($INBOX_DIR/description missing)
 help / color / Atom feed
* overlayfs: overlapping upperdir path
@ 2021-04-01 13:37 Miklos Szeredi
  2021-04-01 14:30 ` Amir Goldstein
  0 siblings, 1 reply; 4+ messages in thread
From: Miklos Szeredi @ 2021-04-01 13:37 UTC (permalink / raw)
  To: Amir Goldstein; +Cc: linux-fsdevel, overlayfs, Vivek Goyal

Commit 146d62e5a586 ("ovl: detect overlapping layers") made sure we
don't have overapping layers, but it also broke the arguably valid use
case of

 mount -olowerdir=/,upperdir=/subdir,..

where subdir also resides on the root fs.

I also see that we check for a trap at lookup time, so the question is
what does the up-front layer check buy us?

Thanks,
Miklos

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: overlayfs: overlapping upperdir path
  2021-04-01 13:37 overlayfs: overlapping upperdir path Miklos Szeredi
@ 2021-04-01 14:30 ` Amir Goldstein
  2021-04-01 15:09   ` Miklos Szeredi
  0 siblings, 1 reply; 4+ messages in thread
From: Amir Goldstein @ 2021-04-01 14:30 UTC (permalink / raw)
  To: Miklos Szeredi; +Cc: linux-fsdevel, overlayfs, Vivek Goyal

On Thu, Apr 1, 2021 at 4:37 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> Commit 146d62e5a586 ("ovl: detect overlapping layers") made sure we
> don't have overapping layers, but it also broke the arguably valid use
> case of
>
>  mount -olowerdir=/,upperdir=/subdir,..
>
> where subdir also resides on the root fs.

How is 'ls /merged/subdir' expected to behave in that use case?
Error?

>
> I also see that we check for a trap at lookup time, so the question is
> what does the up-front layer check buy us?
>

I'm not sure. I know it bought us silence from syzbot that started
mutating many overlapping layers repos....
Will the lookup trap have stopped it too? maybe. We did not try.

In general I think that if we can error out to user on mount time
it is preferred, but if we need to make that use case work, I'd try
to relax as minimum as possible from the check.

Thanks,
Amir.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: overlayfs: overlapping upperdir path
  2021-04-01 14:30 ` Amir Goldstein
@ 2021-04-01 15:09   ` Miklos Szeredi
  2021-04-01 15:19     ` Amir Goldstein
  0 siblings, 1 reply; 4+ messages in thread
From: Miklos Szeredi @ 2021-04-01 15:09 UTC (permalink / raw)
  To: Amir Goldstein; +Cc: linux-fsdevel, overlayfs, Vivek Goyal


[-- Attachment #1: Type: text/plain, Size: 1231 bytes --]

On Thu, Apr 1, 2021 at 4:30 PM Amir Goldstein <amir73il@gmail.com> wrote:
>
> On Thu, Apr 1, 2021 at 4:37 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > Commit 146d62e5a586 ("ovl: detect overlapping layers") made sure we
> > don't have overapping layers, but it also broke the arguably valid use
> > case of
> >
> >  mount -olowerdir=/,upperdir=/subdir,..
> >
> > where subdir also resides on the root fs.
>
> How is 'ls /merged/subdir' expected to behave in that use case?
> Error?

-ELOOP is the error returned.

>
> >
> > I also see that we check for a trap at lookup time, so the question is
> > what does the up-front layer check buy us?
> >
>
> I'm not sure. I know it bought us silence from syzbot that started
> mutating many overlapping layers repos....
> Will the lookup trap have stopped it too? maybe. We did not try.
>
> In general I think that if we can error out to user on mount time
> it is preferred, but if we need to make that use case work, I'd try
> to relax as minimum as possible from the check.

Certainly.  Like lower inside upper makes zero sense, OTOH upper
inside lower does.   So I think we just need to relax the
upperdir/workdir layer check in this case.

Like attached patch.

Thanks,
Miklos

[-- Attachment #2: ovl-allow-upperdir-inside-lowerdir.patch --]
[-- Type: text/x-patch, Size: 1882 bytes --]

diff --git a/fs/overlayfs/super.c b/fs/overlayfs/super.c
index fdd72f1a9c5e..8cf343335029 100644
--- a/fs/overlayfs/super.c
+++ b/fs/overlayfs/super.c
@@ -1826,7 +1826,8 @@ static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
  * - upper/work dir of any overlayfs instance
  */
 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
-			   struct dentry *dentry, const char *name)
+			   struct dentry *dentry, const char *name,
+			   bool is_lower)
 {
 	struct dentry *next = dentry, *parent;
 	int err = 0;
@@ -1838,7 +1839,7 @@ static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
 
 	/* Walk back ancestors to root (inclusive) looking for traps */
 	while (!err && parent != next) {
-		if (ovl_lookup_trap_inode(sb, parent)) {
+		if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
 			err = -ELOOP;
 			pr_err("overlapping %s path\n", name);
 		} else if (ovl_is_inuse(parent)) {
@@ -1864,7 +1865,7 @@ static int ovl_check_overlapping_layers(struct super_block *sb,
 
 	if (ovl_upper_mnt(ofs)) {
 		err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
-				      "upperdir");
+				      "upperdir", false);
 		if (err)
 			return err;
 
@@ -1875,7 +1876,8 @@ static int ovl_check_overlapping_layers(struct super_block *sb,
 		 * workbasedir.  In that case, we already have their traps in
 		 * inode cache and we will catch that case on lookup.
 		 */
-		err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
+		err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
+				      false);
 		if (err)
 			return err;
 	}
@@ -1883,7 +1885,7 @@ static int ovl_check_overlapping_layers(struct super_block *sb,
 	for (i = 1; i < ofs->numlayer; i++) {
 		err = ovl_check_layer(sb, ofs,
 				      ofs->layers[i].mnt->mnt_root,
-				      "lowerdir");
+				      "lowerdir", true);
 		if (err)
 			return err;
 	}

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: overlayfs: overlapping upperdir path
  2021-04-01 15:09   ` Miklos Szeredi
@ 2021-04-01 15:19     ` Amir Goldstein
  0 siblings, 0 replies; 4+ messages in thread
From: Amir Goldstein @ 2021-04-01 15:19 UTC (permalink / raw)
  To: Miklos Szeredi; +Cc: linux-fsdevel, overlayfs, Vivek Goyal

On Thu, Apr 1, 2021 at 6:09 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Thu, Apr 1, 2021 at 4:30 PM Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Thu, Apr 1, 2021 at 4:37 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > >
> > > Commit 146d62e5a586 ("ovl: detect overlapping layers") made sure we
> > > don't have overapping layers, but it also broke the arguably valid use
> > > case of
> > >
> > >  mount -olowerdir=/,upperdir=/subdir,..
> > >
> > > where subdir also resides on the root fs.
> >
> > How is 'ls /merged/subdir' expected to behave in that use case?
> > Error?
>
> -ELOOP is the error returned.
>
> >
> > >
> > > I also see that we check for a trap at lookup time, so the question is
> > > what does the up-front layer check buy us?
> > >
> >
> > I'm not sure. I know it bought us silence from syzbot that started
> > mutating many overlapping layers repos....
> > Will the lookup trap have stopped it too? maybe. We did not try.
> >
> > In general I think that if we can error out to user on mount time
> > it is preferred, but if we need to make that use case work, I'd try
> > to relax as minimum as possible from the check.
>
> Certainly.  Like lower inside upper makes zero sense, OTOH upper
> inside lower does.   So I think we just need to relax the
> upperdir/workdir layer check in this case.
>
> Like attached patch.
>

Fine by me.
Let's let syzbot have fun ;-)

Thanks,
Amir.

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, back to index

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-01 13:37 overlayfs: overlapping upperdir path Miklos Szeredi
2021-04-01 14:30 ` Amir Goldstein
2021-04-01 15:09   ` Miklos Szeredi
2021-04-01 15:19     ` Amir Goldstein

($INBOX_DIR/description missing)

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-unionfs/0 linux-unionfs/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-unionfs linux-unionfs/ https://lore.kernel.org/linux-unionfs \
		linux-unionfs@vger.kernel.org
	public-inbox-index linux-unionfs

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-unionfs


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git