On Tue, Jul 26, 2022 at 04:27:56PM -0400, Jeff Layton wrote: > On Tue, 2022-07-26 at 16:23 -0400, Jeff Layton wrote: > > NFS server is exporting a sticky directory (mode 01777) with root > > squashing enabled. Client has protect_regular enabled and then tries to > > open a file as root in that directory. File is created (with ownership > > set to nobody:nobody) but the open syscall returns an error. > > > > The problem is may_create_in_sticky, which rejects the open even though > > the file has already been created/opened. Only call may_create_in_sticky > > if the file hasn't already been opened or created. > > > > Cc: Christian Brauner <brauner@xxxxxxxxxx> > > Link: https://bugzilla.redhat.com/show_bug.cgi?id=1976829 > > Reported-by: Yongchen Yang <yoyang@xxxxxxxxxx> > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx> > > --- > > fs/namei.c | 13 +++++++++---- > > 1 file changed, 9 insertions(+), 4 deletions(-) > > > > diff --git a/fs/namei.c b/fs/namei.c > > index 1f28d3f463c3..7480b6dc8d27 100644 > > --- a/fs/namei.c > > +++ b/fs/namei.c > > @@ -3495,10 +3495,15 @@ static int do_open(struct nameidata *nd, > > return -EEXIST; > > if (d_is_dir(nd->path.dentry)) > > return -EISDIR; > > - error = may_create_in_sticky(mnt_userns, nd, > > - d_backing_inode(nd->path.dentry)); > > - if (unlikely(error)) > > - return error; > > + if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { > > + error = may_create_in_sticky(mnt_userns, nd, > > + d_backing_inode(nd->path.dentry)); > > + if (unlikely(error)) { > > + printk("%s: f_mode=0x%x oflag=0x%x\n", > > + __func__, file->f_mode, open_flag); > > + return error; > > + } > > + } > > } > > if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) > > return -ENOTDIR; > > I'm pretty sure this patch is the wrong approach, actually, since it > doesn't fix the regular (non-atomic) open codepath. Any thoughts on what Hey Jeff, I haven't quite understood why that won't work for the regular open codepaths. I'm probably missing something obvious. > the right fix might be? When an actual creation has taken place - and not just a lookup - then may_create_in_sticky() assumes that the owner of the inode matches current_fsuid(). That'd would also be problematic on fat or in fact on any fs where the actual inode->i_{g,u}id are based on e.g. uid/gid mount options and not on current_fsuid(), I think? So in order to improve this we would need to work around that assumption in some way. Either by skipping may_create_in_sticky() if the file got created or by adapting may_create_in_sticky(). I only wonder whether skipping may_create_in_sticky() altogether might be a bit too lax. One possibility that came to my mind might be to relax this assumption when the file has been created and the creator has CAP_FOWNER. So (not compile tested or in any way) sm like: diff --git a/fs/namei.c b/fs/namei.c index 1f28d3f463c3..239e9f423346 100644 --- a/fs/namei.c +++ b/fs/namei.c @@ -1221,7 +1221,8 @@ int may_linkat(struct user_namespace *mnt_userns, struct path *link) * Returns 0 if the open is allowed, -ve on error. */ static int may_create_in_sticky(struct user_namespace *mnt_userns, - struct nameidata *nd, struct inode *const inode) + struct nameidata *nd, struct inode *const inode, + bool created) { umode_t dir_mode = nd->dir_mode; kuid_t dir_uid = nd->dir_uid; @@ -1230,7 +1231,9 @@ static int may_create_in_sticky(struct user_namespace *mnt_userns, (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || likely(!(dir_mode & S_ISVTX)) || uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || - uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) + uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)) || + (created && + capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER))) return 0; if (likely(dir_mode & 0002) || @@ -3496,7 +3499,8 @@ static int do_open(struct nameidata *nd, if (d_is_dir(nd->path.dentry)) return -EISDIR; error = may_create_in_sticky(mnt_userns, nd, - d_backing_inode(nd->path.dentry)); + d_backing_inode(nd->path.dentry), + (file->f_mode & FMODE_CREATED)); if (unlikely(error)) return error; }
