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/rwlock.h>
57 #include <sys/vnode.h>
59 #include <geom/geom.h>
61 #include <ufs/ufs/extattr.h>
62 #include <ufs/ufs/quota.h>
63 #include <ufs/ufs/ufsmount.h>
64 #include <ufs/ufs/inode.h>
65 #include <ufs/ufs/ufs_extern.h>
67 #include <ufs/ffs/fs.h>
68 #include <ufs/ffs/ffs_extern.h>
70 #define KERNCRED thread0.td_ucred
75 #ifdef NO_FFS_SNAPSHOT
77 ffs_snapshot(mp, snapfile)
85 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
92 struct workhead *wkhd;
104 ffs_snapshot_mount(mp)
110 ffs_snapshot_unmount(mp)
122 ffs_copyonwrite(devvp, bp)
130 ffs_sync_snap(mp, waitfor)
137 FEATURE(ffs_snapshot, "FFS snapshot support");
139 LIST_HEAD(, snapdata) snapfree;
140 static struct mtx snapfree_lock;
141 MTX_SYSINIT(ffs_snapfree, &snapfree_lock, "snapdata free list", MTX_DEF);
143 static int cgaccount(int, struct vnode *, struct buf *, int);
144 static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
145 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
146 ufs_lbn_t, int), int, int);
147 static int indiracct_ufs1(struct vnode *, struct vnode *, int,
148 ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
149 int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
150 ufs_lbn_t, int), int);
151 static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
152 struct fs *, ufs_lbn_t, int);
153 static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
154 struct fs *, ufs_lbn_t, int);
155 static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
156 struct fs *, ufs_lbn_t, int);
157 static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
158 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
159 ufs_lbn_t, int), int, int);
160 static int indiracct_ufs2(struct vnode *, struct vnode *, int,
161 ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
162 int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
163 ufs_lbn_t, int), int);
164 static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
165 struct fs *, ufs_lbn_t, int);
166 static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
167 struct fs *, ufs_lbn_t, int);
168 static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
169 struct fs *, ufs_lbn_t, int);
170 static int readblock(struct vnode *vp, struct buf *, ufs2_daddr_t);
171 static void try_free_snapdata(struct vnode *devvp);
172 static struct snapdata *ffs_snapdata_acquire(struct vnode *devvp);
173 static int ffs_bp_snapblk(struct vnode *, struct buf *);
176 * To ensure the consistency of snapshots across crashes, we must
177 * synchronously write out copied blocks before allowing the
178 * originals to be modified. Because of the rather severe speed
179 * penalty that this imposes, the code normally only ensures
180 * persistence for the filesystem metadata contained within a
181 * snapshot. Setting the following flag allows this crash
182 * persistence to be enabled for file contents.
184 int dopersistence = 0;
187 #include <sys/sysctl.h>
188 SYSCTL_INT(_debug, OID_AUTO, dopersistence, CTLFLAG_RW, &dopersistence, 0, "");
189 static int snapdebug = 0;
190 SYSCTL_INT(_debug, OID_AUTO, snapdebug, CTLFLAG_RW, &snapdebug, 0, "");
191 int collectsnapstats = 0;
192 SYSCTL_INT(_debug, OID_AUTO, collectsnapstats, CTLFLAG_RW, &collectsnapstats,
197 * Create a snapshot file and initialize it for the filesystem.
200 ffs_snapshot(mp, snapfile)
204 ufs2_daddr_t numblks, blkno, *blkp, *snapblklist;
205 int error, cg, snaploc;
206 int i, size, len, loc;
207 ufs2_daddr_t blockno;
209 struct timespec starttime = {0, 0}, endtime;
211 long redo = 0, snaplistsize = 0;
214 struct fs *copy_fs = NULL, *fs;
215 struct thread *td = curthread;
216 struct inode *ip, *xp;
217 struct buf *bp, *nbp, *ibp;
221 struct vnode *vp, *xvp, *mvp, *devvp;
225 struct ufsmount *ump;
231 * At the moment, journaled soft updates cannot support
234 if (MOUNTEDSUJ(mp)) {
235 vfs_mount_error(mp, "%s: Snapshots are not yet supported when "
236 "running with journaled soft updates", fs->fs_fsmnt);
243 * Need to serialize access to snapshot code per filesystem.
246 * Assign a snapshot slot in the superblock.
249 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
250 if (fs->fs_snapinum[snaploc] == 0)
253 if (snaploc == FSMAXSNAP)
256 * Create the snapshot file.
259 NDINIT(&nd, CREATE, LOCKPARENT | LOCKLEAF | NOCACHE, UIO_SYSSPACE,
261 if ((error = namei(&nd)) != 0)
263 if (nd.ni_vp != NULL) {
267 if (nd.ni_dvp->v_mount != mp)
270 NDFREE(&nd, NDF_ONLY_PNBUF);
271 if (nd.ni_dvp == nd.ni_vp)
279 vat.va_mode = S_IRUSR;
280 vat.va_vaflags |= VA_EXCLUSIVE;
281 if (VOP_GETWRITEMOUNT(nd.ni_dvp, &wrtmp))
284 panic("ffs_snapshot: mount mismatch");
286 if (vn_start_write(NULL, &wrtmp, V_NOWAIT) != 0) {
287 NDFREE(&nd, NDF_ONLY_PNBUF);
289 if ((error = vn_start_write(NULL, &wrtmp,
290 V_XSLEEP | PCATCH)) != 0)
294 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vat);
295 VOP_UNLOCK(nd.ni_dvp, 0);
297 NDFREE(&nd, NDF_ONLY_PNBUF);
298 vn_finished_write(wrtmp);
303 vp->v_vflag |= VV_SYSTEM;
307 * Allocate and copy the last block contents so as to be able
308 * to set size to that of the filesystem.
310 numblks = howmany(fs->fs_size, fs->fs_frag);
311 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
312 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
315 ip->i_size = lblktosize(fs, (off_t)numblks);
316 DIP_SET(ip, i_size, ip->i_size);
317 ip->i_flag |= IN_CHANGE | IN_UPDATE;
318 error = readblock(vp, bp, numblks - 1);
323 * Preallocate critical data structures so that we can copy
324 * them in without further allocation after we suspend all
325 * operations on the filesystem. We would like to just release
326 * the allocated buffers without writing them since they will
327 * be filled in below once we are ready to go, but this upsets
328 * the soft update code, so we go ahead and write the new buffers.
330 * Allocate all indirect blocks and mark all of them as not
331 * needing to be copied.
333 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
334 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
335 fs->fs_bsize, td->td_ucred, BA_METAONLY, &ibp);
341 * Allocate copies for the superblock and its summary information.
343 error = UFS_BALLOC(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
348 blkno = fragstoblks(fs, fs->fs_csaddr);
349 len = howmany(fs->fs_cssize, fs->fs_bsize);
350 for (loc = 0; loc < len; loc++) {
351 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(blkno + loc)),
352 fs->fs_bsize, KERNCRED, 0, &nbp);
358 * Allocate all cylinder group blocks.
360 for (cg = 0; cg < fs->fs_ncg; cg++) {
361 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
362 fs->fs_bsize, KERNCRED, 0, &nbp);
367 ffs_syncvnode(vp, MNT_WAIT, 0);
370 * Copy all the cylinder group maps. Although the
371 * filesystem is still active, we hope that only a few
372 * cylinder groups will change between now and when we
373 * suspend operations. Thus, we will be able to quickly
374 * touch up the few cylinder groups that changed during
375 * the suspension period.
377 len = howmany(fs->fs_ncg, NBBY);
378 space = malloc(len, M_DEVBUF, M_WAITOK|M_ZERO);
380 fs->fs_active = space;
382 for (cg = 0; cg < fs->fs_ncg; cg++) {
383 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
384 fs->fs_bsize, KERNCRED, 0, &nbp);
387 error = cgaccount(cg, vp, nbp, 1);
390 ffs_syncvnode(vp, MNT_WAIT, 0);
395 * Change inode to snapshot type file.
397 ip->i_flags |= SF_SNAPSHOT;
398 DIP_SET(ip, i_flags, ip->i_flags);
399 ip->i_flag |= IN_CHANGE | IN_UPDATE;
401 * Ensure that the snapshot is completely on disk.
402 * Since we have marked it as a snapshot it is safe to
403 * unlock it as no process will be allowed to write to it.
405 if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
409 * All allocations are done, so we can now snapshot the system.
411 * Recind nice scheduling while running with the filesystem suspended.
413 if (td->td_proc->p_nice > 0) {
418 saved_nice = p->p_nice;
423 * Suspend operation on filesystem.
426 vn_finished_write(wrtmp);
427 if ((error = vfs_write_suspend(vp->v_mount, 0)) != 0) {
428 vn_start_write(NULL, &wrtmp, V_WAIT);
429 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
432 if (mp->mnt_kern_flag & MNTK_SUSPENDED)
434 vn_start_write(NULL, &wrtmp, V_WAIT);
436 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
437 if (ip->i_effnlink == 0) {
438 error = ENOENT; /* Snapshot file unlinked */
441 if (collectsnapstats)
442 nanotime(&starttime);
444 /* The last block might have changed. Copy it again to be sure. */
445 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(numblks - 1)),
446 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
449 error = readblock(vp, bp, numblks - 1);
450 bp->b_flags |= B_VALIDSUSPWRT;
455 * First, copy all the cylinder group maps that have changed.
457 for (cg = 0; cg < fs->fs_ncg; cg++) {
458 if ((ACTIVECGNUM(fs, cg) & ACTIVECGOFF(cg)) != 0)
461 error = UFS_BALLOC(vp, lfragtosize(fs, cgtod(fs, cg)),
462 fs->fs_bsize, KERNCRED, 0, &nbp);
465 error = cgaccount(cg, vp, nbp, 2);
471 * Grab a copy of the superblock and its summary information.
472 * We delay writing it until the suspension is released below.
474 copy_fs = malloc((u_long)fs->fs_bsize, M_UFSMNT, M_WAITOK);
475 bcopy(fs, copy_fs, fs->fs_sbsize);
476 if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
477 copy_fs->fs_clean = 1;
478 size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
479 if (fs->fs_sbsize < size)
480 bzero(&((char *)copy_fs)[fs->fs_sbsize],
481 size - fs->fs_sbsize);
482 size = blkroundup(fs, fs->fs_cssize);
483 if (fs->fs_contigsumsize > 0)
484 size += fs->fs_ncg * sizeof(int32_t);
485 space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
486 copy_fs->fs_csp = space;
487 bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
488 space = (char *)space + fs->fs_cssize;
489 loc = howmany(fs->fs_cssize, fs->fs_fsize);
490 i = fs->fs_frag - loc % fs->fs_frag;
491 len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
493 if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
494 len, KERNCRED, &bp)) != 0) {
496 free(copy_fs->fs_csp, M_UFSMNT);
497 free(copy_fs, M_UFSMNT);
501 bcopy(bp->b_data, space, (u_int)len);
502 space = (char *)space + len;
503 bp->b_flags |= B_INVAL | B_NOCACHE;
506 if (fs->fs_contigsumsize > 0) {
507 copy_fs->fs_maxcluster = lp = space;
508 for (i = 0; i < fs->fs_ncg; i++)
509 *lp++ = fs->fs_contigsumsize;
512 * We must check for active files that have been unlinked
513 * (e.g., with a zero link count). We have to expunge all
514 * trace of these files from the snapshot so that they are
515 * not reclaimed prematurely by fsck or unnecessarily dumped.
516 * We turn off the MNTK_SUSPENDED flag to avoid a panic from
517 * spec_strategy about writing on a suspended filesystem.
518 * Note that we skip unlinked snapshot files as they will
519 * be handled separately below.
521 * We also calculate the needed size for the snapshot list.
523 snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
524 FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
526 mp->mnt_kern_flag &= ~MNTK_SUSPENDED;
529 MNT_VNODE_FOREACH_ALL(xvp, mp, mvp) {
530 if ((xvp->v_usecount == 0 &&
531 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) ||
532 xvp->v_type == VNON ||
533 IS_SNAPSHOT(VTOI(xvp))) {
538 * We can skip parent directory vnode because it must have
539 * this snapshot file in it.
541 if (xvp == nd.ni_dvp) {
546 if (vn_lock(xvp, LK_EXCLUSIVE | LK_INTERLOCK) != 0) {
547 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
552 if (xvp->v_usecount == 0 &&
553 (xvp->v_iflag & (VI_OWEINACT | VI_DOINGINACT)) == 0) {
561 vprint("ffs_snapshot: busy vnode", xvp);
562 if (VOP_GETATTR(xvp, &vat, td->td_ucred) == 0 &&
569 if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
575 * If there is a fragment, clear it here.
578 loc = howmany(xp->i_size, fs->fs_bsize) - 1;
580 len = fragroundup(fs, blkoff(fs, xp->i_size));
581 if (len != 0 && len < fs->fs_bsize) {
582 ffs_blkfree(ump, copy_fs, vp,
583 DIP(xp, i_db[loc]), len, xp->i_number,
585 blkno = DIP(xp, i_db[loc]);
586 DIP_SET(xp, i_db[loc], 0);
590 if (xp->i_ump->um_fstype == UFS1)
591 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
594 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
597 DIP_SET(xp, i_db[loc], blkno);
599 error = ffs_freefile(ump, copy_fs, vp, xp->i_number,
604 free(copy_fs->fs_csp, M_UFSMNT);
605 free(copy_fs, M_UFSMNT);
607 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
612 * Erase the journal file from the snapshot.
614 if (fs->fs_flags & FS_SUJ) {
615 error = softdep_journal_lookup(mp, &xvp);
617 free(copy_fs->fs_csp, M_UFSMNT);
618 free(copy_fs, M_UFSMNT);
623 if (xp->i_ump->um_fstype == UFS1)
624 error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
627 error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
632 * Acquire a lock on the snapdata structure, creating it if necessary.
634 sn = ffs_snapdata_acquire(devvp);
636 * Change vnode to use shared snapshot lock instead of the original
639 vp->v_vnlock = &sn->sn_lock;
640 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
641 xp = TAILQ_FIRST(&sn->sn_head);
643 * If this is the first snapshot on this filesystem, then we need
644 * to allocate the space for the list of preallocated snapshot blocks.
645 * This list will be refined below, but this preliminary one will
646 * keep us out of deadlock until the full one is ready.
649 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
651 blkp = &snapblklist[1];
652 *blkp++ = lblkno(fs, fs->fs_sblockloc);
653 blkno = fragstoblks(fs, fs->fs_csaddr);
654 for (cg = 0; cg < fs->fs_ncg; cg++) {
655 if (fragstoblks(fs, cgtod(fs, cg) > blkno))
657 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
659 len = howmany(fs->fs_cssize, fs->fs_bsize);
660 for (loc = 0; loc < len; loc++)
661 *blkp++ = blkno + loc;
662 for (; cg < fs->fs_ncg; cg++)
663 *blkp++ = fragstoblks(fs, cgtod(fs, cg));
664 snapblklist[0] = blkp - snapblklist;
666 if (sn->sn_blklist != NULL)
667 panic("ffs_snapshot: non-empty list");
668 sn->sn_blklist = snapblklist;
669 sn->sn_listsize = blkp - snapblklist;
673 * Record snapshot inode. Since this is the newest snapshot,
674 * it must be placed at the end of the list.
677 fs->fs_snapinum[snaploc] = ip->i_number;
678 if (ip->i_nextsnap.tqe_prev != 0)
679 panic("ffs_snapshot: %ju already on list",
680 (uintmax_t)ip->i_number);
681 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
682 devvp->v_vflag |= VV_COPYONWRITE;
684 ASSERT_VOP_LOCKED(vp, "ffs_snapshot vp");
686 KASSERT((sn != NULL && copy_fs != NULL && error == 0) ||
687 (sn == NULL && copy_fs == NULL && error != 0),
688 ("email phk@ and mckusick@"));
690 * Resume operation on filesystem.
692 vfs_write_resume(vp->v_mount, VR_START_WRITE | VR_NO_SUSPCLR);
693 if (collectsnapstats && starttime.tv_sec > 0) {
695 timespecsub(&endtime, &starttime);
696 printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
697 vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
698 endtime.tv_nsec / 1000000, redo, fs->fs_ncg);
703 * Copy allocation information from all the snapshots in
704 * this snapshot and then expunge them from its view.
706 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap) {
709 if (xp->i_ump->um_fstype == UFS1)
710 error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
713 error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
715 if (error == 0 && xp->i_effnlink == 0) {
716 error = ffs_freefile(ump,
723 fs->fs_snapinum[snaploc] = 0;
728 * Allocate space for the full list of preallocated snapshot blocks.
730 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
732 ip->i_snapblklist = &snapblklist[1];
734 * Expunge the blocks used by the snapshots from the set of
735 * blocks marked as used in the snapshot bitmaps. Also, collect
736 * the list of allocated blocks in i_snapblklist.
738 if (ip->i_ump->um_fstype == UFS1)
739 error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1,
742 error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2,
745 fs->fs_snapinum[snaploc] = 0;
746 free(snapblklist, M_UFSMNT);
749 if (snaplistsize < ip->i_snapblklist - snapblklist)
750 panic("ffs_snapshot: list too small");
751 snaplistsize = ip->i_snapblklist - snapblklist;
752 snapblklist[0] = snaplistsize;
753 ip->i_snapblklist = 0;
755 * Write out the list of allocated blocks to the end of the snapshot.
757 auio.uio_iov = &aiov;
759 aiov.iov_base = (void *)snapblklist;
760 aiov.iov_len = snaplistsize * sizeof(daddr_t);
761 auio.uio_resid = aiov.iov_len;
762 auio.uio_offset = ip->i_size;
763 auio.uio_segflg = UIO_SYSSPACE;
764 auio.uio_rw = UIO_WRITE;
766 if ((error = VOP_WRITE(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
767 fs->fs_snapinum[snaploc] = 0;
768 free(snapblklist, M_UFSMNT);
772 * Write the superblock and its summary information
775 blkno = fragstoblks(fs, fs->fs_csaddr);
776 len = howmany(fs->fs_cssize, fs->fs_bsize);
777 space = copy_fs->fs_csp;
778 for (loc = 0; loc < len; loc++) {
779 error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
782 fs->fs_snapinum[snaploc] = 0;
783 free(snapblklist, M_UFSMNT);
786 bcopy(space, nbp->b_data, fs->fs_bsize);
787 space = (char *)space + fs->fs_bsize;
790 error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
795 loc = blkoff(fs, fs->fs_sblockloc);
796 bcopy((char *)copy_fs, &nbp->b_data[loc], (u_int)fs->fs_sbsize);
800 * As this is the newest list, it is the most inclusive, so
801 * should replace the previous list.
804 space = sn->sn_blklist;
805 sn->sn_blklist = snapblklist;
806 sn->sn_listsize = snaplistsize;
809 free(space, M_UFSMNT);
811 * Preallocate all the direct blocks in the snapshot inode so
812 * that we never have to write the inode itself to commit an
813 * update to the contents of the snapshot. Note that once
814 * created, the size of the snapshot will never change, so
815 * there will never be a need to write the inode except to
816 * update the non-integrity-critical time fields and
817 * allocated-block count.
819 for (blockno = 0; blockno < NDADDR; blockno++) {
820 if (DIP(ip, i_db[blockno]) != 0)
822 error = UFS_BALLOC(vp, lblktosize(fs, blockno),
823 fs->fs_bsize, KERNCRED, BA_CLRBUF, &bp);
826 error = readblock(vp, bp, blockno);
832 free(copy_fs->fs_csp, M_UFSMNT);
833 free(copy_fs, M_UFSMNT);
836 NDFREE(&nd, NDF_ONLY_PNBUF);
837 if (saved_nice > 0) {
842 sched_nice(td->td_proc, saved_nice);
843 PROC_UNLOCK(td->td_proc);
846 if (fs->fs_active != 0) {
847 free(fs->fs_active, M_DEVBUF);
852 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
855 (void) ffs_truncate(vp, (off_t)0, 0, NOCRED);
856 (void) ffs_syncvnode(vp, MNT_WAIT, 0);
862 vn_finished_write(wrtmp);
863 process_deferred_inactive(mp);
868 * Copy a cylinder group map. All the unallocated blocks are marked
869 * BLK_NOCOPY so that the snapshot knows that it need not copy them
870 * if they are later written. If passno is one, then this is a first
871 * pass, so only setting needs to be done. If passno is 2, then this
872 * is a revision to a previous pass which must be undone as the
873 * replacement pass is done.
876 cgaccount(cg, vp, nbp, passno)
882 struct buf *bp, *ibp;
886 ufs2_daddr_t base, numblks;
887 int error, len, loc, indiroff;
891 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
892 (int)fs->fs_cgsize, KERNCRED, &bp);
897 cgp = (struct cg *)bp->b_data;
898 if (!cg_chkmagic(cgp)) {
905 * Recomputation of summary information might not have been performed
906 * at mount time. Sync up summary information for current cylinder
907 * group while data is in memory to ensure that result of background
908 * fsck is slightly more consistent.
910 fs->fs_cs(fs, cg) = cgp->cg_cs;
911 UFS_UNLOCK(ip->i_ump);
912 bcopy(bp->b_data, nbp->b_data, fs->fs_cgsize);
913 if (fs->fs_cgsize < fs->fs_bsize)
914 bzero(&nbp->b_data[fs->fs_cgsize],
915 fs->fs_bsize - fs->fs_cgsize);
916 cgp = (struct cg *)nbp->b_data;
919 nbp->b_flags |= B_VALIDSUSPWRT;
920 numblks = howmany(fs->fs_size, fs->fs_frag);
921 len = howmany(fs->fs_fpg, fs->fs_frag);
922 base = cgbase(fs, cg) / fs->fs_frag;
923 if (base + len >= numblks)
924 len = numblks - base - 1;
927 for ( ; loc < NDADDR; loc++) {
928 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
929 DIP_SET(ip, i_db[loc], BLK_NOCOPY);
930 else if (passno == 2 && DIP(ip, i_db[loc])== BLK_NOCOPY)
931 DIP_SET(ip, i_db[loc], 0);
932 else if (passno == 1 && DIP(ip, i_db[loc])== BLK_NOCOPY)
933 panic("ffs_snapshot: lost direct block");
936 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)(base + loc)),
937 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
941 indiroff = (base + loc - NDADDR) % NINDIR(fs);
942 for ( ; loc < len; loc++, indiroff++) {
943 if (indiroff >= NINDIR(fs)) {
945 ibp->b_flags |= B_VALIDSUSPWRT;
947 error = UFS_BALLOC(vp,
948 lblktosize(fs, (off_t)(base + loc)),
949 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
955 if (ip->i_ump->um_fstype == UFS1) {
956 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
957 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
959 else if (passno == 2 && ((ufs1_daddr_t *)(ibp->b_data))
960 [indiroff] == BLK_NOCOPY)
961 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = 0;
962 else if (passno == 1 && ((ufs1_daddr_t *)(ibp->b_data))
963 [indiroff] == BLK_NOCOPY)
964 panic("ffs_snapshot: lost indirect block");
967 if (ffs_isblock(fs, cg_blksfree(cgp), loc))
968 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = BLK_NOCOPY;
969 else if (passno == 2 &&
970 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
971 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = 0;
972 else if (passno == 1 &&
973 ((ufs2_daddr_t *)(ibp->b_data)) [indiroff] == BLK_NOCOPY)
974 panic("ffs_snapshot: lost indirect block");
977 ibp->b_flags |= B_VALIDSUSPWRT;
983 * Before expunging a snapshot inode, note all the
984 * blocks that it claims with BLK_SNAP so that fsck will
985 * be able to account for those blocks properly and so
986 * that this snapshot knows that it need not copy them
987 * if the other snapshot holding them is freed. This code
988 * is reproduced once each for UFS1 and UFS2.
991 expunge_ufs1(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
992 struct vnode *snapvp;
993 struct inode *cancelip;
995 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
996 struct fs *, ufs_lbn_t, int);
1000 int i, error, indiroff;
1001 ufs_lbn_t lbn, rlbn;
1002 ufs2_daddr_t len, blkno, numblks, blksperindir;
1003 struct ufs1_dinode *dip;
1004 struct thread *td = curthread;
1008 * Prepare to expunge the inode. If its inode block has not
1009 * yet been copied, then allocate and fill the copy.
1011 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1014 blkno = VTOI(snapvp)->i_din1->di_db[lbn];
1016 if (DOINGSOFTDEP(snapvp))
1017 softdep_prealloc(snapvp, MNT_WAIT);
1018 td->td_pflags |= TDP_COWINPROGRESS;
1019 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1020 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1021 td->td_pflags &= ~TDP_COWINPROGRESS;
1024 indiroff = (lbn - NDADDR) % NINDIR(fs);
1025 blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
1029 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1032 error = ffs_balloc_ufs1(snapvp, lblktosize(fs, (off_t)lbn),
1033 fs->fs_bsize, KERNCRED, 0, &bp);
1036 if ((error = readblock(snapvp, bp, lbn)) != 0)
1040 * Set a snapshot inode to be a zero length file, regular files
1041 * or unlinked snapshots to be completely unallocated.
1043 dip = (struct ufs1_dinode *)bp->b_data +
1044 ino_to_fsbo(fs, cancelip->i_number);
1045 if (clearmode || cancelip->i_effnlink == 0)
1049 dip->di_flags &= ~SF_SNAPSHOT;
1050 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
1053 * Now go through and expunge all the blocks in the file
1054 * using the function requested.
1056 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1057 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_db[0],
1058 &cancelip->i_din1->di_db[NDADDR], fs, 0, expungetype)))
1060 if ((error = (*acctfunc)(snapvp, &cancelip->i_din1->di_ib[0],
1061 &cancelip->i_din1->di_ib[NIADDR], fs, -1, expungetype)))
1065 len = numblks - NDADDR;
1067 for (i = 0; len > 0 && i < NIADDR; i++) {
1068 error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
1069 cancelip->i_din1->di_ib[i], lbn, rlbn, len,
1070 blksperindir, fs, acctfunc, expungetype);
1073 blksperindir *= NINDIR(fs);
1074 lbn -= blksperindir + 1;
1075 len -= blksperindir;
1076 rlbn += blksperindir;
1082 * Descend an indirect block chain for vnode cancelvp accounting for all
1083 * its indirect blocks in snapvp.
1086 indiracct_ufs1(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1087 blksperindir, fs, acctfunc, expungetype)
1088 struct vnode *snapvp;
1089 struct vnode *cancelvp;
1095 ufs_lbn_t blksperindir;
1097 int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
1098 struct fs *, ufs_lbn_t, int);
1102 ufs_lbn_t subblksperindir;
1103 struct indir indirs[NIADDR + 2];
1104 ufs1_daddr_t last, *bap;
1108 if (expungetype == BLK_NOCOPY)
1110 panic("indiracct_ufs1: missing indir");
1112 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1114 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1115 panic("indiracct_ufs1: botched params");
1117 * We have to expand bread here since it will deadlock looking
1118 * up the block number for any blocks that are not in the cache.
1120 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1121 bp->b_blkno = fsbtodb(fs, blkno);
1122 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1123 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1128 * Account for the block pointers in this indirect block.
1130 last = howmany(remblks, blksperindir);
1131 if (last > NINDIR(fs))
1133 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1134 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1136 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1137 level == 0 ? rlbn : -1, expungetype);
1138 if (error || level == 0)
1141 * Account for the block pointers in each of the indirect blocks
1142 * in the levels below us.
1144 subblksperindir = blksperindir / NINDIR(fs);
1145 for (lbn++, level--, i = 0; i < last; i++) {
1146 error = indiracct_ufs1(snapvp, cancelvp, level, bap[i], lbn,
1147 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1150 rlbn += blksperindir;
1151 lbn -= blksperindir;
1152 remblks -= blksperindir;
1155 free(bap, M_DEVBUF);
1160 * Do both snap accounting and map accounting.
1163 fullacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1165 ufs1_daddr_t *oldblkp, *lastblkp;
1168 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1172 if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1174 return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1178 * Identify a set of blocks allocated in a snapshot inode.
1181 snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1183 ufs1_daddr_t *oldblkp, *lastblkp;
1186 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1188 struct inode *ip = VTOI(vp);
1189 ufs1_daddr_t blkno, *blkp;
1194 for ( ; oldblkp < lastblkp; oldblkp++) {
1196 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1198 lbn = fragstoblks(fs, blkno);
1200 blkp = &ip->i_din1->di_db[lbn];
1201 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1203 error = ffs_balloc_ufs1(vp, lblktosize(fs, (off_t)lbn),
1204 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1207 blkp = &((ufs1_daddr_t *)(ibp->b_data))
1208 [(lbn - NDADDR) % NINDIR(fs)];
1211 * If we are expunging a snapshot vnode and we
1212 * find a block marked BLK_NOCOPY, then it is
1213 * one that has been allocated to this snapshot after
1214 * we took our current snapshot and can be ignored.
1216 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1221 panic("snapacct_ufs1: bad block");
1222 *blkp = expungetype;
1231 * Account for a set of blocks allocated in a snapshot inode.
1234 mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1236 ufs1_daddr_t *oldblkp, *lastblkp;
1247 inum = ip->i_number;
1252 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1254 if (blkno == 0 || blkno == BLK_NOCOPY)
1256 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1257 *ip->i_snapblklist++ = lblkno;
1258 if (blkno == BLK_SNAP)
1259 blkno = blkstofrags(fs, lblkno);
1260 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1267 * Before expunging a snapshot inode, note all the
1268 * blocks that it claims with BLK_SNAP so that fsck will
1269 * be able to account for those blocks properly and so
1270 * that this snapshot knows that it need not copy them
1271 * if the other snapshot holding them is freed. This code
1272 * is reproduced once each for UFS1 and UFS2.
1275 expunge_ufs2(snapvp, cancelip, fs, acctfunc, expungetype, clearmode)
1276 struct vnode *snapvp;
1277 struct inode *cancelip;
1279 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1280 struct fs *, ufs_lbn_t, int);
1284 int i, error, indiroff;
1285 ufs_lbn_t lbn, rlbn;
1286 ufs2_daddr_t len, blkno, numblks, blksperindir;
1287 struct ufs2_dinode *dip;
1288 struct thread *td = curthread;
1292 * Prepare to expunge the inode. If its inode block has not
1293 * yet been copied, then allocate and fill the copy.
1295 lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1298 blkno = VTOI(snapvp)->i_din2->di_db[lbn];
1300 if (DOINGSOFTDEP(snapvp))
1301 softdep_prealloc(snapvp, MNT_WAIT);
1302 td->td_pflags |= TDP_COWINPROGRESS;
1303 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1304 fs->fs_bsize, KERNCRED, BA_METAONLY, &bp);
1305 td->td_pflags &= ~TDP_COWINPROGRESS;
1308 indiroff = (lbn - NDADDR) % NINDIR(fs);
1309 blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
1313 if ((error = bread(snapvp, lbn, fs->fs_bsize, KERNCRED, &bp)))
1316 error = ffs_balloc_ufs2(snapvp, lblktosize(fs, (off_t)lbn),
1317 fs->fs_bsize, KERNCRED, 0, &bp);
1320 if ((error = readblock(snapvp, bp, lbn)) != 0)
1324 * Set a snapshot inode to be a zero length file, regular files
1325 * to be completely unallocated.
1327 dip = (struct ufs2_dinode *)bp->b_data +
1328 ino_to_fsbo(fs, cancelip->i_number);
1329 if (clearmode || cancelip->i_effnlink == 0)
1333 dip->di_flags &= ~SF_SNAPSHOT;
1334 bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1337 * Now go through and expunge all the blocks in the file
1338 * using the function requested.
1340 numblks = howmany(cancelip->i_size, fs->fs_bsize);
1341 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_db[0],
1342 &cancelip->i_din2->di_db[NDADDR], fs, 0, expungetype)))
1344 if ((error = (*acctfunc)(snapvp, &cancelip->i_din2->di_ib[0],
1345 &cancelip->i_din2->di_ib[NIADDR], fs, -1, expungetype)))
1349 len = numblks - NDADDR;
1351 for (i = 0; len > 0 && i < NIADDR; i++) {
1352 error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
1353 cancelip->i_din2->di_ib[i], lbn, rlbn, len,
1354 blksperindir, fs, acctfunc, expungetype);
1357 blksperindir *= NINDIR(fs);
1358 lbn -= blksperindir + 1;
1359 len -= blksperindir;
1360 rlbn += blksperindir;
1366 * Descend an indirect block chain for vnode cancelvp accounting for all
1367 * its indirect blocks in snapvp.
1370 indiracct_ufs2(snapvp, cancelvp, level, blkno, lbn, rlbn, remblks,
1371 blksperindir, fs, acctfunc, expungetype)
1372 struct vnode *snapvp;
1373 struct vnode *cancelvp;
1379 ufs_lbn_t blksperindir;
1381 int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
1382 struct fs *, ufs_lbn_t, int);
1386 ufs_lbn_t subblksperindir;
1387 struct indir indirs[NIADDR + 2];
1388 ufs2_daddr_t last, *bap;
1392 if (expungetype == BLK_NOCOPY)
1394 panic("indiracct_ufs2: missing indir");
1396 if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
1398 if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1399 panic("indiracct_ufs2: botched params");
1401 * We have to expand bread here since it will deadlock looking
1402 * up the block number for any blocks that are not in the cache.
1404 bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0, 0);
1405 bp->b_blkno = fsbtodb(fs, blkno);
1406 if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
1407 (error = readblock(cancelvp, bp, fragstoblks(fs, blkno)))) {
1412 * Account for the block pointers in this indirect block.
1414 last = howmany(remblks, blksperindir);
1415 if (last > NINDIR(fs))
1417 bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1418 bcopy(bp->b_data, (caddr_t)bap, fs->fs_bsize);
1420 error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
1421 level == 0 ? rlbn : -1, expungetype);
1422 if (error || level == 0)
1425 * Account for the block pointers in each of the indirect blocks
1426 * in the levels below us.
1428 subblksperindir = blksperindir / NINDIR(fs);
1429 for (lbn++, level--, i = 0; i < last; i++) {
1430 error = indiracct_ufs2(snapvp, cancelvp, level, bap[i], lbn,
1431 rlbn, remblks, subblksperindir, fs, acctfunc, expungetype);
1434 rlbn += blksperindir;
1435 lbn -= blksperindir;
1436 remblks -= blksperindir;
1439 free(bap, M_DEVBUF);
1444 * Do both snap accounting and map accounting.
1447 fullacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)
1449 ufs2_daddr_t *oldblkp, *lastblkp;
1452 int exptype; /* BLK_SNAP or BLK_NOCOPY */
1456 if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
1458 return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
1462 * Identify a set of blocks allocated in a snapshot inode.
1465 snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1467 ufs2_daddr_t *oldblkp, *lastblkp;
1470 int expungetype; /* BLK_SNAP or BLK_NOCOPY */
1472 struct inode *ip = VTOI(vp);
1473 ufs2_daddr_t blkno, *blkp;
1478 for ( ; oldblkp < lastblkp; oldblkp++) {
1480 if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
1482 lbn = fragstoblks(fs, blkno);
1484 blkp = &ip->i_din2->di_db[lbn];
1485 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1487 error = ffs_balloc_ufs2(vp, lblktosize(fs, (off_t)lbn),
1488 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1491 blkp = &((ufs2_daddr_t *)(ibp->b_data))
1492 [(lbn - NDADDR) % NINDIR(fs)];
1495 * If we are expunging a snapshot vnode and we
1496 * find a block marked BLK_NOCOPY, then it is
1497 * one that has been allocated to this snapshot after
1498 * we took our current snapshot and can be ignored.
1500 if (expungetype == BLK_SNAP && *blkp == BLK_NOCOPY) {
1505 panic("snapacct_ufs2: bad block");
1506 *blkp = expungetype;
1515 * Account for a set of blocks allocated in a snapshot inode.
1518 mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, expungetype)
1520 ufs2_daddr_t *oldblkp, *lastblkp;
1531 inum = ip->i_number;
1536 for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1538 if (blkno == 0 || blkno == BLK_NOCOPY)
1540 if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1541 *ip->i_snapblklist++ = lblkno;
1542 if (blkno == BLK_SNAP)
1543 blkno = blkstofrags(fs, lblkno);
1544 ffs_blkfree(ip->i_ump, fs, vp, blkno, fs->fs_bsize, inum,
1551 * Decrement extra reference on snapshot when last name is removed.
1552 * It will not be freed until the last open reference goes away.
1561 struct snapdata *sn;
1562 struct ufsmount *ump;
1565 * Find snapshot in incore list.
1568 sn = ip->i_devvp->v_rdev->si_snapdata;
1570 TAILQ_FOREACH(xp, &sn->sn_head, i_nextsnap)
1576 printf("ffs_snapgone: lost snapshot vnode %ju\n",
1577 (uintmax_t)ip->i_number);
1579 * Delete snapshot inode from superblock. Keep list dense.
1584 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
1585 if (fs->fs_snapinum[snaploc] == ip->i_number)
1587 if (snaploc < FSMAXSNAP) {
1588 for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
1589 if (fs->fs_snapinum[snaploc] == 0)
1591 fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
1593 fs->fs_snapinum[snaploc - 1] = 0;
1599 * Prepare a snapshot file for being removed.
1606 struct vnode *devvp;
1609 ufs2_daddr_t numblks, blkno, dblk;
1610 int error, loc, last;
1611 struct snapdata *sn;
1615 devvp = ip->i_devvp;
1617 * If active, delete from incore list (this snapshot may
1618 * already have been in the process of being deleted, so
1619 * would not have been active).
1621 * Clear copy-on-write flag if last snapshot.
1624 if (ip->i_nextsnap.tqe_prev != 0) {
1625 sn = devvp->v_rdev->si_snapdata;
1626 TAILQ_REMOVE(&sn->sn_head, ip, i_nextsnap);
1627 ip->i_nextsnap.tqe_prev = 0;
1629 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
1630 KASSERT(vp->v_vnlock == &sn->sn_lock,
1631 ("ffs_snapremove: lost lock mutation"));
1632 vp->v_vnlock = &vp->v_lock;
1634 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
1635 try_free_snapdata(devvp);
1639 * Clear all BLK_NOCOPY fields. Pass any block claims to other
1640 * snapshots that want them (see ffs_snapblkfree below).
1642 for (blkno = 1; blkno < NDADDR; blkno++) {
1643 dblk = DIP(ip, i_db[blkno]);
1646 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1647 DIP_SET(ip, i_db[blkno], 0);
1648 else if ((dblk == blkstofrags(fs, blkno) &&
1649 ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
1650 ip->i_number, vp->v_type, NULL))) {
1651 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) -
1652 btodb(fs->fs_bsize));
1653 DIP_SET(ip, i_db[blkno], 0);
1656 numblks = howmany(ip->i_size, fs->fs_bsize);
1657 for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1658 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)blkno),
1659 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1662 if (fs->fs_size - blkno > NINDIR(fs))
1665 last = fs->fs_size - blkno;
1666 for (loc = 0; loc < last; loc++) {
1667 if (ip->i_ump->um_fstype == UFS1) {
1668 dblk = ((ufs1_daddr_t *)(ibp->b_data))[loc];
1671 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1672 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1673 else if ((dblk == blkstofrags(fs, blkno) &&
1674 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1675 fs->fs_bsize, ip->i_number, vp->v_type,
1677 ip->i_din1->di_blocks -=
1678 btodb(fs->fs_bsize);
1679 ((ufs1_daddr_t *)(ibp->b_data))[loc]= 0;
1683 dblk = ((ufs2_daddr_t *)(ibp->b_data))[loc];
1686 if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
1687 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1688 else if ((dblk == blkstofrags(fs, blkno) &&
1689 ffs_snapblkfree(fs, ip->i_devvp, dblk,
1690 fs->fs_bsize, ip->i_number, vp->v_type, NULL))) {
1691 ip->i_din2->di_blocks -= btodb(fs->fs_bsize);
1692 ((ufs2_daddr_t *)(ibp->b_data))[loc] = 0;
1698 * Clear snapshot flag and drop reference.
1700 ip->i_flags &= ~SF_SNAPSHOT;
1701 DIP_SET(ip, i_flags, ip->i_flags);
1702 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1704 * The dirtied indirects must be written out before
1705 * softdep_setup_freeblocks() is called. Otherwise indir_trunc()
1706 * may find indirect pointers using the magic BLK_* values.
1708 if (DOINGSOFTDEP(vp))
1709 ffs_syncvnode(vp, MNT_WAIT, 0);
1712 * Reenable disk quotas for ex-snapshot file.
1714 if (!getinoquota(ip))
1715 (void) chkdq(ip, DIP(ip, i_blocks), KERNCRED, FORCE);
1720 * Notification that a block is being freed. Return zero if the free
1721 * should be allowed to proceed. Return non-zero if the snapshot file
1722 * wants to claim the block. The block will be claimed if it is an
1723 * uncopied part of one of the snapshots. It will be freed if it is
1724 * either a BLK_NOCOPY or has already been copied in all of the snapshots.
1725 * If a fragment is being freed, then all snapshots that care about
1726 * it must make a copy since a snapshot file can only claim full sized
1727 * blocks. Note that if more than one snapshot file maps the block,
1728 * we can pick one at random to claim it. Since none of the snapshots
1729 * can change, we are assurred that they will all see the same unmodified
1730 * image. When deleting a snapshot file (see ffs_snapremove above), we
1731 * must push any of these claimed blocks to one of the other snapshots
1732 * that maps it. These claimed blocks are easily identified as they will
1733 * have a block number equal to their logical block number within the
1734 * snapshot. A copied block can never have this property because they
1735 * must always have been allocated from a BLK_NOCOPY location.
1738 ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, wkhd)
1740 struct vnode *devvp;
1745 struct workhead *wkhd;
1747 struct buf *ibp, *cbp, *savedcbp = NULL;
1748 struct thread *td = curthread;
1750 struct vnode *vp = NULL;
1753 int indiroff = 0, error = 0, claimedblk = 0;
1754 struct snapdata *sn;
1756 lbn = fragstoblks(fs, bno);
1759 sn = devvp->v_rdev->si_snapdata;
1764 if (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
1765 VI_MTX(devvp)) != 0)
1767 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
1769 if (DOINGSOFTDEP(vp))
1770 softdep_prealloc(vp, MNT_WAIT);
1772 * Lookup block being written.
1775 blkno = DIP(ip, i_db[lbn]);
1777 td->td_pflags |= TDP_COWINPROGRESS;
1778 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1779 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
1780 td->td_pflags &= ~TDP_COWINPROGRESS;
1783 indiroff = (lbn - NDADDR) % NINDIR(fs);
1784 if (ip->i_ump->um_fstype == UFS1)
1785 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
1787 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
1790 * Check to see if block needs to be copied.
1794 * A block that we map is being freed. If it has not
1795 * been claimed yet, we will claim or copy it (below).
1798 } else if (blkno == BLK_SNAP) {
1800 * No previous snapshot claimed the block,
1801 * so it will be freed and become a BLK_NOCOPY
1802 * (don't care) for us.
1805 panic("snapblkfree: inconsistent block type");
1807 DIP_SET(ip, i_db[lbn], BLK_NOCOPY);
1808 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1809 } else if (ip->i_ump->um_fstype == UFS1) {
1810 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] =
1814 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] =
1819 } else /* BLK_NOCOPY or default */ {
1821 * If the snapshot has already copied the block
1822 * (default), or does not care about the block,
1830 * If this is a full size block, we will just grab it
1831 * and assign it to the snapshot inode. Otherwise we
1832 * will proceed to copy it. See explanation for this
1833 * routine as to why only a single snapshot needs to
1836 if (size == fs->fs_bsize) {
1839 printf("%s %ju lbn %jd from inum %ju\n",
1840 "Grabonremove: snapino",
1841 (uintmax_t)ip->i_number,
1842 (intmax_t)lbn, (uintmax_t)inum);
1845 * If journaling is tracking this write we must add
1846 * the work to the inode or indirect being written.
1850 softdep_inode_append(ip,
1851 curthread->td_ucred, wkhd);
1853 softdep_buf_append(ibp, wkhd);
1856 DIP_SET(ip, i_db[lbn], bno);
1857 } else if (ip->i_ump->um_fstype == UFS1) {
1858 ((ufs1_daddr_t *)(ibp->b_data))[indiroff] = bno;
1861 ((ufs2_daddr_t *)(ibp->b_data))[indiroff] = bno;
1864 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + btodb(size));
1865 ip->i_flag |= IN_CHANGE | IN_UPDATE;
1866 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1872 * Allocate the block into which to do the copy. Note that this
1873 * allocation will never require any additional allocations for
1874 * the snapshot inode.
1876 td->td_pflags |= TDP_COWINPROGRESS;
1877 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
1878 fs->fs_bsize, KERNCRED, 0, &cbp);
1879 td->td_pflags &= ~TDP_COWINPROGRESS;
1884 printf("%s%ju lbn %jd %s %ju size %ld to blkno %jd\n",
1885 "Copyonremove: snapino ", (uintmax_t)ip->i_number,
1886 (intmax_t)lbn, "for inum", (uintmax_t)inum, size,
1887 (intmax_t)cbp->b_blkno);
1890 * If we have already read the old block contents, then
1891 * simply copy them to the new block. Note that we need
1892 * to synchronously write snapshots that have not been
1893 * unlinked, and hence will be visible after a crash,
1894 * to ensure their integrity. At a minimum we ensure the
1895 * integrity of the filesystem metadata, but use the
1896 * dopersistence sysctl-setable flag to decide on the
1897 * persistence needed for file content data.
1899 if (savedcbp != 0) {
1900 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
1902 if ((vtype == VDIR || dopersistence) &&
1904 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
1908 * Otherwise, read the old block contents into the buffer.
1910 if ((error = readblock(vp, cbp, lbn)) != 0) {
1911 bzero(cbp->b_data, fs->fs_bsize);
1913 if ((vtype == VDIR || dopersistence) &&
1915 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
1921 * Note that we need to synchronously write snapshots that
1922 * have not been unlinked, and hence will be visible after
1923 * a crash, to ensure their integrity. At a minimum we
1924 * ensure the integrity of the filesystem metadata, but
1925 * use the dopersistence sysctl-setable flag to decide on
1926 * the persistence needed for file content data.
1929 vp = savedcbp->b_vp;
1931 if ((vtype == VDIR || dopersistence) &&
1932 VTOI(vp)->i_effnlink > 0)
1933 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
1936 * If we have been unable to allocate a block in which to do
1937 * the copy, then return non-zero so that the fragment will
1938 * not be freed. Although space will be lost, the snapshot
1939 * will stay consistent.
1941 if (error != 0 && wkhd != NULL)
1942 softdep_freework(wkhd);
1943 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
1948 * Associate snapshot files when mounting.
1951 ffs_snapshot_mount(mp)
1954 struct ufsmount *ump = VFSTOUFS(mp);
1955 struct vnode *devvp = ump->um_devvp;
1956 struct fs *fs = ump->um_fs;
1957 struct thread *td = curthread;
1958 struct snapdata *sn;
1960 struct vnode *lastvp;
1966 daddr_t snaplistsize;
1967 int error, snaploc, loc;
1970 * XXX The following needs to be set before ffs_truncate or
1971 * VOP_READ can be called.
1973 mp->mnt_stat.f_iosize = fs->fs_bsize;
1975 * Process each snapshot listed in the superblock.
1980 for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
1981 if (fs->fs_snapinum[snaploc] == 0)
1983 if ((error = ffs_vget(mp, fs->fs_snapinum[snaploc],
1984 LK_EXCLUSIVE, &vp)) != 0){
1985 printf("ffs_snapshot_mount: vget failed %d\n", error);
1989 if (!IS_SNAPSHOT(ip) || ip->i_size ==
1990 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag))) {
1991 if (!IS_SNAPSHOT(ip)) {
1992 reason = "non-snapshot";
1994 reason = "old format snapshot";
1995 (void)ffs_truncate(vp, (off_t)0, 0, NOCRED);
1996 (void)ffs_syncvnode(vp, MNT_WAIT, 0);
1998 printf("ffs_snapshot_mount: %s inode %d\n",
1999 reason, fs->fs_snapinum[snaploc]);
2002 for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
2003 if (fs->fs_snapinum[loc] == 0)
2005 fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
2007 fs->fs_snapinum[loc - 1] = 0;
2012 * Acquire a lock on the snapdata structure, creating it if
2015 sn = ffs_snapdata_acquire(devvp);
2017 * Change vnode to use shared snapshot lock instead of the
2018 * original private lock.
2020 vp->v_vnlock = &sn->sn_lock;
2021 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2023 * Link it onto the active snapshot list.
2026 if (ip->i_nextsnap.tqe_prev != 0)
2027 panic("ffs_snapshot_mount: %ju already on list",
2028 (uintmax_t)ip->i_number);
2030 TAILQ_INSERT_TAIL(&sn->sn_head, ip, i_nextsnap);
2031 vp->v_vflag |= VV_SYSTEM;
2038 * No usable snapshots found.
2040 if (sn == NULL || vp == NULL)
2043 * Allocate the space for the block hints list. We always want to
2044 * use the list from the newest snapshot.
2046 auio.uio_iov = &aiov;
2047 auio.uio_iovcnt = 1;
2048 aiov.iov_base = (void *)&snaplistsize;
2049 aiov.iov_len = sizeof(snaplistsize);
2050 auio.uio_resid = aiov.iov_len;
2052 lblktosize(fs, howmany(fs->fs_size, fs->fs_frag));
2053 auio.uio_segflg = UIO_SYSSPACE;
2054 auio.uio_rw = UIO_READ;
2056 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2057 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2058 printf("ffs_snapshot_mount: read_1 failed %d\n", error);
2062 snapblklist = malloc(snaplistsize * sizeof(daddr_t),
2063 M_UFSMNT, M_WAITOK);
2064 auio.uio_iovcnt = 1;
2065 aiov.iov_base = snapblklist;
2066 aiov.iov_len = snaplistsize * sizeof (daddr_t);
2067 auio.uio_resid = aiov.iov_len;
2068 auio.uio_offset -= sizeof(snaplistsize);
2069 if ((error = VOP_READ(vp, &auio, IO_UNIT, td->td_ucred)) != 0) {
2070 printf("ffs_snapshot_mount: read_2 failed %d\n", error);
2072 free(snapblklist, M_UFSMNT);
2077 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_mount");
2078 sn->sn_listsize = snaplistsize;
2079 sn->sn_blklist = (daddr_t *)snapblklist;
2080 devvp->v_vflag |= VV_COPYONWRITE;
2085 * Disassociate snapshot files when unmounting.
2088 ffs_snapshot_unmount(mp)
2091 struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
2092 struct snapdata *sn;
2097 sn = devvp->v_rdev->si_snapdata;
2098 while (sn != NULL && (xp = TAILQ_FIRST(&sn->sn_head)) != NULL) {
2100 TAILQ_REMOVE(&sn->sn_head, xp, i_nextsnap);
2101 xp->i_nextsnap.tqe_prev = 0;
2102 lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE,
2104 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
2105 KASSERT(vp->v_vnlock == &sn->sn_lock,
2106 ("ffs_snapshot_unmount: lost lock mutation"));
2107 vp->v_vnlock = &vp->v_lock;
2108 lockmgr(&vp->v_lock, LK_RELEASE, NULL);
2109 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2110 if (xp->i_effnlink > 0)
2113 sn = devvp->v_rdev->si_snapdata;
2115 try_free_snapdata(devvp);
2116 ASSERT_VOP_LOCKED(devvp, "ffs_snapshot_unmount");
2120 * Check the buffer block to be belong to device buffer that shall be
2121 * locked after snaplk. devvp shall be locked on entry, and will be
2122 * leaved locked upon exit.
2125 ffs_bp_snapblk(devvp, bp)
2126 struct vnode *devvp;
2129 struct snapdata *sn;
2131 ufs2_daddr_t lbn, *snapblklist;
2132 int lower, upper, mid;
2134 ASSERT_VI_LOCKED(devvp, "ffs_bp_snapblk");
2135 KASSERT(devvp->v_type == VCHR, ("Not a device %p", devvp));
2136 sn = devvp->v_rdev->si_snapdata;
2137 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) == NULL)
2139 fs = TAILQ_FIRST(&sn->sn_head)->i_fs;
2140 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2141 snapblklist = sn->sn_blklist;
2142 upper = sn->sn_listsize - 1;
2144 while (lower <= upper) {
2145 mid = (lower + upper) / 2;
2146 if (snapblklist[mid] == lbn)
2148 if (snapblklist[mid] < lbn)
2164 struct vnode *vp, *devvp;
2168 if (bo->bo_dirty.bv_cnt <= dirtybufthresh)
2173 devvp = bo->__bo_vnode;
2174 KASSERT(vp == devvp, ("devvp != vp %p %p", bo, bp));
2177 bp_bdskip = ffs_bp_snapblk(devvp, bp);
2181 if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10 && !bp_bdskip) {
2182 (void) VOP_FSYNC(vp, MNT_NOWAIT, td);
2187 * Try to find a buffer to flush.
2189 TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) {
2190 if ((nbp->b_vflags & BV_BKGRDINPROG) ||
2192 LK_EXCLUSIVE | LK_NOWAIT, NULL))
2195 panic("bdwrite: found ourselves");
2198 * Don't countdeps with the bo lock
2201 if (buf_countdeps(nbp, 0)) {
2208 if (!ffs_bp_snapblk(vp, nbp)) {
2216 if (nbp->b_flags & B_CLUSTEROK) {
2217 vfs_bio_awrite(nbp);
2222 dirtybufferflushes++;
2231 * Check for need to copy block that is about to be written,
2232 * copying the block if necessary.
2235 ffs_copyonwrite(devvp, bp)
2236 struct vnode *devvp;
2239 struct snapdata *sn;
2240 struct buf *ibp, *cbp, *savedcbp = NULL;
2241 struct thread *td = curthread;
2244 struct vnode *vp = NULL;
2245 ufs2_daddr_t lbn, blkno, *snapblklist;
2246 int lower, upper, mid, indiroff, error = 0;
2247 int launched_async_io, prev_norunningbuf;
2248 long saved_runningbufspace;
2250 if (devvp != bp->b_vp && IS_SNAPSHOT(VTOI(bp->b_vp)))
2251 return (0); /* Update on a snapshot file */
2252 if (td->td_pflags & TDP_COWINPROGRESS)
2253 panic("ffs_copyonwrite: recursive call");
2255 * First check to see if it is in the preallocated list.
2256 * By doing this check we avoid several potential deadlocks.
2259 sn = devvp->v_rdev->si_snapdata;
2261 TAILQ_EMPTY(&sn->sn_head)) {
2263 return (0); /* No snapshot */
2265 ip = TAILQ_FIRST(&sn->sn_head);
2267 lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
2268 snapblklist = sn->sn_blklist;
2269 upper = sn->sn_listsize - 1;
2271 while (lower <= upper) {
2272 mid = (lower + upper) / 2;
2273 if (snapblklist[mid] == lbn)
2275 if (snapblklist[mid] < lbn)
2280 if (lower <= upper) {
2284 launched_async_io = 0;
2285 prev_norunningbuf = td->td_pflags & TDP_NORUNNINGBUF;
2287 * Since I/O on bp isn't yet in progress and it may be blocked
2288 * for a long time waiting on snaplk, back it out of
2289 * runningbufspace, possibly waking other threads waiting for space.
2291 saved_runningbufspace = bp->b_runningbufspace;
2292 if (saved_runningbufspace != 0)
2293 runningbufwakeup(bp);
2295 * Not in the precomputed list, so check the snapshots.
2297 while (lockmgr(&sn->sn_lock, LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2298 VI_MTX(devvp)) != 0) {
2300 sn = devvp->v_rdev->si_snapdata;
2302 TAILQ_EMPTY(&sn->sn_head)) {
2304 if (saved_runningbufspace != 0) {
2305 bp->b_runningbufspace = saved_runningbufspace;
2306 atomic_add_long(&runningbufspace,
2307 bp->b_runningbufspace);
2309 return (0); /* Snapshot gone */
2312 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2314 if (DOINGSOFTDEP(vp))
2315 softdep_prealloc(vp, MNT_WAIT);
2317 * We ensure that everything of our own that needs to be
2318 * copied will be done at the time that ffs_snapshot is
2319 * called. Thus we can skip the check here which can
2320 * deadlock in doing the lookup in UFS_BALLOC.
2325 * Check to see if block needs to be copied. We do not have
2326 * to hold the snapshot lock while doing this lookup as it
2327 * will never require any additional allocations for the
2331 blkno = DIP(ip, i_db[lbn]);
2333 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2334 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2335 fs->fs_bsize, KERNCRED, BA_METAONLY, &ibp);
2336 td->td_pflags &= ~TDP_COWINPROGRESS;
2339 indiroff = (lbn - NDADDR) % NINDIR(fs);
2340 if (ip->i_ump->um_fstype == UFS1)
2341 blkno=((ufs1_daddr_t *)(ibp->b_data))[indiroff];
2343 blkno=((ufs2_daddr_t *)(ibp->b_data))[indiroff];
2347 if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
2348 panic("ffs_copyonwrite: bad copy block");
2353 * Allocate the block into which to do the copy. Since
2354 * multiple processes may all try to copy the same block,
2355 * we have to recheck our need to do a copy if we sleep
2356 * waiting for the lock.
2358 * Because all snapshots on a filesystem share a single
2359 * lock, we ensure that we will never be in competition
2360 * with another process to allocate a block.
2362 td->td_pflags |= TDP_COWINPROGRESS | TDP_NORUNNINGBUF;
2363 error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn),
2364 fs->fs_bsize, KERNCRED, 0, &cbp);
2365 td->td_pflags &= ~TDP_COWINPROGRESS;
2370 printf("Copyonwrite: snapino %ju lbn %jd for ",
2371 (uintmax_t)ip->i_number, (intmax_t)lbn);
2372 if (bp->b_vp == devvp)
2373 printf("fs metadata");
2376 (uintmax_t)VTOI(bp->b_vp)->i_number);
2377 printf(" lblkno %jd to blkno %jd\n",
2378 (intmax_t)bp->b_lblkno, (intmax_t)cbp->b_blkno);
2382 * If we have already read the old block contents, then
2383 * simply copy them to the new block. Note that we need
2384 * to synchronously write snapshots that have not been
2385 * unlinked, and hence will be visible after a crash,
2386 * to ensure their integrity. At a minimum we ensure the
2387 * integrity of the filesystem metadata, but use the
2388 * dopersistence sysctl-setable flag to decide on the
2389 * persistence needed for file content data.
2391 if (savedcbp != 0) {
2392 bcopy(savedcbp->b_data, cbp->b_data, fs->fs_bsize);
2394 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2395 dopersistence) && ip->i_effnlink > 0)
2396 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
2398 launched_async_io = 1;
2402 * Otherwise, read the old block contents into the buffer.
2404 if ((error = readblock(vp, cbp, lbn)) != 0) {
2405 bzero(cbp->b_data, fs->fs_bsize);
2407 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2408 dopersistence) && ip->i_effnlink > 0)
2409 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
2411 launched_async_io = 1;
2417 * Note that we need to synchronously write snapshots that
2418 * have not been unlinked, and hence will be visible after
2419 * a crash, to ensure their integrity. At a minimum we
2420 * ensure the integrity of the filesystem metadata, but
2421 * use the dopersistence sysctl-setable flag to decide on
2422 * the persistence needed for file content data.
2425 vp = savedcbp->b_vp;
2427 if ((devvp == bp->b_vp || bp->b_vp->v_type == VDIR ||
2428 dopersistence) && VTOI(vp)->i_effnlink > 0)
2429 (void) ffs_syncvnode(vp, MNT_WAIT, NO_INO_UPDT);
2431 launched_async_io = 1;
2433 lockmgr(vp->v_vnlock, LK_RELEASE, NULL);
2434 td->td_pflags = (td->td_pflags & ~TDP_NORUNNINGBUF) |
2436 if (launched_async_io && (td->td_pflags & TDP_NORUNNINGBUF) == 0)
2437 waitrunningbufspace();
2439 * I/O on bp will now be started, so count it in runningbufspace.
2441 if (saved_runningbufspace != 0) {
2442 bp->b_runningbufspace = saved_runningbufspace;
2443 atomic_add_long(&runningbufspace, bp->b_runningbufspace);
2449 * sync snapshots to force freework records waiting on snapshots to claim
2453 ffs_sync_snap(mp, waitfor)
2457 struct snapdata *sn;
2458 struct vnode *devvp;
2462 devvp = VFSTOUFS(mp)->um_devvp;
2463 if ((devvp->v_vflag & VV_COPYONWRITE) == 0)
2467 sn = devvp->v_rdev->si_snapdata;
2472 if (lockmgr(&sn->sn_lock,
2473 LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
2474 VI_MTX(devvp)) == 0)
2477 TAILQ_FOREACH(ip, &sn->sn_head, i_nextsnap) {
2479 ffs_syncvnode(vp, waitfor, NO_INO_UPDT);
2481 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2485 * Read the specified block into the given buffer.
2486 * Much of this boiler-plate comes from bwrite().
2489 readblock(vp, bp, lbn)
2494 struct inode *ip = VTOI(vp);
2497 bip = g_alloc_bio();
2498 bip->bio_cmd = BIO_READ;
2499 bip->bio_offset = dbtob(fsbtodb(ip->i_fs, blkstofrags(ip->i_fs, lbn)));
2500 bip->bio_data = bp->b_data;
2501 bip->bio_length = bp->b_bcount;
2502 bip->bio_done = NULL;
2504 g_io_request(bip, ip->i_devvp->v_bufobj.bo_private);
2505 bp->b_error = biowait(bip, "snaprdb");
2507 return (bp->b_error);
2513 * Process file deletes that were deferred by ufs_inactive() due to
2514 * the file system being suspended. Transfer IN_LAZYACCESS into
2515 * IN_MODIFIED for vnodes that were accessed during suspension.
2518 process_deferred_inactive(struct mount *mp)
2520 struct vnode *vp, *mvp;
2526 (void) vn_start_secondary_write(NULL, &mp, V_WAIT);
2528 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2530 * IN_LAZYACCESS is checked here without holding any
2531 * vnode lock, but this flag is set only while holding
2534 if (vp->v_type == VNON ||
2535 ((VTOI(vp)->i_flag & IN_LAZYACCESS) == 0 &&
2536 ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0))) {
2541 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2544 if (error == ENOENT)
2545 continue; /* vnode recycled */
2546 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2550 if ((ip->i_flag & IN_LAZYACCESS) != 0) {
2551 ip->i_flag &= ~IN_LAZYACCESS;
2552 ip->i_flag |= IN_MODIFIED;
2555 if ((vp->v_iflag & VI_OWEINACT) == 0 || vp->v_usecount > 0) {
2562 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2563 ("process_deferred_inactive: got VI_OWEINACT"));
2568 vn_finished_secondary_write(mp);
2571 #ifndef NO_FFS_SNAPSHOT
2573 static struct snapdata *
2574 ffs_snapdata_alloc(void)
2576 struct snapdata *sn;
2579 * Fetch a snapdata from the free list if there is one available.
2581 mtx_lock(&snapfree_lock);
2582 sn = LIST_FIRST(&snapfree);
2584 LIST_REMOVE(sn, sn_link);
2585 mtx_unlock(&snapfree_lock);
2589 * If there were no free snapdatas allocate one.
2591 sn = malloc(sizeof *sn, M_UFSMNT, M_WAITOK | M_ZERO);
2592 TAILQ_INIT(&sn->sn_head);
2593 lockinit(&sn->sn_lock, PVFS, "snaplk", VLKTIMEOUT,
2594 LK_CANRECURSE | LK_NOSHARE);
2599 * The snapdata is never freed because we can not be certain that
2600 * there are no threads sleeping on the snap lock. Persisting
2601 * them permanently avoids costly synchronization in ffs_lock().
2604 ffs_snapdata_free(struct snapdata *sn)
2606 mtx_lock(&snapfree_lock);
2607 LIST_INSERT_HEAD(&snapfree, sn, sn_link);
2608 mtx_unlock(&snapfree_lock);
2611 /* Try to free snapdata associated with devvp */
2613 try_free_snapdata(struct vnode *devvp)
2615 struct snapdata *sn;
2616 ufs2_daddr_t *snapblklist;
2618 ASSERT_VI_LOCKED(devvp, "try_free_snapdata");
2619 sn = devvp->v_rdev->si_snapdata;
2621 if (sn == NULL || TAILQ_FIRST(&sn->sn_head) != NULL ||
2622 (devvp->v_vflag & VV_COPYONWRITE) == 0) {
2627 devvp->v_rdev->si_snapdata = NULL;
2628 devvp->v_vflag &= ~VV_COPYONWRITE;
2629 lockmgr(&sn->sn_lock, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp));
2630 snapblklist = sn->sn_blklist;
2631 sn->sn_blklist = NULL;
2632 sn->sn_listsize = 0;
2633 lockmgr(&sn->sn_lock, LK_RELEASE, NULL);
2634 if (snapblklist != NULL)
2635 free(snapblklist, M_UFSMNT);
2636 ffs_snapdata_free(sn);
2639 static struct snapdata *
2640 ffs_snapdata_acquire(struct vnode *devvp)
2642 struct snapdata *nsn;
2643 struct snapdata *sn;
2646 * Allocate a free snapdata. This is done before acquiring the
2647 * devvp lock to avoid allocation while the devvp interlock is
2650 nsn = ffs_snapdata_alloc();
2652 * If there snapshots already exist on this filesystem grab a
2653 * reference to the shared lock. Otherwise this is the first
2654 * snapshot on this filesystem and we need to use our
2655 * pre-allocated snapdata.
2658 if (devvp->v_rdev->si_snapdata == NULL) {
2659 devvp->v_rdev->si_snapdata = nsn;
2662 sn = devvp->v_rdev->si_snapdata;
2664 * Acquire the snapshot lock.
2666 lockmgr(&sn->sn_lock,
2667 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, VI_MTX(devvp));
2669 * Free any unused snapdata.
2672 ffs_snapdata_free(nsn);