Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
flag and introduced the d_weak_revalidate dentry operation instead.
We duly removed the flag from NFS superblocks and NFSv4 superblocks,
and added the new dentry operation to NFS dentries .... but not to NFSv4
dentries.
And nobody noticed.
Until today.
A customer reports a situation where mount(....,MS_REMOUNT,..) on an NFS
filesystem hangs because the network has been deconfigured. This makes
perfect sense and I suggested a code change to fix the problem.
However when a colleague was trying to reproduce the problem to validate
the fix, he couldn't. Then nor could I.
The problem is trivially reproducible with NFSv3, and not at all with
NFSv4. The reason is the missing d_weak_revalidate.
We could simply add d_weak_revalidate for NFSv4, but given that it
has been missing for 4.5 years, and the only time anyone noticed was
when the ommission resulted in a better user experience, I do wonder if
we need to. Can we just discard d_weak_revalidate? What purpose does
it serve? I couldn't find one.
Thanks,
NeilBrown
For reference, see
Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a d_weak_revalidate dentry op")
To reproduce the problem at home, on a system that uses systemd:
1/ place (or find) a filesystem image in a file on an NFS filesystem.
2/ mount the nfs filesystem with "noac" - choose v3 or v4
3/ loop-mount the filesystem image read-only somewhere
4/ reboot
If you choose v4, the reboot will succeed, possibly after a 90second timeout.
If you choose v3, the reboot will hang indefinitely in systemd-shutdown while
remounting the nfs filesystem read-only.
If you don't use "noac" it can still hang, but only if something slows
down the reboot enough that attributes have timed out by the time that
systemd-shutdown runs. This happens for our customer.
If the loop-mounted filesystem is not read-only, you get other problems.
We really want systemd to figure out that the loop-mount needs to be
unmounted first. I have ideas concerning that, but it is messy. But
that isn't the only bug here.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On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
> > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
> > flag and introduced the d_weak_revalidate dentry operation instead.
> > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
> > and added the new dentry operation to NFS dentries .... but not to
> > NFSv4
> > dentries.
> >
> > And nobody noticed.
> >
> > Until today.
> >
> > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
> > NFS
> > filesystem hangs because the network has been deconfigured. This
> > makes
> > perfect sense and I suggested a code change to fix the problem.
> > However when a colleague was trying to reproduce the problem to
> > validate
> > the fix, he couldn't. Then nor could I.
> >
> > The problem is trivially reproducible with NFSv3, and not at all with
> > NFSv4. The reason is the missing d_weak_revalidate.
> >
> > We could simply add d_weak_revalidate for NFSv4, but given that it
> > has been missing for 4.5 years, and the only time anyone noticed was
> > when the ommission resulted in a better user experience, I do wonder
> > if
> > we need to. Can we just discard d_weak_revalidate? What purpose
> > does
> > it serve? I couldn't find one.
> >
> > Thanks,
> > NeilBrown
> >
> > For reference, see
> > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
> > d_weak_revalidate dentry op")
> >
> >
> >
> > To reproduce the problem at home, on a system that uses systemd:
> > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
> > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
> > 3/ loop-mount the filesystem image read-only somewhere
> > 4/ reboot
> >
> > If you choose v4, the reboot will succeed, possibly after a 90second
> > timeout.
> > If you choose v3, the reboot will hang indefinitely in systemd-
> > shutdown while
> > remounting the nfs filesystem read-only.
> >
> > If you don't use "noac" it can still hang, but only if something
> > slows
> > down the reboot enough that attributes have timed out by the time
> > that
> > systemd-shutdown runs. This happens for our customer.
> >
> > If the loop-mounted filesystem is not read-only, you get other
> > problems.
> >
> > We really want systemd to figure out that the loop-mount needs to be
> > unmounted first. I have ideas concerning that, but it is messy. But
> > that isn't the only bug here.
>
> The main purpose of d_weak_revalidate() was to catch the issues that
> arise when someone changes the contents of the current working
> directory or its parent on the server. Since '.' and '..' are treated
> specially in the lookup code, they would not be revalidated without
> special treatment. That leads to issues when looking up files as
> ./<filename> or ../<filename>, since the client won't detect that its
> dcache is stale until it tries to use the cached dentry+inode.
>
> The one thing that has changed since its introduction is, I believe,
> the ESTALE handling in the VFS layer. That might fix a lot of the
> dcache lookup bugs that were previously handled by d_weak_revalidate().
> I haven't done an audit to figure out if it actually can handle all of
> them.
>
It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
vfs: allow umount to handle mountpoints without revalidating them
Possibly the fact that we no longer try to revalidate during unmount
means that this is no longer necessary?
The original patch that added d_weak_revalidate had a reproducer in the
patch description. Have you verified that that problem is still not
reproducible when you remove d_weak_revalidate?
--
Jeff Layton <[email protected]>
On Fri, Aug 11 2017, Trond Myklebust wrote:
> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>> flag and introduced the d_weak_revalidate dentry operation instead.
>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>> and added the new dentry operation to NFS dentries .... but not to
>> NFSv4
>> dentries.
>>
>> And nobody noticed.
>>
>> Until today.
>>
>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>> NFS
>> filesystem hangs because the network has been deconfigured. This
>> makes
>> perfect sense and I suggested a code change to fix the problem.
>> However when a colleague was trying to reproduce the problem to
>> validate
>> the fix, he couldn't. Then nor could I.
>>
>> The problem is trivially reproducible with NFSv3, and not at all with
>> NFSv4. The reason is the missing d_weak_revalidate.
>>
>> We could simply add d_weak_revalidate for NFSv4, but given that it
>> has been missing for 4.5 years, and the only time anyone noticed was
>> when the ommission resulted in a better user experience, I do wonder
>> if
>> we need to. Can we just discard d_weak_revalidate? What purpose
>> does
>> it serve? I couldn't find one.
>>
>> Thanks,
>> NeilBrown
>>
>> For reference, see
>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>> d_weak_revalidate dentry op")
>>
>>
>>
>> To reproduce the problem at home, on a system that uses systemd:
>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>> 3/ loop-mount the filesystem image read-only somewhere
>> 4/ reboot
>>
>> If you choose v4, the reboot will succeed, possibly after a 90second
>> timeout.
>> If you choose v3, the reboot will hang indefinitely in systemd-
>> shutdown while
>> remounting the nfs filesystem read-only.
>>
>> If you don't use "noac" it can still hang, but only if something
>> slows
>> down the reboot enough that attributes have timed out by the time
>> that
>> systemd-shutdown runs. This happens for our customer.
>>
>> If the loop-mounted filesystem is not read-only, you get other
>> problems.
>>
>> We really want systemd to figure out that the loop-mount needs to be
>> unmounted first. I have ideas concerning that, but it is messy. But
>> that isn't the only bug here.
>
> The main purpose of d_weak_revalidate() was to catch the issues that
> arise when someone changes the contents of the current working
> directory or its parent on the server. Since '.' and '..' are treated
> specially in the lookup code, they would not be revalidated without
> special treatment. That leads to issues when looking up files as
> ./<filename> or ../<filename>, since the client won't detect that its
> dcache is stale until it tries to use the cached dentry+inode.
I don't think that is quite right.
d_weak_revalidate() is only called from complete_walk() if LOOKUP_JUMPED
is set. The happens when the final component of a path:
- is a mount point
- is ".."
or if the whole path is "/". I thought "." was treated specially too,
but I cannot find that in the code.
After a path walk completes, the operation that acts on the path will
revalidate the inode one way or another so having an extra early
validation seems hard to justify.
If the inode has been removed, ESTALE is returned. The slightly earlier
return of ESTALE might change some behavior.... All I can think of is
that if the directory under a mountpoint gets deleted, the mountpoint is
automatically removed.., but that happens in d_invalidate() which isn't
called when d_weak_revalidate() is called.
>
> The one thing that has changed since its introduction is, I believe,
> the ESTALE handling in the VFS layer. That might fix a lot of the
> dcache lookup bugs that were previously handled by d_weak_revalidate().
> I haven't done an audit to figure out if it actually can handle all of
> them.
I agree that seems like it might be relevant, but I don't see how it
would relate to any of the three cases that d_weak_revalidate affects.
Maybe there is some other change that we don't remember.
Thanks,
NeilBrown
>
> --
> Trond Myklebust
> Linux NFS client maintainer, PrimaryData
> [email protected]
On Fri, Aug 11 2017, Jeff Layton wrote:
> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>> > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>> > flag and introduced the d_weak_revalidate dentry operation instead.
>> > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>> > and added the new dentry operation to NFS dentries .... but not to
>> > NFSv4
>> > dentries.
>> >
>> > And nobody noticed.
>> >
>> > Until today.
>> >
>> > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>> > NFS
>> > filesystem hangs because the network has been deconfigured. This
>> > makes
>> > perfect sense and I suggested a code change to fix the problem.
>> > However when a colleague was trying to reproduce the problem to
>> > validate
>> > the fix, he couldn't. Then nor could I.
>> >
>> > The problem is trivially reproducible with NFSv3, and not at all with
>> > NFSv4. The reason is the missing d_weak_revalidate.
>> >
>> > We could simply add d_weak_revalidate for NFSv4, but given that it
>> > has been missing for 4.5 years, and the only time anyone noticed was
>> > when the ommission resulted in a better user experience, I do wonder
>> > if
>> > we need to. Can we just discard d_weak_revalidate? What purpose
>> > does
>> > it serve? I couldn't find one.
>> >
>> > Thanks,
>> > NeilBrown
>> >
>> > For reference, see
>> > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>> > d_weak_revalidate dentry op")
>> >
>> >
>> >
>> > To reproduce the problem at home, on a system that uses systemd:
>> > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>> > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>> > 3/ loop-mount the filesystem image read-only somewhere
>> > 4/ reboot
>> >
>> > If you choose v4, the reboot will succeed, possibly after a 90second
>> > timeout.
>> > If you choose v3, the reboot will hang indefinitely in systemd-
>> > shutdown while
>> > remounting the nfs filesystem read-only.
>> >
>> > If you don't use "noac" it can still hang, but only if something
>> > slows
>> > down the reboot enough that attributes have timed out by the time
>> > that
>> > systemd-shutdown runs. This happens for our customer.
>> >
>> > If the loop-mounted filesystem is not read-only, you get other
>> > problems.
>> >
>> > We really want systemd to figure out that the loop-mount needs to be
>> > unmounted first. I have ideas concerning that, but it is messy. But
>> > that isn't the only bug here.
>>
>> The main purpose of d_weak_revalidate() was to catch the issues that
>> arise when someone changes the contents of the current working
>> directory or its parent on the server. Since '.' and '..' are treated
>> specially in the lookup code, they would not be revalidated without
>> special treatment. That leads to issues when looking up files as
>> ./<filename> or ../<filename>, since the client won't detect that its
>> dcache is stale until it tries to use the cached dentry+inode.
>>
>> The one thing that has changed since its introduction is, I believe,
>> the ESTALE handling in the VFS layer. That might fix a lot of the
>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>> I haven't done an audit to figure out if it actually can handle all of
>> them.
>>
>
> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>
> vfs: allow umount to handle mountpoints without revalidating them
You say in the comment for that commit:
but there
are cases where we do want to revalidate the root of the fs.
Do you happen to remember what those cases are?
>
> Possibly the fact that we no longer try to revalidate during unmount
> means that this is no longer necessary?
>
> The original patch that added d_weak_revalidate had a reproducer in the
> patch description. Have you verified that that problem is still not
> reproducible when you remove d_weak_revalidate?
I did try the reproducer and it works as expected both with and without
d_weak_revalidate.
On reflection, the problem it displayed was caused by d_revalidate()
being called when the dentry name was irrelevant. We remove that
(fixing the problem) and introduce d_weak_revalidate because we thought
that minimum functionality was still useful. I'm currently not
convinced that even that is needed.
If we discarded d_weak_revalidate(), we could get rid of the special
handling of umount....
Thanks,
NeilBrown
On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
> On Fri, Aug 11 2017, Jeff Layton wrote:
>
> > On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
> > > On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
> > > > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
> > > > flag and introduced the d_weak_revalidate dentry operation instead.
> > > > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
> > > > and added the new dentry operation to NFS dentries .... but not to
> > > > NFSv4
> > > > dentries.
> > > >
> > > > And nobody noticed.
> > > >
> > > > Until today.
> > > >
> > > > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
> > > > NFS
> > > > filesystem hangs because the network has been deconfigured. This
> > > > makes
> > > > perfect sense and I suggested a code change to fix the problem.
> > > > However when a colleague was trying to reproduce the problem to
> > > > validate
> > > > the fix, he couldn't. Then nor could I.
> > > >
> > > > The problem is trivially reproducible with NFSv3, and not at all with
> > > > NFSv4. The reason is the missing d_weak_revalidate.
> > > >
> > > > We could simply add d_weak_revalidate for NFSv4, but given that it
> > > > has been missing for 4.5 years, and the only time anyone noticed was
> > > > when the ommission resulted in a better user experience, I do wonder
> > > > if
> > > > we need to. Can we just discard d_weak_revalidate? What purpose
> > > > does
> > > > it serve? I couldn't find one.
> > > >
> > > > Thanks,
> > > > NeilBrown
> > > >
> > > > For reference, see
> > > > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
> > > > d_weak_revalidate dentry op")
> > > >
> > > >
> > > >
> > > > To reproduce the problem at home, on a system that uses systemd:
> > > > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
> > > > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
> > > > 3/ loop-mount the filesystem image read-only somewhere
> > > > 4/ reboot
> > > >
> > > > If you choose v4, the reboot will succeed, possibly after a 90second
> > > > timeout.
> > > > If you choose v3, the reboot will hang indefinitely in systemd-
> > > > shutdown while
> > > > remounting the nfs filesystem read-only.
> > > >
> > > > If you don't use "noac" it can still hang, but only if something
> > > > slows
> > > > down the reboot enough that attributes have timed out by the time
> > > > that
> > > > systemd-shutdown runs. This happens for our customer.
> > > >
> > > > If the loop-mounted filesystem is not read-only, you get other
> > > > problems.
> > > >
> > > > We really want systemd to figure out that the loop-mount needs to be
> > > > unmounted first. I have ideas concerning that, but it is messy. But
> > > > that isn't the only bug here.
> > >
> > > The main purpose of d_weak_revalidate() was to catch the issues that
> > > arise when someone changes the contents of the current working
> > > directory or its parent on the server. Since '.' and '..' are treated
> > > specially in the lookup code, they would not be revalidated without
> > > special treatment. That leads to issues when looking up files as
> > > ./<filename> or ../<filename>, since the client won't detect that its
> > > dcache is stale until it tries to use the cached dentry+inode.
> > >
> > > The one thing that has changed since its introduction is, I believe,
> > > the ESTALE handling in the VFS layer. That might fix a lot of the
> > > dcache lookup bugs that were previously handled by d_weak_revalidate().
> > > I haven't done an audit to figure out if it actually can handle all of
> > > them.
> > >
> >
> > It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
> >
> > vfs: allow umount to handle mountpoints without revalidating them
>
> You say in the comment for that commit:
>
> but there
> are cases where we do want to revalidate the root of the fs.
>
> Do you happen to remember what those cases are?
>
Not exactly, but I _think_ I might have been assuming that we needed to
ensure that the inode attrs on the root were up to date after the
pathwalk.
I think that was probably wrong. d_revalidate is really intended to
ensure that the dentry in question still points to the same inode. In
the case of the root of the mount though, we don't really care about the
dentry on the server at all. We're attaching the root of the mount to an
inode and don't care of the dentry name changes. If we do need to ensure
the inode attrs are updated, we'll just revalidate them at that point.
> >
> > Possibly the fact that we no longer try to revalidate during unmount
> > means that this is no longer necessary?
> >
> > The original patch that added d_weak_revalidate had a reproducer in the
> > patch description. Have you verified that that problem is still not
> > reproducible when you remove d_weak_revalidate?
>
> I did try the reproducer and it works as expected both with and without
> d_weak_revalidate.
> On reflection, the problem it displayed was caused by d_revalidate()
> being called when the dentry name was irrelevant. We remove that
> (fixing the problem) and introduce d_weak_revalidate because we thought
> that minimum functionality was still useful. I'm currently not
> convinced that even that is needed.
>
> If we discarded d_weak_revalidate(), we could get rid of the special
> handling of umount....
I like idea. I say go for it and we can see what (if anything) breaks?
--
Jeff Layton <[email protected]>
On Mon, Aug 14 2017, Jeff Layton wrote:
> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>
>> > On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>> > > On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>> > > > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>> > > > flag and introduced the d_weak_revalidate dentry operation instead.
>> > > > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>> > > > and added the new dentry operation to NFS dentries .... but not to
>> > > > NFSv4
>> > > > dentries.
>> > > >
>> > > > And nobody noticed.
>> > > >
>> > > > Until today.
>> > > >
>> > > > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>> > > > NFS
>> > > > filesystem hangs because the network has been deconfigured. This
>> > > > makes
>> > > > perfect sense and I suggested a code change to fix the problem.
>> > > > However when a colleague was trying to reproduce the problem to
>> > > > validate
>> > > > the fix, he couldn't. Then nor could I.
>> > > >
>> > > > The problem is trivially reproducible with NFSv3, and not at all with
>> > > > NFSv4. The reason is the missing d_weak_revalidate.
>> > > >
>> > > > We could simply add d_weak_revalidate for NFSv4, but given that it
>> > > > has been missing for 4.5 years, and the only time anyone noticed was
>> > > > when the ommission resulted in a better user experience, I do wonder
>> > > > if
>> > > > we need to. Can we just discard d_weak_revalidate? What purpose
>> > > > does
>> > > > it serve? I couldn't find one.
>> > > >
>> > > > Thanks,
>> > > > NeilBrown
>> > > >
>> > > > For reference, see
>> > > > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>> > > > d_weak_revalidate dentry op")
>> > > >
>> > > >
>> > > >
>> > > > To reproduce the problem at home, on a system that uses systemd:
>> > > > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>> > > > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>> > > > 3/ loop-mount the filesystem image read-only somewhere
>> > > > 4/ reboot
>> > > >
>> > > > If you choose v4, the reboot will succeed, possibly after a 90second
>> > > > timeout.
>> > > > If you choose v3, the reboot will hang indefinitely in systemd-
>> > > > shutdown while
>> > > > remounting the nfs filesystem read-only.
>> > > >
>> > > > If you don't use "noac" it can still hang, but only if something
>> > > > slows
>> > > > down the reboot enough that attributes have timed out by the time
>> > > > that
>> > > > systemd-shutdown runs. This happens for our customer.
>> > > >
>> > > > If the loop-mounted filesystem is not read-only, you get other
>> > > > problems.
>> > > >
>> > > > We really want systemd to figure out that the loop-mount needs to be
>> > > > unmounted first. I have ideas concerning that, but it is messy. But
>> > > > that isn't the only bug here.
>> > >
>> > > The main purpose of d_weak_revalidate() was to catch the issues that
>> > > arise when someone changes the contents of the current working
>> > > directory or its parent on the server. Since '.' and '..' are treated
>> > > specially in the lookup code, they would not be revalidated without
>> > > special treatment. That leads to issues when looking up files as
>> > > ./<filename> or ../<filename>, since the client won't detect that its
>> > > dcache is stale until it tries to use the cached dentry+inode.
>> > >
>> > > The one thing that has changed since its introduction is, I believe,
>> > > the ESTALE handling in the VFS layer. That might fix a lot of the
>> > > dcache lookup bugs that were previously handled by d_weak_revalidate().
>> > > I haven't done an audit to figure out if it actually can handle all of
>> > > them.
>> > >
>> >
>> > It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>> >
>> > vfs: allow umount to handle mountpoints without revalidating them
>>
>> You say in the comment for that commit:
>>
>> but there
>> are cases where we do want to revalidate the root of the fs.
>>
>> Do you happen to remember what those cases are?
>>
>
> Not exactly, but I _think_ I might have been assuming that we needed to
> ensure that the inode attrs on the root were up to date after the
> pathwalk.
>
> I think that was probably wrong. d_revalidate is really intended to
> ensure that the dentry in question still points to the same inode. In
> the case of the root of the mount though, we don't really care about the
> dentry on the server at all. We're attaching the root of the mount to an
> inode and don't care of the dentry name changes. If we do need to ensure
> the inode attrs are updated, we'll just revalidate them at that point.
>
>> >
>> > Possibly the fact that we no longer try to revalidate during unmount
>> > means that this is no longer necessary?
>> >
>> > The original patch that added d_weak_revalidate had a reproducer in the
>> > patch description. Have you verified that that problem is still not
>> > reproducible when you remove d_weak_revalidate?
>>
>> I did try the reproducer and it works as expected both with and without
>> d_weak_revalidate.
>> On reflection, the problem it displayed was caused by d_revalidate()
>> being called when the dentry name was irrelevant. We remove that
>> (fixing the problem) and introduce d_weak_revalidate because we thought
>> that minimum functionality was still useful. I'm currently not
>> convinced that even that is needed.
>>
>> If we discarded d_weak_revalidate(), we could get rid of the special
>> handling of umount....
>
> I like idea. I say go for it and we can see what (if anything) breaks?
Getting rid of d_weak_revalidate is easy enough - hardly any users.
Getting rid of filename_mountpoint() isn't so easy unfortunately.
autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
in autofs4_d_manage()? It would be a shame to keep this infrastructure
around just so that one part of autofs4 can talk to another part of
autofs4.
Do you know if the fact that filename_mountpoint() skips ->d_manage is
important for sys_umount ??
Thanks,
NeilBrown
> --
> Jeff Layton <[email protected]>
On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
> On Mon, Aug 14 2017, Jeff Layton wrote:
>
> > On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
> > > On Fri, Aug 11 2017, Jeff Layton wrote:
> > >
> > > > On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
> > > > > On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
> > > > > > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
> > > > > > flag and introduced the d_weak_revalidate dentry operation instead.
> > > > > > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
> > > > > > and added the new dentry operation to NFS dentries .... but not to
> > > > > > NFSv4
> > > > > > dentries.
> > > > > >
> > > > > > And nobody noticed.
> > > > > >
> > > > > > Until today.
> > > > > >
> > > > > > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
> > > > > > NFS
> > > > > > filesystem hangs because the network has been deconfigured. This
> > > > > > makes
> > > > > > perfect sense and I suggested a code change to fix the problem.
> > > > > > However when a colleague was trying to reproduce the problem to
> > > > > > validate
> > > > > > the fix, he couldn't. Then nor could I.
> > > > > >
> > > > > > The problem is trivially reproducible with NFSv3, and not at all with
> > > > > > NFSv4. The reason is the missing d_weak_revalidate.
> > > > > >
> > > > > > We could simply add d_weak_revalidate for NFSv4, but given that it
> > > > > > has been missing for 4.5 years, and the only time anyone noticed was
> > > > > > when the ommission resulted in a better user experience, I do wonder
> > > > > > if
> > > > > > we need to. Can we just discard d_weak_revalidate? What purpose
> > > > > > does
> > > > > > it serve? I couldn't find one.
> > > > > >
> > > > > > Thanks,
> > > > > > NeilBrown
> > > > > >
> > > > > > For reference, see
> > > > > > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
> > > > > > d_weak_revalidate dentry op")
> > > > > >
> > > > > >
> > > > > >
> > > > > > To reproduce the problem at home, on a system that uses systemd:
> > > > > > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
> > > > > > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
> > > > > > 3/ loop-mount the filesystem image read-only somewhere
> > > > > > 4/ reboot
> > > > > >
> > > > > > If you choose v4, the reboot will succeed, possibly after a 90second
> > > > > > timeout.
> > > > > > If you choose v3, the reboot will hang indefinitely in systemd-
> > > > > > shutdown while
> > > > > > remounting the nfs filesystem read-only.
> > > > > >
> > > > > > If you don't use "noac" it can still hang, but only if something
> > > > > > slows
> > > > > > down the reboot enough that attributes have timed out by the time
> > > > > > that
> > > > > > systemd-shutdown runs. This happens for our customer.
> > > > > >
> > > > > > If the loop-mounted filesystem is not read-only, you get other
> > > > > > problems.
> > > > > >
> > > > > > We really want systemd to figure out that the loop-mount needs to be
> > > > > > unmounted first. I have ideas concerning that, but it is messy. But
> > > > > > that isn't the only bug here.
> > > > >
> > > > > The main purpose of d_weak_revalidate() was to catch the issues that
> > > > > arise when someone changes the contents of the current working
> > > > > directory or its parent on the server. Since '.' and '..' are treated
> > > > > specially in the lookup code, they would not be revalidated without
> > > > > special treatment. That leads to issues when looking up files as
> > > > > ./<filename> or ../<filename>, since the client won't detect that its
> > > > > dcache is stale until it tries to use the cached dentry+inode.
> > > > >
> > > > > The one thing that has changed since its introduction is, I believe,
> > > > > the ESTALE handling in the VFS layer. That might fix a lot of the
> > > > > dcache lookup bugs that were previously handled by d_weak_revalidate().
> > > > > I haven't done an audit to figure out if it actually can handle all of
> > > > > them.
> > > > >
> > > >
> > > > It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
> > > >
> > > > vfs: allow umount to handle mountpoints without revalidating them
> > >
> > > You say in the comment for that commit:
> > >
> > > but there
> > > are cases where we do want to revalidate the root of the fs.
> > >
> > > Do you happen to remember what those cases are?
> > >
> >
> > Not exactly, but I _think_ I might have been assuming that we needed to
> > ensure that the inode attrs on the root were up to date after the
> > pathwalk.
> >
> > I think that was probably wrong. d_revalidate is really intended to
> > ensure that the dentry in question still points to the same inode. In
> > the case of the root of the mount though, we don't really care about the
> > dentry on the server at all. We're attaching the root of the mount to an
> > inode and don't care of the dentry name changes. If we do need to ensure
> > the inode attrs are updated, we'll just revalidate them at that point.
> >
> > > >
> > > > Possibly the fact that we no longer try to revalidate during unmount
> > > > means that this is no longer necessary?
> > > >
> > > > The original patch that added d_weak_revalidate had a reproducer in the
> > > > patch description. Have you verified that that problem is still not
> > > > reproducible when you remove d_weak_revalidate?
> > >
> > > I did try the reproducer and it works as expected both with and without
> > > d_weak_revalidate.
> > > On reflection, the problem it displayed was caused by d_revalidate()
> > > being called when the dentry name was irrelevant. We remove that
> > > (fixing the problem) and introduce d_weak_revalidate because we thought
> > > that minimum functionality was still useful. I'm currently not
> > > convinced that even that is needed.
> > >
> > > If we discarded d_weak_revalidate(), we could get rid of the special
> > > handling of umount....
> >
> > I like idea. I say go for it and we can see what (if anything) breaks?
>
> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>
> Getting rid of filename_mountpoint() isn't so easy unfortunately.
> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
> in autofs4_d_manage()? It would be a shame to keep this infrastructure
> around just so that one part of autofs4 can talk to another part of
> autofs4.
>
> Do you know if the fact that filename_mountpoint() skips ->d_manage is
> important for sys_umount ??
>
> Thanks,
> NeilBrown
>
(cc'ing David and Ian)
I'm less familiar with the automounting machinery, but I imagine you
don't really want to go triggering new mounts when your intent is to
unmount something. As long as that doesn't happen I'd think we'd be ok
here.
--
Jeff Layton <[email protected]>
On Wed, Aug 16 2017, Jeff Layton wrote:
> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>
>> > On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>> > > On Fri, Aug 11 2017, Jeff Layton wrote:
>> > >
>> > > > On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>> > > > > On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>> > > > > > Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>> > > > > > flag and introduced the d_weak_revalidate dentry operation instead.
>> > > > > > We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>> > > > > > and added the new dentry operation to NFS dentries .... but not to
>> > > > > > NFSv4
>> > > > > > dentries.
>> > > > > >
>> > > > > > And nobody noticed.
>> > > > > >
>> > > > > > Until today.
>> > > > > >
>> > > > > > A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>> > > > > > NFS
>> > > > > > filesystem hangs because the network has been deconfigured. This
>> > > > > > makes
>> > > > > > perfect sense and I suggested a code change to fix the problem.
>> > > > > > However when a colleague was trying to reproduce the problem to
>> > > > > > validate
>> > > > > > the fix, he couldn't. Then nor could I.
>> > > > > >
>> > > > > > The problem is trivially reproducible with NFSv3, and not at all with
>> > > > > > NFSv4. The reason is the missing d_weak_revalidate.
>> > > > > >
>> > > > > > We could simply add d_weak_revalidate for NFSv4, but given that it
>> > > > > > has been missing for 4.5 years, and the only time anyone noticed was
>> > > > > > when the ommission resulted in a better user experience, I do wonder
>> > > > > > if
>> > > > > > we need to. Can we just discard d_weak_revalidate? What purpose
>> > > > > > does
>> > > > > > it serve? I couldn't find one.
>> > > > > >
>> > > > > > Thanks,
>> > > > > > NeilBrown
>> > > > > >
>> > > > > > For reference, see
>> > > > > > Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>> > > > > > d_weak_revalidate dentry op")
>> > > > > >
>> > > > > >
>> > > > > >
>> > > > > > To reproduce the problem at home, on a system that uses systemd:
>> > > > > > 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>> > > > > > 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>> > > > > > 3/ loop-mount the filesystem image read-only somewhere
>> > > > > > 4/ reboot
>> > > > > >
>> > > > > > If you choose v4, the reboot will succeed, possibly after a 90second
>> > > > > > timeout.
>> > > > > > If you choose v3, the reboot will hang indefinitely in systemd-
>> > > > > > shutdown while
>> > > > > > remounting the nfs filesystem read-only.
>> > > > > >
>> > > > > > If you don't use "noac" it can still hang, but only if something
>> > > > > > slows
>> > > > > > down the reboot enough that attributes have timed out by the time
>> > > > > > that
>> > > > > > systemd-shutdown runs. This happens for our customer.
>> > > > > >
>> > > > > > If the loop-mounted filesystem is not read-only, you get other
>> > > > > > problems.
>> > > > > >
>> > > > > > We really want systemd to figure out that the loop-mount needs to be
>> > > > > > unmounted first. I have ideas concerning that, but it is messy. But
>> > > > > > that isn't the only bug here.
>> > > > >
>> > > > > The main purpose of d_weak_revalidate() was to catch the issues that
>> > > > > arise when someone changes the contents of the current working
>> > > > > directory or its parent on the server. Since '.' and '..' are treated
>> > > > > specially in the lookup code, they would not be revalidated without
>> > > > > special treatment. That leads to issues when looking up files as
>> > > > > ./<filename> or ../<filename>, since the client won't detect that its
>> > > > > dcache is stale until it tries to use the cached dentry+inode.
>> > > > >
>> > > > > The one thing that has changed since its introduction is, I believe,
>> > > > > the ESTALE handling in the VFS layer. That might fix a lot of the
>> > > > > dcache lookup bugs that were previously handled by d_weak_revalidate().
>> > > > > I haven't done an audit to figure out if it actually can handle all of
>> > > > > them.
>> > > > >
>> > > >
>> > > > It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>> > > >
>> > > > vfs: allow umount to handle mountpoints without revalidating them
>> > >
>> > > You say in the comment for that commit:
>> > >
>> > > but there
>> > > are cases where we do want to revalidate the root of the fs.
>> > >
>> > > Do you happen to remember what those cases are?
>> > >
>> >
>> > Not exactly, but I _think_ I might have been assuming that we needed to
>> > ensure that the inode attrs on the root were up to date after the
>> > pathwalk.
>> >
>> > I think that was probably wrong. d_revalidate is really intended to
>> > ensure that the dentry in question still points to the same inode. In
>> > the case of the root of the mount though, we don't really care about the
>> > dentry on the server at all. We're attaching the root of the mount to an
>> > inode and don't care of the dentry name changes. If we do need to ensure
>> > the inode attrs are updated, we'll just revalidate them at that point.
>> >
>> > > >
>> > > > Possibly the fact that we no longer try to revalidate during unmount
>> > > > means that this is no longer necessary?
>> > > >
>> > > > The original patch that added d_weak_revalidate had a reproducer in the
>> > > > patch description. Have you verified that that problem is still not
>> > > > reproducible when you remove d_weak_revalidate?
>> > >
>> > > I did try the reproducer and it works as expected both with and without
>> > > d_weak_revalidate.
>> > > On reflection, the problem it displayed was caused by d_revalidate()
>> > > being called when the dentry name was irrelevant. We remove that
>> > > (fixing the problem) and introduce d_weak_revalidate because we thought
>> > > that minimum functionality was still useful. I'm currently not
>> > > convinced that even that is needed.
>> > >
>> > > If we discarded d_weak_revalidate(), we could get rid of the special
>> > > handling of umount....
>> >
>> > I like idea. I say go for it and we can see what (if anything) breaks?
>>
>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>
>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>> around just so that one part of autofs4 can talk to another part of
>> autofs4.
>>
>> Do you know if the fact that filename_mountpoint() skips ->d_manage is
>> important for sys_umount ??
>>
>> Thanks,
>> NeilBrown
>>
>
> (cc'ing David and Ian)
>
> I'm less familiar with the automounting machinery, but I imagine you
> don't really want to go triggering new mounts when your intent is to
> unmount something. As long as that doesn't happen I'd think we'd be ok
> here.
New mounts don't get triggered unless asked for (LOOKUP_AUTOMOUNT) so
that isn't a problem. But the d_manage() operation does get called, and
it can block. I think the purpose of the interface is to avoid races
with expiring automounts. Once the decision has been made to expire the
mount, you want to be able to stop anyone getting a new access.
d_manage() can do that. umount() probably doesn't need d_manage() to be
skipped.
autofs *does* for certain ioctls which are used to re-establish
ownership of an automount (e.g. when restarting automountd). The owner
never gets blocked, but if the new automountd isn't the owner yet, it
needs help.
The current code causes the lookup to never call d_manage(), which I
think is overkill. There might be some other filesystem which uses
d_manage() (not actually possible at present) and those shouldn't be
ignored.
So I came up with this. It isn't exactly elegant, but it should avoid
the deadlock that caused autofs4 to start using kern_path_mount().
If we apply this change, and get rid of d_weak_revalidate(), then I
think we can get rid of all the "mountpoint" pathname lookup code.
Ian/David - you do have any thoughts on this approach?
Thanks,
NeilBrown
diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
index beef981aa54f..478a9aafd6d3 100644
--- a/fs/autofs4/autofs_i.h
+++ b/fs/autofs4/autofs_i.h
@@ -271,3 +271,4 @@ static inline void autofs4_del_expiring(struct dentry *dentry)
}
void autofs4_kill_sb(struct super_block *);
+int is_autofs_finder(void);
diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
index dd9f1bebb5a3..f3e84bfed85a 100644
--- a/fs/autofs4/dev-ioctl.c
+++ b/fs/autofs4/dev-ioctl.c
@@ -200,6 +200,13 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
return 0;
}
+static DEFINE_MUTEX(autofs_find);
+static struct task_struct *autofs_finder = NULL;
+int is_autofs_finder(void)
+{
+ return autofs_finder == current;
+}
+
/* Find the topmost mount satisfying test() */
static int find_autofs_mount(const char *pathname,
struct path *res,
@@ -209,7 +216,12 @@ static int find_autofs_mount(const char *pathname,
struct path path;
int err;
- err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
+ mutex_lock(&autofs_find);
+ autofs_finder = current;
+ err = kern_path(pathname, 0, &path);
+ autofs_finder = NULL;
+ mutex_unlock(&autofs_find);
+
if (err)
return err;
err = -ENOENT;
diff --git a/fs/autofs4/root.c b/fs/autofs4/root.c
index d79ced925861..e93a3112d60c 100644
--- a/fs/autofs4/root.c
+++ b/fs/autofs4/root.c
@@ -438,7 +438,7 @@ static int autofs4_d_manage(const struct path *path, bool rcu_walk)
pr_debug("dentry=%p %pd\n", dentry, dentry);
/* The daemon never waits. */
- if (autofs4_oz_mode(sbi)) {
+ if (autofs4_oz_mode(sbi) || is_autofs_finder()) {
if (!path_is_mountpoint(path))
return -EISDIR;
return 0;
On 16/08/17 19:34, Jeff Layton wrote:
> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>
>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>
>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>> NFSv4
>>>>>>> dentries.
>>>>>>>
>>>>>>> And nobody noticed.
>>>>>>>
>>>>>>> Until today.
>>>>>>>
>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>> NFS
>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>> makes
>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>> validate
>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>
>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>
>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>> if
>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>> does
>>>>>>> it serve? I couldn't find one.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> NeilBrown
>>>>>>>
>>>>>>> For reference, see
>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>> d_weak_revalidate dentry op")
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>> 4/ reboot
>>>>>>>
>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>> timeout.
>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>> shutdown while
>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>
>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>> slows
>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>> that
>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>
>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>> problems.
>>>>>>>
>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>> that isn't the only bug here.
>>>>>>
>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>> arise when someone changes the contents of the current working
>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>> special treatment. That leads to issues when looking up files as
>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>
>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>> them.
>>>>>>
>>>>>
>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>
>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>
>>>> You say in the comment for that commit:
>>>>
>>>> but there
>>>> are cases where we do want to revalidate the root of the fs.
>>>>
>>>> Do you happen to remember what those cases are?
>>>>
>>>
>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>> ensure that the inode attrs on the root were up to date after the
>>> pathwalk.
>>>
>>> I think that was probably wrong. d_revalidate is really intended to
>>> ensure that the dentry in question still points to the same inode. In
>>> the case of the root of the mount though, we don't really care about the
>>> dentry on the server at all. We're attaching the root of the mount to an
>>> inode and don't care of the dentry name changes. If we do need to ensure
>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>
>>>>>
>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>> means that this is no longer necessary?
>>>>>
>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>> patch description. Have you verified that that problem is still not
>>>>> reproducible when you remove d_weak_revalidate?
>>>>
>>>> I did try the reproducer and it works as expected both with and without
>>>> d_weak_revalidate.
>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>> being called when the dentry name was irrelevant. We remove that
>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>> that minimum functionality was still useful. I'm currently not
>>>> convinced that even that is needed.
>>>>
>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>> handling of umount....
>>>
>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>
>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>
>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>> around just so that one part of autofs4 can talk to another part of
>> autofs4.
When this was first implemented autofs didn't use kern_path_mountpoint()
(it didn't exist) it used a path lookup on the parent and a separate
lookup for the last component.
It's used for two operations, first to open a file handle to a (possibly)
covered autofs mount, and second to get mounted information about a path
without following past a (possibly covered) autofs mount.
It's less about not triggering an automount or getting stuck in ->d_manage()
and more about resolving paths that are not accessible via normal vfs walks.
I never thought about re-validation for either of these cases and altering
it to the way it was before filename_mountpoint() shouldn't be a problem.
>>
>> Do you know if the fact that filename_mountpoint() skips ->d_manage is
>> important for sys_umount ??
That's a lot harder to answer, d_manage() is needed by autofs to detect
if a path walk needs to block because of some concurrent activity, usually
to avoid concurrent expire (indirectly a umount) to mount races. But it
isn't the umount it's the autofs expire that sets the flag used to block
walks (sends it to the ref-walk path), the (potential) umount comes later
as a result of the expire and the flag is cleared after expire completion.
I can talk about why ->d_manage() isn't (shouldn't) be needed for other vfs
automount users such as NFS etc. if needed.
So filename_mountpoint() skiping ->d_manage shouldn't matter to autofs at
all. If it goes away then autofs will need to revert to an alternate way
(similar to the way it was originally done) of doing the lookup.
>>
>> Thanks,
>> NeilBrown
>>
>
> (cc'ing David and Ian)
>
> I'm less familiar with the automounting machinery, but I imagine you
> don't really want to go triggering new mounts when your intent is to
> unmount something. As long as that doesn't happen I'd think we'd be ok
> here.
Ian
On Thu, Aug 17 2017, Ian Kent wrote:
> On 16/08/17 19:34, Jeff Layton wrote:
>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>
>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>
>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>> NFSv4
>>>>>>>> dentries.
>>>>>>>>
>>>>>>>> And nobody noticed.
>>>>>>>>
>>>>>>>> Until today.
>>>>>>>>
>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>> NFS
>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>> makes
>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>> validate
>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>
>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>
>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>> if
>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>> does
>>>>>>>> it serve? I couldn't find one.
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> NeilBrown
>>>>>>>>
>>>>>>>> For reference, see
>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>> 4/ reboot
>>>>>>>>
>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>> timeout.
>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>> shutdown while
>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>
>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>> slows
>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>> that
>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>
>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>> problems.
>>>>>>>>
>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>> that isn't the only bug here.
>>>>>>>
>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>> arise when someone changes the contents of the current working
>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>
>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>> them.
>>>>>>>
>>>>>>
>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>
>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>
>>>>> You say in the comment for that commit:
>>>>>
>>>>> but there
>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>
>>>>> Do you happen to remember what those cases are?
>>>>>
>>>>
>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>> ensure that the inode attrs on the root were up to date after the
>>>> pathwalk.
>>>>
>>>> I think that was probably wrong. d_revalidate is really intended to
>>>> ensure that the dentry in question still points to the same inode. In
>>>> the case of the root of the mount though, we don't really care about the
>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>
>>>>>>
>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>> means that this is no longer necessary?
>>>>>>
>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>> patch description. Have you verified that that problem is still not
>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>
>>>>> I did try the reproducer and it works as expected both with and without
>>>>> d_weak_revalidate.
>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>> being called when the dentry name was irrelevant. We remove that
>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>> that minimum functionality was still useful. I'm currently not
>>>>> convinced that even that is needed.
>>>>>
>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>> handling of umount....
>>>>
>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>
>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>
>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>> around just so that one part of autofs4 can talk to another part of
>>> autofs4.
>
> When this was first implemented autofs didn't use kern_path_mountpoint()
> (it didn't exist) it used a path lookup on the parent and a separate
> lookup for the last component.
This was before commit 4e44b6852e03 ("Get rid of path_lookup in
autofs4"). This used kern_path().
I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
mount lookup") which replaced the use of kern_path() with
kern_path_mountpoint().
>
> It's used for two operations, first to open a file handle to a (possibly)
> covered autofs mount, and second to get mounted information about a path
> without following past a (possibly covered) autofs mount.
>
> It's less about not triggering an automount or getting stuck in ->d_manage()
> and more about resolving paths that are not accessible via normal vfs walks.
>
> I never thought about re-validation for either of these cases and altering
> it to the way it was before filename_mountpoint() shouldn't be a
> problem.
If it shouldn't be a problem, what justified ac8387199656??
Thanks,
NeilBrown
On 18/08/17 13:24, NeilBrown wrote:
> On Thu, Aug 17 2017, Ian Kent wrote:
>
>> On 16/08/17 19:34, Jeff Layton wrote:
>>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>>
>>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>>
>>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>>> NFSv4
>>>>>>>>> dentries.
>>>>>>>>>
>>>>>>>>> And nobody noticed.
>>>>>>>>>
>>>>>>>>> Until today.
>>>>>>>>>
>>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>>> NFS
>>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>>> makes
>>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>>> validate
>>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>>
>>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>>
>>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>>> if
>>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>>> does
>>>>>>>>> it serve? I couldn't find one.
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> NeilBrown
>>>>>>>>>
>>>>>>>>> For reference, see
>>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>>> 4/ reboot
>>>>>>>>>
>>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>>> timeout.
>>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>>> shutdown while
>>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>>
>>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>>> slows
>>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>>> that
>>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>>
>>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>>> problems.
>>>>>>>>>
>>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>>> that isn't the only bug here.
>>>>>>>>
>>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>>> arise when someone changes the contents of the current working
>>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>>
>>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>>> them.
>>>>>>>>
>>>>>>>
>>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>>
>>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>>
>>>>>> You say in the comment for that commit:
>>>>>>
>>>>>> but there
>>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>>
>>>>>> Do you happen to remember what those cases are?
>>>>>>
>>>>>
>>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>>> ensure that the inode attrs on the root were up to date after the
>>>>> pathwalk.
>>>>>
>>>>> I think that was probably wrong. d_revalidate is really intended to
>>>>> ensure that the dentry in question still points to the same inode. In
>>>>> the case of the root of the mount though, we don't really care about the
>>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>>
>>>>>>>
>>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>>> means that this is no longer necessary?
>>>>>>>
>>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>>> patch description. Have you verified that that problem is still not
>>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>>
>>>>>> I did try the reproducer and it works as expected both with and without
>>>>>> d_weak_revalidate.
>>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>>> being called when the dentry name was irrelevant. We remove that
>>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>>> that minimum functionality was still useful. I'm currently not
>>>>>> convinced that even that is needed.
>>>>>>
>>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>>> handling of umount....
>>>>>
>>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>>
>>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>>
>>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>>> around just so that one part of autofs4 can talk to another part of
>>>> autofs4.
>>
>> When this was first implemented autofs didn't use kern_path_mountpoint()
>> (it didn't exist) it used a path lookup on the parent and a separate
>> lookup for the last component.
>
> This was before commit 4e44b6852e03 ("Get rid of path_lookup in
> autofs4"). This used kern_path().
I have to plead not guilty of this one.
IIRC it broke the requirement of "lookup parent then lookup last component"
rather it walked the whole path then followed up to find the mount point
struct path.
Like it says in the description of ac8387199656 the caller might not yet
"own" the autofs mount which causes a mount to be triggered during the
walk that can't be satisfied because of the deadlock that occurs.
>
> I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
> mount lookup") which replaced the use of kern_path() with
> kern_path_mountpoint().
Probably should have had a Fixes: 4e44b6852e03 ...
>
>>
>> It's used for two operations, first to open a file handle to a (possibly)
>> covered autofs mount, and second to get mounted information about a path
>> without following past a (possibly covered) autofs mount.
>>
>> It's less about not triggering an automount or getting stuck in ->d_manage()
>> and more about resolving paths that are not accessible via normal vfs walks.
>>
>> I never thought about re-validation for either of these cases and altering
>> it to the way it was before filename_mountpoint() shouldn't be a
>> problem.
>
> If it shouldn't be a problem, what justified ac8387199656??
>
> Thanks,
> NeilBrown
>
On 18/08/17 14:47, Ian Kent wrote:
> On 18/08/17 13:24, NeilBrown wrote:
>> On Thu, Aug 17 2017, Ian Kent wrote:
>>
>>> On 16/08/17 19:34, Jeff Layton wrote:
>>>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>>>
>>>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>>>
>>>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>>>> NFSv4
>>>>>>>>>> dentries.
>>>>>>>>>>
>>>>>>>>>> And nobody noticed.
>>>>>>>>>>
>>>>>>>>>> Until today.
>>>>>>>>>>
>>>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>>>> NFS
>>>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>>>> makes
>>>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>>>> validate
>>>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>>>
>>>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>>>
>>>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>>>> if
>>>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>>>> does
>>>>>>>>>> it serve? I couldn't find one.
>>>>>>>>>>
>>>>>>>>>> Thanks,
>>>>>>>>>> NeilBrown
>>>>>>>>>>
>>>>>>>>>> For reference, see
>>>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>>>> 4/ reboot
>>>>>>>>>>
>>>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>>>> timeout.
>>>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>>>> shutdown while
>>>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>>>
>>>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>>>> slows
>>>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>>>> that
>>>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>>>
>>>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>>>> problems.
>>>>>>>>>>
>>>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>>>> that isn't the only bug here.
>>>>>>>>>
>>>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>>>> arise when someone changes the contents of the current working
>>>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>>>
>>>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>>>> them.
>>>>>>>>>
>>>>>>>>
>>>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>>>
>>>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>>>
>>>>>>> You say in the comment for that commit:
>>>>>>>
>>>>>>> but there
>>>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>>>
>>>>>>> Do you happen to remember what those cases are?
>>>>>>>
>>>>>>
>>>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>>>> ensure that the inode attrs on the root were up to date after the
>>>>>> pathwalk.
>>>>>>
>>>>>> I think that was probably wrong. d_revalidate is really intended to
>>>>>> ensure that the dentry in question still points to the same inode. In
>>>>>> the case of the root of the mount though, we don't really care about the
>>>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>>>
>>>>>>>>
>>>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>>>> means that this is no longer necessary?
>>>>>>>>
>>>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>>>> patch description. Have you verified that that problem is still not
>>>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>>>
>>>>>>> I did try the reproducer and it works as expected both with and without
>>>>>>> d_weak_revalidate.
>>>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>>>> being called when the dentry name was irrelevant. We remove that
>>>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>>>> that minimum functionality was still useful. I'm currently not
>>>>>>> convinced that even that is needed.
>>>>>>>
>>>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>>>> handling of umount....
>>>>>>
>>>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>>>
>>>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>>>
>>>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>>>> around just so that one part of autofs4 can talk to another part of
>>>>> autofs4.
>>>
>>> When this was first implemented autofs didn't use kern_path_mountpoint()
>>> (it didn't exist) it used a path lookup on the parent and a separate
>>> lookup for the last component.
>>
>> This was before commit 4e44b6852e03 ("Get rid of path_lookup in
>> autofs4"). This used kern_path().
>
> I have to plead not guilty of this one.
>
> IIRC it broke the requirement of "lookup parent then lookup last component"
> rather it walked the whole path then followed up to find the mount point
> struct path.
>
> Like it says in the description of ac8387199656 the caller might not yet
> "own" the autofs mount which causes a mount to be triggered during the
> walk that can't be satisfied because of the deadlock that occurs.
Also, 4e44b6852e03 fixed another mistake I had made with:
- if (nd.path.mnt->mnt_mountpoint != nd.path.mnt->mnt_root) {
- if (follow_down(&nd.path.mnt, &nd.path.dentry)) {
- struct inode *inode = nd.path.dentry->d_inode;
- magic = inode->i_sb->s_magic;
- }
+ if (path.mnt->mnt_mountpoint != path.mnt->mnt_root) {
+ if (follow_down(&path.mnt, &path.dentry))
+ magic = path.mnt->mnt_sb->s_magic;
}
>
>>
>> I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
>> mount lookup") which replaced the use of kern_path() with
>> kern_path_mountpoint().
>
> Probably should have had a Fixes: 4e44b6852e03 ...
>
>>
>>>
>>> It's used for two operations, first to open a file handle to a (possibly)
>>> covered autofs mount, and second to get mounted information about a path
>>> without following past a (possibly covered) autofs mount.
>>>
>>> It's less about not triggering an automount or getting stuck in ->d_manage()
>>> and more about resolving paths that are not accessible via normal vfs walks.
>>>
>>> I never thought about re-validation for either of these cases and altering
>>> it to the way it was before filename_mountpoint() shouldn't be a
>>> problem.
>>
>> If it shouldn't be a problem, what justified ac8387199656??
>>
>> Thanks,
>> NeilBrown
>>
>
On Fri, Aug 18 2017, Ian Kent wrote:
> On 18/08/17 13:24, NeilBrown wrote:
>> On Thu, Aug 17 2017, Ian Kent wrote:
>>
>>> On 16/08/17 19:34, Jeff Layton wrote:
>>>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>>>
>>>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>>>
>>>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>>>> NFSv4
>>>>>>>>>> dentries.
>>>>>>>>>>
>>>>>>>>>> And nobody noticed.
>>>>>>>>>>
>>>>>>>>>> Until today.
>>>>>>>>>>
>>>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>>>> NFS
>>>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>>>> makes
>>>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>>>> validate
>>>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>>>
>>>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>>>
>>>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>>>> if
>>>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>>>> does
>>>>>>>>>> it serve? I couldn't find one.
>>>>>>>>>>
>>>>>>>>>> Thanks,
>>>>>>>>>> NeilBrown
>>>>>>>>>>
>>>>>>>>>> For reference, see
>>>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>>>> 4/ reboot
>>>>>>>>>>
>>>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>>>> timeout.
>>>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>>>> shutdown while
>>>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>>>
>>>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>>>> slows
>>>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>>>> that
>>>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>>>
>>>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>>>> problems.
>>>>>>>>>>
>>>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>>>> that isn't the only bug here.
>>>>>>>>>
>>>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>>>> arise when someone changes the contents of the current working
>>>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>>>
>>>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>>>> them.
>>>>>>>>>
>>>>>>>>
>>>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>>>
>>>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>>>
>>>>>>> You say in the comment for that commit:
>>>>>>>
>>>>>>> but there
>>>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>>>
>>>>>>> Do you happen to remember what those cases are?
>>>>>>>
>>>>>>
>>>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>>>> ensure that the inode attrs on the root were up to date after the
>>>>>> pathwalk.
>>>>>>
>>>>>> I think that was probably wrong. d_revalidate is really intended to
>>>>>> ensure that the dentry in question still points to the same inode. In
>>>>>> the case of the root of the mount though, we don't really care about the
>>>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>>>
>>>>>>>>
>>>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>>>> means that this is no longer necessary?
>>>>>>>>
>>>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>>>> patch description. Have you verified that that problem is still not
>>>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>>>
>>>>>>> I did try the reproducer and it works as expected both with and without
>>>>>>> d_weak_revalidate.
>>>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>>>> being called when the dentry name was irrelevant. We remove that
>>>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>>>> that minimum functionality was still useful. I'm currently not
>>>>>>> convinced that even that is needed.
>>>>>>>
>>>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>>>> handling of umount....
>>>>>>
>>>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>>>
>>>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>>>
>>>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>>>> around just so that one part of autofs4 can talk to another part of
>>>>> autofs4.
>>>
>>> When this was first implemented autofs didn't use kern_path_mountpoint()
>>> (it didn't exist) it used a path lookup on the parent and a separate
>>> lookup for the last component.
>>
>> This was before commit 4e44b6852e03 ("Get rid of path_lookup in
>> autofs4"). This used kern_path().
>
> I have to plead not guilty of this one.
>
> IIRC it broke the requirement of "lookup parent then lookup last component"
> rather it walked the whole path then followed up to find the mount point
> struct path.
>
> Like it says in the description of ac8387199656 the caller might not yet
> "own" the autofs mount which causes a mount to be triggered during the
> walk that can't be satisfied because of the deadlock that occurs.
A mount isn't triggered by kern_path(pathname, 0, &path).
That '0' would need to include one of
LOOKUP_PARENT | LOOKUP_DIRECTORY |
LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
to trigger an automount (otherwise you just get -EISDIR).
That is why I assumed that ->d_managed was the problem.
>
>>
>> I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
>> mount lookup") which replaced the use of kern_path() with
>> kern_path_mountpoint().
>
> Probably should have had a Fixes: 4e44b6852e03 ...
Still a bit confused as to exactly what was fixed...
Thanks,
NeilBrown
On 21/08/17 14:23, NeilBrown wrote:
> On Fri, Aug 18 2017, Ian Kent wrote:
>
>> On 18/08/17 13:24, NeilBrown wrote:
>>> On Thu, Aug 17 2017, Ian Kent wrote:
>>>
>>>> On 16/08/17 19:34, Jeff Layton wrote:
>>>>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>>>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>>>>
>>>>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>>>>
>>>>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>>>>> NFSv4
>>>>>>>>>>> dentries.
>>>>>>>>>>>
>>>>>>>>>>> And nobody noticed.
>>>>>>>>>>>
>>>>>>>>>>> Until today.
>>>>>>>>>>>
>>>>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>>>>> NFS
>>>>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>>>>> makes
>>>>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>>>>> validate
>>>>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>>>>
>>>>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>>>>
>>>>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>>>>> if
>>>>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>>>>> does
>>>>>>>>>>> it serve? I couldn't find one.
>>>>>>>>>>>
>>>>>>>>>>> Thanks,
>>>>>>>>>>> NeilBrown
>>>>>>>>>>>
>>>>>>>>>>> For reference, see
>>>>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>>>>> 4/ reboot
>>>>>>>>>>>
>>>>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>>>>> timeout.
>>>>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>>>>> shutdown while
>>>>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>>>>
>>>>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>>>>> slows
>>>>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>>>>> that
>>>>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>>>>
>>>>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>>>>> problems.
>>>>>>>>>>>
>>>>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>>>>> that isn't the only bug here.
>>>>>>>>>>
>>>>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>>>>> arise when someone changes the contents of the current working
>>>>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>>>>
>>>>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>>>>> them.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>>>>
>>>>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>>>>
>>>>>>>> You say in the comment for that commit:
>>>>>>>>
>>>>>>>> but there
>>>>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>>>>
>>>>>>>> Do you happen to remember what those cases are?
>>>>>>>>
>>>>>>>
>>>>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>>>>> ensure that the inode attrs on the root were up to date after the
>>>>>>> pathwalk.
>>>>>>>
>>>>>>> I think that was probably wrong. d_revalidate is really intended to
>>>>>>> ensure that the dentry in question still points to the same inode. In
>>>>>>> the case of the root of the mount though, we don't really care about the
>>>>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>>>>
>>>>>>>>>
>>>>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>>>>> means that this is no longer necessary?
>>>>>>>>>
>>>>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>>>>> patch description. Have you verified that that problem is still not
>>>>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>>>>
>>>>>>>> I did try the reproducer and it works as expected both with and without
>>>>>>>> d_weak_revalidate.
>>>>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>>>>> being called when the dentry name was irrelevant. We remove that
>>>>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>>>>> that minimum functionality was still useful. I'm currently not
>>>>>>>> convinced that even that is needed.
>>>>>>>>
>>>>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>>>>> handling of umount....
>>>>>>>
>>>>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>>>>
>>>>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>>>>
>>>>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>>>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>>>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>>>>> around just so that one part of autofs4 can talk to another part of
>>>>>> autofs4.
>>>>
>>>> When this was first implemented autofs didn't use kern_path_mountpoint()
>>>> (it didn't exist) it used a path lookup on the parent and a separate
>>>> lookup for the last component.
>>>
>>> This was before commit 4e44b6852e03 ("Get rid of path_lookup in
>>> autofs4"). This used kern_path().
>>
>> I have to plead not guilty of this one.
>>
>> IIRC it broke the requirement of "lookup parent then lookup last component"
>> rather it walked the whole path then followed up to find the mount point
>> struct path.
>>
>> Like it says in the description of ac8387199656 the caller might not yet
>> "own" the autofs mount which causes a mount to be triggered during the
>> walk that can't be satisfied because of the deadlock that occurs.
>
> A mount isn't triggered by kern_path(pathname, 0, &path).
> That '0' would need to include one of
> LOOKUP_PARENT | LOOKUP_DIRECTORY |
> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>
> to trigger an automount (otherwise you just get -EISDIR).
It's perfectly sensible to think that but there is a case where a
a mount is triggered when using kern_path().
The EISDIR return occurs for positive dentrys, negative dentrys
will still trigger an automount (which is autofs specific,
indirect mount map using nobrowse option, the install default).
>
> That is why I assumed that ->d_managed was the problem.
>
>>
>>>
>>> I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
>>> mount lookup") which replaced the use of kern_path() with
>>> kern_path_mountpoint().
>>
>> Probably should have had a Fixes: 4e44b6852e03 ...
>
> Still a bit confused as to exactly what was fixed...
Hopefully also considering the negative dentry case will clear that up.
Ian
On Mon, Aug 21 2017, Ian Kent wrote:
> On 21/08/17 14:23, NeilBrown wrote:
>> On Fri, Aug 18 2017, Ian Kent wrote:
>>
>>> On 18/08/17 13:24, NeilBrown wrote:
>>>> On Thu, Aug 17 2017, Ian Kent wrote:
>>>>
>>>>> On 16/08/17 19:34, Jeff Layton wrote:
>>>>>> On Wed, 2017-08-16 at 12:43 +1000, NeilBrown wrote:
>>>>>>> On Mon, Aug 14 2017, Jeff Layton wrote:
>>>>>>>
>>>>>>>> On Mon, 2017-08-14 at 09:36 +1000, NeilBrown wrote:
>>>>>>>>> On Fri, Aug 11 2017, Jeff Layton wrote:
>>>>>>>>>
>>>>>>>>>> On Fri, 2017-08-11 at 05:55 +0000, Trond Myklebust wrote:
>>>>>>>>>>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>>>>>>>>>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>>>>>>>>>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>>>>>>>>>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>>>>>>>>>>> and added the new dentry operation to NFS dentries .... but not to
>>>>>>>>>>>> NFSv4
>>>>>>>>>>>> dentries.
>>>>>>>>>>>>
>>>>>>>>>>>> And nobody noticed.
>>>>>>>>>>>>
>>>>>>>>>>>> Until today.
>>>>>>>>>>>>
>>>>>>>>>>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>>>>>>>>>>> NFS
>>>>>>>>>>>> filesystem hangs because the network has been deconfigured. This
>>>>>>>>>>>> makes
>>>>>>>>>>>> perfect sense and I suggested a code change to fix the problem.
>>>>>>>>>>>> However when a colleague was trying to reproduce the problem to
>>>>>>>>>>>> validate
>>>>>>>>>>>> the fix, he couldn't. Then nor could I.
>>>>>>>>>>>>
>>>>>>>>>>>> The problem is trivially reproducible with NFSv3, and not at all with
>>>>>>>>>>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>>>>>>>>>>
>>>>>>>>>>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>>>>>>>>>>> has been missing for 4.5 years, and the only time anyone noticed was
>>>>>>>>>>>> when the ommission resulted in a better user experience, I do wonder
>>>>>>>>>>>> if
>>>>>>>>>>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>>>>>>>>>>> does
>>>>>>>>>>>> it serve? I couldn't find one.
>>>>>>>>>>>>
>>>>>>>>>>>> Thanks,
>>>>>>>>>>>> NeilBrown
>>>>>>>>>>>>
>>>>>>>>>>>> For reference, see
>>>>>>>>>>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>>>>>>>>>>> d_weak_revalidate dentry op")
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> To reproduce the problem at home, on a system that uses systemd:
>>>>>>>>>>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>>>>>>>>>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>>>>>>>>>>> 3/ loop-mount the filesystem image read-only somewhere
>>>>>>>>>>>> 4/ reboot
>>>>>>>>>>>>
>>>>>>>>>>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>>>>>>>>>>> timeout.
>>>>>>>>>>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>>>>>>>>>>> shutdown while
>>>>>>>>>>>> remounting the nfs filesystem read-only.
>>>>>>>>>>>>
>>>>>>>>>>>> If you don't use "noac" it can still hang, but only if something
>>>>>>>>>>>> slows
>>>>>>>>>>>> down the reboot enough that attributes have timed out by the time
>>>>>>>>>>>> that
>>>>>>>>>>>> systemd-shutdown runs. This happens for our customer.
>>>>>>>>>>>>
>>>>>>>>>>>> If the loop-mounted filesystem is not read-only, you get other
>>>>>>>>>>>> problems.
>>>>>>>>>>>>
>>>>>>>>>>>> We really want systemd to figure out that the loop-mount needs to be
>>>>>>>>>>>> unmounted first. I have ideas concerning that, but it is messy. But
>>>>>>>>>>>> that isn't the only bug here.
>>>>>>>>>>>
>>>>>>>>>>> The main purpose of d_weak_revalidate() was to catch the issues that
>>>>>>>>>>> arise when someone changes the contents of the current working
>>>>>>>>>>> directory or its parent on the server. Since '.' and '..' are treated
>>>>>>>>>>> specially in the lookup code, they would not be revalidated without
>>>>>>>>>>> special treatment. That leads to issues when looking up files as
>>>>>>>>>>> ./<filename> or ../<filename>, since the client won't detect that its
>>>>>>>>>>> dcache is stale until it tries to use the cached dentry+inode.
>>>>>>>>>>>
>>>>>>>>>>> The one thing that has changed since its introduction is, I believe,
>>>>>>>>>>> the ESTALE handling in the VFS layer. That might fix a lot of the
>>>>>>>>>>> dcache lookup bugs that were previously handled by d_weak_revalidate().
>>>>>>>>>>> I haven't done an audit to figure out if it actually can handle all of
>>>>>>>>>>> them.
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> It may also be related to 8033426e6bdb2690d302872ac1e1fadaec1a5581:
>>>>>>>>>>
>>>>>>>>>> vfs: allow umount to handle mountpoints without revalidating them
>>>>>>>>>
>>>>>>>>> You say in the comment for that commit:
>>>>>>>>>
>>>>>>>>> but there
>>>>>>>>> are cases where we do want to revalidate the root of the fs.
>>>>>>>>>
>>>>>>>>> Do you happen to remember what those cases are?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Not exactly, but I _think_ I might have been assuming that we needed to
>>>>>>>> ensure that the inode attrs on the root were up to date after the
>>>>>>>> pathwalk.
>>>>>>>>
>>>>>>>> I think that was probably wrong. d_revalidate is really intended to
>>>>>>>> ensure that the dentry in question still points to the same inode. In
>>>>>>>> the case of the root of the mount though, we don't really care about the
>>>>>>>> dentry on the server at all. We're attaching the root of the mount to an
>>>>>>>> inode and don't care of the dentry name changes. If we do need to ensure
>>>>>>>> the inode attrs are updated, we'll just revalidate them at that point.
>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Possibly the fact that we no longer try to revalidate during unmount
>>>>>>>>>> means that this is no longer necessary?
>>>>>>>>>>
>>>>>>>>>> The original patch that added d_weak_revalidate had a reproducer in the
>>>>>>>>>> patch description. Have you verified that that problem is still not
>>>>>>>>>> reproducible when you remove d_weak_revalidate?
>>>>>>>>>
>>>>>>>>> I did try the reproducer and it works as expected both with and without
>>>>>>>>> d_weak_revalidate.
>>>>>>>>> On reflection, the problem it displayed was caused by d_revalidate()
>>>>>>>>> being called when the dentry name was irrelevant. We remove that
>>>>>>>>> (fixing the problem) and introduce d_weak_revalidate because we thought
>>>>>>>>> that minimum functionality was still useful. I'm currently not
>>>>>>>>> convinced that even that is needed.
>>>>>>>>>
>>>>>>>>> If we discarded d_weak_revalidate(), we could get rid of the special
>>>>>>>>> handling of umount....
>>>>>>>>
>>>>>>>> I like idea. I say go for it and we can see what (if anything) breaks?
>>>>>>>
>>>>>>> Getting rid of d_weak_revalidate is easy enough - hardly any users.
>>>>>>>
>>>>>>> Getting rid of filename_mountpoint() isn't so easy unfortunately.
>>>>>>> autofs4 uses kern_path_mountpoint() - presumably to avoid getting stuck
>>>>>>> in autofs4_d_manage()? It would be a shame to keep this infrastructure
>>>>>>> around just so that one part of autofs4 can talk to another part of
>>>>>>> autofs4.
>>>>>
>>>>> When this was first implemented autofs didn't use kern_path_mountpoint()
>>>>> (it didn't exist) it used a path lookup on the parent and a separate
>>>>> lookup for the last component.
>>>>
>>>> This was before commit 4e44b6852e03 ("Get rid of path_lookup in
>>>> autofs4"). This used kern_path().
>>>
>>> I have to plead not guilty of this one.
>>>
>>> IIRC it broke the requirement of "lookup parent then lookup last component"
>>> rather it walked the whole path then followed up to find the mount point
>>> struct path.
>>>
>>> Like it says in the description of ac8387199656 the caller might not yet
>>> "own" the autofs mount which causes a mount to be triggered during the
>>> walk that can't be satisfied because of the deadlock that occurs.
>>
>> A mount isn't triggered by kern_path(pathname, 0, &path).
>> That '0' would need to include one of
>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>
>> to trigger an automount (otherwise you just get -EISDIR).
>
> It's perfectly sensible to think that but there is a case where a
> a mount is triggered when using kern_path().
>
> The EISDIR return occurs for positive dentrys, negative dentrys
> will still trigger an automount (which is autofs specific,
> indirect mount map using nobrowse option, the install default).
Ahh - light dawns. Thanks :-)
NeilBrown
>
>>
>> That is why I assumed that ->d_managed was the problem.
>>
>>>
>>>>
>>>> I'm more interested in commit ac8387199656 ("autofs4 - fix device ioctl
>>>> mount lookup") which replaced the use of kern_path() with
>>>> kern_path_mountpoint().
>>>
>>> Probably should have had a Fixes: 4e44b6852e03 ...
>>
>> Still a bit confused as to exactly what was fixed...
>
> Hopefully also considering the negative dentry case will clear that up.
>
> Ian
On Mon, Aug 21 2017, Ian Kent wrote:
>>
>> A mount isn't triggered by kern_path(pathname, 0, &path).
>> That '0' would need to include one of
>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>
>> to trigger an automount (otherwise you just get -EISDIR).
>
> It's perfectly sensible to think that but there is a case where a
> a mount is triggered when using kern_path().
>
> The EISDIR return occurs for positive dentrys, negative dentrys
> will still trigger an automount (which is autofs specific,
> indirect mount map using nobrowse option, the install default).
Ok, I understand this better now. This difference between direct and
indirect mounts is slightly awkward. It is visible from user-space, but
not elegant to document.
When you use O_PATH to open a direct automount that has not already been
triggered, the open returns the underlying directory (and fstatfs
confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
an indirect automount, it *will* trigger the automount when "nobrowse" is
in effect, but it won't when "browse" is in effect.
So we cannot just say "O_PATH doesn't trigger automounts", which is
essentially what I said in
https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
It might be possible to modify automount so that it was more consistent
- i.e. if the point is triggered by a mkdir has been done, just to the
mkdir. If it is triggered after a mkdir has been done, do the mount. I
guess that might be racy, and in any case is hard to justify.
Maybe I should change it to be about "direct automounts", and add a note
that indirect automounts aren't so predictable.
But back to my original issue of wanting to discard
kern_path_mountpoint, what would you think of the following approach -
slight revised from before.
Thanks,
NeilBrown
diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
index beef981aa54f..7663ea82e68d 100644
--- a/fs/autofs4/autofs_i.h
+++ b/fs/autofs4/autofs_i.h
@@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
/* autofs4_oz_mode(): do we see the man behind the curtain? (The
* processes which do manipulations for us in user space sees the raw
* filesystem without "magic".)
+ * A process performing certain ioctls can get temporary oz status.
*/
+extern struct task_struct *autofs_tmp_oz;
static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
{
- return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
+ return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
+ autofs_tmp_oz == current;
}
struct inode *autofs4_get_inode(struct super_block *, umode_t);
diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
index dd9f1bebb5a3..d76401669a20 100644
--- a/fs/autofs4/dev-ioctl.c
+++ b/fs/autofs4/dev-ioctl.c
@@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
return 0;
}
+struct task_struct *autofs_tmp_oz;
+int kern_path_oz(const char *pathname, int flags, struct path *path)
+{
+ static DEFINE_MUTEX(autofs_oz);
+ int err;
+
+ mutex_lock(&autofs_oz);
+ autofs_tmp_oz = current;
+ err = kern_path(pathname, flags, path);
+ autofs_tmp_oz = NULL;
+ mutex_unlock(&autofs_oz);
+ return err;
+}
+
/* Find the topmost mount satisfying test() */
static int find_autofs_mount(const char *pathname,
struct path *res,
@@ -209,7 +223,8 @@ static int find_autofs_mount(const char *pathname,
struct path path;
int err;
- err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
+ err = kern_path_oz(pathname, 0, &path);
+
if (err)
return err;
err = -ENOENT;
@@ -552,8 +567,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
if (!fp || param->ioctlfd == -1) {
if (autofs_type_any(type))
- err = kern_path_mountpoint(AT_FDCWD,
- name, &path, LOOKUP_FOLLOW);
+ err = kern_path_oz(name, LOOKUP_FOLLOW, &path);
else
err = find_autofs_mount(name, &path,
test_by_type, &type);
On 23/08/17 09:06, NeilBrown wrote:
> On Mon, Aug 21 2017, Ian Kent wrote:
>
>>>
>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>> That '0' would need to include one of
>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>
>>> to trigger an automount (otherwise you just get -EISDIR).
>>
>> It's perfectly sensible to think that but there is a case where a
>> a mount is triggered when using kern_path().
>>
>> The EISDIR return occurs for positive dentrys, negative dentrys
>> will still trigger an automount (which is autofs specific,
>> indirect mount map using nobrowse option, the install default).
>
> Ok, I understand this better now. This difference between direct and
> indirect mounts is slightly awkward. It is visible from user-space, but
> not elegant to document.
> When you use O_PATH to open a direct automount that has not already been
> triggered, the open returns the underlying directory (and fstatfs
> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
> an indirect automount, it *will* trigger the automount when "nobrowse" is
> in effect, but it won't when "browse" is in effect.
That inconsistency has bothered me for quite a while now.
It was carried over from the autofs module behavior when automounting
support was added to the VFS. What's worse is it prevents the use of
the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
statx().
There is some risk in changing that so it does work but it really does
need to work to enable userspace to not trigger an automount by using
this flag.
So that's (hopefully) going to change soonish, see:
http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
The result should be that stat family calls don't trigger automounts except
for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>
> So we cannot just say "O_PATH doesn't trigger automounts", which is
> essentially what I said in
>
> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>
> It might be possible to modify automount so that it was more consistent
> - i.e. if the point is triggered by a mkdir has been done, just to the
> mkdir. If it is triggered after a mkdir has been done, do the mount. I
> guess that might be racy, and in any case is hard to justify.
>
> Maybe I should change it to be about "direct automounts", and add a note
> that indirect automounts aren't so predictable.
Right and the semantics should be much more consistent in the near future.
I hope (and expect) this semantic change won't cause problems.
>
> But back to my original issue of wanting to discard
> kern_path_mountpoint, what would you think of the following approach -
> slight revised from before.
>
> Thanks,
> NeilBrown
>
> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
> index beef981aa54f..7663ea82e68d 100644
> --- a/fs/autofs4/autofs_i.h
> +++ b/fs/autofs4/autofs_i.h
> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
> * processes which do manipulations for us in user space sees the raw
> * filesystem without "magic".)
> + * A process performing certain ioctls can get temporary oz status.
> */
> +extern struct task_struct *autofs_tmp_oz;
> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
> {
> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
> + autofs_tmp_oz == current;
> }
>
> struct inode *autofs4_get_inode(struct super_block *, umode_t);
> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
> index dd9f1bebb5a3..d76401669a20 100644
> --- a/fs/autofs4/dev-ioctl.c
> +++ b/fs/autofs4/dev-ioctl.c
> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
> return 0;
> }
>
> +struct task_struct *autofs_tmp_oz;
> +int kern_path_oz(const char *pathname, int flags, struct path *path)
> +{
> + static DEFINE_MUTEX(autofs_oz);
> + int err;
> +
> + mutex_lock(&autofs_oz);
> + autofs_tmp_oz = current;
> + err = kern_path(pathname, flags, path);
> + autofs_tmp_oz = NULL;
> + mutex_unlock(&autofs_oz);
> + return err;
> +}
> +
It's simple enough but does look like it will attract criticism as being
a hack!
The kern_path_locked() function is very similar to what was originally
done, along with code to look down the mount stack (rather than up the
way it does now) to get the mount point. In this case, to be valid the
dentry can't be a symlink so that fits kern_path_locked() too.
So maybe it is worth going back to the way it was in the beginning and
be done with it .... OTOH Al must have had a reason for changing the
way it was done that I didn't get.
> /* Find the topmost mount satisfying test() */
> static int find_autofs_mount(const char *pathname,
> struct path *res,
> @@ -209,7 +223,8 @@ static int find_autofs_mount(const char *pathname,
> struct path path;
> int err;
>
> - err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
> + err = kern_path_oz(pathname, 0, &path);
> +
> if (err)
> return err;
> err = -ENOENT;
> @@ -552,8 +567,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
>
> if (!fp || param->ioctlfd == -1) {
> if (autofs_type_any(type))
> - err = kern_path_mountpoint(AT_FDCWD,
> - name, &path, LOOKUP_FOLLOW);
> + err = kern_path_oz(name, LOOKUP_FOLLOW, &path);
> else
> err = find_autofs_mount(name, &path,
> test_by_type, &type);
>
On 23/08/17 10:32, Ian Kent wrote:
> On 23/08/17 09:06, NeilBrown wrote:
>> On Mon, Aug 21 2017, Ian Kent wrote:
>>
>>>>
>>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>>> That '0' would need to include one of
>>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>>
>>>> to trigger an automount (otherwise you just get -EISDIR).
>>>
>>> It's perfectly sensible to think that but there is a case where a
>>> a mount is triggered when using kern_path().
>>>
>>> The EISDIR return occurs for positive dentrys, negative dentrys
>>> will still trigger an automount (which is autofs specific,
>>> indirect mount map using nobrowse option, the install default).
>>
>> Ok, I understand this better now. This difference between direct and
>> indirect mounts is slightly awkward. It is visible from user-space, but
>> not elegant to document.
>> When you use O_PATH to open a direct automount that has not already been
>> triggered, the open returns the underlying directory (and fstatfs
>> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
>> an indirect automount, it *will* trigger the automount when "nobrowse" is
>> in effect, but it won't when "browse" is in effect.
>
> That inconsistency has bothered me for quite a while now.
>
> It was carried over from the autofs module behavior when automounting
> support was added to the VFS. What's worse is it prevents the use of
> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
> statx().
>
> There is some risk in changing that so it does work but it really does
> need to work to enable userspace to not trigger an automount by using
> this flag.
>
> So that's (hopefully) going to change soonish, see:
> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>
> The result should be that stat family calls don't trigger automounts except
> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>
>>
>> So we cannot just say "O_PATH doesn't trigger automounts", which is
>> essentially what I said in
>>
>> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>>
>> It might be possible to modify automount so that it was more consistent
>> - i.e. if the point is triggered by a mkdir has been done, just to the
>> mkdir. If it is triggered after a mkdir has been done, do the mount. I
>> guess that might be racy, and in any case is hard to justify.
>>
>> Maybe I should change it to be about "direct automounts", and add a note
>> that indirect automounts aren't so predictable.
>
> Right and the semantics should be much more consistent in the near future.
> I hope (and expect) this semantic change won't cause problems.
>
>>
>> But back to my original issue of wanting to discard
>> kern_path_mountpoint, what would you think of the following approach -
>> slight revised from before.
>>
>> Thanks,
>> NeilBrown
>>
>> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
>> index beef981aa54f..7663ea82e68d 100644
>> --- a/fs/autofs4/autofs_i.h
>> +++ b/fs/autofs4/autofs_i.h
>> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
>> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
>> * processes which do manipulations for us in user space sees the raw
>> * filesystem without "magic".)
>> + * A process performing certain ioctls can get temporary oz status.
>> */
>> +extern struct task_struct *autofs_tmp_oz;
>> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
>> {
>> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
>> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
>> + autofs_tmp_oz == current;
>> }
>>
>> struct inode *autofs4_get_inode(struct super_block *, umode_t);
>> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
>> index dd9f1bebb5a3..d76401669a20 100644
>> --- a/fs/autofs4/dev-ioctl.c
>> +++ b/fs/autofs4/dev-ioctl.c
>> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
>> return 0;
>> }
>>
>> +struct task_struct *autofs_tmp_oz;
>> +int kern_path_oz(const char *pathname, int flags, struct path *path)
>> +{
>> + static DEFINE_MUTEX(autofs_oz);
>> + int err;
>> +
>> + mutex_lock(&autofs_oz);
>> + autofs_tmp_oz = current;
>> + err = kern_path(pathname, flags, path);
>> + autofs_tmp_oz = NULL;
>> + mutex_unlock(&autofs_oz);
>> + return err;
>> +}
>> +
>
> It's simple enough but does look like it will attract criticism as being
> a hack!
>
> The kern_path_locked() function is very similar to what was originally
> done, along with code to look down the mount stack (rather than up the
> way it does now) to get the mount point. In this case, to be valid the
> dentry can't be a symlink so that fits kern_path_locked() too.
Oh wait, that __lookup_hash() tries too hard to resolve the dentry,
that won't quite work, and maybe d_lookup() can't be used safely in
this context either ....
>
> So maybe it is worth going back to the way it was in the beginning and
> be done with it .... OTOH Al must have had a reason for changing the
> way it was done that I didn't get.
>
>> /* Find the topmost mount satisfying test() */
>> static int find_autofs_mount(const char *pathname,
>> struct path *res,
>> @@ -209,7 +223,8 @@ static int find_autofs_mount(const char *pathname,
>> struct path path;
>> int err;
>>
>> - err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
>> + err = kern_path_oz(pathname, 0, &path);
>> +
>> if (err)
>> return err;
>> err = -ENOENT;
>> @@ -552,8 +567,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
>>
>> if (!fp || param->ioctlfd == -1) {
>> if (autofs_type_any(type))
>> - err = kern_path_mountpoint(AT_FDCWD,
>> - name, &path, LOOKUP_FOLLOW);
>> + err = kern_path_oz(name, LOOKUP_FOLLOW, &path);
>> else
>> err = find_autofs_mount(name, &path,
>> test_by_type, &type);
>>
>
On 23/08/17 10:40, Ian Kent wrote:
> On 23/08/17 10:32, Ian Kent wrote:
>> On 23/08/17 09:06, NeilBrown wrote:
>>> On Mon, Aug 21 2017, Ian Kent wrote:
>>>
>>>>>
>>>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>>>> That '0' would need to include one of
>>>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>>>
>>>>> to trigger an automount (otherwise you just get -EISDIR).
>>>>
>>>> It's perfectly sensible to think that but there is a case where a
>>>> a mount is triggered when using kern_path().
>>>>
>>>> The EISDIR return occurs for positive dentrys, negative dentrys
>>>> will still trigger an automount (which is autofs specific,
>>>> indirect mount map using nobrowse option, the install default).
>>>
>>> Ok, I understand this better now. This difference between direct and
>>> indirect mounts is slightly awkward. It is visible from user-space, but
>>> not elegant to document.
>>> When you use O_PATH to open a direct automount that has not already been
>>> triggered, the open returns the underlying directory (and fstatfs
>>> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
>>> an indirect automount, it *will* trigger the automount when "nobrowse" is
>>> in effect, but it won't when "browse" is in effect.
>>
>> That inconsistency has bothered me for quite a while now.
>>
>> It was carried over from the autofs module behavior when automounting
>> support was added to the VFS. What's worse is it prevents the use of
>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>> statx().
>>
>> There is some risk in changing that so it does work but it really does
>> need to work to enable userspace to not trigger an automount by using
>> this flag.
>>
>> So that's (hopefully) going to change soonish, see:
>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>
>> The result should be that stat family calls don't trigger automounts except
>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>
>>>
>>> So we cannot just say "O_PATH doesn't trigger automounts", which is
>>> essentially what I said in
>>>
>>> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>>>
>>> It might be possible to modify automount so that it was more consistent
>>> - i.e. if the point is triggered by a mkdir has been done, just to the
>>> mkdir. If it is triggered after a mkdir has been done, do the mount. I
>>> guess that might be racy, and in any case is hard to justify.
>>>
>>> Maybe I should change it to be about "direct automounts", and add a note
>>> that indirect automounts aren't so predictable.
>>
>> Right and the semantics should be much more consistent in the near future.
>> I hope (and expect) this semantic change won't cause problems.
>>
>>>
>>> But back to my original issue of wanting to discard
>>> kern_path_mountpoint, what would you think of the following approach -
>>> slight revised from before.
>>>
>>> Thanks,
>>> NeilBrown
>>>
>>> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
>>> index beef981aa54f..7663ea82e68d 100644
>>> --- a/fs/autofs4/autofs_i.h
>>> +++ b/fs/autofs4/autofs_i.h
>>> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
>>> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
>>> * processes which do manipulations for us in user space sees the raw
>>> * filesystem without "magic".)
>>> + * A process performing certain ioctls can get temporary oz status.
>>> */
>>> +extern struct task_struct *autofs_tmp_oz;
>>> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
>>> {
>>> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
>>> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
>>> + autofs_tmp_oz == current;
>>> }
>>>
>>> struct inode *autofs4_get_inode(struct super_block *, umode_t);
>>> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
>>> index dd9f1bebb5a3..d76401669a20 100644
>>> --- a/fs/autofs4/dev-ioctl.c
>>> +++ b/fs/autofs4/dev-ioctl.c
>>> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
>>> return 0;
>>> }
>>>
>>> +struct task_struct *autofs_tmp_oz;
>>> +int kern_path_oz(const char *pathname, int flags, struct path *path)
>>> +{
>>> + static DEFINE_MUTEX(autofs_oz);
>>> + int err;
>>> +
>>> + mutex_lock(&autofs_oz);
>>> + autofs_tmp_oz = current;
>>> + err = kern_path(pathname, flags, path);
>>> + autofs_tmp_oz = NULL;
>>> + mutex_unlock(&autofs_oz);
>>> + return err;
>>> +}
>>> +
>>
>> It's simple enough but does look like it will attract criticism as being
>> a hack!
>>
>> The kern_path_locked() function is very similar to what was originally
>> done, along with code to look down the mount stack (rather than up the
>> way it does now) to get the mount point. In this case, to be valid the
>> dentry can't be a symlink so that fits kern_path_locked() too.
>
> Oh wait, that __lookup_hash() tries too hard to resolve the dentry,
> that won't quite work, and maybe d_lookup() can't be used safely in
> this context either ....
Umm .. d_lookup() does look ok so maybe path_parentat() + d_lookup()
would be ok.
>
>>
>> So maybe it is worth going back to the way it was in the beginning and
>> be done with it .... OTOH Al must have had a reason for changing the
>> way it was done that I didn't get.
>>
>>> /* Find the topmost mount satisfying test() */
>>> static int find_autofs_mount(const char *pathname,
>>> struct path *res,
>>> @@ -209,7 +223,8 @@ static int find_autofs_mount(const char *pathname,
>>> struct path path;
>>> int err;
>>>
>>> - err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
>>> + err = kern_path_oz(pathname, 0, &path);
>>> +
>>> if (err)
>>> return err;
>>> err = -ENOENT;
>>> @@ -552,8 +567,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
>>>
>>> if (!fp || param->ioctlfd == -1) {
>>> if (autofs_type_any(type))
>>> - err = kern_path_mountpoint(AT_FDCWD,
>>> - name, &path, LOOKUP_FOLLOW);
>>> + err = kern_path_oz(name, LOOKUP_FOLLOW, &path);
>>> else
>>> err = find_autofs_mount(name, &path,
>>> test_by_type, &type);
>>>
>>
>
On 23/08/17 10:54, Ian Kent wrote:
> On 23/08/17 10:40, Ian Kent wrote:
>> On 23/08/17 10:32, Ian Kent wrote:
>>> On 23/08/17 09:06, NeilBrown wrote:
>>>> On Mon, Aug 21 2017, Ian Kent wrote:
>>>>
>>>>>>
>>>>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>>>>> That '0' would need to include one of
>>>>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>>>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>>>>
>>>>>> to trigger an automount (otherwise you just get -EISDIR).
>>>>>
>>>>> It's perfectly sensible to think that but there is a case where a
>>>>> a mount is triggered when using kern_path().
>>>>>
>>>>> The EISDIR return occurs for positive dentrys, negative dentrys
>>>>> will still trigger an automount (which is autofs specific,
>>>>> indirect mount map using nobrowse option, the install default).
>>>>
>>>> Ok, I understand this better now. This difference between direct and
>>>> indirect mounts is slightly awkward. It is visible from user-space, but
>>>> not elegant to document.
>>>> When you use O_PATH to open a direct automount that has not already been
>>>> triggered, the open returns the underlying directory (and fstatfs
>>>> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
>>>> an indirect automount, it *will* trigger the automount when "nobrowse" is
>>>> in effect, but it won't when "browse" is in effect.
>>>
>>> That inconsistency has bothered me for quite a while now.
>>>
>>> It was carried over from the autofs module behavior when automounting
>>> support was added to the VFS. What's worse is it prevents the use of
>>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>>> statx().
>>>
>>> There is some risk in changing that so it does work but it really does
>>> need to work to enable userspace to not trigger an automount by using
>>> this flag.
>>>
>>> So that's (hopefully) going to change soonish, see:
>>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>>
>>> The result should be that stat family calls don't trigger automounts except
>>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>>
>>>>
>>>> So we cannot just say "O_PATH doesn't trigger automounts", which is
>>>> essentially what I said in
>>>>
>>>> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>>>>
>>>> It might be possible to modify automount so that it was more consistent
>>>> - i.e. if the point is triggered by a mkdir has been done, just to the
>>>> mkdir. If it is triggered after a mkdir has been done, do the mount. I
>>>> guess that might be racy, and in any case is hard to justify.
>>>>
>>>> Maybe I should change it to be about "direct automounts", and add a note
>>>> that indirect automounts aren't so predictable.
>>>
>>> Right and the semantics should be much more consistent in the near future.
>>> I hope (and expect) this semantic change won't cause problems.
>>>
>>>>
>>>> But back to my original issue of wanting to discard
>>>> kern_path_mountpoint, what would you think of the following approach -
>>>> slight revised from before.
>>>>
>>>> Thanks,
>>>> NeilBrown
>>>>
>>>> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
>>>> index beef981aa54f..7663ea82e68d 100644
>>>> --- a/fs/autofs4/autofs_i.h
>>>> +++ b/fs/autofs4/autofs_i.h
>>>> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
>>>> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
>>>> * processes which do manipulations for us in user space sees the raw
>>>> * filesystem without "magic".)
>>>> + * A process performing certain ioctls can get temporary oz status.
>>>> */
>>>> +extern struct task_struct *autofs_tmp_oz;
>>>> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
>>>> {
>>>> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
>>>> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
>>>> + autofs_tmp_oz == current;
>>>> }
>>>>
>>>> struct inode *autofs4_get_inode(struct super_block *, umode_t);
>>>> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
>>>> index dd9f1bebb5a3..d76401669a20 100644
>>>> --- a/fs/autofs4/dev-ioctl.c
>>>> +++ b/fs/autofs4/dev-ioctl.c
>>>> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
>>>> return 0;
>>>> }
>>>>
>>>> +struct task_struct *autofs_tmp_oz;
>>>> +int kern_path_oz(const char *pathname, int flags, struct path *path)
>>>> +{
>>>> + static DEFINE_MUTEX(autofs_oz);
>>>> + int err;
>>>> +
>>>> + mutex_lock(&autofs_oz);
>>>> + autofs_tmp_oz = current;
>>>> + err = kern_path(pathname, flags, path);
>>>> + autofs_tmp_oz = NULL;
>>>> + mutex_unlock(&autofs_oz);
>>>> + return err;
>>>> +}
>>>> +
>>>
>>> It's simple enough but does look like it will attract criticism as being
>>> a hack!
>>>
>>> The kern_path_locked() function is very similar to what was originally
>>> done, along with code to look down the mount stack (rather than up the
>>> way it does now) to get the mount point. In this case, to be valid the
>>> dentry can't be a symlink so that fits kern_path_locked() too.
>>
>> Oh wait, that __lookup_hash() tries too hard to resolve the dentry,
>> that won't quite work, and maybe d_lookup() can't be used safely in
>> this context either ....
>
> Umm .. d_lookup() does look ok so maybe path_parentat() + d_lookup()
> would be ok.
Double Umm ... with the patch above kern_path() with flags 0 or
LOOKUP_FOLLOW should get either EISDIR or ENOENT ... maybe I should
think occasionally !!
>
>>
>>>
>>> So maybe it is worth going back to the way it was in the beginning and
>>> be done with it .... OTOH Al must have had a reason for changing the
>>> way it was done that I didn't get.
>>>
>>>> /* Find the topmost mount satisfying test() */
>>>> static int find_autofs_mount(const char *pathname,
>>>> struct path *res,
>>>> @@ -209,7 +223,8 @@ static int find_autofs_mount(const char *pathname,
>>>> struct path path;
>>>> int err;
>>>>
>>>> - err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
>>>> + err = kern_path_oz(pathname, 0, &path);
>>>> +
>>>> if (err)
>>>> return err;
>>>> err = -ENOENT;
>>>> @@ -552,8 +567,7 @@ static int autofs_dev_ioctl_ismountpoint(struct file *fp,
>>>>
>>>> if (!fp || param->ioctlfd == -1) {
>>>> if (autofs_type_any(type))
>>>> - err = kern_path_mountpoint(AT_FDCWD,
>>>> - name, &path, LOOKUP_FOLLOW);
>>>> + err = kern_path_oz(name, LOOKUP_FOLLOW, &path);
>>>> else
>>>> err = find_autofs_mount(name, &path,
>>>> test_by_type, &type);
>>>>
>>>
>>
>
On Wed, Aug 23 2017, Ian Kent wrote:
> On 23/08/17 10:32, Ian Kent wrote:
>> On 23/08/17 09:06, NeilBrown wrote:
>>> On Mon, Aug 21 2017, Ian Kent wrote:
>>>
>>>>>
>>>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>>>> That '0' would need to include one of
>>>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>>>
>>>>> to trigger an automount (otherwise you just get -EISDIR).
>>>>
>>>> It's perfectly sensible to think that but there is a case where a
>>>> a mount is triggered when using kern_path().
>>>>
>>>> The EISDIR return occurs for positive dentrys, negative dentrys
>>>> will still trigger an automount (which is autofs specific,
>>>> indirect mount map using nobrowse option, the install default).
>>>
>>> Ok, I understand this better now. This difference between direct and
>>> indirect mounts is slightly awkward. It is visible from user-space, but
>>> not elegant to document.
>>> When you use O_PATH to open a direct automount that has not already been
>>> triggered, the open returns the underlying directory (and fstatfs
>>> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
>>> an indirect automount, it *will* trigger the automount when "nobrowse" is
>>> in effect, but it won't when "browse" is in effect.
>>
>> That inconsistency has bothered me for quite a while now.
>>
>> It was carried over from the autofs module behavior when automounting
>> support was added to the VFS. What's worse is it prevents the use of
>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>> statx().
>>
>> There is some risk in changing that so it does work but it really does
>> need to work to enable userspace to not trigger an automount by using
>> this flag.
>>
>> So that's (hopefully) going to change soonish, see:
>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>
>> The result should be that stat family calls don't trigger automounts except
>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>
>>>
>>> So we cannot just say "O_PATH doesn't trigger automounts", which is
>>> essentially what I said in
>>>
>>> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>>>
>>> It might be possible to modify automount so that it was more consistent
>>> - i.e. if the point is triggered by a mkdir has been done, just to the
>>> mkdir. If it is triggered after a mkdir has been done, do the mount. I
>>> guess that might be racy, and in any case is hard to justify.
>>>
>>> Maybe I should change it to be about "direct automounts", and add a note
>>> that indirect automounts aren't so predictable.
>>
>> Right and the semantics should be much more consistent in the near future.
>> I hope (and expect) this semantic change won't cause problems.
>>
>>>
>>> But back to my original issue of wanting to discard
>>> kern_path_mountpoint, what would you think of the following approach -
>>> slight revised from before.
>>>
>>> Thanks,
>>> NeilBrown
>>>
>>> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
>>> index beef981aa54f..7663ea82e68d 100644
>>> --- a/fs/autofs4/autofs_i.h
>>> +++ b/fs/autofs4/autofs_i.h
>>> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
>>> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
>>> * processes which do manipulations for us in user space sees the raw
>>> * filesystem without "magic".)
>>> + * A process performing certain ioctls can get temporary oz status.
>>> */
>>> +extern struct task_struct *autofs_tmp_oz;
>>> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
>>> {
>>> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
>>> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
>>> + autofs_tmp_oz == current;
>>> }
>>>
>>> struct inode *autofs4_get_inode(struct super_block *, umode_t);
>>> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
>>> index dd9f1bebb5a3..d76401669a20 100644
>>> --- a/fs/autofs4/dev-ioctl.c
>>> +++ b/fs/autofs4/dev-ioctl.c
>>> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
>>> return 0;
>>> }
>>>
>>> +struct task_struct *autofs_tmp_oz;
>>> +int kern_path_oz(const char *pathname, int flags, struct path *path)
>>> +{
>>> + static DEFINE_MUTEX(autofs_oz);
>>> + int err;
>>> +
>>> + mutex_lock(&autofs_oz);
>>> + autofs_tmp_oz = current;
>>> + err = kern_path(pathname, flags, path);
>>> + autofs_tmp_oz = NULL;
>>> + mutex_unlock(&autofs_oz);
>>> + return err;
>>> +}
>>> +
>>
>> It's simple enough but does look like it will attract criticism as being
>> a hack!
>>
>> The kern_path_locked() function is very similar to what was originally
>> done, along with code to look down the mount stack (rather than up the
>> way it does now) to get the mount point. In this case, to be valid the
>> dentry can't be a symlink so that fits kern_path_locked() too.
>
> Oh wait, that __lookup_hash() tries too hard to resolve the dentry,
> that won't quite work, and maybe d_lookup() can't be used safely in
> this context either ....
>
Why do you think that __look_hash() tries too hard?
It does call into the filesystem ->lookup() if the name isn't in the
cache, which probably isn't strictly needed, but that isn't harmful and
the current code does that.
Some the following seems sensible to me (though I haven't tested it).
Thanks,
NeilBrown
diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
index dd9f1bebb5a3..859c198d0163 100644
--- a/fs/autofs4/dev-ioctl.c
+++ b/fs/autofs4/dev-ioctl.c
@@ -208,12 +208,16 @@ static int find_autofs_mount(const char *pathname,
{
struct path path;
int err;
+ struct dentry *de, *parent;
+
+ de = kern_path_locked(pathname, &path);
+ if (IS_ERR(de))
+ return PTR_ERR(de);
+ parent = path.dentry;
+ path.dentry = de;
- err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
- if (err)
- return err;
err = -ENOENT;
- while (path.dentry == path.mnt->mnt_root) {
+ do {
if (path.dentry->d_sb->s_magic == AUTOFS_SUPER_MAGIC) {
if (test(&path, data)) {
path_get(&path);
@@ -222,10 +226,11 @@ static int find_autofs_mount(const char *pathname,
break;
}
}
- if (!follow_up(&path))
- break;
- }
+ } while (follow_down_one(&path));
+
path_put(&path);
+ inode_unlock(d_inode(parent));
+ dput(parent);
return err;
}
On Wed, Aug 23 2017, Ian Kent wrote:
>
> That inconsistency has bothered me for quite a while now.
>
> It was carried over from the autofs module behavior when automounting
> support was added to the VFS. What's worse is it prevents the use of
> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
> statx().
>
> There is some risk in changing that so it does work but it really does
> need to work to enable userspace to not trigger an automount by using
> this flag.
>
> So that's (hopefully) going to change soonish, see:
> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>
> The result should be that stat family calls don't trigger automounts except
> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>
oooh, yes. That's much better - thanks.
We should make sure that change gets into the man pages...
First however, we should probably correct the man page!
stat.2 says:
NOTES
On Linux, lstat() will generally not trigger automounter
action, whereas stat() will (but see the description of
fstatat() AT_NO_AUTOMOUNT fag, above).
which is wrong: lstat and stat treat automounts the same.
@Michael: do you recall why you inserted that text? The commit message
in commit 1ef5b2805471 ("stat.2: Cosmetic reworking of timestamp
discussion in NOTES") is not very helpful.
I propose correcting to
NOTES:
On Linux, lstat() nor stat() act as though AT_NO_AUTOMOUNT was set
and will not trigger automounter action for direct automount
points, though they may (prior to 4.14) for indirect automount
points.
The more precise details, that automount action for indirect automount
points is not triggered when the 'browse' option is used, is probably
not necessary.
Ian: if you agree with that text, and Michael doesn't provide alternate
evidence, I'll submit a formal patch for the man page.... or should we
just wait until the patch actually lands?
Thanks,
NeilBrown
On 24/08/17 11:21, NeilBrown wrote:
> On Wed, Aug 23 2017, Ian Kent wrote:
>
>> On 23/08/17 10:32, Ian Kent wrote:
>>> On 23/08/17 09:06, NeilBrown wrote:
>>>> On Mon, Aug 21 2017, Ian Kent wrote:
>>>>
>>>>>>
>>>>>> A mount isn't triggered by kern_path(pathname, 0, &path).
>>>>>> That '0' would need to include one of
>>>>>> LOOKUP_PARENT | LOOKUP_DIRECTORY |
>>>>>> LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT
>>>>>>
>>>>>> to trigger an automount (otherwise you just get -EISDIR).
>>>>>
>>>>> It's perfectly sensible to think that but there is a case where a
>>>>> a mount is triggered when using kern_path().
>>>>>
>>>>> The EISDIR return occurs for positive dentrys, negative dentrys
>>>>> will still trigger an automount (which is autofs specific,
>>>>> indirect mount map using nobrowse option, the install default).
>>>>
>>>> Ok, I understand this better now. This difference between direct and
>>>> indirect mounts is slightly awkward. It is visible from user-space, but
>>>> not elegant to document.
>>>> When you use O_PATH to open a direct automount that has not already been
>>>> triggered, the open returns the underlying directory (and fstatfs
>>>> confirms that it is AUTOFS_SUPER_MAGIC). When you use O_PATH on
>>>> an indirect automount, it *will* trigger the automount when "nobrowse" is
>>>> in effect, but it won't when "browse" is in effect.
>>>
>>> That inconsistency has bothered me for quite a while now.
>>>
>>> It was carried over from the autofs module behavior when automounting
>>> support was added to the VFS. What's worse is it prevents the use of
>>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>>> statx().
>>>
>>> There is some risk in changing that so it does work but it really does
>>> need to work to enable userspace to not trigger an automount by using
>>> this flag.
>>>
>>> So that's (hopefully) going to change soonish, see:
>>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>>
>>> The result should be that stat family calls don't trigger automounts except
>>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>>
>>>>
>>>> So we cannot just say "O_PATH doesn't trigger automounts", which is
>>>> essentially what I said in
>>>>
>>>> https://git.kernel.org/pub/scm/docs/man-pages/man-pages.git/commit/?id=97a45d02e6671482e8b2cdcce3951930bf6bdb94
>>>>
>>>> It might be possible to modify automount so that it was more consistent
>>>> - i.e. if the point is triggered by a mkdir has been done, just to the
>>>> mkdir. If it is triggered after a mkdir has been done, do the mount. I
>>>> guess that might be racy, and in any case is hard to justify.
>>>>
>>>> Maybe I should change it to be about "direct automounts", and add a note
>>>> that indirect automounts aren't so predictable.
>>>
>>> Right and the semantics should be much more consistent in the near future.
>>> I hope (and expect) this semantic change won't cause problems.
>>>
>>>>
>>>> But back to my original issue of wanting to discard
>>>> kern_path_mountpoint, what would you think of the following approach -
>>>> slight revised from before.
>>>>
>>>> Thanks,
>>>> NeilBrown
>>>>
>>>> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
>>>> index beef981aa54f..7663ea82e68d 100644
>>>> --- a/fs/autofs4/autofs_i.h
>>>> +++ b/fs/autofs4/autofs_i.h
>>>> @@ -135,10 +135,13 @@ static inline struct autofs_info *autofs4_dentry_ino(struct dentry *dentry)
>>>> /* autofs4_oz_mode(): do we see the man behind the curtain? (The
>>>> * processes which do manipulations for us in user space sees the raw
>>>> * filesystem without "magic".)
>>>> + * A process performing certain ioctls can get temporary oz status.
>>>> */
>>>> +extern struct task_struct *autofs_tmp_oz;
>>>> static inline int autofs4_oz_mode(struct autofs_sb_info *sbi)
>>>> {
>>>> - return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp;
>>>> + return sbi->catatonic || task_pgrp(current) == sbi->oz_pgrp ||
>>>> + autofs_tmp_oz == current;
>>>> }
>>>>
>>>> struct inode *autofs4_get_inode(struct super_block *, umode_t);
>>>> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
>>>> index dd9f1bebb5a3..d76401669a20 100644
>>>> --- a/fs/autofs4/dev-ioctl.c
>>>> +++ b/fs/autofs4/dev-ioctl.c
>>>> @@ -200,6 +200,20 @@ static int autofs_dev_ioctl_protosubver(struct file *fp,
>>>> return 0;
>>>> }
>>>>
>>>> +struct task_struct *autofs_tmp_oz;
>>>> +int kern_path_oz(const char *pathname, int flags, struct path *path)
>>>> +{
>>>> + static DEFINE_MUTEX(autofs_oz);
>>>> + int err;
>>>> +
>>>> + mutex_lock(&autofs_oz);
>>>> + autofs_tmp_oz = current;
>>>> + err = kern_path(pathname, flags, path);
>>>> + autofs_tmp_oz = NULL;
>>>> + mutex_unlock(&autofs_oz);
>>>> + return err;
>>>> +}
>>>> +
>>>
>>> It's simple enough but does look like it will attract criticism as being
>>> a hack!
>>>
>>> The kern_path_locked() function is very similar to what was originally
>>> done, along with code to look down the mount stack (rather than up the
>>> way it does now) to get the mount point. In this case, to be valid the
>>> dentry can't be a symlink so that fits kern_path_locked() too.
>>
>> Oh wait, that __lookup_hash() tries too hard to resolve the dentry,
>> that won't quite work, and maybe d_lookup() can't be used safely in
>> this context either ....
>>
>
> Why do you think that __look_hash() tries too hard?
> It does call into the filesystem ->lookup() if the name isn't in the
> cache, which probably isn't strictly needed, but that isn't harmful and
> the current code does that.
Only that the execution path can encounter the negative dentry case in
follow_automount().
Also, my saying that the dentry can't be a symlink is wrong.
For the covered fd case it's reasonable but for the mount point check
ioctl the path could be anything, eg. a symlink to an automount dentry,
and I want that ioctl to be callable by anyone.
>
> Some the following seems sensible to me (though I haven't tested it).
>
> Thanks,
> NeilBrown
>
>
> diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
> index dd9f1bebb5a3..859c198d0163 100644
> --- a/fs/autofs4/dev-ioctl.c
> +++ b/fs/autofs4/dev-ioctl.c
> @@ -208,12 +208,16 @@ static int find_autofs_mount(const char *pathname,
> {
> struct path path;
> int err;
> + struct dentry *de, *parent;
> +
> + de = kern_path_locked(pathname, &path);
> + if (IS_ERR(de))
> + return PTR_ERR(de);
> + parent = path.dentry;
> + path.dentry = de;
>
> - err = kern_path_mountpoint(AT_FDCWD, pathname, &path, 0);
> - if (err)
> - return err;
> err = -ENOENT;
> - while (path.dentry == path.mnt->mnt_root) {
> + do {
> if (path.dentry->d_sb->s_magic == AUTOFS_SUPER_MAGIC) {
> if (test(&path, data)) {
> path_get(&path);
> @@ -222,10 +226,11 @@ static int find_autofs_mount(const char *pathname,
> break;
> }
> }
> - if (!follow_up(&path))
> - break;
> - }
> + } while (follow_down_one(&path));
> +
> path_put(&path);
> + inode_unlock(d_inode(parent));
> + dput(parent);
> return err;
> }
>
>
On 24/08/17 12:07, NeilBrown wrote:
> On Wed, Aug 23 2017, Ian Kent wrote:
>
>>
>> That inconsistency has bothered me for quite a while now.
>>
>> It was carried over from the autofs module behavior when automounting
>> support was added to the VFS. What's worse is it prevents the use of
>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>> statx().
>>
>> There is some risk in changing that so it does work but it really does
>> need to work to enable userspace to not trigger an automount by using
>> this flag.
>>
>> So that's (hopefully) going to change soonish, see:
>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>
>> The result should be that stat family calls don't trigger automounts except
>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>
>
> oooh, yes. That's much better - thanks.
>
> We should make sure that change gets into the man pages...
Yes, I was wondering who to contact for that.
>
> First however, we should probably correct the man page!
> stat.2 says:
>
>
> NOTES
> On Linux, lstat() will generally not trigger automounter
> action, whereas stat() will (but see the description of
> fstatat() AT_NO_AUTOMOUNT fag, above).
>
> which is wrong: lstat and stat treat automounts the same.
> @Michael: do you recall why you inserted that text? The commit message
> in commit 1ef5b2805471 ("stat.2: Cosmetic reworking of timestamp
> discussion in NOTES") is not very helpful.
>
> I propose correcting to
>
> NOTES:
> On Linux, lstat() nor stat() act as though AT_NO_AUTOMOUNT was set
> and will not trigger automounter action for direct automount
> points, though they may (prior to 4.14) for indirect automount
> points.
Shouldn't that be "lstat() and stat() act as though AT_NO_AUTOMOUNT is
set ..."
>
>
> The more precise details, that automount action for indirect automount
> points is not triggered when the 'browse' option is used, is probably
> not necessary.
>
> Ian: if you agree with that text, and Michael doesn't provide alternate
> evidence, I'll submit a formal patch for the man page.... or should we
> just wait until the patch actually lands?
I thought the fstatat() description needed attention too, doubly so with
the AT_NO_AUTOMOUNT change.
The "The fstatat() system call operates in exactly the same way as stat()"
is wrong in the same way as the stat() description was wrong.
After the change fstatat() will trigger automounts if the AT_NO_AUTOMOUNT
flag is not given which is different from lstat() and stat().
The updated NOTE above probably needs to be referred to in order to clarify
what's meant by "in exactly the same way" since that probably refers to the
information returned rather than whether an automount will be done.
Ian
On 24/08/17 12:07, NeilBrown wrote:
>
>
> The more precise details, that automount action for indirect automount
> points is not triggered when the 'browse' option is used, is probably
> not necessary.
>
> Ian: if you agree with that text, and Michael doesn't provide alternate
> evidence, I'll submit a formal patch for the man page.... or should we
> just wait until the patch actually lands?
So far only David commented about using ENOENT rather than EREMOTE.
I prefer ENOENT for this case myself and he didn't object when I
explained why, David, any concerns?
Al has been silent so far so either he hasn't seen it or he's ok with
it, Al, any concerns?
And I guess if there are no concerns there's a good chance Andrew is
ok with it for the next merge window, Andrew?
If everyone agrees then we could go ahead immediately so there's a
better chance of getting it into released man pages closer to the
change being merged.
Ian
On Mon, Aug 14 2017, NeilBrown wrote:
> On Fri, Aug 11 2017, Trond Myklebust wrote:
>
>> On Fri, 2017-08-11 at 14:31 +1000, NeilBrown wrote:
>>> Funny story. 4.5 years ago we discarded the FS_REVAL_DOT superblock
>>> flag and introduced the d_weak_revalidate dentry operation instead.
>>> We duly removed the flag from NFS superblocks and NFSv4 superblocks,
>>> and added the new dentry operation to NFS dentries .... but not to
>>> NFSv4
>>> dentries.
>>>
>>> And nobody noticed.
>>>
>>> Until today.
>>>
>>> A customer reports a situation where mount(....,MS_REMOUNT,..) on an
>>> NFS
>>> filesystem hangs because the network has been deconfigured. This
>>> makes
>>> perfect sense and I suggested a code change to fix the problem.
>>> However when a colleague was trying to reproduce the problem to
>>> validate
>>> the fix, he couldn't. Then nor could I.
>>>
>>> The problem is trivially reproducible with NFSv3, and not at all with
>>> NFSv4. The reason is the missing d_weak_revalidate.
>>>
>>> We could simply add d_weak_revalidate for NFSv4, but given that it
>>> has been missing for 4.5 years, and the only time anyone noticed was
>>> when the ommission resulted in a better user experience, I do wonder
>>> if
>>> we need to. Can we just discard d_weak_revalidate? What purpose
>>> does
>>> it serve? I couldn't find one.
>>>
>>> Thanks,
>>> NeilBrown
>>>
>>> For reference, see
>>> Commit: ecf3d1f1aa74 ("vfs: kill FS_REVAL_DOT by adding a
>>> d_weak_revalidate dentry op")
>>>
>>>
>>>
>>> To reproduce the problem at home, on a system that uses systemd:
>>> 1/ place (or find) a filesystem image in a file on an NFS filesystem.
>>> 2/ mount the nfs filesystem with "noac" - choose v3 or v4
>>> 3/ loop-mount the filesystem image read-only somewhere
>>> 4/ reboot
>>>
>>> If you choose v4, the reboot will succeed, possibly after a 90second
>>> timeout.
>>> If you choose v3, the reboot will hang indefinitely in systemd-
>>> shutdown while
>>> remounting the nfs filesystem read-only.
>>>
>>> If you don't use "noac" it can still hang, but only if something
>>> slows
>>> down the reboot enough that attributes have timed out by the time
>>> that
>>> systemd-shutdown runs. This happens for our customer.
>>>
>>> If the loop-mounted filesystem is not read-only, you get other
>>> problems.
>>>
>>> We really want systemd to figure out that the loop-mount needs to be
>>> unmounted first. I have ideas concerning that, but it is messy. But
>>> that isn't the only bug here.
>>
>> The main purpose of d_weak_revalidate() was to catch the issues that
>> arise when someone changes the contents of the current working
>> directory or its parent on the server. Since '.' and '..' are treated
>> specially in the lookup code, they would not be revalidated without
>> special treatment. That leads to issues when looking up files as
>> ./<filename> or ../<filename>, since the client won't detect that its
>> dcache is stale until it tries to use the cached dentry+inode.
>
> I don't think that is quite right.
> d_weak_revalidate() is only called from complete_walk() if LOOKUP_JUMPED
> is set. The happens when the final component of a path:
> - is a mount point
> - is ".."
> or if the whole path is "/". I thought "." was treated specially too,
> but I cannot find that in the code.
Actually, you were very close to the right answer, and I was missing
something important.
The issue (or, at least "an" issue) happens when you open "." or ".." or
a mount point, or a /proc/*/fd/* symlink.
In each case LOOKUP_JUMPED is set. "." doesn't set it, but it doesn't
clear it either and it is always set at the start of a path lookup.
When you open a file (or directory) on NFS you need to validate the
attributes to ensure close-to-open consistency rules are met.
When you open any path that ends with a LAST_NORM name, d_revalidate will
be passed the LOOKUP_OPEN flag and so nfs_lookup_verify_inode() will
force a revalidate with __nfs_revalidate_inode().
When you open something that ends with LOOKUP_JUMPED, the task of
forcing the revalidate falls to d_weak_revalidate(). Unfortunately it
doesn't actually do that. With NFSv4, there is no d_weak_revalidate().
With NFSv3 there is - but it doesn't know if LOOKUP_JUMPED is set, and
doesn't force the revalidate.
This means that if you
echo *
or
echo ../*
there might be no communication with the server, and you might get stale
data.
These command *do* work as expected only when the directory being listed
is a mountpoint. This is because nfs_opendir() contains:
if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
/* This is a mountpoint, so d_revalidate will never
* have been called, so we need to refresh the
* inode (for close-open consistency) ourselves.
*/
__nfs_revalidate_inode(NFS_SERVER(inode), inode);
}
which I put there some years ago, when things worked differently.
There are various ways we could fix this.
The simplest would be to change complete_walk() to only call
d_weak_revalidate if (nd->flags & LOOKUP_OPEN), and change
d_weak_revalidate to call __nfs_revalidate_inode() unconditionally. And
to get NFSv4 to call this too.
However I would like to take a different approach. I'd like to change
nfs_lookup_revalidate to check LOOKUP_JUMPED itself, and to consider
only the inode when the flag is set.
When we can discard d_weak_revalidate() and call d_revalidate (with
LOOKUP_JUMPED set) in complete_walk(). Maybe this is too intrusive on
other filesystems that don't differentiate revalidate on open ... nfs is
the only filesystem which tests LOOKUP_OPEN in d_revalidate.
Or maybe the LAST_JUMPED flag could be passed to ->open (atomic_open
doesn't need it) - but that could get messy. It would have to go
through vfs_open
Either approach will mean that umount can go back to using
user_path_at(), as the final dentry will only be revalidated on open,
not on other accesses.
The LOOKUP_JUMPED flag and d_weak_revalidate() trace their history back
to FS_REVAL_DOT, and the issue has always been about handling open()
correctly when the path doesn't ends LAST_NORM.
NeilBrown
Hi Neil,
On 24 August 2017 at 06:07, NeilBrown <[email protected]> wrote:
> On Wed, Aug 23 2017, Ian Kent wrote:
>
>>
>> That inconsistency has bothered me for quite a while now.
>>
>> It was carried over from the autofs module behavior when automounting
>> support was added to the VFS. What's worse is it prevents the use of
>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>> statx().
>>
>> There is some risk in changing that so it does work but it really does
>> need to work to enable userspace to not trigger an automount by using
>> this flag.
>>
>> So that's (hopefully) going to change soonish, see:
>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>
>> The result should be that stat family calls don't trigger automounts except
>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>
>
> oooh, yes. That's much better - thanks.
>
> We should make sure that change gets into the man pages...
>
> First however, we should probably correct the man page!
> stat.2 says:
>
>
> NOTES
> On Linux, lstat() will generally not trigger automounter
> action, whereas stat() will (but see the description of
> fstatat() AT_NO_AUTOMOUNT fag, above).
>
> which is wrong: lstat and stat treat automounts the same.
> @Michael: do you recall why you inserted that text? The commit message
> in commit 1ef5b2805471 ("stat.2: Cosmetic reworking of timestamp
> discussion in NOTES") is not very helpful.
That commit really was just cosmetic changes. The change that
introduced the text was 82d2be3d9d66b7, based on a note from Peter
Anvin:
[[
> > Additionally, you may want to make a note in the stat/lstat man page tha
t on
> > Linux, lstat(2) will generally not trigger automounter action, whereas
> > stat(2) will.
>
> I don't understand this last piece. Can you say some more. (I'm not
> familiar with automounter details.)
An automounter (either an explicit one, like autofs, or an implicit
one, such as are used by AFS or NFSv4) is something that triggers
a mount when something is touched.
However, it's undesirable to automount, say, everyone's home
directory just because someone opened up /home in their GUI
browser or typed "ls -l /home". The early automounters simply
didn't list the contents until you accessed it by name;
this is still the case when you can't enumerate a mapping
(say, all DNS names under /net). However, this is extremely
inconvenient, too.
The solution we ended up settling on is to create something
that looks like a directory (i.e. reports S_IFDIR in stat()),
but behaves somewhat like a symlink. In particular, when it is
accessed in a way where a symlink would be dereferenced,
the automount triggers and the directory is mounted. However,
system calls which do *not* cause a symlink to be dereferenced,
like lstat(), also do not cause the automounter to trigger.
This means that "ls -l", or a GUI file browser, can see a list
of directories without causing each one of them to be automounted.
-hpa
]]
Cheers,
Michael
> I propose correcting to
>
> NOTES:
> On Linux, lstat() nor stat() act as though AT_NO_AUTOMOUNT was set
> and will not trigger automounter action for direct automount
> points, though they may (prior to 4.14) for indirect automount
> points.
>
>
> The more precise details, that automount action for indirect automount
> points is not triggered when the 'browse' option is used, is probably
> not necessary.
>
> Ian: if you agree with that text, and Michael doesn't provide alternate
> evidence, I'll submit a formal patch for the man page.... or should we
> just wait until the patch actually lands?
>
> Thanks,
> NeilBrown
>
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
On 24/08/17 19:03, Michael Kerrisk (man-pages) wrote:
> Hi Neil,
>
> On 24 August 2017 at 06:07, NeilBrown <[email protected]> wrote:
>> On Wed, Aug 23 2017, Ian Kent wrote:
>>
>>>
>>> That inconsistency has bothered me for quite a while now.
>>>
>>> It was carried over from the autofs module behavior when automounting
>>> support was added to the VFS. What's worse is it prevents the use of
>>> the AT_NO_AUTOMOUNT flag from working properly with fstatat(2) and with
>>> statx().
>>>
>>> There is some risk in changing that so it does work but it really does
>>> need to work to enable userspace to not trigger an automount by using
>>> this flag.
>>>
>>> So that's (hopefully) going to change soonish, see:
>>> http://ozlabs.org/~akpm/mmotm/broken-out/autofs-fix-at_no_automount-not-being-honored.patch
>>>
>>> The result should be that stat family calls don't trigger automounts except
>>> for fstatat(2) and statx() which will require the AT_NO_AUTOMOUNT flag.
>>>
>>
>> oooh, yes. That's much better - thanks.
>>
>> We should make sure that change gets into the man pages...
>>
>> First however, we should probably correct the man page!
>> stat.2 says:
>>
>>
>> NOTES
>> On Linux, lstat() will generally not trigger automounter
>> action, whereas stat() will (but see the description of
>> fstatat() AT_NO_AUTOMOUNT fag, above).
>>
>> which is wrong: lstat and stat treat automounts the same.
>> @Michael: do you recall why you inserted that text? The commit message
>> in commit 1ef5b2805471 ("stat.2: Cosmetic reworking of timestamp
>> discussion in NOTES") is not very helpful.
>
> That commit really was just cosmetic changes. The change that
> introduced the text was 82d2be3d9d66b7, based on a note from Peter
> Anvin:
Indeed, that was correct for autofs v3 but we're at autofs v5 now and
a lot has changed over time (the commit is from 2008).
All I can do is apologize for not also checking the man pages and trying
to keep them up to date.
Let's just work on making them accurate now.
>
> [[
> > > Additionally, you may want to make a note in the stat/lstat man page tha
> t on
> > > Linux, lstat(2) will generally not trigger automounter action, whereas
> > > stat(2) will.
> >
> > I don't understand this last piece. Can you say some more. (I'm not
> > familiar with automounter details.)
>
> An automounter (either an explicit one, like autofs, or an implicit
> one, such as are used by AFS or NFSv4) is something that triggers
> a mount when something is touched.
>
> However, it's undesirable to automount, say, everyone's home
> directory just because someone opened up /home in their GUI
> browser or typed "ls -l /home". The early automounters simply
> didn't list the contents until you accessed it by name;
> this is still the case when you can't enumerate a mapping
> (say, all DNS names under /net). However, this is extremely
> inconvenient, too.
>
> The solution we ended up settling on is to create something
> that looks like a directory (i.e. reports S_IFDIR in stat()),
> but behaves somewhat like a symlink. In particular, when it is
> accessed in a way where a symlink would be dereferenced,
> the automount triggers and the directory is mounted. However,
> system calls which do *not* cause a symlink to be dereferenced,
> like lstat(), also do not cause the automounter to trigger.
> This means that "ls -l", or a GUI file browser, can see a list
> of directories without causing each one of them to be automounted.
>
> -hpa
> ]]
>
> Cheers,
>
> Michael
>
>> I propose correcting to
>>
>> NOTES:
>> On Linux, lstat() nor stat() act as though AT_NO_AUTOMOUNT was set
>> and will not trigger automounter action for direct automount
>> points, though they may (prior to 4.14) for indirect automount
>> points.
>>
>>
>> The more precise details, that automount action for indirect automount
>> points is not triggered when the 'browse' option is used, is probably
>> not necessary.
>>
>> Ian: if you agree with that text, and Michael doesn't provide alternate
>> evidence, I'll submit a formal patch for the man page.... or should we
>> just wait until the patch actually lands?
>>
>> Thanks,
>> NeilBrown
>>
>
>
>
Expand on the relationship between fstatat() and the other
three functions, and improve the description of AT_NO_AUTOMOUNT.
Specifically, both stat() and lstat() act the same way
with respect to automounts, and that behavior matches
fstatat with the AT_NO_AUTOMOUNT flag.
The text in the NOTES is removed and places with the text for
AT_NO_AUTOMOUNT to improve cohesion.
New text for a difference to be introduced in 4.14.
Cc: Ian Kent <[email protected]>
Signed-off-by: NeilBrown <[email protected]>
---
Thanks Ian and Michael. I considered your input and read
through the whole again, and came up with this which is
quite different to what I suggested before.
If this patch is applied, the result probably shouldn't be released
until the relevant patch actually lands in Linus's tree.
NeilBrown
man2/stat.2 | 37 ++++++++++++++++++++++++-------------
1 file changed, 24 insertions(+), 13 deletions(-)
diff --git a/man2/stat.2 b/man2/stat.2
index d8a9e76b3d9f..c6dddfe0d3a7 100644
--- a/man2/stat.2
+++ b/man2/stat.2
@@ -260,9 +260,12 @@ For further information on the above fields, see
.SS fstatat()
The
.BR fstatat ()
-system call operates in exactly the same way as
+system call is a more general interface for accessing file information
+which can still provide exactly the behavior of each of
.BR stat (),
-except for the differences described here.
+.BR lstat (),
+and
+.BR fstat ().
.PP
If the pathname given in
.I pathname
@@ -272,6 +275,8 @@ referred to by the file descriptor
(rather than relative to the current working directory of
the calling process, as is done by
.BR stat ()
+and
+.BR lstat ()
for a relative pathname).
.PP
If
@@ -284,7 +289,9 @@ then
.I pathname
is interpreted relative to the current working
directory of the calling process (like
-.BR stat ()).
+.BR stat ()
+and
+.BR lstat ()).
.PP
If
.I pathname
@@ -307,7 +314,11 @@ is an empty string, operate on the file referred to by
flag).
In this case,
.I dirfd
-can refer to any type of file, not just a directory.
+can refer to any type of file, not just a directory, and
+the behavior of
+.BR fstatat ()
+is similar to that of
+.BR fstat ().
If
.I dirfd
is
@@ -324,6 +335,8 @@ Don't automount the terminal ("basename") component of
if it is a directory that is an automount point.
This allows the caller to gather attributes of an automount point
(rather than the location it would mount).
+Since Linux 4.14, also don't instantiate a non-existent name in an
+on-demand directory such as used for automounter indirect maps.
This flag can be used in tools that scan directories
to prevent mass-automounting of a directory of automount points.
The
@@ -333,6 +346,13 @@ This flag is Linux-specific; define
.B _GNU_SOURCE
.\" Before glibc 2.16, defining _ATFILE_SOURCE sufficed
to obtain its definition.
+Both
+.BR stat ()
+and
+.BR lstat ()
+act as though
+.B AT_NO_AUTOMOUNT
+was set.
.TP
.B AT_SYMLINK_NOFOLLOW
If
@@ -474,15 +494,6 @@ fields may be less portable.
The interpretation differs between systems,
and possibly on a single system when NFS mounts are involved.)
.SH NOTES
-On Linux,
-.BR lstat ()
-will generally not trigger automounter action, whereas
-.BR stat ()
-will (but see the description of the
-.BR fstatat ()
-.B AT_NO_AUTOMOUNT
-fag, above).
-.\"
.SS Timestamp fields
Older kernels and older standards did not support nanosecond timestamp
fields.
--
2.14.0.rc0.dirty
Ian Kent <[email protected]> wrote:
> So far only David commented about using ENOENT rather than EREMOTE.
>
> I prefer ENOENT for this case myself and he didn't object when I
> explained why, David, any concerns?
Not really - it just seems EREMOTE is a better fit since there is something
there, we're just not allowed to follow it. This is different to a dangling
symlink IMO.
David
Hi Neil,
On 25 August 2017 at 07:32, NeilBrown <[email protected]> wrote:
>
> Expand on the relationship between fstatat() and the other
> three functions, and improve the description of AT_NO_AUTOMOUNT.
> Specifically, both stat() and lstat() act the same way
> with respect to automounts, and that behavior matches
> fstatat with the AT_NO_AUTOMOUNT flag.
>
> The text in the NOTES is removed and places with the text for
> AT_NO_AUTOMOUNT to improve cohesion.
>
> New text for a difference to be introduced in 4.14.
Has the 4.14 piece gone in?
Cheers,
Michael
> Cc: Ian Kent <[email protected]>
> Signed-off-by: NeilBrown <[email protected]>
> ---
>
> Thanks Ian and Michael. I considered your input and read
> through the whole again, and came up with this which is
> quite different to what I suggested before.
>
> If this patch is applied, the result probably shouldn't be released
> until the relevant patch actually lands in Linus's tree.
>
> NeilBrown
>
>
> man2/stat.2 | 37 ++++++++++++++++++++++++-------------
> 1 file changed, 24 insertions(+), 13 deletions(-)
>
> diff --git a/man2/stat.2 b/man2/stat.2
> index d8a9e76b3d9f..c6dddfe0d3a7 100644
> --- a/man2/stat.2
> +++ b/man2/stat.2
> @@ -260,9 +260,12 @@ For further information on the above fields, see
> .SS fstatat()
> The
> .BR fstatat ()
> -system call operates in exactly the same way as
> +system call is a more general interface for accessing file information
> +which can still provide exactly the behavior of each of
> .BR stat (),
> -except for the differences described here.
> +.BR lstat (),
> +and
> +.BR fstat ().
> .PP
> If the pathname given in
> .I pathname
> @@ -272,6 +275,8 @@ referred to by the file descriptor
> (rather than relative to the current working directory of
> the calling process, as is done by
> .BR stat ()
> +and
> +.BR lstat ()
> for a relative pathname).
> .PP
> If
> @@ -284,7 +289,9 @@ then
> .I pathname
> is interpreted relative to the current working
> directory of the calling process (like
> -.BR stat ()).
> +.BR stat ()
> +and
> +.BR lstat ()).
> .PP
> If
> .I pathname
> @@ -307,7 +314,11 @@ is an empty string, operate on the file referred to by
> flag).
> In this case,
> .I dirfd
> -can refer to any type of file, not just a directory.
> +can refer to any type of file, not just a directory, and
> +the behavior of
> +.BR fstatat ()
> +is similar to that of
> +.BR fstat ().
> If
> .I dirfd
> is
> @@ -324,6 +335,8 @@ Don't automount the terminal ("basename") component of
> if it is a directory that is an automount point.
> This allows the caller to gather attributes of an automount point
> (rather than the location it would mount).
> +Since Linux 4.14, also don't instantiate a non-existent name in an
> +on-demand directory such as used for automounter indirect maps.
> This flag can be used in tools that scan directories
> to prevent mass-automounting of a directory of automount points.
> The
> @@ -333,6 +346,13 @@ This flag is Linux-specific; define
> .B _GNU_SOURCE
> .\" Before glibc 2.16, defining _ATFILE_SOURCE sufficed
> to obtain its definition.
> +Both
> +.BR stat ()
> +and
> +.BR lstat ()
> +act as though
> +.B AT_NO_AUTOMOUNT
> +was set.
> .TP
> .B AT_SYMLINK_NOFOLLOW
> If
> @@ -474,15 +494,6 @@ fields may be less portable.
> The interpretation differs between systems,
> and possibly on a single system when NFS mounts are involved.)
> .SH NOTES
> -On Linux,
> -.BR lstat ()
> -will generally not trigger automounter action, whereas
> -.BR stat ()
> -will (but see the description of the
> -.BR fstatat ()
> -.B AT_NO_AUTOMOUNT
> -fag, above).
> -.\"
> .SS Timestamp fields
> Older kernels and older standards did not support nanosecond timestamp
> fields.
> --
> 2.14.0.rc0.dirty
>
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/
On Thu, Sep 14 2017, Michael Kerrisk (man-pages) wrote:
> Hi Neil,
>
> On 25 August 2017 at 07:32, NeilBrown <[email protected]> wrote:
>>
>> Expand on the relationship between fstatat() and the other
>> three functions, and improve the description of AT_NO_AUTOMOUNT.
>> Specifically, both stat() and lstat() act the same way
>> with respect to automounts, and that behavior matches
>> fstatat with the AT_NO_AUTOMOUNT flag.
>>
>> The text in the NOTES is removed and places with the text for
>> AT_NO_AUTOMOUNT to improve cohesion.
>>
>> New text for a difference to be introduced in 4.14.
>
> Has the 4.14 piece gone in?
>
Yes.
Commit: 42f461482178 ("autofs: fix AT_NO_AUTOMOUNT not being honored")
Thanks,
NeilBrown
On 09/15/2017 12:25 AM, NeilBrown wrote:
> On Thu, Sep 14 2017, Michael Kerrisk (man-pages) wrote:
>
>> Hi Neil,
>>
>> On 25 August 2017 at 07:32, NeilBrown <[email protected]> wrote:
>>>
>>> Expand on the relationship between fstatat() and the other
>>> three functions, and improve the description of AT_NO_AUTOMOUNT.
>>> Specifically, both stat() and lstat() act the same way
>>> with respect to automounts, and that behavior matches
>>> fstatat with the AT_NO_AUTOMOUNT flag.
>>>
>>> The text in the NOTES is removed and places with the text for
>>> AT_NO_AUTOMOUNT to improve cohesion.
>>>
>>> New text for a difference to be introduced in 4.14.
>>
>> Has the 4.14 piece gone in?
>>
>
> Yes.
>
> Commit: 42f461482178 ("autofs: fix AT_NO_AUTOMOUNT not being honored")
Thanks. I've applied the patch.
Cheers,
Michael
--
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/