Module Name: src
Committed By: bouyer
Date: Sat Jun 18 17:00:26 UTC 2011
Modified Files:
src/sbin/dump [netbsd-5]: snapshot.c
src/share/man/man4 [netbsd-5]: fss.4
src/sys/dev [netbsd-5]: fss.c fssvar.h
src/sys/kern [netbsd-5]: vfs_wapbl.c
src/sys/ufs/ffs [netbsd-5]: ffs_balloc.c ffs_snapshot.c
src/usr.sbin/fssconfig [netbsd-5]: fssconfig.c
Log Message:
Pull up following revision(s) (requested by hannken in ticket #1627):
sys/kern/vfs_wapbl.c: revisions 1.41-1.42
sbin/dump/snapshot.c: revisions 1.6 (patch)
share/man/man4/fss.4: revisions 1.15 (patch)
sys/dev/fss.c: revisions 1.73 (patch)
sys/dev/fssvar.h: revisions 1.25
usr.sbin/fssconfig/fssconfig.c: revisions 1.7
sys/ufs/ffs/ffs_balloc.c: revisions 1.54
sys/ufs/ffs/ffs_snapshot.c: revisions 1.90, 1.98,
1.100-1.101, 1.103-1.110, 1.111, 1.112-1.115 (patch)
- Try to keep snapshot indirect blocks contiguous. This speeds up snapshot
creation by a factor of ~3 and reduces the file system suspension time by
a factor of ~5.
- Refine the scope of WAPBL transactions and the limit for deallocations in
one transaction so we should no longer get a "wapbl_flush: current
transaction too big to flush" panic when creating or removing snapshots
on larger logging disks.
- fss(4): Allow FSSIOCSET to set the initial flags. Add a new flag
"FSS_UNLINK_ON_CREATE" to unlink the backing store before the snapshot
gets created. With this change dump(8) no longer dumps the zero-sized,
but named snapshot it is working on.
To generate a diff of this commit:
cvs rdiff -u -r1.4 -r1.4.4.1 src/sbin/dump/snapshot.c
cvs rdiff -u -r1.12 -r1.12.2.1 src/share/man/man4/fss.4
cvs rdiff -u -r1.60.4.3 -r1.60.4.4 src/sys/dev/fss.c
cvs rdiff -u -r1.23 -r1.23.4.1 src/sys/dev/fssvar.h
cvs rdiff -u -r1.3.8.5 -r1.3.8.6 src/sys/kern/vfs_wapbl.c
cvs rdiff -u -r1.51 -r1.51.4.1 src/sys/ufs/ffs/ffs_balloc.c
cvs rdiff -u -r1.82.4.3 -r1.82.4.4 src/sys/ufs/ffs/ffs_snapshot.c
cvs rdiff -u -r1.6 -r1.6.6.1 src/usr.sbin/fssconfig/fssconfig.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sbin/dump/snapshot.c
diff -u src/sbin/dump/snapshot.c:1.4 src/sbin/dump/snapshot.c:1.4.4.1
--- src/sbin/dump/snapshot.c:1.4 Mon Apr 28 20:23:08 2008
+++ src/sbin/dump/snapshot.c Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: snapshot.c,v 1.4 2008/04/28 20:23:08 martin Exp $ */
+/* $NetBSD: snapshot.c,v 1.4.4.1 2011/06/18 17:00:25 bouyer Exp $ */
/*-
* Copyright (c) 2005 The NetBSD Foundation, Inc.
@@ -56,7 +56,7 @@
int
snap_open(char *mountpoint, char *backup, time_t *snap_date, char **snap_dev)
{
- int i, fd, israw, fsinternal, dounlink, flags;
+ int i, fd, israw, fsinternal, dounlink;
char path[MAXPATHLEN], fss_dev[14];
dev_t mountdev;
struct fss_set fss;
@@ -118,6 +118,9 @@
if (close(fd) < 0)
goto fail;
+ fss.fss_flags = FSS_UNCONFIG_ON_CLOSE;
+ if (dounlink)
+ fss.fss_flags |= FSS_UNLINK_ON_CREATE;
/*
* Create the snapshot on the first free snapshot device.
*/
@@ -126,9 +129,6 @@
if ((fd = open(fss_dev, O_RDWR, 0)) < 0)
goto fail;
- if (ioctl(fd, FSSIOFGET, &flags) < 0)
- goto fail;
-
if (ioctl(fd, FSSIOCSET, &fss) < 0) {
if (errno != EBUSY)
goto fail;
@@ -136,6 +136,7 @@
fd = -1;
continue;
}
+ dounlink = 0;
if (snap_dev != NULL) {
*snap_dev = strdup(fss_dev);
@@ -145,10 +146,7 @@
}
}
- flags |= FSS_UNCONFIG_ON_CLOSE;
- if (ioctl(fd, FSSIOCGET, &fsg) < 0 ||
- ioctl(fd, FSSIOFSET, &flags) < 0 ||
- (!israw && unlink(fss.fss_bstore) < 0)) {
+ if (ioctl(fd, FSSIOCGET, &fsg) < 0) {
ioctl(fd, FSSIOCCLR);
goto fail;
}
Index: src/share/man/man4/fss.4
diff -u src/share/man/man4/fss.4:1.12 src/share/man/man4/fss.4:1.12.2.1
--- src/share/man/man4/fss.4:1.12 Sun Sep 21 10:25:07 2008
+++ src/share/man/man4/fss.4 Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-.\" $NetBSD: fss.4,v 1.12 2008/09/21 10:25:07 hannken Exp $ */
+.\" $NetBSD: fss.4,v 1.12.2.1 2011/06/18 17:00:25 bouyer Exp $ */
.\"
.\"
.\" Copyright (c) 2003 The NetBSD Foundation, Inc.
@@ -28,7 +28,7 @@
.\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
.\" POSSIBILITY OF SUCH DAMAGE.
.\"
-.Dd September 21, 2008
+.Dd February 24, 2011
.Dt FSS 4
.Os
.Sh NAME
@@ -65,6 +65,7 @@
char *fss_mount;
char *fss_bstore;
blksize_t fss_csize;
+ int fss_flags;
};
.Ed
.Pp
@@ -78,6 +79,9 @@
The struct element
.Va fss_csize
is the preferred size of this data.
+The struct element
+.Va fss_flags
+is the initial set of flags.
.It Dv FSSIOCGET(struct fss_get)
Gets the status of a
.Nm
@@ -120,6 +124,10 @@
Unconfigure the
.Nm
device on the last close.
+.It Dv FSS_UNLINK_ON_CREATE
+Unlink the backing file before the
+.Nm
+device is created.
.El
.It Dv FSSIOFGET(int)
Gets the flags of a
Index: src/sys/dev/fss.c
diff -u src/sys/dev/fss.c:1.60.4.3 src/sys/dev/fss.c:1.60.4.4
--- src/sys/dev/fss.c:1.60.4.3 Sun Mar 28 17:26:26 2010
+++ src/sys/dev/fss.c Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: fss.c,v 1.60.4.3 2010/03/28 17:26:26 snj Exp $ */
+/* $NetBSD: fss.c,v 1.60.4.4 2011/06/18 17:00:25 bouyer Exp $ */
/*-
* Copyright (c) 2003 The NetBSD Foundation, Inc.
@@ -36,7 +36,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.60.4.3 2010/03/28 17:26:26 snj Exp $");
+__KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.60.4.4 2011/06/18 17:00:25 bouyer Exp $");
#include <sys/param.h>
#include <sys/systm.h>
@@ -58,6 +58,7 @@
#include <sys/kthread.h>
#include <sys/fstrans.h>
#include <sys/simplelock.h>
+#include <sys/vfs_syscalls.h> /* For do_sys_unlink(). */
#include <miscfs/specfs/specdev.h>
@@ -301,6 +302,9 @@
struct fss_get *fsg = (struct fss_get *)data;
switch (cmd) {
+ case FSSIOCSET50:
+ fss->fss_flags = 0;
+ /* Fall through */
case FSSIOCSET:
mutex_enter(&sc->sc_lock);
if ((flag & FWRITE) == 0)
@@ -309,6 +313,8 @@
error = EBUSY;
else
error = fss_create_snapshot(sc, fss, l);
+ if (error == 0)
+ sc->sc_uflags = fss->fss_flags;
mutex_exit(&sc->sc_lock);
break;
@@ -604,7 +610,7 @@
* Check for file system internal snapshot.
*/
- NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_USERSPACE, fss->fss_bstore);
+ NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore);
if ((error = namei(&nd)) != 0)
return error;
@@ -618,14 +624,20 @@
sc->sc_bs_bshift++)
if (FSS_FSBSIZE(sc) == fsbsize)
break;
- if (sc->sc_bs_bshift >= bits) {
- VOP_UNLOCK(sc->sc_bs_vp, 0);
+ if (sc->sc_bs_bshift >= bits)
return EINVAL;
- }
sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1;
sc->sc_clshift = 0;
+ if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) {
+ error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE);
+ if (error)
+ return error;
+ }
+ error = vn_lock(nd.ni_vp, LK_EXCLUSIVE);
+ if (error != 0)
+ return error;
error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts);
TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts);
@@ -633,7 +645,7 @@
return error;
}
- vput(nd.ni_vp);
+ vrele(nd.ni_vp);
/*
* Get the block device it is mounted on.
@@ -676,6 +688,11 @@
if (nd.ni_vp->v_type != VREG && nd.ni_vp->v_type != VCHR)
return EINVAL;
+ if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) {
+ error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE);
+ if (error)
+ return error;
+ }
if (sc->sc_bs_vp->v_type == VREG) {
error = VOP_GETATTR(sc->sc_bs_vp, &va, l->l_cred);
if (error != 0)
@@ -803,7 +820,7 @@
fss_softc_free(sc);
if (sc->sc_bs_vp != NULL) {
if (sc->sc_flags & FSS_PERSISTENT)
- vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
+ vrele(sc->sc_bs_vp);
else
vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
}
@@ -830,7 +847,7 @@
fss_softc_free(sc);
if (sc->sc_flags & FSS_PERSISTENT)
- vn_close(sc->sc_bs_vp, FREAD, l->l_cred);
+ vrele(sc->sc_bs_vp);
else
vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred);
sc->sc_bs_vp = NULL;
Index: src/sys/dev/fssvar.h
diff -u src/sys/dev/fssvar.h:1.23 src/sys/dev/fssvar.h:1.23.4.1
--- src/sys/dev/fssvar.h:1.23 Wed Sep 17 14:49:25 2008
+++ src/sys/dev/fssvar.h Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: fssvar.h,v 1.23 2008/09/17 14:49:25 hannken Exp $ */
+/* $NetBSD: fssvar.h,v 1.23.4.1 2011/06/18 17:00:25 bouyer Exp $ */
/*-
* Copyright (c) 2003, 2007 The NetBSD Foundation, Inc.
@@ -35,11 +35,13 @@
#include <sys/simplelock.h>
#define FSS_UNCONFIG_ON_CLOSE 0x01 /* Unconfigure on last close */
+#define FSS_UNLINK_ON_CREATE 0x02 /* Unlink backing store on create */
struct fss_set {
char *fss_mount; /* Mount point of file system */
char *fss_bstore; /* Path of backing store */
blksize_t fss_csize; /* Preferred cluster size */
+ int fss_flags; /* Initial flags */
};
struct fss_get {
@@ -50,11 +52,12 @@
blkcnt_t fsg_bs_size; /* # clusters on backing store */
};
-#define FSSIOCSET _IOW('F', 0, struct fss_set) /* Configure */
+#define FSSIOCSET _IOW('F', 5, struct fss_set) /* Configure */
#define FSSIOCGET _IOR('F', 1, struct fss_get) /* Status */
#define FSSIOCCLR _IO('F', 2) /* Unconfigure */
#define FSSIOFSET _IOW('F', 3, int) /* Set flags */
#define FSSIOFGET _IOR('F', 4, int) /* Get flags */
+#define FSSIOCSET50 _IOW('F', 0, struct fss_set) /* Old configure */
#ifdef _KERNEL
Index: src/sys/kern/vfs_wapbl.c
diff -u src/sys/kern/vfs_wapbl.c:1.3.8.5 src/sys/kern/vfs_wapbl.c:1.3.8.6
--- src/sys/kern/vfs_wapbl.c:1.3.8.5 Mon Mar 7 04:09:28 2011
+++ src/sys/kern/vfs_wapbl.c Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: vfs_wapbl.c,v 1.3.8.5 2011/03/07 04:09:28 riz Exp $ */
+/* $NetBSD: vfs_wapbl.c,v 1.3.8.6 2011/06/18 17:00:25 bouyer Exp $ */
/*-
* Copyright (c) 2003, 2008, 2009 The NetBSD Foundation, Inc.
@@ -33,7 +33,7 @@
* This implements file system independent write ahead filesystem logging.
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.3.8.5 2011/03/07 04:09:28 riz Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_wapbl.c,v 1.3.8.6 2011/06/18 17:00:25 bouyer Exp $");
#include <sys/param.h>
@@ -401,7 +401,7 @@
wl->wl_bufcount_max = (nbuf / 2) * 1024;
/* XXX tie this into resource estimation */
- wl->wl_dealloclim = 2 * btodb(wl->wl_bufbytes_max);
+ wl->wl_dealloclim = wl->wl_bufbytes_max / mp->mnt_stat.f_bsize / 2;
#if WAPBL_UVM_ALLOC
wl->wl_deallocblks = (void *) uvm_km_zalloc(kernel_map,
@@ -830,8 +830,7 @@
((wl->wl_bufcount + (lockcount * 10)) >
wl->wl_bufcount_max / 2) ||
(wapbl_transaction_len(wl) > wl->wl_circ_size / 2) ||
- (wl->wl_dealloccnt >=
- (wl->wl_dealloclim - (wl->wl_dealloclim >> 8)));
+ (wl->wl_dealloccnt >= (wl->wl_dealloclim / 2));
mutex_exit(&wl->wl_mtx);
if (doflush) {
Index: src/sys/ufs/ffs/ffs_balloc.c
diff -u src/sys/ufs/ffs/ffs_balloc.c:1.51 src/sys/ufs/ffs/ffs_balloc.c:1.51.4.1
--- src/sys/ufs/ffs/ffs_balloc.c:1.51 Thu Jul 31 05:38:06 2008
+++ src/sys/ufs/ffs/ffs_balloc.c Sat Jun 18 17:00:25 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: ffs_balloc.c,v 1.51 2008/07/31 05:38:06 simonb Exp $ */
+/* $NetBSD: ffs_balloc.c,v 1.51.4.1 2011/06/18 17:00:25 bouyer Exp $ */
/*
* Copyright (c) 2002 Networks Associates Technology, Inc.
@@ -41,7 +41,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ffs_balloc.c,v 1.51 2008/07/31 05:38:06 simonb Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ffs_balloc.c,v 1.51.4.1 2011/06/18 17:00:25 bouyer Exp $");
#if defined(_KERNEL_OPT)
#include "opt_quota.h"
@@ -340,6 +340,10 @@
goto fail;
}
mutex_enter(&ump->um_lock);
+ /* Try to keep snapshot indirect blocks contiguous. */
+ if (i == num && (ip->i_flags & SF_SNAPSHOT) != 0)
+ pref = ffs_blkpref_ufs1(ip, lbn, indirs[i-1].in_off,
+ flags | B_METAONLY, &bap[0]);
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, flags | B_METAONLY,
NULL);
@@ -932,6 +936,10 @@
goto fail;
}
mutex_enter(&ump->um_lock);
+ /* Try to keep snapshot indirect blocks contiguous. */
+ if (i == num && (ip->i_flags & SF_SNAPSHOT) != 0)
+ pref = ffs_blkpref_ufs2(ip, lbn, indirs[i-1].in_off,
+ flags | B_METAONLY, &bap[0]);
if (pref == 0)
pref = ffs_blkpref_ufs2(ip, lbn, 0, flags | B_METAONLY,
NULL);
Index: src/sys/ufs/ffs/ffs_snapshot.c
diff -u src/sys/ufs/ffs/ffs_snapshot.c:1.82.4.3 src/sys/ufs/ffs/ffs_snapshot.c:1.82.4.4
--- src/sys/ufs/ffs/ffs_snapshot.c:1.82.4.3 Sun Mar 28 17:28:33 2010
+++ src/sys/ufs/ffs/ffs_snapshot.c Sat Jun 18 17:00:26 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: ffs_snapshot.c,v 1.82.4.3 2010/03/28 17:28:33 snj Exp $ */
+/* $NetBSD: ffs_snapshot.c,v 1.82.4.4 2011/06/18 17:00:26 bouyer Exp $ */
/*
* Copyright 2000 Marshall Kirk McKusick. All Rights Reserved.
@@ -38,7 +38,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ffs_snapshot.c,v 1.82.4.3 2010/03/28 17:28:33 snj Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ffs_snapshot.c,v 1.82.4.4 2011/06/18 17:00:26 bouyer Exp $");
#if defined(_KERNEL_OPT)
#include "opt_ffs.h"
@@ -79,6 +79,7 @@
struct snap_info {
kmutex_t si_lock; /* Lock this snapinfo */
kmutex_t si_snaplock; /* Snapshot vnode common lock */
+ lwp_t *si_owner; /* Sanplock owner */
TAILQ_HEAD(inodelst, inode) si_snapshots; /* List of active snapshots */
daddr_t *si_snapblklist; /* Snapshot block hints list */
uint32_t si_gen; /* Incremented on change */
@@ -114,6 +115,9 @@
static int rwfsblk(struct vnode *, int, void *, daddr_t);
static int syncsnap(struct vnode *);
static int wrsnapblk(struct vnode *, void *, daddr_t);
+#if !defined(FFS_NO_SNAPSHOT)
+static int blocks_in_journal(struct fs *);
+#endif
static inline bool is_active_snapshot(struct snap_info *, struct inode *);
static inline daddr_t db_get(struct inode *, int);
@@ -139,6 +143,7 @@
TAILQ_INIT(&si->si_snapshots);
mutex_init(&si->si_lock, MUTEX_DEFAULT, IPL_NONE);
mutex_init(&si->si_snaplock, MUTEX_DEFAULT, IPL_NONE);
+ si->si_owner = NULL;
si->si_gen = 0;
si->si_snapblklist = NULL;
@@ -172,7 +177,6 @@
}
#else /* defined(FFS_NO_SNAPSHOT) */
bool suspended = false;
- bool snapshot_locked = false;
int error, redo = 0, snaploc;
void *sbbuf = NULL;
daddr_t *snaplist = NULL, snaplistsize = 0;
@@ -213,12 +217,6 @@
if (error)
goto out;
/*
- * Change inode to snapshot type file.
- */
- ip->i_flags |= SF_SNAPSHOT;
- DIP_ASSIGN(ip, flags, ip->i_flags);
- ip->i_flag |= IN_CHANGE | IN_UPDATE;
- /*
* Copy all the cylinder group maps. Although the
* filesystem is still active, we hope that only a few
* cylinder groups will change between now and when we
@@ -267,15 +265,11 @@
if (error)
goto out;
/*
- * Acquire the snapshot lock.
- */
- mutex_enter(&si->si_snaplock);
- snapshot_locked = true;
- /*
* Record snapshot inode. Since this is the newest snapshot,
* it must be placed at the end of the list.
*/
- fs->fs_snapinum[snaploc] = ip->i_number;
+ if (ip->i_ffs_effnlink > 0)
+ fs->fs_snapinum[snaploc] = ip->i_number;
mutex_enter(&si->si_lock);
if (is_active_snapshot(si, ip))
@@ -374,15 +368,15 @@
si->si_gen++;
mutex_exit(&si->si_lock);
- if (snapshot_locked)
- mutex_exit(&si->si_snaplock);
if (suspended) {
+ VOP_UNLOCK(vp, 0);
vfs_resume(vp->v_mount);
+ vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
#ifdef DEBUG
getmicrotime(&endtime);
timersub(&endtime, &starttime, &endtime);
- printf("%s: suspended %ld.%03ld sec, redo %d of %d\n",
- mp->mnt_stat.f_mntonname, (long)endtime.tv_sec,
+ printf("%s: suspended %lld.%03ld sec, redo %d of %d\n",
+ mp->mnt_stat.f_mntonname, (long long)endtime.tv_sec,
endtime.tv_usec / 1000, redo, fs->fs_ncg);
#endif
}
@@ -391,7 +385,7 @@
(void) ffs_truncate(vp, (off_t)0, 0, NOCRED);
UFS_WAPBL_END(mp);
}
- } else
+ } else if (ip->i_ffs_effnlink > 0)
vref(vp);
return (error);
}
@@ -402,11 +396,13 @@
static int
snapshot_setup(struct mount *mp, struct vnode *vp)
{
- int error, i, len, loc;
+ int error, n, len, loc, cg;
daddr_t blkno, numblks;
struct buf *ibp, *nbp;
struct fs *fs = VFSTOUFS(mp)->um_fs;
struct lwp *l = curlwp;
+ const int wbreak = blocks_in_journal(fs)/8;
+ struct inode *ip = VTOI(vp);
/*
* Check mount, exclusive reference and owner.
@@ -425,6 +421,11 @@
if (error)
return error;
}
+
+ /* Change inode to snapshot type file. */
+ ip->i_flags |= SF_SNAPSHOT;
+ DIP_ASSIGN(ip, flags, ip->i_flags);
+ ip->i_flag |= IN_CHANGE | IN_UPDATE;
/*
* Write an empty list of preallocated blocks to the end of
* the snapshot to set size to at least that of the filesystem.
@@ -451,7 +452,7 @@
error = UFS_WAPBL_BEGIN(mp);
if (error)
return error;
- for (blkno = NDADDR, i = 0; blkno < numblks; blkno += NINDIR(fs)) {
+ for (blkno = NDADDR, n = 0; blkno < numblks; blkno += NINDIR(fs)) {
error = ffs_balloc(vp, lblktosize(fs, (off_t)blkno),
fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
if (error)
@@ -460,7 +461,7 @@
bawrite(ibp);
else
brelse(ibp, 0);
- if ((++i % 16) == 0) {
+ if (wbreak > 0 && (++n % wbreak) == 0) {
UFS_WAPBL_END(mp);
error = UFS_WAPBL_BEGIN(mp);
if (error)
@@ -483,6 +484,28 @@
if (error)
goto out;
bawrite(nbp);
+ if (wbreak > 0 && (++n % wbreak) == 0) {
+ UFS_WAPBL_END(mp);
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
+ }
+ }
+ /*
+ * Allocate all cylinder group blocks.
+ */
+ for (cg = 0; cg < fs->fs_ncg; cg++) {
+ error = ffs_balloc(vp, lfragtosize(fs, cgtod(fs, cg)),
+ fs->fs_bsize, l->l_cred, 0, &nbp);
+ if (error)
+ goto out;
+ bawrite(nbp);
+ if (wbreak > 0 && (++n % wbreak) == 0) {
+ UFS_WAPBL_END(mp);
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
+ }
}
out:
@@ -563,8 +586,7 @@
snapshot_expunge(struct mount *mp, struct vnode *vp, struct fs *copy_fs,
daddr_t *snaplistsize, daddr_t **snaplist)
{
- bool has_wapbl = false;
- int cg, error, len, loc;
+ int cg, error = 0, len, loc;
daddr_t blkno, *blkp;
struct fs *fs = VFSTOUFS(mp)->um_fs;
struct inode *xp;
@@ -595,10 +617,6 @@
*/
*snaplistsize = fs->fs_ncg + howmany(fs->fs_cssize, fs->fs_bsize) +
FSMAXSNAP + 1 /* superblock */ + 1 /* last block */ + 1 /* size */;
- error = UFS_WAPBL_BEGIN(mp);
- if (error)
- goto out;
- has_wapbl = true;
mutex_enter(&mntvnode_lock);
/*
* NOTE: not using the TAILQ_FOREACH here since in this loop vgone()
@@ -650,19 +668,30 @@
if (loc < NDADDR) {
len = fragroundup(fs, blkoff(fs, xp->i_size));
if (len > 0 && len < fs->fs_bsize) {
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error) {
+ (void)vunmark(mvp);
+ goto out;
+ }
ffs_blkfree(copy_fs, vp, db_get(xp, loc),
len, xp->i_number);
blkno = db_get(xp, loc);
db_assign(xp, loc, 0);
+ UFS_WAPBL_END(mp);
}
}
*snaplistsize += 1;
error = expunge(vp, xp, copy_fs, fullacct, BLK_NOCOPY);
if (blkno)
db_assign(xp, loc, blkno);
- if (!error)
- error = ffs_freefile(copy_fs, vp, xp->i_number,
- xp->i_mode);
+ if (!error) {
+ error = UFS_WAPBL_BEGIN(mp);
+ if (!error) {
+ error = ffs_freefile(copy_fs, vp,
+ xp->i_number, xp->i_mode);
+ UFS_WAPBL_END(mp);
+ }
+ }
if (error) {
(void)vunmark(mvp);
goto out;
@@ -687,10 +716,9 @@
*blkp++ = blkno + loc;
for (; cg < fs->fs_ncg; cg++)
*blkp++ = fragstoblks(fs, cgtod(fs, cg));
+ (*snaplist)[0] = blkp - &(*snaplist)[0];
out:
- if (has_wapbl)
- UFS_WAPBL_END(mp);
if (mvp != NULL)
vnfree(mvp);
if (logvp != NULL)
@@ -712,25 +740,26 @@
snapshot_expunge_snap(struct mount *mp, struct vnode *vp,
struct fs *copy_fs, daddr_t snaplistsize)
{
- int error, i;
+ int error = 0, i;
daddr_t numblks, *snaplist = NULL;
struct fs *fs = VFSTOUFS(mp)->um_fs;
struct inode *ip = VTOI(vp), *xp;
struct lwp *l = curlwp;
struct snap_info *si = VFSTOUFS(mp)->um_snapinfo;
- error = UFS_WAPBL_BEGIN(mp);
- if (error)
- return error;
TAILQ_FOREACH(xp, &si->si_snapshots, i_nextsnap) {
- if (xp == ip)
- break;
- error = expunge(vp, xp, fs, snapacct, BLK_SNAP);
- if (error)
- break;
+ if (xp != ip) {
+ error = expunge(vp, xp, fs, snapacct, BLK_SNAP);
+ if (error)
+ break;
+ }
if (xp->i_ffs_effnlink != 0)
continue;
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ break;
error = ffs_freefile(copy_fs, vp, xp->i_number, xp->i_mode);
+ UFS_WAPBL_END(mp);
if (error)
break;
}
@@ -762,12 +791,10 @@
snaplist[i] = ufs_rw64(snaplist[i], UFS_FSNEEDSWAP(fs));
error = vn_rdwr(UIO_WRITE, vp, (void *)snaplist,
snaplistsize * sizeof(daddr_t), lblktosize(fs, (off_t)numblks),
- UIO_SYSSPACE, IO_NODELOCKED | IO_JOURNALLOCKED | IO_UNIT,
- l->l_cred, NULL, NULL);
+ UIO_SYSSPACE, IO_NODELOCKED | IO_UNIT, l->l_cred, NULL, NULL);
for (i = 0; i < snaplistsize; i++)
snaplist[i] = ufs_rw64(snaplist[i], UFS_FSNEEDSWAP(fs));
out:
- UFS_WAPBL_END(mp);
if (error && snaplist != NULL) {
free(snaplist, M_UFSMNT);
ip->i_snapblklist = NULL;
@@ -860,13 +887,10 @@
static int
cgaccount(struct vnode *vp, int passno, int *redo)
{
- int cg, error;
+ int cg, error = 0;
struct buf *nbp;
struct fs *fs = VTOI(vp)->i_fs;
- error = UFS_WAPBL_BEGIN(vp->v_mount);
- if (error)
- return error;
if (redo != NULL)
*redo = 0;
if (passno == 1)
@@ -875,18 +899,24 @@
for (cg = 0; cg < fs->fs_ncg; cg++) {
if (passno == 2 && ACTIVECG_ISSET(fs, cg))
continue;
+
if (redo != NULL)
*redo += 1;
+ error = UFS_WAPBL_BEGIN(vp->v_mount);
+ if (error)
+ return error;
error = ffs_balloc(vp, lfragtosize(fs, cgtod(fs, cg)),
fs->fs_bsize, curlwp->l_cred, 0, &nbp);
- if (error)
+ if (error) {
+ UFS_WAPBL_END(vp->v_mount);
break;
+ }
error = cgaccount1(cg, vp, nbp->b_data, passno);
bawrite(nbp);
+ UFS_WAPBL_END(vp->v_mount);
if (error)
break;
}
- UFS_WAPBL_END(vp->v_mount);
return error;
}
@@ -993,8 +1023,14 @@
struct lwp *l = curlwp;
void *bap;
struct buf *bp;
+ struct mount *mp;
ns = UFS_FSNEEDSWAP(fs);
+ mp = snapvp->v_mount;
+
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
/*
* Prepare to expunge the inode. If its inode block has not
* yet been copied, then allocate and fill the copy.
@@ -1012,8 +1048,10 @@
if (! error)
error = rwfsblk(snapvp, B_READ, bp->b_data, lbn);
}
- if (error)
+ if (error) {
+ UFS_WAPBL_END(mp);
return error;
+ }
/*
* Set a snapshot inode to be a zero length file, regular files
* or unlinked snapshots to be completely unallocated.
@@ -1040,6 +1078,7 @@
bzero(&dip2->di_db[0], (NDADDR + NIADDR) * sizeof(int64_t));
}
bdwrite(bp);
+ UFS_WAPBL_END(mp);
/*
* Now go through and expunge all the blocks in the file
* using the function requested.
@@ -1049,13 +1088,15 @@
bap = &cancelip->i_ffs1_db[0];
else
bap = &cancelip->i_ffs2_db[0];
- if ((error = (*acctfunc)(snapvp, bap, 0, NDADDR, fs, 0, expungetype)))
+ error = (*acctfunc)(snapvp, bap, 0, NDADDR, fs, 0, expungetype);
+ if (error)
return (error);
if (fs->fs_magic == FS_UFS1_MAGIC)
bap = &cancelip->i_ffs1_ib[0];
else
bap = &cancelip->i_ffs2_ib[0];
- if ((error = (*acctfunc)(snapvp, bap, 0, NIADDR, fs, -1, expungetype)))
+ error = (*acctfunc)(snapvp, bap, 0, NIADDR, fs, -1, expungetype);
+ if (error)
return (error);
blksperindir = 1;
lbn = -NDADDR;
@@ -1171,12 +1212,17 @@
{
struct inode *ip = VTOI(vp);
struct lwp *l = curlwp;
+ struct mount *mp = vp->v_mount;
daddr_t blkno;
daddr_t lbn;
struct buf *ibp;
- int error;
+ int error, n;
+ const int wbreak = blocks_in_journal(VFSTOUFS(mp)->um_fs)/8;
- for ( ; oldblkp < lastblkp; oldblkp++) {
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
+ for ( n = 0; oldblkp < lastblkp; oldblkp++) {
blkno = idb_get(ip, bap, oldblkp);
if (blkno == 0 || blkno == BLK_NOCOPY || blkno == BLK_SNAP)
continue;
@@ -1188,7 +1234,7 @@
error = ffs_balloc(vp, lblktosize(fs, (off_t)lbn),
fs->fs_bsize, l->l_cred, B_METAONLY, &ibp);
if (error)
- return (error);
+ break;
blkno = idb_get(ip, ibp->b_data,
(lbn - NDADDR) % NINDIR(fs));
}
@@ -1212,8 +1258,15 @@
bdwrite(ibp);
}
}
+ if (wbreak > 0 && (++n % wbreak) == 0) {
+ UFS_WAPBL_END(mp);
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
+ }
}
- return (0);
+ UFS_WAPBL_END(mp);
+ return error;
}
/*
@@ -1225,16 +1278,21 @@
{
daddr_t blkno;
struct inode *ip;
+ struct mount *mp = vp->v_mount;
ino_t inum;
- int acctit;
+ int acctit, error, n;
+ const int wbreak = blocks_in_journal(VFSTOUFS(mp)->um_fs)/8;
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
ip = VTOI(vp);
inum = ip->i_number;
if (lblkno == -1)
acctit = 0;
else
acctit = 1;
- for ( ; oldblkp < lastblkp; oldblkp++, lblkno++) {
+ for ( n = 0; oldblkp < lastblkp; oldblkp++, lblkno++) {
blkno = idb_get(ip, bap, oldblkp);
if (blkno == 0 || blkno == BLK_NOCOPY)
continue;
@@ -1243,9 +1301,43 @@
if (blkno == BLK_SNAP)
blkno = blkstofrags(fs, lblkno);
ffs_blkfree(fs, vp, blkno, fs->fs_bsize, inum);
+ if (wbreak > 0 && (++n % wbreak) == 0) {
+ UFS_WAPBL_END(mp);
+ error = UFS_WAPBL_BEGIN(mp);
+ if (error)
+ return error;
+ }
}
+ UFS_WAPBL_END(mp);
return (0);
}
+
+/*
+ * Number of blocks that fit into the journal or zero if not logging.
+ */
+static int
+blocks_in_journal(struct fs *fs)
+{
+ off_t bpj;
+
+ if ((fs->fs_flags & FS_DOWAPBL) == 0)
+ return 0;
+ bpj = 1;
+ if (fs->fs_journal_version == UFS_WAPBL_VERSION) {
+ switch (fs->fs_journal_location) {
+ case UFS_WAPBL_JOURNALLOC_END_PARTITION:
+ bpj = (off_t)fs->fs_journallocs[UFS_WAPBL_EPART_BLKSZ]*
+ fs->fs_journallocs[UFS_WAPBL_EPART_COUNT];
+ break;
+ case UFS_WAPBL_JOURNALLOC_IN_FILESYSTEM:
+ bpj = (off_t)fs->fs_journallocs[UFS_WAPBL_INFS_BLKSZ]*
+ fs->fs_journallocs[UFS_WAPBL_INFS_COUNT];
+ break;
+ }
+ }
+ bpj /= fs->fs_bsize;
+ return (bpj > 0 ? bpj : 1);
+}
#endif /* defined(FFS_NO_SNAPSHOT) */
/*
@@ -1322,6 +1414,7 @@
*
* Clear copy-on-write flag if last snapshot.
*/
+ mutex_enter(&si->si_snaplock);
mutex_enter(&si->si_lock);
if (is_active_snapshot(si, ip)) {
TAILQ_REMOVE(&si->si_snapshots, ip, i_nextsnap);
@@ -1331,18 +1424,22 @@
si->si_snapblklist = xp->i_snapblklist;
si->si_gen++;
mutex_exit(&si->si_lock);
+ mutex_exit(&si->si_snaplock);
} else {
si->si_snapblklist = 0;
si->si_gen++;
mutex_exit(&si->si_lock);
+ mutex_exit(&si->si_snaplock);
fscow_disestablish(mp, ffs_copyonwrite, devvp);
}
if (ip->i_snapblklist != NULL) {
free(ip->i_snapblklist, M_UFSMNT);
ip->i_snapblklist = NULL;
}
- } else
+ } else {
mutex_exit(&si->si_lock);
+ mutex_exit(&si->si_snaplock);
+ }
/*
* Clear all BLK_NOCOPY fields. Pass any block claims to other
* snapshots that want them (see ffs_snapblkfree below).
@@ -1380,6 +1477,9 @@
}
}
bawrite(ibp);
+ UFS_WAPBL_END(mp);
+ error = UFS_WAPBL_BEGIN(mp);
+ KASSERT(error == 0);
}
/*
* Clear snapshot flag and drop reference.
@@ -1420,25 +1520,18 @@
daddr_t lbn;
daddr_t blkno;
uint32_t gen;
- int indiroff = 0, snapshot_locked = 0, error = 0, claimedblk = 0;
+ int indiroff = 0, error = 0, claimedblk = 0;
si = VFSTOUFS(mp)->um_snapinfo;
lbn = fragstoblks(fs, bno);
+ mutex_enter(&si->si_snaplock);
mutex_enter(&si->si_lock);
+ si->si_owner = curlwp;
+
retry:
gen = si->si_gen;
TAILQ_FOREACH(ip, &si->si_snapshots, i_nextsnap) {
vp = ITOV(ip);
- if (snapshot_locked == 0) {
- if (!mutex_tryenter(&si->si_snaplock)) {
- mutex_exit(&si->si_lock);
- mutex_enter(&si->si_snaplock);
- mutex_enter(&si->si_lock);
- }
- snapshot_locked = 1;
- if (gen != si->si_gen)
- goto retry;
- }
/*
* Lookup block being written.
*/
@@ -1535,6 +1628,9 @@
error = syncsnap(vp);
else
error = 0;
+ mutex_enter(&si->si_lock);
+ si->si_owner = NULL;
+ mutex_exit(&si->si_lock);
mutex_exit(&si->si_snaplock);
return (error == 0);
}
@@ -1574,7 +1670,9 @@
if (gen != si->si_gen)
goto retry;
}
+ si->si_owner = NULL;
mutex_exit(&si->si_lock);
+ mutex_exit(&si->si_snaplock);
if (saved_data)
free(saved_data, M_UFSMNT);
/*
@@ -1583,8 +1681,6 @@
* not be freed. Although space will be lost, the snapshot
* will stay consistent.
*/
- if (snapshot_locked)
- mutex_exit(&si->si_snaplock);
return (error);
}
@@ -1723,12 +1819,11 @@
mutex_enter(&si->si_lock);
while ((xp = TAILQ_FIRST(&si->si_snapshots)) != 0) {
vp = ITOV(xp);
- vp->v_vnlock = &vp->v_lock;
TAILQ_REMOVE(&si->si_snapshots, xp, i_nextsnap);
if (xp->i_snapblklist == si->si_snapblklist)
si->si_snapblklist = NULL;
FREE(xp->i_snapblklist, M_UFSMNT);
- if (xp->i_ffs_effnlink > 0) {
+ if (xp->i_ffs_effnlink != 0) {
si->si_gen++;
mutex_exit(&si->si_lock);
vrele(vp);
@@ -1798,6 +1893,15 @@
/*
* Not in the precomputed list, so check the snapshots.
*/
+ if (si->si_owner != curlwp) {
+ if (!mutex_tryenter(&si->si_snaplock)) {
+ mutex_exit(&si->si_lock);
+ mutex_enter(&si->si_snaplock);
+ mutex_enter(&si->si_lock);
+ }
+ si->si_owner = curlwp;
+ snapshot_locked = 1;
+ }
if (data_valid && bp->b_bcount == fs->fs_bsize)
saved_data = bp->b_data;
retry:
@@ -1819,6 +1923,7 @@
blkno = db_get(ip, lbn);
} else {
mutex_exit(&si->si_lock);
+ blkno = 0; /* XXX: GCC */
if ((error = snapblkaddr(vp, lbn, &blkno)) != 0) {
mutex_enter(&si->si_lock);
break;
@@ -1838,34 +1943,8 @@
error = ENOMEM;
break;
}
-
- if (snapshot_locked == 0) {
- if (!mutex_tryenter(&si->si_snaplock)) {
- mutex_exit(&si->si_lock);
- mutex_enter(&si->si_snaplock);
- mutex_enter(&si->si_lock);
- }
- snapshot_locked = 1;
- if (gen != si->si_gen)
- goto retry;
-
- /* Check again if block still needs to be copied */
- if (lbn < NDADDR) {
- blkno = db_get(ip, lbn);
- } else {
- mutex_exit(&si->si_lock);
- if ((error = snapblkaddr(vp, lbn, &blkno)) != 0) {
- mutex_enter(&si->si_lock);
- break;
- }
- mutex_enter(&si->si_lock);
- if (gen != si->si_gen)
- goto retry;
- }
-
- if (blkno != 0)
- continue;
- }
+ /* Only one level of recursion allowed. */
+ KASSERT(snapshot_locked);
/*
* Allocate the block into which to do the copy. Since
* multiple processes may all try to copy the same block,
@@ -1920,11 +1999,14 @@
* have not been unlinked, and hence will be visible after
* a crash, to ensure their integrity.
*/
- mutex_exit(&si->si_lock);
+ if (snapshot_locked) {
+ si->si_owner = NULL;
+ mutex_exit(&si->si_lock);
+ mutex_exit(&si->si_snaplock);
+ } else
+ mutex_exit(&si->si_lock);
if (saved_data && saved_data != bp->b_data)
free(saved_data, M_UFSMNT);
- if (snapshot_locked)
- mutex_exit(&si->si_snaplock);
return error;
}
@@ -2084,9 +2166,14 @@
mutex_enter(&bufcache_lock);
while ((bp = LIST_FIRST(&vp->v_dirtyblkhd))) {
- KASSERT((bp->b_cflags & BC_BUSY) == 0);
+ error = bbusy(bp, false, 0, NULL);
+ if (error == EPASSTHROUGH)
+ continue;
+ else if (error != 0) {
+ mutex_exit(&bufcache_lock);
+ return error;
+ }
KASSERT(bp->b_bcount == fs->fs_bsize);
- bp->b_cflags |= BC_BUSY;
mutex_exit(&bufcache_lock);
error = rwfsblk(vp, B_WRITE, bp->b_data,
fragstoblks(fs, dbtofsb(fs, bp->b_blkno)));
Index: src/usr.sbin/fssconfig/fssconfig.c
diff -u src/usr.sbin/fssconfig/fssconfig.c:1.6 src/usr.sbin/fssconfig/fssconfig.c:1.6.6.1
--- src/usr.sbin/fssconfig/fssconfig.c:1.6 Mon Apr 28 20:24:16 2008
+++ src/usr.sbin/fssconfig/fssconfig.c Sat Jun 18 17:00:26 2011
@@ -1,4 +1,4 @@
-/* $NetBSD: fssconfig.c,v 1.6 2008/04/28 20:24:16 martin Exp $ */
+/* $NetBSD: fssconfig.c,v 1.6.6.1 2011/06/18 17:00:26 bouyer Exp $ */
/*-
* Copyright (c) 2003 The NetBSD Foundation, Inc.
@@ -168,15 +168,16 @@
err(1, "open: %s", argv[0]);
}
+ if ((xflag || istmp) && isreg)
+ fss.fss_flags |= FSS_UNLINK_ON_CREATE;
+ else
+ fss.fss_flags = 0;
if (ioctl(fd, FSSIOCSET, &fss) < 0) {
if (istmp)
unlink(fss.fss_bstore);
err(1, "%s: FSSIOCSET", full);
}
- if ((xflag || istmp) && isreg && unlink(fss.fss_bstore) < 0)
- err(1, "unlink: %s", fss.fss_bstore);
-
if (vflag)
list(1, argv);
}
