If initiate_file_draining returned NFS4ERR_DELAY, all the lsegs of
a file might be released before the retrying cb_layout request arriving
at the client. In this situation,the list lo->plh_segs is empty,
layoutget request of the file will use open stateid to obtain a new
layout stateid. And if the retrying cb_layout request arrived at the
client after the layoutget reply, new layout stateid would be overwrite
by one out of date.
Signed-off-by: shaobingqing <[email protected]>
---
fs/nfs/callback_proc.c | 9 +++++++--
fs/nfs/pnfs.c | 3 ++-
fs/nfs/pnfs.h | 1 +
3 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index ae2e87b..5b20ace 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -170,10 +170,15 @@ static u32 initiate_file_draining(struct nfs_client *clp,
spin_lock(&ino->i_lock);
if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags) ||
pnfs_mark_matching_lsegs_invalid(lo, &free_me_list,
- &args->cbl_range))
+ &args->cbl_range)) {
+ if (!test_and_set_bit(NFS_LAYOUT_BC_DELAY, &lo->plh_flags))
+ pnfs_get_layout_hdr(lo);
rv = NFS4ERR_DELAY;
- else
+ } else {
+ if (test_and_clear_bit(NFS_LAYOUT_BC_DELAY, &lo->plh_flags))
+ pnfs_put_layout_hdr(lo);
rv = NFS4ERR_NOMATCHING_LAYOUT;
+ }
pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
spin_unlock(&ino->i_lock);
pnfs_free_lseg_list(&free_me_list);
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index cb34024..998bc82 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -721,7 +721,8 @@ pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
status = -EAGAIN;
} else if (!nfs4_valid_open_stateid(open_state)) {
status = -EBADF;
- } else if (list_empty(&lo->plh_segs)) {
+ } else if (list_empty(&lo->plh_segs) &&
+ !test_bit(NFS_LAYOUT_BC_DELAY, &lo->plh_flags)) {
int seq;
do {
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index a4f4181..fb9b37e 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -63,6 +63,7 @@ enum {
NFS_LAYOUT_BULK_RECALL, /* bulk recall affecting layout */
NFS_LAYOUT_ROC, /* some lseg had roc bit set */
NFS_LAYOUT_RETURN, /* Return this layout ASAP */
+ NFS_LAYOUT_BC_DELAY, /*cb_layout are delayed*/
};
enum layoutdriver_policy_flags {
--
1.7.4.2
On Feb 17, 2014, at 2:08, shaobingqing <[email protected]> wrote:
> If initiate_file_draining returned NFS4ERR_DELAY, all the lsegs of
> a file might be released before the retrying cb_layout request arriving
> at the client. In this situation,the list lo->plh_segs is empty,
> layoutget request of the file will use open stateid to obtain a new
> layout stateid. And if the retrying cb_layout request arrived at the
> client after the layoutget reply, new layout stateid would be overwrite
> by one out of date.
I have very serious doubts about adding layout state to represent what the server is _trying_ to do. What if the server decides not to retry the recall? What if it reboots?
Let?s start with looking at the motivations: Have you ever run into the above problem in practice? If so, what was the effect on overall performance?
_________________________________
Trond Myklebust
Linux NFS client maintainer, PrimaryData
[email protected]