On Fri, 5 Aug 2022, Matthew Wilcox wrote:
> On Mon, Aug 01, 2022 at 11:01:40AM -0400, Mikulas Patocka wrote:
> >
> > In most cases, the buffer is set uptodate while it is locked, so that
> > there is no race on the uptodate flag (the race exists on the locked
> > flag). Are there any cases where the uptodate flag is modified on unlocked
> > buffer, so that it needs special treatment too?
>
> I think you misunderstand the purpose of locked/uptodate. At least
> for pages, the lock flag does not order access to the data in the page.
> Indeed, the contents of the page can be changed while you hold the lock.
> But the uptodate flag does order access to the data. At the point where
> you can observe the uptodate flag set, you know the contents of the page
> have been completely read from storage. And you don't need to hold the
> lock to check the uptodate flag. So this is wrong:
>
> buffer_lock()
> *data = 0x12345678;
> buffer_set_uptodate_not_ordered()
> buffer_unlock_ordered()
>
> because a reader can do:
>
> while (!buffer_test_uptodate()) {
> buffer_lock();
> buffer_unlock();
> }
> x = *data;
>
> and get x != 0x12345678 because the compiler can move the
> buffer_set_uptodate_not_ordered() before the store to *data.
Thanks for explanation. Would you like this patch?
From: Mikulas Patocka <mpatocks@xxxxxxxxxx>
Let's have a look at this piece of code in __bread_slow:
get_bh(bh);
bh->b_end_io = end_buffer_read_sync;
submit_bh(REQ_OP_READ, 0, bh);
wait_on_buffer(bh);
if (buffer_uptodate(bh))
return bh;
Neither wait_on_buffer nor buffer_uptodate contain any memory barrier.
Consequently, if someone calls sb_bread and then reads the buffer data,
the read of buffer data may be executed before wait_on_buffer(bh) on
architectures with weak memory ordering and it may return invalid data.
Fix this bug by adding a write memory barrier to set_buffer_uptodate and a
read memory barrier to buffer_uptodate (in the same way as
folio_test_uptodate and folio_mark_uptodate).
We also add a barrier to buffer_locked - it pairs with a barrier in
unlock_buffer.
Signed-off-by: Mikulas Patocka <mpatocka@xxxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Index: linux-2.6/include/linux/buffer_head.h
===================================================================
--- linux-2.6.orig/include/linux/buffer_head.h
+++ linux-2.6/include/linux/buffer_head.h
@@ -117,10 +117,8 @@ static __always_inline int test_clear_bu
* of the form "mark_buffer_foo()". These are higher-level functions which
* do something in addition to setting a b_state bit.
*/
-BUFFER_FNS(Uptodate, uptodate)
BUFFER_FNS(Dirty, dirty)
TAS_BUFFER_FNS(Dirty, dirty)
-BUFFER_FNS(Lock, locked)
BUFFER_FNS(Req, req)
TAS_BUFFER_FNS(Req, req)
BUFFER_FNS(Mapped, mapped)
@@ -135,6 +133,49 @@ BUFFER_FNS(Meta, meta)
BUFFER_FNS(Prio, prio)
BUFFER_FNS(Defer_Completion, defer_completion)
+static __always_inline void set_buffer_uptodate(struct buffer_head *bh)
+{
+ /*
+ * make it consistent with folio_mark_uptodate
+ * pairs with smp_acquire__after_ctrl_dep in buffer_uptodate
+ */
+ smp_wmb();
+ set_bit(BH_Uptodate, &bh->b_state);
+}
+
+static __always_inline void clear_buffer_uptodate(struct buffer_head *bh)
+{
+ clear_bit(BH_Uptodate, &bh->b_state);
+}
+
+static __always_inline int buffer_uptodate(const struct buffer_head *bh)
+{
+ bool ret = test_bit(BH_Uptodate, &bh->b_state);
+ /*
+ * make it consistent with folio_test_uptodate
+ * pairs with smp_wmb in set_buffer_uptodate
+ */
+ if (ret)
+ smp_acquire__after_ctrl_dep();
+ return ret;
+}
+
+static __always_inline void set_buffer_locked(struct buffer_head *bh)
+{
+ set_bit(BH_Lock, &bh->b_state);
+}
+
+static __always_inline int buffer_locked(const struct buffer_head *bh)
+{
+ bool ret = test_bit(BH_Lock, &bh->b_state);
+ /*
+ * pairs with smp_mb__after_atomic in unlock_buffer
+ */
+ if (!ret)
+ smp_acquire__after_ctrl_dep();
+ return ret;
+}
+
#define bh_offset(bh) ((unsigned long)(bh)->b_data & ~PAGE_MASK)
/* If we *know* page->private refers to buffer_heads */
