2023-03-26 16:26:03

by Immad Mir

[permalink] [raw]
Subject: [PATCH] FS: JFS: Fix null-ptr-deref Read in txBegin

From: Immad Mir <[email protected]>

syzkaller reported the following issue:

BUG: KASAN: null-ptr-deref in instrument_atomic_read include/linux/instrumented.h:72 [inline]
BUG: KASAN: null-ptr-deref in _test_bit include/asm-generic/bitops/instrumented-non-atomic.h:141 [inline]
BUG: KASAN: null-ptr-deref in txBegin+0x131/0x6c0 fs/jfs/jfs_txnmgr.c:366
Read of size 8 at addr 0000000000000040 by task syz-executor.0/5098

CPU: 0 PID: 5098 Comm: syz-executor.0 Not tainted 6.3.0-rc3-syzkaller-00005-g7d31677bb7b1 #0
Hardware name: Google Compute Engine/Google Compute Engine, BIOS Google 03/02/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0x1e7/0x2d0 lib/dump_stack.c:106
print_report+0xe6/0x540 mm/kasan/report.c:433
kasan_report+0x176/0x1b0 mm/kasan/report.c:536
kasan_check_range+0x283/0x290 mm/kasan/generic.c:187
instrument_atomic_read include/linux/instrumented.h:72 [inline]
_test_bit include/asm-generic/bitops/instrumented-non-atomic.h:141 [inline]
txBegin+0x131/0x6c0 fs/jfs/jfs_txnmgr.c:366
jfs_link+0x1ac/0x5e0 fs/jfs/namei.c:802
vfs_link+0x4ed/0x680 fs/namei.c:4522
do_linkat+0x5cc/0x9e0 fs/namei.c:4593
__do_sys_linkat fs/namei.c:4621 [inline]
__se_sys_linkat fs/namei.c:4618 [inline]
__x64_sys_linkat+0xdd/0xf0 fs/namei.c:4618
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd

The issue can be resolved by checking whethere "log"
for a given superblock exists in the jfs_link function
before beginning a transaction.

Tested with syzbot.
Reported-by: [email protected]
Link: https://syzkaller.appspot.com/bug?id=be7e52c50c5182cc09a09ea6fc456446b2039de3

Signed-off-by: Immad Mir <[email protected]>
---
fs/jfs/namei.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
index b29d68b5e..cd43b68e2 100644
--- a/fs/jfs/namei.c
+++ b/fs/jfs/namei.c
@@ -799,6 +799,8 @@ static int jfs_link(struct dentry *old_dentry,
if (rc)
goto out;

+ if (!(JFS_SBI(ip->i_sb)->log))
+ goto out;
tid = txBegin(ip->i_sb, 0);

mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
--
2.40.0


2023-06-23 13:47:42

by Immad Mir

[permalink] [raw]
Subject: [PATCH] FS: JFS: Fix null-ptr-deref Read in txBegin

From: Immad Mir <[email protected]>

Syzkaller reported an issue where txBegin may be called
on a superblock in a read-only mounted filesystem which leads
to NULL pointer deref. This could be solved by checking if
the filesystem is read-only before calling txBegin, and returning
with appropiate error code.

Reported-By: [email protected]
Link: https://syzkaller.appspot.com/bug?id=be7e52c50c5182cc09a09ea6fc456446b2039de3

Signed-off-by: Immad Mir <[email protected]>
---
fs/jfs/namei.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
index b29d68b5e..12e95431c 100644
--- a/fs/jfs/namei.c
+++ b/fs/jfs/namei.c
@@ -798,8 +798,13 @@ static int jfs_link(struct dentry *old_dentry,
rc = dquot_initialize(dir);
if (rc)
goto out;
-
- tid = txBegin(ip->i_sb, 0);
+ if (!isReadOnly(ip)) {
+ tid = txBegin(ip->i_sb, 0);
+ } else {
+ jfs_error(ip->i_sb, "read-only filesystem\n");
+ rc = -EROFS;
+ goto out;
+ }

mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
--
2.40.0


2023-06-23 14:05:50

by Dave Kleikamp

[permalink] [raw]
Subject: Re: [PATCH] FS: JFS: Fix null-ptr-deref Read in txBegin

On 6/23/23 8:44AM, [email protected] wrote:
> From: Immad Mir <[email protected]>
>
> Syzkaller reported an issue where txBegin may be called
> on a superblock in a read-only mounted filesystem which leads
> to NULL pointer deref. This could be solved by checking if
> the filesystem is read-only before calling txBegin, and returning
> with appropiate error code.

Looks good. I'm going to change it to just an if clause without the else
and push it to jfs-next.

Thanks,
>
> Reported-By: [email protected]
> Link: https://syzkaller.appspot.com/bug?id=be7e52c50c5182cc09a09ea6fc456446b2039de3
>
> Signed-off-by: Immad Mir <[email protected]>
> ---
> fs/jfs/namei.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> index b29d68b5e..12e95431c 100644
> --- a/fs/jfs/namei.c
> +++ b/fs/jfs/namei.c
> @@ -798,8 +798,13 @@ static int jfs_link(struct dentry *old_dentry,
> rc = dquot_initialize(dir);
> if (rc)
> goto out;
> -
> - tid = txBegin(ip->i_sb, 0);
> + if (!isReadOnly(ip)) {
> + tid = txBegin(ip->i_sb, 0);
> + } else {
> + jfs_error(ip->i_sb, "read-only filesystem\n");
> + rc = -EROFS;
> + goto out;
> + }
>
> mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
> mutex_lock_nested(&JFS_IP(ip)->commit_mutex, COMMIT_MUTEX_CHILD);
> --
> 2.40.0
>

2023-06-23 14:05:52

by Dave Kleikamp

[permalink] [raw]
Subject: Re: [PATCH] FS: JFS: Fix null-ptr-deref Read in txBegin

On 6/23/23 8:40AM, Immad Mir wrote:
> Thanks for the feedback. I've prepared two patches to fix the bug.
>
> >Does this test case attempt to remount a read-only file system as
> read-write? I see a potential bug there.
> I'm not really sure about this.
>
> > Should be setting rc to an error here. I suggest -EROFS, but anything
> is better than returning zero. Calling jfs_error() might also be in
> order, as that would explicitly mark the file system to read-only. (The
> default behavior at least.)
>
> I've incorporated your suggested changes.
>
> > It'd be nice if we could move the check to txBegin(), but it is
> assumed to always succeed, so there's no good error recovery there
> without changing all of the callers. Maybe we can call jfs_error() there
> in case we get there from another syscall.
>
> I am not sure what to do here. I am calling jfs_error and returning 0
> which is not what the caller would expect.

jfs_error returns void. What it does is mark the superblock dirty and
either panics, marks sets the filesystem read-only, or does nothing
else, depending on a mount flag which I doubt anyone uses, so the
default action is to set it read-only. You still have to set a return
code, etc. after calling it.

I'm sorry I wasn't more specific the first time.

>
> Thanks,
> Immad.
>
>
> On Thu, Jun 22, 2023 at 8:38 PM Dave Kleikamp <[email protected]
> <mailto:[email protected]>> wrote:
>
> On 6/20/23 10:53PM, Immad Mir wrote:
>
>>
>>
>> Hi. May I please request a review on this patch.
>
> Sorry for the delay. See below.
>
>>
>> Thanks,
>> Immad
>>
>> ------------------------------------------------------------------------
>> *From:* [email protected] <mailto:[email protected]>
>> <[email protected]> <mailto:[email protected]>
>> *Sent:* Sunday, March 26, 2023 9:51:15 PM
>> *Cc:* [email protected] <mailto:[email protected]>
>> <[email protected]> <mailto:[email protected]>;
>> [email protected] <mailto:[email protected]>
>> <[email protected]> <mailto:[email protected]>;
>> Immad Mir <[email protected]> <mailto:[email protected]>;
>> [email protected]
>> <mailto:[email protected]>
>> <[email protected]>
>> <mailto:[email protected]>;
>> Dave Kleikamp <[email protected]> <mailto:[email protected]>;
>> [email protected]
>> <mailto:[email protected]>
>> <[email protected]>
>> <mailto:[email protected]>;
>> [email protected] <mailto:[email protected]>
>> <[email protected]> <mailto:[email protected]>
>> *Subject:* [PATCH] FS: JFS: Fix null-ptr-deref Read in txBegin
>>
>> From: Immad Mir <[email protected]> <mailto:[email protected]>
>>
>> syzkaller reported the following issue:
>>
>> BUG: KASAN: null-ptr-deref in instrument_atomic_read
>> include/linux/instrumented.h:72 [inline]
>> BUG: KASAN: null-ptr-deref in _test_bit
>> include/asm-generic/bitops/instrumented-non-atomic.h:141 [inline]
>> BUG: KASAN: null-ptr-deref in txBegin+0x131/0x6c0
>> fs/jfs/jfs_txnmgr.c:366
>> Read of size 8 at addr 0000000000000040 by task syz-executor.0/5098
>>
>> CPU: 0 PID: 5098 Comm: syz-executor.0 Not tainted
>> 6.3.0-rc3-syzkaller-00005-g7d31677bb7b1 #0
>> Hardware name: Google Compute Engine/Google Compute Engine, BIOS
>> Google 03/02/2023
>> Call Trace:
>>  <TASK>
>>  __dump_stack lib/dump_stack.c:88 [inline]
>>  dump_stack_lvl+0x1e7/0x2d0 lib/dump_stack.c:106
>>  print_report+0xe6/0x540 mm/kasan/report.c:433
>>  kasan_report+0x176/0x1b0 mm/kasan/report.c:536
>>  kasan_check_range+0x283/0x290 mm/kasan/generic.c:187
>>  instrument_atomic_read include/linux/instrumented.h:72 [inline]
>>  _test_bit
>> include/asm-generic/bitops/instrumented-non-atomic.h:141 [inline]
>>  txBegin+0x131/0x6c0 fs/jfs/jfs_txnmgr.c:366
>>  jfs_link+0x1ac/0x5e0 fs/jfs/namei.c:802
>>  vfs_link+0x4ed/0x680 fs/namei.c:4522
>>  do_linkat+0x5cc/0x9e0 fs/namei.c:4593
>>  __do_sys_linkat fs/namei.c:4621 [inline]
>>  __se_sys_linkat fs/namei.c:4618 [inline]
>>  __x64_sys_linkat+0xdd/0xf0 fs/namei.c:4618
>>  do_syscall_x64 arch/x86/entry/common.c:50 [inline]
>>  do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
>>  entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>
>> The issue can be resolved by checking whethere "log"
>> for a given superblock exists in the jfs_link function
>> before beginning a transaction.
>
> I'm not sure how we got here. log should only be null if the file
> system is mounted read-only. Does this test case attempt to remount
> a read-only file system as read-write? I see a potential bug there.
>
>>
>> Tested with syzbot.
>> Reported-by: [email protected]
>> <mailto:[email protected]>
>> Link:
>> https://syzkaller.appspot.com/bug?id=be7e52c50c5182cc09a09ea6fc456446b2039de3 <https://syzkaller.appspot.com/bug?id=be7e52c50c5182cc09a09ea6fc456446b2039de3>
>>
>> Signed-off-by: Immad Mir <[email protected]>
>> <mailto:[email protected]>
>> ---
>>  fs/jfs/namei.c | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
>> index b29d68b5e..cd43b68e2 100644
>> --- a/fs/jfs/namei.c
>> +++ b/fs/jfs/namei.c
>> @@ -799,6 +799,8 @@ static int jfs_link(struct dentry *old_dentry,
>>          if (rc)
>>                  goto out;
>>
>> +       if (!(JFS_SBI(ip->i_sb)->log))
>> +               goto out;
>
> Should be setting rc to an error here. I suggest -EROFS, but
> anything is better than returning zero. Calling jfs_error() might
> also be in order, as that would explicitly mark the file system to
> read-only. (The default behavior at least.)
>
>>          tid = txBegin(ip->i_sb, 0);
> It'd be nice if we could move the check to txBegin(), but it is
> assumed to always succeed, so there's no good error recovery there
> without changing all of the callers. Maybe we can call jfs_error()
> there in case we get there from another syscall.
>>
>> mutex_lock_nested(&JFS_IP(dir)->commit_mutex, COMMIT_MUTEX_PARENT);
>> --
>> 2.40.0
>>
>>