2003-03-25 17:29:42

by Stephen C. Tweedie

[permalink] [raw]
Subject: [Patch 8/8] 2.4: Fix flushtime ordering on BUF_DIRTY list

[From Andrew Morton.] Set the flushtime on committing buffers when they
are actually sent to the dirty list, not when they are first dirtied.
Avoids some starvation scenarios when the flushtimes on BUF_DIRTY get
out-of-order.

--- linux-2.4-ext3push/fs/jbd/transaction.c.=K0007=.orig 2003-03-25 10:59:15.000000000 +0000
+++ linux-2.4-ext3push/fs/jbd/transaction.c 2003-03-25 10:59:15.000000000 +0000
@@ -1138,7 +1138,6 @@ int journal_dirty_metadata (handle_t *ha

spin_lock(&journal_datalist_lock);
set_bit(BH_JBDDirty, &bh->b_state);
- set_buffer_flushtime(bh);

J_ASSERT_JH(jh, jh->b_transaction != NULL);

@@ -2090,6 +2089,13 @@ void journal_file_buffer(struct journal_
spin_unlock(&journal_datalist_lock);
}

+static void jbd_refile_buffer(struct buffer_head *bh)
+{
+ if (buffer_dirty(bh) && (bh->b_list != BUF_DIRTY))
+ set_buffer_flushtime(bh);
+ refile_buffer(bh);
+}
+
/*
* Remove a buffer from its current buffer list in preparation for
* dropping it from its current transaction entirely. If the buffer has
@@ -2110,7 +2116,7 @@ void __journal_refile_buffer(struct jour
__journal_unfile_buffer(jh);
jh->b_transaction = NULL;
/* Onto BUF_DIRTY for writeback */
- refile_buffer(jh2bh(jh));
+ jbd_refile_buffer(jh2bh(jh));
return;
}