Re: [PATCH v4 07/12] NFS add support for asynchronous COPY

[Olga Kornievskaia <aglo@umich.edu>]

On Thu, Sep 28, 2017 at 3:33 PM, Anna Schumaker
<schumaker.anna@gmail.com> wrote:
> Hi Olga,
>
> On 09/28/2017 01:28 PM, Olga Kornievskaia wrote:
>> Change xdr to always send COPY asynchronously.
>>
>> Keep the list copies send in a list under a server structure.
>> Once copy is sent, it waits on a completion structure that will
>> be signalled by the callback thread that receives CB_OFFLOAD.
>>
>> Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
>> ---
>>  fs/nfs/callback_proc.c    | 38 ++++++++++++++++++++++++++++++-
>>  fs/nfs/client.c           |  1 +
>>  fs/nfs/nfs42proc.c        | 57 ++++++++++++++++++++++++++++++++++++++++++-----
>>  fs/nfs/nfs42xdr.c         |  8 ++++---
>>  include/linux/nfs_fs.h    |  9 ++++++++
>>  include/linux/nfs_fs_sb.h |  1 +
>>  include/linux/nfs_xdr.h   |  1 +
>>  7 files changed, 106 insertions(+), 9 deletions(-)
>>
>> diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
>> index eef2728..d3e7b61 100644
>> --- a/fs/nfs/callback_proc.c
>> +++ b/fs/nfs/callback_proc.c
>> @@ -657,9 +657,45 @@ __be32 nfs4_callback_notify_lock(void *argp, void *resp,
>>  }
>>  #endif /* CONFIG_NFS_V4_1 */
>>  #ifdef CONFIG_NFS_V4_2
>> -__be32 nfs4_callback_offload(void *args, void *dummy,
>> +static void nfs4_copy_cb_args(struct nfs4_copy_state *cp_state,
>> +                             struct cb_offloadargs *args)
>> +{
>> +     cp_state->count = args->wr_count;
>> +     cp_state->error = args->error;
>> +     if (!args->error) {
>> +             cp_state->verf.committed = args->wr_writeverf.committed;
>> +             memcpy(&cp_state->verf.verifier.data[0],
>> +                     &args->wr_writeverf.verifier.data[0],
>> +                     NFS4_VERIFIER_SIZE);
>> +     }
>> +}
>> +
>> +__be32 nfs4_callback_offload(void *data, void *dummy,
>>                            struct cb_process_state *cps)
>>  {
>> +     struct cb_offloadargs *args = data;
>> +     struct nfs_server *server;
>> +     struct nfs4_copy_state *copy;
>> +
>> +     rcu_read_lock();
>> +     list_for_each_entry_rcu(server, &cps->clp->cl_superblocks,
>> +                             client_link) {
>> +             spin_lock(&server->nfs_client->cl_lock);
>> +             list_for_each_entry(copy, &server->ss_copies, copies) {
>> +                     if (memcmp(args->coa_stateid.other,
>> +                                     copy->stateid.other,
>> +                                     sizeof(args->coa_stateid.other)))
>> +                             continue;
>> +                     nfs4_copy_cb_args(copy, args);
>> +                     complete(&copy->completion);
>> +                     spin_unlock(&server->nfs_client->cl_lock);
>> +                     goto out;
>> +             }
>> +             spin_unlock(&server->nfs_client->cl_lock);
>> +     }
>> +out:
>> +     rcu_read_unlock();
>> +
>>       return 0;
>>  }
>>  #endif /* CONFIG_NFS_V4_2 */
>> diff --git a/fs/nfs/client.c b/fs/nfs/client.c
>> index efebe6c..7b338be 100644
>> --- a/fs/nfs/client.c
>> +++ b/fs/nfs/client.c
>> @@ -876,6 +876,7 @@ struct nfs_server *nfs_alloc_server(void)
>>       INIT_LIST_HEAD(&server->delegations);
>>       INIT_LIST_HEAD(&server->layouts);
>>       INIT_LIST_HEAD(&server->state_owners_lru);
>> +     INIT_LIST_HEAD(&server->ss_copies);
>>
>>       atomic_set(&server->active, 0);
>>
>> diff --git a/fs/nfs/nfs42proc.c b/fs/nfs/nfs42proc.c
>> index 5c42e71..c8ca353 100644
>> --- a/fs/nfs/nfs42proc.c
>> +++ b/fs/nfs/nfs42proc.c
>> @@ -129,6 +129,39 @@ int nfs42_proc_deallocate(struct file *filep, loff_t offset, loff_t len)
>>       return err;
>>  }
>>
>> +static int handle_async_copy(struct nfs42_copy_res *res,
>> +                          struct nfs_server *server,
>> +                          struct file *src,
>> +                          struct file *dst,
>> +                          nfs4_stateid *src_stateid,
>> +                          uint64_t *ret_count)
>> +{
>> +     struct nfs4_copy_state *copy;
>> +     int status;> +
>> +     copy = kzalloc(sizeof(struct nfs4_copy_state), GFP_NOFS);
>> +     if (!copy)
>> +             return -ENOMEM;
>> +     memcpy(&copy->stateid, &res->write_res.stateid, NFS4_STATEID_SIZE);
>> +     init_completion(&copy->completion);
>> +
>> +     spin_lock(&server->nfs_client->cl_lock);
>> +     list_add_tail(&copy->copies, &server->ss_copies);
>> +     spin_unlock(&server->nfs_client->cl_lock);
>> +
>> +     wait_for_completion_interruptible(&copy->completion);
>> +     spin_lock(&server->nfs_client->cl_lock);
>> +     list_del_init(&copy->copies);
>> +     spin_unlock(&server->nfs_client->cl_lock);
>> +     *ret_count = copy->count;
>> +     memcpy(&res->write_res.verifier, &copy->verf, sizeof(copy->verf));
>> +     if (copy->count <= 0)
>> +             status = -copy->error;
>
> If copy->count > 0 then status won't be set before this function returns.  GCC complains about this when I try compiling.

Got it. I'll initialize it.

>
> Anna
>
>> +
>> +     kfree(copy);
>> +     return status;
>> +}
>> +
>>  static ssize_t _nfs42_proc_copy(struct file *src,
>>                               struct nfs_lock_context *src_lock,
>>                               struct file *dst,
>> @@ -167,9 +200,13 @@ static ssize_t _nfs42_proc_copy(struct file *src,
>>       if (status)
>>               return status;
>>
>> -     res->commit_res.verf = kzalloc(sizeof(struct nfs_writeverf), GFP_NOFS);
>> -     if (!res->commit_res.verf)
>> -             return -ENOMEM;
>> +     res->commit_res.verf = NULL;
>> +     if (args->sync) {
>> +             res->commit_res.verf =
>> +                     kzalloc(sizeof(struct nfs_writeverf), GFP_NOFS);
>> +             if (!res->commit_res.verf)
>> +                     return -ENOMEM;
>> +     }
>>       status = nfs4_call_sync(server->client, server, &msg,
>>                               &args->seq_args, &res->seq_res, 0);
>>       if (status == -ENOTSUPP)
>> @@ -177,18 +214,27 @@ static ssize_t _nfs42_proc_copy(struct file *src,
>>       if (status)
>>               goto out;
>>
>> -     if (nfs_write_verifier_cmp(&res->write_res.verifier.verifier,
>> +     if (args->sync &&
>> +             !nfs_write_verifier_cmp(&res->write_res.verifier.verifier,
>>                                   &res->commit_res.verf->verifier)) {
>>               status = -EAGAIN;
>>               goto out;
>>       }
>>
>> +     if (!res->synchronous) {
>> +             status = handle_async_copy(res, server, src, dst,
>> +                             &args->src_stateid, &res->write_res.count);
>> +             if (status)
>> +                     return status;
>> +     }
>> +
>>       truncate_pagecache_range(dst_inode, pos_dst,
>>                                pos_dst + res->write_res.count);
>>
>>       status = res->write_res.count;
>>  out:
>> -     kfree(res->commit_res.verf);
>> +     if (args->sync)
>> +             kfree(res->commit_res.verf);
>>       return status;
>>  }
>>
>> @@ -205,6 +251,7 @@ ssize_t nfs42_proc_copy(struct file *src, loff_t pos_src,
>>               .dst_fh         = NFS_FH(file_inode(dst)),
>>               .dst_pos        = pos_dst,
>>               .count          = count,
>> +             .sync           = false,
>>       };
>>       struct nfs42_copy_res res;
>>       struct nfs4_exception src_exception = {
>> diff --git a/fs/nfs/nfs42xdr.c b/fs/nfs/nfs42xdr.c
>> index 9e557ad..2c37c66 100644
>> --- a/fs/nfs/nfs42xdr.c
>> +++ b/fs/nfs/nfs42xdr.c
>> @@ -160,7 +160,7 @@ static void encode_copy(struct xdr_stream *xdr,
>>       encode_uint64(xdr, args->count);
>>
>>       encode_uint32(xdr, 1); /* consecutive = true */
>> -     encode_uint32(xdr, 1); /* synchronous = true */
>> +     encode_uint32(xdr, args->sync);
>>       encode_uint32(xdr, 0); /* src server list */
>>  }
>>
>> @@ -291,7 +291,8 @@ static void nfs4_xdr_enc_copy(struct rpc_rqst *req,
>>       encode_savefh(xdr, &hdr);
>>       encode_putfh(xdr, args->dst_fh, &hdr);
>>       encode_copy(xdr, args, &hdr);
>> -     encode_copy_commit(xdr, args, &hdr);
>> +     if (args->sync)
>> +             encode_copy_commit(xdr, args, &hdr);
>>       encode_nops(&hdr);
>>  }
>>
>> @@ -613,7 +614,8 @@ static int nfs4_xdr_dec_copy(struct rpc_rqst *rqstp,
>>       status = decode_copy(xdr, res);
>>       if (status)
>>               goto out;
>> -     status = decode_commit(xdr, &res->commit_res);
>> +     if (res->commit_res.verf)
>> +             status = decode_commit(xdr, &res->commit_res);
>>  out:
>>       return status;
>>  }
>> diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
>> index a0282ce..580a8d9 100644
>> --- a/include/linux/nfs_fs.h
>> +++ b/include/linux/nfs_fs.h
>> @@ -183,6 +183,15 @@ struct nfs_inode {
>>       struct inode            vfs_inode;
>>  };
>>
>> +struct nfs4_copy_state {
>> +     struct list_head        copies;
>> +     nfs4_stateid            stateid;
>> +     struct completion       completion;
>> +     uint64_t                count;
>> +     struct nfs_writeverf    verf;
>> +     int                     error;
>> +};
>> +
>>  /*
>>   * Cache validity bit flags
>>   */
>> diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
>> index de1eafe..511eefb 100644
>> --- a/include/linux/nfs_fs_sb.h
>> +++ b/include/linux/nfs_fs_sb.h
>> @@ -206,6 +206,7 @@ struct nfs_server {
>>       struct list_head        state_owners_lru;
>>       struct list_head        layouts;
>>       struct list_head        delegations;
>> +     struct list_head        ss_copies;
>>
>>       unsigned long           mig_gen;
>>       unsigned long           mig_status;
>> diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
>> index 784ec8d..020d958 100644
>> --- a/include/linux/nfs_xdr.h
>> +++ b/include/linux/nfs_xdr.h
>> @@ -1384,6 +1384,7 @@ struct nfs42_copy_args {
>>       u64                             dst_pos;
>>
>>       u64                             count;
>> +     bool                            sync;
>>  };
>>
>>  struct nfs42_write_res {
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html