Re: [PATCH 2/4] mm/damon/sysfs: check contexts->nr before clear_schemes_tried_regions

[SeongJae Park <sj@kernel.org>]

On Thu, 19 Mar 2026 15:57:40 +0000 Josh Law <objecting@objecting.org> wrote:

> The CLEAR_SCHEMES_TRIED_REGIONS command accesses contexts_arr[0]
> without verifying nr_contexts >= 1, causing a NULL pointer dereference
> when no context is configured. Add the missing check.

Nice catch, thank you!

Privileged users can trigger this using DAMON sysfs interface.  E.g.,

    # cd /sys/kernel/mm/damon/admin/kdamonds/
    # echo 1 > nr_kdamonds
    # echo clear_schemes_tried_regions > state
    killed
    # dmesg
    [...]
    [63541.377604] BUG: kernel NULL pointer dereference, address: 0000000000000000
    [...]

Privileged users can do anything even worse than this, but they might also do
this by a mistake.

So this deserves Fixes: and Cc stable.

>
> Signed-off-by: Josh Law <objecting@objecting.org>
> ---
>  mm/damon/sysfs.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/mm/damon/sysfs.c b/mm/damon/sysfs.c
> index b573b9d60784..36ad2e8956c9 100644
> --- a/mm/damon/sysfs.c
> +++ b/mm/damon/sysfs.c
> @@ -1769,6 +1769,8 @@ static int damon_sysfs_handle_cmd(enum damon_sysfs_cmd cmd,
>       case DAMON_SYSFS_CMD_UPDATE_SCHEMES_TRIED_REGIONS:
>               return damon_sysfs_update_schemes_tried_regions(kdamond, false);
>       case DAMON_SYSFS_CMD_CLEAR_SCHEMES_TRIED_REGIONS:
> +             if (kdamond->contexts->nr != 1)
> +                     return -EINVAL;
>               return damon_sysfs_schemes_clear_regions(
>                       kdamond->contexts->contexts_arr[0]->schemes);
>       case DAMON_SYSFS_CMD_UPDATE_SCHEMES_EFFECTIVE_QUOTAS:
> --
> 2.34.1

So this patch looks good as an individual fix for the individual bug, but...

Sashiko commented.

# review url: https://sashiko.dev/#/patchset/20260319155742.186627-3-objecting@objecting.org

: Does this missing check also affect other manual commands?
:
: If a user writes UPDATE_SCHEMES_STATS, UPDATE_SCHEMES_EFFECTIVE_QUOTAS,
: or UPDATE_TUNED_INTERVALS to the state file after setting nr_contexts
: to 0, damon_sysfs_handle_cmd() queues the corresponding callback via
: damon_sysfs_damon_call().
:
: When the kdamond thread executes the callback, it appears functions like
: damon_sysfs_upd_schemes_stats() access contexts_arr[0] without verifying
: contexts->nr:
:
: static int damon_sysfs_upd_schemes_stats(void *data)
: {
:         struct damon_sysfs_kdamond *kdamond = data;
:         struct damon_ctx *ctx = kdamond->damon_ctx;
:
:         damon_sysfs_schemes_update_stats(
:                         kdamond->contexts->contexts_arr[0]->schemes, ctx);
:         return 0;
: }
:
: Could this result in a similar NULL pointer dereference if these commands
: are triggered while no context is configured?

Sashiko is correct.  Privileged users can trigger the issues like below.

# damo start
# cd /sys/kernel/mm/damon/admin/kdamonds/0
# echo 0 > contexts/nr_contexts
# echo update_schemes_stats > state
# echo update_schemes_effective_quotas > state
# echo update_tuned_intervals > state

Not necessarily blocker of this patch, but seems all the issues are in a same
category.  The third patch of this series is also fixing one of the category
bugs.  How about fixing all at once by checking kdamond->contexts->nr at the
beginning of damon_sysfs_handle_cmd(), like below?

--- a/mm/damon/sysfs.c
+++ b/mm/damon/sysfs.c
@@ -2404,6 +2404,9 @@ static int damon_sysfs_update_schemes_tried_regions(
 static int damon_sysfs_handle_cmd(enum damon_sysfs_cmd cmd,
                struct damon_sysfs_kdamond *kdamond)
 {
+       if (cmd != DAMON_SYSFS_CMD_OFF && kdamond->contexts->nr != 1)
+               return -EINVAL;
+
        switch (cmd) {
        case DAMON_SYSFS_CMD_ON:
                return damon_sysfs_turn_damon_on(kdamond);

If we pick this, Fixes: would be deserve to the oldest buggy commit that
introduced the first bug of this category.  It is indeed quite old.

Fixes: 0ac32b8affb5 ("mm/damon/sysfs: support DAMOS stats")
Cc: <stable@vger.kernel.org> # 5.18.x


Thanks,
SJ