| Message ID | 20240319002925.1228063-3-xiubli@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ceph: skip copying the data extends the file EOF | expand |
On Tue, Mar 19, 2024 at 1:32 AM <xiubli@redhat.com> wrote: > > From: Xiubo Li <xiubli@redhat.com> > > In case of hitting the file EOF the ceph_read_iter() needs to > retrieve the file size from MDS, and the Fr caps is not a neccessary. > > URL: https://patchwork.kernel.org/project/ceph-devel/list/?series=819323 > Reported-by: Frank Hsiao 蕭法宣 <frankhsiao@qnap.com> > Signed-off-by: Xiubo Li <xiubli@redhat.com> > Tested-by: Frank Hsiao 蕭法宣 <frankhsiao@qnap.com> > --- > fs/ceph/file.c | 8 +++++--- > 1 file changed, 5 insertions(+), 3 deletions(-) > > diff --git a/fs/ceph/file.c b/fs/ceph/file.c > index c35878427985..a85f95c941fc 100644 > --- a/fs/ceph/file.c > +++ b/fs/ceph/file.c > @@ -2191,14 +2191,16 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) > int statret; > struct page *page = NULL; > loff_t i_size; > + int mask = CEPH_STAT_CAP_SIZE; > if (retry_op == READ_INLINE) { > page = __page_cache_alloc(GFP_KERNEL); > if (!page) > return -ENOMEM; > } > > - statret = __ceph_do_getattr(inode, page, > - CEPH_STAT_CAP_INLINE_DATA, !!page); > + if (retry_op == READ_INLINE) > + mask = CEPH_STAT_CAP_INLINE_DATA; Hi Xiubo, This introduces an additional retry_op == READ_INLINE branch right below an existing one. Should this be: int mask = CEPH_STAT_CAP_SIZE; if (retry_op == READ_INLINE) { page = __page_cache_alloc(GFP_KERNEL); if (!page) return -ENOMEM; mask = CEPH_STAT_CAP_INLINE_DATA; } Thanks, Ilya
On 3/19/24 21:35, Ilya Dryomov wrote: > On Tue, Mar 19, 2024 at 1:32 AM <xiubli@redhat.com> wrote: >> From: Xiubo Li <xiubli@redhat.com> >> >> In case of hitting the file EOF the ceph_read_iter() needs to >> retrieve the file size from MDS, and the Fr caps is not a neccessary. >> >> URL: https://patchwork.kernel.org/project/ceph-devel/list/?series=819323 >> Reported-by: Frank Hsiao 蕭法宣 <frankhsiao@qnap.com> >> Signed-off-by: Xiubo Li <xiubli@redhat.com> >> Tested-by: Frank Hsiao 蕭法宣 <frankhsiao@qnap.com> >> --- >> fs/ceph/file.c | 8 +++++--- >> 1 file changed, 5 insertions(+), 3 deletions(-) >> >> diff --git a/fs/ceph/file.c b/fs/ceph/file.c >> index c35878427985..a85f95c941fc 100644 >> --- a/fs/ceph/file.c >> +++ b/fs/ceph/file.c >> @@ -2191,14 +2191,16 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) >> int statret; >> struct page *page = NULL; >> loff_t i_size; >> + int mask = CEPH_STAT_CAP_SIZE; >> if (retry_op == READ_INLINE) { >> page = __page_cache_alloc(GFP_KERNEL); >> if (!page) >> return -ENOMEM; >> } >> >> - statret = __ceph_do_getattr(inode, page, >> - CEPH_STAT_CAP_INLINE_DATA, !!page); >> + if (retry_op == READ_INLINE) >> + mask = CEPH_STAT_CAP_INLINE_DATA; > Hi Xiubo, > > This introduces an additional retry_op == READ_INLINE branch right > below an existing one. Should this be: > > int mask = CEPH_STAT_CAP_SIZE; > if (retry_op == READ_INLINE) { > page = __page_cache_alloc(GFP_KERNEL); > if (!page) > return -ENOMEM; > > mask = CEPH_STAT_CAP_INLINE_DATA; > } Look good to me. Thanks Ilya. - Xiubo > Thanks, > > Ilya >
diff --git a/fs/ceph/file.c b/fs/ceph/file.c index c35878427985..a85f95c941fc 100644 --- a/fs/ceph/file.c +++ b/fs/ceph/file.c @@ -2191,14 +2191,16 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) int statret; struct page *page = NULL; loff_t i_size; + int mask = CEPH_STAT_CAP_SIZE; if (retry_op == READ_INLINE) { page = __page_cache_alloc(GFP_KERNEL); if (!page) return -ENOMEM; } - statret = __ceph_do_getattr(inode, page, - CEPH_STAT_CAP_INLINE_DATA, !!page); + if (retry_op == READ_INLINE) + mask = CEPH_STAT_CAP_INLINE_DATA; + statret = __ceph_do_getattr(inode, page, mask, !!page); if (statret < 0) { if (page) __free_page(page); @@ -2239,7 +2241,7 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) /* hit EOF or hole? */ if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && ret < len) { - doutc(cl, "hit hole, ppos %lld < size %lld, reading more\n", + doutc(cl, "may hit hole, ppos %lld < size %lld, reading more\n", iocb->ki_pos, i_size); read += ret;