[AUDIT] ratelimit printk messages audit
Eric Paris [Thu, 24 Jan 2008 03:55:05 +0000 (22:55 -0500)]
some printk messages from the audit system can become excessive.  This
patch ratelimits those messages.  It was found that messages, such as
the audit backlog lost printk message could flood the logs to the point
that a machine could take an nmi watchdog hit or otherwise become
unresponsive.

Signed-off-by: Eric Paris <eparis@redhat.com>

kernel/audit.c

index 9e3e457..c8555b1 100644 (file)
@@ -166,7 +166,8 @@ void audit_panic(const char *message)
        case AUDIT_FAIL_SILENT:
                break;
        case AUDIT_FAIL_PRINTK:
-               printk(KERN_ERR "audit: %s\n", message);
+               if (printk_ratelimit())
+                       printk(KERN_ERR "audit: %s\n", message);
                break;
        case AUDIT_FAIL_PANIC:
                panic("audit: %s\n", message);
@@ -234,11 +235,13 @@ void audit_log_lost(const char *message)
        }
 
        if (print) {
-               printk(KERN_WARNING
-                      "audit: audit_lost=%d audit_rate_limit=%d audit_backlog_limit=%d\n",
-                      atomic_read(&audit_lost),
-                      audit_rate_limit,
-                      audit_backlog_limit);
+               if (printk_ratelimit())
+                       printk(KERN_WARNING
+                               "audit: audit_lost=%d audit_rate_limit=%d "
+                               "audit_backlog_limit=%d\n",
+                               atomic_read(&audit_lost),
+                               audit_rate_limit,
+                               audit_backlog_limit);
                audit_panic(message);
        }
 }
@@ -352,7 +355,11 @@ static int kauditd_thread(void *dummy)
                                        audit_pid = 0;
                                }
                        } else {
-                               printk(KERN_NOTICE "%s\n", skb->data + NLMSG_SPACE(0));
+                               if (printk_ratelimit())
+                                       printk(KERN_NOTICE "%s\n", skb->data +
+                                               NLMSG_SPACE(0));
+                               else
+                                       audit_log_lost("printk limit exceeded\n");
                                kfree_skb(skb);
                        }
                } else {
@@ -1066,7 +1073,7 @@ struct audit_buffer *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask,
                        remove_wait_queue(&audit_backlog_wait, &wait);
                        continue;
                }
-               if (audit_rate_check())
+               if (audit_rate_check() && printk_ratelimit())
                        printk(KERN_WARNING
                               "audit: audit_backlog=%d > "
                               "audit_backlog_limit=%d\n",
@@ -1349,9 +1356,11 @@ void audit_log_end(struct audit_buffer *ab)
                        skb_queue_tail(&audit_skb_queue, ab->skb);
                        ab->skb = NULL;
                        wake_up_interruptible(&kauditd_wait);
-               } else {
+               } else if (printk_ratelimit()) {
                        struct nlmsghdr *nlh = nlmsg_hdr(ab->skb);
                        printk(KERN_NOTICE "type=%d %s\n", nlh->nlmsg_type, ab->skb->data + NLMSG_SPACE(0));
+               } else {
+                       audit_log_lost("printk limit exceeded\n");
                }
        }
        audit_buffer_free(ab);