Re: [PATCH ghak90 (was ghak32) V4 01/10] audit: collect audit task parameters

On Thu, Nov 1, 2018 at 6:07 PM Richard Guy Briggs <rgb(a)redhat.com&gt; wrote: &gt; On 2018-10-19 19:15, Paul Moore wrote: &gt; &gt; On Sun, Aug 5, 2018 at 4:32 AM Richard Guy Briggs <rgb(a)redhat.com&gt; wrote: &gt; &gt; &gt; The audit-related parameters in struct task_struct should ideally be &gt; &gt; &gt; collected together and accessed through a standard audit API. &gt; &gt; &gt; &gt; &gt; &gt; Collect the existing loginuid, sessionid and audit_context together in a &gt; &gt; &gt; new struct audit_task_info called "audit" in struct task_struct. &gt; &gt; &gt; &gt; &gt; &gt; Use kmem_cache to manage this pool of memory. &gt; &gt; &gt; Un-inline audit_free() to be able to always recover that memory. &gt; &gt; &gt; &gt; &gt; &gt; See: https://github.com/linux-audit/audit-kernel/issues/81 &gt; &gt; &gt; &gt; &gt; &gt; Signed-off-by: Richard Guy Briggs <rgb(a)redhat.com&gt; &gt; &gt; &gt; --- &gt; &gt; &gt; include/linux/audit.h | 34 ++++++++++++++++++++++++---------- &gt; &gt; &gt; include/linux/sched.h | 5 +---- &gt; &gt; &gt; init/init_task.c | 3 +-- &gt; &gt; &gt; init/main.c | 2 ++ &gt; &gt; &gt; kernel/auditsc.c | 51 ++++++++++++++++++++++++++++++++++++++++++--------- &gt; &gt; &gt; kernel/fork.c | 4 +++- &gt; &gt; &gt; 6 files changed, 73 insertions(+), 26 deletions(-) &gt; &gt; &gt; &gt; ... &gt; &gt; &gt; &gt; &gt; diff --git a/include/linux/sched.h b/include/linux/sched.h &gt; &gt; &gt; index 87bf02d..e117272 100644 &gt; &gt; &gt; --- a/include/linux/sched.h &gt; &gt; &gt; +++ b/include/linux/sched.h &gt; &gt; &gt; @@ -873,10 +872,8 @@ struct task_struct { &gt; &gt; &gt; &gt; &gt; &gt; struct callback_head *task_works; &gt; &gt; &gt; &gt; &gt; &gt; - struct audit_context *audit_context; &gt; &gt; &gt; #ifdef CONFIG_AUDITSYSCALL &gt; &gt; &gt; - kuid_t loginuid; &gt; &gt; &gt; - unsigned int sessionid; &gt; &gt; &gt; + struct audit_task_info *audit; &gt; &gt; &gt; #endif &gt; &gt; &gt; struct seccomp seccomp; &gt; &gt; &gt; &gt; Prior to this patch audit_context was available regardless of &gt; &gt; CONFIG_AUDITSYSCALL, after this patch the corresponding audit_context &gt; &gt; is only available when CONFIG_AUDITSYSCALL is defined. &gt; &gt; This was intentional since audit_context is not used when AUDITSYSCALL is &gt; disabled. audit_alloc() was stubbed in that case to return 0. audit_context() &gt; returned NULL. &gt; &gt; The fact that audit_context was still present in struct task_struct was an &gt; oversight in the two patches already accepted: &gt; ("audit: use inline function to get audit context") &gt; ("audit: use inline function to get audit context") &gt; that failed to hide or remove it from struct task_struct when it was no longer &gt; relevant. Okay, in that case let's pull this out and fix this separately from the audit container ID patchset. &gt; On further digging, loginuid and sessionid (and audit_log_session_info) should &gt; be part of CONFIG_AUDIT scope and not CONFIG_AUDITSYSCALL since it is used in &gt; CONFIG_CHANGE, ANOM_LINK, FEATURE_CHANGE(, INTEGRITY_RULE), none of which are &gt; otherwise dependent on AUDITSYSCALL. This looks like something else we should fix independently from this patchset. &gt; Looking ahead, contid should be treated like loginuid and sessionid, which are &gt; currently only available when syscall auditting is. That seems reasonable. Eventually it would be great if we got rid of CONFIG_AUDITSYSCALL, but that is a separate issue, and something that is going to require work from the different arch/ABI folks to ensure everything is working properly. &gt; Converting records from standalone to syscall and checking audit_dummy_context &gt; changes the nature of CONFIG_AUDIT/!CONFIG_AUDITSYSCALL separation. &gt; eg: ANOM_LINK accompanied by PATH record (which needed CWD addition to be &gt; complete anyways) &gt; &gt; &gt; &gt; diff --git a/init/main.c b/init/main.c &gt; &gt; &gt; index 3b4ada1..6aba171 100644 &gt; &gt; &gt; --- a/init/main.c &gt; &gt; &gt; +++ b/init/main.c &gt; &gt; &gt; @@ -92,6 +92,7 @@ &gt; &gt; &gt; #include <linux rodata_test.h=""> &gt; &gt; &gt; #include <linux jump_label.h=""> &gt; &gt; &gt; #include <linux mem_encrypt.h=""> &gt; &gt; &gt; +#include <linux audit.h=""> &gt; &gt; &gt; &gt; &gt; &gt; #include <asm io.h=""> &gt; &gt; &gt; #include <asm bugs.h=""> &gt; &gt; &gt; @@ -721,6 +722,7 @@ asmlinkage __visible void __init start_kernel(void) &gt; &gt; &gt; nsfs_init(); &gt; &gt; &gt; cpuset_init(); &gt; &gt; &gt; cgroup_init(); &gt; &gt; &gt; + audit_task_init(); &gt; &gt; &gt; taskstats_init_early(); &gt; &gt; &gt; delayacct_init(); &gt; &gt; &gt; &gt; It seems like we would need either init_struct_audit or &gt; &gt; audit_task_init(), but not both, yes? &gt; &gt; One sets initial values of init task via an included struct, other makes a call &gt; to create the kmem cache. Both seem appropriate to me unless we move the &gt; initialization from a struct to assignments in audit_task_init(), but I'm not &gt; that comfortable separating the audit init values from the rest of the &gt; task_struct init task initializers (though there are other subsystems that need &gt; to do so dynamically). My original thinking was focused on the use of init_struct_audit as an initializer when audit_task_init() was already creating a kmem_cache pool and a zero'd/init'd audit_task_info could be obtained via the usual kmem_cache functions. Alternatively, although I don't believe it would be recommended for this case, would be to use init_struct_audit as an init helper if we included the audit_task_info struct directly in the task_struct, as opposed to a pointer. What I missed was the simple fact that you're only using init_struct_audit for the init_task, which pretty much makes my original question rather silly :) -- paul moore www.paul-moore.com