| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20230420075222epcms2p2a0a00e26b5c70d90950eed750822e5fc@epcms2p2>
Date: Thu, 20 Apr 2023 16:52:22 +0900
From: Daejun Park <daejun7.park@...sung.com>
To: Jaegeuk Kim <jaegeuk@...nel.org>,
Daejun Park <daejun7.park@...sung.com>
CC: "chao@...nel.org" <chao@...nel.org>,
"corbet@....net" <corbet@....net>,
"rostedt@...dmis.org" <rostedt@...dmis.org>,
"mhiramat@...nel.org" <mhiramat@...nel.org>,
"linux-f2fs-devel@...ts.sourceforge.net"
<linux-f2fs-devel@...ts.sourceforge.net>,
"linux-doc@...r.kernel.org" <linux-doc@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"linux-trace-kernel@...r.kernel.org"
<linux-trace-kernel@...r.kernel.org>,
Seokhwan Kim <sukka.kim@...sung.com>,
beomsu kim <beomsu7.kim@...sung.com>,
Yonggil Song <yonggil.song@...sung.com>
Subject: RE:(2) [PATCH] f2fs: add async reset zone command support
Hi Jeageuk,
Thanks for helpful review.
>
> On 04/14, Daejun Park wrote:
> > This patch enables submit reset zone command asynchornously. It helps
> > decrease average latency of write IOs in high utilization scenario by
> > faster checkpointing.
> >
> > Signed-off-by: Daejun Park <daejun7.park@...sung.com>
> > ---
> > Documentation/filesystems/f2fs.rst | 4 ++
> > fs/f2fs/f2fs.h | 1 +
> > fs/f2fs/segment.c | 92 +++++++++++++++++++++++++++++-
> > fs/f2fs/super.c | 8 +++
> > include/trace/events/f2fs.h | 18 +++++-
> > 5 files changed, 119 insertions(+), 4 deletions(-)
> >
> > diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
> > index 2055e72871fe..4cfabf831a79 100644
> > --- a/Documentation/filesystems/f2fs.rst
> > +++ b/Documentation/filesystems/f2fs.rst
> > @@ -342,6 +342,10 @@ discard_unit=%s Control discard unit, the argument can be "block", "segment"
> > default, it is helpful for large sized SMR or ZNS devices to
> > reduce memory cost by getting rid of fs metadata supports small
> > discard.
> > +async_reset_zone Enable the RESET WRITE POINTER command to be submitted asynchronously.
> > + With this option, the RESET WRITE POINTER command can be processed by
> > + the discard thread like a discard command. It can reduce checkpoint
> > + latency by asynchronously checking for completion of the RESET WRITE POINTER command.
>
> Do we need to set this by default instead of mount option?
I think it is good idea. I will make this feature by default.
>
> > memory=%s Control memory mode. This supports "normal" and "low" modes.
> > "low" mode is introduced to support low memory devices.
> > Because of the nature of low memory devices, in this mode, f2fs
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index 4e2596dacbf1..021e55c5d1a8 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -167,6 +167,7 @@ struct f2fs_mount_info {
> > * be aligned to this unit: block,
> > * segment or section
> > */
> > + bool async_zone_reset; /* async zone reset */
> > struct fscrypt_dummy_policy dummy_enc_policy; /* test dummy encryption */
> > block_t unusable_cap_perc; /* percentage for cap */
> > block_t unusable_cap; /* Amount of space allowed to be
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index 45128694eefa..60cfe97c9dbd 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -1189,6 +1189,46 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
> > static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
> > struct block_device *bdev, block_t lstart,
> > block_t start, block_t len);
> > +
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > +static int __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
> > + struct discard_cmd *dc, blk_opf_t flag,
> > + struct list_head *wait_list,
> > + unsigned int *issued)
> > +{
> > + struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> > + struct block_device *bdev = dc->bdev;
> > + struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
> > + unsigned long flags;
> > +
> > + trace_f2fs_issue_reset_zone(bdev, SECTOR_FROM_BLOCK(dc->di.start));
> > +
> > + spin_lock_irqsave(&dc->lock, flags);
> > + dc->state = D_SUBMIT;
> > + dc->bio_ref++;
> > + spin_unlock_irqrestore(&dc->lock, flags);
> > +
> > + if (issued)
> > + (*issued)++;
> > +
> > + atomic_inc(&dcc->queued_discard);
> > + dc->queued++;
> > + list_move_tail(&dc->list, wait_list);
> > +
> > + /* sanity check on discard range */
> > + __check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
> > +
> > + bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
> > + bio->bi_private = dc;
> > + bio->bi_end_io = f2fs_submit_discard_endio;
> > + submit_bio(bio);
> > +
> > + atomic_inc(&dcc->issued_discard);
> > +
> > + return 0;
> > +}
> > +#endif
> > +
> > /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
> > static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > struct discard_policy *dpolicy,
> > @@ -1210,6 +1250,11 @@ static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
> > if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
> > return 0;
> >
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > + if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
> > + return __submit_zone_reset_cmd(sbi, dc, flag, wait_list, issued);
> > +#endif
> > +
> > trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
> >
> > lstart = dc->di.lstart;
> > @@ -1454,21 +1499,42 @@ static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
> > }
> > }
> >
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > +static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
> > + struct block_device *bdev, block_t blkstart, block_t blklen)
> > +{
> > + block_t lblkstart = blkstart;
> > +
> > + if (f2fs_is_multi_device(sbi)) {
> > + int devi = f2fs_target_device_index(sbi, blkstart);
> > +
> > + blkstart -= FDEV(devi).start_blk;
> > + }
> > +
> > + trace_f2fs_queue_reset_zone(bdev, blkstart);
> > +
> > + mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
> > + __insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
> > + mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
> > +}
> > +#endif
> > +
> > static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
> > struct block_device *bdev, block_t blkstart, block_t blklen)
> > {
> > block_t lblkstart = blkstart;
> >
> > + trace_f2fs_queue_discard(bdev, blkstart, blklen);
> > +
> > if (!f2fs_bdev_support_discard(bdev))
> > return;
> >
> > - trace_f2fs_queue_discard(bdev, blkstart, blklen);
> > -
> > if (f2fs_is_multi_device(sbi)) {
> > int devi = f2fs_target_device_index(sbi, blkstart);
> >
> > blkstart -= FDEV(devi).start_blk;
> > }
> > +
> > mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
> > __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
> > mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
> > @@ -1719,7 +1785,22 @@ static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
> > dc = __lookup_discard_cmd(sbi, blkaddr);
> > if (dc) {
> > if (dc->state == D_PREP) {
> > +#ifdef CONFIG_BLK_DEV_ZONED
> > + if (f2fs_sb_has_blkzoned(sbi) &&
> > + bdev_is_zoned(dc->bdev)) {
> > + struct discard_policy dpolicy;
> > + /* force submit zone reset */
> > + __init_discard_policy(sbi, &dpolicy,
> > + DPOLICY_FORCE, 1);
> > + __submit_discard_cmd(sbi, &dpolicy, dc, NULL);
> > + dc->ref++;
> > + need_wait = true;
> > + } else {
> > + __punch_discard_cmd(sbi, dc, blkaddr);
>
> Can be consolidated in both cases below.
How do you think this modification as below.
if (dc) {
if (dc->state == D_PREP) {
if (f2fs_sb_has_blkzoned(sbi) &&
bdev_is_zoned(dc->bdev)) {
struct discard_policy dpolicy;
/* force submit zone reset */
__init_discard_policy(sbi, &dpolicy,
DPOLICY_FORCE, 1);
__submit_discard_cmd(sbi, &dpolicy, dc, NULL);
dc->ref++;
need_wait = true;
} else {
__punch_discard_cmd(sbi, dc, blkaddr);
}
} else {
dc->ref++;
need_wait = true;
}
}
>
> > + }
> > +#else
> > __punch_discard_cmd(sbi, dc, blkaddr);
> > +#endif
> > } else {
> > dc->ref++;
> > need_wait = true;
> > @@ -1869,6 +1950,13 @@ static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
> > blkstart, blklen);
> > return -EIO;
> > }
> > +
> > + if (likely(!is_sbi_flag_set(sbi, SBI_POR_DOING)) &&
> > + F2FS_OPTION(sbi).async_zone_reset) {
> > + __queue_zone_reset_cmd(sbi, bdev, lblkstart, blklen);
> > + return 0;
> > + }
>
> Can make a sigle case above and below, if we set this by default.
I can remove "F2FS_OPTION(sbi).async_zone_reset", but I think the checking
"SBI_POR_DOING" statement can not be removed. Because check_zone_write_pointer()
and fix_curseg_write_pointer() use this __f2fs_issue_discard_zone() for sending
reset write pointer command synchronously.
> > +
> > trace_f2fs_issue_reset_zone(bdev, blkstart);
> > return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
> > sector, nr_sects, GFP_NOFS);
> > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > index 7d0202f7b317..48198112cbbc 100644
> > --- a/fs/f2fs/super.c
> > +++ b/fs/f2fs/super.c
> > @@ -162,6 +162,7 @@ enum {
> > Opt_gc_merge,
> > Opt_nogc_merge,
> > Opt_discard_unit,
> > + Opt_async_zone_reset,
> > Opt_memory_mode,
> > Opt_age_extent_cache,
> > Opt_err,
> > @@ -241,6 +242,7 @@ static match_table_t f2fs_tokens = {
> > {Opt_gc_merge, "gc_merge"},
> > {Opt_nogc_merge, "nogc_merge"},
> > {Opt_discard_unit, "discard_unit=%s"},
> > + {Opt_async_zone_reset, "async_zone_reset"},
> > {Opt_memory_mode, "memory=%s"},
> > {Opt_age_extent_cache, "age_extent_cache"},
> > {Opt_err, NULL},
> > @@ -1249,6 +1251,9 @@ static int parse_options(struct super_block *sb, char *options, bool is_remount)
> > }
> > kfree(name);
> > break;
> > + case Opt_async_zone_reset:
> > + F2FS_OPTION(sbi).async_zone_reset = true;
> > + break;
> > case Opt_memory_mode:
> > name = match_strdup(&args[0]);
> > if (!name)
> > @@ -2047,6 +2052,9 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
> > if (test_opt(sbi, ATGC))
> > seq_puts(seq, ",atgc");
> >
> > + if (F2FS_OPTION(sbi).async_zone_reset)
> > + seq_puts(seq, ",async_zone_reset");
> > +
> > if (F2FS_OPTION(sbi).memory_mode == MEMORY_MODE_NORMAL)
> > seq_printf(seq, ",memory=%s", "normal");
> > else if (F2FS_OPTION(sbi).memory_mode == MEMORY_MODE_LOW)
> > diff --git a/include/trace/events/f2fs.h b/include/trace/events/f2fs.h
> > index 99cbc5949e3c..ee1477de8324 100644
> > --- a/include/trace/events/f2fs.h
> > +++ b/include/trace/events/f2fs.h
> > @@ -1512,7 +1512,7 @@ DEFINE_EVENT(f2fs_discard, f2fs_remove_discard,
> > TP_ARGS(dev, blkstart, blklen)
> > );
> >
> > -TRACE_EVENT(f2fs_issue_reset_zone,
> > +DECLARE_EVENT_CLASS(f2fs_reset_zone,
> >
> > TP_PROTO(struct block_device *dev, block_t blkstart),
> >
> > @@ -1528,11 +1528,25 @@ TRACE_EVENT(f2fs_issue_reset_zone,
> > __entry->blkstart = blkstart;
> > ),
> >
> > - TP_printk("dev = (%d,%d), reset zone at block = 0x%llx",
> > + TP_printk("dev = (%d,%d), zone at block = 0x%llx",
> > show_dev(__entry->dev),
> > (unsigned long long)__entry->blkstart)
> > );
> >
> > +DEFINE_EVENT(f2fs_reset_zone, f2fs_queue_reset_zone,
> > +
> > + TP_PROTO(struct block_device *dev, block_t blkstart),
> > +
> > + TP_ARGS(dev, blkstart)
> > +);
> > +
> > +DEFINE_EVENT(f2fs_reset_zone, f2fs_issue_reset_zone,
> > +
> > + TP_PROTO(struct block_device *dev, block_t blkstart),
> > +
> > + TP_ARGS(dev, blkstart)
> > +);
> > +
> > TRACE_EVENT(f2fs_issue_flush,
> >
> > TP_PROTO(struct block_device *dev, unsigned int nobarrier,
> > --
> > 2.25.1
>
Thanks,
Daejun
Powered by blists - more mailing lists