Skip to contentSkip to conversation

mm/damon/core: document damos_commit_dests() failure semantics

PATCH In review Josh Law 2 participants 4 messages Mar 19

Subsystems (2)
lore
Share & tools
lore

Josh Law

author
#0 lore raw

PATCH 1/1 mm/damon/core: document damos_commit_dests() failure semantics

Add a kernel-doc comment to damos_commit_dests() documenting its allocation failure contract: on -ENOMEM, the destination structure is left in a partially torn-down state that is safe to deallocate via damon_destroy_scheme(), but must not be reused for further commits.

This was unclear from the code alone and led to a separate patch attempting to reset nr_dests on failure. Make the intended usage explicit so future readers do not repeat the confusion.

---
 mm/damon/core.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
mm/damon/core.c +17 -0
index e233eb84a2d5..c884bb31c9b8 100644
@@ -1054,6 +1054,23 @@ static void damos_set_filters_default_reject(struct damos *s)
 		damos_filters_default_reject(&s->ops_filters);
 }
 
+/**
+ * damos_commit_dests() - Copy migration destinations from @src to @dst.
+ * @dst:	Destination structure to update.
+ * @src:	Source structure to copy from.
+ *
+ * If the number of destinations has changed, the old arrays in @dst are freed
+ * and new ones are allocated.  On success, @dst contains a full copy of
+ * @src's arrays and count.
+ *
+ * On allocation failure, @dst is left in a partially torn-down state: its
+ * arrays may be NULL and @nr_dests may not reflect the actual allocation
+ * sizes.  The structure remains safe to deallocate via damon_destroy_scheme(),
+ * but callers must not reuse @dst for further commits — it should be
+ * discarded.
+ *
+ * Return: 0 on success, -ENOMEM on allocation failure.
+ */
 static int damos_commit_dests(struct damos_migrate_dests *dst,
 		struct damos_migrate_dests *src)
 {
Signature · 1 line
2.34.1
maintainer Reviewed-by on v1 1/1
#1 lore raw

Hello Josh,

Quoted 4 lines · On Thu, 19 Mar 2026 07:13:32 +0000 Josh Law <[email protected]> wrote:

Add a kernel-doc comment to damos_commit_dests() documenting its allocation failure contract: on -ENOMEM, the destination structure is left in a partially torn-down state that is safe to deallocate via damon_destroy_scheme(), but must not be reused for further commits.

Nice patch. We intentionally do not use kernel-doc comment for static functions, to not encourage API document readers use the non-public functions. We still use kernel-doc like comment for static functions, to help developers. That is, starting the comment with '/*' instead of '/**'. Maybe 'damon_call()' is an example.

Quoted 3 lines

This was unclear from the code alone and led to a separate patch attempting to reset nr_dests on failure.

Good point. Adding a link [1] would be nice.

Quoted 4 lines

Make the intended usage
explicit so future readers do not repeat the confusion.

Other than the above trivial things,

I added this patch to my tree, after addressing my above comments. If you don't mind, I will post it tomorrow as a v2, for mm.git inclusion. If you have different opinions to my comments or you prefer to post v2 on your own, please feel free to do so :)

Quoted 34 lines
---
 mm/damon/core.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
mm/damon/core.c +17 -0
index e233eb84a2d5..c884bb31c9b8 100644
@@ -1054,6 +1054,23 @@ static void damos_set_filters_default_reject(struct damos *s)
 		damos_filters_default_reject(&s->ops_filters);
 }
 
+/**
+ * damos_commit_dests() - Copy migration destinations from @src to @dst.
+ * @dst:	Destination structure to update.
+ * @src:	Source structure to copy from.
+ *
+ * If the number of destinations has changed, the old arrays in @dst are freed
+ * and new ones are allocated.  On success, @dst contains a full copy of
+ * @src's arrays and count.
+ *
+ * On allocation failure, @dst is left in a partially torn-down state: its
+ * arrays may be NULL and @nr_dests may not reflect the actual allocation
+ * sizes.  The structure remains safe to deallocate via damon_destroy_scheme(),
+ * but callers must not reuse @dst for further commits — it should be
+ * discarded.
+ *
+ * Return: 0 on success, -ENOMEM on allocation failure.
+ */
 static int damos_commit_dests(struct damos_migrate_dests *dst,
 		struct damos_migrate_dests *src)
 {
Signature · 1 line
2.34.1

And Sashiko comment [2]. TL; DR: Good finding. We aware the issue and working on the fix. Nonetheless, that's out of the scope of this patch.

: Does leaving dst in a torn-down state cause a NULL pointer dereference in : the background kdamond thread?
:
: If damon_commit_ctx() fails during damos_commit_dests(), the sysfs update : aborts. However, the broken scheme remains active in kdamond->damon_ctx : because damon_commit_schemes() does not remove or destroy dst_scheme on : error.
:
: When the background thread calls kdamond_apply_schemes() and invokes : damos_va_migrate_dests_add(), it iterates based on the unmodified nr_dests:
:
: for (i = 0; i < dests->nr_dests; i++)
: weight_total += dests->weight_arr[i];
:
: Since weight_arr is NULL but nr_dests retains its old non-zero value, will : this unconditionally dereference the NULL pointer? It appears the previous : patch attempting to reset nr_dests on failure might be necessary to prevent : this crash.

Nice catch. We actually figured out [3] this issue from the discussion on Josh's previous patch that motivated this one. And I'm working on it. Because the issue is arguably orthogonal to this patch, I think this patch is good to go.

[1] https://lore.kernel.org/[email protected]
[2] review url: https://sashiko.dev/#/patchset/[email protected]
[3] https://lore.kernel.org/[email protected]

Thanks,
SJ

[...]

maintainer
#2 lore raw
Quoted 3 lines · On Thu, 19 Mar 2026 07:31:14 -0700 SeongJae Park <[email protected]> wrote:

Hello Josh,

On Thu, 19 Mar 2026 07:13:32 +0000 Josh Law <[email protected]> wrote:

[...]

Quoted 5 lines

And Sashiko comment [2]. TL; DR: Good finding. We aware the issue and working on the fix. Nonetheless, that's out of the scope of this patch.

: Does leaving dst in a torn-down state cause a NULL pointer dereference in : the background kdamond thread?

[...]

Quoted 2 lines

Nice catch. We actually figured out [3] this issue from the discussion on Josh's previous patch that motivated this one. And I'm working on it.

I just posted the patch:
https://lore.kernel.org/[email protected]

Thanks,
SJ

[...]

author
#3 lore raw
Quoted 15 lines · SeongJae Park
Quoted 3 lines · On Thu, 19 Mar 2026 07:31:14 -0700 SeongJae Park <[email protected]> wrote:

Hello Josh,

On Thu, 19 Mar 2026 07:13:32 +0000 Josh Law <[email protected]> wrote:

[...]

Quoted 5 lines

And Sashiko comment [2]. TL; DR: Good finding. We aware the issue and working on the fix. Nonetheless, that's out of the scope of this patch.

: Does leaving dst in a torn-down state cause a NULL pointer dereference in : the background kdamond thread?

[...]

Quoted 2 lines

Nice catch. We actually figured out [3] this issue from the discussion on Josh's previous patch that motivated this one. And I'm working on it.

Quoted 8 lines

I just posted the patch:
https://lore.kernel.org/[email protected]

Thanks,
SJ

[...]

Hello, this morning i found some big bugs in sysfs.c, ill post the patche(s) soon