Re: [f2fs-dev] [PATCH 3/6] f2fs: skip truncate when verity in progress in ->write_begin()

[Eric Biggers <ebiggers@kernel.org>]

On Tue, Aug 13, 2019 at 10:40:39AM +0800, Chao Yu wrote:
> Hi Eric,
> 
> On 2019/8/13 6:58, Eric Biggers wrote:
> > Hi Chao,
> > 
> > On Mon, Aug 12, 2019 at 08:25:33PM +0800, Chao Yu wrote:
> >> Hi Eric,
> >>
> >> On 2019/8/12 5:35, Eric Biggers wrote:
> >>> From: Eric Biggers <ebiggers@google.com>
> >>>
> >>> When an error (e.g. ENOSPC) occurs during f2fs_write_begin() when called
> >>> from f2fs_write_merkle_tree_block(), skip truncating the file.  i_size
> >>> is not meaningful in this case, and the truncation is handled by
> >>> f2fs_end_enable_verity() instead.
> >>>
> >>> Fixes: 60d7bf0f790f ("f2fs: add fs-verity support")
> >>> Signed-off-by: Eric Biggers <ebiggers@google.com>
> >>> ---
> >>>  fs/f2fs/data.c | 2 +-
> >>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>
> >>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> >>> index 3f525f8a3a5fa..00b03fb87bd9b 100644
> >>> --- a/fs/f2fs/data.c
> >>> +++ b/fs/f2fs/data.c
> >>> @@ -2476,7 +2476,7 @@ static void f2fs_write_failed(struct address_space *mapping, loff_t to)
> >>>  	struct inode *inode = mapping->host;
> >>>  	loff_t i_size = i_size_read(inode);
> >>>  
> >>> -	if (to > i_size) {
> >>
> >> Maybe adding one single line comment to mention that it's redundant/unnecessary
> >> truncation here is better, if I didn't misunderstand this.
> >>
> >> Thanks,
> >>
> >>> +	if (to > i_size && !f2fs_verity_in_progress(inode)) {
> >>>  		down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
> >>>  		down_write(&F2FS_I(inode)->i_mmap_sem);
> >>>  
> > 
> > Do you mean add a comment instead of the !f2fs_verity_in_progress() check, or in
> > addition to it?  ->write_begin(), ->writepages(), and ->write_end() are all
> 
> Sorry, I didn't make this very clear, I meant adding the comment in addition on
> above change.
> 
> > supposed to ignore i_size when verity is in progress, so I don't think this
> > particular part should be different, even if technically it's still correct to
> > truncate twice.  Also, ext4 needs this check in its ->write_begin() for locking
> > reasons; we should keep f2fs similar.
> 
> Agreed.
> 
> > 
> > How about having both a comment and the check, like:
> > 
> >         /* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
> >         if (to > i_size && !f2fs_verity_in_progress(inode)) {
> 
> The comment looks good to me. :)
> 
> Thanks,
> 
> > 
> > - Eric
> > .
> > 
> 

Okay, this is what I applied:

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 3f525f8a3a5f..54cad80acb7d 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2476,7 +2476,8 @@ static void f2fs_write_failed(struct address_space *mapping, loff_t to)
 	struct inode *inode = mapping->host;
 	loff_t i_size = i_size_read(inode);
 
-	if (to > i_size) {
+	/* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
+	if (to > i_size && !f2fs_verity_in_progress(inode)) {
 		down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
 		down_write(&F2FS_I(inode)->i_mmap_sem);