All readahead is currently capped to a maximum of the device readahead limit, which defaults to 128k. For heuristics-based readahead this makes perfect sense, too, but unfortunately the limit is also applied to the explicit readahead() or madvise(MADV_WILLNEED) syscalls, and 128k is an awfully low limit, particularly for bigger machines. It's not unreasonable for a user on a 100G machine to say, read this 1G file, and read it now, I'm going to access the whole thing shortly.
Since both readahead() and MADV_WILLNEED take an explicit length parameter, it seems weird to truncate that request quietly. Just do what the user asked for and leave the limiting to the heuristics. Signed-off-by: Johannes Weiner <[email protected]> --- mm/readahead.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/mm/readahead.c b/mm/readahead.c index 20e58e8..6d182db 100644 --- a/mm/readahead.c +++ b/mm/readahead.c @@ -212,7 +212,6 @@ int force_page_cache_readahead(struct address_space *mapping, struct file *filp, if (unlikely(!mapping->a_ops->readpage && !mapping->a_ops->readpages)) return -EINVAL; - nr_to_read = min(nr_to_read, inode_to_bdi(mapping->host)->ra_pages); while (nr_to_read) { int err; @@ -485,6 +484,7 @@ void page_cache_sync_readahead(struct address_space *mapping, /* be dumb */ if (filp && (filp->f_mode & FMODE_RANDOM)) { + req_size = min(req_size, inode_to_bdi(mapping->host)->ra_pages); force_page_cache_readahead(mapping, filp, offset, req_size); return; } -- 2.7.1
