M2 data path: BSD netisr workqueue + lock reentrancy fixes + diagnostics

- hooks: packets are queued to pfk_netisr ordered wq (process context) and
  re-injected with an skb->mark guard (PFK_MARK_PASSED); softirq-safe,
  documented divergence vs OpenBSD in-place ip_input continuation
- pfk_rwlock: per-task writer recursion (NET/PF lock nesting like OpenBSD)
- pfk_net: holder-task+depth reentrancy (pid-based hack removed)
- ifnet events deferred to ordered workqueue (rtnl AB-BA fix)
- idr iteration start index fix (kifs were never attached -> kif==NULL drop)
- pfk_lock_trace module param for lock diagnostics
- tests/vm: 01-state 02-block 03-tables 04-anchors + lib; vm-run-tests.sh
- docs/en/install.md: apply/build guide (core patch set currently EMPTY)

module/compat/pfk_kern.c

@@ -100,27 +100,61 @@ taskq_create(const char *name, unsigned int nthreads, int ipl,
 /* ---------------- NET_LOCK ---------------- */
 static DECLARE_RWSEM(pfk_net_sem);
 static atomic_t pfk_net_readers = ATOMIC_INIT(0);
-static atomic_t pfk_net_owner = ATOMIC_INIT(0);	/* task pid holding write */
+
+/*
+ * PFLINUX: OpenBSD's NET_LOCK nesting is friendly for the holder (e.g.
+ * pfioctl takes NET_LOCK and pf_status_read() does too).  Linux rw_semaphores
+ * are not recursive, so track the holder task + depth: nested lock/unlock by
+ * the same task balance without touching the rwsem.  (Tracking was by task
+ * pointer on purpose: pids can be recycled.)
+ */
+static DEFINE_RAW_SPINLOCK(pfk_net_owner_lock);
+static struct task_struct *pfk_net_owner;
+static int pfk_net_depth;
 
 void pfk_net_lock_init(void) { }
+
 void
 pfk_net_lock(void)
 {
-	/* Re-entrant for the holder: OpenBSD's NET_LOCK is friendly with
-	 * nested calls in the same path (e.g. our netdev notifier firing
-	 * while pfioctl holds the lock). */
-	if (atomic_read(&pfk_net_owner) == current->pid)
+	unsigned long flags;
+	bool nested;
+
+	raw_spin_lock_irqsave(&pfk_net_owner_lock, flags);
+	nested = (pfk_net_owner == current);
+	if (nested)
+		pfk_net_depth++;
+	raw_spin_unlock_irqrestore(&pfk_net_owner_lock, flags);
+	if (nested)
 		return;
 	down_write(&pfk_net_sem);
-	atomic_set(&pfk_net_owner, current->pid);
+	raw_spin_lock_irqsave(&pfk_net_owner_lock, flags);
+	pfk_net_owner = current;
+	pfk_net_depth = 1;
+	raw_spin_unlock_irqrestore(&pfk_net_owner_lock, flags);
 }
+
 void
 pfk_net_unlock(void)
 {
-	if (atomic_read(&pfk_net_owner) != current->pid)
-		return;		/* nested no-op for non-holders */
-	atomic_set(&pfk_net_owner, 0);
-	up_write(&pfk_net_sem);
+	unsigned long flags;
+	int depth;
+
+	raw_spin_lock_irqsave(&pfk_net_owner_lock, flags);
+	if (pfk_net_owner != current) {
+		/* not the holder: unbalanced unlock — ignore loudly */
+		raw_spin_unlock_irqrestore(&pfk_net_owner_lock, flags);
+		WARN_ON_ONCE(1);
+		return;
+	}
+	depth = --pfk_net_depth;
+	raw_spin_unlock_irqrestore(&pfk_net_owner_lock, flags);
+	if (depth == 0) {
+		raw_spin_lock_irqsave(&pfk_net_owner_lock, flags);
+		pfk_net_owner = NULL;
+		raw_spin_unlock_irqrestore(&pfk_net_owner_lock, flags);
+		up_write(&pfk_net_sem);
+	}
 }
 void pfk_net_read_enter(void)  { down_read(&pfk_net_sem);
 				 atomic_inc(&pfk_net_readers); }