Re: [PATCH 1/1] ext4: Prevent an infinite loop in the lazyinit thread.

[Jan Kara <jack@suse.cz>]

On Wed 06-11-24 14:47:41, Mathieu Othacehe wrote:
> Use ktime_get_ns instead of ktime_get_real_ns when computing the lr_timeout
> not to be affected by system time jumps.
> 
> Use a boolean instead of the MAX_JIFFY_OFFSET value to determine whether
> the next_wakeup value has been set. Comparing elr->lr_next_sched to
> MAX_JIFFY_OFFSET can cause the lazyinit thread to loop indefinitely.
> 
> Co-developed-by: Lukas Skupinski <lukas.skupinski@landisgyr.com>
> Signed-off-by: Lukas Skupinski <lukas.skupinski@landisgyr.com>
> Signed-off-by: Mathieu Othacehe <othacehe@gnu.org>

Nice catch! The patch looks good so feel free to add:

Reviewed-by: Jan Kara <jack@suse.cz>

But the analysis you've done in patch 0/1 would ideally be in the changelog
of this patch so that we can easily get back to it in the future in git logs.
Maybe Ted can handle that on commit?

								Honza

> ---
>  fs/ext4/super.c | 20 +++++++++++++-------
>  1 file changed, 13 insertions(+), 7 deletions(-)
> 
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index 9fcf44064c6a6..b4839ccd83ad5 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -3712,12 +3712,12 @@ static int ext4_run_li_request(struct ext4_li_request *elr)
>  		ret = 1;
>  
>  	if (!ret) {
> -		start_time = ktime_get_real_ns();
> +		start_time = ktime_get_ns();
>  		ret = ext4_init_inode_table(sb, group,
>  					    elr->lr_timeout ? 0 : 1);
>  		trace_ext4_lazy_itable_init(sb, group);
>  		if (elr->lr_timeout == 0) {
> -			elr->lr_timeout = nsecs_to_jiffies((ktime_get_real_ns() - start_time) *
> +			elr->lr_timeout = nsecs_to_jiffies((ktime_get_ns() - start_time) *
>  				EXT4_SB(elr->lr_super)->s_li_wait_mult);
>  		}
>  		elr->lr_next_sched = jiffies + elr->lr_timeout;
> @@ -3777,8 +3777,9 @@ static int ext4_lazyinit_thread(void *arg)
>  
>  cont_thread:
>  	while (true) {
> -		next_wakeup = MAX_JIFFY_OFFSET;
> +		bool next_wakeup_initialized = false;
>  
> +		next_wakeup = 0;
>  		mutex_lock(&eli->li_list_mtx);
>  		if (list_empty(&eli->li_request_list)) {
>  			mutex_unlock(&eli->li_list_mtx);
> @@ -3791,8 +3792,11 @@ static int ext4_lazyinit_thread(void *arg)
>  					 lr_request);
>  
>  			if (time_before(jiffies, elr->lr_next_sched)) {
> -				if (time_before(elr->lr_next_sched, next_wakeup))
> +				if (!next_wakeup_initialized ||
> +				    time_before(elr->lr_next_sched, next_wakeup)) {
>  					next_wakeup = elr->lr_next_sched;
> +					next_wakeup_initialized = true;
> +				}
>  				continue;
>  			}
>  			if (down_read_trylock(&elr->lr_super->s_umount)) {
> @@ -3820,16 +3824,18 @@ static int ext4_lazyinit_thread(void *arg)
>  				elr->lr_next_sched = jiffies +
>  					get_random_u32_below(EXT4_DEF_LI_MAX_START_DELAY * HZ);
>  			}
> -			if (time_before(elr->lr_next_sched, next_wakeup))
> +			if (!next_wakeup_initialized ||
> +			    time_before(elr->lr_next_sched, next_wakeup)) {
>  				next_wakeup = elr->lr_next_sched;
> +				next_wakeup_initialized = true;
> +			}
>  		}
>  		mutex_unlock(&eli->li_list_mtx);
>  
>  		try_to_freeze();
>  
>  		cur = jiffies;
> -		if ((time_after_eq(cur, next_wakeup)) ||
> -		    (MAX_JIFFY_OFFSET == next_wakeup)) {
> +		if (!next_wakeup_initialized || time_after_eq(cur, next_wakeup)) {
>  			cond_resched();
>  			continue;
>  		}
> -- 
> 2.46.0
> 
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR