linux-unionfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] ovl: fix oops in ovl_rename
@ 2021-09-24  1:16 Zheng Liang
  2021-09-24  7:12 ` 139
  2021-09-24 19:01 ` Miklos Szeredi
  0 siblings, 2 replies; 4+ messages in thread
From: Zheng Liang @ 2021-09-24  1:16 UTC (permalink / raw)
  To: miklos; +Cc: linux-unionfs, zhengliang6, yi.zhang

We find a kernel NULL pointer dereference problem in overlayfs.
The problem can appear in the following scene:

mkdir lower upper work merge
touch lower/old
touch lower/new
mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merge
rm merge/new
------------------------------------------------------------------------------------------
process A(rename file in merge)                                process B(delete file in upper)
renameat2(AT_FDCWD, "old", AT_FDCWD, "new", 0)
(new is whiteout file in upper)
  do_renameat2
    vfs_rename
      ovl_rename
      (overwrite=true,ovl_lower_positive(old)=true,
      ovl_dentry_is_whiteout(new)=true)
      (we can add some delay after "flags|=RENAME_EXCHANGE",
      it can make the problem appear more easy)
      ……                                                       unlink(new)
      ……                                                       (delete whiteout in upper)
      (newdentry is negative)
        ovl_do_rename

So,before commencing with ovl_do_rename that the flags maybe attach RENAME_EXCHANGE
and the newdentry is negative in ovl_rename.If we enabled selinux,it
will lead to kernel panic.such as the following log:
PID: 2552045  TASK: ffff8880302faf00  CPU: 2   COMMAND: "fsstress"
 #0 [ffff888080e772a0] machine_kexec at ffffffff856adedc
 #1 [ffff888080e773a8] __crash_kexec at ffffffff8585cd20
 #2 [ffff888080e774c0] panic at ffffffff8572b288
 #3 [ffff888080e77590] oops_end at ffffffff85641f6e
 #4 [ffff888080e775f0] __do_page_fault at ffffffff856cd55b
 #5 [ffff888080e77668] do_page_fault at ffffffff856cd834
 #6 [ffff888080e776a0] async_page_fault at ffffffff8660125e
    [exception RIP: __inode_security_revalidate+34]
    RIP: ffffffff85c43452  RSP: ffff888080e77758  RFLAGS: 00010202
    RAX: 0000000000000000  RBX: 0000000000000000  RCX: ffffffff8593ae7e
    RDX: 0000000000000000  RSI: 0000000000000297  RDI: 0000000000000297
    RBP: ffff8881984e6628   R8: ffffed10115e3f39   R9: ffffed10115e3f39
    R10: 0000000000000001  R11: ffffed10115e3f38  R12: 0000000000000001
    R13: 0000000000000000  R14: ffff88808350a000  R15: 1ffff110101ceef5
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
 #7 [ffff888080e77780] selinux_inode_rename at ffffffff85c4418d
 #8 [ffff888080e77858] security_inode_rename at ffffffff85c35f24
 #9 [ffff888080e77890] vfs_rename at ffffffff85b01209
 #10 [ffff888080e779a8] ovl_do_rename at ffffffffc0a44c22 [overlay]
 #11 [ffff888080e779d8] ovl_rename at ffffffffc0a46575 [overlay]
 #12 [ffff888080e77b48] vfs_rename at ffffffff85b0155a
 #13 [ffff888080e77c60] do_renameat2 at ffffffff85b06e65
 #14 [ffff888080e77f00] __x64_sys_renameat2 at ffffffff85b06fb2
 #15 [ffff888080e77f30] do_syscall_64 at ffffffff85606243
 #16 [ffff888080e77f50] entry_SYSCALL_64_after_hwframe at ffffffff866000ad

We can add some check in ovl_rename for this scene and return error to avoid kernel panic.

Signed-off-by: Zheng Liang <zhengliang6@huawei.com>
---
 fs/overlayfs/dir.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
index 1fefb2b8960e..93c7c267de93 100644
--- a/fs/overlayfs/dir.c
+++ b/fs/overlayfs/dir.c
@@ -1219,9 +1219,13 @@ static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
 				goto out_dput;
 		}
 	} else {
-		if (!d_is_negative(newdentry) &&
-		    (!new_opaque || !ovl_is_whiteout(newdentry)))
-			goto out_dput;
+		if (!d_is_negative(newdentry)) {
+			if (!new_opaque || !ovl_is_whiteout(newdentry))
+				goto out_dput;
+		} else {
+			if (flags & RENAME_EXCHANGE)
+				goto out_dput;
+		}
 	}
 
 	if (olddentry == trap)
-- 
2.25.4


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

* Re: [PATCH] ovl: fix oops in ovl_rename
  2021-09-24  1:16 [PATCH] ovl: fix oops in ovl_rename Zheng Liang
@ 2021-09-24  7:12 ` 139
  2021-09-24  8:50   ` zhengliang (A)
  2021-09-24 19:01 ` Miklos Szeredi
  1 sibling, 1 reply; 4+ messages in thread
From: 139 @ 2021-09-24  7:12 UTC (permalink / raw)
  To: Zheng Liang; +Cc: miklos, linux-unionfs, yi.zhang

在 2021年9月24日,09:09,Zheng Liang <zhengliang6@huawei.com> 写道:
> 
> We find a kernel NULL pointer dereference problem in overlayfs.
> The problem can appear in the following scene:
> 
> mkdir lower upper work merge
> touch lower/old
> touch lower/new
> mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merge
> rm merge/new
> ------------------------------------------------------------------------------------------
> process A(rename file in merge)                                process B(delete file in upper)
> renameat2(AT_FDCWD, "old", AT_FDCWD, "new", 0)
> (new is whiteout file in upper)
>  do_renameat2
>    vfs_rename
>      ovl_rename
>      (overwrite=true,ovl_lower_positive(old)=true,
>      ovl_dentry_is_whiteout(new)=true)
>      (we can add some delay after "flags|=RENAME_EXCHANGE",
>      it can make the problem appear more easy)
>      ……                                                       unlink(new)
>      ……                                                       (delete whiteout in upper)
>      (newdentry is negative)
>        ovl_do_rename

Is it a real use case? I believe there will be many unexpected behaviors if you delete the files in upper layer deliberately.

Thanks,
Chengguang



> 
> So,before commencing with ovl_do_rename that the flags maybe attach RENAME_EXCHANGE
> and the newdentry is negative in ovl_rename.If we enabled selinux,it
> will lead to kernel panic.such as the following log:
> PID: 2552045  TASK: ffff8880302faf00  CPU: 2   COMMAND: "fsstress"
> #0 [ffff888080e772a0] machine_kexec at ffffffff856adedc
> #1 [ffff888080e773a8] __crash_kexec at ffffffff8585cd20
> #2 [ffff888080e774c0] panic at ffffffff8572b288
> #3 [ffff888080e77590] oops_end at ffffffff85641f6e
> #4 [ffff888080e775f0] __do_page_fault at ffffffff856cd55b
> #5 [ffff888080e77668] do_page_fault at ffffffff856cd834
> #6 [ffff888080e776a0] async_page_fault at ffffffff8660125e
>    [exception RIP: __inode_security_revalidate+34]
>    RIP: ffffffff85c43452  RSP: ffff888080e77758  RFLAGS: 00010202
>    RAX: 0000000000000000  RBX: 0000000000000000  RCX: ffffffff8593ae7e
>    RDX: 0000000000000000  RSI: 0000000000000297  RDI: 0000000000000297
>    RBP: ffff8881984e6628   R8: ffffed10115e3f39   R9: ffffed10115e3f39
>    R10: 0000000000000001  R11: ffffed10115e3f38  R12: 0000000000000001
>    R13: 0000000000000000  R14: ffff88808350a000  R15: 1ffff110101ceef5
>    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
> #7 [ffff888080e77780] selinux_inode_rename at ffffffff85c4418d
> #8 [ffff888080e77858] security_inode_rename at ffffffff85c35f24
> #9 [ffff888080e77890] vfs_rename at ffffffff85b01209
> #10 [ffff888080e779a8] ovl_do_rename at ffffffffc0a44c22 [overlay]
> #11 [ffff888080e779d8] ovl_rename at ffffffffc0a46575 [overlay]
> #12 [ffff888080e77b48] vfs_rename at ffffffff85b0155a
> #13 [ffff888080e77c60] do_renameat2 at ffffffff85b06e65
> #14 [ffff888080e77f00] __x64_sys_renameat2 at ffffffff85b06fb2
> #15 [ffff888080e77f30] do_syscall_64 at ffffffff85606243
> #16 [ffff888080e77f50] entry_SYSCALL_64_after_hwframe at ffffffff866000ad
> 
> We can add some check in ovl_rename for this scene and return error to avoid kernel panic.
> 
> Signed-off-by: Zheng Liang <zhengliang6@huawei.com>
> ---
> fs/overlayfs/dir.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
> index 1fefb2b8960e..93c7c267de93 100644
> --- a/fs/overlayfs/dir.c
> +++ b/fs/overlayfs/dir.c
> @@ -1219,9 +1219,13 @@ static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
>                goto out_dput;
>        }
>    } else {
> -        if (!d_is_negative(newdentry) &&
> -            (!new_opaque || !ovl_is_whiteout(newdentry)))
> -            goto out_dput;
> +        if (!d_is_negative(newdentry)) {
> +            if (!new_opaque || !ovl_is_whiteout(newdentry))
> +                goto out_dput;
> +        } else {
> +            if (flags & RENAME_EXCHANGE)
> +                goto out_dput;
> +        }
>    }
> 
>    if (olddentry == trap)
> -- 
> 2.25.4
> 



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

* [PATCH] ovl: fix oops in ovl_rename
  2021-09-24  7:12 ` 139
@ 2021-09-24  8:50   ` zhengliang (A)
  0 siblings, 0 replies; 4+ messages in thread
From: zhengliang (A) @ 2021-09-24  8:50 UTC (permalink / raw)
  To: 139; +Cc: miklos, linux-unionfs, zhangyi (F)



> We find a kernel NULL pointer dereference problem in overlayfs.
> The problem can appear in the following scene:
> 
> mkdir lower upper work merge
> touch lower/old
> touch lower/new
> mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work 
> merge rm merge/new
> ------------------------------------------------------------------------------------------
> process A(rename file in merge)                                process B(delete file in upper)
> renameat2(AT_FDCWD, "old", AT_FDCWD, "new", 0) (new is whiteout file 
> in upper)
>  do_renameat2
>    vfs_rename
>      ovl_rename
>      (overwrite=true,ovl_lower_positive(old)=true,
>      ovl_dentry_is_whiteout(new)=true)
>      (we can add some delay after "flags|=RENAME_EXCHANGE",
>      it can make the problem appear more easy)
>      ……                                                       unlink(new)
>      ……                                                       (delete whiteout in upper)
>      (newdentry is negative)
>        ovl_do_rename

Is it a real use case? I believe there will be many unexpected behaviors if you delete the files in upper layer deliberately.

Thanks,
Chengguang


As described in the documentation (Documentation/filesystems/overlayfs.rst)
       Changes to underlying filesystems
       ---------------------------------
      Changes to the underlying filesystems while part of a mounted overlay
      filesystem are not allowed.  If the underlying filesystem is changed,
      the behavior of the overlay is undefined, though it will not result in
      a crash or deadlock.
So, We should probably fix this crash problem.

> 
> So,before commencing with ovl_do_rename that the flags maybe attach 
> RENAME_EXCHANGE and the newdentry is negative in ovl_rename.If we 
> enabled selinux,it will lead to kernel panic.such as the following log:
> PID: 2552045  TASK: ffff8880302faf00  CPU: 2   COMMAND: "fsstress"
> #0 [ffff888080e772a0] machine_kexec at ffffffff856adedc
> #1 [ffff888080e773a8] __crash_kexec at ffffffff8585cd20
> #2 [ffff888080e774c0] panic at ffffffff8572b288
> #3 [ffff888080e77590] oops_end at ffffffff85641f6e
> #4 [ffff888080e775f0] __do_page_fault at ffffffff856cd55b
> #5 [ffff888080e77668] do_page_fault at ffffffff856cd834
> #6 [ffff888080e776a0] async_page_fault at ffffffff8660125e
>    [exception RIP: __inode_security_revalidate+34]
>    RIP: ffffffff85c43452  RSP: ffff888080e77758  RFLAGS: 00010202
>    RAX: 0000000000000000  RBX: 0000000000000000  RCX: ffffffff8593ae7e
>    RDX: 0000000000000000  RSI: 0000000000000297  RDI: 0000000000000297
>    RBP: ffff8881984e6628   R8: ffffed10115e3f39   R9: ffffed10115e3f39
>    R10: 0000000000000001  R11: ffffed10115e3f38  R12: 0000000000000001
>    R13: 0000000000000000  R14: ffff88808350a000  R15: 1ffff110101ceef5
>    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
> #7 [ffff888080e77780] selinux_inode_rename at ffffffff85c4418d
> #8 [ffff888080e77858] security_inode_rename at ffffffff85c35f24
> #9 [ffff888080e77890] vfs_rename at ffffffff85b01209
> #10 [ffff888080e779a8] ovl_do_rename at ffffffffc0a44c22 [overlay]
> #11 [ffff888080e779d8] ovl_rename at ffffffffc0a46575 [overlay]
> #12 [ffff888080e77b48] vfs_rename at ffffffff85b0155a
> #13 [ffff888080e77c60] do_renameat2 at ffffffff85b06e65
> #14 [ffff888080e77f00] __x64_sys_renameat2 at ffffffff85b06fb2
> #15 [ffff888080e77f30] do_syscall_64 at ffffffff85606243
> #16 [ffff888080e77f50] entry_SYSCALL_64_after_hwframe at 
> ffffffff866000ad
> 
> We can add some check in ovl_rename for this scene and return error to avoid kernel panic.
> 
> Signed-off-by: Zheng Liang <zhengliang6@huawei.com>
> ---
> fs/overlayfs/dir.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c index 
> 1fefb2b8960e..93c7c267de93 100644
> --- a/fs/overlayfs/dir.c
> +++ b/fs/overlayfs/dir.c
> @@ -1219,9 +1219,13 @@ static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
>                goto out_dput;
>        }
>    } else {
> -        if (!d_is_negative(newdentry) &&
> -            (!new_opaque || !ovl_is_whiteout(newdentry)))
> -            goto out_dput;
> +        if (!d_is_negative(newdentry)) {
> +            if (!new_opaque || !ovl_is_whiteout(newdentry))
> +                goto out_dput;
> +        } else {
> +            if (flags & RENAME_EXCHANGE)
> +                goto out_dput;
> +        }
>    }
> 
>    if (olddentry == trap)
> --
> 2.25.4
> 



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

* Re: [PATCH] ovl: fix oops in ovl_rename
  2021-09-24  1:16 [PATCH] ovl: fix oops in ovl_rename Zheng Liang
  2021-09-24  7:12 ` 139
@ 2021-09-24 19:01 ` Miklos Szeredi
  1 sibling, 0 replies; 4+ messages in thread
From: Miklos Szeredi @ 2021-09-24 19:01 UTC (permalink / raw)
  To: Zheng Liang; +Cc: overlayfs, zhangyi (F)

On Fri, 24 Sept 2021 at 03:09, Zheng Liang <zhengliang6@huawei.com> wrote:
>
> We find a kernel NULL pointer dereference problem in overlayfs.
> The problem can appear in the following scene:
>
> mkdir lower upper work merge
> touch lower/old
> touch lower/new
> mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merge
> rm merge/new
> ------------------------------------------------------------------------------------------
> process A(rename file in merge)                                process B(delete file in upper)
> renameat2(AT_FDCWD, "old", AT_FDCWD, "new", 0)
> (new is whiteout file in upper)
>   do_renameat2
>     vfs_rename
>       ovl_rename
>       (overwrite=true,ovl_lower_positive(old)=true,
>       ovl_dentry_is_whiteout(new)=true)
>       (we can add some delay after "flags|=RENAME_EXCHANGE",
>       it can make the problem appear more easy)
>       ……                                                       unlink(new)
>       ……                                                       (delete whiteout in upper)
>       (newdentry is negative)
>         ovl_do_rename
>
> So,before commencing with ovl_do_rename that the flags maybe attach RENAME_EXCHANGE
> and the newdentry is negative in ovl_rename.If we enabled selinux,it
> will lead to kernel panic.such as the following log:
> PID: 2552045  TASK: ffff8880302faf00  CPU: 2   COMMAND: "fsstress"
>  #0 [ffff888080e772a0] machine_kexec at ffffffff856adedc
>  #1 [ffff888080e773a8] __crash_kexec at ffffffff8585cd20
>  #2 [ffff888080e774c0] panic at ffffffff8572b288
>  #3 [ffff888080e77590] oops_end at ffffffff85641f6e
>  #4 [ffff888080e775f0] __do_page_fault at ffffffff856cd55b
>  #5 [ffff888080e77668] do_page_fault at ffffffff856cd834
>  #6 [ffff888080e776a0] async_page_fault at ffffffff8660125e
>     [exception RIP: __inode_security_revalidate+34]
>     RIP: ffffffff85c43452  RSP: ffff888080e77758  RFLAGS: 00010202
>     RAX: 0000000000000000  RBX: 0000000000000000  RCX: ffffffff8593ae7e
>     RDX: 0000000000000000  RSI: 0000000000000297  RDI: 0000000000000297
>     RBP: ffff8881984e6628   R8: ffffed10115e3f39   R9: ffffed10115e3f39
>     R10: 0000000000000001  R11: ffffed10115e3f38  R12: 0000000000000001
>     R13: 0000000000000000  R14: ffff88808350a000  R15: 1ffff110101ceef5
>     ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
>  #7 [ffff888080e77780] selinux_inode_rename at ffffffff85c4418d
>  #8 [ffff888080e77858] security_inode_rename at ffffffff85c35f24
>  #9 [ffff888080e77890] vfs_rename at ffffffff85b01209
>  #10 [ffff888080e779a8] ovl_do_rename at ffffffffc0a44c22 [overlay]
>  #11 [ffff888080e779d8] ovl_rename at ffffffffc0a46575 [overlay]
>  #12 [ffff888080e77b48] vfs_rename at ffffffff85b0155a
>  #13 [ffff888080e77c60] do_renameat2 at ffffffff85b06e65
>  #14 [ffff888080e77f00] __x64_sys_renameat2 at ffffffff85b06fb2
>  #15 [ffff888080e77f30] do_syscall_64 at ffffffff85606243
>  #16 [ffff888080e77f50] entry_SYSCALL_64_after_hwframe at ffffffff866000ad
>
> We can add some check in ovl_rename for this scene and return error to avoid kernel panic.

Thanks.  Patch looks good; pushed to vfs.git#overlayfs-next (with a
cleaned up commit message).

Miklos

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

end of thread, other threads:[~2021-09-24 19:01 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-09-24  1:16 [PATCH] ovl: fix oops in ovl_rename Zheng Liang
2021-09-24  7:12 ` 139
2021-09-24  8:50   ` zhengliang (A)
2021-09-24 19:01 ` Miklos Szeredi

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).