On Wednesday, February 20, 2019 5:51:00 AM IST Eric Biggers wrote:
> Hi Chandan,
>
> On Mon, Feb 18, 2019 at 03:34:29PM +0530, Chandan Rajendra wrote:
> > Ext4 and F2FS currently use a non-NULL value stored at bio->bi_private
> > to determine if the contents of the bio need to be "post processed"
> > i.e. whether its contents need to be decrypted and/or verified. For
> > block size < page size scenario, bio->bi_private would hold a pointer to
> > buffer_head. Hence, this commit adds the new flag REQ_POST_READ_PROC to
> > be able to decisively check for post process requirement for a bio.
> >
> > Signed-off-by: Chandan Rajendra <chandan@xxxxxxxxxxxxx>
> > ---
> > fs/ext4/readpage.c | 11 +++++++++--
> > fs/post_read_process.c | 2 +-
> > include/linux/blk_types.h | 2 ++
> > 3 files changed, 12 insertions(+), 3 deletions(-)
> >
> > diff --git a/fs/ext4/readpage.c b/fs/ext4/readpage.c
> > index 8943fc41fd33..c7dbab35deaa 100644
> > --- a/fs/ext4/readpage.c
> > +++ b/fs/ext4/readpage.c
> > @@ -245,6 +245,7 @@ int ext4_mpage_readpages(struct address_space *mapping,
> > }
> > if (bio == NULL) {
> > struct bio_post_read_ctx *ctx;
> > + unsigned int op_flags = 0;
> >
> > bio = bio_alloc(GFP_KERNEL,
> > min_t(int, nr_pages, BIO_MAX_PAGES));
> > @@ -259,8 +260,14 @@ int ext4_mpage_readpages(struct address_space *mapping,
> > bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
> > bio->bi_end_io = mpage_end_io;
> > bio->bi_private = ctx;
> > - bio_set_op_attrs(bio, REQ_OP_READ,
> > - is_readahead ? REQ_RAHEAD : 0);
> > +
> > + if (is_readahead)
> > + op_flags |= REQ_RAHEAD;
> > +
> > + if (ctx)
> > + op_flags |= REQ_POST_READ_PROC;
> > +
> > + bio_set_op_attrs(bio, REQ_OP_READ, op_flags);
> > }
> >
> > length = first_hole << blkbits;
> > diff --git a/fs/post_read_process.c b/fs/post_read_process.c
> > index 1f8663d70247..66c1c6e57e70 100644
> > --- a/fs/post_read_process.c
> > +++ b/fs/post_read_process.c
> > @@ -104,7 +104,7 @@ void put_bio_post_read_ctx(struct bio_post_read_ctx *ctx)
> >
> > bool bio_post_read_required(struct bio *bio)
> > {
> > - return bio->bi_private && !bio->bi_status;
> > + return bio->bi_opf & REQ_POST_READ_PROC;
> > }
> >
> > static int __init bio_init_post_read_processing(void)
> > diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
> > index 5c7e7f859a24..6904945c8c40 100644
> > --- a/include/linux/blk_types.h
> > +++ b/include/linux/blk_types.h
> > @@ -320,6 +320,7 @@ enum req_flag_bits {
> > __REQ_RAHEAD, /* read ahead, can fail anytime */
> > __REQ_BACKGROUND, /* background IO */
> > __REQ_NOWAIT, /* Don't wait if request will block */
> > + __REQ_POST_READ_PROC,
> >
> > /* command specific flags for REQ_OP_WRITE_ZEROES: */
> > __REQ_NOUNMAP, /* do not free blocks when zeroing */
> > @@ -346,6 +347,7 @@ enum req_flag_bits {
> > #define REQ_RAHEAD (1ULL << __REQ_RAHEAD)
> > #define REQ_BACKGROUND (1ULL << __REQ_BACKGROUND)
> > #define REQ_NOWAIT (1ULL << __REQ_NOWAIT)
> > +#define REQ_POST_READ_PROC (1ULL << __REQ_POST_READ_PROC)
> > #define REQ_NOUNMAP (1ULL << __REQ_NOUNMAP)
> > #define REQ_HIPRI (1ULL << __REQ_HIPRI)
> >
>
> I don't think this is an appropriate use of a request flag, as request flags are
> meant for the block layer.
>
> Also doesn't the bio still need a pointer to the bio_post_read_ctx anyway? So I
> don't see how this would solve the problem, if ->bi_private is already used.
I had glanced across block_read_full_page() function which implements reading
non-contiguous blocks mapped by the page in block size < page size
scenario. Here bio->bi_private would point to the buffer head that represents
the block on which read I/O was performed. In such a case,
bio_post_read_required() would always return true. Hence I decided to add this
request flag.
Now, I believe we can actually save away the ctx pointer in bh->b_private
member and later decide on whether the buffer head requires post processing
based on bh->bi_private's non-NULL value. I will have to read up the code more
thoroughly to confirm this.
--
chandan
