On Mon, 2025-02-24 at 22:09 +0000, Viacheslav Dubeyko wrote:
> On Mon, 2025-02-24 at 13:15 +1100, NeilBrown wrote:
> > On Fri, 21 Feb 2025, Viacheslav Dubeyko wrote:
> > > On Fri, 2025-02-21 at 10:36 +1100, NeilBrown wrote:
> > > > ceph already splices the correct dentry (in splice_dentry()) from the
> > > > result of mkdir but does nothing more with it.
> > > >
> > > > Now that ->mkdir can return a dentry, return the correct dentry.
> > > >
> > > > Signed-off-by: NeilBrown <neilb@xxxxxxx>
> > > > ---
> > > > fs/ceph/dir.c | 9 ++++++++-
> > > > 1 file changed, 8 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
> > > > index 39e0f240de06..c1a1c168bb27 100644
> > > > --- a/fs/ceph/dir.c
> > > > +++ b/fs/ceph/dir.c
> > > > @@ -1099,6 +1099,7 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
> > > > struct ceph_client *cl = mdsc->fsc->client;
> > > > struct ceph_mds_request *req;
> > > > struct ceph_acl_sec_ctx as_ctx = {};
> > > > + struct dentry *ret = NULL;
> > >
> > > I believe that it makes sense to initialize pointer by error here and always
> > > return ret as output. If something goes wrong in the logic, then we already have
> > > error.
> >
> > I'm not certain that I understand, but I have made a change which seems
> > to be consistent with the above and included it below. Please let me
> > know if it is what you intended.
> >
> > >
> > > > int err;
> > > > int op;
> > > >
> > > > @@ -1166,14 +1167,20 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
> > > > !req->r_reply_info.head->is_dentry)
> > > > err = ceph_handle_notrace_create(dir, dentry);
> > > > out_req:
> > > > + if (!err && req->r_dentry != dentry)
> > > > + /* Some other dentry was spliced in */
> > > > + ret = dget(req->r_dentry);
> > > > ceph_mdsc_put_request(req);
> > > > out:
> > > > if (!err)
> > > > + /* Should this use 'ret' ?? */
> > >
> > > Could we make a decision should or shouldn't? :)
> > > It looks not good to leave this comment instead of proper implementation. Do we
> > > have some obstacles to make this decision?
> >
> > I suspect we should use ret, but I didn't want to make a change which
> > wasn't directly required by my needed. So I highlighted this which
> > looks to me like a possible bug, hoping that someone more familiar with
> > the code would give an opinion. Do you agree that 'ret' (i.e.
> > ->r_dentry) should be used when ret is not NULL?
> >
>
> I think if we are going to return ret as a dentry, then it makes sense to call
> the ceph_init_inode_acls() for d_inode(ret). I don't see the point to call
> ceph_init_inode_acls() for d_inode(dentry) then.
>
My assumption when looking at this was that they should point to the
same inode. That said, working with d_inode(ret) after that point is
less confusing to the casual reader.
> > >
> > > > ceph_init_inode_acls(d_inode(dentry), &as_ctx);
> > > > else
> > > > d_drop(dentry);
> > > > ceph_release_acl_sec_ctx(&as_ctx);
> > > > - return ERR_PTR(err);
> > > > + if (err)
> > > > + return ERR_PTR(err);
> > > > + return ret;
> > >
> > > What's about this?
> > >
> > > return err ? ERR_PTR(err) : ret;
> >
> > We could do that, but you said above that you thought we should always
> > return 'ret' - which does make some sense.
> >
> > What do you think of the following alternate patch?
> >
>
> Patch looks good to me. Thanks.
>
> Reviewed-by: Viacheslav Dubeyko <Slava.Dubeyko@xxxxxxx>
>
> > Thanks,
> > NeilBrown
> >
> > diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
> > index 39e0f240de06..d2e5c557df83 100644
> > --- a/fs/ceph/dir.c
> > +++ b/fs/ceph/dir.c
> > @@ -1099,6 +1099,7 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
> > struct ceph_client *cl = mdsc->fsc->client;
> > struct ceph_mds_request *req;
> > struct ceph_acl_sec_ctx as_ctx = {};
> > + struct dentry *ret;
> > int err;
> > int op;
> >
> > @@ -1116,32 +1117,32 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
> > ceph_vinop(dir), dentry, dentry, mode);
> > op = CEPH_MDS_OP_MKDIR;
> > } else {
> > - err = -EROFS;
> > + ret = ERR_PTR(-EROFS);
> > goto out;
> > }
> >
> > if (op == CEPH_MDS_OP_MKDIR &&
> > ceph_quota_is_max_files_exceeded(dir)) {
> > - err = -EDQUOT;
> > + ret = ERR_PTR(-EDQUOT);
> > goto out;
> > }
> > if ((op == CEPH_MDS_OP_MKSNAP) && IS_ENCRYPTED(dir) &&
> > !fscrypt_has_encryption_key(dir)) {
> > - err = -ENOKEY;
> > + ret = ERR_PTR(-ENOKEY);
> > goto out;
> > }
> >
> >
> > req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
> > if (IS_ERR(req)) {
> > - err = PTR_ERR(req);
> > + ret = ERR_CAST(req);
> > goto out;
> > }
> >
> > mode |= S_IFDIR;
> > req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx);
> > if (IS_ERR(req->r_new_inode)) {
> > - err = PTR_ERR(req->r_new_inode);
> > + ret = ERR_CAST(req->r_new_inode);
> > req->r_new_inode = NULL;
> > goto out_req;
> > }
> > @@ -1165,15 +1166,23 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
> > !req->r_reply_info.head->is_target &&
> > !req->r_reply_info.head->is_dentry)
> > err = ceph_handle_notrace_create(dir, dentry);
> > + ret = ERR_PTR(err);
> > out_req:
> > + if (!IS_ERR(ret) && req->r_dentry != dentry)
> > + /* Some other dentry was spliced in */
> > + ret = dget(req->r_dentry);
> > ceph_mdsc_put_request(req);
> > out:
> > - if (!err)
> > - ceph_init_inode_acls(d_inode(dentry), &as_ctx);
> > - else
> > + if (!IS_ERR(ret)) {
> > + if (ret)
> > + ceph_init_inode_acls(d_inode(ret), &as_ctx);
> > + else
> > + ceph_init_inode_acls(d_inode(dentry), &as_ctx);
> > + } else {
> > d_drop(dentry);
> > + }
> > ceph_release_acl_sec_ctx(&as_ctx);
> > - return ERR_PTR(err);
> > + return ret;
> > }
> >
> > static int ceph_link(struct dentry *old_dentry, struct inode *dir,
> >
>
> Thanks,
> Slava.
>
--
Jeff Layton <jlayton@xxxxxxxxxx>
