When cachefiles_shorten_object() calls fallocate() to shape the cache file
to match the DIO size, it passes the total file size it wants to achieve,
not the amount of zeros that should be inserted.  Since this is meant to
preallocate that amount of storage for the file, it can cause the cache to
fill up the disk and hit ENOSPC.

Fix this by passing the length actually required to go from the current EOF
to the desired EOF.

Fixes: 7623ed6772de ("cachefiles: Implement cookie resize for truncate")
Reported-by: Jeffle Xu <jefflexu@linux.alibaba.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: linux-cachefs@redhat.com
---

 fs/cachefiles/interface.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/cachefiles/interface.c b/fs/cachefiles/interface.c
index 51c968cd00a6..ae93cee9d25d 100644
--- a/fs/cachefiles/interface.c
+++ b/fs/cachefiles/interface.c
@@ -254,7 +254,7 @@ static bool cachefiles_shorten_object(struct cachefiles_object *object,
 		ret = cachefiles_inject_write_error();
 		if (ret == 0)
 			ret = vfs_fallocate(file, FALLOC_FL_ZERO_RANGE,
-					    new_size, dio_size);
+					    new_size, dio_size - new_size);
 		if (ret < 0) {
 			trace_cachefiles_io_error(object, file_inode(file), ret,
 						  cachefiles_trace_fallocate_error);

On Thu, 2022-03-03 at 11:55 +0000, David Howells wrote:
> When cachefiles_shorten_object() calls fallocate() to shape the cache file
> to match the DIO size, it passes the total file size it wants to achieve,
> not the amount of zeros that should be inserted.  Since this is meant to
> preallocate that amount of storage for the file, it can cause the cache to
> fill up the disk and hit ENOSPC.
> 
> Fix this by passing the length actually required to go from the current EOF
> to the desired EOF.
> 
> Fixes: 7623ed6772de ("cachefiles: Implement cookie resize for truncate")
> Reported-by: Jeffle Xu <jefflexu@linux.alibaba.com>
> Signed-off-by: David Howells <dhowells@redhat.com>
> cc: linux-cachefs@redhat.com
> ---
> 
>  fs/cachefiles/interface.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/cachefiles/interface.c b/fs/cachefiles/interface.c
> index 51c968cd00a6..ae93cee9d25d 100644
> --- a/fs/cachefiles/interface.c
> +++ b/fs/cachefiles/interface.c
> @@ -254,7 +254,7 @@ static bool cachefiles_shorten_object(struct cachefiles_object *object,
>  		ret = cachefiles_inject_write_error();
>  		if (ret == 0)
>  			ret = vfs_fallocate(file, FALLOC_FL_ZERO_RANGE,
> -					    new_size, dio_size);
> +					    new_size, dio_size - new_size);
>  		if (ret < 0) {
>  			trace_cachefiles_io_error(object, file_inode(file), ret,
>  						  cachefiles_trace_fallocate_error);
> 
> 

Looks good!

I could often force the cache to fill up with the right fsstress run on
ceph, but with this in place I'm on the 5th run of xfstest generic/013
and it hasn't happened yet. You can add these if you like:

Tested-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Jeff Layton <jlayton@kernel.org>