diff mbox series

nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes

Message ID 20180820065608.11860-1-tigran.mkrtchyan@desy.de (mailing list archive)
State New, archived
Headers show
Series nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes | expand

Commit Message

Mkrtchyan, Tigran Aug. 20, 2018, 6:56 a.m. UTC
for tightly coupled DSes client must ignore provided synthetic uid and
gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.

Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
---
 fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Rick Macklem Aug. 20, 2018, 12:51 p.m. UTC | #1
Thanks. I'll test this later to-day. (I did a slightly more complex version
of the fix outside of the ff_layout_get_ds_cred() call, but yours is
definitely simpler).

There is also the "stateid", which I believe is supposed to be the one in
the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
My patch is here and has the changes for stated as well as cred.

http://people.freebsd.org/~rmacklem/flexfile.patch
(Sorry, I don't know git;-)

Thanks for doing this and I'll post if my testing finds problems, rick
Rick Macklem Aug. 20, 2018, 1:18 p.m. UTC | #2
I just put a patch for the stated part (stripped out my version of the cred
stuff, which I noticed I missed commit in it;) so it might be easier to read.
It is here:
http://people.freebsd.org/~rmacklem/flexfilestateid.patch

Thanks for doing this, rick
Mkrtchyan, Tigran Aug. 20, 2018, 8:41 p.m. UTC | #3
Hi Rick,

draft-19 says:

   For tight coupling, ffds_stateid provides the stateid to be used by
   the client to access the file.  For loose coupling and a NFSv4
   storage device, the client will have to use an anonymous stateid to
   perform I/O on the storage device.  With no control protocol, the
   metadata server stateid can not be used to provide a global stateid
   model.  Thus the server MUST set the ffds_stateid to be the anonymous
   stateid.

To me, the last sentence sounds like a clear statement, that it's server
responsibility to provide either global or anonymous stateid and client
should just use it as is. IOW,

+	stateid = nfs4_ff_layout_select_ds_stateid(lseg, idx);
+	if (stateid)
+		nfs4_stateid_copy(&hdr->args.stateid, stateid);


must happen independent from ds being loose or tight coupled. As our DSes
use the same stateid's as MDS, we did see that provided stateid is not used.

Trond, Tom, can you comment on it?

Thanks,
    Tigran.

----- Original Message -----
> From: "Rick Macklem" <rmacklem@uoguelph.ca>
> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs" <linux-nfs@vger.kernel.org>
> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
> Sent: Monday, August 20, 2018 3:18:00 PM
> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes

> I just put a patch for the stated part (stripped out my version of the cred
> stuff, which I noticed I missed commit in it;) so it might be easier to read.
> It is here:
> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
> 
> Thanks for doing this, rick
> 
> ________________________________________
> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
> behalf of Rick Macklem <rmacklem@uoguelph.ca>
> Sent: Monday, August 20, 2018 8:51:14 AM
> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
> coupled DSes
> 
> Thanks. I'll test this later to-day. (I did a slightly more complex version
> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
> definitely simpler).
> 
> There is also the "stateid", which I believe is supposed to be the one in
> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
> My patch is here and has the changes for stated as well as cred.
> 
> http://people.freebsd.org/~rmacklem/flexfile.patch
> (Sorry, I don't know git;-)
> 
> Thanks for doing this and I'll post if my testing finds problems, rick
> 
> 
> ________________________________________
> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
> Sent: Monday, August 20, 2018 2:56:08 AM
> To: linux-nfs@vger.kernel.org
> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
> DSes
> 
> for tightly coupled DSes client must ignore provided synthetic uid and
> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
> 
> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
> ---
> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> index d62279d3fc5d..290625cfd369 100644
> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
> ds_idx,
>        struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>        struct rpc_cred *cred;
> 
> -       if (mirror) {
> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>                cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>                if (!cred)
>                        cred = get_rpccred(mdscred);
> --
> 2.17.1
Thomas Haynes Aug. 20, 2018, 8:52 p.m. UTC | #4
> On Aug 20, 2018, at 1:41 PM, Mkrtchyan, Tigran <tigran.mkrtchyan@desy.de> wrote:
> 
> Hi Rick,
> 
> draft-19 says:
> 
>   For tight coupling, ffds_stateid provides the stateid to be used by
>   the client to access the file.  For loose coupling and a NFSv4
>   storage device, the client will have to use an anonymous stateid to
>   perform I/O on the storage device.  With no control protocol, the
>   metadata server stateid can not be used to provide a global stateid
>   model.  Thus the server MUST set the ffds_stateid to be the anonymous
>   stateid.
> 
> To me, the last sentence sounds like a clear statement, that it's server
> responsibility to provide either global or anonymous stateid and client
> should just use it as is. IOW,
> 
> +	stateid = nfs4_ff_layout_select_ds_stateid(lseg, idx);
> +	if (stateid)
> +		nfs4_stateid_copy(&hdr->args.stateid, stateid);
> 
> 
> must happen independent from ds being loose or tight coupled. As our DSes
> use the same stateid's as MDS, we did see that provided stateid is not used.
> 
> Trond, Tom, can you comment on it?


Yes, I agree, the server always provides the stateid.


> 
> Thanks,
>    Tigran.
> 
> ----- Original Message -----
>> From: "Rick Macklem" <rmacklem@uoguelph.ca>
>> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs" <linux-nfs@vger.kernel.org>
>> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
>> Sent: Monday, August 20, 2018 3:18:00 PM
>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes
> 
>> I just put a patch for the stated part (stripped out my version of the cred
>> stuff, which I noticed I missed commit in it;) so it might be easier to read.
>> It is here:
>> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
>> 
>> Thanks for doing this, rick
>> 
>> ________________________________________
>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>> behalf of Rick Macklem <rmacklem@uoguelph.ca>
>> Sent: Monday, August 20, 2018 8:51:14 AM
>> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>> coupled DSes
>> 
>> Thanks. I'll test this later to-day. (I did a slightly more complex version
>> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
>> definitely simpler).
>> 
>> There is also the "stateid", which I believe is supposed to be the one in
>> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
>> My patch is here and has the changes for stated as well as cred.
>> 
>> http://people.freebsd.org/~rmacklem/flexfile.patch
>> (Sorry, I don't know git;-)
>> 
>> Thanks for doing this and I'll post if my testing finds problems, rick
>> 
>> 
>> ________________________________________
>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>> Sent: Monday, August 20, 2018 2:56:08 AM
>> To: linux-nfs@vger.kernel.org
>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
>> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
>> DSes
>> 
>> for tightly coupled DSes client must ignore provided synthetic uid and
>> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
>> 
>> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>> ---
>> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> index d62279d3fc5d..290625cfd369 100644
>> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
>> ds_idx,
>>       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>>       struct rpc_cred *cred;
>> 
>> -       if (mirror) {
>> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>>               cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>>               if (!cred)
>>                       cred = get_rpccred(mdscred);
>> --
>> 2.17.1
Rick Macklem Aug. 20, 2018, 9:16 p.m. UTC | #5
Tom Haynes wrote:
>> On Aug 20, 2018, at 1:41 PM, Mkrtchyan, Tigran <tigran.mkrtchyan@desy.de> wrote:
>>
>> Hi Rick,
>>
>> draft-19 says:
>>
>>   For tight coupling, ffds_stateid provides the stateid to be used by
>>   the client to access the file.  For loose coupling and a NFSv4
>>   storage device, the client will have to use an anonymous stateid to
>>   perform I/O on the storage device.  With no control protocol, the
>>   metadata server stateid can not be used to provide a global stateid
>>   model.  Thus the server MUST set the ffds_stateid to be the anonymous
>>   stateid.
>>
>> To me, the last sentence sounds like a clear statement, that it's server
>> responsibility to provide either global or anonymous stateid and client
>> should just use it as is. IOW,
>>
>> +     stateid = nfs4_ff_layout_select_ds_stateid(lseg, idx);
>> +     if (stateid)
>> +             nfs4_stateid_copy(&hdr->args.stateid, stateid);
>>
If you do this unconditionally, I think you can get rid of the code that looks like:
             if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
                                          hdr->args.lock_context, FMODE_READ) == -EIO)
                            rpc_exit(task, -EIO);  /* lost lock, terminate I/O */
at the end of ff_layout_read_prepare_v4() and similar but with FMODE_WRITE
at the end of ff_layout_write_prepare_v4().

>>
>> must happen independent from ds being loose or tight coupled. As our DSes
>> use the same stateid's as MDS, we did see that provided stateid is not used.
>>
>> Trond, Tom, can you comment on it?
>
>
>Yes, I agree, the server always provides the stateid.
Sure. Since the above states that the loosely coupled server must fill it in as
the anonymous stated, the client will end up using the anonymous stated
for loosely coupled, whether it uses ffds_stateid or just sets it to the
anonymous stated. (Unpatched, the client always use the anonymous stated, including for tightly coupled.)

Doing it unconditionally makes the patch even simpler.

rick

>
> Thanks,
>    Tigran.
>
> ----- Original Message -----
>> From: "Rick Macklem" <rmacklem@uoguelph.ca>
>> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs" <linux-nfs@vger.kernel.org>
>> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
>> Sent: Monday, August 20, 2018 3:18:00 PM
>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes
>
>> I just put a patch for the stated part (stripped out my version of the cred
>> stuff, which I noticed I missed commit in it;) so it might be easier to read.
>> It is here:
>> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
>>
>> Thanks for doing this, rick
>>
>> ________________________________________
>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>> behalf of Rick Macklem <rmacklem@uoguelph.ca>
>> Sent: Monday, August 20, 2018 8:51:14 AM
>> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>> coupled DSes
>>
>> Thanks. I'll test this later to-day. (I did a slightly more complex version
>> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
>> definitely simpler).
>>
>> There is also the "stateid", which I believe is supposed to be the one in
>> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
>> My patch is here and has the changes for stated as well as cred.
>>
>> http://people.freebsd.org/~rmacklem/flexfile.patch
>> (Sorry, I don't know git;-)
>>
>> Thanks for doing this and I'll post if my testing finds problems, rick
>>
>>
>> ________________________________________
>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>> Sent: Monday, August 20, 2018 2:56:08 AM
>> To: linux-nfs@vger.kernel.org
>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
>> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
>> DSes
>>
>> for tightly coupled DSes client must ignore provided synthetic uid and
>> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
>>
>> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>> ---
>> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> index d62279d3fc5d..290625cfd369 100644
>> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
>> ds_idx,
>>       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>>       struct rpc_cred *cred;
>>
>> -       if (mirror) {
>> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>>               cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>>               if (!cred)
>>                       cred = get_rpccred(mdscred);
>> --
>> 2.17.1
Mkrtchyan, Tigran Aug. 20, 2018, 9:28 p.m. UTC | #6
----- Original Message -----
> From: "Rick Macklem" <rmacklem@uoguelph.ca>
> To: "Tom Haynes" <loghyr@gmail.com>, "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>
> Cc: "linux-nfs" <linux-nfs@vger.kernel.org>, "trondmy" <trondmy@hammerspace.com>, "Anna Schumaker"
> <Anna.Schumaker@Netapp.com>
> Sent: Monday, August 20, 2018 11:16:22 PM
> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes

> Tom Haynes wrote:
>>> On Aug 20, 2018, at 1:41 PM, Mkrtchyan, Tigran <tigran.mkrtchyan@desy.de> wrote:
>>>
>>> Hi Rick,
>>>
>>> draft-19 says:
>>>
>>>   For tight coupling, ffds_stateid provides the stateid to be used by
>>>   the client to access the file.  For loose coupling and a NFSv4
>>>   storage device, the client will have to use an anonymous stateid to
>>>   perform I/O on the storage device.  With no control protocol, the
>>>   metadata server stateid can not be used to provide a global stateid
>>>   model.  Thus the server MUST set the ffds_stateid to be the anonymous
>>>   stateid.
>>>
>>> To me, the last sentence sounds like a clear statement, that it's server
>>> responsibility to provide either global or anonymous stateid and client
>>> should just use it as is. IOW,
>>>
>>> +     stateid = nfs4_ff_layout_select_ds_stateid(lseg, idx);
>>> +     if (stateid)
>>> +             nfs4_stateid_copy(&hdr->args.stateid, stateid);
>>>
> If you do this unconditionally, I think you can get rid of the code that looks
> like:
>             if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
>                                          hdr->args.lock_context, FMODE_READ) == -EIO)
>                            rpc_exit(task, -EIO);  /* lost lock, terminate I/O */
> at the end of ff_layout_read_prepare_v4() and similar but with FMODE_WRITE
> at the end of ff_layout_write_prepare_v4().
> 
>>>
>>> must happen independent from ds being loose or tight coupled. As our DSes
>>> use the same stateid's as MDS, we did see that provided stateid is not used.
>>>
>>> Trond, Tom, can you comment on it?
>>
>>
>>Yes, I agree, the server always provides the stateid.
> Sure. Since the above states that the loosely coupled server must fill it in as
> the anonymous stated, the client will end up using the anonymous stated
> for loosely coupled, whether it uses ffds_stateid or just sets it to the
> anonymous stated. (Unpatched, the client always use the anonymous stated,
> including for tightly coupled.)

I am not sure that's true, as our DSes will reject such IO requests. And we
do use flex_files in production:

Frame 96: 264 bytes on wire (2112 bits), 264 bytes captured (2112 bits) on interface 0
Linux cooked capture
Internet Protocol Version 4, Src: 131.169.xx, Dst: 131.169.xx
Transmission Control Protocol, Src Port: 946, Dst Port: 32049, Seq: 925, Ack: 569, Len: 196
Remote Procedure Call, Type:Call XID:0xa09b9261
Network File System, Ops(3): SEQUENCE, PUTFH, READ
    [Program Version: 4]
    [V4 Procedure: COMPOUND (1)]
    Tag: <EMPTY>
        length: 0
        contents: <EMPTY>
    minorversion: 1
    Operations (count: 3): SEQUENCE, PUTFH, READ
        Opcode: SEQUENCE (53)
            sessionid: 5b7b31b9000000060000000000000001
            seqid: 0x00000002
            slot id: 0
            high slot id: 0
            cache this?: No
        Opcode: PUTFH (22)
            FileHandle
                length: 27
                [hash (CRC-32): 0x4c4a3909]
                FileHandle: 01caffee000000008d61f80c000d00000800000000002887...
        Opcode: READ (25)
            StateID
                [StateID Hash: 0x71b8]
                StateID seqid: 0
                StateID Other: 5b7b31b60000000500000001
                [StateID Other hash: 0xbecf]
            offset: 0
            count: 1015
    [Main Opcode: READ (25)]


Tigran.

> 
> Doing it unconditionally makes the patch even simpler.
> 
> rick
> 
>>
>> Thanks,
>>    Tigran.
>>
>> ----- Original Message -----
>>> From: "Rick Macklem" <rmacklem@uoguelph.ca>
>>> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs"
>>> <linux-nfs@vger.kernel.org>
>>> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
>>> Sent: Monday, August 20, 2018 3:18:00 PM
>>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>>> coupled DSes
>>
>>> I just put a patch for the stated part (stripped out my version of the cred
>>> stuff, which I noticed I missed commit in it;) so it might be easier to read.
>>> It is here:
>>> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
>>>
>>> Thanks for doing this, rick
>>>
>>> ________________________________________
>>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>>> behalf of Rick Macklem <rmacklem@uoguelph.ca>
>>> Sent: Monday, August 20, 2018 8:51:14 AM
>>> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
>>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
>>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>>> coupled DSes
>>>
>>> Thanks. I'll test this later to-day. (I did a slightly more complex version
>>> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
>>> definitely simpler).
>>>
>>> There is also the "stateid", which I believe is supposed to be the one in
>>> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
>>> My patch is here and has the changes for stated as well as cred.
>>>
>>> http://people.freebsd.org/~rmacklem/flexfile.patch
>>> (Sorry, I don't know git;-)
>>>
>>> Thanks for doing this and I'll post if my testing finds problems, rick
>>>
>>>
>>> ________________________________________
>>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>>> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>>> Sent: Monday, August 20, 2018 2:56:08 AM
>>> To: linux-nfs@vger.kernel.org
>>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
>>> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
>>> DSes
>>>
>>> for tightly coupled DSes client must ignore provided synthetic uid and
>>> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
>>>
>>> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>>> ---
>>> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> index d62279d3fc5d..290625cfd369 100644
>>> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
>>> ds_idx,
>>>       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>>>       struct rpc_cred *cred;
>>>
>>> -       if (mirror) {
>>> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>>>               cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>>>               if (!cred)
>>>                       cred = get_rpccred(mdscred);
>>> --
> >> 2.17.1
Rick Macklem Aug. 20, 2018, 9:46 p.m. UTC | #7
Mkrtchyan, Tigran wrote:
[stuff snipped for brevity]
>Rick Macklem wrote:
>> Sure. Since the above states that the loosely coupled server must fill it in as
>> the anonymous stated, the client will end up using the anonymous stated
>> for loosely coupled, whether it uses ffds_stateid or just sets it to the
>> anonymous stated. (Unpatched, the client always use the anonymous stated,
>> including for tightly coupled.)
>
>I am not sure that's true, as our DSes will reject such IO requests. And we
>do use flex_files in production:
Oops, yes, you are correct. (It has been a while since I did this.)
The unpatched driver uses whatever it would use for the MDS (an open/lock/…
stated). If the server sends the same stateid in the layout, then you wouldn't notice.
(I'd guess Primary Data is using NFSv3 data servers, so they wouldn't notice either.)

Oh, and ignore my comment about taking out the code at the end of
ff_layout_read_prepare_v4() and ff_layout_write_prepare_v4() unless you know
of a better way to handle the failure case than filling in the stateid that would
be used against the MDS when it can't get the ffds_stateid.

rick

Frame 96: 264 bytes on wire (2112 bits), 264 bytes captured (2112 bits) on interface 0
Linux cooked capture
Internet Protocol Version 4, Src: 131.169.xx, Dst: 131.169.xx
Transmission Control Protocol, Src Port: 946, Dst Port: 32049, Seq: 925, Ack: 569, Len: 196
Remote Procedure Call, Type:Call XID:0xa09b9261
Network File System, Ops(3): SEQUENCE, PUTFH, READ
    [Program Version: 4]
    [V4 Procedure: COMPOUND (1)]
    Tag: <EMPTY>
        length: 0
        contents: <EMPTY>
    minorversion: 1
    Operations (count: 3): SEQUENCE, PUTFH, READ
        Opcode: SEQUENCE (53)
            sessionid: 5b7b31b9000000060000000000000001
            seqid: 0x00000002
            slot id: 0
            high slot id: 0
            cache this?: No
        Opcode: PUTFH (22)
            FileHandle
                length: 27
                [hash (CRC-32): 0x4c4a3909]
                FileHandle: 01caffee000000008d61f80c000d00000800000000002887...
        Opcode: READ (25)
            StateID
                [StateID Hash: 0x71b8]
                StateID seqid: 0
                StateID Other: 5b7b31b60000000500000001
                [StateID Other hash: 0xbecf]
            offset: 0
            count: 1015
    [Main Opcode: READ (25)]


Tigran.

>
> Doing it unconditionally makes the patch even simpler.
>
> rick
>
>>
>> Thanks,
>>    Tigran.
>>
>> ----- Original Message -----
>>> From: "Rick Macklem" <rmacklem@uoguelph.ca>
>>> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs"
>>> <linux-nfs@vger.kernel.org>
>>> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
>>> Sent: Monday, August 20, 2018 3:18:00 PM
>>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>>> coupled DSes
>>
>>> I just put a patch for the stated part (stripped out my version of the cred
>>> stuff, which I noticed I missed commit in it;) so it might be easier to read.
>>> It is here:
>>> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
>>>
>>> Thanks for doing this, rick
>>>
>>> ________________________________________
>>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>>> behalf of Rick Macklem <rmacklem@uoguelph.ca>
>>> Sent: Monday, August 20, 2018 8:51:14 AM
>>> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
>>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
>>> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
>>> coupled DSes
>>>
>>> Thanks. I'll test this later to-day. (I did a slightly more complex version
>>> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
>>> definitely simpler).
>>>
>>> There is also the "stateid", which I believe is supposed to be the one in
>>> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
>>> My patch is here and has the changes for stated as well as cred.
>>>
>>> http://people.freebsd.org/~rmacklem/flexfile.patch
>>> (Sorry, I don't know git;-)
>>>
>>> Thanks for doing this and I'll post if my testing finds problems, rick
>>>
>>>
>>> ________________________________________
>>> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
>>> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>>> Sent: Monday, August 20, 2018 2:56:08 AM
>>> To: linux-nfs@vger.kernel.org
>>> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
>>> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
>>> DSes
>>>
>>> for tightly coupled DSes client must ignore provided synthetic uid and
>>> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
>>>
>>> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
>>> ---
>>> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> index d62279d3fc5d..290625cfd369 100644
>>> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
>>> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
>>> ds_idx,
>>>       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>>>       struct rpc_cred *cred;
>>>
>>> -       if (mirror) {
>>> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>>>               cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>>>               if (!cred)
>>>                       cred = get_rpccred(mdscred);
>>> --
> >> 2.17.1
Rick Macklem Aug. 21, 2018, 1:32 a.m. UTC | #8
Mkrtchyan, Tigran wrote:
>draft-19 says:
>
>   For tight coupling, ffds_stateid provides the stateid to be used by
>   the client to access the file.  For loose coupling and a NFSv4
>   storage device, the client will have to use an anonymous stateid to
>   perform I/O on the storage device.  With no control protocol, the
>   metadata server stateid can not be used to provide a global stateid
>   model.  Thus the server MUST set the ffds_stateid to be the anonymous
>   stateid.
>
>To me, the last sentence sounds like a clear statement, that it's server
>responsibility to provide either global or anonymous stateid and client
>should just use it as is. IOW,
>
>+       stateid = nfs4_ff_layout_select_ds_stateid(lseg, idx);
>+       if (stateid)
>+               nfs4_stateid_copy(&hdr->args.stateid, stateid);
>
>
>must happen independent from ds being loose or tight coupled. As our DSes
>use the same stateid's as MDS, we did see that provided stateid is not used.
>
>Trond, Tom, can you comment on it?
I've updated http://people.freebsd.org/~rmacklem/flexfilestateid.patch
to do the above for both loose and tightly coupled as you suggested.
When used with your little cred patch, everything seems ok against the
FreeBSD server (which is tightly coupled and uses ffds_stateids that are
different from the open/lock... ones).

rick

Thanks,
    Tigran.

----- Original Message -----
> From: "Rick Macklem" <rmacklem@uoguelph.ca>
> To: "Tigran Mkrtchyan" <tigran.mkrtchyan@desy.de>, "linux-nfs" <linux-nfs@vger.kernel.org>
> Cc: trondmy@hammerspace.com, "Anna Schumaker" <Anna.Schumaker@Netapp.com>
> Sent: Monday, August 20, 2018 3:18:00 PM
> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled DSes

> I just put a patch for the stated part (stripped out my version of the cred
> stuff, which I noticed I missed commit in it;) so it might be easier to read.
> It is here:
> http://people.freebsd.org/~rmacklem/flexfilestateid.patch
>
> Thanks for doing this, rick
>
> ________________________________________
> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
> behalf of Rick Macklem <rmacklem@uoguelph.ca>
> Sent: Monday, August 20, 2018 8:51:14 AM
> To: Tigran Mkrtchyan; linux-nfs@vger.kernel.org
> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com
> Subject: Re: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly
> coupled DSes
>
> Thanks. I'll test this later to-day. (I did a slightly more complex version
> of the fix outside of the ff_layout_get_ds_cred() call, but yours is
> definitely simpler).
>
> There is also the "stateid", which I believe is supposed to be the one in
> the layout for the tightly coupled case (for NFSv4 I/O ops to the DS).
> My patch is here and has the changes for stated as well as cred.
>
> http://people.freebsd.org/~rmacklem/flexfile.patch
> (Sorry, I don't know git;-)
>
> Thanks for doing this and I'll post if my testing finds problems, rick
>
>
> ________________________________________
> From: linux-nfs-owner@vger.kernel.org <linux-nfs-owner@vger.kernel.org> on
> behalf of Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
> Sent: Monday, August 20, 2018 2:56:08 AM
> To: linux-nfs@vger.kernel.org
> Cc: trondmy@hammerspace.com; Anna.Schumaker@Netapp.com; Tigran Mkrtchyan
> Subject: [PATCH] nfs4: flex_file: ignore synthetic uid/gid for tightly coupled
> DSes
>
> for tightly coupled DSes client must ignore provided synthetic uid and
> gid as stated in draft-ietf-nfsv4-flex-files-19#section-5.1.
>
> Signed-off-by: Tigran Mkrtchyan <tigran.mkrtchyan@desy.de>
> ---
> fs/nfs/flexfilelayout/flexfilelayoutdev.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> index d62279d3fc5d..290625cfd369 100644
> --- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> +++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
> @@ -452,7 +452,7 @@ ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32
> ds_idx,
>        struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
>        struct rpc_cred *cred;
>
> -       if (mirror) {
> +       if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
>                cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
>                if (!cred)
>                        cred = get_rpccred(mdscred);
> --
> 2.17.1
diff mbox series

Patch

diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
index d62279d3fc5d..290625cfd369 100644
--- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
+++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
@@ -452,7 +452,7 @@  ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
 	struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
 	struct rpc_cred *cred;
 
-	if (mirror) {
+	if (mirror && !mirror->mirror_ds->ds_versions[0].tightly_coupled) {
 		cred = ff_layout_get_mirror_cred(mirror, lseg->pls_range.iomode);
 		if (!cred)
 			cred = get_rpccred(mdscred);