Re: [PATCH v2] seq_buf: add seq_buf_strlen(), seq_buf_init_append(), seq_buf_puts_trunc()

[Bill Wendling]

On Wed, Sep 16, 2026 at 4:47 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, 16 Sep 2026 22:21:25 +0000 Bill Wendling <morbo@xxxxxxxxxx> wrote:
>
> > Converting some strlcat() call sites seq_buf need behavior seq_buf doesn't
> > currently provide, either directly or without introducing subtle bugs:
> >
> >  - seq_buf_strlen(): seq_buf_used() reports the full buffer size when the
> >    buffer is completely filled, even though seq_buf_str() then overwrites the
> >    final byte with a NUL terminator, leaving only size - 1 bytes of
> >    actual content. Callers that need the true length of the
> >    NUL-terminated string have to fall back to strlen(seq_buf_str(s)).
> >    seq_buf_strlen() mirrors seq_buf_str()'s NUL-termination logic but returns
> >    the resulting string's length directly.
> >
> >  - seq_buf_init_append(): seq_buf_init() always clears the buffer it's given
> >    via seq_buf_clear(). Code migrating from strlcat(buf, ...), which
> >    appends to whatever @buf already contains, can't use seq_buf_init()
> >    without silently discarding that existing content. seq_buf_init_append()
> >    preserves it and positions the seq_buf to append after it.
> >
> >  - seq_buf_puts_trunc(): seq_buf_puts() (like seq_buf_printf() and friends)
> >    writes nothing at all if the string doesn't fully fit, whereas strlcat()
> >    always copies as much of the source as there is room for. Converting a
> >    strlcat() call site that relied on that partial-copy behavior to
> >    plain seq_buf_puts() can silently drop content that used to survive
> >    truncated. seq_buf_puts_trunc() keeps the leading bytes of the string
> >    that fit.
> >
> > ...
> >
> >  include/linux/seq_buf.h | 60 +++++++++++++++++++++++++++++++++++++++++
> >  lib/seq_buf.c           | 35 ++++++++++++++++++++++++
> >  2 files changed, 95 insertions(+)
>
> This seems a lot of code for enabling some strlcat removals.  How many
> is "some"?  If "3" then perhaps do something different at those
> callsites?
>
In the series of patches I have in the wings (I sent one version, but
it was a bit scattered and need to be resent), there are roughly 7
patches which use them, some more than once. While the impetus for
these new functions are to help streamline the removal of the
remaining strlcat, I believe Kees suggested these in order to keep the
mathematics involved in keeping the seq_bufs correct was behind the
API, rather than relying upon each user to get the math "correct".
(Kees, please correct me if I'm wrong.)

> Sashiko had a couple of comments:
>         https://sashiko.dev/#/patchset/20260916222125.1259631-1-morbo@xxxxxxxxxx
>
I'll look into them.

> Should these new functions be added to lib/tests/seq_buf_kunit.c?

Yes. I'll send a followup with the Sashiko comments addressed and the
KUnit tests added.

Share and enjoy!
-bw