kernel-5.18.8 breaks cifs mounts

kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Hi list,

it appears that kernel 5.18.8 breaks cifs mounts on my machine. With 
5.18.7, everything works fine. With 5.18.8, I am getting:

$ sudo mount /mnt/openmediavault/
mount error(22): Invalid argument
Refer to the mount.cifs(8) manual page (e.g. man mount.cifs) and kernel 
log messages (dmesg)

The relevant /etc/fstab line is:

//odroidxu4.local/julian /mnt/openmediavault    cifs 
credentials=/home/julas/.credentials,uid=julas,gid=julas,vers=3.1.1,nobrl,_netdev,auto 
0 0

Is this a known problem?

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[ronnie sahlberg]

On Fri, 1 Jul 2022 at 03:00, Julian Sikorski <belegdol@gmail.com> wrote:
>
> Hi list,
>
> it appears that kernel 5.18.8 breaks cifs mounts on my machine. With
> 5.18.7, everything works fine. With 5.18.8, I am getting:
>
> $ sudo mount /mnt/openmediavault/
> mount error(22): Invalid argument
> Refer to the mount.cifs(8) manual page (e.g. man mount.cifs) and kernel
> log messages (dmesg)
>
> The relevant /etc/fstab line is:
>
> //odroidxu4.local/julian /mnt/openmediavault    cifs
> credentials=/home/julas/.credentials,uid=julas,gid=julas,vers=3.1.1,nobrl,_netdev,auto
> 0 0
>
> Is this a known problem?

What is the output in dmesg ?

>
> Best regards,
> Julian

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Am 30.06.22 um 19:02 schrieb ronnie sahlberg:
> On Fri, 1 Jul 2022 at 03:00, Julian Sikorski <belegdol@gmail.com> wrote:
>>
>> Hi list,
>>
>> it appears that kernel 5.18.8 breaks cifs mounts on my machine. With
>> 5.18.7, everything works fine. With 5.18.8, I am getting:
>>
>> $ sudo mount /mnt/openmediavault/
>> mount error(22): Invalid argument
>> Refer to the mount.cifs(8) manual page (e.g. man mount.cifs) and kernel
>> log messages (dmesg)
>>
>> The relevant /etc/fstab line is:
>>
>> //odroidxu4.local/julian /mnt/openmediavault    cifs
>> credentials=/home/julas/.credentials,uid=julas,gid=julas,vers=3.1.1,nobrl,_netdev,auto
>> 0 0
>>
>> Is this a known problem?
> 
> What is the output in dmesg ?
> 
Not much is there:

Jun 30 18:19:34 snowball3 kernel: CIFS: Attempting to mount 
\\odroidxu4.local\julian

Jun 30 18:19:34 snowball3 kernel: CIFS: VFS: cifs_mount failed w/return 
code = -22


I tried reverting 16d5d91 as it was the only commit referencing smb, but 
it did not help unfortunately.

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[Paulo Alcantara]

Julian Sikorski <belegdol@gmail.com> writes:

> Not much is there:
>
> Jun 30 18:19:34 snowball3 kernel: CIFS: Attempting to mount 
> \\odroidxu4.local\julian
>
> Jun 30 18:19:34 snowball3 kernel: CIFS: VFS: cifs_mount failed w/return 
> code = -22
>
>
> I tried reverting 16d5d91 as it was the only commit referencing smb, but 
> it did not help unfortunately.

Could you please run

	echo 'module cifs +p' > /sys/kernel/debug/dynamic_debug/control
	echo 'file fs/cifs/* +p' > /sys/kernel/debug/dynamic_debug/control
	echo 1 > /proc/fs/cifs/cifsFYI
	echo 1 > /sys/module/dns_resolver/parameters/debug
	dmesg --clear
	tcpdump -s 0 -w trace.pcap port 445 & pid=$!
	mount ...
	kill $pid
	dmesg > trace.log

and then send trace.log and trace.pcap.

What SMB server and version?

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

[-- Attachment #1: Type: text/plain, Size: 1011 bytes --]

Am 30.06.22 um 19:55 schrieb Paulo Alcantara:
> Julian Sikorski <belegdol@gmail.com> writes:
> 
>> Not much is there:
>>
>> Jun 30 18:19:34 snowball3 kernel: CIFS: Attempting to mount
>> \\odroidxu4.local\julian
>>
>> Jun 30 18:19:34 snowball3 kernel: CIFS: VFS: cifs_mount failed w/return
>> code = -22
>>
>>
>> I tried reverting 16d5d91 as it was the only commit referencing smb, but
>> it did not help unfortunately.
> 
> Could you please run
> 
> 	echo 'module cifs +p' > /sys/kernel/debug/dynamic_debug/control
> 	echo 'file fs/cifs/* +p' > /sys/kernel/debug/dynamic_debug/control
> 	echo 1 > /proc/fs/cifs/cifsFYI
> 	echo 1 > /sys/module/dns_resolver/parameters/debug
> 	dmesg --clear
> 	tcpdump -s 0 -w trace.pcap port 445 & pid=$!
> 	mount ...
> 	kill $pid
> 	dmesg > trace.log
> 
> and then send trace.log and trace.pcap.

Attached.

> 
> What SMB server and version?

Openmediavault 6.0.30 running on top of armbian bullseye. The samba 
package version is 4.13.13+dfsg-1~deb11u3.

Best regards,
Julian

[-- Attachment #2: trace.log --]
[-- Type: text/x-log, Size: 2868 bytes --]

[  862.910364] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'source'
[  862.910381] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'ip'
[  862.910387] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'unc'
[  862.910391] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'vers'
[  862.910395] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'nostrictsync'
[  862.910400] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'uid'
[  862.910405] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'gid'
[  862.910408] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'user'
[  862.910411] CIFS: fs/cifs/fs_context.c: CIFS: parsing cifs mount option 'pass'
[  862.910416] CIFS: fs/cifs/cifsfs.c: Devname: \\odroidxu4.local\julian flags: 0
[  862.910422] CIFS: fs/cifs/connect.c: Username: julas
[  862.910425] CIFS: fs/cifs/connect.c: file mode: 0755  dir mode: 0755
[  862.910431] CIFS: fs/cifs/connect.c: VFS: in mount_get_conns as Xid: 4 with uid: 0
[  862.910435] CIFS: fs/cifs/connect.c: UNC: \\odroidxu4.local\julian
[  862.910441] CIFS: fs/cifs/connect.c: generic_ip_connect: connecting to 192.168.0.220:445
[  862.910462] CIFS: fs/cifs/connect.c: Socket created
[  862.910464] CIFS: fs/cifs/connect.c: sndbuf 16384 rcvbuf 131072 rcvtimeo 0x1b58
[  862.916973] CIFS: fs/cifs/connect.c: cifs_get_tcp_session: next dns resolution scheduled for 600 seconds in the future
[  862.916977] CIFS: fs/cifs/connect.c: Demultiplex PID: 4568
[  862.916982] CIFS: fs/cifs/connect.c: VFS: in cifs_get_smb_ses as Xid: 5 with uid: 0
[  862.916986] CIFS: fs/cifs/connect.c: Existing smb sess not found
[  862.916994] CIFS: fs/cifs/smb2pdu.c: Negotiate protocol
[  862.917003] CIFS: fs/cifs/transport.c: wait_for_free_credits: remove 1 credits total=0
[  862.917023] CIFS: fs/cifs/transport.c: Sending smb: smb_len=236
[  862.964239] CIFS: fs/cifs/connect.c: RFC1002 header 0x49
[  862.964269] CIFS: fs/cifs/smb2misc.c: SMB2 data length 0 offset 0
[  862.964273] CIFS: fs/cifs/smb2misc.c: SMB2 len 73
[  862.964279] CIFS: fs/cifs/smb2ops.c: smb2_add_credits: added 1 credits total=1
[  862.964379] CIFS: fs/cifs/transport.c: cifs_sync_mid_result: cmd=0 mid=0 state=4
[  862.964401] CIFS: fs/cifs/smb2maperror.c: Mapping SMB2 status code 0xc000000d to POSIX err -22
[  862.964416] CIFS: fs/cifs/misc.c: Null buffer passed to cifs_small_buf_release
[  862.964431] CIFS: fs/cifs/connect.c: VFS: leaving cifs_get_smb_ses (xid = 5) rc = -22
[  862.964442] CIFS: fs/cifs/dfs_cache.c: cache_refresh_path: search path: \odroidxu4.local\julian
[  862.964451] CIFS: fs/cifs/dfs_cache.c: get_dfs_referral: get an DFS referral for \odroidxu4.local\julian
[  862.964478] CIFS: fs/cifs/connect.c: VFS: leaving mount_put_conns (xid = 4) rc = 0
[  862.964482] CIFS: VFS: cifs_mount failed w/return code = -22

[-- Attachment #3: trace.pcap --]
[-- Type: application/vnd.tcpdump.pcap, Size: 1173 bytes --]

Re: kernel-5.18.8 breaks cifs mounts

[Paulo Alcantara]

Julian Sikorski <belegdol@gmail.com> writes:

> Am 30.06.22 um 19:55 schrieb Paulo Alcantara:
>> Julian Sikorski <belegdol@gmail.com> writes:
>> 
>>> Not much is there:
>>>
>>> Jun 30 18:19:34 snowball3 kernel: CIFS: Attempting to mount
>>> \\odroidxu4.local\julian
>>>
>>> Jun 30 18:19:34 snowball3 kernel: CIFS: VFS: cifs_mount failed w/return
>>> code = -22
>>>
>>>
>>> I tried reverting 16d5d91 as it was the only commit referencing smb, but
>>> it did not help unfortunately.
>> 
>> Could you please run
>> 
>> 	echo 'module cifs +p' > /sys/kernel/debug/dynamic_debug/control
>> 	echo 'file fs/cifs/* +p' > /sys/kernel/debug/dynamic_debug/control
>> 	echo 1 > /proc/fs/cifs/cifsFYI
>> 	echo 1 > /sys/module/dns_resolver/parameters/debug
>> 	dmesg --clear
>> 	tcpdump -s 0 -w trace.pcap port 445 & pid=$!
>> 	mount ...
>> 	kill $pid
>> 	dmesg > trace.log
>> 
>> and then send trace.log and trace.pcap.
>
> Attached.
>
>> 
>> What SMB server and version?
>
> Openmediavault 6.0.30 running on top of armbian bullseye. The samba 
> package version is 4.13.13+dfsg-1~deb11u3.

Thanks!

So, it is failing very early during the negproto request:

	SMB2 302 Negotiate Protocol Request
	SMB2 143 Negotiate Protocol Response, Error: STATUS_INVALID_PARAMETER

which seems related to these commits in v5.18.8

	16d5d9100927 smb3: use netname when available on secondary channels
	ca83f50b43a0 smb3: fix empty netname context on secondary channels

Have you tried reverting both in v5.18.8 to see if it works?

I built v5.18.8 but couldn't reproduce it against w22 server, though.

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Am 30.06.22 um 20:28 schrieb Paulo Alcantara:
> Julian Sikorski <belegdol@gmail.com> writes:
> 
>> Am 30.06.22 um 19:55 schrieb Paulo Alcantara:
>>> Julian Sikorski <belegdol@gmail.com> writes:
>>>
>>>> Not much is there:
>>>>
>>>> Jun 30 18:19:34 snowball3 kernel: CIFS: Attempting to mount
>>>> \\odroidxu4.local\julian
>>>>
>>>> Jun 30 18:19:34 snowball3 kernel: CIFS: VFS: cifs_mount failed w/return
>>>> code = -22
>>>>
>>>>
>>>> I tried reverting 16d5d91 as it was the only commit referencing smb, but
>>>> it did not help unfortunately.
>>>
>>> Could you please run
>>>
>>> 	echo 'module cifs +p' > /sys/kernel/debug/dynamic_debug/control
>>> 	echo 'file fs/cifs/* +p' > /sys/kernel/debug/dynamic_debug/control
>>> 	echo 1 > /proc/fs/cifs/cifsFYI
>>> 	echo 1 > /sys/module/dns_resolver/parameters/debug
>>> 	dmesg --clear
>>> 	tcpdump -s 0 -w trace.pcap port 445 & pid=$!
>>> 	mount ...
>>> 	kill $pid
>>> 	dmesg > trace.log
>>>
>>> and then send trace.log and trace.pcap.
>>
>> Attached.
>>
>>>
>>> What SMB server and version?
>>
>> Openmediavault 6.0.30 running on top of armbian bullseye. The samba
>> package version is 4.13.13+dfsg-1~deb11u3.
> 
> Thanks!
> 
> So, it is failing very early during the negproto request:
> 
> 	SMB2 302 Negotiate Protocol Request
> 	SMB2 143 Negotiate Protocol Response, Error: STATUS_INVALID_PARAMETER
> 
> which seems related to these commits in v5.18.8
> 
> 	16d5d9100927 smb3: use netname when available on secondary channels
> 	ca83f50b43a0 smb3: fix empty netname context on secondary channels
> 
> Have you tried reverting both in v5.18.8 to see if it works?
> 
> I built v5.18.8 but couldn't reproduce it against w22 server, though.
Good catch! I only noticed

16d5d9100927 smb3: use netname when available on secondary channels

initally, and reverting it did not help. Reverting

ca83f50b43a0 smb3: fix empty netname context on secondary channels

makes the mounts work again.

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[Enzo Matsumiya]

On 06/30, Julian Sikorski wrote:
>>>Openmediavault 6.0.30 running on top of armbian bullseye. The samba
>>>package version is 4.13.13+dfsg-1~deb11u3.

[ ... ]

>16d5d9100927 smb3: use netname when available on secondary channels
>
>initally, and reverting it did not help. Reverting
>
>ca83f50b43a0 smb3: fix empty netname context on secondary channels
>
>makes the mounts work again.

It's possible that server is not discarding/ignoring SMB2_NETNAME_NEGOTIATE_CONTEXT_ID
negotiate context as it should, and rather treating it as an error. 

I took a quick look in SLE15-SP3 samba 4.15 code and I didn't spot any
obvious mishandling there (it seems it's not supported yet). Also I
couldn't reproduce the issue with that server.

Maybe it's possible it's a vendor modification?


Cheers,

Enzo

Re: kernel-5.18.8 breaks cifs mounts

[Shyam Prasad N]

On Fri, Jul 1, 2022 at 2:07 AM Enzo Matsumiya <ematsumiya@suse.de> wrote:
>
> On 06/30, Julian Sikorski wrote:
> >>>Openmediavault 6.0.30 running on top of armbian bullseye. The samba
> >>>package version is 4.13.13+dfsg-1~deb11u3.
>
> [ ... ]
>
> >16d5d9100927 smb3: use netname when available on secondary channels
> >
> >initally, and reverting it did not help. Reverting
> >
> >ca83f50b43a0 smb3: fix empty netname context on secondary channels
> >
> >makes the mounts work again.
>
> It's possible that server is not discarding/ignoring SMB2_NETNAME_NEGOTIATE_CONTEXT_ID
> negotiate context as it should, and rather treating it as an error.
>
> I took a quick look in SLE15-SP3 samba 4.15 code and I didn't spot any
> obvious mishandling there (it seems it's not supported yet). Also I
> couldn't reproduce the issue with that server.
>
> Maybe it's possible it's a vendor modification?
>
>
> Cheers,
>
> Enzo

I tried mounting a samba share, and it seems to work fine.

From the network capture attached, I see that the netname is not
empty, and seems fine.
Negotiate Context: SMB2_NETNAME_NEGOTIATE_CONTEXT_ID
    Type: SMB2_NETNAME_NEGOTIATE_CONTEXT_ID (0x0005)
    DataLength: 30
    Reserved: 00000000
    Netname: odroidxu4.local

Julian,
Will you be able to capture network traces with the two changes
reverted? i.e. when it works for you?

-- 
Regards,
Shyam

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
> Am 03.07.22 um 07:01 schrieb Steve French:
>> I lean toward thinking that this is a Samba bug (although I don't see
>> it on my local system - it works to samba for me, although I was
>> trying against a slightly different version, Samba 4.15.5-Ubuntu).
>>
>> Looking at the traces in more detail they look the same (failing vs.
>> working) other than the order of the negotiate context, which fails
>> with POSIX as the 3rd context, and netname as the 4th, but works with
>> the order reversed (although same contexts, and same overall length)
>> ie with POSIX context as the fourth one and netname context as the
>> third one.
>>
>> The failing server code in Samba is in
>> smbd_smb2_request_process_negprot but I don't see changes to it
>> recently around this error.
>>
>> Does this fail to anyone else's Samba version?
>>
>> This is probably a Samba server bug but ... seems odd since it doesn't
>> fail to Samba for me.
>>
>> Jeremy/Metze,
>> Does this look familiar?
> 
> Maybe this one:
> 
> https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a 
> 
> 
> that went only into 4.15 and higher.
> 
> metze

Nice catch, I can confirm that adding this patch to debian samba 
2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we 
get this patch into debian?

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

[-- Attachment #1: Type: text/plain, Size: 1578 bytes --]

Am 04.07.22 um 10:12 schrieb Shyam Prasad N:
> On Fri, Jul 1, 2022 at 2:07 AM Enzo Matsumiya <ematsumiya@suse.de> wrote:
>>
>> On 06/30, Julian Sikorski wrote:
>>>>> Openmediavault 6.0.30 running on top of armbian bullseye. The samba
>>>>> package version is 4.13.13+dfsg-1~deb11u3.
>>
>> [ ... ]
>>
>>> 16d5d9100927 smb3: use netname when available on secondary channels
>>>
>>> initally, and reverting it did not help. Reverting
>>>
>>> ca83f50b43a0 smb3: fix empty netname context on secondary channels
>>>
>>> makes the mounts work again.
>>
>> It's possible that server is not discarding/ignoring SMB2_NETNAME_NEGOTIATE_CONTEXT_ID
>> negotiate context as it should, and rather treating it as an error.
>>
>> I took a quick look in SLE15-SP3 samba 4.15 code and I didn't spot any
>> obvious mishandling there (it seems it's not supported yet). Also I
>> couldn't reproduce the issue with that server.
>>
>> Maybe it's possible it's a vendor modification?
>>
>>
>> Cheers,
>>
>> Enzo
> 
> I tried mounting a samba share, and it seems to work fine.
> 
>  From the network capture attached, I see that the netname is not
> empty, and seems fine.
> Negotiate Context: SMB2_NETNAME_NEGOTIATE_CONTEXT_ID
>      Type: SMB2_NETNAME_NEGOTIATE_CONTEXT_ID (0x0005)
>      DataLength: 30
>      Reserved: 00000000
>      Netname: odroidxu4.local
> 
> Julian,
> Will you be able to capture network traces with the two changes
> reverted? i.e. when it works for you?
> 
Looks like some messages were not CCed to the list, apologies. Please 
find the traces attached.

Best regards,
Julian

[-- Attachment #2: trace_reverted.pcap --]
[-- Type: application/vnd.tcpdump.pcap, Size: 17839 bytes --]

Re: kernel-5.18.8 breaks cifs mounts

[Shyam Prasad N]

On Mon, Jul 4, 2022 at 10:03 PM Julian Sikorski <belegdol@gmail.com> wrote:
>
>
>
> Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
> > Am 03.07.22 um 07:01 schrieb Steve French:
> >> I lean toward thinking that this is a Samba bug (although I don't see
> >> it on my local system - it works to samba for me, although I was
> >> trying against a slightly different version, Samba 4.15.5-Ubuntu).
> >>
> >> Looking at the traces in more detail they look the same (failing vs.
> >> working) other than the order of the negotiate context, which fails
> >> with POSIX as the 3rd context, and netname as the 4th, but works with
> >> the order reversed (although same contexts, and same overall length)
> >> ie with POSIX context as the fourth one and netname context as the
> >> third one.
> >>
> >> The failing server code in Samba is in
> >> smbd_smb2_request_process_negprot but I don't see changes to it
> >> recently around this error.
> >>
> >> Does this fail to anyone else's Samba version?
> >>
> >> This is probably a Samba server bug but ... seems odd since it doesn't
> >> fail to Samba for me.
> >>
> >> Jeremy/Metze,
> >> Does this look familiar?
> >
> > Maybe this one:
> >
> > https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a
> >
> >
> > that went only into 4.15 and higher.
> >
> > metze
>
> Nice catch, I can confirm that adding this patch to debian samba
> 2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we
> get this patch into debian?
>
> Best regards,
> Julian

Hi Metze,
I went through the above patch, and it looks like an issue with
parsing garbage at the end of the buffer, rather than negotiate count.
I'm not sure how the netname negotiate context patches above are being
affected by this samba server patch. The only difference from the
client side for single channel is that the netname context appears as
the 4th element in the list of 4, rather than the 3rd element in the
list of 4. Do you have a possible explanation?

Julian,
Thanks for the repro attempt. So I assume that with the latest samba
server, things work as expected without reverting any changes in
5.18.8 kernel, correct?

-- 
Regards,
Shyam

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Am 05.07.22 um 07:28 schrieb Shyam Prasad N:
> On Mon, Jul 4, 2022 at 10:03 PM Julian Sikorski <belegdol@gmail.com> wrote:
>>
>>
>>
>> Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
>>> Am 03.07.22 um 07:01 schrieb Steve French:
>>>> I lean toward thinking that this is a Samba bug (although I don't see
>>>> it on my local system - it works to samba for me, although I was
>>>> trying against a slightly different version, Samba 4.15.5-Ubuntu).
>>>>
>>>> Looking at the traces in more detail they look the same (failing vs.
>>>> working) other than the order of the negotiate context, which fails
>>>> with POSIX as the 3rd context, and netname as the 4th, but works with
>>>> the order reversed (although same contexts, and same overall length)
>>>> ie with POSIX context as the fourth one and netname context as the
>>>> third one.
>>>>
>>>> The failing server code in Samba is in
>>>> smbd_smb2_request_process_negprot but I don't see changes to it
>>>> recently around this error.
>>>>
>>>> Does this fail to anyone else's Samba version?
>>>>
>>>> This is probably a Samba server bug but ... seems odd since it doesn't
>>>> fail to Samba for me.
>>>>
>>>> Jeremy/Metze,
>>>> Does this look familiar?
>>>
>>> Maybe this one:
>>>
>>> https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a
>>>
>>>
>>> that went only into 4.15 and higher.
>>>
>>> metze
>>
>> Nice catch, I can confirm that adding this patch to debian samba
>> 2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we
>> get this patch into debian?
>>
>> Best regards,
>> Julian
> 
> Hi Metze,
> I went through the above patch, and it looks like an issue with
> parsing garbage at the end of the buffer, rather than negotiate count.
> I'm not sure how the netname negotiate context patches above are being
> affected by this samba server patch. The only difference from the
> client side for single channel is that the netname context appears as
> the 4th element in the list of 4, rather than the 3rd element in the
> list of 4. Do you have a possible explanation?
> 
> Julian,
> Thanks for the repro attempt. So I assume that with the latest samba
> server, things work as expected without reverting any changes in
> 5.18.8 kernel, correct?
> 
I have no way of testing a newer samba server unfortunately, but several 
people in this email discussion have mentioned that newer servers work 
for them:
- Paulo with W22
- Enzo and Steve with 4.15
- Steve with 4.16 without 
https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[Julian Sikorski]

Am 04.07.22 um 18:29 schrieb Julian Sikorski:
> 
> 
> Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
>> Am 03.07.22 um 07:01 schrieb Steve French:
>>> I lean toward thinking that this is a Samba bug (although I don't see
>>> it on my local system - it works to samba for me, although I was
>>> trying against a slightly different version, Samba 4.15.5-Ubuntu).
>>>
>>> Looking at the traces in more detail they look the same (failing vs.
>>> working) other than the order of the negotiate context, which fails
>>> with POSIX as the 3rd context, and netname as the 4th, but works with
>>> the order reversed (although same contexts, and same overall length)
>>> ie with POSIX context as the fourth one and netname context as the
>>> third one.
>>>
>>> The failing server code in Samba is in
>>> smbd_smb2_request_process_negprot but I don't see changes to it
>>> recently around this error.
>>>
>>> Does this fail to anyone else's Samba version?
>>>
>>> This is probably a Samba server bug but ... seems odd since it doesn't
>>> fail to Samba for me.
>>>
>>> Jeremy/Metze,
>>> Does this look familiar?
>>
>> Maybe this one:
>>
>> https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a 
>>
>>
>> that went only into 4.15 and higher.
>>
>> metze
> 
> Nice catch, I can confirm that adding this patch to debian samba 
> 2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we 
> get this patch into debian?
> 
> Best regards,
> Julian
I have now filed a bug against debian samba package:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1014453

Best regards,
Julian

Re: kernel-5.18.8 breaks cifs mounts

[Steve French]

I was able to reproduce it to older Samba server (4.12.5) and could
workaround the Samba server bug by using mount option "compress" on
the client (which won't do anything different since it is not
supported but  interestingly it avoids the problem by adding another
context at the end).

On Wed, Jul 6, 2022 at 6:18 AM Julian Sikorski <belegdol@gmail.com> wrote:
>
> Am 04.07.22 um 18:29 schrieb Julian Sikorski:
> >
> >
> > Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
> >> Am 03.07.22 um 07:01 schrieb Steve French:
> >>> I lean toward thinking that this is a Samba bug (although I don't see
> >>> it on my local system - it works to samba for me, although I was
> >>> trying against a slightly different version, Samba 4.15.5-Ubuntu).
> >>>
> >>> Looking at the traces in more detail they look the same (failing vs.
> >>> working) other than the order of the negotiate context, which fails
> >>> with POSIX as the 3rd context, and netname as the 4th, but works with
> >>> the order reversed (although same contexts, and same overall length)
> >>> ie with POSIX context as the fourth one and netname context as the
> >>> third one.
> >>>
> >>> The failing server code in Samba is in
> >>> smbd_smb2_request_process_negprot but I don't see changes to it
> >>> recently around this error.
> >>>
> >>> Does this fail to anyone else's Samba version?
> >>>
> >>> This is probably a Samba server bug but ... seems odd since it doesn't
> >>> fail to Samba for me.
> >>>
> >>> Jeremy/Metze,
> >>> Does this look familiar?
> >>
> >> Maybe this one:
> >>
> >> https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a
> >>
> >>
> >> that went only into 4.15 and higher.
> >>
> >> metze
> >
> > Nice catch, I can confirm that adding this patch to debian samba
> > 2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we
> > get this patch into debian?
> >
> > Best regards,
> > Julian
> I have now filed a bug against debian samba package:
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1014453
>
> Best regards,
> Julian



-- 
Thanks,

Steve

Re: kernel-5.18.8 breaks cifs mounts

[Georg Müller]

Hi,

Am 12.07.22 um 06:36 schrieb Steve French:
> I was able to reproduce it to older Samba server (4.12.5) and could
> workaround the Samba server bug by using mount option "compress" on
> the client (which won't do anything different since it is not
> supported but  interestingly it avoids the problem by adding another
> context at the end).


The general problem might be solved by just moving the now optional hostname
context in fs/cifs/smb2pdu.c back between build_encrypt_ctxt and build_posix_ctxt.

I encounter the issue with a Synology Diskstation, and there might be
much more people running into the problem when this kernel becomes more
common...


> On Wed, Jul 6, 2022 at 6:18 AM Julian Sikorski <belegdol@gmail.com> wrote:
>>
>> Am 04.07.22 um 18:29 schrieb Julian Sikorski:
>>>
>>>
>>> Am 03.07.22 um 19:51 schrieb Stefan Metzmacher:
>>>> Am 03.07.22 um 07:01 schrieb Steve French:
>>>>> I lean toward thinking that this is a Samba bug (although I don't see
>>>>> it on my local system - it works to samba for me, although I was
>>>>> trying against a slightly different version, Samba 4.15.5-Ubuntu).
>>>>>
>>>>> Looking at the traces in more detail they look the same (failing vs.
>>>>> working) other than the order of the negotiate context, which fails
>>>>> with POSIX as the 3rd context, and netname as the 4th, but works with
>>>>> the order reversed (although same contexts, and same overall length)
>>>>> ie with POSIX context as the fourth one and netname context as the
>>>>> third one.
>>>>>
>>>>> The failing server code in Samba is in
>>>>> smbd_smb2_request_process_negprot but I don't see changes to it
>>>>> recently around this error.
>>>>>
>>>>> Does this fail to anyone else's Samba version?
>>>>>
>>>>> This is probably a Samba server bug but ... seems odd since it doesn't
>>>>> fail to Samba for me.
>>>>>
>>>>> Jeremy/Metze,
>>>>> Does this look familiar?
>>>>
>>>> Maybe this one:
>>>>
>>>> https://git.samba.org/?p=samba.git;a=commitdiff;h=147dd9d58a429695a3b6c6e45c8b0eaafc67908a
>>>>
>>>>
>>>> that went only into 4.15 and higher.
>>>>
>>>> metze
>>>
>>> Nice catch, I can confirm that adding this patch to debian samba
>>> 2:4.13.13+dfsg-1~deb11u3 package makes the mounts work again. How do we
>>> get this patch into debian?
>>>
>>> Best regards,
>>> Julian
>> I have now filed a bug against debian samba package:
>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1014453
>>
>> Best regards,
>> Julian
>
>
>