Hey Samuel,
Oh wow...thank you so much, you have no idea how happy i am that this got
in.
Thank you so so much.
I will send the separate small patches to address the comments, but this
one is just to thank you and to try to answer the question.
sync should absolutely trigger a sync, in fact you can see in pager.c in
diskfs_sync_everything we call
journal_commit_running_transaction ();
Which calls sync more than once (perhaps too many times)...
I just tested it without any delay like this
echo c > c && sync
and added a log warn statement in flush_to_disk and inside journal.c and
inside diskfs_sync_everything
void
diskfs_sync_everything (int wait)
{
error_t sync_one (void *v_p)
{
struct pager *p = v_p;
pager_sync (p, wait);
return 0;
}
write_all_disknodes ();
JRNL_LOG_WARN("About to commit");
/* We only commit if there is a journal and we have a running transaction
*/
journal_commit_running_transaction ();
ports_bucket_iterate (file_pager_bucket, sync_one);
/* Do things on the the disk pager. */
sync_global (wait);
if (wait && !ext2_journal)
{
error_t err = store_sync (store);
if (err && err != EOPNOTSUPP && err != D_INVALID_OPERATION)
ext2_warning ("device flush failed: %s", strerror (err));
}
}
```
static void
flush_to_disk (void)
{
error_t err = store_sync (store);
JRNL_LOG_WARN("Calling store_sync");
/* Ignore EOPNOTSUPP (drivers), but warn on real I/O errors */
if (err && err != EOPNOTSUPP)
JRNL_LOG_WARN ("Device flush failed: %s", strerror (err));
}
```
my log is:
```
ext2fs: part:5:device:wd0: warning: [JOURNAL] About to commit
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
ext2fs: part:5:device:wd0: warning: [JOURNAL] Calling store_sync
```
let me know if i misunderstood something, its been a bit of time and i have
started to forget some of the details.
What can happen if there is a delay when echo is executed and sync is
called the background thread could have committed in the meantime.
Thanks again,
Milos
On Wed, May 27, 2026 at 6:00 PM Samuel Thibault <[email protected]>
wrote:
> Do we really have enough sync calls?
>
> I am implementing the flush call for ide/ahci drivers in gnumach, and I
> am not seeing that many sync requests. Notably,
>
> $ echo a > a
> $ sync
>
> does not trigger a sync. Shouldn't it?
>
> Samuel
>
