Skip to contentSkip to conversation

SMP Race in brelse

DISCUSSION Daniel Phillips 2 participants 2 messages Feb 2, 2001

lore
Share & tools
lore
#0 lore raw

There is a rare SMP race in brelse:

1138 void __brelse(struct buffer_head * buf)
1139 {
1140 if (atomic_read(&buf->b_count)) {
1141 atomic_dec(&buf->b_count);
1142 return;
1143 }
1144 printk("VFS: brelse: Trying to free free buffer\n");
1145 }

cpu1 cpu2

Starting with buf->b_count = 1, if we have:

if (atomic_read(&buf->b_count))

					 if (atomic_read(&buf->b_count))
       atomic_dec(&buf->b_count);
					      atomic_dec(&buf->b_count);

buf->b_count is now 0, but it should be -1, we fail to to report an erroneous extra brelse.

Signature · 5 lines
Daniel
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
#1 lore raw

Hi Daniel,

That is very well known (I posted about it many years ago :) but, as Ingo (or someone else? maybe sct or Alan? actually, I think it was Andrea) explained it is not a bug -- for that if() is only for purpose of catching bad callers (which, in perfect world, shouldn't exist). The whole brelse() could just contain a single atomic_dec() and that is all.

Regards,
Tigran

PS. Having thought about it -- it was neither sct, nor Alan, nor even Andrea -- it was Linus who explained it :)

Quoted 15 lines · Daniel Phillips

There is a rare SMP race in brelse:

1138 void __brelse(struct buffer_head * buf)
1139 {
1140 if (atomic_read(&buf->b_count)) {
1141 atomic_dec(&buf->b_count);
1142 return;
1143 }
1144 printk("VFS: brelse: Trying to free free buffer\n");
1145 }

cpu1 cpu2

Starting with buf->b_count = 1, if we have:

Quoted 9 lines
   if (atomic_read(&buf->b_count))
					 if (atomic_read(&buf->b_count))
       atomic_dec(&buf->b_count);
					      atomic_dec(&buf->b_count);

buf->b_count is now 0, but it should be -1, we fail to to report an erroneous extra brelse.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/