Re: [PATCH next 3/3] debugfs: introduce debugfs_create_seq[,_data]

[Greg KH <gregkh@linuxfoundation.org>]

On Fri, Nov 29, 2019 at 11:03:47PM +0800, Kefeng Wang wrote:
> 
> 
> On 2019/11/29 22:22, Greg KH wrote:
> > On Fri, Nov 29, 2019 at 05:27:52PM +0800, Kefeng Wang wrote:
> >> Like proc_create_seq[,_data] in procfs, introduce a similar
> >> debugfs_create_seq[,_data] function, which could drastically
> >> reduces the boilerplate code.
> >>
> >> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >> ---
> >>  fs/debugfs/file.c       | 62 ++++++++++++++++++++++++++++++++++++++++-
> >>  include/linux/debugfs.h | 16 +++++++++++
> >>  2 files changed, 77 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/fs/debugfs/file.c b/fs/debugfs/file.c
> >> index 68f0c4b19bef..77522717c9fb 100644
> >> --- a/fs/debugfs/file.c
> >> +++ b/fs/debugfs/file.c
> >> @@ -1107,7 +1107,10 @@ EXPORT_SYMBOL_GPL(debugfs_create_regset32);
> >>  #endif /* CONFIG_HAS_IOMEM */
> >>  
> >>  struct debugfs_entry {
> >> -	int (*read)(struct seq_file *seq, void *data);
> >> +	union {
> >> +		const struct seq_operations *seq_ops;
> >> +		int (*read)(struct seq_file *seq, void *data);
> >> +	};
> >>  	void *data;
> >>  };
> >>  
> >> @@ -1196,3 +1199,60 @@ struct dentry *debugfs_create_single_data(const char *name, umode_t mode,
> >>  				   &debugfs_entry_ops);
> >>  }
> >>  EXPORT_SYMBOL_GPL(debugfs_create_single_data);
> >> +
> >> +static int debugfs_entry_seq_open(struct inode *inode, struct file *file)
> >> +{
> >> +	struct debugfs_entry *entry = inode->i_private;
> >> +	int ret;
> >> +
> >> +	entry = debugfs_clear_lowest_bit(entry);
> >> +
> >> +	ret = seq_open(file, entry->seq_ops);
> >> +	if (!ret && entry->data) {
> >> +		struct seq_file *seq = file->private_data;
> >> +		seq->private = entry->data;
> >> +	}
> >> +
> >> +	return ret;
> >> +}
> >> +
> >> +static const struct file_operations debugfs_seq_fops = {
> >> +	.open		= debugfs_entry_seq_open,
> >> +	.read		= seq_read,
> >> +	.llseek		= seq_lseek,
> >> +	.release	= seq_release,
> >> +};
> >> +
> >> +/**
> >> + * debugfs_create_seq_data - create a file in the debugfs filesystem
> >> + * @name: a pointer to a string containing the name of the file to create.
> >> + * @mode: the permission that the file should have.
> >> + * @parent: a pointer to the parent dentry for this file.  This should be a
> >> + *          directory dentry if set.  If this parameter is NULL, then the
> >> + *          file will be created in the root of the debugfs filesystem.
> >> + * @data: a pointer to something that the caller will want to get to later
> >> + *        on.  The inode.i_private pointer will point to this value on
> >> + *        the open() call.
> >> + * @seq_ops: seq_operations pointer of the seqfile.
> >> + *
> >> + * This function creates a file in debugfs with the extra data and a seq_ops.
> >> + */
> >> +struct dentry *debugfs_create_seq_data(const char *name, umode_t mode,
> >> +				       struct dentry *parent, void *data,
> >> +				       const struct seq_operations *seq_ops)
> >> +{
> >> +	struct debugfs_entry *entry;
> >> +
> >> +	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
> >> +	if (!entry)
> >> +		return ERR_PTR(-ENOMEM);
> >> +
> >> +	entry->seq_ops = seq_ops;
> >> +	entry->data = data;
> >> +
> >> +	entry = debugfs_set_lowest_bit(entry);
> >> +
> >> +	return debugfs_create_file(name, mode, parent, entry,
> >> +				   &debugfs_seq_fops);
> >> +}
> >> +EXPORT_SYMBOL_GPL(debugfs_create_seq_data);
> >> diff --git a/include/linux/debugfs.h b/include/linux/debugfs.h
> >> index 82171f183e93..d32d49983547 100644
> >> --- a/include/linux/debugfs.h
> >> +++ b/include/linux/debugfs.h
> >> @@ -147,6 +147,10 @@ struct dentry *debugfs_create_single_data(const char *name, umode_t mode,
> >>  					  int (*read_fn)(struct seq_file *s,
> >>  							 void *data));
> >>  
> >> +struct dentry *debugfs_create_seq_data(const char *name, umode_t mode,
> >> +				       struct dentry *parent, void *data,
> >> +				       const struct seq_operations *seq_ops);
> >> +
> >>  bool debugfs_initialized(void);
> >>  
> >>  ssize_t debugfs_read_file_bool(struct file *file, char __user *user_buf,
> > 
> > If you notice, I have been removing the return value of the
> > debugfs_create_* functions over the past few kernel versions (look at
> > 5.5-rc1 for a lot more).  Please don't add any new functions that also
> > return a dentry that I then need to go and remove.
> 
> Hi Greg, see function debugfs_create_seq_data()and debugfs_create_single_data(),
> they are wrapper function of debugfs_create_file(), when remove the debugfs file
> called debugfs_remove(), some drivers could do such thing, we must return a dentry
> to the caller.

Again, no, we should not return a dentry.  Files should go in
directories and then removed all at once if they need to, which is why I
have been making these types of changes.

Yes, for some existing files that does not work, but again, please do
not create a new api that does this.  Only use it for the files that do
not need their dentries saved.

thanks,

greg k-h