[v2] ceph: implement updated ceph_mds_request_head structure

When we added the btime feature in mainline ceph, we had to extend
struct ceph_mds_request_args so that it could be set. Implement the same
in the kernel client.

Rename ceph_mds_request_head with a _old extension, and a union
ceph_mds_request_args_ext to allow for the extended size of the new
header format.

Add the appropriate code to handle both formats in struct
create_request_message and key the behavior on whether the peer supports
CEPH_FEATURE_FS_BTIME.

The gid_list field in the payload is now populated from the saved
credential. For now, we don't add any support for setting the btime via
setattr, but this does enable us to add that in the future.

[ idryomov: break unnecessarily long lines ]

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 fs/ceph/mds_client.c         | 101 ++++++++++++++++++++++++++---------
 include/linux/ceph/ceph_fs.h |  32 ++++++++++-
 2 files changed, 108 insertions(+), 25 deletions(-)

 v2: fix encoding of unsafe request resends
     add encode_payload_tail helper
     rework find_old_request_head to take a "legacy" flag argument
     
Ilya,

I'll go ahead and merge this into testing, but your call on whether we
should take this for v5.11, or wait for v5.12. We don't have anything
blocked on this just yet.

I dropped your SoB and Xiubo Reviewed-by tags as well, as the patch is
a bit different from the original.

On 2020/12/17 5:04, Jeff Layton wrote:
> On Wed, 2020-12-16 at 21:16 +0100, Ilya Dryomov wrote:

>> On Wed, Dec 16, 2020 at 8:50 PM Jeff Layton <jlayton@kernel.org> wrote:

>>> When we added the btime feature in mainline ceph, we had to extend

>>> struct ceph_mds_request_args so that it could be set. Implement the same

>>> in the kernel client.

>>>

>>> Rename ceph_mds_request_head with a _old extension, and a union

>>> ceph_mds_request_args_ext to allow for the extended size of the new

>>> header format.

>>>

>>> Add the appropriate code to handle both formats in struct

>>> create_request_message and key the behavior on whether the peer supports

>>> CEPH_FEATURE_FS_BTIME.

>>>

>>> The gid_list field in the payload is now populated from the saved

>>> credential. For now, we don't add any support for setting the btime via

>>> setattr, but this does enable us to add that in the future.

>>>

>>> [ idryomov: break unnecessarily long lines ]

>>>

>>> Signed-off-by: Jeff Layton <jlayton@kernel.org>

>>> ---

>>>   fs/ceph/mds_client.c         | 101 ++++++++++++++++++++++++++---------

>>>   include/linux/ceph/ceph_fs.h |  32 ++++++++++-

>>>   2 files changed, 108 insertions(+), 25 deletions(-)

>>>

>>>   v2: fix encoding of unsafe request resends

>>>       add encode_payload_tail helper

>>>       rework find_old_request_head to take a "legacy" flag argument

>>>

>>> Ilya,

>>>

>>> I'll go ahead and merge this into testing, but your call on whether we

>>> should take this for v5.11, or wait for v5.12. We don't have anything

>>> blocked on this just yet.

>>>

>>> I dropped your SoB and Xiubo Reviewed-by tags as well, as the patch is

>>> a bit different from the original.

>>>

>>> diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c

>>> index 0ff76f21466a..cd0cc5d8c4f0 100644

>>> --- a/fs/ceph/mds_client.c

>>> +++ b/fs/ceph/mds_client.c

>>> @@ -2475,15 +2475,46 @@ static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,

>>>          return r;

>>>   }

>>>

>>> +static struct ceph_mds_request_head_old *

>>> +find_old_request_head(void *p, bool legacy)

>>> +{

>>> +       struct ceph_mds_request_head *new_head;

>>> +

>>> +       if (legacy)

>>> +               return (struct ceph_mds_request_head_old *)p;

>>> +       new_head = (struct ceph_mds_request_head *)p;

>>> +       return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;

>>> +}

>>> +

>>> +static void encode_payload_tail(void **p, struct ceph_mds_request *req, bool legacy)

>>> +{

>>> +       struct ceph_timespec ts;

>>> +

>>> +       ceph_encode_timespec64(&ts, &req->r_stamp);

>>> +       ceph_encode_copy(p, &ts, sizeof(ts));

>>> +

>>> +       /* gid list */

>>> +       if (!legacy) {

>>> +               int i;

>>> +

>>> +               ceph_encode_32(p, req->r_cred->group_info->ngroups);

>>> +               for (i = 0; i < req->r_cred->group_info->ngroups; i++)

>>> +                       ceph_encode_64(p, from_kgid(&init_user_ns,

>>> +                                      req->r_cred->group_info->gid[i]));

>>> +       }

>>> +}

>>> +

>>>   /*

>>>    * called under mdsc->mutex

>>>    */

>>> -static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,

>>> +static struct ceph_msg *create_request_message(struct ceph_mds_session *session,

>>>                                                 struct ceph_mds_request *req,

>>> -                                              int mds, bool drop_cap_releases)

>>> +                                              bool drop_cap_releases)

>>>   {

>>> +       int mds = session->s_mds;

>>> +       struct ceph_mds_client *mdsc = session->s_mdsc;

>>>          struct ceph_msg *msg;

>>> -       struct ceph_mds_request_head *head;

>>> +       struct ceph_mds_request_head_old *head;

>>>          const char *path1 = NULL;

>>>          const char *path2 = NULL;

>>>          u64 ino1 = 0, ino2 = 0;

>>> @@ -2493,6 +2524,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,

>>>          u16 releases;

>>>          void *p, *end;

>>>          int ret;

>>> +       bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);

>>>

>>>          ret = set_request_path_attr(req->r_inode, req->r_dentry,

>>>                                req->r_parent, req->r_path1, req->r_ino1.ino,

>>> @@ -2514,14 +2546,23 @@ static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,

>>>                  goto out_free1;

>>>          }

>>>

>>> -       len = sizeof(*head) +

>>> -               pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +

>>> +       if (legacy) {

>>> +               /* Old style */

>>> +               len = sizeof(*head);

>>> +       } else {

>>> +               /* New style: add gid_list and any later fields */

>>> +               len = sizeof(struct ceph_mds_request_head) + sizeof(u32) +

>>> +                     (sizeof(u64) * req->r_cred->group_info->ngroups);

>>> +       }

>>> +

>>> +       len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +

>>>                  sizeof(struct ceph_timespec);

>>>

>>>          /* calculate (max) length for cap releases */

>>>          len += sizeof(struct ceph_mds_request_release) *

>>>                  (!!req->r_inode_drop + !!req->r_dentry_drop +

>>>                   !!req->r_old_inode_drop + !!req->r_old_dentry_drop);

>>> +

>>>          if (req->r_dentry_drop)

>>>                  len += pathlen1;

>>>          if (req->r_old_dentry_drop)

>>> @@ -2533,11 +2574,25 @@ static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,

>>>                  goto out_free2;

>>>          }

>>>

>>> -       msg->hdr.version = cpu_to_le16(3);

>>>          msg->hdr.tid = cpu_to_le64(req->r_tid);

>>>

>>> -       head = msg->front.iov_base;

>>> -       p = msg->front.iov_base + sizeof(*head);

>>> +       /*

>>> +        * The old ceph_mds_request_header didn't contain a version field, and

>>> +        * one was added when we moved the message version from 3->4.

>>> +        */

>>> +       if (legacy) {

>>> +               msg->hdr.version = cpu_to_le16(3);

>>> +               p = msg->front.iov_base + sizeof(*head);

>>> +       } else {

>>> +               struct ceph_mds_request_head *new_head = msg->front.iov_base;

>>> +

>>> +               msg->hdr.version = cpu_to_le16(4);

>>> +               new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);

>>> +               p = msg->front.iov_base + sizeof(*new_head);

>>> +       }

>>> +

>>> +       head = find_old_request_head(msg->front.iov_base, legacy);

>>> +

>>>          end = msg->front.iov_base + msg->front.iov_len;

>>>

>>>          head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);

>>> @@ -2583,12 +2638,7 @@ static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,

>>>

>>>          head->num_releases = cpu_to_le16(releases);

>>>

>>> -       /* time stamp */

>>> -       {

>>> -               struct ceph_timespec ts;

>>> -               ceph_encode_timespec64(&ts, &req->r_stamp);

>>> -               ceph_encode_copy(&p, &ts, sizeof(ts));

>>> -       }

>>> +       encode_payload_tail(&p, req, legacy);

>>>

>>>          if (WARN_ON_ONCE(p > end)) {

>>>                  ceph_msg_put(msg);

>>> @@ -2642,9 +2692,10 @@ static int __prepare_send_request(struct ceph_mds_session *session,

>>>   {

>>>          int mds = session->s_mds;

>>>          struct ceph_mds_client *mdsc = session->s_mdsc;

>>> -       struct ceph_mds_request_head *rhead;

>>> +       struct ceph_mds_request_head_old *rhead;

>>>          struct ceph_msg *msg;

>>>          int flags = 0;

>>> +       bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);

>>>

>>>          req->r_attempts++;

>>>          if (req->r_inode) {

>>> @@ -2661,6 +2712,7 @@ static int __prepare_send_request(struct ceph_mds_session *session,

>>>

>>>          if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {

>>>                  void *p;

>>> +

>>>                  /*

>>>                   * Replay.  Do not regenerate message (and rebuild

>>>                   * paths, etc.); just use the original message.

>>> @@ -2668,7 +2720,7 @@ static int __prepare_send_request(struct ceph_mds_session *session,

>>>                   * d_move mangles the src name.

>>>                   */

>>>                  msg = req->r_request;

>>> -               rhead = msg->front.iov_base;

>>> +               rhead = find_old_request_head(msg->front.iov_base, legacy);

>>>

>>>                  flags = le32_to_cpu(rhead->flags);

>>>                  flags |= CEPH_MDS_FLAG_REPLAY;

>>> @@ -2682,13 +2734,14 @@ static int __prepare_send_request(struct ceph_mds_session *session,

>>>                  /* remove cap/dentry releases from message */

>>>                  rhead->num_releases = 0;

>>>

>>> -               /* time stamp */

>>> +               /* verify that we haven't got mixed-feature MDSs */

>>> +               if (legacy)

>>> +                       WARN_ON_ONCE(le16_to_cpu(msg->hdr.version) >= 4);

>>> +               else

>>> +                       WARN_ON_ONCE(le16_to_cpu(msg->hdr.version) < 4);

>> As mentioned in the ticket, I already did a minimal encode_gid_list()

>> helper -- just haven't pushed because I had to leave.  Looking at this,

>> I not clear on why even bother with v3 vs v4 and the legacy branch.

>> The only that is conditional on CEPH_FEATURE_FS_BTIME is the head, we

>> can encode gids and set the version to 4 unconditionally.

>>

> We can't because legacy servers won't understand the newer struct

> ceph_mds_request_head. I'm pretty sure anything pre-luminous will barf

> on it, which is why we needed the feature bit test.

>

Yeah, agree. For the old ceph versions it may something like dropping 
the message to ground or crash the mds deamons as I hit before for the 
metric features.