[5.10] [Bugfix] mm/pagecache_limit: reject reclaim_interval_ms values above UINT_MAX · Pull Request !16454 · anolis/cloud-kernel

bugzilla: https://bugzilla.openanolis.cn/show_bug.cgi?id=54819

The memory.pagecache_limit.reclaim_interval_ms cgroup v1 interface parses
its input with kstrtou64() but passes the value directly to
msecs_to_jiffies(), whose parameter type is unsigned int. Values above
UINT_MAX lose their high bits silently: writing 4294967296 (2^32)
succeeds and stores/reads back 0, which disables the reclaim worker's
interval-based re-arming, and writing 2^32+1000 is treated as 1000 ms.
The saturation guard inside __msecs_to_jiffies() only sees the
already-narrowed value and cannot catch this.

Fix: reject inputs above UINT_MAX with -EINVAL in the write callback, so
over-range configurations fail loudly instead of silently changing the
configured interval. All values within [0, UINT_MAX] keep their exact
previous behavior; no struct or interface change, so there is no KABI
impact.

Note: values in [2^31, 2^32) still hit the pre-existing
msecs_to_jiffies() "infinite timeout" saturation and read back as a very
large interval; that is the generic msecs_to_jiffies() semantics and is
unchanged by this fix.

Verified: writing 4294967296 now returns -EINVAL; writing 1000 reads
back 1000. Incremental build of mm/ passes on aarch64.

类型	指派人员	状态
审查	xunlei Dust.Li baolin.wang Guixin Liu XiangZao	已完成 (0/0人)
测试	小龙已测试通过	已完成 (1/1人)

GVPanolis / cloud-kernel

搜索帮助

GVP anolis / cloud-kernel