qosify, branch master

qosify: expose default traffic classes in statistics

2026-06-22T11:49:55Z

Report the TCP and UDP default class counters, the class map entries past the configured classes, as separate tcp_default and udp_default entries in the classes statistics table, so default and best-effort traffic is visible alongside the named classes. Signed-off-by: Felix Fietkau

qosify: count all traffic in per-DSCP statistics

2026-06-22T11:49:49Z

The per-DSCP accounting ran at the end of classify(), but traffic mapped to a class without the PRESENT flag, notably the best-effort TCP/UDP port defaults, returned early from dscp_lookup_class() and never reached it. The bulk of traffic was therefore left uncounted. Account every IP packet on exit, keyed on the DSCP value actually present in the packet as it leaves the classifier, and count the default port classes in their class map entries. Marking behaviour is unchanged: the default and unclassified paths still pass the packet without rewriting its DSCP field. Shared counters are incremented atomically to avoid lost updates across CPUs. Signed-off-by: Felix Fietkau

qosify-bpf: remove unused variable in classify()

2026-06-22T11:44:19Z

Signed-off-by: Felix Fietkau

qosify: account for GSO segment count in packet statistics

2026-06-19T10:24:28Z

A GSO/GRO super-packet is a single skb but represents skb->gso_segs packets on the wire. Count it as such in the traffic class, per-DSCP and per-DNS-pattern packet counters instead of incrementing by one per skb, matching the segment accounting already done by the flow bulk/prio checks. Byte counters keep using skb->len, which is already the aggregate length. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

qosify: report eBPF map entry count and last reload time

2026-06-19T10:24:28Z

Expose the combined number of active IPv4 and IPv6 map entries and the wall-clock time of the last map reload as top-level fields in the get_stats reply. last_reload_time is initialised at startup so it reflects the initial load until the first explicit reload. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

qosify: track per-DNS-pattern traffic statistics

2026-06-19T10:24:28Z

Assign a stable 32-bit pattern id to every DNS map entry and tag the IP map entries it resolves with that id. Account packets and bytes per pattern id in a dedicated per-CPU stats map, so the cost scales with the number of DNS patterns rather than the number of resolved IP addresses. User space owns the lifecycle of the stats entries: it creates the entry when a pattern id is assigned and removes it when the pattern is freed, while the eBPF classifier only updates entries that already exist. The map is the source of truth and persists across IP entry timeouts; user space reports the per-pattern hit count and the summed per-CPU packet/byte totals via the "dns" table in get_stats, and a reset explicitly zeroes the counters. The cname cache carries the pattern id alongside the cached dscp so that attributions inherited through a cname chain are preserved on cache hits; since pattern ids are never reused, a stale cached id can at most lose an attribution, never misattribute it. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

dns: track cname cache hit/miss statistics

2026-06-18T08:49:38Z

Count lookups served from the cname cache, lookups that miss it and the current number of cached entries, and report them via the "dns_cache" table in the get_stats reply. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

qosify: add per-DSCP traffic statistics

2026-06-18T08:49:38Z

Add a dscp_stats BPF array map indexed by the final DSCP value and account every classified packet there. Expose the per-codepoint packet and byte counters via the "dscp" table in the get_stats reply. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

qosify: group traffic class stats under a "classes" table

2026-06-18T08:49:35Z

Wrap the per-class statistics in a "classes" table in the get_stats reply so that the upcoming dns, dscp and dns_cache statistic groups and the top-level scalar fields can be added as unambiguous siblings. Note: this changes the get_stats reply layout. The per-class tables, previously emitted at the top level, now live under "classes". Signed-off-by: John Crispin Signed-off-by: Felix Fietkau

qosify: account for bytes in traffic class statistics

2026-06-18T08:49:35Z

Extend the per-class statistics with a byte counter next to the existing packet counter. Signed-off-by: John Crispin Signed-off-by: Felix Fietkau