* Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
* Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
*
- * This program 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; only version 2 of the License.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2 only,
+ * as published by the Free Software Foundation.
*
* This program 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
+ * 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
- * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
- * Place - Suite 330, Boston, MA 02111-1307, USA.
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
-#define _GNU_SOURCE
+#define _LGPL_SOURCE
#include <errno.h>
#include <limits.h>
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <sched.h>
-#include <sys/signal.h>
+#include <signal.h>
+#include <assert.h>
+#include <signal.h>
-#include <common/error.h>
-#include <common/compat/mman.h>
-#include <common/compat/clone.h>
+#include <common/lttng-kernel.h>
+#include <common/common.h>
+#include <common/utils.h>
+#include <common/compat/getenv.h>
+#include <common/compat/prctl.h>
+#include <common/unix.h>
+#include <common/defaults.h>
+#include <common/lttng-elf.h>
+
+#include <lttng/constant.h>
#include "runas.h"
-#define RUNAS_CHILD_STACK_SIZE 10485760
+struct run_as_data;
+struct run_as_ret;
+typedef int (*run_as_fct)(struct run_as_data *data, struct run_as_ret *ret_value);
-#ifndef MAP_STACK
-#define MAP_STACK 0
-#endif
+struct run_as_mkdir_data {
+ char path[PATH_MAX];
+ mode_t mode;
+};
+
+struct run_as_open_data {
+ char path[PATH_MAX];
+ int flags;
+ mode_t mode;
+};
+
+struct run_as_unlink_data {
+ char path[PATH_MAX];
+};
+
+struct run_as_rmdir_recursive_data {
+ char path[PATH_MAX];
+};
+
+struct run_as_extract_elf_symbol_offset_data {
+ char function[LTTNG_SYMBOL_NAME_LEN];
+};
+
+struct run_as_extract_sdt_probe_offsets_data {
+ char probe_name[LTTNG_SYMBOL_NAME_LEN];
+ char provider_name[LTTNG_SYMBOL_NAME_LEN];
+};
+
+struct run_as_mkdir_ret {
+ int ret;
+};
+
+struct run_as_open_ret {
+ int ret;
+};
+
+struct run_as_unlink_ret {
+ int ret;
+};
+
+struct run_as_rmdir_recursive_ret {
+ int ret;
+};
+
+struct run_as_extract_elf_symbol_offset_ret {
+ uint64_t offset;
+};
+
+struct run_as_extract_sdt_probe_offsets_ret {
+ uint32_t num_offset;
+ uint64_t offsets[LTTNG_KERNEL_MAX_UPROBE_NUM];
+};
+
+enum run_as_cmd {
+ RUN_AS_MKDIR,
+ RUN_AS_OPEN,
+ RUN_AS_UNLINK,
+ RUN_AS_RMDIR_RECURSIVE,
+ RUN_AS_MKDIR_RECURSIVE,
+ RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET,
+ RUN_AS_EXTRACT_SDT_PROBE_OFFSETS,
+};
struct run_as_data {
- int (*cmd)(void *data);
- void *data;
+ enum run_as_cmd cmd;
+ int fd;
+ union {
+ struct run_as_mkdir_data mkdir;
+ struct run_as_open_data open;
+ struct run_as_unlink_data unlink;
+ struct run_as_rmdir_recursive_data rmdir_recursive;
+ struct run_as_extract_elf_symbol_offset_data extract_elf_symbol_offset;
+ struct run_as_extract_sdt_probe_offsets_data extract_sdt_probe_offsets;
+ } u;
uid_t uid;
gid_t gid;
- int retval_pipe;
};
-struct run_as_mkdir_data {
- const char *path;
- mode_t mode;
+/*
+ * The run_as_ret structure holds the returned value and status of the command.
+ *
+ * The `u` union field holds the return value of the command; in most cases it
+ * represents the success or the failure of the command. In more complex
+ * commands, it holds a computed value.
+ *
+ * The _errno field is the errno recorded after the execution of the command.
+ *
+ * The _error fields is used the signify that return status of the command. For
+ * simple commands returning `int` the _error field will be the same as the
+ * ret_int field. In complex commands, it signify the success or failure of the
+ * command.
+ *
+ */
+struct run_as_ret {
+ int fd;
+ union {
+ struct run_as_mkdir_ret mkdir;
+ struct run_as_open_ret open;
+ struct run_as_unlink_ret unlink;
+ struct run_as_rmdir_recursive_ret rmdir_recursive;
+ struct run_as_extract_elf_symbol_offset_ret extract_elf_symbol_offset;
+ struct run_as_extract_sdt_probe_offsets_ret extract_sdt_probe_offsets;
+ } u;
+ int _errno;
+ bool _error;
};
-struct run_as_open_data {
- const char *path;
- int flags;
- mode_t mode;
+struct run_as_worker {
+ pid_t pid; /* Worker PID. */
+ int sockpair[2];
+ char *procname;
};
+/* Single global worker per process (for now). */
+static struct run_as_worker *global_worker;
+/* Lock protecting the worker. */
+static pthread_mutex_t worker_lock = PTHREAD_MUTEX_INITIALIZER;
+
+#ifdef VALGRIND
+static
+int use_clone(void)
+{
+ return 0;
+}
+#else
+static
+int use_clone(void)
+{
+ return !lttng_secure_getenv("LTTNG_DEBUG_NOCLONE");
+}
+#endif
+
+LTTNG_HIDDEN
+int _utils_mkdir_recursive_unsafe(const char *path, mode_t mode);
+
/*
* Create recursively directory using the FULL path.
*/
static
-int _mkdir_recursive(void *_data)
+int _mkdir_recursive(struct run_as_data *data, struct run_as_ret *ret_value)
{
- struct run_as_mkdir_data *data = _data;
const char *path;
- char *p, tmp[PATH_MAX];
- struct stat statbuf;
mode_t mode;
- size_t len;
- int ret;
- path = data->path;
- mode = data->mode;
+ path = data->u.mkdir.path;
+ mode = data->u.mkdir.mode;
+
+ /* Safe to call as we have transitioned to the requested uid/gid. */
+ ret_value->u.mkdir.ret = _utils_mkdir_recursive_unsafe(path, mode);
+ ret_value->_errno = errno;
+ ret_value->_error = (ret_value->u.mkdir.ret) ? true : false;
+ return ret_value->u.mkdir.ret;
+}
+
+static
+int _mkdir(struct run_as_data *data, struct run_as_ret *ret_value)
+{
+ ret_value->u.mkdir.ret = mkdir(data->u.mkdir.path, data->u.mkdir.mode);
+ ret_value->_errno = errno;
+ ret_value->_error = (ret_value->u.mkdir.ret) ? true : false;
+ return ret_value->u.mkdir.ret;
+}
+
+static
+int _open(struct run_as_data *data, struct run_as_ret *ret_value)
+{
+ ret_value->u.open.ret = open(data->u.open.path, data->u.open.flags, data->u.open.mode);
+ ret_value->fd = ret_value->u.open.ret;
+ ret_value->_errno = errno;
+ ret_value->_error = (ret_value->u.open.ret) ? true : false;
+ return ret_value->u.open.ret;
+}
+
+static
+int _unlink(struct run_as_data *data, struct run_as_ret *ret_value)
+{
+ ret_value->u.unlink.ret = unlink(data->u.unlink.path);
+ ret_value->_errno = errno;
+ ret_value->_error = (ret_value->u.unlink.ret) ? true : false;
+ return ret_value->u.unlink.ret;
+}
+
+static
+int _rmdir_recursive(struct run_as_data *data, struct run_as_ret *ret_value)
+{
+ ret_value->u.rmdir_recursive.ret = utils_recursive_rmdir(data->u.rmdir_recursive.path);
+ ret_value->_errno = errno;
+ ret_value->_error = (ret_value->u.rmdir_recursive.ret) ? true : false;
+ return ret_value->u.rmdir_recursive.ret;
+}
+
+static
+int _extract_elf_symbol_offset(struct run_as_data *data,
+ struct run_as_ret *ret_value)
+{
+ int ret = 0;
+ ret_value->_error = false;
+
+ ret = lttng_elf_get_symbol_offset(data->fd,
+ data->u.extract_elf_symbol_offset.function,
+ &ret_value->u.extract_elf_symbol_offset.offset);
+ if (ret) {
+ DBG("Failed to extract ELF function offset");
+ ret_value->_error = true;
+ }
+
+ return ret;
+}
+
+static
+int _extract_sdt_probe_offsets(struct run_as_data *data,
+ struct run_as_ret *ret_value)
+{
+ int ret = 0;
+ uint64_t *offsets = NULL;
+ uint32_t num_offset;
+
+ ret_value->_error = false;
+
+ /* On success, this call allocates the offsets paramater. */
+ ret = lttng_elf_get_sdt_probe_offsets(data->fd,
+ data->u.extract_sdt_probe_offsets.provider_name,
+ data->u.extract_sdt_probe_offsets.probe_name,
+ &offsets, &num_offset);
+
+ if (ret) {
+ DBG("Failed to extract SDT probe offsets");
+ ret_value->_error = true;
+ goto end;
+ }
+
+ if (num_offset <= 0 || num_offset > LTTNG_KERNEL_MAX_UPROBE_NUM) {
+ DBG("Wrong number of probes.");
+ ret = -1;
+ ret_value->_error = true;
+ goto free_offset;
+ }
+
+ /* Copy the content of the offsets array to the ret struct. */
+ memcpy(ret_value->u.extract_sdt_probe_offsets.offsets,
+ offsets, num_offset * sizeof(uint64_t));
+
+ ret_value->u.extract_sdt_probe_offsets.num_offset = num_offset;
+
+free_offset:
+ free(offsets);
+end:
+ return ret;
+}
+
+static
+run_as_fct run_as_enum_to_fct(enum run_as_cmd cmd)
+{
+ switch (cmd) {
+ case RUN_AS_MKDIR:
+ return _mkdir;
+ case RUN_AS_OPEN:
+ return _open;
+ case RUN_AS_UNLINK:
+ return _unlink;
+ case RUN_AS_RMDIR_RECURSIVE:
+ return _rmdir_recursive;
+ case RUN_AS_MKDIR_RECURSIVE:
+ return _mkdir_recursive;
+ case RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET:
+ return _extract_elf_symbol_offset;
+ case RUN_AS_EXTRACT_SDT_PROBE_OFFSETS:
+ return _extract_sdt_probe_offsets;
+ default:
+ ERR("Unknown command %d", (int) cmd);
+ return NULL;
+ }
+}
- ret = snprintf(tmp, sizeof(tmp), "%s", path);
+static
+int do_send_fd(int sock, int fd)
+{
+ ssize_t len;
+
+ if (fd < 0) {
+ ERR("Invalid file description");
+ return 0;
+ }
+
+ len = lttcomm_send_fds_unix_sock(sock, &fd, 1);
+ if (len < 0) {
+ PERROR("lttcomm_send_fds_unix_sock");
+ return -1;
+ }
+ return 0;
+}
+
+static
+int do_recv_fd(int sock, int *fd)
+{
+ ssize_t len;
+
+ if (*fd < 0) {
+ ERR("Invalid file description");
+ return 0;
+ }
+
+ len = lttcomm_recv_fds_unix_sock(sock, fd, 1);
+
+ if (!len) {
+ return -1;
+ } else if (len < 0) {
+ PERROR("lttcomm_recv_fds_unix_sock");
+ return -1;
+ }
+ return 0;
+}
+
+static
+int send_fd_to_worker(struct run_as_worker *worker, enum run_as_cmd cmd, int fd)
+{
+ int ret = 0;
+
+ switch (cmd) {
+ case RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET:
+ case RUN_AS_EXTRACT_SDT_PROBE_OFFSETS:
+ break;
+ default:
+ return 0;
+ }
+
+ ret = do_send_fd(worker->sockpair[0], fd);
if (ret < 0) {
- PERROR("snprintf mkdir");
- goto error;
- }
-
- len = ret;
- if (tmp[len - 1] == '/') {
- tmp[len - 1] = 0;
- }
-
- for (p = tmp + 1; *p; p++) {
- if (*p == '/') {
- *p = 0;
- ret = stat(tmp, &statbuf);
- if (ret < 0) {
- ret = mkdir(tmp, mode);
- if (ret < 0) {
- if (!(errno == EEXIST)) {
- PERROR("mkdir recursive");
- ret = -errno;
- goto error;
- }
- }
- }
- *p = '/';
- }
+ PERROR("do_send_fd");
+ ret = -1;
+ }
+
+ return ret;
+}
+
+static
+int send_fd_to_master(struct run_as_worker *worker, enum run_as_cmd cmd, int fd)
+{
+ int ret = 0, ret_close = 0;
+
+ switch (cmd) {
+ case RUN_AS_OPEN:
+ break;
+ default:
+ return 0;
}
- ret = mkdir(tmp, mode);
+ ret = do_send_fd(worker->sockpair[1], fd);
if (ret < 0) {
- if (!(errno == EEXIST)) {
- PERROR("mkdir recursive last piece");
- ret = -errno;
- } else {
- ret = 0;
- }
+ PERROR("do_send_fd error");
+ ret = -1;
+ }
+
+ ret_close = close(fd);
+ if (ret_close < 0) {
+ PERROR("close");
}
-error:
return ret;
}
static
-int _mkdir(void *_data)
+int recv_fd_from_worker(struct run_as_worker *worker, enum run_as_cmd cmd, int *fd)
{
- struct run_as_mkdir_data *data = _data;
- return mkdir(data->path, data->mode);
+ int ret = 0;
+
+ switch (cmd) {
+ case RUN_AS_OPEN:
+ break;
+ default:
+ return 0;
+ }
+
+ ret = do_recv_fd(worker->sockpair[0], fd);
+ if (ret < 0) {
+ PERROR("do_recv_fd error");
+ ret = -1;
+ }
+
+ return ret;
}
static
-int _open(void *_data)
+int recv_fd_from_master(struct run_as_worker *worker, enum run_as_cmd cmd, int *fd)
{
- struct run_as_open_data *data = _data;
- return open(data->path, data->flags, data->mode);
+ int ret = 0;
+
+ switch (cmd) {
+ case RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET:
+ case RUN_AS_EXTRACT_SDT_PROBE_OFFSETS:
+ break;
+ default:
+ return 0;
+ }
+
+ ret = do_recv_fd(worker->sockpair[1], fd);
+ if (ret < 0) {
+ PERROR("do_recv_fd error");
+ ret = -1;
+ }
+
+ return ret;
}
static
-int child_run_as(void *_data)
+int cleanup_received_fd(enum run_as_cmd cmd, int fd)
{
- struct run_as_data *data = _data;
- size_t writelen, writeleft, index;
- union {
- int i;
- char c[sizeof(int)];
- } sendret;
- int ret;
+ int ret = 0;
+
+ switch (cmd) {
+ case RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET:
+ case RUN_AS_EXTRACT_SDT_PROBE_OFFSETS:
+ break;
+ default:
+ return 0;
+ }
+
+ ret = close(fd);
+ if (ret < 0) {
+ PERROR("close error");
+ ret = -1;
+ }
+
+ return ret;
+}
+
+/*
+ * Return < 0 on error, 0 if OK, 1 on hangup.
+ */
+static
+int handle_one_cmd(struct run_as_worker *worker)
+{
+ int ret = 0;
+ struct run_as_data data;
+ ssize_t readlen, writelen;
+ struct run_as_ret sendret;
+ run_as_fct cmd;
+ uid_t prev_euid;
/*
- * Child: it is safe to drop egid and euid while sharing the
- * file descriptors with the parent process, since we do not
- * drop "uid": therefore, the user we are dropping egid/euid to
- * cannot attach to this process with, e.g. ptrace, nor map this
- * process memory.
+ * Stage 1: Receive run_as_data struct from the master.
+ * The structure contains the command type and all the parameters needed for
+ * its execution
*/
- if (data->gid != getegid()) {
- ret = setegid(data->gid);
+ readlen = lttcomm_recv_unix_sock(worker->sockpair[1], &data,
+ sizeof(data));
+ if (readlen == 0) {
+ /* hang up */
+ ret = 1;
+ goto end;
+ }
+ if (readlen < sizeof(data)) {
+ PERROR("lttcomm_recv_unix_sock error");
+ ret = -1;
+ goto end;
+ }
+
+ cmd = run_as_enum_to_fct(data.cmd);
+ if (!cmd) {
+ ret = -1;
+ goto end;
+ }
+
+ /*
+ * Stage 2: Receive file descriptor from master.
+ * Some commands need a file descriptor as input so if it's needed we
+ * receive the fd using the Unix socket.
+ */
+ ret = recv_fd_from_master(worker, data.cmd, &data.fd);
+ if (ret < 0) {
+ PERROR("recv_fd_from_master error");
+ ret = -1;
+ goto end;
+ }
+
+ prev_euid = getuid();
+ if (data.gid != getegid()) {
+ ret = setegid(data.gid);
if (ret < 0) {
- perror("setegid");
- return EXIT_FAILURE;
+ PERROR("setegid");
+ goto write_return;
}
}
- if (data->uid != geteuid()) {
- ret = seteuid(data->uid);
+ if (data.uid != prev_euid) {
+ ret = seteuid(data.uid);
if (ret < 0) {
- perror("seteuid");
- return EXIT_FAILURE;
+ PERROR("seteuid");
+ goto write_return;
}
}
+
/*
* Also set umask to 0 for mkdir executable bit.
*/
umask(0);
- sendret.i = (*data->cmd)(data->data);
- /* send back return value */
- writeleft = sizeof(sendret);
- index = 0;
- do {
- writelen = write(data->retval_pipe, &sendret.c[index],
- writeleft);
- if (writelen < 0) {
- perror("write");
- return EXIT_FAILURE;
+
+ /*
+ * Stage 3: Execute the command
+ */
+ ret = (*cmd)(&data, &sendret);
+ if (ret < 0) {
+ DBG("Execution of command returned an error");
+ }
+
+write_return:
+ ret = cleanup_received_fd(data.cmd, data.fd);
+ if (ret < 0) {
+ ERR("Error cleaning up FD");
+ goto end;
+ }
+
+ /*
+ * Stage 4: Send run_as_ret structure to the master.
+ * This structure contain the return value of the command and the errno.
+ */
+ writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
+ sizeof(sendret));
+ if (writelen < sizeof(sendret)) {
+ PERROR("lttcomm_send_unix_sock error");
+ ret = -1;
+ goto end;
+ }
+
+ /*
+ * Stage 5: Send file descriptor to the master
+ * Some commands return a file descriptor so if it's needed we pass it back
+ * to the master using the Unix socket.
+ */
+ ret = send_fd_to_master(worker, data.cmd, sendret.fd);
+ if (ret < 0) {
+ DBG("Sending FD to master returned an error");
+ goto end;
+ }
+
+ if (seteuid(prev_euid) < 0) {
+ PERROR("seteuid");
+ ret = -1;
+ goto end;
+ }
+ ret = 0;
+end:
+ return ret;
+}
+
+static
+int run_as_worker(struct run_as_worker *worker)
+{
+ int ret;
+ ssize_t writelen;
+ struct run_as_ret sendret;
+ size_t proc_orig_len;
+
+ /*
+ * Initialize worker. Set a different process cmdline.
+ */
+ proc_orig_len = strlen(worker->procname);
+ memset(worker->procname, 0, proc_orig_len);
+ strncpy(worker->procname, DEFAULT_RUN_AS_WORKER_NAME, proc_orig_len);
+
+ ret = lttng_prctl(PR_SET_NAME,
+ (unsigned long) DEFAULT_RUN_AS_WORKER_NAME, 0, 0, 0);
+ if (ret && ret != -ENOSYS) {
+ /* Don't fail as this is not essential. */
+ PERROR("prctl PR_SET_NAME");
+ }
+
+ memset(&sendret, 0, sizeof(sendret));
+
+ writelen = lttcomm_send_unix_sock(worker->sockpair[1], &sendret,
+ sizeof(sendret));
+ if (writelen < sizeof(sendret)) {
+ PERROR("lttcomm_send_unix_sock error");
+ ret = EXIT_FAILURE;
+ goto end;
+ }
+
+ for (;;) {
+ ret = handle_one_cmd(worker);
+ if (ret < 0) {
+ ret = EXIT_FAILURE;
+ goto end;
+ } else if (ret > 0) {
+ break;
+ } else {
+ continue; /* Next command. */
}
- writeleft -= writelen;
- index += writelen;
- } while (writeleft > 0);
- return EXIT_SUCCESS;
+ }
+ ret = EXIT_SUCCESS;
+end:
+ return ret;
}
static
-int run_as(int (*cmd)(void *data), void *data, uid_t uid, gid_t gid)
+int run_as_cmd(struct run_as_worker *worker,
+ enum run_as_cmd cmd,
+ struct run_as_data *data,
+ struct run_as_ret *ret_value,
+ uid_t uid, gid_t gid)
{
- struct run_as_data run_as_data;
int ret = 0;
- int status;
- pid_t pid;
- int retval_pipe[2];
- ssize_t readlen, readleft, index;
- void *child_stack;
- union {
- int i;
- char c[sizeof(int)];
- } retval;
+ ssize_t readlen, writelen;
/*
* If we are non-root, we can only deal with our own uid.
*/
if (geteuid() != 0) {
if (uid != geteuid()) {
+ ret = -1;
+ ret_value->_errno = EPERM;
ERR("Client (%d)/Server (%d) UID mismatch (and sessiond is not root)",
- uid, geteuid());
- return -EPERM;
+ (int) uid, (int) geteuid());
+ goto end;
}
}
- ret = pipe(retval_pipe);
- if (ret < 0) {
- perror("pipe");
+ data->cmd = cmd;
+ data->uid = uid;
+ data->gid = gid;
+
+ /*
+ * Stage 1: Send the run_as_data struct to the worker process
+ */
+ writelen = lttcomm_send_unix_sock(worker->sockpair[0], data,
+ sizeof(*data));
+ if (writelen < sizeof(*data)) {
+ PERROR("Error writing message to run_as");
+ ret = -1;
+ ret_value->_errno = EIO;
goto end;
}
- run_as_data.data = data;
- run_as_data.cmd = cmd;
- run_as_data.uid = uid;
- run_as_data.gid = gid;
- run_as_data.retval_pipe = retval_pipe[1]; /* write end */
- child_stack = mmap(NULL, RUNAS_CHILD_STACK_SIZE,
- PROT_WRITE | PROT_READ,
- MAP_PRIVATE | MAP_GROWSDOWN | MAP_ANONYMOUS | MAP_STACK,
- -1, 0);
- if (child_stack == MAP_FAILED) {
- perror("mmap");
- ret = -ENOMEM;
- goto close_pipe;
+
+ /*
+ * Stage 2: Send file descriptor to the worker process if needed
+ */
+ ret = send_fd_to_worker(worker, data->cmd, data->fd);
+ if (ret) {
+ PERROR("do_send_fd error");
+ ret = -1;
+ ret_value->_errno = EIO;
+ goto end;
}
+
/*
- * Pointing to the middle of the stack to support architectures
- * where the stack grows up (HPPA).
+ * Stage 3: Wait for the execution of the command
*/
- pid = clone(child_run_as, child_stack + (RUNAS_CHILD_STACK_SIZE / 2),
- CLONE_FILES | SIGCHLD,
- &run_as_data, NULL);
- if (pid < 0) {
- perror("clone");
- ret = pid;
- goto unmap_stack;
- }
- /* receive return value */
- readleft = sizeof(retval);
- index = 0;
- do {
- readlen = read(retval_pipe[0], &retval.c[index], readleft);
- if (readlen < 0) {
- perror("read");
- ret = -1;
- break;
- }
- readleft -= readlen;
- index += readlen;
- } while (readleft > 0);
/*
- * Parent: wait for child to return, in which case the
- * shared memory map will have been created.
+ * Stage 4: Receive the run_as_ret struct containing the return value and
+ * errno
*/
- pid = waitpid(pid, &status, 0);
- if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
- perror("wait");
+ readlen = lttcomm_recv_unix_sock(worker->sockpair[0], ret_value,
+ sizeof(*ret_value));
+ if (!readlen) {
+ ERR("Run-as worker has hung-up during run_as_cmd");
ret = -1;
+ ret_value->_errno = EIO;
+ goto end;
+ } else if (readlen < sizeof(*ret_value)) {
+ PERROR("Error reading response from run_as");
+ ret = -1;
+ ret_value->_errno = errno;
}
-unmap_stack:
- ret = munmap(child_stack, RUNAS_CHILD_STACK_SIZE);
+
+ /*
+ * Stage 5: Receive file descriptor if needed
+ */
+ ret = recv_fd_from_worker(worker, data->cmd, &ret_value->fd);
if (ret < 0) {
- perror("munmap");
+ ERR("Error receiving fd");
+ ret = -1;
+ ret_value->_errno = EIO;
}
-close_pipe:
- close(retval_pipe[0]);
- close(retval_pipe[1]);
+
end:
- return retval.i;
+ return ret;
+}
+
+/*
+ * This is for debugging ONLY and should not be considered secure.
+ */
+static
+int run_as_noworker(enum run_as_cmd cmd,
+ struct run_as_data *data, struct run_as_ret *ret_value,
+ uid_t uid, gid_t gid)
+{
+ int ret, saved_errno;
+ mode_t old_mask;
+ run_as_fct fct;
+
+ fct = run_as_enum_to_fct(cmd);
+ if (!fct) {
+ errno = -ENOSYS;
+ ret = -1;
+ goto end;
+ }
+ old_mask = umask(0);
+ ret = fct(data, ret_value);
+ saved_errno = ret_value->_errno;
+ umask(old_mask);
+ errno = saved_errno;
+end:
+ return ret;
+}
+
+static
+int run_as_restart_worker(struct run_as_worker *worker)
+{
+ int ret = 0;
+ char *procname = NULL;
+
+ procname = worker->procname;
+
+ /* Close socket to run_as worker process and clean up the zombie process */
+ run_as_destroy_worker();
+
+ /* Create a new run_as worker process*/
+ ret = run_as_create_worker(procname);
+ if (ret < 0 ) {
+ ERR("Restarting the worker process failed");
+ ret = -1;
+ goto err;
+ }
+err:
+ return ret;
}
+static
+int run_as(enum run_as_cmd cmd, struct run_as_data *data,
+ struct run_as_ret *ret_value, uid_t uid, gid_t gid)
+{
+ int ret, saved_errno;
+
+ if (use_clone()) {
+ DBG("Using run_as worker");
+ pthread_mutex_lock(&worker_lock);
+ assert(global_worker);
+
+ ret = run_as_cmd(global_worker, cmd, data, ret_value, uid, gid);
+ saved_errno = ret_value->_errno;
+
+ pthread_mutex_unlock(&worker_lock);
+ /*
+ * If the worker thread crashed the errno is set to EIO. we log
+ * the error and start a new worker process.
+ */
+ if (ret == -1 && saved_errno == EIO) {
+ DBG("Socket closed unexpectedly... "
+ "Restarting the worker process");
+ ret = run_as_restart_worker(global_worker);
+
+ if (ret == -1) {
+ ERR("Failed to restart worker process.");
+ goto err;
+ }
+ }
+ } else {
+ DBG("Using run_as without worker");
+ ret = run_as_noworker(cmd, data, ret_value, uid, gid);
+ }
+err:
+ return ret;
+}
+
+LTTNG_HIDDEN
int run_as_mkdir_recursive(const char *path, mode_t mode, uid_t uid, gid_t gid)
{
- struct run_as_mkdir_data data;
+ struct run_as_data data;
+ struct run_as_ret ret;
+ memset(&data, 0, sizeof(data));
+ memset(&ret, 0, sizeof(ret));
DBG3("mkdir() recursive %s with mode %d for uid %d and gid %d",
- path, mode, uid, gid);
- data.path = path;
- data.mode = mode;
- return run_as(_mkdir_recursive, &data, uid, gid);
+ path, (int) mode, (int) uid, (int) gid);
+ strncpy(data.u.mkdir.path, path, PATH_MAX - 1);
+ data.u.mkdir.path[PATH_MAX - 1] = '\0';
+ data.u.mkdir.mode = mode;
+
+ run_as(RUN_AS_MKDIR_RECURSIVE, &data, &ret, uid, gid);
+ errno = ret._errno;
+ return ret.u.mkdir.ret;
}
+LTTNG_HIDDEN
int run_as_mkdir(const char *path, mode_t mode, uid_t uid, gid_t gid)
{
- struct run_as_mkdir_data data;
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ memset(&data, 0, sizeof(data));
+ memset(&ret, 0, sizeof(ret));
DBG3("mkdir() %s with mode %d for uid %d and gid %d",
- path, mode, uid, gid);
- data.path = path;
- data.mode = mode;
- return run_as(_mkdir, &data, uid, gid);
+ path, (int) mode, (int) uid, (int) gid);
+ strncpy(data.u.mkdir.path, path, PATH_MAX - 1);
+ data.u.mkdir.path[PATH_MAX - 1] = '\0';
+ data.u.mkdir.mode = mode;
+ run_as(RUN_AS_MKDIR, &data, &ret, uid, gid);
+ errno = ret._errno;
+ return ret.u.mkdir.ret;
}
-/*
- * Note: open_run_as is currently not working. We'd need to pass the fd
- * opened in the child to the parent.
- */
+LTTNG_HIDDEN
int run_as_open(const char *path, int flags, mode_t mode, uid_t uid, gid_t gid)
{
- struct run_as_open_data data;
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ memset(&data, 0, sizeof(data));
+ memset(&ret, 0, sizeof(ret));
DBG3("open() %s with flags %X mode %d for uid %d and gid %d",
- path, flags, mode, uid, gid);
- data.path = path;
- data.flags = flags;
- data.mode = mode;
- return run_as(_open, &data, uid, gid);
+ path, flags, (int) mode, (int) uid, (int) gid);
+ strncpy(data.u.open.path, path, PATH_MAX - 1);
+ data.u.open.path[PATH_MAX - 1] = '\0';
+ data.u.open.flags = flags;
+ data.u.open.mode = mode;
+ run_as(RUN_AS_OPEN, &data, &ret, uid, gid);
+ errno = ret._errno;
+ ret.u.open.ret = ret.fd;
+ return ret.u.open.ret;
+}
+
+LTTNG_HIDDEN
+int run_as_unlink(const char *path, uid_t uid, gid_t gid)
+{
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ memset(&data, 0, sizeof(data));
+ memset(&ret, 0, sizeof(ret));
+
+ DBG3("unlink() %s with for uid %d and gid %d",
+ path, (int) uid, (int) gid);
+ strncpy(data.u.unlink.path, path, PATH_MAX - 1);
+ data.u.unlink.path[PATH_MAX - 1] = '\0';
+ run_as(RUN_AS_UNLINK, &data, &ret, uid, gid);
+ errno = ret._errno;
+ return ret.u.unlink.ret;
+}
+
+LTTNG_HIDDEN
+int run_as_rmdir_recursive(const char *path, uid_t uid, gid_t gid)
+{
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ memset(&data, 0, sizeof(data));
+ memset(&ret, 0, sizeof(ret));
+
+ DBG3("rmdir_recursive() %s with for uid %d and gid %d",
+ path, (int) uid, (int) gid);
+ strncpy(data.u.rmdir_recursive.path, path, PATH_MAX - 1);
+ data.u.rmdir_recursive.path[PATH_MAX - 1] = '\0';
+ run_as(RUN_AS_RMDIR_RECURSIVE, &data, &ret, uid, gid);
+ errno = ret._errno;
+ return ret.u.rmdir_recursive.ret;
+}
+
+LTTNG_HIDDEN
+int run_as_extract_elf_symbol_offset(int fd, const char* function,
+ uid_t uid, gid_t gid, uint64_t *offset)
+{
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ DBG3("extract_elf_symbol_offset() on fd=%d and function=%s "
+ "with for uid %d and gid %d", fd, function, (int) uid, (int) gid);
+
+ data.fd = fd;
+
+ strncpy(data.u.extract_elf_symbol_offset.function, function, LTTNG_SYMBOL_NAME_LEN - 1);
+
+ data.u.extract_elf_symbol_offset.function[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
+
+ run_as(RUN_AS_EXTRACT_ELF_SYMBOL_OFFSET, &data, &ret, uid, gid);
+
+ errno = ret._errno;
+
+ if (ret._error) {
+ return -1;
+ }
+
+ *offset = ret.u.extract_elf_symbol_offset.offset;
+ return 0;
+}
+
+LTTNG_HIDDEN
+int run_as_extract_sdt_probe_offsets(int fd, const char* provider_name,
+ const char* probe_name, uid_t uid, gid_t gid,
+ uint64_t **offsets, uint32_t *num_offset)
+{
+ struct run_as_data data;
+ struct run_as_ret ret;
+
+ DBG3("extract_sdt_probe_offsets() on fd=%d, probe_name=%s and "
+ "provider_name=%s with for uid %d and gid %d", fd, probe_name,
+ provider_name, (int) uid, (int) gid);
+
+ data.fd = fd;
+
+ strncpy(data.u.extract_sdt_probe_offsets.probe_name, probe_name, LTTNG_SYMBOL_NAME_LEN - 1);
+ strncpy(data.u.extract_sdt_probe_offsets.provider_name, provider_name, LTTNG_SYMBOL_NAME_LEN - 1);
+
+ data.u.extract_sdt_probe_offsets.probe_name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
+ data.u.extract_sdt_probe_offsets.provider_name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
+
+ run_as(RUN_AS_EXTRACT_SDT_PROBE_OFFSETS, &data, &ret, uid, gid);
+
+ errno = ret._errno;
+
+ if (ret._error) {
+ return -1;
+ }
+
+ *num_offset = ret.u.extract_sdt_probe_offsets.num_offset;
+
+ *offsets = zmalloc(*num_offset * sizeof(uint64_t));
+ if (!*offsets) {
+ return -ENOMEM;
+ }
+
+ memcpy(*offsets, ret.u.extract_sdt_probe_offsets.offsets, *num_offset * sizeof(uint64_t));
+ return 0;
+}
+
+static
+int reset_sighandler(void)
+{
+ int sig;
+
+ DBG("Resetting run_as worker signal handlers to default");
+ for (sig = 1; sig <= 31; sig++) {
+ (void) signal(sig, SIG_DFL);
+ }
+ return 0;
+}
+
+static
+void worker_sighandler(int sig)
+{
+ const char *signame;
+
+ /*
+ * The worker will inherit its parent's signals since they are part of
+ * the same process group. However, in the case of SIGINT and SIGTERM,
+ * we want to give the worker a chance to teardown gracefully when its
+ * parent closes the command socket.
+ */
+ switch (sig) {
+ case SIGINT:
+ signame = "SIGINT";
+ break;
+ case SIGTERM:
+ signame = "SIGTERM";
+ break;
+ default:
+ signame = NULL;
+ }
+
+ if (signame) {
+ DBG("run_as worker received signal %s", signame);
+ } else {
+ DBG("run_as_worker received signal %d", sig);
+ }
+}
+
+static
+int set_worker_sighandlers(void)
+{
+ int ret = 0;
+ sigset_t sigset;
+ struct sigaction sa;
+
+ if ((ret = sigemptyset(&sigset)) < 0) {
+ PERROR("sigemptyset");
+ goto end;
+ }
+
+ sa.sa_handler = worker_sighandler;
+ sa.sa_mask = sigset;
+ sa.sa_flags = 0;
+ if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
+ PERROR("sigaction SIGINT");
+ goto end;
+ }
+
+ if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
+ PERROR("sigaction SIGTERM");
+ goto end;
+ }
+
+ DBG("run_as signal handler set for SIGTERM and SIGINT");
+end:
+ return ret;
+}
+
+LTTNG_HIDDEN
+int run_as_create_worker(char *procname)
+{
+ pid_t pid;
+ int i, ret = 0;
+ ssize_t readlen;
+ struct run_as_ret recvret;
+ struct run_as_worker *worker;
+
+ pthread_mutex_lock(&worker_lock);
+ assert(!global_worker);
+ if (!use_clone()) {
+ /*
+ * Don't initialize a worker, all run_as tasks will be performed
+ * in the current process.
+ */
+ ret = 0;
+ goto end;
+ }
+ worker = zmalloc(sizeof(*worker));
+ if (!worker) {
+ ret = -ENOMEM;
+ goto end;
+ }
+ worker->procname = procname;
+ /* Create unix socket. */
+ if (lttcomm_create_anon_unix_socketpair(worker->sockpair) < 0) {
+ ret = -1;
+ goto error_sock;
+ }
+
+ /* Fork worker. */
+ pid = fork();
+ if (pid < 0) {
+ PERROR("fork");
+ ret = -1;
+ goto error_fork;
+ } else if (pid == 0) {
+ /* Child */
+
+ reset_sighandler();
+
+ set_worker_sighandlers();
+
+ /* The child has no use for this lock. */
+ pthread_mutex_unlock(&worker_lock);
+ /* Just close, no shutdown. */
+ if (close(worker->sockpair[0])) {
+ PERROR("close");
+ exit(EXIT_FAILURE);
+ }
+
+ /*
+ * Close all FDs aside from STDIN, STDOUT, STDERR and sockpair[1]
+ * Sockpair[1] is used as a control channel with the master
+ */
+ for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
+ if (i != worker->sockpair[1]) {
+ (void) close(i);
+ }
+ }
+
+ worker->sockpair[0] = -1;
+ ret = run_as_worker(worker);
+ if (lttcomm_close_unix_sock(worker->sockpair[1])) {
+ PERROR("close");
+ ret = -1;
+ }
+ worker->sockpair[1] = -1;
+ LOG(ret ? PRINT_ERR : PRINT_DBG, "run_as worker exiting (ret = %d)", ret);
+ exit(ret ? EXIT_FAILURE : EXIT_SUCCESS);
+ } else {
+ /* Parent */
+
+ /* Just close, no shutdown. */
+ if (close(worker->sockpair[1])) {
+ PERROR("close");
+ ret = -1;
+ goto error_fork;
+ }
+ worker->sockpair[1] = -1;
+ worker->pid = pid;
+ /* Wait for worker to become ready. */
+ readlen = lttcomm_recv_unix_sock(worker->sockpair[0],
+ &recvret, sizeof(recvret));
+ if (readlen < sizeof(recvret)) {
+ ERR("readlen: %zd", readlen);
+ PERROR("Error reading response from run_as at creation");
+ ret = -1;
+ goto error_fork;
+ }
+ global_worker = worker;
+ }
+end:
+ pthread_mutex_unlock(&worker_lock);
+ return ret;
+
+ /* Error handling. */
+error_fork:
+ for (i = 0; i < 2; i++) {
+ if (worker->sockpair[i] < 0) {
+ continue;
+ }
+ if (lttcomm_close_unix_sock(worker->sockpair[i])) {
+ PERROR("close");
+ }
+ worker->sockpair[i] = -1;
+ }
+error_sock:
+ free(worker);
+ pthread_mutex_unlock(&worker_lock);
+ return ret;
+}
+
+LTTNG_HIDDEN
+void run_as_destroy_worker(void)
+{
+ struct run_as_worker *worker = global_worker;
+
+ DBG("Destroying run_as worker");
+ pthread_mutex_lock(&worker_lock);
+ if (!worker) {
+ goto end;
+ }
+ /* Close unix socket */
+ DBG("Closing run_as worker socket");
+ if (lttcomm_close_unix_sock(worker->sockpair[0])) {
+ PERROR("close");
+ }
+ worker->sockpair[0] = -1;
+ /* Wait for worker. */
+ for (;;) {
+ int status;
+ pid_t wait_ret;
+
+ wait_ret = waitpid(worker->pid, &status, 0);
+ if (wait_ret < 0) {
+ if (errno == EINTR) {
+ continue;
+ }
+ PERROR("waitpid");
+ break;
+ }
+
+ if (WIFEXITED(status)) {
+ LOG(WEXITSTATUS(status) == 0 ? PRINT_DBG : PRINT_ERR,
+ DEFAULT_RUN_AS_WORKER_NAME " terminated with status code %d",
+ WEXITSTATUS(status));
+ break;
+ } else if (WIFSIGNALED(status)) {
+ ERR(DEFAULT_RUN_AS_WORKER_NAME " was killed by signal %d",
+ WTERMSIG(status));
+ break;
+ }
+ }
+ free(worker);
+ global_worker = NULL;
+end:
+ pthread_mutex_unlock(&worker_lock);
}