shmem: fix recovery on rename failures

maple_tree insertions can fail if we are seriously short on memory;
simple_offset_rename() does not recover well if it runs into that.
The same goes for simple_offset_rename_exchange().

Moreover, shmem_whiteout() expects that if it succeeds, the caller will
progress to d_move(), i.e. that shmem_rename2() won't fail past the
successful call of shmem_whiteout().

Not hard to fix, fortunately - mtree_store() can't fail if the index we
are trying to store into is already present in the tree as a singleton.

For simple_offset_rename_exchange() that's enough - we just need to be
careful about the order of operations.

For simple_offset_rename() solution is to preinsert the target into the
tree for new_dir; the rest can be done without any potentially failing
operations.

That preinsertion has to be done in shmem_rename2() rather than in
simple_offset_rename() itself - otherwise we'd need to deal with the
possibility of failure after successful shmem_whiteout().

Fixes: a2e4595 ("shmem: stable directory offsets")
Reviewed-by: Christian Brauner <brauner@kernel.org>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Author: Al Viro

fs/libfs.c

@@ -346,22 +346,22 @@ void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry)
  * User space expects the directory offset value of the replaced
  * (new) directory entry to be unchanged after a rename.
  *
- * Returns zero on success, a negative errno value on failure.
+ * Caller must have grabbed a slot for new_dentry in the maple_tree
+ * associated with new_dir, even if dentry is negative.
  */
-int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
-			 struct inode *new_dir, struct dentry *new_dentry)
+void simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
+			  struct inode *new_dir, struct dentry *new_dentry)
 {
 	struct offset_ctx *old_ctx = old_dir->i_op->get_offset_ctx(old_dir);
 	struct offset_ctx *new_ctx = new_dir->i_op->get_offset_ctx(new_dir);
 	long new_offset = dentry2offset(new_dentry);
 
-	simple_offset_remove(old_ctx, old_dentry);
+	if (WARN_ON(!new_offset))
+		return;
 
-	if (new_offset) {
-		offset_set(new_dentry, 0);
-		return simple_offset_replace(new_ctx, old_dentry, new_offset);
-	}
-	return simple_offset_add(new_ctx, old_dentry);
+	simple_offset_remove(old_ctx, old_dentry);
+	offset_set(new_dentry, 0);
+	WARN_ON(simple_offset_replace(new_ctx, old_dentry, new_offset));
 }
 
 /**
@@ -388,31 +388,23 @@ int simple_offset_rename_exchange(struct inode *old_dir,
 	long new_index = dentry2offset(new_dentry);
 	int ret;
 
-	simple_offset_remove(old_ctx, old_dentry);
-	simple_offset_remove(new_ctx, new_dentry);
+	if (WARN_ON(!old_index || !new_index))
+		return -EINVAL;
 
-	ret = simple_offset_replace(new_ctx, old_dentry, new_index);
-	if (ret)
-		goto out_restore;
+	ret = mtree_store(&new_ctx->mt, new_index, old_dentry, GFP_KERNEL);
+	if (WARN_ON(ret))
+		return ret;
 
-	ret = simple_offset_replace(old_ctx, new_dentry, old_index);
-	if (ret) {
-		simple_offset_remove(new_ctx, old_dentry);
-		goto out_restore;
+	ret = mtree_store(&old_ctx->mt, old_index, new_dentry, GFP_KERNEL);
+	if (WARN_ON(ret)) {
+		mtree_store(&new_ctx->mt, new_index, new_dentry, GFP_KERNEL);
+		return ret;
 	}
 
-	ret = simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
-	if (ret) {
-		simple_offset_remove(new_ctx, old_dentry);
-		simple_offset_remove(old_ctx, new_dentry);
-		goto out_restore;
-	}
+	offset_set(old_dentry, new_index);
+	offset_set(new_dentry, old_index);
+	simple_rename_exchange(old_dir, old_dentry, new_dir, new_dentry);
 	return 0;
-
-out_restore:
-	(void)simple_offset_replace(old_ctx, old_dentry, old_index);
-	(void)simple_offset_replace(new_ctx, new_dentry, new_index);
-	return ret;
 }
 
 /**

include/linux/fs.h

@@ -3247,7 +3247,7 @@ struct offset_ctx {
 void simple_offset_init(struct offset_ctx *octx);
 int simple_offset_add(struct offset_ctx *octx, struct dentry *dentry);
 void simple_offset_remove(struct offset_ctx *octx, struct dentry *dentry);
-int simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
+void simple_offset_rename(struct inode *old_dir, struct dentry *old_dentry,
 			 struct inode *new_dir, struct dentry *new_dentry);
 int simple_offset_rename_exchange(struct inode *old_dir,
 				  struct dentry *old_dentry,

mm/shmem.c

@@ -4038,6 +4038,7 @@ static int shmem_rename2(struct mnt_idmap *idmap,
 {
 	struct inode *inode = d_inode(old_dentry);
 	int they_are_dirs = S_ISDIR(inode->i_mode);
+	bool had_offset = false;
 	int error;
 
 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
@@ -4050,16 +4051,23 @@ static int shmem_rename2(struct mnt_idmap *idmap,
 	if (!simple_empty(new_dentry))
 		return -ENOTEMPTY;
 
+	error = simple_offset_add(shmem_get_offset_ctx(new_dir), new_dentry);
+	if (error == -EBUSY)
+		had_offset = true;
+	else if (unlikely(error))
+		return error;
+
 	if (flags & RENAME_WHITEOUT) {
 		error = shmem_whiteout(idmap, old_dir, old_dentry);
-		if (error)
+		if (error) {
+			if (!had_offset)
+				simple_offset_remove(shmem_get_offset_ctx(new_dir),
+						     new_dentry);
 			return error;
+		}
 	}
 
-	error = simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
-	if (error)
-		return error;
-
+	simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry);
 	if (d_really_is_positive(new_dentry)) {
 		(void) shmem_unlink(new_dir, new_dentry);
 		if (they_are_dirs) {