From 04d75529dc0f9be78786162ebab7424af4644df2 Mon Sep 17 00:00:00 2001 From: Paul Moses Date: Mon, 23 Feb 2026 15:05:44 +0000 Subject: net/sched: act_gate: snapshot parameters with RCU on replace commit 62413a9c3cb183afb9bb6e94dd68caf4e4145f4c upstream. The gate action can be replaced while the hrtimer callback or dump path is walking the schedule list. Convert the parameters to an RCU-protected snapshot and swap updates under tcf_lock, freeing the previous snapshot via call_rcu(). When REPLACE omits the entry list, preserve the existing schedule so the effective state is unchanged. Fixes: a51c328df310 ("net: qos: introduce a gate control flow action") Cc: stable@vger.kernel.org Signed-off-by: Paul Moses Tested-by: Vladimir Oltean Acked-by: Jamal Hadi Salim Reviewed-by: Victor Nogueira Link: https://patch.msgid.link/20260223150512.2251594-2-p@1g4.org Signed-off-by: Jakub Kicinski Signed-off-by: Greg Kroah-Hartman --- net/sched/act_gate.c | 265 ++++++++++++++++++++++++++++++++++++--------------- 1 file changed, 186 insertions(+), 79 deletions(-) (limited to 'net/sched') diff --git a/net/sched/act_gate.c b/net/sched/act_gate.c index c1f75f272757..d09013ae1892 100644 --- a/net/sched/act_gate.c +++ b/net/sched/act_gate.c @@ -32,9 +32,12 @@ static ktime_t gate_get_time(struct tcf_gate *gact) return KTIME_MAX; } -static void gate_get_start_time(struct tcf_gate *gact, ktime_t *start) +static void tcf_gate_params_free_rcu(struct rcu_head *head); + +static void gate_get_start_time(struct tcf_gate *gact, + const struct tcf_gate_params *param, + ktime_t *start) { - struct tcf_gate_params *param = &gact->param; ktime_t now, base, cycle; u64 n; @@ -69,12 +72,14 @@ static enum hrtimer_restart gate_timer_func(struct hrtimer *timer) { struct tcf_gate *gact = container_of(timer, struct tcf_gate, hitimer); - struct tcf_gate_params *p = &gact->param; struct tcfg_gate_entry *next; + struct tcf_gate_params *p; ktime_t close_time, now; spin_lock(&gact->tcf_lock); + p = rcu_dereference_protected(gact->param, + lockdep_is_held(&gact->tcf_lock)); next = gact->next_entry; /* cycle start, clear pending bit, clear total octets */ @@ -225,6 +230,35 @@ static void release_entry_list(struct list_head *entries) } } +static int tcf_gate_copy_entries(struct tcf_gate_params *dst, + const struct tcf_gate_params *src, + struct netlink_ext_ack *extack) +{ + struct tcfg_gate_entry *entry; + int i = 0; + + list_for_each_entry(entry, &src->entries, list) { + struct tcfg_gate_entry *new; + + new = kzalloc(sizeof(*new), GFP_ATOMIC); + if (!new) { + NL_SET_ERR_MSG(extack, "Not enough memory for entry"); + return -ENOMEM; + } + + new->index = entry->index; + new->gate_state = entry->gate_state; + new->interval = entry->interval; + new->ipv = entry->ipv; + new->maxoctets = entry->maxoctets; + list_add_tail(&new->list, &dst->entries); + i++; + } + + dst->num_entries = i; + return 0; +} + static int parse_gate_list(struct nlattr *list_attr, struct tcf_gate_params *sched, struct netlink_ext_ack *extack) @@ -270,24 +304,44 @@ release_list: return err; } -static void gate_setup_timer(struct tcf_gate *gact, u64 basetime, - enum tk_offsets tko, s32 clockid, - bool do_init) +static bool gate_timer_needs_cancel(u64 basetime, u64 old_basetime, + enum tk_offsets tko, + enum tk_offsets old_tko, + s32 clockid, s32 old_clockid) { - if (!do_init) { - if (basetime == gact->param.tcfg_basetime && - tko == gact->tk_offset && - clockid == gact->param.tcfg_clockid) - return; + return basetime != old_basetime || + clockid != old_clockid || + tko != old_tko; +} - spin_unlock_bh(&gact->tcf_lock); - hrtimer_cancel(&gact->hitimer); - spin_lock_bh(&gact->tcf_lock); +static int gate_clock_resolve(s32 clockid, enum tk_offsets *tko, + struct netlink_ext_ack *extack) +{ + switch (clockid) { + case CLOCK_REALTIME: + *tko = TK_OFFS_REAL; + return 0; + case CLOCK_MONOTONIC: + *tko = TK_OFFS_MAX; + return 0; + case CLOCK_BOOTTIME: + *tko = TK_OFFS_BOOT; + return 0; + case CLOCK_TAI: + *tko = TK_OFFS_TAI; + return 0; + default: + NL_SET_ERR_MSG(extack, "Invalid 'clockid'"); + return -EINVAL; } - gact->param.tcfg_basetime = basetime; - gact->param.tcfg_clockid = clockid; - gact->tk_offset = tko; - hrtimer_setup(&gact->hitimer, gate_timer_func, clockid, HRTIMER_MODE_ABS_SOFT); +} + +static void gate_setup_timer(struct tcf_gate *gact, s32 clockid, + enum tk_offsets tko) +{ + WRITE_ONCE(gact->tk_offset, tko); + hrtimer_setup(&gact->hitimer, gate_timer_func, clockid, + HRTIMER_MODE_ABS_SOFT); } static int tcf_gate_init(struct net *net, struct nlattr *nla, @@ -296,15 +350,22 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, struct netlink_ext_ack *extack) { struct tc_action_net *tn = net_generic(net, act_gate_ops.net_id); - enum tk_offsets tk_offset = TK_OFFS_TAI; + u64 cycletime = 0, basetime = 0, cycletime_ext = 0; + struct tcf_gate_params *p = NULL, *old_p = NULL; + enum tk_offsets old_tk_offset = TK_OFFS_TAI; + const struct tcf_gate_params *cur_p = NULL; bool bind = flags & TCA_ACT_FLAGS_BIND; struct nlattr *tb[TCA_GATE_MAX + 1]; + enum tk_offsets tko = TK_OFFS_TAI; struct tcf_chain *goto_ch = NULL; - u64 cycletime = 0, basetime = 0; - struct tcf_gate_params *p; + s32 timer_clockid = CLOCK_TAI; + bool use_old_entries = false; + s32 old_clockid = CLOCK_TAI; + bool need_cancel = false; s32 clockid = CLOCK_TAI; struct tcf_gate *gact; struct tc_gate *parm; + u64 old_basetime = 0; int ret = 0, err; u32 gflags = 0; s32 prio = -1; @@ -321,26 +382,8 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, if (!tb[TCA_GATE_PARMS]) return -EINVAL; - if (tb[TCA_GATE_CLOCKID]) { + if (tb[TCA_GATE_CLOCKID]) clockid = nla_get_s32(tb[TCA_GATE_CLOCKID]); - switch (clockid) { - case CLOCK_REALTIME: - tk_offset = TK_OFFS_REAL; - break; - case CLOCK_MONOTONIC: - tk_offset = TK_OFFS_MAX; - break; - case CLOCK_BOOTTIME: - tk_offset = TK_OFFS_BOOT; - break; - case CLOCK_TAI: - tk_offset = TK_OFFS_TAI; - break; - default: - NL_SET_ERR_MSG(extack, "Invalid 'clockid'"); - return -EINVAL; - } - } parm = nla_data(tb[TCA_GATE_PARMS]); index = parm->index; @@ -366,6 +409,60 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, return -EEXIST; } + gact = to_gate(*a); + + err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); + if (err < 0) + goto release_idr; + + p = kzalloc(sizeof(*p), GFP_KERNEL); + if (!p) { + err = -ENOMEM; + goto chain_put; + } + INIT_LIST_HEAD(&p->entries); + + use_old_entries = !tb[TCA_GATE_ENTRY_LIST]; + if (!use_old_entries) { + err = parse_gate_list(tb[TCA_GATE_ENTRY_LIST], p, extack); + if (err < 0) + goto err_free; + use_old_entries = !err; + } + + if (ret == ACT_P_CREATED && use_old_entries) { + NL_SET_ERR_MSG(extack, "The entry list is empty"); + err = -EINVAL; + goto err_free; + } + + if (ret != ACT_P_CREATED) { + rcu_read_lock(); + cur_p = rcu_dereference(gact->param); + + old_basetime = cur_p->tcfg_basetime; + old_clockid = cur_p->tcfg_clockid; + old_tk_offset = READ_ONCE(gact->tk_offset); + + basetime = old_basetime; + cycletime_ext = cur_p->tcfg_cycletime_ext; + prio = cur_p->tcfg_priority; + gflags = cur_p->tcfg_flags; + + if (!tb[TCA_GATE_CLOCKID]) + clockid = old_clockid; + + err = 0; + if (use_old_entries) { + err = tcf_gate_copy_entries(p, cur_p, extack); + if (!err && !tb[TCA_GATE_CYCLE_TIME]) + cycletime = cur_p->tcfg_cycletime; + } + rcu_read_unlock(); + if (err) + goto err_free; + } + if (tb[TCA_GATE_PRIORITY]) prio = nla_get_s32(tb[TCA_GATE_PRIORITY]); @@ -375,25 +472,26 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, if (tb[TCA_GATE_FLAGS]) gflags = nla_get_u32(tb[TCA_GATE_FLAGS]); - gact = to_gate(*a); - if (ret == ACT_P_CREATED) - INIT_LIST_HEAD(&gact->param.entries); + if (tb[TCA_GATE_CYCLE_TIME]) + cycletime = nla_get_u64(tb[TCA_GATE_CYCLE_TIME]); - err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack); - if (err < 0) - goto release_idr; + if (tb[TCA_GATE_CYCLE_TIME_EXT]) + cycletime_ext = nla_get_u64(tb[TCA_GATE_CYCLE_TIME_EXT]); - spin_lock_bh(&gact->tcf_lock); - p = &gact->param; + err = gate_clock_resolve(clockid, &tko, extack); + if (err) + goto err_free; + timer_clockid = clockid; - if (tb[TCA_GATE_CYCLE_TIME]) - cycletime = nla_get_u64(tb[TCA_GATE_CYCLE_TIME]); + need_cancel = ret != ACT_P_CREATED && + gate_timer_needs_cancel(basetime, old_basetime, + tko, old_tk_offset, + timer_clockid, old_clockid); - if (tb[TCA_GATE_ENTRY_LIST]) { - err = parse_gate_list(tb[TCA_GATE_ENTRY_LIST], p, extack); - if (err < 0) - goto chain_put; - } + if (need_cancel) + hrtimer_cancel(&gact->hitimer); + + spin_lock_bh(&gact->tcf_lock); if (!cycletime) { struct tcfg_gate_entry *entry; @@ -402,22 +500,20 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, list_for_each_entry(entry, &p->entries, list) cycle = ktime_add_ns(cycle, entry->interval); cycletime = cycle; - if (!cycletime) { - err = -EINVAL; - goto chain_put; - } } p->tcfg_cycletime = cycletime; + p->tcfg_cycletime_ext = cycletime_ext; - if (tb[TCA_GATE_CYCLE_TIME_EXT]) - p->tcfg_cycletime_ext = - nla_get_u64(tb[TCA_GATE_CYCLE_TIME_EXT]); - - gate_setup_timer(gact, basetime, tk_offset, clockid, - ret == ACT_P_CREATED); + if (need_cancel || ret == ACT_P_CREATED) + gate_setup_timer(gact, timer_clockid, tko); p->tcfg_priority = prio; p->tcfg_flags = gflags; - gate_get_start_time(gact, &start); + p->tcfg_basetime = basetime; + p->tcfg_clockid = timer_clockid; + gate_get_start_time(gact, p, &start); + + old_p = rcu_replace_pointer(gact->param, p, + lockdep_is_held(&gact->tcf_lock)); gact->current_close_time = start; gact->current_gate_status = GATE_ACT_GATE_OPEN | GATE_ACT_PENDING; @@ -434,11 +530,15 @@ static int tcf_gate_init(struct net *net, struct nlattr *nla, if (goto_ch) tcf_chain_put_by_act(goto_ch); + if (old_p) + call_rcu(&old_p->rcu, tcf_gate_params_free_rcu); + return ret; +err_free: + release_entry_list(&p->entries); + kfree(p); chain_put: - spin_unlock_bh(&gact->tcf_lock); - if (goto_ch) tcf_chain_put_by_act(goto_ch); release_idr: @@ -446,21 +546,29 @@ release_idr: * without taking tcf_lock. */ if (ret == ACT_P_CREATED) - gate_setup_timer(gact, gact->param.tcfg_basetime, - gact->tk_offset, gact->param.tcfg_clockid, - true); + gate_setup_timer(gact, timer_clockid, tko); + tcf_idr_release(*a, bind); return err; } +static void tcf_gate_params_free_rcu(struct rcu_head *head) +{ + struct tcf_gate_params *p = container_of(head, struct tcf_gate_params, rcu); + + release_entry_list(&p->entries); + kfree(p); +} + static void tcf_gate_cleanup(struct tc_action *a) { struct tcf_gate *gact = to_gate(a); struct tcf_gate_params *p; - p = &gact->param; hrtimer_cancel(&gact->hitimer); - release_entry_list(&p->entries); + p = rcu_dereference_protected(gact->param, 1); + if (p) + call_rcu(&p->rcu, tcf_gate_params_free_rcu); } static int dumping_entry(struct sk_buff *skb, @@ -509,10 +617,9 @@ static int tcf_gate_dump(struct sk_buff *skb, struct tc_action *a, struct nlattr *entry_list; struct tcf_t t; - spin_lock_bh(&gact->tcf_lock); - opt.action = gact->tcf_action; - - p = &gact->param; + rcu_read_lock(); + opt.action = READ_ONCE(gact->tcf_action); + p = rcu_dereference(gact->param); if (nla_put(skb, TCA_GATE_PARMS, sizeof(opt), &opt)) goto nla_put_failure; @@ -552,12 +659,12 @@ static int tcf_gate_dump(struct sk_buff *skb, struct tc_action *a, tcf_tm_dump(&t, &gact->tcf_tm); if (nla_put_64bit(skb, TCA_GATE_TM, sizeof(t), &t, TCA_GATE_PAD)) goto nla_put_failure; - spin_unlock_bh(&gact->tcf_lock); + rcu_read_unlock(); return skb->len; nla_put_failure: - spin_unlock_bh(&gact->tcf_lock); + rcu_read_unlock(); nlmsg_trim(skb, b); return -1; } -- cgit v1.2.3 From 380ad8b7c65ea7aa10ef2258297079ed5ac1f5b6 Mon Sep 17 00:00:00 2001 From: Victor Nogueira Date: Wed, 25 Feb 2026 10:43:48 -0300 Subject: net/sched: Only allow act_ct to bind to clsact/ingress qdiscs and shared blocks commit 11cb63b0d1a0685e0831ae3c77223e002ef18189 upstream. As Paolo said earlier [1]: "Since the blamed commit below, classify can return TC_ACT_CONSUMED while the current skb being held by the defragmentation engine. As reported by GangMin Kim, if such packet is that may cause a UaF when the defrag engine later on tries to tuch again such packet." act_ct was never meant to be used in the egress path, however some users are attaching it to egress today [2]. Attempting to reach a middle ground, we noticed that, while most qdiscs are not handling TC_ACT_CONSUMED, clsact/ingress qdiscs are. With that in mind, we address the issue by only allowing act_ct to bind to clsact/ingress qdiscs and shared blocks. That way it's still possible to attach act_ct to egress (albeit only with clsact). [1] https://lore.kernel.org/netdev/674b8cbfc385c6f37fb29a1de08d8fe5c2b0fbee.1771321118.git.pabeni@redhat.com/ [2] https://lore.kernel.org/netdev/cc6bfb4a-4a2b-42d8-b9ce-7ef6644fb22b@ovn.org/ Reported-by: GangMin Kim Fixes: 3f14b377d01d ("net/sched: act_ct: fix skb leak and crash on ooo frags") CC: stable@vger.kernel.org Signed-off-by: Victor Nogueira Acked-by: Jamal Hadi Salim Link: https://patch.msgid.link/20260225134349.1287037-1-victor@mojatatu.com Signed-off-by: Jakub Kicinski Signed-off-by: Greg Kroah-Hartman --- include/net/act_api.h | 1 + net/sched/act_ct.c | 6 ++++++ net/sched/cls_api.c | 7 +++++++ 3 files changed, 14 insertions(+) (limited to 'net/sched') diff --git a/include/net/act_api.h b/include/net/act_api.h index 91a24b5e0b93..2ba40eb45aad 100644 --- a/include/net/act_api.h +++ b/include/net/act_api.h @@ -70,6 +70,7 @@ struct tc_action { #define TCA_ACT_FLAGS_REPLACE (1U << (TCA_ACT_FLAGS_USER_BITS + 2)) #define TCA_ACT_FLAGS_NO_RTNL (1U << (TCA_ACT_FLAGS_USER_BITS + 3)) #define TCA_ACT_FLAGS_AT_INGRESS (1U << (TCA_ACT_FLAGS_USER_BITS + 4)) +#define TCA_ACT_FLAGS_AT_INGRESS_OR_CLSACT (1U << (TCA_ACT_FLAGS_USER_BITS + 5)) /* Update lastuse only if needed, to avoid dirtying a cache line. * We use a temp variable to avoid fetching jiffies twice. diff --git a/net/sched/act_ct.c b/net/sched/act_ct.c index 6749a4a9a9cd..b3c160ad590d 100644 --- a/net/sched/act_ct.c +++ b/net/sched/act_ct.c @@ -1358,6 +1358,12 @@ static int tcf_ct_init(struct net *net, struct nlattr *nla, return -EINVAL; } + if (bind && !(flags & TCA_ACT_FLAGS_AT_INGRESS_OR_CLSACT)) { + NL_SET_ERR_MSG_MOD(extack, + "Attaching ct to a non ingress/clsact qdisc is unsupported"); + return -EOPNOTSUPP; + } + err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack); if (err < 0) return err; diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c index ecec0a1e1c1a..bac9cd71ff8e 100644 --- a/net/sched/cls_api.c +++ b/net/sched/cls_api.c @@ -2228,6 +2228,11 @@ static bool is_qdisc_ingress(__u32 classid) return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS)); } +static bool is_ingress_or_clsact(struct tcf_block *block, struct Qdisc *q) +{ + return tcf_block_shared(block) || (q && !!(q->flags & TCQ_F_INGRESS)); +} + static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n, struct netlink_ext_ack *extack) { @@ -2420,6 +2425,8 @@ replay: flags |= TCA_ACT_FLAGS_NO_RTNL; if (is_qdisc_ingress(parent)) flags |= TCA_ACT_FLAGS_AT_INGRESS; + if (is_ingress_or_clsact(block, q)) + flags |= TCA_ACT_FLAGS_AT_INGRESS_OR_CLSACT; err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh, flags, extack); if (err == 0) { -- cgit v1.2.3