[BACK]Return to ffs_snapshot.c CVS log [TXT][DIR] Up to [cvs.NetBSD.org] / src / sys / ufs / ffs

Annotation of src/sys/ufs/ffs/ffs_snapshot.c, Revision 1.48

1.48    ! hannken     1: /*     $NetBSD: ffs_snapshot.c,v 1.47 2007/08/09 07:34:28 hannken Exp $        */
1.18      thorpej     2:
1.1       hannken     3: /*
                      4:  * Copyright 2000 Marshall Kirk McKusick. All Rights Reserved.
                      5:  *
                      6:  * Further information about snapshots can be obtained from:
                      7:  *
                      8:  *     Marshall Kirk McKusick          http://www.mckusick.com/softdep/
                      9:  *     1614 Oxford Street              mckusick@mckusick.com
                     10:  *     Berkeley, CA 94709-1608         +1-510-843-9542
                     11:  *     USA
                     12:  *
                     13:  * Redistribution and use in source and binary forms, with or without
                     14:  * modification, are permitted provided that the following conditions
                     15:  * are met:
                     16:  *
                     17:  * 1. Redistributions of source code must retain the above copyright
                     18:  *    notice, this list of conditions and the following disclaimer.
                     19:  * 2. Redistributions in binary form must reproduce the above copyright
                     20:  *    notice, this list of conditions and the following disclaimer in the
                     21:  *    documentation and/or other materials provided with the distribution.
                     22:  *
                     23:  * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
                     24:  * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
                     25:  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
                     26:  * DISCLAIMED.  IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
                     27:  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
                     28:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
                     29:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
                     30:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
                     31:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
                     32:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
                     33:  * SUCH DAMAGE.
                     34:  *
                     35:  *     @(#)ffs_snapshot.c      8.11 (McKusick) 7/23/00
                     36:  *
                     37:  *     from FreeBSD: ffs_snapshot.c,v 1.79 2004/02/13 02:02:06 kuriyama Exp
                     38:  */
                     39:
                     40: #include <sys/cdefs.h>
1.48    ! hannken    41: __KERNEL_RCSID(0, "$NetBSD: ffs_snapshot.c,v 1.47 2007/08/09 07:34:28 hannken Exp $");
1.8       hannken    42:
                     43: #if defined(_KERNEL_OPT)
                     44: #include "opt_ffs.h"
                     45: #endif
1.1       hannken    46:
                     47: #include <sys/param.h>
                     48: #include <sys/kernel.h>
                     49: #include <sys/systm.h>
                     50: #include <sys/conf.h>
                     51: #include <sys/buf.h>
                     52: #include <sys/proc.h>
                     53: #include <sys/namei.h>
                     54: #include <sys/sched.h>
                     55: #include <sys/stat.h>
                     56: #include <sys/malloc.h>
                     57: #include <sys/mount.h>
                     58: #include <sys/resource.h>
                     59: #include <sys/resourcevar.h>
                     60: #include <sys/vnode.h>
1.29      elad       61: #include <sys/kauth.h>
1.40      hannken    62: #include <sys/fstrans.h>
1.1       hannken    63:
                     64: #include <miscfs/specfs/specdev.h>
                     65:
                     66: #include <ufs/ufs/quota.h>
                     67: #include <ufs/ufs/ufsmount.h>
                     68: #include <ufs/ufs/inode.h>
                     69: #include <ufs/ufs/ufs_extern.h>
                     70: #include <ufs/ufs/ufs_bswap.h>
                     71:
                     72: #include <ufs/ffs/fs.h>
                     73: #include <ufs/ffs/ffs_extern.h>
                     74:
                     75: /* FreeBSD -> NetBSD conversion */
1.31      ad         76: #define KERNCRED       lwp0.l_cred
1.1       hannken    77: #define ufs1_daddr_t   int32_t
                     78: #define ufs2_daddr_t   int64_t
                     79: #define ufs_lbn_t      daddr_t
                     80: #define VI_MTX(v)      (&(v)->v_interlock)
                     81: #define VI_LOCK(v)     simple_lock(&(v)->v_interlock)
                     82: #define VI_UNLOCK(v)   simple_unlock(&(v)->v_interlock)
                     83: #define MNT_ILOCK(v)   simple_lock(&mntvnode_slock)
                     84: #define MNT_IUNLOCK(v) simple_unlock(&mntvnode_slock)
                     85:
1.10      hannken    86: #if !defined(FFS_NO_SNAPSHOT)
1.43      christos   87: static int cgaccount(int, struct vnode *, void *, int);
1.1       hannken    88: static int expunge_ufs1(struct vnode *, struct inode *, struct fs *,
                     89:     int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
                     90:     ufs_lbn_t, int), int);
                     91: static int indiracct_ufs1(struct vnode *, struct vnode *, int,
                     92:     ufs1_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
                     93:     int (*)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *, struct fs *,
                     94:     ufs_lbn_t, int), int);
                     95: static int fullacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
                     96:     struct fs *, ufs_lbn_t, int);
                     97: static int snapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
                     98:     struct fs *, ufs_lbn_t, int);
                     99: static int mapacct_ufs1(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
                    100:     struct fs *, ufs_lbn_t, int);
                    101: static int expunge_ufs2(struct vnode *, struct inode *, struct fs *,
                    102:     int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
                    103:     ufs_lbn_t, int), int);
                    104: static int indiracct_ufs2(struct vnode *, struct vnode *, int,
                    105:     ufs2_daddr_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, ufs_lbn_t, struct fs *,
                    106:     int (*)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *, struct fs *,
                    107:     ufs_lbn_t, int), int);
                    108: static int fullacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
                    109:     struct fs *, ufs_lbn_t, int);
                    110: static int snapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
                    111:     struct fs *, ufs_lbn_t, int);
                    112: static int mapacct_ufs2(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
                    113:     struct fs *, ufs_lbn_t, int);
1.43      christos  114: static int readvnblk(struct vnode *, void *, ufs2_daddr_t);
1.10      hannken   115: #endif /* !defined(FFS_NO_SNAPSHOT) */
                    116:
1.47      hannken   117: static void si_mount_dtor(void *);
                    118: static struct snap_info *si_mount_init(struct mount *);
1.1       hannken   119: static int ffs_copyonwrite(void *, struct buf *);
1.43      christos  120: static int readfsblk(struct vnode *, void *, ufs2_daddr_t);
                    121: static int writevnblk(struct vnode *, void *, ufs2_daddr_t);
1.4       hannken   122: static inline int cow_enter(void);
                    123: static inline void cow_leave(int);
1.1       hannken   124: static inline ufs2_daddr_t db_get(struct inode *, int);
                    125: static inline void db_assign(struct inode *, int, ufs2_daddr_t);
1.43      christos  126: static inline ufs2_daddr_t idb_get(struct inode *, void *, int);
                    127: static inline void idb_assign(struct inode *, void *, int, ufs2_daddr_t);
1.1       hannken   128:
1.47      hannken   129: struct snap_info {
                    130:        TAILQ_HEAD(inodelst, inode) si_snapshots; /* List of active snapshots */
                    131:        daddr_t *si_snapblklist;                /* Snapshot block hints list */
                    132: };
                    133:
1.1       hannken   134: #ifdef DEBUG
                    135: static int snapdebug = 0;
                    136: #endif
1.47      hannken   137: static kmutex_t si_mount_init_lock;
                    138: static specificdata_key_t si_mount_data_key;
                    139:
                    140: void
                    141: ffs_snapshot_init(void)
                    142: {
                    143:        int error;
                    144:
                    145:        error = mount_specific_key_create(&si_mount_data_key, si_mount_dtor);
                    146:        KASSERT(error == 0);
                    147:        mutex_init(&si_mount_init_lock, MUTEX_DEFAULT, IPL_NONE);
                    148: }
                    149:
                    150: void
                    151: ffs_snapshot_fini(void)
                    152: {
                    153:        mount_specific_key_delete(si_mount_data_key);
                    154:        mutex_destroy(&si_mount_init_lock);
                    155: }
                    156:
                    157: static void
                    158: si_mount_dtor(void *arg)
                    159: {
                    160:        struct snap_info *si = arg;
                    161:
                    162:        KASSERT(TAILQ_EMPTY(&si->si_snapshots));
                    163:        KASSERT(si->si_snapblklist == NULL);
                    164:        free(si, M_MOUNT);
                    165: }
                    166:
                    167: static struct snap_info *
                    168: si_mount_init(struct mount *mp)
                    169: {
                    170:        struct snap_info *new;
                    171:
                    172:        mutex_enter(&si_mount_init_lock);
                    173:
                    174:        if ((new = mount_getspecific(mp, si_mount_data_key)) != NULL) {
                    175:                mutex_exit(&si_mount_init_lock);
                    176:                return new;
                    177:        }
                    178:
                    179:        new = malloc(sizeof(*new), M_MOUNT, M_WAITOK);
                    180:        TAILQ_INIT(&new->si_snapshots);
                    181:        new->si_snapblklist = NULL;
                    182:        mount_setspecific(mp, si_mount_data_key, new);
                    183:        mutex_exit(&si_mount_init_lock);
                    184:        return new;
                    185: }
1.1       hannken   186:
                    187: /*
                    188:  * Create a snapshot file and initialize it for the filesystem.
1.4       hannken   189:  * Vnode is locked on entry and return.
1.1       hannken   190:  */
                    191: int
1.36      christos  192: ffs_snapshot(struct mount *mp, struct vnode *vp,
                    193:     struct timespec *ctime)
1.1       hannken   194: {
1.10      hannken   195: #if defined(FFS_NO_SNAPSHOT)
                    196:        return EOPNOTSUPP;
                    197: }
                    198: #else /* defined(FFS_NO_SNAPSHOT) */
1.1       hannken   199:        ufs2_daddr_t numblks, blkno, *blkp, snaplistsize = 0, *snapblklist;
                    200:        int error, ns, cg, snaploc;
1.15      hannken   201:        int i, s, size, len, loc;
1.1       hannken   202:        int flag = mp->mnt_flag;
                    203:        struct timeval starttime;
                    204: #ifdef DEBUG
                    205:        struct timeval endtime;
                    206: #endif
                    207:        struct timespec ts;
                    208:        long redo = 0;
                    209:        int32_t *lp;
                    210:        void *space;
1.43      christos  211:        void *sbbuf = NULL;
1.47      hannken   212:        struct fs *copy_fs = NULL, *fs = VFSTOUFS(mp)->um_fs;
1.23      christos  213:        struct lwp *l = curlwp;
1.1       hannken   214:        struct inode *ip, *xp;
1.15      hannken   215:        struct buf *bp, *ibp, *nbp;
1.1       hannken   216:        struct vattr vat;
1.35      reinoud   217:        struct vnode *xvp, *nvp, *devvp;
1.47      hannken   218:        struct snap_info *si;
1.1       hannken   219:
                    220:        ns = UFS_FSNEEDSWAP(fs);
1.47      hannken   221:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                    222:                si = si_mount_init(mp);
1.1       hannken   223:        /*
                    224:         * Need to serialize access to snapshot code per filesystem.
                    225:         */
                    226:        /*
                    227:         * If the vnode already is a snapshot, return.
                    228:         */
                    229:        if (VTOI(vp)->i_flags & SF_SNAPSHOT) {
                    230:                if (ctime) {
                    231:                        ctime->tv_sec = DIP(VTOI(vp), mtime);
                    232:                        ctime->tv_nsec = DIP(VTOI(vp), mtimensec);
                    233:                }
                    234:                return 0;
                    235:        }
                    236:        /*
1.9       hannken   237:         * Check mount, exclusive reference and owner.
1.1       hannken   238:         */
1.4       hannken   239:        if (vp->v_mount != mp)
1.1       hannken   240:                return EXDEV;
1.4       hannken   241:        if (vp->v_usecount != 1 || vp->v_writecount != 0)
1.1       hannken   242:                return EBUSY;
1.31      ad        243:        if (kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
1.39      elad      244:            NULL) != 0 &&
1.31      ad        245:            VTOI(vp)->i_uid != kauth_cred_geteuid(l->l_cred))
1.9       hannken   246:                return EACCES;
                    247:
1.1       hannken   248:        if (vp->v_size != 0) {
1.23      christos  249:                error = ffs_truncate(vp, 0, 0, NOCRED, l);
1.4       hannken   250:                if (error)
1.1       hannken   251:                        return error;
                    252:        }
                    253:        /*
                    254:         * Assign a snapshot slot in the superblock.
                    255:         */
                    256:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
                    257:                if (fs->fs_snapinum[snaploc] == 0)
                    258:                        break;
                    259:        if (snaploc == FSMAXSNAP)
                    260:                return (ENOSPC);
                    261:        ip = VTOI(vp);
                    262:        devvp = ip->i_devvp;
                    263:        /*
1.16      hannken   264:         * Write an empty list of preallocated blocks to the end of
                    265:         * the snapshot to set size to at least that of the filesystem.
1.1       hannken   266:         */
                    267:        numblks = howmany(fs->fs_size, fs->fs_frag);
1.16      hannken   268:        blkno = 1;
                    269:        blkno = ufs_rw64(blkno, ns);
1.1       hannken   270:        error = vn_rdwr(UIO_WRITE, vp,
1.43      christos  271:            (void *)&blkno, sizeof(blkno), lblktosize(fs, (off_t)numblks),
1.31      ad        272:            UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, l->l_cred, NULL, NULL);
1.1       hannken   273:        if (error)
                    274:                goto out;
                    275:        /*
                    276:         * Preallocate critical data structures so that we can copy
                    277:         * them in without further allocation after we suspend all
                    278:         * operations on the filesystem. We would like to just release
                    279:         * the allocated buffers without writing them since they will
                    280:         * be filled in below once we are ready to go, but this upsets
                    281:         * the soft update code, so we go ahead and write the new buffers.
                    282:         *
                    283:         * Allocate all indirect blocks and mark all of them as not
                    284:         * needing to be copied.
                    285:         */
                    286:        for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1.22      yamt      287:                error = ffs_balloc(vp, lblktosize(fs, (off_t)blkno),
1.31      ad        288:                    fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
1.1       hannken   289:                if (error)
                    290:                        goto out;
1.15      hannken   291:                bawrite(ibp);
1.1       hannken   292:        }
                    293:        /*
                    294:         * Allocate copies for the superblock and its summary information.
                    295:         */
1.22      yamt      296:        error = ffs_balloc(vp, fs->fs_sblockloc, fs->fs_sbsize, KERNCRED,
1.15      hannken   297:            0, &nbp);
                    298:        if (error)
                    299:                goto out;
                    300:        bawrite(nbp);
1.1       hannken   301:        blkno = fragstoblks(fs, fs->fs_csaddr);
1.15      hannken   302:        len = howmany(fs->fs_cssize, fs->fs_bsize);
                    303:        for (loc = 0; loc < len; loc++) {
1.22      yamt      304:                error = ffs_balloc(vp, lblktosize(fs, (off_t)(blkno + loc)),
1.15      hannken   305:                    fs->fs_bsize, KERNCRED, 0, &nbp);
                    306:                if (error)
1.1       hannken   307:                        goto out;
1.15      hannken   308:                bawrite(nbp);
                    309:        }
1.1       hannken   310:        /*
                    311:         * Copy all the cylinder group maps. Although the
                    312:         * filesystem is still active, we hope that only a few
                    313:         * cylinder groups will change between now and when we
                    314:         * suspend operations. Thus, we will be able to quickly
                    315:         * touch up the few cylinder groups that changed during
                    316:         * the suspension period.
                    317:         */
                    318:        len = howmany(fs->fs_ncg, NBBY);
1.25      christos  319:        fs->fs_active = malloc(len, M_DEVBUF, M_WAITOK | M_ZERO);
1.1       hannken   320:        for (cg = 0; cg < fs->fs_ncg; cg++) {
1.22      yamt      321:                if ((error = ffs_balloc(vp, lfragtosize(fs, cgtod(fs, cg)),
1.15      hannken   322:                    fs->fs_bsize, KERNCRED, 0, &nbp)) != 0)
1.1       hannken   323:                        goto out;
1.15      hannken   324:                error = cgaccount(cg, vp, nbp->b_data, 1);
                    325:                bawrite(nbp);
                    326:                if (error)
1.1       hannken   327:                        goto out;
                    328:        }
                    329:        /*
                    330:         * Change inode to snapshot type file.
                    331:         */
                    332:        ip->i_flags |= SF_SNAPSHOT;
                    333:        DIP_ASSIGN(ip, flags, ip->i_flags);
                    334:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                    335:        /*
                    336:         * Ensure that the snapshot is completely on disk.
                    337:         * Since we have marked it as a snapshot it is safe to
                    338:         * unlock it as no process will be allowed to write to it.
                    339:         */
1.23      christos  340:        if ((error = VOP_FSYNC(vp, KERNCRED, FSYNC_WAIT, 0, 0, l)) != 0)
1.1       hannken   341:                goto out;
                    342:        VOP_UNLOCK(vp, 0);
                    343:        /*
                    344:         * All allocations are done, so we can now snapshot the system.
                    345:         *
                    346:         * Suspend operation on filesystem.
                    347:         */
1.40      hannken   348:        if ((error = vfs_suspend(vp->v_mount, 0)) != 0) {
1.1       hannken   349:                vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
                    350:                goto out;
                    351:        }
                    352:        vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1.30      kardel    353:        getmicrotime(&starttime);
1.1       hannken   354:        /*
                    355:         * First, copy all the cylinder group maps that have changed.
                    356:         */
                    357:        for (cg = 0; cg < fs->fs_ncg; cg++) {
                    358:                if (ACTIVECG_ISSET(fs, cg))
                    359:                        continue;
                    360:                redo++;
1.22      yamt      361:                if ((error = ffs_balloc(vp, lfragtosize(fs, cgtod(fs, cg)),
1.15      hannken   362:                    fs->fs_bsize, KERNCRED, 0, &nbp)) != 0)
1.1       hannken   363:                        goto out1;
1.15      hannken   364:                error = cgaccount(cg, vp, nbp->b_data, 2);
                    365:                bawrite(nbp);
                    366:                if (error)
1.1       hannken   367:                        goto out1;
                    368:        }
                    369:        /*
                    370:         * Grab a copy of the superblock and its summary information.
                    371:         * We delay writing it until the suspension is released below.
                    372:         */
1.16      hannken   373:        sbbuf = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.1       hannken   374:        loc = blkoff(fs, fs->fs_sblockloc);
                    375:        if (loc > 0)
1.43      christos  376:                memset(sbbuf, 0, loc);
                    377:        copy_fs = (struct fs *)((char *)sbbuf + loc);
1.1       hannken   378:        bcopy(fs, copy_fs, fs->fs_sbsize);
                    379:        size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
                    380:        if (fs->fs_sbsize < size)
1.43      christos  381:                memset((char *)sbbuf + loc + fs->fs_sbsize, 0,
                    382:                    size - fs->fs_sbsize);
1.1       hannken   383:        size = blkroundup(fs, fs->fs_cssize);
                    384:        if (fs->fs_contigsumsize > 0)
                    385:                size += fs->fs_ncg * sizeof(int32_t);
                    386:        space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
                    387:        copy_fs->fs_csp = space;
                    388:        bcopy(fs->fs_csp, copy_fs->fs_csp, fs->fs_cssize);
1.12      yamt      389:        space = (char *)space + fs->fs_cssize;
1.1       hannken   390:        loc = howmany(fs->fs_cssize, fs->fs_fsize);
                    391:        i = fs->fs_frag - loc % fs->fs_frag;
                    392:        len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
                    393:        if (len > 0) {
                    394:                if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
                    395:                    len, KERNCRED, &bp)) != 0) {
                    396:                        brelse(bp);
                    397:                        free(copy_fs->fs_csp, M_UFSMNT);
                    398:                        goto out1;
                    399:                }
                    400:                bcopy(bp->b_data, space, (u_int)len);
1.12      yamt      401:                space = (char *)space + len;
1.1       hannken   402:                bp->b_flags |= B_INVAL | B_NOCACHE;
                    403:                brelse(bp);
                    404:        }
                    405:        if (fs->fs_contigsumsize > 0) {
                    406:                copy_fs->fs_maxcluster = lp = space;
                    407:                for (i = 0; i < fs->fs_ncg; i++)
                    408:                        *lp++ = fs->fs_contigsumsize;
                    409:        }
                    410:        /*
                    411:         * We must check for active files that have been unlinked
                    412:         * (e.g., with a zero link count). We have to expunge all
                    413:         * trace of these files from the snapshot so that they are
                    414:         * not reclaimed prematurely by fsck or unnecessarily dumped.
                    415:         * We turn off the MNTK_SUSPENDED flag to avoid a panic from
                    416:         * spec_strategy about writing on a suspended filesystem.
                    417:         * Note that we skip unlinked snapshot files as they will
                    418:         * be handled separately below.
                    419:         *
                    420:         * We also calculate the needed size for the snapshot list.
                    421:         */
                    422:        snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
                    423:            FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
                    424:        MNT_ILOCK(mp);
                    425: loop:
1.35      reinoud   426:        /*
                    427:         * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
                    428:         * and vclean() can be called indirectly
                    429:         */
                    430:        for (xvp = TAILQ_FIRST(&mp->mnt_vnodelist); xvp; xvp = nvp) {
1.1       hannken   431:                /*
                    432:                 * Make sure this vnode wasn't reclaimed in getnewvnode().
                    433:                 * Start over if it has (it won't be on the list anymore).
                    434:                 */
                    435:                if (xvp->v_mount != mp)
                    436:                        goto loop;
                    437:                VI_LOCK(xvp);
1.35      reinoud   438:                nvp = TAILQ_NEXT(xvp, v_mntvnodes);
1.1       hannken   439:                MNT_IUNLOCK(mp);
                    440:                if ((xvp->v_flag & VXLOCK) ||
                    441:                    xvp->v_usecount == 0 || xvp->v_type == VNON ||
                    442:                    (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
                    443:                        VI_UNLOCK(xvp);
                    444:                        MNT_ILOCK(mp);
                    445:                        continue;
                    446:                }
1.40      hannken   447:                VI_UNLOCK(xvp);
1.1       hannken   448: #ifdef DEBUG
                    449:                if (snapdebug)
                    450:                        vprint("ffs_snapshot: busy vnode", xvp);
                    451: #endif
1.31      ad        452:                if (VOP_GETATTR(xvp, &vat, l->l_cred, l) == 0 &&
1.1       hannken   453:                    vat.va_nlink > 0) {
                    454:                        MNT_ILOCK(mp);
                    455:                        continue;
                    456:                }
                    457:                xp = VTOI(xvp);
                    458:                if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
                    459:                        MNT_ILOCK(mp);
                    460:                        continue;
                    461:                }
                    462:                /*
                    463:                 * If there is a fragment, clear it here.
                    464:                 */
                    465:                blkno = 0;
                    466:                loc = howmany(xp->i_size, fs->fs_bsize) - 1;
                    467:                if (loc < NDADDR) {
                    468:                        len = fragroundup(fs, blkoff(fs, xp->i_size));
1.5       hannken   469:                        if (len > 0 && len < fs->fs_bsize) {
1.1       hannken   470:                                ffs_blkfree(copy_fs, vp, db_get(xp, loc),
                    471:                                    len, xp->i_number);
                    472:                                blkno = db_get(xp, loc);
                    473:                                db_assign(xp, loc, 0);
                    474:                        }
                    475:                }
                    476:                snaplistsize += 1;
                    477:                if (xp->i_ump->um_fstype == UFS1)
                    478:                        error = expunge_ufs1(vp, xp, copy_fs, fullacct_ufs1,
                    479:                            BLK_NOCOPY);
                    480:                else
                    481:                        error = expunge_ufs2(vp, xp, copy_fs, fullacct_ufs2,
                    482:                            BLK_NOCOPY);
                    483:                if (blkno)
                    484:                        db_assign(xp, loc, blkno);
1.6       hannken   485:                if (!error)
                    486:                        error = ffs_freefile(copy_fs, vp, xp->i_number,
                    487:                            xp->i_mode);
1.1       hannken   488:                if (error) {
                    489:                        free(copy_fs->fs_csp, M_UFSMNT);
                    490:                        goto out1;
                    491:                }
                    492:                MNT_ILOCK(mp);
                    493:        }
                    494:        MNT_IUNLOCK(mp);
                    495:        /*
                    496:         * If there already exist snapshots on this filesystem, grab a
                    497:         * reference to their shared lock. If this is the first snapshot
                    498:         * on this filesystem, we need to allocate a lock for the snapshots
                    499:         * to share. In either case, acquire the snapshot lock and give
                    500:         * up our original private lock.
                    501:         */
                    502:        VI_LOCK(devvp);
1.47      hannken   503:        if ((xp = TAILQ_FIRST(&si->si_snapshots)) != NULL) {
1.1       hannken   504:                struct lock *lkp;
                    505:
                    506:                lkp = ITOV(xp)->v_vnlock;
                    507:                VI_UNLOCK(devvp);
                    508:                VI_LOCK(vp);
                    509:                vp->v_vnlock = lkp;
                    510:        } else {
                    511:                struct lock *lkp;
                    512:
                    513:                VI_UNLOCK(devvp);
                    514:                MALLOC(lkp, struct lock *, sizeof(struct lock), M_UFSMNT,
                    515:                    M_WAITOK);
                    516:                lockinit(lkp, PVFS, "snaplk", 0, LK_CANRECURSE);
                    517:                VI_LOCK(vp);
                    518:                vp->v_vnlock = lkp;
                    519:        }
                    520:        vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY);
                    521:        transferlockers(&vp->v_lock, vp->v_vnlock);
                    522:        lockmgr(&vp->v_lock, LK_RELEASE, NULL);
                    523:        /*
                    524:         * If this is the first snapshot on this filesystem, then we need
                    525:         * to allocate the space for the list of preallocated snapshot blocks.
                    526:         * This list will be refined below, but this preliminary one will
                    527:         * keep us out of deadlock until the full one is ready.
                    528:         */
                    529:        if (xp == NULL) {
1.25      christos  530:                snapblklist = malloc(
1.1       hannken   531:                    snaplistsize * sizeof(ufs2_daddr_t), M_UFSMNT, M_WAITOK);
                    532:                blkp = &snapblklist[1];
1.15      hannken   533:                *blkp++ = lblkno(fs, fs->fs_sblockloc);
1.1       hannken   534:                blkno = fragstoblks(fs, fs->fs_csaddr);
                    535:                for (cg = 0; cg < fs->fs_ncg; cg++) {
1.15      hannken   536:                        if (fragstoblks(fs, cgtod(fs, cg)) > blkno)
1.1       hannken   537:                                break;
1.15      hannken   538:                        *blkp++ = fragstoblks(fs, cgtod(fs, cg));
1.1       hannken   539:                }
                    540:                len = howmany(fs->fs_cssize, fs->fs_bsize);
                    541:                for (loc = 0; loc < len; loc++)
1.15      hannken   542:                        *blkp++ = blkno + loc;
1.1       hannken   543:                for (; cg < fs->fs_ncg; cg++)
1.15      hannken   544:                        *blkp++ = fragstoblks(fs, cgtod(fs, cg));
                    545:                snapblklist[0] = blkp - snapblklist;
1.1       hannken   546:                VI_LOCK(devvp);
1.47      hannken   547:                if (si->si_snapblklist != NULL)
1.1       hannken   548:                        panic("ffs_snapshot: non-empty list");
1.47      hannken   549:                si->si_snapblklist = snapblklist;
1.1       hannken   550:                VI_UNLOCK(devvp);
                    551:        }
                    552:        /*
                    553:         * Record snapshot inode. Since this is the newest snapshot,
                    554:         * it must be placed at the end of the list.
                    555:         */
                    556:        VI_LOCK(devvp);
                    557:        fs->fs_snapinum[snaploc] = ip->i_number;
                    558:        if (ip->i_nextsnap.tqe_prev != 0)
1.19      christos  559:                panic("ffs_snapshot: %llu already on list",
                    560:                    (unsigned long long)ip->i_number);
1.47      hannken   561:        TAILQ_INSERT_TAIL(&si->si_snapshots, ip, i_nextsnap);
1.1       hannken   562:        VI_UNLOCK(devvp);
                    563:        if (xp == NULL)
                    564:                vn_cow_establish(devvp, ffs_copyonwrite, devvp);
                    565:        vp->v_flag |= VSYSTEM;
                    566: out1:
                    567:        /*
                    568:         * Resume operation on filesystem.
                    569:         */
1.40      hannken   570:        vfs_resume(vp->v_mount);
1.1       hannken   571:        /*
                    572:         * Set the mtime to the time the snapshot has been taken.
                    573:         */
1.21      yamt      574:        TIMEVAL_TO_TIMESPEC(&starttime, &ts);
1.1       hannken   575:        if (ctime)
                    576:                *ctime = ts;
                    577:        DIP_ASSIGN(ip, mtime, ts.tv_sec);
                    578:        DIP_ASSIGN(ip, mtimensec, ts.tv_nsec);
                    579:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                    580:
                    581: #ifdef DEBUG
                    582:        if (starttime.tv_sec > 0) {
1.30      kardel    583:                getmicrotime(&endtime);
1.1       hannken   584:                timersub(&endtime, &starttime, &endtime);
                    585:                printf("%s: suspended %ld.%03ld sec, redo %ld of %d\n",
                    586:                    vp->v_mount->mnt_stat.f_mntonname, (long)endtime.tv_sec,
                    587:                    endtime.tv_usec / 1000, redo, fs->fs_ncg);
                    588:        }
                    589: #endif
                    590:        if (error)
                    591:                goto out;
                    592:        /*
                    593:         * Copy allocation information from all the snapshots in
                    594:         * this snapshot and then expunge them from its view.
                    595:         */
1.47      hannken   596:        TAILQ_FOREACH(xp, &si->si_snapshots, i_nextsnap) {
1.1       hannken   597:                if (xp == ip)
                    598:                        break;
                    599:                if (xp->i_ump->um_fstype == UFS1)
                    600:                        error = expunge_ufs1(vp, xp, fs, snapacct_ufs1,
                    601:                            BLK_SNAP);
                    602:                else
                    603:                        error = expunge_ufs2(vp, xp, fs, snapacct_ufs2,
                    604:                            BLK_SNAP);
1.48    ! hannken   605:                if (error == 0 && xp->i_ffs_effnlink == 0)
        !           606:                        error = ffs_freefile(copy_fs, vp,
        !           607:                            xp->i_number, xp->i_mode);
1.1       hannken   608:                if (error) {
                    609:                        fs->fs_snapinum[snaploc] = 0;
                    610:                        goto done;
                    611:                }
                    612:        }
                    613:        /*
                    614:         * Allocate space for the full list of preallocated snapshot blocks.
                    615:         */
1.25      christos  616:        snapblklist = malloc(snaplistsize * sizeof(ufs2_daddr_t),
1.1       hannken   617:            M_UFSMNT, M_WAITOK);
                    618:        ip->i_snapblklist = &snapblklist[1];
                    619:        /*
                    620:         * Expunge the blocks used by the snapshots from the set of
                    621:         * blocks marked as used in the snapshot bitmaps. Also, collect
                    622:         * the list of allocated blocks in i_snapblklist.
                    623:         */
                    624:        if (ip->i_ump->um_fstype == UFS1)
                    625:                error = expunge_ufs1(vp, ip, copy_fs, mapacct_ufs1, BLK_SNAP);
                    626:        else
                    627:                error = expunge_ufs2(vp, ip, copy_fs, mapacct_ufs2, BLK_SNAP);
                    628:        if (error) {
                    629:                fs->fs_snapinum[snaploc] = 0;
                    630:                FREE(snapblklist, M_UFSMNT);
                    631:                goto done;
                    632:        }
                    633:        if (snaplistsize < ip->i_snapblklist - snapblklist)
                    634:                panic("ffs_snapshot: list too small");
                    635:        snaplistsize = ip->i_snapblklist - snapblklist;
1.15      hannken   636:        snapblklist[0] = snaplistsize;
                    637:        ip->i_snapblklist = &snapblklist[0];
1.1       hannken   638:        /*
                    639:         * Write out the list of allocated blocks to the end of the snapshot.
                    640:         */
1.15      hannken   641:        for (i = 0; i < snaplistsize; i++)
                    642:                snapblklist[i] = ufs_rw64(snapblklist[i], ns);
1.43      christos  643:        error = vn_rdwr(UIO_WRITE, vp, (void *)snapblklist,
1.16      hannken   644:            snaplistsize*sizeof(ufs2_daddr_t), lblktosize(fs, (off_t)numblks),
1.31      ad        645:            UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, l->l_cred, NULL, NULL);
1.15      hannken   646:        for (i = 0; i < snaplistsize; i++)
                    647:                snapblklist[i] = ufs_rw64(snapblklist[i], ns);
1.1       hannken   648:        if (error) {
                    649:                fs->fs_snapinum[snaploc] = 0;
                    650:                FREE(snapblklist, M_UFSMNT);
                    651:                goto done;
                    652:        }
                    653:        /*
                    654:         * Write the superblock and its summary information
                    655:         * to the snapshot.
                    656:         */
                    657:        blkno = fragstoblks(fs, fs->fs_csaddr);
                    658:        len = howmany(fs->fs_cssize, fs->fs_bsize);
                    659:        space = copy_fs->fs_csp;
1.8       hannken   660: #ifdef FFS_EI
1.1       hannken   661:        if (ns) {
                    662:                ffs_sb_swap(copy_fs, copy_fs);
                    663:                ffs_csum_swap(space, space, fs->fs_cssize);
                    664:        }
1.8       hannken   665: #endif
1.1       hannken   666:        for (loc = 0; loc < len; loc++) {
1.15      hannken   667:                error = bread(vp, blkno + loc, fs->fs_bsize, KERNCRED, &nbp);
                    668:                if (error) {
                    669:                        brelse(nbp);
1.1       hannken   670:                        fs->fs_snapinum[snaploc] = 0;
                    671:                        FREE(snapblklist, M_UFSMNT);
                    672:                        goto done;
                    673:                }
1.15      hannken   674:                bcopy(space, nbp->b_data, fs->fs_bsize);
1.1       hannken   675:                space = (char *)space + fs->fs_bsize;
1.15      hannken   676:                bawrite(nbp);
1.1       hannken   677:        }
                    678:        /*
                    679:         * As this is the newest list, it is the most inclusive, so
1.15      hannken   680:         * should replace the previous list. If this is the first snapshot
                    681:         * free the preliminary list.
1.1       hannken   682:         */
                    683:        VI_LOCK(devvp);
1.47      hannken   684:        space = si->si_snapblklist;
                    685:        si->si_snapblklist = snapblklist;
1.1       hannken   686:        VI_UNLOCK(devvp);
1.47      hannken   687:        if (TAILQ_FIRST(&si->si_snapshots) == ip)
1.1       hannken   688:                FREE(space, M_UFSMNT);
                    689: done:
                    690:        free(copy_fs->fs_csp, M_UFSMNT);
1.15      hannken   691:        if (!error) {
                    692:                error = bread(vp, lblkno(fs, fs->fs_sblockloc), fs->fs_bsize,
                    693:                    KERNCRED, &nbp);
                    694:                if (error) {
                    695:                        brelse(nbp);
                    696:                        fs->fs_snapinum[snaploc] = 0;
                    697:                }
1.16      hannken   698:                bcopy(sbbuf, nbp->b_data, fs->fs_bsize);
1.15      hannken   699:                bawrite(nbp);
                    700:        }
1.1       hannken   701: out:
1.4       hannken   702:        /*
1.15      hannken   703:         * Invalidate and free all pages on the snapshot vnode.
                    704:         * All metadata has been written through the buffer cache.
                    705:         * Clean all dirty buffers now to avoid UBC inconsistencies.
1.4       hannken   706:         */
                    707:        if (!error) {
                    708:                simple_lock(&vp->v_interlock);
                    709:                error = VOP_PUTPAGES(vp, 0, 0,
                    710:                    PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
                    711:        }
1.15      hannken   712:        if (!error) {
                    713:                s = splbio();
                    714:                for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
                    715:                        nbp = LIST_NEXT(bp, b_vnbufs);
                    716:                        simple_lock(&bp->b_interlock);
                    717:                        splx(s);
                    718:                        if ((bp->b_flags & (B_DELWRI|B_BUSY)) != B_DELWRI)
                    719:                                panic("ffs_snapshot: not dirty or busy, bp %p",
                    720:                                    bp);
                    721:                        bp->b_flags |= B_BUSY|B_VFLUSH;
                    722:                        if (LIST_FIRST(&bp->b_dep) == NULL)
                    723:                                bp->b_flags |= B_NOCACHE;
                    724:                        simple_unlock(&bp->b_interlock);
                    725:                        bwrite(bp);
                    726:                        s = splbio();
                    727:                }
                    728:                simple_lock(&global_v_numoutput_slock);
                    729:                while (vp->v_numoutput) {
                    730:                        vp->v_flag |= VBWAIT;
1.43      christos  731:                        ltsleep((void *)&vp->v_numoutput, PRIBIO+1,
1.15      hannken   732:                            "snapflushbuf", 0, &global_v_numoutput_slock);
                    733:                }
                    734:                simple_unlock(&global_v_numoutput_slock);
                    735:                splx(s);
                    736:        }
1.16      hannken   737:        if (sbbuf)
                    738:                free(sbbuf, M_UFSMNT);
1.1       hannken   739:        if (fs->fs_active != 0) {
                    740:                FREE(fs->fs_active, M_DEVBUF);
                    741:                fs->fs_active = 0;
                    742:        }
                    743:        mp->mnt_flag = flag;
                    744:        if (error)
1.23      christos  745:                (void) ffs_truncate(vp, (off_t)0, 0, NOCRED, l);
1.1       hannken   746:        else
                    747:                vref(vp);
                    748:        return (error);
                    749: }
                    750:
                    751: /*
                    752:  * Copy a cylinder group map. All the unallocated blocks are marked
                    753:  * BLK_NOCOPY so that the snapshot knows that it need not copy them
                    754:  * if they are later written. If passno is one, then this is a first
                    755:  * pass, so only setting needs to be done. If passno is 2, then this
                    756:  * is a revision to a previous pass which must be undone as the
                    757:  * replacement pass is done.
                    758:  */
                    759: static int
1.43      christos  760: cgaccount(int cg, struct vnode *vp, void *data, int passno)
1.1       hannken   761: {
                    762:        struct buf *bp, *ibp;
                    763:        struct inode *ip;
                    764:        struct cg *cgp;
                    765:        struct fs *fs;
                    766:        ufs2_daddr_t base, numblks;
                    767:        int error, len, loc, ns, indiroff;
                    768:
                    769:        ip = VTOI(vp);
                    770:        fs = ip->i_fs;
                    771:        ns = UFS_FSNEEDSWAP(fs);
                    772:        error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
                    773:                (int)fs->fs_cgsize, KERNCRED, &bp);
                    774:        if (error) {
                    775:                brelse(bp);
                    776:                return (error);
                    777:        }
                    778:        cgp = (struct cg *)bp->b_data;
                    779:        if (!cg_chkmagic(cgp, ns)) {
                    780:                brelse(bp);
                    781:                return (EIO);
                    782:        }
                    783:        ACTIVECG_SET(fs, cg);
                    784:
                    785:        bcopy(bp->b_data, data, fs->fs_cgsize);
                    786:        brelse(bp);
                    787:        if (fs->fs_cgsize < fs->fs_bsize)
1.43      christos  788:                memset((char *)data + fs->fs_cgsize, 0,
1.1       hannken   789:                    fs->fs_bsize - fs->fs_cgsize);
                    790:        numblks = howmany(fs->fs_size, fs->fs_frag);
                    791:        len = howmany(fs->fs_fpg, fs->fs_frag);
                    792:        base = cg * fs->fs_fpg / fs->fs_frag;
                    793:        if (base + len >= numblks)
                    794:                len = numblks - base - 1;
                    795:        loc = 0;
                    796:        if (base < NDADDR) {
                    797:                for ( ; loc < NDADDR; loc++) {
                    798:                        if (ffs_isblock(fs, cg_blksfree(cgp, ns), loc))
                    799:                                db_assign(ip, loc, BLK_NOCOPY);
                    800:                        else if (db_get(ip, loc) == BLK_NOCOPY) {
                    801:                                if (passno == 2)
                    802:                                        db_assign(ip, loc, 0);
                    803:                                else if (passno == 1)
                    804:                                        panic("ffs_snapshot: lost direct block");
                    805:                        }
                    806:                }
                    807:        }
1.22      yamt      808:        if ((error = ffs_balloc(vp, lblktosize(fs, (off_t)(base + loc)),
1.11      perry     809:            fs->fs_bsize, KERNCRED, B_METAONLY, &ibp)) != 0)
1.1       hannken   810:                return (error);
                    811:        indiroff = (base + loc - NDADDR) % NINDIR(fs);
                    812:        for ( ; loc < len; loc++, indiroff++) {
                    813:                if (indiroff >= NINDIR(fs)) {
1.15      hannken   814:                        bawrite(ibp);
1.22      yamt      815:                        if ((error = ffs_balloc(vp,
1.1       hannken   816:                            lblktosize(fs, (off_t)(base + loc)),
                    817:                            fs->fs_bsize, KERNCRED, B_METAONLY, &ibp)) != 0)
                    818:                                return (error);
                    819:                        indiroff = 0;
                    820:                }
                    821:                if (ffs_isblock(fs, cg_blksfree(cgp, ns), loc))
                    822:                        idb_assign(ip, ibp->b_data, indiroff, BLK_NOCOPY);
                    823:                else if (idb_get(ip, ibp->b_data, indiroff) == BLK_NOCOPY) {
                    824:                        if (passno == 2)
                    825:                                idb_assign(ip, ibp->b_data, indiroff, 0);
                    826:                        else if (passno == 1)
                    827:                                panic("ffs_snapshot: lost indirect block");
                    828:                }
                    829:        }
1.15      hannken   830:        bdwrite(ibp);
1.1       hannken   831:        return (0);
                    832: }
                    833:
                    834: /*
                    835:  * Before expunging a snapshot inode, note all the
                    836:  * blocks that it claims with BLK_SNAP so that fsck will
                    837:  * be able to account for those blocks properly and so
                    838:  * that this snapshot knows that it need not copy them
                    839:  * if the other snapshot holding them is freed. This code
                    840:  * is reproduced once each for UFS1 and UFS2.
                    841:  */
                    842: static int
1.18      thorpej   843: expunge_ufs1(struct vnode *snapvp, struct inode *cancelip, struct fs *fs,
                    844:     int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
                    845:                    struct fs *, ufs_lbn_t, int),
                    846:     int expungetype)
1.1       hannken   847: {
1.4       hannken   848:        int i, s, error, ns, indiroff;
1.1       hannken   849:        ufs_lbn_t lbn, rlbn;
                    850:        ufs2_daddr_t len, blkno, numblks, blksperindir;
                    851:        struct ufs1_dinode *dip;
                    852:        struct buf *bp;
1.43      christos  853:        void *bf;
1.1       hannken   854:
                    855:        ns = UFS_FSNEEDSWAP(fs);
                    856:        /*
                    857:         * Prepare to expunge the inode. If its inode block has not
                    858:         * yet been copied, then allocate and fill the copy.
                    859:         */
                    860:        lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
                    861:        blkno = 0;
                    862:        if (lbn < NDADDR) {
                    863:                blkno = db_get(VTOI(snapvp), lbn);
                    864:        } else {
1.4       hannken   865:                s = cow_enter();
1.22      yamt      866:                error = ffs_balloc(snapvp, lblktosize(fs, (off_t)lbn),
1.1       hannken   867:                   fs->fs_bsize, KERNCRED, B_METAONLY, &bp);
1.4       hannken   868:                cow_leave(s);
1.1       hannken   869:                if (error)
                    870:                        return (error);
                    871:                indiroff = (lbn - NDADDR) % NINDIR(fs);
                    872:                blkno = idb_get(VTOI(snapvp), bp->b_data, indiroff);
                    873:                brelse(bp);
                    874:        }
1.17      christos  875:        bf = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.1       hannken   876:        if (blkno != 0)
1.17      christos  877:                error = readvnblk(snapvp, bf, lbn);
1.1       hannken   878:        else
1.17      christos  879:                error = readfsblk(snapvp, bf, lbn);
1.1       hannken   880:        if (error) {
1.17      christos  881:                free(bf, M_UFSMNT);
1.1       hannken   882:                return error;
                    883:        }
                    884:        /*
                    885:         * Set a snapshot inode to be a zero length file, regular files
1.48    ! hannken   886:         * or unlinked snapshots to be completely unallocated.
1.1       hannken   887:         */
1.17      christos  888:        dip = (struct ufs1_dinode *)bf + ino_to_fsbo(fs, cancelip->i_number);
1.48    ! hannken   889:        if (expungetype == BLK_NOCOPY || cancelip->i_ffs_effnlink == 0)
1.1       hannken   890:                dip->di_mode = 0;
                    891:        dip->di_size = 0;
                    892:        dip->di_blocks = 0;
                    893:        dip->di_flags =
                    894:            ufs_rw32(ufs_rw32(dip->di_flags, ns) & ~SF_SNAPSHOT, ns);
                    895:        bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs1_daddr_t));
1.17      christos  896:        error = writevnblk(snapvp, bf, lbn);
                    897:        free(bf, M_UFSMNT);
1.1       hannken   898:        if (error)
                    899:                return error;
                    900:        /*
                    901:         * Now go through and expunge all the blocks in the file
                    902:         * using the function requested.
                    903:         */
                    904:        numblks = howmany(cancelip->i_size, fs->fs_bsize);
                    905:        if ((error = (*acctfunc)(snapvp, &cancelip->i_ffs1_db[0],
                    906:            &cancelip->i_ffs1_db[NDADDR], fs, 0, expungetype)))
                    907:                return (error);
                    908:        if ((error = (*acctfunc)(snapvp, &cancelip->i_ffs1_ib[0],
                    909:            &cancelip->i_ffs1_ib[NIADDR], fs, -1, expungetype)))
                    910:                return (error);
                    911:        blksperindir = 1;
                    912:        lbn = -NDADDR;
                    913:        len = numblks - NDADDR;
                    914:        rlbn = NDADDR;
                    915:        for (i = 0; len > 0 && i < NIADDR; i++) {
                    916:                error = indiracct_ufs1(snapvp, ITOV(cancelip), i,
                    917:                    ufs_rw32(cancelip->i_ffs1_ib[i], ns), lbn, rlbn, len,
                    918:                    blksperindir, fs, acctfunc, expungetype);
                    919:                if (error)
                    920:                        return (error);
                    921:                blksperindir *= NINDIR(fs);
                    922:                lbn -= blksperindir + 1;
                    923:                len -= blksperindir;
                    924:                rlbn += blksperindir;
                    925:        }
                    926:        return (0);
                    927: }
                    928:
                    929: /*
                    930:  * Descend an indirect block chain for vnode cancelvp accounting for all
                    931:  * its indirect blocks in snapvp.
1.11      perry     932:  */
1.1       hannken   933: static int
1.18      thorpej   934: indiracct_ufs1(struct vnode *snapvp, struct vnode *cancelvp, int level,
                    935:     ufs1_daddr_t blkno, ufs_lbn_t lbn, ufs_lbn_t rlbn, ufs_lbn_t remblks,
                    936:     ufs_lbn_t blksperindir, struct fs *fs,
                    937:     int (*acctfunc)(struct vnode *, ufs1_daddr_t *, ufs1_daddr_t *,
                    938:                    struct fs *, ufs_lbn_t, int),
                    939:     int expungetype)
1.1       hannken   940: {
                    941:        int error, ns, num, i;
                    942:        ufs_lbn_t subblksperindir;
                    943:        struct indir indirs[NIADDR + 2];
                    944:        ufs1_daddr_t last, *bap;
                    945:        struct buf *bp;
                    946:
                    947:        ns = UFS_FSNEEDSWAP(fs);
                    948:
                    949:        if (blkno == 0) {
                    950:                if (expungetype == BLK_NOCOPY)
                    951:                        return (0);
                    952:                panic("indiracct_ufs1: missing indir");
                    953:        }
                    954:        if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
                    955:                return (error);
                    956:        if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
                    957:                panic("indiracct_ufs1: botched params");
                    958:        /*
                    959:         * We have to expand bread here since it will deadlock looking
                    960:         * up the block number for any blocks that are not in the cache.
                    961:         */
                    962:        bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0);
                    963:        bp->b_blkno = fsbtodb(fs, blkno);
                    964:        if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
                    965:            (error = readfsblk(bp->b_vp, bp->b_data, fragstoblks(fs, blkno)))) {
                    966:                brelse(bp);
                    967:                return (error);
                    968:        }
                    969:        /*
                    970:         * Account for the block pointers in this indirect block.
                    971:         */
                    972:        last = howmany(remblks, blksperindir);
                    973:        if (last > NINDIR(fs))
                    974:                last = NINDIR(fs);
1.25      christos  975:        bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1.43      christos  976:        bcopy(bp->b_data, (void *)bap, fs->fs_bsize);
1.1       hannken   977:        brelse(bp);
                    978:        error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
                    979:            level == 0 ? rlbn : -1, expungetype);
                    980:        if (error || level == 0)
                    981:                goto out;
                    982:        /*
                    983:         * Account for the block pointers in each of the indirect blocks
                    984:         * in the levels below us.
                    985:         */
                    986:        subblksperindir = blksperindir / NINDIR(fs);
                    987:        for (lbn++, level--, i = 0; i < last; i++) {
                    988:                error = indiracct_ufs1(snapvp, cancelvp, level,
                    989:                    ufs_rw32(bap[i], ns), lbn, rlbn, remblks, subblksperindir,
                    990:                    fs, acctfunc, expungetype);
                    991:                if (error)
                    992:                        goto out;
                    993:                rlbn += blksperindir;
                    994:                lbn -= blksperindir;
                    995:                remblks -= blksperindir;
                    996:        }
                    997: out:
                    998:        FREE(bap, M_DEVBUF);
                    999:        return (error);
                   1000: }
                   1001:
                   1002: /*
                   1003:  * Do both snap accounting and map accounting.
                   1004:  */
                   1005: static int
1.18      thorpej  1006: fullacct_ufs1(struct vnode *vp, ufs1_daddr_t *oldblkp, ufs1_daddr_t *lastblkp,
                   1007:     struct fs *fs, ufs_lbn_t lblkno,
                   1008:     int exptype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1009: {
                   1010:        int error;
                   1011:
                   1012:        if ((error = snapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
                   1013:                return (error);
                   1014:        return (mapacct_ufs1(vp, oldblkp, lastblkp, fs, lblkno, exptype));
                   1015: }
                   1016:
                   1017: /*
                   1018:  * Identify a set of blocks allocated in a snapshot inode.
                   1019:  */
                   1020: static int
1.18      thorpej  1021: snapacct_ufs1(struct vnode *vp, ufs1_daddr_t *oldblkp, ufs1_daddr_t *lastblkp,
1.36      christos 1022:     struct fs *fs, ufs_lbn_t lblkno,
1.18      thorpej  1023:     int expungetype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1024: {
                   1025:        struct inode *ip = VTOI(vp);
                   1026:        ufs1_daddr_t blkno, *blkp;
                   1027:        ufs_lbn_t lbn;
                   1028:        struct buf *ibp;
                   1029:        int error, ns;
                   1030:
                   1031:        ns = UFS_FSNEEDSWAP(fs);
                   1032:
                   1033:        for ( ; oldblkp < lastblkp; oldblkp++) {
                   1034:                blkno = ufs_rw32(*oldblkp, ns);
                   1035:                if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
                   1036:                        continue;
                   1037:                lbn = fragstoblks(fs, blkno);
                   1038:                if (lbn < NDADDR) {
                   1039:                        blkp = &ip->i_ffs1_db[lbn];
                   1040:                        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1041:                } else {
1.22      yamt     1042:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.1       hannken  1043:                            fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
                   1044:                        if (error)
                   1045:                                return (error);
                   1046:                        blkp = &((ufs1_daddr_t *)(ibp->b_data))
                   1047:                            [(lbn - NDADDR) % NINDIR(fs)];
                   1048:                }
                   1049:                /*
                   1050:                 * If we are expunging a snapshot vnode and we
                   1051:                 * find a block marked BLK_NOCOPY, then it is
                   1052:                 * one that has been allocated to this snapshot after
                   1053:                 * we took our current snapshot and can be ignored.
                   1054:                 */
                   1055:                blkno = ufs_rw32(*blkp, ns);
                   1056:                if (expungetype == BLK_SNAP && blkno == BLK_NOCOPY) {
                   1057:                        if (lbn >= NDADDR)
                   1058:                                brelse(ibp);
                   1059:                } else {
                   1060:                        if (blkno != 0)
                   1061:                                panic("snapacct_ufs1: bad block");
                   1062:                        *blkp = ufs_rw32(expungetype, ns);
                   1063:                        if (lbn >= NDADDR)
1.15      hannken  1064:                                bdwrite(ibp);
1.1       hannken  1065:                }
                   1066:        }
                   1067:        return (0);
                   1068: }
                   1069:
                   1070: /*
                   1071:  * Account for a set of blocks allocated in a snapshot inode.
                   1072:  */
                   1073: static int
1.18      thorpej  1074: mapacct_ufs1(struct vnode *vp, ufs1_daddr_t *oldblkp, ufs1_daddr_t *lastblkp,
                   1075:     struct fs *fs, ufs_lbn_t lblkno, int expungetype)
1.1       hannken  1076: {
                   1077:        ufs1_daddr_t blkno;
                   1078:        struct inode *ip;
                   1079:        ino_t inum;
                   1080:        int acctit, ns;
                   1081:
                   1082:        ns = UFS_FSNEEDSWAP(fs);
                   1083:        ip = VTOI(vp);
                   1084:        inum = ip->i_number;
                   1085:        if (lblkno == -1)
                   1086:                acctit = 0;
                   1087:        else
                   1088:                acctit = 1;
                   1089:        for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
                   1090:                blkno = ufs_rw32(*oldblkp, ns);
                   1091:                if (blkno == 0 || blkno == BLK_NOCOPY)
                   1092:                        continue;
                   1093:                if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1.15      hannken  1094:                        *ip->i_snapblklist++ = lblkno;
1.1       hannken  1095:                if (blkno == BLK_SNAP)
                   1096:                        blkno = blkstofrags(fs, lblkno);
                   1097:                ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
                   1098:        }
                   1099:        return (0);
                   1100: }
                   1101:
                   1102: /*
                   1103:  * Before expunging a snapshot inode, note all the
                   1104:  * blocks that it claims with BLK_SNAP so that fsck will
                   1105:  * be able to account for those blocks properly and so
                   1106:  * that this snapshot knows that it need not copy them
                   1107:  * if the other snapshot holding them is freed. This code
                   1108:  * is reproduced once each for UFS1 and UFS2.
                   1109:  */
                   1110: static int
1.18      thorpej  1111: expunge_ufs2(struct vnode *snapvp, struct inode *cancelip, struct fs *fs,
                   1112:     int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
                   1113:                    struct fs *, ufs_lbn_t, int),
                   1114:     int expungetype)
1.1       hannken  1115: {
1.4       hannken  1116:        int i, s, error, ns, indiroff;
1.1       hannken  1117:        ufs_lbn_t lbn, rlbn;
                   1118:        ufs2_daddr_t len, blkno, numblks, blksperindir;
                   1119:        struct ufs2_dinode *dip;
                   1120:        struct buf *bp;
1.43      christos 1121:        void *bf;
1.1       hannken  1122:
                   1123:        ns = UFS_FSNEEDSWAP(fs);
                   1124:        /*
                   1125:         * Prepare to expunge the inode. If its inode block has not
                   1126:         * yet been copied, then allocate and fill the copy.
                   1127:         */
                   1128:        lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
                   1129:        blkno = 0;
                   1130:        if (lbn < NDADDR) {
                   1131:                blkno = db_get(VTOI(snapvp), lbn);
                   1132:        } else {
1.4       hannken  1133:                s = cow_enter();
1.22      yamt     1134:                error = ffs_balloc(snapvp, lblktosize(fs, (off_t)lbn),
1.1       hannken  1135:                   fs->fs_bsize, KERNCRED, B_METAONLY, &bp);
1.4       hannken  1136:                cow_leave(s);
1.1       hannken  1137:                if (error)
                   1138:                        return (error);
                   1139:                indiroff = (lbn - NDADDR) % NINDIR(fs);
                   1140:                blkno = idb_get(VTOI(snapvp), bp->b_data, indiroff);
                   1141:                brelse(bp);
                   1142:        }
1.17      christos 1143:        bf = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.1       hannken  1144:        if (blkno != 0)
1.17      christos 1145:                error = readvnblk(snapvp, bf, lbn);
1.1       hannken  1146:        else
1.17      christos 1147:                error = readfsblk(snapvp, bf, lbn);
1.1       hannken  1148:        if (error) {
1.17      christos 1149:                free(bf, M_UFSMNT);
1.1       hannken  1150:                return error;
                   1151:        }
                   1152:        /*
                   1153:         * Set a snapshot inode to be a zero length file, regular files
1.48    ! hannken  1154:         * or unlinked snapshots to be completely unallocated.
1.1       hannken  1155:         */
1.17      christos 1156:        dip = (struct ufs2_dinode *)bf + ino_to_fsbo(fs, cancelip->i_number);
1.48    ! hannken  1157:        if (expungetype == BLK_NOCOPY || cancelip->i_ffs_effnlink == 0)
1.1       hannken  1158:                dip->di_mode = 0;
                   1159:        dip->di_size = 0;
                   1160:        dip->di_blocks = 0;
                   1161:        dip->di_flags =
                   1162:            ufs_rw32(ufs_rw32(dip->di_flags, ns) & ~SF_SNAPSHOT, ns);
                   1163:        bzero(&dip->di_db[0], (NDADDR + NIADDR) * sizeof(ufs2_daddr_t));
1.17      christos 1164:        error = writevnblk(snapvp, bf, lbn);
                   1165:        free(bf, M_UFSMNT);
1.1       hannken  1166:        if (error)
                   1167:                return error;
                   1168:        /*
                   1169:         * Now go through and expunge all the blocks in the file
                   1170:         * using the function requested.
                   1171:         */
                   1172:        numblks = howmany(cancelip->i_size, fs->fs_bsize);
                   1173:        if ((error = (*acctfunc)(snapvp, &cancelip->i_ffs2_db[0],
                   1174:            &cancelip->i_ffs2_db[NDADDR], fs, 0, expungetype)))
                   1175:                return (error);
                   1176:        if ((error = (*acctfunc)(snapvp, &cancelip->i_ffs2_ib[0],
                   1177:            &cancelip->i_ffs2_ib[NIADDR], fs, -1, expungetype)))
                   1178:                return (error);
                   1179:        blksperindir = 1;
                   1180:        lbn = -NDADDR;
                   1181:        len = numblks - NDADDR;
                   1182:        rlbn = NDADDR;
                   1183:        for (i = 0; len > 0 && i < NIADDR; i++) {
                   1184:                error = indiracct_ufs2(snapvp, ITOV(cancelip), i,
                   1185:                    ufs_rw64(cancelip->i_ffs2_ib[i], ns), lbn, rlbn, len,
                   1186:                    blksperindir, fs, acctfunc, expungetype);
                   1187:                if (error)
                   1188:                        return (error);
                   1189:                blksperindir *= NINDIR(fs);
                   1190:                lbn -= blksperindir + 1;
                   1191:                len -= blksperindir;
                   1192:                rlbn += blksperindir;
                   1193:        }
                   1194:        return (0);
                   1195: }
                   1196:
                   1197: /*
                   1198:  * Descend an indirect block chain for vnode cancelvp accounting for all
                   1199:  * its indirect blocks in snapvp.
1.11      perry    1200:  */
1.1       hannken  1201: static int
1.18      thorpej  1202: indiracct_ufs2(struct vnode *snapvp, struct vnode *cancelvp, int level,
                   1203:     ufs2_daddr_t blkno, ufs_lbn_t lbn, ufs_lbn_t rlbn, ufs_lbn_t remblks,
                   1204:     ufs_lbn_t blksperindir, struct fs *fs,
                   1205:     int (*acctfunc)(struct vnode *, ufs2_daddr_t *, ufs2_daddr_t *,
                   1206:                    struct fs *, ufs_lbn_t, int),
                   1207:     int expungetype)
1.1       hannken  1208: {
                   1209:        int error, ns, num, i;
                   1210:        ufs_lbn_t subblksperindir;
                   1211:        struct indir indirs[NIADDR + 2];
                   1212:        ufs2_daddr_t last, *bap;
                   1213:        struct buf *bp;
                   1214:
                   1215:        ns = UFS_FSNEEDSWAP(fs);
                   1216:
                   1217:        if (blkno == 0) {
                   1218:                if (expungetype == BLK_NOCOPY)
                   1219:                        return (0);
                   1220:                panic("indiracct_ufs2: missing indir");
                   1221:        }
                   1222:        if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
                   1223:                return (error);
                   1224:        if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
                   1225:                panic("indiracct_ufs2: botched params");
                   1226:        /*
                   1227:         * We have to expand bread here since it will deadlock looking
                   1228:         * up the block number for any blocks that are not in the cache.
                   1229:         */
                   1230:        bp = getblk(cancelvp, lbn, fs->fs_bsize, 0, 0);
                   1231:        bp->b_blkno = fsbtodb(fs, blkno);
                   1232:        if ((bp->b_flags & (B_DONE | B_DELWRI)) == 0 &&
                   1233:            (error = readfsblk(bp->b_vp, bp->b_data, fragstoblks(fs, blkno)))) {
                   1234:                brelse(bp);
                   1235:                return (error);
                   1236:        }
                   1237:        /*
                   1238:         * Account for the block pointers in this indirect block.
                   1239:         */
                   1240:        last = howmany(remblks, blksperindir);
                   1241:        if (last > NINDIR(fs))
                   1242:                last = NINDIR(fs);
1.25      christos 1243:        bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1.43      christos 1244:        bcopy(bp->b_data, (void *)bap, fs->fs_bsize);
1.1       hannken  1245:        brelse(bp);
                   1246:        error = (*acctfunc)(snapvp, &bap[0], &bap[last], fs,
                   1247:            level == 0 ? rlbn : -1, expungetype);
                   1248:        if (error || level == 0)
                   1249:                goto out;
                   1250:        /*
                   1251:         * Account for the block pointers in each of the indirect blocks
                   1252:         * in the levels below us.
                   1253:         */
                   1254:        subblksperindir = blksperindir / NINDIR(fs);
                   1255:        for (lbn++, level--, i = 0; i < last; i++) {
                   1256:                error = indiracct_ufs2(snapvp, cancelvp, level,
                   1257:                    ufs_rw64(bap[i], ns), lbn, rlbn, remblks, subblksperindir,
                   1258:                    fs, acctfunc, expungetype);
                   1259:                if (error)
                   1260:                        goto out;
                   1261:                rlbn += blksperindir;
                   1262:                lbn -= blksperindir;
                   1263:                remblks -= blksperindir;
                   1264:        }
                   1265: out:
                   1266:        FREE(bap, M_DEVBUF);
                   1267:        return (error);
                   1268: }
                   1269:
                   1270: /*
                   1271:  * Do both snap accounting and map accounting.
                   1272:  */
                   1273: static int
1.18      thorpej  1274: fullacct_ufs2(struct vnode *vp, ufs2_daddr_t *oldblkp, ufs2_daddr_t *lastblkp,
                   1275:     struct fs *fs, ufs_lbn_t lblkno,
                   1276:     int exptype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1277: {
                   1278:        int error;
                   1279:
                   1280:        if ((error = snapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype)))
                   1281:                return (error);
                   1282:        return (mapacct_ufs2(vp, oldblkp, lastblkp, fs, lblkno, exptype));
                   1283: }
                   1284:
                   1285: /*
                   1286:  * Identify a set of blocks allocated in a snapshot inode.
                   1287:  */
                   1288: static int
1.18      thorpej  1289: snapacct_ufs2(struct vnode *vp, ufs2_daddr_t *oldblkp, ufs2_daddr_t *lastblkp,
1.36      christos 1290:     struct fs *fs, ufs_lbn_t lblkno,
1.18      thorpej  1291:     int expungetype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1292: {
                   1293:        struct inode *ip = VTOI(vp);
                   1294:        ufs2_daddr_t blkno, *blkp;
                   1295:        ufs_lbn_t lbn;
                   1296:        struct buf *ibp;
                   1297:        int error, ns;
                   1298:
                   1299:        ns = UFS_FSNEEDSWAP(fs);
                   1300:
                   1301:        for ( ; oldblkp < lastblkp; oldblkp++) {
                   1302:                blkno = ufs_rw64(*oldblkp, ns);
                   1303:                if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
                   1304:                        continue;
                   1305:                lbn = fragstoblks(fs, blkno);
                   1306:                if (lbn < NDADDR) {
                   1307:                        blkp = &ip->i_ffs2_db[lbn];
                   1308:                        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1309:                } else {
1.22      yamt     1310:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.1       hannken  1311:                            fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
                   1312:                        if (error)
                   1313:                                return (error);
                   1314:                        blkp = &((ufs2_daddr_t *)(ibp->b_data))
                   1315:                            [(lbn - NDADDR) % NINDIR(fs)];
                   1316:                }
                   1317:                /*
                   1318:                 * If we are expunging a snapshot vnode and we
                   1319:                 * find a block marked BLK_NOCOPY, then it is
                   1320:                 * one that has been allocated to this snapshot after
                   1321:                 * we took our current snapshot and can be ignored.
                   1322:                 */
                   1323:                blkno = ufs_rw64(*blkp, ns);
                   1324:                if (expungetype == BLK_SNAP && blkno == BLK_NOCOPY) {
                   1325:                        if (lbn >= NDADDR)
                   1326:                                brelse(ibp);
                   1327:                } else {
                   1328:                        if (blkno != 0)
                   1329:                                panic("snapacct_ufs2: bad block");
                   1330:                        *blkp = ufs_rw64(expungetype, ns);
                   1331:                        if (lbn >= NDADDR)
1.15      hannken  1332:                                bdwrite(ibp);
1.1       hannken  1333:                }
                   1334:        }
                   1335:        return (0);
                   1336: }
                   1337:
                   1338: /*
                   1339:  * Account for a set of blocks allocated in a snapshot inode.
                   1340:  */
                   1341: static int
1.18      thorpej  1342: mapacct_ufs2(struct vnode *vp, ufs2_daddr_t *oldblkp, ufs2_daddr_t *lastblkp,
                   1343:     struct fs *fs, ufs_lbn_t lblkno, int expungetype)
1.1       hannken  1344: {
                   1345:        ufs2_daddr_t blkno;
                   1346:        struct inode *ip;
                   1347:        ino_t inum;
                   1348:        int acctit, ns;
                   1349:
                   1350:        ns = UFS_FSNEEDSWAP(fs);
                   1351:        ip = VTOI(vp);
                   1352:        inum = ip->i_number;
                   1353:        if (lblkno == -1)
                   1354:                acctit = 0;
                   1355:        else
                   1356:                acctit = 1;
                   1357:        for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
                   1358:                blkno = ufs_rw64(*oldblkp, ns);
                   1359:                if (blkno == 0 || blkno == BLK_NOCOPY)
                   1360:                        continue;
                   1361:                if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1.15      hannken  1362:                        *ip->i_snapblklist++ = lblkno;
1.1       hannken  1363:                if (blkno == BLK_SNAP)
                   1364:                        blkno = blkstofrags(fs, lblkno);
                   1365:                ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
                   1366:        }
                   1367:        return (0);
                   1368: }
1.10      hannken  1369: #endif /* defined(FFS_NO_SNAPSHOT) */
1.1       hannken  1370:
                   1371: /*
                   1372:  * Decrement extra reference on snapshot when last name is removed.
                   1373:  * It will not be freed until the last open reference goes away.
                   1374:  */
                   1375: void
1.18      thorpej  1376: ffs_snapgone(struct inode *ip)
1.1       hannken  1377: {
1.47      hannken  1378:        struct mount *mp = ip->i_devvp->v_specmountpoint;
1.1       hannken  1379:        struct inode *xp;
                   1380:        struct fs *fs;
1.47      hannken  1381:        struct snap_info *si;
1.1       hannken  1382:        int snaploc;
                   1383:
1.47      hannken  1384:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1385:                return;
1.1       hannken  1386:        /*
                   1387:         * Find snapshot in incore list.
                   1388:         */
1.47      hannken  1389:        TAILQ_FOREACH(xp, &si->si_snapshots, i_nextsnap)
1.1       hannken  1390:                if (xp == ip)
                   1391:                        break;
                   1392:        if (xp != NULL)
                   1393:                vrele(ITOV(ip));
                   1394: #ifdef DEBUG
                   1395:        else if (snapdebug)
1.19      christos 1396:                printf("ffs_snapgone: lost snapshot vnode %llu\n",
                   1397:                    (unsigned long long)ip->i_number);
1.1       hannken  1398: #endif
                   1399:        /*
                   1400:         * Delete snapshot inode from superblock. Keep list dense.
                   1401:         */
                   1402:        fs = ip->i_fs;
                   1403:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
                   1404:                if (fs->fs_snapinum[snaploc] == ip->i_number)
                   1405:                        break;
                   1406:        if (snaploc < FSMAXSNAP) {
                   1407:                for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
                   1408:                        if (fs->fs_snapinum[snaploc] == 0)
                   1409:                                break;
                   1410:                        fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
                   1411:                }
                   1412:                fs->fs_snapinum[snaploc - 1] = 0;
                   1413:        }
                   1414: }
                   1415:
                   1416: /*
                   1417:  * Prepare a snapshot file for being removed.
                   1418:  */
                   1419: void
1.18      thorpej  1420: ffs_snapremove(struct vnode *vp)
1.1       hannken  1421: {
1.15      hannken  1422:        struct inode *ip = VTOI(vp), *xp;
1.1       hannken  1423:        struct vnode *devvp = ip->i_devvp;
                   1424:        struct fs *fs = ip->i_fs;
1.47      hannken  1425:        struct mount *mp = devvp->v_specmountpoint;
1.1       hannken  1426:        struct lock *lkp;
                   1427:        struct buf *ibp;
1.47      hannken  1428:        struct snap_info *si;
1.15      hannken  1429:        ufs2_daddr_t numblks, blkno, dblk;
1.1       hannken  1430:        int error, ns, loc, last;
                   1431:
1.47      hannken  1432:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1433:                return;
1.1       hannken  1434:        ns = UFS_FSNEEDSWAP(fs);
                   1435:        /*
                   1436:         * If active, delete from incore list (this snapshot may
                   1437:         * already have been in the process of being deleted, so
                   1438:         * would not have been active).
                   1439:         *
                   1440:         * Clear copy-on-write flag if last snapshot.
                   1441:         */
                   1442:        if (ip->i_nextsnap.tqe_prev != 0) {
                   1443:                VI_LOCK(devvp);
                   1444:                lockmgr(&vp->v_lock, LK_INTERLOCK | LK_EXCLUSIVE,
                   1445:                    VI_MTX(devvp));
                   1446:                VI_LOCK(devvp);
1.47      hannken  1447:                TAILQ_REMOVE(&si->si_snapshots, ip, i_nextsnap);
1.1       hannken  1448:                ip->i_nextsnap.tqe_prev = 0;
                   1449:                lkp = vp->v_vnlock;
                   1450:                vp->v_vnlock = &vp->v_lock;
                   1451:                lockmgr(lkp, LK_RELEASE, NULL);
1.47      hannken  1452:                if (TAILQ_FIRST(&si->si_snapshots) != 0) {
1.15      hannken  1453:                        /* Roll back the list of preallocated blocks. */
1.47      hannken  1454:                        xp = TAILQ_LAST(&si->si_snapshots, inodelst);
                   1455:                        si->si_snapblklist = xp->i_snapblklist;
1.1       hannken  1456:                        VI_UNLOCK(devvp);
                   1457:                } else {
1.47      hannken  1458:                        si->si_snapblklist = 0;
1.1       hannken  1459:                        lockmgr(lkp, LK_DRAIN|LK_INTERLOCK, VI_MTX(devvp));
                   1460:                        lockmgr(lkp, LK_RELEASE, NULL);
                   1461:                        vn_cow_disestablish(devvp, ffs_copyonwrite, devvp);
                   1462:                        FREE(lkp, M_UFSMNT);
                   1463:                }
1.15      hannken  1464:                FREE(ip->i_snapblklist, M_UFSMNT);
                   1465:                ip->i_snapblklist = NULL;
1.1       hannken  1466:        }
                   1467:        /*
                   1468:         * Clear all BLK_NOCOPY fields. Pass any block claims to other
                   1469:         * snapshots that want them (see ffs_snapblkfree below).
                   1470:         */
                   1471:        for (blkno = 1; blkno < NDADDR; blkno++) {
                   1472:                dblk = db_get(ip, blkno);
                   1473:                if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
                   1474:                        db_assign(ip, blkno, 0);
                   1475:                else if ((dblk == blkstofrags(fs, blkno) &&
                   1476:                     ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
                   1477:                     ip->i_number))) {
                   1478:                        DIP_ADD(ip, blocks, -btodb(fs->fs_bsize));
                   1479:                        db_assign(ip, blkno, 0);
                   1480:                }
                   1481:        }
                   1482:        numblks = howmany(ip->i_size, fs->fs_bsize);
                   1483:        for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1.22      yamt     1484:                error = ffs_balloc(vp, lblktosize(fs, (off_t)blkno),
1.1       hannken  1485:                    fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
                   1486:                if (error)
                   1487:                        continue;
                   1488:                if (fs->fs_size - blkno > NINDIR(fs))
                   1489:                        last = NINDIR(fs);
                   1490:                else
                   1491:                        last = fs->fs_size - blkno;
                   1492:                for (loc = 0; loc < last; loc++) {
                   1493:                        dblk = idb_get(ip, ibp->b_data, loc);
                   1494:                        if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
                   1495:                                idb_assign(ip, ibp->b_data, loc, 0);
                   1496:                        else if (dblk == blkstofrags(fs, blkno) &&
                   1497:                            ffs_snapblkfree(fs, ip->i_devvp, dblk,
                   1498:                            fs->fs_bsize, ip->i_number)) {
                   1499:                                DIP_ADD(ip, blocks, -btodb(fs->fs_bsize));
                   1500:                                idb_assign(ip, ibp->b_data, loc, 0);
                   1501:                        }
                   1502:                }
1.15      hannken  1503:                bawrite(ibp);
1.1       hannken  1504:        }
                   1505:        /*
                   1506:         * Clear snapshot flag and drop reference.
                   1507:         */
                   1508:        ip->i_flags &= ~SF_SNAPSHOT;
                   1509:        DIP_ASSIGN(ip, flags, ip->i_flags);
                   1510:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1511: }
                   1512:
                   1513: /*
                   1514:  * Notification that a block is being freed. Return zero if the free
                   1515:  * should be allowed to proceed. Return non-zero if the snapshot file
                   1516:  * wants to claim the block. The block will be claimed if it is an
                   1517:  * uncopied part of one of the snapshots. It will be freed if it is
                   1518:  * either a BLK_NOCOPY or has already been copied in all of the snapshots.
                   1519:  * If a fragment is being freed, then all snapshots that care about
                   1520:  * it must make a copy since a snapshot file can only claim full sized
                   1521:  * blocks. Note that if more than one snapshot file maps the block,
                   1522:  * we can pick one at random to claim it. Since none of the snapshots
                   1523:  * can change, we are assurred that they will all see the same unmodified
                   1524:  * image. When deleting a snapshot file (see ffs_snapremove above), we
                   1525:  * must push any of these claimed blocks to one of the other snapshots
                   1526:  * that maps it. These claimed blocks are easily identified as they will
                   1527:  * have a block number equal to their logical block number within the
                   1528:  * snapshot. A copied block can never have this property because they
                   1529:  * must always have been allocated from a BLK_NOCOPY location.
                   1530:  */
                   1531: int
1.18      thorpej  1532: ffs_snapblkfree(struct fs *fs, struct vnode *devvp, ufs2_daddr_t bno,
1.36      christos 1533:     long size, ino_t inum)
1.1       hannken  1534: {
1.47      hannken  1535:        struct mount *mp = devvp->v_specmountpoint;
1.1       hannken  1536:        struct buf *ibp;
                   1537:        struct inode *ip;
                   1538:        struct vnode *vp = NULL, *saved_vp = NULL;
1.47      hannken  1539:        struct snap_info *si;
1.43      christos 1540:        void *saved_data = NULL;
1.1       hannken  1541:        ufs_lbn_t lbn;
                   1542:        ufs2_daddr_t blkno;
1.4       hannken  1543:        int s, indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
1.1       hannken  1544:
1.47      hannken  1545:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1546:                return 0;
1.1       hannken  1547:        lbn = fragstoblks(fs, bno);
                   1548: retry:
                   1549:        VI_LOCK(devvp);
1.47      hannken  1550:        TAILQ_FOREACH(ip, &si->si_snapshots, i_nextsnap) {
1.1       hannken  1551:                vp = ITOV(ip);
                   1552:                /*
                   1553:                 * Lookup block being written.
                   1554:                 */
                   1555:                if (lbn < NDADDR) {
                   1556:                        blkno = db_get(ip, lbn);
                   1557:                } else {
                   1558:                        if (snapshot_locked == 0 &&
                   1559:                            lockmgr(vp->v_vnlock,
                   1560:                              LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
                   1561:                              VI_MTX(devvp)) != 0)
                   1562:                                goto retry;
                   1563:                        snapshot_locked = 1;
1.4       hannken  1564:                        s = cow_enter();
1.22      yamt     1565:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.1       hannken  1566:                            fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
1.4       hannken  1567:                        cow_leave(s);
1.1       hannken  1568:                        if (error)
                   1569:                                break;
                   1570:                        indiroff = (lbn - NDADDR) % NINDIR(fs);
                   1571:                        blkno = idb_get(ip, ibp->b_data, indiroff);
                   1572:                }
                   1573:                /*
                   1574:                 * Check to see if block needs to be copied.
                   1575:                 */
                   1576:                if (blkno == 0) {
                   1577:                        /*
                   1578:                         * A block that we map is being freed. If it has not
                   1579:                         * been claimed yet, we will claim or copy it (below).
                   1580:                         */
                   1581:                        claimedblk = 1;
                   1582:                } else if (blkno == BLK_SNAP) {
                   1583:                        /*
                   1584:                         * No previous snapshot claimed the block,
                   1585:                         * so it will be freed and become a BLK_NOCOPY
                   1586:                         * (don't care) for us.
                   1587:                         */
                   1588:                        if (claimedblk)
                   1589:                                panic("snapblkfree: inconsistent block type");
                   1590:                        if (snapshot_locked == 0 &&
                   1591:                            lockmgr(vp->v_vnlock,
                   1592:                              LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
                   1593:                              VI_MTX(devvp)) != 0) {
1.32      christos 1594: #if 0 /* CID-2949: dead code */
1.1       hannken  1595:                                if (lbn >= NDADDR)
                   1596:                                        brelse(ibp);
1.32      christos 1597: #endif
1.1       hannken  1598:                                vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL);
                   1599:                                goto retry;
                   1600:                        }
                   1601:                        snapshot_locked = 1;
                   1602:                        if (lbn < NDADDR) {
                   1603:                                db_assign(ip, lbn, BLK_NOCOPY);
                   1604:                                ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1605:                        } else {
                   1606:                                idb_assign(ip, ibp->b_data, indiroff,
                   1607:                                    BLK_NOCOPY);
                   1608:                                bwrite(ibp);
                   1609:                        }
                   1610:                        continue;
                   1611:                } else /* BLK_NOCOPY or default */ {
                   1612:                        /*
                   1613:                         * If the snapshot has already copied the block
                   1614:                         * (default), or does not care about the block,
                   1615:                         * it is not needed.
                   1616:                         */
1.27      bouyer   1617:                        if (lbn >= NDADDR)
1.1       hannken  1618:                                brelse(ibp);
                   1619:                        continue;
                   1620:                }
                   1621:                /*
                   1622:                 * If this is a full size block, we will just grab it
                   1623:                 * and assign it to the snapshot inode. Otherwise we
                   1624:                 * will proceed to copy it. See explanation for this
                   1625:                 * routine as to why only a single snapshot needs to
                   1626:                 * claim this block.
                   1627:                 */
                   1628:                if (snapshot_locked == 0 &&
                   1629:                    lockmgr(vp->v_vnlock,
                   1630:                      LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT,
                   1631:                      VI_MTX(devvp)) != 0) {
                   1632:                        vn_lock(vp, LK_EXCLUSIVE | LK_SLEEPFAIL);
                   1633:                        goto retry;
                   1634:                }
                   1635:                snapshot_locked = 1;
                   1636:                if (size == fs->fs_bsize) {
                   1637: #ifdef DEBUG
                   1638:                        if (snapdebug)
1.19      christos 1639:                                printf("%s %llu lbn %" PRId64
                   1640:                                    "from inum %llu\n",
                   1641:                                    "Grabonremove: snapino",
                   1642:                                    (unsigned long long)ip->i_number,
                   1643:                                    lbn, (unsigned long long)inum);
1.1       hannken  1644: #endif
                   1645:                        if (lbn < NDADDR) {
                   1646:                                db_assign(ip, lbn, bno);
                   1647:                        } else {
                   1648:                                idb_assign(ip, ibp->b_data, indiroff, bno);
                   1649:                                bwrite(ibp);
                   1650:                        }
                   1651:                        DIP_ADD(ip, blocks, btodb(size));
                   1652:                        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1653:                        VOP_UNLOCK(vp, 0);
                   1654:                        return (1);
                   1655:                }
                   1656:                if (lbn >= NDADDR)
                   1657:                        brelse(ibp);
                   1658: #ifdef DEBUG
                   1659:                if (snapdebug)
1.19      christos 1660:                        printf("%s%llu lbn %" PRId64 " %s %llu size %ld\n",
                   1661:                            "Copyonremove: snapino ",
                   1662:                            (unsigned long long)ip->i_number,
                   1663:                            lbn, "for inum", (unsigned long long)inum, size);
1.1       hannken  1664: #endif
                   1665:                /*
                   1666:                 * If we have already read the old block contents, then
                   1667:                 * simply copy them to the new block. Note that we need
                   1668:                 * to synchronously write snapshots that have not been
                   1669:                 * unlinked, and hence will be visible after a crash,
                   1670:                 * to ensure their integrity.
                   1671:                 */
                   1672:                if (saved_data) {
                   1673:                        error = writevnblk(vp, saved_data, lbn);
                   1674:                        if (error)
                   1675:                                break;
                   1676:                        continue;
                   1677:                }
                   1678:                /*
                   1679:                 * Otherwise, read the old block contents into the buffer.
                   1680:                 */
                   1681:                saved_data = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
                   1682:                saved_vp = vp;
                   1683:                if ((error = readfsblk(vp, saved_data, lbn)) != 0) {
                   1684:                        free(saved_data, M_UFSMNT);
                   1685:                        saved_data = NULL;
                   1686:                        break;
                   1687:                }
                   1688:        }
                   1689:        /*
                   1690:         * Note that we need to synchronously write snapshots that
                   1691:         * have not been unlinked, and hence will be visible after
                   1692:         * a crash, to ensure their integrity.
                   1693:         */
                   1694:        if (saved_data) {
                   1695:                error = writevnblk(saved_vp, saved_data, lbn);
                   1696:                free(saved_data, M_UFSMNT);
                   1697:        }
                   1698:        /*
                   1699:         * If we have been unable to allocate a block in which to do
                   1700:         * the copy, then return non-zero so that the fragment will
                   1701:         * not be freed. Although space will be lost, the snapshot
                   1702:         * will stay consistent.
                   1703:         */
                   1704:        if (snapshot_locked)
                   1705:                VOP_UNLOCK(vp, 0);
                   1706:        else
                   1707:                VI_UNLOCK(devvp);
                   1708:        return (error);
                   1709: }
                   1710:
                   1711: /*
                   1712:  * Associate snapshot files when mounting.
                   1713:  */
                   1714: void
1.18      thorpej  1715: ffs_snapshot_mount(struct mount *mp)
1.1       hannken  1716: {
1.47      hannken  1717:        struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
                   1718:        struct fs *fs = VFSTOUFS(mp)->um_fs;
1.31      ad       1719:        struct lwp *l = curlwp;
1.1       hannken  1720:        struct vnode *vp;
                   1721:        struct inode *ip, *xp;
1.47      hannken  1722:        struct snap_info *si;
1.1       hannken  1723:        ufs2_daddr_t snaplistsize, *snapblklist;
1.15      hannken  1724:        int i, error, ns, snaploc, loc;
1.1       hannken  1725:
1.46      hannken  1726:        /*
                   1727:         * No persistent snapshots on apple ufs file systems.
                   1728:         */
1.47      hannken  1729:        if (UFS_MPISAPPLEUFS(VFSTOUFS(mp)))
1.46      hannken  1730:                return;
                   1731:
1.47      hannken  1732:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1733:                si = si_mount_init(mp);
1.1       hannken  1734:        ns = UFS_FSNEEDSWAP(fs);
                   1735:        /*
1.22      yamt     1736:         * XXX The following needs to be set before ffs_truncate or
1.1       hannken  1737:         * VOP_READ can be called.
                   1738:         */
                   1739:        mp->mnt_stat.f_iosize = fs->fs_bsize;
                   1740:        /*
                   1741:         * Process each snapshot listed in the superblock.
                   1742:         */
                   1743:        vp = NULL;
                   1744:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
                   1745:                if (fs->fs_snapinum[snaploc] == 0)
                   1746:                        break;
                   1747:                if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc],
                   1748:                    &vp)) != 0) {
                   1749:                        printf("ffs_snapshot_mount: vget failed %d\n", error);
                   1750:                        continue;
                   1751:                }
                   1752:                ip = VTOI(vp);
                   1753:                if ((ip->i_flags & SF_SNAPSHOT) == 0) {
                   1754:                        printf("ffs_snapshot_mount: non-snapshot inode %d\n",
                   1755:                            fs->fs_snapinum[snaploc]);
                   1756:                        vput(vp);
                   1757:                        vp = NULL;
                   1758:                        for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
                   1759:                                if (fs->fs_snapinum[loc] == 0)
                   1760:                                        break;
                   1761:                                fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
                   1762:                        }
                   1763:                        fs->fs_snapinum[loc - 1] = 0;
                   1764:                        snaploc--;
                   1765:                        continue;
                   1766:                }
1.15      hannken  1767:
                   1768:                /*
                   1769:                 * Read the block hints list. Use an empty list on
                   1770:                 * read errors.
                   1771:                 */
                   1772:                error = vn_rdwr(UIO_READ, vp,
1.43      christos 1773:                    (void *)&snaplistsize, sizeof(snaplistsize),
1.15      hannken  1774:                    lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)),
                   1775:                    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT,
1.31      ad       1776:                    l->l_cred, NULL, NULL);
1.15      hannken  1777:                if (error) {
                   1778:                        printf("ffs_snapshot_mount: read_1 failed %d\n", error);
                   1779:                        snaplistsize = 1;
                   1780:                } else
                   1781:                        snaplistsize = ufs_rw64(snaplistsize, ns);
1.25      christos 1782:                snapblklist = malloc(
1.15      hannken  1783:                    snaplistsize * sizeof(ufs2_daddr_t), M_UFSMNT, M_WAITOK);
                   1784:                if (error)
                   1785:                        snapblklist[0] = 1;
                   1786:                else {
1.43      christos 1787:                        error = vn_rdwr(UIO_READ, vp, (void *)snapblklist,
1.15      hannken  1788:                            snaplistsize * sizeof(ufs2_daddr_t),
                   1789:                            lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)),
                   1790:                            UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT,
1.31      ad       1791:                            l->l_cred, NULL, NULL);
1.15      hannken  1792:                        for (i = 0; i < snaplistsize; i++)
                   1793:                                snapblklist[i] = ufs_rw64(snapblklist[i], ns);
                   1794:                        if (error) {
                   1795:                                printf("ffs_snapshot_mount: read_2 failed %d\n",
                   1796:                                    error);
                   1797:                                snapblklist[0] = 1;
                   1798:                        }
                   1799:                }
                   1800:                ip->i_snapblklist = &snapblklist[0];
                   1801:
1.1       hannken  1802:                /*
                   1803:                 * If there already exist snapshots on this filesystem, grab a
                   1804:                 * reference to their shared lock. If this is the first snapshot
                   1805:                 * on this filesystem, we need to allocate a lock for the
                   1806:                 * snapshots to share. In either case, acquire the snapshot
                   1807:                 * lock and give up our original private lock.
                   1808:                 */
                   1809:                VI_LOCK(devvp);
1.47      hannken  1810:                if ((xp = TAILQ_FIRST(&si->si_snapshots)) != NULL) {
1.1       hannken  1811:                        struct lock *lkp;
                   1812:
                   1813:                        lkp = ITOV(xp)->v_vnlock;
                   1814:                        VI_UNLOCK(devvp);
                   1815:                        VI_LOCK(vp);
                   1816:                        vp->v_vnlock = lkp;
                   1817:                } else {
                   1818:                        struct lock *lkp;
                   1819:
                   1820:                        VI_UNLOCK(devvp);
                   1821:                        MALLOC(lkp, struct lock *, sizeof(struct lock),
                   1822:                            M_UFSMNT, M_WAITOK);
                   1823:                        lockinit(lkp, PVFS, "snaplk", 0, LK_CANRECURSE);
                   1824:                        VI_LOCK(vp);
                   1825:                        vp->v_vnlock = lkp;
                   1826:                }
                   1827:                vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY);
                   1828:                transferlockers(&vp->v_lock, vp->v_vnlock);
                   1829:                lockmgr(&vp->v_lock, LK_RELEASE, NULL);
                   1830:                /*
                   1831:                 * Link it onto the active snapshot list.
                   1832:                 */
                   1833:                VI_LOCK(devvp);
                   1834:                if (ip->i_nextsnap.tqe_prev != 0)
1.19      christos 1835:                        panic("ffs_snapshot_mount: %llu already on list",
                   1836:                            (unsigned long long)ip->i_number);
1.1       hannken  1837:                else
1.47      hannken  1838:                        TAILQ_INSERT_TAIL(&si->si_snapshots, ip, i_nextsnap);
1.1       hannken  1839:                vp->v_flag |= VSYSTEM;
                   1840:                VI_UNLOCK(devvp);
                   1841:                VOP_UNLOCK(vp, 0);
                   1842:        }
                   1843:        /*
                   1844:         * No usable snapshots found.
                   1845:         */
                   1846:        if (vp == NULL)
                   1847:                return;
                   1848:        /*
1.15      hannken  1849:         * Attach the block hints list. We always want to
1.1       hannken  1850:         * use the list from the newest snapshot.
1.15      hannken  1851:        */
1.47      hannken  1852:        xp = TAILQ_LAST(&si->si_snapshots, inodelst);
1.1       hannken  1853:        VI_LOCK(devvp);
1.47      hannken  1854:        si->si_snapblklist = xp->i_snapblklist;
1.1       hannken  1855:        VI_UNLOCK(devvp);
                   1856:        vn_cow_establish(devvp, ffs_copyonwrite, devvp);
                   1857: }
                   1858:
                   1859: /*
                   1860:  * Disassociate snapshot files when unmounting.
                   1861:  */
                   1862: void
1.18      thorpej  1863: ffs_snapshot_unmount(struct mount *mp)
1.1       hannken  1864: {
1.47      hannken  1865:        struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
1.1       hannken  1866:        struct lock *lkp = NULL;
                   1867:        struct inode *xp;
                   1868:        struct vnode *vp;
1.47      hannken  1869:        struct snap_info *si;
1.1       hannken  1870:
1.47      hannken  1871:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1872:                return;
1.1       hannken  1873:        VI_LOCK(devvp);
1.47      hannken  1874:        while ((xp = TAILQ_FIRST(&si->si_snapshots)) != 0) {
1.1       hannken  1875:                vp = ITOV(xp);
                   1876:                lkp = vp->v_vnlock;
                   1877:                vp->v_vnlock = &vp->v_lock;
1.47      hannken  1878:                TAILQ_REMOVE(&si->si_snapshots, xp, i_nextsnap);
1.1       hannken  1879:                xp->i_nextsnap.tqe_prev = 0;
1.47      hannken  1880:                if (xp->i_snapblklist == si->si_snapblklist)
                   1881:                        si->si_snapblklist = NULL;
1.15      hannken  1882:                VI_UNLOCK(devvp);
                   1883:                FREE(xp->i_snapblklist, M_UFSMNT);
                   1884:                if (xp->i_ffs_effnlink > 0)
1.1       hannken  1885:                        vrele(vp);
1.15      hannken  1886:                VI_LOCK(devvp);
1.1       hannken  1887:        }
                   1888:        VI_UNLOCK(devvp);
                   1889:        if (lkp != NULL) {
                   1890:                vn_cow_disestablish(devvp, ffs_copyonwrite, devvp);
                   1891:                FREE(lkp, M_UFSMNT);
                   1892:        }
                   1893: }
                   1894:
                   1895: /*
                   1896:  * Check for need to copy block that is about to be written,
                   1897:  * copying the block if necessary.
                   1898:  */
                   1899: static int
1.18      thorpej  1900: ffs_copyonwrite(void *v, struct buf *bp)
1.1       hannken  1901: {
                   1902:        struct buf *ibp;
                   1903:        struct fs *fs;
                   1904:        struct inode *ip;
                   1905:        struct vnode *devvp = v, *vp = 0, *saved_vp = NULL;
1.47      hannken  1906:        struct mount *mp = devvp->v_specmountpoint;
                   1907:        struct snap_info *si;
1.43      christos 1908:        void *saved_data = NULL;
1.1       hannken  1909:        ufs2_daddr_t lbn, blkno, *snapblklist;
1.4       hannken  1910:        int lower, upper, mid, s, ns, indiroff, snapshot_locked = 0, error = 0;
1.1       hannken  1911:
                   1912:        /*
                   1913:         * Check for valid snapshots.
                   1914:         */
1.47      hannken  1915:        if ((si = mount_getspecific(mp, si_mount_data_key)) == NULL)
                   1916:                return 0;
1.1       hannken  1917:        VI_LOCK(devvp);
1.47      hannken  1918:        ip = TAILQ_FIRST(&si->si_snapshots);
1.1       hannken  1919:        if (ip == NULL) {
                   1920:                VI_UNLOCK(devvp);
1.11      perry    1921:                return 0;
1.1       hannken  1922:        }
                   1923:        /*
                   1924:         * First check to see if it is in the preallocated list.
                   1925:         * By doing this check we avoid several potential deadlocks.
                   1926:         */
                   1927:        fs = ip->i_fs;
                   1928:        ns = UFS_FSNEEDSWAP(fs);
                   1929:        lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
1.47      hannken  1930:        snapblklist = si->si_snapblklist;
                   1931:        upper = si->si_snapblklist[0] - 1;
1.1       hannken  1932:        lower = 1;
                   1933:        while (lower <= upper) {
                   1934:                mid = (lower + upper) / 2;
1.15      hannken  1935:                if (snapblklist[mid] == lbn)
1.1       hannken  1936:                        break;
1.15      hannken  1937:                if (snapblklist[mid] < lbn)
1.1       hannken  1938:                        lower = mid + 1;
                   1939:                else
                   1940:                        upper = mid - 1;
                   1941:        }
                   1942:        if (lower <= upper) {
                   1943:                VI_UNLOCK(devvp);
                   1944:                return 0;
                   1945:        }
                   1946:        /*
                   1947:         * Not in the precomputed list, so check the snapshots.
                   1948:         */
                   1949: retry:
1.47      hannken  1950:        TAILQ_FOREACH(ip, &si->si_snapshots, i_nextsnap) {
1.1       hannken  1951:                vp = ITOV(ip);
                   1952:                /*
                   1953:                 * We ensure that everything of our own that needs to be
                   1954:                 * copied will be done at the time that ffs_snapshot is
                   1955:                 * called. Thus we can skip the check here which can
1.22      yamt     1956:                 * deadlock in doing the lookup in ffs_balloc.
1.1       hannken  1957:                 */
                   1958:                if (bp->b_vp == vp)
                   1959:                        continue;
                   1960:                /*
                   1961:                 * Check to see if block needs to be copied. We do not have
                   1962:                 * to hold the snapshot lock while doing this lookup as it
                   1963:                 * will never require any additional allocations for the
                   1964:                 * snapshot inode.
                   1965:                 */
                   1966:                if (lbn < NDADDR) {
                   1967:                        blkno = db_get(ip, lbn);
                   1968:                } else {
                   1969:                        if (snapshot_locked == 0 &&
                   1970:                            lockmgr(vp->v_vnlock,
                   1971:                              LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
                   1972:                              VI_MTX(devvp)) != 0) {
                   1973:                                VI_LOCK(devvp);
                   1974:                                goto retry;
                   1975:                        }
                   1976:                        snapshot_locked = 1;
1.4       hannken  1977:                        s = cow_enter();
1.22      yamt     1978:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.1       hannken  1979:                           fs->fs_bsize, KERNCRED, B_METAONLY, &ibp);
1.4       hannken  1980:                        cow_leave(s);
1.1       hannken  1981:                        if (error)
                   1982:                                break;
                   1983:                        indiroff = (lbn - NDADDR) % NINDIR(fs);
                   1984:                        blkno = idb_get(ip, ibp->b_data, indiroff);
                   1985:                        brelse(ibp);
                   1986:                }
                   1987: #ifdef DIAGNOSTIC
                   1988:                if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
                   1989:                        panic("ffs_copyonwrite: bad copy block");
                   1990: #endif
                   1991:                if (blkno != 0)
                   1992:                        continue;
1.4       hannken  1993: #ifdef DIAGNOSTIC
1.41      ad       1994:                if (curlwp->l_pflag & LP_UFSCOW)
1.4       hannken  1995:                        printf("ffs_copyonwrite: recursive call\n");
                   1996: #endif
1.1       hannken  1997:                /*
                   1998:                 * Allocate the block into which to do the copy. Since
                   1999:                 * multiple processes may all try to copy the same block,
                   2000:                 * we have to recheck our need to do a copy if we sleep
                   2001:                 * waiting for the lock.
                   2002:                 *
                   2003:                 * Because all snapshots on a filesystem share a single
                   2004:                 * lock, we ensure that we will never be in competition
                   2005:                 * with another process to allocate a block.
                   2006:                 */
                   2007:                if (snapshot_locked == 0 &&
                   2008:                    lockmgr(vp->v_vnlock,
                   2009:                      LK_INTERLOCK | LK_EXCLUSIVE | LK_SLEEPFAIL,
                   2010:                      VI_MTX(devvp)) != 0) {
                   2011:                        VI_LOCK(devvp);
                   2012:                        goto retry;
                   2013:                }
                   2014:                snapshot_locked = 1;
                   2015: #ifdef DEBUG
                   2016:                if (snapdebug) {
1.19      christos 2017:                        printf("Copyonwrite: snapino %llu lbn %" PRId64 " for ",
                   2018:                            (unsigned long long)ip->i_number, lbn);
1.1       hannken  2019:                        if (bp->b_vp == devvp)
                   2020:                                printf("fs metadata");
                   2021:                        else
1.19      christos 2022:                                printf("inum %llu", (unsigned long long)
                   2023:                                    VTOI(bp->b_vp)->i_number);
1.4       hannken  2024:                        printf(" lblkno %" PRId64 "\n", bp->b_lblkno);
1.1       hannken  2025:                }
                   2026: #endif
                   2027:                /*
                   2028:                 * If we have already read the old block contents, then
                   2029:                 * simply copy them to the new block. Note that we need
                   2030:                 * to synchronously write snapshots that have not been
                   2031:                 * unlinked, and hence will be visible after a crash,
                   2032:                 * to ensure their integrity.
                   2033:                 */
                   2034:                if (saved_data) {
                   2035:                        error = writevnblk(vp, saved_data, lbn);
                   2036:                        if (error)
                   2037:                                break;
                   2038:                        continue;
                   2039:                }
                   2040:                /*
                   2041:                 * Otherwise, read the old block contents into the buffer.
                   2042:                 */
                   2043:                saved_data = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
                   2044:                saved_vp = vp;
                   2045:                if ((error = readfsblk(vp, saved_data, lbn)) != 0) {
                   2046:                        free(saved_data, M_UFSMNT);
                   2047:                        saved_data = NULL;
                   2048:                        break;
                   2049:                }
                   2050:        }
                   2051:        /*
                   2052:         * Note that we need to synchronously write snapshots that
                   2053:         * have not been unlinked, and hence will be visible after
                   2054:         * a crash, to ensure their integrity.
                   2055:         */
                   2056:        if (saved_data) {
                   2057:                error = writevnblk(saved_vp, saved_data, lbn);
                   2058:                free(saved_data, M_UFSMNT);
                   2059:        }
                   2060:        if (snapshot_locked)
                   2061:                VOP_UNLOCK(vp, 0);
                   2062:        else
                   2063:                VI_UNLOCK(devvp);
                   2064:        return error;
                   2065: }
                   2066:
                   2067: /*
                   2068:  * Read the specified block from disk. Vp is usually a snapshot vnode.
                   2069:  */
                   2070: static int
1.43      christos 2071: readfsblk(struct vnode *vp, void *data, ufs2_daddr_t lbn)
1.1       hannken  2072: {
1.24      yamt     2073:        int error;
1.1       hannken  2074:        struct inode *ip = VTOI(vp);
                   2075:        struct fs *fs = ip->i_fs;
                   2076:        struct buf *nbp;
                   2077:
1.24      yamt     2078:        nbp = getiobuf();
1.1       hannken  2079:        nbp->b_flags = B_READ;
                   2080:        nbp->b_bcount = nbp->b_bufsize = fs->fs_bsize;
                   2081:        nbp->b_error = 0;
                   2082:        nbp->b_data = data;
                   2083:        nbp->b_blkno = nbp->b_rawblkno = fsbtodb(fs, blkstofrags(fs, lbn));
                   2084:        nbp->b_proc = NULL;
                   2085:        nbp->b_dev = ip->i_devvp->v_rdev;
                   2086:        nbp->b_vp = NULLVP;
                   2087:
1.44      ad       2088:        bdev_strategy(nbp);
1.1       hannken  2089:
                   2090:        error = biowait(nbp);
                   2091:
1.24      yamt     2092:        putiobuf(nbp);
1.1       hannken  2093:
                   2094:        return error;
                   2095: }
                   2096:
1.37      christos 2097: #if !defined(FFS_NO_SNAPSHOT)
1.1       hannken  2098: /*
1.4       hannken  2099:  * Read the specified block. Bypass UBC to prevent deadlocks.
1.1       hannken  2100:  */
                   2101: static int
1.43      christos 2102: readvnblk(struct vnode *vp, void *data, ufs2_daddr_t lbn)
1.1       hannken  2103: {
1.4       hannken  2104:        int error;
                   2105:        daddr_t bn;
                   2106:        off_t offset;
1.1       hannken  2107:        struct inode *ip = VTOI(vp);
                   2108:        struct fs *fs = ip->i_fs;
                   2109:
1.4       hannken  2110:        error = VOP_BMAP(vp, lbn, NULL, &bn, NULL);
                   2111:        if (error)
                   2112:                return error;
1.1       hannken  2113:
1.4       hannken  2114:        if (bn != (daddr_t)-1) {
                   2115:                offset = dbtob(bn);
                   2116:                simple_lock(&vp->v_interlock);
                   2117:                error = VOP_PUTPAGES(vp, trunc_page(offset),
                   2118:                    round_page(offset+fs->fs_bsize),
                   2119:                    PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
1.1       hannken  2120:                if (error)
1.4       hannken  2121:                        return error;
                   2122:
                   2123:                return readfsblk(vp, data, fragstoblks(fs, dbtofsb(fs, bn)));
1.1       hannken  2124:        }
                   2125:
1.4       hannken  2126:        bzero(data, fs->fs_bsize);
                   2127:
                   2128:        return 0;
1.1       hannken  2129: }
1.37      christos 2130: #endif /* !defined(FFS_NO_SNAPSHOT) */
1.1       hannken  2131:
                   2132: /*
1.4       hannken  2133:  * Write the specified block. Bypass UBC to prevent deadlocks.
1.1       hannken  2134:  */
                   2135: static int
1.43      christos 2136: writevnblk(struct vnode *vp, void *data, ufs2_daddr_t lbn)
1.1       hannken  2137: {
1.4       hannken  2138:        int s, error;
                   2139:        off_t offset;
                   2140:        struct buf *bp;
1.1       hannken  2141:        struct inode *ip = VTOI(vp);
                   2142:        struct fs *fs = ip->i_fs;
                   2143:
1.4       hannken  2144:        offset = lblktosize(fs, (off_t)lbn);
                   2145:        s = cow_enter();
                   2146:        simple_lock(&vp->v_interlock);
                   2147:        error = VOP_PUTPAGES(vp, trunc_page(offset),
                   2148:            round_page(offset+fs->fs_bsize), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE);
                   2149:        if (error == 0)
1.22      yamt     2150:                error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.4       hannken  2151:                    fs->fs_bsize, KERNCRED, B_SYNC, &bp);
                   2152:        cow_leave(s);
1.1       hannken  2153:        if (error)
                   2154:                return error;
                   2155:
1.4       hannken  2156:        bcopy(data, bp->b_data, fs->fs_bsize);
                   2157:        bp->b_flags |= B_NOCACHE;
                   2158:
                   2159:        return bwrite(bp);
                   2160: }
                   2161:
                   2162: /*
1.41      ad       2163:  * Set/reset lwp's LP_UFSCOW flag.
1.4       hannken  2164:  * May be called recursive.
                   2165:  */
                   2166: static inline int
                   2167: cow_enter(void)
                   2168: {
                   2169:        struct lwp *l = curlwp;
                   2170:
1.41      ad       2171:        if (l->l_pflag & LP_UFSCOW) {
1.4       hannken  2172:                return 0;
                   2173:        } else {
1.41      ad       2174:                l->l_pflag |= LP_UFSCOW;
                   2175:                return LP_UFSCOW;
1.1       hannken  2176:        }
1.4       hannken  2177: }
                   2178:
                   2179: static inline void
                   2180: cow_leave(int flag)
                   2181: {
                   2182:        struct lwp *l = curlwp;
1.1       hannken  2183:
1.41      ad       2184:        l->l_pflag &= ~flag;
1.1       hannken  2185: }
                   2186:
                   2187: /*
                   2188:  * Get/Put direct block from inode or buffer containing disk addresses. Take
                   2189:  * care for fs type (UFS1/UFS2) and byte swapping. These functions should go
                   2190:  * into a global include.
                   2191:  */
                   2192: static inline ufs2_daddr_t
                   2193: db_get(struct inode *ip, int loc)
                   2194: {
                   2195:        if (ip->i_ump->um_fstype == UFS1)
1.2       hannken  2196:                return ufs_rw32(ip->i_ffs1_db[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2197:        else
1.2       hannken  2198:                return ufs_rw64(ip->i_ffs2_db[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2199: }
                   2200:
                   2201: static inline void
                   2202: db_assign(struct inode *ip, int loc, ufs2_daddr_t val)
                   2203: {
                   2204:        if (ip->i_ump->um_fstype == UFS1)
1.2       hannken  2205:                ip->i_ffs1_db[loc] = ufs_rw32(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2206:        else
1.2       hannken  2207:                ip->i_ffs2_db[loc] = ufs_rw64(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2208: }
                   2209:
                   2210: static inline ufs2_daddr_t
1.43      christos 2211: idb_get(struct inode *ip, void *bf, int loc)
1.1       hannken  2212: {
                   2213:        if (ip->i_ump->um_fstype == UFS1)
1.17      christos 2214:                return ufs_rw32(((ufs1_daddr_t *)(bf))[loc],
1.2       hannken  2215:                    UFS_IPNEEDSWAP(ip));
1.1       hannken  2216:        else
1.17      christos 2217:                return ufs_rw64(((ufs2_daddr_t *)(bf))[loc],
1.2       hannken  2218:                    UFS_IPNEEDSWAP(ip));
1.1       hannken  2219: }
                   2220:
                   2221: static inline void
1.43      christos 2222: idb_assign(struct inode *ip, void *bf, int loc, ufs2_daddr_t val)
1.1       hannken  2223: {
                   2224:        if (ip->i_ump->um_fstype == UFS1)
1.17      christos 2225:                ((ufs1_daddr_t *)(bf))[loc] =
1.2       hannken  2226:                    ufs_rw32(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2227:        else
1.17      christos 2228:                ((ufs2_daddr_t *)(bf))[loc] =
1.2       hannken  2229:                    ufs_rw64(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2230: }

CVSweb <webmaster@jp.NetBSD.org>