACE, snapfile, td);
250 if ((error = namei(&nd)) != 0)
252 if (nd.ni_vp != NULL) {
256 if (nd.ni_dvp->v_mount != mp)
259 NDFREE(&nd, NDF_ONLY_PNBUF);
260 if (nd.ni_dvp == nd.ni_vp)
268 vat.va_mode = S_IRUSR;
269 vat.va_vaflags |= VA_EXCLUSIVE;
270 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
273 panic("ffs_snapshot: mount mismatch");
275 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
276 NDFREE(&nd, NDF_ONLY_PNBUF);
278 if ((error = vn_start_write(NULL, &wrtmp,
279 V_XSLEEP | PCATCH)) != 0)
283 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
284 VOP_UNLOCK(nd.ni_dvp, 0);
286 NDFREE(&nd, NDF_ONLY_PNBUF);
287 vn_finished_write(wrtmp);
292 vp->v_vflag |= VV_SYSTEM;
296 * Allocate and copy the last block contents so as to be able
297 * to set size to that of the filesystem.
299 numblks = howmany(fs->fs_size, fs->fs_frag);
300 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
301 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
304 ip->i_size = lblktosize(fs, (off_t)numblks);
305 DIP_SET(ip, i_size, ip->i_size);
306 ip->i_flag |= IN_CHANGE | IN_UPDATE;
307 error = readblock(vp, bp, numblks - 1);
312 * Preallocate critical data structures so that we can copy
313 * them in without further allocation after we suspend all
314 * operations on the filesystem. We would like to just release
315 * the allocated buffers without writing them since they will
316 * be filled in below once we are ready to go, but this upsets
317 * the soft update code, so we go ahead and write the new buffers.
319 * Allocate all indirect blocks and mark all of them as not
320 * needing to be copied.
322 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
323 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
324 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp);
330 * Allocate copies for the superblock and its summary information.
332 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
337 blkno = fragstoblks(fs, fs->fs_csaddr);
338 len = howmany(fs->fs_cssize, fs->fs_bsize);
339 for (loc = 0; loc < len; loc++) {
340 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
341 fs->fs_bsize, KERNCRED, 0, &nbp);
347 * Allocate all cylinder group blocks.
349 for (cg = 0; cg < fs->fs_ncg; cg++) {
350 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
351 fs->fs_bsize, KERNCRED, 0, &nbp);
356 ffs_syncvnode(vp, MNT_WAIT);
359 * Copy all the cylinder group maps. Although the
360 * filesystem is still active, we hope that only a few
361 * cylinder groups will change between now and when we
362 * suspend operations. Thus, we will be able to quickly
363 * touch up the few cylinder groups that changed during
364 * the suspension period.
366 len = howmany(fs->fs_ncg, NBBY);
367 space = malloc(len, M_DEVBUF, M_WAITOK|M_ZERO);
369 fs->fs_active = space;
371 for (cg = 0; cg < fs->fs_ncg; cg++) {
372 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
373 fs->fs_bsize, KERNCRED, 0, &nbp);
376 error = cgaccount(cg, vp, nbp, 1);
379 ffs_syncvnode(vp, MNT_WAIT);
384 * Change inode to snapshot type file.
386 ip->i_flags |= SF_SNAPSHOT;
387 DIP_SET(ip, i_flags, ip->i_flags);
388 ip->i_flag |= IN_CHANGE | IN_UPDATE;
390 * Ensure that the snapshot is completely on disk.
391 * Since we have marked it as a snapshot it is safe to
392 * unlock it as no process will be allowed to write to it.
394 if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0)
398 * All allocations are done, so we can now snapshot the system.
400 * Recind nice scheduling while running with the filesystem suspended.
402 if (td->td_proc->p_nice > 0) {
407 saved_nice = p->p_nice;
412 * Suspend operation on filesystem.
415 vn_finished_write(wrtmp);
416 if ((error = vfs_write_suspend(vp->v_mount)) != 0) {
417 vn_start_write(NULL, &wrtmp, V_WAIT);
418 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
421 if (mp->mnt_kern_flag & MNTK_SUSPENDED)
423 vn_start_write(NULL, &wrtmp, V_WAIT);
425 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
426 if (ip->i_effnlink == 0) {
427 error = ENOENT; /* Snapshot file unlinked */
430 if (collectsnapstats)
431 nanotime(&starttime);
433 /* The last block might have changed. Copy it again to be sure. */
434 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
435 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
438 error = readblock(vp, bp, numblks - 1);
439 bp->b_flags |= B_VALIDSUSPWRT;
444 * First, copy all the cylinder group maps that have changed.
446 for (cg = 0; cg < fs->fs_ncg; cg++) {
447 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0)
450 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
451 fs->fs_bsize, KERNCRED, 0, &nbp);
454 error = cgaccount(cg, vp, nbp, 2);
460 * Grab a copy of the superblock and its summary information.
461 * We delay writing it until the suspension is released below.
463 copy_fs = malloc((u_long)fs->fs_bsize, M_UFSMNT, M_WAITOK);
464 bcopy(fs, copy_fs, fs->fs_sbsize);
465 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
466 copy_fs->fs_clean = 1;
467 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
468 if (fs->fs_sbsize < size)
469 bzero(&((char *)copy_fs)[fs->fs_sbsize],
470 size - fs->fs_sbsize);
471 size = blkroundup(fs, fs->fs_cssize);
472 if (fs->fs_contigsumsize > 0)
473 size += fs->fs_ncg * sizeof(int32_t);
474 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
475 copy_fs->fs_csp = space;
476 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
477 space = (char *)space + fs->fs_cssize;
478 loc = howmany(fs->fs_cssize, fs->fs_fsize);
479 i = fs->fs_frag - loc % fs->fs_frag;
480 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
482 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
483 len, KERNCRED, &bp)) != 0) {
485 free(copy_fs->fs_csp, M_UFSMNT);
486 free(copy_fs, M_UFSMNT);
490 bcopy(bp->b_data, space, (u_int)len);
491 space = (char *)space + len;
492 bp->b_flags |= B_INVAL | B_NOCACHE;
495 if (fs->fs_contigsumsize > 0) {
496 copy_fs->fs_maxcluster = lp = space;
497 for (i = 0; i < fs->fs_ncg; i++)
498 *lp++ = fs->fs_contigsumsize;
501 * We must check for active files that have been unlinked
502 * (e.g., with a zero link count). We have to expunge all
503 * trace of these files from the snapshot so that they are
504 * not reclaimed prematurely by fsck or unnecessarily dumped.
505 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
506 * spec_strategy about writing on a suspended filesystem.
507 * Note that we skip unlinked snapshot files as they will
508 * be handled separately below.
510 * We also calculate the needed size for the snapshot list.
512 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
513 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
515 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
517 MNT_VNODE_FOREACH(xvp, mp, mvp) {
520 if ((xvp->v_iflag & VI_DOOMED) ||
521 (xvp->v_usecount == 0 &&
522 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) ||
523 xvp->v_type == VNON ||
524 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
530 * We can skip parent directory vnode because it must have
531 * this snapshot file in it.
533 if (xvp == nd.ni_dvp) {
539 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) {
541 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
546 if (xvp->v_usecount == 0 &&
547 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) {
556 vprint("ffs_snapshot: busy vnode", xvp);
557 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 &&
565 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
572 * If there is a fragment, clear it here.
575 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
577 len = fragroundup(fs, blkoff(fs, xp->i_size));
578 if (len != 0 && len < fs->fs_bsize) {
579 ffs_blkfree(ump, copy_fs, vp,
580 DIP(xp, i_db[loc]), len, xp->i_number,
582 blkno = DIP(xp, i_db[loc]);
583 DIP_SET(xp, i_db[loc], 0);
587 if (xp->i_ump->um_fstype == UFS1)
588 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
591 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
594 DIP_SET(xp, i_db[loc], blkno);
596 error = ffs_freefile(ump, copy_fs, vp, xp->i_number,
601 free(copy_fs->fs_csp, M_UFSMNT);
602 free(copy_fs, M_UFSMNT);
604 MNT_VNODE_FOREACH_ABORT(mp, mvp);
611 * Erase the journal file from the snapshot.
613 if (fs->fs_flags & FS_SUJ) {
614 error = softdep_journal_lookup(mp, &xvp);
616 free(copy_fs->fs_csp, M_UFSMNT);
617 free(copy_fs, M_UFSMNT);
622 if (xp->i_ump->um_fstype == UFS1)
623 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
626 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
631 * Acquire a lock on the snapdata structure, creating it if necessary.
633 sn = ffs_snapdata_acquire(devvp);
635 * Change vnode to use shared snapshot lock instead of the original
638 vp->v_vnlock = &sn->sn_lock;
639 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
640 xp = TAILQ_FIRST(&sn->sn_head);
642 * If this is the first snapshot on this filesystem, then we need
643 * to allocate the space for the list of preallocated snapshot blocks.
644 * This list will be refined below, but this preliminary one will
645 * keep us out of deadlock until the full one is ready.
648 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
650 blkp = &snapblklist[1];
651 *blkp++ = lblkno(fs, fs->fs_sblockloc);
652 blkno = fragstoblks(fs, fs->fs_csaddr);
653 for (cg = 0; cg < fs->fs_ncg; cg++) {
654 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
656 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
658 len = howmany(fs->fs_cssize, fs->fs_bsize);
659 for (loc = 0; loc < len; loc++)
660 *blkp++ = blkno + loc;
661 for (; cg < fs->fs_ncg; cg++)
662 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
663 snapblklist[0] = blkp - snapblklist;
665 if (sn->sn_blklist != NULL)
666 panic("ffs_snapshot: non-empty list");
667 sn->sn_blklist = snapblklist;
668 sn->sn_listsize = blkp - snapblklist;
672 * Record snapshot inode. Since this is the newest snapshot,
673 * it must be placed at the end of the list.
676 fs->fs_snapinum[snaploc] = ip->i_number;
677 if (ip->i_nextsnap.tqe_prev != 0)
678 panic("ffs_snapshot: %d already on list", ip->i_number);
679 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
680 devvp->v_vflag |= VV_COPYONWRITE;
682 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
684 KASSERT((sn != NULL && copy_fs != NULL && error == 0) ||
685 (sn == NULL && copy_fs == NULL && error != 0),
686 ("email phk@ and mckusick@"));
688 * Resume operation on filesystem.
690 vfs_write_resume(vp->v_mount);
691 vn_start_write(NULL, &wrtmp, V_WAIT);
692 if (collectsnapstats && starttime.tv_sec > 0) {
694 timespecsub(&endtime, &starttime);
695 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
696 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
697 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
702 * Copy allocation information from all the snapshots in
703 * this snapshot and then expunge them from its view.
705 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) {
708 if (xp->i_ump->um_fstype == UFS1)
709 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
712 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
714 if (error == 0 && xp->i_effnlink == 0) {
715 error = ffs_freefile(ump,
722 fs->fs_snapinum[snaploc] = 0;
727 * Allocate space for the full list of preallocated snapshot blocks.
729 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
731 ip->i_snapblklist = &snapblklist[1];
733 * Expunge the blocks used by the snapshots from the set of
734 * blocks marked as used in the snapshot bitmaps. Also, collect
735 * the list of allocated blocks in i_snapblklist.
737 if (ip->i_ump->um_fstype == UFS1)
738 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1,
741 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2,
744 fs->fs_snapinum[snaploc] = 0;
745 free(snapblklist, M_UFSMNT);
748 if (snaplistsize < ip->i_snapblklist - snapblklist)
749 panic("ffs_snapshot: list too small");
750 snaplistsize = ip->i_snapblklist - snapblklist;
751 snapblklist[0] = snaplistsize;
752 ip->i_snapblklist = 0;
754 * Write out the list of allocated blocks to the end of the snapshot.
756 auio.uio_iov = &aiov;
758 aiov.iov_base = (void *)snapblklist;
759 aiov.iov_len = snaplistsize * sizeof(daddr_t);
760 auio.uio_resid = aiov.iov_len;
761 auio.uio_offset = ip->i_size;
762 auio.uio_segflg = UIO_SYSSPACE;
763 auio.uio_rw = UIO_WRITE;
765 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
766 fs->fs_snapinum[snaploc] = 0;
767 free(snapblklist, M_UFSMNT);
771 * Write the superblock and its summary information
774 blkno = fragstoblks(fs, fs->fs_csaddr);
775 len = howmany(fs->fs_cssize, fs->fs_bsize);
776 space = copy_fs->fs_csp;
777 for (loc = 0; loc < len; loc++) {
778 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
781 fs->fs_snapinum[snaploc] = 0;
782 free(snapblklist, M_UFSMNT);
785 bcopy(space, nbp->b_data, fs->fs_bsize);
786 space = (char *)space + fs->fs_bsize;
789 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
794 loc = blkoff(fs, fs->fs_sblockloc);
795 bcopy((char *)copy_fs, &nbp->b_data[loc], fs->fs_bsize);
799 * As this is the newest list, it is the most inclusive, so
800 * should replace the previous list.
803 space = sn->sn_blklist;
804 sn->sn_blklist = snapblklist;
805 sn->sn_listsize = snaplistsize;
808 free(space, M_UFSMNT);
810 * If another process is currently writing the buffer containing
811 * the inode for this snapshot then a deadlock can occur. Drop
812 * the snapshot lock until the buffer has been written.
814 VREF(vp); /* Protect against ffs_snapgone() */
816 (void) bread(ip->i_devvp,
817 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
818 (int) fs->fs_bsize, NOCRED, &nbp);
820 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
821 if (ip->i_effnlink == 0)
822 error = ENOENT; /* Snapshot file unlinked */
824 vrele(vp); /* Drop extra reference */
826 free(copy_fs->fs_csp, M_UFSMNT);
827 free(copy_fs, M_UFSMNT);
830 NDFREE(&nd, NDF_ONLY_PNBUF);
831 if (saved_nice > 0) {
836 sched_nice(td->td_proc, saved_nice);
837 PROC_UNLOCK(td->td_proc);
840 if (fs->fs_active != 0) {
841 free(fs->fs_active, M_DEVBUF);
846 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
849 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED, td);
850 (void) ffs_syncvnode(vp, MNT_WAIT);
856 vn_finished_write(wrtmp);
857 process_deferred_inactive(mp);
862 * Copy a cylinder group map. All the unallocated blocks are marked
863 * BLK_NOCOPY so that the snapshot knows that it need not copy them
864 * if they are later written. If passno is one, then this is a first
865 * pass, so only setting needs to be done. If passno is 2, then this
866 * is a revision to a previous pass which must be undone as the
867 * replacement pass is done.
870 cgaccount(cg, vp, nbp, passno)
876 struct buf *bp, *ibp;
880 ufs2_daddr_t base, numblks;
881 int error, len, loc, indiroff;
885 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
886 (int)fs->fs_cgsize, KERNCRED, &bp);
891 cgp = (struct cg *)bp->b_data;
892 if (!cg_chkmagic(cgp)) {
899 * Recomputation of summary information might not have been performed
900 * at mount time. Sync up summary information for current cylinder
901 * group while data is in memory to ensure that result of background
902 * fsck is slightly more consistent.
904 fs->fs_cs(fs, cg) = cgp->cg_cs;
905 UFS_UNLOCK(ip->i_ump);
906 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
907 if (fs->fs_cgsize < fs->fs_bsize)
908 bzero(&nbp->b_data[fs->fs_cgsize],
909 fs->fs_bsize - fs->fs_cgsize);
910 cgp = (struct cg *)nbp->b_data;
913 nbp->b_flags |= B_VALIDSUSPWRT;
914 numblks = howmany(fs->fs_size, fs->fs_frag);
915 len = howmany(fs->fs_fpg, fs->fs_frag);
916 base = cgbase(fs, cg) / fs->fs_frag;
917 if (base + len >= numblks)
918 len = numblks - base - 1;
921 for ( ; loc < NDADDR; loc++) {
922 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
923 DIP_SET(ip, i_db[loc], BLK_NOCOPY);
924 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
925 DIP_SET(ip, i_db[loc], 0);
926 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
927 panic("ffs_snapshot: lost direct block");
930 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
931 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
935 indiroff = (base + loc - NDADDR) % NINDIR(fs);
936 for ( ; loc < len; loc++, indiroff++) {
937 if (indiroff >= NINDIR(fs)) {
939 ibp->b_flags |= B_VALIDSUSPWRT;
941 error = UFS_BALLOC(vp,
942 lblktosize(fs, (off_t)(base + loc)),
943 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
949 if (ip->i_ump->um_fstype == UFS1) {
950 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
951 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
953 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
954 [indiroff] == BLK_NOCOPY)
955 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
956 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
957 [indiroff] == BLK_NOCOPY)
958 panic("ffs_snapshot: lost indirect block");
961 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
962 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
963 else if (passno == 2 &&
964 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
965 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
966 else if (passno == 1 &&
967 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
968 panic("ffs_snapshot: lost indirect block");
971 ibp->b_flags |= B_VALIDSUSPWRT;
977 * Before expunging a snapshot inode, note all the
978 * blocks that it claims with BLK_SNAP so that fsck will
979 * be able to account for those blocks properly and so
980 * that this snapshot knows that it need not copy them
981 * if the other snapshot holding them is freed. This code
982 * is reproduced once each for UFS1 and UFS2.
985 expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
986 struct vnode *snapvp;
987 struct inode *cancelip;
989 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
990 struct fs *, ufs_lbn_t, int);
994 int i, error, indiroff;
996 ufs2_daddr_t len, blkno, numblks, blksperindir;
997 struct ufs1_dinode *dip;
998 struct thread *td = curthread;
1002 * Prepare to expunge the inode. If its inode block has not
1003 * yet been copied, then allocate and fill the copy.
1005 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1008 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
1010 if (DOINGSOFTDEP(snapvp))
1011 softdep_prealloc(snapvp, MNT_WAIT);
1012 td->td_pflags |= TDP_COWINPROGRESS;
1013 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1014 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1015 td->td_pflags &= ~TDP_COWINPROGRESS;
1018 indiroff = (lbn - NDADDR) % NINDIR(fs);
1019 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
1023 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1026 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1027 fs->fs_bsize, KERNCRED, 0, &bp);
1030 if ((error = readblock(snapvp, bp, lbn)) != 0)
1034 * Set a snapshot inode to be a zero length file, regular files
1035 * or unlinked snapshots to be completely unallocated.
1037 dip = (struct ufs1_dinode *)bp->b_data +
1038 ino_to_fsbo(fs, cancelip->i_number);
1039 if (clearmode || cancelip->i_effnlink == 0)
1043 dip->di_flags &= ~SF_SNAPSHOT;
1044 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
1047 * Now go through and expunge all the blocks in the file
1048 * using the function requested.
1050 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1051 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
1052 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
1054 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
1055 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
1059 len = numblks - NDADDR;
1061 for (i = 0; len > 0 && i < NIADDR; i++) {
1062 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
1063 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
1064 blksperindir, fs, acctfunc, expungetype);
1067 blksperindir *= NINDIR(fs);
1068 lbn -= blksperindir + 1;
1069 len -= blksperindir;
1070 rlbn += blksperindir;
1076 * Descend an indirect block chain for vnode cancelvp accounting for all
1077 * its indirect blocks in snapvp.
1080 indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1081 blksperindir, fs, acctfunc, expungetype)
1082 struct vnode *snapvp;
1083 struct vnode *cancelvp;
1089 ufs_lbn_t blksperindir;
1091 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
1092 struct fs *, ufs_lbn_t, int);
1096 ufs_lbn_t subblksperindir;
1097 struct indir indirs[NIADDR + 2];
1098 ufs1_daddr_t last, *bap;
1102 if (expungetype == BLK_NOCOPY)
1104 panic("indiracct_ufs1: missing indir");
1106 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1108 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1109 panic("indiracct_ufs1: botched params");
1111 * We have to expand bread here since it will deadlock looking
1112 * up the block number for any blocks that are not in the cache.
1114 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1115 bp->b_blkno = fsbtodb(fs, blkno);
1116 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1117 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1122 * Account for the block pointers in this indirect block.
1124 last = howmany(remblks, blksperindir);
1125 if (last > NINDIR(fs))
1127 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1128 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1130 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1131 level == 0 ? rlbn : -1, expungetype);
1132 if (error || level == 0)
1135 * Account for the block pointers in each of the indirect blocks
1136 * in the levels below us.
1138 subblksperindir = blksperindir / NINDIR(fs);
1139 for (lbn++, level--, i = 0; i < last; i++) {
1140 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
1141 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1144 rlbn += blksperindir;
1145 lbn -= blksperindir;
1146 remblks -= blksperindir;
1149 free(bap, M_DEVBUF);
1154 * Do both snap accounting and map accounting.
1157 fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1159 ufs1_daddr_t *oldblkp, *lastblkp;
1162 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1166 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1168 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1172 * Identify a set of blocks allocated in a snapshot inode.
1175 snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1177 ufs1_daddr_t *oldblkp, *lastblkp;
1180 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1182 struct inode *ip = VTOI(vp);
1183 ufs1_daddr_t blkno, *blkp;
1188 for ( ; oldblkp < lastblkp; oldblkp++) {
1190 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1192 lbn = fragstoblks(fs, blkno);
1194 blkp = &ip->i_din1->di_db[lbn];
1195 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1197 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn),
1198 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1201 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1202 [(lbn - NDADDR) % NINDIR(fs)];
1205 * If we are expunging a snapshot vnode and we
1206 * find a block marked BLK_NOCOPY, then it is
1207 * one that has been allocated to this snapshot after
1208 * we took our current snapshot and can be ignored.
1210 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1215 panic("snapacct_ufs1: bad block");
1216 *blkp = expungetype;
1225 * Account for a set of blocks allocated in a snapshot inode.
1228 mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1230 ufs1_daddr_t *oldblkp, *lastblkp;
1241 inum = ip->i_number;
1246 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1248 if (blkno == 0 || blkno == BLK_NOCOPY)
1250 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1251 *ip->i_snapblklist++ = lblkno;
1252 if (blkno == BLK_SNAP)
1253 blkno = blkstofrags(fs, lblkno);
1254 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1261 * Before expunging a snapshot inode, note all the
1262 * blocks that it claims with BLK_SNAP so that fsck will
1263 * be able to account for those blocks properly and so
1264 * that this snapshot knows that it need not copy them
1265 * if the other snapshot holding them is freed. This code
1266 * is reproduced once each for UFS1 and UFS2.
1269 expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
1270 struct vnode *snapvp;
1271 struct inode *cancelip;
1273 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1274 struct fs *, ufs_lbn_t, int);
1278 int i, error, indiroff;
1279 ufs_lbn_t lbn, rlbn;
1280 ufs2_daddr_t len, blkno, numblks, blksperindir;
1281 struct ufs2_dinode *dip;
1282 struct thread *td = curthread;
1286 * Prepare to expunge the inode. If its inode block has not
1287 * yet been copied, then allocate and fill the copy.
1289 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1292 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1294 if (DOINGSOFTDEP(snapvp))
1295 softdep_prealloc(snapvp, MNT_WAIT);
1296 td->td_pflags |= TDP_COWINPROGRESS;
1297 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1298 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1299 td->td_pflags &= ~TDP_COWINPROGRESS;
1302 indiroff = (lbn - NDADDR) % NINDIR(fs);
1303 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1307 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1310 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1311 fs->fs_bsize, KERNCRED, 0, &bp);
1314 if ((error = readblock(snapvp, bp, lbn)) != 0)
1318 * Set a snapshot inode to be a zero length file, regular files
1319 * to be completely unallocated.
1321 dip = (struct ufs2_dinode *)bp->b_data +
1322 ino_to_fsbo(fs, cancelip->i_number);
1323 if (clearmode || cancelip->i_effnlink == 0)
1327 dip->di_flags &= ~SF_SNAPSHOT;
1328 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1331 * Now go through and expunge all the blocks in the file
1332 * using the function requested.
1334 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1335 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1336 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1338 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1339 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1343 len = numblks - NDADDR;
1345 for (i = 0; len > 0 && i < NIADDR; i++) {
1346 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1347 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1348 blksperindir, fs, acctfunc, expungetype);
1351 blksperindir *= NINDIR(fs);
1352 lbn -= blksperindir + 1;
1353 len -= blksperindir;
1354 rlbn += blksperindir;
1360 * Descend an indirect block chain for vnode cancelvp accounting for all
1361 * its indirect blocks in snapvp.
1364 indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1365 blksperindir, fs, acctfunc, expungetype)
1366 struct vnode *snapvp;
1367 struct vnode *cancelvp;
1373 ufs_lbn_t blksperindir;
1375 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1376 struct fs *, ufs_lbn_t, int);
1380 ufs_lbn_t subblksperindir;
1381 struct indir indirs[NIADDR + 2];
1382 ufs2_daddr_t last, *bap;
1386 if (expungetype == BLK_NOCOPY)
1388 panic("indiracct_ufs2: missing indir");
1390 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1392 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1393 panic("indiracct_ufs2: botched params");
1395 * We have to expand bread here since it will deadlock looking
1396 * up the block number for any blocks that are not in the cache.
1398 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1399 bp->b_blkno = fsbtodb(fs, blkno);
1400 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1401 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1406 * Account for the block pointers in this indirect block.
1408 last = howmany(remblks, blksperindir);
1409 if (last > NINDIR(fs))
1411 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1412 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1414 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1415 level == 0 ? rlbn : -1, expungetype);
1416 if (error || level == 0)
1419 * Account for the block pointers in each of the indirect blocks
1420 * in the levels below us.
1422 subblksperindir = blksperindir / NINDIR(fs);
1423 for (lbn++, level--, i = 0; i < last; i++) {
1424 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1425 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1428 rlbn += blksperindir;
1429 lbn -= blksperindir;
1430 remblks -= blksperindir;
1433 free(bap, M_DEVBUF);
1438 * Do both snap accounting and map accounting.
1441 fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1443 ufs2_daddr_t *oldblkp, *lastblkp;
1446 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1450 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1452 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1456 * Identify a set of blocks allocated in a snapshot inode.
1459 snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1461 ufs2_daddr_t *oldblkp, *lastblkp;
1464 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1466 struct inode *ip = VTOI(vp);
1467 ufs2_daddr_t blkno, *blkp;
1472 for ( ; oldblkp < lastblkp; oldblkp++) {
1474 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1476 lbn = fragstoblks(fs, blkno);
1478 blkp = &ip->i_din2->di_db[lbn];
1479 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1481 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn),
1482 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1485 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1486 [(lbn - NDADDR) % NINDIR(fs)];
1489 * If we are expunging a snapshot vnode and we
1490 * find a block marked BLK_NOCOPY, then it is
1491 * one that has been allocated to this snapshot after
1492 * we took our current snapshot and can be ignored.
1494 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1499 panic("snapacct_ufs2: bad block");
1500 *blkp = expungetype;
1509 * Account for a set of blocks allocated in a snapshot inode.
1512 mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1514 ufs2_daddr_t *oldblkp, *lastblkp;
1525 inum = ip->i_number;
1530 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1532 if (blkno == 0 || blkno == BLK_NOCOPY)
1534 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1535 *ip->i_snapblklist++ = lblkno;
1536 if (blkno == BLK_SNAP)
1537 blkno = blkstofrags(fs, lblkno);
1538 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1545 * Decrement extra reference on snapshot when last name is removed.
1546 * It will not be freed until the last open reference goes away.
1555 struct snapdata *sn;
1556 struct ufsmount *ump;
1559 * Find snapshot in incore list.
1562 sn = ip->i_devvp->v_rdev->si_snapdata;
1564 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap)
1570 printf("ffs_snapgone: lost snapshot vnode %d\n",
1573 * Delete snapshot inode from superblock. Keep list dense.
1578 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1579 if (fs->fs_snapinum[snaploc] == ip->i_number)
1581 if (snaploc < FSMAXSNAP) {
1582 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1583 if (fs->fs_snapinum[snaploc] == 0)
1585 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1587 fs->fs_snapinum[snaploc - 1] = 0;
1593 * Prepare a snapshot file for being removed.
1600 struct vnode *devvp;
1603 ufs2_daddr_t numblks, blkno, dblk;
1604 int error, loc, last;
1605 struct snapdata *sn;
1609 devvp = ip->i_devvp;
1611 * If active, delete from incore list (this snapshot may
1612 * already have been in the process of being deleted, so
1613 * would not have been active).
1615 * Clear copy-on-write flag if last snapshot.
1618 if (ip->i_nextsnap.tqe_prev != 0) {
1619 sn = devvp->v_rdev->si_snapdata;
1620 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap);
1621 ip->i_nextsnap.tqe_prev = 0;
1623 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
1624 KASSERT(vp->v_vnlock == &sn->sn_lock,
1625 ("ffs_snapremove: lost lock mutation"));
1626 vp->v_vnlock = &vp->v_lock;
1628 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
1629 try_free_snapdata(devvp);
1633 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1634 * snapshots that want them (see ffs_snapblkfree below).
1636 for (blkno = 1; blkno < NDADDR; blkno++) {
1637 dblk = DIP(ip, i_db[blkno]);
1640 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1641 DIP_SET(ip, i_db[blkno], 0);
1642 else if ((dblk == blkstofrags(fs, blkno) &&
1643 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1644 ip->i_number, vp->v_type, NULL))) {
1645 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
1646 btodb(fs->fs_bsize));
1647 DIP_SET(ip, i_db[blkno], 0);
1650 numblks = howmany(ip->i_size, fs->fs_bsize);
1651 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1652 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1653 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1656 if (fs->fs_size - blkno > NINDIR(fs))
1659 last = fs->fs_size - blkno;
1660 for (loc = 0; loc < last; loc++) {
1661 if (ip->i_ump->um_fstype == UFS1) {
1662 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1665 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1666 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1667 else if ((dblk == blkstofrags(fs, blkno) &&
1668 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1669 fs->fs_bsize, ip->i_number, vp->v_type,
1671 ip->i_din1->di_blocks -=
1672 btodb(fs->fs_bsize);
1673 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1677 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1680 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1681 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1682 else if ((dblk == blkstofrags(fs, blkno) &&
1683 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1684 fs->fs_bsize, ip->i_number, vp->v_type, NULL))) {
1685 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1686 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1692 * Clear snapshot flag and drop reference.
1694 ip->i_flags &= ~SF_SNAPSHOT;
1695 DIP_SET(ip, i_flags, ip->i_flags);
1696 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1698 * The dirtied indirects must be written out before
1699 * softdep_setup_freeblocks() is called. Otherwise indir_trunc()
1700 * may find indirect pointers using the magic BLK_* values.
1702 if (DOINGSOFTDEP(vp))
1703 ffs_syncvnode(vp, MNT_WAIT);
1706 * Reenable disk quotas for ex-snapshot file.
1708 if (!getinoquota(ip))
1709 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE);
1714 * Notification that a block is being freed. Return zero if the free
1715 * should be allowed to proceed. Return non-zero if the snapshot file
1716 * wants to claim the block. The block will be claimed if it is an
1717 * uncopied part of one of the snapshots. It will be freed if it is
1718 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1719 * If a fragment is being freed, then all snapshots that care about
1720 * it must make a copy since a snapshot file can only claim full sized
1721 * blocks. Note that if more than one snapshot file maps the block,
1722 * we can pick one at random to claim it. Since none of the snapshots
1723 * can change, we are assurred that they will all see the same unmodified
1724 * image. When deleting a snapshot file (see ffs_snapremove above), we
1725 * must push any of these claimed blocks to one of the other snapshots
1726 * that maps it. These claimed blocks are easily identified as they will
1727 * have a block number equal to their logical block number within the
1728 * snapshot. A copied block can never have this property because they
1729 * must always have been allocated from a BLK_NOCOPY location.
1732 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
1734 struct vnode *devvp;
1739 struct workhead *wkhd;
1741 struct buf *ibp, *cbp, *savedcbp = 0;
1742 struct thread *td = curthread;
1744 struct vnode *vp = NULL;
1747 int indiroff = 0, error = 0, claimedblk = 0;
1748 struct snapdata *sn;
1750 lbn = fragstoblks(fs, bno);
1753 sn = devvp->v_rdev->si_snapdata;
1758 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1759 VI_MTX(devvp)) != 0)
1761 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1763 if (DOINGSOFTDEP(vp))
1764 softdep_prealloc(vp, MNT_WAIT);
1766 * Lookup block being written.
1769 blkno = DIP(ip, i_db[lbn]);
1771 td->td_pflags |= TDP_COWINPROGRESS;
1772 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1773 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1774 td->td_pflags &= ~TDP_COWINPROGRESS;
1777 indiroff = (lbn - NDADDR) % NINDIR(fs);
1778 if (ip->i_ump->um_fstype == UFS1)
1779 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1781 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1784 * Check to see if block needs to be copied.
1788 * A block that we map is being freed. If it has not
1789 * been claimed yet, we will claim or copy it (below).
1792 } else if (blkno == BLK_SNAP) {
1794 * No previous snapshot claimed the block,
1795 * so it will be freed and become a BLK_NOCOPY
1796 * (don't care) for us.
1799 panic("snapblkfree: inconsistent block type");
1801 DIP_SET(ip, i_db[lbn], BLK_NOCOPY);
1802 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1803 } else if (ip->i_ump->um_fstype == UFS1) {
1804 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1808 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1813 } else /* BLK_NOCOPY or default */ {
1815 * If the snapshot has already copied the block
1816 * (default), or does not care about the block,
1824 * If this is a full size block, we will just grab it
1825 * and assign it to the snapshot inode. Otherwise we
1826 * will proceed to copy it. See explanation for this
1827 * routine as to why only a single snapshot needs to
1830 if (size == fs->fs_bsize) {
1833 printf("%s %d lbn %jd from inum %d\n",
1834 "Grabonremove: snapino", ip->i_number,
1835 (intmax_t)lbn, inum);
1838 * If journaling is tracking this write we must add
1839 * the work to the inode or indirect being written.
1843 softdep_inode_append(ip,
1844 curthread->td_ucred, wkhd);
1846 softdep_buf_append(ibp, wkhd);
1849 DIP_SET(ip, i_db[lbn], bno);
1850 } else if (ip->i_ump->um_fstype == UFS1) {
1851 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1854 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1857 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size));
1858 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1859 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1865 * Allocate the block into which to do the copy. Note that this
1866 * allocation will never require any additional allocations for
1867 * the snapshot inode.
1869 td->td_pflags |= TDP_COWINPROGRESS;
1870 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1871 fs->fs_bsize, KERNCRED, 0, &cbp);
1872 td->td_pflags &= ~TDP_COWINPROGRESS;
1877 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n",
1878 "Copyonremove: snapino ", ip->i_number,
1879 (intmax_t)lbn, "for inum", inum, size,
1880 (intmax_t)cbp->b_blkno);
1883 * If we have already read the old block contents, then
1884 * simply copy them to the new block. Note that we need
1885 * to synchronously write snapshots that have not been
1886 * unlinked, and hence will be visible after a crash,
1887 * to ensure their integrity. At a minimum we ensure the
1888 * integrity of the filesystem metadata, but use the
1889 * dopersistence sysctl-setable flag to decide on the
1890 * persistence needed for file content data.
1892 if (savedcbp != 0) {
1893 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1895 if ((vtype == VDIR || dopersistence) &&
1897 (void) ffs_syncvnode(vp, MNT_WAIT);
1901 * Otherwise, read the old block contents into the buffer.
1903 if ((error = readblock(vp, cbp, lbn)) != 0) {
1904 bzero(cbp->b_data, fs->fs_bsize);
1906 if ((vtype == VDIR || dopersistence) &&
1908 (void) ffs_syncvnode(vp, MNT_WAIT);
1914 * Note that we need to synchronously write snapshots that
1915 * have not been unlinked, and hence will be visible after
1916 * a crash, to ensure their integrity. At a minimum we
1917 * ensure the integrity of the filesystem metadata, but
1918 * use the dopersistence sysctl-setable flag to decide on
1919 * the persistence needed for file content data.
1922 vp = savedcbp->b_vp;
1924 if ((vtype == VDIR || dopersistence) &&
1925 VTOI(vp)->i_effnlink > 0)
1926 (void) ffs_syncvnode(vp, MNT_WAIT);
1929 * If we have been unable to allocate a block in which to do
1930 * the copy, then return non-zero so that the fragment will
1931 * not be freed. Although space will be lost, the snapshot
1932 * will stay consistent.
1934 if (error != 0 && wkhd != NULL)
1935 softdep_freework(wkhd);
1936 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1941 * Associate snapshot files when mounting.
1944 ffs_snapshot_mount(mp)
1947 struct ufsmount *ump = VFSTOUFS(mp);
1948 struct vnode *devvp = ump->um_devvp;
1949 struct fs *fs = ump->um_fs;
1950 struct thread *td = curthread;
1951 struct snapdata *sn;
1953 struct vnode *lastvp;
1959 daddr_t snaplistsize;
1960 int error, snaploc, loc;
1963 * XXX The following needs to be set before ffs_truncate or
1964 * VOP_READ can be called.
1966 mp->mnt_stat.f_iosize = fs->fs_bsize;
1968 * Process each snapshot listed in the superblock.
1973 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1974 if (fs->fs_snapinum[snaploc] == 0)
1976 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc],
1977 LK_EXCLUSIVE, &vp)) != 0){
1978 printf("ffs_snapshot_mount: vget failed %d\n", error);
1982 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size ==
1983 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1984 if ((ip->i_flags & SF_SNAPSHOT) == 0) {
1985 reason = "non-snapshot";
1987 reason = "old format snapshot";
1988 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED, td);
1989 (void)ffs_syncvnode(vp, MNT_WAIT);
1991 printf("ffs_snapshot_mount: %s inode %d\n",
1992 reason, fs->fs_snapinum[snaploc]);
1995 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
1996 if (fs->fs_snapinum[loc] == 0)
1998 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
2000 fs->fs_snapinum[loc - 1] = 0;
2005 * Acquire a lock on the snapdata structure, creating it if
2008 sn = ffs_snapdata_acquire(devvp);
2010 * Change vnode to use shared snapshot lock instead of the
2011 * original private lock.
2013 vp->v_vnlock = &sn->sn_lock;
2014 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2016 * Link it onto the active snapshot list.
2019 if (ip->i_nextsnap.tqe_prev != 0)
2020 panic("ffs_snapshot_mount: %d already on list",
2023 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
2024 vp->v_vflag |= VV_SYSTEM;
2031 * No usable snapshots found.
2033 if (sn == NULL || vp == NULL)
2036 * Allocate the space for the block hints list. We always want to
2037 * use the list from the newest snapshot.
2039 auio.uio_iov = &aiov;
2040 auio.uio_iovcnt = 1;
2041 aiov.iov_base = (void *)&snaplistsize;
2042 aiov.iov_len = sizeof(snaplistsize);
2043 auio.uio_resid = aiov.iov_len;
2045 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
2046 auio.uio_segflg = UIO_SYSSPACE;
2047 auio.uio_rw = UIO_READ;
2049 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2050 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2051 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
2055 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
2056 M_UFSMNT, M_WAITOK);
2057 auio.uio_iovcnt = 1;
2058 aiov.iov_base = snapblklist;
2059 aiov.iov_len = snaplistsize * sizeof (daddr_t);
2060 auio.uio_resid = aiov.iov_len;
2061 auio.uio_offset -= sizeof(snaplistsize);
2062 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2063 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
2065 free(snapblklist, M_UFSMNT);
2070 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
2071 sn->sn_listsize = snaplistsize;
2072 sn->sn_blklist = (daddr_t *)snapblklist;
2073 devvp->v_vflag |= VV_COPYONWRITE;
2078 * Disassociate snapshot files when unmounting.
2081 ffs_snapshot_unmount(mp)
2084 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
2085 struct snapdata *sn;
2090 sn = devvp->v_rdev->si_snapdata;
2091 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) {
2093 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap);
2094 xp->i_nextsnap.tqe_prev = 0;
2095 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE,
2097 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
2098 KASSERT(vp->v_vnlock == &sn->sn_lock,
2099 ("ffs_snapshot_unmount: lost lock mutation"));
2100 vp->v_vnlock = &vp->v_lock;
2101 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2102 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2103 if (xp->i_effnlink > 0)
2106 sn = devvp->v_rdev->si_snapdata;
2108 try_free_snapdata(devvp);
2109 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
2113 * Check the buffer block to be belong to device buffer that shall be
2114 * locked after snaplk. devvp shall be locked on entry, and will be
2115 * leaved locked upon exit.
2118 ffs_bp_snapblk(devvp, bp)
2119 struct vnode *devvp;
2122 struct snapdata *sn;
2124 ufs2_daddr_t lbn, *snapblklist;
2125 int lower, upper, mid;
2127 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk");
2128 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp));
2129 sn = devvp->v_rdev->si_snapdata;
2130 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL)
2132 fs = TAILQ_FIRST(&sn->sn_head)->i_fs;
2133 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2134 snapblklist = sn->sn_blklist;
2135 upper = sn->sn_listsize - 1;
2137 while (lower <= upper) {
2138 mid = (lower + upper) / 2;
2139 if (snapblklist[mid] == lbn)
2141 if (snapblklist[mid] < lbn)
2157 struct vnode *vp, *devvp;
2161 if (bo->bo_dirty.bv_cnt <= dirtybufthresh)
2166 devvp = bo->__bo_vnode;
2167 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp));
2170 bp_bdskip = ffs_bp_snapblk(devvp, bp);
2174 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) {
2175 (void) VOP_FSYNC(vp, MNT_NOWAIT, td);
2180 * Try to find a buffer to flush.
2182 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) {
2183 if ((nbp->b_vflags & BV_BKGRDINPROG) ||
2185 LK_EXCLUSIVE | LK_NOWAIT, NULL))
2188 panic("bdwrite: found ourselves");
2191 * Don't countdeps with the bo lock
2194 if (buf_countdeps(nbp, 0)) {
2201 if (!ffs_bp_snapblk(vp, nbp)) {
2202 if (BO_MTX(bo) != VI_MTX(vp)) {
2211 if (nbp->b_flags & B_CLUSTEROK) {
2212 vfs_bio_awrite(nbp);
2217 dirtybufferflushes++;
2226 * Check for need to copy block that is about to be written,
2227 * copying the block if necessary.
2230 ffs_copyonwrite(devvp, bp)
2231 struct vnode *devvp;
2234 struct snapdata *sn;
2235 struct buf *ibp, *cbp, *savedcbp = 0;
2236 struct thread *td = curthread;
2239 struct vnode *vp = 0;
2240 ufs2_daddr_t lbn, blkno, *snapblklist;
2241 int lower, upper, mid, indiroff, error = 0;
2242 int launched_async_io, prev_norunningbuf;
2243 long saved_runningbufspace;
2245 if (devvp != bp->b_vp && (VTOI(bp->b_vp)->i_flags & SF_SNAPSHOT) != 0)
2246 return (0); /* Update on a snapshot file */
2247 if (td->td_pflags & TDP_COWINPROGRESS)
2248 panic("ffs_copyonwrite: recursive call");
2250 * First check to see if it is in the preallocated list.
2251 * By doing this check we avoid several potential deadlocks.
2254 sn = devvp->v_rdev->si_snapdata;
2256 TAILQ_EMPTY(&sn->sn_head)) {
2258 return (0); /* No snapshot */
2260 ip = TAILQ_FIRST(&sn->sn_head);
2262 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2263 snapblklist = sn->sn_blklist;
2264 upper = sn->sn_listsize - 1;
2266 while (lower <= upper) {
2267 mid = (lower + upper) / 2;
2268 if (snapblklist[mid] == lbn)
2270 if (snapblklist[mid] < lbn)
2275 if (lower <= upper) {
2279 launched_async_io = 0;
2280 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF;
2282 * Since I/O on bp isn't yet in progress and it may be blocked
2283 * for a long time waiting on snaplk, back it out of
2284 * runningbufspace, possibly waking other threads waiting for space.
2286 saved_runningbufspace = bp->b_runningbufspace;
2287 if (saved_runningbufspace != 0)
2288 runningbufwakeup(bp);
2290 * Not in the precomputed list, so check the snapshots.
2292 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2293 VI_MTX(devvp)) != 0) {
2295 sn = devvp->v_rdev->si_snapdata;
2297 TAILQ_EMPTY(&sn->sn_head)) {
2299 if (saved_runningbufspace != 0) {
2300 bp->b_runningbufspace = saved_runningbufspace;
2301 atomic_add_long(&runningbufspace,
2302 bp->b_runningbufspace);
2304 return (0); /* Snapshot gone */
2307 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2309 if (DOINGSOFTDEP(vp))
2310 softdep_prealloc(vp, MNT_WAIT);
2312 * We ensure that everything of our own that needs to be
2313 * copied will be done at the time that ffs_snapshot is
2314 * called. Thus we can skip the check here which can
2315 * deadlock in doing the lookup in UFS_BALLOC.
2320 * Check to see if block needs to be copied. We do not have
2321 * to hold the snapshot lock while doing this lookup as it
2322 * will never require any additional allocations for the
2326 blkno = DIP(ip, i_db[lbn]);
2328 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2329 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2330 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
2331 td->td_pflags &= ~TDP_COWINPROGRESS;
2334 indiroff = (lbn - NDADDR) % NINDIR(fs);
2335 if (ip->i_ump->um_fstype == UFS1)
2336 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2338 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2342 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2343 panic("ffs_copyonwrite: bad copy block");
2348 * Allocate the block into which to do the copy. Since
2349 * multiple processes may all try to copy the same block,
2350 * we have to recheck our need to do a copy if we sleep
2351 * waiting for the lock.
2353 * Because all snapshots on a filesystem share a single
2354 * lock, we ensure that we will never be in competition
2355 * with another process to allocate a block.
2357 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2358 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2359 fs->fs_bsize, KERNCRED, 0, &cbp);
2360 td->td_pflags &= ~TDP_COWINPROGRESS;
2365 printf("Copyonwrite: snapino %d lbn %jd for ",
2366 ip->i_number, (intmax_t)lbn);
2367 if (bp->b_vp == devvp)
2368 printf("fs metadata");
2370 printf("inum %d", VTOI(bp->b_vp)->i_number);
2371 printf(" lblkno %jd to blkno %jd\n",
2372 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2376 * If we have already read the old block contents, then
2377 * simply copy them to the new block. Note that we need
2378 * to synchronously write snapshots that have not been
2379 * unlinked, and hence will be visible after a crash,
2380 * to ensure their integrity. At a minimum we ensure the
2381 * integrity of the filesystem metadata, but use the
2382 * dopersistence sysctl-setable flag to decide on the
2383 * persistence needed for file content data.
2385 if (savedcbp != 0) {
2386 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2388 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2389 dopersistence) && ip->i_effnlink > 0)
2390 (void) ffs_syncvnode(vp, MNT_WAIT);
2392 launched_async_io = 1;
2396 * Otherwise, read the old block contents into the buffer.
2398 if ((error = readblock(vp, cbp, lbn)) != 0) {
2399 bzero(cbp->b_data, fs->fs_bsize);
2401 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2402 dopersistence) && ip->i_effnlink > 0)
2403 (void) ffs_syncvnode(vp, MNT_WAIT);
2405 launched_async_io = 1;
2411 * Note that we need to synchronously write snapshots that
2412 * have not been unlinked, and hence will be visible after
2413 * a crash, to ensure their integrity. At a minimum we
2414 * ensure the integrity of the filesystem metadata, but
2415 * use the dopersistence sysctl-setable flag to decide on
2416 * the persistence needed for file content data.
2419 vp = savedcbp->b_vp;
2421 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2422 dopersistence) && VTOI(vp)->i_effnlink > 0)
2423 (void) ffs_syncvnode(vp, MNT_WAIT);
2425 launched_async_io = 1;
2427 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
2428 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) |
2430 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0)
2431 waitrunningbufspace();
2433 * I/O on bp will now be started, so count it in runningbufspace.
2435 if (saved_runningbufspace != 0) {
2436 bp->b_runningbufspace = saved_runningbufspace;
2437 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
2443 * sync snapshots to force freework records waiting on snapshots to claim
2447 ffs_sync_snap(mp, waitfor)
2451 struct snapdata *sn;
2452 struct vnode *devvp;
2456 devvp = VFSTOUFS(mp)->um_devvp;
2457 if ((devvp->v_vflag & VV_COPYONWRITE) == 0)
2461 sn = devvp->v_rdev->si_snapdata;
2466 if (lockmgr(&sn->sn_lock,
2467 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2468 VI_MTX(devvp)) == 0)
2471 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2473 ffs_syncvnode(vp, waitfor);
2475 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2479 * Read the specified block into the given buffer.
2480 * Much of this boiler-plate comes from bwrite().
2483 readblock(vp, bp, lbn)
2488 struct inode *ip = VTOI(vp);
2491 bip = g_alloc_bio();
2492 bip->bio_cmd = BIO_READ;
2493 bip->bio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2494 bip->bio_data = bp->b_data;
2495 bip->bio_length = bp->b_bcount;
2496 bip->bio_done = NULL;
2498 g_io_request(bip, ip->i_devvp->v_bufobj.bo_private);
2499 bp->b_error = biowait(bip, "snaprdb");
2501 return (bp->b_error);
2507 * Process file deletes that were deferred by ufs_inactive() due to
2508 * the file system being suspended. Transfer IN_LAZYACCESS into
2509 * IN_MODIFIED for vnodes that were accessed during suspension.
2512 process_deferred_inactive(struct mount *mp)
2514 struct vnode *vp, *mvp;
2520 (void) vn_start_secondary_write(NULL, &mp, V_WAIT);
2523 MNT_VNODE_FOREACH(vp, mp, mvp) {
2526 * IN_LAZYACCESS is checked here without holding any
2527 * vnode lock, but this flag is set only while holding
2530 if (vp->v_type == VNON || (vp->v_iflag & VI_DOOMED) != 0 ||
2531 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 &&
2532 ((vp->v_iflag & VI_OWEINACT) == 0 ||
2533 vp->v_usecount > 0))) {
2539 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2543 if (error == ENOENT)
2544 continue; /* vnode recycled */
2545 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2549 if ((ip->i_flag & IN_LAZYACCESS) != 0) {
2550 ip->i_flag &= ~IN_LAZYACCESS;
2551 ip->i_flag |= IN_MODIFIED;
2554 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) {
2562 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2563 ("process_deferred_inactive: "
2564 "recursed on VI_DOINGINACT"));
2565 vp->v_iflag |= VI_DOINGINACT;
2566 vp->v_iflag &= ~VI_OWEINACT;
2568 (void) VOP_INACTIVE(vp, td);
2570 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2571 ("process_deferred_inactive: lost VI_DOINGINACT"));
2572 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2573 ("process_deferred_inactive: got VI_OWEINACT"));
2574 vp->v_iflag &= ~VI_DOINGINACT;
2581 vn_finished_secondary_write(mp);
2584 #ifndef NO_FFS_SNAPSHOT
2586 static struct snapdata *
2587 ffs_snapdata_alloc(void)
2589 struct snapdata *sn;
2592 * Fetch a snapdata from the free list if there is one available.
2594 mtx_lock(&snapfree_lock);
2595 sn = LIST_FIRST(&snapfree);
2597 LIST_REMOVE(sn, sn_link);
2598 mtx_unlock(&snapfree_lock);
2602 * If there were no free snapdatas allocate one.
2604 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
2605 TAILQ_INIT(&sn->sn_head);
2606 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
2607 LK_CANRECURSE | LK_NOSHARE);
2612 * The snapdata is never freed because we can not be certain that
2613 * there are no threads sleeping on the snap lock. Persisting
2614 * them permanently avoids costly synchronization in ffs_lock().
2617 ffs_snapdata_free(struct snapdata *sn)
2619 mtx_lock(&snapfree_lock);
2620 LIST_INSERT_HEAD(&snapfree, sn, sn_link);
2621 mtx_unlock(&snapfree_lock);
2624 /* Try to free snapdata associated with devvp */
2626 try_free_snapdata(struct vnode *devvp)
2628 struct snapdata *sn;
2629 ufs2_daddr_t *snapblklist;
2631 ASSERT_VI_LOCKED(devvp, "try_free_snapdata");
2632 sn = devvp->v_rdev->si_snapdata;
2634 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL ||
2635 (devvp->v_vflag & VV_COPYONWRITE) == 0) {
2640 devvp->v_rdev->si_snapdata = NULL;
2641 devvp->v_vflag &= ~VV_COPYONWRITE;
2642 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp));
2643 snapblklist = sn->sn_blklist;
2644 sn->sn_blklist = NULL;
2645 sn->sn_listsize = 0;
2646 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2647 if (snapblklist != NULL)
2648 free(snapblklist, M_UFSMNT);
2649 ffs_snapdata_free(sn);
2652 static struct snapdata *
2653 ffs_snapdata_acquire(struct vnode *devvp)
2655 struct snapdata *nsn;
2656 struct snapdata *sn;
2659 * Allocate a free snapdata. This is done before acquiring the
2660 * devvp lock to avoid allocation while the devvp interlock is
2663 nsn = ffs_snapdata_alloc();
2665 * If there snapshots already exist on this filesystem grab a
2666 * reference to the shared lock. Otherwise this is the first
2667 * snapshot on this filesystem and we need to use our
2668 * pre-allocated snapdata.
2671 if (devvp->v_rdev->si_snapdata == NULL) {
2672 devvp->v_rdev->si_snapdata = nsn;
2675 sn = devvp->v_rdev->si_snapdata;
2677 * Acquire the snapshot lock.
2679 lockmgr(&sn->sn_lock,
2680 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, VI_MTX(devvp));
2682 * Free any unused snapdata.
2685 ffs_snapdata_free(nsn);