[B.A.T.M.A.N.] [PATCH] batman-adv: compat: Substitute compat code for netlink constification
Substitutes the compat-patches approach for the netlink const changes to the prefered way of using compat-includes.
Signed-off-by: Linus Lüssing linus.luessing@c0d3.blue --- compat-include/net/genetlink.h | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)
diff --git a/compat-include/net/genetlink.h b/compat-include/net/genetlink.h index 72a8991..b195b47 100644 --- a/compat-include/net/genetlink.h +++ b/compat-include/net/genetlink.h @@ -28,6 +28,15 @@
#include <linux/export.h>
+/* multicast groups */ +enum __batadv_netlink_multicast_groups { + __BATADV_NL_MCGRP_TPMETER, +}; + +static struct genl_multicast_group __batadv_netlink_mcgrps[] = { + [__BATADV_NL_MCGRP_TPMETER] = { .name = BATADV_NL_MCAST_GROUP_TPMETER }, +}; + struct batadv_genl_family { /* data handled by the actual kernel */ struct genl_family family; @@ -140,12 +149,14 @@ static inline int batadv_genl_register_family(struct genl_family *family) static inline int batadv_genl_register_family_with_ops_grps(struct genl_family *family, struct genl_ops *ops, size_t n_ops, - struct genl_multicast_group *mcgrps, + const struct genl_multicast_group *mcgrps, size_t n_mcgrps) { + BUILD_BUG_ON(ARRAY_SIZE(__batadv_netlink_mcgrps) != n_mcgrps); + family->ops = ops; family->n_ops = n_ops; - family->mcgrps = mcgrps; + family->mcgrps = __batadv_netlink_mcgrps; family->n_mcgrps = n_mcgrps; family->module = THIS_MODULE;
On Tue, Oct 18, 2016 at 11:21:25PM +0200, Linus Lüssing wrote:
Substitutes the compat-patches approach for the netlink const changes to the prefered way of using compat-includes.
Signed-off-by: Linus Lüssing linus.luessing@c0d3.blue
Actually, it isn't a substitution yet, the compat-patches thing is still there.
Let me know what you think about this approach and if you like it what I should do about the replacements.sh.
On Dienstag, 18. Oktober 2016 23:21:25 CEST Linus Lüssing wrote:
Substitutes the compat-patches approach for the netlink const changes to the prefered way of using compat-includes.
Signed-off-by: Linus Lüssing linus.luessing@c0d3.blue
compat-include/net/genetlink.h | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)
diff --git a/compat-include/net/genetlink.h b/compat-include/net/genetlink.h index 72a8991..b195b47 100644 --- a/compat-include/net/genetlink.h +++ b/compat-include/net/genetlink.h @@ -28,6 +28,15 @@
#include <linux/export.h>
+/* multicast groups */ +enum __batadv_netlink_multicast_groups {
- __BATADV_NL_MCGRP_TPMETER,
+};
+static struct genl_multicast_group __batadv_netlink_mcgrps[] = {
- [__BATADV_NL_MCGRP_TPMETER] = { .name = BATADV_NL_MCAST_GROUP_TPMETER },
+};
Doesn't seem to scale. Especially when we think about batadv_netlink_ops which should also be const. It is currently not const because of the Linux <= 3.13 workaround.
Kind regards, Sven
On Tue, Oct 18, 2016 at 11:28:23PM +0200, Sven Eckelmann wrote:
Doesn't seem to scale. Especially when we think about batadv_netlink_ops which should also be const. It is currently not const because of the Linux <= 3.13 workaround.
Hm, okay, batadv_netlink_ops is a little more tricky/larger, yes. What about a memcpy'ing approach with BUILD_BUG_ON()'s as safe-guards like this:
diff --git a/compat-include/net/genetlink.h b/compat-include/net/genetlink.h index 72a8991..f16bf35 100644 --- a/compat-include/net/genetlink.h +++ b/compat-include/net/genetlink.h @@ -28,6 +28,9 @@
#include <linux/export.h>
+static struct genl_ops __batadv_netlink_ops[12]; +static struct genl_multicast_group __batadv_netlink_mcgrps[1]; + struct batadv_genl_family { /* data handled by the actual kernel */ struct genl_family family; @@ -137,15 +140,23 @@ static inline int batadv_genl_register_family(struct genl_family *family) return ret; }
-static inline int +static inline int __init batadv_genl_register_family_with_ops_grps(struct genl_family *family, - struct genl_ops *ops, size_t n_ops, - struct genl_multicast_group *mcgrps, + const struct genl_ops *ops, size_t n_ops, + const struct genl_multicast_group *mcgrps, size_t n_mcgrps) { - family->ops = ops; + BUILD_BUG_ON(ARRAY_SIZE(__batadv_netlink_ops) != n_ops); + BUILD_BUG_ON(ARRAY_SIZE(__batadv_netlink_mcgrps) != n_mcgrps); + BUILD_BUG_ON(sizeof(__batadv_netlink_ops) != sizeof(*ops) * n_ops); + BUILD_BUG_ON(sizeof(__batadv_netlink_mcgrps) != sizeof(*mcgrps) * n_mcgrps); + + memcpy(__batadv_netlink_ops, ops, sizeof(__batadv_netlink_ops)); + memcpy(__batadv_netlink_mcgrps, mcgrps, sizeof(__batadv_netlink_mcgrps)); + + family->ops = __batadv_netlink_ops; family->n_ops = n_ops; - family->mcgrps = mcgrps; + family->mcgrps = __batadv_netlink_mcgrps; family->n_mcgrps = n_mcgrps; family->module = THIS_MODULE;
diff --git a/net/batman-adv/netlink.c b/net/batman-adv/netlink.c index 64cb6ac..aee20a3 100644 --- a/net/batman-adv/netlink.c +++ b/net/batman-adv/netlink.c @@ -534,7 +534,7 @@ batadv_netlink_dump_hardifs(struct sk_buff *msg, struct netlink_callback *cb) return msg->len; }
-static struct genl_ops batadv_netlink_ops[] = { +static const struct genl_ops batadv_netlink_ops[] = { { .cmd = BATADV_CMD_GET_MESH_INFO, .flags = GENL_ADMIN_PERM,
On Mittwoch, 19. Oktober 2016 01:22:39 CEST Linus Lüssing wrote:
On Tue, Oct 18, 2016 at 11:28:23PM +0200, Sven Eckelmann wrote:
Doesn't seem to scale. Especially when we think about batadv_netlink_ops which should also be const. It is currently not const because of the Linux <= 3.13 workaround.
Hm, okay, batadv_netlink_ops is a little more tricky/larger, yes. What about a memcpy'ing approach with BUILD_BUG_ON()'s as safe-guards like this:
Yes, this would also be an idea. But I would personally just use the coccinelle approach because it needs less extra hacks. Or do you have any problems with coccinelle? My personal preferences are (at the moment):
1. code which doesn't need compat code 2. code which can supported with clean (not too hacky) compat-include 3. code which can be supported with coccinelle 4. code which can be supported with compat.h hacks 5. code which can be supported with small patches
The replace.sh was just a quick way to make sure we can ship 2016.4 without to many extra modifications.
Btw. this is right now only something for later. It is more important to get your other changes in master. And also to get the patchwork queue smaller - so reviews are more than welcome.
Kind regards, Sven
On Wed, Oct 19, 2016 at 08:31:03AM +0200, Sven Eckelmann wrote:
On Mittwoch, 19. Oktober 2016 01:22:39 CEST Linus Lüssing wrote:
On Tue, Oct 18, 2016 at 11:28:23PM +0200, Sven Eckelmann wrote:
Doesn't seem to scale. Especially when we think about batadv_netlink_ops which should also be const. It is currently not const because of the Linux <= 3.13 workaround.
Hm, okay, batadv_netlink_ops is a little more tricky/larger, yes. What about a memcpy'ing approach with BUILD_BUG_ON()'s as safe-guards like this:
Yes, this would also be an idea. But I would personally just use the coccinelle approach because it needs less extra hacks. Or do you have any problems with coccinelle?
Hm, no, dunno. Just haven't really looked at coccinelle yet and thought your point 3. (cocinelle) were on place 4. instead. Besides, I had fun creating this compat patch and was marvelled that the BUILD_BUG_ON() worked even with function parameters :D (maybe, actually because they were const now? You clever compiler :D).
Also, it looks like the whole compat infrastructure is growing and I'm wondering whether it is easier to maintain slightly larger compat-includes vs. then two more places where compat changes could come from.
But I don't feel strongly about it, if others think coccinelle is easier to review and maintain, then I'm fine with it and will have a closer look at coccinelle :).
b.a.t.m.a.n@lists.open-mesh.org
-
Linus Lüssing -
Sven Eckelmann