close() behaviour changes depending on whether file exists or not?

[Chris Friesen]

I'm running 3.14 on my laptop, and I'm seeing some unexpected behaviour when I 
write out to a big file.  Basically if the

If I run

strace dd if=/dev/zero of=bigfile bs=1M count=1000;cat /proc/meminfo |egrep 
"Dirty|Writeback"

I see the final close() of "bigfile" (fd 1 in the strace output) complete right 
away, and I get speeds of 1.5GB/s.  When it completes I see "Dirty" in 
/proc/meminfo right at about a GB of data.

However, if I then immediately repeat the test (so the "bigfile" file exists 
already) I see the final close() delay until the writeback is done, almost as 
though there is an implicit fsync() call.  When it completes /proc/meminfo has a 
"Dirty" value of basically nothing and a "Writeback" value of about 67MB.

Is this a known issue?  I checked a 3.10 kernel that I had hanging around and it 
showed the same behaviour.

Thanks,
Chris


PS: Please CC me on reply, I'm not subscribed to the list.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/