Skip to contentSkip to conversation

fs: make all new mount api fds cloexec by default

PATCH In review Christian Brauner 2 participants 3 messages May 8, 2019 VFS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fs: make all new mount api fds cloexec by default

This makes file descriptors returned from the new syscalls of the new mount api cloexec by default.

From a userspace perspective it is rarely the case that fds are supposed to be inherited across exec. In fact, most of the time userspace either needs to remember to pass the <SPECIFIC>_CLOEXEC flag along or needs to invoke fcntl() on fd to prevent accidentally leaking the fd. This is a much bigger issue than accidentally forgetting to remove the cloexec flag to inherit the fd.
For old file descriptor types we can't break userspace but new ones should
- whenever reasonable - be cloexec by default. Examples of this policy are the new seccomp notify fds and also pidfds. If userspace wants to inherit fds across exec they can remove the O_CLOEXEC flag and need to opt in to inheritance explicitly.

Note, this also has the advantage that we can get rid of all the special flags per file descriptor type for the new mount api. In total this lets us remove 4 flags:
- FSMOUNT_CLOEXEC
- FSOPEN_CLOEXEC
- FSPICK_CLOEXEC
- OPEN_TREE_CLOEXEC

Ideally, this would be changed before rc1 is out since this would otherwise a UAPI break.

---
 fs/fsopen.c                | 13 ++++++-------
 fs/namespace.c             | 11 ++++-------
 include/uapi/linux/mount.h | 18 +++---------------
 3 files changed, 13 insertions(+), 29 deletions(-)
fs/fsopen.c +6 -7
index 3bb9c0c8cbcc..a38fa8c616cf 100644
@@ -88,12 +88,12 @@ const struct file_operations fscontext_fops = {
 /*
  * Attach a filesystem context to a file and an fd.
  */
-static int fscontext_create_fd(struct fs_context *fc, unsigned int o_flags)
+static int fscontext_create_fd(struct fs_context *fc)
 {
 	int fd;
 
 	fd = anon_inode_getfd("fscontext", &fscontext_fops, fc,
-			      O_RDWR | o_flags);
+			      O_RDWR | O_CLOEXEC);
 	if (fd < 0)
 		put_fs_context(fc);
 	return fd;
@@ -126,7 +126,7 @@ SYSCALL_DEFINE2(fsopen, const char __user *, _fs_name, unsigned int, flags)
 	if (!ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if (flags & ~FSOPEN_CLOEXEC)
+	if (flags)
 		return -EINVAL;
 
 	fs_name = strndup_user(_fs_name, PAGE_SIZE);
@@ -149,7 +149,7 @@ SYSCALL_DEFINE2(fsopen, const char __user *, _fs_name, unsigned int, flags)
 	if (ret < 0)
 		goto err_fc;
 
-	return fscontext_create_fd(fc, flags & FSOPEN_CLOEXEC ? O_CLOEXEC : 0);
+	return fscontext_create_fd(fc);
 
 err_fc:
 	put_fs_context(fc);
@@ -169,8 +169,7 @@ SYSCALL_DEFINE3(fspick, int, dfd, const char __user *, path, unsigned int, flags
 	if (!ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if ((flags & ~(FSPICK_CLOEXEC |
-		       FSPICK_SYMLINK_NOFOLLOW |
+	if ((flags & ~(FSPICK_SYMLINK_NOFOLLOW |
 		       FSPICK_NO_AUTOMOUNT |
 		       FSPICK_EMPTY_PATH)) != 0)
 		return -EINVAL;
@@ -203,7 +202,7 @@ SYSCALL_DEFINE3(fspick, int, dfd, const char __user *, path, unsigned int, flags
 		goto err_fc;
 
 	path_put(&target);
-	return fscontext_create_fd(fc, flags & FSPICK_CLOEXEC ? O_CLOEXEC : 0);
+	return fscontext_create_fd(fc);
 
 err_fc:
 	put_fs_context(fc);
fs/namespace.c +4 -7
index 3357c3d65475..ab8cea5d745f 100644
@@ -2369,11 +2369,8 @@ SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags)
 	int error;
 	int fd;
 
-	BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
-
 	if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
-		      AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
-		      OPEN_TREE_CLOEXEC))
+		      AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE))
 		return -EINVAL;
 
 	if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
@@ -2389,7 +2386,7 @@ SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags)
 	if (detached && !may_mount())
 		return -EPERM;
 
-	fd = get_unused_fd_flags(flags & O_CLOEXEC);
+	fd = get_unused_fd_flags(flags | O_CLOEXEC);
 	if (fd < 0)
 		return fd;
 
@@ -3352,7 +3349,7 @@ SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
 	if (!may_mount())
 		return -EPERM;
 
-	if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
+	if (flags)
 		return -EINVAL;
 
 	if (attr_flags & ~(MOUNT_ATTR_RDONLY |
@@ -3457,7 +3454,7 @@ SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
 	}
 	file->f_mode |= FMODE_NEED_UNMOUNT;
 
-	ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
+	ret = get_unused_fd_flags(flags | O_CLOEXEC);
 	if (ret >= 0)
 		fd_install(ret, file);
 	else
include/uapi/linux/mount.h +3 -15
index 96a0240f23fe..c688e4ac843b 100644
@@ -59,7 +59,6 @@
  * open_tree() flags.
  */
 #define OPEN_TREE_CLONE		1		/* Clone the target tree and attach the clone */
-#define OPEN_TREE_CLOEXEC	O_CLOEXEC	/* Close the file on execve() */
 
 /*
  * move_mount() flags.
@@ -72,18 +71,12 @@
 #define MOVE_MOUNT_T_EMPTY_PATH		0x00000040 /* Empty to path permitted */
 #define MOVE_MOUNT__MASK		0x00000077
 
-/*
- * fsopen() flags.
- */
-#define FSOPEN_CLOEXEC		0x00000001
-
 /*
  * fspick() flags.
  */
-#define FSPICK_CLOEXEC		0x00000001
-#define FSPICK_SYMLINK_NOFOLLOW	0x00000002
-#define FSPICK_NO_AUTOMOUNT	0x00000004
-#define FSPICK_EMPTY_PATH	0x00000008
+#define FSPICK_SYMLINK_NOFOLLOW	0x00000001
+#define FSPICK_NO_AUTOMOUNT	0x00000002
+#define FSPICK_EMPTY_PATH	0x00000004
 
 /*
  * The type of fsconfig() call made.
@@ -99,11 +92,6 @@ enum fsconfig_command {
 	FSCONFIG_CMD_RECONFIGURE = 7,	/* Invoke superblock reconfiguration */
 };
 
-/*
- * fsmount() flags.
- */
-#define FSMOUNT_CLOEXEC		0x00000001
-
 /*
  * Mount attributes.
  */
Signature · 1 line
2.21.0
#1 lore raw

fs/namespace.c SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags) · line 2389

Quoted 2 lines · Christian Brauner
fs/namespace.c +1 -1
-	fd = get_unused_fd_flags(flags & O_CLOEXEC);
+	fd = get_unused_fd_flags(flags | O_CLOEXEC);

That'll break if there are any flags other than O_CLOEXEC.

fs/namespace.c SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, · line 3457

Quoted 2 lines
fs/namespace.c +1 -1
-	ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
+	ret = get_unused_fd_flags(flags | O_CLOEXEC);

That'll break because flags is not compatible with what get_unused_fd_flags() is expecting.

David

#2 lore raw
Quoted 7 lines · David Howells

fs/namespace.c SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags) · line 2389

Quoted 2 lines · Christian Brauner
fs/namespace.c +1 -1
-	fd = get_unused_fd_flags(flags & O_CLOEXEC);
+	fd = get_unused_fd_flags(flags | O_CLOEXEC);

That'll break if there are any flags other than O_CLOEXEC.

Quoted 7 lines

Review

Quoted 1 line
(quoted diff) +0 -1
-	ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC :

0);

fs/namespace.c SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, · line 3457

Quoted 1 line
fs/namespace.c +1 -0
+	ret = get_unused_fd_flags(flags | O_CLOEXEC);

That'll break because flags is not compatible with what get_unused_fd_flags()
is expecting.

Technically only when new flags are added.
Right now it works correctly.
I'll send a v1 soon.

Thanks for catching that, David.
Christian