Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org --- fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct cifs_ses *ses, const char *tree, iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
- /* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */ + /* + * 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 + * unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1 + */ if ((ses->server->dialect == SMB311_PROT_ID) && - !smb3_encryption_required(tcon)) + !smb3_encryption_required(tcon) && + !(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL))) req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED;
memset(&rqst, 0, sizeof(struct smb_rqst));
Ronnie Sahlberg lsahlber@redhat.com writes:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2.
We could add a
Fixes: 6188f28bf608 ("Tree connect for SMB3.1.1 must be signed for non-encrypted shares")
tag in the commit message.
- /* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */
- /*
* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL)))memset(&rqst, 0, sizeof(struct smb_rqst));
I suspect there might more (and/or better) places to check. If you grep for SMB2_FLAGS_SIGNED there are a couple of other places that should be considered:
smb2_hdr_assemble() { ... if (tcon->ses && tcon->ses->server && tcon->ses->server->sign && !smb3_encryption_required(tcon)) shdr->Flags |= SMB2_FLAGS_SIGNED; ... }
cifs_get_smb_ses() { ... ses->sectype = volume_info->sectype; ses->sign = volume_info->sign;
mutex_lock(&ses->session_mutex); rc = cifs_negotiate_protocol(xid, ses); if (!rc) rc = cifs_setup_session(xid, ses, volume_info->local_nls); mutex_unlock(&ses->session_mutex); if (rc) goto get_ses_fail; ... }
After negprot and before sess setup we already know the protocol version so I guess there could be some patch there as well.
See also SMB2_logoff(), SMB2_ioctl_init(), cifs_enable_signing()
Cheers,
I added the commit tag and cleaned up the related checkpatch warning.
I don't know of any other locations where we have to make a similar change as this is the only place where we (on the client) forced signing to be on for a request when the user (or server) didn't request it (due to the requirement in MS-SMB2 that tree connect be signed). In the other places you noted, the client or server would have forced signing on ("sign" mount option or "server signing = mandatory" for example).
On Thu, Mar 21, 2019 at 6:46 AM Aurélien Aptel aaptel@suse.com wrote:
Ronnie Sahlberg lsahlber@redhat.com writes:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2.
We could add a
Fixes: 6188f28bf608 ("Tree connect for SMB3.1.1 must be signed for non-encrypted shares")tag in the commit message.
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL))) req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));I suspect there might more (and/or better) places to check. If you grep for SMB2_FLAGS_SIGNED there are a couple of other places that should be considered:
smb2_hdr_assemble() { ... if (tcon->ses && tcon->ses->server && tcon->ses->server->sign && !smb3_encryption_required(tcon)) shdr->Flags |= SMB2_FLAGS_SIGNED; ... }
cifs_get_smb_ses() { ... ses->sectype = volume_info->sectype; ses->sign = volume_info->sign;
mutex_lock(&ses->session_mutex); rc = cifs_negotiate_protocol(xid, ses); if (!rc) rc = cifs_setup_session(xid, ses, volume_info->local_nls); mutex_unlock(&ses->session_mutex); if (rc) goto get_ses_fail;... }
After negprot and before sess setup we already know the protocol version so I guess there could be some patch there as well.
See also SMB2_logoff(), SMB2_ioctl_init(), cifs_enable_signing()
Cheers,
Aurélien Aptel / SUSE Labs Samba Team GPG: 1839 CB5F 9F5B FB9B AA97 8C99 03C8 A49B 521B D5D3 SUSE Linux GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg lsahlber@redhat.com wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct cifs_ses *ses, const char *tree, iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL))) req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel (https://people.canonical.com/~tyhicks/disco-cifs.2/ specifically) but it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
dmesg: http://paste.ubuntu.com/p/JGhCsgBVcb/
server logs (debug level 5, samba 4.10.0): log.: http://paste.ubuntu.com/p/jMDJ8DBfRM/ log.smbd: http://paste.ubuntu.com/p/Z9W5z28BP9/ smb.conf: http://paste.ubuntu.com/p/9HpSyFq8n8/
On Fri, Mar 22, 2019 at 3:13 AM Andreas Hasenack andreas@canonical.com wrote:
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg lsahlber@redhat.com wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct cifs_ses *ses, const char *tree, iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL))) req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel (https://people.canonical.com/~tyhicks/disco-cifs.2/ specifically) but it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
Try with the current version of Steve's for-next branch plus this patch. I could reproduce the failure with 3.11 on for-next but when I added this patch then the mount was successful.
At least that would verify that the current for-net works for you (or not). There may be other things missing in older kernels that broke 3.11 guest mounts, but lets check if for-next works first.
Regards Ronnie Sahlberg
dmesg: http://paste.ubuntu.com/p/JGhCsgBVcb/
server logs (debug level 5, samba 4.10.0): log.: http://paste.ubuntu.com/p/jMDJ8DBfRM/ log.smbd: http://paste.ubuntu.com/p/Z9W5z28BP9/ smb.conf: http://paste.ubuntu.com/p/9HpSyFq8n8/
I tried to Samba with guest mount and noted that Samba server did not set the guest or null flags in the session setup response so we still marked tcon as signed even with guest mount in my quick test (to Samba 4.10)
On Thu, Mar 21, 2019 at 2:54 PM ronnie sahlberg ronniesahlberg@gmail.com wrote:
On Fri, Mar 22, 2019 at 3:13 AM Andreas Hasenack andreas@canonical.com wrote:
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg lsahlber@redhat.com wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct cifs_ses *ses, const char *tree, iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB2 3.2.4.1.1* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags & (SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL))) req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel (https://people.canonical.com/~tyhicks/disco-cifs.2/ specifically) but it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
Try with the current version of Steve's for-next branch plus this patch. I could reproduce the failure with 3.11 on for-next but when I added this patch then the mount was successful.
At least that would verify that the current for-net works for you (or not). There may be other things missing in older kernels that broke 3.11 guest mounts, but lets check if for-next works first.
Regards Ronnie Sahlberg
dmesg: http://paste.ubuntu.com/p/JGhCsgBVcb/
server logs (debug level 5, samba 4.10.0): log.: http://paste.ubuntu.com/p/jMDJ8DBfRM/ log.smbd: http://paste.ubuntu.com/p/Z9W5z28BP9/ smb.conf: http://paste.ubuntu.com/p/9HpSyFq8n8/
-----Original Message----- From: linux-cifs-owner@vger.kernel.org linux-cifs-owner@vger.kernel.org On Behalf Of Steve French Sent: Friday, March 22, 2019 7:30 PM To: ronnie sahlberg ronniesahlberg@gmail.com Cc: Andreas Hasenack andreas@canonical.com; Ronnie Sahlberg lsahlber@redhat.com; linux-cifs linux-cifs@vger.kernel.org; Stable stable@vger.kernel.org Subject: Re: [PATCH] cifs: allow guest mounts to work for smb3.11
I tried to Samba with guest mount and noted that Samba server did not set the guest or null flags in the session setup response so we still marked tcon as signed even with guest mount in my quick test (to Samba 4.10)
This gets to the point of my earlier comment. The flags and dialect etc are all fine to check, but the code is setting the SIGNED bit before it actually signs. Shouldn't it be verifying that it has a valid signing key before it does this?
Another point to make here is that the SESSION_IS_GUEST is entirely the server's choice. Basically, what it means is the server arbitrarily authenticated the user as a guest, regardless of how the client attempted to log in. If it's set, then absolutely the client should forget that it ever attempted to authenticate as a given principal (and not sign, encrypt, etc). However, if it's clear, and the server made the user "guest", then:
a) the server is arguably buggy b) the signing failure is by protocol design.
Tom.
On Thu, Mar 21, 2019 at 2:54 PM ronnie sahlberg ronniesahlberg@gmail.com wrote:
On Fri, Mar 22, 2019 at 3:13 AM Andreas Hasenack
andreas@canonical.com wrote:
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg lsahlber@redhat.com
wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct
cifs_ses *ses, const char *tree,
iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1
* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags &(SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL)))
req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel
(https://nam06.safelinks.protection.outlook.com/?url=https:%2F%2Fpeople.can onical.com%2F~tyhicks%2Fdisco- cifs.2%2F&data=02%7C01%7Cttalpey%40microsoft.com%7C3cc97bf9482 747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0 %7C636888942385347937&sdata=i0kUZ4yinqjgvsDDc0N6LBTPlLt8DeNnt WW1PhS0xVg%3D&reserved=0 specifically) but
it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
Try with the current version of Steve's for-next branch plus this patch. I could reproduce the failure with 3.11 on for-next but when I added this patch then the mount was successful.
At least that would verify that the current for-net works for you (or not). There may be other things missing in older kernels that broke 3.11 guest
mounts,
but lets check if for-next works first.
Regards Ronnie Sahlberg
dmesg:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FJGhCsgBVcb%2F&data=02%7C01%7Cttalpey%40micros oft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d 7cd011db47%7C1%7C0%7C636888942385347937&sdata=r5eMhXks%2F quoNzzhFahCKjMOE22fFlVCluwmRLWpmOc%3D&reserved=0
server logs (debug level 5, samba 4.10.0): log.:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FjMDJ8DBfRM%2F&data=02%7C01%7Cttalpey%40micr osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab 2d7cd011db47%7C1%7C0%7C636888942385347937&sdata=4owqPDbgc FOjQU%2BEt2aB1P77hFHbIuzoxOkdpX5X2eE%3D&reserved=0
log.smbd:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FZ9W5z28BP9%2F&data=02%7C01%7Cttalpey%40micr osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab 2d7cd011db47%7C1%7C0%7C636888942385357938&sdata=66Ng0qrdbt AwVirxqN3uWjEaFrRi1w6XGcX%2BZITmkOM%3D&reserved=0
smb.conf:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2F9HpSyFq8n8%2F&data=02%7C01%7Cttalpey%40micro soft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2 d7cd011db47%7C1%7C0%7C636888942385357938&sdata=U4dqv23dZN KgaDAR2TbEmsrnkR6EnqejdVTPVLpIjKk%3D&reserved=0
-- Thanks,
Steve
The reason we didn't see this before SMB3.1.1 is that in smb3_validate_negotiate we had this check:
if (tcon->ses->user_name == NULL) { cifs_dbg(FYI, "Can't validate negotiate: null user mount\n"); return 0; /* validation requires signing */ }
Sounds like we should add the identical check in the smb3.1.1 tcon check that Ronnie's earlier patch fixed (but only for Windows)
On Fri, Mar 22, 2019 at 7:20 PM Tom Talpey ttalpey@microsoft.com wrote:
-----Original Message----- From: linux-cifs-owner@vger.kernel.org linux-cifs-owner@vger.kernel.org On Behalf Of Steve French Sent: Friday, March 22, 2019 7:30 PM To: ronnie sahlberg ronniesahlberg@gmail.com Cc: Andreas Hasenack andreas@canonical.com; Ronnie Sahlberg lsahlber@redhat.com; linux-cifs linux-cifs@vger.kernel.org; Stable stable@vger.kernel.org Subject: Re: [PATCH] cifs: allow guest mounts to work for smb3.11
I tried to Samba with guest mount and noted that Samba server did not set the guest or null flags in the session setup response so we still marked tcon as signed even with guest mount in my quick test (to Samba 4.10)
This gets to the point of my earlier comment. The flags and dialect etc are all fine to check, but the code is setting the SIGNED bit before it actually signs. Shouldn't it be verifying that it has a valid signing key before it does this?
Another point to make here is that the SESSION_IS_GUEST is entirely the server's choice. Basically, what it means is the server arbitrarily authenticated the user as a guest, regardless of how the client attempted to log in. If it's set, then absolutely the client should forget that it ever attempted to authenticate as a given principal (and not sign, encrypt, etc). However, if it's clear, and the server made the user "guest", then:
a) the server is arguably buggy b) the signing failure is by protocol design.
Tom.
On Thu, Mar 21, 2019 at 2:54 PM ronnie sahlberg ronniesahlberg@gmail.com wrote:
On Fri, Mar 22, 2019 at 3:13 AM Andreas Hasenack
andreas@canonical.com wrote:
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg lsahlber@redhat.com
wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct
cifs_ses *ses, const char *tree,
iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1
* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags &(SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL)))
req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel
(https://nam06.safelinks.protection.outlook.com/?url=https:%2F%2Fpeople.can onical.com%2F~tyhicks%2Fdisco- cifs.2%2F&data=02%7C01%7Cttalpey%40microsoft.com%7C3cc97bf9482 747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0 %7C636888942385347937&sdata=i0kUZ4yinqjgvsDDc0N6LBTPlLt8DeNnt WW1PhS0xVg%3D&reserved=0 specifically) but
it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
Try with the current version of Steve's for-next branch plus this patch. I could reproduce the failure with 3.11 on for-next but when I added this patch then the mount was successful.
At least that would verify that the current for-net works for you (or not). There may be other things missing in older kernels that broke 3.11 guest
mounts,
but lets check if for-next works first.
Regards Ronnie Sahlberg
dmesg:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FJGhCsgBVcb%2F&data=02%7C01%7Cttalpey%40micros oft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d 7cd011db47%7C1%7C0%7C636888942385347937&sdata=r5eMhXks%2F quoNzzhFahCKjMOE22fFlVCluwmRLWpmOc%3D&reserved=0
server logs (debug level 5, samba 4.10.0): log.:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FjMDJ8DBfRM%2F&data=02%7C01%7Cttalpey%40micr osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab 2d7cd011db47%7C1%7C0%7C636888942385347937&sdata=4owqPDbgc FOjQU%2BEt2aB1P77hFHbIuzoxOkdpX5X2eE%3D&reserved=0
log.smbd:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2FZ9W5z28BP9%2F&data=02%7C01%7Cttalpey%40micr osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab 2d7cd011db47%7C1%7C0%7C636888942385357938&sdata=66Ng0qrdbt AwVirxqN3uWjEaFrRi1w6XGcX%2BZITmkOM%3D&reserved=0
smb.conf:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub untu.com%2Fp%2F9HpSyFq8n8%2F&data=02%7C01%7Cttalpey%40micro soft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2 d7cd011db47%7C1%7C0%7C636888942385357938&sdata=U4dqv23dZN KgaDAR2TbEmsrnkR6EnqejdVTPVLpIjKk%3D&reserved=0
-- Thanks,
Steve
-----Original Message----- From: Steve French smfrench@gmail.com Sent: Friday, March 22, 2019 8:42 PM To: Tom Talpey ttalpey@microsoft.com Cc: ronnie sahlberg ronniesahlberg@gmail.com; Andreas Hasenack andreas@canonical.com; Ronnie Sahlberg lsahlber@redhat.com; linux-cifs linux-cifs@vger.kernel.org; Stable stable@vger.kernel.org Subject: Re: [PATCH] cifs: allow guest mounts to work for smb3.11
The reason we didn't see this before SMB3.1.1 is that in smb3_validate_negotiate we had this check:
if (tcon->ses->user_name == NULL) { cifs_dbg(FYI, "Can't validate negotiate: null user mount\n"); return 0; /* validation requires signing */ }Sounds like we should add the identical check in the smb3.1.1 tcon check that Ronnie's earlier patch fixed (but only for Windows)
Technically a NULL username is a _null_ session, not _guest_. The difference is whether it was the client or server which decided to not use a named principal. Sending a blank user by the client leads to a null session, while the server choosing to ignore the client-provided user leads to guest.
Either one should be avoided, of course.
Tom.
On Fri, Mar 22, 2019 at 7:20 PM Tom Talpey ttalpey@microsoft.com wrote:
-----Original Message----- From: linux-cifs-owner@vger.kernel.org linux-cifs-owner@vger.kernel.org
On
Behalf Of Steve French Sent: Friday, March 22, 2019 7:30 PM To: ronnie sahlberg ronniesahlberg@gmail.com Cc: Andreas Hasenack andreas@canonical.com; Ronnie Sahlberg lsahlber@redhat.com; linux-cifs linux-cifs@vger.kernel.org; Stable stable@vger.kernel.org Subject: Re: [PATCH] cifs: allow guest mounts to work for smb3.11
I tried to Samba with guest mount and noted that Samba server did not set the guest or null flags in the session setup response so we still marked tcon as signed even with guest mount in my quick test (to Samba 4.10)
This gets to the point of my earlier comment. The flags and dialect etc are all fine to check, but the code is setting the SIGNED bit before it actually signs. Shouldn't it be verifying that it has a valid signing key before it does this?
Another point to make here is that the SESSION_IS_GUEST is entirely the server's choice. Basically, what it means is the server arbitrarily authenticated the user as a guest, regardless of how the client attempted to log in. If it's set, then absolutely the client should forget that it ever attempted to authenticate as a given principal (and not sign, encrypt, etc). However, if it's clear, and the server made the user "guest", then:
a) the server is arguably buggy b) the signing failure is by protocol design.
Tom.
On Thu, Mar 21, 2019 at 2:54 PM ronnie sahlberg ronniesahlberg@gmail.com wrote:
On Fri, Mar 22, 2019 at 3:13 AM Andreas Hasenack
andreas@canonical.com wrote:
Hello Ronnie,
On Thu, Mar 21, 2019 at 1:59 AM Ronnie Sahlberg
wrote:
Fix Guest/Anonymous sessions so that they work with SMB 3.11.
In git commit 6188f28 tightened the conditions and forced signing for the SMB2-TreeConnect commands as per MS-SMB2. However, this should only apply to normal user sessions and not for Guest/Anonumous sessions.
Signed-off-by: Ronnie Sahlberg lsahlber@redhat.com CC: Stable stable@vger.kernel.org
fs/cifs/smb2pdu.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c index c399e09b76e6..8e4a1da95418 100644 --- a/fs/cifs/smb2pdu.c +++ b/fs/cifs/smb2pdu.c @@ -1628,9 +1628,13 @@ SMB2_tcon(const unsigned int xid, struct
cifs_ses *ses, const char *tree,
iov[1].iov_base = unc_path; iov[1].iov_len = unc_path_len;
/* 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1 */
/** 3.11 tcon req must be signed if not encrypted. See MS-SMB23.2.4.1.1
* unless it is guest or anonymous user. See MS-SMB2 3.2.5.3.1*/ if ((ses->server->dialect == SMB311_PROT_ID) &&
!smb3_encryption_required(tcon))
!smb3_encryption_required(tcon) &&!(ses->session_flags &(SMB2_SESSION_FLAG_IS_GUEST|SMB2_SESSION_FLAG_IS_NULL)))
req->sync_hdr.Flags |= SMB2_FLAGS_SIGNED; memset(&rqst, 0, sizeof(struct smb_rqst));-- 2.13.6
I tried this patch with an ubuntu kernel
(https://nam06.safelinks.protection.outlook.com/?url=https:%2F%2Fpeople.can
onical.com%2F~tyhicks%2Fdisco-
cifs.2%2F&data=02%7C01%7Cttalpey%40microsoft.com%7C3cc97bf9482
747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0
%7C636888942385347937&sdata=i0kUZ4yinqjgvsDDc0N6LBTPlLt8DeNnt
WW1PhS0xVg%3D&reserved=0 specifically) but
it didn't work, I'm still getting failures with smb3.11 and a guest mount.
Maybe I'm missing some other fix, or a more up-to-date kernel? Shall I try with a self-compiled upstream one?
Try with the current version of Steve's for-next branch plus this patch. I could reproduce the failure with 3.11 on for-next but when I added this patch then the mount was successful.
At least that would verify that the current for-net works for you (or not). There may be other things missing in older kernels that broke 3.11 guest
mounts,
but lets check if for-next works first.
Regards Ronnie Sahlberg
dmesg:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub
untu.com%2Fp%2FJGhCsgBVcb%2F&data=02%7C01%7Cttalpey%40micros
oft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2d
7cd011db47%7C1%7C0%7C636888942385347937&sdata=r5eMhXks%2F
quoNzzhFahCKjMOE22fFlVCluwmRLWpmOc%3D&reserved=0
server logs (debug level 5, samba 4.10.0): log.:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub
untu.com%2Fp%2FjMDJ8DBfRM%2F&data=02%7C01%7Cttalpey%40micr
osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab
2d7cd011db47%7C1%7C0%7C636888942385347937&sdata=4owqPDbgc
FOjQU%2BEt2aB1P77hFHbIuzoxOkdpX5X2eE%3D&reserved=0
log.smbd:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub
untu.com%2Fp%2FZ9W5z28BP9%2F&data=02%7C01%7Cttalpey%40micr
osoft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab
2d7cd011db47%7C1%7C0%7C636888942385357938&sdata=66Ng0qrdbt
AwVirxqN3uWjEaFrRi1w6XGcX%2BZITmkOM%3D&reserved=0
smb.conf:
https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fpaste.ub
untu.com%2Fp%2F9HpSyFq8n8%2F&data=02%7C01%7Cttalpey%40micro
soft.com%7C3cc97bf9482747afd16d08d6af1e62ff%7C72f988bf86f141af91ab2
d7cd011db47%7C1%7C0%7C636888942385357938&sdata=U4dqv23dZN
KgaDAR2TbEmsrnkR6EnqejdVTPVLpIjKk%3D&reserved=0
-- Thanks,
Steve
-- Thanks,
Steve
linux-stable-mirror@lists.linaro.org
-
Andreas Hasenack -
Aurélien Aptel -
Ronnie Sahlberg -
ronnie sahlberg -
Steve French -
Tom Talpey