Re: [PATCH] memfd: Check for non-NULL file_seals in memfd_create() syscall

[Roberto Sassu <roberto.sassu@huaweicloud.com>]

On 6/8/2023 11:05 PM, Roberto Sassu wrote:
> On 6/7/2023 7:21 PM, Andrew Morton wrote:
>> On Wed,  7 Jun 2023 15:24:27 +0200 Roberto Sassu 
>> <roberto.sassu@huaweicloud.com> wrote:
>>
>>> From: Roberto Sassu <roberto.sassu@huawei.com>
>>>
>>> Ensure that file_seals is non-NULL before using it in the memfd_create()
>>> syscall. One situation in which memfd_file_seals_ptr() could return a 
>>> NULL
>>> pointer is when CONFIG_SHMEM=n.
>>
>> Thanks.  Has thie crash actually been demonstrated?
> 
> Welcome. Yes, I noticed it when booting Fedora 38:
> 
> Jun 07 11:45:17 localhost kernel: BUG: kernel NULL pointer dereference, 
> address: 0000000000000000
> Jun 07 11:45:17 localhost kernel: #PF: supervisor write access in kernel 
> mode
> Jun 07 11:45:17 localhost kernel: #PF: error_code(0x0002) - not-present 
> page
> Jun 07 11:45:17 localhost kernel: PGD 0 P4D 0
> Jun 07 11:45:17 localhost kernel: Oops: 0002 [#1] PREEMPT SMP NOPTI
> Jun 07 11:45:17 localhost kernel: CPU: 0 PID: 752 Comm: dbus-broker-lau 
> Not tainted 6.4.0-rc1+ #596
> Jun 07 11:45:17 localhost kernel: Hardware name: QEMU Standard PC (Q35 + 
> ICH9, 2009), BIOS 1.13.0-1ubuntu1.1 04/01/2014
> Jun 07 11:45:17 localhost kernel: RIP: 
> 0010:__do_sys_memfd_create+0x2a4/0x320
> Jun 07 11:45:17 localhost kernel: Code: ff 83 e3 02 0f 84 6a ff ff ff 49 
> 81 7d 28 00 cd 24 82 74 0c 4c 89 ef e8 5a 6c 27 00 84 c0 74 29 49 8b 45 
> 20 48 05 88 04 00 0>
> Jun 07 11:45:17 localhost kernel: RSP: 0018:ffffc900007d3ef8 EFLAGS: 
> 00010246
> Jun 07 11:45:17 localhost kernel: RAX: 0000000000000000 RBX: 
> 0000000000000002 RCX: 0000000000000000
> Jun 07 11:45:17 localhost kernel: RDX: 0000000000000003 RSI: 
> 0000000000000004 RDI: ffff888012ab1600
> Jun 07 11:45:17 localhost kernel: RBP: ffff888025eb4d60 R08: 
> 0000000000000000 R09: 0000000000000000
> Jun 07 11:45:17 localhost kernel: R10: ffffffff83524c90 R11: 
> 0000000000000000 R12: 000000000000000c
> Jun 07 11:45:17 localhost kernel: R13: ffff888012ab1600 R14: 
> ffff888025eb4d66 R15: 0000000000000000
> Jun 07 11:45:17 localhost kernel: FS:  00007f4ef466de80(0000) 
> GS:ffff88807dc00000(0000) knlGS:0000000000000000
> Jun 07 11:45:17 localhost kernel: CS:  0010 DS: 0000 ES: 0000 CR0: 
> 0000000080050033
> Jun 07 11:45:17 localhost kernel: CR2: 0000000000000000 CR3: 
> 000000000e7fe000 CR4: 0000000000350ef0
> Jun 07 11:45:17 localhost kernel: Call Trace:
> Jun 07 11:45:17 localhost kernel:  <TASK>
> Jun 07 11:45:17 localhost kernel:  do_syscall_64+0x3b/0x90
> Jun 07 11:45:17 localhost kernel:  entry_SYSCALL_64_after_hwframe+0x72/0xdc
> Jun 07 11:45:17 localhost kernel: RIP: 0033:0x7f4ef4a0dd2d
> Jun 07 11:45:17 localhost kernel: Code: c3 66 2e 0f 1f 84 00 00 00 00 00 
> 66 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 
> 4c 8b 4c 24 08 0f 0>
> Jun 07 11:45:17 localhost kernel: RSP: 002b:00007fff308c5b58 EFLAGS: 
> 00000246 ORIG_RAX: 000000000000013f
> Jun 07 11:45:17 localhost kernel: RAX: ffffffffffffffda RBX: 
> 00005597c14742d8 RCX: 00007f4ef4a0dd2d
> Jun 07 11:45:17 localhost kernel: RDX: 00007f4ef49dfa5b RSI: 
> 0000000000000003 RDI: 00005597c0b8f778
> Jun 07 11:45:17 localhost kernel: RBP: 00007fff308c5b80 R08: 
> 0000000000000000 R09: 00007fff308c5fa0
> Jun 07 11:45:17 localhost kernel: R10: 0000000000000000 R11: 
> 0000000000000246 R12: 0000000000000000
> Jun 07 11:45:17 localhost kernel: R13: 00007fff308c5e60 R14: 
> 00007fff308c5c20 R15: 00007fff308c5c00
> Jun 07 11:45:17 localhost kernel:  </TASK>
> Jun 07 11:45:17 localhost kernel: Modules linked in: sunrpc 
> intel_rapl_msr intel_rapl_common kvm_amd snd_hda_codec_generic ccp 
> snd_hda_intel snd_intel_dspcfg iTCO_w>
> Jun 07 11:45:17 localhost kernel: CR2: 0000000000000000
> Jun 07 11:45:17 localhost kernel: ---[ end trace 0000000000000000 ]---
> 
> Thanks for picking the patches. If it is still possible, it seems
> that the Fixes tag is incorrect. The fixed commit should be:
> 
> c3b1b1cbf002 ("ramfs: add support for "mode=" mount option")

Ops, sorry. The comment above was for the patch: shmem: Use 
ramfs_kill_sb() for kill_sb method of ramfs-based tmpfs.

Thanks

Roberto

> Roberto
> 
>>> --- a/mm/memfd.c
>>> +++ b/mm/memfd.c
>>> @@ -371,12 +371,15 @@ SYSCALL_DEFINE2(memfd_create,
>>>           inode->i_mode &= ~0111;
>>>           file_seals = memfd_file_seals_ptr(file);
>>> -        *file_seals &= ~F_SEAL_SEAL;
>>> -        *file_seals |= F_SEAL_EXEC;
>>> +        if (file_seals) {
>>> +            *file_seals &= ~F_SEAL_SEAL;
>>> +            *file_seals |= F_SEAL_EXEC;
>>> +        }
>>>       } else if (flags & MFD_ALLOW_SEALING) {
>>>           /* MFD_EXEC and MFD_ALLOW_SEALING are set */
>>>           file_seals = memfd_file_seals_ptr(file);
>>> -        *file_seals &= ~F_SEAL_SEAL;
>>> +        if (file_seals)
>>> +            *file_seals &= ~F_SEAL_SEAL;
>>>       }
>>>       fd_install(fd, file);