net/sched: account classifier filter allocations to memcg
Subsystems (6)
Jamal Hadi Salim
maintainerPATCH 1/1 net/sched: account classifier filter allocations to memcg
Allocations in the tc classifier *_change() paths (filter objects, per-CPU counters, and per-filter aux data) use plain GFP_KERNEL without __GFP_ACCOUNT, allowing unprivileged users to pin kernel memory outside memcg charging. The shared tcf_exts_init_ex() action array allocation in cls_api.c was also uncharged; this patch closes it along with the per-classifier filter-object/percpu/aux allocations that remain unaccounted.
Add GFP_KERNEL_ACCOUNT to:
- the shared tcf_exts_init_ex() action array (cls_api.c), common to every filter of every classifier (32 pointers, 256 bytes);
- the filter-object, per-CPU-counter, and per-filter aux allocations in cls_basic, cls_bpf, cls_cgroup, cls_flow, cls_flower, cls_fw, cls_matchall, cls_route and cls_u32;
- the u32_init_knode() replace-path knode allocation (cls_u32.c), which allocates the same struct tc_u_knode + sel.keys on every replace of an existing knode and was missed by the create-path-only conversion.
Also fix the cls_basic error path: basic_change() inserts fnew into the IDR before allocating the per-CPU counter. If alloc_percpu() fails the errout path kfree'd fnew without idr_remove, leaving a dangling pointer in the IDR. With GFP_KERNEL_ACCOUNT the percpu alloc becomes failable on demand (memcg at memory.max), making the dead path attacker-reachable and burning the handle permanently. Add the idr_remove on the percpu failure path, matching the basic_set_parms failure-path pattern.
Note: [email protected] provided a poc for basic_cls, but it was easy to extend to the other classifiers.
Conditions to recreate the bug:
- CONFIG_NET_SCHED, CONFIG_NET_CLS_* (the classifier being used), CONFIG_NET_CLS_ACT, CONFIG_MEMCG, CONFIG_USER_NS, CONFIG_NET_NS.
- Unprivileged user in a fresh user+network namespace (unshare -Urn), or root with CAP_NET_ADMIN.
- Create a large number of tc filters (e.g. tc filter add dev lo ingress ... <classifier> ...) while watching a memcg-limited cgroup:
system slab grows far faster than memory.current, pinning kernel memory outside memcg charging.
--- net/sched/cls_api.c | 3 ++- net/sched/cls_basic.c | 6 ++++-- net/sched/cls_bpf.c | 6 +++--- net/sched/cls_cgroup.c | 2 +- net/sched/cls_flow.c | 2 +- net/sched/cls_flower.c | 4 ++-- net/sched/cls_fw.c | 4 ++-- net/sched/cls_matchall.c | 4 ++-- net/sched/cls_route.c | 4 ++-- net/sched/cls_u32.c | 11 ++++++----- 10 files changed, 25 insertions(+), 21 deletions(-)
net/sched/cls_api.c +2 -1
@@ -3364,7 +3364,8 @@ int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,* This reference might be taken later from tcf_exts_get_net(). */ exts->net = net; - exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO); + exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO, +GFP_KERNEL_ACCOUNT );if (!exts->actions)return -ENOMEM ;#endif
net/sched/cls_basic.c +4 -2
@@ -193,7 +193,7 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,return -EINVAL ; } - fnew =kzalloc_obj (*fnew); + fnew =kzalloc_obj (*fnew,GFP_KERNEL_ACCOUNT );if (!fnew)return -ENOBUFS ;
@@ -212,9 +212,11 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,if (err) goto errout; fnew->handle = handle; - fnew->pf = alloc_percpu(struct tc_basic_pcnt); + fnew->pf = alloc_percpu_gfp(struct tc_basic_pcnt, GFP_KERNEL_ACCOUNT);if (!fnew->pf) { err = -ENOMEM ; +if (!fold) +idr_remove (&head->handle_idr, fnew->handle); goto errout; }
net/sched/cls_bpf.c +3 -3
@@ -352,7 +352,7 @@ static int cls_bpf_prog_from_ops(struct nlattr **tb, struct cls_bpf_prog *prog)if (bpf_size !=nla_len (tb[TCA_BPF_OPS ]))return -EINVAL ; - bpf_ops =kmemdup (nla_data (tb[TCA_BPF_OPS ]), bpf_size,GFP_KERNEL ); + bpf_ops =kmemdup (nla_data (tb[TCA_BPF_OPS ]), bpf_size,GFP_KERNEL_ACCOUNT );if (bpf_ops ==NULL )return -ENOMEM ;
@@ -403,7 +403,7 @@ static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog, }if (tb[TCA_BPF_NAME ]) { - name =nla_memdup (tb[TCA_BPF_NAME ],GFP_KERNEL ); + name =nla_memdup (tb[TCA_BPF_NAME ],GFP_KERNEL_ACCOUNT );if (!name) {bpf_prog_put (fp);return -ENOMEM ;
@@ -443,7 +443,7 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,if (ret <0 )return ret; - prog =kzalloc_obj (*prog); + prog =kzalloc_obj (*prog,GFP_KERNEL_ACCOUNT );if (!prog)return -ENOBUFS ;
net/sched/cls_cgroup.c +1 -1
@@ -95,7 +95,7 @@ static int cls_cgroup_change(struct net *net, struct sk_buff *in_skb,if (head && handle != head->handle)return -ENOENT ; - new =kzalloc_obj (*head); + new =kzalloc_obj (*head,GFP_KERNEL_ACCOUNT );if (!new)return -ENOBUFS ;
net/sched/cls_flow.c +1 -1
@@ -438,7 +438,7 @@ static int flow_change(struct net *net, struct sk_buff *in_skb,return -EOPNOTSUPP ; } - fnew =kzalloc_obj (*fnew); + fnew =kzalloc_obj (*fnew,GFP_KERNEL_ACCOUNT );if (!fnew)return -ENOBUFS ;
net/sched/cls_flower.c +2 -2
@@ -2233,7 +2233,7 @@ static struct fl_flow_mask *fl_create_new_mask(struct cls_fl_head *head,struct fl_flow_mask *newmask;int err; - newmask =kzalloc_obj (*newmask); + newmask =kzalloc_obj (*newmask,GFP_KERNEL_ACCOUNT );if (!newmask)return ERR_PTR (-ENOMEM );
@@ -2394,7 +2394,7 @@ static int fl_change(struct net *net, struct sk_buff *in_skb, goto errout_tb; } - fnew =kzalloc_obj (*fnew); + fnew =kzalloc_obj (*fnew,GFP_KERNEL_ACCOUNT );if (!fnew) { err = -ENOBUFS ; goto errout_tb;
net/sched/cls_fw.c +2 -2
@@ -276,7 +276,7 @@ static int fw_change(struct net *net, struct sk_buff *in_skb,if (f->id != handle && handle)return -EINVAL ; - fnew = kzalloc_obj(struct fw_filter); + fnew = kzalloc_obj(struct fw_filter, GFP_KERNEL_ACCOUNT);if (!fnew)return -ENOBUFS ;
@@ -330,7 +330,7 @@ static int fw_change(struct net *net, struct sk_buff *in_skb,rcu_assign_pointer (tp->root, head); } - f = kzalloc_obj(struct fw_filter); + f = kzalloc_obj(struct fw_filter, GFP_KERNEL_ACCOUNT);if (f ==NULL )return -ENOBUFS ;
net/sched/cls_matchall.c +2 -2
@@ -189,7 +189,7 @@ static int mall_change(struct net *net, struct sk_buff *in_skb,return -EINVAL ; } - new =kzalloc_obj (*new); + new =kzalloc_obj (*new,GFP_KERNEL_ACCOUNT );if (!new)return -ENOBUFS ;
@@ -201,7 +201,7 @@ static int mall_change(struct net *net, struct sk_buff *in_skb, handle =1 ; new->handle = handle; new->flags = userflags; - new->pf = alloc_percpu(struct tc_matchall_pcnt); + new->pf = alloc_percpu_gfp(struct tc_matchall_pcnt, GFP_KERNEL_ACCOUNT);if (!new->pf) { err = -ENOMEM ; goto err_alloc_percpu;
net/sched/cls_route.c +2 -2
@@ -455,7 +455,7 @@ static int route4_set_parms(struct net *net, struct tcf_proto *tp, h1 =to_hash (nhandle); b =rtnl_dereference (head->table[h1]);if (!b) { - b = kzalloc_obj(struct route4_bucket); + b = kzalloc_obj(struct route4_bucket, GFP_KERNEL_ACCOUNT);if (b ==NULL )return -ENOBUFS ;
@@ -524,7 +524,7 @@ static int route4_change(struct net *net, struct sk_buff *in_skb,return -EINVAL ; err = -ENOBUFS ; - f = kzalloc_obj(struct route4_filter); + f = kzalloc_obj(struct route4_filter, GFP_KERNEL_ACCOUNT);if (!f) goto errout;
net/sched/cls_u32.c +6 -5
@@ -825,7 +825,7 @@ static struct tc_u_knode *u32_init_knode(struct net *net, struct tcf_proto *tp,struct tc_u32_sel *s = &n->sel;struct tc_u_knode *new; - new =kzalloc_flex (*new, sel.keys, s->nkeys); + new =kzalloc_flex (*new, sel.keys, s->nkeys,GFP_KERNEL_ACCOUNT );if (!new)return NULL ;
@@ -1114,15 +1114,16 @@ static int u32_change(struct net *net, struct sk_buff *in_skb, goto erridr; } - n =kzalloc_flex (*n, sel.keys, s->nkeys); + n =kzalloc_flex (*n, sel.keys, s->nkeys,GFP_KERNEL_ACCOUNT );if (n ==NULL ) { err = -ENOBUFS ; goto erridr; }#ifdef CONFIG_CLS_U32_PERF - n->pf =__alloc_percpu (struct_size (n->pf, kcnts, s->nkeys), - __alignof__(struct tc_u32_pcnt )); + n->pf =__alloc_percpu_gfp (struct_size (n->pf, kcnts, s->nkeys), + __alignof__(struct tc_u32_pcnt ), +GFP_KERNEL_ACCOUNT );if (!n->pf) { err = -ENOBUFS ; goto errfree;
@@ -1144,7 +1145,7 @@ static int u32_change(struct net *net, struct sk_buff *in_skb, goto errout;#ifdef CONFIG_CLS_U32_MARK - n->pcpu_success =alloc_percpu (u32); + n->pcpu_success =alloc_percpu_gfp (u32,GFP_KERNEL_ACCOUNT );if (!n->pcpu_success) { err = -ENOMEM ; goto errout;
Signature · 1 line
2.43.0
Quoted 39 lines · Jamal Hadi Salim
Allocations in the tc classifier
*_change()paths (filter objects, per-CPU counters, and per-filter aux data) use plainGFP_KERNELwithout__GFP_ACCOUNT, allowing unprivileged users to pin kernel memory outside memcg charging. The sharedtcf_exts_init_ex()action array allocation incls_api.cwas also uncharged; this patch closes it along with the per-classifier filter-object/percpu/aux allocations that remain unaccounted.Add
GFP_KERNEL_ACCOUNTto:
- the sharedtcf_exts_init_ex()action array (cls_api.c), common to every filter of every classifier (32 pointers, 256 bytes);
- the filter-object, per-CPU-counter, and per-filter aux allocations incls_basic,cls_bpf,cls_cgroup,cls_flow,cls_flower,cls_fw,cls_matchall,cls_routeandcls_u32;
- theu32_init_knode()replace-path knode allocation (cls_u32.c), which allocates the same structtc_u_knode+ sel.keys on every replace of an existing knode and was missed by the create-path-only conversion.Also fix the
cls_basicerror path:basic_change()inserts fnew into the IDR before allocating the per-CPU counter. Ifalloc_percpu()fails the errout path kfree'd fnew withoutidr_remove, leaving a dangling pointer in the IDR. WithGFP_KERNEL_ACCOUNTthe percpu alloc becomes failable on demand (memcg at memory.max), making the dead path attacker-reachable and burning the handle permanently. Add theidr_removeon the percpu failure path, matching thebasic_set_parmsfailure-path pattern.Note: [email protected] provided a poc for
basic_cls, but it was easy to extend to the other classifiers.Conditions to recreate the bug:
-CONFIG_NET_SCHED, CONFIG_NET_CLS_* (the classifier being used),CONFIG_NET_CLS_ACT,CONFIG_MEMCG,CONFIG_USER_NS,CONFIG_NET_NS.
- Unprivileged user in a fresh user+network namespace (unshare -Urn), or root withCAP_NET_ADMIN.
- Create a large number of tc filters (e.g. tc filter add dev lo ingress ... <classifier> ...) while watching a memcg-limited cgroup:
system slab grows far faster than memory.current, pinning kernel memory outside memcg charging.
Quoted 3 lines
Fixes 0da974f4f303 ("[NET]: Conversions from kmalloc+memset to k(z|c)alloc.")Reported-by [email protected]Signed-off-by Jamal Hadi Salim <[email protected]>
Jakub Kicinski
maintainerThis is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net/sched: account classifier filter allocations to memcg
This adds GFP_KERNEL_ACCOUNT to the filter objects, per-CPU counters and per-filter aux allocations in the tc classifier *_change() paths, plus the shared exts->actions array in tcf_exts_init_ex(). It also adds an idr_remove() to the cls_basic percpu-allocation failure path.
Quoted 19 lines
net/sched/cls_api.c +2
-1@@ -3364,7 +3364,8 @@ int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,* This reference might be taken later from tcf_exts_get_net(). */ exts->net = net; - exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO); + exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO, +GFP_KERNEL_ACCOUNT );if (!exts->actions)return -ENOMEM ;#endif net/sched/cls_basic.c +0
-0
net/sched/cls_basic.c static int basic_change(struct net *net, struct sk_buff *in_skb, · line 199
Quoted 8 lines
net/sched/cls_basic.c +1
-1@@ -193,7 +193,7 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,return -EINVAL ; } - fnew =kzalloc_obj (*fnew); + fnew =kzalloc_obj (*fnew,GFP_KERNEL_ACCOUNT );if (!fnew)return -ENOBUFS ;
[Severity: Medium]
The changelog says this "closes" the bypass that lets "unprivileged users to pin kernel memory outside memcg charging". Is that accurate when the larger, user-sized allocations on the very same basic_change() request are still uncharged?
basic_change() charges fnew and fnew->pf here, then reaches the ematch tree through basic_set_parms():
net/sched/cls_basic.c:basic_set_parms()
err = tcf_em_tree_validate(tp, tb[TCA_BASIC_EMATCHES], &f->ematches);
net/sched/ematch.c:tcf_em_tree_validate()
matches_len = tree_hdr->nmatches * sizeof(*em);
tree->matches = kzalloc(matches_len, GFP_KERNEL);and per-ematch payloads in tcf_em_validate():
void *v = kmemdup(data, data_len, GFP_KERNEL);
nmatches and data_len come from the netlink message, so the uncharged part of a single filter can be much larger than the charged part. The same tcf_em_tree_validate() path is used by flow_change() and cls_cgroup_change(), whose filter objects this patch also charges.
Should the u32 hash tables be converted too? u32_change() charges the knode, but the TCA_U32_DIVISOR branch in the same function still uses the default gfp for the up-to-257-pointer table that is published in tp_c->hlist:
net/sched/cls_u32.c:u32_change()
ht = kzalloc_flex(*ht, ht, divisor + 1); if (ht == NULL) return -ENOBUFS;
The same object type is allocated as root_ht in u32_init(), also uncharged.
There are a few other siblings on these paths that keep plain GFP_KERNEL:
- fw_change() allocates head = kzalloc_obj(*head) (fw_head with its 256-entry hash table) a few lines above the fw_filter allocation this patch converts.
- route4_init() allocates route4_head, which carries table[256], while route4_filter and route4_bucket in the same file are converted.
- fl_init() allocates the cls_fl_head plus the rhashtable that grows with each inserted filter, and fl_tmplt_create() allocates the template object; both stay uncharged while fnew and newmask are charged.
- tcf_proto_create(), tcf_chain_create() and tcf_block_create() in cls_api.c, and the IDR nodes for the handles being charged (idr_alloc_u32(..., GFP_KERNEL) in basic_change() and u32_change()).
With the boundary running through the middle of one netlink request, a memcg-limited task can still see slab grow faster than memory.current, which is the signal the changelog uses as the reproducer. Was leaving these out intentional, or is the intent to convert them in a follow-up?
Quoted 13 lines
net/sched/cls_basic.c +3
-1@@ -212,9 +212,11 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,if (err) goto errout; fnew->handle = handle; - fnew->pf = alloc_percpu(struct tc_basic_pcnt); + fnew->pf = alloc_percpu_gfp(struct tc_basic_pcnt, GFP_KERNEL_ACCOUNT);if (!fnew->pf) { err = -ENOMEM ; +if (!fold) +idr_remove (&head->handle_idr, fnew->handle); goto errout; }
[ ... ]
Quoted 15 lines
net/sched/cls_u32.c +1
-1@@ -1114,15 +1114,16 @@ static int u32_change(struct net *net, struct sk_buff *in_skb, goto erridr; } - n =kzalloc_flex (*n, sel.keys, s->nkeys); + n =kzalloc_flex (*n, sel.keys, s->nkeys,GFP_KERNEL_ACCOUNT );if (n ==NULL ) { err = -ENOBUFS ; goto erridr; }
[ ... ]
Jamal Hadi Salim
maintainerQuoted 11 lines · Jakub Kicinski
This is an AI-generated review of your patch. The human sending this email has considered the AI review valid, or at least plausible.
---
net/sched: account classifier filter allocations to memcgThis adds
GFP_KERNEL_ACCOUNTto the filter objects, per-CPU counters and per-filter aux allocations in the tc classifier*_change()paths, plus the sharedexts->actionsarray intcf_exts_init_ex(). It also adds anidr_remove()to thecls_basicpercpu-allocation failure path.
Thanks for merging.
I have a list of follow-up items for this patch - mostly pre-existing issues and others pointing to similar patterns identified by the sashikos.
cheers,
jamal
Quoted 119 lines
Quoted 19 lines
net/sched/cls_api.c +2
-1@@ -3364,7 +3364,8 @@ int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,* This reference might be taken later from tcf_exts_get_net(). */ exts->net = net; - exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO); + exts->actions = kzalloc_objs(struct tc_action *, TCA_ACT_MAX_PRIO, +GFP_KERNEL_ACCOUNT );if (!exts->actions)return -ENOMEM ;#endif net/sched/cls_basic.c +0
-0net/sched/cls_basic.c static int basic_change(struct net *net, struct sk_buff *in_skb, · line 199
Quoted 8 lines
net/sched/cls_basic.c +1
-1@@ -193,7 +193,7 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,return -EINVAL ; } - fnew =kzalloc_obj (*fnew); + fnew =kzalloc_obj (*fnew,GFP_KERNEL_ACCOUNT );if (!fnew)return -ENOBUFS ;[Severity: Medium]
The changelog says this "closes" the bypass that lets "unprivileged users to pin kernel memory outside memcg charging". Is that accurate when the larger, user-sized allocations on the very samebasic_change()request are still uncharged?
basic_change()charges fnew andfnew->pfhere, then reaches the ematch tree throughbasic_set_parms():net/sched/cls_basic.c:basic_set_parms()
err = tcf_em_tree_validate(tp, tb[TCA_BASIC_EMATCHES], &f->ematches);net/sched/ematch.c:tcf_em_tree_validate()
matches_len = tree_hdr->nmatches * sizeof(*em); tree->matches = kzalloc(matches_len, GFP_KERNEL);and per-ematch payloads in
tcf_em_validate():void *v = kmemdup(data, data_len, GFP_KERNEL);nmatches and
data_lencome from the netlink message, so the uncharged part of a single filter can be much larger than the charged part. The sametcf_em_tree_validate()path is used byflow_change()andcls_cgroup_change(), whose filter objects this patch also charges.Should the u32 hash tables be converted too?
u32_change()charges the knode, but theTCA_U32_DIVISORbranch in the same function still uses the default gfp for the up-to-257-pointer table that is published intp_c->hlist:net/sched/cls_u32.c:u32_change()
ht = kzalloc_flex(*ht, ht, divisor + 1); if (ht == NULL) return -ENOBUFS;The same object type is allocated as
root_htinu32_init(), also uncharged.There are a few other siblings on these paths that keep plain
GFP_KERNEL:-
fw_change()allocates head =kzalloc_obj(*head)(fw_headwith its 256-entry hash table) a few lines above thefw_filterallocation this patch converts.
-route4_init()allocatesroute4_head, which carries table[256], whileroute4_filterandroute4_bucketin the same file are converted.
-fl_init()allocates thecls_fl_headplus the rhashtable that grows with each inserted filter, andfl_tmplt_create()allocates the template object; both stay uncharged while fnew and newmask are charged.
-tcf_proto_create(),tcf_chain_create()andtcf_block_create()incls_api.c, and the IDR nodes for the handles being charged (idr_alloc_u32(..., GFP_KERNEL)inbasic_change()andu32_change()).With the boundary running through the middle of one netlink request, a memcg-limited task can still see slab grow faster than memory.current, which is the signal the changelog uses as the reproducer. Was leaving these out intentional, or is the intent to convert them in a follow-up?
Quoted 13 lines
net/sched/cls_basic.c +3
-1@@ -212,9 +212,11 @@ static int basic_change(struct net *net, struct sk_buff *in_skb,if (err) goto errout; fnew->handle = handle; - fnew->pf = alloc_percpu(struct tc_basic_pcnt); + fnew->pf = alloc_percpu_gfp(struct tc_basic_pcnt, GFP_KERNEL_ACCOUNT);if (!fnew->pf) { err = -ENOMEM ; +if (!fold) +idr_remove (&head->handle_idr, fnew->handle); goto errout; }[ ... ]
Quoted 15 lines
net/sched/cls_u32.c +1
-1@@ -1114,15 +1114,16 @@ static int u32_change(struct net *net, struct sk_buff *in_skb, goto erridr; } - n =kzalloc_flex (*n, sel.keys, s->nkeys); + n =kzalloc_flex (*n, sel.keys, s->nkeys,GFP_KERNEL_ACCOUNT );if (n ==NULL ) { err = -ENOBUFS ; goto erridr; }[ ... ]
Jakub Kicinski
maintainerQuoted 3 lines · On Sat, 22 Aug 2026 16:24:21 -0400 Jamal Hadi Salim wrote:
I have a list of follow-up items for this patch - mostly pre-existing issues and others pointing to similar patterns identified by the sashikos.
roger
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:
Quoted 9 lines · On Wed, 19 Aug 2026 10:37:33 -0400 you wrote:
Allocations in the tc classifier
*_change()paths (filter objects, per-CPU counters, and per-filter aux data) use plainGFP_KERNELwithout__GFP_ACCOUNT, allowing unprivileged users to pin kernel memory outside memcg charging. The sharedtcf_exts_init_ex()action array allocation incls_api.cwas also uncharged; this patch closes it along with the per-classifier filter-object/percpu/aux allocations that remain unaccounted.[...]
Here is the summary with links:
- [net] net/sched: account classifier filter allocations to memcg
https://git.kernel.org/netdev/net/c/1beb81947eb4
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html