Re: [PATCH v2] f2fs: another way to set large folio by remembering inode number

[Jaegeuk Kim <jaegeuk@kernel.org>]

By the way, is it worth to add some generic apis such as
1) reclaim a specifc inode object when closing the last file
2) add another fadvise hint for large folio

On 04/10, Jaegeuk Kim wrote:
> enum {
>        F2FS_XATTR_FADV_LARGEFOLIO,
> };
> 
> unsigned int value = BIT(F2FS_XATTR_FADV_LARGEFOLIO);
> 
> 1. setxattr(file, "user.fadvise", &value, sizeof(unsigned int), 0)
>  -> register the inode number for large folio
> 2. chmod(0400, file)
>  -> make Read-Only
> 3. fsync() && close() && open(READ)
>  -> f2fs_iget() with large folio
> 4. open(WRITE), mkwrite on mmap, chmod(WRITE)
>  -> return error
> 5. close() and open()
>  -> goto #3
> 6. unlink
>  -> deregister the inode number
> 
> Suggested-by: Akilesh Kailash <akailash@google.com>
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>  
>   Log from v1:
>    - add a condition in f2fs_drop_inode
>    - add Doc
> 
>  Documentation/filesystems/f2fs.rst | 41 ++++++++++++++++++++++++++----
>  fs/f2fs/checkpoint.c               |  2 +-
>  fs/f2fs/data.c                     |  2 +-
>  fs/f2fs/f2fs.h                     |  1 +
>  fs/f2fs/file.c                     | 11 ++++++--
>  fs/f2fs/inode.c                    | 19 +++++++++++---
>  fs/f2fs/super.c                    |  7 +++++
>  fs/f2fs/xattr.c                    | 35 ++++++++++++++++++++++++-
>  fs/f2fs/xattr.h                    |  6 +++++
>  9 files changed, 111 insertions(+), 13 deletions(-)
> 
> diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
> index 7e4031631286..de899d0d3088 100644
> --- a/Documentation/filesystems/f2fs.rst
> +++ b/Documentation/filesystems/f2fs.rst
> @@ -1044,11 +1044,14 @@ page allocation for significant performance gains. To minimize code complexity,
>  this support is currently excluded from the write path, which requires handling
>  complex optimizations such as compression and block allocation modes.
>  
> -This optional feature is triggered only when a file's immutable bit is set.
> -Consequently, F2FS will return EOPNOTSUPP if a user attempts to open a cached
> -file with write permissions, even immediately after clearing the bit. Write
> -access is only restored once the cached inode is dropped. The usage flow is
> -demonstrated below:
> +This optional feature is triggered by two mechanisms: the file's immutable bit
> +or a specific xattr flag. In both cases, F2FS ensures data integrity by
> +restricting the file to a read-only state while large folios are active.
> +
> +1. Immutable Bit Approach:
> +Triggered when the FS_IMMUTABLE_FL is set. This is a strict enforcement
> +where the file cannot be modified at all until the bit is cleared and
> +the cached inode is dropped.
>  
>  .. code-block::
>  
> @@ -1078,3 +1081,31 @@ demonstrated below:
>     Written 4096 bytes with pattern = zero, total_time = 29 us, max_latency = 28 us
>  
>     # rm /data/testfile_read_seq
> +
> +2. XATTR fadvise Approach:
> +A more flexible registration via extended attributes.
> +
> +.. code-block::
> +
> +    enum {
> +        F2FS_XATTR_FADV_LARGEFOLIO,
> +    };
> +    unsigned int value = BIT(F2FS_XATTR_FADV_LARGEFOLIO);
> +
> +    /* Registers the inode number for large folio support in the subsystem.*/
> +    # setxattr(file, "user.fadvise", &value, sizeof(unsigned int), 0)
> +
> +    /* The file must be made Read-Only to transition into the large folio path. */
> +    # fchmod(0400, fd)
> +
> +    /* clean up dirty inode state. */
> +    # fsync(fd)
> +
> +    /* Drop the inode cache.
> +    # close(fd)
> +
> +    /* f2fs_iget() instantiates the inode with large folio support.*/
> +    # open()
> +
> +    /* Returns -EOPNOTSUPP or error to protect the large folio cache.*/
> +    # open(WRITE), mkwrite on mmap, or chmod(WRITE)
> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> index 01e1ba77263e..fdd62ddc3ed6 100644
> --- a/fs/f2fs/checkpoint.c
> +++ b/fs/f2fs/checkpoint.c
> @@ -778,7 +778,7 @@ void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
>  	__remove_ino_entry(sbi, ino, type);
>  }
>  
> -/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
> +/* mode should be APPEND_INO, UPDATE_INO, LARGE_FOLIO_IO, or TRANS_DIR_INO */
>  bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
>  {
>  	struct inode_management *im = &sbi->im[mode];
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 965d4e6443c6..5e46230398d7 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -2494,7 +2494,7 @@ static int f2fs_read_data_large_folio(struct inode *inode,
>  	int ret = 0;
>  	bool folio_in_bio;
>  
> -	if (!IS_IMMUTABLE(inode) || f2fs_compressed_file(inode)) {
> +	if (f2fs_compressed_file(inode)) {
>  		if (folio)
>  			folio_unlock(folio);
>  		return -EOPNOTSUPP;
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index e40b6b2784ee..02bc6eb96a59 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -381,6 +381,7 @@ enum {
>  /* for the list of ino */
>  enum {
>  	ORPHAN_INO,		/* for orphan ino list */
> +	LARGE_FOLIO_INO,	/* for large folio case */
>  	APPEND_INO,		/* for append ino list */
>  	UPDATE_INO,		/* for update ino list */
>  	TRANS_DIR_INO,		/* for transactions dir ino list */
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index c0220cd7b332..64ba900410fc 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -2068,9 +2068,16 @@ static long f2fs_fallocate(struct file *file, int mode,
>  
>  static int f2fs_release_file(struct inode *inode, struct file *filp)
>  {
> -	if (atomic_dec_and_test(&F2FS_I(inode)->open_count))
> +	if (atomic_dec_and_test(&F2FS_I(inode)->open_count)) {
>  		f2fs_remove_donate_inode(inode);
> -
> +		/*
> +		 * In order to get large folio as soon as possible, let's drop
> +		 * inode cache asap. See also f2fs_drop_inode.
> +		 */
> +		if (f2fs_exist_written_data(F2FS_I_SB(inode),
> +					    inode->i_ino, LARGE_FOLIO_INO))
> +                       d_drop(filp->f_path.dentry);
> +	}
>  	/*
>  	 * f2fs_release_file is called at every close calls. So we should
>  	 * not drop any inmemory pages by close called by other process.
> diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> index 89240be8cc59..e100bc5a378c 100644
> --- a/fs/f2fs/inode.c
> +++ b/fs/f2fs/inode.c
> @@ -565,6 +565,20 @@ static bool is_meta_ino(struct f2fs_sb_info *sbi, unsigned int ino)
>  		ino == F2FS_COMPRESS_INO(sbi);
>  }
>  
> +static void f2fs_mapping_set_large_folio(struct inode *inode)
> +{
> +	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> +
> +	if (f2fs_compressed_file(inode))
> +		return;
> +	if (f2fs_quota_file(sbi, inode->i_ino))
> +		return;
> +	if (IS_IMMUTABLE(inode) ||
> +	    (f2fs_exist_written_data(sbi, inode->i_ino, LARGE_FOLIO_INO) &&
> +	     !(inode->i_mode & S_IWUGO)))
> +	    mapping_set_folio_min_order(inode->i_mapping, 0);
> +}
> +
>  struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
>  {
>  	struct f2fs_sb_info *sbi = F2FS_SB(sb);
> @@ -620,9 +634,7 @@ struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
>  		inode->i_op = &f2fs_file_inode_operations;
>  		inode->i_fop = &f2fs_file_operations;
>  		inode->i_mapping->a_ops = &f2fs_dblock_aops;
> -		if (IS_IMMUTABLE(inode) && !f2fs_compressed_file(inode) &&
> -		    !f2fs_quota_file(sbi, inode->i_ino))
> -			mapping_set_folio_min_order(inode->i_mapping, 0);
> +		f2fs_mapping_set_large_folio(inode);
>  	} else if (S_ISDIR(inode->i_mode)) {
>  		inode->i_op = &f2fs_dir_inode_operations;
>  		inode->i_fop = &f2fs_dir_operations;
> @@ -895,6 +907,7 @@ void f2fs_evict_inode(struct inode *inode)
>  	f2fs_remove_ino_entry(sbi, inode->i_ino, APPEND_INO);
>  	f2fs_remove_ino_entry(sbi, inode->i_ino, UPDATE_INO);
>  	f2fs_remove_ino_entry(sbi, inode->i_ino, FLUSH_INO);
> +	f2fs_remove_ino_entry(sbi, inode->i_ino, LARGE_FOLIO_INO);
>  
>  	if (!is_sbi_flag_set(sbi, SBI_IS_FREEZING)) {
>  		sb_start_intwrite(inode->i_sb);
> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> index ccf806b676f5..11d1e0c99ac1 100644
> --- a/fs/f2fs/super.c
> +++ b/fs/f2fs/super.c
> @@ -1844,6 +1844,13 @@ static int f2fs_drop_inode(struct inode *inode)
>  			return 1;
>  		}
>  	}
> +	/*
> +	 * In order to get large folio as soon as possible, let's drop
> +	 * inode cache asap. See also f2fs_release_file.
> +	 */
> +	if (f2fs_exist_written_data(sbi, inode->i_ino, LARGE_FOLIO_INO) &&
> +	    !is_inode_flag_set(inode, FI_DIRTY_INODE))
> +		return 1;
>  
>  	/*
>  	 * This is to avoid a deadlock condition like below.
> diff --git a/fs/f2fs/xattr.c b/fs/f2fs/xattr.c
> index 941dc62a6d6f..0c0e44c2dcdd 100644
> --- a/fs/f2fs/xattr.c
> +++ b/fs/f2fs/xattr.c
> @@ -44,6 +44,16 @@ static void xattr_free(struct f2fs_sb_info *sbi, void *xattr_addr,
>  		kfree(xattr_addr);
>  }
>  
> +static int f2fs_xattr_fadvise_get(struct inode *inode, void *buffer)
> +{
> +	if (!buffer)
> +		goto out;
> +	if (mapping_large_folio_support(inode->i_mapping))
> +		*((unsigned int *)buffer) |= BIT(F2FS_XATTR_FADV_LARGEFOLIO);
> +out:
> +	return sizeof(unsigned int);
> +}
> +
>  static int f2fs_xattr_generic_get(const struct xattr_handler *handler,
>  		struct dentry *unused, struct inode *inode,
>  		const char *name, void *buffer, size_t size)
> @@ -61,10 +71,29 @@ static int f2fs_xattr_generic_get(const struct xattr_handler *handler,
>  	default:
>  		return -EINVAL;
>  	}
> +	if (handler->flags == F2FS_XATTR_INDEX_USER &&
> +	    !strcmp(name, "fadvise"))
> +		return f2fs_xattr_fadvise_get(inode, buffer);
> +
>  	return f2fs_getxattr(inode, handler->flags, name,
>  			     buffer, size, NULL);
>  }
>  
> +static int f2fs_xattr_fadvise_set(struct inode *inode, const void *value)
> +{
> +	unsigned int new_fadvise;
> +
> +	new_fadvise = *(unsigned int *)value;
> +
> +	if (new_fadvise & BIT(F2FS_XATTR_FADV_LARGEFOLIO))
> +		f2fs_add_ino_entry(F2FS_I_SB(inode),
> +				inode->i_ino, LARGE_FOLIO_INO);
> +	else
> +		f2fs_remove_ino_entry(F2FS_I_SB(inode),
> +				inode->i_ino, LARGE_FOLIO_INO);
> +	return 0;
> +}
> +
>  static int f2fs_xattr_generic_set(const struct xattr_handler *handler,
>  		struct mnt_idmap *idmap,
>  		struct dentry *unused, struct inode *inode,
> @@ -84,6 +113,10 @@ static int f2fs_xattr_generic_set(const struct xattr_handler *handler,
>  	default:
>  		return -EINVAL;
>  	}
> +	if (handler->flags == F2FS_XATTR_INDEX_USER &&
> +	    !strcmp(name, "fadvise"))
> +		return f2fs_xattr_fadvise_set(inode, value);
> +
>  	return f2fs_setxattr(inode, handler->flags, name,
>  					value, size, NULL, flags);
>  }
> @@ -842,4 +875,4 @@ int __init f2fs_init_xattr_cache(void)
>  void f2fs_destroy_xattr_cache(void)
>  {
>  	kmem_cache_destroy(inline_xattr_slab);
> -}
> \ No newline at end of file
> +}
> diff --git a/fs/f2fs/xattr.h b/fs/f2fs/xattr.h
> index bce3d93e4755..455f460d014e 100644
> --- a/fs/f2fs/xattr.h
> +++ b/fs/f2fs/xattr.h
> @@ -24,6 +24,7 @@
>  #define F2FS_XATTR_REFCOUNT_MAX         1024
>  
>  /* Name indexes */
> +#define F2FS_USER_FADVISE_NAME			"user.fadvise"
>  #define F2FS_SYSTEM_ADVISE_NAME			"system.advise"
>  #define F2FS_XATTR_INDEX_USER			1
>  #define F2FS_XATTR_INDEX_POSIX_ACL_ACCESS	2
> @@ -39,6 +40,11 @@
>  #define F2FS_XATTR_NAME_ENCRYPTION_CONTEXT	"c"
>  #define F2FS_XATTR_NAME_VERITY			"v"
>  
> +/* used for F2FS_USER_FADVISE_NAME */
> +enum {
> +	F2FS_XATTR_FADV_LARGEFOLIO,
> +};
> +
>  struct f2fs_xattr_header {
>  	__le32  h_magic;        /* magic number for identification */
>  	__le32  h_refcount;     /* reference count */
> -- 
> 2.53.0.1213.gd9a14994de-goog
>