[bug report] coredump: hand a pidfd to the usermode coredump helper

[bug report] coredump: hand a pidfd to the usermode coredump helper

[Dan Carpenter]

Hello Christian Brauner,

Commit 4268b86fe0c7 ("coredump: hand a pidfd to the usermode coredump
helper") from Apr 14, 2025 (linux-next), leads to the following
Smatch static checker warning:

	fs/coredump.c:556 umh_coredump_setup()
	warn: re-assigning __cleanup__ ptr 'pidfs_file'

fs/coredump.c
    536 static int umh_coredump_setup(struct subprocess_info *info, struct cred *new)
    537 {
    538         struct file *files[2];
    539         struct coredump_params *cp = (struct coredump_params *)info->data;
    540         int err;
    541 
    542         if (cp->pid) {
    543                 struct file *pidfs_file __free(fput) = NULL;
    544 
    545                 pidfs_file = pidfs_alloc_file(cp->pid, 0);

We allocate pidfs_file.

    546                 if (IS_ERR(pidfs_file))
    547                         return PTR_ERR(pidfs_file);
    548 
    549                 /*
    550                  * Usermode helpers are childen of either
    551                  * system_unbound_wq or of kthreadd. So we know that
    552                  * we're starting off with a clean file descriptor
    553                  * table. So we should always be able to use
    554                  * COREDUMP_PIDFD_NUMBER as our file descriptor value.
    555                  */
--> 556                 VFS_WARN_ON_ONCE((pidfs_file = fget_raw(COREDUMP_PIDFD_NUMBER)) != NULL);

Then we set it to NULL without calling fput() on it.

    557 
    558                 err = replace_fd(COREDUMP_PIDFD_NUMBER, pidfs_file, 0);

pidfs_file is NULL so then we basically do:

	err = close_fd(COREDUMP_PIDFD_NUMBER);

I'm so confused...  Should the WARN_ON be for == NULL?

    559                 if (err < 0)
    560                         return err;
    561         }
    562 
    563         err = create_pipe_files(files, 0);
    564         if (err)
    565                 return err;
    566 
    567         cp->file = files[1];
    568 
    569         err = replace_fd(0, files[0], 0);
    570         fput(files[0]);
    571         if (err < 0)
    572                 return err;
    573 
    574         /* and disallow core files too */
    575         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
    576 
    577         return 0;
    578 }

regards,
dan carpenter

Re: [bug report] coredump: hand a pidfd to the usermode coredump helper

[Christian Brauner]

On Fri, Apr 25, 2025 at 05:23:38PM +0300, Dan Carpenter wrote:
> Hello Christian Brauner,
> 
> Commit 4268b86fe0c7 ("coredump: hand a pidfd to the usermode coredump
> helper") from Apr 14, 2025 (linux-next), leads to the following
> Smatch static checker warning:
> 
> 	fs/coredump.c:556 umh_coredump_setup()
> 	warn: re-assigning __cleanup__ ptr 'pidfs_file'

It's only with CONFIG_DEBUG_VFS and I've already removed that assert
in-tree and should've shown up in linux-next. Thanks!