Re: [PATCH v4 2nd 2/4] pstore: support multiple pmsg instances

[Kees Cook <keescook@chromium.org>]

On Mon, Jan 30, 2017 at 5:58 PM, Nobuhiro Iwamatsu
<nobuhiro.iwamatsu.kw@hitachi.com> wrote:
> From: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
>
> This enables pmsg to deal with multiple instances. One possible
> use is content priority control on limited persistent store space. By
> using different buffers, we can prevent important messages from being
> overwritten by less important messages.
>
> When pstore backend module specifies the number of instances (num_pmsg)
> in pstore_info, multiple /dev/pmsg[ID] appear. (ID is an integer
> starting at 0. It corresponds to minor number of the each char device.)
> Writes to each /dev/pmsg[ID] are isolated each other. After reboot, the
> contents are available in /sys/fs/pstore/pmsg-[backend]-[ID].
>
> Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
> Signed-off-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.kw@hitachi.com>
> Cc: Anton Vorontsov <anton@enomsg.org>
> Cc: Colin Cross <ccross@android.com>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: Mark Salyzyn <salyzyn@android.com>
> Cc: Seiji Aguchi <seiji.aguchi.tr@hitachi.com>
> Cc: Tony Luck <tony.luck@intel.com>
>
> V4:
>     Rebase.
> V3:
>     rebase.
>     merged device_create().
> ---
>  fs/pstore/pmsg.c       | 23 ++++++++++++++++-------
>  include/linux/pstore.h |  1 +
>  2 files changed, 17 insertions(+), 7 deletions(-)
>
> diff --git a/fs/pstore/pmsg.c b/fs/pstore/pmsg.c
> index 78f6176c020f..5da5cba4b387 100644
> --- a/fs/pstore/pmsg.c
> +++ b/fs/pstore/pmsg.c
> @@ -34,7 +34,8 @@ static ssize_t write_pmsg(struct file *file, const char __user *buf,
>                 return -EFAULT;
>
>         mutex_lock(&pmsg_lock);
> -       ret = psinfo->write_buf_user(PSTORE_TYPE_PMSG, 0, &id, 0, buf, 0, count,
> +       ret = psinfo->write_buf_user(PSTORE_TYPE_PMSG, 0, &id,
> +                                    iminor(file->f_inode), buf, 0, count,
>                                      psinfo);
>         mutex_unlock(&pmsg_lock);
>         return ret ? ret : count;
> @@ -61,7 +62,7 @@ static char *pmsg_devnode(struct device *dev, umode_t *mode)
>
>  void pstore_register_pmsg(void)
>  {
> -       struct device *pmsg_device;
> +       int i;
>
>         pmsg_major = register_chrdev(0, PMSG_NAME, &pmsg_fops);
>         if (pmsg_major < 0) {
> @@ -76,15 +77,23 @@ void pstore_register_pmsg(void)
>         }
>         pmsg_class->devnode = pmsg_devnode;
>
> -       pmsg_device = device_create(pmsg_class, NULL, MKDEV(pmsg_major, 0),
> -                                       NULL, "%s%d", PMSG_NAME, 0);
> -       if (IS_ERR(pmsg_device)) {
> -               pr_err("failed to create device\n");
> -               goto err_device;
> +       /* allocate additional /dev/pmsg[ID] */

This isn't "additional" any more, but rather "Allocate each pmsg
device" or something.

> +       for (i = 0; i < psinfo->num_pmsg; i++) {
> +               struct device *pmsg_device;
> +
> +               pmsg_device = device_create(pmsg_class, NULL,
> +                                           MKDEV(pmsg_major, i), NULL, "%s%d",
> +                                           PMSG_NAME, i);
> +               if (IS_ERR(pmsg_device)) {
> +                       pr_err("failed to create device\n");
> +                       goto err_device;
> +               }
>         }
>         return;
>
>  err_device:
> +       for (i--; i >= 0; i--)
> +               device_destroy(pmsg_class, MKDEV(pmsg_major, i));
>         class_destroy(pmsg_class);
>  err_class:
>         unregister_chrdev(pmsg_major, PMSG_NAME);
> diff --git a/include/linux/pstore.h b/include/linux/pstore.h
> index 0da29cae009b..7a5db4833b8a 100644
> --- a/include/linux/pstore.h
> +++ b/include/linux/pstore.h
> @@ -55,6 +55,7 @@ struct pstore_info {
>         size_t          bufsize;
>         struct mutex    read_mutex;     /* serialize open/read/close */
>         int             flags;
> +       unsigned int    num_pmsg;

Something in this patch needs to set num_pmsg to 1 unconditionally, so
that bisectability is retained. I.e. after this patch, I should still
get a /dev/pmsg0 device. Right now, it will be skipped since num_pmsg
will always == 0.

>         int             (*open)(struct pstore_info *psi);
>         int             (*close)(struct pstore_info *psi);
>         ssize_t         (*read)(u64 *id, enum pstore_type_id *type,
> --
> 2.11.0
>

Thanks for the rebasing!

-Kees

-- 
Kees Cook
Pixel Security