execute: support basic filesystem namespacing

This commit is contained in:
Lennart Poettering 2010-04-21 22:15:06 +02:00
parent 020379a7f7
commit 15ae422b74
15 changed files with 675 additions and 14 deletions

1
.gitignore vendored
View File

@ -1,3 +1,4 @@
test-ns
systemd-initctl.service
systemd-logger.service
systemd-cgroups-agent

View File

@ -52,7 +52,8 @@ pkglibexec_PROGRAMS = \
noinst_PROGRAMS = \
test-engine \
test-job-type
test-job-type \
test-ns
dbuspolicy_DATA = \
org.freedesktop.systemd1.conf
@ -161,7 +162,9 @@ COMMON_SOURCES= \
unit-name.c \
unit-name.h \
fdset.c \
fdset.h
fdset.h \
namespace.h \
namespace.c
systemd_SOURCES = \
$(COMMON_SOURCES) \
@ -192,6 +195,14 @@ test_job_type_SOURCES = \
test_job_type_CPPFLAGS = $(systemd_CPPFLAGS)
test_job_type_LDADD = $(systemd_LDADD)
test_ns_SOURCES = \
$(BASIC_SOURCES) \
test-ns.c \
namespace.c
test_ns_CPPFLAGS = $(systemd_CPPFLAGS)
test_ns_LDADD = $(systemd_LDADD)
systemd_logger_SOURCES = \
$(BASIC_SOURCES) \
logger.c

View File

@ -329,7 +329,7 @@ int config_parse_path(
assert(rvalue);
assert(data);
if (*rvalue != '/') {
if (!path_is_absolute(rvalue)) {
log_error("[%s:%u] Not an absolute path: %s", filename, line, rvalue);
return -EINVAL;
}
@ -394,3 +394,67 @@ fail:
return -ENOMEM;
}
int config_parse_path_strv(
const char *filename,
unsigned line,
const char *section,
const char *lvalue,
const char *rvalue,
void *data,
void *userdata) {
char*** sv = data;
char **n;
char *w;
unsigned k;
size_t l;
char *state;
int r;
assert(filename);
assert(lvalue);
assert(rvalue);
assert(data);
k = strv_length(*sv);
FOREACH_WORD_QUOTED(w, l, rvalue, state)
k++;
if (!(n = new(char*, k+1)))
return -ENOMEM;
k = 0;
if (*sv)
for (; (*sv)[k]; k++)
n[k] = (*sv)[k];
FOREACH_WORD_QUOTED(w, l, rvalue, state) {
if (!(n[k] = strndup(w, l))) {
r = -ENOMEM;
goto fail;
}
if (!path_is_absolute(n[k])) {
log_error("[%s:%u] Not an absolute path: %s", filename, line, rvalue);
r = -EINVAL;
goto fail;
}
k++;
}
n[k] = NULL;
free(*sv);
*sv = n;
return 0;
fail:
free(n[k]);
for (; k > 0; k--)
free(n[k-1]);
free(n);
return r;
}

View File

@ -50,5 +50,6 @@ int config_parse_bool(const char *filename, unsigned line, const char *section,
int config_parse_string(const char *filename, unsigned line, const char *section, const char *lvalue, const char *rvalue, void *data, void *userdata);
int config_parse_path(const char *filename, unsigned line, const char *section, const char *lvalue, const char *rvalue, void *data, void *userdata);
int config_parse_strv(const char *filename, unsigned line, const char *section, const char *lvalue, const char *rvalue, void *data, void *userdata);
int config_parse_path_strv(const char *filename, unsigned line, const char *section, const char *lvalue, const char *rvalue, void *data, void *userdata);
#endif

View File

@ -34,6 +34,7 @@
#include <sys/stat.h>
#include <grp.h>
#include <pwd.h>
#include <sys/mount.h>
#include "execute.h"
#include "strv.h"
@ -43,6 +44,7 @@
#include "ioprio.h"
#include "securebits.h"
#include "cgroup.h"
#include "namespace.h"
/* This assumes there is a 'tty' group */
#define TTY_MODE 0620
@ -794,8 +796,6 @@ int exec_spawn(ExecCommand *command,
goto fail;
}
umask(context->umask);
if (confirm_spawn) {
char response;
@ -900,6 +900,19 @@ int exec_spawn(ExecCommand *command,
goto fail;
}
if (strv_length(context->read_write_dirs) > 0 ||
strv_length(context->read_only_dirs) > 0 ||
strv_length(context->inaccessible_dirs) > 0 ||
context->mount_flags != MS_SHARED ||
context->private_tmp)
if ((r = setup_namespace(
context->read_write_dirs,
context->read_only_dirs,
context->inaccessible_dirs,
context->private_tmp,
context->mount_flags)) < 0)
goto fail;
if (context->user) {
username = context->user;
if (get_user_creds(&username, &uid, &gid, &home) < 0) {
@ -920,6 +933,8 @@ int exec_spawn(ExecCommand *command,
goto fail;
}
umask(context->umask);
if (apply_chroot) {
if (context->root_directory)
if (chroot(context->root_directory) < 0) {
@ -1066,6 +1081,7 @@ void exec_context_init(ExecContext *c) {
c->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 0);
c->cpu_sched_policy = SCHED_OTHER;
c->syslog_priority = LOG_DAEMON|LOG_INFO;
c->mount_flags = MS_SHARED;
}
void exec_context_done(ExecContext *c) {
@ -1105,6 +1121,15 @@ void exec_context_done(ExecContext *c) {
cap_free(c->capabilities);
c->capabilities = NULL;
}
strv_free(c->read_only_dirs);
c->read_only_dirs = NULL;
strv_free(c->read_write_dirs);
c->read_write_dirs = NULL;
strv_free(c->inaccessible_dirs);
c->inaccessible_dirs = NULL;
}
void exec_command_done(ExecCommand *c) {
@ -1143,6 +1168,15 @@ void exec_command_free_array(ExecCommand **c, unsigned n) {
}
}
static void strv_fprintf(FILE *f, char **l) {
char **g;
assert(f);
STRV_FOREACH(g, l)
fprintf(f, " %s", *g);
}
void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
char ** e;
unsigned i;
@ -1157,11 +1191,13 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
"%sUMask: %04o\n"
"%sWorkingDirectory: %s\n"
"%sRootDirectory: %s\n"
"%sNonBlocking: %s\n",
"%sNonBlocking: %s\n"
"%sPrivateTmp: %s\n",
prefix, c->umask,
prefix, c->working_directory ? c->working_directory : "/",
prefix, c->root_directory ? c->root_directory : "/",
prefix, yes_no(c->non_blocking));
prefix, yes_no(c->non_blocking),
prefix, yes_no(c->private_tmp));
if (c->environment)
for (e = c->environment; *e; e++)
@ -1269,14 +1305,27 @@ void exec_context_dump(ExecContext *c, FILE* f, const char *prefix) {
if (c->group)
fprintf(f, "%sGroup: %s", prefix, c->group);
if (c->supplementary_groups) {
char **g;
if (strv_length(c->supplementary_groups) > 0) {
fprintf(f, "%sSupplementaryGroups:", prefix);
strv_fprintf(f, c->supplementary_groups);
fputs("\n", f);
}
STRV_FOREACH(g, c->supplementary_groups)
fprintf(f, " %s", *g);
if (strv_length(c->read_write_dirs) > 0) {
fprintf(f, "%sReadWriteDirs:", prefix);
strv_fprintf(f, c->read_write_dirs);
fputs("\n", f);
}
if (strv_length(c->read_only_dirs) > 0) {
fprintf(f, "%sReadOnlyDirs:", prefix);
strv_fprintf(f, c->read_only_dirs);
fputs("\n", f);
}
if (strv_length(c->inaccessible_dirs) > 0) {
fprintf(f, "%sInaccessibleDirs:", prefix);
strv_fprintf(f, c->inaccessible_dirs);
fputs("\n", f);
}
}

View File

@ -109,6 +109,9 @@ struct ExecContext {
char *group;
char **supplementary_groups;
char **read_write_dirs, **read_only_dirs, **inaccessible_dirs;
unsigned long mount_flags;
uint64_t capability_bounding_set_drop;
cap_t capabilities;
@ -116,6 +119,7 @@ struct ExecContext {
bool cpu_sched_reset_on_fork;
bool non_blocking;
bool private_tmp;
bool oom_adjust_set:1;
bool nice_set:1;

View File

@ -27,6 +27,7 @@
#include <fcntl.h>
#include <sched.h>
#include <sys/prctl.h>
#include <sys/mount.h>
#include "unit.h"
#include "strv.h"
@ -909,6 +910,43 @@ static int config_parse_sysv_priority(
DEFINE_CONFIG_PARSE_ENUM(config_parse_kill_mode, kill_mode, KillMode, "Failed to parse kill mode");
static int config_parse_mount_flags(
const char *filename,
unsigned line,
const char *section,
const char *lvalue,
const char *rvalue,
void *data,
void *userdata) {
ExecContext *c = data;
char *w;
size_t l;
char *state;
unsigned long flags = 0;
assert(filename);
assert(lvalue);
assert(rvalue);
assert(data);
FOREACH_WORD(w, l, rvalue, state) {
if (strncmp(w, "shared", l) == 0)
flags |= MS_SHARED;
else if (strncmp(w, "slave", l) == 0)
flags |= MS_SLAVE;
else if (strncmp(w, "private", l) == 0)
flags |= MS_PRIVATE;
else {
log_error("[%s:%u] Failed to parse mount flags: %s", filename, line, rvalue);
return -EINVAL;
}
}
c->mount_flags = flags;
return 0;
}
#define FOLLOW_MAX 8
static int open_follow(char **filename, FILE **_f, Set *names, char **_final) {
@ -1149,7 +1187,12 @@ static int load_from_path(Unit *u, const char *path) {
{ "LimitNICE", config_parse_limit, &(context).rlimit[RLIMIT_NICE], section }, \
{ "LimitRTPRIO", config_parse_limit, &(context).rlimit[RLIMIT_RTPRIO], section }, \
{ "LimitRTTIME", config_parse_limit, &(context).rlimit[RLIMIT_RTTIME], section }, \
{ "ControlGroup", config_parse_cgroup, u, section }
{ "ControlGroup", config_parse_cgroup, u, section }, \
{ "ReadWriteDirectories", config_parse_path_strv, &(context).read_write_dirs, section }, \
{ "ReadOnlyDirectories", config_parse_path_strv, &(context).read_only_dirs, section }, \
{ "InaccessibleDirectories",config_parse_path_strv, &(context).inaccessible_dirs, section }, \
{ "PrivateTmp", config_parse_bool, &(context).private_tmp, section }, \
{ "MountFlags", config_parse_mount_flags, &(context), section }
const ConfigItem items[] = {
{ "Names", config_parse_names, u, "Unit" },

1
main.c
View File

@ -701,7 +701,6 @@ int main(int argc, char *argv[]) {
break;
case MANAGER_REEXECUTE:
if (prepare_reexecute(m, &serialization, &fds) < 0)
goto finish;

View File

@ -1,12 +1,39 @@
/*-*- Mode: C; c-basic-offset: 8 -*-*/
#ifndef foomissinghfoo
#define foomissinghfoo
/***
This file is part of systemd.
Copyright 2010 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
systemd is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
/* Missing glibc definitions to access certain kernel APIs */
#include <sys/resource.h>
#include <sys/syscall.h>
#ifndef RLIMIT_RTTIME
#define RLIMIT_RTTIME 15
#endif
static inline int pivot_root(const char *new_root, const char *put_old) {
return syscall(SYS_pivot_root, new_root, put_old);
}
#endif

334
namespace.c Normal file
View File

@ -0,0 +1,334 @@
/*-*- Mode: C; c-basic-offset: 8 -*-*/
/***
This file is part of systemd.
Copyright 2010 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
systemd is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <errno.h>
#include <sys/mount.h>
#include <string.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sched.h>
#include <sys/syscall.h>
#include <limits.h>
#include "strv.h"
#include "util.h"
#include "namespace.h"
#include "missing.h"
typedef enum PathMode {
/* This is ordered by priority! */
INACCESSIBLE,
READONLY,
PRIVATE,
READWRITE
} PathMode;
typedef struct Path {
const char *path;
PathMode mode;
} Path;
static int append_paths(Path **p, char **strv, PathMode mode) {
char **i;
STRV_FOREACH(i, strv) {
if (!path_is_absolute(*i))
return -EINVAL;
(*p)->path = *i;
(*p)->mode = mode;
(*p)++;
}
return 0;
}
static int path_compare(const void *a, const void *b) {
const Path *p = a, *q = b;
if (path_equal(p->path, q->path)) {
/* If the paths are equal, check the mode */
if (p->mode < q->mode)
return -1;
if (p->mode > q->mode)
return 1;
return 0;
}
/* If the paths are not equal, then order prefixes first */
if (path_startswith(p->path, q->path))
return 1;
if (path_startswith(q->path, p->path))
return -1;
return 0;
}
static void drop_duplicates(Path *p, unsigned *n, bool *need_inaccessible, bool *need_private) {
Path *f, *t, *previous;
assert(p);
assert(n);
assert(need_inaccessible);
assert(need_private);
for (f = p, t = p, previous = NULL; f < p+*n; f++) {
if (previous && path_equal(f->path, previous->path))
continue;
t->path = f->path;
t->mode = f->mode;
if (t->mode == PRIVATE)
*need_private = true;
if (t->mode == INACCESSIBLE)
*need_inaccessible = true;
previous = t;
t++;
}
*n = t - p;
}
static int apply_mount(Path *p, const char *root_dir, const char *inaccessible_dir, const char *private_dir, unsigned long flags) {
const char *what;
char *where;
int r;
bool read_only = false;
assert(p);
assert(root_dir);
assert(inaccessible_dir);
assert(private_dir);
if (!(where = strappend(root_dir, p->path)))
return -ENOMEM;
switch (p->mode) {
case INACCESSIBLE:
what = inaccessible_dir;
read_only = true;
break;
case READONLY:
read_only = true;
/* Fall through */
case READWRITE:
what = p->path;
break;
case PRIVATE:
what = private_dir;
break;
}
if ((r = mount(what, where, NULL, MS_BIND|MS_REC, NULL)) >= 0) {
log_debug("Successfully mounted %s to %s", what, where);
/* The bind mount will always inherit the original
* flags. If we want to set any flag we need
* to do so in a second indepdant step. */
if (flags)
r = mount(NULL, where, NULL, MS_REMOUNT|MS_REC|flags, NULL);
/* Avoid expontial growth of trees */
if (r >= 0 && path_equal(p->path, "/"))
r = mount(NULL, where, NULL, MS_REMOUNT|MS_UNBINDABLE, NULL);
if (r >= 0 && read_only)
r = mount(NULL, where, NULL, MS_REMOUNT|MS_RDONLY, NULL);
if (r < 0) {
r = -errno;
umount2(where, MNT_DETACH);
}
}
free(where);
return r;
}
int setup_namespace(
char **writable,
char **readable,
char **inaccessible,
bool private_tmp,
unsigned long flags) {
char
tmp_dir[] = "/tmp/systemd-namespace-XXXXXX",
root_dir[] = "/tmp/systemd-namespace-XXXXXX/root",
old_root_dir[] = "/tmp/systemd-namespace-XXXXXX/root/tmp/old-root-XXXXXX",
inaccessible_dir[] = "/tmp/systemd-namespace-XXXXXX/inaccessible",
private_dir[] = "/tmp/systemd-namespace-XXXXXX/private";
Path *paths, *p;
unsigned n;
bool need_private = false, need_inaccessible = false;
bool remove_tmp = false, remove_root = false, remove_old_root = false, remove_inaccessible = false, remove_private = false;
int r;
const char *t;
n =
strv_length(writable) +
strv_length(readable) +
strv_length(inaccessible) +
(private_tmp ? 2 : 1);
if (!(paths = new(Path, n)))
return -ENOMEM;
p = paths;
if ((r = append_paths(&p, writable, READWRITE)) < 0 ||
(r = append_paths(&p, readable, READONLY)) < 0 ||
(r = append_paths(&p, inaccessible, INACCESSIBLE)) < 0)
goto fail;
if (private_tmp) {
p->path = "/tmp";
p->mode = PRIVATE;
p++;
}
p->path = "/";
p->mode = READWRITE;
p++;
assert(paths + n == p);
qsort(paths, n, sizeof(Path), path_compare);
drop_duplicates(paths, &n, &need_inaccessible, &need_private);
if (!mkdtemp(tmp_dir)) {
r = -errno;
goto fail;
}
remove_tmp = true;
memcpy(root_dir, tmp_dir, sizeof(tmp_dir)-1);
if (mkdir(root_dir, 0777) < 0) {
r = -errno;
goto fail;
}
remove_root = true;
if (need_inaccessible) {
memcpy(inaccessible_dir, tmp_dir, sizeof(tmp_dir)-1);
if (mkdir(inaccessible_dir, 0) < 0) {
r = -errno;
goto fail;
}
remove_inaccessible = true;
}
if (need_private) {
memcpy(private_dir, tmp_dir, sizeof(tmp_dir)-1);
if (mkdir(private_dir, 0777 + S_ISVTX) < 0) {
r = -errno;
goto fail;
}
remove_private = true;
}
if (unshare(CLONE_NEWNS) < 0) {
r = -errno;
goto fail;
}
/* We assume that by default mount events from us won't be
* propagated to the root namespace. */
for (p = paths; p < paths + n; p++)
if ((r = apply_mount(p, root_dir, inaccessible_dir, private_dir, flags)) < 0)
goto undo_mounts;
memcpy(old_root_dir, tmp_dir, sizeof(tmp_dir)-1);
if (!mkdtemp(old_root_dir)) {
r = -errno;
goto undo_mounts;
}
remove_old_root = true;
if (chdir(root_dir) < 0) {
r = -errno;
goto undo_mounts;
}
if (pivot_root(root_dir, old_root_dir) < 0) {
r = -errno;
goto undo_mounts;
}
t = old_root_dir + sizeof(root_dir) - 1;
if (umount2(t, MNT_DETACH) < 0)
/* At this point it's too late to turn anything back,
* since we are already in the new root. */
return -errno;
if (rmdir(t) < 0)
return -errno;
return 0;
undo_mounts:
for (p--; p >= paths; p--) {
char full_path[PATH_MAX];
snprintf(full_path, sizeof(full_path), "%s%s", root_dir, p->path);
char_array_0(full_path);
umount2(full_path, MNT_DETACH);
}
fail:
if (remove_old_root)
rmdir(old_root_dir);
if (remove_inaccessible)
rmdir(inaccessible_dir);
if (remove_private)
rmdir(private_dir);
if (remove_root)
rmdir(root_dir);
if (remove_tmp)
rmdir(tmp_dir);
free(paths);
return r;
}

34
namespace.h Normal file
View File

@ -0,0 +1,34 @@
/*-*- Mode: C; c-basic-offset: 8 -*-*/
#ifndef foonamespacehfoo
#define foonamespacehfoo
/***
This file is part of systemd.
Copyright 2010 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
systemd is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <stdbool.h>
int setup_namespace(
char **writable,
char **readable,
char **inaccessible,
bool private_tmp,
unsigned long flags);
#endif

3
strv.h
View File

@ -22,6 +22,9 @@
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <stdarg.h>
#include <stdbool.h>
#include "macro.h"
char *strv_find(char **l, const char *name);

57
test-ns.c Normal file
View File

@ -0,0 +1,57 @@
/*-*- Mode: C; c-basic-offset: 8 -*-*/
/***
This file is part of systemd.
Copyright 2010 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
systemd is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/mount.h>
#include "namespace.h"
#include "log.h"
int main(int argc, char *argv[]) {
const char * const writable[] = {
"/home",
NULL
};
const char * const readable[] = {
"/var",
NULL
};
const char * const inaccessible[] = {
"/home/lennart/projects",
NULL
};
int r;
if ((r = setup_namespace((char**) writable, (char**) readable, (char**) inaccessible, true, MS_SHARED)) < 0) {
log_error("Failed to setup namespace: %s", strerror(-r));
return 1;
}
execl("/bin/sh", "/bin/sh", NULL);
log_error("execl(): %m");
return 1;
}

33
util.c
View File

@ -1140,6 +1140,39 @@ bool path_startswith(const char *path, const char *prefix) {
}
}
bool path_equal(const char *a, const char *b) {
assert(a);
assert(b);
if ((a[0] == '/') != (b[0] == '/'))
return false;
for (;;) {
size_t j, k;
a += strspn(a, "/");
b += strspn(b, "/");
if (*a == 0 && *b == 0)
return true;
if (*a == 0 || *b == 0)
return false;
j = strcspn(a, "/");
k = strcspn(b, "/");
if (j != k)
return false;
if (memcmp(a, b, j) != 0)
return false;
a += j;
b += k;
}
}
char *ascii_strlower(char *t) {
char *p;

1
util.h
View File

@ -157,6 +157,7 @@ char *cunescape(const char *s);
char *path_kill_slashes(char *path);
bool path_startswith(const char *path, const char *prefix);
bool path_equal(const char *a, const char *b);
char *ascii_strlower(char *path);