NFS error handling

[Trond Myklebust]

On Fri, 2007-11-09 at 17:15 +0100, Zoltan Menyhart wrote:
> I have got some kernel crashes in the following situation:
> 
> __do_page_cache_readahead():
> 	read_pages(... &page_pool, ...):
> 		mapping->a_ops->readpages(... pages, ...):
> 			nfs_readpages(... pages, ...):
> 				if (NFS_STALE(inode))
> 					return -ESTALE;
> 	BUG_ON(!list_empty(&page_pool));
> 
> I can understand that not much can be done to a stale file.
> However, crashing the system is too much.
> I thing the NFS layer should empty the list of the pages
> and mark each page with the PG_error bit.
> Other NFS errors should be handled in the same way.

Is the put_pages_list() call in read_pages() somehow buggy?

Trond