Re: [PATCH v2] erofs: accept source file descriptor via fsconfig

Gao Xiang <[email protected]> writes:

> On 2026/7/14 14:36, Giuseppe Scrivano wrote:
>> Gao Xiang <[email protected]> writes:
>> 
>>> On 2026/7/14 08:49, Aleksa Sarai wrote:
>>>> On 2026-07-13, Giuseppe Scrivano <[email protected]> wrote:
>>>>> thanks for the hints.
>>>>>
>>>>> I'll prepare a v3 if you are fine with the version below:
>>>> No worries, and this seems more reasonable at a first glance.
>>>>
>>>>> diff --git a/fs/erofs/super.c b/fs/erofs/super.c
>>>>> index 86fa5c6a0c70..72c85cc53085 100644
>>>>> --- a/fs/erofs/super.c
>>>>> +++ b/fs/erofs/super.c
>>>> ...
>>>>> @@ -437,6 +439,38 @@ static bool erofs_fc_set_dax_mode(struct fs_context *fc, unsigned int mode)
>>>>>    	return false;
>>>>>    }
>>>>>    +static int erofs_fc_parse_source(struct fs_context *fc,
>>>>> +				 struct fs_parameter *param)
>>>>> +{
>>>>> +	struct erofs_sb_info *sbi = fc->s_fs_info;
>>>>> +
>>>>> +	if (fc->source || sbi->dif0.file)
>>>>> +		return invalf(fc, "Multiple sources");
>>>>> +
>>>>> +	switch (param->type) {
>>>>> +	case fs_value_is_string:
>>>>> +		fc->source = param->string;
>>>>> +		param->string = NULL;
>>>>> +		return 0;
>>>>> +	case fs_value_is_file: {
>>>>> +		char *buf, *p;
>>>>> +
>>>>> +		sbi->dif0.file = get_file(param->file);
>>>> A very minor nit, but you can actually steal the file reference here
>>>> with
>>>> 		sbi->dif0.file = no_free_ptr(param->file);
>>>> A few other places do this. (You'll also need to change the
>>>> param->file
>>>> reference below.)
>>>>
>>>>> +		buf = kmalloc(PATH_MAX, GFP_KERNEL);
>>>>> +		if (!buf)
>>>>> +			return -ENOMEM;
>>>>> +		p = file_path(param->file, buf, PATH_MAX);
>>>>> +		fc->source = kstrdup(IS_ERR(p) ? "(fd)" : p, GFP_KERNEL);
>>>> I think that /proc/self/fd/%d would be a more useful name for
>>>> debugging
>>>> if file_path() fails (not that it is really possible here AFAICS). But
>>>> I'm not really too fussed.
>>>
>>> Not quite sure if we should get in agreement with the format of this
>>> one in advance (IOWs, users use source_fd and how fc->source looks like;
>>> since other fses may follow the same practice if source_fd becomes common
>>> later) since it's a user-visible field and I believe we shouldn't treat
>>> this one as a dontcare field as some pseudo fses (since those fses don't
>>> rely on `fc->source` by design but typically EROFS can rely on.)
>>>
>>> I hope Christian and others could share move thought on this part too
>>> before I land this feature for the next cycle.
>> would it be better to just return the error from file_path without
>> any
>> fallback?
>
> I hope Christian or other vfs folks can decide how to handle fc->source
> string here, since in the long term, how to deal with source_fd should
> be unique among different fses: just our current short-term
> implementation lands into erofs directly for file-backed mounts to
> fulfill composefs needs.
>

overlayfs is already handling such a case, and it resolves the path using d_path:

static int ovl_parse_layer(struct fs_context *fc, struct fs_parameter *param,
			   enum ovl_opt layer)
{
	struct path layer_path __free(path_put) = {};
	int err = 0;

	switch (param->type) {
	case fs_value_is_string:
		err = ovl_kern_path(param->string, &layer_path, layer);
		if (err)
			return err;
		err = ovl_do_parse_layer(fc, param->string, &layer_path, layer);
		break;
	case fs_value_is_file: {
		char *buf __free(kfree);
		char *layer_name;

		buf = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
		if (!buf)
			return -ENOMEM;

		layer_path = param->file->f_path;
		path_get(&layer_path);

		layer_name = d_path(&layer_path, buf, PATH_MAX);
		if (IS_ERR(layer_name))
			return PTR_ERR(layer_name);

		err = ovl_do_parse_layer(fc, layer_name, &layer_path, layer);
		break;
	}
	default:
		WARN_ON_ONCE(true);
		err = -EINVAL;
	}

	return err;
}

If there are no objections, I'll change the code to match what overlay
does and return the error instead of using a fallback.

Regards,
Giuseppe