Skip to contentSkip to conversation

Mount failure due to restricted access to a point along the mount path

DISCUSSION Suresh Jayaraman 3 participants 3 messages Feb 4, 2013 2 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi all,

I have been looking at the $SUBJECT. And there is already a bug reported in the samba.org bugzilla

https://bugzilla.samba.org/show_bug.cgi?id=8950

cifs gets "Access denied" while trying to get the root inode. Suppose if you are trying to mount /share/dir/subdir and if the user doesn't have access to any of point along the path ie. /share or /dir, mount will fail.

As noted by Jeff Layton in the bug report, this seems to be a known problem since we started using the shared superblock model. Though we can work around this problem by providing neccessary permissions for the entire path, users consider this as a regression. I tend to agree with them and think that may be we should have addressed this problem before pushing the shared superblock changes.

Quoted 1 line

From the changelog it appears that this results in better performance

when we use SMB2.1 leases. How much better performance we are seeing? Is that worth living with this regression?

Does this affect the CIFS users who are not using SMB2.1 in anyway?

Thanks

Signature · 1 line
Suresh Jayaraman

A message that is not in the archive (sent off-list or not yet received): 510FD4FD.3010001-IBi9RG/[email protected]

#2 lore raw
Quoted 18 lines · On Mon, 04 Feb 2013 21:04:21 +0530 Suresh Jayaraman…

Hi all,

I have been looking at the $SUBJECT. And there is already a bug reported in the samba.org bugzilla

https://bugzilla.samba.org/show_bug.cgi?id=8950

cifs gets "Access denied" while trying to get the root inode. Suppose if you are trying to mount /share/dir/subdir and if the user doesn't have access to any of point along the path ie. /share or /dir, mount will fail.

As noted by Jeff Layton in the bug report, this seems to be a known problem since we started using the shared superblock model. Though we can work around this problem by providing neccessary permissions for the entire path, users consider this as a regression. I tend to agree with them and think that may be we should have addressed this problem before pushing the shared superblock changes.

Quoted 9 lines

From the changelog it appears that this results in better performance when we use SMB2.1 leases. How much better performance we are seeing? Is that worth living with this regression?

Does this affect the CIFS users who are not using SMB2.1 in anyway?

Thanks

Well, more than that, it's a cache coherency thing. If you have an inode that's reachable via 2 mounts, you want the caches to be coherent between the two. Note that superblock sharing was a requirement for the fscache patches NFS, so I was never clear why that wasn't the case for CIFS...

In any case, yes this is a problem regardless of CIFS version in use.

What we probably ought to do is come up with a way to instantiate the intermediate dentries with "dummy" inodes at mount time. Those will need to be revalidated (and rehashed) if they ever come to life suddenly though.

Signature · 1 line
Jeff Layton <[email protected]>

A message that is not in the archive (sent off-list or not yet received): 20130204144152.1c839176-9yPaYZwiELC+kQycOl6kW4xkIHaj4LzF@public.gmane.org

#4 lore raw
Quoted 39 lines · Jeff Layton
Quoted 19 lines · On Mon, 04 Feb 2013 21:04:21 +0530 Suresh Jayaraman…

Hi all,

I have been looking at the $SUBJECT. And there is already a bug reported in the samba.org bugzilla

https://bugzilla.samba.org/show_bug.cgi?id=8950

cifs gets "Access denied" while trying to get the root inode. Suppose if you are trying to mount /share/dir/subdir and if the user doesn't have access to any of point along the path ie. /share or /dir, mount will fail.

As noted by Jeff Layton in the bug report, this seems to be a known problem since we started using the shared superblock model. Though we can work around this problem by providing neccessary permissions for the entire path, users consider this as a regression. I tend to agree with them and think that may be we should have addressed this problem before pushing the shared superblock changes.

Quoted 9 lines

From the changelog it appears that this results in better performance when we use SMB2.1 leases. How much better performance we are seeing? Is that worth living with this regression?

Does this affect the CIFS users who are not using SMB2.1 in anyway?

Thanks

Well, more than that, it's a cache coherency thing. If you have an inode that's reachable via 2 mounts, you want the caches to be coherent between the two. Note that superblock sharing was a requirement for the fscache patches NFS, so I was never clear why that wasn't the case for CIFS...

Quoted 6 lines

In any case, yes this is a problem regardless of CIFS version in use.

What we probably ought to do is come up with a way to instantiate the intermediate dentries with "dummy" inodes at mount time. Those will need to be revalidated (and rehashed) if they ever come to life suddenly though.

Agreed

--
Thanks,

Steve