Hi,
On 04/06/15 11:27, Andrew Price wrote:
On 04/06/15 09:42, Steven Whitehouse wrote:
Hi,
Will glibc do the fallback path, or just return this as an error? I
think thats worth checking as it would be nice it it would transparently
fall back in this case,
You only get the fallback with posix_fallocate() but many applications
will use fallocate() directly for various reasons¹.
On 03/06/15 22:30, Abhi Das wrote:
We cannot provide an efficient implementation due to the headers
on the data blocks, so there doesn't seem much point in having it.
I'm not sure I like the idea that fallocate() could work in one
directory and fail in another... What exactly is the issue here? Is it
just the journal space required or that writing the data block headers
would be too slow?
Andy
The issue is that data blocks for journaled data have metadata headers
on them. Although I guess that would not affect the normal journaled
data. but it would affect the journaled data used by the various kernel
internal files (i.e. on the meta fs).
The idea of fallocate is to be able to allocate the zero the blocks
efficiently, but if they have to go through the journal anyway, then it
is going to be really slow, so not a lot of point in doing it. We should
definitely try and encourage applications to use posix_fallocate where
possible, so that the fallback will work, although I know that we cannot
cover all situations.
The reality is that virtually nobody uses jdata files anyway, so this is
not likely to affect anybody, otherwise they'd have reported this as not
working before now,
Steve.
¹ https://sourceware.org/bugzilla/show_bug.cgi?id=15661
Resolves: rhbz#1221331
Signed-off-by: Abhi Das <a...@redhat.com>
---
fs/gfs2/file.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index c706c6d..8252115 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -917,7 +917,7 @@ static long gfs2_fallocate(struct file *file, int
mode, loff_t offset, loff_t le
struct gfs2_holder gh;
int ret;
- if (mode & ~FALLOC_FL_KEEP_SIZE)
+ if ((mode & ~FALLOC_FL_KEEP_SIZE) || gfs2_is_jdata(ip))
return -EOPNOTSUPP;
mutex_lock(&inode->i_mutex);
