Re: [Linux-cifsd-devel] [PATCH] cifsd: Do not use 0 as TreeID

[Stefan Metzmacher <metze@samba.org>]

[-- Attachment #1.1: Type: text/plain, Size: 5164 bytes --]

Am 15.05.21 um 16:10 schrieb Namjae Jeon:
> 2021-05-15 17:57 GMT+09:00, Stefan Metzmacher <metze@samba.org>:
>>
>> Am 15.05.21 um 07:18 schrieb Namjae Jeon:
>>> 2021-05-14 22:11 GMT+09:00, Stefan Metzmacher via Linux-cifsd-devel
>>> <linux-cifsd-devel@lists.sourceforge.net>:
>>>>
>>>> Am 14.05.21 um 14:52 schrieb Marios Makassikis:
>>>>> Returning TreeID=0 is valid behaviour according to [MS-SMB2] 2.2.1.2:
>>>>>
>>>>>   TreeId (4 bytes): Uniquely identifies the tree connect for the
>>>>> command.
>>>>>   This MUST be 0 for the SMB2 TREE_CONNECT Request. The TreeId can be
>>>>>   any unsigned 32-bit integer that is received from a previous
>>>>>   SMB2 TREE_CONNECT Response. TreeId SHOULD be set to 0 for the
>>>>>   following commands:
>>>>>    [...]
>>>>>
>>>>> However, some client implementations reject it as invalid. Windows 7/10
>>>>> assigns ids starting from 1, and samba4 returns a random uint32_t
>>>>> which suggests there may be other clients that consider it is
>>>>> invalid behaviour.
>>>>>
>>>>> While here, simplify ksmbd_acquire_smb2_tid. 0xFFFF is a reserved value
>>>>> for CIFS/SMB1:
>>>>>   [MS-CIFS] 2.2.4.50.2
>>>>>
>>>>>   TID (2 bytes): The newly generated Tree ID, used in subsequent CIFS
>>>>>   client requests to refer to a resource relative to the
>>>>>   SMB_Data.Bytes.Path specified in the request. Most access to the
>>>>>   server requires a valid TID, whether the resource is password
>>>>>   protected or not. The value 0xFFFF is reserved; the server MUST NOT
>>>>>   return a TID value of 0xFFFF.
>>>>>
>>>>> Signed-off-by: Marios Makassikis <mmakassikis@freebox.fr>
>>>>> ---
>>>>> Example library that treats zero TreeID as invalid:
>>>>>
>>>>> https://github.com/AgNO3/jcifs-ng/blob/master/src/main/java/jcifs/internal/smb2/tree/Smb2TreeConnectResponse.java#L201
>>>>>
>>>>>  mgmt/ksmbd_ida.c | 9 ++-------
>>>>>  1 file changed, 2 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/mgmt/ksmbd_ida.c b/mgmt/ksmbd_ida.c
>>>>> index 7eb6476..34e0d2e 100644
>>>>> --- a/mgmt/ksmbd_ida.c
>>>>> +++ b/mgmt/ksmbd_ida.c
>>>>> @@ -13,19 +13,14 @@ static inline int __acquire_id(struct ida *ida, int
>>>>> from, int to)
>>>>>  #ifdef CONFIG_SMB_INSECURE_SERVER
>>>>>  int ksmbd_acquire_smb1_tid(struct ida *ida)
>>>>>  {
>>>>> -	return __acquire_id(ida, 0, 0xFFFF);
>>>>> +	return __acquire_id(ida, 1, 0xFFFF);
>>>>>  }
>>>>>  #endif
>>>>>
>>>>>  int ksmbd_acquire_smb2_tid(struct ida *ida)
>>>>>  {
>>>>> -	int id;
>>>>> +	return  __acquire_id(ida, 1, 0);
>>>>
>>>> I think that should be __acquire_id(ida, 1, 0xFFFFFFFF) (or a lower
>>>> constraint)
>>>>
>>>> 0xFFFFFFFF is used for compound requests to inherit the tree id from the
>>>> previous request.
>>> Where is it defined in the specification ? As I know,
>>> SMB2_FLAGS_RELATED_OPERATIONS flags in smb header indicate inherit
>>> tree id in previous request.
>>
>> [MS-SMB2] 3.2.4.1.4 Sending Compounded Requests
>>
>> ...
>>
>>   The client MUST construct the subsequent request as it would do normally.
>> For any subsequent
>>   requests the client MUST set SMB2_FLAGS_RELATED_OPERATIONS in the Flags
>> field of the SMB2
>>   header to indicate that it is using the SessionId, TreeId, and FileId
>> supplied in the previous
>>   request (or generated by the server in processing that request). For an
>> operation compounded
>>   with an SMB2 CREATE request, the FileId field SHOULD be set to {
>> 0xFFFFFFFFFFFFFFFF,
>>   0xFFFFFFFFFFFFFFFF }.
>>
>> This only explicitly talks about FileId and I'm not any client would do
>> that, but in theory it should be possible to
>> compound, the 2nd session setup request (of an anonymous authentication)
>> with a tree connect request
>> and an open.
>>
>> Which means it's the safest behavior for a server to avoid 0 and all F as
>> valid id,
>> there're still enough ids to use....
>>
>> It also makes sure that we don't end up with very confusing network
>> captures.
> Okay, I have checked cifs client code like the following.
> 
>         if (request_type & CHAINED_REQUEST) {
>                 if (!(request_type & END_OF_CHAIN)) {
>                         /* next 8-byte aligned request */
>                         *total_len = DIV_ROUND_UP(*total_len, 8) * 8;
>                         shdr->NextCommand = cpu_to_le32(*total_len);
>                 } else /* END_OF_CHAIN */
>                         shdr->NextCommand = 0;
>                 if (request_type & RELATED_REQUEST) {
>                         shdr->Flags |= SMB2_FLAGS_RELATED_OPERATIONS;
>                         /*
>                          * Related requests use info from previous read request
>                          * in chain.
>                          */
>                         shdr->SessionId = 0xFFFFFFFF;
>                         shdr->TreeId = 0xFFFFFFFF;
>                         req->PersistentFileId = 0xFFFFFFFF;
>                         req->VolatileFileId = 0xFFFFFFFF;
>                 }

Which seems actually wrong and should be 0xFFFFFFFFFFFFFFFF for all but TreeId...

metze



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]