On Thu, 20 Aug 2015 19:18:25 -0700
Peng Tao <bergwolf@xxxxxxxxxxxxxxx> wrote:
> On Thu, Aug 20, 2015 at 6:01 PM, Peng Tao <bergwolf@xxxxxxxxxxxxxxx> wrote:
> > On Thu, Aug 20, 2015 at 4:17 AM, Jeff Layton <jlayton@xxxxxxxxxxxxxxx> wrote:
> >> ...when there are open files to be closed.
> >>
> >> When knfsd does an fput(), it gets queued to a list and a workqueue job
> >> is then scheduled to do the actual __fput work. In the case of knfsd
> >> closing down the file prior to a REMOVE or RENAME, we really want to
> >> ensure that those files are closed prior to returning. When there are
> >> files to be closed, call flush_delayed_fput to ensure this.
> >>
> >> There are deadlock possibilities if you call flush_delayed_fput while
> >> holding locks, however. In the case of nfsd_rename, we don't even do the
> >> lookups of the dentries to be renamed until we've locked for rename.
> >>
> >> Once we've figured out what the target dentry is for a rename, check to
> >> see whether there are cached open files associated with it. If there
> >> are, then unwind all of the locking, close them all, and then reattempt
> >> the rename.
> >>
> >> Signed-off-by: Jeff Layton <jeff.layton@xxxxxxxxxxxxxxx>
> >> ---
> >> fs/file_table.c | 1 +
> >> fs/nfsd/filecache.c | 33 ++++++++++++++++++++++++++++++++-
> >> fs/nfsd/filecache.h | 1 +
> >> fs/nfsd/trace.h | 10 +++++++++-
> >> fs/nfsd/vfs.c | 47 +++++++++++++++++++++++++++++++++++++++--------
> >> 5 files changed, 82 insertions(+), 10 deletions(-)
> >>
> >> diff --git a/fs/file_table.c b/fs/file_table.c
> >> index 7f9d407c7595..33898e72618c 100644
> >> --- a/fs/file_table.c
> >> +++ b/fs/file_table.c
> >> @@ -257,6 +257,7 @@ void flush_delayed_fput(void)
> >> {
> >> delayed_fput(NULL);
> >> }
> >> +EXPORT_SYMBOL_GPL(flush_delayed_fput);
> >>
> >> static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
> >>
> >> diff --git a/fs/nfsd/filecache.c b/fs/nfsd/filecache.c
> >> index 4bd683f03b6e..b62942ba6e7b 100644
> >> --- a/fs/nfsd/filecache.c
> >> +++ b/fs/nfsd/filecache.c
> >> @@ -284,6 +284,34 @@ nfsd_file_find_locked(struct inode *inode, unsigned int may_flags,
> >> }
> >>
> >> /**
> >> + * nfsd_file_is_cached - are there any cached open files for this fh?
> >> + * @inode: inode of the file to check
> >> + *
> >> + * Scan the hashtable for open files that match this fh. Returns true if there
> >> + * are any, and false if not.
> >> + */
> >> +bool
> >> +nfsd_file_is_cached(struct inode *inode)
> >> +{
> >> + bool ret = false;
> >> + struct nfsd_file *nf;
> >> + unsigned int hashval = (unsigned int)hash_ptr(inode, NFSD_FILE_HASH_BITS);
> >> +
> >> + rcu_read_lock();
> >> + hlist_for_each_entry_rcu(nf, &nfsd_file_hashtbl[hashval].nfb_head,
> >> + nf_node) {
> >> + if (inode == nf->nf_inode) {
> >> + ret = true;
> >> + break;
> >> + }
> >> + }
> >> + rcu_read_unlock();
> >> + trace_nfsd_file_is_cached(hashval, inode, (int)ret);
> >> + return ret;
> >> +}
> >> +
> >> +
> >> +/**
> >> * nfsd_file_close_inode - attempt to forcibly close a nfsd_file
> >> * @inode: inode of the file to attempt to remove
> >> *
> >> @@ -305,7 +333,10 @@ nfsd_file_close_inode(struct inode *inode)
> >> }
> >> spin_unlock(&nfsd_file_hashtbl[hashval].nfb_lock);
> >> trace_nfsd_file_close_inode(hashval, inode, !list_empty(&dispose));
> >> - nfsd_file_dispose_list(&dispose);
> >> + if (!list_empty(&dispose)) {
> >> + nfsd_file_dispose_list(&dispose);
> >> + flush_delayed_fput();
> > It looks like flush_delayed_fput() is not exported symbol?
> >
> > And if flush_delayed_fput() is acceptable, it looks like __fput_sync()
> > is a better fit, because knfsd would not try to do all the delayed
> > fput() work, just the dispose list...
> oh, just saw that flush_delayed_fput() is exported in this patch!
> sorry for the noise. But I still think __fput_sync() might be a better
> fit, despite the assertion there... I'm fine with settling with
> flush_delayed_fput() though since calling __fput_sync() from a kernel
> thread might get more objections.
>
> Cheers,
> Tao
>
I looked at __fput_sync when I first rolled this patch, but it's a
little less convenient to use. __fput_sync is a synchronous analogue to
fput -- so you have to ensure that you use it instead of fput.
To make that work here, we'd need a separate set of destruction
routines that uses __fput_sync instead of fput. Certainly we can do
that if necessary, but I don't think it's really worth it.
The downside of course is that we might end up with this thread doing a
little extra __fput work if there happened to be other things queued
onto the delayed_fput_list, but I'm not too concerned about that.
> >
> > Cheers,
> > Tao
> >
> >> + }
> >> }
> >>
> >> __be32
> >> diff --git a/fs/nfsd/filecache.h b/fs/nfsd/filecache.h
> >> index 191cdb25aa66..4a873efb7953 100644
> >> --- a/fs/nfsd/filecache.h
> >> +++ b/fs/nfsd/filecache.h
> >> @@ -27,6 +27,7 @@ void nfsd_file_cache_shutdown(void);
> >> void nfsd_file_put(struct nfsd_file *nf);
> >> struct nfsd_file *nfsd_file_get(struct nfsd_file *nf);
> >> void nfsd_file_close_inode(struct inode *inode);
> >> +bool nfsd_file_is_cached(struct inode *inode);
> >> __be32 nfsd_file_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,
> >> unsigned int may_flags, struct nfsd_file **nfp);
> >> #endif /* _FS_NFSD_FILECACHE_H */
> >> diff --git a/fs/nfsd/trace.h b/fs/nfsd/trace.h
> >> index 95af3b9c7b66..fc6d8ee51a00 100644
> >> --- a/fs/nfsd/trace.h
> >> +++ b/fs/nfsd/trace.h
> >> @@ -140,7 +140,7 @@ TRACE_EVENT(nfsd_file_acquire,
> >> be32_to_cpu(__entry->status))
> >> );
> >>
> >> -TRACE_EVENT(nfsd_file_close_inode,
> >> +DECLARE_EVENT_CLASS(nfsd_file_search_class,
> >> TP_PROTO(unsigned int hash, struct inode *inode, int found),
> >> TP_ARGS(hash, inode, found),
> >> TP_STRUCT__entry(
> >> @@ -156,6 +156,14 @@ TRACE_EVENT(nfsd_file_close_inode,
> >> TP_printk("hash=0x%x inode=0x%p found=%d", __entry->hash,
> >> __entry->inode, __entry->found)
> >> );
> >> +
> >> +#define DEFINE_NFSD_FILE_SEARCH_EVENT(name) \
> >> +DEFINE_EVENT(nfsd_file_search_class, name, \
> >> + TP_PROTO(unsigned int hash, struct inode *inode, int found), \
> >> + TP_ARGS(hash, inode, found))
> >> +
> >> +DEFINE_NFSD_FILE_SEARCH_EVENT(nfsd_file_close_inode);
> >> +DEFINE_NFSD_FILE_SEARCH_EVENT(nfsd_file_is_cached);
> >> #endif /* _NFSD_TRACE_H */
> >>
> >> #undef TRACE_INCLUDE_PATH
> >> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> >> index 98d3b9d96480..4cc78a4ec694 100644
> >> --- a/fs/nfsd/vfs.c
> >> +++ b/fs/nfsd/vfs.c
> >> @@ -1592,6 +1592,17 @@ nfsd_close_cached_files(struct dentry *dentry)
> >> nfsd_file_close_inode(inode);
> >> }
> >>
> >> +static bool
> >> +nfsd_has_cached_files(struct dentry *dentry)
> >> +{
> >> + bool ret = false;
> >> + struct inode *inode = d_inode(dentry);
> >> +
> >> + if (inode && S_ISREG(inode->i_mode))
> >> + ret = nfsd_file_is_cached(inode);
> >> + return ret;
> >> +}
> >> +
> >> /*
> >> * Rename a file
> >> * N.B. After this call _both_ ffhp and tfhp need an fh_put
> >> @@ -1604,6 +1615,7 @@ nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
> >> struct inode *fdir, *tdir;
> >> __be32 err;
> >> int host_err;
> >> + bool has_cached = false;
> >>
> >> err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
> >> if (err)
> >> @@ -1622,6 +1634,7 @@ nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
> >> if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
> >> goto out;
> >>
> >> +retry:
> >> host_err = fh_want_write(ffhp);
> >> if (host_err) {
> >> err = nfserrno(host_err);
> >> @@ -1661,12 +1674,16 @@ nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
> >> if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
> >> goto out_dput_new;
> >>
> >> - nfsd_close_cached_files(ndentry);
> >> - host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
> >> - if (!host_err) {
> >> - host_err = commit_metadata(tfhp);
> >> - if (!host_err)
> >> - host_err = commit_metadata(ffhp);
> >> + if (nfsd_has_cached_files(ndentry)) {
> >> + has_cached = true;
> >> + goto out_dput_old;
> >> + } else {
> >> + host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
> >> + if (!host_err) {
> >> + host_err = commit_metadata(tfhp);
> >> + if (!host_err)
> >> + host_err = commit_metadata(ffhp);
> >> + }
> >> }
> >> out_dput_new:
> >> dput(ndentry);
> >> @@ -1679,12 +1696,26 @@ nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
> >> * as that would do the wrong thing if the two directories
> >> * were the same, so again we do it by hand.
> >> */
> >> - fill_post_wcc(ffhp);
> >> - fill_post_wcc(tfhp);
> >> + if (!has_cached) {
> >> + fill_post_wcc(ffhp);
> >> + fill_post_wcc(tfhp);
> >> + }
> >> unlock_rename(tdentry, fdentry);
> >> ffhp->fh_locked = tfhp->fh_locked = 0;
> >> fh_drop_write(ffhp);
> >>
> >> + /*
> >> + * If the target dentry has cached open files, then we need to try to
> >> + * close them prior to doing the rename. Flushing delayed fput
> >> + * shouldn't be done with locks held however, so we delay it until this
> >> + * point and then reattempt the whole shebang.
> >> + */
> >> + if (has_cached) {
> >> + has_cached = false;
> >> + nfsd_close_cached_files(ndentry);
> >> + dput(ndentry);
> >> + goto retry;
> >> + }
> >> out:
> >> return err;
> >> }
> >> --
> >> 2.4.3
> >>
> >> --
> >> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> >> the body of a message to majordomo@xxxxxxxxxxxxxxx
> >> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Jeff Layton <jlayton@xxxxxxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
