[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 1.87

1.87    ! hannken     1: /*     $NetBSD: ffs_snapshot.c,v 1.86 2008/12/07 18:55:58 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>
1.87    ! hannken    41: __KERNEL_RCSID(0, "$NetBSD: ffs_snapshot.c,v 1.86 2008/12/07 18:55:58 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>
1.75      hannken    63: #include <sys/wapbl.h>
1.1       hannken    64:
                     65: #include <miscfs/specfs/specdev.h>
                     66:
                     67: #include <ufs/ufs/quota.h>
                     68: #include <ufs/ufs/ufsmount.h>
                     69: #include <ufs/ufs/inode.h>
                     70: #include <ufs/ufs/ufs_extern.h>
                     71: #include <ufs/ufs/ufs_bswap.h>
1.75      hannken    72: #include <ufs/ufs/ufs_wapbl.h>
1.1       hannken    73:
                     74: #include <ufs/ffs/fs.h>
                     75: #include <ufs/ffs/ffs_extern.h>
                     76:
1.72      hannken    77: #include <uvm/uvm.h>
                     78:
1.10      hannken    79: #if !defined(FFS_NO_SNAPSHOT)
1.76      hannken    80: typedef int (*acctfunc_t)
                     81:     (struct vnode *, void *, int, int, struct fs *, daddr_t, int);
                     82:
1.79      hannken    83: static int snapshot_setup(struct mount *, struct vnode *);
                     84: static int snapshot_copyfs(struct mount *, struct vnode *, void **);
                     85: static int snapshot_expunge(struct mount *, struct vnode *,
                     86:     struct fs *, daddr_t *, daddr_t **);
                     87: static int snapshot_expunge_snap(struct mount *, struct vnode *,
                     88:     struct fs *, daddr_t);
                     89: static int snapshot_writefs(struct mount *, struct vnode *, void *);
                     90: static int cgaccount(struct vnode *, int, int *);
                     91: static int cgaccount1(int, struct vnode *, void *, int);
1.76      hannken    92: static int expunge(struct vnode *, struct inode *, struct fs *,
                     93:     acctfunc_t, int);
                     94: static int indiracct(struct vnode *, struct vnode *, int, daddr_t,
                     95:     daddr_t, daddr_t, daddr_t, daddr_t, struct fs *, acctfunc_t, int);
                     96: static int fullacct(struct vnode *, void *, int, int, struct fs *,
                     97:     daddr_t, int);
                     98: static int snapacct(struct vnode *, void *, int, int, struct fs *,
                     99:     daddr_t, int);
                    100: static int mapacct(struct vnode *, void *, int, int, struct fs *,
                    101:     daddr_t, int);
1.10      hannken   102: #endif /* !defined(FFS_NO_SNAPSHOT) */
                    103:
1.55      hannken   104: static int ffs_copyonwrite(void *, struct buf *, bool);
1.72      hannken   105: static int snapblkaddr(struct vnode *, daddr_t, daddr_t *);
1.79      hannken   106: static int rwfsblk(struct vnode *, int, void *, daddr_t);
1.75      hannken   107: static int syncsnap(struct vnode *);
1.79      hannken   108: static int wrsnapblk(struct vnode *, void *, daddr_t);
1.76      hannken   109:
1.79      hannken   110: static inline daddr_t db_get(struct inode *, int);
                    111: static inline void db_assign(struct inode *, int, daddr_t);
                    112: static inline daddr_t ib_get(struct inode *, int);
                    113: static inline void ib_assign(struct inode *, int, daddr_t);
                    114: static inline daddr_t idb_get(struct inode *, void *, int);
                    115: static inline void idb_assign(struct inode *, void *, int, daddr_t);
1.1       hannken   116:
1.47      hannken   117: struct snap_info {
1.49      hannken   118:        kmutex_t si_lock;                       /* Lock this snapinfo */
1.74      hannken   119:        kmutex_t si_snaplock;                   /* Snapshot vnode common lock */
1.47      hannken   120:        TAILQ_HEAD(inodelst, inode) si_snapshots; /* List of active snapshots */
                    121:        daddr_t *si_snapblklist;                /* Snapshot block hints list */
1.49      hannken   122:        uint32_t si_gen;                        /* Incremented on change */
1.47      hannken   123: };
                    124:
1.1       hannken   125: #ifdef DEBUG
                    126: static int snapdebug = 0;
                    127: #endif
1.47      hannken   128:
1.66      hannken   129: int
                    130: ffs_snapshot_init(struct ufsmount *ump)
1.47      hannken   131: {
1.66      hannken   132:        struct snap_info *si;
                    133:
                    134:        si = ump->um_snapinfo = kmem_alloc(sizeof(*si), KM_SLEEP);
                    135:        if (si == NULL)
                    136:                return ENOMEM;
                    137:
                    138:        TAILQ_INIT(&si->si_snapshots);
                    139:        mutex_init(&si->si_lock, MUTEX_DEFAULT, IPL_NONE);
1.74      hannken   140:        mutex_init(&si->si_snaplock, MUTEX_DEFAULT, IPL_NONE);
1.66      hannken   141:        si->si_gen = 0;
                    142:        si->si_snapblklist = NULL;
1.47      hannken   143:
1.66      hannken   144:        return 0;
1.47      hannken   145: }
                    146:
                    147: void
1.66      hannken   148: ffs_snapshot_fini(struct ufsmount *ump)
1.47      hannken   149: {
1.66      hannken   150:        struct snap_info *si;
1.47      hannken   151:
1.66      hannken   152:        si = ump->um_snapinfo;
                    153:        ump->um_snapinfo = NULL;
1.47      hannken   154:
                    155:        KASSERT(TAILQ_EMPTY(&si->si_snapshots));
1.49      hannken   156:        mutex_destroy(&si->si_lock);
1.74      hannken   157:        mutex_destroy(&si->si_snaplock);
1.47      hannken   158:        KASSERT(si->si_snapblklist == NULL);
1.63      ad        159:        kmem_free(si, sizeof(*si));
1.47      hannken   160: }
                    161:
1.1       hannken   162: /*
                    163:  * Create a snapshot file and initialize it for the filesystem.
1.4       hannken   164:  * Vnode is locked on entry and return.
1.1       hannken   165:  */
                    166: int
1.79      hannken   167: ffs_snapshot(struct mount *mp, struct vnode *vp, struct timespec *ctime)
1.1       hannken   168: {
1.10      hannken   169: #if defined(FFS_NO_SNAPSHOT)
                    170:        return EOPNOTSUPP;
                    171: }
                    172: #else /* defined(FFS_NO_SNAPSHOT) */
1.79      hannken   173:        bool suspended = false;
                    174:        bool snapshot_locked = false;
                    175:        int error, redo = 0, snaploc;
                    176:        void *sbbuf = NULL;
                    177:        daddr_t *snaplist = NULL, snaplistsize = 0;
                    178:        struct buf *bp, *nbp;
                    179:        struct fs *copy_fs, *fs = VFSTOUFS(mp)->um_fs;
                    180:        struct inode *ip = VTOI(vp);
                    181:        struct lwp *l = curlwp;
                    182:        struct snap_info *si = VFSTOUFS(mp)->um_snapinfo;
                    183:        struct timespec ts;
1.1       hannken   184:        struct timeval starttime;
                    185: #ifdef DEBUG
                    186:        struct timeval endtime;
                    187: #endif
1.79      hannken   188:        struct vnode *devvp = ip->i_devvp;
1.1       hannken   189:
                    190:        /*
                    191:         * If the vnode already is a snapshot, return.
                    192:         */
                    193:        if (VTOI(vp)->i_flags & SF_SNAPSHOT) {
                    194:                if (ctime) {
                    195:                        ctime->tv_sec = DIP(VTOI(vp), mtime);
                    196:                        ctime->tv_nsec = DIP(VTOI(vp), mtimensec);
                    197:                }
                    198:                return 0;
                    199:        }
                    200:        /*
1.79      hannken   201:         * Check for free snapshot slot in the superblock.
                    202:         */
                    203:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
                    204:                if (fs->fs_snapinum[snaploc] == 0)
                    205:                        break;
                    206:        if (snaploc == FSMAXSNAP)
                    207:                return (ENOSPC);
                    208:        /*
                    209:         * Prepare the vnode to become a snapshot.
                    210:         */
                    211:        error = snapshot_setup(mp, vp);
                    212:        if (error)
                    213:                goto out;
                    214:        /*
                    215:         * Change inode to snapshot type file.
                    216:         */
                    217:        ip->i_flags |= SF_SNAPSHOT;
                    218:        DIP_ASSIGN(ip, flags, ip->i_flags);
                    219:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                    220:        /*
                    221:         * Copy all the cylinder group maps. Although the
                    222:         * filesystem is still active, we hope that only a few
                    223:         * cylinder groups will change between now and when we
                    224:         * suspend operations. Thus, we will be able to quickly
                    225:         * touch up the few cylinder groups that changed during
                    226:         * the suspension period.
                    227:         */
                    228:        error = cgaccount(vp, 1, NULL);
                    229:        if (error)
                    230:                goto out;
                    231:        /*
                    232:         * Ensure that the snapshot is completely on disk.
                    233:         * Since we have marked it as a snapshot it is safe to
                    234:         * unlock it as no process will be allowed to write to it.
                    235:         */
                    236:        error = VOP_FSYNC(vp, l->l_cred, FSYNC_WAIT, 0, 0);
                    237:        if (error)
                    238:                goto out;
                    239:        VOP_UNLOCK(vp, 0);
                    240:        /*
                    241:         * All allocations are done, so we can now suspend the filesystem.
                    242:         */
                    243:        error = vfs_suspend(vp->v_mount, 0);
                    244:        vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
                    245:        if (error)
                    246:                goto out;
                    247:        suspended = true;
                    248:        getmicrotime(&starttime);
                    249:        /*
                    250:         * First, copy all the cylinder group maps that have changed.
                    251:         */
                    252:        error = cgaccount(vp, 2, &redo);
                    253:        if (error)
                    254:                goto out;
                    255:        /*
                    256:         * Create a copy of the superblock and its summary information.
                    257:         */
                    258:        error = snapshot_copyfs(mp, vp, &sbbuf);
                    259:        copy_fs = (struct fs *)((char *)sbbuf + blkoff(fs, fs->fs_sblockloc));
                    260:        if (error)
                    261:                goto out;
                    262:        /*
                    263:         * Expunge unlinked files from our view.
                    264:         */
                    265:        error = snapshot_expunge(mp, vp, copy_fs, &snaplistsize, &snaplist);
                    266:        if (error)
                    267:                goto out;
                    268:        /*
                    269:         * Acquire the snapshot lock.
                    270:         */
                    271:        mutex_enter(&si->si_snaplock);
                    272:        snapshot_locked = true;
                    273:        /*
                    274:         * Record snapshot inode. Since this is the newest snapshot,
                    275:         * it must be placed at the end of the list.
                    276:         */
                    277:        fs->fs_snapinum[snaploc] = ip->i_number;
                    278:
                    279:        mutex_enter(&si->si_lock);
                    280:        if (ip->i_nextsnap.tqe_prev != 0)
                    281:                panic("ffs_snapshot: %"PRIu64" already on list", ip->i_number);
                    282:        TAILQ_INSERT_TAIL(&si->si_snapshots, ip, i_nextsnap);
                    283:        if (TAILQ_FIRST(&si->si_snapshots) == ip) {
                    284:                /*
                    285:                 * If this is the first snapshot on this filesystem, put the
                    286:                 * preliminary list in place and establish the cow handler.
                    287:                 */
                    288:                si->si_snapblklist = snaplist;
                    289:                fscow_establish(mp, ffs_copyonwrite, devvp);
                    290:        }
                    291:        si->si_gen++;
                    292:        mutex_exit(&si->si_lock);
                    293:
                    294:        vp->v_vflag |= VV_SYSTEM;
                    295:        /*
                    296:         * Set the mtime to the time the snapshot has been taken.
                    297:         */
                    298:        TIMEVAL_TO_TIMESPEC(&starttime, &ts);
                    299:        if (ctime)
                    300:                *ctime = ts;
                    301:        DIP_ASSIGN(ip, mtime, ts.tv_sec);
                    302:        DIP_ASSIGN(ip, mtimensec, ts.tv_nsec);
                    303:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                    304:        /*
                    305:         * Copy allocation information from all snapshots and then
                    306:         * expunge them from our view.
                    307:         */
                    308:        error = snapshot_expunge_snap(mp, vp, copy_fs, snaplistsize);
                    309:        if (error)
                    310:                goto out;
                    311:        /*
                    312:         * Write the superblock and its summary information to the snapshot.
                    313:         */
                    314:        error = snapshot_writefs(mp, vp, sbbuf);
                    315:        if (error)
                    316:                goto out;
                    317:        /*
                    318:         * We're nearly done, ensure that the snapshot is completely on disk.
                    319:         */
                    320:        error = VOP_FSYNC(vp, l->l_cred, FSYNC_WAIT, 0, 0);
                    321:        if (error)
                    322:                goto out;
                    323:        /*
                    324:         * Invalidate and free all pages on the snapshot vnode.
                    325:         * We will read and write through the buffercache.
                    326:         */
                    327:        mutex_enter(&vp->v_interlock);
                    328:        error = VOP_PUTPAGES(vp, 0, 0,
                    329:                    PGO_ALLPAGES | PGO_CLEANIT | PGO_SYNCIO | PGO_FREE);
                    330:        if (error)
                    331:                goto out;
                    332:        /*
                    333:         * Invalidate short ( < fs_bsize ) buffers.  We will always read
                    334:         * full size buffers later.
                    335:         */
                    336:        mutex_enter(&bufcache_lock);
                    337:        KASSERT(LIST_FIRST(&vp->v_dirtyblkhd) == NULL);
                    338:        for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
                    339:                nbp = LIST_NEXT(bp, b_vnbufs);
                    340:                KASSERT((bp->b_cflags & BC_BUSY) == 0);
                    341:                if (bp->b_bcount < fs->fs_bsize) {
                    342:                        bp->b_cflags |= BC_BUSY;
                    343:                        brelsel(bp, BC_INVAL | BC_VFLUSH);
                    344:                }
                    345:        }
                    346:        mutex_exit(&bufcache_lock);
                    347:
                    348: out:
                    349:        if (sbbuf != NULL) {
                    350:                free(copy_fs->fs_csp, M_UFSMNT);
                    351:                free(sbbuf, M_UFSMNT);
                    352:        }
                    353:        if (fs->fs_active != NULL) {
                    354:                free(fs->fs_active, M_DEVBUF);
                    355:                fs->fs_active = NULL;
                    356:        }
                    357:
                    358:        mutex_enter(&si->si_lock);
                    359:        if (snaplist != NULL) {
                    360:                if (si->si_snapblklist == snaplist)
                    361:                        si->si_snapblklist = NULL;
                    362:                free(snaplist, M_UFSMNT);
                    363:        }
                    364:        if (error) {
                    365:                fs->fs_snapinum[snaploc] = 0;
                    366:        } else {
                    367:                /*
                    368:                 * As this is the newest list, it is the most inclusive, so
                    369:                 * should replace the previous list.
                    370:                 */
                    371:                si->si_snapblklist = ip->i_snapblklist;
                    372:        }
                    373:        si->si_gen++;
                    374:        mutex_exit(&si->si_lock);
                    375:
                    376:        if (snapshot_locked)
                    377:                mutex_exit(&si->si_snaplock);
                    378:        if (suspended) {
                    379:                vfs_resume(vp->v_mount);
                    380: #ifdef DEBUG
                    381:                getmicrotime(&endtime);
                    382:                timersub(&endtime, &starttime, &endtime);
                    383:                printf("%s: suspended %ld.%03ld sec, redo %d of %d\n",
                    384:                    mp->mnt_stat.f_mntonname, (long)endtime.tv_sec,
                    385:                    endtime.tv_usec / 1000, redo, fs->fs_ncg);
                    386: #endif
                    387:        }
                    388:        if (error) {
                    389:                if (!UFS_WAPBL_BEGIN(mp)) {
                    390:                        (void) ffs_truncate(vp, (off_t)0, 0, NOCRED);
                    391:                        UFS_WAPBL_END(mp);
                    392:                }
                    393:        } else
                    394:                vref(vp);
                    395:        return (error);
                    396: }
                    397:
                    398: /*
                    399:  * Prepare vnode to become a snapshot.
                    400:  */
                    401: static int
                    402: snapshot_setup(struct mount *mp, struct vnode *vp)
                    403: {
                    404:        int error, i, len, loc;
                    405:        daddr_t blkno, numblks;
                    406:        struct buf *ibp, *nbp;
                    407:        struct fs *fs = VFSTOUFS(mp)->um_fs;
                    408:        struct lwp *l = curlwp;
                    409:
                    410:        /*
1.9       hannken   411:         * Check mount, exclusive reference and owner.
1.1       hannken   412:         */
1.4       hannken   413:        if (vp->v_mount != mp)
1.1       hannken   414:                return EXDEV;
1.4       hannken   415:        if (vp->v_usecount != 1 || vp->v_writecount != 0)
1.1       hannken   416:                return EBUSY;
1.31      ad        417:        if (kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER,
1.39      elad      418:            NULL) != 0 &&
1.31      ad        419:            VTOI(vp)->i_uid != kauth_cred_geteuid(l->l_cred))
1.9       hannken   420:                return EACCES;
                    421:
1.1       hannken   422:        if (vp->v_size != 0) {
1.56      pooka     423:                error = ffs_truncate(vp, 0, 0, NOCRED);
1.4       hannken   424:                if (error)
1.1       hannken   425:                        return error;
                    426:        }
                    427:        /*
1.16      hannken   428:         * Write an empty list of preallocated blocks to the end of
                    429:         * the snapshot to set size to at least that of the filesystem.
1.1       hannken   430:         */
                    431:        numblks = howmany(fs->fs_size, fs->fs_frag);
1.16      hannken   432:        blkno = 1;
1.79      hannken   433:        blkno = ufs_rw64(blkno, UFS_FSNEEDSWAP(fs));
1.1       hannken   434:        error = vn_rdwr(UIO_WRITE, vp,
1.43      christos  435:            (void *)&blkno, sizeof(blkno), lblktosize(fs, (off_t)numblks),
1.31      ad        436:            UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, l->l_cred, NULL, NULL);
1.1       hannken   437:        if (error)
1.79      hannken   438:                return error;
1.1       hannken   439:        /*
                    440:         * Preallocate critical data structures so that we can copy
                    441:         * them in without further allocation after we suspend all
                    442:         * operations on the filesystem. We would like to just release
                    443:         * the allocated buffers without writing them since they will
                    444:         * be filled in below once we are ready to go, but this upsets
                    445:         * the soft update code, so we go ahead and write the new buffers.
                    446:         *
                    447:         * Allocate all indirect blocks and mark all of them as not
                    448:         * needing to be copied.
                    449:         */
1.75      hannken   450:        error = UFS_WAPBL_BEGIN(mp);
                    451:        if (error)
1.79      hannken   452:                return error;
1.75      hannken   453:        for (blkno = NDADDR, i = 0; blkno < numblks; blkno += NINDIR(fs)) {
1.22      yamt      454:                error = ffs_balloc(vp, lblktosize(fs, (off_t)blkno),
1.31      ad        455:                    fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
1.79      hannken   456:                if (error)
1.1       hannken   457:                        goto out;
1.72      hannken   458:                if (DOINGSOFTDEP(vp))
                    459:                        bawrite(ibp);
                    460:                else
                    461:                        brelse(ibp, 0);
1.75      hannken   462:                if ((++i % 16) == 0) {
                    463:                        UFS_WAPBL_END(mp);
                    464:                        error = UFS_WAPBL_BEGIN(mp);
                    465:                        if (error)
1.79      hannken   466:                                return error;
1.75      hannken   467:                }
1.1       hannken   468:        }
                    469:        /*
                    470:         * Allocate copies for the superblock and its summary information.
                    471:         */
1.79      hannken   472:        error = ffs_balloc(vp, fs->fs_sblockloc, fs->fs_sbsize, l->l_cred,
1.15      hannken   473:            0, &nbp);
1.79      hannken   474:        if (error)
1.15      hannken   475:                goto out;
                    476:        bawrite(nbp);
1.1       hannken   477:        blkno = fragstoblks(fs, fs->fs_csaddr);
1.15      hannken   478:        len = howmany(fs->fs_cssize, fs->fs_bsize);
                    479:        for (loc = 0; loc < len; loc++) {
1.22      yamt      480:                error = ffs_balloc(vp, lblktosize(fs, (off_t)(blkno + loc)),
1.79      hannken   481:                    fs->fs_bsize, l->l_cred, 0, &nbp);
                    482:                if (error)
1.1       hannken   483:                        goto out;
1.15      hannken   484:                bawrite(nbp);
                    485:        }
1.79      hannken   486:
                    487: out:
1.75      hannken   488:        UFS_WAPBL_END(mp);
1.79      hannken   489:        return error;
                    490: }
                    491:
                    492: /*
                    493:  * Create a copy of the superblock and its summary information.
                    494:  * It is up to the caller to free copyfs and copy_fs->fs_csp.
                    495:  */
                    496: static int
                    497: snapshot_copyfs(struct mount *mp, struct vnode *vp, void **sbbuf)
                    498: {
                    499:        int error, i, len, loc, size;
                    500:        void *space;
                    501:        int32_t *lp;
                    502:        struct buf *bp;
                    503:        struct fs *copyfs, *fs = VFSTOUFS(mp)->um_fs;
                    504:        struct lwp *l = curlwp;
                    505:        struct vnode *devvp = VTOI(vp)->i_devvp;
                    506:
1.1       hannken   507:        /*
                    508:         * Grab a copy of the superblock and its summary information.
                    509:         * We delay writing it until the suspension is released below.
                    510:         */
1.79      hannken   511:        *sbbuf = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.1       hannken   512:        loc = blkoff(fs, fs->fs_sblockloc);
                    513:        if (loc > 0)
1.79      hannken   514:                memset(*sbbuf, 0, loc);
                    515:        copyfs = (struct fs *)((char *)(*sbbuf) + loc);
                    516:        bcopy(fs, copyfs, fs->fs_sbsize);
1.1       hannken   517:        size = fs->fs_bsize < SBLOCKSIZE ? fs->fs_bsize : SBLOCKSIZE;
                    518:        if (fs->fs_sbsize < size)
1.79      hannken   519:                memset((char *)(*sbbuf) + loc + fs->fs_sbsize, 0,
1.43      christos  520:                    size - fs->fs_sbsize);
1.1       hannken   521:        size = blkroundup(fs, fs->fs_cssize);
                    522:        if (fs->fs_contigsumsize > 0)
                    523:                size += fs->fs_ncg * sizeof(int32_t);
1.79      hannken   524:        space = malloc(size, M_UFSMNT, M_WAITOK);
                    525:        copyfs->fs_csp = space;
                    526:        bcopy(fs->fs_csp, copyfs->fs_csp, fs->fs_cssize);
1.12      yamt      527:        space = (char *)space + fs->fs_cssize;
1.1       hannken   528:        loc = howmany(fs->fs_cssize, fs->fs_fsize);
                    529:        i = fs->fs_frag - loc % fs->fs_frag;
                    530:        len = (i == fs->fs_frag) ? 0 : i * fs->fs_fsize;
                    531:        if (len > 0) {
                    532:                if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + loc),
1.79      hannken   533:                    len, l->l_cred, 0, &bp)) != 0) {
1.52      ad        534:                        brelse(bp, 0);
1.79      hannken   535:                        free(copyfs->fs_csp, M_UFSMNT);
                    536:                        free(*sbbuf, M_UFSMNT);
                    537:                        *sbbuf = NULL;
                    538:                        return error;
1.1       hannken   539:                }
                    540:                bcopy(bp->b_data, space, (u_int)len);
1.12      yamt      541:                space = (char *)space + len;
1.52      ad        542:                brelse(bp, BC_INVAL | BC_NOCACHE);
1.1       hannken   543:        }
                    544:        if (fs->fs_contigsumsize > 0) {
1.79      hannken   545:                copyfs->fs_maxcluster = lp = space;
1.1       hannken   546:                for (i = 0; i < fs->fs_ncg; i++)
                    547:                        *lp++ = fs->fs_contigsumsize;
                    548:        }
1.79      hannken   549:        if (mp->mnt_wapbl)
                    550:                copyfs->fs_flags &= ~FS_DOWAPBL;
                    551:        return 0;
                    552: }
                    553:
                    554: /*
                    555:  * We must check for active files that have been unlinked (e.g., with a zero
                    556:  * link count). We have to expunge all trace of these files from the snapshot
                    557:  * so that they are not reclaimed prematurely by fsck or unnecessarily dumped.
                    558:  * Note that we skip unlinked snapshot files as they will be handled separately.
                    559:  * Calculate the snapshot list size and create a preliminary list.
                    560:  */
                    561: static int
                    562: snapshot_expunge(struct mount *mp, struct vnode *vp, struct fs *copy_fs,
                    563:     daddr_t *snaplistsize, daddr_t **snaplist)
                    564: {
1.80      hannken   565:        bool has_wapbl = false;
1.79      hannken   566:        int cg, error, len, loc;
                    567:        daddr_t blkno, *blkp;
                    568:        struct fs *fs = VFSTOUFS(mp)->um_fs;
                    569:        struct inode *xp;
                    570:        struct lwp *l = curlwp;
                    571:        struct vattr vat;
                    572:        struct vnode *logvp = NULL, *mvp = NULL, *xvp;
                    573:
                    574:        *snaplist = NULL;
1.1       hannken   575:        /*
1.79      hannken   576:         * Get the log inode if any.
                    577:         */
                    578:        if ((fs->fs_flags & FS_DOWAPBL) &&
                    579:            fs->fs_journal_location == UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM) {
                    580:                error = VFS_VGET(mp,
                    581:                    fs->fs_journallocs[UFS_WAPBL_INFS_INO], &logvp);
                    582:                if (error)
                    583:                        goto out;
                    584:        }
                    585:        /*
                    586:         * Allocate a marker vnode.
1.1       hannken   587:         */
1.58      pooka     588:        if ((mvp = vnalloc(mp)) == NULL) {
1.57      ad        589:                error = ENOMEM;
1.75      hannken   590:                goto out;
1.57      ad        591:        }
1.79      hannken   592:        /*
                    593:         * We also calculate the needed size for the snapshot list.
                    594:         */
                    595:        *snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
                    596:            FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
1.80      hannken   597:        error = UFS_WAPBL_BEGIN(mp);
                    598:        if (error)
                    599:                goto out;
                    600:        has_wapbl = true;
1.79      hannken   601:        mutex_enter(&mntvnode_lock);
1.35      reinoud   602:        /*
                    603:         * NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
                    604:         * and vclean() can be called indirectly
                    605:         */
1.57      ad        606:        for (xvp = TAILQ_FIRST(&mp->mnt_vnodelist); xvp; xvp = vunmark(mvp)) {
1.59      hannken   607:                vmark(mvp, xvp);
1.1       hannken   608:                /*
                    609:                 * Make sure this vnode wasn't reclaimed in getnewvnode().
                    610:                 * Start over if it has (it won't be on the list anymore).
                    611:                 */
1.57      ad        612:                if (xvp->v_mount != mp || vismarker(xvp))
                    613:                        continue;
1.79      hannken   614:                mutex_enter(&xvp->v_interlock);
1.53      ad        615:                if ((xvp->v_iflag & VI_XLOCK) ||
1.1       hannken   616:                    xvp->v_usecount == 0 || xvp->v_type == VNON ||
1.61      hannken   617:                    VTOI(xvp) == NULL ||
1.1       hannken   618:                    (VTOI(xvp)->i_flags & SF_SNAPSHOT)) {
1.79      hannken   619:                        mutex_exit(&xvp->v_interlock);
1.1       hannken   620:                        continue;
                    621:                }
1.79      hannken   622:                mutex_exit(&mntvnode_lock);
1.57      ad        623:                /*
                    624:                 * XXXAD should increase vnode ref count to prevent it
                    625:                 * disappearing or being recycled.
                    626:                 */
1.79      hannken   627:                mutex_exit(&xvp->v_interlock);
1.1       hannken   628: #ifdef DEBUG
                    629:                if (snapdebug)
                    630:                        vprint("ffs_snapshot: busy vnode", xvp);
                    631: #endif
                    632:                xp = VTOI(xvp);
1.79      hannken   633:                if (xvp != logvp) {
                    634:                        if (VOP_GETATTR(xvp, &vat, l->l_cred) == 0 &&
                    635:                            vat.va_nlink > 0) {
                    636:                                mutex_enter(&mntvnode_lock);
                    637:                                continue;
                    638:                        }
                    639:                        if (ffs_checkfreefile(copy_fs, vp, xp->i_number)) {
                    640:                                mutex_enter(&mntvnode_lock);
                    641:                                continue;
                    642:                        }
1.1       hannken   643:                }
                    644:                /*
                    645:                 * If there is a fragment, clear it here.
                    646:                 */
                    647:                blkno = 0;
                    648:                loc = howmany(xp->i_size, fs->fs_bsize) - 1;
                    649:                if (loc < NDADDR) {
                    650:                        len = fragroundup(fs, blkoff(fs, xp->i_size));
1.5       hannken   651:                        if (len > 0 && len < fs->fs_bsize) {
1.83      joerg     652:                                ffs_blkfree_snap(copy_fs, vp, db_get(xp, loc),
1.1       hannken   653:                                    len, xp->i_number);
                    654:                                blkno = db_get(xp, loc);
                    655:                                db_assign(xp, loc, 0);
                    656:                        }
                    657:                }
1.79      hannken   658:                *snaplistsize += 1;
1.76      hannken   659:                error = expunge(vp, xp, copy_fs, fullacct, BLK_NOCOPY);
1.1       hannken   660:                if (blkno)
                    661:                        db_assign(xp, loc, blkno);
1.6       hannken   662:                if (!error)
1.84      joerg     663:                        error = ffs_freefile_snap(copy_fs, vp, xp->i_number,
1.6       hannken   664:                            xp->i_mode);
1.1       hannken   665:                if (error) {
1.57      ad        666:                        (void)vunmark(mvp);
1.75      hannken   667:                        goto out;
1.1       hannken   668:                }
1.79      hannken   669:                mutex_enter(&mntvnode_lock);
1.65      hannken   670:        }
1.79      hannken   671:        mutex_exit(&mntvnode_lock);
1.1       hannken   672:        /*
1.79      hannken   673:         * Create a preliminary list of preallocated snapshot blocks.
1.1       hannken   674:         */
1.79      hannken   675:        *snaplist = malloc(*snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK);
                    676:        blkp = &(*snaplist)[1];
                    677:        *blkp++ = lblkno(fs, fs->fs_sblockloc);
                    678:        blkno = fragstoblks(fs, fs->fs_csaddr);
                    679:        for (cg = 0; cg < fs->fs_ncg; cg++) {
                    680:                if (fragstoblks(fs, cgtod(fs, cg)) > blkno)
                    681:                        break;
                    682:                *blkp++ = fragstoblks(fs, cgtod(fs, cg));
                    683:        }
                    684:        len = howmany(fs->fs_cssize, fs->fs_bsize);
                    685:        for (loc = 0; loc < len; loc++)
                    686:                *blkp++ = blkno + loc;
                    687:        for (; cg < fs->fs_ncg; cg++)
                    688:                *blkp++ = fragstoblks(fs, cgtod(fs, cg));
                    689:
                    690: out:
1.80      hannken   691:        if (has_wapbl)
                    692:                UFS_WAPBL_END(mp);
1.79      hannken   693:        if (mvp != NULL)
                    694:                vnfree(mvp);
                    695:        if (logvp != NULL)
                    696:                vput(logvp);
                    697:        if (error && *snaplist != NULL) {
                    698:                free(*snaplist, M_UFSMNT);
                    699:                *snaplist = NULL;
                    700:        }
                    701:
                    702:        return error;
                    703: }
                    704:
                    705: /*
                    706:  * Copy allocation information from all the snapshots in this snapshot and
                    707:  * then expunge them from its view. Also, collect the list of allocated
                    708:  * blocks in i_snapblklist.
                    709:  */
                    710: static int
                    711: snapshot_expunge_snap(struct mount *mp, struct vnode *vp,
                    712:     struct fs *copy_fs, daddr_t snaplistsize)
                    713: {
                    714:        int error, i;
                    715:        daddr_t numblks, *snaplist = NULL;
                    716:        struct fs *fs = VFSTOUFS(mp)->um_fs;
                    717:        struct inode *ip = VTOI(vp), *xp;
                    718:        struct lwp *l = curlwp;
                    719:        struct snap_info *si = VFSTOUFS(mp)->um_snapinfo;
                    720:
                    721:        error = UFS_WAPBL_BEGIN(mp);
                    722:        if (error)
                    723:                return error;
1.47      hannken   724:        TAILQ_FOREACH(xp, &si->si_snapshots, i_nextsnap) {
1.1       hannken   725:                if (xp == ip)
                    726:                        break;
1.79      hannken   727:                error = expunge(vp, xp, fs, snapacct, BLK_SNAP);
                    728:                if (error)
                    729:                        break;
                    730:                if (xp->i_ffs_effnlink != 0)
                    731:                        continue;
1.84      joerg     732:                error = ffs_freefile_snap(copy_fs, vp, xp->i_number, xp->i_mode);
1.79      hannken   733:                if (error)
                    734:                        break;
1.1       hannken   735:        }
1.79      hannken   736:        if (error)
                    737:                goto out;
1.1       hannken   738:        /*
                    739:         * Allocate space for the full list of preallocated snapshot blocks.
                    740:         */
1.79      hannken   741:        snaplist = malloc(snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK);
                    742:        ip->i_snapblklist = &snaplist[1];
1.1       hannken   743:        /*
                    744:         * Expunge the blocks used by the snapshots from the set of
                    745:         * blocks marked as used in the snapshot bitmaps. Also, collect
                    746:         * the list of allocated blocks in i_snapblklist.
                    747:         */
1.79      hannken   748:        error = expunge(vp, ip, copy_fs, mapacct, BLK_SNAP);
                    749:        if (error)
                    750:                goto out;
                    751:        if (snaplistsize < ip->i_snapblklist - snaplist)
1.1       hannken   752:                panic("ffs_snapshot: list too small");
1.79      hannken   753:        snaplistsize = ip->i_snapblklist - snaplist;
                    754:        snaplist[0] = snaplistsize;
                    755:        ip->i_snapblklist = &snaplist[0];
1.1       hannken   756:        /*
                    757:         * Write out the list of allocated blocks to the end of the snapshot.
                    758:         */
1.79      hannken   759:        numblks = howmany(fs->fs_size, fs->fs_frag);
1.15      hannken   760:        for (i = 0; i < snaplistsize; i++)
1.79      hannken   761:                snaplist[i] = ufs_rw64(snaplist[i], UFS_FSNEEDSWAP(fs));
                    762:        error = vn_rdwr(UIO_WRITE, vp, (void *)snaplist,
                    763:            snaplistsize * sizeof(daddr_t), lblktosize(fs, (off_t)numblks),
                    764:            UIO_SYSSPACE, IO_NODELOCKED | IO_JOURNALLOCKED | IO_UNIT,
                    765:            l->l_cred, NULL, NULL);
1.15      hannken   766:        for (i = 0; i < snaplistsize; i++)
1.79      hannken   767:                snaplist[i] = ufs_rw64(snaplist[i], UFS_FSNEEDSWAP(fs));
                    768: out:
1.80      hannken   769:        UFS_WAPBL_END(mp);
1.79      hannken   770:        if (error && snaplist != NULL) {
                    771:                free(snaplist, M_UFSMNT);
                    772:                ip->i_snapblklist = NULL;
1.1       hannken   773:        }
1.79      hannken   774:        return error;
                    775: }
                    776:
                    777: /*
                    778:  * Write the superblock and its summary information to the snapshot.
                    779:  * Make sure, the first NDADDR blocks get copied to the snapshot.
                    780:  */
                    781: static int
                    782: snapshot_writefs(struct mount *mp, struct vnode *vp, void *sbbuf)
                    783: {
                    784:        int error, len, loc;
                    785:        void *space;
                    786:        daddr_t blkno;
                    787:        struct buf *bp;
                    788:        struct fs *copyfs, *fs = VFSTOUFS(mp)->um_fs;
                    789:        struct inode *ip = VTOI(vp);
                    790:        struct lwp *l = curlwp;
                    791:
                    792:        copyfs = (struct fs *)((char *)sbbuf + blkoff(fs, fs->fs_sblockloc));
                    793:
1.1       hannken   794:        /*
                    795:         * Write the superblock and its summary information
                    796:         * to the snapshot.
                    797:         */
                    798:        blkno = fragstoblks(fs, fs->fs_csaddr);
                    799:        len = howmany(fs->fs_cssize, fs->fs_bsize);
1.79      hannken   800:        space = copyfs->fs_csp;
1.8       hannken   801: #ifdef FFS_EI
1.79      hannken   802:        if (UFS_FSNEEDSWAP(fs)) {
                    803:                ffs_sb_swap(copyfs, copyfs);
1.1       hannken   804:                ffs_csum_swap(space, space, fs->fs_cssize);
                    805:        }
1.8       hannken   806: #endif
1.75      hannken   807:        error = UFS_WAPBL_BEGIN(mp);
1.79      hannken   808:        if (error)
                    809:                return error;
1.1       hannken   810:        for (loc = 0; loc < len; loc++) {
1.79      hannken   811:                error = bread(vp, blkno + loc, fs->fs_bsize, l->l_cred,
                    812:                    B_MODIFY, &bp);
1.15      hannken   813:                if (error) {
1.79      hannken   814:                        brelse(bp, 0);
                    815:                        break;
1.1       hannken   816:                }
1.79      hannken   817:                bcopy(space, bp->b_data, fs->fs_bsize);
1.1       hannken   818:                space = (char *)space + fs->fs_bsize;
1.79      hannken   819:                bawrite(bp);
                    820:        }
                    821:        if (error)
                    822:                goto out;
                    823:        error = bread(vp, lblkno(fs, fs->fs_sblockloc),
                    824:            fs->fs_bsize, l->l_cred, B_MODIFY, &bp);
                    825:        if (error) {
                    826:                brelse(bp, 0);
                    827:                goto out;
                    828:        } else {
                    829:                bcopy(sbbuf, bp->b_data, fs->fs_bsize);
                    830:                bawrite(bp);
1.1       hannken   831:        }
                    832:        /*
1.75      hannken   833:         * Copy the first NDADDR blocks to the snapshot so ffs_copyonwrite()
                    834:         * and ffs_snapblkfree() will always work on indirect blocks.
                    835:         */
                    836:        for (loc = 0; loc < NDADDR; loc++) {
                    837:                if (db_get(ip, loc) != 0)
                    838:                        continue;
                    839:                error = ffs_balloc(vp, lblktosize(fs, (off_t)loc),
1.79      hannken   840:                    fs->fs_bsize, l->l_cred, 0, &bp);
1.75      hannken   841:                if (error)
                    842:                        break;
1.79      hannken   843:                error = rwfsblk(vp, B_READ, bp->b_data, loc);
1.75      hannken   844:                if (error) {
1.79      hannken   845:                        brelse(bp, 0);
                    846:                        break;
1.75      hannken   847:                }
1.79      hannken   848:                bawrite(bp);
1.75      hannken   849:        }
1.79      hannken   850:
                    851: out:
1.75      hannken   852:        UFS_WAPBL_END(mp);
1.79      hannken   853:        return error;
                    854: }
                    855:
                    856: /*
                    857:  * Copy all cylinder group maps.
                    858:  */
                    859: static int
                    860: cgaccount(struct vnode *vp, int passno, int *redo)
                    861: {
                    862:        int cg, error;
                    863:        struct buf *nbp;
                    864:        struct fs *fs = VTOI(vp)->i_fs;
                    865:
1.80      hannken   866:        error = UFS_WAPBL_BEGIN(vp->v_mount);
                    867:        if (error)
                    868:                return error;
1.79      hannken   869:        if (redo != NULL)
                    870:                *redo = 0;
                    871:        if (passno == 1)
                    872:                fs->fs_active = malloc(howmany(fs->fs_ncg, NBBY),
                    873:                    M_DEVBUF, M_WAITOK | M_ZERO);
                    874:        for (cg = 0; cg < fs->fs_ncg; cg++) {
                    875:                if (passno == 2 && ACTIVECG_ISSET(fs, cg))
                    876:                        continue;
                    877:                if (redo != NULL)
                    878:                        *redo += 1;
                    879:                error = ffs_balloc(vp, lfragtosize(fs, cgtod(fs, cg)),
                    880:                    fs->fs_bsize, curlwp->l_cred, 0, &nbp);
                    881:                if (error)
1.80      hannken   882:                        break;
1.79      hannken   883:                error = cgaccount1(cg, vp, nbp->b_data, passno);
                    884:                bawrite(nbp);
1.75      hannken   885:                if (error)
1.80      hannken   886:                        break;
1.15      hannken   887:        }
1.80      hannken   888:        UFS_WAPBL_END(vp->v_mount);
                    889:        return error;
1.1       hannken   890: }
                    891:
                    892: /*
                    893:  * Copy a cylinder group map. All the unallocated blocks are marked
                    894:  * BLK_NOCOPY so that the snapshot knows that it need not copy them
                    895:  * if they are later written. If passno is one, then this is a first
                    896:  * pass, so only setting needs to be done. If passno is 2, then this
                    897:  * is a revision to a previous pass which must be undone as the
                    898:  * replacement pass is done.
                    899:  */
                    900: static int
1.79      hannken   901: cgaccount1(int cg, struct vnode *vp, void *data, int passno)
1.1       hannken   902: {
                    903:        struct buf *bp, *ibp;
                    904:        struct inode *ip;
                    905:        struct cg *cgp;
                    906:        struct fs *fs;
1.79      hannken   907:        struct lwp *l = curlwp;
                    908:        daddr_t base, numblks;
1.1       hannken   909:        int error, len, loc, ns, indiroff;
                    910:
                    911:        ip = VTOI(vp);
                    912:        fs = ip->i_fs;
                    913:        ns = UFS_FSNEEDSWAP(fs);
                    914:        error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1.79      hannken   915:                (int)fs->fs_cgsize, l->l_cred, 0, &bp);
1.1       hannken   916:        if (error) {
1.52      ad        917:                brelse(bp, 0);
1.1       hannken   918:                return (error);
                    919:        }
                    920:        cgp = (struct cg *)bp->b_data;
                    921:        if (!cg_chkmagic(cgp, ns)) {
1.52      ad        922:                brelse(bp, 0);
1.1       hannken   923:                return (EIO);
                    924:        }
                    925:        ACTIVECG_SET(fs, cg);
                    926:
                    927:        bcopy(bp->b_data, data, fs->fs_cgsize);
1.52      ad        928:        brelse(bp, 0);
1.1       hannken   929:        if (fs->fs_cgsize < fs->fs_bsize)
1.43      christos  930:                memset((char *)data + fs->fs_cgsize, 0,
1.1       hannken   931:                    fs->fs_bsize - fs->fs_cgsize);
                    932:        numblks = howmany(fs->fs_size, fs->fs_frag);
                    933:        len = howmany(fs->fs_fpg, fs->fs_frag);
                    934:        base = cg * fs->fs_fpg / fs->fs_frag;
                    935:        if (base + len >= numblks)
                    936:                len = numblks - base - 1;
                    937:        loc = 0;
                    938:        if (base < NDADDR) {
                    939:                for ( ; loc < NDADDR; loc++) {
                    940:                        if (ffs_isblock(fs, cg_blksfree(cgp, ns), loc))
                    941:                                db_assign(ip, loc, BLK_NOCOPY);
                    942:                        else if (db_get(ip, loc) == BLK_NOCOPY) {
                    943:                                if (passno == 2)
                    944:                                        db_assign(ip, loc, 0);
                    945:                                else if (passno == 1)
                    946:                                        panic("ffs_snapshot: lost direct block");
                    947:                        }
                    948:                }
                    949:        }
1.22      yamt      950:        if ((error = ffs_balloc(vp, lblktosize(fs, (off_t)(base + loc)),
1.79      hannken   951:            fs->fs_bsize, l->l_cred, B_METAONLY, &ibp)) != 0)
1.1       hannken   952:                return (error);
                    953:        indiroff = (base + loc - NDADDR) % NINDIR(fs);
                    954:        for ( ; loc < len; loc++, indiroff++) {
                    955:                if (indiroff >= NINDIR(fs)) {
1.15      hannken   956:                        bawrite(ibp);
1.22      yamt      957:                        if ((error = ffs_balloc(vp,
1.1       hannken   958:                            lblktosize(fs, (off_t)(base + loc)),
1.79      hannken   959:                            fs->fs_bsize, l->l_cred, B_METAONLY, &ibp)) != 0)
1.1       hannken   960:                                return (error);
                    961:                        indiroff = 0;
                    962:                }
                    963:                if (ffs_isblock(fs, cg_blksfree(cgp, ns), loc))
                    964:                        idb_assign(ip, ibp->b_data, indiroff, BLK_NOCOPY);
                    965:                else if (idb_get(ip, ibp->b_data, indiroff) == BLK_NOCOPY) {
                    966:                        if (passno == 2)
                    967:                                idb_assign(ip, ibp->b_data, indiroff, 0);
                    968:                        else if (passno == 1)
                    969:                                panic("ffs_snapshot: lost indirect block");
                    970:                }
                    971:        }
1.15      hannken   972:        bdwrite(ibp);
1.1       hannken   973:        return (0);
                    974: }
                    975:
                    976: /*
                    977:  * Before expunging a snapshot inode, note all the
                    978:  * blocks that it claims with BLK_SNAP so that fsck will
                    979:  * be able to account for those blocks properly and so
                    980:  * that this snapshot knows that it need not copy them
1.76      hannken   981:  * if the other snapshot holding them is freed.
1.1       hannken   982:  */
                    983: static int
1.76      hannken   984: expunge(struct vnode *snapvp, struct inode *cancelip, struct fs *fs,
                    985:     acctfunc_t acctfunc, int expungetype)
1.1       hannken   986: {
1.71      hannken   987:        int i, error, ns;
1.76      hannken   988:        daddr_t lbn, rlbn;
                    989:        daddr_t len, blkno, numblks, blksperindir;
                    990:        struct ufs1_dinode *dip1;
                    991:        struct ufs2_dinode *dip2;
1.79      hannken   992:        struct lwp *l = curlwp;
1.76      hannken   993:        void *bap;
1.1       hannken   994:        struct buf *bp;
                    995:
                    996:        ns = UFS_FSNEEDSWAP(fs);
                    997:        /*
                    998:         * Prepare to expunge the inode. If its inode block has not
                    999:         * yet been copied, then allocate and fill the copy.
                   1000:         */
                   1001:        lbn = fragstoblks(fs, ino_to_fsba(fs, cancelip->i_number));
1.72      hannken  1002:        error = snapblkaddr(snapvp, lbn, &blkno);
1.71      hannken  1003:        if (error)
                   1004:                return error;
                   1005:        if (blkno != 0) {
1.79      hannken  1006:                error = bread(snapvp, lbn, fs->fs_bsize, l->l_cred,
1.71      hannken  1007:                    B_MODIFY, &bp);
1.1       hannken  1008:        } else {
1.22      yamt     1009:                error = ffs_balloc(snapvp, lblktosize(fs, (off_t)lbn),
1.79      hannken  1010:                    fs->fs_bsize, l->l_cred, 0, &bp);
1.71      hannken  1011:                if (! error)
1.75      hannken  1012:                        error = rwfsblk(snapvp, B_READ, bp->b_data, lbn);
1.1       hannken  1013:        }
1.71      hannken  1014:        if (error)
1.1       hannken  1015:                return error;
                   1016:        /*
                   1017:         * Set a snapshot inode to be a zero length file, regular files
1.48      hannken  1018:         * or unlinked snapshots to be completely unallocated.
1.1       hannken  1019:         */
1.76      hannken  1020:        if (fs->fs_magic == FS_UFS1_MAGIC) {
                   1021:                dip1 = (struct ufs1_dinode *)bp->b_data +
                   1022:                    ino_to_fsbo(fs, cancelip->i_number);
                   1023:                if (expungetype == BLK_NOCOPY || cancelip->i_ffs_effnlink == 0)
                   1024:                        dip1->di_mode = 0;
                   1025:                dip1->di_size = 0;
                   1026:                dip1->di_blocks = 0;
                   1027:                dip1->di_flags =
                   1028:                    ufs_rw32(ufs_rw32(dip1->di_flags, ns) & ~SF_SNAPSHOT, ns);
                   1029:                bzero(&dip1->di_db[0], (NDADDR + NIADDR) * sizeof(int32_t));
                   1030:        } else {
                   1031:                dip2 = (struct ufs2_dinode *)bp->b_data +
                   1032:                    ino_to_fsbo(fs, cancelip->i_number);
                   1033:                if (expungetype == BLK_NOCOPY || cancelip->i_ffs_effnlink == 0)
                   1034:                        dip2->di_mode = 0;
                   1035:                dip2->di_size = 0;
                   1036:                dip2->di_blocks = 0;
                   1037:                dip2->di_flags =
                   1038:                    ufs_rw32(ufs_rw32(dip2->di_flags, ns) & ~SF_SNAPSHOT, ns);
                   1039:                bzero(&dip2->di_db[0], (NDADDR + NIADDR) * sizeof(int64_t));
                   1040:        }
1.71      hannken  1041:        bdwrite(bp);
1.1       hannken  1042:        /*
                   1043:         * Now go through and expunge all the blocks in the file
                   1044:         * using the function requested.
                   1045:         */
                   1046:        numblks = howmany(cancelip->i_size, fs->fs_bsize);
1.76      hannken  1047:        if (fs->fs_magic == FS_UFS1_MAGIC)
                   1048:                bap = &cancelip->i_ffs1_db[0];
                   1049:        else
                   1050:                bap = &cancelip->i_ffs2_db[0];
                   1051:        if ((error = (*acctfunc)(snapvp, bap, 0, NDADDR, fs, 0, expungetype)))
1.1       hannken  1052:                return (error);
1.76      hannken  1053:        if (fs->fs_magic == FS_UFS1_MAGIC)
                   1054:                bap = &cancelip->i_ffs1_ib[0];
                   1055:        else
                   1056:                bap = &cancelip->i_ffs2_ib[0];
                   1057:        if ((error = (*acctfunc)(snapvp, bap, 0, NIADDR, fs, -1, expungetype)))
1.1       hannken  1058:                return (error);
                   1059:        blksperindir = 1;
                   1060:        lbn = -NDADDR;
                   1061:        len = numblks - NDADDR;
                   1062:        rlbn = NDADDR;
                   1063:        for (i = 0; len > 0 && i < NIADDR; i++) {
1.76      hannken  1064:                error = indiracct(snapvp, ITOV(cancelip), i,
                   1065:                    ib_get(cancelip, i), lbn, rlbn, len,
1.1       hannken  1066:                    blksperindir, fs, acctfunc, expungetype);
                   1067:                if (error)
                   1068:                        return (error);
                   1069:                blksperindir *= NINDIR(fs);
                   1070:                lbn -= blksperindir + 1;
                   1071:                len -= blksperindir;
                   1072:                rlbn += blksperindir;
                   1073:        }
                   1074:        return (0);
                   1075: }
                   1076:
                   1077: /*
                   1078:  * Descend an indirect block chain for vnode cancelvp accounting for all
                   1079:  * its indirect blocks in snapvp.
1.11      perry    1080:  */
1.1       hannken  1081: static int
1.76      hannken  1082: indiracct(struct vnode *snapvp, struct vnode *cancelvp, int level,
                   1083:     daddr_t blkno, daddr_t lbn, daddr_t rlbn, daddr_t remblks,
                   1084:     daddr_t blksperindir, struct fs *fs, acctfunc_t acctfunc, int expungetype)
1.1       hannken  1085: {
1.76      hannken  1086:        int error, num, i;
                   1087:        daddr_t subblksperindir;
1.1       hannken  1088:        struct indir indirs[NIADDR + 2];
1.76      hannken  1089:        daddr_t last;
                   1090:        void *bap;
1.1       hannken  1091:        struct buf *bp;
                   1092:
                   1093:        if (blkno == 0) {
                   1094:                if (expungetype == BLK_NOCOPY)
                   1095:                        return (0);
1.76      hannken  1096:                panic("indiracct: missing indir");
1.1       hannken  1097:        }
                   1098:        if ((error = ufs_getlbns(cancelvp, rlbn, indirs, &num)) != 0)
                   1099:                return (error);
                   1100:        if (lbn != indirs[num - 1 - level].in_lbn || num < 2)
1.76      hannken  1101:                panic("indiracct: botched params");
1.1       hannken  1102:        /*
                   1103:         * We have to expand bread here since it will deadlock looking
                   1104:         * up the block number for any blocks that are not in the cache.
                   1105:         */
1.69      hannken  1106:        error = ffs_getblk(cancelvp, lbn, fsbtodb(fs, blkno), fs->fs_bsize,
                   1107:            false, &bp);
                   1108:        if (error)
                   1109:                return error;
1.72      hannken  1110:        if ((bp->b_oflags & (BO_DONE | BO_DELWRI)) == 0 && (error =
1.75      hannken  1111:            rwfsblk(bp->b_vp, B_READ, bp->b_data, fragstoblks(fs, blkno)))) {
1.52      ad       1112:                brelse(bp, 0);
1.1       hannken  1113:                return (error);
                   1114:        }
                   1115:        /*
                   1116:         * Account for the block pointers in this indirect block.
                   1117:         */
                   1118:        last = howmany(remblks, blksperindir);
                   1119:        if (last > NINDIR(fs))
                   1120:                last = NINDIR(fs);
1.25      christos 1121:        bap = malloc(fs->fs_bsize, M_DEVBUF, M_WAITOK);
1.43      christos 1122:        bcopy(bp->b_data, (void *)bap, fs->fs_bsize);
1.52      ad       1123:        brelse(bp, 0);
1.76      hannken  1124:        error = (*acctfunc)(snapvp, bap, 0, last,
                   1125:            fs, level == 0 ? rlbn : -1, expungetype);
1.1       hannken  1126:        if (error || level == 0)
                   1127:                goto out;
                   1128:        /*
                   1129:         * Account for the block pointers in each of the indirect blocks
                   1130:         * in the levels below us.
                   1131:         */
                   1132:        subblksperindir = blksperindir / NINDIR(fs);
                   1133:        for (lbn++, level--, i = 0; i < last; i++) {
1.76      hannken  1134:                error = indiracct(snapvp, cancelvp, level,
                   1135:                    idb_get(VTOI(snapvp), bap, i), lbn, rlbn, remblks,
                   1136:                    subblksperindir, fs, acctfunc, expungetype);
1.1       hannken  1137:                if (error)
                   1138:                        goto out;
                   1139:                rlbn += blksperindir;
                   1140:                lbn -= blksperindir;
                   1141:                remblks -= blksperindir;
                   1142:        }
                   1143: out:
                   1144:        FREE(bap, M_DEVBUF);
                   1145:        return (error);
                   1146: }
                   1147:
                   1148: /*
                   1149:  * Do both snap accounting and map accounting.
                   1150:  */
                   1151: static int
1.76      hannken  1152: fullacct(struct vnode *vp, void *bap, int oldblkp, int lastblkp,
                   1153:     struct fs *fs, daddr_t lblkno,
1.18      thorpej  1154:     int exptype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1155: {
                   1156:        int error;
                   1157:
1.76      hannken  1158:        if ((error = snapacct(vp, bap, oldblkp, lastblkp, fs, lblkno, exptype)))
1.1       hannken  1159:                return (error);
1.76      hannken  1160:        return (mapacct(vp, bap, oldblkp, lastblkp, fs, lblkno, exptype));
1.1       hannken  1161: }
                   1162:
                   1163: /*
                   1164:  * Identify a set of blocks allocated in a snapshot inode.
                   1165:  */
                   1166: static int
1.76      hannken  1167: snapacct(struct vnode *vp, void *bap, int oldblkp, int lastblkp,
                   1168:     struct fs *fs, daddr_t lblkno,
1.18      thorpej  1169:     int expungetype /* BLK_SNAP or BLK_NOCOPY */)
1.1       hannken  1170: {
                   1171:        struct inode *ip = VTOI(vp);
1.79      hannken  1172:        struct lwp *l = curlwp;
1.76      hannken  1173:        daddr_t blkno;
                   1174:        daddr_t lbn;
1.1       hannken  1175:        struct buf *ibp;
                   1176:        int error;
                   1177:
                   1178:        for ( ; oldblkp < lastblkp; oldblkp++) {
1.76      hannken  1179:                blkno = idb_get(ip, bap, oldblkp);
1.1       hannken  1180:                if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
                   1181:                        continue;
                   1182:                lbn = fragstoblks(fs, blkno);
                   1183:                if (lbn < NDADDR) {
1.76      hannken  1184:                        blkno = db_get(ip, lbn);
1.1       hannken  1185:                        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1186:                } else {
1.22      yamt     1187:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.79      hannken  1188:                            fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
1.1       hannken  1189:                        if (error)
                   1190:                                return (error);
1.76      hannken  1191:                        blkno = idb_get(ip, ibp->b_data,
                   1192:                            (lbn - NDADDR) % NINDIR(fs));
1.1       hannken  1193:                }
                   1194:                /*
                   1195:                 * If we are expunging a snapshot vnode and we
                   1196:                 * find a block marked BLK_NOCOPY, then it is
                   1197:                 * one that has been allocated to this snapshot after
                   1198:                 * we took our current snapshot and can be ignored.
                   1199:                 */
                   1200:                if (expungetype == BLK_SNAP && blkno == BLK_NOCOPY) {
                   1201:                        if (lbn >= NDADDR)
1.52      ad       1202:                                brelse(ibp, 0);
1.1       hannken  1203:                } else {
                   1204:                        if (blkno != 0)
1.76      hannken  1205:                                panic("snapacct: bad block");
                   1206:                        if (lbn < NDADDR)
                   1207:                                db_assign(ip, lbn, expungetype);
                   1208:                        else {
                   1209:                                idb_assign(ip, ibp->b_data,
                   1210:                                    (lbn - NDADDR) % NINDIR(fs), expungetype);
1.15      hannken  1211:                                bdwrite(ibp);
1.76      hannken  1212:                        }
1.1       hannken  1213:                }
                   1214:        }
                   1215:        return (0);
                   1216: }
                   1217:
                   1218: /*
                   1219:  * Account for a set of blocks allocated in a snapshot inode.
                   1220:  */
                   1221: static int
1.76      hannken  1222: mapacct(struct vnode *vp, void *bap, int oldblkp, int lastblkp,
                   1223:     struct fs *fs, daddr_t lblkno, int expungetype)
1.1       hannken  1224: {
1.76      hannken  1225:        daddr_t blkno;
1.1       hannken  1226:        struct inode *ip;
                   1227:        ino_t inum;
1.76      hannken  1228:        int acctit;
1.1       hannken  1229:
                   1230:        ip = VTOI(vp);
                   1231:        inum = ip->i_number;
                   1232:        if (lblkno == -1)
                   1233:                acctit = 0;
                   1234:        else
                   1235:                acctit = 1;
                   1236:        for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
1.76      hannken  1237:                blkno = idb_get(ip, bap, oldblkp);
1.1       hannken  1238:                if (blkno == 0 || blkno == BLK_NOCOPY)
                   1239:                        continue;
                   1240:                if (acctit && expungetype == BLK_SNAP && blkno != BLK_SNAP)
1.15      hannken  1241:                        *ip->i_snapblklist++ = lblkno;
1.1       hannken  1242:                if (blkno == BLK_SNAP)
                   1243:                        blkno = blkstofrags(fs, lblkno);
1.83      joerg    1244:                ffs_blkfree_snap(fs, vp, blkno, fs->fs_bsize, inum);
1.1       hannken  1245:        }
                   1246:        return (0);
                   1247: }
1.10      hannken  1248: #endif /* defined(FFS_NO_SNAPSHOT) */
1.1       hannken  1249:
                   1250: /*
                   1251:  * Decrement extra reference on snapshot when last name is removed.
                   1252:  * It will not be freed until the last open reference goes away.
                   1253:  */
                   1254: void
1.18      thorpej  1255: ffs_snapgone(struct inode *ip)
1.1       hannken  1256: {
1.47      hannken  1257:        struct mount *mp = ip->i_devvp->v_specmountpoint;
1.1       hannken  1258:        struct inode *xp;
                   1259:        struct fs *fs;
1.47      hannken  1260:        struct snap_info *si;
1.1       hannken  1261:        int snaploc;
                   1262:
1.66      hannken  1263:        si = VFSTOUFS(mp)->um_snapinfo;
                   1264:
1.1       hannken  1265:        /*
                   1266:         * Find snapshot in incore list.
                   1267:         */
1.49      hannken  1268:        mutex_enter(&si->si_lock);
1.47      hannken  1269:        TAILQ_FOREACH(xp, &si->si_snapshots, i_nextsnap)
1.1       hannken  1270:                if (xp == ip)
                   1271:                        break;
1.49      hannken  1272:        mutex_exit(&si->si_lock);
1.1       hannken  1273:        if (xp != NULL)
                   1274:                vrele(ITOV(ip));
                   1275: #ifdef DEBUG
                   1276:        else if (snapdebug)
1.19      christos 1277:                printf("ffs_snapgone: lost snapshot vnode %llu\n",
                   1278:                    (unsigned long long)ip->i_number);
1.1       hannken  1279: #endif
                   1280:        /*
                   1281:         * Delete snapshot inode from superblock. Keep list dense.
                   1282:         */
1.49      hannken  1283:        mutex_enter(&si->si_lock);
1.1       hannken  1284:        fs = ip->i_fs;
                   1285:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++)
                   1286:                if (fs->fs_snapinum[snaploc] == ip->i_number)
                   1287:                        break;
                   1288:        if (snaploc < FSMAXSNAP) {
                   1289:                for (snaploc++; snaploc < FSMAXSNAP; snaploc++) {
                   1290:                        if (fs->fs_snapinum[snaploc] == 0)
                   1291:                                break;
                   1292:                        fs->fs_snapinum[snaploc - 1] = fs->fs_snapinum[snaploc];
                   1293:                }
                   1294:                fs->fs_snapinum[snaploc - 1] = 0;
                   1295:        }
1.49      hannken  1296:        si->si_gen++;
                   1297:        mutex_exit(&si->si_lock);
1.1       hannken  1298: }
                   1299:
                   1300: /*
                   1301:  * Prepare a snapshot file for being removed.
                   1302:  */
                   1303: void
1.18      thorpej  1304: ffs_snapremove(struct vnode *vp)
1.1       hannken  1305: {
1.15      hannken  1306:        struct inode *ip = VTOI(vp), *xp;
1.1       hannken  1307:        struct vnode *devvp = ip->i_devvp;
                   1308:        struct fs *fs = ip->i_fs;
1.47      hannken  1309:        struct mount *mp = devvp->v_specmountpoint;
1.1       hannken  1310:        struct buf *ibp;
1.47      hannken  1311:        struct snap_info *si;
1.79      hannken  1312:        struct lwp *l = curlwp;
                   1313:        daddr_t numblks, blkno, dblk;
1.76      hannken  1314:        int error, loc, last;
1.1       hannken  1315:
1.66      hannken  1316:        si = VFSTOUFS(mp)->um_snapinfo;
1.74      hannken  1317:        mutex_enter(&si->si_snaplock);
1.1       hannken  1318:        /*
                   1319:         * If active, delete from incore list (this snapshot may
                   1320:         * already have been in the process of being deleted, so
                   1321:         * would not have been active).
                   1322:         *
                   1323:         * Clear copy-on-write flag if last snapshot.
                   1324:         */
                   1325:        if (ip->i_nextsnap.tqe_prev != 0) {
1.49      hannken  1326:                mutex_enter(&si->si_lock);
1.47      hannken  1327:                TAILQ_REMOVE(&si->si_snapshots, ip, i_nextsnap);
1.1       hannken  1328:                ip->i_nextsnap.tqe_prev = 0;
1.47      hannken  1329:                if (TAILQ_FIRST(&si->si_snapshots) != 0) {
1.15      hannken  1330:                        /* Roll back the list of preallocated blocks. */
1.47      hannken  1331:                        xp = TAILQ_LAST(&si->si_snapshots, inodelst);
                   1332:                        si->si_snapblklist = xp->i_snapblklist;
1.1       hannken  1333:                } else {
1.47      hannken  1334:                        si->si_snapblklist = 0;
1.49      hannken  1335:                        si->si_gen++;
                   1336:                        mutex_exit(&si->si_lock);
1.51      hannken  1337:                        fscow_disestablish(mp, ffs_copyonwrite, devvp);
1.49      hannken  1338:                        mutex_enter(&si->si_lock);
1.1       hannken  1339:                }
1.49      hannken  1340:                si->si_gen++;
                   1341:                mutex_exit(&si->si_lock);
1.79      hannken  1342:                if (ip->i_snapblklist != NULL) {
                   1343:                        free(ip->i_snapblklist, M_UFSMNT);
                   1344:                        ip->i_snapblklist = NULL;
                   1345:                }
1.1       hannken  1346:        }
1.74      hannken  1347:        mutex_exit(&si->si_snaplock);
1.1       hannken  1348:        /*
                   1349:         * Clear all BLK_NOCOPY fields. Pass any block claims to other
                   1350:         * snapshots that want them (see ffs_snapblkfree below).
                   1351:         */
                   1352:        for (blkno = 1; blkno < NDADDR; blkno++) {
                   1353:                dblk = db_get(ip, blkno);
                   1354:                if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
                   1355:                        db_assign(ip, blkno, 0);
                   1356:                else if ((dblk == blkstofrags(fs, blkno) &&
                   1357:                     ffs_snapblkfree(fs, ip->i_devvp, dblk, fs->fs_bsize,
                   1358:                     ip->i_number))) {
                   1359:                        DIP_ADD(ip, blocks, -btodb(fs->fs_bsize));
                   1360:                        db_assign(ip, blkno, 0);
                   1361:                }
                   1362:        }
                   1363:        numblks = howmany(ip->i_size, fs->fs_bsize);
                   1364:        for (blkno = NDADDR; blkno < numblks; blkno += NINDIR(fs)) {
1.22      yamt     1365:                error = ffs_balloc(vp, lblktosize(fs, (off_t)blkno),
1.79      hannken  1366:                    fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
1.1       hannken  1367:                if (error)
                   1368:                        continue;
                   1369:                if (fs->fs_size - blkno > NINDIR(fs))
                   1370:                        last = NINDIR(fs);
                   1371:                else
                   1372:                        last = fs->fs_size - blkno;
                   1373:                for (loc = 0; loc < last; loc++) {
                   1374:                        dblk = idb_get(ip, ibp->b_data, loc);
                   1375:                        if (dblk == BLK_NOCOPY || dblk == BLK_SNAP)
                   1376:                                idb_assign(ip, ibp->b_data, loc, 0);
                   1377:                        else if (dblk == blkstofrags(fs, blkno) &&
                   1378:                            ffs_snapblkfree(fs, ip->i_devvp, dblk,
                   1379:                            fs->fs_bsize, ip->i_number)) {
                   1380:                                DIP_ADD(ip, blocks, -btodb(fs->fs_bsize));
                   1381:                                idb_assign(ip, ibp->b_data, loc, 0);
                   1382:                        }
                   1383:                }
1.15      hannken  1384:                bawrite(ibp);
1.1       hannken  1385:        }
                   1386:        /*
                   1387:         * Clear snapshot flag and drop reference.
                   1388:         */
                   1389:        ip->i_flags &= ~SF_SNAPSHOT;
                   1390:        DIP_ASSIGN(ip, flags, ip->i_flags);
                   1391:        ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1392: }
                   1393:
                   1394: /*
                   1395:  * Notification that a block is being freed. Return zero if the free
                   1396:  * should be allowed to proceed. Return non-zero if the snapshot file
                   1397:  * wants to claim the block. The block will be claimed if it is an
                   1398:  * uncopied part of one of the snapshots. It will be freed if it is
                   1399:  * either a BLK_NOCOPY or has already been copied in all of the snapshots.
                   1400:  * If a fragment is being freed, then all snapshots that care about
                   1401:  * it must make a copy since a snapshot file can only claim full sized
                   1402:  * blocks. Note that if more than one snapshot file maps the block,
                   1403:  * we can pick one at random to claim it. Since none of the snapshots
                   1404:  * can change, we are assurred that they will all see the same unmodified
                   1405:  * image. When deleting a snapshot file (see ffs_snapremove above), we
                   1406:  * must push any of these claimed blocks to one of the other snapshots
                   1407:  * that maps it. These claimed blocks are easily identified as they will
                   1408:  * have a block number equal to their logical block number within the
                   1409:  * snapshot. A copied block can never have this property because they
                   1410:  * must always have been allocated from a BLK_NOCOPY location.
                   1411:  */
                   1412: int
1.79      hannken  1413: ffs_snapblkfree(struct fs *fs, struct vnode *devvp, daddr_t bno,
1.36      christos 1414:     long size, ino_t inum)
1.1       hannken  1415: {
1.47      hannken  1416:        struct mount *mp = devvp->v_specmountpoint;
1.1       hannken  1417:        struct buf *ibp;
                   1418:        struct inode *ip;
1.49      hannken  1419:        struct vnode *vp = NULL;
1.47      hannken  1420:        struct snap_info *si;
1.43      christos 1421:        void *saved_data = NULL;
1.79      hannken  1422:        daddr_t lbn;
                   1423:        daddr_t blkno;
1.49      hannken  1424:        uint32_t gen;
1.67      hannken  1425:        int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
1.1       hannken  1426:
1.66      hannken  1427:        si = VFSTOUFS(mp)->um_snapinfo;
1.1       hannken  1428:        lbn = fragstoblks(fs, bno);
1.49      hannken  1429:        mutex_enter(&si->si_lock);
1.1       hannken  1430: retry:
1.49      hannken  1431:        gen = si->si_gen;
1.47      hannken  1432:        TAILQ_FOREACH(ip, &si->si_snapshots, i_nextsnap) {
1.1       hannken  1433:                vp = ITOV(ip);
1.49      hannken  1434:                if (snapshot_locked == 0) {
1.74      hannken  1435:                        if (!mutex_tryenter(&si->si_snaplock)) {
1.64      ad       1436:                                mutex_exit(&si->si_lock);
1.74      hannken  1437:                                mutex_enter(&si->si_snaplock);
1.49      hannken  1438:                                mutex_enter(&si->si_lock);
                   1439:                        }
                   1440:                        snapshot_locked = 1;
                   1441:                        if (gen != si->si_gen)
                   1442:                                goto retry;
                   1443:                }
1.1       hannken  1444:                /*
                   1445:                 * Lookup block being written.
                   1446:                 */
                   1447:                if (lbn < NDADDR) {
                   1448:                        blkno = db_get(ip, lbn);
                   1449:                } else {
1.49      hannken  1450:                        mutex_exit(&si->si_lock);
1.22      yamt     1451:                        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
1.79      hannken  1452:                            fs->fs_bsize, FSCRED, B_METAONLY, &ibp);
1.49      hannken  1453:                        if (error) {
                   1454:                                mutex_enter(&si->si_lock);
1.1       hannken  1455:                                break;
1.49      hannken  1456:                        }
1.1       hannken  1457:                        indiroff = (lbn - NDADDR) % NINDIR(fs);
                   1458:                        blkno = idb_get(ip, ibp->b_data, indiroff);
1.49      hannken  1459:                        mutex_enter(&si->si_lock);
                   1460:                        if (gen != si->si_gen) {
1.52      ad       1461:                                brelse(ibp, 0);
1.49      hannken  1462:                                goto retry;
                   1463:                        }
1.1       hannken  1464:                }
                   1465:                /*
                   1466:                 * Check to see if block needs to be copied.
                   1467:                 */
                   1468:                if (blkno == 0) {
                   1469:                        /*
                   1470:                         * A block that we map is being freed. If it has not
                   1471:                         * been claimed yet, we will claim or copy it (below).
                   1472:                         */
                   1473:                        claimedblk = 1;
                   1474:                } else if (blkno == BLK_SNAP) {
                   1475:                        /*
                   1476:                         * No previous snapshot claimed the block,
                   1477:                         * so it will be freed and become a BLK_NOCOPY
                   1478:                         * (don't care) for us.
                   1479:                         */
                   1480:                        if (claimedblk)
                   1481:                                panic("snapblkfree: inconsistent block type");
                   1482:                        if (lbn < NDADDR) {
                   1483:                                db_assign(ip, lbn, BLK_NOCOPY);
                   1484:                                ip->i_flag |= IN_CHANGE | IN_UPDATE;
                   1485:                        } else {
                   1486:                                idb_assign(ip, ibp->b_data, indiroff,
                   1487:                                    BLK_NOCOPY);
1.49      hannken  1488:                                mutex_exit(&si->si_lock);
1.72      hannken  1489:                                if (ip->i_ffs_effnlink > 0)
                   1490:                                        bwrite(ibp);
                   1491:                                else
                   1492:                                        bdwrite(ibp);
1.49      hannken  1493:                                mutex_enter(&si->si_lock);
                   1494:                                if (gen != si->si_gen)
                   1495:                                        goto retry;
1.1       hannken  1496:                        }
                   1497:                        continue;
                   1498:                } else /* BLK_NOCOPY or default */ {
                   1499:                        /*
                   1500:                         * If the snapshot has already copied the block
                   1501:                         * (default), or does not care about the block,
                   1502:                         * it is not needed.
                   1503:                         */
1.27      bouyer   1504:                        if (lbn >= NDADDR)
1.52      ad       1505:                                brelse(ibp, 0);
1.1       hannken  1506:                        continue;
                   1507:                }
                   1508:                /*
                   1509:                 * If this is a full size block, we will just grab it
                   1510:                 * and assign it to the snapshot inode. Otherwise we
                   1511:                 * will proceed to copy it. See explanation for this
                   1512:                 * routine as to why only a single snapshot needs to
                   1513:                 * claim this block.
                   1514:                 */
                   1515:                if (size == fs->fs_bsize) {
                   1516: #ifdef DEBUG
                   1517:                        if (snapdebug)
1.19      christos 1518:                                printf("%s %llu lbn %" PRId64
                   1519:                                    "from inum %llu\n",
                   1520:                                    "Grabonremove: snapino",
                   1521:                                    (unsigned long long)ip->i_number,
                   1522:                                    lbn, (unsigned long long)inum);
1.1       hannken  1523: #endif
1.49      hannken  1524:                        mutex_exit(&si->si_lock);
1.1       hannken  1525:                        if (lbn < NDADDR) {
                   1526:                                db_assign(ip, lbn, bno);
                   1527:                        } else {
                   1528:                                idb_assign(ip, ibp->b_data, indiroff, bno);
1.72      hannken  1529:                                if (ip->i_ffs_effnlink > 0)
                   1530:                                        bwrite(ibp);
                   1531:                                else
                   1532:                                        bdwrite(ibp);
1.1       hannken  1533:                        }
                   1534:                        DIP_ADD(ip, blocks, btodb(size));
                   1535:                        ip->i_flag |= IN_CHANGE | IN_UPDATE;
1.75      hannken  1536:                        if (ip->i_ffs_effnlink > 0 && mp->mnt_wapbl)
                   1537:                                error = syncsnap(vp);
                   1538:                        else
                   1539:                                error = 0;
1.74      hannken  1540:                        mutex_exit(&si->si_snaplock);
1.75      hannken  1541:                        return (error == 0);
1.1       hannken  1542:                }
                   1543:                if (lbn >= NDADDR)
1.52      ad       1544:                        brelse(ibp, 0);
1.1       hannken  1545: #ifdef DEBUG
                   1546:                if (snapdebug)
1.19      christos 1547:                        printf("%s%llu lbn %" PRId64 " %s %llu size %ld\n",
                   1548:                            "Copyonremove: snapino ",
                   1549:                            (unsigned long long)ip->i_number,
                   1550:                            lbn, "for inum", (unsigned long long)inum, size);
1.1       hannken  1551: #endif
                   1552:                /*
                   1553:                 * If we have already read the old block contents, then
                   1554:                 * simply copy them to the new block. Note that we need
                   1555:                 * to synchronously write snapshots that have not been
                   1556:                 * unlinked, and hence will be visible after a crash,
                   1557:                 * to ensure their integrity.
                   1558:                 */
1.49      hannken  1559:                mutex_exit(&si->si_lock);
                   1560:                if (saved_data == NULL) {
                   1561:                        saved_data = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.75      hannken  1562:                        error = rwfsblk(vp, B_READ, saved_data, lbn);
1.72      hannken  1563:                        if (error) {
1.49      hannken  1564:                                free(saved_data, M_UFSMNT);
                   1565:                                saved_data = NULL;
                   1566:                                mutex_enter(&si->si_lock);
1.1       hannken  1567:                                break;
1.49      hannken  1568:                        }
1.1       hannken  1569:                }
1.72      hannken  1570:                error = wrsnapblk(vp, saved_data, lbn);
1.75      hannken  1571:                if (error == 0 && ip->i_ffs_effnlink > 0 && mp->mnt_wapbl)
                   1572:                        error = syncsnap(vp);
1.49      hannken  1573:                mutex_enter(&si->si_lock);
                   1574:                if (error)
1.1       hannken  1575:                        break;
1.49      hannken  1576:                if (gen != si->si_gen)
                   1577:                        goto retry;
1.1       hannken  1578:        }
1.49      hannken  1579:        mutex_exit(&si->si_lock);
                   1580:        if (saved_data)
1.1       hannken  1581:                free(saved_data, M_UFSMNT);
                   1582:        /*
                   1583:         * If we have been unable to allocate a block in which to do
                   1584:         * the copy, then return non-zero so that the fragment will
                   1585:         * not be freed. Although space will be lost, the snapshot
                   1586:         * will stay consistent.
                   1587:         */
                   1588:        if (snapshot_locked)
1.74      hannken  1589:                mutex_exit(&si->si_snaplock);
1.1       hannken  1590:        return (error);
                   1591: }
                   1592:
                   1593: /*
                   1594:  * Associate snapshot files when mounting.
                   1595:  */
                   1596: void
1.18      thorpej  1597: ffs_snapshot_mount(struct mount *mp)
1.1       hannken  1598: {
1.47      hannken  1599:        struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
                   1600:        struct fs *fs = VFSTOUFS(mp)->um_fs;
1.31      ad       1601:        struct lwp *l = curlwp;
1.1       hannken  1602:        struct vnode *vp;
                   1603:        struct inode *ip, *xp;
1.47      hannken  1604:        struct snap_info *si;
1.79      hannken  1605:        daddr_t snaplistsize, *snapblklist;
1.15      hannken  1606:        int i, error, ns, snaploc, loc;
1.1       hannken  1607:
1.46      hannken  1608:        /*
                   1609:         * No persistent snapshots on apple ufs file systems.
                   1610:         */
1.47      hannken  1611:        if (UFS_MPISAPPLEUFS(VFSTOUFS(mp)))
1.46      hannken  1612:                return;
                   1613:
1.66      hannken  1614:        si = VFSTOUFS(mp)->um_snapinfo;
1.1       hannken  1615:        ns = UFS_FSNEEDSWAP(fs);
                   1616:        /*
1.22      yamt     1617:         * XXX The following needs to be set before ffs_truncate or
1.1       hannken  1618:         * VOP_READ can be called.
                   1619:         */
                   1620:        mp->mnt_stat.f_iosize = fs->fs_bsize;
                   1621:        /*
                   1622:         * Process each snapshot listed in the superblock.
                   1623:         */
                   1624:        vp = NULL;
1.49      hannken  1625:        mutex_enter(&si->si_lock);
1.1       hannken  1626:        for (snaploc = 0; snaploc < FSMAXSNAP; snaploc++) {
                   1627:                if (fs->fs_snapinum[snaploc] == 0)
                   1628:                        break;
                   1629:                if ((error = VFS_VGET(mp, fs->fs_snapinum[snaploc],
                   1630:                    &vp)) != 0) {
                   1631:                        printf("ffs_snapshot_mount: vget failed %d\n", error);
                   1632:                        continue;
                   1633:                }
                   1634:                ip = VTOI(vp);
                   1635:                if ((ip->i_flags & SF_SNAPSHOT) == 0) {
                   1636:                        printf("ffs_snapshot_mount: non-snapshot inode %d\n",
                   1637:                            fs->fs_snapinum[snaploc]);
                   1638:                        vput(vp);
                   1639:                        vp = NULL;
                   1640:                        for (loc = snaploc + 1; loc < FSMAXSNAP; loc++) {
                   1641:                                if (fs->fs_snapinum[loc] == 0)
                   1642:                                        break;
                   1643:                                fs->fs_snapinum[loc - 1] = fs->fs_snapinum[loc];
                   1644:                        }
                   1645:                        fs->fs_snapinum[loc - 1] = 0;
                   1646:                        snaploc--;
                   1647:                        continue;
                   1648:                }
1.15      hannken  1649:
                   1650:                /*
                   1651:                 * Read the block hints list. Use an empty list on
                   1652:                 * read errors.
                   1653:                 */
                   1654:                error = vn_rdwr(UIO_READ, vp,
1.43      christos 1655:                    (void *)&snaplistsize, sizeof(snaplistsize),
1.15      hannken  1656:                    lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)),
1.87    ! hannken  1657:                    UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT|IO_ALTSEMANTICS,
1.31      ad       1658:                    l->l_cred, NULL, NULL);
1.15      hannken  1659:                if (error) {
                   1660:                        printf("ffs_snapshot_mount: read_1 failed %d\n", error);
                   1661:                        snaplistsize = 1;
                   1662:                } else
                   1663:                        snaplistsize = ufs_rw64(snaplistsize, ns);
1.25      christos 1664:                snapblklist = malloc(
1.79      hannken  1665:                    snaplistsize * sizeof(daddr_t), M_UFSMNT, M_WAITOK);
1.15      hannken  1666:                if (error)
                   1667:                        snapblklist[0] = 1;
                   1668:                else {
1.43      christos 1669:                        error = vn_rdwr(UIO_READ, vp, (void *)snapblklist,
1.79      hannken  1670:                            snaplistsize * sizeof(daddr_t),
1.15      hannken  1671:                            lblktosize(fs, howmany(fs->fs_size, fs->fs_frag)),
1.87    ! hannken  1672:                            UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT|IO_ALTSEMANTICS,
1.31      ad       1673:                            l->l_cred, NULL, NULL);
1.15      hannken  1674:                        for (i = 0; i < snaplistsize; i++)
                   1675:                                snapblklist[i] = ufs_rw64(snapblklist[i], ns);
                   1676:                        if (error) {
                   1677:                                printf("ffs_snapshot_mount: read_2 failed %d\n",
                   1678:                                    error);
                   1679:                                snapblklist[0] = 1;
                   1680:                        }
                   1681:                }
                   1682:                ip->i_snapblklist = &snapblklist[0];
                   1683:
1.1       hannken  1684:                /*
                   1685:                 * Link it onto the active snapshot list.
                   1686:                 */
                   1687:                if (ip->i_nextsnap.tqe_prev != 0)
1.19      christos 1688:                        panic("ffs_snapshot_mount: %llu already on list",
                   1689:                            (unsigned long long)ip->i_number);
1.1       hannken  1690:                else
1.47      hannken  1691:                        TAILQ_INSERT_TAIL(&si->si_snapshots, ip, i_nextsnap);
1.53      ad       1692:                vp->v_vflag |= VV_SYSTEM;
1.1       hannken  1693:                VOP_UNLOCK(vp, 0);
                   1694:        }
                   1695:        /*
                   1696:         * No usable snapshots found.
                   1697:         */
1.49      hannken  1698:        if (vp == NULL) {
                   1699:                mutex_exit(&si->si_lock);
1.1       hannken  1700:                return;
1.49      hannken  1701:        }
1.1       hannken  1702:        /*
1.15      hannken  1703:         * Attach the block hints list. We always want to
1.1       hannken  1704:         * use the list from the newest snapshot.
1.15      hannken  1705:        */
1.47      hannken  1706:        xp = TAILQ_LAST(&si->si_snapshots, inodelst);
                   1707:        si->si_snapblklist = xp->i_snapblklist;
1.51      hannken  1708:        fscow_establish(mp, ffs_copyonwrite, devvp);
1.49      hannken  1709:        si->si_gen++;
                   1710:        mutex_exit(&si->si_lock);
1.1       hannken  1711: }
                   1712:
                   1713: /*
                   1714:  * Disassociate snapshot files when unmounting.
                   1715:  */
                   1716: void
1.18      thorpej  1717: ffs_snapshot_unmount(struct mount *mp)
1.1       hannken  1718: {
1.47      hannken  1719:        struct vnode *devvp = VFSTOUFS(mp)->um_devvp;
1.1       hannken  1720:        struct inode *xp;
1.49      hannken  1721:        struct vnode *vp = NULL;
1.47      hannken  1722:        struct snap_info *si;
1.1       hannken  1723:
1.66      hannken  1724:        si = VFSTOUFS(mp)->um_snapinfo;
1.49      hannken  1725:        mutex_enter(&si->si_lock);
1.47      hannken  1726:        while ((xp = TAILQ_FIRST(&si->si_snapshots)) != 0) {
1.1       hannken  1727:                vp = ITOV(xp);
                   1728:                vp->v_vnlock = &vp->v_lock;
1.47      hannken  1729:                TAILQ_REMOVE(&si->si_snapshots, xp, i_nextsnap);
1.1       hannken  1730:                xp->i_nextsnap.tqe_prev = 0;
1.47      hannken  1731:                if (xp->i_snapblklist == si->si_snapblklist)
                   1732:                        si->si_snapblklist = NULL;
1.15      hannken  1733:                FREE(xp->i_snapblklist, M_UFSMNT);
1.49      hannken  1734:                if (xp->i_ffs_effnlink > 0) {
                   1735:                        si->si_gen++;
                   1736:                        mutex_exit(&si->si_lock);
1.1       hannken  1737:                        vrele(vp);
1.49      hannken  1738:                        mutex_enter(&si->si_lock);
                   1739:                }
1.1       hannken  1740:        }
1.49      hannken  1741:        if (vp)
1.51      hannken  1742:                fscow_disestablish(mp, ffs_copyonwrite, devvp);
1.49      hannken  1743:        si->si_gen++;
                   1744:        mutex_exit(&si->si_lock);
1.1       hannken  1745: }
                   1746:
                   1747: /*
                   1748:  * Check for need to copy block that is about to be written,
                   1749:  * copying the block if necessary.
                   1750:  */
                   1751: static int
1.55      hannken  1752: ffs_copyonwrite(void *v, struct buf *bp, bool data_valid)
1.1       hannken  1753: {
                   1754:        struct fs *fs;
                   1755:        struct inode *ip;
1.49      hannken  1756:        struct vnode *devvp = v, *vp = NULL;
1.47      hannken  1757:        struct mount *mp = devvp->v_specmountpoint;
                   1758:        struct snap_info *si;
1.43      christos 1759:        void *saved_data = NULL;
1.79      hannken  1760:        daddr_t lbn, blkno, *snapblklist;
1.49      hannken  1761:        uint32_t gen;
1.76      hannken  1762:        int lower, upper, mid, snapshot_locked = 0, error = 0;
1.1       hannken  1763:
                   1764:        /*
                   1765:         * Check for valid snapshots.
                   1766:         */
1.66      hannken  1767:        si = VFSTOUFS(mp)->um_snapinfo;
1.49      hannken  1768:        mutex_enter(&si->si_lock);
1.47      hannken  1769:        ip = TAILQ_FIRST(&si->si_snapshots);
1.1       hannken  1770:        if (ip == NULL) {
1.49      hannken  1771:                mutex_exit(&si->si_lock);
1.11      perry    1772:                return 0;
1.1       hannken  1773:        }
                   1774:        /*
1.75      hannken  1775:         * First check to see if it is after the file system or
                   1776:         * in the preallocated list.
1.1       hannken  1777:         * By doing this check we avoid several potential deadlocks.
                   1778:         */
                   1779:        fs = ip->i_fs;
                   1780:        lbn = fragstoblks(fs, dbtofsb(fs, bp->b_blkno));
1.75      hannken  1781:        if (bp->b_blkno >= fsbtodb(fs, fs->fs_size)) {
                   1782:                mutex_exit(&si->si_lock);
                   1783:                return 0;
                   1784:        }
1.47      hannken  1785:        snapblklist = si->si_snapblklist;
1.87    ! hannken  1786:        upper = (snapblklist != NULL ? snapblklist[0] - 1 : 0);
1.1       hannken  1787:        lower = 1;
                   1788:        while (lower <= upper) {
                   1789:                mid = (lower + upper) / 2;
1.15      hannken  1790:                if (snapblklist[mid] == lbn)
1.1       hannken  1791:                        break;
1.15      hannken  1792:                if (snapblklist[mid] < lbn)
1.1       hannken  1793:                        lower = mid + 1;
                   1794:                else
                   1795:                        upper = mid - 1;
                   1796:        }
                   1797:        if (lower <= upper) {
1.49      hannken  1798:                mutex_exit(&si->si_lock);
1.1       hannken  1799:                return 0;
                   1800:        }
                   1801:        /*
                   1802:         * Not in the precomputed list, so check the snapshots.
                   1803:         */
1.55      hannken  1804:         if (data_valid && bp->b_bcount == fs->fs_bsize)
                   1805:                saved_data = bp->b_data;
1.1       hannken  1806: retry:
1.49      hannken  1807:        gen = si->si_gen;
1.47      hannken  1808:        TAILQ_FOREACH(ip, &si->si_snapshots, i_nextsnap) {
1.1       hannken  1809:                vp = ITOV(ip);
                   1810:                /*
                   1811:                 * We ensure that everything of our own that needs to be
                   1812:                 * copied will be done at the time that ffs_snapshot is
                   1813:                 * called. Thus we can skip the check here which can
1.22      yamt     1814:                 * deadlock in doing the lookup in ffs_balloc.
1.1       hannken  1815:                 */
                   1816:                if (bp->b_vp == vp)
                   1817:                        continue;
                   1818:                /*
1.68      hannken  1819:                 * Check to see if block needs to be copied.
1.1       hannken  1820:                 */
                   1821:                if (lbn < NDADDR) {
                   1822:                        blkno = db_get(ip, lbn);
                   1823:                } else {
1.49      hannken  1824:                        mutex_exit(&si->si_lock);
1.72      hannken  1825:                        if ((error = snapblkaddr(vp, lbn, &blkno)) != 0) {
1.49      hannken  1826:                                mutex_enter(&si->si_lock);
1.1       hannken  1827:                                break;
1.49      hannken  1828:                        }
                   1829:                        mutex_enter(&si->si_lock);
                   1830:                        if (gen != si->si_gen)
                   1831:                                goto retry;
1.1       hannken  1832:                }
                   1833: #ifdef DIAGNOSTIC
                   1834:                if (blkno == BLK_SNAP && bp->b_lblkno >= 0)
                   1835:                        panic("ffs_copyonwrite: bad copy block");
                   1836: #endif
                   1837:                if (blkno != 0)
                   1838:                        continue;
1.72      hannken  1839:
1.74      hannken  1840:                if (curlwp == uvm.pagedaemon_lwp) {
                   1841:                        error = ENOMEM;
                   1842:                        break;
                   1843:                }
                   1844:
1.72      hannken  1845:                if (snapshot_locked == 0) {
1.74      hannken  1846:                        if (!mutex_tryenter(&si->si_snaplock)) {
1.72      hannken  1847:                                mutex_exit(&si->si_lock);
1.74      hannken  1848:                                mutex_enter(&si->si_snaplock);
1.72      hannken  1849:                                mutex_enter(&si->si_lock);
                   1850:                        }
                   1851:                        snapshot_locked = 1;
                   1852:                        if (gen != si->si_gen)
                   1853:                                goto retry;
                   1854:
                   1855:                        /* Check again if block still needs to be copied */
                   1856:                        if (lbn < NDADDR) {
                   1857:                                blkno = db_get(ip, lbn);
                   1858:                        } else {
                   1859:                                mutex_exit(&si->si_lock);
                   1860:                                if ((error = snapblkaddr(vp, lbn, &blkno)) != 0) {
                   1861:                                        mutex_enter(&si->si_lock);
                   1862:                                        break;
                   1863:                                }
                   1864:                                mutex_enter(&si->si_lock);
                   1865:                                if (gen != si->si_gen)
                   1866:                                        goto retry;
                   1867:                        }
                   1868:
                   1869:                        if (blkno != 0)
                   1870:                                continue;
                   1871:                }
1.1       hannken  1872:                /*
                   1873:                 * Allocate the block into which to do the copy. Since
                   1874:                 * multiple processes may all try to copy the same block,
                   1875:                 * we have to recheck our need to do a copy if we sleep
                   1876:                 * waiting for the lock.
                   1877:                 *
                   1878:                 * Because all snapshots on a filesystem share a single
                   1879:                 * lock, we ensure that we will never be in competition
                   1880:                 * with another process to allocate a block.
                   1881:                 */
                   1882: #ifdef DEBUG
                   1883:                if (snapdebug) {
1.19      christos 1884:                        printf("Copyonwrite: snapino %llu lbn %" PRId64 " for ",
                   1885:                            (unsigned long long)ip->i_number, lbn);
1.1       hannken  1886:                        if (bp->b_vp == devvp)
                   1887:                                printf("fs metadata");
                   1888:                        else
1.19      christos 1889:                                printf("inum %llu", (unsigned long long)
                   1890:                                    VTOI(bp->b_vp)->i_number);
1.4       hannken  1891:                        printf(" lblkno %" PRId64 "\n", bp->b_lblkno);
1.1       hannken  1892:                }
                   1893: #endif
                   1894:                /*
                   1895:                 * If we have already read the old block contents, then
                   1896:                 * simply copy them to the new block. Note that we need
                   1897:                 * to synchronously write snapshots that have not been
                   1898:                 * unlinked, and hence will be visible after a crash,
                   1899:                 * to ensure their integrity.
                   1900:                 */
1.49      hannken  1901:                mutex_exit(&si->si_lock);
                   1902:                if (saved_data == NULL) {
                   1903:                        saved_data = malloc(fs->fs_bsize, M_UFSMNT, M_WAITOK);
1.75      hannken  1904:                        error = rwfsblk(vp, B_READ, saved_data, lbn);
1.72      hannken  1905:                        if (error) {
1.49      hannken  1906:                                free(saved_data, M_UFSMNT);
                   1907:                                saved_data = NULL;
                   1908:                                mutex_enter(&si->si_lock);
1.1       hannken  1909:                                break;
1.49      hannken  1910:                        }
1.1       hannken  1911:                }
1.72      hannken  1912:                error = wrsnapblk(vp, saved_data, lbn);
1.75      hannken  1913:                if (error == 0 && ip->i_ffs_effnlink > 0 && mp->mnt_wapbl)
                   1914:                        error = syncsnap(vp);
1.49      hannken  1915:                mutex_enter(&si->si_lock);
                   1916:                if (error)
1.1       hannken  1917:                        break;
1.49      hannken  1918:                if (gen != si->si_gen)
                   1919:                        goto retry;
1.1       hannken  1920:        }
                   1921:        /*
                   1922:         * Note that we need to synchronously write snapshots that
                   1923:         * have not been unlinked, and hence will be visible after
                   1924:         * a crash, to ensure their integrity.
                   1925:         */
1.49      hannken  1926:        mutex_exit(&si->si_lock);
1.55      hannken  1927:        if (saved_data && saved_data != bp->b_data)
1.1       hannken  1928:                free(saved_data, M_UFSMNT);
                   1929:        if (snapshot_locked)
1.74      hannken  1930:                mutex_exit(&si->si_snaplock);
                   1931:        return error;
                   1932: }
                   1933:
                   1934: /*
                   1935:  * Read from a snapshot.
                   1936:  */
                   1937: int
                   1938: ffs_snapshot_read(struct vnode *vp, struct uio *uio, int ioflag)
                   1939: {
                   1940:        struct inode *ip = VTOI(vp);
                   1941:        struct fs *fs = ip->i_fs;
                   1942:        struct snap_info *si = VFSTOUFS(vp->v_mount)->um_snapinfo;
                   1943:        struct buf *bp;
                   1944:        daddr_t lbn, nextlbn;
1.81      hannken  1945:        off_t fsbytes, bytesinfile;
1.74      hannken  1946:        long size, xfersize, blkoffset;
                   1947:        int error;
                   1948:
                   1949:        fstrans_start(vp->v_mount, FSTRANS_SHARED);
                   1950:        mutex_enter(&si->si_snaplock);
                   1951:
1.87    ! hannken  1952:        if (ioflag & IO_ALTSEMANTICS)
        !          1953:                fsbytes = ip->i_size;
        !          1954:        else
        !          1955:                fsbytes = lfragtosize(fs, fs->fs_size);
1.74      hannken  1956:        for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
1.81      hannken  1957:                bytesinfile = fsbytes - uio->uio_offset;
1.74      hannken  1958:                if (bytesinfile <= 0)
                   1959:                        break;
                   1960:                lbn = lblkno(fs, uio->uio_offset);
                   1961:                nextlbn = lbn + 1;
1.82      hannken  1962:                size = fs->fs_bsize;
1.74      hannken  1963:                blkoffset = blkoff(fs, uio->uio_offset);
                   1964:                xfersize = MIN(MIN(fs->fs_bsize - blkoffset, uio->uio_resid),
                   1965:                    bytesinfile);
                   1966:
1.82      hannken  1967:                if (lblktosize(fs, nextlbn + 1) >= fsbytes) {
                   1968:                        if (lblktosize(fs, lbn) + size > fsbytes)
1.87    ! hannken  1969:                                size = fragroundup(fs,
        !          1970:                                    fsbytes - lblktosize(fs, lbn));
1.74      hannken  1971:                        error = bread(vp, lbn, size, NOCRED, 0, &bp);
1.82      hannken  1972:                } else {
                   1973:                        int nextsize = fs->fs_bsize;
1.74      hannken  1974:                        error = breadn(vp, lbn,
                   1975:                            size, &nextlbn, &nextsize, 1, NOCRED, 0, &bp);
                   1976:                }
                   1977:                if (error)
                   1978:                        break;
                   1979:
                   1980:                /*
                   1981:                 * We should only get non-zero b_resid when an I/O error
                   1982:                 * has occurred, which should cause us to break above.
                   1983:                 * However, if the short read did not cause an error,
                   1984:                 * then we want to ensure that we do not uiomove bad
                   1985:                 * or uninitialized data.
                   1986:                 */
                   1987:                size -= bp->b_resid;
1.82      hannken  1988:                if (size < blkoffset + xfersize) {
                   1989:                        xfersize = size - blkoffset;
                   1990:                        if (xfersize <= 0)
1.74      hannken  1991:                                break;
                   1992:                }
                   1993:                error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
                   1994:                if (error)
                   1995:                        break;
1.75      hannken  1996:                brelse(bp, BC_AGE);
1.74      hannken  1997:        }
                   1998:        if (bp != NULL)
1.75      hannken  1999:                brelse(bp, BC_AGE);
1.74      hannken  2000:
                   2001:        mutex_exit(&si->si_snaplock);
                   2002:        fstrans_done(vp->v_mount);
1.1       hannken  2003:        return error;
                   2004: }
                   2005:
                   2006: /*
1.79      hannken  2007:  * Lookup a snapshots data block address.
                   2008:  * Simpler than UFS_BALLOC() as we know all metadata is already allocated
                   2009:  * and safe even for the pagedaemon where we cannot bread().
                   2010:  */
                   2011: static int
                   2012: snapblkaddr(struct vnode *vp, daddr_t lbn, daddr_t *res)
                   2013: {
                   2014:        struct indir indirs[NIADDR + 2];
                   2015:        struct inode *ip = VTOI(vp);
                   2016:        struct fs *fs = ip->i_fs;
                   2017:        struct buf *bp;
                   2018:        int error, num;
                   2019:
                   2020:        KASSERT(lbn >= 0);
                   2021:
                   2022:        if (lbn < NDADDR) {
                   2023:                *res = db_get(ip, lbn);
                   2024:                return 0;
                   2025:        }
                   2026:        if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
                   2027:                return error;
                   2028:        if (curlwp == uvm.pagedaemon_lwp) {
                   2029:                mutex_enter(&bufcache_lock);
                   2030:                bp = incore(vp, indirs[num-1].in_lbn);
                   2031:                if (bp && (bp->b_oflags & (BO_DONE | BO_DELWRI))) {
                   2032:                        *res = idb_get(ip, bp->b_data, indirs[num-1].in_off);
                   2033:                        error = 0;
                   2034:                } else
                   2035:                        error = ENOMEM;
                   2036:                mutex_exit(&bufcache_lock);
                   2037:                return error;
                   2038:        }
                   2039:        error = bread(vp, indirs[num-1].in_lbn, fs->fs_bsize, NOCRED, 0, &bp);
                   2040:        if (error == 0)
                   2041:                *res = idb_get(ip, bp->b_data, indirs[num-1].in_off);
                   2042:        brelse(bp, 0);
                   2043:
                   2044:        return error;
                   2045: }
                   2046:
                   2047: /*
1.75      hannken  2048:  * Read or write the specified block of the filesystem vp resides on
                   2049:  * from or to the disk bypassing the buffer cache.
1.1       hannken  2050:  */
                   2051: static int
1.79      hannken  2052: rwfsblk(struct vnode *vp, int flags, void *data, daddr_t lbn)
1.1       hannken  2053: {
1.24      yamt     2054:        int error;
1.1       hannken  2055:        struct inode *ip = VTOI(vp);
                   2056:        struct fs *fs = ip->i_fs;
                   2057:        struct buf *nbp;
                   2058:
1.57      ad       2059:        nbp = getiobuf(NULL, true);
1.75      hannken  2060:        nbp->b_flags = flags;
1.1       hannken  2061:        nbp->b_bcount = nbp->b_bufsize = fs->fs_bsize;
                   2062:        nbp->b_error = 0;
                   2063:        nbp->b_data = data;
                   2064:        nbp->b_blkno = nbp->b_rawblkno = fsbtodb(fs, blkstofrags(fs, lbn));
                   2065:        nbp->b_proc = NULL;
                   2066:        nbp->b_dev = ip->i_devvp->v_rdev;
1.70      reinoud  2067:        SET(nbp->b_cflags, BC_BUSY);    /* mark buffer busy */
1.1       hannken  2068:
1.44      ad       2069:        bdev_strategy(nbp);
1.1       hannken  2070:
                   2071:        error = biowait(nbp);
                   2072:
1.24      yamt     2073:        putiobuf(nbp);
1.1       hannken  2074:
                   2075:        return error;
                   2076: }
                   2077:
                   2078: /*
1.75      hannken  2079:  * Write all dirty buffers to disk and invalidate them.
                   2080:  */
                   2081: static int
                   2082: syncsnap(struct vnode *vp)
                   2083: {
                   2084:        int error;
                   2085:        buf_t *bp;
                   2086:        struct fs *fs = VTOI(vp)->i_fs;
                   2087:
                   2088:        mutex_enter(&bufcache_lock);
                   2089:        while ((bp = LIST_FIRST(&vp->v_dirtyblkhd))) {
                   2090:                KASSERT((bp->b_cflags & BC_BUSY) == 0);
                   2091:                KASSERT(bp->b_bcount == fs->fs_bsize);
                   2092:                bp->b_cflags |= BC_BUSY;
                   2093:                mutex_exit(&bufcache_lock);
                   2094:                error = rwfsblk(vp, B_WRITE, bp->b_data,
                   2095:                    fragstoblks(fs, dbtofsb(fs, bp->b_blkno)));
                   2096:                brelse(bp, BC_INVAL | BC_VFLUSH);
                   2097:                if (error)
                   2098:                        return error;
                   2099:                mutex_enter(&bufcache_lock);
                   2100:        }
                   2101:        mutex_exit(&bufcache_lock);
                   2102:
                   2103:        return 0;
                   2104: }
                   2105:
                   2106: /*
1.72      hannken  2107:  * Write the specified block to a snapshot.
1.1       hannken  2108:  */
                   2109: static int
1.79      hannken  2110: wrsnapblk(struct vnode *vp, void *data, daddr_t lbn)
1.1       hannken  2111: {
                   2112:        struct inode *ip = VTOI(vp);
                   2113:        struct fs *fs = ip->i_fs;
1.74      hannken  2114:        struct buf *bp;
                   2115:        int error;
1.1       hannken  2116:
1.74      hannken  2117:        error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
1.79      hannken  2118:            FSCRED, (ip->i_ffs_effnlink > 0 ? B_SYNC : 0), &bp);
1.74      hannken  2119:        if (error)
                   2120:                return error;
                   2121:        bcopy(data, bp->b_data, fs->fs_bsize);
1.72      hannken  2122:        if (ip->i_ffs_effnlink > 0)
1.74      hannken  2123:                error = bwrite(bp);
                   2124:        else
                   2125:                bawrite(bp);
1.4       hannken  2126:
1.72      hannken  2127:        return error;
1.4       hannken  2128: }
                   2129:
                   2130: /*
1.1       hannken  2131:  * Get/Put direct block from inode or buffer containing disk addresses. Take
                   2132:  * care for fs type (UFS1/UFS2) and byte swapping. These functions should go
                   2133:  * into a global include.
                   2134:  */
1.79      hannken  2135: static inline daddr_t
1.1       hannken  2136: db_get(struct inode *ip, int loc)
                   2137: {
                   2138:        if (ip->i_ump->um_fstype == UFS1)
1.2       hannken  2139:                return ufs_rw32(ip->i_ffs1_db[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2140:        else
1.2       hannken  2141:                return ufs_rw64(ip->i_ffs2_db[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2142: }
                   2143:
                   2144: static inline void
1.79      hannken  2145: db_assign(struct inode *ip, int loc, daddr_t val)
1.1       hannken  2146: {
                   2147:        if (ip->i_ump->um_fstype == UFS1)
1.2       hannken  2148:                ip->i_ffs1_db[loc] = ufs_rw32(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2149:        else
1.2       hannken  2150:                ip->i_ffs2_db[loc] = ufs_rw64(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2151: }
                   2152:
1.79      hannken  2153: static inline daddr_t
1.76      hannken  2154: ib_get(struct inode *ip, int loc)
                   2155: {
                   2156:        if (ip->i_ump->um_fstype == UFS1)
                   2157:                return ufs_rw32(ip->i_ffs1_ib[loc], UFS_IPNEEDSWAP(ip));
                   2158:        else
                   2159:                return ufs_rw64(ip->i_ffs2_ib[loc], UFS_IPNEEDSWAP(ip));
                   2160: }
                   2161:
                   2162: static inline void
1.79      hannken  2163: ib_assign(struct inode *ip, int loc, daddr_t val)
1.76      hannken  2164: {
                   2165:        if (ip->i_ump->um_fstype == UFS1)
                   2166:                ip->i_ffs1_ib[loc] = ufs_rw32(val, UFS_IPNEEDSWAP(ip));
                   2167:        else
                   2168:                ip->i_ffs2_ib[loc] = ufs_rw64(val, UFS_IPNEEDSWAP(ip));
                   2169: }
                   2170:
1.79      hannken  2171: static inline daddr_t
1.43      christos 2172: idb_get(struct inode *ip, void *bf, int loc)
1.1       hannken  2173: {
                   2174:        if (ip->i_ump->um_fstype == UFS1)
1.79      hannken  2175:                return ufs_rw32(((int32_t *)(bf))[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2176:        else
1.79      hannken  2177:                return ufs_rw64(((int64_t *)(bf))[loc], UFS_IPNEEDSWAP(ip));
1.1       hannken  2178: }
                   2179:
                   2180: static inline void
1.79      hannken  2181: idb_assign(struct inode *ip, void *bf, int loc, daddr_t val)
1.1       hannken  2182: {
                   2183:        if (ip->i_ump->um_fstype == UFS1)
1.79      hannken  2184:                ((int32_t *)(bf))[loc] = ufs_rw32(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2185:        else
1.79      hannken  2186:                ((int64_t *)(bf))[loc] = ufs_rw64(val, UFS_IPNEEDSWAP(ip));
1.1       hannken  2187: }

CVSweb <webmaster@jp.NetBSD.org>