Merge pull request #2664 from brauner/2018-09-30/syscall_wrappers

syscalls: add wrappers and explicit raw syscalls

src/lxc/Makefile.am

@@ -23,6 +23,7 @@ noinst_HEADERS = api_extensions.h \
 		 macro.h \
 		 monitor.h \
 		 namespace.h \
+		 raw_syscalls.h \
 		 start.h \
 		 state.h \
 		 storage/btrfs.h \
@@ -116,6 +117,7 @@ liblxc_la_SOURCES = af_unix.c af_unix.h \
 		    network.c network.h \
 		    monitor.c monitor.h \
 		    parse.c parse.h \
+		    raw_syscalls.c raw_syscalls.h \
 		    ringbuf.c ringbuf.h \
 		    rtnl.c rtnl.h \
 		    state.c state.h \
@@ -339,21 +341,31 @@ endif
 if ENABLE_COMMANDS
 # Binaries shipping with liblxc
 init_lxc_SOURCES = cmd/lxc_init.c \
+		   compiler.h \
+		   error.h \
 		   initutils.c initutils.h \
+		   log.c log.h \
+		   parse.c parse.h \
+		   raw_syscalls.c raw_syscalls.h \
 		   string_utils.c string_utils.h
-lxc_monitord_SOURCES = cmd/lxc_monitord.c
+lxc_monitord_SOURCES = cmd/lxc_monitord.c \
+		       af_unix.c af_unix.h \
+		       log.c log.h \
+		       mainloop.c mainloop.h \
+		       monitor.c monitor.h \
+		       raw_syscalls.c raw_syscalls.h \
+		       utils.c utils.h
 lxc_user_nic_SOURCES = cmd/lxc_user_nic.c \
 		       ../include/netns_ifaddrs.c ../include/netns_ifaddrs.h \
 		       log.c log.h \
-		       namespace.c namespace.h \
 		       network.c network.h \
-		       parse.c parse.h
+		       parse.c parse.h \
+		       raw_syscalls.c raw_syscalls.h
 lxc_usernsexec_SOURCES = cmd/lxc_usernsexec.c \
 			 conf.c conf.h \
 			 list.h \
 			 log.c log.h \
 			 macro.h \
-			 namespace.c namespace.h \
 			 file_utils.c file_utils.h \
 			 string_utils.c string_utils.h \
 			 utils.c utils.h

src/lxc/af_unix.c

@@ -37,6 +37,7 @@
 
 #include "config.h"
 #include "log.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 #ifndef HAVE_STRLCPY

src/lxc/attach.c

@@ -59,6 +59,7 @@
 #include "macro.h"
 #include "mainloop.h"
 #include "namespace.h"
+#include "raw_syscalls.h"
 #include "terminal.h"
 #include "utils.h"
 

src/lxc/cmd/lxc_init.c

@@ -46,8 +46,8 @@
 #include "error.h"
 #include "initutils.h"
 #include "log.h"
-#include "namespace.h"
 #include "parse.h"
+#include "raw_syscalls.h"
 #include "string_utils.h"
 
 /* option keys for long only options */

src/lxc/cmd/lxc_monitord.c

@@ -49,6 +49,7 @@
 #include "log.h"
 #include "mainloop.h"
 #include "monitor.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 #define CLIENTFDS_CHUNK 64

src/lxc/cmd/lxc_user_nic.c

@@ -49,9 +49,9 @@
 
 #include "config.h"
 #include "log.h"
-#include "namespace.h"
 #include "network.h"
 #include "parse.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 #ifndef HAVE_STRLCPY

src/lxc/cmd/lxc_usernsexec.c

@@ -47,7 +47,8 @@
 #include "list.h"
 #include "log.h"
 #include "macro.h"
-#include "namespace.h"
+#include "file_utils.h"
+#include "string_utils.h"
 #include "utils.h"
 
 extern int lxc_log_fd;

src/lxc/conf.c

@@ -70,10 +70,12 @@
 #include "namespace.h"
 #include "network.h"
 #include "parse.h"
+#include "raw_syscalls.h"
 #include "ringbuf.h"
 #include "start.h"
 #include "storage.h"
 #include "storage/overlay.h"
+#include "syscall_wrappers.h"
 #include "terminal.h"
 #include "utils.h"
 
@@ -124,21 +126,6 @@ thread_local struct lxc_conf *current_config;
 struct lxc_conf *current_config;
 #endif
 
-/* Define pivot_root() if missing from the C library */
-#ifndef HAVE_PIVOT_ROOT
-static int pivot_root(const char *new_root, const char *put_old)
-{
-#ifdef __NR_pivot_root
-	return syscall(__NR_pivot_root, new_root, put_old);
-#else
-	errno = ENOSYS;
-	return -1;
-#endif
-}
-#else
-extern int pivot_root(const char *new_root, const char *put_old);
-#endif
-
 char *lxchook_names[NUM_LXC_HOOKS] = {
 	"pre-start",
 	"pre-mount",
@@ -3549,21 +3536,11 @@ static bool verify_start_hooks(struct lxc_conf *conf)
 
 static bool execveat_supported(void)
 {
-#ifdef __NR_execveat
-	/*
-	 * We use the syscall here, because it was introduced in kernel 3.19,
-	 * while glibc got support for using the syscall much later, in 2.27.
-	 * We don't want to use glibc because it falls back to /proc, and the
-	 * container may not have /proc mounted depending on its configuration.
-	 */
-	syscall(__NR_execveat, -1, "", NULL, NULL, AT_EMPTY_PATH);
+	lxc_raw_execveat(-1, "", NULL, NULL, AT_EMPTY_PATH);
 	if (errno == ENOSYS)
 		return false;
 
 	return true;
-#else
-	return false;
-#endif
 }
 
 int lxc_setup(struct lxc_handler *handler)

src/lxc/execute.c

@@ -35,6 +35,7 @@
 #include "config.h"
 #include "log.h"
 #include "start.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 lxc_log_define(execute, start);
@@ -122,11 +123,7 @@ static int execute_start(struct lxc_handler *handler, void* data)
 	NOTICE("Exec'ing \"%s\"", my_args->argv[0]);
 
 	if (my_args->init_fd >= 0)
-#ifdef __NR_execveat
-		syscall(__NR_execveat, my_args->init_fd, "", argv, environ, AT_EMPTY_PATH);
-#else
-		ERROR("System seems to be missing execveat syscall number");
-#endif
+		lxc_raw_execveat(my_args->init_fd, "", argv, environ, AT_EMPTY_PATH);
 	else
 		execvp(argv[0], argv);
 	SYSERROR("Failed to exec %s", argv[0]);

src/lxc/lsm/apparmor.c

@@ -37,6 +37,7 @@
 #include "log.h"
 #include "lsm.h"
 #include "parse.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 lxc_log_define(apparmor, lsm);

src/lxc/lxccontainer.c

@@ -65,6 +65,7 @@
 #include "namespace.h"
 #include "network.h"
 #include "parse.h"
+#include "raw_syscalls.h"
 #include "start.h"
 #include "state.h"
 #include "storage.h"

src/lxc/namespace.c

@@ -75,80 +75,6 @@ pid_t lxc_clone(int (*fn)(void *), void *arg, int flags)
 	return ret;
 }
 
-/**
- * This is based on raw_clone in systemd but adapted to our needs. This uses
- * copy on write semantics and doesn't pass a stack. CLONE_VM is tricky and
- * doesn't really matter to us so disallow it.
- *
- * The nice thing about this is that we get fork() behavior. That is
- * lxc_raw_clone() returns 0 in the child and the child pid in the parent.
- */
-pid_t lxc_raw_clone(unsigned long flags)
-{
-
-	/* These flags don't interest at all so we don't jump through any hoopes
-	 * of retrieving them and passing them to the kernel.
-	 */
-	errno = EINVAL;
-	if ((flags & (CLONE_VM | CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
-		      CLONE_CHILD_CLEARTID | CLONE_SETTLS)))
-		return -EINVAL;
-
-#if defined(__s390x__) || defined(__s390__) || defined(__CRIS__)
-	/* On s390/s390x and cris the order of the first and second arguments
-	 * of the system call is reversed.
-	 */
-	return (int)syscall(__NR_clone, NULL, flags | SIGCHLD);
-#elif defined(__sparc__) && defined(__arch64__)
-	{
-		/**
-		 * sparc64 always returns the other process id in %o0, and
-		 * a boolean flag whether this is the child or the parent in
-		 * %o1. Inline assembly is needed to get the flag returned
-		 * in %o1.
-		 */
-		int in_child;
-		int child_pid;
-		asm volatile("mov %2, %%g1\n\t"
-			     "mov %3, %%o0\n\t"
-			     "mov 0 , %%o1\n\t"
-			     "t 0x6d\n\t"
-			     "mov %%o1, %0\n\t"
-			     "mov %%o0, %1"
-			     : "=r"(in_child), "=r"(child_pid)
-			     : "i"(__NR_clone), "r"(flags | SIGCHLD)
-			     : "%o1", "%o0", "%g1");
-
-		if (in_child)
-			return 0;
-		else
-			return child_pid;
-	}
-#elif defined(__ia64__)
-	/* On ia64 the stack and stack size are passed as separate arguments. */
-	return (int)syscall(__NR_clone, flags | SIGCHLD, NULL, 0);
-#else
-	return (int)syscall(__NR_clone, flags | SIGCHLD, NULL);
-#endif
-}
-
-pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args, unsigned long flags)
-{
-	pid_t pid;
-
-	pid = lxc_raw_clone(flags);
-	if (pid < 0)
-		return -1;
-
-	/* exit() is not thread-safe and might mess with the parent's signal
-	 * handlers and other stuff when exec() fails.
-	 */
-	if (pid == 0)
-		_exit(fn(args));
-
-	return pid;
-}
-
 /* Leave the user namespace at the first position in the array of structs so
  * that we always attach to it first when iterating over the struct and using
  * setns() to switch namespaces. This especially affects lxc_attach(): Suppose

src/lxc/namespace.h

@@ -128,67 +128,16 @@ int clone(int (*fn)(void *), void *child_stack,
  *   corresponding libc wrapper. glibc currently does not run pthread_atfork()
  *   handlers but does not guarantee that they are not. Other libcs might or
  *   might not run pthread_atfork() handlers. If you require guarantees please
- *   refer to the lxc_raw_clone*() functions below.
+ *   refer to the lxc_raw_clone*() functions in raw_syscalls.{c,h}.
  *
  * - should call lxc_raw_getpid():
  *   The child should use lxc_raw_getpid() to retrieve its pid.
  */
 extern pid_t lxc_clone(int (*fn)(void *), void *arg, int flags);
 
-/**
- * lxc_raw_clone() - create a new process
- *
- * - fork() behavior:
- *   This function returns 0 in the child and > 0 in the parent.
- *
- * - copy-on-write:
- *   This function does not allocate a new stack and relies on copy-on-write
- *   semantics.
- *
- * - supports subset of ClONE_* flags:
- *   lxc_raw_clone() intentionally only supports a subset of the flags available
- *   to the actual system call. Please refer to the implementation what flags
- *   cannot be used. Also, please don't assume that just because a flag isn't
- *   explicitly checked for as being unsupported that it is supported. If in
- *   doubt or not sufficiently familiar with process creation in the kernel and
- *   interactions with libcs this function should be used.
- *
- * - no pthread_atfork() handlers:
- *   This function circumvents - as much as this this is possible - any libc
- *   wrappers and thus does not run any pthread_atfork() handlers. Make sure
- *   that this is safe to do in the context you are trying to call this
- *   function.
- *
- * - must call lxc_raw_getpid():
- *   The child must use lxc_raw_getpid() to retrieve its pid.
- */
-extern pid_t lxc_raw_clone(unsigned long flags);
-/**
- * lxc_raw_clone_cb() - create a new process
- *
- * - non-fork() behavior:
- *   Function does return pid of the child or -1 on error. Pass in a callback
- *   function via the "fn" argument that gets executed in the child process. The
- *   "args" argument is passed to "fn".
- *
- * All other comments that apply to lxc_raw_clone() apply to lxc_raw_clone_cb()
- * as well.
- */
-extern pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args,
-			      unsigned long flags);
-
 extern int lxc_namespace_2_cloneflag(const char *namespace);
 extern int lxc_namespace_2_ns_idx(const char *namespace);
 extern int lxc_namespace_2_std_identifiers(char *namespaces);
 extern int lxc_fill_namespace_flags(char *flaglist, int *flags);
 
-/**
- * Because of older glibc's pid cache (up to 2.25) whenever clone() is called
- * the child must must retrieve it's own pid via lxc_raw_getpid().
- */
-static inline pid_t lxc_raw_getpid(void)
-{
-	return (pid_t) syscall(SYS_getpid);
-}
-
 #endif

src/lxc/network.c

@@ -56,6 +56,7 @@
 #include "macro.h"
 #include "network.h"
 #include "nl.h"
+#include "raw_syscalls.h"
 #include "utils.h"
 
 #ifndef HAVE_STRLCPY