> I don't see "no_const attribute used on type that is not constified" > message here, and I don't know why you got it.
Ok, this is my attempt compiling aufs using "-fplugin-arg-constify_plugin-no-constify" in EXTRA_CFLAGS. For my system informations please refer to: https://393559.bugs.gentoo.org/attachment.cgi?id=295143 zrouter aufs3-standalone # grep EXTRA_CFLAGS Makefile EXTRA_CFLAGS := -I${CURDIR}/include EXTRA_CFLAGS += ${AUFS_DEF_CONFIG} -fplugin-arg-constify_plugin-no-constify MakeMod = ${MAKE} -C ${KDIR} M=${CURDIR}/fs/aufs EXTRA_CFLAGS="${EXTRA_CFLAGS}" @echo ${EXTRA_CFLAGS} zrouter aufs3-standalone # make -I/var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/include -DCONFIG_AUFS_FS_MODULE -UCONFIG_AUFS -DCONFIG_AUFS_BRANCH_MAX_127 -DCONFIG_AUFS_HNOTIFY -DCONFIG_AUFS_HFSNOTIFY -DCONFIG_AUFS_SBILIST -DCONFIG_AUFS_RDU -fplugin-arg-constify_plugin-no-constify make -C /lib/modules/3.0.4-hardened-r5-zlive-1/build M=/var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/fs/aufs EXTRA_CFLAGS="-I/var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/include -DCONFIG_AUFS_FS_MODULE -UCONFIG_AUFS -DCONFIG_AUFS_BRANCH_MAX_127 -DCONFIG_AUFS_HNOTIFY -DCONFIG_AUFS_HFSNOTIFY -DCONFIG_AUFS_SBILIST -DCONFIG_AUFS_RDU -fplugin-arg-constify_plugin-no-constify" modules make[1]: Entering directory `/usr/src/linux-3.0.4-hardened-r5' CC [M] /var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/fs/aufs/module.o In file included from /var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/fs/aufs/module.c:24:0: include/linux/seq_file.h:35:16: error: 'no_const' attribute used on type that is not constified make[2]: *** [/var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/fs/aufs/module.o] Error 1 make[1]: *** [_module_/var/tmp/portage/sys-fs/aufs3-3_p20111114/work/aufs3-standalone/fs/aufs] Error 2 make[1]: Leaving directory `/usr/src/linux-3.0.4-hardened-r5' make: *** [fs/aufs/aufs.ko] Error 2 I found the typedef causing problem in "/usr/src/linux/include/linux/seq_file.h". So added a "no_const" typedef to our "a.c": zrouter constest # cat a.c #define NULL (void *)0 struct ops3 { void (* foo1)(void); void (* foo)(void); void (* foo2)(void); void (* foo3)(void); } __attribute((no_const)); struct ops { void (* foo1)(void); void (* foo)(void); void (* foo2)(void); void (* foo3)(void); }; #if 1 #define __no_const __attribute__((no_const)) typedef struct ops __no_const ops_no_const; #endif #if 1 struct { const struct ops3 *ops3; const struct ops *ops; } no; #endif struct ops3 a, aa; struct ops b, bb; int main() { a = aa; b = bb; a.foo2 = NULL; b.foo = NULL; return 0; } zrouter constest # gcc -fplugin=/usr/src/linux/tools/gcc/constify_plugin.so -fplugin-arg-constify_plugin-no-constify a.c a.c:19:16: error: 'no_const' attribute used on type that is not constified Got it! Should be disabled but is still there listening attribute events.. > > If the grsec patch modifies the declaration of "struct fsnotify_ops" as > "const", then the message you got > hfsnotify.c:208:2: error: assignment of read-only member 'br_hfsn_ops' > might be correct, since > - "struct au_branch" is not const. > - the variable "br" is not const either. > - the variable "br->br_hfsn_ops" is not declared as const either, but > its type "struct fsnotify_ops" is handled as const. > This is just my guess. > Think so, and read somewhere so: that plugin makes kernel structs "const" if their members are functions and there is no "const" or "__no_const" attribute specification. > If the constify_plugin handles > - "struct au_branch" is const > or > - the variable "br" is const > or > - the variable "br->br_hfsn_ops" is declared as const > then I am afraid there is something wrong in the constify_plugin. > But this is not your case, right? > Don't understand your question here... Given the previous fact "br->br_hfsn_ops" is const for the compiler. The problem could lay in struct fsnotify_ops defined in include/linux/fsnotify_backend.h. Infact adding a __no_const attribute is a solution. But I don't know if this is the correct solution... Neither I know who should apply it. > So I'd suggest you to dive into these as next step. > - Does the grsec patch (or someone else) modifies the declaration of > "struct fsnotify_ops" as "const"? As I said this is the case. > - Does gcc have another option not to produce "no_const attribute used > ..."? Disabling the plugin changes module signature (lacks CONSTIFY_PLUGIN beside GRSEC and kernel version information) the compiled plugin fails to load (look my previous post)... Don't know and can't test the effect if "-fplugin-arg-constify_plugin-no-constify" should work as expected. For me (has I didn't find any documentation for that plugin option) "'no_const' attribute used..." is a bug as I think that option should disable the plugin. > > > J. R. Okajima A. Zuccherelli ------------------------------------------------------------------------------ Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex infrastructure or vast IT resources to deliver seamless, secure access to virtual desktops. With this all-in-one solution, easily deploy virtual desktops for less than the cost of PCs and save 60% on VDI infrastructure costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox
