New locking/refcounting for fs_struct

* all changes of current->fs are done under task_lock and write_lock of
  old fs->lock
* refcount is not atomic anymore (same protection)
* its decrements are done when removing reference from current; at the
  same time we decide whether to free it.
* put_fs_struct() is gone
* new field - ->in_exec.  Set by check_unsafe_exec() if we are trying to do
  execve() and only subthreads share fs_struct.  Cleared when finishing exec
  (success and failure alike).  Makes CLONE_FS fail with -EAGAIN if set.
* check_unsafe_exec() may fail with -EAGAIN if another execve() from subthread
  is in progress.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs/compat.c

@@ -51,6 +51,7 @@
 #include <linux/poll.h>
 #include <linux/mm.h>
 #include <linux/eventpoll.h>
+#include <linux/fs_struct.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
@@ -1441,12 +1442,15 @@ int compat_do_execve(char * filename,
 	bprm->cred = prepare_exec_creds();
 	if (!bprm->cred)
 		goto out_unlock;
-	check_unsafe_exec(bprm);
+
+	retval = check_unsafe_exec(bprm);
+	if (retval)
+		goto out_unlock;
 
 	file = open_exec(filename);
 	retval = PTR_ERR(file);
 	if (IS_ERR(file))
-		goto out_unlock;
+		goto out_unmark;
 
 	sched_exec();
 
@@ -1488,6 +1492,9 @@ int compat_do_execve(char * filename,
 		goto out;
 
 	/* execve succeeded */
+	write_lock(&current->fs->lock);
+	current->fs->in_exec = 0;
+	write_unlock(&current->fs->lock);
 	current->in_execve = 0;
 	mutex_unlock(&current->cred_exec_mutex);
 	acct_update_integrals(current);
@@ -1506,6 +1513,11 @@ out_file:
 		fput(bprm->file);
 	}
 
+out_unmark:
+	write_lock(&current->fs->lock);
+	current->fs->in_exec = 0;
+	write_unlock(&current->fs->lock);
+
 out_unlock:
 	current->in_execve = 0;
 	mutex_unlock(&current->cred_exec_mutex);

fs/exec.c

@@ -1056,16 +1056,18 @@ EXPORT_SYMBOL(install_exec_creds);
  * - the caller must hold current->cred_exec_mutex to protect against
  *   PTRACE_ATTACH
  */
-void check_unsafe_exec(struct linux_binprm *bprm)
+int check_unsafe_exec(struct linux_binprm *bprm)
 {
 	struct task_struct *p = current, *t;
 	unsigned long flags;
 	unsigned n_fs, n_sighand;
+	int res = 0;
 
 	bprm->unsafe = tracehook_unsafe_exec(p);
 
 	n_fs = 1;
 	n_sighand = 1;
+	write_lock(&p->fs->lock);
 	lock_task_sighand(p, &flags);
 	for (t = next_thread(p); t != p; t = next_thread(t)) {
 		if (t->fs == p->fs)
@@ -1073,11 +1075,19 @@ void check_unsafe_exec(struct linux_binprm *bprm)
 		n_sighand++;
 	}
 
-	if (atomic_read(&p->fs->count) > n_fs ||
-	    atomic_read(&p->sighand->count) > n_sighand)
+	if (p->fs->users > n_fs ||
+	    atomic_read(&p->sighand->count) > n_sighand) {
 		bprm->unsafe |= LSM_UNSAFE_SHARE;
+	} else {
+		if (p->fs->in_exec)
+			res = -EAGAIN;
+		p->fs->in_exec = 1;
+	}
 
 	unlock_task_sighand(p, &flags);
+	write_unlock(&p->fs->lock);
+
+	return res;
 }
 
 /* 
@@ -1296,12 +1306,15 @@ int do_execve(char * filename,
 	bprm->cred = prepare_exec_creds();
 	if (!bprm->cred)
 		goto out_unlock;
-	check_unsafe_exec(bprm);
+
+	retval = check_unsafe_exec(bprm);
+	if (retval)
+		goto out_unlock;
 
 	file = open_exec(filename);
 	retval = PTR_ERR(file);
 	if (IS_ERR(file))
-		goto out_unlock;
+		goto out_unmark;
 
 	sched_exec();
 
@@ -1344,6 +1357,9 @@ int do_execve(char * filename,
 		goto out;
 
 	/* execve succeeded */
+	write_lock(&current->fs->lock);
+	current->fs->in_exec = 0;
+	write_unlock(&current->fs->lock);
 	current->in_execve = 0;
 	mutex_unlock(&current->cred_exec_mutex);
 	acct_update_integrals(current);
@@ -1362,6 +1378,11 @@ out_file:
 		fput(bprm->file);
 	}
 
+out_unmark:
+	write_lock(&current->fs->lock);
+	current->fs->in_exec = 0;
+	write_unlock(&current->fs->lock);
+
 out_unlock:
 	current->in_execve = 0;
 	mutex_unlock(&current->cred_exec_mutex);

fs/fs_struct.c

@@ -72,25 +72,27 @@ void chroot_fs_refs(struct path *old_root, struct path *new_root)
 		path_put(old_root);
 }
 
-void put_fs_struct(struct fs_struct *fs)
+void free_fs_struct(struct fs_struct *fs)
 {
-	/* No need to hold fs->lock if we are killing it */
-	if (atomic_dec_and_test(&fs->count)) {
 	path_put(&fs->root);
 	path_put(&fs->pwd);
 	kmem_cache_free(fs_cachep, fs);
-	}
 }
 
 void exit_fs(struct task_struct *tsk)
 {
-	struct fs_struct * fs = tsk->fs;
+	struct fs_struct *fs = tsk->fs;
 
 	if (fs) {
+		int kill;
 		task_lock(tsk);
+		write_lock(&fs->lock);
 		tsk->fs = NULL;
+		kill = !--fs->users;
+		write_unlock(&fs->lock);
 		task_unlock(tsk);
-		put_fs_struct(fs);
+		if (kill)
+			free_fs_struct(fs);
 	}
 }
 
@@ -99,7 +101,8 @@ struct fs_struct *copy_fs_struct(struct fs_struct *old)
 	struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 	/* We don't need to lock fs - think why ;-) */
 	if (fs) {
-		atomic_set(&fs->count, 1);
+		fs->users = 1;
+		fs->in_exec = 0;
 		rwlock_init(&fs->lock);
 		fs->umask = old->umask;
 		read_lock(&old->lock);
@@ -114,28 +117,54 @@ struct fs_struct *copy_fs_struct(struct fs_struct *old)
 
 int unshare_fs_struct(void)
 {
-	struct fs_struct *fsp = copy_fs_struct(current->fs);
-	if (!fsp)
+	struct fs_struct *fs = current->fs;
+	struct fs_struct *new_fs = copy_fs_struct(fs);
+	int kill;
+
+	if (!new_fs)
 		return -ENOMEM;
-	exit_fs(current);
-	current->fs = fsp;
+
+	task_lock(current);
+	write_lock(&fs->lock);
+	kill = !--fs->users;
+	current->fs = new_fs;
+	write_unlock(&fs->lock);
+	task_unlock(current);
+
+	if (kill)
+		free_fs_struct(fs);
+
 	return 0;
 }
 EXPORT_SYMBOL_GPL(unshare_fs_struct);
 
 /* to be mentioned only in INIT_TASK */
 struct fs_struct init_fs = {
-	.count		= ATOMIC_INIT(1),
+	.users		= 1,
 	.lock		= __RW_LOCK_UNLOCKED(init_fs.lock),
 	.umask		= 0022,
 };
 
 void daemonize_fs_struct(void)
 {
-	struct fs_struct *fs;
+	struct fs_struct *fs = current->fs;
 
-	exit_fs(current);	/* current->fs->count--; */
-	fs = &init_fs;
-	current->fs = fs;
-	atomic_inc(&fs->count);
+	if (fs) {
+		int kill;
+
+		task_lock(current);
+
+		write_lock(&init_fs.lock);
+		init_fs.users++;
+		write_unlock(&init_fs.lock);
+
+		write_lock(&fs->lock);
+		current->fs = &init_fs;
+		kill = !--fs->users;
+		write_unlock(&fs->lock);
+
+		task_unlock(current);
+		if (kill)
+			free_fs_struct(fs);
+	}
 }

fs/internal.h

@@ -44,7 +44,7 @@ extern void __init chrdev_init(void);
 /*
  * exec.c
  */
-extern void check_unsafe_exec(struct linux_binprm *);
+extern int check_unsafe_exec(struct linux_binprm *);
 
 /*
  * namespace.c

fs/proc/task_nommu.c

@@ -49,7 +49,7 @@ void task_mem(struct seq_file *m, struct mm_struct *mm)
 	else
 		bytes += kobjsize(mm);
 	
-	if (current->fs && atomic_read(&current->fs->count) > 1)
+	if (current->fs && current->fs->users > 1)
 		sbytes += kobjsize(current->fs);
 	else
 		bytes += kobjsize(current->fs);

include/linux/fs_struct.h

@@ -4,12 +4,10 @@
 #include <linux/path.h>
 
 struct fs_struct {
-	atomic_t count;	/* This usage count is used by check_unsafe_exec() for
-			 * security checking purposes - therefore it may not be
-			 * incremented, except by clone(CLONE_FS).
-			 */
+	int users;
 	rwlock_t lock;
 	int umask;
+	int in_exec;
 	struct path root, pwd;
 };
 
@@ -19,7 +17,7 @@ extern void exit_fs(struct task_struct *);
 extern void set_fs_root(struct fs_struct *, struct path *);
 extern void set_fs_pwd(struct fs_struct *, struct path *);
 extern struct fs_struct *copy_fs_struct(struct fs_struct *);
-extern void put_fs_struct(struct fs_struct *);
+extern void free_fs_struct(struct fs_struct *);
 extern void daemonize_fs_struct(void);
 extern int unshare_fs_struct(void);
 

kernel/fork.c

@@ -683,11 +683,19 @@ fail_nomem:
 
 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 {
+	struct fs_struct *fs = current->fs;
 	if (clone_flags & CLONE_FS) {
-		atomic_inc(&current->fs->count);
+		/* tsk->fs is already what we want */
+		write_lock(&fs->lock);
+		if (fs->in_exec) {
+			write_unlock(&fs->lock);
+			return -EAGAIN;
+		}
+		fs->users++;
+		write_unlock(&fs->lock);
 		return 0;
 	}
-	tsk->fs = copy_fs_struct(current->fs);
+	tsk->fs = copy_fs_struct(fs);
 	if (!tsk->fs)
 		return -ENOMEM;
 	return 0;
@@ -1518,12 +1526,16 @@ static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
 {
 	struct fs_struct *fs = current->fs;
 
-	if ((unshare_flags & CLONE_FS) &&
-	    (fs && atomic_read(&fs->count) > 1)) {
-		*new_fsp = copy_fs_struct(current->fs);
+	if (!(unshare_flags & CLONE_FS) || !fs)
+		return 0;
+
+	/* don't need lock here; in the worst case we'll do useless copy */
+	if (fs->users == 1)
+		return 0;
+
+	*new_fsp = copy_fs_struct(fs);
 	if (!*new_fsp)
 		return -ENOMEM;
-	}
 
 	return 0;
 }
@@ -1639,8 +1651,13 @@ SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
 
 		if (new_fs) {
 			fs = current->fs;
+			write_lock(&fs->lock);
 			current->fs = new_fs;
+			if (--fs->users)
+				new_fs = NULL;
+			else
 				new_fs = fs;
+			write_unlock(&fs->lock);
 		}
 
 		if (new_mm) {
@@ -1679,7 +1696,7 @@ bad_unshare_cleanup_sigh:
 
 bad_unshare_cleanup_fs:
 	if (new_fs)
-		put_fs_struct(new_fs);
+		free_fs_struct(new_fs);
 
 bad_unshare_cleanup_thread:
 bad_unshare_out: