2011-04-12 21:27:37

by Weston Andros Adamson

Subject: [PATCH] Don't mark_inode_dirty_sync() while holding lock

mark_inode_dirty_sync() grabs the same inode lock!

race conditions between holding the lock in pnfs_set_layoutcommit() and in
mark_inode_dirty_sync() can result in a second call to pnfs_layoutcommit_inode(), but
this will be a noop as NFS_INO_LAYOUTCOMMIT won't be set in the second call
---
fs/nfs/pnfs.c | 8 +++++++-
1 files changed, 7 insertions(+), 1 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index d9ab972..d71997e 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1004,6 +1004,7 @@ pnfs_set_layoutcommit(struct nfs_write_data *wdata)
{
struct nfs_inode *nfsi = NFS_I(wdata->inode);
loff_t end_pos = wdata->args.offset + wdata->res.count;
+ int mark_as_dirty = false;

spin_lock(&nfsi->vfs_inode.i_lock);
if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
@@ -1011,13 +1012,18 @@ pnfs_set_layoutcommit(struct nfs_write_data *wdata)
get_lseg(wdata->lseg);
wdata->lseg->pls_lc_cred =
get_rpccred(wdata->args.context->state->owner->so_cred);
- mark_inode_dirty_sync(wdata->inode);
+ mark_as_dirty = true;
dprintk("%s: Set layoutcommit for inode %lu ",
__func__, wdata->inode->i_ino);
}
if (end_pos > wdata->lseg->pls_end_pos)
wdata->lseg->pls_end_pos = end_pos;
spin_unlock(&nfsi->vfs_inode.i_lock);
+
+ /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
+ * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
+ if (mark_as_dirty)
+ mark_inode_dirty_sync(wdata->inode);
}
EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

--
1.7.4.2

2011-04-12 22:23:47

by Myklebust, Trond

Subject: Re: [PATCH] Don't mark_inode_dirty_sync() while holding lock

On Tue, 2011-04-12 at 17:17 -0400, Weston Andros Adamson wrote:
> mark_inode_dirty_sync() grabs the same inode lock!
>
> race conditions between holding the lock in pnfs_set_layoutcommit() and in
> mark_inode_dirty_sync() can result in a second call to pnfs_layoutcommit_inode(), but
> this will be a noop as NFS_INO_LAYOUTCOMMIT won't be set in the second call


I'm missing a "Signed-off-by:" here.

> ---
> fs/nfs/pnfs.c | 8 +++++++-
> 1 files changed, 7 insertions(+), 1 deletions(-)
>
> diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
> index d9ab972..d71997e 100644
> --- a/fs/nfs/pnfs.c
> +++ b/fs/nfs/pnfs.c
> @@ -1004,6 +1004,7 @@ pnfs_set_layoutcommit(struct nfs_write_data *wdata)
> {
> struct nfs_inode *nfsi = NFS_I(wdata->inode);
> loff_t end_pos = wdata->args.offset + wdata->res.count;
> + int mark_as_dirty = false;

Strictly speaking, an 'int' does not take values in the set 'true' and
'false'.

> spin_lock(&nfsi->vfs_inode.i_lock);
> if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
> @@ -1011,13 +1012,18 @@ pnfs_set_layoutcommit(struct nfs_write_data *wdata)
> get_lseg(wdata->lseg);
> wdata->lseg->pls_lc_cred =
> get_rpccred(wdata->args.context->state->owner->so_cred);
> - mark_inode_dirty_sync(wdata->inode);
> + mark_as_dirty = true;
> dprintk("%s: Set layoutcommit for inode %lu ",
> __func__, wdata->inode->i_ino);
> }
> if (end_pos > wdata->lseg->pls_end_pos)
> wdata->lseg->pls_end_pos = end_pos;
> spin_unlock(&nfsi->vfs_inode.i_lock);
> +
> + /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
> + * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
> + if (mark_as_dirty)
> + mark_inode_dirty_sync(wdata->inode);
> }
> EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);

Otherwise this looks fine. Please correct and resend.

--
Trond Myklebust
Linux NFS client maintainer

NetApp
[email protected]
http://www.netapp.com