Groups > linux.kernel > #1412605 > unrolled thread

NFS/d_splice_alias breakage

Started by	Oleg Drokin <green@linuxhacker.ru>
First post	2016-06-03 01:10 +0200
Last post	2016-06-04 00:20 +0200
Articles	20 on this page of 41 — 6 participants

Back to article view | Back to linux.kernel

  NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 01:10 +0200
    [PATCH] Allow d_splice_alias to accept hashed dentries green@linuxhacker.ru - 2016-06-03 02:10 +0200
      Re: [PATCH] Allow d_splice_alias to accept hashed dentries Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 02:30 +0200
    Re: NFS/d_splice_alias breakage Trond Myklebust <trondmy@primarydata.com> - 2016-06-03 02:50 +0200
      Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 03:00 +0200
        Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 05:30 +0200
          Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 05:40 +0200
      Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 05:30 +0200
    Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 05:40 +0200
      Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 05:50 +0200
        Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 06:30 +0200
          Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 06:50 +0200
            Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 07:00 +0200
          Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 07:00 +0200
            Re: NFS/d_splice_alias breakage Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 08:00 +0200
              Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-07 01:40 +0200
                Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-10 03:40 +0200
                  Re: NFS/d_splice_alias breakage Oleg Drokin <green@linuxhacker.ru> - 2016-06-10 18:50 +0200
          Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 18:40 +0200
            Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 20:30 +0200
              Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 20:40 +0200
                Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 22:10 +0200
                  Re: Dcache oops Linus Torvalds <torvalds@linux-foundation.org> - 2016-06-03 23:20 +0200
                    Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 23:30 +0200
                      Re: Dcache oops Linus Torvalds <torvalds@linux-foundation.org> - 2016-06-04 00:10 +0200
                        Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 00:30 +0200
                        Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 00:30 +0200
                          Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 00:40 +0200
                            Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-04 00:50 +0200
                            Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-04 02:00 +0200
                              Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 03:00 +0200
                                Re: Dcache oops Jeff Layton <jlayton@poochiereds.net> - 2016-06-04 14:30 +0200
                                Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-04 18:20 +0200
                                  [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim. green@linuxhacker.ru - 2016-06-04 18:30 +0200
                                    Re: [PATCH] nfs4: Fix potential use after free of state in  nfs4_do_reclaim. Jeff Layton <jlayton@poochiereds.net> - 2016-06-04 22:00 +0200
                          Re: Dcache oops Linus Torvalds <torvalds@linux-foundation.org> - 2016-06-04 00:40 +0200
                            Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-04 00:50 +0200
                            Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 00:50 +0200
                  Re: Dcache oops Oleg Drokin <green@linuxhacker.ru> - 2016-06-03 23:20 +0200
                    Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-03 23:50 +0200
                      Re: Dcache oops Al Viro <viro@ZenIV.linux.org.uk> - 2016-06-04 00:20 +0200

Page 2 of 3 — ← Prev page 1 [2] 3 Next page →

#1413387 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-03 20:40 +0200
Subject	Re: Dcache oops
Message-ID	<rG2aK-6Nw-1@gated-at.bofh.it>
In reply to	#1413384

On Jun 3, 2016, at 2:22 PM, Al Viro wrote:

> On Fri, Jun 03, 2016 at 12:38:40PM -0400, Oleg Drokin wrote:
>> I am dropping NFS people since it seems to be converting into a generic VFS/dcache bug even though you need NFS or the like to trigger it - the lookup_open path.
> 
> NFS bug is real; there might very well be something else, but that d_drop()
> in nfs_atomic_open() needs to be restored.

Yes, that's what I mean - after the nfs fix, there seems to be another bug in
dcache that I can now trigger once nfs bug is no longer blocking the way.

>> [ 2642.364383] BUG: unable to handle kernel paging request at ffff880113f82000
>> [ 2642.365014] IP: [<ffffffff817f87d4>] bad_gs+0xd1d/0x1ba9
> 
> *ow*
> Could you dump your vmlinux (and System.map) somewhere on anonftp?
> This 'bad_gs' is there simply because it's one of the few labels in
> .fixup - to say anything useful we'll need to find out where we'd
> really come from.

I see.
vmlinux with debug symbols: http://knox.linuxhacker.ru/tmp/dcache/vmlinux.gz
System.map: http://knox.linuxhacker.ru/tmp/dcache/System.map.gz

>> Hm, somehow crashdumping support is broken for the newish kernels on my test box, I guess
>> I'll try to fix it and then re-reproduce to better understand what's going on here,
>> this trace is all I have for now in case anybody has any immediate ideas.
> 
> PS: Oleg, fix your MUA, please - long lines in mail are bloody annoying.

Huh, sorry about that. I kind of hoped the era of 80 columns text terminals was
mostly behind us. Time to dust off mutt, I guess.

[toc] | [prev] | [next] | [standalone]

#1413434 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-03 22:10 +0200
Subject	Re: Dcache oops
Message-ID	<rG3zP-7MZ-1@gated-at.bofh.it>
In reply to	#1413387

On Fri, Jun 03, 2016 at 02:35:41PM -0400, Oleg Drokin wrote:

> >> [ 2642.364383] BUG: unable to handle kernel paging request at ffff880113f82000
> >> [ 2642.365014] IP: [<ffffffff817f87d4>] bad_gs+0xd1d/0x1ba9
> > 
> > *ow*
> > Could you dump your vmlinux (and System.map) somewhere on anonftp?
> > This 'bad_gs' is there simply because it's one of the few labels in
> > .fixup - to say anything useful we'll need to find out where we'd
> > really come from.
> 
> I see.
> vmlinux with debug symbols: http://knox.linuxhacker.ru/tmp/dcache/vmlinux.gz
> System.map: http://knox.linuxhacker.ru/tmp/dcache/System.map.gz

OK...
ffffffff817f87cd:       48 8d 0a                lea    (%rdx),%rcx
ffffffff817f87d0:       48 83 e1 f8             and    $0xfffffffffffffff8,%rcx
ffffffff817f87d4:       4c 8b 01                mov    (%rcx),%r8
ffffffff817f87d7:       8d 0a                   lea    (%rdx),%ecx
ffffffff817f87d9:       83 e1 07                and    $0x7,%ecx
ffffffff817f87dc:       c1 e1 03                shl    $0x3,%ecx
ffffffff817f87df:       49 d3 e8                shr    %cl,%r8
ffffffff817f87e2:       e9 9b b3 a4 ff          jmpq   ffffffff81243b82 <__d_lookup+0x132>

Aha...  It's load_unaligned_zeropad() from dentry_string_cmp(), hitting
a genuinely unmapped address.  That sends it into fixup, where it tries to
load an aligned word containing the address in question, in hope that
fault was on attempt to cross into the next page.  No such luck, address
was aligned in the first place (it's in %rdx - 0xffff880113f82000), so
we still oops.

The unexpected part is that unmapped address did *NOT* come from a dentry;
it's .name of qstr we were looking for.  And your call chain was
__d_lookup() <- d_lookup() <- lookup_open(), so in lookup_open() it was
nd->last.name...

Can the same thing be reproduced (with NFS fix) on v4.6, ede4090, 7f427d3,
4e8440b?

[toc] | [prev] | [next] | [standalone]

#1413489 — Re: Dcache oops

From	Linus Torvalds <torvalds@linux-foundation.org>
Date	2016-06-03 23:20 +0200
Subject	Re: Dcache oops
Message-ID	<rG4Fz-8oJ-7@gated-at.bofh.it>
In reply to	#1413434

On Fri, Jun 3, 2016 at 1:07 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>
> Aha...  It's load_unaligned_zeropad() from dentry_string_cmp(), hitting
> a genuinely unmapped address.  That sends it into fixup, where it tries to
> load an aligned word containing the address in question, in hope that
> fault was on attempt to cross into the next page.  No such luck, address
> was aligned in the first place (it's in %rdx - 0xffff880113f82000), so
> we still oops.

Hmm. We do end up comparing the string length racily with the name
address, so I suspect there could be a real bug in dentry_cmp: the
fact that the name pointer and length aren't necessarily atomic wrt
each other means that we could overrun the dentry pointer to the next
page, even though it's aligned.

So maybe we need to do a careful load of the aligned dentry string
value too, not just the possibly unaligned qstr name. It's a *very*
unlikely race to hit, though. You'd have to shrink the name in the
dentry, get the old (longer name length) to match the one you look up,
and when have the memory unmapped at the end of the new (short)
length.

However, this is not that theoretical race case for two reasons:

 (1) this happens in __d_lookup(), which has taken the dentry lock. So
it's not the racy unlocked RCU case to begin with.

 (2) as you point out, since it is the load_unaligned_zeropad() case,
it isn't the possibly racy dentry name at all, but trhe qstr we're
comparing against (which may have the unaligned case, but not the
confusion about length).

> The unexpected part is that unmapped address did *NOT* come from a dentry;
> it's .name of qstr we were looking for.  And your call chain was
> __d_lookup() <- d_lookup() <- lookup_open(), so in lookup_open() it was
> nd->last.name...

That should have no such issues. It should be a normal qstr, and the
length should be reliable.

So something must have corrupted the qstr.

The remaining length *should* in %edi, judging by the

   0xffffffff81243b82 <+306>: cmp    $0x7,%edi

in the __d_lookup() disassembly. And %rdi contains 2, so there were
supposed to be two more characters at 'ct' (which is %rdx).

Why would nd->last.name be bogus? I don't see anything.

           Linus

[toc] | [prev] | [next] | [standalone]

#1413496 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-03 23:30 +0200
Subject	Re: Dcache oops
Message-ID	<rG4Pf-8ss-11@gated-at.bofh.it>
In reply to	#1413489

On Fri, Jun 03, 2016 at 02:18:15PM -0700, Linus Torvalds wrote:

> So something must have corrupted the qstr.
> 
> The remaining length *should* in %edi, judging by the
> 
>    0xffffffff81243b82 <+306>: cmp    $0x7,%edi
> 
> in the __d_lookup() disassembly. And %rdi contains 2, so there were
> supposed to be two more characters at 'ct' (which is %rdx).

... and since r8 and rsi are 0, we couldn't have consumed anything.
> 
> Why would nd->last.name be bogus? I don't see anything.

An interesting part is that it's page-aligned.  Which is impossible for
a short name obtained by getname(), but is quite likely for a symlink body.
So at a guess, we have a page containing a symlink body freed under us.

[toc] | [prev] | [next] | [standalone]

#1413522 — Re: Dcache oops

From	Linus Torvalds <torvalds@linux-foundation.org>
Date	2016-06-04 00:10 +0200
Subject	Re: Dcache oops
Message-ID	<rG5rX-uq-3@gated-at.bofh.it>
In reply to	#1413496

On Fri, Jun 3, 2016 at 2:26 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>>
>> in the __d_lookup() disassembly. And %rdi contains 2, so there were
>> supposed to be two more characters at 'ct' (which is %rdx).
>
> ... and since r8 and rsi are 0, we couldn't have consumed anything.

Right you are. So it really started out page-aligned.

>> Why would nd->last.name be bogus? I don't see anything.
>
> An interesting part is that it's page-aligned.  Which is impossible for
> a short name obtained by getname(), but is quite likely for a symlink body.
> So at a guess, we have a page containing a symlink body freed under us.

Hmm. Good point.

Is perhaps the "delayed_call" logic broken, and the symlink is free'd too early?

That whole set_delayed_call/do_delayed_call thing came in 4.5. Maybe
something broke that logic, and we've executed the delayed freeing
before we should have.

Normally it's done at terminate_walk() time. But I note that in
walk_component(), we do put_link(nd) which does a do_delayed_call(),
but does *not* do a clear_delayed_call(), so now I think a subsequent
terminate_walk() might drop it *again*.

I'm probably missing something, but I have to say that the old
explicit "put_link()" callback logic was more obvious than the new
delayed calls are.

          Linus

[toc] | [prev] | [next] | [standalone]

#1413539 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-04 00:30 +0200
Subject	Re: Dcache oops
Message-ID	<rG5Lk-AU-41@gated-at.bofh.it>
In reply to	#1413522

On Fri, Jun 03, 2016 at 11:23:55PM +0100, Al Viro wrote:

> It's not that.  It's explicit put_link() in do_last(), followed by
> ESTALEOPEN and subsequent misbegotten "retry the last step on ESTALEOPEN"
> looking at now-freed nd->last.name.  IOW, the bug predates delayed_call
> stuff.

FWIW, I'd stepped into that in d63ff28f "namei: make should_follow_link() store
the link in nd->link", so it's 4.1+ mess.  delayed_call stuff is 4.4+...

[toc] | [prev] | [next] | [standalone]

#1413543 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-04 00:30 +0200
Subject	Re: Dcache oops
Message-ID	<rG5Lk-AU-43@gated-at.bofh.it>
In reply to	#1413522

On Fri, Jun 03, 2016 at 03:00:02PM -0700, Linus Torvalds wrote:

> Is perhaps the "delayed_call" logic broken, and the symlink is free'd too early?
> 
> That whole set_delayed_call/do_delayed_call thing came in 4.5. Maybe
> something broke that logic, and we've executed the delayed freeing
> before we should have.
> 
> Normally it's done at terminate_walk() time. But I note that in
> walk_component(), we do put_link(nd) which does a do_delayed_call(),
> but does *not* do a clear_delayed_call(), so now I think a subsequent
> terminate_walk() might drop it *again*.

Nope - put_link() also decrements nd->depth.  No double calls there...

> I'm probably missing something, but I have to say that the old
> explicit "put_link()" callback logic was more obvious than the new
> delayed calls are.

It's not that.  It's explicit put_link() in do_last(), followed by
ESTALEOPEN and subsequent misbegotten "retry the last step on ESTALEOPEN"
looking at now-freed nd->last.name.  IOW, the bug predates delayed_call
stuff.

[toc] | [prev] | [next] | [standalone]

#1413551 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-04 00:40 +0200
Subject	Re: Dcache oops
Message-ID	<rG5UZ-Ei-11@gated-at.bofh.it>
In reply to	#1413543

On Fri, Jun 03, 2016 at 11:23:55PM +0100, Al Viro wrote:

> It's not that.  It's explicit put_link() in do_last(), followed by
> ESTALEOPEN and subsequent misbegotten "retry the last step on ESTALEOPEN"
> looking at now-freed nd->last.name.  IOW, the bug predates delayed_call
> stuff.

EOPENSTALE, that is...  Oleg, could you check if the following works?

diff --git a/fs/namei.c b/fs/namei.c
index 4c4f95a..3d9511e 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -3166,9 +3166,7 @@ static int do_last(struct nameidata *nd,
 	int acc_mode = op->acc_mode;
 	unsigned seq;
 	struct inode *inode;
-	struct path save_parent = { .dentry = NULL, .mnt = NULL };
 	struct path path;
-	bool retried = false;
 	int error;
 
 	nd->flags &= ~LOOKUP_PARENT;
@@ -3211,7 +3209,6 @@ static int do_last(struct nameidata *nd,
 			return -EISDIR;
 	}
 
-retry_lookup:
 	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
 		error = mnt_want_write(nd->path.mnt);
 		if (!error)
@@ -3292,23 +3289,14 @@ finish_lookup:
 	if (unlikely(error))
 		return error;
 
-	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
-		path_to_nameidata(&path, nd);
-	} else {
-		save_parent.dentry = nd->path.dentry;
-		save_parent.mnt = mntget(path.mnt);
-		nd->path.dentry = path.dentry;
-
-	}
+	path_to_nameidata(&path, nd);
 	nd->inode = inode;
 	nd->seq = seq;
 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
 finish_open:
 	error = complete_walk(nd);
-	if (error) {
-		path_put(&save_parent);
+	if (error)
 		return error;
-	}
 	audit_inode(nd->name, nd->path.dentry, 0);
 	error = -EISDIR;
 	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
@@ -3331,13 +3319,9 @@ finish_open_created:
 		goto out;
 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 	error = vfs_open(&nd->path, file, current_cred());
-	if (!error) {
-		*opened |= FILE_OPENED;
-	} else {
-		if (error == -EOPENSTALE)
-			goto stale_open;
+	if (error)
 		goto out;
-	}
+	*opened |= FILE_OPENED;
 opened:
 	error = open_check_o_direct(file);
 	if (!error)
@@ -3353,26 +3337,7 @@ out:
 	}
 	if (got_write)
 		mnt_drop_write(nd->path.mnt);
-	path_put(&save_parent);
 	return error;
-
-stale_open:
-	/* If no saved parent or already retried then can't retry */
-	if (!save_parent.dentry || retried)
-		goto out;
-
-	BUG_ON(save_parent.dentry != dir);
-	path_put(&nd->path);
-	nd->path = save_parent;
-	nd->inode = dir->d_inode;
-	save_parent.mnt = NULL;
-	save_parent.dentry = NULL;
-	if (got_write) {
-		mnt_drop_write(nd->path.mnt);
-		got_write = false;
-	}
-	retried = true;
-	goto retry_lookup;
 }
 
 static int do_tmpfile(struct nameidata *nd, unsigned flags,

[toc] | [prev] | [next] | [standalone]

#1413553 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-04 00:50 +0200
Subject	Re: Dcache oops
Message-ID	<rG64F-Hz-3@gated-at.bofh.it>
In reply to	#1413551

On Jun 3, 2016, at 6:37 PM, Al Viro wrote:

> On Fri, Jun 03, 2016 at 11:23:55PM +0100, Al Viro wrote:
> 
>> It's not that.  It's explicit put_link() in do_last(), followed by
>> ESTALEOPEN and subsequent misbegotten "retry the last step on ESTALEOPEN"
>> looking at now-freed nd->last.name.  IOW, the bug predates delayed_call
>> stuff.
> 
> EOPENSTALE, that is...  Oleg, could you check if the following works?

Ok, I'll try this one too.
The other one with goto stale_open also was appearing to work.

We'll see what else is going to show up next...

> diff --git a/fs/namei.c b/fs/namei.c
> index 4c4f95a..3d9511e 100644
> --- a/fs/namei.c
> +++ b/fs/namei.c
> @@ -3166,9 +3166,7 @@ static int do_last(struct nameidata *nd,
> 	int acc_mode = op->acc_mode;
> 	unsigned seq;
> 	struct inode *inode;
> -	struct path save_parent = { .dentry = NULL, .mnt = NULL };
> 	struct path path;
> -	bool retried = false;
> 	int error;
> 
> 	nd->flags &= ~LOOKUP_PARENT;
> @@ -3211,7 +3209,6 @@ static int do_last(struct nameidata *nd,
> 			return -EISDIR;
> 	}
> 
> -retry_lookup:
> 	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
> 		error = mnt_want_write(nd->path.mnt);
> 		if (!error)
> @@ -3292,23 +3289,14 @@ finish_lookup:
> 	if (unlikely(error))
> 		return error;
> 
> -	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
> -		path_to_nameidata(&path, nd);
> -	} else {
> -		save_parent.dentry = nd->path.dentry;
> -		save_parent.mnt = mntget(path.mnt);
> -		nd->path.dentry = path.dentry;
> -
> -	}
> +	path_to_nameidata(&path, nd);
> 	nd->inode = inode;
> 	nd->seq = seq;
> 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
> finish_open:
> 	error = complete_walk(nd);
> -	if (error) {
> -		path_put(&save_parent);
> +	if (error)
> 		return error;
> -	}
> 	audit_inode(nd->name, nd->path.dentry, 0);
> 	error = -EISDIR;
> 	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
> @@ -3331,13 +3319,9 @@ finish_open_created:
> 		goto out;
> 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
> 	error = vfs_open(&nd->path, file, current_cred());
> -	if (!error) {
> -		*opened |= FILE_OPENED;
> -	} else {
> -		if (error == -EOPENSTALE)
> -			goto stale_open;
> +	if (error)
> 		goto out;
> -	}
> +	*opened |= FILE_OPENED;
> opened:
> 	error = open_check_o_direct(file);
> 	if (!error)
> @@ -3353,26 +3337,7 @@ out:
> 	}
> 	if (got_write)
> 		mnt_drop_write(nd->path.mnt);
> -	path_put(&save_parent);
> 	return error;
> -
> -stale_open:
> -	/* If no saved parent or already retried then can't retry */
> -	if (!save_parent.dentry || retried)
> -		goto out;
> -
> -	BUG_ON(save_parent.dentry != dir);
> -	path_put(&nd->path);
> -	nd->path = save_parent;
> -	nd->inode = dir->d_inode;
> -	save_parent.mnt = NULL;
> -	save_parent.dentry = NULL;
> -	if (got_write) {
> -		mnt_drop_write(nd->path.mnt);
> -		got_write = false;
> -	}
> -	retried = true;
> -	goto retry_lookup;
> }
> 
> static int do_tmpfile(struct nameidata *nd, unsigned flags,

[toc] | [prev] | [next] | [standalone]

#1413593 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-04 02:00 +0200
Subject	Re: Dcache oops
Message-ID	<rG7ap-1iQ-7@gated-at.bofh.it>
In reply to	#1413551

On Jun 3, 2016, at 6:37 PM, Al Viro wrote:

> On Fri, Jun 03, 2016 at 11:23:55PM +0100, Al Viro wrote:
> 
>> It's not that.  It's explicit put_link() in do_last(), followed by
>> ESTALEOPEN and subsequent misbegotten "retry the last step on ESTALEOPEN"
>> looking at now-freed nd->last.name.  IOW, the bug predates delayed_call
>> stuff.
> 
> EOPENSTALE, that is...  Oleg, could you check if the following works?

Yes, this one lasted for an hour with no crashing, so it must be good.
Thanks.
(note, I am not equipped to verify correctness of NFS operations, though).

> diff --git a/fs/namei.c b/fs/namei.c
> index 4c4f95a..3d9511e 100644
> --- a/fs/namei.c
> +++ b/fs/namei.c
> @@ -3166,9 +3166,7 @@ static int do_last(struct nameidata *nd,
> 	int acc_mode = op->acc_mode;
> 	unsigned seq;
> 	struct inode *inode;
> -	struct path save_parent = { .dentry = NULL, .mnt = NULL };
> 	struct path path;
> -	bool retried = false;
> 	int error;
> 
> 	nd->flags &= ~LOOKUP_PARENT;
> @@ -3211,7 +3209,6 @@ static int do_last(struct nameidata *nd,
> 			return -EISDIR;
> 	}
> 
> -retry_lookup:
> 	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
> 		error = mnt_want_write(nd->path.mnt);
> 		if (!error)
> @@ -3292,23 +3289,14 @@ finish_lookup:
> 	if (unlikely(error))
> 		return error;
> 
> -	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
> -		path_to_nameidata(&path, nd);
> -	} else {
> -		save_parent.dentry = nd->path.dentry;
> -		save_parent.mnt = mntget(path.mnt);
> -		nd->path.dentry = path.dentry;
> -
> -	}
> +	path_to_nameidata(&path, nd);
> 	nd->inode = inode;
> 	nd->seq = seq;
> 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
> finish_open:
> 	error = complete_walk(nd);
> -	if (error) {
> -		path_put(&save_parent);
> +	if (error)
> 		return error;
> -	}
> 	audit_inode(nd->name, nd->path.dentry, 0);
> 	error = -EISDIR;
> 	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
> @@ -3331,13 +3319,9 @@ finish_open_created:
> 		goto out;
> 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
> 	error = vfs_open(&nd->path, file, current_cred());
> -	if (!error) {
> -		*opened |= FILE_OPENED;
> -	} else {
> -		if (error == -EOPENSTALE)
> -			goto stale_open;
> +	if (error)
> 		goto out;
> -	}
> +	*opened |= FILE_OPENED;
> opened:
> 	error = open_check_o_direct(file);
> 	if (!error)
> @@ -3353,26 +3337,7 @@ out:
> 	}
> 	if (got_write)
> 		mnt_drop_write(nd->path.mnt);
> -	path_put(&save_parent);
> 	return error;
> -
> -stale_open:
> -	/* If no saved parent or already retried then can't retry */
> -	if (!save_parent.dentry || retried)
> -		goto out;
> -
> -	BUG_ON(save_parent.dentry != dir);
> -	path_put(&nd->path);
> -	nd->path = save_parent;
> -	nd->inode = dir->d_inode;
> -	save_parent.mnt = NULL;
> -	save_parent.dentry = NULL;
> -	if (got_write) {
> -		mnt_drop_write(nd->path.mnt);
> -		got_write = false;
> -	}
> -	retried = true;
> -	goto retry_lookup;
> }
> 
> static int do_tmpfile(struct nameidata *nd, unsigned flags,

[toc] | [prev] | [next] | [standalone]

#1413617 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-04 03:00 +0200
Subject	Re: Dcache oops
Message-ID	<rG86t-1Sw-7@gated-at.bofh.it>
In reply to	#1413593

On Fri, Jun 03, 2016 at 07:58:37PM -0400, Oleg Drokin wrote:

> > EOPENSTALE, that is...  Oleg, could you check if the following works?
> 
> Yes, this one lasted for an hour with no crashing, so it must be good.
> Thanks.
> (note, I am not equipped to verify correctness of NFS operations, though).

I suspect that Jeff Layton might have relevant regression tests.  Incidentally,
we really need a consolidated regression testsuite, including the tests you'd
been running.  Right now there's some stuff in xfstests, LTP and cthon; if
anything, this mess shows just why we need all of that and then some in
a single place.  Lustre stuff has caught a 3 years old NFS bug (missing
d_drop() in nfs_atomic_open()) and a year-old bug in handling of EOPENSTALE
retries on the last component of a trailing non-embedded symlink.  Neither
is hard to trigger; it's just that relevant tests hadn't been run on NFS,
period.

Jeff, could you verify that the following does not cause regressions in
stale fhandles treatment?  I want to rip the damn retry logics out of
do_last() and if the staleness had only been discovered inside of
nfs4_file_open() just have the upper-level logics handle it by doing
a normal LOOKUP_REVAL pass from scratch.  To hell with trying to be clever;
a few roundtrips it saves us in some cases is not worth the complexity and
potential for bugs.  I'm fairly sure that the time spent debugging this
particular turd exceeds the total amount of time it has ever saved,
and do_last() is in dire need of simplification.  All talk about "enough eyes"
isn't worth much when the readers of code in question feel like ripping their
eyes out...

diff --git a/fs/namei.c b/fs/namei.c
index 4c4f95a..3d9511e 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -3166,9 +3166,7 @@ static int do_last(struct nameidata *nd,
 	int acc_mode = op->acc_mode;
 	unsigned seq;
 	struct inode *inode;
-	struct path save_parent = { .dentry = NULL, .mnt = NULL };
 	struct path path;
-	bool retried = false;
 	int error;

 	nd->flags &= ~LOOKUP_PARENT;
@@ -3211,7 +3209,6 @@ static int do_last(struct nameidata *nd,
 			return -EISDIR;
 	}

-retry_lookup:
 	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
 		error = mnt_want_write(nd->path.mnt);
 		if (!error)
@@ -3292,23 +3289,14 @@ finish_lookup:
 	if (unlikely(error))
 		return error;

-	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
-		path_to_nameidata(&path, nd);
-	} else {
-		save_parent.dentry = nd->path.dentry;
-		save_parent.mnt = mntget(path.mnt);
-		nd->path.dentry = path.dentry;
-
-	}
+	path_to_nameidata(&path, nd);
 	nd->inode = inode;
 	nd->seq = seq;
 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
 finish_open:
 	error = complete_walk(nd);
-	if (error) {
-		path_put(&save_parent);
+	if (error)
 		return error;
-	}
 	audit_inode(nd->name, nd->path.dentry, 0);
 	error = -EISDIR;
 	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
@@ -3331,13 +3319,9 @@ finish_open_created:
 		goto out;
 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
 	error = vfs_open(&nd->path, file, current_cred());
-	if (!error) {
-		*opened |= FILE_OPENED;
-	} else {
-		if (error == -EOPENSTALE)
-			goto stale_open;
+	if (error)
 		goto out;
-	}
+	*opened |= FILE_OPENED;
 opened:
 	error = open_check_o_direct(file);
 	if (!error)
@@ -3353,26 +3337,7 @@ out:
 	}
 	if (got_write)
 		mnt_drop_write(nd->path.mnt);
-	path_put(&save_parent);
 	return error;
-
-stale_open:
-	/* If no saved parent or already retried then can't retry */
-	if (!save_parent.dentry || retried)
-		goto out;
-
-	BUG_ON(save_parent.dentry != dir);
-	path_put(&nd->path);
-	nd->path = save_parent;
-	nd->inode = dir->d_inode;
-	save_parent.mnt = NULL;
-	save_parent.dentry = NULL;
-	if (got_write) {
-		mnt_drop_write(nd->path.mnt);
-		got_write = false;
-	}
-	retried = true;
-	goto retry_lookup;
 }

 static int do_tmpfile(struct nameidata *nd, unsigned flags,

[toc] | [prev] | [next] | [standalone]

#1413745 — Re: Dcache oops

From	Jeff Layton <jlayton@poochiereds.net>
Date	2016-06-04 14:30 +0200
Subject	Re: Dcache oops
Message-ID	<rGiSe-A9-5@gated-at.bofh.it>
In reply to	#1413617

On Sat, 2016-06-04 at 01:56 +0100, Al Viro wrote:
> On Fri, Jun 03, 2016 at 07:58:37PM -0400, Oleg Drokin wrote:
> 
> > 
> > > 
> > > EOPENSTALE, that is...  Oleg, could you check if the following works?
> > Yes, this one lasted for an hour with no crashing, so it must be good.
> > Thanks.
> > (note, I am not equipped to verify correctness of NFS operations, though).
> I suspect that Jeff Layton might have relevant regression tests.  Incidentally,
> we really need a consolidated regression testsuite, including the tests you'd
> been running.  Right now there's some stuff in xfstests, LTP and cthon; if
> anything, this mess shows just why we need all of that and then some in
> a single place.  Lustre stuff has caught a 3 years old NFS bug (missing
> d_drop() in nfs_atomic_open()) and a year-old bug in handling of EOPENSTALE
> retries on the last component of a trailing non-embedded symlink.  Neither
> is hard to trigger; it's just that relevant tests hadn't been run on NFS,
> period.
> 
> Jeff, could you verify that the following does not cause regressions in
> stale fhandles treatment?  I want to rip the damn retry logics out of
> do_last() and if the staleness had only been discovered inside of
> nfs4_file_open() just have the upper-level logics handle it by doing
> a normal LOOKUP_REVAL pass from scratch.  To hell with trying to be clever;
> a few roundtrips it saves us in some cases is not worth the complexity and
> potential for bugs.  I'm fairly sure that the time spent debugging this
> particular turd exceeds the total amount of time it has ever saved,
> and do_last() is in dire need of simplification.  All talk about "enough eyes"
> isn't worth much when the readers of code in question feel like ripping their
> eyes out...
> 

Agreed. I see no need to optimize an error case here. Any performance
hit that we'd get here is almost certainly acceptable in this
situation. The main thing is that we prevent the ESTALE from bubbling
up into userland if we can avoid it by retrying.

No, I didn't have the test for this anymore unfortunately. RHQA might
have one though.

Either way, I cooked one up that does this on the server:

#!/bin/bash

while true; do
	rm -rf foo
	mkdir foo
	echo foo > foo/bar
	usleep 100000
done

...and then this on the client after mounting the fs with
lookupcache=none and noac.

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <stdio.h>
#include <unistd.h>

int main(int argc, char **argv)
{
	int fd;

	while(1) {
		fd = open(argv[1], O_RDONLY);
		if (fd < 0) {
			if (errno == ESTALE) {
				printf("ESTALE");
				return 1;
			}
			continue;
		}
		close(fd);
	}
	return 0;
}

I did see some of the OPEN compounds come back with NFS4ERR_STALE on
the PUTFH op but no corresponding ESTALE error in userland. So, this
patch does seem to do the right thing.

Reviewed-and-Tested-by: Jeff Layton <jlayton@poochiereds.net>

[toc] | [prev] | [next] | [standalone]

#1413788 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-04 18:20 +0200
Subject	Re: Dcache oops
Message-ID	<rGmsN-3kP-9@gated-at.bofh.it>
In reply to	#1413617

On Jun 3, 2016, at 8:56 PM, Al Viro wrote:

> On Fri, Jun 03, 2016 at 07:58:37PM -0400, Oleg Drokin wrote:
> 
>>> EOPENSTALE, that is...  Oleg, could you check if the following works?
>> 
>> Yes, this one lasted for an hour with no crashing, so it must be good.
>> Thanks.
>> (note, I am not equipped to verify correctness of NFS operations, though).
> 
> I suspect that Jeff Layton might have relevant regression tests.  Incidentally,
> we really need a consolidated regression testsuite, including the tests you'd
> been running.  Right now there's some stuff in xfstests, LTP and cthon; if
> anything, this mess shows just why we need all of that and then some in
> a single place.  Lustre stuff has caught a 3 years old NFS bug (missing
> d_drop() in nfs_atomic_open()) and a year-old bug in handling of EOPENSTALE
> retries on the last component of a trailing non-embedded symlink.  Neither
> is hard to trigger; it's just that relevant tests hadn't been run on NFS,
> period.

BTW, the nets also have brought in another use after free in nfs4 state
tracking code (this is the one I was trying to hunt down from the start).
I'll submit a patch shortly.
And also there's a mysterious ext4 data corruption that I do not really fully
understand but only hit once so far.

[toc] | [prev] | [next] | [standalone]

#1413791 — [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.

From	green@linuxhacker.ru
Date	2016-06-04 18:30 +0200
Subject	[PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.
Message-ID	<rGmCt-3nX-11@gated-at.bofh.it>
In reply to	#1413788

From: Oleg Drokin <green@linuxhacker.ru>

Commit e8d975e73e5f ("fixing infinite OPEN loop in 4.0 stateid recovery")
introduced access to state after it was just potentially freed by
nfs4_put_open_state leading to a random data corruption somewhere.

BUG: unable to handle kernel paging request at ffff88004941ee40
IP: [<ffffffff813baf01>] nfs4_do_reclaim+0x461/0x740
PGD 3501067 PUD 3504067 PMD 6ff37067 PTE 800000004941e060
Oops: 0002 [#1] SMP DEBUG_PAGEALLOC
Modules linked in: loop rpcsec_gss_krb5 acpi_cpufreq tpm_tis joydev i2c_piix4 pcspkr tpm virtio_console nfsd ttm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops floppy serio_raw virtio_blk drm
CPU: 6 PID: 2161 Comm: 192.168.10.253- Not tainted 4.7.0-rc1-vm-nfs+ #112
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
task: ffff8800463dcd00 ti: ffff88003ff48000 task.ti: ffff88003ff48000
RIP: 0010:[<ffffffff813baf01>]  [<ffffffff813baf01>] nfs4_do_reclaim+0x461/0x740
RSP: 0018:ffff88003ff4bd68  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffffffff81a49900 RCX: 00000000000000e8
RDX: 00000000000000e8 RSI: ffff8800418b9930 RDI: ffff880040c96c88
RBP: ffff88003ff4bdf8 R08: 0000000000000001 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: ffff880040c96c98
R13: ffff88004941ee20 R14: ffff88004941ee40 R15: ffff88004941ee00
FS:  0000000000000000(0000) GS:ffff88006d000000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffff88004941ee40 CR3: 0000000060b0b000 CR4: 00000000000006e0
Stack:
 ffffffff813baad5 ffff8800463dcd00 ffff880000000001 ffffffff810e6b68
 ffff880043ddbc88 ffff8800418b9800 ffff8800418b98c8 ffff88004941ee48
 ffff880040c96c90 ffff880040c96c00 ffff880040c96c20 ffff880040c96c40
Call Trace:
 [<ffffffff813baad5>] ? nfs4_do_reclaim+0x35/0x740
 [<ffffffff810e6b68>] ? trace_hardirqs_on_caller+0x128/0x1b0
 [<ffffffff813bb7cd>] nfs4_run_state_manager+0x5ed/0xa40
 [<ffffffff813bb1e0>] ? nfs4_do_reclaim+0x740/0x740
 [<ffffffff813bb1e0>] ? nfs4_do_reclaim+0x740/0x740
 [<ffffffff810af0d1>] kthread+0x101/0x120
 [<ffffffff810e6b68>] ? trace_hardirqs_on_caller+0x128/0x1b0
 [<ffffffff818843af>] ret_from_fork+0x1f/0x40
 [<ffffffff810aefd0>] ? kthread_create_on_node+0x250/0x250
Code: 65 80 4c 8b b5 78 ff ff ff e8 fc 88 4c 00 48 8b 7d 88 e8 13 67 d2 ff 49 8b 47 40 a8 02 0f 84 d3 01 00 00 4c 89 ff e8 7f f9 ff ff <f0> 41 80 26 7f 48 8b 7d c8 e8 b1 84 4c 00 e9 39 fd ff ff 3d e6
RIP  [<ffffffff813baf01>] nfs4_do_reclaim+0x461/0x740
 RSP <ffff88003ff4bd68>
CR2: ffff88004941ee40

Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
---
 fs/nfs/nfs4state.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 9679f47..834b875 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -1488,9 +1488,9 @@ restart:
 					}
 					spin_unlock(&state->state_lock);
 				}
-				nfs4_put_open_state(state);
 				clear_bit(NFS_STATE_RECLAIM_NOGRACE,
 					&state->flags);
+				nfs4_put_open_state(state);
 				spin_lock(&sp->so_lock);
 				goto restart;
 			}
-- 
2.5.5

[toc] | [prev] | [next] | [standalone]

#1413855 — Re: [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.

From	Jeff Layton <jlayton@poochiereds.net>
Date	2016-06-04 22:00 +0200
Subject	Re: [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.
Message-ID	<rGpTH-5eI-21@gated-at.bofh.it>
In reply to	#1413791

On Sat, 2016-06-04 at 12:21 -0400, green@linuxhacker.ru wrote:
> From: Oleg Drokin <green@linuxhacker.ru>
> 
> Commit e8d975e73e5f ("fixing infinite OPEN loop in 4.0 stateid recovery")
> introduced access to state after it was just potentially freed by
> nfs4_put_open_state leading to a random data corruption somewhere.
> 
> BUG: unable to handle kernel paging request at ffff88004941ee40
> IP: [] nfs4_do_reclaim+0x461/0x740
> PGD 3501067 PUD 3504067 PMD 6ff37067 PTE 800000004941e060
> Oops: 0002 [#1] SMP DEBUG_PAGEALLOC
> Modules linked in: loop rpcsec_gss_krb5 acpi_cpufreq tpm_tis joydev i2c_piix4 pcspkr tpm virtio_console nfsd ttm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops floppy serio_raw virtio_blk drm
> CPU: 6 PID: 2161 Comm: 192.168.10.253- Not tainted 4.7.0-rc1-vm-nfs+ #112
> Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
> task: ffff8800463dcd00 ti: ffff88003ff48000 task.ti: ffff88003ff48000
> RIP: 0010:[]  [] nfs4_do_reclaim+0x461/0x740
> RSP: 0018:ffff88003ff4bd68  EFLAGS: 00010246
> RAX: 0000000000000000 RBX: ffffffff81a49900 RCX: 00000000000000e8
> RDX: 00000000000000e8 RSI: ffff8800418b9930 RDI: ffff880040c96c88
> RBP: ffff88003ff4bdf8 R08: 0000000000000001 R09: 0000000000000000
> R10: 0000000000000000 R11: 0000000000000000 R12: ffff880040c96c98
> R13: ffff88004941ee20 R14: ffff88004941ee40 R15: ffff88004941ee00
> FS:  0000000000000000(0000) GS:ffff88006d000000(0000) knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: ffff88004941ee40 CR3: 0000000060b0b000 CR4: 00000000000006e0
> Stack:
>  ffffffff813baad5 ffff8800463dcd00 ffff880000000001 ffffffff810e6b68
>  ffff880043ddbc88 ffff8800418b9800 ffff8800418b98c8 ffff88004941ee48
>  ffff880040c96c90 ffff880040c96c00 ffff880040c96c20 ffff880040c96c40
> Call Trace:
>  [] ? nfs4_do_reclaim+0x35/0x740
>  [] ? trace_hardirqs_on_caller+0x128/0x1b0
>  [] nfs4_run_state_manager+0x5ed/0xa40
>  [] ? nfs4_do_reclaim+0x740/0x740
>  [] ? nfs4_do_reclaim+0x740/0x740
>  [] kthread+0x101/0x120
>  [] ? trace_hardirqs_on_caller+0x128/0x1b0
>  [] ret_from_fork+0x1f/0x40
>  [] ? kthread_create_on_node+0x250/0x250
> Code: 65 80 4c 8b b5 78 ff ff ff e8 fc 88 4c 00 48 8b 7d 88 e8 13 67 d2 ff 49 8b 47 40 a8 02 0f 84 d3 01 00 00 4c 89 ff e8 7f f9 ff ff  41 80 26 7f 48 8b 7d c8 e8 b1 84 4c 00 e9 39 fd ff ff 3d e6
> RIP  [] nfs4_do_reclaim+0x461/0x740
>  RSP 
> CR2: ffff88004941ee40
> 
> Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
> ---
>  fs/nfs/nfs4state.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
> index 9679f47..834b875 100644
> --- a/fs/nfs/nfs4state.c
> +++ b/fs/nfs/nfs4state.c
> @@ -1488,9 +1488,9 @@ restart:
>  					}
>  					spin_unlock(&state->state_lock);
>  				}
> -				nfs4_put_open_state(state);
>  				clear_bit(NFS_STATE_RECLAIM_NOGRACE,
>  					&state->flags);
> +				nfs4_put_open_state(state);
>  				spin_lock(&sp->so_lock);
>  				goto restart;
>  			}

Nice catch.

Reviewed-by: Jeff Layton <jlayton@poochiereds.net>

[toc] | [prev] | [next] | [standalone]

#1413552 — Re: Dcache oops

From	Linus Torvalds <torvalds@linux-foundation.org>
Date	2016-06-04 00:40 +0200
Subject	Re: Dcache oops
Message-ID	<rG5UZ-Ei-15@gated-at.bofh.it>
In reply to	#1413543

On Fri, Jun 3, 2016 at 3:23 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> On Fri, Jun 03, 2016 at 03:00:02PM -0700, Linus Torvalds wrote:
>>>
>> Normally it's done at terminate_walk() time. But I note that in
>> walk_component(), we do put_link(nd) which does a do_delayed_call(),
>> but does *not* do a clear_delayed_call(), so now I think a subsequent
>> terminate_walk() might drop it *again*.
>
> Nope - put_link() also decrements nd->depth.  No double calls there...

Yeah, I figured that out, and then continued to try to look at other cases..

Happy to hear that you seem to have figured it out.

But why did it apparently only start happening now?

             Linus

[toc] | [prev] | [next] | [standalone]

#1413554 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-04 00:50 +0200
Subject	Re: Dcache oops
Message-ID	<rG64F-Hz-5@gated-at.bofh.it>
In reply to	#1413552

On Jun 3, 2016, at 6:36 PM, Linus Torvalds wrote:

> On Fri, Jun 3, 2016 at 3:23 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>> On Fri, Jun 03, 2016 at 03:00:02PM -0700, Linus Torvalds wrote:
>>>> 
>>> Normally it's done at terminate_walk() time. But I note that in
>>> walk_component(), we do put_link(nd) which does a do_delayed_call(),
>>> but does *not* do a clear_delayed_call(), so now I think a subsequent
>>> terminate_walk() might drop it *again*.
>> 
>> Nope - put_link() also decrements nd->depth.  No double calls there...
> 
> Yeah, I figured that out, and then continued to try to look at other cases..
> 
> Happy to hear that you seem to have figured it out.
> 
> But why did it apparently only start happening now?

Apparently nobody runs NFS in a serious manner these days anymore.
EOPENSTALE is only used by NFS.

[toc] | [prev] | [next] | [standalone]

#1413555 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-04 00:50 +0200
Subject	Re: Dcache oops
Message-ID	<rG64F-Hz-7@gated-at.bofh.it>
In reply to	#1413552

On Fri, Jun 03, 2016 at 03:36:22PM -0700, Linus Torvalds wrote:

> Happy to hear that you seem to have figured it out.
> 
> But why did it apparently only start happening now?

Oleg has started to use Lustre torture tests on NFS, that's all.  Note, BTW,
that first they'd triggered an oopsable bug (fairly easy to reproduce) in
nfs_atomic_open() that had been there for 3 years ;-/

[toc] | [prev] | [next] | [standalone]

#1413492 — Re: Dcache oops

From	Oleg Drokin <green@linuxhacker.ru>
Date	2016-06-03 23:20 +0200
Subject	Re: Dcache oops
Message-ID	<rG4Fz-8oJ-15@gated-at.bofh.it>
In reply to	#1413434

On Jun 3, 2016, at 4:07 PM, Al Viro wrote:

> On Fri, Jun 03, 2016 at 02:35:41PM -0400, Oleg Drokin wrote:
> 
>>>> [ 2642.364383] BUG: unable to handle kernel paging request at ffff880113f82000
>>>> [ 2642.365014] IP: [<ffffffff817f87d4>] bad_gs+0xd1d/0x1ba9
>>> 
>>> *ow*
>>> Could you dump your vmlinux (and System.map) somewhere on anonftp?
>>> This 'bad_gs' is there simply because it's one of the few labels in
>>> .fixup - to say anything useful we'll need to find out where we'd
>>> really come from.
>> 
>> I see.
>> vmlinux with debug symbols: http://knox.linuxhacker.ru/tmp/dcache/vmlinux.gz
>> System.map: http://knox.linuxhacker.ru/tmp/dcache/System.map.gz
> 
> OK...
> ffffffff817f87cd:       48 8d 0a                lea    (%rdx),%rcx
> ffffffff817f87d0:       48 83 e1 f8             and    $0xfffffffffffffff8,%rcx
> ffffffff817f87d4:       4c 8b 01                mov    (%rcx),%r8
> ffffffff817f87d7:       8d 0a                   lea    (%rdx),%ecx
> ffffffff817f87d9:       83 e1 07                and    $0x7,%ecx
> ffffffff817f87dc:       c1 e1 03                shl    $0x3,%ecx
> ffffffff817f87df:       49 d3 e8                shr    %cl,%r8
> ffffffff817f87e2:       e9 9b b3 a4 ff          jmpq   ffffffff81243b82 <__d_lookup+0x132>
> 
> Aha...  It's load_unaligned_zeropad() from dentry_string_cmp(), hitting
> a genuinely unmapped address.  That sends it into fixup, where it tries to
> load an aligned word containing the address in question, in hope that
> fault was on attempt to cross into the next page.  No such luck, address
> was aligned in the first place (it's in %rdx - 0xffff880113f82000), so
> we still oops.
> 
> The unexpected part is that unmapped address did *NOT* come from a dentry;
> it's .name of qstr we were looking for.  And your call chain was
> __d_lookup() <- d_lookup() <- lookup_open(), so in lookup_open() it was
> nd->last.name...
> 
> Can the same thing be reproduced (with NFS fix) on v4.6, ede4090, 7f427d3,
> 4e8440b?

Well, that was faster than I expected. 4e8440b triggers right away, so I guess
there's no point in trying the later ones?
BTW, just to confirm you are noticing - this is a DEBUG_PAGEALLOC build,
so all freed memory is unmapped which is likely causing this oops - as a sign
of use after free.

[   54.990119] BUG: unable to handle kernel paging request at ffff8800d2b7f000
[   54.990423] IP: [<ffffffff817f91b6>] bad_gs+0xcff/0x1b89
[   54.990598] PGD 2dca067 PUD 11f900067 PMD 11f86a067 PTE 80000000d2b7f060
[   54.990942] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
[   54.991044] Modules linked in: loop rpcsec_gss_krb5 joydev pcspkr i2c_piix4 acpi_cpufreq tpm_tis tpm nfsd drm_kms_helper ttm drm serio_raw virtio_blk
[   54.992301] CPU: 7 PID: 5550 Comm: file_concat.sh Not tainted 4.6.0-4e8440b+ #1
[   54.993019] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[   54.993407] task: ffff8800d6ec6200 ti: ffff880110f44000 task.ti: ffff880110f44000
[   54.994112] RIP: 0010:[<ffffffff817f91b6>]  [<ffffffff817f91b6>] bad_gs+0xcff/0x1b89
[   54.994870] RSP: 0018:ffff880110f47ba0  EFLAGS: 00010286
[   54.995288] RAX: ffff8800d2760f30 RBX: ffff8800d2760f00 RCX: ffff8800d2b7f000
[   54.995736] RDX: ffff8800d2b7f000 RSI: 0000000000000000 RDI: 0000000000000001
[   54.996154] RBP: ffff880110f47be8 R08: 0000000000000000 R09: ffff8800d2760ed0
[   54.996628] R10: 0000000000000059 R11: ffff8800d6ec6d78 R12: ffff8800d2694ed0
[   54.997044] R13: ffff880110f47df0 R14: ffff8800d2760f50 R15: 00000000b761b37e
[   54.997450] FS:  00007fe8804d2700(0000) GS:ffff88011f5c0000(0000) knlGS:0000000000000000
[   54.998164] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   54.998624] CR2: ffff8800d2b7f000 CR3: 000000011931f000 CR4: 00000000000006e0
[   54.999225] Stack:
[   54.999596]  ffffffff81243965 0000000000000096 ffff8800d2b7f000 ffffffff00000001
[   55.000632]  000000000000042c ffff880110f47df0 ffff8800d2694ed0 ffff880110f47de0
[   55.001497]  ffff880110f47df0 ffff880110f47c18 ffffffff81243b94 ffffffff81234b3e
[   55.002353] Call Trace:
[   55.002709]  [<ffffffff81243965>] ? __d_lookup+0x5/0x1b0
[   55.003104]  [<ffffffff81243b94>] d_lookup+0x84/0xb0
[   55.003487]  [<ffffffff81234b3e>] ? lookup_open+0xfe/0x7a0
[   55.003878]  [<ffffffff81234b3e>] lookup_open+0xfe/0x7a0
[   55.004272]  [<ffffffff8123717e>] path_openat+0x4de/0xfc0
[   55.004665]  [<ffffffff8123925e>] do_filp_open+0x7e/0xe0
[   55.005061]  [<ffffffff812330a0>] ? lock_rename+0x100/0x100
[   55.005452]  [<ffffffff817f5367>] ? _raw_spin_unlock+0x27/0x40
[   55.005846]  [<ffffffff8124868c>] ? __alloc_fd+0xbc/0x170
[   55.006246]  [<ffffffff81226896>] do_sys_open+0x116/0x1f0
[   55.006635]  [<ffffffff8122698e>] SyS_open+0x1e/0x20
[   55.007036]  [<ffffffff817f5b36>] entry_SYSCALL_64_fastpath+0x1e/0xad
[   55.007434] Code: e1 03 49 d3 e8 e9 8f a3 a4 ff 48 8d 0a 48 83 e1 f8 4c 8b 01 8d 0a 83 e1 07 c1 e1 03 49 d3 e8 e9 4e a4 a4 ff 48 8d 0a 48 83 e1 f8 <4c> 8b 01 8d 0a 83 e1 07 c1 e1 03 49 d3 e8 e9 c9 a8 a4 ff b9 f2 
[   55.011146] RIP  [<ffffffff817f91b6>] bad_gs+0xcff/0x1b89
[   55.011584]  RSP <ffff880110f47ba0>
[   55.011943] CR2: ffff8800d2b7f000

[toc] | [prev] | [next] | [standalone]

#1413516 — Re: Dcache oops

From	Al Viro <viro@ZenIV.linux.org.uk>
Date	2016-06-03 23:50 +0200
Subject	Re: Dcache oops
Message-ID	<rG58C-7y-21@gated-at.bofh.it>
In reply to	#1413492

On Fri, Jun 03, 2016 at 05:17:06PM -0400, Oleg Drokin wrote:

> > Can the same thing be reproduced (with NFS fix) on v4.6, ede4090, 7f427d3,
> > 4e8440b?
> 
> Well, that was faster than I expected. 4e8440b triggers right away, so I guess
> there's no point in trying the later ones?
> BTW, just to confirm you are noticing - this is a DEBUG_PAGEALLOC build,
> so all freed memory is unmapped which is likely causing this oops - as a sign
> of use after free.
 
> [   54.990119] BUG: unable to handle kernel paging request at ffff8800d2b7f000

Again a page-aligned nd->last.name and even smaller nd->last.len.  It smells
like a page that used to contain a symlink body, but got freed under us.

[toc] | [prev] | [next] | [standalone]

Page 2 of 3 — ← Prev page 1 [2] 3 Next page →

csiph-web

NFS/d_splice_alias breakage

Contents

#1413387 — Re: Dcache oops

#1413434 — Re: Dcache oops

#1413489 — Re: Dcache oops

#1413496 — Re: Dcache oops

#1413522 — Re: Dcache oops

#1413539 — Re: Dcache oops

#1413543 — Re: Dcache oops

#1413551 — Re: Dcache oops

#1413553 — Re: Dcache oops

#1413593 — Re: Dcache oops

#1413617 — Re: Dcache oops

#1413745 — Re: Dcache oops

#1413788 — Re: Dcache oops

#1413791 — [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.

#1413855 — Re: [PATCH] nfs4: Fix potential use after free of state in nfs4_do_reclaim.

#1413552 — Re: Dcache oops

#1413554 — Re: Dcache oops

#1413555 — Re: Dcache oops

#1413492 — Re: Dcache oops

#1413516 — Re: Dcache oops