Add a security hook that allows a LSM to be notified when a task gets a fatal signal. This patch is a previous step on the way to compute the task crash period by the "brute" LSM (linux security module to detect and mitigate fork brute force attack against vulnerable userspace processes).

Signed-off-by: John Wood john.wood@gmx.com Reviewed-by: Kees Cook keescook@chromium.org --- include/linux/lsm_hook_defs.h | 1 + include/linux/lsm_hooks.h | 4 ++++ include/linux/security.h | 4 ++++ kernel/signal.c | 1 + security/security.c | 5 +++++ 5 files changed, 15 insertions(+)

diff --git a/include/linux/lsm_hook_defs.h b/include/linux/lsm_hook_defs.h index 04c01794de83..e28468e84300 100644 --- a/include/linux/lsm_hook_defs.h +++ b/include/linux/lsm_hook_defs.h @@ -225,6 +225,7 @@ LSM_HOOK(int, -ENOSYS, task_prctl, int option, unsigned long arg2, unsigned long arg3, unsigned long arg4, unsigned long arg5) LSM_HOOK(void, LSM_RET_VOID, task_to_inode, struct task_struct *p, struct inode *inode) +LSM_HOOK(void, LSM_RET_VOID, task_fatal_signal, const kernel_siginfo_t *siginfo) LSM_HOOK(int, 0, ipc_permission, struct kern_ipc_perm *ipcp, short flag) LSM_HOOK(void, LSM_RET_VOID, ipc_getsecid, struct kern_ipc_perm *ipcp, u32 *secid) diff --git a/include/linux/lsm_hooks.h b/include/linux/lsm_hooks.h index 5c4c5c0602cb..fc8bef0f15d9 100644 --- a/include/linux/lsm_hooks.h +++ b/include/linux/lsm_hooks.h @@ -799,6 +799,10 @@ * security attributes, e.g. for /proc/pid inodes. * @p contains the task_struct for the task. * @inode contains the inode structure for the inode. + * @task_fatal_signal: + * This hook allows security modules to be notified when a task gets a + * fatal signal. + * @siginfo contains the signal information. * * Security hooks for Netlink messaging. * diff --git a/include/linux/security.h b/include/linux/security.h index 06f7c50ce77f..609c76c6c764 100644 --- a/include/linux/security.h +++ b/include/linux/security.h @@ -433,6 +433,7 @@ int security_task_kill(struct task_struct *p, struct kernel_siginfo *info, int security_task_prctl(int option, unsigned long arg2, unsigned long arg3, unsigned long arg4, unsigned long arg5); void security_task_to_inode(struct task_struct *p, struct inode *inode); +void security_task_fatal_signal(const kernel_siginfo_t *siginfo); int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag); void security_ipc_getsecid(struct kern_ipc_perm *ipcp, u32 *secid); int security_msg_msg_alloc(struct msg_msg *msg); @@ -1183,6 +1184,9 @@ static inline int security_task_prctl(int option, unsigned long arg2, static inline void security_task_to_inode(struct task_struct *p, struct inode *inode) { }

+static inline void security_task_fatal_signal(const kernel_siginfo_t *siginfo) +{ } + static inline int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag) { diff --git a/kernel/signal.c b/kernel/signal.c index f7c6ffcbd044..4380763b3d8d 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -2804,6 +2804,7 @@ bool get_signal(struct ksignal *ksig) /* * Anything else is fatal, maybe with a core dump. */ + security_task_fatal_signal(&ksig->info); current->flags |= PF_SIGNALED;

if (sig_kernel_coredump(signr)) { diff --git a/security/security.c b/security/security.c index b38155b2de83..208e3e7d4284 100644 --- a/security/security.c +++ b/security/security.c @@ -1891,6 +1891,11 @@ void security_task_to_inode(struct task_struct *p, struct inode *inode) call_void_hook(task_to_inode, p, inode); }

+void security_task_fatal_signal(const kernel_siginfo_t *siginfo) +{ + call_void_hook(task_fatal_signal, siginfo); +} + int security_ipc_permission(struct kern_ipc_perm *ipcp, short flag) { return call_int_hook(ipc_permission, 0, ipcp, flag); -- 2.25.1

For a correct management of a fork brute force attack it is necessary to track all the information related to the application crashes. To do so, use the extended attributes (xattr) of the executable files and define a statistical data structure to hold all the necessary information shared by all the fork hierarchy processes. This info is the number of crashes, the last crash timestamp and the crash period's moving average.

The same can be achieved using a pointer to the fork hierarchy statistical data held by the task_struct structure. But this has an important drawback: a brute force attack that happens through the execve system call losts the faults info since these statistics are freed when the fork hierarchy disappears. Using this method makes not possible to manage this attack type that can be successfully treated using extended attributes.

Also, to avoid false positives during the attack detection it is necessary to narrow the possible cases. So, only the following scenarios are taken into account:

1.- Launching (fork()/exec()) a setuid/setgid process repeatedly until a desirable memory layout is got (e.g. Stack Clash). 2.- Connecting to an exec()ing network daemon (e.g. xinetd) repeatedly until a desirable memory layout is got (e.g. what CTFs do for simple network service). 3.- Launching processes without exec() (e.g. Android Zygote) and exposing state to attack a sibling. 4.- Connecting to a fork()ing network daemon (e.g. apache) repeatedly until the previously shared memory layout of all the other children is exposed (e.g. kind of related to HeartBleed).

In each case, a privilege boundary has been crossed:

Case 1: setuid/setgid process Case 2: network to local Case 3: privilege changes Case 4: network to local

To mark that a privilege boundary has been crossed it is only necessary to create a new stats for the executable file via the extended attribute and only if it has no previous statistical data. This is done using four different LSM hooks, one per privilege boundary:

setuid/setgid process --> bprm_creds_from_file hook (based on secureexec flag). network to local -------> socket_accept hook (taking into account only external connections). privilege changes ------> task_fix_setuid and task_fix_setgid hooks.

To detect a brute force attack it is necessary that the executable file statistics be updated in every fatal crash and the most important data to update is the application crash period. To do so, use the new "task_fatal_signal" LSM hook added in a previous step.

The application crash period must be a value that is not prone to change due to spurious data and follows the real crash period. So, to compute it, the exponential moving average (EMA) is used.

Based on the updated statistics two different attacks can be handled. A slow brute force attack that is detected if the maximum number of faults per fork hierarchy is reached and a fast brute force attack that is detected if the application crash period falls below a certain threshold.

Moreover, only the signals delivered by the kernel are taken into account with the exception of the SIGABRT signal since the latter is used by glibc for stack canary, malloc, etc failures, which may indicate that a mitigation has been triggered.

Signed-off-by: John Wood john.wood@gmx.com --- include/uapi/linux/xattr.h | 3 + security/brute/brute.c | 500 +++++++++++++++++++++++++++++++++++++ 2 files changed, 503 insertions(+)

diff --git a/include/uapi/linux/xattr.h b/include/uapi/linux/xattr.h index 9463db2dfa9d..ce1c8497dceb 100644 --- a/include/uapi/linux/xattr.h +++ b/include/uapi/linux/xattr.h @@ -76,6 +76,9 @@ #define XATTR_CAPS_SUFFIX "capability" #define XATTR_NAME_CAPS XATTR_SECURITY_PREFIX XATTR_CAPS_SUFFIX

+#define XATTR_BRUTE_SUFFIX "brute" +#define XATTR_NAME_BRUTE XATTR_SECURITY_PREFIX XATTR_BRUTE_SUFFIX + #define XATTR_POSIX_ACL_ACCESS "posix_acl_access" #define XATTR_NAME_POSIX_ACL_ACCESS XATTR_SYSTEM_PREFIX XATTR_POSIX_ACL_ACCESS #define XATTR_POSIX_ACL_DEFAULT "posix_acl_default" diff --git a/security/brute/brute.c b/security/brute/brute.c index 0edb89a58ab0..03bebfd1ed1f 100644 --- a/security/brute/brute.c +++ b/security/brute/brute.c @@ -2,8 +2,85 @@

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

+#include <linux/binfmts.h> #include <linux/lsm_hooks.h> #include <linux/sysctl.h> +#include <linux/xattr.h> +#include <net/ipv6.h> +#include <net/sock.h> + +/** + * struct brute_stats - Fork brute force attack statistics. + * @faults: Number of crashes. + * @nsecs: Last crash timestamp as the number of nanoseconds in the + * International Atomic Time (TAI) reference. + * @period: Crash period's moving average. + * @flags: Statistics flags as a whole. + * @not_allowed: Not allowed executable file flag. + * @unused: Remaining unused flags. + * + * This structure holds the statistical data shared by all the fork hierarchy + * processes. + */ +struct brute_stats { + u32 faults; + u64 nsecs; + u64 period; + union { + u8 flags; + struct { + u8 not_allowed : 1; + u8 unused : 7; + }; + }; +}; + +/** + * struct brute_raw_stats - Raw fork brute force attack statistics. + * @faults: Number of crashes. + * @nsecs: Last crash timestamp as the number of nanoseconds in the + * International Atomic Time (TAI) reference. + * @period: Crash period's moving average. + * @flags: Statistics flags. + * + * This structure holds the statistical data on disk as an extended attribute. + * Since the filesystems on which extended attributes are stored might also be + * used on architectures with a different byte order and machine word size, care + * should be taken to store attribute values in an architecture-independent + * format. + */ +struct brute_raw_stats { + __le32 faults; + __le64 nsecs; + __le64 period; + u8 flags; +} __packed; + +/** + * brute_get_current_exe_file() - Get the current task's executable file. + * + * Since all the kernel threads associated with a task share the same executable + * file, get the thread group leader's executable file. + * + * Context: The file must be released via fput(). + * Return: NULL if the current task has no associated executable file. A pointer + * to the executable file otherwise. + */ +static struct file *brute_get_current_exe_file(void) +{ + struct task_struct *task = current; + struct file *exe_file; + + rcu_read_lock(); + if (!thread_group_leader(task)) + task = rcu_dereference(task->group_leader); + get_task_struct(task); + rcu_read_unlock(); + + exe_file = get_task_exe_file(task); + put_task_struct(task); + return exe_file; +}

/** * DOC: brute_ema_weight_numerator @@ -19,6 +96,18 @@ static unsigned int brute_ema_weight_numerator __read_mostly = 7; */ static unsigned int brute_ema_weight_denominator __read_mostly = 10;

+/** + * brute_mul_by_ema_weight() - Multiply by EMA weight. + * @value: Value to multiply by EMA weight. + * + * Return: The result of the multiplication operation. + */ +static inline u64 brute_mul_by_ema_weight(u64 value) +{ + return mul_u64_u32_div(value, brute_ema_weight_numerator, + brute_ema_weight_denominator); +} + /** * DOC: brute_max_faults * @@ -30,6 +119,56 @@ static unsigned int brute_ema_weight_denominator __read_mostly = 10; */ static unsigned int brute_max_faults __read_mostly = 200;

+/** + * brute_update_crash_period() - Update the application crash period. + * @stats: Statistics that hold the application crash period to update. Cannot + * be NULL. + * + * The application crash period must be a value that is not prone to change due + * to spurious data and follows the real crash period. So, to compute it, the + * exponential moving average (EMA) is used. + * + * This kind of average defines a weight (between 0 and 1) for the new value to + * add and applies the remainder of the weight to the current average value. + * This way, some spurious data will not excessively modify the average and only + * if the new values are persistent, the moving average will tend towards them. + * + * Mathematically the application crash period's EMA can be expressed as + * follows: + * + * period_ema = period * weight + period_ema * (1 - weight) + * + * If the operations are applied: + * + * period_ema = period * weight + period_ema - period_ema * weight + * + * If the operands are ordered: + * + * period_ema = period_ema - period_ema * weight + period * weight + * + * Finally, this formula can be written as follows: + * + * period_ema -= period_ema * weight; + * period_ema += period * weight; + */ +static void brute_update_crash_period(struct brute_stats *stats) +{ + u64 current_period; + u64 now = ktime_get_clocktai_ns(); + + if (stats->faults >= (u32)brute_max_faults) + return; + + if (stats->nsecs) { + current_period = now > stats->nsecs ? now - stats->nsecs : 0; + stats->period -= brute_mul_by_ema_weight(stats->period); + stats->period += brute_mul_by_ema_weight(current_period); + } + + stats->nsecs = now; + stats->faults += 1; +} + /** * DOC: brute_min_faults * @@ -51,6 +190,365 @@ static unsigned int brute_min_faults __read_mostly = 5; */ static unsigned int brute_crash_period_threshold __read_mostly = 30;

+/** + * brute_attack_running() - Test if a brute force attack is happening. + * @stats: Statistical data shared by all the fork hierarchy processes. Cannot + * be NULL. + * + * The decision if a brute force attack is running is based on the statistical + * data shared by all the fork hierarchy processes. + * + * There are two types of brute force attacks that can be detected using the + * statistical data. The first one is a slow brute force attack that is detected + * if the maximum number of faults per fork hierarchy is reached. The second + * type is a fast brute force attack that is detected if the application crash + * period falls below a certain threshold. + * + * Moreover, it is important to note that no attacks will be detected until a + * minimum number of faults have occurred. This allows to have a trend in the + * crash period when the EMA is used. + * + * Return: True if a brute force attack is happening. False otherwise. + */ +static bool brute_attack_running(const struct brute_stats *stats) +{ + u64 threshold; + + if (stats->faults < (u32)brute_min_faults) + return false; + + if (stats->faults >= (u32)brute_max_faults) + return true; + + threshold = (u64)brute_crash_period_threshold * (u64)NSEC_PER_SEC; + return stats->period < threshold; +} + +/** + * brute_print_attack_running() - Warn about a fork brute force attack. + */ +static inline void brute_print_attack_running(void) +{ + pr_warn("fork brute force attack detected [pid %d: %s]\n", current->pid, + current->comm); +} + +/** + * brute_get_xattr_stats() - Get the stats from an extended attribute. + * @dentry: The dentry of the file to get the extended attribute. + * @inode: The inode of the file to get the extended attribute. + * @stats: The stats where to store the info obtained from the extended + * attribute. Cannot be NULL. + * + * Return: An error code if it is not possible to get the statistical data. Zero + * otherwise. + */ +static int brute_get_xattr_stats(struct dentry *dentry, struct inode *inode, + struct brute_stats *stats) +{ + int rc; + struct brute_raw_stats raw_stats; + + rc = __vfs_getxattr(dentry, inode, XATTR_NAME_BRUTE, &raw_stats, + sizeof(raw_stats)); + if (rc < 0) + return rc; + + stats->faults = le32_to_cpu(raw_stats.faults); + stats->nsecs = le64_to_cpu(raw_stats.nsecs); + stats->period = le64_to_cpu(raw_stats.period); + stats->flags = raw_stats.flags; + return 0; +} + +/** + * brute_set_xattr_stats() - Set the stats to an extended attribute. + * @dentry: The dentry of the file to set the extended attribute. + * @inode: The inode of the file to set the extended attribute. + * @stats: The stats from where to extract the info to set the extended attribute. + * Cannot be NULL. + * + * Return: An error code if it is not possible to set the statistical data. Zero + * otherwise. + */ +static int brute_set_xattr_stats(struct dentry *dentry, struct inode *inode, + const struct brute_stats *stats) +{ + struct brute_raw_stats raw_stats; + + raw_stats.faults = cpu_to_le32(stats->faults); + raw_stats.nsecs = cpu_to_le64(stats->nsecs); + raw_stats.period = cpu_to_le64(stats->period); + raw_stats.flags = stats->flags; + + return __vfs_setxattr(&init_user_ns, dentry, inode, XATTR_NAME_BRUTE, + &raw_stats, sizeof(raw_stats), 0); +} + +/** + * brute_update_xattr_stats() - Update the stats of a file. + * @file: The file that holds the statistical data to update. Cannot be NULL. + * + * For a correct management of a fork brute force attack it is only necessary to + * update the statistics and test if an attack is happening based on these data. + * It is important to note that if the file has no stats nothing is updated nor + * created. This way, the scenario where an application has not crossed any + * privilege boundary is avoided since the existence of the extended attribute + * denotes the crossing of bounds. + */ +static void brute_update_xattr_stats(const struct file *file) +{ + struct dentry *dentry = file_dentry(file); + struct inode *inode = file_inode(file); + struct brute_stats stats; + int rc; + + inode_lock(inode); + rc = brute_get_xattr_stats(dentry, inode, &stats); + WARN_ON_ONCE(rc && rc != -ENODATA); + if (rc) { + inode_unlock(inode); + return; + } + + brute_update_crash_period(&stats); + if (brute_attack_running(&stats)) { + brute_print_attack_running(); + stats.not_allowed = true; + } + + rc = brute_set_xattr_stats(dentry, inode, &stats); + WARN_ON_ONCE(rc); + inode_unlock(inode); +} + +/** + * brute_reset_stats() - Reset the statistical data. + * @stats: Statistics to be reset. Cannot be NULL. + */ +static inline void brute_reset_stats(struct brute_stats *stats) +{ + memset(stats, 0, sizeof(*stats)); +} + +/** + * brute_new_xattr_stats() - New statistics for a file. + * @file: The file in which to create the new statistical data. Cannot be NULL. + * + * Only if the file has no statistical data create it. This function will be + * called to mark that a privilege boundary has been crossed so, if new stats + * are required, they do not contain any useful data. The existence of the + * extended attribute denotes the crossing of privilege bounds. + * + * Return: An error code if it is not possible to get or set the statistical + * data. Zero otherwise. + */ +static int brute_new_xattr_stats(const struct file *file) +{ + struct dentry *dentry = file_dentry(file); + struct inode *inode = file_inode(file); + struct brute_stats stats; + int rc; + + inode_lock(inode); + rc = brute_get_xattr_stats(dentry, inode, &stats); + if (rc && rc != -ENODATA) + goto unlock; + + if (rc == -ENODATA) { + brute_reset_stats(&stats); + rc = brute_set_xattr_stats(dentry, inode, &stats); + if (rc) + goto unlock; + } + +unlock: + inode_unlock(inode); + return rc; +} + +/** + * brute_current_new_xattr_stats() - New stats for the current task's exe file. + * + * Return: An error code if it is not possible to get or set the statistical + * data. Zero otherwise. + */ +static int brute_current_new_xattr_stats(void) +{ + struct file *exe_file; + int rc; + + exe_file = brute_get_current_exe_file(); + if (WARN_ON_ONCE(!exe_file)) + return -ENOENT; + + rc = brute_new_xattr_stats(exe_file); + WARN_ON_ONCE(rc); + fput(exe_file); + return rc; +} + +/** + * brute_signal_from_user() - Test if a signal is coming from userspace. + * @siginfo: Contains the signal information. + * + * To avoid false positives during the attack detection it is necessary to + * narrow the possible cases. So, only the signals delivered by the kernel are + * taken into account with the exception of the SIGABRT signal since the latter + * is used by glibc for stack canary, malloc, etc failures, which may indicate + * that a mitigation has been triggered. + * + * Return: True if the signal is coming from usersapce. False otherwise. + */ +static inline bool brute_signal_from_user(const kernel_siginfo_t *siginfo) +{ + return siginfo->si_signo == SIGKILL && siginfo->si_code != SIGABRT; +} + +/** + * brute_task_fatal_signal() - Target for the task_fatal_signal hook. + * @siginfo: Contains the signal information. + * + * To detect a brute force attack it is necessary, as a first step, to test in + * every fatal crash if the signal is delibered by the kernel. If so, update the + * statistics and act based on these data. + */ +static void brute_task_fatal_signal(const kernel_siginfo_t *siginfo) +{ + struct file *exe_file; + + if (brute_signal_from_user(siginfo)) + return; + + exe_file = brute_get_current_exe_file(); + if (WARN_ON_ONCE(!exe_file)) + return; + + brute_update_xattr_stats(exe_file); + fput(exe_file); +} + +/** + * brute_task_execve() - Target for the bprm_creds_from_file hook. + * @bprm: Contains the linux_binprm structure. + * @file: Binary that will be executed without an interpreter. + * + * This hook is useful to mark that a privilege boundary (setuid/setgid process) + * has been crossed. This is done based on the "secureexec" flag. + * + * To be defensive return an error code if it is not possible to get or set the + * stats using an extended attribute since this blocks the execution of the + * file. This scenario is treated as an attack. + * + * It is important to note that here the brute_new_xattr_stats function could be + * used with a previous test of the secureexec flag. However it is better to use + * the basic xattr functions since in a future commit a test if the execution is + * allowed (via the brute_stats::not_allowed flag) will be necessary. This way, + * the stats of the file will be get only once. + * + * Return: An error code if it is not possible to get or set the statistical + * data. Zero otherwise. + */ +static int brute_task_execve(struct linux_binprm *bprm, struct file *file) +{ + struct dentry *dentry = file_dentry(bprm->file); + struct inode *inode = file_inode(bprm->file); + struct brute_stats stats; + int rc; + + inode_lock(inode); + rc = brute_get_xattr_stats(dentry, inode, &stats); + if (WARN_ON_ONCE(rc && rc != -ENODATA)) + goto unlock; + + if (rc == -ENODATA && bprm->secureexec) { + brute_reset_stats(&stats); + rc = brute_set_xattr_stats(dentry, inode, &stats); + if (WARN_ON_ONCE(rc)) + goto unlock; + } + + rc = 0; +unlock: + inode_unlock(inode); + return rc; +} + +/** + * brute_task_change_priv() - Target for the task_fix_setid hooks. + * @new: The set of credentials that will be installed. + * @old: The set of credentials that are being replaced. + * @flags: Contains one of the LSM_SETID_* values. + * + * This hook is useful to mark that a privilege boundary (privilege changes) has + * been crossed. + * + * Return: An error code if it is not possible to get or set the statistical + * data. Zero otherwise. + */ +static int brute_task_change_priv(struct cred *new, const struct cred *old, int flags) +{ + return brute_current_new_xattr_stats(); +} + +#ifdef CONFIG_IPV6 +/** + * brute_local_ipv6_rcv_saddr() - Test if an ipv6 rcv_saddr is local. + * @sk: The sock that contains the ipv6 address. + * + * Return: True if the ipv6 rcv_saddr is local. False otherwise. + */ +static inline bool brute_local_ipv6_rcv_saddr(const struct sock *sk) +{ + return ipv6_addr_equal(&sk->sk_v6_rcv_saddr, &in6addr_loopback); +} +#else +static inline bool brute_local_ipv6_rcv_saddr(const struct sock *sk) +{ + return false; +} +#endif /* CONFIG_IPV6 */ + +#ifdef CONFIG_SECURITY_NETWORK +/** + * brute_socket_accept() - Target for the socket_accept hook. + * @sock: Contains the listening socket structure. + * @newsock: Contains the newly created server socket for connection. + * + * This hook is useful to mark that a privilege boundary (network to local) has + * been crossed. This is done only if the listening socket accepts external + * connections. The sockets for inter-process communication (IPC) and those that + * are listening on loopback addresses are not taken into account. + * + * Return: An error code if it is not possible to get or set the statistical + * data. Zero otherwise. + */ +static int brute_socket_accept(struct socket *sock, struct socket *newsock) +{ + struct sock *sk = sock->sk; + + if (sk->sk_family == AF_UNIX || sk->sk_family == AF_NETLINK || + sk->sk_rcv_saddr == htonl(INADDR_LOOPBACK) || + brute_local_ipv6_rcv_saddr(sk)) + return 0; + + return brute_current_new_xattr_stats(); +} +#endif /* CONFIG_SECURITY_NETWORK */ + +/* + * brute_hooks - Targets for the LSM's hooks. + */ +static struct security_hook_list brute_hooks[] __lsm_ro_after_init = { + LSM_HOOK_INIT(task_fatal_signal, brute_task_fatal_signal), + LSM_HOOK_INIT(bprm_creds_from_file, brute_task_execve), + LSM_HOOK_INIT(task_fix_setuid, brute_task_change_priv), + LSM_HOOK_INIT(task_fix_setgid, brute_task_change_priv), +#ifdef CONFIG_SECURITY_NETWORK + LSM_HOOK_INIT(socket_accept, brute_socket_accept), +#endif +}; + #ifdef CONFIG_SYSCTL static unsigned int uint_max = UINT_MAX; #define SYSCTL_UINT_MAX (&uint_max) @@ -137,6 +635,8 @@ static inline void brute_init_sysctl(void) { } static int __init brute_init(void) { pr_info("becoming mindful\n"); + security_add_hooks(brute_hooks, ARRAY_SIZE(brute_hooks), + KBUILD_MODNAME); brute_init_sysctl(); return 0; } -- 2.25.1

Add a new SIGCHLD si_code to notify to userspace, using the "waitid" system call, that a task has been killed by Brute LSM to mitigate a brute force attack.

This is useful to supervisors in order to decide if a process that has been killed to avoid an attack needs to be respawned. This way, it is possible to avoid the scenario where a brute force attack can be continued due to the respawn of a process. Although the xattr of the executable is accessible from userspace, in complex daemons this file may not be visible directly by the supervisor as it may be run through some wrapper. So, the waitid notification is necessary.

To achieve this, use the task_struct security blob to hold a flag that shows when a task has been killed by Brute LSM, and also, test this flag in the "wait_task_zombie" and "do_notify_parent" functions.

Suggested-by: Andi Kleen ak@linux.intel.com Signed-off-by: John Wood john.wood@gmx.com --- arch/x86/kernel/signal_compat.c | 2 +- include/brute/brute.h | 16 ++++++++ include/uapi/asm-generic/siginfo.h | 3 +- kernel/exit.c | 6 ++- kernel/signal.c | 4 +- security/brute/brute.c | 59 +++++++++++++++++++++++++++++- 6 files changed, 85 insertions(+), 5 deletions(-) create mode 100644 include/brute/brute.h

diff --git a/arch/x86/kernel/signal_compat.c b/arch/x86/kernel/signal_compat.c index 06743ec054d2..d4656f1b6341 100644 --- a/arch/x86/kernel/signal_compat.c +++ b/arch/x86/kernel/signal_compat.c @@ -30,7 +30,7 @@ static inline void signal_compat_build_tests(void) BUILD_BUG_ON(NSIGSEGV != 9); BUILD_BUG_ON(NSIGBUS != 5); BUILD_BUG_ON(NSIGTRAP != 6); - BUILD_BUG_ON(NSIGCHLD != 6); + BUILD_BUG_ON(NSIGCHLD != 7); BUILD_BUG_ON(NSIGSYS != 2);

/* This is part of the ABI and can never change in size: */ diff --git a/include/brute/brute.h b/include/brute/brute.h new file mode 100644 index 000000000000..8531a7038711 --- /dev/null +++ b/include/brute/brute.h @@ -0,0 +1,16 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _BRUTE_H_ +#define _BRUTE_H_ + +#include <linux/sched.h> + +#ifdef CONFIG_SECURITY_FORK_BRUTE +bool brute_task_killed(const struct task_struct *task); +#else +static inline bool brute_task_killed(const struct task_struct *task) +{ + return false; +} +#endif + +#endif /* _BRUTE_H_ */ diff --git a/include/uapi/asm-generic/siginfo.h b/include/uapi/asm-generic/siginfo.h index 5a3c221f4c9d..ffc3ed2d4bce 100644 --- a/include/uapi/asm-generic/siginfo.h +++ b/include/uapi/asm-generic/siginfo.h @@ -274,7 +274,8 @@ typedef struct siginfo { #define CLD_TRAPPED 4 /* traced child has trapped */ #define CLD_STOPPED 5 /* child has stopped */ #define CLD_CONTINUED 6 /* stopped child has continued */ -#define NSIGCHLD 6 +#define CLD_BRUTE 7 /* child was killed by brute LSM */ +#define NSIGCHLD 7

/* * SIGPOLL (or any other signal without signal specific si_codes) si_codes diff --git a/kernel/exit.c b/kernel/exit.c index fd1c04193e18..69bcbd00d277 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -69,6 +69,8 @@ #include <asm/unistd.h> #include <asm/mmu_context.h>

+#include <brute/brute.h> + static void __unhash_process(struct task_struct *p, bool group_dead) { nr_threads--; @@ -1001,6 +1003,7 @@ static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) pid_t pid = task_pid_vnr(p); uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); struct waitid_info *infop; + bool killed_by_brute = brute_task_killed(p);

if (!likely(wo->wo_flags & WEXITED)) return 0; @@ -1114,7 +1117,8 @@ static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) infop->cause = CLD_EXITED; infop->status = status >> 8; } else { - infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; + infop->cause = (status & 0x80) ? CLD_DUMPED : + killed_by_brute ? CLD_BRUTE : CLD_KILLED; infop->status = status & 0x7f; } infop->pid = pid; diff --git a/kernel/signal.c b/kernel/signal.c index 4380763b3d8d..c85c091ecc27 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -55,6 +55,8 @@ #include <asm/siginfo.h> #include <asm/cacheflush.h>

+#include <brute/brute.h> + /* * SLAB caches for signal bits. */ @@ -2012,7 +2014,7 @@ bool do_notify_parent(struct task_struct *tsk, int sig) if (tsk->exit_code & 0x80) info.si_code = CLD_DUMPED; else if (tsk->exit_code & 0x7f) - info.si_code = CLD_KILLED; + info.si_code = brute_task_killed(tsk) ? CLD_BRUTE : CLD_KILLED; else { info.si_code = CLD_EXITED; info.si_status = tsk->exit_code >> 8; diff --git a/security/brute/brute.c b/security/brute/brute.c index 4e0fd23990c8..e5c9098addf9 100644 --- a/security/brute/brute.c +++ b/security/brute/brute.c @@ -56,6 +56,59 @@ struct brute_raw_stats { u8 flags; } __packed;

+/** + * struct brute_task - Task info. + * @killed: Task killed to mitigate a brute force attack. + */ +struct brute_task { + u8 killed : 1; +}; + +/* + * brute_blob_sizes - LSM blob sizes. + */ +static struct lsm_blob_sizes brute_blob_sizes __lsm_ro_after_init = { + .lbs_task = sizeof(struct brute_task), +}; + +/** + * brute_task() - Get the task info. + * @task: The task to get the info. + * + * Return: A pointer to the brute_task structure. + */ +static inline struct brute_task *brute_task(const struct task_struct *task) +{ + return task->security + brute_blob_sizes.lbs_task; +} + +/** + * brute_set_task_killed() - Set task killed to mitigate a brute force attack. + * @task: The task to set. + */ +static inline void brute_set_task_killed(struct task_struct *task) +{ + struct brute_task *task_info; + + task_info = brute_task(task); + task_info->killed = true; +} + +/** + * brute_task_killed() - Test if a task has been killed to mitigate an attack. + * @task: The task to test. + * + * Return: True if the task has been killed to mitigate a brute force attack. + * False otherwise. + */ +inline bool brute_task_killed(const struct task_struct *task) +{ + struct brute_task *task_info; + + task_info = brute_task(task); + return task_info->killed; +} + /** * brute_get_current_exe_file() - Get the current task's executable file. * @@ -296,8 +349,10 @@ static void brute_kill_offending_tasks(const struct file *file)

read_lock(&tasklist_lock); for_each_process(task) { - if (task->group_leader == current->group_leader) + if (task->group_leader == current->group_leader) { + brute_set_task_killed(task); continue; + }

exe_file = get_task_exe_file(task); if (!exe_file) @@ -311,6 +366,7 @@ static void brute_kill_offending_tasks(const struct file *file) do_send_sig_info(SIGKILL, SEND_SIG_PRIV, task, PIDTYPE_PID); pr_warn_ratelimited("offending process %d [%s] killed\n", task->pid, task->comm); + brute_set_task_killed(task); } read_unlock(&tasklist_lock); } @@ -735,4 +791,5 @@ static int __init brute_init(void) DEFINE_LSM(brute) = { .name = KBUILD_MODNAME, .init = brute_init, + .blobs = &brute_blob_sizes, }; -- 2.25.1

Add some info detailing what is the Brute LSM, its motivation, weak points of existing implementations, proposed solutions, notifications, enabling, disabling, configuration and self-tests.

Signed-off-by: John Wood john.wood@gmx.com --- Documentation/admin-guide/LSM/Brute.rst | 359 ++++++++++++++++++++++++ Documentation/admin-guide/LSM/index.rst | 1 + security/brute/Kconfig | 3 +- 3 files changed, 362 insertions(+), 1 deletion(-) create mode 100644 Documentation/admin-guide/LSM/Brute.rst

diff --git a/Documentation/admin-guide/LSM/Brute.rst b/Documentation/admin-guide/LSM/Brute.rst new file mode 100644 index 000000000000..087da9c07374 --- /dev/null +++ b/Documentation/admin-guide/LSM/Brute.rst @@ -0,0 +1,359 @@ +.. SPDX-License-Identifier: GPL-2.0 + +===== +Brute +===== + +Brute is a Linux Security Module that detects and mitigates fork brute force +attacks against vulnerable userspace processes. + + +Motivation +========== + +Attacks against vulnerable userspace applications with the purpose to break ASLR +or bypass canaries traditionally use some level of brute force with the help of +the fork system call. This is possible since when creating a new process using +fork its memory contents are the same as those of the parent process (the +process that called the fork system call). So, the attacker can test the memory +infinite times to find the correct memory values or the correct memory addresses +without worrying about crashing the application. + +Based on the above scenario it would be nice to have this detected and +mitigated, and this is the goal of this implementation. Specifically the +following attacks are expected to be detected: + + 1. Launching (fork()/exec()) a setuid/setgid process repeatedly until a + desirable memory layout is got (e.g. Stack Clash). + 2. Connecting to an exec()ing network daemon (e.g. xinetd) repeatedly until a + desirable memory layout is got (e.g. what CTFs do for simple network + service). + 3. Launching processes without exec() (e.g. Android Zygote) and exposing state + to attack a sibling. + 4. Connecting to a fork()ing network daemon (e.g. apache) repeatedly until the + previously shared memory layout of all the other children is exposed (e.g. + kind of related to HeartBleed). + +In each case, a privilege boundary has been crossed: + + | Case 1: setuid/setgid process + | Case 2: network to local + | Case 3: privilege changes + | Case 4: network to local + +So, what really needs to be detected are fork/exec brute force attacks that +cross any of the commented bounds. + + +Other implementations +===================== + +The public version of grsecurity, as a summary, is based on the idea of delaying +the fork system call if a child died due to some fatal signal (``SIGSEGV``, +``SIGBUS``, ``SIGKILL`` or ``SIGILL``). This has some issues: + +Bad practices +------------- + +Adding delays to the kernel is, in general, a bad idea. + +Scenarios not detected (false negatives) +---------------------------------------- + +This protection acts only when the fork system call is called after a child has +crashed. So, it would still be possible for an attacker to fork a big amount of +children (in the order of thousands), then probe all of them, and finally wait +the protection time before repeating the steps. + +Moreover, this method is based on the idea that the protection doesn't act if +the parent crashes. So, it would still be possible for an attacker to fork a +process and probe itself. Then, fork the child process and probe itself again. +This way, these steps can be repeated infinite times without any mitigation. + +Scenarios detected (false positives) +------------------------------------ + +Scenarios where an application rarely fails for reasons unrelated to a real +attack. + + +This implementation +=================== + +The main idea behind this implementation is to improve the existing ones +focusing on the weak points annotated before. Basically, the adopted solution is +to detect a fast crash rate instead of only one simple crash and to detect both +the crash of parent and child processes. Also, fine tune the detection focusing +on privilege boundary crossing. And finally, as a mitigation method, kill all +the offending tasks involved in the attack and mark the executable as "not +allowed" (to block the following executions) instead of using delays. + +To achieve this goal, and going into more details, this implementation is based +on the use of some statistical data shared across all the processes that can +have the same memory contents. Or in other words, a statistical data shared +between all the fork hierarchy processes after an execve system call. + +The purpose of these statistics is, basically, collect all the necessary info to +compute the application crash period in order to detect an attack. To track all +this information, the extended attributes (xattr) of the executable files are +used. More specifically, the name of the attribute is "brute" and uses the +"security" namespace. So, the full xattr name for the Brute LSM is: + + ``security.brute`` + +The same can be achieved using a pointer to the fork hierarchy statistical data +held by the ``task_struct`` structure, but this has an important drawback: a +brute force attack that happens through the execve system call losts the faults +info since these statistics are freed when the fork hierarchy disappears. Using +the last method (pointer in the ``task_struct`` structure) makes not possible to +manage this attack type that can be successfully treated using extended +attributes. + +To detect a brute force attack it is necessary that the statistics shared by all +the fork hierarchy processes be updated in every fatal crash and the most +important data to update is the application crash period. + +The crash period is the time between two consecutive faults, but this also has a +drawback: if an application crashes twice in a short period of time for some +reason unrelated to a real attack, a false positive will be triggered. To avoid +this scenario the exponential moving average (EMA) is used. This way, the +application crash period will be a value that is not prone to change due to +spurious data and follows the real crash period. + +These statistics are stored in the executables using the extended attributes +feature. So, the detection and mitigation of brute force attacks using this LSM +it is only feasible in filesystems that support xattr. + +.. kernel-doc:: security/brute/brute.c + :identifiers: brute_raw_stats + +This is a fixed sized struct with a very small footprint. So, in reference to +memory usage, it is not expected to have problems storing it as an extended +attribute. + +Concerning to access rights to this statistical data, as stated above, the +"security" namespace is used. Since no custom policy, related to this extended +attribute, has been implemented for the Brute LSM, all processes have read +access to these statistics, and write access is limited to processes that have +the ``CAP_SYS_ADMIN`` capability. + +Attack detection +---------------- + +There are two types of brute force attacks that need to be detected. The first +one is an attack that happens through the fork system call and the second one is +an attack that happens through the execve system call. Moreover, these two +attack types have two variants. A slow brute force attack that is detected if a +maximum number of faults per fork hierarchy is reached and a fast brute force +attack that is detected if the application crash period falls below a certain +threshold. + +Attack mitigation +----------------- + +Once an attack has been detected, this is mitigated killing all the offending +tasks involved. Or in other words, once an attack has been detected, this is +mitigated killing all the processes that are executing the same file that is +running during the brute force attack. Also, to prevent the executable involved +in the attack from being respawned by a supervisor, and thus prevent a brute +force attack from being started again, the file is marked as "not allowed" and +the following executions are avoided based on this mark. This method allows +supervisors to implement their own policy: they can read the statistics, know if +the executable is blocked by the Brute LSM and why, and act based on this +information. If they want to respawn the offending executable it is only +necessary to remove the "``security.brute``" extended attribute and thus remove +the statistical data. + +Fine tuning the attack detection +-------------------------------- + +To avoid false positives during the attack detection it is necessary to narrow +the possible cases. To do so, and based on the threat scenarios that we want to +detect, this implementation also focuses on the crossing of privilege bounds. + +To be precise, only the following privilege bounds are taken into account: + + 1. setuid/setgid process + 2. network to local + 3. privilege changes + +Moreover, only the fatal signals delivered by the kernel are taken into account +avoiding the fatal signals sent by userspace applications (with the exception of +the ``SIGABRT`` user signal since this is used by glibc for stack canary, +malloc, etc. failures, which may indicate that a mitigation has been triggered). + +Userspace notification via waitid() system call +----------------------------------------------- + +Although the xattr of the executable is accessible from userspace, in complex +daemons this file may not be visible directly by the supervisor as it may be run +through some wrapper. So, an extension to the ``waitid()`` system call has been +added. + + ``int waitid(idtype_t idtype, id_t id, siginfo_t *infop, int options);`` + +Upon successful return, ``waitid()`` fills in the ``siginfo_t`` structure +pointed to by ``infop``, but now, the ``si_code`` field can be: + + ``CLD_BRUTE``: child was killed by brute LSM. Defined as value 7. + +in addition to the following codes: + + | ``CLD_EXITED``: child has called exit. Defined as value 1. + | ``CLD_KILLED``: child was killed by signal. Defined as value 2. + | ``CLD_DUMPED``: child terminated abnormally. Defined as value 3. + | ``CLD_TRAPPED``: traced child has trapped. Defined as value 4. + | ``CLD_STOPPED``: child has stopped. Defined as value 5. + | ``CLD_CONTINUED``: stopped child has continued. Defined as value 6. + +Exponential moving average (EMA) +-------------------------------- + +This kind of average defines a weight (between 0 and 1) for the new value to add +and applies the remainder of the weight to the current average value. This way, +some spurious data will not excessively modify the average and only if the new +values are persistent, the moving average will tend towards them. + +Mathematically the application crash period's EMA can be expressed as follows: + + period_ema = period * weight + period_ema * (1 - weight) + +Related to the attack detection, the EMA must guarantee that not many crashes +are needed. To demonstrate this, the scenario where an application has failed +and then has been running without any crashes for a month, will be used. + +The period's EMA can be written now as: + + period_ema[i] = period[i] * weight + period_ema[i - 1] * (1 - weight) + +If the new crash periods have insignificant values related to the first crash +period (a month in this case), the formula can be rewritten as: + + period_ema[i] = period_ema[i - 1] * (1 - weight) + +And by extension: + + | period_ema[i - 1] = period_ema[i - 2] * (1 - weight) + | period_ema[i - 2] = period_ema[i - 3] * (1 - weight) + | period_ema[i - 3] = period_ema[i - 4] * (1 - weight) + +So, if the substitution is made: + + | period_ema[i] = period_ema[i - 1] * (1 - weight) + | period_ema[i] = period_ema[i - 2] * (1 - weight)\ :sup:`2` + | period_ema[i] = period_ema[i - 3] * (1 - weight)\ :sup:`3` + | period_ema[i] = period_ema[i - 4] * (1 - weight)\ :sup:`4` + +And in a more generic form: + + period_ema[i] = period_ema[i - n] * (1 - weight)\ :sup:`n` + +Where "n" represents the number of iterations to obtain an EMA value. Or in +other words, the number of crashes to detect an attack. + +So, if we isolate the number of crashes: + + | period_ema[i] / period_ema[i - n] = (1 - weight)\ :sup:`n` + | log(period_ema[i] / period_ema[i - n]) = log((1 - weight)\ :sup:`n`) + | log(period_ema[i] / period_ema[i - n]) = n * log(1 - weight) + | n = log(period_ema[i] / period_ema[i - n]) / log(1 - weight) + +Then, in the commented scenario (an application has failed and then has been +running without any crashes for a month), the approximate number of crashes to +detect an attack (using the default implementation values for the weight and the +crash period threshold) is: + + | weight = 7 / 10 + | crash_period_threshold = 30 seconds + + | n = log(crash_period_threshold / seconds_per_month) / log(1 - weight) + | n = log(30 / (30 * 24 * 3600)) / log(1 - 0.7) + | n = 9.44 + +So, with 10 crashes for this scenario an attack will be detected. If these steps +are repeated for different scenarios and the results are collected: + + ======================== ===================================== + time without any crashes number of crashes to detect an attack + ======================== ===================================== + 1 month 9.44 + 1 year 11.50 + 10 years 13.42 + ======================== ===================================== + +However, this computation has a drawback. The first data added to the EMA not +obtains a real average showing a trend. So the solution is simple, the EMA needs +a minimum number of data to be able to be interpreted. This way, the case where +a few first faults are fast enough followed by no crashes is avoided. + +Per system enabling/disabling +----------------------------- + +This feature can be enabled at build time using the +``CONFIG_SECURITY_FORK_BRUTE`` option or using the visual config application +under the following menu: + + Security options ``--->`` Fork brute force attack detection and mitigation + +Also, at boot time, this feature can be disable too, by changing the "``lsm=``" +boot parameter. + +Per system configuration +------------------------ + +To customize the detection's sensibility there are five new sysctl attributes +for the Brute LSM that are accessible through the following path: + + ``/proc/sys/kernel/brute/`` + +More specifically, the files and their description are: + +**ema_weight_numerator** + + .. kernel-doc:: security/brute/brute.c + :doc: brute_ema_weight_numerator + +**ema_weight_denominator** + + .. kernel-doc:: security/brute/brute.c + :doc: brute_ema_weight_denominator + +**max_faults** + + .. kernel-doc:: security/brute/brute.c + :doc: brute_max_faults + +**min_faults** + + .. kernel-doc:: security/brute/brute.c + :doc: brute_min_faults + +**crash_period_threshold** + + .. kernel-doc:: security/brute/brute.c + :doc: brute_crash_period_threshold + +Kernel selftests +---------------- + +To validate all the expectations about this implementation, there is a set of +selftests. This tests cover fork/exec brute force attacks crossing the following +privilege boundaries: + + 1. setuid process + 2. privilege changes + 3. network to local + +Also, there are some tests to check that fork/exec brute force attacks without +crossing any privilege boundary already commented doesn't trigger the detection +and mitigation stage. Moreover, a test to verify the userspace notification via +the ``waitid()`` system call has also been added. + +To build the tests: + ``make -C tools/testing/selftests/ TARGETS=brute`` + +To run the tests: + ``make -C tools/testing/selftests TARGETS=brute run_tests`` + +To package the tests: + ``make -C tools/testing/selftests TARGETS=brute gen_tar`` diff --git a/Documentation/admin-guide/LSM/index.rst b/Documentation/admin-guide/LSM/index.rst index a6ba95fbaa9f..1f68982bb330 100644 --- a/Documentation/admin-guide/LSM/index.rst +++ b/Documentation/admin-guide/LSM/index.rst @@ -41,6 +41,7 @@ subdirectories. :maxdepth: 1

apparmor + Brute LoadPin SELinux Smack diff --git a/security/brute/Kconfig b/security/brute/Kconfig index 5da314d221aa..d2dd33b08642 100644 --- a/security/brute/Kconfig +++ b/security/brute/Kconfig @@ -9,6 +9,7 @@ config SECURITY_FORK_BRUTE offending tasks are killed. Also, the executable file involved in the attack will be marked as "not allowed" and new execve system calls using this file will fail. Like capabilities, this security module - stacks with other LSMs. + stacks with other LSMs. Further information can be found in + Documentation/admin-guide/LSM/Brute.rst.

If you are unsure how to answer this question, answer N. -- 2.25.1