Re: [PATCH v5 6/6] mm,thp: avoid writes to file with THP in pagecache

[Song Liu <songliubraving@fb.com>]

> On Jun 21, 2019, at 6:39 AM, Kirill A. Shutemov <kirill@shutemov.name> wrote:
> 
> On Fri, Jun 21, 2019 at 01:10:54PM +0000, Song Liu wrote:
>> 
>> 
>>> On Jun 21, 2019, at 6:07 AM, Kirill A. Shutemov <kirill@shutemov.name> wrote:
>>> 
>>> On Thu, Jun 20, 2019 at 01:53:48PM -0700, Song Liu wrote:
>>>> In previous patch, an application could put part of its text section in
>>>> THP via madvise(). These THPs will be protected from writes when the
>>>> application is still running (TXTBSY). However, after the application
>>>> exits, the file is available for writes.
>>>> 
>>>> This patch avoids writes to file THP by dropping page cache for the file
>>>> when the file is open for write. A new counter nr_thps is added to struct
>>>> address_space. In do_last(), if the file is open for write and nr_thps
>>>> is non-zero, we drop page cache for the whole file.
>>>> 
>>>> Signed-off-by: Song Liu <songliubraving@fb.com>
>>>> ---
>>>> fs/inode.c         |  3 +++
>>>> fs/namei.c         | 22 +++++++++++++++++++++-
>>>> include/linux/fs.h | 31 +++++++++++++++++++++++++++++++
>>>> mm/filemap.c       |  1 +
>>>> mm/khugepaged.c    |  4 +++-
>>>> 5 files changed, 59 insertions(+), 2 deletions(-)
>>>> 
>>>> diff --git a/fs/inode.c b/fs/inode.c
>>>> index df6542ec3b88..518113a4e219 100644
>>>> --- a/fs/inode.c
>>>> +++ b/fs/inode.c
>>>> @@ -181,6 +181,9 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
>>>> 	mapping->flags = 0;
>>>> 	mapping->wb_err = 0;
>>>> 	atomic_set(&mapping->i_mmap_writable, 0);
>>>> +#ifdef CONFIG_READ_ONLY_THP_FOR_FS
>>>> +	atomic_set(&mapping->nr_thps, 0);
>>>> +#endif
>>>> 	mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
>>>> 	mapping->private_data = NULL;
>>>> 	mapping->writeback_index = 0;
>>>> diff --git a/fs/namei.c b/fs/namei.c
>>>> index 20831c2fbb34..de64f24b58e9 100644
>>>> --- a/fs/namei.c
>>>> +++ b/fs/namei.c
>>>> @@ -3249,6 +3249,22 @@ static int lookup_open(struct nameidata *nd, struct path *path,
>>>> 	return error;
>>>> }
>>>> 
>>>> +/*
>>>> + * The file is open for write, so it is not mmapped with VM_DENYWRITE. If
>>>> + * it still has THP in page cache, drop the whole file from pagecache
>>>> + * before processing writes. This helps us avoid handling write back of
>>>> + * THP for now.
>>>> + */
>>>> +static inline void release_file_thp(struct file *file)
>>>> +{
>>>> +#ifdef CONFIG_READ_ONLY_THP_FOR_FS
>>>> +	struct inode *inode = file_inode(file);
>>>> +
>>>> +	if (inode_is_open_for_write(inode) && filemap_nr_thps(inode->i_mapping))
>>>> +		truncate_pagecache(inode, 0);
>>>> +#endif
>>>> +}
>>>> +
>>>> /*
>>>> * Handle the last step of open()
>>>> */
>>>> @@ -3418,7 +3434,11 @@ static int do_last(struct nameidata *nd,
>>>> 		goto out;
>>>> opened:
>>>> 	error = ima_file_check(file, op->acc_mode);
>>>> -	if (!error && will_truncate)
>>>> +	if (error)
>>>> +		goto out;
>>>> +
>>>> +	release_file_thp(file);
>>> 
>>> What protects against re-fill the file with THP in parallel?
>> 
>> khugepaged would only process vma with VM_DENYWRITE. So once the
>> file is open for write (i_write_count > 0), khugepage will not 
>> collapse the pages. 
> 
> I have not look at the patch very closely. Do you only create THP by
> khugepaged? Not in fault path?

Right. My set only creates THP in khugepaged. William Kucharski is
working on the fault path. 

Thanks,
Song