We must not call timekeeping functions unless they are available. If we dumpWould nulling out the timestamp be better done in do_gettimeofday()? That way we don't have to export timekeeping internals and users would get something more sane for this corner case.
before they have resumed, avoid a WARN_ON by setting the timestamp to 0.
Since the "ram" pstore driver can be a module, we must have
timekeeping_suspended exported.
Reported-by: Doug Anderson <dianders@xxxxxxxxxxxx>
Cc: Anton Vorontsov <cbouatmailru@xxxxxxxxx>
Cc: John Stultz <johnstul@xxxxxxxxxx>
Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
---
v2:
- export needed for timekeeping_suspended (thanks to Fengguang Wu).
---
fs/pstore/ram.c | 8 +++++++-
kernel/time/timekeeping.c | 1 +
2 files changed, 8 insertions(+), 1 deletion(-)
diff --git a/fs/pstore/ram.c b/fs/pstore/ram.c
index 1a4f6da..6d014e0 100644
--- a/fs/pstore/ram.c
+++ b/fs/pstore/ram.c
@@ -171,7 +171,13 @@ static size_t ramoops_write_kmsg_hdr(struct persistent_ram_zone *prz)
struct timeval timestamp;
size_t len;
- do_gettimeofday(×tamp);
+ /* Handle dumping before timekeeping has resumed. */
+ if (unlikely(timekeeping_suspended)) {
+ timestamp.tv_sec = 0;
+ timestamp.tv_usec = 0;
+ } else
+ do_gettimeofday(×tamp);
+