Kernel: T9010: update existing patches to remove "hunk off" notices - #1226

scripts/package-build/linux-kernel/patches/kernel/0001-linkstate-ip-device-attribute.patch (2)

145-151: 🩺 Stability & Availability | 🔴 Critical | ⚡ Quick win

Guard __in6_dev_get(dev) before dereference to avoid crash

At Line 148, __in6_dev_get(dev) is dereferenced unconditionally. If it returns NULL, this is a kernel NULL-pointer dereference in route scoring path.

Proposed fix

 static inline int rt6_link_filter(const struct fib6_nh *nh)
 {
 	const struct net_device *dev = nh->fib_nh_dev;
-	int linkf = __in6_dev_get(dev)->cnf.link_filter;
+	const struct inet6_dev *idev = __in6_dev_get(dev);
+	int linkf;
+
+	if (!idev)
+		return 0;
+
+	linkf = idev->cnf.link_filter;
 	return (linkf && !netif_running(dev))
 		|| (linkf > 1 && !netif_carrier_ok(dev));
 }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@scripts/package-build/linux-kernel/patches/kernel/0001-linkstate-ip-device-attribute.patch`
around lines 145 - 151, The rt6_link_filter helper dereferences
__in6_dev_get(dev) without checking for NULL, which can crash in the route
scoring path. Update rt6_link_filter to first store the __in6_dev_get(dev)
result in a local pointer and guard the cnf.link_filter access with a NULL check
before using it; keep the existing netif_running and netif_carrier_ok logic
unchanged.

---

37-46: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Enforce documented link_filter range (0..2) in sysctl handlers

Lines 37-46 document only 0/1/2, but Line 107 and Lines 127-133 register unrestricted integer writes. Out-of-range values are currently accepted, creating a doc/UAPI contract break and undefined admin behavior.

Proposed fix (IPv6 table pattern)

 	{
 		.procname	= "link_filter",
 		.data		= &ipv6_devconf.link_filter,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
-		.proc_handler	= proc_dointvec,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= (void *)SYSCTL_ZERO,
+		.extra2		= (void *)SYSCTL_TWO,
 	},

Apply equivalent min/max enforcement for the IPv4 link_filter sysctl entry as well.

Also applies to: 107-107, 127-133

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@scripts/package-build/linux-kernel/patches/kernel/0001-linkstate-ip-device-attribute.patch`
around lines 37 - 46, The `link_filter` sysctl accepts unrestricted integers in
the IPv4 and IPv6 handlers, which conflicts with the documented 0..2 range.
Update the sysctl registration/handler setup around the `link_filter` entries to
enforce minimum 0 and maximum 2 for both tables, following the IPv6 table
pattern and applying the same bounds to the IPv4 `link_filter` entry as well.

scripts/package-build/linux-kernel/patches/kernel/0002-build-linux-perf-package.patch (2)

69-77: 🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Guard binary-perf target creation on package presence.

binary-targets includes binary-perf unconditionally (Line 69), but perf-package is conditional (Line 76). When no linux-perf-* package is emitted, this target can execute with an empty package binding.

Proposed fix

-binary-targets := $(addprefix binary-, image image-dbg headers libc-dev perf)
+binary-targets := $(addprefix binary-, image image-dbg headers libc-dev)
@@
 perf-package = $(filter linux-perf-%, $(all-packages))
+ifneq ($(strip $(perf-package)),)
+binary-targets += binary-perf
+endif

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@scripts/package-build/linux-kernel/patches/kernel/0002-build-linux-perf-package.patch`
around lines 69 - 77, Guard the unconditional binary-perf target in the
packaging make logic: binary-targets currently always includes binary-perf while
perf-package in the same block may be empty, so adjust the target list or the
perf package selection in the kernel packaging rules to only create binary-perf
when a linux-perf-* package exists. Use the existing binary-targets and
perf-package symbols in the build rules to keep the perf target conditional
alongside the other package-specific targets.

---

91-95: 🎯 Functional Correctness | 🟠 Major

Use debhelper substvars for linux-perf-${KERNELRELEASE} dependencies. scripts/package-build/linux-kernel/patches/kernel/0002-build-linux-perf-package.patch:91-95 hardcodes runtime libs only; add ${shlibs:Depends} and ${misc:Depends} so perf tracks its actual ELF and package metadata dependencies.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@scripts/package-build/linux-kernel/patches/kernel/0002-build-linux-perf-package.patch`
around lines 91 - 95, The linux-perf package stanza is missing
debhelper-generated dependency substitution, so update the package metadata in
the linux-perf patch to use substvars instead of only hardcoded runtime
libraries. In the package definition for linux-perf-${KERNELRELEASE}, add
${shlibs:Depends} and ${misc:Depends} to the Depends field alongside the
existing libs so the package tracks ELF and metadata dependencies correctly.