Messages in this thread | | | Date | Fri, 13 Jul 2018 17:44:45 -0400 | From | Richard Guy Briggs <> | Subject | Re: [RFC PATCH ghak59 V1 4/6] audit: hand taken context to audit_kill_trees for syscall logging |
| |
On 2018-06-28 18:23, Paul Moore wrote: > On Thu, Jun 14, 2018 at 4:23 PM Richard Guy Briggs <rgb@redhat.com> wrote: > > Since the context is taken from the task in __audit_syscall_exit() and > > __audit_free(), hand it to audit_kill_trees() so it can be used to > > associate with a syscall record. This requires adding the context > > parameter to kill_rules() rather than using the current audit_context > > (which has been taken). > > > > The callers of trim_marked() and evict_chunk() still have their context. > > > > See: https://github.com/linux-audit/audit-kernel/issues/50 > > See: https://github.com/linux-audit/audit-kernel/issues/59 > > Signed-off-by: Richard Guy Briggs <rgb@redhat.com> > > --- > > kernel/audit.h | 4 ++-- > > kernel/audit_tree.c | 18 ++++++++++-------- > > kernel/auditsc.c | 4 ++-- > > 3 files changed, 14 insertions(+), 12 deletions(-) > > > > diff --git a/kernel/audit.h b/kernel/audit.h > > index 214e149..f39f7aa 100644 > > --- a/kernel/audit.h > > +++ b/kernel/audit.h > > @@ -312,7 +312,7 @@ extern void audit_log_d_path_exe(struct audit_buffer *ab, > > extern int audit_tag_tree(char *old, char *new); > > extern const char *audit_tree_path(struct audit_tree *tree); > > extern void audit_put_tree(struct audit_tree *tree); > > -extern void audit_kill_trees(struct list_head *list); > > +extern void audit_kill_trees(struct audit_context *context); > > #else > > #define audit_remove_tree_rule(rule) BUG() > > #define audit_add_tree_rule(rule) -EINVAL > > @@ -321,7 +321,7 @@ extern void audit_log_d_path_exe(struct audit_buffer *ab, > > #define audit_put_tree(tree) (void)0 > > #define audit_tag_tree(old, new) -EINVAL > > #define audit_tree_path(rule) "" /* never called */ > > -#define audit_kill_trees(list) BUG() > > +#define audit_kill_trees(context) BUG() > > #endif > > > > extern char *audit_unpack_string(void **bufp, size_t *remain, size_t len); > > diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c > > index a01b9da..2d3e1071 100644 > > --- a/kernel/audit_tree.c > > +++ b/kernel/audit_tree.c > > @@ -493,13 +493,13 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree) > > return 0; > > } > > > > -static void audit_tree_log_remove_rule(struct audit_krule *rule) > > +static void audit_tree_log_remove_rule(struct audit_context *context, struct audit_krule *rule) > > { > > struct audit_buffer *ab; > > > > if (!audit_enabled) > > return; > > - ab = audit_log_start(audit_context(), GFP_KERNEL, AUDIT_CONFIG_CHANGE); > > + ab = audit_log_start(context, GFP_KERNEL, AUDIT_CONFIG_CHANGE); > > if (unlikely(!ab)) > > return; > > audit_log_format(ab, "op=remove_rule"); > > @@ -510,7 +510,7 @@ static void audit_tree_log_remove_rule(struct audit_krule *rule) > > audit_log_end(ab); > > } > > > > -static void kill_rules(struct audit_tree *tree) > > +static void kill_rules(struct audit_context *context, struct audit_tree *tree) > > { > > struct audit_krule *rule, *next; > > struct audit_entry *entry; > > @@ -521,7 +521,7 @@ static void kill_rules(struct audit_tree *tree) > > list_del_init(&rule->rlist); > > if (rule->tree) { > > /* not a half-baked one */ > > - audit_tree_log_remove_rule(rule); > > + audit_tree_log_remove_rule(context, rule); > > if (entry->rule.exe) > > audit_remove_mark(entry->rule.exe); > > rule->tree = NULL; > > @@ -584,7 +584,7 @@ static void trim_marked(struct audit_tree *tree) > > tree->goner = 1; > > spin_unlock(&hash_lock); > > mutex_lock(&audit_filter_mutex); > > - kill_rules(tree); > > + kill_rules(audit_context(), tree); > > list_del_init(&tree->list); > > mutex_unlock(&audit_filter_mutex); > > prune_one(tree); > > @@ -924,8 +924,10 @@ static void audit_schedule_prune(void) > > * ... and that one is done if evict_chunk() decides to delay until the end > > * of syscall. Runs synchronously. > > */ > > -void audit_kill_trees(struct list_head *list) > > +void audit_kill_trees(struct audit_context *context) > > { > > + struct list_head *list = &context->killed_trees; > > + > > audit_ctl_lock(); > > mutex_lock(&audit_filter_mutex); > > > > @@ -933,7 +935,7 @@ void audit_kill_trees(struct list_head *list) > > struct audit_tree *victim; > > > > victim = list_entry(list->next, struct audit_tree, list); > > - kill_rules(victim); > > + kill_rules(context, victim); > > list_del_init(&victim->list); > > > > mutex_unlock(&audit_filter_mutex); > > @@ -972,7 +974,7 @@ static void evict_chunk(struct audit_chunk *chunk) > > list_del_init(&owner->same_root); > > spin_unlock(&hash_lock); > > if (!postponed) { > > - kill_rules(owner); > > + kill_rules(audit_context(), owner); > > list_move(&owner->list, &prune_list); > > need_prune = 1; > > } else { > > diff --git a/kernel/auditsc.c b/kernel/auditsc.c > > index ceb1c45..2590c9e 100644 > > --- a/kernel/auditsc.c > > +++ b/kernel/auditsc.c > > @@ -1490,7 +1490,7 @@ void __audit_free(struct task_struct *tsk) > > if (context->in_syscall && context->current_state == AUDIT_RECORD_CONTEXT) > > audit_log_exit(context, tsk); > > if (!list_empty(&context->killed_trees)) > > - audit_kill_trees(&context->killed_trees); > > + audit_kill_trees(context); > > See my comment below about the ordering of audit_kill_trees() and > audit_log_exit(). > > > audit_free_context(context); > > } > > @@ -1577,7 +1577,7 @@ void __audit_syscall_exit(int success, long return_code) > > context->prio = context->state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0; > > > > if (!list_empty(&context->killed_trees)) > > - audit_kill_trees(&context->killed_trees); > > + audit_kill_trees(context); > > I wonder if we should move the kill_trees if-block above the > audit_log_exit() block so that any records that are emitted will be > before the SYSCALL record. I didn't chase down all the code paths, > but it seems like it should be safe, no?
Interesting. I thought I had looked at re-ordering them and rejected that approach due to that information being needed for audit_log_exit(), but I don't find any such dependency this pass through the code.
I guess the only concern I have then is that if the state is anything other than AUDIT_RECORD_CONTEXT it would be an orphan record, but it would be regardless with the existing code or with my proposed changes. Perhaps that is a bug to start with, though I'm not sure it is at all serious, so I'm not concerned about it.
I think re-ordering should be safe and that eliminates the seeming complexity introduced by the next patch, which is a good thing.
> > audit_free_names(context); > > unroll_tree_refs(context, NULL, 0); > > paul moore
- RGB
-- Richard Guy Briggs <rgb@redhat.com> Sr. S/W Engineer, Kernel Security, Base Operating Systems Remote, Ottawa, Red Hat Canada IRC: rgb, SunRaycer Voice: +1.647.777.2635, Internal: (81) 32635
| |