From: Greg KH <gregkh@linuxfoundation.org>
To: Mohammad Mosafer <mohsafer@gmail.com>
Cc: linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org,
syzbot+6227549bd2c8a1ec8ba0@syzkaller.appspotmail.com
Subject: Re: [PATCH v2] usb: gadget: f_fs: fix use-after-free in ffs_closed() on umount
Date: Fri, 20 Nov 2026 11:18:50 +0100 [thread overview]
Message-ID: <2026112030-aflame-skid-859a@gregkh> (raw)
In-Reply-To: <20261006214940.158352-1-mohsafer@gmail.com>
On Tue, Oct 06, 2026 at 04:49:40PM -0500, Mohammad Mosafer wrote:
> ffs_free_inst() releases the ffs_dev with ffs_release_dev() and only
> then re-acquires ffs_dev_lock to free it with _ffs_free_dev(). In
> between, the dev is still linked on the ffs_devices list while already
> marked unmounted, so a concurrent mount(2) of functionfs finds it by
> name in ffs_acquire_dev() and links a fresh ffs_data to the doomed dev
> (ffs_data->private_data = dev). _ffs_free_dev() then kfrees the dev,
> and when that mount is torn down, ffs_closed() dereferences the stale
> ffs->private_data:
>
> BUG: KASAN: slab-use-after-free in ffs_data_clear+0x438/0x530
> Write of size 1 at addr ffff88810594664a by task repro/116
> ffs_data_clear+0x438/0x530
> ffs_fs_kill_sb+0x7b/0x510
> deactivate_locked_super+0xa9/0x200
> cleanup_mnt+0x255/0x380
> ... reached via umount(2)
> Freed by task 112:
> kfree+0x127/0x3b0
> ffs_free_inst+0x10c/0x1a0
> usb_put_function_instance+0x8a/0xc0
> configfs_rmdir+0x773/0x9c0
> Allocated by task 113:
> ffs_alloc_inst+0x109/0x360
> function_make+0x138/0x330
> configfs_mkdir+0x48b/0x1090
>
> Hold ffs_dev_lock across the release and the free so that a released
> dev is never findable, splitting ffs_release_dev() into a lock-assuming
> _ffs_release_dev() (matching the _ffs_* convention in this file). For
> the same reason, re-read ffs->private_data under ffs_dev_lock in
> ffs_data_put(): the argument read there was unlocked, so a concurrent
> ffs_free_inst() could unlink and free the dev while the putter waited
> on the mutex, and the subsequent ffs_release_dev() would dereference
> the freed dev. The now unreached ffs_release_dev() wrapper is dropped;
> its remaining callers use _ffs_release_dev() with the lock held.
>
> The race was reproduced with a multi-threaded harness racing configfs
> mkdir/rmdir of the ffs instance against mount/umount of functionfs on
> a KASAN kernel: the unpatched kernel reports the use-after-free
> reliably (2/2 runs), the patched kernel survives an extended soak with
> identical churn (2/2 runs clean).
>
> Reported-by: syzbot+6227549bd2c8a1ec8ba0@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=6227549bd2c8a1ec8ba0
> Fixes: 5920cda627688c ("usb: gadget: FunctionFS: convert to new function interface with backward compatibility")
> Signed-off-by: Mohammad Mosafer <mohsafer@gmail.com>
> ---
>
> Changes in v2:
> - Re-read ffs->private_data under ffs_dev_lock in ffs_data_put(): the
> unlocked argument read raced with a concurrent ffs_free_inst() freeing
> the dev while the putter waited on ffs_dev_lock (flagged by the
> Sashiko review bot).
> - Drop the now-unused ffs_release_dev() wrapper.
> drivers/usb/gadget/function/f_fs.c | 31 ++++++++++++++++++++++--------
> 1 file changed, 23 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/usb/gadget/function/f_fs.c b/drivers/usb/gadget/function/f_fs.c
> index c64a268e98a4..960b73dc06b3 100644
> --- a/drivers/usb/gadget/function/f_fs.c
> +++ b/drivers/usb/gadget/function/f_fs.c
> @@ -288,7 +288,7 @@ static struct ffs_dev *_ffs_find_dev(const char *name);
> static struct ffs_dev *_ffs_alloc_dev(void);
> static void _ffs_free_dev(struct ffs_dev *dev);
> static int ffs_acquire_dev(const char *dev_name, struct ffs_data *ffs_data);
> -static void ffs_release_dev(struct ffs_dev *ffs_dev);
> +static void _ffs_release_dev(struct ffs_dev *ffs_dev);
> static int ffs_ready(struct ffs_data *ffs);
> static void ffs_closed(struct ffs_data *ffs);
> static void ffs_reset_work(struct work_struct *work);
> @@ -2231,7 +2231,14 @@ static void ffs_data_put(struct ffs_data *ffs)
> if (refcount_dec_and_test(&ffs->ref)) {
> pr_info("%s(): freeing\n", __func__);
> ffs_data_clear(ffs);
> - ffs_release_dev(ffs->private_data);
> + /*
> + * ffs->private_data must be re-read under ffs_dev_lock:
> + * a concurrent ffs_free_inst() may have NULLed it and
> + * freed the dev it pointed to.
> + */
> + ffs_dev_lock();
> + _ffs_release_dev(ffs->private_data);
> + ffs_dev_unlock();
> BUG_ON(waitqueue_active(&ffs->ev.waitq) ||
> swait_active(&ffs->ep0req_completion.wait) ||
> waitqueue_active(&ffs->wait));
> @@ -4147,8 +4154,17 @@ static void ffs_free_inst(struct usb_function_instance *f)
> struct f_fs_opts *opts;
>
> opts = to_f_fs_opts(f);
> - ffs_release_dev(opts->dev);
> +
> + /*
> + * Release and free the dev under a single ffs_dev_lock critical
> + * section. Between ffs_release_dev() and _ffs_free_dev() the dev
> + * would still be on the ffs_devices list while already unmounted,
> + * so a concurrent ffs_acquire_dev() could link a fresh ffs_data to
> + * the doomed dev, leaving it with a dangling ->private_data that is
> + * dereferenced in ffs_closed() when that mount is torn down.
> + */
That comment isn't needed, right?
> ffs_dev_lock();
> + _ffs_release_dev(opts->dev);
> _ffs_free_dev(opts->dev);
> ffs_dev_unlock();
> kfree(opts);
> @@ -4363,10 +4379,11 @@ static int ffs_acquire_dev(const char *dev_name, struct ffs_data *ffs_data)
> return ret;
> }
>
> -static void ffs_release_dev(struct ffs_dev *ffs_dev)
> +/*
> + * ffs_dev_lock must be taken by the caller
Then enforce it please.
thanks,
greg k-h
prev parent reply other threads:[~2026-11-20 10:18 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-06 21:05 [PATCH] " Mohammad Mosafer
2026-10-06 21:49 ` [PATCH v2] " Mohammad Mosafer
2026-11-20 10:18 ` Greg KH [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2026112030-aflame-skid-859a@gregkh \
--to=gregkh@linuxfoundation.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-usb@vger.kernel.org \
--cc=mohsafer@gmail.com \
--cc=syzbot+6227549bd2c8a1ec8ba0@syzkaller.appspotmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®