Eric Dumazet says: ==================== net: netns refcount tracking, base series We have 100+ syzbot reports about netns being dismantled too soon, still unresolved as of today. We think a missing get_net() or an extra put_net() is the root cause. In order to find the bug(s), and be able to spot future ones, this patch adds CONFIG_NET_NS_REFCNT_TRACKER and new helpers to precisely pair all put_net() with corresponding get_net(). To use these helpers, each data structure owning a refcount should also use a "netns_tracker" to pair the get() and put(). Small sections of codes where the get()/put() are in sight do not need to have a tracker, because they are short lived, but in theory it is also possible to declare an on-stack tracker. v2: Include core networking patches only. ====================
Unified · Split
| @@ -173,6 +173,7 @@ struct channel { · 172 unmodified lines | ||
| 173 | 173 | spinlock_t downl; /* protects `chan', file.xq dequeue */ |
| 174 | 174 | struct ppp *ppp; /* ppp unit we're connected to */ |
| 175 | 175 | struct net *chan_net; /* the net channel belongs to */ |
| 176 | + netns_tracker ns_tracker; | |
| 176 | 177 | struct list_head clist; /* link in list of channels per unit */ |
| 177 | 178 | rwlock_t upl; /* protects `ppp' and 'bridge' */ |
| 178 | 179 | struct channel __rcu *bridge; /* "bridged" ppp channel */ |
| @@ -2879,7 +2880,7 @@ int ppp_register_net_channel(struct net *net, struct ppp_channel *chan) · 2700 unmodified lines | ||
| 2879 | 2880 | |
| 2880 | 2881 | pch->ppp = NULL; |
| 2881 | 2882 | pch->chan = chan; |
| 2882 | − pch->chan_net = get_net(net); | |
| 2883 | + pch->chan_net = get_net_track(net, &pch->ns_tracker, GFP_KERNEL); | |
| 2883 | 2884 | chan->ppp = pch; |
| 2884 | 2885 | init_ppp_file(&pch->file, CHANNEL); |
| 2885 | 2886 | pch->file.hdrlen = chan->hdrlen; |
| @@ -3519,7 +3520,7 @@ ppp_disconnect_channel(struct channel *pch) · 633 unmodified lines | ||
| 3519 | 3520 | */ |
| 3520 | 3521 | static void ppp_destroy_channel(struct channel *pch) |
| 3521 | 3522 | { |
| 3522 | − put_net(pch->chan_net); | |
| 3523 | + put_net_track(pch->chan_net, &pch->ns_tracker); | |
| 3523 | 3524 | pch->chan_net = NULL; |
| 3524 | 3525 | |
| 3525 | 3526 | atomic_dec(&channel_count); |
| @@ -61,15 +61,27 @@ static int seq_open_net(struct inode *inode, struct file *file) · 60 unmodified lines | ||
| 61 | 61 | } |
| 62 | 62 | #ifdef CONFIG_NET_NS |
| 63 | 63 | p->net = net; |
| 64 | + netns_tracker_alloc(net, &p->ns_tracker, GFP_KERNEL); | |
| 64 | 65 | #endif |
| 65 | 66 | return 0; |
| 66 | 67 | } |
| 67 | 68 | |
| 69 | +static void seq_file_net_put_net(struct seq_file *seq) | |
| 70 | +{ | |
| 71 | +#ifdef CONFIG_NET_NS | |
| 72 | + struct seq_net_private *priv = seq->private; | |
| 73 | + | |
| 74 | + put_net_track(priv->net, &priv->ns_tracker); | |
| 75 | +#else | |
| 76 | + put_net(&init_net); | |
| 77 | +#endif | |
| 78 | +} | |
| 79 | + | |
| 68 | 80 | static int seq_release_net(struct inode *ino, struct file *f) |
| 69 | 81 | { |
| 70 | 82 | struct seq_file *seq = f->private_data; |
| 71 | 83 | |
| 72 | − put_net(seq_file_net(seq)); | |
| 84 | + seq_file_net_put_net(seq); | |
| 73 | 85 | seq_release_private(ino, f); |
| 74 | 86 | return 0; |
| 75 | 87 | } |
| @@ -87,7 +99,8 @@ int bpf_iter_init_seq_net(void *priv_data, struct bpf_iter_aux_info *aux) · 11 unmodified lines | ||
| 87 | 99 | #ifdef CONFIG_NET_NS |
| 88 | 100 | struct seq_net_private *p = priv_data; |
| 89 | 101 | |
| 90 | − p->net = get_net(current->nsproxy->net_ns); | |
| 102 | + p->net = get_net_track(current->nsproxy->net_ns, &p->ns_tracker, | |
| 103 | + GFP_KERNEL); | |
| 91 | 104 | #endif |
| 92 | 105 | return 0; |
| 93 | 106 | } |
| @@ -97,7 +110,7 @@ void bpf_iter_fini_seq_net(void *priv_data) · 3 unmodified lines | ||
| 97 | 110 | #ifdef CONFIG_NET_NS |
| 98 | 111 | struct seq_net_private *p = priv_data; |
| 99 | 112 | |
| 100 | − put_net(p->net); | |
| 113 | + put_net_track(p->net, &p->ns_tracker); | |
| 101 | 114 | #endif |
| 102 | 115 | } |
| 103 | 116 | |
| @@ -48,7 +48,7 @@ · 47 unmodified lines | ||
| 48 | 48 | #include <uapi/linux/pkt_cls.h> |
| 49 | 49 | #include <linux/hashtable.h> |
| 50 | 50 | #include <linux/rbtree.h> |
| 51 | −#include <linux/ref_tracker.h> | |
| 51 | +#include <net/net_trackers.h> | |
| 52 | 52 | |
| 53 | 53 | struct netpoll_info; |
| 54 | 54 | struct device; |
| @@ -300,13 +300,6 @@ enum netdev_state_t { · 245 unmodified lines | ||
| 300 | 300 | __LINK_STATE_TESTING, |
| 301 | 301 | }; |
| 302 | 302 | |
| 303 | − | |
| 304 | −#ifdef CONFIG_NET_DEV_REFCNT_TRACKER | |
| 305 | −typedef struct ref_tracker *netdevice_tracker; | |
| 306 | −#else | |
| 307 | −typedef struct {} netdevice_tracker; | |
| 308 | −#endif | |
| 309 | − | |
| 310 | 303 | struct gro_list { |
| 311 | 304 | struct list_head list; |
| 312 | 305 | int count; |
| @@ -9,7 +9,8 @@ extern struct net init_net; · 8 unmodified lines | ||
| 9 | 9 | |
| 10 | 10 | struct seq_net_private { |
| 11 | 11 | #ifdef CONFIG_NET_NS |
| 12 | − struct net *net; | |
| 12 | + struct net *net; | |
| 13 | + netns_tracker ns_tracker; | |
| 13 | 14 | #endif |
| 14 | 15 | }; |
| 15 | 16 | |
| @@ -34,6 +34,7 @@ · 33 unmodified lines | ||
| 34 | 34 | #include <net/netns/smc.h> |
| 35 | 35 | #include <net/netns/bpf.h> |
| 36 | 36 | #include <net/netns/mctp.h> |
| 37 | +#include <net/net_trackers.h> | |
| 37 | 38 | #include <linux/ns_common.h> |
| 38 | 39 | #include <linux/idr.h> |
| 39 | 40 | #include <linux/skbuff.h> |
| @@ -87,6 +88,7 @@ struct net { · 47 unmodified lines | ||
| 87 | 88 | struct idr netns_ids; |
| 88 | 89 | |
| 89 | 90 | struct ns_common ns; |
| 91 | + struct ref_tracker_dir refcnt_tracker; | |
| 90 | 92 | |
| 91 | 93 | struct list_head dev_base_head; |
| 92 | 94 | struct proc_dir_entry *proc_net; |
| @@ -240,6 +242,7 @@ void ipx_unregister_sysctl(void); · 147 unmodified lines | ||
| 240 | 242 | #ifdef CONFIG_NET_NS |
| 241 | 243 | void __put_net(struct net *net); |
| 242 | 244 | |
| 245 | +/* Try using get_net_track() instead */ | |
| 243 | 246 | static inline struct net *get_net(struct net *net) |
| 244 | 247 | { |
| 245 | 248 | refcount_inc(&net->ns.count); |
| @@ -258,6 +261,7 @@ static inline struct net *maybe_get_net(struct net *net) · 12 unmodified lines | ||
| 258 | 261 | return net; |
| 259 | 262 | } |
| 260 | 263 | |
| 264 | +/* Try using put_net_track() instead */ | |
| 261 | 265 | static inline void put_net(struct net *net) |
| 262 | 266 | { |
| 263 | 267 | if (refcount_dec_and_test(&net->ns.count)) |
| @@ -308,6 +312,36 @@ static inline int check_net(const struct net *net) · 44 unmodified lines | ||
| 308 | 312 | #endif |
| 309 | 313 | |
| 310 | 314 | |
| 315 | +static inline void netns_tracker_alloc(struct net *net, | |
| 316 | + netns_tracker *tracker, gfp_t gfp) | |
| 317 | +{ | |
| 318 | +#ifdef CONFIG_NET_NS_REFCNT_TRACKER | |
| 319 | + ref_tracker_alloc(&net->refcnt_tracker, tracker, gfp); | |
| 320 | +#endif | |
| 321 | +} | |
| 322 | + | |
| 323 | +static inline void netns_tracker_free(struct net *net, | |
| 324 | + netns_tracker *tracker) | |
| 325 | +{ | |
| 326 | +#ifdef CONFIG_NET_NS_REFCNT_TRACKER | |
| 327 | + ref_tracker_free(&net->refcnt_tracker, tracker); | |
| 328 | +#endif | |
| 329 | +} | |
| 330 | + | |
| 331 | +static inline struct net *get_net_track(struct net *net, | |
| 332 | + netns_tracker *tracker, gfp_t gfp) | |
| 333 | +{ | |
| 334 | + get_net(net); | |
| 335 | + netns_tracker_alloc(net, tracker, gfp); | |
| 336 | + return net; | |
| 337 | +} | |
| 338 | + | |
| 339 | +static inline void put_net_track(struct net *net, netns_tracker *tracker) | |
| 340 | +{ | |
| 341 | + netns_tracker_free(net, tracker); | |
| 342 | + put_net(net); | |
| 343 | +} | |
| 344 | + | |
| 311 | 345 | typedef struct { |
| 312 | 346 | #ifdef CONFIG_NET_NS |
| 313 | 347 | struct net *net; |
| @@ -0,0 +1,18 @@ | ||
| 1 | +/* SPDX-License-Identifier: GPL-2.0 */ | |
| 2 | +#ifndef __NET_NET_TRACKERS_H | |
| 3 | +#define __NET_NET_TRACKERS_H | |
| 4 | +#include <linux/ref_tracker.h> | |
| 5 | + | |
| 6 | +#ifdef CONFIG_NET_DEV_REFCNT_TRACKER | |
| 7 | +typedef struct ref_tracker *netdevice_tracker; | |
| 8 | +#else | |
| 9 | +typedef struct {} netdevice_tracker; | |
| 10 | +#endif | |
| 11 | + | |
| 12 | +#ifdef CONFIG_NET_NS_REFCNT_TRACKER | |
| 13 | +typedef struct ref_tracker *netns_tracker; | |
| 14 | +#else | |
| 15 | +typedef struct {} netns_tracker; | |
| 16 | +#endif | |
| 17 | + | |
| 18 | +#endif /* __NET_NET_TRACKERS_H */ | |
| @@ -202,7 +202,8 @@ struct tcf_exts { · 201 unmodified lines | ||
| 202 | 202 | __u32 type; /* for backward compat(TCA_OLD_COMPAT) */ |
| 203 | 203 | int nr_actions; |
| 204 | 204 | struct tc_action **actions; |
| 205 | − struct net *net; | |
| 205 | + struct net *net; | |
| 206 | + netns_tracker ns_tracker; | |
| 206 | 207 | #endif |
| 207 | 208 | /* Map to export classifier specific extension TLV types to the |
| 208 | 209 | * generic extensions API. Unsupported extensions must be set to 0. |
| @@ -218,6 +219,7 @@ static inline int tcf_exts_init(struct tcf_exts *exts, struct net *net, · 9 unmodified lines | ||
| 218 | 219 | exts->type = 0; |
| 219 | 220 | exts->nr_actions = 0; |
| 220 | 221 | exts->net = net; |
| 222 | + netns_tracker_alloc(net, &exts->ns_tracker, GFP_KERNEL); | |
| 221 | 223 | exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *), |
| 222 | 224 | GFP_KERNEL); |
| 223 | 225 | if (!exts->actions) |
| @@ -236,6 +238,8 @@ static inline bool tcf_exts_get_net(struct tcf_exts *exts) · 12 unmodified lines | ||
| 236 | 238 | { |
| 237 | 239 | #ifdef CONFIG_NET_CLS_ACT |
| 238 | 240 | exts->net = maybe_get_net(exts->net); |
| 241 | + if (exts->net) | |
| 242 | + netns_tracker_alloc(exts->net, &exts->ns_tracker, GFP_KERNEL); | |
| 239 | 243 | return exts->net != NULL; |
| 240 | 244 | #else |
| 241 | 245 | return true; |
| @@ -246,7 +250,7 @@ static inline void tcf_exts_put_net(struct tcf_exts *exts) · 4 unmodified lines | ||
| 246 | 250 | { |
| 247 | 251 | #ifdef CONFIG_NET_CLS_ACT |
| 248 | 252 | if (exts->net) |
| 249 | − put_net(exts->net); | |
| 253 | + put_net_track(exts->net, &exts->ns_tracker); | |
| 250 | 254 | #endif |
| 251 | 255 | } |
| 252 | 256 | |
| @@ -350,6 +350,7 @@ struct bpf_local_storage; · 349 unmodified lines | ||
| 350 | 350 | * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME |
| 351 | 351 | * @sk_txtime_report_errors: set report errors mode for SO_TXTIME |
| 352 | 352 | * @sk_txtime_unused: unused txtime flags |
| 353 | + * @ns_tracker: tracker for netns reference | |
| 353 | 354 | */ |
| 354 | 355 | struct sock { |
| 355 | 356 | /* |
| @@ -538,6 +539,7 @@ struct sock { · 182 unmodified lines | ||
| 538 | 539 | struct bpf_local_storage __rcu *sk_bpf_storage; |
| 539 | 540 | #endif |
| 540 | 541 | struct rcu_head sk_rcu; |
| 542 | + netns_tracker ns_tracker; | |
| 541 | 543 | }; |
| 542 | 544 | |
| 543 | 545 | enum sk_pacing { |
| @@ -8,3 +8,12 @@ config NET_DEV_REFCNT_TRACKER · 7 unmodified lines | ||
| 8 | 8 | help |
| 9 | 9 | Enable debugging feature to track device references. |
| 10 | 10 | This adds memory and cpu costs. |
| 11 | + | |
| 12 | +config NET_NS_REFCNT_TRACKER | |
| 13 | + bool "Enable networking namespace refcount tracking" | |
| 14 | + depends on DEBUG_KERNEL && STACKTRACE_SUPPORT | |
| 15 | + select REF_TRACKER | |
| 16 | + default n | |
| 17 | + help | |
| 18 | + Enable debugging feature to track netns references. | |
| 19 | + This adds memory and cpu costs. | |
| @@ -311,6 +311,8 @@ static __net_init int setup_net(struct net *net, struct user_namespace *user_ns) · 310 unmodified lines | ||
| 311 | 311 | LIST_HEAD(net_exit_list); |
| 312 | 312 | |
| 313 | 313 | refcount_set(&net->ns.count, 1); |
| 314 | + ref_tracker_dir_init(&net->refcnt_tracker, 128); | |
| 315 | + | |
| 314 | 316 | refcount_set(&net->passive, 1); |
| 315 | 317 | get_random_bytes(&net->hash_mix, sizeof(u32)); |
| 316 | 318 | preempt_disable(); |
| @@ -635,6 +637,7 @@ static DECLARE_WORK(net_cleanup_work, cleanup_net); · 318 unmodified lines | ||
| 635 | 637 | |
| 636 | 638 | void __put_net(struct net *net) |
| 637 | 639 | { |
| 640 | + ref_tracker_dir_exit(&net->refcnt_tracker); | |
| 638 | 641 | /* Cleanup the network namespace in process context */ |
| 639 | 642 | if (llist_add(&net->cleanup_list, &cleanup_list)) |
| 640 | 643 | queue_work(netns_wq, &net_cleanup_work); |
| @@ -1983,7 +1983,7 @@ struct sock *sk_alloc(struct net *net, int family, gfp_t priority, · 1982 unmodified lines | ||
| 1983 | 1983 | sock_lock_init(sk); |
| 1984 | 1984 | sk->sk_net_refcnt = kern ? 0 : 1; |
| 1985 | 1985 | if (likely(sk->sk_net_refcnt)) { |
| 1986 | − get_net(net); | |
| 1986 | + get_net_track(net, &sk->ns_tracker, priority); | |
| 1987 | 1987 | sock_inuse_add(net, 1); |
| 1988 | 1988 | } |
| 1989 | 1989 | |
| @@ -2039,7 +2039,7 @@ static void __sk_destruct(struct rcu_head *head) · 49 unmodified lines | ||
| 2039 | 2039 | put_pid(sk->sk_peer_pid); |
| 2040 | 2040 | |
| 2041 | 2041 | if (likely(sk->sk_net_refcnt)) |
| 2042 | − put_net(sock_net(sk)); | |
| 2042 | + put_net_track(sock_net(sk), &sk->ns_tracker); | |
| 2043 | 2043 | sk_prot_free(sk->sk_prot_creator, sk); |
| 2044 | 2044 | } |
| 2045 | 2045 | |
| @@ -2126,7 +2126,7 @@ struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) · 80 unmodified lines | ||
| 2126 | 2126 | |
| 2127 | 2127 | /* SANITY */ |
| 2128 | 2128 | if (likely(newsk->sk_net_refcnt)) { |
| 2129 | − get_net(sock_net(newsk)); | |
| 2129 | + get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); | |
| 2130 | 2130 | sock_inuse_add(sock_net(newsk), 1); |
| 2131 | 2131 | } |
| 2132 | 2132 | sk_node_init(&newsk->sk_node); |
| @@ -32,7 +32,8 @@ · 31 unmodified lines | ||
| 32 | 32 | static struct dentry *rootdir; |
| 33 | 33 | |
| 34 | 34 | struct l2tp_dfs_seq_data { |
| 35 | − struct net *net; | |
| 35 | + struct net *net; | |
| 36 | + netns_tracker ns_tracker; | |
| 36 | 37 | int tunnel_idx; /* current tunnel */ |
| 37 | 38 | int session_idx; /* index of session within current tunnel */ |
| 38 | 39 | struct l2tp_tunnel *tunnel; |
| @@ -281,7 +282,7 @@ static int l2tp_dfs_seq_open(struct inode *inode, struct file *file) · 242 unmodified lines | ||
| 281 | 282 | rc = PTR_ERR(pd->net); |
| 282 | 283 | goto err_free_pd; |
| 283 | 284 | } |
| 284 | − | |
| 285 | + netns_tracker_alloc(pd->net, &pd->ns_tracker, GFP_KERNEL); | |
| 285 | 286 | rc = seq_open(file, &l2tp_dfs_seq_ops); |
| 286 | 287 | if (rc) |
| 287 | 288 | goto err_free_net; |
| @@ -293,7 +294,7 @@ out: · 5 unmodified lines | ||
| 293 | 294 | return rc; |
| 294 | 295 | |
| 295 | 296 | err_free_net: |
| 296 | − put_net(pd->net); | |
| 297 | + put_net_track(pd->net, &pd->ns_tracker); | |
| 297 | 298 | err_free_pd: |
| 298 | 299 | kfree(pd); |
| 299 | 300 | goto out; |
| @@ -307,7 +308,7 @@ static int l2tp_dfs_seq_release(struct inode *inode, struct file *file) · 7 unmodified lines | ||
| 307 | 308 | seq = file->private_data; |
| 308 | 309 | pd = seq->private; |
| 309 | 310 | if (pd->net) |
| 310 | − put_net(pd->net); | |
| 311 | + put_net_track(pd->net, &pd->ns_tracker); | |
| 311 | 312 | kfree(pd); |
| 312 | 313 | seq_release(inode, file); |
| 313 | 314 | |