Re: [PATCH bpf-next] libbpf: Optimized return value in libbpf_strerror when errno is libbpf errno

From: Xin Liu
Date: Fri Dec 09 2022 - 03:41:32 EST


On Wed, 7 Dec 2022 6:31:34 PM Andrii Nakryiko wrote:
> On Wed, Dec 7, 2022 at 1:09 AM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
> >
> > On 12/7/22 1:00 AM, Andrii Nakryiko wrote:
> > > On Mon, Dec 5, 2022 at 1:11 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
> > >>
> > >> On 12/3/22 10:37 AM, Xin Liu wrote:
> > >>> This is a small improvement in libbpf_strerror. When libbpf_strerror
> > >>> is used to obtain the system error description, if the length of the
> > >>> buf is insufficient, libbpf_sterror returns ERANGE and sets errno to
> > >>> ERANGE.
> > >>>
> > >>> However, this processing is not performed when the error code
> > >>> customized by libbpf is obtained. Make some minor improvements here,
> > >>> return -ERANGE and set errno to ERANGE when buf is not enough for
> > >>> custom description.
> > >>>
> > >>> Signed-off-by: Xin Liu <liuxin350@xxxxxxxxxx>
> > >>> ---
> > >>> tools/lib/bpf/libbpf_errno.c | 6 ++++++
> > >>> 1 file changed, 6 insertions(+)
> > >>>
> > >>> diff --git a/tools/lib/bpf/libbpf_errno.c b/tools/lib/bpf/libbpf_errno.c
> > >>> index 96f67a772a1b..48ce7d5b5bf9 100644
> > >>> --- a/tools/lib/bpf/libbpf_errno.c
> > >>> +++ b/tools/lib/bpf/libbpf_errno.c
> > >>> @@ -54,10 +54,16 @@ int libbpf_strerror(int err, char *buf, size_t size)
> > >>>
> > >>> if (err < __LIBBPF_ERRNO__END) {
> > >>> const char *msg;
> > >>> + size_t msg_size;
> > >>>
> > >>> msg = libbpf_strerror_table[ERRNO_OFFSET(err)];
> > >>> snprintf(buf, size, "%s", msg);
> > >>> buf[size - 1] = '\0';
> > >>> +
> > >>> + msg_size = strlen(msg);
> > >>> + if (msg_size >= size)
> > >>> + return libbpf_err(-ERANGE);
> > >>
> > >> Given this is related to libbpf_strerror_table[] where the error strings are known
> > >> lets do compile-time error instead. All callers should pass in a buffer of STRERR_BUFSIZE
> > >> size in libbpf.
> > >
> > > That sounds a bit too pessimistic?.. If the actual error message fits
> > > in the buffer, why return -ERANGE just because theoretically some
> > > error descriptions might fit?
> > >
> > > But I don't think we need to calculate strlen(). snprintf above
> > > returns the number of bytes required to print a full string, even if
> > > it was truncated. So just comparing snprintf's result to size should
> > > be enough.
> >
> > I meant sth like below. For example if we were to shrink STRERR_BUFSIZE down
> > to 32 for testing, you'd then get:
>
> Sure, but I'm not sure why do we need to do this? Array of pointers to
> string will overall use less memory, as each string will take as much
> space as needed and no more.
>
> I guess I'm missing which problem you are trying to solve. I believe
> Xin was addressing a concern of extern (to libbpf) callers not getting
> -ERANGE in cases when they provide too small a buffer, which is just a
> simple snprintf() use adjustment to be a proper fix.
>
> >
> > # make libbpf_errno.o
> > gcc -g -O2 -std=gnu89 -Wbad-function-cast -Wdeclaration-after-statement -Wformat-security -Wformat-y2k -Winit-self -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wno-system-headers -Wold-style-definition -Wpacked -Wredundant-decls -Wstrict-prototypes -Wswitch-default -Wswitch-enum -Wundef -Wwrite-strings -Wformat -Wno-type-limits -Wstrict-aliasing=3 -Wshadow -Wno-switch-enum -Werror -Wall -I. -I/home/darkstar/trees/bpf-next/tools/include -I/home/darkstar/trees/bpf-next/tools/include/uapi -fvisibility=hidden -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -c -o libbpf_errno.o libbpf_errno.c
> > libbpf_errno.c:27:31: error: initializer-string for array of chars is too long [-Werror]
> > 27 | [ERRCODE_OFFSET(KVERSION)] = "'version' section incorrect or lost",
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > libbpf_errno.c:27:31: note: (near initialization for ‘libbpf_strerror_table[2]’)
> > libbpf_errno.c:31:29: error: initializer-string for array of chars is too long [-Werror]
> > 31 | [ERRCODE_OFFSET(VERIFY)] = "Kernel verifier blocks program loading",
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > libbpf_errno.c:31:29: note: (near initialization for ‘libbpf_strerror_table[7]’)
> > libbpf_errno.c:34:31: error: initializer-string for array of chars is too long [-Werror]
> > 34 | [ERRCODE_OFFSET(PROGTYPE)] = "Kernel doesn't support this program type",
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > libbpf_errno.c:34:31: note: (near initialization for ‘libbpf_strerror_table[10]’)
> > libbpf_errno.c:37:30: error: initializer-string for array of chars is too long [-Werror]
> > 37 | [ERRCODE_OFFSET(NLPARSE)] = "Incorrect netlink message parsing",
> > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > libbpf_errno.c:37:30: note: (near initialization for ‘libbpf_strerror_table[13]’)
> > cc1: all warnings being treated as errors
> > make: *** [<builtin>: libbpf_errno.o] Error 1
> >
> >
> >
> > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> > index 2a82f49ce16f..2e5df1624f79 100644
> > --- a/tools/lib/bpf/libbpf.c
> > +++ b/tools/lib/bpf/libbpf.c
> > @@ -265,8 +265,6 @@ static void pr_perm_msg(int err)
> > buf);
> > }
> >
> > -#define STRERR_BUFSIZE 128
> > -
> > /* Copied from tools/perf/util/util.h */
> > #ifndef zfree
> > # define zfree(ptr) ({ free(*ptr); *ptr = NULL; })
> > diff --git a/tools/lib/bpf/libbpf_errno.c b/tools/lib/bpf/libbpf_errno.c
> > index 96f67a772a1b..2f03f861b8b6 100644
> > --- a/tools/lib/bpf/libbpf_errno.c
> > +++ b/tools/lib/bpf/libbpf_errno.c
> > @@ -21,7 +21,7 @@
> > #define ERRCODE_OFFSET(c) ERRNO_OFFSET(LIBBPF_ERRNO__##c)
> > #define NR_ERRNO (__LIBBPF_ERRNO__END - __LIBBPF_ERRNO__START)
> >
> > -static const char *libbpf_strerror_table[NR_ERRNO] = {
> > +static const char libbpf_strerror_table[NR_ERRNO][STRERR_BUFSIZE] = {
> > [ERRCODE_OFFSET(LIBELF)] = "Something wrong in libelf",
> > [ERRCODE_OFFSET(FORMAT)] = "BPF object format invalid",
> > [ERRCODE_OFFSET(KVERSION)] = "'version' section incorrect or lost",
> > diff --git a/tools/lib/bpf/libbpf_internal.h b/tools/lib/bpf/libbpf_internal.h
> > index 377642ff51fc..d4dc4fe945a6 100644
> > --- a/tools/lib/bpf/libbpf_internal.h
> > +++ b/tools/lib/bpf/libbpf_internal.h
> > @@ -57,6 +57,8 @@
> > #define ELF64_ST_VISIBILITY(o) ((o) & 0x03)
> > #endif
> >
> > +#define STRERR_BUFSIZE 128
> > +
> > #define BTF_INFO_ENC(kind, kind_flag, vlen) \
> > ((!!(kind_flag) << 31) | ((kind) << 24) | ((vlen) & BTF_MAX_VLEN))
> > #define BTF_TYPE_ENC(name, info, size_or_type) (name), (info), (size_or_type)

I also think it would be more appropriate to modify snprintf slightly
here to keep memory usage as low as possible, and I will submit an
optimized version of v2, thanks to Daniel and Andrii for their
suggestions.