fs: fs_struct use seqlock

Use a seqlock in the fs_struct to enable us to take an atomic copy of the
complete cwd and root paths. Use this in the RCU lookup path to avoid a
thread-shared spinlock in RCU lookup operations.

Multi-threaded apps may now perform path lookups with scalability matching
multi-process apps. Operations such as stat(2) become very scalable for
multi-threaded workload.

Signed-off-by: Nick Piggin <[e-mail]>

fs/fs_struct.c

···
 	struct path old_root;
 
 	spin_lock(&fs->lock);
+	write_seqcount_begin(&fs->seq);
 	old_root = fs->root;
 	fs->root = *path;
 	path_get(path);
+	write_seqcount_end(&fs->seq);
 	spin_unlock(&fs->lock);
 	if (old_root.dentry)
 		path_put(&old_root);
···
 	struct path old_pwd;
 
 	spin_lock(&fs->lock);
+	write_seqcount_begin(&fs->seq);
 	old_pwd = fs->pwd;
 	fs->pwd = *path;
 	path_get(path);
+	write_seqcount_end(&fs->seq);
 	spin_unlock(&fs->lock);
 
 	if (old_pwd.dentry)
···
 		fs = p->fs;
 		if (fs) {
 			spin_lock(&fs->lock);
+			write_seqcount_begin(&fs->seq);
 			if (fs->root.dentry == old_root->dentry
 			    && fs->root.mnt == old_root->mnt) {
 				path_get(new_root);
···
 				fs->pwd = *new_root;
 				count++;
 			}
+			write_seqcount_end(&fs->seq);
 			spin_unlock(&fs->lock);
 		}
 		task_unlock(p);
···
 		int kill;
 		task_lock(tsk);
 		spin_lock(&fs->lock);
+		write_seqcount_begin(&fs->seq);
 		tsk->fs = NULL;
 		kill = !--fs->users;
+		write_seqcount_end(&fs->seq);
 		spin_unlock(&fs->lock);
 		task_unlock(tsk);
 		if (kill)
···
 		fs->users = 1;
 		fs->in_exec = 0;
 		spin_lock_init(&fs->lock);
+		seqcount_init(&fs->seq);
 		fs->umask = old->umask;
 		get_fs_root_and_pwd(old, &fs->root, &fs->pwd);
 	}
···
 struct fs_struct init_fs = {
 	.users		= 1,
 	.lock		= __SPIN_LOCK_UNLOCKED(init_fs.lock),
+	.seq		= SEQCNT_ZERO,
 	.umask		= 0022,
 };
 

fs/namei.c

···
 {
 	if (!nd->root.mnt) {
 		struct fs_struct *fs = current->fs;
-		spin_lock(&fs->lock);
-		nd->root = fs->root;
-		spin_unlock(&fs->lock);
+		unsigned seq;
+
+		do {
+			seq = read_seqcount_begin(&fs->seq);
+			nd->root = fs->root;
+		} while (read_seqcount_retry(&fs->seq, seq));
 	}
 }
 
···
 
 	if (*name=='/') {
 		struct fs_struct *fs = current->fs;
+		unsigned seq;
 
 		br_read_lock(vfsmount_lock);
 		rcu_read_lock();
 
-		spin_lock(&fs->lock);
-		nd->root = fs->root;
-		nd->path = nd->root;
-		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
-		spin_unlock(&fs->lock);
+		do {
+			seq = read_seqcount_begin(&fs->seq);
+			nd->root = fs->root;
+			nd->path = nd->root;
+			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
+		} while (read_seqcount_retry(&fs->seq, seq));
 
 	} else if (dfd == AT_FDCWD) {
 		struct fs_struct *fs = current->fs;
+		unsigned seq;
 
 		br_read_lock(vfsmount_lock);
 		rcu_read_lock();
 
-		spin_lock(&fs->lock);
-		nd->path = fs->pwd;
-		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
-		spin_unlock(&fs->lock);
+		do {
+			seq = read_seqcount_begin(&fs->seq);
+			nd->path = fs->pwd;
+			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
+		} while (read_seqcount_retry(&fs->seq, seq));
+
 	} else {
 		struct dentry *dentry;
 
···
 		if (fput_needed)
 			nd->file = file;
 
-		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
+		nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
 		br_read_lock(vfsmount_lock);
 		rcu_read_lock();
 	}

include/linux/fs_struct.h

···
 #define _LINUX_FS_STRUCT_H
 
 #include <linux/path.h>
+#include <linux/spinlock.h>
+#include <linux/seqlock.h>
 
 struct fs_struct {
 	int users;
 	spinlock_t lock;
+	seqcount_t seq;
 	int umask;
 	int in_exec;
 	struct path root, pwd;