Nandfs use of MNT_VNODE_FOREACH
Ronald Klop
ronald-freebsd8 at klop.yi.org
Mon Dec 31 10:04:32 UTC 2012
On Fri, 28 Dec 2012 00:02:23 +0100, Konstantin Belousov
<kostikbel at gmail.com> wrote:
> I took a look at removing MNT_VNODE_FOREACH interface in the HEAD,
> and apparently we still have one user of the said interface, probably,
> due to cross-commits.
>
> Namely, nandfs utilizes the interface. First, it is obsoleted and
> is going to be removed. Second, I believe that MNT_VNODE_FOREACH_ACTIVE
> would be better choice there, because intent of the loop is to do
> something with each dirty vnode, and dirty vnode must be active, because
> there are dirty buffers attached to it.
>
> That said, use of vget(LK_RETRY) and immediate check for VI_DOOMED
> is not useful, I removed the LK_RETRY from the lockflags. Another issue
> is
> that intent was to avoid sleep for locked vnode, but the check is racy.
> I used LK_NOWAIT instead.
>
> Check for mnt_syncer is also usually done as vp->v_type == VNON, but
> lets postpone this.
>
> Could someone who uses the filesystem test the patch below, please ?
>
> diff --git a/sys/fs/nandfs/nandfs_segment.c
> b/sys/fs/nandfs/nandfs_segment.c
> index 836bead..a73b7f2 100644
> --- a/sys/fs/nandfs/nandfs_segment.c
> +++ b/sys/fs/nandfs/nandfs_segment.c
> @@ -481,36 +481,20 @@ nandfs_iterate_dirty_vnodes(struct mount *mp,
> struct nandfs_seginfo *seginfo)
> int error, lockreq, update;
> td = curthread;
> - lockreq = LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY;
> + lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
> - MNT_ILOCK(mp);
> -
> - MNT_VNODE_FOREACH(vp, mp, mvp) {
> + MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
> update = 0;
> if (mp->mnt_syncer == vp)
> continue;
> - if (VOP_ISLOCKED(vp))
> - continue;
> -
> - VI_LOCK(vp);
> - MNT_IUNLOCK(mp);
> - if (vp->v_iflag & VI_DOOMED) {
> + if (VOP_ISLOCKED(vp)) {
> VI_UNLOCK(vp);
> - MNT_ILOCK(mp);
> - continue;
> - }
> -
> - if ((error = vget(vp, lockreq, td)) != 0) {
> - MNT_ILOCK(mp);
> continue;
> }
> - if (vp->v_iflag & VI_DOOMED) {
> - vput(vp);
> - MNT_ILOCK(mp);
> + if (vget(vp, lockreq, td) != 0)
> continue;
> - }
> nandfs_node = VTON(vp);
> if (nandfs_node->nn_flags & IN_MODIFIED) {
> @@ -532,12 +516,8 @@ nandfs_iterate_dirty_vnodes(struct mount *mp,
> struct nandfs_seginfo *seginfo)
> if (update)
> nandfs_node_update(nandfs_node);
> -
> - MNT_ILOCK(mp);
> }
> - MNT_IUNLOCK(mp);
> -
> return (0);
> }
>
It looks like it hangs. No DHCP yet, so no ping also. But I have remote
debugging.
... [kernel boot messages] ...
Trying to mount root from nandfs:/dev/gnand0s.root []...
WARNING: NANDFS is considered to be a highly experimental feature in
FreeBSD.
after vfs_mountroot()
start_init: trying /sbin/init
sys_execve returned 0
Dec 31 09:57:01 init: init started
(probe0:umass-sim0:0:0:0): Down reving Protocol Version from 2 to 0?
pass0 at umass-sim0 bus 0 scbus0 target 0 lun 0
pass0: <USB 2.0 USB Flash Drive 1100> Removable Direct Access SCSI-0 device
pass0: Serial Number AA00000000002542
pass0: 40.000MB/s transfers
GEOM: new disk da0
da0 at umass-sim0 bus 0 scbus0 target 0 lun 0
da0: <USB 2.0 USB Flash Drive 1100> Removable Direct Access SCSI-0 device
da0: Serial Number AA00000000002542
da0: 40.000MB/s transfers
da0: 3894MB (7975296 512 byte sectors: 255H 63S/T 496C)
Setting hostuuid: 64f53bc5-bfde-11d3-902f-005043016d4c.
Setting hostid: 0x2afd1481.
No suitable dump device was found.
Entropy harvesting: interrupts ethernet
point_to_point(da0:umass-sim0:0:0:0): PREVENT ALLOW MEDIUM REMOVAL. CDB:
1e 0 0 0 1 0
(da0:umass-sim0:0:0:0): CAM status: Auto-Sense Retrieval Failed
(da0:umass-sim0:0:0:0): Error 5, Unretryable error
sha256: /kernel: No such file or directory
kickstart.
Starting file system checks:
/dev/da0s2: FILE SYSTEM CLEAN; SKIPPING CHECKS
/dev/da0s2: clean, 894198 free (2526 frags, 111459 blocks, 0.3%
fragmentation)
/dev/da0s1: FILESYSTEM CLEAN; SKIPPING CHECKS
Mounting local file systems:.
Writing entropy file:.
Setting hostname: sh10.klop.ws.
KDB: enter: Break to debugger
[ thread pid 0 tid 100042 ]
Stopped at kdb_enter+0x48: ldrb r15, [r15, r15, ror r15]!
db> help
alltrace b break bt c call
capture continue countfreebufsd delete dhwatch
dump dwatch examine findstack gdb halt
hwatch kill match next p panic
print ps reboot reset run s
script scripts search set show step
t textdump thread trace unscript until
w watch watchdog where write x
db>
Da0 is not interesting. It was used for installing and is not mounted
automatically.
What debug info can I provide?
Ronald.
More information about the freebsd-fs
mailing list