[net-next,RFC,v2,4/9] bpf, security: Add Checmate security LSM and BPF program type

Message ID	20160829114705.GA20877@ircssh.c.rugged-nimbus-611.internal (mailing list archive)
State	New, archived
Headers	show Return-Path: <linux-security-module-owner@kernel.org> Date: Mon, 29 Aug 2016 04:47:07 -0700 From: Sargun Dhillon <sargun@sargun.me> To: netdev@vger.kernel.org Cc: cgroups@vger.kernel.org, linux-security-module@vger.kernel.org, daniel@iogearbox.net, ast@fb.com Subject: [net-next RFC v2 4/9] bpf, security: Add Checmate security LSM and BPF program type Message-ID: <20160829114705.GA20877@ircssh.c.rugged-nimbus-611.internal> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.21 (2010-09-15) Sender: owner-linux-security-module@vger.kernel.org Precedence: bulk

On Mon, Aug 29, 2016 at 01:01:18PM -0400, Tejun Heo wrote: > Hello, > > On Mon, Aug 29, 2016 at 04:47:07AM -0700, Sargun Dhillon wrote: > > This patch adds a minor LSM, Checmate. Checmate is a flexible programmable, > > extensible minor LSM that's coupled with cgroups and BPF. It is designed to > > enforce container-specific policies. It is also a cgroupv2 controller. By > > itself, it doesn't do very much, but in conjunction with a orchestrator > > complex policies can be installed on the cgroup hierarchy. > > > > These cgroup programs are tied to the kernel ABI version. If one tries > > to load a BPF program compiled against a different kernel version, > > an error will be thrown. > > First of all, please talk with people working on network cgroup bpf > and landlock. I don't think it's a good idea to have N different ways > to implement cgroup-aware bpf mechanism. There can be multiple > consumers but there gotta be a common mechanism instead of several > independent controllers. > I've talked to Daniel Mack, and Alexei. I agree with you that it makes sense not to have an infinite number of these cgroup + bpf + lsm subsystems in the kernel. I think that making sure we don't sacrifice capability is important. > > diff --git a/include/linux/checmate.h b/include/linux/checmate.h > > new file mode 100644 > > index 0000000..4c4db4a > > --- /dev/null > > +++ b/include/linux/checmate.h > > @@ -0,0 +1,108 @@ > > +#ifndef _LINUX_CHECMATE_H_ > > +#define _LINUX_CHECMATE_H_ 1 > > +#include <linux/security.h> > > + > > +enum checmate_hook_num { > > + /* CONFIG_SECURITY_NET hooks */ > > + CHECMATE_HOOK_UNIX_STREAM_CONNECT, > > + CHECMATE_HOOK_UNIX_MAY_SEND, > > + CHECMATE_HOOK_SOCKET_CREATE, > > + CHECMATE_HOOK_SOCKET_POST_CREATE, > > + CHECMATE_HOOK_SOCKET_BIND, > > + CHECMATE_HOOK_SOCKET_CONNECT, > > + CHECMATE_HOOK_SOCKET_LISTEN, > > + CHECMATE_HOOK_SOCKET_ACCEPT, > > + CHECMATE_HOOK_SOCKET_SENDMSG, > > + CHECMATE_HOOK_SOCKET_RECVMSG, > > + CHECMATE_HOOK_SOCKET_GETSOCKNAME, > > + CHECMATE_HOOK_SOCKET_GETPEERNAME, > > + CHECMATE_HOOK_SOCKET_GETSOCKOPT, > > + CHECMATE_HOOK_SOCKET_SETSOCKOPT, > > + CHECMATE_HOOK_SOCKET_SHUTDOWN, > > + CHECMATE_HOOK_SOCKET_SOCK_RCV_SKB, > > + CHECMATE_HOOK_SK_FREE_SECURITY, > > + __CHECMATE_HOOK_MAX, > > +}; > > Do we really want a separate hook for each call? A logical extension > of this would be having a separate hook per syscall which feels kinda > horrible. > It would be a separate hook per LSM hook. Why wouldn't we want a separate bpf hook per lsm hook? I think if one program has to handle them all, the first program would be looking up the hook program in a bpf prog array. If you think it's better to have this logic in the BPF program, that makes sense. I had a version of this patch that allowed you to attach a prog array instead, but I think that it's cleaner attaching a program per lsm hook. In addition, there's a performance impact that comes from these hooks, so I wouldn't want to execute unneccessary code if it's avoidable. The prog array approach also makes stacking filters difficult. If people want multiple filters per hook, the orchestrator would have to rewrite the existing filters to be cooperative. > > +/* CONFIG_SECURITY_NET contexts */ > > +struct checmate_unix_stream_connect_ctx { > > + struct sock *sock; > > + struct sock *other; > > + struct sock *newsk; > > +}; > ... > > +struct checmate_sk_free_security_ctx { > > + struct sock *sk; > > +}; > ... > > +struct checmate_ctx { > > + int hook; > > + union { > > +/* CONFIG_SECURITY_NET contexts */ > > + struct checmate_unix_stream_connect_ctx unix_stream_connect; > > + struct checmate_unix_may_send_ctx unix_may_send; > > + struct checmate_socket_create_ctx socket_create; > > + struct checmate_socket_bind_ctx socket_bind; > > + struct checmate_socket_connect_ctx socket_connect; > > + struct checmate_socket_listen_ctx socket_listen; > > + struct checmate_socket_accept_ctx socket_accept; > > + struct checmate_socket_sendmsg_ctx socket_sendmsg; > > + struct checmate_socket_recvmsg_ctx socket_recvmsg; > > + struct checmate_socket_sock_rcv_skb_ctx socket_sock_rcv_skb; > > + struct checmate_sk_free_security_ctx sk_free_security; > > + }; > > +}; > > I'm not convinced about the approach. It's an approach which pretty > much requires future extensions while being rigid. Not a good > combination. > Do you have an alternative recommendation? Maybe just a set of 5 u64s as the context object along with the hook ID? > > +/* > > Please use /** for function comments. > > > + * checmate_instance_cleanup_rcu - Cleans up a Checmate program instance > > + * @rp: rcu_head pointer to a Checmate instance > > + */ > > +static void checmate_instance_cleanup_rcu(struct rcu_head *rp) > > +{ > > + struct checmate_instance *instance; > > + > > + instance = container_of(rp, struct checmate_instance, rcu); > > + bpf_prog_put(instance->prog); > > + kfree(instance); > > +} > > +static struct cftype checmate_files[] = { > > +#ifdef CONFIG_SECURITY_NETWORK > > + CHECMATE_CFTYPE(CHECMATE_HOOK_UNIX_STREAM_CONNECT, > > + "unix_stream_connect"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_UNIX_MAY_SEND, > > + "unix_may_send"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_CREATE, "socket_create"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_BIND, "socket_bind"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_CONNECT, "socket_connect"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_LISTEN, "socket_listen"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_ACCEPT, "socket_accept"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SENDMSG, "socket_sendmsg"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_RECVMSG, "socket_recvmsg"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SHUTDOWN, "socket_shutdown"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SOCK_RCV_SKB, > > + "socket_sock_rcv_skb"), > > + CHECMATE_CFTYPE(CHECMATE_HOOK_SK_FREE_SECURITY, "sk_free_security"), > > +#endif /* CONFIG_SECURITY_NETWORK */ > > + {} > > +}; > > I don't think this is a good interface approach. > Any other recommendations? > > +struct cgroup_subsys checmate_cgrp_subsys = { > > + .css_alloc = checmate_css_alloc, > > + .css_free = checmate_css_free, > > + .dfl_cftypes = checmate_files, > > +}; > > Unless this is properly delegatable, IOW, it's safe to fully delegate > to a lesser security domain for all operations including program > loading and assignment (I can't see how that'd be the case), making it > an explicit controller doens't work in terms of userland interface. > It's fine for bpf / lsm / whatever to attach to cgroups by extending > struct cgroup itself or implementing an implicit controller but to be > visible as an explicit controller it must be able to follow cgroup > interface rules including delegation. If not, it's best to come > through the interface which enforces the required permission checks > and then talk to cgroup from there. This was also an issue with > network cgroup bpf programs that Daniel Mack is working on. Please > chat with him. > Program assignment is possible by lesser security domains. Program loading is limited to CAP_SYS_ADMIN in init_user_ns. We could make it accessible to CAP_SYS_ADMIN in any userns, but it the reasoning behind this is that Checmate BPF programs can leak kernel pointers. Could we potentially restrict it to only CAP_MAC_OVERRIDE, while still meeting cgroup delegation requirements? Filters which are higher up in the heirarchy will still be enforced during delegation. This was an explicit design, as the "Orchestrator in Orchestrator" use case needs to be supported. > > +static struct cgroup_subsys_state *css_from_sk(struct sock *sk) > > +{ > > + struct cgroup_subsys_state *css; > > + struct cgroup *cgrp; > > + > > + if (!sk_fullsock(sk)) > > + return ERR_PTR(-EINVAL); > > + cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); > > + > > + rcu_read_lock(); > > + do { > > + css = rcu_dereference(cgrp->subsys[checmate_cgrp_id]); > > + if (css) > > + goto out; > > + cgrp = cgroup_parent(cgrp); > > + } while (cgrp); > > + > > +out: > > + rcu_read_unlock(); > > + > > + return css; > > This pattern cannot be right. What protects the returned @css? > Thanks, I'll fix this and keep a reference. > > +static struct cgroup_subsys_state *css_from_current(void) > > +{ > > + struct cgroup_subsys_state *css; > > + > > + if (unlikely(in_interrupt())) > > + return ERR_PTR(-ENOENT); > > + > > + rcu_read_lock(); > > + css = task_css(current, checmate_cgrp_id); > > + rcu_read_unlock(); > > + > > + return css; > > Ditto here. RCU readlock is what was protecting @css. Returning @css > after releasing RCU readlock makes no sense. > > Thanks. > > -- > tejun Thanks for your feedback Tejun. I'll look at making fixes to the above, and touch base with Daniel, and Alexei. -- To unsubscribe from this list: send the line "unsubscribe linux-security-module" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html

diff --git a/include/linux/cgroup_subsys.h b/include/linux/cgroup_subsys.h index 0df0336a..fbb7aa7 100644 --- a/include/linux/cgroup_subsys.h +++ b/include/linux/cgroup_subsys.h @@ -56,6 +56,10 @@ SUBSYS(hugetlb) SUBSYS(pids) #endif +#if IS_ENABLED(CONFIG_SECURITY_CHECMATE) +SUBSYS(checmate) +#endif + /* * The following subsystems are not supported on the default hierarchy. */ diff --git a/include/linux/checmate.h b/include/linux/checmate.h new file mode 100644 index 0000000..4c4db4a --- /dev/null +++ b/include/linux/checmate.h @@ -0,0 +1,108 @@ +#ifndef _LINUX_CHECMATE_H_ +#define _LINUX_CHECMATE_H_ 1 +#include <linux/security.h> + +enum checmate_hook_num { + /* CONFIG_SECURITY_NET hooks */ + CHECMATE_HOOK_UNIX_STREAM_CONNECT, + CHECMATE_HOOK_UNIX_MAY_SEND, + CHECMATE_HOOK_SOCKET_CREATE, + CHECMATE_HOOK_SOCKET_POST_CREATE, + CHECMATE_HOOK_SOCKET_BIND, + CHECMATE_HOOK_SOCKET_CONNECT, + CHECMATE_HOOK_SOCKET_LISTEN, + CHECMATE_HOOK_SOCKET_ACCEPT, + CHECMATE_HOOK_SOCKET_SENDMSG, + CHECMATE_HOOK_SOCKET_RECVMSG, + CHECMATE_HOOK_SOCKET_GETSOCKNAME, + CHECMATE_HOOK_SOCKET_GETPEERNAME, + CHECMATE_HOOK_SOCKET_GETSOCKOPT, + CHECMATE_HOOK_SOCKET_SETSOCKOPT, + CHECMATE_HOOK_SOCKET_SHUTDOWN, + CHECMATE_HOOK_SOCKET_SOCK_RCV_SKB, + CHECMATE_HOOK_SK_FREE_SECURITY, + __CHECMATE_HOOK_MAX, +}; + +/* CONFIG_SECURITY_NET contexts */ +struct checmate_unix_stream_connect_ctx { + struct sock *sock; + struct sock *other; + struct sock *newsk; +}; + +struct checmate_unix_may_send_ctx { + struct socket *sock; + struct socket *other; +}; + +struct checmate_socket_create_ctx { + int family; + int type; + int protocol; + int kern; +}; + +struct checmate_socket_bind_ctx { + struct socket *sock; + struct sockaddr *address; + int addrlen; +}; + +struct checmate_socket_connect_ctx { + struct socket *sock; + struct sockaddr *address; + int addrlen; +}; + +struct checmate_socket_listen_ctx { + struct socket *sock; + int backlog; +}; + +struct checmate_socket_accept_ctx { + struct socket *sock; + struct socket *newsock; +}; + +struct checmate_socket_sendmsg_ctx { + struct socket *sock; + struct msghdr *msg; + int size; +}; + +struct checmate_socket_recvmsg_ctx { + struct socket *sock; + struct msghdr *msg; + int size; + int flags; +}; + +struct checmate_socket_sock_rcv_skb_ctx { + struct sock *sk; + struct sk_buff *skb; +}; + +struct checmate_sk_free_security_ctx { + struct sock *sk; +}; + +struct checmate_ctx { + int hook; + union { +/* CONFIG_SECURITY_NET contexts */ + struct checmate_unix_stream_connect_ctx unix_stream_connect; + struct checmate_unix_may_send_ctx unix_may_send; + struct checmate_socket_create_ctx socket_create; + struct checmate_socket_bind_ctx socket_bind; + struct checmate_socket_connect_ctx socket_connect; + struct checmate_socket_listen_ctx socket_listen; + struct checmate_socket_accept_ctx socket_accept; + struct checmate_socket_sendmsg_ctx socket_sendmsg; + struct checmate_socket_recvmsg_ctx socket_recvmsg; + struct checmate_socket_sock_rcv_skb_ctx socket_sock_rcv_skb; + struct checmate_sk_free_security_ctx sk_free_security; + }; +}; + +#endif /* _LINUX_CHECMATE_H_ */ diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index e4c5a1b..91bc92f 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -95,6 +95,7 @@ enum bpf_prog_type { BPF_PROG_TYPE_SCHED_ACT, BPF_PROG_TYPE_TRACEPOINT, BPF_PROG_TYPE_XDP, + BPF_PROG_TYPE_CHECMATE, }; #define BPF_PSEUDO_MAP_FD 1 diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index 228f962..6f4f7b0 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -741,7 +741,7 @@ static int bpf_prog_load(union bpf_attr *attr) if (attr->insn_cnt >= BPF_MAXINSNS) return -EINVAL; - if (type == BPF_PROG_TYPE_KPROBE && + if ((type == BPF_PROG_TYPE_KPROBE || type == BPF_PROG_TYPE_CHECMATE) && attr->kern_version != LINUX_VERSION_CODE) return -EINVAL; diff --git a/security/Kconfig b/security/Kconfig index df28f2b..c819539 100644 --- a/security/Kconfig +++ b/security/Kconfig @@ -152,6 +152,7 @@ source security/tomoyo/Kconfig source security/apparmor/Kconfig source security/loadpin/Kconfig source security/yama/Kconfig +source security/checmate/Kconfig source security/integrity/Kconfig diff --git a/security/Makefile b/security/Makefile index f2d71cd..6cc3342 100644 --- a/security/Makefile +++ b/security/Makefile @@ -8,6 +8,7 @@ subdir-$(CONFIG_SECURITY_SMACK) += smack subdir-$(CONFIG_SECURITY_TOMOYO) += tomoyo subdir-$(CONFIG_SECURITY_APPARMOR) += apparmor subdir-$(CONFIG_SECURITY_YAMA) += yama +subdir-$(CONFIG_SECURITY_CHECMATE) += checmate subdir-$(CONFIG_SECURITY_LOADPIN) += loadpin # always enable default capabilities @@ -25,6 +26,7 @@ obj-$(CONFIG_SECURITY_APPARMOR) += apparmor/ obj-$(CONFIG_SECURITY_YAMA) += yama/ obj-$(CONFIG_SECURITY_LOADPIN) += loadpin/ obj-$(CONFIG_CGROUP_DEVICE) += device_cgroup.o +obj-$(CONFIG_SECURITY_CHECMATE) += checmate/ # Object integrity file lists subdir-$(CONFIG_INTEGRITY) += integrity diff --git a/security/checmate/Kconfig b/security/checmate/Kconfig new file mode 100644 index 0000000..9dc76d1 --- /dev/null +++ b/security/checmate/Kconfig @@ -0,0 +1,11 @@ +config SECURITY_CHECMATE + bool "Checmate support" + depends on SECURITY && SOCK_CGROUP_DATA + default n + help + This selects Checmate, which is an LSM that works in conjuncion with + cgroups and BPF in order to provide programmable, flexible, and + extensible security policies. Further information can be found in + Documentation/security/Checmate.txt + + If you are unsure how to answer this question, answer N. diff --git a/security/checmate/Makefile b/security/checmate/Makefile new file mode 100644 index 0000000..c676773 --- /dev/null +++ b/security/checmate/Makefile @@ -0,0 +1,3 @@ +obj-$(CONFIG_SECURITY_CHECMATE) := checmate.o + +checmate-y := checmate_bpf.o checmate_lsm.o diff --git a/security/checmate/checmate_bpf.c b/security/checmate/checmate_bpf.c new file mode 100644 index 0000000..001225c --- /dev/null +++ b/security/checmate/checmate_bpf.c @@ -0,0 +1,68 @@ +/* + * Checmate Linux Security Module + * + * Copyright (C) 2016 Sargun Dhillon <sargun@sargun.me> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2, as + * published by the Free Software Foundation. + * + */ + +#include <linux/bpf.h> +#include <linux/checmate.h> + +static const struct bpf_func_proto * +checmate_prog_func_proto(enum bpf_func_id func_id) +{ + switch (func_id) { + case BPF_FUNC_map_lookup_elem: + return &bpf_map_lookup_elem_proto; + case BPF_FUNC_map_update_elem: + return &bpf_map_update_elem_proto; + case BPF_FUNC_map_delete_elem: + return &bpf_map_delete_elem_proto; + case BPF_FUNC_probe_read: + return &bpf_probe_read_proto; + case BPF_FUNC_tail_call: + return &bpf_tail_call_proto; + case BPF_FUNC_get_current_pid_tgid: + return &bpf_get_current_pid_tgid_proto; + case BPF_FUNC_get_current_task: + return &bpf_get_current_task_proto; + case BPF_FUNC_get_current_uid_gid: + return &bpf_get_current_uid_gid_proto; + case BPF_FUNC_get_current_comm: + return &bpf_get_current_comm_proto; + case BPF_FUNC_trace_printk: + return bpf_get_trace_printk_proto(); + default: + return NULL; + } +} + +static bool checmate_prog_is_valid_access(int off, int size, + enum bpf_access_type type, + enum bpf_reg_type *reg_type) +{ + if (type != BPF_READ) + return false; + if (off < 0 || off >= sizeof(struct checmate_ctx)) + return false; + return true; +} + +static const struct bpf_verifier_ops checmate_prog_ops = { + .get_func_proto = checmate_prog_func_proto, + .is_valid_access = checmate_prog_is_valid_access, +}; + +static struct bpf_prog_type_list checmate_tl = { + .ops = &checmate_prog_ops, + .type = BPF_PROG_TYPE_CHECMATE, +}; + +void register_checmate_prog_ops(void) +{ + bpf_register_prog_type(&checmate_tl); +} diff --git a/security/checmate/checmate_lsm.c b/security/checmate/checmate_lsm.c new file mode 100644 index 0000000..ef8514d --- /dev/null +++ b/security/checmate/checmate_lsm.c @@ -0,0 +1,610 @@ +/* + * Checmate Linux Security Module + * + * Copyright (C) 2016 Sargun Dhillon <sargun@sargun.me> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2, as + * published by the Free Software Foundation. + * + */ + +#include <linux/prctl.h> +#include <linux/checmate.h> +#include <linux/lsm_hooks.h> +#include <linux/mutex.h> +#include <linux/bpf.h> +#include <linux/filter.h> +#include <linux/cgroup.h> +#include <net/sock.h> +#include <net/request_sock.h> + +#define MAX_CHECMATE_INSTANCES 32 + +/* Global mutex for any Checmate hook manipulation operations */ +DEFINE_MUTEX(checmate_mutex); + +#define CHECMATE_CFTYPE(HOOK, NAME) \ + { \ + .name = NAME, \ + .private = HOOK, \ + .read_u64 = checmate_read_u64, \ + .write_s64 = checmate_write_s64, \ + .flags = CFTYPE_NOT_ON_ROOT \ + } + +extern void register_checmate_prog_ops(void); + +struct checmate_instance { + struct list_head list; + struct rcu_head rcu; + struct bpf_prog *prog; +}; + +struct checmate_hook { + struct list_head instances; + int count; +}; + +struct checmate_css { + struct cgroup_subsys_state css; + struct checmate_hook hooks[__CHECMATE_HOOK_MAX]; +}; + +static struct checmate_css *css_checmate(struct cgroup_subsys_state *css) +{ + return container_of(css, struct checmate_css, css); +} + +static struct checmate_css *parent_checmate(struct checmate_css *checmate) +{ + return css_checmate(checmate->css.parent); +} + +static struct cgroup_subsys_state * +checmate_css_alloc(struct cgroup_subsys_state *parent) +{ + struct checmate_css *checmate; + int i; + + checmate = kzalloc(sizeof(*checmate), GFP_KERNEL); + if (!checmate) + return ERR_PTR(-ENOMEM); + + for (i = 0; i < ARRAY_SIZE(checmate->hooks); i++) + INIT_LIST_HEAD(&checmate->hooks[i].instances); + + return &checmate->css; +} + +/* + * checmate_hook_free - Deallocate, and release resources for a given hook + * @hook: The hook + * + * Always succeeds. Only to be used when hook is out of use, and therefore + * doesn't use the RCU mechanism to cleanup he hook. Only use it for + * retirement of a hook. + */ +static void checmate_hook_free(struct checmate_hook *hook) +{ + struct checmate_instance *instance, *next; + + list_for_each_entry_safe(instance, next, &hook->instances, list) { + list_del(&instance->list); + bpf_prog_put(instance->prog); + kfree(instance); + } +} + +/* + * checmate_css_free - Callback for css_free + * @css: The cgroup_subsys_state to be freed + */ +static void checmate_css_free(struct cgroup_subsys_state *css) +{ + struct checmate_css *checmate = css_checmate(css); + int i; + + mutex_lock(&checmate_mutex); + for (i = 0; i < ARRAY_SIZE(checmate->hooks); i++) + checmate_hook_free(&checmate->hooks[i]); + + kfree(checmate); + mutex_unlock(&checmate_mutex); +} + +/* + * checmate_instance_add - Add BPF program instance to a Checmate hook + * @hook: The hook + * @prog: A checmate BPF program + * + * Checks if the program is already part of the hook, and only adds new + * programs. + * + * Returns 0 on success. -errno on failure. + * + * Requires that the Checmate mutex is held during the operation. + */ +static int checmate_instance_add(struct checmate_hook *hook, + struct bpf_prog *prog) +{ + struct checmate_instance *instance; + int rc = 0; + + if (hook->count >= MAX_CHECMATE_INSTANCES) + return -ENOSPC; + + list_for_each_entry(instance, &hook->instances, list) { + if (instance->prog == prog) { + bpf_prog_put(prog); + rc = -EEXIST; + goto err; + } + } + + instance = kmalloc(sizeof(*instance), GFP_KERNEL); + if (!instance) { + rc = -ENOMEM; + goto err; + } + + instance->prog = prog; + list_add_tail_rcu(&instance->list, &hook->instances); + hook->count++; + return rc; + +err: + bpf_prog_put(prog); + return rc; +} + +/* + * checmate_instance_cleanup_rcu - Cleans up a Checmate program instance + * @rp: rcu_head pointer to a Checmate instance + */ +static void checmate_instance_cleanup_rcu(struct rcu_head *rp) +{ + struct checmate_instance *instance; + + instance = container_of(rp, struct checmate_instance, rcu); + bpf_prog_put(instance->prog); + kfree(instance); +} + +/* + * checmate_instance_remove - Remove Checmate program instance from a hook + * @hook: The hook + * @prog: A Checmate BPF program referred to by the instance. + * + * Returns 0 on success. -errno on failure. + * + * Requires that the Checmate mutex is held during the operation. + */ +static int checmate_instance_remove(struct checmate_hook *hook, + struct bpf_prog *prog) +{ + struct checmate_instance *instance, *next; + int rc = -ENOENT; + + list_for_each_entry_safe(instance, next, &hook->instances, list) { + if (instance->prog == prog) { + list_del_rcu(&instance->list); + call_rcu(&instance->rcu, checmate_instance_cleanup_rcu); + rc = 0; + hook->count--; + break; + } + } + bpf_prog_put(prog); + + return rc; +} + +/* + * checmate_hook_reset - Remove all program instances from a Checmate hook + * @hook: The hook + * + * Always succeeds. + * + * Requires that the Checmate mutex is held during the operation. + */ +static void checmate_hook_reset(struct checmate_hook *hook) +{ + struct checmate_instance *instance, *next; + + list_for_each_entry_safe(instance, next, &hook->instances, list) { + list_del_rcu(&instance->list); + call_rcu(&instance->rcu, checmate_instance_cleanup_rcu); + } + hook->count = 0; +} + +/* + * checmate_write_s64 - Handle a write to the checmate cgroup control file + * @css: The given cgroup state that own's the hook + * @cft: The given cftype that is being referenced, used to get the hook id. + * @val: The bpf program fd that is involved in the operation, or 0. + * + * val == 0: Reset all programs in hook. + * val > 0: Add the given program. + * val < 0: Remove the given program. + * + * Returns 0 on success. -errno on failure. + */ +static int checmate_write_s64(struct cgroup_subsys_state *css, + struct cftype *cft, s64 val) +{ + struct checmate_css *checmate = css_checmate(css); + struct checmate_hook *hook; + struct bpf_prog *prog; + int rc = 0; + + hook = &checmate->hooks[cft->private]; + mutex_lock(&checmate_mutex); + if (val == 0) { + checmate_hook_reset(hook); + goto out; + } + + /* If we're not resetting, we have to load, and check the program */ + prog = bpf_prog_get_type(abs(val), BPF_PROG_TYPE_CHECMATE); + if (IS_ERR(prog)) + return PTR_ERR(prog); + + if (val > 0) + rc = checmate_instance_add(hook, prog); + else + rc = checmate_instance_remove(hook, prog); + +out: + mutex_unlock(&checmate_mutex); + return rc; +} + +/* + * checmate_read_u64 - Read the number of programs loaded into a given hook + * @css: The given cgroup state that own's the hook + * @cft: The given cftype that is being referenced, used to get the hook id. + * + * + * Returns number of programs loaded into hook. Always succeeds. + */ +static u64 checmate_read_u64(struct cgroup_subsys_state *css, + struct cftype *cft) +{ + struct checmate_css *checmate = css_checmate(css); + struct checmate_hook *hook; + + hook = &checmate->hooks[cft->private]; + return hook->count; +} + +static struct cftype checmate_files[] = { +#ifdef CONFIG_SECURITY_NETWORK + CHECMATE_CFTYPE(CHECMATE_HOOK_UNIX_STREAM_CONNECT, + "unix_stream_connect"), + CHECMATE_CFTYPE(CHECMATE_HOOK_UNIX_MAY_SEND, + "unix_may_send"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_CREATE, "socket_create"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_BIND, "socket_bind"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_CONNECT, "socket_connect"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_LISTEN, "socket_listen"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_ACCEPT, "socket_accept"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SENDMSG, "socket_sendmsg"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_RECVMSG, "socket_recvmsg"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SHUTDOWN, "socket_shutdown"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SOCKET_SOCK_RCV_SKB, + "socket_sock_rcv_skb"), + CHECMATE_CFTYPE(CHECMATE_HOOK_SK_FREE_SECURITY, "sk_free_security"), +#endif /* CONFIG_SECURITY_NETWORK */ + {} +}; + +struct cgroup_subsys checmate_cgrp_subsys = { + .css_alloc = checmate_css_alloc, + .css_free = checmate_css_free, + .dfl_cftypes = checmate_files, +}; + +/* + * check_checmate_filters - Run all the BPF programs associated with a hook + * @css: A pointer to the Checmate css + * @ctx: A pointer to the Checmate ctx + * + * Return 0 on success, on first hook returning non-0, the error is returned + * to the caller. + */ +static int checmate_check_filters(struct checmate_css *checmate, + struct checmate_ctx *ctx) +{ + struct checmate_instance *instance; + struct checmate_hook *hook; + int rc = 0; + + hook = &checmate->hooks[ctx->hook]; + + rcu_read_lock(); + list_for_each_entry_rcu(instance, &hook->instances, list) { + rc = BPF_PROG_RUN(instance->prog, (void *)ctx); + if (rc) + break; + } + rcu_read_unlock(); + + return rc; +} + +/* + * call_bpf_int_hook - Walk the cgroup hierarchy, running filters up the chain + * @hook: The Hook ID + * @css: A pointer to the Checmate css + * @cgrp: A pointer to the cgroup we're in, may be null or err + * + * Return 0 on success, on first hook erroring, the error is returned + * to the caller. + * + * Requires that the context struct is populated before passing, but + * the actual ctx->hook number is set by the function. + */ +static int call_bpf_int_hook(int hook, struct cgroup_subsys_state *css, + struct checmate_ctx *ctx) +{ + struct checmate_css *checmate; + int rc = 0; + + /* Fail open if we can't find the css / cgroup */ + if (unlikely(IS_ERR_OR_NULL(css))) + goto out; + + ctx->hook = hook; + + for (checmate = css_checmate(css); parent_checmate(checmate); + checmate = parent_checmate(checmate)) { + rc = checmate_check_filters(checmate, ctx); + if (rc) + break; + } + +out: + return rc; +} + +/* + * call_bpf_void_hook - Run all the BPF programs associated with a hook + * Wrapper around call_bpf_int_hook. + */ +static void call_bpf_void_hook(int hook, struct cgroup_subsys_state *css, + struct checmate_ctx *ctx) +{ + call_bpf_int_hook(hook, css, ctx); +} + +/* + * css_from_sk - Get the Checmate CSS for an sk + * @sk: The struct sock we're trying to get the CSS for. + * + * Return Checmate CSS on success, or NULL / ERR_PTR on failure. It will try + * to return the effective CSS. + */ +static struct cgroup_subsys_state *css_from_sk(struct sock *sk) +{ + struct cgroup_subsys_state *css; + struct cgroup *cgrp; + + if (!sk_fullsock(sk)) + return ERR_PTR(-EINVAL); + cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); + + rcu_read_lock(); + do { + css = rcu_dereference(cgrp->subsys[checmate_cgrp_id]); + if (css) + goto out; + cgrp = cgroup_parent(cgrp); + } while (cgrp); + +out: + rcu_read_unlock(); + + return css; +} + +/* + * css_from_sock - Get the Checmate CSS for a socket + * @sock: The struct socket we're trying to get the CSS for. + * + * Return CSS on success. NULL / ERR_PTR on failure. It's a wrapper around + * css_from_sk. + */ +static struct cgroup_subsys_state *css_from_sock(struct socket *sock) +{ + struct sock *sk; + + sk = sock->sk; + if (!sk) + return ERR_PTR(-ENOENT); + + return css_from_sk(sk); +} + +/* + * css_from_sock - Get the checmate CSS for the current task context. + * + * Return CSS success on success. ERR_PTR on failure. It checks to see if it's + * being called from an interrupt as well. + */ +static struct cgroup_subsys_state *css_from_current(void) +{ + struct cgroup_subsys_state *css; + + if (unlikely(in_interrupt())) + return ERR_PTR(-ENOENT); + + rcu_read_lock(); + css = task_css(current, checmate_cgrp_id); + rcu_read_unlock(); + + return css; +} + +/* Checmate hooks */ +#ifdef CONFIG_SECURITY_NETWORK +static int checmate_unix_stream_connect(struct sock *sock, struct sock *other, + struct sock *newsk) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sk(sock); + ctx.unix_stream_connect.sock = sock; + ctx.unix_stream_connect.other = other; + ctx.unix_stream_connect.newsk = newsk; + return call_bpf_int_hook(CHECMATE_HOOK_UNIX_STREAM_CONNECT, css, &ctx); +} + +static int checmate_unix_may_send(struct socket *sock, struct socket *other) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.unix_may_send.sock = sock; + ctx.unix_may_send.other = other; + return call_bpf_int_hook(CHECMATE_HOOK_UNIX_MAY_SEND, css, &ctx); +} + +static int checmate_socket_create(int family, int type, int protocol, int kern) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_current(); + ctx.socket_create.family = family; + ctx.socket_create.type = type; + ctx.socket_create.protocol = protocol; + ctx.socket_create.kern = kern; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_CREATE, css, &ctx); +} + +static int checmate_socket_bind(struct socket *sock, struct sockaddr *address, + int addrlen) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_bind.sock = sock; + ctx.socket_bind.address = address; + ctx.socket_bind.addrlen = addrlen; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_BIND, css, &ctx); +} + +static int checmate_socket_connect(struct socket *sock, + struct sockaddr *address, int addrlen) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_connect.sock = sock; + ctx.socket_connect.address = address; + ctx.socket_connect.addrlen = addrlen; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_CONNECT, css, &ctx); +} + +static int checmate_socket_listen(struct socket *sock, int backlog) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_listen.sock = sock; + ctx.socket_listen.backlog = backlog; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_LISTEN, css, &ctx); +} + +static int checmate_socket_accept(struct socket *sock, struct socket *newsock) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_accept.sock = sock; + ctx.socket_accept.newsock = newsock; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_ACCEPT, css, &ctx); +} + +static int checmate_socket_sendmsg(struct socket *sock, struct msghdr *msg, + int size) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_sendmsg.sock = sock; + ctx.socket_sendmsg.msg = msg; + ctx.socket_sendmsg.size = size; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_SENDMSG, css, &ctx); +} + +static int checmate_socket_recvmsg(struct socket *sock, struct msghdr *msg, + int size, int flags) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sock(sock); + ctx.socket_recvmsg.sock = sock; + ctx.socket_recvmsg.msg = msg; + ctx.socket_recvmsg.size = size; + ctx.socket_recvmsg.flags = flags; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_RECVMSG, css, &ctx); +} + +static int checmate_socket_sock_rcv_skb(struct sock *sk, struct sk_buff *skb) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sk(sk); + ctx.socket_sock_rcv_skb.sk = sk; + ctx.socket_sock_rcv_skb.skb = skb; + return call_bpf_int_hook(CHECMATE_HOOK_SOCKET_SOCK_RCV_SKB, css, &ctx); +} + +static void checmate_sk_free_security(struct sock *sk) +{ + struct cgroup_subsys_state *css; + struct checmate_ctx ctx; + + css = css_from_sk(sk); + ctx.sk_free_security.sk = sk; + return call_bpf_void_hook(CHECMATE_HOOK_SK_FREE_SECURITY, css, &ctx); +} + +#endif /* CONFIG_SECURITY_NETWORK */ + +static struct security_hook_list checmate_hooks[] = { +#ifdef CONFIG_SECURITY_NETWORK + LSM_HOOK_INIT(unix_stream_connect, checmate_unix_stream_connect), + LSM_HOOK_INIT(unix_may_send, checmate_unix_may_send), + LSM_HOOK_INIT(socket_create, checmate_socket_create), + LSM_HOOK_INIT(socket_bind, checmate_socket_bind), + LSM_HOOK_INIT(socket_connect, checmate_socket_connect), + LSM_HOOK_INIT(socket_listen, checmate_socket_listen), + LSM_HOOK_INIT(socket_accept, checmate_socket_accept), + LSM_HOOK_INIT(socket_sendmsg, checmate_socket_sendmsg), + LSM_HOOK_INIT(socket_recvmsg, checmate_socket_recvmsg), + LSM_HOOK_INIT(socket_sock_rcv_skb, checmate_socket_sock_rcv_skb), + LSM_HOOK_INIT(sk_free_security, checmate_sk_free_security), +#endif /* CONFIG_SECURITY_NETWORK */ +}; + +static int __init checmate_setup(void) +{ + pr_info("Checmate activating.\n"); + register_checmate_prog_ops(); + security_add_hooks(checmate_hooks, ARRAY_SIZE(checmate_hooks)); + return 0; +} +late_initcall(checmate_setup);

[net-next,RFC,v2,4/9] bpf, security: Add Checmate security LSM and BPF program type

Commit Message

Comments

Patch