Am 2012-06-08 20:44, schrieb Jens Axboe:

On 2012-06-08 19:20, Georg Schönberger wrote:

Am 2012-06-08 14:47, schrieb Jens Axboe:

On 06/01/2012 12:06 PM, Georg Schönberger wrote:

Hello again,

I am having some troubles with running the fio example "ssd-test". I
changed the global section to decrease the size:
[global]
bs=4k
ioengine=libaio
iodepth=4
size=100M
direct=1
runtime=60
filename=ssd.test.file

If I am running Fio I get the following error:
----------------------------------------
fio.git/examples$ fio ssd-test
seq-read: (g=0): rw=read, bs=4K-4K/4K-4K, ioengine=libaio, iodepth=4
rand-read: (g=1): rw=randread, bs=4K-4K/4K-4K, ioengine=libaio, iodepth=4
seq-write: (g=2): rw=write, bs=4K-4K/4K-4K, ioengine=libaio, iodepth=4
rand-write: (g=3): rw=randwrite, bs=4K-4K/4K-4K, ioengine=libaio, iodepth=4
fio-2.0.8
Starting 4 processes
fio: pid=17997, err=22/file:filesetup.c:529, func=open(ssd.test.file),
error=Invalid argument
fio: pid=17998, err=22/file:filesetup.c:529, func=open(ssd.test.file),
error=Invalid argument
fio: pid=17999, err=22/file:filesetup.c:529, func=open(ssd.test.file),
error=Invalid argument
fio: pid=18000, err=22/file:filesetup.c:529, func=open(ssd.test.file),
error=Invalid argument

Looks like the file system hosting the test file does not support
O_DIRECT.

The file system is ext4.
Kernel Version 3.2.0-24-generic

And you aren't running the job file in some other location? ext4 would
definitely work. Please do:

# strace -o strace.out -f fio ssd-test

and send me strace.out (gzip'ed). Also include a pwd from where you are
running it, and output of the 'mount' command.

Here is the solution to my problem (answered from Jens Axboe):
ecryptfs does not support O_DIRECT IO. So that was the explanation for the problem.

Also a patch so that FIO gives some further information now:
Sure, I think it'd be a good idea if fio also complains that this is the likely explanation. Fio does this if the first IO errors, but it doesn't catch cases where O_DIRECT simply isn't supported. Can you try this patch, see if it complains as it should?

The patch has already been commited (also from Jens):
diff --git a/filesetup.c b/filesetup.c
index a1ad026..b1f73f7 100644
--- a/filesetup.c
+++ b/filesetup.c
@@ -526,6 +526,11 @@ open_again:

         snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name);

+        if (__e == EINVAL && (flags & OS_O_DIRECT)) {
+            log_err("fio: looks like your file system does not " \
+                "support direct=1/buffered=0\n");
+        }
+
         td_verror(td, __e, buf);
     }