Jonah Sherman <jsherman@stuy.edu> wrote:
>
> I have come across a bug in the loop driver.
So you have. See, the `dd' fills the machine up with dirty memory and the
loop driver wants to take a copy of all that memory to write it out. It takes
this copy inside page reclaim, where we are allowed to use *all* memory.
So obviously, as there is so much memory to be written, we just run out of
the stuff.
The loop driver has a tendency to do this sort of thing. Whenever it does,
we slap another bandaid on it.
diff -puN drivers/block/loop.c~loop-hack drivers/block/loop.c
--- 25/drivers/block/loop.c~loop-hack 2003-02-24 21:21:11.000000000 -0800
+++ 25-akpm/drivers/block/loop.c 2003-02-24 21:45:13.000000000 -0800
@@ -447,7 +447,22 @@ static struct bio *loop_get_buffer(struc
goto out_bh;
}
- bio = bio_copy(rbh, GFP_NOIO, rbh->bi_rw & WRITE);
+ /*
+ * When called on the page reclaim -> writepage path, this code can
+ * trivially consume all memory. So we drop PF_MEMALLOC to avoid
+ * stealing all the page reserves and throttle to the writeout rate.
+ * pdflush will have been woken by page reclaim. Let it do its work.
+ */
+ do {
+ int flags = current->flags;
+
+ current->flags &= ~PF_MEMALLOC;
+ bio = bio_copy(rbh, (GFP_ATOMIC & ~__GFP_HIGH) | __GFP_NOWARN,
+ rbh->bi_rw & WRITE);
+ current->flags = flags;
+ if (bio == NULL)
+ blk_congestion_wait(WRITE, HZ/10);
+ } while (bio == NULL);
bio->bi_end_io = loop_end_io_transfer;
bio->bi_private = rbh;
_
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Feb 28 2003 - 22:00:26 EST