The fallback skb pool fronted by find_skb() is netconsole's only client: every other netpoll goes through __netpoll_setup() / netpoll_send_skb() without ever touching np->skb_pool. Today __netpoll_setup() and __netpoll_cleanup() create and destroy the pool for everyone, paying ~48 KB of pre-allocated skbs per netpoll instance that only netconsole uses, what a waste! Move the responsibility to netconsole. __netpoll_setup() did this under the RTNL, but netconsole enables targets from enabled_store() / alloc_param_target() without it, while the teardown path flushes the pool (cancel_work_sync() + skb_queue_purge()) under the RTNL from netconsole_process_cleanups_core(). Initialising the queue head and the refill work on every enable would therefore race that flush. They only need initialising once: after a flush the queue head is left valid and empty and cancel_work_sync() leaves the work re-armable. Set them up in alloc_and_init(), while the target is not yet reachable, and let the enable paths only refill the pool via refill_skbs(), which serialises with the flush through skb_pool.lock. See discussions in [1] Link: https://lore.kernel.org/all/alDMvD5S7TZnoD_V@gmail.com/ [1] Signed-off-by: Breno Leitao --- drivers/net/netconsole.c | 53 ++++++++++++++++++++++++++++++++++++++++++++++-- net/core/netpoll.c | 12 +---------- 2 files changed, 52 insertions(+), 13 deletions(-) diff --git a/drivers/net/netconsole.c b/drivers/net/netconsole.c index a939daa07cf9d..1f75c4bbea8b6 100644 --- a/drivers/net/netconsole.c +++ b/drivers/net/netconsole.c @@ -292,11 +292,33 @@ static void netcons_release_dev(struct netconsole_target *nt) memset(&nt->np.dev_name, 0, IFNAMSIZ); } +/* Seed the per-target skb pool that find_skb() falls back to. The queue + * head and refill work are set up once in alloc_and_init(); this only + * (re)fills the pool. Pair with netconsole_skb_pool_flush(). + */ +static void netconsole_skb_pool_init(struct netconsole_target *nt) +{ + refill_skbs(&nt->np); +} + +static void netconsole_skb_pool_flush(struct netconsole_target *nt) +{ + skb_pool_flush(&nt->np); +} + /* Attempts to resume logging to a deactivated target. */ static void resume_target(struct netconsole_target *nt) { + /* Initialise the skb pool before netpoll_setup() makes nt->np.dev + * visible to target_list walkers (e.g. netconsole_netdev_event), + * which otherwise may move the target to the cleanup list and + * call netconsole_skb_pool_flush() on uninitialised state. + */ + netconsole_skb_pool_init(nt); + if (netpoll_setup(&nt->np)) { /* netpoll fails setup once, do not try again. */ + netconsole_skb_pool_flush(nt); nt->state = STATE_DISABLED; return; } @@ -358,6 +380,7 @@ static void process_resume_target(struct work_struct *work) rtnl_lock(); if (nt->state == STATE_ENABLED && nt->np.dev && nt->np.dev->reg_state != NETREG_REGISTERED) { + netconsole_skb_pool_flush(nt); netcons_release_dev(nt); nt->state = STATE_DISABLED; } @@ -398,6 +421,9 @@ static struct netconsole_target *alloc_and_init(void) eth_broadcast_addr(nt->np.remote_mac); nt->state = STATE_DISABLED; INIT_WORK(&nt->resume_wq, process_resume_target); + /* Set up the skb pool primitives once; enabling only refills it. */ + skb_queue_head_init(&nt->np.skb_pool); + INIT_WORK(&nt->np.refill_wq, refill_skbs_work_handler); return nt; } @@ -417,6 +443,7 @@ static void netconsole_process_cleanups_core(void) list_for_each_entry_safe(nt, tmp, &target_cleanup_list, list) { /* all entries in the cleanup_list needs to be disabled */ WARN_ON_ONCE(nt->state == STATE_ENABLED); + netconsole_skb_pool_flush(nt); netcons_release_dev(nt); /* moved the cleaned target to target_list. Need to hold both * locks @@ -758,9 +785,19 @@ static ssize_t enabled_store(struct config_item *item, */ netconsole_print_banner(&nt->np); + /* Initialise the skb pool before netpoll_setup() so the pool + * is valid as soon as nt->np.dev becomes visible to + * target_list walkers (netconsole_netdev_event), which would + * otherwise call netconsole_skb_pool_flush() on uninitialised + * state. + */ + netconsole_skb_pool_init(nt); + ret = netpoll_setup(&nt->np); - if (ret) + if (ret) { + netconsole_skb_pool_flush(nt); goto out_unlock; + } nt->state = STATE_ENABLED; pr_info("network logging started\n"); @@ -1514,8 +1551,10 @@ static void drop_netconsole_target(struct config_group *group, * netpoll_cleanup() is idempotent (it skips when np->dev is NULL), so * it is safe even if the cleanup worker already tore the netpoll down. */ - if (needs_cleanup) + if (needs_cleanup) { + netconsole_skb_pool_flush(nt); netpoll_cleanup(&nt->np); + } config_item_put(&nt->group.cg_item); } @@ -2330,10 +2369,18 @@ static struct netconsole_target *alloc_param_target(char *target_config, if (err) goto fail; + /* Initialise the skb pool before netpoll_setup() so the pool is + * valid as soon as nt->np.dev becomes visible. The target is not + * yet on target_list, so a netdev event cannot reach it here, but + * mirror the configfs path for symmetry. + */ + netconsole_skb_pool_init(nt); + err = netpoll_setup(&nt->np); if (err) { pr_err("Not enabling netconsole for %s%d. Netpoll setup failed\n", NETCONSOLE_PARAM_TARGET_PREFIX, cmdline_count); + netconsole_skb_pool_flush(nt); if (!IS_ENABLED(CONFIG_NETCONSOLE_DYNAMIC)) /* only fail if dynamic reconfiguration is set, * otherwise, keep the target in the list, but disabled. @@ -2355,6 +2402,8 @@ static struct netconsole_target *alloc_param_target(char *target_config, static void free_param_target(struct netconsole_target *nt) { cancel_work_sync(&nt->resume_wq); + if (nt->state == STATE_ENABLED) + netconsole_skb_pool_flush(nt); netpoll_cleanup(&nt->np); #ifdef CONFIG_NETCONSOLE_DYNAMIC kfree(nt->userdata); diff --git a/net/core/netpoll.c b/net/core/netpoll.c index e062d88d10a37..58f30a4d5eb0f 100644 --- a/net/core/netpoll.c +++ b/net/core/netpoll.c @@ -377,9 +377,6 @@ int __netpoll_setup(struct netpoll *np, struct net_device *ndev) const struct net_device_ops *ops; int err; - skb_queue_head_init(&np->skb_pool); - INIT_WORK(&np->refill_wq, refill_skbs_work_handler); - if (ndev->priv_flags & IFF_DISABLE_NETPOLL) { np_err(np, "%s doesn't support polling, aborting\n", ndev->name); @@ -414,9 +411,6 @@ int __netpoll_setup(struct netpoll *np, struct net_device *ndev) np->dev = ndev; strscpy(np->dev_name, ndev->name, IFNAMSIZ); - /* fill up the skb queue */ - refill_skbs(np); - /* last thing to do is link it to the net device structure */ rcu_assign_pointer(ndev->npinfo, npinfo); @@ -606,7 +600,7 @@ int netpoll_setup(struct netpoll *np) err = __netpoll_setup(np, ndev); if (err) - goto flush; + goto put; rtnl_unlock(); /* Make sure all NAPI polls which started before dev->npinfo @@ -617,8 +611,6 @@ int netpoll_setup(struct netpoll *np) return 0; -flush: - skb_pool_flush(np); put: DEBUG_NET_WARN_ON_ONCE(np->dev); if (ip_overwritten) @@ -662,8 +654,6 @@ static void __netpoll_cleanup(struct netpoll *np) disable_delayed_work_sync(&npinfo->tx_work); call_rcu(&npinfo->rcu, rcu_cleanup_netpoll_info); } - - skb_pool_flush(np); } void __netpoll_free(struct netpoll *np) -- 2.53.0-Meta