Re: [PATCH v2 3/9] mm/mshare: make msharefs writable and support directories

[Khalid Aziz <khalid.aziz@oracle.com>]

On 6/30/22 15:34, Darrick J. Wong wrote:
> On Wed, Jun 29, 2022 at 04:53:54PM -0600, Khalid Aziz wrote:
>> Make msharefs filesystem writable and allow creating directories
>> to support better access control to mshare'd regions defined in
>> msharefs.
>>
>> Signed-off-by: Khalid Aziz <khalid.aziz@oracle.com>
>> ---
>>   mm/mshare.c | 195 +++++++++++++++++++++++++++++++++++++++++++++++++---
>>   1 file changed, 186 insertions(+), 9 deletions(-)
>>
>> diff --git a/mm/mshare.c b/mm/mshare.c
>> index 3e448e11c742..2d5924d39221 100644
>> --- a/mm/mshare.c
>> +++ b/mm/mshare.c
>> @@ -21,11 +21,21 @@
>>   #include <linux/fileattr.h>
>>   #include <uapi/linux/magic.h>
>>   #include <uapi/linux/limits.h>
>> +#include <uapi/linux/mman.h>
>>   
>>   static struct super_block *msharefs_sb;
>>   
>> +static const struct inode_operations msharefs_dir_inode_ops;
>> +static const struct inode_operations msharefs_file_inode_ops;
>> +
>> +static int
>> +msharefs_open(struct inode *inode, struct file *file)
>> +{
>> +	return simple_open(inode, file);
>> +}
>> +
>>   static const struct file_operations msharefs_file_operations = {
>> -	.open		= simple_open,
>> +	.open		= msharefs_open,
>>   	.llseek		= no_llseek,
>>   };
>>   
>> @@ -42,6 +52,113 @@ msharefs_d_hash(const struct dentry *dentry, struct qstr *qstr)
>>   	return 0;
>>   }
>>   
>> +static struct dentry
>> +*msharefs_alloc_dentry(struct dentry *parent, const char *name)
>> +{
>> +	struct dentry *d;
>> +	struct qstr q;
>> +	int err;
>> +
>> +	q.name = name;
>> +	q.len = strlen(name);
>> +
>> +	err = msharefs_d_hash(parent, &q);
>> +	if (err)
>> +		return ERR_PTR(err);
>> +
>> +	d = d_alloc(parent, &q);
>> +	if (d)
>> +		return d;
>> +
>> +	return ERR_PTR(-ENOMEM);
>> +}
>> +
>> +static struct inode
>> +*msharefs_get_inode(struct super_block *sb, const struct inode *dir,
>> +			umode_t mode)
>> +{
>> +	struct inode *inode = new_inode(sb);
>> +
>> +	if (inode) {
> 
> Not sure why you wouldn't go with the less-indently version:
> 
> 	if (!inode)
> 		return ERR_PTR(-ENOMEM);
> 
> 	inode->i_ino = get_next_ino();
> 	<etc>
> 

Yeah, good idea. I will change it.

>> +		inode->i_ino = get_next_ino();
>> +		inode_init_owner(&init_user_ns, inode, dir, mode);
>> +
>> +		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
>> +
>> +		switch (mode & S_IFMT) {
> 
> Shouldn't we set the mode somewhere?

mode is passed in as parameter to msharefs_get_inode() which uses this value to determine its actions.

> 
>> +		case S_IFREG:
>> +			inode->i_op = &msharefs_file_inode_ops;
>> +			inode->i_fop = &msharefs_file_operations;
>> +			break;
>> +		case S_IFDIR:
>> +			inode->i_op = &msharefs_dir_inode_ops;
>> +			inode->i_fop = &simple_dir_operations;
>> +			inc_nlink(inode);
>> +			break;
>> +		case S_IFLNK:
>> +			inode->i_op = &page_symlink_inode_operations;
>> +			break;
>> +		default:
>> +			discard_new_inode(inode);
>> +			inode = NULL;
>> +			break;
>> +		}
>> +	}
>> +
>> +	return inode;
>> +}
>> +
>> +static int
>> +msharefs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
>> +		struct dentry *dentry, umode_t mode, dev_t dev)
>> +{
>> +	struct inode *inode;
>> +	int err = 0;
>> +
>> +	inode = msharefs_get_inode(dir->i_sb, dir, mode);
>> +	if (IS_ERR(inode))
>> +		return PTR_ERR(inode);
> 
> ...and if @inode is NULL?

Oh right, IS_ERR() does not check for NULL value. I will add a check for that and return ENOMEM.

> 
>> +
>> +	d_instantiate(dentry, inode);
>> +	dget(dentry);
>> +	dir->i_mtime = dir->i_ctime = current_time(dir);
>> +
>> +	return err;
>> +}
>> +
>> +static int
>> +msharefs_create(struct user_namespace *mnt_userns, struct inode *dir,
>> +		struct dentry *dentry, umode_t mode, bool excl)
>> +{
>> +	return msharefs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
>> +}
>> +
>> +static int
>> +msharefs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
>> +		struct dentry *dentry, umode_t mode)
>> +{
>> +	int ret = msharefs_mknod(&init_user_ns, dir, dentry, mode | S_IFDIR, 0);
>> +
>> +	if (!ret)
>> +		inc_nlink(dir);
>> +	return ret;
>> +}
>> +
>> +static const struct inode_operations msharefs_file_inode_ops = {
>> +	.setattr	= simple_setattr,
>> +	.getattr	= simple_getattr,
>> +};
>> +static const struct inode_operations msharefs_dir_inode_ops = {
>> +	.create		= msharefs_create,
>> +	.lookup		= simple_lookup,
>> +	.link		= simple_link,
>> +	.unlink		= simple_unlink,
>> +	.mkdir		= msharefs_mkdir,
>> +	.rmdir		= simple_rmdir,
>> +	.mknod		= msharefs_mknod,
>> +	.rename		= simple_rename,
>> +};
>> +
>>   static void
>>   mshare_evict_inode(struct inode *inode)
>>   {
>> @@ -58,7 +175,7 @@ mshare_info_read(struct file *file, char __user *buf, size_t nbytes,
>>   {
>>   	char s[80];
>>   
>> -	sprintf(s, "%ld", PGDIR_SIZE);
>> +	sprintf(s, "%ld\n", PGDIR_SIZE);
> 
> Changing this already?

Possibly. There is one suggestion to change it to PMD and it might be a better choice.

> 
>>   	return simple_read_from_buffer(buf, nbytes, ppos, s, strlen(s));
>>   }
>>   
>> @@ -72,6 +189,38 @@ static const struct super_operations mshare_s_ops = {
>>   	.evict_inode = mshare_evict_inode,
>>   };
>>   
>> +static int
>> +prepopulate_files(struct super_block *s, struct inode *dir,
>> +			struct dentry *root, const struct tree_descr *files)
>> +{
>> +	int i;
>> +	struct inode *inode;
>> +	struct dentry *dentry;
>> +
>> +	for (i = 0; !files->name || files->name[0]; i++, files++) {
>> +		if (!files->name)
>> +			continue;
> 
> What ends the array?  NULL name or empty name?
> Do we have to erase all of these when the fs gets unmounted?

This code is very similar to simple_fill_super() and I reused the code from there. inodes and dentries will need to be 
erased on unmount through evict_inode.

Thanks,
Khalid