linux-kernel - Re: [PATCH v2] fs/netfs3: add a boundary check for EA

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <YyH0dkkuJnFNE9xT@kadam>
Date:   Wed, 14 Sep 2022 18:34:14 +0300
From:   Dan Carpenter <dan.carpenter@...cle.com>
To:     eadavis@...a.com
Cc:     syzbot+c4d950787fd5553287b7@...kaller.appspotmail.com,
        almaz.alexandrovich@...agon-software.com,
        linux-kernel@...r.kernel.org, llvm@...ts.linux.dev,
        nathan@...nel.org, ndesaulniers@...gle.com, ntfs3@...ts.linux.dev,
        syzkaller-bugs@...glegroups.com, trix@...hat.com
Subject: Re: [PATCH v2] fs/netfs3: add a boundary check for EA_FULL

On Mon, Sep 12, 2022 at 02:54:31PM +0800, eadavis@...a.com wrote:
> From: Edward Adam Davis <eadavis@...a.com>
> 
> the root cause is: 
> The remaining space after the offset is less than the space needed to 
> accommodate the next EA_FULL struct.
> 

This needs to be checked on the first iteration as well before calling
unpacked_ea_size(ea).

	if (bytes - *off < sizeof(*ea))
		return false;

> Link: https://syzkaller.appspot.com/bug?extid=c4d950787fd5553287b7
> Reported-by: syzbot+c4d950787fd5553287b7@...kaller.appspotmail.com
> Signed-off-by: Edward Adam Davis <eadavis@...a.com>
> ---
>  fs/ntfs3/xattr.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/fs/ntfs3/xattr.c b/fs/ntfs3/xattr.c
> index 7de8718c68a9..c90cc453390d 100644
> --- a/fs/ntfs3/xattr.c
> +++ b/fs/ntfs3/xattr.c
> @@ -52,6 +52,7 @@ static inline bool find_ea(const struct EA_FULL *ea_all, u32 bytes,
>  	for (;;) {
>  		const struct EA_FULL *ea = Add2Ptr(ea_all, *off);
>  		u32 next_off = *off + unpacked_ea_size(ea);
> +		u32 next_len = 0;
>  
>  		if (next_off > bytes)
>  			return false;
> @@ -63,6 +64,13 @@ static inline bool find_ea(const struct EA_FULL *ea_all, u32 bytes,
>  		*off = next_off;
>  		if (next_off >= bytes)
>  			return false;
> +
> +		next_len = next_off + 8;

8 is a magic number.  Use sizeof(*ea).

> +		if (next_len >= bytes ||
> +		    ((!ea->size) &&
> +		     (next_len + ea->name_len + 
> +		      le16_to_cpu(ea->elength) >= bytes)))

This is open coding unpacked_ea_size() but slightly different/incorrect.
No need to check this anyway, because it gets checked at the start of
the iteration.

If we add the if (bytes - *off < sizeof(*ea)) check to the start of the
iteration that means we can delete the "if (next_off == bytes)" check
from the end.

The second issue with this code is that unpacked_ea_size() is a user
controlled u32.  We're adding it to "*off" which is also a u32 so that
can have an integer overflow...

See diff below.  (Untested).

regards,
dan carpenter

diff --git a/fs/ntfs3/xattr.c b/fs/ntfs3/xattr.c
index 7de8718c68a9..c3dbe06fb784 100644
--- a/fs/ntfs3/xattr.c
+++ b/fs/ntfs3/xattr.c
@@ -44,14 +44,20 @@ static inline size_t packed_ea_size(const struct EA_FULL *ea)
 static inline bool find_ea(const struct EA_FULL *ea_all, u32 bytes,
 			   const char *name, u8 name_len, u32 *off)
 {
+	const struct EA_FULL *ea;
+	u32 next_off;
+
 	*off = 0;
 
-	if (!ea_all || !bytes)
+	if (!ea_all)
 		return false;
 
 	for (;;) {
-		const struct EA_FULL *ea = Add2Ptr(ea_all, *off);
-		u32 next_off = *off + unpacked_ea_size(ea);
+		if (bytes - *off < sizeof(*ea))
+			return false;
+
+		ea = Add2Ptr(ea_all, *off);
+		next_off = size_add(*off, unpacked_ea_size(ea));
 
 		if (next_off > bytes)
 			return false;
@@ -61,8 +67,6 @@ static inline bool find_ea(const struct EA_FULL *ea_all, u32 bytes,
 			return true;
 
 		*off = next_off;
-		if (next_off >= bytes)
-			return false;
 	}
 }