Jeff Layton <jlayton@xxxxxxxxxx> writes:
> On Thu, 2022-04-07 at 15:38 +0100, Luís Henriques wrote:
>> When doing a direct/sync write, we need to invalidate the page cache in
>> the range being written to. If we don't do this, the cache will include
>> invalid data as we just did a write that avoided the page cache.
>>
>> Signed-off-by: Luís Henriques <lhenriques@xxxxxxx>
>> ---
>> fs/ceph/file.c | 9 +++++++++
>> 1 file changed, 9 insertions(+)
>>
>> Ok, here's a new attempt. After discussion in this thread and on IRC, I
>> think this is the right fix. generic/647 now passes with and without
>> encryption. Thanks!
>>
>> Changes since v2:
>> - Invalidation needs to be done after a write
>>
>> Changes since v1:
>> - Replaced truncate_inode_pages_range() by invalidate_inode_pages2_range
>> - Call fscache_invalidate with FSCACHE_INVAL_DIO_WRITE if we're doing DIO
>>
>> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
>> index 5072570c2203..63e67eb60310 100644
>> --- a/fs/ceph/file.c
>> +++ b/fs/ceph/file.c
>> @@ -1938,6 +1938,15 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
>> break;
>> }
>> ceph_clear_error_write(ci);
>> + ret = invalidate_inode_pages2_range(
>> + inode->i_mapping,
>> + pos >> PAGE_SHIFT,
>> + (pos + len - 1) >> PAGE_SHIFT);
>> + if (ret < 0) {
>> + dout("invalidate_inode_pages2_range returned %d\n",
>> + ret);
>> + ret = 0;
>> + }
>> pos += len;
>> written += len;
>> dout("sync_write written %d\n", written);
>
> Looks good. I suspect we can also remove the
> invalidate_indode_pages2_range call earlier in this function too. I may
> roll that into this patch.
Right, that occurred to me as well but I wasn't really sure that would be
safe.
> I'll give this an xfstests run with fscrypt enabled and see how it does.
I'll do the same here, I just run a few tests on a vstart cluster. But
I'll definitely give it some more testing.
Cheers,
--
Luís
