Fix: worker structure is leaked in run_as process
[lttng-tools.git] / src / common / runas.c
index 5ab42717a9058b4a5083516aa142a3c1efda226d..4fc229b0d0317452df0b3db08c78ab4c998a97d3 100644 (file)
@@ -16,7 +16,7 @@
  * 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/lttng-kernel.h>
 #include <common/common.h>
 #include <common/utils.h>
-#include <common/compat/mman.h>
-#include <common/compat/clone.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;
+};
 
-#ifdef __FreeBSD__
-/* FreeBSD MAP_STACK always return -ENOMEM */
-#define LTTNG_MAP_STACK                0
-#else
-#define LTTNG_MAP_STACK                MAP_STACK
-#endif
+struct run_as_open_data {
+       char path[PATH_MAX];
+       int flags;
+       mode_t mode;
+};
 
-#ifndef MAP_GROWSDOWN
-#define MAP_GROWSDOWN          0
-#endif
+struct run_as_unlink_data {
+       char path[PATH_MAX];
+};
 
-#ifndef MAP_ANONYMOUS
-#define MAP_ANONYMOUS          MAP_ANON
-#endif
+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)
@@ -87,255 +177,1079 @@ int use_clone(void)
 static
 int use_clone(void)
 {
-       return !getenv("LTTNG_DEBUG_NOCLONE");
+       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;
        mode_t mode;
 
-       path = data->path;
-       mode = data->mode;
+       path = data->u.mkdir.path;
+       mode = data->u.mkdir.mode;
 
-       return utils_mkdir_recursive(path, 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(void *_data)
+int _mkdir(struct run_as_data *data, struct run_as_ret *ret_value)
 {
-       int ret;
-       struct run_as_mkdir_data *data = _data;
+       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 < 0;
+       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;
+       }
+}
+
+static
+int do_send_fd(int sock, int fd)
+{
+       ssize_t len;
+
+       if (fd < 0) {
+               ERR("Attempt to send invalid file descriptor to master (fd = %i)", fd);
+               /* Return 0 as this is not a fatal error. */
+               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;
 
-       ret = mkdir(data->path, data->mode);
+       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;
+       }
+       if (*fd < 0) {
+               ERR("Invalid file descriptor received from worker (fd = %i)", *fd);
+               /* Return 0 as this is not a fatal error. */
+               return 0;
+       }
+
+       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;
+       }
+
+       if (fd < 0) {
+               ERR("Refusing to send invalid fd to worker (fd = %i)", fd);
+               return -1;
+       }
+
+       ret = do_send_fd(worker->sockpair[0], fd);
        if (ret < 0) {
-               ret = -errno;
+               PERROR("do_send_fd");
+               ret = -1;
        }
 
        return ret;
 }
 
 static
-int _open(void *_data)
+int send_fd_to_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, ret_close = 0;
+
+       switch (cmd) {
+       case RUN_AS_OPEN:
+               break;
+       default:
+               return 0;
+       }
+
+       if (fd < 0) {
+               DBG("Not sending file descriptor to master as it is invalid (fd = %i)", fd);
+               return 0;
+       }
+       ret = do_send_fd(worker->sockpair[1], fd);
+       if (ret < 0) {
+               PERROR("do_send_fd error");
+               ret = -1;
+       }
+
+       ret_close = close(fd);
+       if (ret_close < 0) {
+               PERROR("close");
+       }
+
+       return ret;
 }
 
 static
-int child_run_as(void *_data)
+int recv_fd_from_worker(struct run_as_worker *worker, enum run_as_cmd cmd, int *fd)
 {
-       int ret;
-       struct run_as_data *data = _data;
-       ssize_t writelen;
-       int sendret;
+       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 recv_fd_from_master(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_recv_fd(worker->sockpair[1], fd);
+       if (ret < 0) {
+               PERROR("do_recv_fd error");
+               ret = -1;
+       }
+
+       return ret;
+}
+
+static
+int cleanup_received_fd(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;
+       }
+
+       if (fd < 0) {
+               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;
+
+       memset(&sendret, 0, sizeof(sendret));
+       sendret.fd = -1;
+
+       /*
+        * Stage 1: Receive run_as_data struct from the master.
+        * The structure contains the command type and all the parameters needed for
+        * its execution
+        */
+       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;
+       }
 
        /*
-        * 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 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.
         */
-       if (data->gid != getegid()) {
-               ret = setegid(data->gid);
+       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) {
+                       sendret._error = true;
+                       sendret._errno = errno;
                        PERROR("setegid");
-                       sendret = -1;
                        goto write_return;
                }
        }
-       if (data->uid != geteuid()) {
-               ret = seteuid(data->uid);
+       if (data.uid != prev_euid) {
+               ret = seteuid(data.uid);
                if (ret < 0) {
+                       sendret._error = true;
+                       sendret._errno = errno;
                        PERROR("seteuid");
-                       sendret = -1;
                        goto write_return;
                }
        }
+
        /*
         * Also set umask to 0 for mkdir executable bit.
         */
        umask(0);
-       sendret = (*data->cmd)(data->data);
+
+       /*
+        * Stage 3: Execute the command
+        */
+       ret = (*cmd)(&data, &sendret);
+       if (ret < 0) {
+               DBG("Execution of command returned an error");
+       }
 
 write_return:
-       /* send back return value */
-       writelen = lttng_write(data->retval_pipe, &sendret, sizeof(sendret));
+       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("lttng_write error");
-               return EXIT_FAILURE;
-       } else {
-               return EXIT_SUCCESS;
+               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_clone(int (*cmd)(void *data), void *data, uid_t uid, gid_t gid)
+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. */
+               }
+       }
+       ret = EXIT_SUCCESS;
+end:
+       return ret;
+}
+
+static
+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;
-       ssize_t readlen;
-       int status;
-       pid_t pid;
-       int retval_pipe[2];
-       void *child_stack;
-       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");
-               retval = ret;
+       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 | LTTNG_MAP_STACK,
-               -1, 0);
-       if (child_stack == MAP_FAILED) {
-               PERROR("mmap");
-               retval = -ENOMEM;
-               goto close_pipe;
-       }
+
        /*
-        * Pointing to the middle of the stack to support architectures
-        * where the stack grows up (HPPA).
+        * Stage 2: Send file descriptor to the worker process if needed
         */
-       pid = lttng_clone_files(child_run_as, child_stack + (RUNAS_CHILD_STACK_SIZE / 2),
-               &run_as_data);
-       if (pid < 0) {
-               PERROR("clone");
-               retval = pid;
-               goto unmap_stack;
-       }
-       /* receive return value */
-       readlen = lttng_read(retval_pipe[0], &retval, sizeof(retval));
-       if (readlen < sizeof(retval)) {
+       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;
        }
 
        /*
-        * Parent: wait for child to return, in which case the
-        * shared memory map will have been created.
+        * Stage 3: Wait for the execution of the command
         */
-       pid = waitpid(pid, &status, 0);
-       if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
-               PERROR("wait");
-               retval = -1;
-       }
-unmap_stack:
-       ret = munmap(child_stack, RUNAS_CHILD_STACK_SIZE);
-       if (ret < 0) {
-               PERROR("munmap");
-               retval = ret;
+
+       /*
+        * Stage 4: Receive the run_as_ret struct containing the return value and
+        * errno
+        */
+       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;
+               goto end;
        }
-close_pipe:
-       ret = close(retval_pipe[0]);
-       if (ret) {
-               PERROR("close");
+
+       if (ret_value->_error) {
+               /* Skip stage 5 on error as there will be no fd to receive. */
+               goto end;
        }
-       ret = close(retval_pipe[1]);
-       if (ret) {
-               PERROR("close");
+
+       /*
+        * Stage 5: Receive file descriptor if needed
+        */
+       ret = recv_fd_from_worker(worker, data->cmd, &ret_value->fd);
+       if (ret < 0) {
+               ERR("Error receiving fd");
+               ret = -1;
+               ret_value->_errno = EIO;
        }
+
 end:
-       return retval;
+       return ret;
 }
 
 /*
- * To be used on setups where gdb has issues debugging programs using
- * clone/rfork. Note that this is for debuging ONLY, and should not be
- * considered secure.
+ * This is for debugging ONLY and should not be considered secure.
  */
 static
-int run_as_noclone(int (*cmd)(void *data), void *data, uid_t uid, gid_t gid)
+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;
+       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 = cmd(data);
+       ret = fct(data, ret_value);
+       saved_errno = ret_value->_errno;
        umask(old_mask);
+       errno = saved_errno;
+end:
+       return ret;
+}
+
+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;
+}
+
+static
+int run_as_create_worker_no_lock(const char *procname)
+{
+       pid_t pid;
+       int i, ret = 0;
+       ssize_t readlen;
+       struct run_as_ret recvret;
+       struct run_as_worker *worker;
+
+       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 = strdup(procname);
+       if (!worker->procname) {
+               ret = -ENOMEM;
+               goto error_procname_alloc;
+       }
+       /* 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();
+
+               /* 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;
+               free(worker->procname);
+               free(worker);
+               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:
+       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->procname);
+error_procname_alloc:
+       free(worker);
        return ret;
 }
 
 static
-int run_as(int (*cmd)(void *data), void *data, uid_t uid, gid_t gid)
+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_no_lock(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;
+
+       pthread_mutex_lock(&worker_lock);
        if (use_clone()) {
-               int ret;
+               DBG("Using run_as worker");
+
+               assert(global_worker);
 
-               DBG("Using run_as_clone");
-               pthread_mutex_lock(&lttng_libc_state_lock);
-               ret = run_as_clone(cmd, data, uid, gid);
-               pthread_mutex_unlock(&lttng_libc_state_lock);
-               return ret;
+               ret = run_as_cmd(global_worker, cmd, data, ret_value, uid, gid);
+               saved_errno = ret_value->_errno;
+
+               /*
+                * 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_noclone");
-               return run_as_noclone(cmd, data, uid, gid);
+               DBG("Using run_as without worker");
+               ret = run_as_noworker(cmd, data, ret_value, uid, gid);
        }
+err:
+       pthread_mutex_unlock(&worker_lock);
+       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;
+
+       memset(&data, 0, sizeof(data));
+       memset(&ret, 0, sizeof(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;
+
+       memset(&data, 0, sizeof(data));
+       memset(&ret, 0, sizeof(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;
+}
+
+LTTNG_HIDDEN
+int run_as_create_worker(const char *procname)
+{
+       int ret;
+
+       pthread_mutex_lock(&worker_lock);
+       ret = run_as_create_worker_no_lock(procname);
+       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->procname);
+       free(worker);
+       global_worker = NULL;
+end:
+       pthread_mutex_unlock(&worker_lock);
 }
This page took 0.037639 seconds and 5 git commands to generate.