| Message ID | 20191018155114.7423-1-phil@nwl.cc |
|---|---|
| State | Accepted |
| Delegated to: | Pablo Neira |
| Headers | show |
| Series | [iptables] nft: Use ARRAY_SIZE() macro in nft_strerror() | expand |
On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > Variable 'table' is an array of type struct table_struct, so this is a > classical use-case for ARRAY_SIZE() macro. > > Signed-off-by: Phil Sutter <phil@nwl.cc> Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
On Wed, Oct 23, 2019 at 01:20:24PM +0200, Pablo Neira Ayuso wrote: > On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > > Variable 'table' is an array of type struct table_struct, so this is a > > classical use-case for ARRAY_SIZE() macro. > > > > Signed-off-by: Phil Sutter <phil@nwl.cc> > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org> BTW, probably good to add the array check? https://sourceforge.net/p/libhx/libhx/ci/master/tree/include/libHX/defs.h#l152
Hi Pablo, On Wed, Oct 23, 2019 at 01:23:11PM +0200, Pablo Neira Ayuso wrote: > On Wed, Oct 23, 2019 at 01:20:24PM +0200, Pablo Neira Ayuso wrote: > > On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > > > Variable 'table' is an array of type struct table_struct, so this is a > > > classical use-case for ARRAY_SIZE() macro. > > > > > > Signed-off-by: Phil Sutter <phil@nwl.cc> > > > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org> > > BTW, probably good to add the array check? > > https://sourceforge.net/p/libhx/libhx/ci/master/tree/include/libHX/defs.h#l152 Copying from kernel sources, do you think that's fine? | # ifndef ARRAY_SIZE | -# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) | +# define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:(-!!(e)); })) | +# define __same_type(a, b) \ | + __builtin_types_compatible_p(typeof(a), typeof(b)) | +/* &a[0] degrades to a pointer: a different type from an array */ | +# define __must_be_array(a) \ | + BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0])) | +# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) + __must_be_array(x) | # endif Cheers, Phil
On Wed, Oct 23, 2019 at 02:16:27PM +0200, Phil Sutter wrote: > Hi Pablo, > > On Wed, Oct 23, 2019 at 01:23:11PM +0200, Pablo Neira Ayuso wrote: > > On Wed, Oct 23, 2019 at 01:20:24PM +0200, Pablo Neira Ayuso wrote: > > > On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > > > > Variable 'table' is an array of type struct table_struct, so this is a > > > > classical use-case for ARRAY_SIZE() macro. > > > > > > > > Signed-off-by: Phil Sutter <phil@nwl.cc> > > > > > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org> > > > > BTW, probably good to add the array check? > > > > https://sourceforge.net/p/libhx/libhx/ci/master/tree/include/libHX/defs.h#l152 > > Copying from kernel sources, do you think that's fine? > > | # ifndef ARRAY_SIZE > | -# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) > | +# define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:(-!!(e)); })) > | +# define __same_type(a, b) \ > | + __builtin_types_compatible_p(typeof(a), typeof(b)) > | +/* &a[0] degrades to a pointer: a different type from an array */ > | +# define __must_be_array(a) \ > | + BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0])) > | +# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) + __must_be_array(x) > | # endif At quick glance I would say that's fine.
Hi, On Wed, Oct 23, 2019 at 10:41:49PM +0200, Pablo Neira Ayuso wrote: > On Wed, Oct 23, 2019 at 02:16:27PM +0200, Phil Sutter wrote: > > Hi Pablo, > > > > On Wed, Oct 23, 2019 at 01:23:11PM +0200, Pablo Neira Ayuso wrote: > > > On Wed, Oct 23, 2019 at 01:20:24PM +0200, Pablo Neira Ayuso wrote: > > > > On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > > > > > Variable 'table' is an array of type struct table_struct, so this is a > > > > > classical use-case for ARRAY_SIZE() macro. > > > > > > > > > > Signed-off-by: Phil Sutter <phil@nwl.cc> > > > > > > > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org> > > > > > > BTW, probably good to add the array check? > > > > > > https://sourceforge.net/p/libhx/libhx/ci/master/tree/include/libHX/defs.h#l152 > > > > Copying from kernel sources, do you think that's fine? > > > > | # ifndef ARRAY_SIZE > > | -# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) > > | +# define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:(-!!(e)); })) > > | +# define __same_type(a, b) \ > > | + __builtin_types_compatible_p(typeof(a), typeof(b)) > > | +/* &a[0] degrades to a pointer: a different type from an array */ > > | +# define __must_be_array(a) \ > > | + BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0])) > > | +# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) + __must_be_array(x) > > | # endif > > At quick glance I would say that's fine. While testing it, I noticed that gcc has a builtin check already: | ../include/xtables.h:640:36: warning: division 'sizeof (const uint32_t * {aka const unsigned int *}) / sizeof (uint32_t {aka const unsigned int})' does not compute the number of array elements [-Wsizeof-pointer-div] | 640 | # define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) | | ^ | nft.c:914:18: note: in expansion of macro 'ARRAY_SIZE' | 914 | for (i = 1; i < ARRAY_SIZE(multp); i++) { | | ^~~~~~~~~~ | nft.c:906:25: note: first 'sizeof' operand was declared here | 906 | static const uint32_t *multp = mult; | | ^~~~~ AFAICT, the only benefit the above brings is that it causes an error instead of warning. Do you think we still need it? Maybe instead enable -Werror? ;) Cheers, Phil
On Thu, Oct 24, 2019 at 10:45:03AM +0200, Phil Sutter wrote: > Hi, > > On Wed, Oct 23, 2019 at 10:41:49PM +0200, Pablo Neira Ayuso wrote: > > On Wed, Oct 23, 2019 at 02:16:27PM +0200, Phil Sutter wrote: > > > Hi Pablo, > > > > > > On Wed, Oct 23, 2019 at 01:23:11PM +0200, Pablo Neira Ayuso wrote: > > > > On Wed, Oct 23, 2019 at 01:20:24PM +0200, Pablo Neira Ayuso wrote: > > > > > On Fri, Oct 18, 2019 at 05:51:14PM +0200, Phil Sutter wrote: > > > > > > Variable 'table' is an array of type struct table_struct, so this is a > > > > > > classical use-case for ARRAY_SIZE() macro. > > > > > > > > > > > > Signed-off-by: Phil Sutter <phil@nwl.cc> > > > > > > > > > > Acked-by: Pablo Neira Ayuso <pablo@netfilter.org> > > > > > > > > BTW, probably good to add the array check? > > > > > > > > https://sourceforge.net/p/libhx/libhx/ci/master/tree/include/libHX/defs.h#l152 > > > > > > Copying from kernel sources, do you think that's fine? > > > > > > | # ifndef ARRAY_SIZE > > > | -# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) > > > | +# define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:(-!!(e)); })) > > > | +# define __same_type(a, b) \ > > > | + __builtin_types_compatible_p(typeof(a), typeof(b)) > > > | +/* &a[0] degrades to a pointer: a different type from an array */ > > > | +# define __must_be_array(a) \ > > > | + BUILD_BUG_ON_ZERO(__same_type((a), &(a)[0])) > > > | +# define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) + __must_be_array(x) > > > | # endif > > > > At quick glance I would say that's fine. > > While testing it, I noticed that gcc has a builtin check already: > > | ../include/xtables.h:640:36: warning: division 'sizeof (const uint32_t * {aka const unsigned int *}) / sizeof (uint32_t {aka const unsigned int})' does not compute the number of array elements [-Wsizeof-pointer-div] > | 640 | # define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x))) > | | ^ > | nft.c:914:18: note: in expansion of macro 'ARRAY_SIZE' > | 914 | for (i = 1; i < ARRAY_SIZE(multp); i++) { > | | ^~~~~~~~~~ > | nft.c:906:25: note: first 'sizeof' operand was declared here > | 906 | static const uint32_t *multp = mult; > | | ^~~~~ > > AFAICT, the only benefit the above brings is that it causes an error > instead of warning. Do you think we still need it? Maybe instead enable > -Werror? ;) If gcc is already checking for this. Warning should be fine. Regarding -Werror, we would at least need to keep the autogenerated C code by bison away from it. IIRC I enabled this in conntrack-tools long time ago, and I started getting reports on it breaking compilation with new gcc versions that were actually spewing new warnings. That was stopping users to install latest, probably -Werror is too agressive?
Hey, On Thu, Oct 24, 2019 at 11:29:03AM +0200, Pablo Neira Ayuso wrote: [...] > If gcc is already checking for this. Warning should be fine. > > Regarding -Werror, we would at least need to keep the autogenerated C > code by bison away from it. In nftables there is libparser_la_CFLAGS which holds quite some exclusions already and could be used to pass -Wno-error as well. (Maybe a good idea to add this regardless of whether we set -Werror by default or not.) > IIRC I enabled this in conntrack-tools long time ago, and I started > getting reports on it breaking compilation with new gcc versions that > were actually spewing new warnings. That was stopping users to install > latest, probably -Werror is too agressive? Yes, I wasn't completely serious. Breaking users' builds for things they may not be in control of is not the best idea. We could instead add a configure option to enable strict mode, but checking for warnings is something I usually do so probably not that important after all. Cheers, Phil
On Thu, Oct 24, 2019 at 11:51:01AM +0200, Phil Sutter wrote: > Hey, > > On Thu, Oct 24, 2019 at 11:29:03AM +0200, Pablo Neira Ayuso wrote: > [...] > > If gcc is already checking for this. Warning should be fine. > > > > Regarding -Werror, we would at least need to keep the autogenerated C > > code by bison away from it. > > In nftables there is libparser_la_CFLAGS which holds quite some > exclusions already and could be used to pass -Wno-error as well. (Maybe > a good idea to add this regardless of whether we set -Werror by default > or not.) > > > IIRC I enabled this in conntrack-tools long time ago, and I started > > getting reports on it breaking compilation with new gcc versions that > > were actually spewing new warnings. That was stopping users to install > > latest, probably -Werror is too agressive? > > Yes, I wasn't completely serious. Breaking users' builds for things they > may not be in control of is not the best idea. We could instead add a > configure option to enable strict mode, but checking for warnings is > something I usually do so probably not that important after all. Indeed. So let's leave things as is then.
diff --git a/iptables/nft.c b/iptables/nft.c index 89b1c7a808f57..3c230c121f8b9 100644 --- a/iptables/nft.c +++ b/iptables/nft.c @@ -2888,7 +2888,7 @@ const char *nft_strerror(int err) { NULL, ENOENT, "No chain/target/match by that name" }, }; - for (i = 0; i < sizeof(table)/sizeof(struct table_struct); i++) { + for (i = 0; i < ARRAY_SIZE(table); i++) { if ((!table[i].fn || table[i].fn == nft_fn) && table[i].err == err) return table[i].message;
Variable 'table' is an array of type struct table_struct, so this is a classical use-case for ARRAY_SIZE() macro. Signed-off-by: Phil Sutter <phil@nwl.cc> --- iptables/nft.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)