linux-kernel - Re: [PATCH -next V3] ubi: fix race condition between ctrl_cdev_ioctl and ubi_cdev

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <b8f6bbf3-cca4-9bfa-f70a-393f04edd21f@huawei.com>
Date:   Tue, 28 Dec 2021 15:48:12 +0800
From:   Zhihao Cheng <chengzhihao1@...wei.com>
To:     Richard Weinberger <richard@....at>,
        "libaokun (A)" <libaokun1@...wei.com>
CC:     Miquel Raynal <miquel.raynal@...tlin.com>,
        Vignesh Raghavendra <vigneshr@...com>,
        linux-mtd <linux-mtd@...ts.infradead.org>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        <yukuai3@...wei.com>, Hulk Robot <hulkci@...wei.com>
Subject: Re: [PATCH -next V3] ubi: fix race condition between ctrl_cdev_ioctl
 and ubi_cdev_ioctl

在 2021/12/24 5:06, Richard Weinberger 写道:
> ----- Ursprüngliche Mail -----


>>> ubi_get_device is used to prevent devices from being deleted during
>>> sysfs execution. However, now kernfs ensures that devices will not
>>> be deleted before all reference counting are released.
>>> The key process is shown in the following stack.
> 
> ubi_get_device() in dev_attribute_show() is used to detect whether
> the ubi device got detached while the sysfs file is open.
> 
> Hmm. I thought for sysfs this is not the case since sysfs does not implement
> a release() method. So kernfs_drain_open_files() will return early.
> But there is a good chance that I don't got all kernfs/sysfs details.
> 

kernfs_drain() will wait 'root->deactivate_waitq' if 
atomic_read(&kn->active) not equals to KN_DEACTIVATED_BIAS.

The UBI seq_show callback is invoked with avtive cnt taken:
vfs_read
   kernfs_fop_read_iter
     seq_read_iter
       m->op->start (kernfs_seq_start)   // kernfs_get_active(of->kn)
       kernfs_seq_show
         dev_attribute_show [ubi]
       m->op->stop (kernfs_seq_stop)     // kernfs_put_active(of->kn)

The kernfs_drain() is stuck at wait_event() until sysfs reading 
finished, in my local test.