Skip to contentSkip to conversation

mm/damon/core: validate ranges in damon_set_regions()

RFC PATCH In review SJ Park 2 participants 3 messages Jun 28

Subsystems (3)
lore
Share & tools
lore

SJ Park

author maintainer
#0 lore raw

RFC PATCH 1/1 mm/damon/core: validate ranges in damon_set_regions()

DAMON core logic assumes zero length regions don't exist. However, a few DAMON API callers including DAMON_SYSFS, DAMON_RECLAIM and DAMON_LRU_SORT allow users to set empty monitoring target regions. This could result in WARN_ONCE() on CONFIG_DAMON_DEBUG_SANITY enabled kernel, and divide-by-zero from damon_merge_two_regions().

For example, the WANR_ONCE() can be triggered like below.

    # grep DAMON_DEBUG_SANITY /boot/config-$(uname -r)
    # CONFIG_DAMON_DEBUG_SANITY=y
    # damo start
    # cd /sys/kernel/mm/damon/admin/kdamonds/0
    # echo 0 > contexts/0/targets/0/regions/0/start
    # echo 0 > contexts/0/targets/0/regions/0/end
    # echo commit > state
    # dmesg
    [....]
    [   73.705780] ------------[ cut here ]------------
    [   73.707552] start 0 >= end 0
    [   73.708452] WARNING: mm/damon/core.c:359 at damon_new_region+0x6e/0x80, CPU#1: kdamond.0/758
    [...]

All DAMON API callers eventually use damon_set_regions() to setup the regions. Add the validation logic in the function.

Fixes 43b0536cb471 ("mm/damon: introduce DAMON-based Reclamation (DAMON_RECLAIM)")
Cc <[email protected]> # 5.16.x

---
Changes from RFC v1
- RFC v1: https://lore.kernel.org/[email protected]
- Fixup the commit message for how the fix is made.
- Do the validation with min_region_sz-aligned addresses.

FYI, this fix cannot be applied as is to the commit that introduced this class of bugs, because damon_set_regions() was introduced after the bug. I considered making three fixes for each caller to make the backporting on the old kernels easy. However, the first LTS kernel having the bug is 6.1.y, which has damon_set_regions() and all the callers are using it. So porting this to necessary stable kernels should be easy enough.

 mm/damon/core.c | 6 ++++++
 1 file changed, 6 insertions(+)
mm/damon/core.c +6 -0
index d99f7a297fdde..df0cc699494fe 100644
@@ -356,6 +356,12 @@ int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
 	unsigned int i;
 	int err;
 
+	for (i = 0; i < nr_ranges; i++) {
+		if (ALIGN_DOWN(ranges[i].start, min_region_sz) >=
+				ALIGN_DOWN(ranges[i].end, min_region_sz))
+			return -EINVAL;
+	}
+
 	/* Remove regions which are not in the new ranges */
 	damon_for_each_region_safe(r, next, t) {
 		for (i = 0; i < nr_ranges; i++) {

base-commit: 7c001190b88a32f80c93a6ac302af59a9756309c

Signature · 1 line
2.47.3
#1 lore raw
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

New issues:
- [Critical] The new validation logic incorrectly uses ALIGN_DOWN for the region's end address, falsely rejecting valid sub-block monitoring regions and breaking the user API.
- [High] Integer overflow in ALIGN() bypasses the newly added validation check, failing to prevent the creation of zero-length regions.

Pre-existing issues:
- [High] sysfs interface allows addr_unit to be set such that min_region_sz is not a power of two, breaking alignment macros and corrupting boundaries. --

commit a7d14189013c68fcb3a5a9644321f7f319e6fe5f
Author: SJ Park <[email protected]>

mm/damon/core: validate ranges in damon_set_regions()

DAMON core logic assumes zero length regions don't exist, but API callers allow empty monitoring target regions. This patch adds validation in damon_set_regions() to ensure start is strictly less than end after alignment.

mm/damon/core.c int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges, · line 363

Quoted 11 lines
mm/damon/core.c +3 -0
index 7e4b9affc5b06..cbdf29d119c93 100644
@@ -358,6 +358,12 @@ int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
 	unsigned int i;
 	int err;
 
+	for (i = 0; i < nr_ranges; i++) {
+		if (ALIGN_DOWN(ranges[i].start, min_region_sz) >=
+				ALIGN_DOWN(ranges[i].end, min_region_sz))

[Severity: Critical]
Does this ALIGN_DOWN() on the end address incorrectly reject valid regions?

If a user configures a target region with start and end within the same alignment boundary (for example, start=4096 and end=8191 with a min_region_sz=4096), the check evaluates to 4096 >= 4096 and returns -EINVAL. The previous behavior successfully expanded this to [4096, 8192) using ALIGN(end). This seems to reject previously valid inputs.

[Severity: High]
Can an integer overflow here bypass the new validation check?

If a user specifies a region with an end address near ULONG_MAX, ALIGN_DOWN(ULONG_MAX, min_region_sz) will evaluate to a large positive value, which passes the start >= end check.

However, the region creation logic later uses ALIGN() on the end address:

mm/damon/core.c:damon_set_regions() {
    ...
		for (i = 0; i < nr_ranges; i++) {
			r = damon_new_region(
					ALIGN_DOWN(ranges[i].start,
						min_region_sz),
					ALIGN(ranges[i].end, min_region_sz));
    ...
}

ALIGN(ULONG_MAX, min_region_sz) overflows to 0, which means damon_new_region() is still called with an end address of 0, bypassing this validation and creating the exact zero-length region this patch intends to stop.

[Severity: High]
This is a pre-existing issue, but can min_region_sz be a non-power-of-two, corrupting boundaries when passed to ALIGN_DOWN here?

The ALIGN and ALIGN_DOWN macros rely on bitwise masking, which requires the alignment value to be a power of two.

Looking at how min_region_sz is calculated in damon_sysfs_apply_inputs():

mm/damon/sysfs.c:damon_sysfs_apply_inputs() {
    ...
		ctx->min_region_sz = max(
				DAMON_MIN_REGION_SZ / sys_ctx->addr_unit, 1);
    ...
}

If a user writes a value like 3 to addr_unit via sysfs, min_region_sz becomes 1365 (4096 / 3). When 1365 is passed to ALIGN_DOWN here, the macro produces garbage addresses.

Quoted 5 lines
mm/damon/core.c +3 -0
+			return -EINVAL;
+	}
+
 	/* Remove regions which are not in the new ranges */
 	damon_for_each_region_safe(r, next, t) {
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
author maintainer
#2 lore raw
Quoted 34 lines · On Sun, 28 Jun 2026 01:06:30 +0000 [email protected] wrote:

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

New issues:
- [Critical] The new validation logic incorrectly uses ALIGN_DOWN for the region's end address, falsely rejecting valid sub-block monitoring regions and breaking the user API.
- [High] Integer overflow in ALIGN() bypasses the newly added validation check, failing to prevent the creation of zero-length regions.

Pre-existing issues:
- [High] sysfs interface allows addr_unit to be set such that min_region_sz is not a power of two, breaking alignment macros and corrupting boundaries. --

commit a7d14189013c68fcb3a5a9644321f7f319e6fe5f
Author: SJ Park <[email protected]>

mm/damon/core: validate ranges in damon_set_regions()

DAMON core logic assumes zero length regions don't exist, but API callers allow empty monitoring target regions. This patch adds validation in damon_set_regions() to ensure start is strictly less than end after alignment.

mm/damon/core.c int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges, · line 363

Quoted 11 lines
mm/damon/core.c +3 -0
index 7e4b9affc5b06..cbdf29d119c93 100644
@@ -358,6 +358,12 @@ int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges,
 	unsigned int i;
 	int err;
 
+	for (i = 0; i < nr_ranges; i++) {
+		if (ALIGN_DOWN(ranges[i].start, min_region_sz) >=
+				ALIGN_DOWN(ranges[i].end, min_region_sz))

[Severity: Critical]
Does this ALIGN_DOWN() on the end address incorrectly reject valid regions?

Quoted 5 lines

If a user configures a target region with start and end within the same alignment boundary (for example, start=4096 and end=8191 with a min_region_sz=4096), the check evaluates to 4096 >= 4096 and returns -EINVAL. The previous behavior successfully expanded this to [4096, 8192) using ALIGN(end). This seems to reject previously valid inputs.

Correct. I mistakenly used ALIGN_DOWN() instead of ALIGN() for the end address. I will fix this in the next revision.

Quoted 20 lines

[Severity: High]
Can an integer overflow here bypass the new validation check?

If a user specifies a region with an end address near ULONG_MAX, ALIGN_DOWN(ULONG_MAX, min_region_sz) will evaluate to a large positive value, which passes the start >= end check.

However, the region creation logic later uses ALIGN() on the end address:

mm/damon/core.c:damon_set_regions() {
    ...
		for (i = 0; i < nr_ranges; i++) {
			r = damon_new_region(
					ALIGN_DOWN(ranges[i].start,
						min_region_sz),
					ALIGN(ranges[i].end, min_region_sz));
    ...
}
Quoted 3 lines

ALIGN(ULONG_MAX, min_region_sz) overflows to 0, which means damon_new_region() is still called with an end address of 0, bypassing this validation and creating the exact zero-length region this patch intends to stop.

Use of ALIGN_DOWN() for the end address was a mistake. I will use ALIGN() for the end address in the next spin.

Quoted 17 lines

[Severity: High]
This is a pre-existing issue, but can min_region_sz be a non-power-of-two, corrupting boundaries when passed to ALIGN_DOWN here?

The ALIGN and ALIGN_DOWN macros rely on bitwise masking, which requires the alignment value to be a power of two.

Looking at how min_region_sz is calculated in damon_sysfs_apply_inputs():

mm/damon/sysfs.c:damon_sysfs_apply_inputs() {
    ...
		ctx->min_region_sz = max(
				DAMON_MIN_REGION_SZ / sys_ctx->addr_unit, 1);
    ...
}
Quoted 3 lines

If a user writes a value like 3 to addr_unit via sysfs, min_region_sz becomes 1365 (4096 / 3). When 1365 is passed to ALIGN_DOWN here, the macro produces garbage addresses.

Wrong. We have the power of 2 check in damon_commit_ctx() and damon_start(). So this issue cannot happen.

Thanks,
SJ

[...]