Re: Mapped pages handling in shrink_cache()

[Andrew Morton]

[e-mail] wrote:
> 
> Hi, In the following code from shrink_cache()
> 
>           if (PageDirty(page) && is_page_cache_freeable(page) && page->mapping)
> {
>                .
>                .
>                .
> 
>                int (*writepage)(struct page *);
> 
>                writepage = page->mapping->a_ops->writepage;
>                if ((gfp_mask & __GFP_FS) && writepage) {
>                     ClearPageDirty(page);
>                     SetPageLaunder(page);
>                     page_cache_get(page);
>                     spin_unlock(&pagemap_lru_lock);
> 
>                     writepage(page);
>                     page_cache_release(page);
> 
>                     spin_lock(&pagemap_lru_lock);
>                     continue;      <<<<<<< shouldn't the page be unlocked before
>  continuing with the next page <<<<<
>                }
> 

The page is unlocked when IO completes, in interrupt context.
See end_buffer_io_async().