Re: [PATCH 14/33] vfs: Implement a filesystem superblock creation/configuration context [ver #11]

From: Sergey Senozhatsky
Date: Tue Sep 18 2018 - 21:15:28 EST


On (09/18/18 17:39), David Howells wrote:
[..]
> -static int legacy_init_fs_context(struct fs_context *fc, struct dentry *dentry)
> +int legacy_init_fs_context(struct fs_context *fc, struct dentry *dentry)
> {
> + switch (fc->purpose) {
> + default:
> + fc->fs_private = kzalloc(sizeof(struct legacy_fs_context),
> + GFP_KERNEL);
> + if (!fc->fs_private)
> + return -ENOMEM;

ops->reconfigure() invoked for FS_CONTEXT_FOR_UMOUNT or
FS_CONTEXT_FOR_EMERGENCY_RO will never access fc->fs_private?

> + break;
>
> - fc->fs_private = kzalloc(sizeof(struct legacy_fs_context), GFP_KERNEL);
> - if (!fc->fs_private)
> - return -ENOMEM;
> + case FS_CONTEXT_FOR_UMOUNT:
> + case FS_CONTEXT_FOR_EMERGENCY_RO:
> + break;
> + }

So `fc' can either be zeroed-out, when it comes from fc = kzalloc(),
or contain some garbage otherwise. Would it make sense to zero-out `fc'
regardless of its origin?

> down_write(&sb->s_umount);
> - if (!sb_rdonly(sb))
> - /* Might want to call ->init_fs_context(). */
> - ret = reconfigure_super(&fc);
> + if (!sb_rdonly(sb)) {
> + int ret;
> +
> + if (fc.fs_type->init_fs_context)
> + ret = fc.fs_type->init_fs_context(&fc, NULL);
> + else
> + ret = legacy_init_fs_context(&fc, NULL);
> +
> + if (ret == 0) {
> + ret = reconfigure_super(&fc);
> + fc.ops->free(&fc);
^^^^^^^
Is ops->free() always !NULL?

-ss