Content-Type: text/plain;
        charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 12.2 \(3445.102.3\))
Subject: Re: [PATCH v1 20/23] NFS: Account for XDR pad of buf->pages
From:   Chuck Lever <chuck.lever@oracle.com>
In-Reply-To: <CAN-5tyE6rgj90=bCcX82fW34i1SNSSLcLVHbjBo5dmQaXzoTwA@mail.gmail.com>
Date:   Mon, 8 Apr 2019 10:43:04 -0400
Cc:     Anna Schumaker <anna.schumaker@netapp.com>,
        linux-rdma <linux-rdma@vger.kernel.org>,
        Linux NFS Mailing List <linux-nfs@vger.kernel.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <1E4939AD-F0AE-47EF-81E2-B919FB203B76@oracle.com>
References: <20190211161920.2817.20881.stgit@manet.1015granger.net>
 <20190211162525.2817.60645.stgit@manet.1015granger.net>
 <CAN-5tyH1zAuW7AxoZc+psRWWQT0S9J3wDu+iDgLg4=qGTvUufA@mail.gmail.com>
 <0157F163-7554-413C-ACCA-FE6646C04716@oracle.com>
 <CAN-5tyHSH75TVSok+RXdXoq6bOVse__E4AMCzuSnmYrTgzH_Vg@mail.gmail.com>
 <5AC0653D-3D4E-4C0C-BFE3-616BF32589CB@oracle.com>
 <CAN-5tyFN90xdKBvFmH2kYa2gQtW2EmVa59Q+w=M4yvHR9euuVQ@mail.gmail.com>
 <19341226-AE0F-4C94-B7CD-AE14CCB456AF@oracle.com>
 <CAN-5tyE6rgj90=bCcX82fW34i1SNSSLcLVHbjBo5dmQaXzoTwA@mail.gmail.com>
To:     Olga Kornievskaia <aglo@umich.edu>
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk


> On Apr 8, 2019, at 10:36 AM, Olga Kornievskaia <aglo@umich.edu> wrote:
>=20
> On Fri, Apr 5, 2019 at 3:42 PM Chuck Lever <chuck.lever@oracle.com> =
wrote:
>>=20
>>=20
>>=20
>>> On Apr 5, 2019, at 3:27 PM, Olga Kornievskaia <aglo@umich.edu> =
wrote:
>>>=20
>>> On Fri, Apr 5, 2019 at 3:23 PM Chuck Lever <chuck.lever@oracle.com> =
wrote:
>>>>=20
>>>>=20
>>>>=20
>>>>> On Apr 5, 2019, at 3:17 PM, Olga Kornievskaia <aglo@umich.edu> =
wrote:
>>>>>=20
>>>>> On Fri, Apr 5, 2019 at 1:51 PM Chuck Lever =
<chuck.lever@oracle.com> wrote:
>>>>>>=20
>>>>>>=20
>>>>>>=20
>>>>>>> On Apr 5, 2019, at 1:36 PM, Olga Kornievskaia <aglo@umich.edu> =
wrote:
>>>>>>>=20
>>>>>>> Hi Chuck,
>>>>>>>=20
>>>>>>> This patch break ACLs. After applying this patch nfs4_getfacl =
fails
>>>>>>> (it fails within xdr and returns ENOTSUPP). Any ideas why?
>>>>>>=20
>>>>>> Possibly the macro that defines the maximum size of the reply
>>>>>> is incorrect.
>>>>>>=20
>>>>>=20
>>>>> This also breaks FS_LOCATION. I'm going to go on the limb here and =
say
>>>>> that it probably breaks whatever else it modified.
>>>>=20
>>>> It modifies READ, READDIR, and READLINK. Are those broken?
>>>=20
>>> I don't know how to test READLINK.. but I think READ/READDIR work OK
>>> otherwise folks would have noticed it (I gather ACL and FS_LOCATION
>>> testing doesn't happen frequently).
>>=20
>> I guess I don't have any NFSv4 ACL or FS_LOCATIONS regressions
>> tests in my automated unit tests.
>>=20
>>=20
>>>>> The question is: can't we just revert it??
>>>>=20
>>>> Why not "root cause" it first?
>>>=20
>>> I'm trying :-/ I was just fishing to see how important the change =
was.
>>=20
>> Try reverting just this hunk:
>=20
> That doesn't help. It seems to be this piece that's causing issues
> hdrsize +=3D RPC_REPHDRSIZE + req->rq_cred->cr_auth->au_rslack - 1
>=20
> With this there is an extra byte (in front) in the buffer when (ACL)
> operation is decoded.

How do you know there isn't a latent bug in the getfacl decoder?

How are you reproducing this issue? I can try it here later today.


>> diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
>> index d0fa18d..6d9d5e2 100644
>> --- a/fs/nfs/nfs4xdr.c
>> +++ b/fs/nfs/nfs4xdr.c
>> @@ -284,14 +284,14 @@ static int decode_layoutget(struct xdr_stream =
*xdr, struct rpc_rqst *req,
>> #define decode_delegreturn_maxsz (op_decode_hdr_maxsz)
>> #define encode_getacl_maxsz     (encode_getattr_maxsz)
>> #define decode_getacl_maxsz     (op_decode_hdr_maxsz + \
>> -                                nfs4_fattr_bitmap_maxsz + 1)
>> +                                nfs4_fattr_bitmap_maxsz + 1 + 1)
>> #define encode_setacl_maxsz     (op_encode_hdr_maxsz + \
>>                                 encode_stateid_maxsz + 3)
>> #define decode_setacl_maxsz     (decode_setattr_maxsz)
>> #define encode_fs_locations_maxsz \
>>                                (encode_getattr_maxsz)
>> #define decode_fs_locations_maxsz \
>> -                               (0)
>> +                               (1)
>> #define encode_secinfo_maxsz    (op_encode_hdr_maxsz + =
nfs4_name_maxsz)
>> #define decode_secinfo_maxsz    (op_decode_hdr_maxsz + 1 + =
((NFS_MAX_SECFLAVORS * (16 + GSS_OID_MAX_LEN)) / 4))
>>=20
>>=20
>>>>>>> On Mon, Feb 11, 2019 at 11:25 AM Chuck Lever =
<chuck.lever@oracle.com> wrote:
>>>>>>>>=20
>>>>>>>> Certain NFS results (eg. READLINK) might expect a data payload =
that
>>>>>>>> is not an exact multiple of 4 bytes. In this case, XDR encoding
>>>>>>>> is required to pad that payload so its length on the wire is a
>>>>>>>> multiple of 4 bytes. The constants that define the maximum size =
of
>>>>>>>> each NFS result do not appear to account for this extra word.
>>>>>>>>=20
>>>>>>>> In each case where the data payload is to be received into =
pages:
>>>>>>>>=20
>>>>>>>> - 1 word is added to the size of the receive buffer allocated =
by
>>>>>>>> call_allocate
>>>>>>>>=20
>>>>>>>> - rpc_inline_rcv_pages subtracts 1 word from @hdrsize so that =
the
>>>>>>>> extra buffer space falls into the rcv_buf's tail iovec
>>>>>>>>=20
>>>>>>>> - If buf->pagelen is word-aligned, an XDR pad is not needed and
>>>>>>>> is thus removed from the tail
>>>>>>>>=20
>>>>>>>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>>>>>>>> ---
>>>>>>>> fs/nfs/nfs2xdr.c  |    6 +++---
>>>>>>>> fs/nfs/nfs3xdr.c  |   10 +++++-----
>>>>>>>> fs/nfs/nfs4xdr.c  |   15 ++++++++-------
>>>>>>>> net/sunrpc/clnt.c |    6 +++++-
>>>>>>>> net/sunrpc/xdr.c  |    2 ++
>>>>>>>> 5 files changed, 23 insertions(+), 16 deletions(-)
>>>>>>>>=20
>>>>>>>> diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c
>>>>>>>> index 1dcd0fe..a7ed29d 100644
>>>>>>>> --- a/fs/nfs/nfs2xdr.c
>>>>>>>> +++ b/fs/nfs/nfs2xdr.c
>>>>>>>> @@ -56,11 +56,11 @@
>>>>>>>>=20
>>>>>>>> #define NFS_attrstat_sz                (1+NFS_fattr_sz)
>>>>>>>> #define NFS_diropres_sz                =
(1+NFS_fhandle_sz+NFS_fattr_sz)
>>>>>>>> -#define NFS_readlinkres_sz     (2)
>>>>>>>> -#define NFS_readres_sz         (1+NFS_fattr_sz+1)
>>>>>>>> +#define NFS_readlinkres_sz     (2+1)
>>>>>>>> +#define NFS_readres_sz         (1+NFS_fattr_sz+1+1)
>>>>>>>> #define NFS_writeres_sz         (NFS_attrstat_sz)
>>>>>>>> #define NFS_stat_sz            (1)
>>>>>>>> -#define NFS_readdirres_sz      (1)
>>>>>>>> +#define NFS_readdirres_sz      (1+1)
>>>>>>>> #define NFS_statfsres_sz       (1+NFS_info_sz)
>>>>>>>>=20
>>>>>>>> static int nfs_stat_to_errno(enum nfs_stat);
>>>>>>>> diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c
>>>>>>>> index a54dcf4..110358f 100644
>>>>>>>> --- a/fs/nfs/nfs3xdr.c
>>>>>>>> +++ b/fs/nfs/nfs3xdr.c
>>>>>>>> @@ -69,13 +69,13 @@
>>>>>>>> #define NFS3_removeres_sz      (NFS3_setattrres_sz)
>>>>>>>> #define NFS3_lookupres_sz      (1+NFS3_fh_sz+(2 * =
NFS3_post_op_attr_sz))
>>>>>>>> #define NFS3_accessres_sz      (1+NFS3_post_op_attr_sz+1)
>>>>>>>> -#define NFS3_readlinkres_sz    (1+NFS3_post_op_attr_sz+1)
>>>>>>>> -#define NFS3_readres_sz                =
(1+NFS3_post_op_attr_sz+3)
>>>>>>>> +#define NFS3_readlinkres_sz    (1+NFS3_post_op_attr_sz+1+1)
>>>>>>>> +#define NFS3_readres_sz                =
(1+NFS3_post_op_attr_sz+3+1)
>>>>>>>> #define NFS3_writeres_sz       (1+NFS3_wcc_data_sz+4)
>>>>>>>> #define NFS3_createres_sz      =
(1+NFS3_fh_sz+NFS3_post_op_attr_sz+NFS3_wcc_data_sz)
>>>>>>>> #define NFS3_renameres_sz      (1+(2 * NFS3_wcc_data_sz))
>>>>>>>> #define NFS3_linkres_sz                =
(1+NFS3_post_op_attr_sz+NFS3_wcc_data_sz)
>>>>>>>> -#define NFS3_readdirres_sz     (1+NFS3_post_op_attr_sz+2)
>>>>>>>> +#define NFS3_readdirres_sz     (1+NFS3_post_op_attr_sz+2+1)
>>>>>>>> #define NFS3_fsstatres_sz      (1+NFS3_post_op_attr_sz+13)
>>>>>>>> #define NFS3_fsinfores_sz      (1+NFS3_post_op_attr_sz+12)
>>>>>>>> #define NFS3_pathconfres_sz    (1+NFS3_post_op_attr_sz+6)
>>>>>>>> @@ -85,7 +85,7 @@
>>>>>>>> #define ACL3_setaclargs_sz     (NFS3_fh_sz+1+ \
>>>>>>>>                             =
XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE))
>>>>>>>> #define ACL3_getaclres_sz      (1+NFS3_post_op_attr_sz+1+ \
>>>>>>>> -                               =
XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE))
>>>>>>>> +                               =
XDR_QUADLEN(NFS_ACL_INLINE_BUFSIZE)+1)
>>>>>>>> #define ACL3_setaclres_sz      (1+NFS3_post_op_attr_sz)
>>>>>>>>=20
>>>>>>>> static int nfs3_stat_to_errno(enum nfs_stat);
>>>>>>>> @@ -1629,7 +1629,7 @@ static int nfs3_xdr_dec_read3res(struct =
rpc_rqst *req, struct xdr_stream *xdr,
>>>>>>>>     result->op_status =3D status;
>>>>>>>>     if (status !=3D NFS3_OK)
>>>>>>>>             goto out_status;
>>>>>>>> -       result->replen =3D 3 + ((xdr_stream_pos(xdr) - pos) >> =
2);
>>>>>>>> +       result->replen =3D 4 + ((xdr_stream_pos(xdr) - pos) >> =
2);
>>>>>>>>     error =3D decode_read3resok(xdr, result);
>>>>>>>> out:
>>>>>>>>     return error;
>>>>>>>> diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
>>>>>>>> index d0fa18d..6d9d5e2 100644
>>>>>>>> --- a/fs/nfs/nfs4xdr.c
>>>>>>>> +++ b/fs/nfs/nfs4xdr.c
>>>>>>>> @@ -215,14 +215,14 @@ static int decode_layoutget(struct =
xdr_stream *xdr, struct rpc_rqst *req,
>>>>>>>>                              nfs4_fattr_bitmap_maxsz)
>>>>>>>> #define encode_read_maxsz      (op_encode_hdr_maxsz + \
>>>>>>>>                              encode_stateid_maxsz + 3)
>>>>>>>> -#define decode_read_maxsz      (op_decode_hdr_maxsz + 2)
>>>>>>>> +#define decode_read_maxsz      (op_decode_hdr_maxsz + 2 + 1)
>>>>>>>> #define encode_readdir_maxsz   (op_encode_hdr_maxsz + \
>>>>>>>>                              2 + encode_verifier_maxsz + 5 + \
>>>>>>>>                             nfs4_label_maxsz)
>>>>>>>> #define decode_readdir_maxsz   (op_decode_hdr_maxsz + \
>>>>>>>> -                                decode_verifier_maxsz)
>>>>>>>> +                                decode_verifier_maxsz + 1)
>>>>>>>> #define encode_readlink_maxsz  (op_encode_hdr_maxsz)
>>>>>>>> -#define decode_readlink_maxsz  (op_decode_hdr_maxsz + 1)
>>>>>>>> +#define decode_readlink_maxsz  (op_decode_hdr_maxsz + 1 + 1)
>>>>>>>> #define encode_write_maxsz     (op_encode_hdr_maxsz + \
>>>>>>>>                              encode_stateid_maxsz + 4)
>>>>>>>> #define decode_write_maxsz     (op_decode_hdr_maxsz + \
>>>>>>>> @@ -284,14 +284,14 @@ static int decode_layoutget(struct =
xdr_stream *xdr, struct rpc_rqst *req,
>>>>>>>> #define decode_delegreturn_maxsz (op_decode_hdr_maxsz)
>>>>>>>> #define encode_getacl_maxsz    (encode_getattr_maxsz)
>>>>>>>> #define decode_getacl_maxsz    (op_decode_hdr_maxsz + \
>>>>>>>> -                                nfs4_fattr_bitmap_maxsz + 1)
>>>>>>>> +                                nfs4_fattr_bitmap_maxsz + 1 + =
1)
>>>>>>>> #define encode_setacl_maxsz    (op_encode_hdr_maxsz + \
>>>>>>>>                              encode_stateid_maxsz + 3)
>>>>>>>> #define decode_setacl_maxsz    (decode_setattr_maxsz)
>>>>>>>> #define encode_fs_locations_maxsz \
>>>>>>>>                             (encode_getattr_maxsz)
>>>>>>>> #define decode_fs_locations_maxsz \
>>>>>>>> -                               (0)
>>>>>>>> +                               (1)
>>>>>>>> #define encode_secinfo_maxsz   (op_encode_hdr_maxsz + =
nfs4_name_maxsz)
>>>>>>>> #define decode_secinfo_maxsz   (op_decode_hdr_maxsz + 1 + =
((NFS_MAX_SECFLAVORS * (16 + GSS_OID_MAX_LEN)) / 4))
>>>>>>>>=20
>>>>>>>> @@ -392,12 +392,13 @@ static int decode_layoutget(struct =
xdr_stream *xdr, struct rpc_rqst *req,
>>>>>>>>                             1 /* opaque devaddr4 length */ + \
>>>>>>>>                               /* devaddr4 payload is read into =
page */ \
>>>>>>>>                             1 /* notification bitmap length */ =
+ \
>>>>>>>> -                               1 /* notification bitmap, word =
0 */)
>>>>>>>> +                               1 /* notification bitmap, word =
0 */ + \
>>>>>>>> +                               1 /* possible XDR padding */)
>>>>>>>> #define encode_layoutget_maxsz (op_encode_hdr_maxsz + 10 + \
>>>>>>>>                             encode_stateid_maxsz)
>>>>>>>> #define decode_layoutget_maxsz (op_decode_hdr_maxsz + 8 + \
>>>>>>>>                             decode_stateid_maxsz + \
>>>>>>>> -                               =
XDR_QUADLEN(PNFS_LAYOUT_MAXSIZE))
>>>>>>>> +                               =
XDR_QUADLEN(PNFS_LAYOUT_MAXSIZE) + 1)
>>>>>>>> #define encode_layoutcommit_maxsz (op_encode_hdr_maxsz +        =
  \
>>>>>>>>                             2 /* offset */ + \
>>>>>>>>                             2 /* length */ + \
>>>>>>>> diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
>>>>>>>> index f780605..4ea38b0 100644
>>>>>>>> --- a/net/sunrpc/clnt.c
>>>>>>>> +++ b/net/sunrpc/clnt.c
>>>>>>>> @@ -1177,7 +1177,11 @@ void rpc_prepare_reply_pages(struct =
rpc_rqst *req, struct page **pages,
>>>>>>>>                          unsigned int base, unsigned int len,
>>>>>>>>                          unsigned int hdrsize)
>>>>>>>> {
>>>>>>>> -       hdrsize +=3D RPC_REPHDRSIZE + =
req->rq_cred->cr_auth->au_rslack;
>>>>>>>> +       /* Subtract one to force an extra word of buffer space =
for the
>>>>>>>> +        * payload's XDR pad to fall into the rcv_buf's tail =
iovec.
>>>>>>>> +        */
>>>>>>>> +       hdrsize +=3D RPC_REPHDRSIZE + =
req->rq_cred->cr_auth->au_rslack - 1;
>>>>>>>> +
>>>>>>>>     xdr_inline_pages(&req->rq_rcv_buf, hdrsize << 2, pages, =
base, len);
>>>>>>>>     trace_rpc_reply_pages(req);
>>>>>>>> }
>>>>>>>> diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
>>>>>>>> index 7cca515..aa8177d 100644
>>>>>>>> --- a/net/sunrpc/xdr.c
>>>>>>>> +++ b/net/sunrpc/xdr.c
>>>>>>>> @@ -189,6 +189,8 @@ __be32 *xdr_encode_opaque(__be32 *p, const =
void *ptr, unsigned int nbytes)
>>>>>>>>=20
>>>>>>>>     tail->iov_base =3D buf + offset;
>>>>>>>>     tail->iov_len =3D buflen - offset;
>>>>>>>> +       if ((xdr->page_len & 3) =3D=3D 0)
>>>>>>>> +               tail->iov_len -=3D sizeof(__be32);
>>>>>>>>=20
>>>>>>>>     xdr->buflen +=3D len;
>>>>>>>> }
>>>>>>>>=20
>>>>>>=20
>>>>>> --
>>>>>> Chuck Lever
>>>>=20
>>>> --
>>>> Chuck Lever
>>=20
>> --
>> Chuck Lever

--
Chuck Lever