On 12/4/2013 1:33 AM, Vyacheslav Dubeyko wrote:
On Tue, 2013-12-03 at 16:29 -0800, Andrew Morton wrote:
[snip]
It converts every printk in nilfs2 into pr_foo_ratelimited (and bloats
nilfs2.ko by 5k in the process). Isn't this rather overkill?
I have converted not every printk() in nilfs2 but I agree that printk()
was changed in many places by ratelimited version. So, yes, it can be
not very good idea. But such replacement was made for code that can emit
really many count of practically identical error messages. And there are
situation of sophisticated issues in nilfs2 when huge amount of error
messages simply hide an important information about the issue. As a
result, my goal was to reduce amount of repeatable error messages.
So, what could you recommend as possible and proper solution?
I think this will help. However, the idea I had in mind originally was
for nilfs to "give up" sooner.
I suspect that my nilfs partition became corrupt for hardware or
hardware-driver reasons. So lets ignore that part for now.
With the data on the drive being corrupt, it appeared that nilfs
encountered an invalid directory (possibly just a long string of NUL
bytes?) and emitted more than a million errors about invalid structures,
triggering the soft-lockup watchdog and rebooting the system. When I
recompiled my kernel with soft-lockup set to 5 minutes, it simply filled
my log files.
[10796.519283] NILFS error (device sdf1): nilfs_check_page: bad entry in
directory #2383620: rec_len is smaller than minimal - offset=1143304192,
inode=0, rec_len=0, name_len=0
I haven't read the code involved, but what I think should happen is that
on the very *first* error, it should return an I/O error to userland.
Also, the partition was set to "errors=remount-ro", so the very first
error should also make the filesystem read-only, correct?
-Mike
--
To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
