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, *sbp = NULL;
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 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
470 loc = blkoff(fs, fs->fs_sblockloc);
471 copy_fs = (struct fs *)(sbp->b_data + loc);
472 bcopy(fs, copy_fs, fs->fs_sbsize);
473 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
474 copy_fs->fs_clean = 1;
475 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
476 if (fs->fs_sbsize < size)
477 bzero(&sbp->b_data[loc + fs->fs_sbsize], size - fs->fs_sbsize);
478 size = blkroundup(fs, fs->fs_cssize);
479 if (fs->fs_contigsumsize > 0)
480 size += fs->fs_ncg * sizeof(int32_t);
481 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
482 copy_fs->fs_csp = space;
483 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
484 space = (char *)space + fs->fs_cssize;
485 loc = howmany(fs->fs_cssize, fs->fs_fsize);
486 i = fs->fs_frag - loc % fs->fs_frag;
487 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
489 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
490 len, KERNCRED, &bp)) != 0) {
492 free(copy_fs->fs_csp, M_UFSMNT);
497 bcopy(bp->b_data, space, (u_int)len);
498 space = (char *)space + len;
499 bp->b_flags |= B_INVAL | B_NOCACHE;
502 if (fs->fs_contigsumsize > 0) {
503 copy_fs->fs_maxcluster = lp = space;
504 for (i = 0; i < fs->fs_ncg; i++)
505 *lp++ = fs->fs_contigsumsize;
508 * We must check for active files that have been unlinked
509 * (e.g., with a zero link count). We have to expunge all
510 * trace of these files from the snapshot so that they are
511 * not reclaimed prematurely by fsck or unnecessarily dumped.
512 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
513 * spec_strategy about writing on a suspended filesystem.
514 * Note that we skip unlinked snapshot files as they will
515 * be handled separately below.
517 * We also calculate the needed size for the snapshot list.
519 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
520 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
522 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
524 MNT_VNODE_FOREACH(xvp, mp, mvp) {
527 if ((xvp->v_iflag & VI_DOOMED) ||
528 (xvp->v_usecount == 0 &&
529 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) ||
530 xvp->v_type == VNON ||
531 (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
537 * We can skip parent directory vnode because it must have
538 * this snapshot file in it.
540 if (xvp == nd.ni_dvp) {
546 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) {
548 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
553 if (xvp->v_usecount == 0 &&
554 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) {
563 vprint("ffs_snapshot: busy vnode", xvp);
564 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 &&
572 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
579 * If there is a fragment, clear it here.
582 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
584 len = fragroundup(fs, blkoff(fs, xp->i_size));
585 if (len != 0 && len < fs->fs_bsize) {
586 ffs_blkfree(ump, copy_fs, vp,
587 DIP(xp, i_db[loc]), len, xp->i_number,
589 blkno = DIP(xp, i_db[loc]);
590 DIP_SET(xp, i_db[loc], 0);
594 if (xp->i_ump->um_fstype == UFS1)
595 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
598 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
601 DIP_SET(xp, i_db[loc], blkno);
603 error = ffs_freefile(ump, copy_fs, vp, xp->i_number,
608 free(copy_fs->fs_csp, M_UFSMNT);
611 MNT_VNODE_FOREACH_ABORT(mp, mvp);
618 * Erase the journal file from the snapshot.
620 if (fs->fs_flags & FS_SUJ) {
621 error = softdep_journal_lookup(mp, &xvp);
623 free(copy_fs->fs_csp, M_UFSMNT);
629 if (xp->i_ump->um_fstype == UFS1)
630 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
633 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
638 * Acquire a lock on the snapdata structure, creating it if necessary.
640 sn = ffs_snapdata_acquire(devvp);
642 * Change vnode to use shared snapshot lock instead of the original
645 vp->v_vnlock = &sn->sn_lock;
646 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
647 xp = TAILQ_FIRST(&sn->sn_head);
649 * If this is the first snapshot on this filesystem, then we need
650 * to allocate the space for the list of preallocated snapshot blocks.
651 * This list will be refined below, but this preliminary one will
652 * keep us out of deadlock until the full one is ready.
655 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
657 blkp = &snapblklist[1];
658 *blkp++ = lblkno(fs, fs->fs_sblockloc);
659 blkno = fragstoblks(fs, fs->fs_csaddr);
660 for (cg = 0; cg < fs->fs_ncg; cg++) {
661 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
663 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
665 len = howmany(fs->fs_cssize, fs->fs_bsize);
666 for (loc = 0; loc < len; loc++)
667 *blkp++ = blkno + loc;
668 for (; cg < fs->fs_ncg; cg++)
669 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
670 snapblklist[0] = blkp - snapblklist;
672 if (sn->sn_blklist != NULL)
673 panic("ffs_snapshot: non-empty list");
674 sn->sn_blklist = snapblklist;
675 sn->sn_listsize = blkp - snapblklist;
679 * Record snapshot inode. Since this is the newest snapshot,
680 * it must be placed at the end of the list.
683 fs->fs_snapinum[snaploc] = ip->i_number;
684 if (ip->i_nextsnap.tqe_prev != 0)
685 panic("ffs_snapshot: %d already on list", ip->i_number);
686 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
687 devvp->v_vflag |= VV_COPYONWRITE;
689 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
691 KASSERT((sn != NULL && sbp != NULL && error == 0) ||
692 (sn == NULL && sbp == NULL && error != 0),
693 ("email phk@ and mckusick@"));
695 * Resume operation on filesystem.
697 vfs_write_resume(vp->v_mount);
698 vn_start_write(NULL, &wrtmp, V_WAIT);
699 if (collectsnapstats && starttime.tv_sec > 0) {
701 timespecsub(&endtime, &starttime);
702 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
703 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
704 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
709 * Copy allocation information from all the snapshots in
710 * this snapshot and then expunge them from its view.
712 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) {
715 if (xp->i_ump->um_fstype == UFS1)
716 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
719 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
721 if (error == 0 && xp->i_effnlink == 0) {
722 error = ffs_freefile(ump,
729 fs->fs_snapinum[snaploc] = 0;
734 * Allocate space for the full list of preallocated snapshot blocks.
736 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
738 ip->i_snapblklist = &snapblklist[1];
740 * Expunge the blocks used by the snapshots from the set of
741 * blocks marked as used in the snapshot bitmaps. Also, collect
742 * the list of allocated blocks in i_snapblklist.
744 if (ip->i_ump->um_fstype == UFS1)
745 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1,
748 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2,
751 fs->fs_snapinum[snaploc] = 0;
752 free(snapblklist, M_UFSMNT);
755 if (snaplistsize < ip->i_snapblklist - snapblklist)
756 panic("ffs_snapshot: list too small");
757 snaplistsize = ip->i_snapblklist - snapblklist;
758 snapblklist[0] = snaplistsize;
759 ip->i_snapblklist = 0;
761 * Write out the list of allocated blocks to the end of the snapshot.
763 auio.uio_iov = &aiov;
765 aiov.iov_base = (void *)snapblklist;
766 aiov.iov_len = snaplistsize * sizeof(daddr_t);
767 auio.uio_resid = aiov.iov_len;
768 auio.uio_offset = ip->i_size;
769 auio.uio_segflg = UIO_SYSSPACE;
770 auio.uio_rw = UIO_WRITE;
772 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
773 fs->fs_snapinum[snaploc] = 0;
774 free(snapblklist, M_UFSMNT);
778 * Write the superblock and its summary information
781 blkno = fragstoblks(fs, fs->fs_csaddr);
782 len = howmany(fs->fs_cssize, fs->fs_bsize);
783 space = copy_fs->fs_csp;
784 for (loc = 0; loc < len; loc++) {
785 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
788 fs->fs_snapinum[snaploc] = 0;
789 free(snapblklist, M_UFSMNT);
792 bcopy(space, nbp->b_data, fs->fs_bsize);
793 space = (char *)space + fs->fs_bsize;
797 * As this is the newest list, it is the most inclusive, so
798 * should replace the previous list.
801 space = sn->sn_blklist;
802 sn->sn_blklist = snapblklist;
803 sn->sn_listsize = snaplistsize;
806 free(space, M_UFSMNT);
808 * If another process is currently writing the buffer containing
809 * the inode for this snapshot then a deadlock can occur. Drop
810 * the snapshot lock until the buffer has been written.
812 VREF(vp); /* Protect against ffs_snapgone() */
814 (void) bread(ip->i_devvp,
815 fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
816 (int) fs->fs_bsize, NOCRED, &nbp);
818 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
819 if (ip->i_effnlink == 0)
820 error = ENOENT; /* Snapshot file unlinked */
822 vrele(vp); /* Drop extra reference */
824 free(copy_fs->fs_csp, M_UFSMNT);
827 NDFREE(&nd, NDF_ONLY_PNBUF);
828 if (saved_nice > 0) {
833 sched_nice(td->td_proc, saved_nice);
834 PROC_UNLOCK(td->td_proc);
837 if (fs->fs_active != 0) {
838 free(fs->fs_active, M_DEVBUF);
843 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
846 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED, td);
847 (void) ffs_syncvnode(vp, MNT_WAIT);
853 vn_finished_write(wrtmp);
854 process_deferred_inactive(mp);
859 * Copy a cylinder group map. All the unallocated blocks are marked
860 * BLK_NOCOPY so that the snapshot knows that it need not copy them
861 * if they are later written. If passno is one, then this is a first
862 * pass, so only setting needs to be done. If passno is 2, then this
863 * is a revision to a previous pass which must be undone as the
864 * replacement pass is done.
867 cgaccount(cg, vp, nbp, passno)
873 struct buf *bp, *ibp;
877 ufs2_daddr_t base, numblks;
878 int error, len, loc, indiroff;
882 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
883 (int)fs->fs_cgsize, KERNCRED, &bp);
888 cgp = (struct cg *)bp->b_data;
889 if (!cg_chkmagic(cgp)) {
896 * Recomputation of summary information might not have been performed
897 * at mount time. Sync up summary information for current cylinder
898 * group while data is in memory to ensure that result of background
899 * fsck is slightly more consistent.
901 fs->fs_cs(fs, cg) = cgp->cg_cs;
902 UFS_UNLOCK(ip->i_ump);
903 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
904 if (fs->fs_cgsize < fs->fs_bsize)
905 bzero(&nbp->b_data[fs->fs_cgsize],
906 fs->fs_bsize - fs->fs_cgsize);
907 cgp = (struct cg *)nbp->b_data;
910 nbp->b_flags |= B_VALIDSUSPWRT;
911 numblks = howmany(fs->fs_size, fs->fs_frag);
912 len = howmany(fs->fs_fpg, fs->fs_frag);
913 base = cgbase(fs, cg) / fs->fs_frag;
914 if (base + len >= numblks)
915 len = numblks - base - 1;
918 for ( ; loc < NDADDR; loc++) {
919 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
920 DIP_SET(ip, i_db[loc], BLK_NOCOPY);
921 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
922 DIP_SET(ip, i_db[loc], 0);
923 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
924 panic("ffs_snapshot: lost direct block");
927 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
928 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
932 indiroff = (base + loc - NDADDR) % NINDIR(fs);
933 for ( ; loc < len; loc++, indiroff++) {
934 if (indiroff >= NINDIR(fs)) {
936 ibp->b_flags |= B_VALIDSUSPWRT;
938 error = UFS_BALLOC(vp,
939 lblktosize(fs, (off_t)(base + loc)),
940 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
946 if (ip->i_ump->um_fstype == UFS1) {
947 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
948 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
950 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
951 [indiroff] == BLK_NOCOPY)
952 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
953 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
954 [indiroff] == BLK_NOCOPY)
955 panic("ffs_snapshot: lost indirect block");
958 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
959 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
960 else if (passno == 2 &&
961 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
962 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
963 else if (passno == 1 &&
964 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
965 panic("ffs_snapshot: lost indirect block");
968 ibp->b_flags |= B_VALIDSUSPWRT;
974 * Before expunging a snapshot inode, note all the
975 * blocks that it claims with BLK_SNAP so that fsck will
976 * be able to account for those blocks properly and so
977 * that this snapshot knows that it need not copy them
978 * if the other snapshot holding them is freed. This code
979 * is reproduced once each for UFS1 and UFS2.
982 expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
983 struct vnode *snapvp;
984 struct inode *cancelip;
986 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
987 struct fs *, ufs_lbn_t, int);
991 int i, error, indiroff;
993 ufs2_daddr_t len, blkno, numblks, blksperindir;
994 struct ufs1_dinode *dip;
995 struct thread *td = curthread;
999 * Prepare to expunge the inode. If its inode block has not
1000 * yet been copied, then allocate and fill the copy.
1002 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1005 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
1007 if (DOINGSOFTDEP(snapvp))
1008 softdep_prealloc(snapvp, MNT_WAIT);
1009 td->td_pflags |= TDP_COWINPROGRESS;
1010 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1011 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1012 td->td_pflags &= ~TDP_COWINPROGRESS;
1015 indiroff = (lbn - NDADDR) % NINDIR(fs);
1016 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
1020 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1023 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1024 fs->fs_bsize, KERNCRED, 0, &bp);
1027 if ((error = readblock(snapvp, bp, lbn)) != 0)
1031 * Set a snapshot inode to be a zero length file, regular files
1032 * or unlinked snapshots to be completely unallocated.
1034 dip = (struct ufs1_dinode *)bp->b_data +
1035 ino_to_fsbo(fs, cancelip->i_number);
1036 if (clearmode || cancelip->i_effnlink == 0)
1040 dip->di_flags &= ~SF_SNAPSHOT;
1041 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
1044 * Now go through and expunge all the blocks in the file
1045 * using the function requested.
1047 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1048 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
1049 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
1051 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
1052 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
1056 len = numblks - NDADDR;
1058 for (i = 0; len > 0 && i < NIADDR; i++) {
1059 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
1060 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
1061 blksperindir, fs, acctfunc, expungetype);
1064 blksperindir *= NINDIR(fs);
1065 lbn -= blksperindir + 1;
1066 len -= blksperindir;
1067 rlbn += blksperindir;
1073 * Descend an indirect block chain for vnode cancelvp accounting for all
1074 * its indirect blocks in snapvp.
1077 indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1078 blksperindir, fs, acctfunc, expungetype)
1079 struct vnode *snapvp;
1080 struct vnode *cancelvp;
1086 ufs_lbn_t blksperindir;
1088 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
1089 struct fs *, ufs_lbn_t, int);
1093 ufs_lbn_t subblksperindir;
1094 struct indir indirs[NIADDR + 2];
1095 ufs1_daddr_t last, *bap;
1099 if (expungetype == BLK_NOCOPY)
1101 panic("indiracct_ufs1: missing indir");
1103 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1105 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1106 panic("indiracct_ufs1: botched params");
1108 * We have to expand bread here since it will deadlock looking
1109 * up the block number for any blocks that are not in the cache.
1111 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1112 bp->b_blkno = fsbtodb(fs, blkno);
1113 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1114 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1119 * Account for the block pointers in this indirect block.
1121 last = howmany(remblks, blksperindir);
1122 if (last > NINDIR(fs))
1124 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1125 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1127 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1128 level == 0 ? rlbn : -1, expungetype);
1129 if (error || level == 0)
1132 * Account for the block pointers in each of the indirect blocks
1133 * in the levels below us.
1135 subblksperindir = blksperindir / NINDIR(fs);
1136 for (lbn++, level--, i = 0; i < last; i++) {
1137 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
1138 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1141 rlbn += blksperindir;
1142 lbn -= blksperindir;
1143 remblks -= blksperindir;
1146 free(bap, M_DEVBUF);
1151 * Do both snap accounting and map accounting.
1154 fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1156 ufs1_daddr_t *oldblkp, *lastblkp;
1159 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1163 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1165 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1169 * Identify a set of blocks allocated in a snapshot inode.
1172 snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1174 ufs1_daddr_t *oldblkp, *lastblkp;
1177 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1179 struct inode *ip = VTOI(vp);
1180 ufs1_daddr_t blkno, *blkp;
1185 for ( ; oldblkp < lastblkp; oldblkp++) {
1187 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1189 lbn = fragstoblks(fs, blkno);
1191 blkp = &ip->i_din1->di_db[lbn];
1192 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1194 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn),
1195 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1198 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1199 [(lbn - NDADDR) % NINDIR(fs)];
1202 * If we are expunging a snapshot vnode and we
1203 * find a block marked BLK_NOCOPY, then it is
1204 * one that has been allocated to this snapshot after
1205 * we took our current snapshot and can be ignored.
1207 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1212 panic("snapacct_ufs1: bad block");
1213 *blkp = expungetype;
1222 * Account for a set of blocks allocated in a snapshot inode.
1225 mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1227 ufs1_daddr_t *oldblkp, *lastblkp;
1238 inum = ip->i_number;
1243 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1245 if (blkno == 0 || blkno == BLK_NOCOPY)
1247 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1248 *ip->i_snapblklist++ = lblkno;
1249 if (blkno == BLK_SNAP)
1250 blkno = blkstofrags(fs, lblkno);
1251 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1258 * Before expunging a snapshot inode, note all the
1259 * blocks that it claims with BLK_SNAP so that fsck will
1260 * be able to account for those blocks properly and so
1261 * that this snapshot knows that it need not copy them
1262 * if the other snapshot holding them is freed. This code
1263 * is reproduced once each for UFS1 and UFS2.
1266 expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
1267 struct vnode *snapvp;
1268 struct inode *cancelip;
1270 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1271 struct fs *, ufs_lbn_t, int);
1275 int i, error, indiroff;
1276 ufs_lbn_t lbn, rlbn;
1277 ufs2_daddr_t len, blkno, numblks, blksperindir;
1278 struct ufs2_dinode *dip;
1279 struct thread *td = curthread;
1283 * Prepare to expunge the inode. If its inode block has not
1284 * yet been copied, then allocate and fill the copy.
1286 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1289 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1291 if (DOINGSOFTDEP(snapvp))
1292 softdep_prealloc(snapvp, MNT_WAIT);
1293 td->td_pflags |= TDP_COWINPROGRESS;
1294 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1295 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1296 td->td_pflags &= ~TDP_COWINPROGRESS;
1299 indiroff = (lbn - NDADDR) % NINDIR(fs);
1300 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1304 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1307 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1308 fs->fs_bsize, KERNCRED, 0, &bp);
1311 if ((error = readblock(snapvp, bp, lbn)) != 0)
1315 * Set a snapshot inode to be a zero length file, regular files
1316 * to be completely unallocated.
1318 dip = (struct ufs2_dinode *)bp->b_data +
1319 ino_to_fsbo(fs, cancelip->i_number);
1320 if (clearmode || cancelip->i_effnlink == 0)
1324 dip->di_flags &= ~SF_SNAPSHOT;
1325 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1328 * Now go through and expunge all the blocks in the file
1329 * using the function requested.
1331 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1332 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1333 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1335 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1336 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1340 len = numblks - NDADDR;
1342 for (i = 0; len > 0 && i < NIADDR; i++) {
1343 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1344 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1345 blksperindir, fs, acctfunc, expungetype);
1348 blksperindir *= NINDIR(fs);
1349 lbn -= blksperindir + 1;
1350 len -= blksperindir;
1351 rlbn += blksperindir;
1357 * Descend an indirect block chain for vnode cancelvp accounting for all
1358 * its indirect blocks in snapvp.
1361 indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1362 blksperindir, fs, acctfunc, expungetype)
1363 struct vnode *snapvp;
1364 struct vnode *cancelvp;
1370 ufs_lbn_t blksperindir;
1372 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1373 struct fs *, ufs_lbn_t, int);
1377 ufs_lbn_t subblksperindir;
1378 struct indir indirs[NIADDR + 2];
1379 ufs2_daddr_t last, *bap;
1383 if (expungetype == BLK_NOCOPY)
1385 panic("indiracct_ufs2: missing indir");
1387 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1389 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1390 panic("indiracct_ufs2: botched params");
1392 * We have to expand bread here since it will deadlock looking
1393 * up the block number for any blocks that are not in the cache.
1395 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1396 bp->b_blkno = fsbtodb(fs, blkno);
1397 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1398 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1403 * Account for the block pointers in this indirect block.
1405 last = howmany(remblks, blksperindir);
1406 if (last > NINDIR(fs))
1408 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1409 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1411 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1412 level == 0 ? rlbn : -1, expungetype);
1413 if (error || level == 0)
1416 * Account for the block pointers in each of the indirect blocks
1417 * in the levels below us.
1419 subblksperindir = blksperindir / NINDIR(fs);
1420 for (lbn++, level--, i = 0; i < last; i++) {
1421 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1422 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1425 rlbn += blksperindir;
1426 lbn -= blksperindir;
1427 remblks -= blksperindir;
1430 free(bap, M_DEVBUF);
1435 * Do both snap accounting and map accounting.
1438 fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1440 ufs2_daddr_t *oldblkp, *lastblkp;
1443 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1447 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1449 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1453 * Identify a set of blocks allocated in a snapshot inode.
1456 snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1458 ufs2_daddr_t *oldblkp, *lastblkp;
1461 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1463 struct inode *ip = VTOI(vp);
1464 ufs2_daddr_t blkno, *blkp;
1469 for ( ; oldblkp < lastblkp; oldblkp++) {
1471 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1473 lbn = fragstoblks(fs, blkno);
1475 blkp = &ip->i_din2->di_db[lbn];
1476 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1478 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn),
1479 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1482 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1483 [(lbn - NDADDR) % NINDIR(fs)];
1486 * If we are expunging a snapshot vnode and we
1487 * find a block marked BLK_NOCOPY, then it is
1488 * one that has been allocated to this snapshot after
1489 * we took our current snapshot and can be ignored.
1491 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1496 panic("snapacct_ufs2: bad block");
1497 *blkp = expungetype;
1506 * Account for a set of blocks allocated in a snapshot inode.
1509 mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1511 ufs2_daddr_t *oldblkp, *lastblkp;
1522 inum = ip->i_number;
1527 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1529 if (blkno == 0 || blkno == BLK_NOCOPY)
1531 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1532 *ip->i_snapblklist++ = lblkno;
1533 if (blkno == BLK_SNAP)
1534 blkno = blkstofrags(fs, lblkno);
1535 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1542 * Decrement extra reference on snapshot when last name is removed.
1543 * It will not be freed until the last open reference goes away.
1552 struct snapdata *sn;
1553 struct ufsmount *ump;
1556 * Find snapshot in incore list.
1559 sn = ip->i_devvp->v_rdev->si_snapdata;
1561 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap)
1567 printf("ffs_snapgone: lost snapshot vnode %d\n",
1570 * Delete snapshot inode from superblock. Keep list dense.
1575 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1576 if (fs->fs_snapinum[snaploc] == ip->i_number)
1578 if (snaploc < FSMAXSNAP) {
1579 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1580 if (fs->fs_snapinum[snaploc] == 0)
1582 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1584 fs->fs_snapinum[snaploc - 1] = 0;
1590 * Prepare a snapshot file for being removed.
1597 struct vnode *devvp;
1600 ufs2_daddr_t numblks, blkno, dblk;
1601 int error, loc, last;
1602 struct snapdata *sn;
1606 devvp = ip->i_devvp;
1608 * If active, delete from incore list (this snapshot may
1609 * already have been in the process of being deleted, so
1610 * would not have been active).
1612 * Clear copy-on-write flag if last snapshot.
1615 if (ip->i_nextsnap.tqe_prev != 0) {
1616 sn = devvp->v_rdev->si_snapdata;
1617 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap);
1618 ip->i_nextsnap.tqe_prev = 0;
1620 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
1621 KASSERT(vp->v_vnlock == &sn->sn_lock,
1622 ("ffs_snapremove: lost lock mutation"));
1623 vp->v_vnlock = &vp->v_lock;
1625 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
1626 try_free_snapdata(devvp);
1630 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1631 * snapshots that want them (see ffs_snapblkfree below).
1633 for (blkno = 1; blkno < NDADDR; blkno++) {
1634 dblk = DIP(ip, i_db[blkno]);
1637 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1638 DIP_SET(ip, i_db[blkno], 0);
1639 else if ((dblk == blkstofrags(fs, blkno) &&
1640 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1641 ip->i_number, vp->v_type, NULL))) {
1642 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
1643 btodb(fs->fs_bsize));
1644 DIP_SET(ip, i_db[blkno], 0);
1647 numblks = howmany(ip->i_size, fs->fs_bsize);
1648 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1649 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1650 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1653 if (fs->fs_size - blkno > NINDIR(fs))
1656 last = fs->fs_size - blkno;
1657 for (loc = 0; loc < last; loc++) {
1658 if (ip->i_ump->um_fstype == UFS1) {
1659 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1662 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1663 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1664 else if ((dblk == blkstofrags(fs, blkno) &&
1665 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1666 fs->fs_bsize, ip->i_number, vp->v_type,
1668 ip->i_din1->di_blocks -=
1669 btodb(fs->fs_bsize);
1670 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1674 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1677 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1678 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1679 else if ((dblk == blkstofrags(fs, blkno) &&
1680 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1681 fs->fs_bsize, ip->i_number, vp->v_type, NULL))) {
1682 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1683 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1689 * Clear snapshot flag and drop reference.
1691 ip->i_flags &= ~SF_SNAPSHOT;
1692 DIP_SET(ip, i_flags, ip->i_flags);
1693 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1695 * The dirtied indirects must be written out before
1696 * softdep_setup_freeblocks() is called. Otherwise indir_trunc()
1697 * may find indirect pointers using the magic BLK_* values.
1699 if (DOINGSOFTDEP(vp))
1700 ffs_syncvnode(vp, MNT_WAIT);
1703 * Reenable disk quotas for ex-snapshot file.
1705 if (!getinoquota(ip))
1706 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE);
1711 * Notification that a block is being freed. Return zero if the free
1712 * should be allowed to proceed. Return non-zero if the snapshot file
1713 * wants to claim the block. The block will be claimed if it is an
1714 * uncopied part of one of the snapshots. It will be freed if it is
1715 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1716 * If a fragment is being freed, then all snapshots that care about
1717 * it must make a copy since a snapshot file can only claim full sized
1718 * blocks. Note that if more than one snapshot file maps the block,
1719 * we can pick one at random to claim it. Since none of the snapshots
1720 * can change, we are assurred that they will all see the same unmodified
1721 * image. When deleting a snapshot file (see ffs_snapremove above), we
1722 * must push any of these claimed blocks to one of the other snapshots
1723 * that maps it. These claimed blocks are easily identified as they will
1724 * have a block number equal to their logical block number within the
1725 * snapshot. A copied block can never have this property because they
1726 * must always have been allocated from a BLK_NOCOPY location.
1729 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
1731 struct vnode *devvp;
1736 struct workhead *wkhd;
1738 struct buf *ibp, *cbp, *savedcbp = 0;
1739 struct thread *td = curthread;
1741 struct vnode *vp = NULL;
1744 int indiroff = 0, error = 0, claimedblk = 0;
1745 struct snapdata *sn;
1747 lbn = fragstoblks(fs, bno);
1750 sn = devvp->v_rdev->si_snapdata;
1755 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1756 VI_MTX(devvp)) != 0)
1758 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1760 if (DOINGSOFTDEP(vp))
1761 softdep_prealloc(vp, MNT_WAIT);
1763 * Lookup block being written.
1766 blkno = DIP(ip, i_db[lbn]);
1768 td->td_pflags |= TDP_COWINPROGRESS;
1769 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1770 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1771 td->td_pflags &= ~TDP_COWINPROGRESS;
1774 indiroff = (lbn - NDADDR) % NINDIR(fs);
1775 if (ip->i_ump->um_fstype == UFS1)
1776 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1778 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1781 * Check to see if block needs to be copied.
1785 * A block that we map is being freed. If it has not
1786 * been claimed yet, we will claim or copy it (below).
1789 } else if (blkno == BLK_SNAP) {
1791 * No previous snapshot claimed the block,
1792 * so it will be freed and become a BLK_NOCOPY
1793 * (don't care) for us.
1796 panic("snapblkfree: inconsistent block type");
1798 DIP_SET(ip, i_db[lbn], BLK_NOCOPY);
1799 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1800 } else if (ip->i_ump->um_fstype == UFS1) {
1801 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1805 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1810 } else /* BLK_NOCOPY or default */ {
1812 * If the snapshot has already copied the block
1813 * (default), or does not care about the block,
1821 * If this is a full size block, we will just grab it
1822 * and assign it to the snapshot inode. Otherwise we
1823 * will proceed to copy it. See explanation for this
1824 * routine as to why only a single snapshot needs to
1827 if (size == fs->fs_bsize) {
1830 printf("%s %d lbn %jd from inum %d\n",
1831 "Grabonremove: snapino", ip->i_number,
1832 (intmax_t)lbn, inum);
1835 * If journaling is tracking this write we must add
1836 * the work to the inode or indirect being written.
1840 softdep_inode_append(ip,
1841 curthread->td_ucred, wkhd);
1843 softdep_buf_append(ibp, wkhd);
1846 DIP_SET(ip, i_db[lbn], bno);
1847 } else if (ip->i_ump->um_fstype == UFS1) {
1848 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1851 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1854 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size));
1855 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1856 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1862 * Allocate the block into which to do the copy. Note that this
1863 * allocation will never require any additional allocations for
1864 * the snapshot inode.
1866 td->td_pflags |= TDP_COWINPROGRESS;
1867 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1868 fs->fs_bsize, KERNCRED, 0, &cbp);
1869 td->td_pflags &= ~TDP_COWINPROGRESS;
1874 printf("%s%d lbn %jd %s %d size %ld to blkno %jd\n",
1875 "Copyonremove: snapino ", ip->i_number,
1876 (intmax_t)lbn, "for inum", inum, size,
1877 (intmax_t)cbp->b_blkno);
1880 * If we have already read the old block contents, then
1881 * simply copy them to the new block. Note that we need
1882 * to synchronously write snapshots that have not been
1883 * unlinked, and hence will be visible after a crash,
1884 * to ensure their integrity. At a minimum we ensure the
1885 * integrity of the filesystem metadata, but use the
1886 * dopersistence sysctl-setable flag to decide on the
1887 * persistence needed for file content data.
1889 if (savedcbp != 0) {
1890 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1892 if ((vtype == VDIR || dopersistence) &&
1894 (void) ffs_syncvnode(vp, MNT_WAIT);
1898 * Otherwise, read the old block contents into the buffer.
1900 if ((error = readblock(vp, cbp, lbn)) != 0) {
1901 bzero(cbp->b_data, fs->fs_bsize);
1903 if ((vtype == VDIR || dopersistence) &&
1905 (void) ffs_syncvnode(vp, MNT_WAIT);
1911 * Note that we need to synchronously write snapshots that
1912 * have not been unlinked, and hence will be visible after
1913 * a crash, to ensure their integrity. At a minimum we
1914 * ensure the integrity of the filesystem metadata, but
1915 * use the dopersistence sysctl-setable flag to decide on
1916 * the persistence needed for file content data.
1919 vp = savedcbp->b_vp;
1921 if ((vtype == VDIR || dopersistence) &&
1922 VTOI(vp)->i_effnlink > 0)
1923 (void) ffs_syncvnode(vp, MNT_WAIT);
1926 * If we have been unable to allocate a block in which to do
1927 * the copy, then return non-zero so that the fragment will
1928 * not be freed. Although space will be lost, the snapshot
1929 * will stay consistent.
1931 if (error != 0 && wkhd != NULL)
1932 softdep_freework(wkhd);
1933 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1938 * Associate snapshot files when mounting.
1941 ffs_snapshot_mount(mp)
1944 struct ufsmount *ump = VFSTOUFS(mp);
1945 struct vnode *devvp = ump->um_devvp;
1946 struct fs *fs = ump->um_fs;
1947 struct thread *td = curthread;
1948 struct snapdata *sn;
1950 struct vnode *lastvp;
1956 daddr_t snaplistsize;
1957 int error, snaploc, loc;
1960 * XXX The following needs to be set before ffs_truncate or
1961 * VOP_READ can be called.
1963 mp->mnt_stat.f_iosize = fs->fs_bsize;
1965 * Process each snapshot listed in the superblock.
1970 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1971 if (fs->fs_snapinum[snaploc] == 0)
1973 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc],
1974 LK_EXCLUSIVE, &vp)) != 0){
1975 printf("ffs_snapshot_mount: vget failed %d\n", error);
1979 if ((ip->i_flags & SF_SNAPSHOT) == 0 || ip->i_size ==
1980 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1981 if ((ip->i_flags & SF_SNAPSHOT) == 0) {
1982 reason = "non-snapshot";
1984 reason = "old format snapshot";
1985 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED, td);
1986 (void)ffs_syncvnode(vp, MNT_WAIT);
1988 printf("ffs_snapshot_mount: %s inode %d\n",
1989 reason, fs->fs_snapinum[snaploc]);
1992 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
1993 if (fs->fs_snapinum[loc] == 0)
1995 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
1997 fs->fs_snapinum[loc - 1] = 0;
2002 * Acquire a lock on the snapdata structure, creating it if
2005 sn = ffs_snapdata_acquire(devvp);
2007 * Change vnode to use shared snapshot lock instead of the
2008 * original private lock.
2010 vp->v_vnlock = &sn->sn_lock;
2011 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2013 * Link it onto the active snapshot list.
2016 if (ip->i_nextsnap.tqe_prev != 0)
2017 panic("ffs_snapshot_mount: %d already on list",
2020 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
2021 vp->v_vflag |= VV_SYSTEM;
2028 * No usable snapshots found.
2030 if (sn == NULL || vp == NULL)
2033 * Allocate the space for the block hints list. We always want to
2034 * use the list from the newest snapshot.
2036 auio.uio_iov = &aiov;
2037 auio.uio_iovcnt = 1;
2038 aiov.iov_base = (void *)&snaplistsize;
2039 aiov.iov_len = sizeof(snaplistsize);
2040 auio.uio_resid = aiov.iov_len;
2042 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
2043 auio.uio_segflg = UIO_SYSSPACE;
2044 auio.uio_rw = UIO_READ;
2046 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2047 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2048 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
2052 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
2053 M_UFSMNT, M_WAITOK);
2054 auio.uio_iovcnt = 1;
2055 aiov.iov_base = snapblklist;
2056 aiov.iov_len = snaplistsize * sizeof (daddr_t);
2057 auio.uio_resid = aiov.iov_len;
2058 auio.uio_offset -= sizeof(snaplistsize);
2059 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2060 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
2062 free(snapblklist, M_UFSMNT);
2067 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
2068 sn->sn_listsize = snaplistsize;
2069 sn->sn_blklist = (daddr_t *)snapblklist;
2070 devvp->v_vflag |= VV_COPYONWRITE;
2075 * Disassociate snapshot files when unmounting.
2078 ffs_snapshot_unmount(mp)
2081 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
2082 struct snapdata *sn;
2087 sn = devvp->v_rdev->si_snapdata;
2088 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) {
2090 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap);
2091 xp->i_nextsnap.tqe_prev = 0;
2092 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE,
2094 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
2095 KASSERT(vp->v_vnlock == &sn->sn_lock,
2096 ("ffs_snapshot_unmount: lost lock mutation"));
2097 vp->v_vnlock = &vp->v_lock;
2098 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2099 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2100 if (xp->i_effnlink > 0)
2103 sn = devvp->v_rdev->si_snapdata;
2105 try_free_snapdata(devvp);
2106 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
2110 * Check the buffer block to be belong to device buffer that shall be
2111 * locked after snaplk. devvp shall be locked on entry, and will be
2112 * leaved locked upon exit.
2115 ffs_bp_snapblk(devvp, bp)
2116 struct vnode *devvp;
2119 struct snapdata *sn;
2121 ufs2_daddr_t lbn, *snapblklist;
2122 int lower, upper, mid;
2124 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk");
2125 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp));
2126 sn = devvp->v_rdev->si_snapdata;
2127 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL)
2129 fs = TAILQ_FIRST(&sn->sn_head)->i_fs;
2130 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2131 snapblklist = sn->sn_blklist;
2132 upper = sn->sn_listsize - 1;
2134 while (lower <= upper) {
2135 mid = (lower + upper) / 2;
2136 if (snapblklist[mid] == lbn)
2138 if (snapblklist[mid] < lbn)
2154 struct vnode *vp, *devvp;
2158 if (bo->bo_dirty.bv_cnt <= dirtybufthresh)
2163 devvp = bo->__bo_vnode;
2164 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp));
2167 bp_bdskip = ffs_bp_snapblk(devvp, bp);
2171 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) {
2172 (void) VOP_FSYNC(vp, MNT_NOWAIT, td);
2177 * Try to find a buffer to flush.
2179 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) {
2180 if ((nbp->b_vflags & BV_BKGRDINPROG) ||
2182 LK_EXCLUSIVE | LK_NOWAIT, NULL))
2185 panic("bdwrite: found ourselves");
2188 * Don't countdeps with the bo lock
2191 if (buf_countdeps(nbp, 0)) {
2198 if (!ffs_bp_snapblk(vp, nbp)) {
2199 if (BO_MTX(bo) != VI_MTX(vp)) {
2208 if (nbp->b_flags & B_CLUSTEROK) {
2209 vfs_bio_awrite(nbp);
2214 dirtybufferflushes++;
2223 * Check for need to copy block that is about to be written,
2224 * copying the block if necessary.
2227 ffs_copyonwrite(devvp, bp)
2228 struct vnode *devvp;
2231 struct snapdata *sn;
2232 struct buf *ibp, *cbp, *savedcbp = 0;
2233 struct thread *td = curthread;
2236 struct vnode *vp = 0;
2237 ufs2_daddr_t lbn, blkno, *snapblklist;
2238 int lower, upper, mid, indiroff, error = 0;
2239 int launched_async_io, prev_norunningbuf;
2240 long saved_runningbufspace;
2242 if (devvp != bp->b_vp && (VTOI(bp->b_vp)->i_flags & SF_SNAPSHOT) != 0)
2243 return (0); /* Update on a snapshot file */
2244 if (td->td_pflags & TDP_COWINPROGRESS)
2245 panic("ffs_copyonwrite: recursive call");
2247 * First check to see if it is in the preallocated list.
2248 * By doing this check we avoid several potential deadlocks.
2251 sn = devvp->v_rdev->si_snapdata;
2253 TAILQ_EMPTY(&sn->sn_head)) {
2255 return (0); /* No snapshot */
2257 ip = TAILQ_FIRST(&sn->sn_head);
2259 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2260 snapblklist = sn->sn_blklist;
2261 upper = sn->sn_listsize - 1;
2263 while (lower <= upper) {
2264 mid = (lower + upper) / 2;
2265 if (snapblklist[mid] == lbn)
2267 if (snapblklist[mid] < lbn)
2272 if (lower <= upper) {
2276 launched_async_io = 0;
2277 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF;
2279 * Since I/O on bp isn't yet in progress and it may be blocked
2280 * for a long time waiting on snaplk, back it out of
2281 * runningbufspace, possibly waking other threads waiting for space.
2283 saved_runningbufspace = bp->b_runningbufspace;
2284 if (saved_runningbufspace != 0)
2285 runningbufwakeup(bp);
2287 * Not in the precomputed list, so check the snapshots.
2289 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2290 VI_MTX(devvp)) != 0) {
2292 sn = devvp->v_rdev->si_snapdata;
2294 TAILQ_EMPTY(&sn->sn_head)) {
2296 if (saved_runningbufspace != 0) {
2297 bp->b_runningbufspace = saved_runningbufspace;
2298 atomic_add_long(&runningbufspace,
2299 bp->b_runningbufspace);
2301 return (0); /* Snapshot gone */
2304 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2306 if (DOINGSOFTDEP(vp))
2307 softdep_prealloc(vp, MNT_WAIT);
2309 * We ensure that everything of our own that needs to be
2310 * copied will be done at the time that ffs_snapshot is
2311 * called. Thus we can skip the check here which can
2312 * deadlock in doing the lookup in UFS_BALLOC.
2317 * Check to see if block needs to be copied. We do not have
2318 * to hold the snapshot lock while doing this lookup as it
2319 * will never require any additional allocations for the
2323 blkno = DIP(ip, i_db[lbn]);
2325 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2326 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2327 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
2328 td->td_pflags &= ~TDP_COWINPROGRESS;
2331 indiroff = (lbn - NDADDR) % NINDIR(fs);
2332 if (ip->i_ump->um_fstype == UFS1)
2333 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2335 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2339 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2340 panic("ffs_copyonwrite: bad copy block");
2345 * Allocate the block into which to do the copy. Since
2346 * multiple processes may all try to copy the same block,
2347 * we have to recheck our need to do a copy if we sleep
2348 * waiting for the lock.
2350 * Because all snapshots on a filesystem share a single
2351 * lock, we ensure that we will never be in competition
2352 * with another process to allocate a block.
2354 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2355 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2356 fs->fs_bsize, KERNCRED, 0, &cbp);
2357 td->td_pflags &= ~TDP_COWINPROGRESS;
2362 printf("Copyonwrite: snapino %d lbn %jd for ",
2363 ip->i_number, (intmax_t)lbn);
2364 if (bp->b_vp == devvp)
2365 printf("fs metadata");
2367 printf("inum %d", VTOI(bp->b_vp)->i_number);
2368 printf(" lblkno %jd to blkno %jd\n",
2369 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2373 * If we have already read the old block contents, then
2374 * simply copy them to the new block. Note that we need
2375 * to synchronously write snapshots that have not been
2376 * unlinked, and hence will be visible after a crash,
2377 * to ensure their integrity. At a minimum we ensure the
2378 * integrity of the filesystem metadata, but use the
2379 * dopersistence sysctl-setable flag to decide on the
2380 * persistence needed for file content data.
2382 if (savedcbp != 0) {
2383 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2385 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2386 dopersistence) && ip->i_effnlink > 0)
2387 (void) ffs_syncvnode(vp, MNT_WAIT);
2389 launched_async_io = 1;
2393 * Otherwise, read the old block contents into the buffer.
2395 if ((error = readblock(vp, cbp, lbn)) != 0) {
2396 bzero(cbp->b_data, fs->fs_bsize);
2398 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2399 dopersistence) && ip->i_effnlink > 0)
2400 (void) ffs_syncvnode(vp, MNT_WAIT);
2402 launched_async_io = 1;
2408 * Note that we need to synchronously write snapshots that
2409 * have not been unlinked, and hence will be visible after
2410 * a crash, to ensure their integrity. At a minimum we
2411 * ensure the integrity of the filesystem metadata, but
2412 * use the dopersistence sysctl-setable flag to decide on
2413 * the persistence needed for file content data.
2416 vp = savedcbp->b_vp;
2418 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2419 dopersistence) && VTOI(vp)->i_effnlink > 0)
2420 (void) ffs_syncvnode(vp, MNT_WAIT);
2422 launched_async_io = 1;
2424 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
2425 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) |
2427 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0)
2428 waitrunningbufspace();
2430 * I/O on bp will now be started, so count it in runningbufspace.
2432 if (saved_runningbufspace != 0) {
2433 bp->b_runningbufspace = saved_runningbufspace;
2434 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
2440 * sync snapshots to force freework records waiting on snapshots to claim
2444 ffs_sync_snap(mp, waitfor)
2448 struct snapdata *sn;
2449 struct vnode *devvp;
2453 devvp = VFSTOUFS(mp)->um_devvp;
2454 if ((devvp->v_vflag & VV_COPYONWRITE) == 0)
2458 sn = devvp->v_rdev->si_snapdata;
2463 if (lockmgr(&sn->sn_lock,
2464 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2465 VI_MTX(devvp)) == 0)
2468 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2470 ffs_syncvnode(vp, waitfor);
2472 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2476 * Read the specified block into the given buffer.
2477 * Much of this boiler-plate comes from bwrite().
2480 readblock(vp, bp, lbn)
2485 struct inode *ip = VTOI(vp);
2488 bip = g_alloc_bio();
2489 bip->bio_cmd = BIO_READ;
2490 bip->bio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2491 bip->bio_data = bp->b_data;
2492 bip->bio_length = bp->b_bcount;
2493 bip->bio_done = NULL;
2495 g_io_request(bip, ip->i_devvp->v_bufobj.bo_private);
2496 bp->b_error = biowait(bip, "snaprdb");
2498 return (bp->b_error);
2504 * Process file deletes that were deferred by ufs_inactive() due to
2505 * the file system being suspended. Transfer IN_LAZYACCESS into
2506 * IN_MODIFIED for vnodes that were accessed during suspension.
2509 process_deferred_inactive(struct mount *mp)
2511 struct vnode *vp, *mvp;
2517 (void) vn_start_secondary_write(NULL, &mp, V_WAIT);
2520 MNT_VNODE_FOREACH(vp, mp, mvp) {
2523 * IN_LAZYACCESS is checked here without holding any
2524 * vnode lock, but this flag is set only while holding
2527 if (vp->v_type == VNON || (vp->v_iflag & VI_DOOMED) != 0 ||
2528 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 &&
2529 ((vp->v_iflag & VI_OWEINACT) == 0 ||
2530 vp->v_usecount > 0))) {
2536 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2540 if (error == ENOENT)
2541 continue; /* vnode recycled */
2542 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2546 if ((ip->i_flag & IN_LAZYACCESS) != 0) {
2547 ip->i_flag &= ~IN_LAZYACCESS;
2548 ip->i_flag |= IN_MODIFIED;
2551 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) {
2559 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2560 ("process_deferred_inactive: "
2561 "recursed on VI_DOINGINACT"));
2562 vp->v_iflag |= VI_DOINGINACT;
2563 vp->v_iflag &= ~VI_OWEINACT;
2565 (void) VOP_INACTIVE(vp, td);
2567 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2568 ("process_deferred_inactive: lost VI_DOINGINACT"));
2569 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2570 ("process_deferred_inactive: got VI_OWEINACT"));
2571 vp->v_iflag &= ~VI_DOINGINACT;
2578 vn_finished_secondary_write(mp);
2581 #ifndef NO_FFS_SNAPSHOT
2583 static struct snapdata *
2584 ffs_snapdata_alloc(void)
2586 struct snapdata *sn;
2589 * Fetch a snapdata from the free list if there is one available.
2591 mtx_lock(&snapfree_lock);
2592 sn = LIST_FIRST(&snapfree);
2594 LIST_REMOVE(sn, sn_link);
2595 mtx_unlock(&snapfree_lock);
2599 * If there were no free snapdatas allocate one.
2601 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
2602 TAILQ_INIT(&sn->sn_head);
2603 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
2604 LK_CANRECURSE | LK_NOSHARE);
2609 * The snapdata is never freed because we can not be certain that
2610 * there are no threads sleeping on the snap lock. Persisting
2611 * them permanently avoids costly synchronization in ffs_lock().
2614 ffs_snapdata_free(struct snapdata *sn)
2616 mtx_lock(&snapfree_lock);
2617 LIST_INSERT_HEAD(&snapfree, sn, sn_link);
2618 mtx_unlock(&snapfree_lock);
2621 /* Try to free snapdata associated with devvp */
2623 try_free_snapdata(struct vnode *devvp)
2625 struct snapdata *sn;
2626 ufs2_daddr_t *snapblklist;
2628 ASSERT_VI_LOCKED(devvp, "try_free_snapdata");
2629 sn = devvp->v_rdev->si_snapdata;
2631 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL ||
2632 (devvp->v_vflag & VV_COPYONWRITE) == 0) {
2637 devvp->v_rdev->si_snapdata = NULL;
2638 devvp->v_vflag &= ~VV_COPYONWRITE;
2639 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp));
2640 snapblklist = sn->sn_blklist;
2641 sn->sn_blklist = NULL;
2642 sn->sn_listsize = 0;
2643 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2644 if (snapblklist != NULL)
2645 free(snapblklist, M_UFSMNT);
2646 ffs_snapdata_free(sn);
2649 static struct snapdata *
2650 ffs_snapdata_acquire(struct vnode *devvp)
2652 struct snapdata *nsn;
2653 struct snapdata *sn;
2656 * Allocate a free snapdata. This is done before acquiring the
2657 * devvp lock to avoid allocation while the devvp interlock is
2660 nsn = ffs_snapdata_alloc();
2662 * If there snapshots already exist on this filesystem grab a
2663 * reference to the shared lock. Otherwise this is the first
2664 * snapshot on this filesystem and we need to use our
2665 * pre-allocated snapdata.
2668 if (devvp->v_rdev->si_snapdata == NULL) {
2669 devvp->v_rdev->si_snapdata = nsn;
2672 sn = devvp->v_rdev->si_snapdata;
2674 * Acquire the snapshot lock.
2676 lockmgr(&sn->sn_lock,
2677 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, VI_MTX(devvp));
2679 * Free any unused snapdata.
2682 ffs_snapdata_free(nsn);