Possible error in fs/buffer.c
While code surfing I found what appears to be a problem with the
following code in fs/buffer.c {function mark_buffer_inode_dirty}:
if (list_empty(&bh->b_assoc_buffers))
buffer_insert_list(&buffer_mapping->private_lock,
bh, &mapping->private_list);
From my research, the job of buffer_insert_list is to place the list contents of bh->b_assoc_buffers onto the list mapping->private list. The above code segment will only call buffer_insert_list if b_assoc_buffers is empty. Why waste the time, and resources attempting to add an empty list to the end of private_list?
If this is a problem, then there are two possibilities for fixing
this code:
1) This was just a mistake
add "!" in the if, this only adds the list if there is something to add.
- if (list_empty(&bh->b_assoc_buffers))
+ if (!list_empty(&bh->b_assoc_buffers))
2) Some of the assoc buffers may already be on the private_list.
If this is a possibility then this routine will need more work.
Please CC me on all responses since I am not a member of this list.
Signature · 4 lines
Cheers Rob Maloy [email protected] (410) 854-6637
Quoted 7 lines · Robert W. Maloy
While code surfing I found what appears to be a problem with the
following code infs/buffer.c{functionmark_buffer_inode_dirty}:if (list_empty(&bh->b_assoc_buffers)) buffer_insert_list(&buffer_mapping->private_lock, bh, &mapping->private_list);
This is saying "is the buffer is not on any list, then place it on the mapping's list now. If the buffer _is_ already on some list then we know it must already be on mapping->private_list, so we have nothing to do".
Re-dirtying an already-dirty buffer is an extremely common case, and here we look for an opportunity to avoid dirtying a cacheline or two.
Quoted 9 lines
Andrew Morton wrote:
This is saying "is the buffer is not on any list, then place it on the mapping's list now. If the buffer _is_ already on some list then we know it must already be onmapping->private_list, so we have nothing to do".Re-dirtying an already-dirty buffer is an extremely common case, and here we look for an opportunity to avoid dirtying a cacheline or two.
This makes some since except that
buffer_insert_list(&buffer_mapping->private_lock,
bh, &mapping->private_list); only adds &bh->b_assoc_buffers to the end of mapping->private_list
not the buffer itself.
Signature · 4 lines
Cheers Rob Maloy [email protected] (410) 854-6637