NFS: Allow multiple commit requests in flight per file

Allow synchronous RPC calls to wait for pending RPC calls to finish,
but also allow asynchronous ones to just fire off another commit.

With this patch, the xfstests generic/074 test completes in 226s
instead of 242s

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index e73693f75dee638af922b74d837e9058337f3191..14f77df79c2585b5a94c9167098d6d2faf487417 100644 (file)
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -721,14 +721,8 @@ static void nfs_direct_commit_complete(struct nfs_commit_data *data)
                nfs_direct_write_complete(dreq, data->inode);
 }
 
-static void nfs_direct_error_cleanup(struct nfs_inode *nfsi)
-{
-       /* There is no lock to clear */
-}
-
 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = {
        .completion = nfs_direct_commit_complete,
-       .error_cleanup = nfs_direct_error_cleanup,
 };
 
 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 93e236429c5d785a1711d643d0b4676dfe4396cf..e6ef80ec699c72d9d1f95e048bd612d2c90bcefa 100644 (file)
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -514,7 +514,7 @@ static void nfs_check_dirty_writeback(struct page *page,
         * so it will not block due to pages that will shortly be freeable.
         */
        nfsi = NFS_I(mapping->host);
-       if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
+       if (atomic_read(&nfsi->commit_info.rpcs_out)) {
                *writeback = true;
                return;
        }

diff --git a/fs/nfs/nfstrace.h b/fs/nfs/nfstrace.h
index 59f838cdc009307f539bb94ee75f3447bfd79a56..9f80a086b612a622b68ef0f21a51aa0117cf5371 100644 (file)
--- a/fs/nfs/nfstrace.h
+++ b/fs/nfs/nfstrace.h
@@ -39,7 +39,6 @@
                        { 1 << NFS_INO_INVALIDATING, "INVALIDATING" }, \
                        { 1 << NFS_INO_FLUSHING, "FLUSHING" }, \
                        { 1 << NFS_INO_FSCACHE, "FSCACHE" }, \
-                       { 1 << NFS_INO_COMMIT, "COMMIT" }, \
                        { 1 << NFS_INO_LAYOUTCOMMIT, "NEED_LAYOUTCOMMIT" }, \
                        { 1 << NFS_INO_LAYOUTCOMMITTING, "LAYOUTCOMMIT" })
 

diff --git a/fs/nfs/pnfs_nfs.c b/fs/nfs/pnfs_nfs.c
index 24655b807d442596e14c360b29c81d4cedca9ef0..3c8e3a44e6ead0533d88e783dc9e089990716c52 100644 (file)
--- a/fs/nfs/pnfs_nfs.c
+++ b/fs/nfs/pnfs_nfs.c
@@ -266,17 +266,14 @@ pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
                } else {
                        nfs_retry_commit(mds_pages, NULL, cinfo, 0);
                        pnfs_generic_retry_commit(cinfo, 0);
-                       cinfo->completion_ops->error_cleanup(NFS_I(inode));
                        return -ENOMEM;
                }
        }
 
        nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
 
-       if (nreq == 0) {
-               cinfo->completion_ops->error_cleanup(NFS_I(inode));
+       if (nreq == 0)
                goto out;
-       }
 
        atomic_add(nreq, &cinfo->mds->rpcs_out);
 

diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 0aa3e6b3db70ad1e7eb35efac0c391b106f7fbb7..ae29f082c9c23eba8f015b3722ad58005d53d494 100644 (file)
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -21,6 +21,8 @@
 #include <linux/nfs_page.h>
 #include <linux/backing-dev.h>
 #include <linux/export.h>
+#include <linux/freezer.h>
+#include <linux/wait.h>
 
 #include <asm/uaccess.h>
 
@@ -1535,27 +1537,29 @@ static void nfs_writeback_result(struct rpc_task *task,
        }
 }
 
+static int nfs_wait_atomic_killable(atomic_t *key)
+{
+       if (fatal_signal_pending(current))
+               return -ERESTARTSYS;
+       freezable_schedule_unsafe();
+       return 0;
+}
 
-static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
+static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
 {
-       int ret;
+       return wait_on_atomic_t(&cinfo->rpcs_out,
+                       nfs_wait_atomic_killable, TASK_KILLABLE);
+}
 
-       if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
-               return 1;
-       if (!may_wait)
-               return 0;
-       ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
-                               NFS_INO_COMMIT,
-                               nfs_wait_bit_killable,
-                               TASK_KILLABLE);
-       return (ret < 0) ? ret : 1;
+static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
+{
+       atomic_inc(&cinfo->rpcs_out);
 }
 
-static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
+static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
 {
-       clear_bit(NFS_INO_COMMIT, &nfsi->flags);
-       smp_mb__after_atomic();
-       wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
+       if (atomic_dec_and_test(&cinfo->rpcs_out))
+               wake_up_atomic_t(&cinfo->rpcs_out);
 }
 
 void nfs_commitdata_release(struct nfs_commit_data *data)
@@ -1693,7 +1697,6 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how,
                                   data->mds_ops, how, 0);
  out_bad:
        nfs_retry_commit(head, NULL, cinfo, 0);
-       cinfo->completion_ops->error_cleanup(NFS_I(inode));
        return -ENOMEM;
 }
 
@@ -1755,8 +1758,7 @@ static void nfs_commit_release_pages(struct nfs_commit_data *data)
                clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
 
        nfs_init_cinfo(&cinfo, data->inode, data->dreq);
-       if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
-               nfs_commit_clear_lock(NFS_I(data->inode));
+       nfs_commit_end(cinfo.mds);
 }
 
 static void nfs_commit_release(void *calldata)
@@ -1775,7 +1777,6 @@ static const struct rpc_call_ops nfs_commit_ops = {
 
 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
        .completion = nfs_commit_release_pages,
-       .error_cleanup = nfs_commit_clear_lock,
 };
 
 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
@@ -1794,30 +1795,25 @@ int nfs_commit_inode(struct inode *inode, int how)
        LIST_HEAD(head);
        struct nfs_commit_info cinfo;
        int may_wait = how & FLUSH_SYNC;
+       int error = 0;
        int res;
 
-       res = nfs_commit_set_lock(NFS_I(inode), may_wait);
-       if (res <= 0)
-               goto out_mark_dirty;
        nfs_init_cinfo_from_inode(&cinfo, inode);
+       nfs_commit_begin(cinfo.mds);
        res = nfs_scan_commit(inode, &head, &cinfo);
-       if (res) {
-               int error;
-
+       if (res)
                error = nfs_generic_commit_list(inode, &head, how, &cinfo);
-               if (error < 0)
-                       return error;
-               if (!may_wait)
-                       goto out_mark_dirty;
-               error = wait_on_bit_action(&NFS_I(inode)->flags,
-                               NFS_INO_COMMIT,
-                               nfs_wait_bit_killable,
-                               TASK_KILLABLE);
-               if (error < 0)
-                       return error;
-       } else
-               nfs_commit_clear_lock(NFS_I(inode));
+       nfs_commit_end(cinfo.mds);
+       if (error < 0)
+               goto out_error;
+       if (!may_wait)
+               goto out_mark_dirty;
+       error = wait_on_commit(cinfo.mds);
+       if (error < 0)
+               return error;
        return res;
+out_error:
+       res = error;
        /* Note: If we exit without ensuring that the commit is complete,
         * we must mark the inode as dirty. Otherwise, future calls to
         * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure

diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
index c0e961474a527058c8d1ac2aa72070c9b15e3db5..ebf0bd72a42bde532394209a4ecfa45d69a8dadc 100644 (file)
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -216,7 +216,6 @@ struct nfs_inode {
 #define NFS_INO_FLUSHING       (4)             /* inode is flushing out data */
 #define NFS_INO_FSCACHE                (5)             /* inode can be cached by FS-Cache */
 #define NFS_INO_FSCACHE_LOCK   (6)             /* FS-Cache cookie management lock */
-#define NFS_INO_COMMIT         (7)             /* inode is committing unstable writes */
 #define NFS_INO_LAYOUTCOMMIT   (9)             /* layoutcommit required */
 #define NFS_INO_LAYOUTCOMMITTING (10)          /* layoutcommit inflight */
 #define NFS_INO_LAYOUTSTATS    (11)            /* layoutstats inflight */

diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index e89dbb14138c260f1a130ed20d25195469dfeca4..a8905b7d4d7f7795ba87064b93b1a2d8f5cb3980 100644 (file)
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -1423,7 +1423,6 @@ struct nfs_mds_commit_info {
 struct nfs_commit_data;
 struct nfs_inode;
 struct nfs_commit_completion_ops {
-       void (*error_cleanup) (struct nfs_inode *nfsi);
        void (*completion) (struct nfs_commit_data *data);
 };