Re: [PATCH v2 3/4] printk/btrfs: Handle more message headers

From: Petr Mladek
Date: Tue Dec 13 2016 - 08:52:57 EST


On Tue 2016-12-13 10:21:27, Geert Uytterhoeven wrote:
> Hi Petr,
>
> On Wed, Nov 9, 2016 at 1:41 PM, Petr Mladek <pmladek@xxxxxxxx> wrote:
> > The commit 4bcc595ccd80decb4245096e ("printk: reinstate KERN_CONT for
> > printing continuation lines") allows to define more message headers
> > for a single message. The motivation is that continuous lines might
> > get mixed. Therefore it make sense to define the right log level
> > for every piece of a cont line.
> >
> > The current btrfs_printk() macros do not support continuous lines
> > at the moment. But better be prepared for a custom messages and
> > avoid potential "lvl" buffer overflow.
> >
> > This patch iterates over the entire message header. It is interested
> > only into the message level like the original code.
> >
> > This patch also introduces PRINTK_MAX_SINGLE_HEADER_LEN. Three bytes
> > are enough for the message level header at the moment. But it used to
> > be three, see the commit 04d2c8c83d0e3ac5f ("printk: convert the format
> > for KERN_<LEVEL> to a 2 byte pattern").
> >
> > Also I fixed the default ratelimit level. It looked very strange
> > when it was different from the default log level.
> >
> > Signed-off-by: Petr Mladek <pmladek@xxxxxxxx>
> > ---
> > fs/btrfs/super.c | 26 +++++++++++++++-----------
> > include/linux/printk.h | 2 ++
> > 2 files changed, 17 insertions(+), 11 deletions(-)
> >
> > diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
> > index 74ed5aae6cea..c083d84eaa32 100644
> > --- a/fs/btrfs/super.c
> > +++ b/fs/btrfs/super.c
> > @@ -202,27 +202,31 @@ void __btrfs_handle_fs_error(struct btrfs_fs_info *fs_info, const char *function
> > void btrfs_printk(const struct btrfs_fs_info *fs_info, const char *fmt, ...)
> > {
> > struct super_block *sb = fs_info->sb;
> > - char lvl[4];
> > + char lvl[PRINTK_MAX_SINGLE_HEADER_LEN + 1];
> > struct va_format vaf;
> > va_list args;
> > - const char *type = logtypes[4];
> > + const char *type = NULL;
> > int kern_level;
> > struct ratelimit_state *ratelimit;
>
> warning: âratelimitâ may be used uninitialized in this function
>
> So this triggered my attention. It seems my gcc (4.1.2) is not smart enough
> to notice that ratelimit will be set to a default value if !type.
>
> Still, IMHO the code is too convoluted for the casual reader.

I did not see this with gcc 4.8.5. But I agree that it might
be confusing. Please, find a proposed fix below.


> > vaf.fmt = fmt;
> > diff --git a/include/linux/printk.h b/include/linux/printk.h
> > index a0859e169bc3..afe8ccec1672 100644
> > --- a/include/linux/printk.h
> > +++ b/include/linux/printk.h
> > @@ -10,6 +10,8 @@
> > extern const char linux_banner[];
> > extern const char linux_proc_banner[];
> >
> > +#define PRINTK_MAX_SINGLE_HEADER_LEN 2
>
> I think you want to use this definition instead of the hardcoded "2" in
> printk_skip_level():
>
> | static inline const char *printk_skip_level(const char *buffer)
> | {
> | if (printk_get_level(buffer))
> | return buffer + 2;
>
> return buffer + PRINTK_MAX_SINGLE_HEADER_LEN;

I do not like hardcoded constants either. But this is with sync
with the hardcoded buffer[0] and buffer[1] in printk_get_level().

PRINTK_MAX_SINGLE_HEADER_LEN is meant as the maximum length in case
of some non-standard lengths in the future. We used 3-byte headers
in the past. The name is confusing for this use case. I probably
should have avoided _MAX_ in the name. I would leave this for now.
It is well localized...


Here is the proposed fix for the warning: