Merge tag 'nfs-for-3.7-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client updates from Trond Myklebust:
 "Features include:

   - Remove CONFIG_EXPERIMENTAL dependency from NFSv4.1
     Aside from the issues discussed at the LKS, distros are shipping
     NFSv4.1 with all the trimmings.
   - Fix fdatasync()/fsync() for the corner case of a server reboot.
   - NFSv4 OPEN access fix: finally distinguish correctly between
     open-for-read and open-for-execute permissions in all situations.
   - Ensure that the TCP socket is closed when we're in CLOSE_WAIT
   - More idmapper bugfixes
   - Lots of pNFS bugfixes and cleanups to remove unnecessary state and
     make the code easier to read.
   - In cases where a pNFS read or write fails, allow the client to
     resume trying layoutgets after two minutes of read/write-
     through-mds.
   - More net namespace fixes to the NFSv4 callback code.
   - More net namespace fixes to the NFSv3 locking code.
   - More NFSv4 migration preparatory patches.
     Including patches to detect network trunking in both NFSv4 and
     NFSv4.1
   - pNFS block updates to optimise LAYOUTGET calls."

* tag 'nfs-for-3.7-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (113 commits)
  pnfsblock: cleanup nfs4_blkdev_get
  NFS41: send real read size in layoutget
  NFS41: send real write size in layoutget
  NFS: track direct IO left bytes
  NFSv4.1: Cleanup ugliness in pnfs_layoutgets_blocked()
  NFSv4.1: Ensure that the layout sequence id stays 'close' to the current
  NFSv4.1: Deal with seqid wraparound in the pNFS return-on-close code
  NFSv4 set open access operation call flag in nfs4_init_opendata_res
  NFSv4.1: Remove the dependency on CONFIG_EXPERIMENTAL
  NFSv4 reduce attribute requests for open reclaim
  NFSv4: nfs4_open_done first must check that GETATTR decoded a file type
  NFSv4.1: Deal with wraparound when updating the layout "barrier" seqid
  NFSv4.1: Deal with wraparound issues when updating the layout stateid
  NFSv4.1: Always set the layout stateid if this is the first layoutget
  NFSv4.1: Fix another refcount issue in pnfs_find_alloc_layout
  NFSv4: don't put ACCESS in OPEN compound if O_EXCL
  NFSv4: don't check MAY_WRITE access bit in OPEN
  NFS: Set key construction data for the legacy upcall
  NFSv4.1: don't do two EXCHANGE_IDs on mount
  NFS: nfs41_walk_client_list(): re-lock before iterating
  ...

1 files changed, 29 insertions, 12 deletions

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index f692be97676d..582bb8866131 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -259,7 +259,7 @@ nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
         struct dentry *dentry = file->f_path.dentry;
         struct nfs_open_context *ctx = nfs_file_open_context(file);
         struct inode *inode = dentry->d_inode;
-        int have_error, status;
+        int have_error, do_resend, status;
         int ret = 0;
 
         dprintk("NFS: fsync file(%s/%s) datasync %d\n",
@@ -267,15 +267,23 @@ nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
                         datasync);
 
         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
+        do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
         have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
         status = nfs_commit_inode(inode, FLUSH_SYNC);
-        if (status >= 0 && ret < 0)
-                status = ret;
         have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
-        if (have_error)
+        if (have_error) {
                 ret = xchg(&ctx->error, 0);
-        if (!ret && status < 0)
+                if (ret)
+                        goto out;
+        }
+        if (status < 0) {
                 ret = status;
+                goto out;
+        }
+        do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
+        if (do_resend)
+                ret = -EAGAIN;
+out:
         return ret;
 }
 EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
@@ -286,13 +294,22 @@ nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
         int ret;
         struct inode *inode = file->f_path.dentry->d_inode;
 
-        ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
-        if (ret != 0)
-                goto out;
-        mutex_lock(&inode->i_mutex);
-        ret = nfs_file_fsync_commit(file, start, end, datasync);
-        mutex_unlock(&inode->i_mutex);
-out:
+        do {
+                ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
+                if (ret != 0)
+                        break;
+                mutex_lock(&inode->i_mutex);
+                ret = nfs_file_fsync_commit(file, start, end, datasync);
+                mutex_unlock(&inode->i_mutex);
+                /*
+                 * If nfs_file_fsync_commit detected a server reboot, then
+                 * resend all dirty pages that might have been covered by
+                 * the NFS_CONTEXT_RESEND_WRITES flag
+                 */
+                start = 0;
+                end = LLONG_MAX;
+        } while (ret == -EAGAIN);
+
         return ret;
 }