NFS/pNFS: Fix up pNFS write reschedule layering violations and bugs
The flexfiles layout in particular, seems to want to poke around in the
O_DIRECT flags when retransmitting.
This patch sets up an interface to allow it to call back into O_DIRECT
to handle retransmission correctly. It also fixes a potential bug whereby
we could change the behaviour of O_DIRECT if an error is already pending.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index 4b1d08f..e73693f 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -117,12 +117,6 @@
return atomic_dec_and_test(&dreq->io_count);
}
-void nfs_direct_set_resched_writes(struct nfs_direct_req *dreq)
-{
- dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
-}
-EXPORT_SYMBOL_GPL(nfs_direct_set_resched_writes);
-
static void
nfs_direct_good_bytes(struct nfs_direct_req *dreq, struct nfs_pgio_header *hdr)
{
@@ -839,10 +833,25 @@
}
}
+static void nfs_direct_write_reschedule_io(struct nfs_pgio_header *hdr)
+{
+ struct nfs_direct_req *dreq = hdr->dreq;
+
+ spin_lock(&dreq->lock);
+ if (dreq->error == 0) {
+ dreq->flags = NFS_ODIRECT_RESCHED_WRITES;
+ /* fake unstable write to let common nfs resend pages */
+ hdr->verf.committed = NFS_UNSTABLE;
+ hdr->good_bytes = hdr->args.count;
+ }
+ spin_unlock(&dreq->lock);
+}
+
static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = {
.error_cleanup = nfs_write_sync_pgio_error,
.init_hdr = nfs_direct_pgio_init,
.completion = nfs_direct_write_completion,
+ .reschedule_io = nfs_direct_write_reschedule_io,
};
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
index 03516c8..df475d4 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -912,18 +912,7 @@
hdr->args.count,
(unsigned long long)hdr->args.offset);
- if (!hdr->dreq) {
- struct nfs_open_context *ctx;
-
- ctx = nfs_list_entry(hdr->pages.next)->wb_context;
- set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
- hdr->completion_ops->error_cleanup(&hdr->pages);
- } else {
- nfs_direct_set_resched_writes(hdr->dreq);
- /* fake unstable write to let common nfs resend pages */
- hdr->verf.committed = NFS_UNSTABLE;
- hdr->good_bytes = hdr->args.count;
- }
+ hdr->completion_ops->reschedule_io(hdr);
return;
}
diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
index 313d554..99a2919 100644
--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -519,7 +519,6 @@
inode_dio_wait(inode);
}
extern ssize_t nfs_dreq_bytes_left(struct nfs_direct_req *dreq);
-extern void nfs_direct_set_resched_writes(struct nfs_direct_req *dreq);
/* nfs4proc.c */
extern void __nfs4_read_done_cb(struct nfs_pgio_header *);
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 7b93164..0aa3e6b 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1326,9 +1326,15 @@
}
}
+static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
+{
+ nfs_async_write_error(&hdr->pages);
+}
+
static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
.error_cleanup = nfs_async_write_error,
.completion = nfs_write_completion,
+ .reschedule_io = nfs_async_write_reschedule_io,
};
void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,