From: Nikolay Borisov <nborisov@suse.com>
To: Omar Sandoval <osandov@osandov.com>, linux-btrfs@vger.kernel.org
Cc: kernel-team@fb.com, Christoph Hellwig <hch@lst.de>
Subject: Re: [PATCH 12/15] btrfs: get rid of one layer of bios in direct I/O
Date: Tue, 17 Mar 2020 18:48:36 +0200 [thread overview]
Message-ID: <a34e17d4-45d2-a745-cc06-f76d529cec54@suse.com> (raw)
In-Reply-To: <f9d0f9e8b8d11ff103654387f4370f50c6c074ae.1583789410.git.osandov@fb.com>
On 9.03.20 г. 23:32 ч., Omar Sandoval wrote:
> From: Omar Sandoval <osandov@fb.com>
>
> In the worst case, there are _4_ layers of bios in the Btrfs direct I/O
> path:
>
> 1. The bio created by the generic direct I/O code (dio_bio).
> 2. A clone of dio_bio we create in btrfs_submit_direct() to represent
> the entire direct I/O range (orig_bio).
> 3. A partial clone of orig_bio limited to the size of a RAID stripe that
> we create in btrfs_submit_direct_hook().
> 4. Clones of each of those split bios for each RAID stripe that we
> create in btrfs_map_bio().
>
> As of the previous commit, the second layer (orig_bio) is no longer
> needed for anything: we can split dio_bio instead, and complete dio_bio
> directly when all of the cloned bios complete. This lets us clean up a
> bunch of cruft, including dip->subio_endio and dip->errors (we can use
> dio_bio->bi_status instead). It also enables the next big cleanup of
> direct I/O read repair.
>
> Signed-off-by: Omar Sandoval <osandov@fb.com>
> ---
> fs/btrfs/inode.c | 213 +++++++++++++++--------------------------------
> 1 file changed, 66 insertions(+), 147 deletions(-)
>
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index 4a2e44f3e66e..40c1562704e9 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -54,11 +54,8 @@ struct btrfs_iget_args {
> struct btrfs_root *root;
> };
>
<snip>
> @@ -7400,6 +7384,29 @@ static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
> return ret;
> }
>
> +static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
> +{
> + /*
> + * This implies a barrier so that stores to dio_bio->bi_status before
> + * this and the following load are fully ordered.
> + */
It's not obvious which load this refers to. It's not obvious where this
ordering matters i.e what are the threads that care?
> + if (!refcount_dec_and_test(&dip->refs))
> + return;
> +
> + if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
> + __endio_write_update_ordered(dip->inode, dip->logical_offset,
> + dip->bytes,
> + !dip->dio_bio->bi_status);
> + } else {
> + unlock_extent(&BTRFS_I(dip->inode)->io_tree,
> + dip->logical_offset,
> + dip->logical_offset + dip->bytes - 1);
> + }
> +
> + dio_end_io(dip->dio_bio);
> + kfree(dip);
> +}
> +
> static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
> struct bio *bio,
> int mirror_num)
<snip>
> @@ -7920,98 +7876,77 @@ static void btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
> struct inode *inode = dip->inode;
> struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
> struct bio *bio;
> - struct bio *orig_bio = dip->orig_bio;
> - u64 start_sector = orig_bio->bi_iter.bi_sector;
> + struct bio *dio_bio = dip->dio_bio;
> + u64 start_sector = dio_bio->bi_iter.bi_sector;
> u64 file_offset = dip->logical_offset;
> int async_submit = 0;
> - u64 submit_len;
> + u64 submit_len = dio_bio->bi_iter.bi_size;
> int clone_offset = 0;
> int clone_len;
> int ret;
> blk_status_t status;
> struct btrfs_io_geometry geom;
>
> - submit_len = orig_bio->bi_iter.bi_size;
> - ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio),
> - start_sector << 9, submit_len, &geom);
> - if (ret)
> - goto out_err;
> -
> - if (geom.len >= submit_len) {
> - bio = orig_bio;
> - dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
> - goto submit;
> - }
> -
> /* async crcs make it difficult to collect full stripe writes. */
> if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
> async_submit = 0;
> else
> async_submit = 1;
>
> - /* bio split */
> ASSERT(geom.len <= INT_MAX);
geom.len now contains some random data since it's no longer initialised,
meaning this ASSERT hasn't triggered by pure luck. It should be
(re)moved inside the do {} while loop.
> do {
> + ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
> + start_sector << 9, submit_len,
> + &geom);
> + if (ret) {
> + status = errno_to_blk_status(ret);
> + goto out_err;
> + }
> +
> clone_len = min_t(int, submit_len, geom.len);
>
> /*
> * This will never fail as it's passing GPF_NOFS and
> * the allocation is backed by btrfs_bioset.
> */
> - bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
> - clone_len);
> + bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
> bio->bi_private = dip;
> bio->bi_end_io = btrfs_end_dio_bio;
> btrfs_io_bio(bio)->logical = file_offset;
>
> ASSERT(submit_len >= clone_len);
> submit_len -= clone_len;
> - if (submit_len == 0)
> - break;
>
> /*
> * Increase the count before we submit the bio so we know
> * the end IO handler won't happen before we increase the
> * count. Otherwise, the dip might get freed before we're
> * done setting it up.
> + *
> + * We transfer the initial reference to the last bio, so we
> + * don't need to increment the reference count for the last one.
> */
> - refcount_inc(&dip->refs);
> + if (submit_len > 0)
> + refcount_inc(&dip->refs);
>
> status = btrfs_submit_dio_bio(bio, inode, file_offset,
> async_submit);
> if (status) {
> bio_put(bio);
> - refcount_dec(&dip->refs);
> + if (submit_len > 0)
> + refcount_dec(&dip->refs);
> goto out_err;
> }
>
> clone_offset += clone_len;
> start_sector += clone_len >> 9;
> file_offset += clone_len;
> -
> - ret = btrfs_get_io_geometry(fs_info, btrfs_op(orig_bio),
> - start_sector << 9, submit_len, &geom);
> - if (ret)
> - goto out_err;
> } while (submit_len > 0);> + return;
<snip>
next prev parent reply other threads:[~2020-03-17 16:48 UTC|newest]
Thread overview: 69+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-03-09 21:32 [PATCH 00/15] btrfs: read repair/direct I/O improvements Omar Sandoval
2020-03-09 21:32 ` [PATCH 01/15] btrfs: fix error handling when submitting direct I/O bio Omar Sandoval
2020-03-11 17:54 ` Josef Bacik
2020-03-17 13:46 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 02/15] btrfs: fix double __endio_write_update_ordered in direct I/O Omar Sandoval
2020-03-10 16:30 ` Christoph Hellwig
2020-03-11 9:03 ` Omar Sandoval
2020-03-17 14:04 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 03/15] btrfs: look at full bi_io_vec for repair decision Omar Sandoval
2020-03-10 16:33 ` Christoph Hellwig
2020-03-11 9:07 ` Omar Sandoval
2020-03-16 10:48 ` Christoph Hellwig
2020-03-17 14:38 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 04/15] btrfs: don't do repair validation for checksum errors Omar Sandoval
2020-03-11 17:55 ` Josef Bacik
2020-03-09 21:32 ` [PATCH 05/15] btrfs: clarify btrfs_lookup_bio_sums documentation Omar Sandoval
2020-03-11 17:56 ` Josef Bacik
2020-03-11 18:23 ` Omar Sandoval
2020-03-11 18:34 ` Josef Bacik
2020-03-17 14:38 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 06/15] btrfs: rename __readpage_endio_check to check_data_csum Omar Sandoval
2020-03-10 14:46 ` Johannes Thumshirn
2020-03-11 17:57 ` Josef Bacik
2020-03-17 14:39 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 07/15] btrfs: make btrfs_check_repairable() static Omar Sandoval
2020-03-10 14:53 ` Johannes Thumshirn
2020-03-11 17:58 ` Josef Bacik
2020-03-17 14:52 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 08/15] btrfs: move btrfs_dio_private to inode.c Omar Sandoval
2020-03-10 14:56 ` Johannes Thumshirn
2020-03-11 8:48 ` Omar Sandoval
2020-03-17 14:53 ` Nikolay Borisov
2020-03-19 16:16 ` David Sterba
2020-03-09 21:32 ` [PATCH 09/15] btrfs: kill btrfs_dio_private->private Omar Sandoval
2020-03-11 17:59 ` Josef Bacik
2020-03-17 14:54 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 10/15] btrfs: convert btrfs_dio_private->pending_bios to refcount_t Omar Sandoval
2020-03-11 18:00 ` Josef Bacik
2020-03-17 15:10 ` Nikolay Borisov
2020-03-09 21:32 ` [PATCH 11/15] btrfs: put direct I/O checksums in btrfs_dio_private instead of bio Omar Sandoval
2020-03-11 18:04 ` Josef Bacik
2020-03-17 16:37 ` Nikolay Borisov
2020-04-03 16:18 ` David Sterba
2020-03-09 21:32 ` [PATCH 12/15] btrfs: get rid of one layer of bios in direct I/O Omar Sandoval
2020-03-10 16:38 ` Christoph Hellwig
2020-03-11 9:19 ` Omar Sandoval
2020-03-16 10:49 ` Christoph Hellwig
2020-03-11 18:07 ` Josef Bacik
2020-03-17 16:48 ` Nikolay Borisov [this message]
2020-03-09 21:32 ` [PATCH 13/15] btrfs: simplify direct I/O read repair Omar Sandoval
2020-03-11 18:16 ` Josef Bacik
2020-04-03 16:40 ` David Sterba
2020-04-03 18:05 ` Omar Sandoval
2020-04-16 10:08 ` David Sterba
2020-03-09 21:32 ` [PATCH 14/15] btrfs: get rid of endio_repair_workers Omar Sandoval
2020-03-11 18:16 ` Josef Bacik
2020-03-09 21:32 ` [PATCH 15/15] btrfs: unify buffered and direct I/O read repair Omar Sandoval
2020-03-11 18:19 ` Josef Bacik
2020-03-19 8:53 ` Nikolay Borisov
2020-03-19 9:03 ` Christoph Hellwig
2020-03-20 21:28 ` Omar Sandoval
2020-03-10 16:39 ` [PATCH 00/15] btrfs: read repair/direct I/O improvements Christoph Hellwig
2020-03-11 9:22 ` Omar Sandoval
2020-03-18 16:33 ` Goldwyn Rodrigues
2020-03-19 14:08 ` David Sterba
2020-03-18 22:07 ` David Sterba
2020-03-20 21:29 ` Omar Sandoval
2020-03-20 14:10 ` Christoph Hellwig
2020-03-20 14:11 ` Christoph Hellwig
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a34e17d4-45d2-a745-cc06-f76d529cec54@suse.com \
--to=nborisov@suse.com \
--cc=hch@lst.de \
--cc=kernel-team@fb.com \
--cc=linux-btrfs@vger.kernel.org \
--cc=osandov@osandov.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).