This patchset includes following function points:
1: Let usermodehelper function possible to set pid namespace
done by: [PATCH v4 1/3] Make call_usermodehelper_exec possible
to set pid namespace.
2: Let pipe_type core_pattern write dump into container's rootfs
done by: [PATCH v4 2/3] Limit dump_pipe program's permission to
init for container.
2: Make separate core_pattern setting for each container
done by: [PATCH v4 3/3] Make core_pattern support namespace
3: Compatibility with current system
also included in: [PATCH v4 3/3] Make core_pattern support namespace
If container hadn't change core_pattern setting, it will keep
same setting with host.
Test:
1: Pass a test script for each function of this patchset
## TEST IN HOST ##
[root@kerneldev dumptest]# ./test_host
Set file core_pattern: OK
./test_host: line 41: 2366 Segmentation fault (core dumped) "$SCRI=
PT_BASE_DIR"/make_dump
Checking dumpfile: OK
Set file core_pattern: OK
./test_host: line 41: 2369 Segmentation fault (core dumped) "$SCRI=
PT_BASE_DIR"/make_dump
Checking dump_pipe triggered: OK
Checking rootfs: OK
Checking dumpfile: OK
Checking namespace: OK
Checking process list: OK
Checking capabilities: OK
## TEST IN GUEST ##
# ./test
Segmentation fault (core dumped)
Checking dump_pipe triggered: OK
Checking rootfs: OK
Checking dumpfile: OK
Checking namespace: OK
Checking process list: OK
Checking cg pids: OK
Checking capabilities: OK
[ 64.940734] make_dump[2432]: segfault at 0 ip 000000000040049d sp 000=
07ffc4af025f0 error 6 in make_dump[400000+a6000]
#
2: Pass other test(which is not easy to do in script) by hand.
Changelog v3.1-v4:
1. remove extra fork pointed out by:
Andrei Vagin <[email protected]>
Changelog v3-v3.1:
1. Switch "pwd" of pipe program to container's root fs.
2. Rebase on top of v4.9-rc1
Changelog v2->v3:
1: Fix problem of setting pid namespace, pointed out by:
Andrei Vagin <[email protected]>
Changelog v1(RFC)->v2:
1: Add [PATCH 2/2] which was todo in [RFC v1].
2: Pass a test script for each function.
3: Rebase on top of v4.7.
Suggested-by: Eric W. Biederman <[email protected]>
Suggested-by: KOSAKI Motohiro <[email protected]>
Signed-off-by: Zhao Lei <[email protected]>
Signed-off-by: Cao Shufeng <[email protected]>
Cao Shufeng (2):
Make call_usermodehelper_exec possible to set namespaces
Limit dump_pipe program's permission to init for container
Zhao Lei (1):
Make core_pattern support namespace
fs/coredump.c | 150 +++++++++++++++++++++++++++++++++++++++---
include/linux/binfmts.h | 2 +
include/linux/kmod.h | 4 ++
include/linux/pid_namespace.h | 3 +
init/do_mounts_initrd.c | 3 +-
kernel/kmod.c | 43 +++++++++---
kernel/pid.c | 2 +
kernel/pid_namespace.c | 2 +
kernel/sysctl.c | 50 ++++++++++++--
lib/kobject_uevent.c | 3 +-
security/keys/request_key.c | 4 +-
11 files changed, 241 insertions(+), 25 deletions(-)
--
2.7.4
From: Zhao Lei <[email protected]>
Currently, each container shared one copy of coredump setting
with the host system, if host system changed the setting, each
running containers will be affected.
Same story happened when container changed core_pattern, both
host and other container will be affected.
For container based on namespace design, it is good to allow
each container keeping their own coredump setting.
It will bring us following benefit:
1: Each container can change their own coredump setting
based on operation on /proc/sys/kernel/core_pattern
2: Coredump setting changed in host will not affect
running containers.
3: Support both case of "putting coredump in guest" and
"putting curedump in host".
Each namespace-based software(lxc, docker, ..) can use this function
to custom their dump setting.
And this function makes each continer working as separate system,
it fit for design goal of namespace.
Test(in lxc):
# In the host
# ----------------
# echo host_core >/proc/sys/kernel/core_pattern
# cat /proc/sys/kernel/core_pattern
host_core
# ulimit -c 1024000
# ./make_dump
Segmentation fault (core dumped)
# ls -l
-rw------- 1 root root 331776 Feb 4 18:02 host_core.2175
-rwxr-xr-x 1 root root 759731 Feb 4 18:01 make_dump
#
# In the container
# ----------------
# cat /proc/sys/kernel/core_pattern
host_core
# echo container_core >/proc/sys/kernel/core_pattern
# ./make_dump
Segmentation fault (core dumped)
# ls -l
-rwxr-xr-x 1 root root 759731 Feb 4 10:45 make_dump
-rw------- 1 root root 331776 Feb 4 10:45 container_core.16
#
# Return to host
# ----------------
# cat /proc/sys/kernel/core_pattern
host_core
# ls
host_core.2175 make_dump make_dump.c
# rm -f host_core.2175
# ./make_dump
Segmentation fault (core dumped)
# ls -l
-rw------- 1 root root 331776 Feb 4 18:49 host_core.2351
-rwxr-xr-x 1 root root 759731 Feb 4 18:01 make_dump
#
Signed-off-by: Zhao Lei <[email protected]>
---
fs/coredump.c | 25 ++++++++++++++++------
include/linux/pid_namespace.h | 3 +++
kernel/pid.c | 2 ++
kernel/pid_namespace.c | 2 ++
kernel/sysctl.c | 50 ++++++++++++++++++++++++++++++++++++++-----
5 files changed, 70 insertions(+), 12 deletions(-)
diff --git a/fs/coredump.c b/fs/coredump.c
index aa2ef6c..f97a987 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -49,7 +49,6 @@
int core_uses_pid;
unsigned int core_pipe_limit;
-char core_pattern[CORENAME_MAX_SIZE] = "core";
static int core_name_size = CORENAME_MAX_SIZE;
struct core_name {
@@ -57,8 +56,6 @@ struct core_name {
int used, size;
};
-/* The maximal length of core_pattern is also specified in sysctl.c */
-
static int expand_corename(struct core_name *cn, int size)
{
char *corename = krealloc(cn->corename, size, GFP_KERNEL);
@@ -183,10 +180,10 @@ static int cn_print_exe_file(struct core_name *cn)
* name into corename, which must have space for at least
* CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
*/
-static int format_corename(struct core_name *cn, struct coredump_params *cprm)
+static int format_corename(struct core_name *cn, const char *pat_ptr,
+ struct coredump_params *cprm)
{
const struct cred *cred = current_cred();
- const char *pat_ptr = core_pattern;
int ispipe = (*pat_ptr == '|');
int pid_in_pattern = 0;
int err = 0;
@@ -663,6 +660,8 @@ void do_coredump(const siginfo_t *siginfo)
*/
.mm_flags = mm->flags,
};
+ struct pid_namespace *pid_ns;
+ char core_pattern[CORENAME_MAX_SIZE];
audit_core_dumps(siginfo->si_signo);
@@ -672,6 +671,18 @@ void do_coredump(const siginfo_t *siginfo)
if (!__get_dumpable(cprm.mm_flags))
goto fail;
+ pid_ns = task_active_pid_ns(current);
+ spin_lock(&pid_ns->core_pattern_lock);
+ while (pid_ns != &init_pid_ns) {
+ if (pid_ns->core_pattern[0])
+ break;
+ spin_unlock(&pid_ns->core_pattern_lock);
+ pid_ns = pid_ns->parent,
+ spin_lock(&pid_ns->core_pattern_lock);
+ }
+ strcpy(core_pattern, pid_ns->core_pattern);
+ spin_unlock(&pid_ns->core_pattern_lock);
+
cred = prepare_creds();
if (!cred)
goto fail;
@@ -693,7 +704,7 @@ void do_coredump(const siginfo_t *siginfo)
old_cred = override_creds(cred);
- ispipe = format_corename(&cn, &cprm);
+ ispipe = format_corename(&cn, core_pattern, &cprm);
if (ispipe) {
int dump_count;
@@ -740,7 +751,7 @@ void do_coredump(const siginfo_t *siginfo)
}
rcu_read_lock();
- vinit_task = find_task_by_vpid(1);
+ vinit_task = find_task_by_pid_ns(1, pid_ns);
rcu_read_unlock();
if (!vinit_task) {
printk(KERN_WARNING "failed getting init task info, skipping core dump\n");
diff --git a/include/linux/pid_namespace.h b/include/linux/pid_namespace.h
index 34cce96..f0b0c21 100644
--- a/include/linux/pid_namespace.h
+++ b/include/linux/pid_namespace.h
@@ -9,6 +9,7 @@
#include <linux/nsproxy.h>
#include <linux/kref.h>
#include <linux/ns_common.h>
+#include <linux/binfmts.h>
struct pidmap {
atomic_t nr_free;
@@ -46,6 +47,8 @@ struct pid_namespace {
int hide_pid;
int reboot; /* group exit code if this pidns was rebooted */
struct ns_common ns;
+ spinlock_t core_pattern_lock;
+ char core_pattern[CORENAME_MAX_SIZE];
};
extern struct pid_namespace init_pid_ns;
diff --git a/kernel/pid.c b/kernel/pid.c
index f66162f..e7ee122 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c
@@ -83,6 +83,8 @@ struct pid_namespace init_pid_ns = {
#ifdef CONFIG_PID_NS
.ns.ops = &pidns_operations,
#endif
+ .core_pattern_lock = __SPIN_LOCK_UNLOCKED(init_pid_ns.core_pattern_lock),
+ .core_pattern = "core",
};
EXPORT_SYMBOL_GPL(init_pid_ns);
diff --git a/kernel/pid_namespace.c b/kernel/pid_namespace.c
index df9e8e9..c70ca5d 100644
--- a/kernel/pid_namespace.c
+++ b/kernel/pid_namespace.c
@@ -137,6 +137,8 @@ static struct pid_namespace *create_pid_namespace(struct user_namespace *user_ns
for (i = 1; i < PIDMAP_ENTRIES; i++)
atomic_set(&ns->pidmap[i].nr_free, BITS_PER_PAGE);
+ spin_lock_init(&ns->core_pattern_lock);
+
return ns;
out_free_map:
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 706309f..c73ced5 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -484,7 +484,7 @@ static struct ctl_table kern_table[] = {
},
{
.procname = "core_pattern",
- .data = core_pattern,
+ .data = NULL,
.maxlen = CORENAME_MAX_SIZE,
.mode = 0644,
.proc_handler = proc_dostring_coredump,
@@ -2401,12 +2401,20 @@ int proc_dointvec_minmax(struct ctl_table *table, int write,
static void validate_coredump_safety(void)
{
#ifdef CONFIG_COREDUMP
+ struct pid_namespace *pid_ns = task_active_pid_ns(current);
+ const char *core_pattern;
+
+ spin_lock(&pid_ns->core_pattern_lock);
+ core_pattern = pid_ns->core_pattern;
+
if (suid_dumpable == SUID_DUMP_ROOT &&
core_pattern[0] != '/' && core_pattern[0] != '|') {
printk(KERN_WARNING "Unsafe core_pattern used with "\
"suid_dumpable=2. Pipe handler or fully qualified "\
"core dump path required.\n");
}
+
+ spin_unlock(&pid_ns->core_pattern_lock);
#endif
}
@@ -2423,10 +2431,42 @@ static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
static int proc_dostring_coredump(struct ctl_table *table, int write,
void __user *buffer, size_t *lenp, loff_t *ppos)
{
- int error = proc_dostring(table, write, buffer, lenp, ppos);
- if (!error)
- validate_coredump_safety();
- return error;
+ int ret;
+ char core_pattern[CORENAME_MAX_SIZE];
+ struct pid_namespace *pid_ns = task_active_pid_ns(current);
+
+ if (write) {
+ if (*ppos && sysctl_writes_strict == SYSCTL_WRITES_WARN)
+ warn_sysctl_write(table);
+
+ ret = _proc_do_string(core_pattern, table->maxlen, write,
+ (char __user *)buffer, lenp, ppos);
+ if (ret)
+ return ret;
+
+ spin_lock(&pid_ns->core_pattern_lock);
+ strcpy(pid_ns->core_pattern, core_pattern);
+ spin_unlock(&pid_ns->core_pattern_lock);
+ } else {
+ spin_lock(&pid_ns->core_pattern_lock);
+ while (pid_ns != &init_pid_ns) {
+ if (pid_ns->core_pattern[0])
+ break;
+ spin_unlock(&pid_ns->core_pattern_lock);
+ pid_ns = pid_ns->parent,
+ spin_lock(&pid_ns->core_pattern_lock);
+ }
+ strcpy(core_pattern, pid_ns->core_pattern);
+ spin_unlock(&pid_ns->core_pattern_lock);
+
+ ret = _proc_do_string(core_pattern, table->maxlen, write,
+ (char __user *)buffer, lenp, ppos);
+ if (ret)
+ return ret;
+ }
+
+ validate_coredump_safety();
+ return 0;
}
#endif
--
2.7.4
Current call_usermodehelper_work() can not set namespaces for
the executed program.
This patch add above function for call_usermodehelper_work().
The init_intermediate is introduced for init works which should
be done before fork(). So that we get a method to set namespaces
for children. The cleanup_intermediate is introduced for cleaning
up what we have done in init_intermediate, like switching back
the namespace.
This function is helpful for coredump to run pipe_program in
specific container environment.
Signed-off-by: Cao Shufeng <[email protected]>
Co-author-by: Zhao Lei <[email protected]>
---
fs/coredump.c | 3 ++-
include/linux/kmod.h | 4 ++++
init/do_mounts_initrd.c | 3 ++-
kernel/kmod.c | 43 +++++++++++++++++++++++++++++++++++--------
lib/kobject_uevent.c | 3 ++-
security/keys/request_key.c | 4 ++--
6 files changed, 47 insertions(+), 13 deletions(-)
diff --git a/fs/coredump.c b/fs/coredump.c
index 281b768..52f2ed6 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -641,7 +641,8 @@ void do_coredump(const siginfo_t *siginfo)
retval = -ENOMEM;
sub_info = call_usermodehelper_setup(helper_argv[0],
helper_argv, NULL, GFP_KERNEL,
- umh_pipe_setup, NULL, &cprm);
+ NULL, NULL, umh_pipe_setup,
+ NULL, &cprm);
if (sub_info)
retval = call_usermodehelper_exec(sub_info,
UMH_WAIT_EXEC);
diff --git a/include/linux/kmod.h b/include/linux/kmod.h
index fcfd2bf..994e429 100644
--- a/include/linux/kmod.h
+++ b/include/linux/kmod.h
@@ -61,6 +61,8 @@ struct subprocess_info {
char **envp;
int wait;
int retval;
+ void (*init_intermediate)(struct subprocess_info *info);
+ void (*cleanup_intermediate)(struct subprocess_info *info);
int (*init)(struct subprocess_info *info, struct cred *new);
void (*cleanup)(struct subprocess_info *info);
void *data;
@@ -71,6 +73,8 @@ call_usermodehelper(char *path, char **argv, char **envp, int wait);
extern struct subprocess_info *
call_usermodehelper_setup(char *path, char **argv, char **envp, gfp_t gfp_mask,
+ void (*init_intermediate)(struct subprocess_info *info),
+ void (*cleanup_intermediate)(struct subprocess_info *info),
int (*init)(struct subprocess_info *info, struct cred *new),
void (*cleanup)(struct subprocess_info *), void *data);
diff --git a/init/do_mounts_initrd.c b/init/do_mounts_initrd.c
index a1000ca..59d11c9 100644
--- a/init/do_mounts_initrd.c
+++ b/init/do_mounts_initrd.c
@@ -72,7 +72,8 @@ static void __init handle_initrd(void)
current->flags |= PF_FREEZER_SKIP;
info = call_usermodehelper_setup("/linuxrc", argv, envp_init,
- GFP_KERNEL, init_linuxrc, NULL, NULL);
+ GFP_KERNEL, NULL, NULL, init_linuxrc,
+ NULL, NULL);
if (!info)
return;
call_usermodehelper_exec(info, UMH_WAIT_PROC);
diff --git a/kernel/kmod.c b/kernel/kmod.c
index 0277d12..42f5a74 100644
--- a/kernel/kmod.c
+++ b/kernel/kmod.c
@@ -91,7 +91,8 @@ static int call_modprobe(char *module_name, int wait)
argv[4] = NULL;
info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
- NULL, free_modprobe_argv, NULL);
+ NULL, NULL, NULL, free_modprobe_argv,
+ NULL);
if (!info)
goto free_module_name;
@@ -301,6 +302,9 @@ static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
/* Restore default kernel sig handler */
kernel_sigaction(SIGCHLD, SIG_IGN);
+ if(sub_info->cleanup_intermediate) {
+ sub_info->cleanup_intermediate(sub_info);
+ }
umh_complete(sub_info);
}
@@ -322,6 +326,9 @@ static void call_usermodehelper_exec_work(struct work_struct *work)
{
struct subprocess_info *sub_info =
container_of(work, struct subprocess_info, work);
+ if(sub_info->init_intermediate) {
+ sub_info->init_intermediate(sub_info);
+ }
if (sub_info->wait & UMH_WAIT_PROC) {
call_usermodehelper_exec_sync(sub_info);
@@ -334,6 +341,11 @@ static void call_usermodehelper_exec_work(struct work_struct *work)
*/
pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
CLONE_PARENT | SIGCHLD);
+
+ if(sub_info->cleanup_intermediate) {
+ sub_info->cleanup_intermediate(sub_info);
+ }
+
if (pid < 0) {
sub_info->retval = pid;
umh_complete(sub_info);
@@ -499,25 +511,38 @@ static void helper_unlock(void)
* @argv: arg vector for process
* @envp: environment for process
* @gfp_mask: gfp mask for memory allocation
- * @cleanup: a cleanup function
+ * @init_intermediate: init function which is called in parent task
+ * @cleanup_intermediate: clean function which is called in parent task
* @init: an init function
+ * @cleanup: a cleanup function
* @data: arbitrary context sensitive data
*
* Returns either %NULL on allocation failure, or a subprocess_info
* structure. This should be passed to call_usermodehelper_exec to
* exec the process and free the structure.
*
- * The init function is used to customize the helper process prior to
- * exec. A non-zero return code causes the process to error out, exit,
- * and return the failure to the calling process
+ * The init_intermediate is called in the parent task of user mode
+ * helper. It's designed for init works which must be done in
+ * parent task, like switching the pid_ns_for_children.
+ *
+ * The cleanup_intermediate is used when we want to cleanup what
+ * we have done in init_intermediate, it is also called in parent
+ * task.
+ *
+ * The init function is called after fork. It is used to customize the
+ * helper process prior to exec. A non-zero return code causes the
+ * process to error out, exit, and return the failure to the
+ * calling process.
*
- * The cleanup function is just before ethe subprocess_info is about to
+ * The cleanup function is just before the subprocess_info is about to
* be freed. This can be used for freeing the argv and envp. The
* Function must be runnable in either a process context or the
* context in which call_usermodehelper_exec is called.
*/
struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
char **envp, gfp_t gfp_mask,
+ void (*init_intermediate)(struct subprocess_info *info),
+ void (*cleanup_intermediate)(struct subprocess_info *info),
int (*init)(struct subprocess_info *info, struct cred *new),
void (*cleanup)(struct subprocess_info *info),
void *data)
@@ -532,8 +557,10 @@ struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
sub_info->argv = argv;
sub_info->envp = envp;
- sub_info->cleanup = cleanup;
+ sub_info->init_intermediate = init_intermediate;
+ sub_info->cleanup_intermediate = cleanup_intermediate;
sub_info->init = init;
+ sub_info->cleanup = cleanup;
sub_info->data = data;
out:
return sub_info;
@@ -619,7 +646,7 @@ int call_usermodehelper(char *path, char **argv, char **envp, int wait)
gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
- NULL, NULL, NULL);
+ NULL, NULL, NULL, NULL, NULL);
if (info == NULL)
return -ENOMEM;
diff --git a/lib/kobject_uevent.c b/lib/kobject_uevent.c
index f6c2c1e..7a7c57a 100644
--- a/lib/kobject_uevent.c
+++ b/lib/kobject_uevent.c
@@ -345,7 +345,8 @@ int kobject_uevent_env(struct kobject *kobj, enum kobject_action action,
retval = -ENOMEM;
info = call_usermodehelper_setup(env->argv[0], env->argv,
env->envp, GFP_KERNEL,
- NULL, cleanup_uevent_env, env);
+ NULL, NULL, NULL,
+ cleanup_uevent_env, env);
if (info) {
retval = call_usermodehelper_exec(info, UMH_NO_WAIT);
env = NULL; /* freed by cleanup_uevent_env */
diff --git a/security/keys/request_key.c b/security/keys/request_key.c
index 43affcf..51dfb38 100644
--- a/security/keys/request_key.c
+++ b/security/keys/request_key.c
@@ -78,8 +78,8 @@ static int call_usermodehelper_keys(char *path, char **argv, char **envp,
struct subprocess_info *info;
info = call_usermodehelper_setup(path, argv, envp, GFP_KERNEL,
- umh_keys_init, umh_keys_cleanup,
- session_keyring);
+ NULL, NULL, umh_keys_init,
+ umh_keys_cleanup, session_keyring);
if (!info)
return -ENOMEM;
--
2.7.4
Currently when we set core_pattern to a pipe, the pipe program is
forked by kthread running with root's permission, and write dumpfile
into host's filesystem.
Same thing happened for container, the dumper and dumpfile are also
in host(not in container).
It have following program:
1: Not consistent with file_type core_pattern
When we set core_pattern to a file, the container will write dump
into container's filesystem instead of host.
2: Not safe for privileged container
In a privileged container, user can destroy host system by following
command:
# # In a container
# echo "|/bin/dd of=/boot/vmlinuz" >/proc/sys/kernel/core_pattern
# make_dump
This patch switch dumper program's environment to init task, so, for
container, dumper program have same environment with init task in
container, which make dumper program put in container's filesystem, and
write coredump into container's filesystem.
The dumper's permission is also limited into subset of container's init
process.
Suggested-by: Eric W. Biederman <[email protected]>
Suggested-by: KOSAKI Motohiro <[email protected]>
Signed-off-by: Cao ShuFeng<[email protected]>
---
fs/coredump.c | 126 +++++++++++++++++++++++++++++++++++++++++++++++-
include/linux/binfmts.h | 2 +
2 files changed, 126 insertions(+), 2 deletions(-)
diff --git a/fs/coredump.c b/fs/coredump.c
index 52f2ed6..aa2ef6c 100644
--- a/fs/coredump.c
+++ b/fs/coredump.c
@@ -502,6 +502,45 @@ static void wait_for_dump_helpers(struct file *file)
}
/*
+ * umh_ns_setup
+ * set the namesapces to the bask task of a container.
+ * we need to switch back to the original namespaces
+ * so that the thread of workqueue is not influlenced.
+ *
+ * this method runs in workqueue kernel thread.
+ */
+static void umh_ns_setup(struct subprocess_info *info)
+{
+ struct coredump_params *cp = (struct coredump_params *)info->data;
+ struct task_struct *base_task = cp->base_task;
+
+ if (base_task) {
+ cp->current_task_nsproxy = current->nsproxy;
+ //prevent current namespace from being freed
+ get_nsproxy(current->nsproxy);
+ /* Set namespaces to base_task */
+ get_nsproxy(base_task->nsproxy);
+ switch_task_namespaces(current, base_task->nsproxy);
+ }
+}
+
+/*
+ * umh_ns_cleanup
+ * cleanup what we have done in umh_ns_setup.
+ *
+ * this method runs in workqueue kernel thread.
+ */
+static void umh_ns_cleanup(struct subprocess_info *info)
+{
+ struct coredump_params *cp = (struct coredump_params *)info->data;
+ struct nsproxy *current_task_nsproxy = cp->current_task_nsproxy;
+ if (current_task_nsproxy) {
+ /* switch workqueue's original namespace back */
+ switch_task_namespaces(current, current_task_nsproxy);
+ }
+}
+
+/*
* umh_pipe_setup
* helper function to customize the process used
* to collect the core in userspace. Specifically
@@ -516,6 +555,8 @@ static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
{
struct file *files[2];
struct coredump_params *cp = (struct coredump_params *)info->data;
+ struct task_struct *base_task;
+
int err = create_pipe_files(files, 0);
if (err)
return err;
@@ -524,10 +565,76 @@ static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
err = replace_fd(0, files[0], 0);
fput(files[0]);
+ if (err)
+ return err;
+
/* and disallow core files too */
current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
- return err;
+ base_task = cp->base_task;
+ if (base_task) {
+ const struct cred *base_cred;
+
+ /* Set fs_root to base_task */
+ spin_lock(&base_task->fs->lock);
+ set_fs_root(current->fs, &base_task->fs->root);
+ set_fs_pwd(current->fs, &base_task->fs->pwd);
+ spin_unlock(&base_task->fs->lock);
+
+ /* Set cgroup to base_task */
+ current->flags &= ~PF_NO_SETAFFINITY;
+ err = cgroup_attach_task_all(base_task, current);
+ if (err < 0)
+ return err;
+
+ /* Set cred to base_task */
+ base_cred = get_task_cred(base_task);
+
+ new->uid = base_cred->uid;
+ new->gid = base_cred->gid;
+ new->suid = base_cred->suid;
+ new->sgid = base_cred->sgid;
+ new->euid = base_cred->euid;
+ new->egid = base_cred->egid;
+ new->fsuid = base_cred->fsuid;
+ new->fsgid = base_cred->fsgid;
+
+ new->securebits = base_cred->securebits;
+
+ new->cap_inheritable = base_cred->cap_inheritable;
+ new->cap_permitted = base_cred->cap_permitted;
+ new->cap_effective = base_cred->cap_effective;
+ new->cap_bset = base_cred->cap_bset;
+ new->cap_ambient = base_cred->cap_ambient;
+
+ security_cred_free(new);
+#ifdef CONFIG_SECURITY
+ new->security = NULL;
+#endif
+ err = security_prepare_creds(new, base_cred, GFP_KERNEL);
+ if (err < 0) {
+ put_cred(base_cred);
+ return err;
+ }
+
+ free_uid(new->user);
+ new->user = base_cred->user;
+ get_uid(new->user);
+
+ put_user_ns(new->user_ns);
+ new->user_ns = base_cred->user_ns;
+ get_user_ns(new->user_ns);
+
+ put_group_info(new->group_info);
+ new->group_info = base_cred->group_info;
+ get_group_info(new->group_info);
+
+ put_cred(base_cred);
+
+ validate_creds(new);
+ }
+
+ return 0;
}
void do_coredump(const siginfo_t *siginfo)
@@ -590,6 +697,7 @@ void do_coredump(const siginfo_t *siginfo)
if (ispipe) {
int dump_count;
+ struct task_struct *vinit_task;
char **helper_argv;
struct subprocess_info *sub_info;
@@ -631,6 +739,15 @@ void do_coredump(const siginfo_t *siginfo)
goto fail_dropcount;
}
+ rcu_read_lock();
+ vinit_task = find_task_by_vpid(1);
+ rcu_read_unlock();
+ if (!vinit_task) {
+ printk(KERN_WARNING "failed getting init task info, skipping core dump\n");
+ goto fail_dropcount;
+ }
+
+
helper_argv = argv_split(GFP_KERNEL, cn.corename, NULL);
if (!helper_argv) {
printk(KERN_WARNING "%s failed to allocate memory\n",
@@ -638,15 +755,20 @@ void do_coredump(const siginfo_t *siginfo)
goto fail_dropcount;
}
+ get_task_struct(vinit_task);
+
+ cprm.base_task = vinit_task;
+
retval = -ENOMEM;
sub_info = call_usermodehelper_setup(helper_argv[0],
helper_argv, NULL, GFP_KERNEL,
- NULL, NULL, umh_pipe_setup,
+ umh_ns_setup, umh_ns_cleanup, umh_pipe_setup,
NULL, &cprm);
if (sub_info)
retval = call_usermodehelper_exec(sub_info,
UMH_WAIT_EXEC);
+ put_task_struct(vinit_task);
argv_free(helper_argv);
if (retval) {
printk(KERN_INFO "Core dump to |%s pipe failed\n",
diff --git a/include/linux/binfmts.h b/include/linux/binfmts.h
index 1303b57..7ba4271 100644
--- a/include/linux/binfmts.h
+++ b/include/linux/binfmts.h
@@ -59,6 +59,8 @@ struct linux_binprm {
/* Function parameter for binfmt->coredump */
struct coredump_params {
+ struct task_struct *base_task;
+ struct nsproxy *current_task_nsproxy;
const siginfo_t *siginfo;
struct pt_regs *regs;
struct file *file;
--
2.7.4
On Tue, Oct 25, 2016 at 03:28:56PM +0800, Cao Shufeng wrote:
> From: Zhao Lei <[email protected]>
> It will bring us following benefit:
> 1: Each container can change their own coredump setting
> based on operation on /proc/sys/kernel/core_pattern
> 2: Coredump setting changed in host will not affect
> running containers.
> 3: Support both case of "putting coredump in guest" and
> "putting curedump in host".
Would you explain more about case #3 here? In particular, I'm curious
what the impact is for systems that have already configured core_pattern
with the understanding that the program might be invoked to handle
either a host or a container core. In particular, is there any way to
specify that the container handler fall back to the host handler?
On the systems that I've configured, /proc/sys is mounted read-only in the
container. The host has a special program run from core_pattern that
determines which container generated the core. It then stores the cores
in a directory that uniquely identifies the container. The cores are
isolated on their own filesystem, and given a quota per-container. The
eventual goal is to have a service evacuate the cores to an object store
where we can make them available to the customer via a web service.
Does your change still allow a global handler in the host to process
cores from containers? Or is that behavior removed completely?
-K
ping
在 2016-10-25二的 15:28 +0800，Cao Shufeng写道：
> This patchset includes following function points:
> 1: Let usermodehelper function possible to set pid namespace
> done by: [PATCH v4 1/3] Make call_usermodehelper_exec possible
> to set pid namespace.
> 2: Let pipe_type core_pattern write dump into container's rootfs
> done by: [PATCH v4 2/3] Limit dump_pipe program's permission to
> init for container.
> 2: Make separate core_pattern setting for each container
> done by: [PATCH v4 3/3] Make core_pattern support namespace
> 3: Compatibility with current system
> also included in: [PATCH v4 3/3] Make core_pattern support namespace
> If container hadn't change core_pattern setting, it will keep
> same setting with host.
>
> Test:
> 1: Pass a test script for each function of this patchset
> ## TEST IN HOST ##
> [root@kerneldev dumptest]# ./test_host
> Set file core_pattern: OK
> ./test_host: line 41: 2366 Segmentation fault (core dumped) "$SCRI=
> PT_BASE_DIR"/make_dump
> Checking dumpfile: OK
> Set file core_pattern: OK
> ./test_host: line 41: 2369 Segmentation fault (core dumped) "$SCRI=
> PT_BASE_DIR"/make_dump
> Checking dump_pipe triggered: OK
> Checking rootfs: OK
> Checking dumpfile: OK
> Checking namespace: OK
> Checking process list: OK
> Checking capabilities: OK
>
> ## TEST IN GUEST ##
> # ./test
> Segmentation fault (core dumped)
> Checking dump_pipe triggered: OK
> Checking rootfs: OK
> Checking dumpfile: OK
> Checking namespace: OK
> Checking process list: OK
> Checking cg pids: OK
> Checking capabilities: OK
> [ 64.940734] make_dump[2432]: segfault at 0 ip 000000000040049d sp 000=
> 07ffc4af025f0 error 6 in make_dump[400000+a6000]
> #
> 2: Pass other test(which is not easy to do in script) by hand.
>
> Changelog v3.1-v4:
> 1. remove extra fork pointed out by:
> Andrei Vagin <[email protected]>
>
> Changelog v3-v3.1:
> 1. Switch "pwd" of pipe program to container's root fs.
> 2. Rebase on top of v4.9-rc1
>
> Changelog v2->v3:
> 1: Fix problem of setting pid namespace, pointed out by:
> Andrei Vagin <[email protected]>
>
> Changelog v1(RFC)->v2:
> 1: Add [PATCH 2/2] which was todo in [RFC v1].
> 2: Pass a test script for each function.
> 3: Rebase on top of v4.7.
>
> Suggested-by: Eric W. Biederman <[email protected]>
> Suggested-by: KOSAKI Motohiro <[email protected]>
> Signed-off-by: Zhao Lei <[email protected]>
> Signed-off-by: Cao Shufeng <[email protected]>
>
> Cao Shufeng (2):
> Make call_usermodehelper_exec possible to set namespaces
> Limit dump_pipe program's permission to init for container
>
> Zhao Lei (1):
> Make core_pattern support namespace
>
> fs/coredump.c | 150 +++++++++++++++++++++++++++++++++++++++---
> include/linux/binfmts.h | 2 +
> include/linux/kmod.h | 4 ++
> include/linux/pid_namespace.h | 3 +
> init/do_mounts_initrd.c | 3 +-
> kernel/kmod.c | 43 +++++++++---
> kernel/pid.c | 2 +
> kernel/pid_namespace.c | 2 +
> kernel/sysctl.c | 50 ++++++++++++--
> lib/kobject_uevent.c | 3 +-
> security/keys/request_key.c | 4 +-
> 11 files changed, 241 insertions(+), 25 deletions(-)
>
--
Best Regards,
Cao Shufeng
--------------------------------------------------
Cao Shufeng
Development Dept.I
Nanjing Fujitsu Nanda Software Tech. Co., Ltd.(FNST)
No.6 Wenzhu Road, Nanjing, 210012, China
TEL: +86+25-86630566-8552
FUJITSU INTERNAL: 7998-8552
EMail: [email protected]