Merge: Backport revert of commit "readahead: properly shorten readahead when falling back to do_page_cache_ra()"

MR: https://gitlab.com/redhat/centos-stream/src/kernel/centos-stream-10/-/merge_requests/166

Upstream found out that this commit can cause an integer underflow in
the fs readahead code, which causes it to read a huge amount of data,
from the storage device, stalling IO on the host,
which in turn can in some extreme cases (e.g using NFS or a hard disk)
even cause I/O timeouts.

In our case this caused VMs after v2v conversion to fail to finish first boot
script and bluescreen or otherwise hang during its execution because they
started to timeout on I/O.

JIRA: https://issues.redhat.com/browse/RHEL-55724

JIRA: https://issues.redhat.com/browse/RHEL-56929

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>

Approved-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Approved-by: Herton R. Krzesinski <herton@redhat.com>
Approved-by: Rafael Aquini <raquini@redhat.com>
Approved-by: CKI KWF Bot <cki-ci-bot+kwf-gitlab-com@redhat.com>

Merged-by: Jan Stancek <jstancek@redhat.com>

Author: jstancek

mm/readahead.c

@@ -453,8 +453,7 @@ void page_cache_ra_order(struct readahead_control *ractl,
 		struct file_ra_state *ra, unsigned int new_order)
 {
 	struct address_space *mapping = ractl->mapping;
-	pgoff_t start = readahead_index(ractl);
-	pgoff_t index = start;
+	pgoff_t index = readahead_index(ractl);
 	unsigned int min_order = mapping_min_folio_order(mapping);
 	pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT;
 	pgoff_t mark = index + ra->size - ra->async_size;
@@ -517,7 +516,7 @@ void page_cache_ra_order(struct readahead_control *ractl,
 	if (!err)
 		return;
 fallback:
-	do_page_cache_ra(ractl, ra->size - (index - start), ra->async_size);
+	do_page_cache_ra(ractl, ra->size, ra->async_size);
 }
 
 static unsigned long ractl_max_pages(struct readahead_control *ractl,