[04/24] lustre: llite: fully disable readahead in kernel I/O path

Commit Message

James Simmons Sept. 6, 2022, 1:55 a.m. UTC

From: Qian Yingjin <[e-mail]>

In the new kernel (rhel9 or ubuntu 2204), the readahead path may
be out of the control of Lustre CLIO engine:

generic_file_read_iter()
  ->filemap_read()
    ->filemap_get_pages()
      ->page_cache_sync_readahead()
        ->page_cache_sync_ra()

void page_cache_sync_ra()
{
    if (!ractl->ra->ra_pages || blk_cgroup_congested()) {
	if (!ractl->file)
        	return;
	req_count = 1;
	do_forced_ra = true;
    }

    /* be dumb */
    if (do_forced_ra) {
	force_page_cache_ra(ractl, req_count);
	return;
    }
    ...
}

From the kernel readahead code, even if read-ahead is disabled
(via @ra_pages == 0), it still issues this request as read-ahead
as we will need it to satisfy the requested range. The forced
read-ahead will do the right thing and limit the read to just
the requested range, which we will set to 1 page for this case.

Thus it can not totally avoid the read-ahead in the kernel I/O
path only by setting @ra_pages with 0.
To fully disable the read-ahead in the Linux kernel I/O path, we
still need to set @io_pages to 0, it will set I/O range to 0 in
@force_page_cache_ra():

void force_page_cache_ra()
{
    ...
    max_pages = = max_t(unsigned long, bdi->io_pages,
			ra->ra_pages);
    nr_to_read = min_t(unsigned long, nr_to_read, max_pages);
    while (nr_to_read) {
	...
    }
    ...
}

After set bdi->io_pages with 0, it can pass the sanity/101j.

WC-bug-id: https://jira.whamcloud.com/browse/LU-16019
Lustre-commit: f0cf7fd3cccb2313f ("LU-16019 llite: fully disable readahead in kernel I/O path")
Signed-off-by: Qian Yingjin <[e-mail]>
Reviewed-on: https://review.whamcloud.com/47993
Reviewed-by: Andreas Dilger <[e-mail]>
Reviewed-by: Li Xi <[e-mail]>
Reviewed-by: Oleg Drokin <[e-mail]>
Signed-off-by: James Simmons <[e-mail]>
---
 fs/lustre/llite/llite_lib.c | 1 +
 1 file changed, 1 insertion(+)

Patch

diff --git a/fs/lustre/llite/llite_lib.c b/fs/lustre/llite/llite_lib.c
index 5daced0..5931258 100644
--- a/fs/lustre/llite/llite_lib.c
+++ b/fs/lustre/llite/llite_lib.c
@@ -1261,6 +1261,7 @@  int ll_fill_super(struct super_block *sb)
 
 	/* disable kernel readahead */
 	sb->s_bdi->ra_pages = 0;
+	sb->s_bdi->io_pages = 0;
 
 	/* Call ll_debugsfs_register_super() before lustre_process_log()
 	 * so that "llite.*.*" params can be processed correctly.