Yesterday I went to LinuxCon Japan 2014 and stopped at Red Hat's booth and Oracle's booth. I explained about this module ( using page 92 of http://I-love.SAKURA.ne.jp/tomoyo/LCJ2014-en.pdf ) and got positive responses from persons who have experienced troubleshooting jobs. I was convinced that I am not the only person who is bothered by lack of process history information in the logs. Therefore, I repost this module toward inclusion into mainline Linux kernel. Changes from previous version ( http://lwn.net/Articles/575044/ ): (1) Assign a value to "u32 *seclen" in addition to "char *secdata" at security_task_getsecid() hook. (2) Make calculation of time stamp a bit faster. Background: When an unexpected system event (e.g. reboot) occurs, the administrator may want to identify which application triggered the event. System call auditing could be used for recording such event. However, the audit log may not be able to provide sufficient information for identifying the application because the audit log does not reflect how the program was executed. I sometimes receive "which application triggered the event" questions on RHEL systems. TOMOYO security module can track how the program was executed, but TOMOYO is not yet available in Fedora/RHEL distributions. Although subj= field is added to the audit log if SELinux is not disabled, SELinux is too difficult to customize as fine grained as I expect in order to reflect how the program was executed. Therefore, I'm currently using AKARI and SystemTap for emulating TOMOYO-like tracing. But AKARI and SystemTap do not help unless the kernel module is loaded before the unexpected system event occurs. Generally, the administrator is failing to record the first event, and has to wait for the same event to occur again after loading the kernel module and/or configuring auditing. I came to think that we want a built-in kernel routine which is automatically started upon boot so that we don't fail to record the first event. What I did: Assuming that multiple concurrent LSM support comes in the near future, I wrote a trivial LSM module which emits TOMOYO-like information into the audit logs. Usage: Just register this LSM module. No configuration is needed. You will get history of current thread in the form of comm name and time stamp pairs in the subj= field of audit logs like examples shown in the patch description. ---------- >From ff68d3a4cd496bd263d2939848777fffc30cbc0b Mon Sep 17 00:00:00 2001 From: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; Date: Fri, 23 May 2014 21:31:56 +0900 Subject: [PATCH] TaskTracker : Simplified thread information tracker. Existing audit logs generated via system call auditing functionality include current thread's comm name. But it is not always sufficient for identifying which application has requested specific operations because comm name does not reflect history of current thread. This security module adds functionality for adding current thread's history information like TOMOYO security module does, expecting that this module can help us getting more information from system call auditing functionality. type=USER_LOGIN msg=audit(1400879947.084:24): pid=4308 uid=0 auid=0 ses=2 subj="swapper/0(2014/05/23-21:17:30)=>init(2014/05/23-21:17:33)=> switch_root(2014/05/23-21:17:34)=>init(2014/05/23-21:17:34)=> sh(2014/05/23-21:17:56)=>mingetty(2014/05/23-21:17:56)=> login(2014/05/23-21:19:05)" msg='op=login id=0 exe="/bin/login" hostname=? addr=? terminal=tty1 res=success' type=SYSCALL msg=audit(1400880014.444:26): arch=40000003 syscall=11 success=yes exit=0 a0=8140f78 a1=812b7d8 a2=812b248 a3=812b7d8 items=2 ppid=4323 pid=4355 auid=0 uid=0 gid=0 euid=0 suid=0 fsuid=0 egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="tail" exe="/usr/bin/tail" subj="swapper/0(2014/05/23-21:17:30)=>init(2014/05/23-21:17:33)=> switch_root(2014/05/23-21:17:34)=>init(2014/05/23-21:17:34)=> sh(2014/05/23-21:17:37)=>rc(2014/05/23-21:17:37)=> S55sshd(2014/05/23-21:17:53)=>sshd(2014/05/23-21:17:53)=> sshd(2014/05/23-21:18:17)=>bash(2014/05/23-21:18:21)=> tail(2014/05/23-21:20:14)" key=(null) Signed-off-by: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; --- security/Kconfig | 6 + security/Makefile | 2 + security/tasktracker/Kconfig | 35 +++++ security/tasktracker/Makefile | 1 + security/tasktracker/tasktracker.c | 282 ++++++++++++++++++++++++++++++++++++ 5 files changed, 326 insertions(+), 0 deletions(-) create mode 100644 security/tasktracker/Kconfig create mode 100644 security/tasktracker/Makefile create mode 100644 security/tasktracker/tasktracker.c diff --git a/security/Kconfig b/security/Kconfig index beb86b5..14e7d27 100644 --- a/security/Kconfig +++ b/security/Kconfig @@ -122,6 +122,7 @@ source security/smack/Kconfig source security/tomoyo/Kconfig source security/apparmor/Kconfig source security/yama/Kconfig +source security/tasktracker/Kconfig source security/integrity/Kconfig @@ -132,6 +133,7 @@ choice default DEFAULT_SECURITY_TOMOYO if SECURITY_TOMOYO default DEFAULT_SECURITY_APPARMOR if SECURITY_APPARMOR default DEFAULT_SECURITY_YAMA if SECURITY_YAMA + default DEFAULT_SECURITY_TT if SECURITY_TT default DEFAULT_SECURITY_DAC help @@ -153,6 +155,9 @@ choice config DEFAULT_SECURITY_YAMA bool "Yama" if SECURITY_YAMA=y + config DEFAULT_SECURITY_TT + bool "TaskTracker" if SECURITY_TT=y + config DEFAULT_SECURITY_DAC bool "Unix Discretionary Access Controls" @@ -165,6 +170,7 @@ config DEFAULT_SECURITY default "tomoyo" if DEFAULT_SECURITY_TOMOYO default "apparmor" if DEFAULT_SECURITY_APPARMOR default "yama" if DEFAULT_SECURITY_YAMA + default "tt" if DEFAULT_SECURITY_TT default "" if DEFAULT_SECURITY_DAC endmenu diff --git a/security/Makefile b/security/Makefile index 05f1c93..28a90ed 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-$(CONFIG_SECURITY_TT) += tasktracker # always enable default capabilities obj-y += commoncap.o @@ -22,6 +23,7 @@ obj-$(CONFIG_AUDIT) += lsm_audit.o obj-$(CONFIG_SECURITY_TOMOYO) += tomoyo/ obj-$(CONFIG_SECURITY_APPARMOR) += apparmor/ obj-$(CONFIG_SECURITY_YAMA) += yama/ +obj-$(CONFIG_SECURITY_TT) += tasktracker/ obj-$(CONFIG_CGROUP_DEVICE) += device_cgroup.o # Object integrity file lists diff --git a/security/tasktracker/Kconfig b/security/tasktracker/Kconfig new file mode 100644 index 0000000..6de5354 --- /dev/null +++ b/security/tasktracker/Kconfig @@ -0,0 +1,35 @@ +config SECURITY_TT + bool "TaskTracker support" + depends on SECURITY + default n + help + Existing audit logs generated via system call auditing functionality + include current thread's comm name. But it is not always sufficient + for identifying which application has requested specific operations + because comm name does not reflect history of current thread. + + This security module adds functionality for adding current thread's + history information like TOMOYO security module does, expecting that + this module can help us getting more information from system call + auditing functionality. + + If you are unsure how to answer this question, answer N. + + Usage: + + Just register this module. No configuration is needed. + + You will get history of current thread in the form of + comm name and time stamp pairs in the subj= field of audit logs + like an example shown below. + + type=SYSCALL msg=audit(1400880014.444:26): arch=40000003 syscall=11 + success=yes exit=0 a0=8140f78 a1=812b7d8 a2=812b248 a3=812b7d8 + items=2 ppid=4323 pid=4355 auid=0 uid=0 gid=0 euid=0 suid=0 fsuid=0 + egid=0 sgid=0 fsgid=0 tty=pts0 ses=1 comm="tail" + exe="/usr/bin/tail" subj="swapper/0(2014/05/23-21:17:30)=> + init(2014/05/23-21:17:33)=>switch_root(2014/05/23-21:17:34)=> + init(2014/05/23-21:17:34)=>sh(2014/05/23-21:17:37)=> + rc(2014/05/23-21:17:37)=>S55sshd(2014/05/23-21:17:53)=> + sshd(2014/05/23-21:17:53)=>sshd(2014/05/23-21:18:17)=> + bash(2014/05/23-21:18:21)=>tail(2014/05/23-21:20:14)" key=(null) diff --git a/security/tasktracker/Makefile b/security/tasktracker/Makefile new file mode 100644 index 0000000..15d03ce --- /dev/null +++ b/security/tasktracker/Makefile @@ -0,0 +1 @@ +obj-$(CONFIG_SECURITY_TT) := tasktracker.o diff --git a/security/tasktracker/tasktracker.c b/security/tasktracker/tasktracker.c new file mode 100644 index 0000000..ec4eb0c --- /dev/null +++ b/security/tasktracker/tasktracker.c @@ -0,0 +1,282 @@ +/* + * tasktracker.c - Simplified thread information tracker. + * + * Copyright (C) 2010-2014 Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; + */ +#include <linux/security.h> +#include <linux/binfmts.h> + +/* Wrapper structure for passing string buffer. */ +struct tt_record { + char history[1024]; +}; + +/* Structure for representing YYYY/MM/DD hh/mm/ss. */ +struct tt_time { + u16 year; + u8 month; + u8 day; + u8 hour; + u8 min; + u8 sec; +}; + +/** + * tt_get_time - Get current time in YYYY/MM/DD hh/mm/ss format. + * + * @stamp: Pointer to "struct tt_time". + * + * Returns nothing. + * + * This function does not handle Y2038 problem. + */ +static void tt_get_time(struct tt_time *stamp) +{ + struct timeval tv; + static const u16 tt_eom[2][12] = { + { 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365 }, + { 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335, 366 } + }; + u16 y = 1970; + u8 m; + bool r; + time_t time; + do_gettimeofday(&tv); + time = tv.tv_sec; + stamp->sec = time % 60; + time /= 60; + stamp->min = time % 60; + time /= 60; + stamp->hour = time % 24; + time /= 24; + if (time >= 16071) { + /* Start from 2014/01/01 rather than 1970/01/01. */ + time -= 16071; + y += 44; + } + while (1) { + const unsigned short days = (y & 3) ? 365 : 366; + if (time < days) + break; + time -= days; + y++; + } + r = (y & 3) == 0; + for (m = 0; m < 11 && time >= tt_eom[r][m]; m++) + ; + if (m) + time -= tt_eom[r][m - 1]; + stamp->year = y; + stamp->month = ++m; + stamp->day = ++time; +} + +/** + * tt_update_record - Update "struct tt_record" for given credential. + * + * @record: Pointer to "struct tt_record". + * + * Returns nothing. + */ +static void tt_update_record(struct tt_record *record) +{ + char *cp; + int i; + struct tt_time stamp; + tt_get_time(&stamp); + /* + * Lockless update because current thread's record is not concurrently + * accessible, for "struct cred"->security is not visible from other + * threads because this function is called upon only boot up and + * successful execve() operation. + */ + cp = record->history; + i = strlen(cp); + while (i >= sizeof(record->history) - (TASK_COMM_LEN * 4 + 30)) { + /* + * Since this record is not for making security decision, + * I don't care by-chance matching "=>" in task's commname. + */ + char *cp2 = strstr(cp + 2, "=>"); + if (!cp2) + return; + memmove(cp + 1, cp2, strlen(cp2) + 1); + i = strlen(cp); + } + if (!i) + *cp++ = '"'; + else { + cp += i - 1; + *cp++ = '='; + *cp++ = '>'; + } + /* + * Lockless read because this is current thread and being unexpectedly + * modified by other thread is not a fatal problem. + */ + for (i = 0; i < TASK_COMM_LEN; i++) { + const unsigned char c = current->comm[i]; + if (!c) + break; + else if (c == '"' || c == '\\' || c < 0x21 || c > 0x7e) { + *cp++ = '\\'; + *cp++ = (c >> 6) + '0'; + *cp++ = ((c >> 3) & 7) + '0'; + *cp++ = (c & 7) + '0'; + } else + *cp++ = c; + } + sprintf(cp, "(%04u/%02u/%02u-%02u:%02u:%02u)\"", stamp.year, + stamp.month, stamp.day, stamp.hour, stamp.min, stamp.sec); +} + +/** + * tt_find_record - Find "struct tt_record" for given credential. + * + * @cred: Pointer to "struct cred". + * + * Returns pointer to "struct tt_record". + */ +static inline struct tt_record *tt_find_record(const struct cred *cred) +{ + return cred->security; +} + +/** + * tt_cred_alloc_blank - Allocate memory for new credentials. + * + * @new: Pointer to "struct cred". + * @gfp: Memory allocation flags. + * + * Returns 0 on success, negative value otherwise. + */ +static int tt_cred_alloc_blank(struct cred *new, gfp_t gfp) +{ + new->security = kzalloc(sizeof(struct tt_record), gfp); + return new->security ? 0 : -ENOMEM; +} + +/** + * tt_cred_prepare - Allocate memory for new credentials. + * + * @new: Pointer to "struct cred". + * @old: Pointer to "struct cred". + * @gfp: Memory allocation flags. + * + * Returns 0 on success, negative value otherwise. + */ +static int tt_cred_prepare(struct cred *new, const struct cred *old, + gfp_t gfp) +{ + if (tt_cred_alloc_blank(new, gfp)) + return -ENOMEM; + strcpy(tt_find_record(new)->history, tt_find_record(old)->history); + return 0; +} + +/** + * tt_cred_free - Release memory used by credentials. + * + * @cred: Pointer to "struct cred". + * + * Returns nothing. + */ +static void tt_cred_free(struct cred *cred) +{ + kfree(cred->security); +} + +/** + * tt_cred_transfer - Transfer "struct tt_record" between credentials. + * + * @new: Pointer to "struct cred". + * @old: Pointer to "struct cred". + * + * Returns nothing. + */ +static void tt_cred_transfer(struct cred *new, const struct cred *old) +{ + strcpy(tt_find_record(new)->history, tt_find_record(old)->history); +} + +/** + * tt_bprm_committing_creds - A hook which is called when do_execve() succeeded. + * + * @bprm: Pointer to "struct linux_binprm". + * + * Returns nothing. + */ +static void tt_bprm_committing_creds(struct linux_binprm *bprm) +{ + tt_update_record(tt_find_record(bprm->cred)); +} + +/** + * tt_task_getsecid - Check whether to audit or not. + * + * @p: Pointer to "struct task_struct". + * @secid: Pointer to flag. + */ +static void tt_task_getsecid(struct task_struct *p, u32 *secid) +{ + *secid = (p == current); +} + +/** + * tt_secid_to_secctx - Allocate memory used for auditing. + * + * @secid: Bool flag to allocate. + * @secdata: Pointer to allocate memory. + * @seclen: Size of allocated memory. + * + * Returns 0 on success, -EINVAL otherwise. + */ +static int tt_secid_to_secctx(u32 secid, char **secdata, u32 *seclen) +{ + struct tt_record *record; + /* Ignore unless current thread's record is requested. */ + if (secid != 1) + return -EINVAL; + /* + * We don't need to duplicate the string because current thread's + * record is updated upon only boot up and successful execve() + * operation, even if current thread's record is shared between + * multiple threads. + */ + record = tt_find_record(current->real_cred); + *secdata = record->history; + *seclen = strlen(record->history); + return 0; +} + +/* List of hooks. */ +static struct security_operations tasktracker_ops = { + .name = "tt", + .secid_to_secctx = tt_secid_to_secctx, + .task_getsecid = tt_task_getsecid, + .cred_prepare = tt_cred_prepare, + .cred_free = tt_cred_free, + .cred_alloc_blank = tt_cred_alloc_blank, + .cred_transfer = tt_cred_transfer, + .bprm_committing_creds = tt_bprm_committing_creds, +}; + +/** + * tt_init - Initialize this module. + * + * Returns 0 on success, panic otherwise. + */ +static int __init tt_init(void) +{ + struct cred *cred = (struct cred *) current_cred(); + if (!security_module_enable(&tasktracker_ops)) + return 0; + if (tt_cred_alloc_blank(cred, GFP_ATOMIC) || + register_security(&tasktracker_ops)) + panic("Failure registering TaskTracker"); + tt_update_record(tt_find_record(cred)); + pr_info("TaskTracker initialized\n"); + return 0; +} + +security_initcall(tt_init); -- 1.7.1

On Monday, June 23, 2014 09:14:35 PM Tetsuo Handa wrote: > Any comments on this proposal? subj= is the wrong way to record this. The subj field name is for process labels. When field names get re-used for different purposes, it causes lots of problems in being able to assign meaning and correctly use it in analysis. I would suggest using phist= for process history or something like that. Please don't re-use subj for this.

Also, the comm file is under control of the user. What if they create a program "sshd=>crond"? Would that throw off the analysis? How do you ensure user supplied names do not contain symbols that you are using to denote parentage?

Also, would you consider adding this information as a auxiliary record rather than as part of a syscall record? The advantage is it can be filtered or searched for. We recently did this for PROCTITLE information. Perhaps this fits better as a PROCHIST auxiliary record?

From d015533ce544feb8922fcbf023017c82bd79a9ac Mon Sep 17 00:00:00 2001

Steve Grubb wrote: > Hello, > > On Sun, 28 Sep 2014 00:13:14 +0900 > Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; wrote: > > Steve Grubb wrote: > > > On Sat, 27 Sep 2014 10:02:44 +0900 > > > Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; wrote: > > > > > > > May I continue proposing this functionality? > > > > > > From the audit perspective, sure. I think we were expecting a > > > revised patch after the comments. Other groups may have different > > > thoughts, though. > > > > > > -Steve > > > > OK, thank you. Before posting a revised patch, can I hear answers > > about specification questions listed below? > > Sure. > > > (Q1) Where can I find which bytes in $value need to be escaped when > > emitting a record like name='$value' ? > > I have written a specification that describes how to write well formed > audit events to help with questions like this. You can find it here: > > http://people.redhat.com/sgrubb/audit/audit-events.txt > > If you know that a field is under user control, it must be escaped so > that they cannot try to trick the parser. > > > Is 0x20 in $value permitted? > > No. That is the separator between fields, so it cannot be allowed. What > we suggest is to use a dash or hyphen between if you are logging a > phrase that cannot be altered by the user. For example, you may have an > op field saying it deleted a rule. You would do it as op=rule-deleted. > However, we do not suggest that for user controlled fields. Just escape > it by calling audit_log_untrustedstring() if in the kernel. There are > examples in the page I mention above. > > > (Q2) Does auxiliary record work with only type=SYSCALL case? > > Auxiliary records compliment a syscall record by adding extra > information. PATH, IPC, CWD, and EXECVE are some examples. They get > emitted in audit_log_exit() if you wanted to look at them in more > detail. > > HTH... Richard Guy Briggs wrote: > On 14/09/28, Tetsuo Handa wrote: > > (Q2) Does auxiliary record work with only type=SYSCALL case? > > Auxiliary records don't work with AUDIT_LOGIN because that record has a > NULL context. Similarly for core dumps (AUDIT_ANOM_ABEND), AUDIT_SECCOMP, > configuration changes (AUDIT_CONFIG_CHANGE, AUDIT_FEATURE_CHANGE), most > (all?) AUDIT_USER_* messages. > I see, thank you. Although I feel that, from the point of view of troubleshooting, emitting history of thread's comm name into NULL-context records would help sysadmin to map login session and operations a user did from that login session, I'm OK with starting history of thread's comm name as auxiliary records (i.e. not emitted into NULL-context records). Adding LKML for reviewers. What else can I do for merging this patch?

---------- >From 2beb93e870e9c1a6391d8706aa84a608b8353c2a Mon Sep 17 00:00:00 2001 From: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; Date: Sun, 28 Sep 2014 19:20:16 +0900 Subject: [PATCH] audit: Emit history of thread's comm name. When an unexpected system event (e.g. reboot) occurs, the administrator may want to identify which application triggered the event. System call auditing could be used for recording such event. However, the audit log may not be able to provide sufficient information for identifying the application because the audit log does not reflect how the program was executed. This patch adds ability to trace how the program was executed and emit it as an auxiliary record in the form of comm name and time stamp pairs as of execve(). type=UNKNOWN[1329] msg=audit(1403741314.019:22): history=' swapper/0(2014/06/26-09:06:04)=>init(2014/06/26-09:06:10)=> switch_root(2014/06/26-09:06:13)=>init(2014/06/26-09:06:13)=> sh(2014/06/26-00:06:27)=>rc(2014/06/26-00:06:27)=> S55sshd(2014/06/26-00:06:35)=>sshd(2014/06/26-00:06:35)=> sshd(2014/06/26-00:06:40)=>bash(2014/06/26-00:06:43)=> tail(2014/06/26-00:08:34)' Note that only char < 0x21, char > 0x7e, '\'', '\\' and '=' are escaped using \ooo style octal value rather than converting all characters to XX style hexadecimal value. Signed-off-by: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; --- fs/exec.c | 1 + include/linux/audit.h | 4 ++ include/linux/init_task.h | 5 ++ include/linux/sched.h | 3 + include/uapi/linux/audit.h | 1 + kernel/audit.c | 1 + kernel/auditsc.c | 113 ++++++++++++++++++++++++++++++++++++++++++++ 7 files changed, 128 insertions(+), 0 deletions(-) diff --git a/fs/exec.c b/fs/exec.c index a2b42a9..1e81709 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -1191,6 +1191,7 @@ void install_exec_creds(struct linux_binprm *bprm) commit_creds(bprm->cred); bprm->cred = NULL; + audit_update_history(); /* * Disable monitoring for regular users * when executing setuid binaries. Must diff --git a/include/linux/audit.h b/include/linux/audit.h index 22cfddb..97d08e1 100644 --- a/include/linux/audit.h +++ b/include/linux/audit.h @@ -206,6 +206,8 @@ static inline void audit_ptrace(struct task_struct *t) __audit_ptrace(t); } +extern void audit_update_history(void); + /* Private API (for audit.c only) */ extern unsigned int audit_serial(void); extern int auditsc_get_stamp(struct audit_context *ctx, @@ -419,6 +421,8 @@ static inline void audit_mmap_fd(int fd, int flags) { } static inline void audit_ptrace(struct task_struct *t) { } +static inline void audit_update_history(void) +{ } #define audit_n_rules 0 #define audit_signals 0 #endif /* CONFIG_AUDITSYSCALL */ diff --git a/include/linux/init_task.h b/include/linux/init_task.h index 2bb4c4f..7a5695b 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -98,8 +98,12 @@ extern struct group_info init_groups; #define INIT_IDS \ .loginuid = INVALID_UID, \ .sessionid = (unsigned int)-1, +extern char init_task_history[]; +#define INIT_THREAD_HISTORY \ + .comm_history = init_task_history, #else #define INIT_IDS +#define INIT_THREAD_HISTORY #endif #ifdef CONFIG_TREE_PREEMPT_RCU @@ -227,6 +231,7 @@ extern struct task_group root_task_group; INIT_CPUSET_SEQ(tsk) \ INIT_RT_MUTEXES(tsk) \ INIT_VTIME(tsk) \ + INIT_THREAD_HISTORY \ } diff --git a/include/linux/sched.h b/include/linux/sched.h index b867a4d..fd3cdaf 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1644,6 +1644,9 @@ struct task_struct { unsigned int sequential_io; unsigned int sequential_io_avg; #endif +#ifdef CONFIG_AUDITSYSCALL + char *comm_history; +#endif }; /* Future-safe accessor for struct task_struct's cpus_allowed. */ diff --git a/include/uapi/linux/audit.h b/include/uapi/linux/audit.h index 3b9ff33..a6a8ee8 100644 --- a/include/uapi/linux/audit.h +++ b/include/uapi/linux/audit.h @@ -110,6 +110,7 @@ #define AUDIT_SECCOMP 1326 /* Secure Computing event */ #define AUDIT_PROCTITLE 1327 /* Proctitle emit event */ #define AUDIT_FEATURE_CHANGE 1328 /* audit log listing feature changes */ +#define AUDIT_PROCHISTORY 1329 /* Commname history emit event */ #define AUDIT_AVC 1400 /* SE Linux avc denial or grant */ #define AUDIT_SELINUX_ERR 1401 /* Internal SE Linux Errors */ diff --git a/kernel/audit.c b/kernel/audit.c index ba2ff5a..252544a 100644 --- a/kernel/audit.c +++ b/kernel/audit.c @@ -1163,6 +1163,7 @@ static int __init audit_init(void) { int i; + audit_update_history(); if (audit_initialized == AUDIT_DISABLED) return 0; diff --git a/kernel/auditsc.c b/kernel/auditsc.c index 21eae3c..2e5ee14 100644 --- a/kernel/auditsc.c +++ b/kernel/auditsc.c @@ -85,6 +85,9 @@ /* max length to print of cmdline/proctitle value during audit */ #define MAX_PROCTITLE_AUDIT_LEN 128 +/* thread's comm name history length */ +#define COMM_HISTORY_SIZE 1024 + /* number of audit rules */ int audit_n_rules; @@ -950,6 +953,11 @@ int audit_alloc(struct task_struct *tsk) enum audit_state state; char *key = NULL; + tsk->comm_history = kmemdup(current->comm_history, COMM_HISTORY_SIZE, + GFP_KERNEL); + if (!tsk->comm_history) + return -ENOMEM; + if (likely(!audit_ever_enabled)) return 0; /* Return if not auditing. */ @@ -960,6 +968,8 @@ int audit_alloc(struct task_struct *tsk) } if (!(context = audit_alloc_context(state))) { + kfree(tsk->comm_history); + tsk->comm_history = NULL; kfree(key); audit_log_lost("out of memory in audit_alloc"); return -ENOMEM; @@ -1349,6 +1359,17 @@ out: audit_log_end(ab); } +static void audit_log_history(struct audit_context *context) +{ + struct audit_buffer *ab; + + ab = audit_log_start(context, GFP_KERNEL, AUDIT_PROCHISTORY); + if (!ab) + return; /* audit_panic or being filtered */ + audit_log_format(ab, "history='%s'", current->comm_history); + audit_log_end(ab); +} + static void audit_log_exit(struct audit_context *context, struct task_struct *tsk) { int i, call_panic = 0; @@ -1467,6 +1488,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts } audit_log_proctitle(tsk, context); + audit_log_history(context); /* Send end of event record to help user space know we are finished */ ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE); @@ -1486,6 +1508,8 @@ void __audit_free(struct task_struct *tsk) { struct audit_context *context; + kfree(tsk->comm_history); + tsk->comm_history = NULL; context = audit_take_context(tsk, 0, 0); if (!context) return; @@ -2503,3 +2527,92 @@ struct list_head *audit_killed_trees(void) return NULL; return &ctx->killed_trees; } + +char init_task_history[COMM_HISTORY_SIZE]; + +/** + * audit_update_history - Update current->comm_history field. + * + * Returns nothing. + * + * Update is done locklessly because current thread's history is updated by + * only current thread upon boot up and successful execve() operation, and + * we don't read other thread's history. + */ +void audit_update_history(void) +{ + static const u16 eom[2][12] = { + { 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365 }, + { 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335, 366 } + }; + u16 year = 1970; + u16 day; + u8 month; + u8 hour; + u8 minute; + u8 second; + bool r; + time_t now = get_seconds(); + char *history = current->comm_history; + int pos = strlen(history); + + /* Make some room by truncating old history. */ + while (pos >= COMM_HISTORY_SIZE - (TASK_COMM_LEN * 4 + 30)) { + char *cp = strchr(history + 1, '='); + + if (unlikely(!cp)) + return; + pos -= cp - history; + memmove(history, cp, pos + 1); + } + if (pos) { + history += pos; + *history++ = '='; + *history++ = '>'; + } + /* + * Read locklessly because this is current thread and being + * unexpectedly modified by other thread is not a fatal problem. + */ + for (pos = 0; pos < TASK_COMM_LEN; pos++) { + const unsigned char c = current->comm[pos]; + + if (!c) + break; + else if (c == '\'' || c == '\\' || c == '=' || c < 0x21 || + c > 0x7e) { + *history++ = '\\'; + *history++ = (c >> 6) + '0'; + *history++ = ((c >> 3) & 7) + '0'; + *history++ = (c & 7) + '0'; + } else + *history++ = c; + } + /* Append current time in "(YYYY/MM/DD-hh:mm:ss)" format. */ + second = now % 60; + now /= 60; + minute = now % 60; + now /= 60; + hour = now % 24; + day = now / 24; + if (day >= 16071) { + /* Start from 2014/01/01 rather than 1970/01/01. */ + day -= 16071; + year += 44; + } + while (1) { + const u16 days = (year & 3) ? 365 : 366; + + if (day < days) + break; + day -= days; + year++; + } + r = (year & 3) == 0; + for (month = 0; month < 11 && day >= eom[r][month]; month++) + ; + if (month) + day -= eom[r][month - 1]; + snprintf(history, 22, "(%04u/%02u/%02u-%02u:%02u:%02u)", year, + month + 1, day + 1, hour, minute, second); +} -- 1.7.1

Thank you for comments. Richard Guy Briggs wrote: > Steve already mentioned any user-influenced fields need to be escaped, > so I'd recommend audit_log_untrustedstring() as being much simpler from > your perspective and much better tested and understood from audit > maintainer's perspective. At least use the existing 'o' printf format > specifier instead of inventing your own. I do acknowledge that the > resulting output from your function is easier to read in its raw format > passed from the kernel, however, it makes your code harder to maintain. I'm not sure whether I should use audit_log_untrustedstring().

This record contains multiple user-influenced comm names. If I use audit_log_untrustedstring(), I would need to split this record into multiple records like history[0]='...' history[1]='...' history[2]='...' in order to avoid matching delimiters (i.e. ';', '=' and '>') used in this record.

This would also change from "char *" in "struct task_struct" to array of struct { "comm name", "pid", "stamp" } in "struct task_struct". I don't know whether such change makes easier to maintain than now.

> As for the date-stamping bits, they seem to be the majority of the code > in audit_update_history(). I'd just emit a number and punt that to > userspace for decoding. Alternatively, I'd use an existing service in > the kernel to do that date formatting, or at least call a new function > to format that date string should a suitable one not already exist, so > you can remove that complexity from audit_update_history(). Since I don't know existing functions for date formatting,

I split it as a new function. If it is acceptable, I'd like to make that function public and replace tomoyo_convert_time() in security/tomoyo/util.c with that function. Regards. ---------------------------------------- >From 50d59b5640a7501b8d5f843fb57283fcb62b1118 Mon Sep 17 00:00:00 2001 From: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; Date: Mon, 12 Jan 2015 14:45:23 +0900 Subject: [PATCH] audit: Emit history of thread's comm name. When an unexpected system event (e.g. reboot) occurs, the administrator may want to identify which application triggered the event. System call auditing could be used for recording such event. However, the audit log may not be able to provide sufficient information for identifying the application because the audit log does not reflect how the program was executed. This patch adds ability to trace how the program was executed and emit it as an auxiliary record in the form of comm name, pid and time stamp pairs as of execve(). type=UNKNOWN[1329] msg=audit(1421039813.810:3693): history=' name=swapper\0570;pid=1;start=20150112140720=>name=init;pid=1; start=20150112140721=>name=systemd;pid=1;start=20150112140722=> name=sshd;pid=2473;start=20150112050733=>name=sshd;pid=9838; start=20150112051105=>name=bash;pid=9840;start=20150112051108=> name=tail;pid=9876;start=20150112051653' Since converting all bytes using audit_log_untrustedstring() makes this record unparsable because this record includes multiple user-influenced comm names which may match delimiters used in this record (i.e. ';', '=' and '>'), only bytes which are not alphabets, numbers, '.', '_' nor '-' are escaped. Signed-off-by: Tetsuo Handa <penguin-kernel(a)I-love.SAKURA.ne.jp&gt; --- fs/exec.c | 1 + include/linux/audit.h | 4 ++ include/linux/init_task.h | 5 ++ include/linux/sched.h | 3 + include/uapi/linux/audit.h | 1 + kernel/audit.c | 1 + kernel/auditsc.c | 155 +++++++++++++++++++++++++++++++++++++++++++++ 7 files changed, 170 insertions(+) diff --git a/fs/exec.c b/fs/exec.c index ad8798e..5e92651 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -1200,6 +1200,7 @@ void install_exec_creds(struct linux_binprm *bprm) commit_creds(bprm->cred); bprm->cred = NULL; + audit_update_history(); /* * Disable monitoring for regular users * when executing setuid binaries. Must diff --git a/include/linux/audit.h b/include/linux/audit.h index af84234..74310a7 100644 --- a/include/linux/audit.h +++ b/include/linux/audit.h @@ -221,6 +221,8 @@ static inline void audit_ptrace(struct task_struct *t) __audit_ptrace(t); } +extern void audit_update_history(void); + /* Private API (for audit.c only) */ extern unsigned int audit_serial(void); extern int auditsc_get_stamp(struct audit_context *ctx, @@ -437,6 +439,8 @@ static inline void audit_mmap_fd(int fd, int flags) { } static inline void audit_ptrace(struct task_struct *t) { } +static inline void audit_update_history(void) +{ } #define audit_n_rules 0 #define audit_signals 0 #endif /* CONFIG_AUDITSYSCALL */ diff --git a/include/linux/init_task.h b/include/linux/init_task.h index 3037fc0..078823a 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -98,8 +98,12 @@ extern struct group_info init_groups; #define INIT_IDS \ .loginuid = INVALID_UID, \ .sessionid = (unsigned int)-1, +extern char init_task_history[]; +#define INIT_THREAD_HISTORY \ + .comm_history = init_task_history, #else #define INIT_IDS +#define INIT_THREAD_HISTORY #endif #ifdef CONFIG_PREEMPT_RCU @@ -247,6 +251,7 @@ extern struct task_group root_task_group; INIT_RT_MUTEXES(tsk) \ INIT_VTIME(tsk) \ INIT_NUMA_BALANCING(tsk) \ + INIT_THREAD_HISTORY \ } diff --git a/include/linux/sched.h b/include/linux/sched.h index 8db31ef..77539e4 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1701,6 +1701,9 @@ struct task_struct { #ifdef CONFIG_DEBUG_ATOMIC_SLEEP unsigned long task_state_change; #endif +#ifdef CONFIG_AUDITSYSCALL + char *comm_history; +#endif }; /* Future-safe accessor for struct task_struct's cpus_allowed. */ diff --git a/include/uapi/linux/audit.h b/include/uapi/linux/audit.h index d3475e1..93ad58c 100644 --- a/include/uapi/linux/audit.h +++ b/include/uapi/linux/audit.h @@ -110,6 +110,7 @@ #define AUDIT_SECCOMP 1326 /* Secure Computing event */ #define AUDIT_PROCTITLE 1327 /* Proctitle emit event */ #define AUDIT_FEATURE_CHANGE 1328 /* audit log listing feature changes

+#define AUDIT_PROCHISTORY 1329 /* Commname history emit event */ #define AUDIT_AVC 1400 /* SE Linux avc denial or grant */ #define AUDIT_SELINUX_ERR 1401 /* Internal SE Linux Errors */ diff --git a/kernel/audit.c b/kernel/audit.c index 72ab759..d45397e 100644 --- a/kernel/audit.c +++ b/kernel/audit.c @@ -1154,6 +1154,7 @@ static int __init audit_init(void) { int i; + audit_update_history(); if (audit_initialized == AUDIT_DISABLED) return 0; diff --git a/kernel/auditsc.c b/kernel/auditsc.c index 072566d..2edeba2 100644 --- a/kernel/auditsc.c +++ b/kernel/auditsc.c @@ -88,6 +88,9 @@ /* max length to print of cmdline/proctitle value during audit */ #define MAX_PROCTITLE_AUDIT_LEN 128 +/* thread's comm name history length */ +#define COMM_HISTORY_SIZE 1024 + /* number of audit rules */ int audit_n_rules; @@ -945,6 +948,11 @@ int audit_alloc(struct task_struct *tsk) enum audit_state state; char *key = NULL; + tsk->comm_history = kmemdup(current->comm_history, COMM_HISTORY_SIZE, + GFP_KERNEL); + if (!tsk->comm_history) + return -ENOMEM; + if (likely(!audit_ever_enabled)) return 0; /* Return if not auditing. */ @@ -955,6 +963,8 @@ int audit_alloc(struct task_struct *tsk) } if (!(context = audit_alloc_context(state))) { + kfree(tsk->comm_history); + tsk->comm_history = NULL; kfree(key); audit_log_lost("out of memory in audit_alloc"); return -ENOMEM; @@ -1344,6 +1354,17 @@ out: audit_log_end(ab); } +static void audit_log_history(struct audit_context *context) +{ + struct audit_buffer *ab; + + ab = audit_log_start(context, GFP_KERNEL, AUDIT_PROCHISTORY); + if (!ab) + return; /* audit_panic or being filtered */ + audit_log_format(ab, "history='%s'", current->comm_history); + audit_log_end(ab); +} + static void audit_log_exit(struct audit_context *context, struct task_struct *tsk) { int i, call_panic = 0; @@ -1462,6 +1483,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts } audit_log_proctitle(tsk, context); + audit_log_history(context); /* Send end of event record to help user space know we are finished */ ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE); @@ -1481,6 +1503,8 @@ void __audit_free(struct task_struct *tsk) { struct audit_context *context; + kfree(tsk->comm_history); + tsk->comm_history = NULL; context = audit_take_context(tsk, 0, 0); if (!context) return; @@ -2535,3 +2559,134 @@ struct list_head *audit_killed_trees(void) return NULL; return &ctx->killed_trees; } + +char init_task_history[COMM_HISTORY_SIZE]; + +/* Structure for representing YYYY/MM/DD hh/mm/ss. */ +struct yyyymmdd_hhmmss { + u16 year; + u8 month; + u8 day; + u8 hour; + u8 min; + u8 sec; +}; + +/** + * tt_get_time - Get current time in YYYY/MM/DD hh/mm/ss format. + * + * @stamp: Pointer to "struct yyyymmdd_hhmmss". + * + * Returns nothing. + * + * This function does not handle Y2038 problem. + */ +static void tt_get_time(struct yyyymmdd_hhmmss *stamp) +{ + static const u16 days_until_end_of_month[2][12] = { + { 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334, 365 }, + { 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335, 366 } + }; + u16 y = 1970; + u8 m; + bool r; + time_t time = get_seconds(); + + stamp->sec = time % 60; + time /= 60; + stamp->min = time % 60; + time /= 60; + stamp->hour = time % 24; + time /= 24; + if (time >= 16436) { + /* Start from 2015/01/01 rather than 1970/01/01. */ + time -= 16436; + y += 45; + } + while (1) { + const unsigned short days = (y & 3) ? 365 : 366; + + if (time < days) + break; + time -= days; + y++; + } + r = (y & 3) == 0; + for (m = 0; m < 11 && time >= days_until_end_of_month[r][m]; m++) + ; + if (m) + time -= days_until_end_of_month[r][m - 1]; + stamp->year = y; + stamp->month = ++m; + stamp->day = ++time; +} + +/** + * audit_update_history - Update current->comm_history field. + * + * Returns nothing. + * + * Update is done locklessly because current thread's history is updated by + * only current thread upon boot up and successful execve() operation, and + * we don't read other thread's history. + */ +void audit_update_history(void) +{ + char *cp; + int i; + int required; + char buf[256]; + + /* + * Lockless read because this is current thread and being unexpectedly + * modified by other thread is not a fatal problem. + */ + cp = buf; + cp += snprintf(buf, sizeof(buf) - 1, "name="); + for (i = 0; i < TASK_COMM_LEN; i++) { + const unsigned char c = current->comm[i]; + + if (!c) + break; + if (isalnum(c) || c == '.' || c == '_' || c == '-') { + *cp++ = c; + continue; + } + *cp++ = '\\'; + *cp++ = (c >> 6) + '0'; + *cp++ = ((c >> 3) & 7) + '0'; + *cp++ = (c & 7) + '0'; + } + /* Append PID. */ + cp += snprintf(cp, buf - cp + sizeof(buf) - 1, ";pid=%u", + current->pid); + /* Append timestamp. */ + { + struct yyyymmdd_hhmmss stamp; + + tt_get_time(&stamp); + cp += snprintf(cp, buf - cp + sizeof(buf) - 1, + ";start=%04u%02u%02u%02u%02u%02u", stamp.year, + stamp.month, stamp.day, stamp.hour, stamp.min, + stamp.sec); + } + /* Terminate the buffer. */ + if (cp >= buf + sizeof(buf)) + cp = buf + sizeof(buf) - 1; + *cp = '\0'; + required = cp - buf; + /* Make some room by truncating old history. */ + cp = current->comm_history; + while (i = strlen(cp), i + required >= COMM_HISTORY_SIZE - 10) { + char *cp2 = memchr(cp + 2, '>', i - 2); + + BUG_ON(!cp2); + cp2--; + memmove(cp, cp2, strlen(cp2) + 1); + } + /* Emit the buffer. */ + if (!i) + sprintf(cp, "%s", buf); + else + sprintf(cp + i, "=>%s", buf); +}

It also occurs to me that the pid_t of that process in init PID namespace would be useful for each entry.

> > As for the date-stamping bits, they seem to be the majority of the code > > in audit_update_history(). I'd just emit a number and punt that to > > userspace for decoding. Alternatively, I'd use an existing service in > > the kernel to do that date formatting, or at least call a new function > > to format that date string should a suitable one not already exist, so > > you can remove that complexity from audit_update_history(). > > Since I don't know existing functions for date formatting, All time in the audit system is "%lu.%03lu", t.tv_sec, t.tv_nsec/1000000. User space tooling expects this.

> This record contains multiple user-influenced comm names. If I use > audit_log_untrustedstring(), I would need to split this record into > multiple records like history[0]='...' history[1]='...' history[2]='...' > in order to avoid matching delimiters (i.e. ';', '=' and '>') used in > this record. That sounds like a good change to me. Audit records are always name=value with a space between fields. We need this to always stay like this because the tooling expects that format. There is nowhere in the audit logs we use =>.

> This would also change from "char *" in "struct task_struct" > to array of struct { "comm name", "pid", "stamp" } in "struct task_struct". > I don't know whether such change makes easier to maintain than now. You can still use char *, just use history[x]= to append with. We faced the same issue regarding argv[] logging. You might look at the execve record generation to get some ideas how that was handled.

This will end up producing a varying number of fields. The userspace tools are looking for a constant number of fields per record type.

On Monday, January 12, 2015 03:13:12 PM Tetsuo Handa wrote: > Thank you for comments. > > Richard Guy Briggs wrote: > > Steve already mentioned any user-influenced fields need to be escaped, > > so I'd recommend audit_log_untrustedstring() as being much simpler from > > your perspective and much better tested and understood from audit > > maintainer's perspective. At least use the existing 'o' printf format > > specifier instead of inventing your own. I do acknowledge that the > > resulting output from your function is easier to read in its raw format > > passed from the kernel, however, it makes your code harder to maintain. > > I'm not sure whether I should use audit_log_untrustedstring(). That is the accepted encoding. We do not want 2 different kinds of encoding functions.

This is where I would seperate them to: audit_log_format(ab, "history="); audit_log_untrustedstring(ab, current->comm_history);

From 637299dfae899270f7b4b2fbae35af680d19704d Mon Sep 17 00:00:00 2001