Re: [PATCH] xfsdump: fix race condition between lseek() and read()/write()

On 07/12/2017 01:46 PM, Eric Sandeen wrote:

Sorry, I was wrong ...

quoted

diff --git a/inventory/inv_idx.c b/inventory/inv_idx.c
index 95529e8..cd9b9cb 100644
--- a/inventory/inv_idx.c
+++ b/inventory/inv_idx.c

@@ -341,8 +341,8 @@ idx_put_sesstime( inv_sestoken_t tok, bool_t whichtime)

        rval = GET_REC_NOLOCK( fd, &ent, sizeof( invt_entry_t ),
                                tok->sd_invtok->d_invindex_off);

This sets the offset to d_invindex_off

... and reads sizeof(invt_entry_t), advancing by that much.

quoted

 			      ent.ie_timeperiod.tp_start,
 			      ent.ie_timeperiod.tp_end );
 #endif
-	rval = PUT_REC_NOLOCK_SEEKCUR( fd, &ent, sizeof( invt_entry_t ),
-				      -(off64_t)(sizeof( invt_entry_t )));

This backed up one entry from the current offset...

which goes back to...

quoted

+	rval = PUT_REC_NOLOCK(fd, &ent, sizeof(invt_entry_t),
+			      tok->sd_invtok->d_invindex_off);

...d_invindex_off.  Sorry for the false alarm.

-eric

but this keeps it at the current offset, no?  Isn't this a different behavior?