On Sat, May 8, 2021 at 8:29 AM Tom Talpey <tom@xxxxxxxxxx> wrote:
>
> On 5/7/2021 9:13 PM, Steve French wrote:
> > 1) we were not setting CAP_MULTICHANNEL on negotiate request
>
> > diff --git a/fs/cifs/smb2pdu.c b/fs/cifs/smb2pdu.c
> > index e36c2a867783..a8bf43184773 100644
> > --- a/fs/cifs/smb2pdu.c
> > +++ b/fs/cifs/smb2pdu.c
> > @@ -841,6 +841,8 @@ SMB2_negotiate(const unsigned int xid, struct cifs_ses *ses)
> > req->SecurityMode = 0;
> >
> > req->Capabilities = cpu_to_le32(server->vals->req_capabilities);
> > + if (ses->chan_max > 1)
> > + req->Capabilities |= cpu_to_le32(SMB2_GLOBAL_CAP_MULTI_CHANNEL);
> >
> > /* ClientGUID must be zero for SMB2.02 dialect */
> > if (server->vals->protocol_id == SMB20_PROT_ID)
> > @@ -1032,6 +1034,9 @@ int smb3_validate_negotiate(const unsigned int xid, struct cifs_tcon *tcon)
> >
> > pneg_inbuf->Capabilities =
> > cpu_to_le32(server->vals->req_capabilities);
> > + if (tcon->ses->chan_max > 1)
> > + pneg_inbuf->Capabilities |= cpu_to_le32(SMB2_GLOBAL_CAP_MULTI_CHANNEL);
> > +
>
> This doesn't look quite right, and it can lead to failed negotiate by
> setting CAP_MULTI_CHANNEL when the server didn't actually send the bit.
> Have you tested this with servers that don't do multichannel?
Yes. Validate negotiate ioctl request is supposed to validate what
the client sent not what the server responded, so according to
MS-SMB2, I must send in the ioctl what I sent before on negprot
request
Section 3.2.5.5 says for validate negotiate "Capabilities is set to
Connection.ClientCapabilities." where
"Connection.ClientCapabilities: The capabilities sent by the client in
the SMB2 NEGOTIATE Request" (not what the server responded with,
what the ClientCapabilities were sent)
I tested it with two cases that don't support multichannel: Samba, and
also an azure server target where multichannel was disabled.
>
> > 2) we were ignoring whether the server set CAP_NEGOTIATE in the response
>
> Is this "CAP_NEGOTIATE" a typo? I think you mean CAP_MULTI_CHANNEL.
Yes - typo
>
> > diff --git a/fs/cifs/sess.c b/fs/cifs/sess.c
> > index 63d517b9f2ff..a391ca3166f3 100644
> > --- a/fs/cifs/sess.c
> > +++ b/fs/cifs/sess.c
> > @@ -97,6 +97,12 @@ int cifs_try_adding_channels(struct cifs_sb_info *cifs_sb, struct cifs_ses *ses)
> > return 0;
> > }
> >
> > + if ((ses->server->capabilities & SMB2_GLOBAL_CAP_MULTI_CHANNEL) == false) {
>
> This compares a bit to a boolean. "false" should be "0"?
I changed it to the more common style if (!(ses->...capabilities & SMB@....))
>
> > + cifs_dbg(VFS, "server does not support CAP_MULTI_CHANNEL, multichannel disabled\n");
>
> The wording could be clearer. Technically speaking, the server does not
> support _multichannel_, which it indicated by not setting CAP_MULTI_CHANNEL.
> Also, wouldn't it be more useful to add the servername to this message?
> "server %s does not support multichannel, using single channel"
> or similar.
Good idea
> > 3) we were silently ignoring multichannel when "max_channels" was > 1
> > but the user forgot to include "multichannel" in mount line.
>
> > diff --git a/fs/cifs/fs_context.c b/fs/cifs/fs_context.c
> > index 3bcf881c3ae9..8f7af6fcdc76 100644
> > --- a/fs/cifs/fs_context.c
> > +++ b/fs/cifs/fs_context.c
> > @@ -1021,6 +1021,9 @@ static int smb3_fs_context_parse_param(struct
> fs_context *fc,
> > goto cifs_parse_mount_err;
> > }
> > ctx->max_channels = result.uint_32;
> > + /* If more than one channel requested ... they want multichan */
> > + if ((ctx->multichannel == false) && (result.uint_32 > 1))
> > + ctx->multichannel = true;
>
> Wouldn't this be clearer and simpler as just "if (result.uint32 > 1)" ?
made that change
Updated two of the patches as described above - attached.
--
Thanks,
Steve
From 1fae9cf8242f7d7028fa95f1cfd24b67942b8b4e Mon Sep 17 00:00:00 2001
From: Steve French <stfrench@xxxxxxxxxxxxx>
Date: Fri, 7 May 2021 19:33:51 -0500
Subject: [PATCH 3/4] smb3: if max_channels set to more than one channel
request multichannel
Mounting with "multichannel" is obviously implied if user requested
more than one channel on mount (ie mount parm max_channels>1).
Currently both have to be specified. Fix that so that if max_channels
is greater than 1 on mount, enable multichannel rather than silently
falling back to non-multichannel.
Signed-off-by: Steve French <stfrench@xxxxxxxxxxxxx>
Reviewed-by: Shyam Prasad N <sprasad@xxxxxxxxxxxxx>
---
fs/cifs/fs_context.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/fs/cifs/fs_context.c b/fs/cifs/fs_context.c
index 3bcf881c3ae9..5d21cd905315 100644
--- a/fs/cifs/fs_context.c
+++ b/fs/cifs/fs_context.c
@@ -1021,6 +1021,9 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
goto cifs_parse_mount_err;
}
ctx->max_channels = result.uint_32;
+ /* If more than one channel requested ... they want multichan */
+ if (result.uint_32 > 1)
+ ctx->multichannel = true;
break;
case Opt_handletimeout:
ctx->handle_timeout = result.uint_32;
--
2.27.0
From f2421e5efcc25e1f7a5661d0ace059c1ddaf4b8d Mon Sep 17 00:00:00 2001
From: Steve French <stfrench@xxxxxxxxxxxxx>
Date: Fri, 7 May 2021 20:00:41 -0500
Subject: [PATCH 2/4] smb3: do not attempt multichannel to server which does
not support it
We were ignoring CAP_MULTI_CHANNEL in the server response - if the
server doesn't support multichannel we should not be attempting it.
See MS-SMB2 section 3.2.5.2
Reviewed-by: Shyam Prasad N <sprasad@xxxxxxxxxxxxx>
Signed-off-by: Steve French <stfrench@xxxxxxxxxxxxx>
---
fs/cifs/sess.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/fs/cifs/sess.c b/fs/cifs/sess.c
index 63d517b9f2ff..a92a1fb7cb52 100644
--- a/fs/cifs/sess.c
+++ b/fs/cifs/sess.c
@@ -97,6 +97,12 @@ int cifs_try_adding_channels(struct cifs_sb_info *cifs_sb, struct cifs_ses *ses)
return 0;
}
+ if (!(ses->server->capabilities & SMB2_GLOBAL_CAP_MULTI_CHANNEL)) {
+ cifs_dbg(VFS, "server %s does not support multichannel\n", ses->server->hostname);
+ ses->chan_max = 1;
+ return 0;
+ }
+
/*
* Make a copy of the iface list at the time and use that
* instead so as to not hold the iface spinlock for opening
--
2.27.0
