From: Chao Yu <chao@kernel.org>
To: Yunlei He <heyunlei@huawei.com>,
jaegeuk@kernel.org, yuchao0@huawei.com,
linux-f2fs-devel@lists.sourceforge.net
Cc: zhangdianfang@huawei.com
Subject: Re: [PATCH v2] f2fs: introduce a method to make nat journal more fresh
Date: Fri, 9 Mar 2018 22:30:01 +0800 [thread overview]
Message-ID: <9d48f1e3-dcb9-fce0-7306-c04c3e9865c1@kernel.org> (raw)
In-Reply-To: <1520589197-30137-1-git-send-email-heyunlei@huawei.com>
On 2018/3/9 17:53, Yunlei He wrote:
> This patch introduce a method to make nat journal more fresh:
> i. sort set list using dirty entry number and cp version
> average value.
> ii. if meet with cache hit, update average version valus with
> current cp version.
>
> With this patch, newly modified nat set will flush to journal,
> and flush old nat set with same dirty entry number to nat area.
>
> Signed-off-by: Yunlei He <heyunlei@huawei.com>
> ---
> fs/f2fs/node.c | 40 ++++++++++++++++++++++++++--------------
> fs/f2fs/node.h | 2 ++
> 2 files changed, 28 insertions(+), 14 deletions(-)
>
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index 177c438..844bdda 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -193,8 +193,8 @@ static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
> __free_nat_entry(e);
> }
>
> -static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
> - struct nat_entry *ne)
> +static void __set_nat_cache_dirty(struct f2fs_sb_info *sbi, bool remove_journal,
> + struct f2fs_nm_info *nm_i, struct nat_entry *ne)
> {
> nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
> struct nat_entry_set *head;
> @@ -207,9 +207,17 @@ static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
> INIT_LIST_HEAD(&head->set_list);
> head->set = set;
> head->entry_cnt = 0;
> + head->to_journal = false;
> + head->cp_ver = cur_cp_version(F2FS_CKPT(sbi));
> f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
> }
>
> + /* journal hit case, try to locate set in journal */
> + if (!remove_journal && head->to_journal)
> + head->cp_ver = div_u64(head->cp_ver, head->entry_cnt + 1) *
> + head->entry_cnt + div_u64(cur_cp_version(F2FS_CKPT(sbi)),
If all entries were flushed into journal in last checkpoint, head->entry_cnt
will always be zero, so as calculating, head->cp_ver will be cur_cp_version().
> + head->entry_cnt + 1);
How about
head->cp_ver = (NAT_ENTRY_PER_BLOCK - 1) *
(head->cp_ver / NAT_ENTRY_PER_BLOCK) +
1 * (cur_cp_version / NAT_ENTRY_PER_BLOCK);
> +
> if (get_nat_flag(ne, IS_DIRTY))
> goto refresh_list;
>
> @@ -357,7 +365,7 @@ static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
> nat_set_blkaddr(e, new_blkaddr);
> if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
> set_nat_flag(e, IS_CHECKPOINTED, false);
> - __set_nat_cache_dirty(nm_i, e);
> + __set_nat_cache_dirty(sbi, false, nm_i, e);
>
> /* update fsync_mark if its inode nat entry is still alive */
> if (ni->nid != ni->ino)
> @@ -2395,14 +2403,14 @@ static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
> spin_unlock(&nm_i->nid_list_lock);
> }
>
> - __set_nat_cache_dirty(nm_i, ne);
> + __set_nat_cache_dirty(sbi, true, nm_i, ne);
> }
> update_nats_in_cursum(journal, -i);
> up_write(&curseg->journal_rwsem);
> }
>
> -static void __adjust_nat_entry_set(struct nat_entry_set *nes,
> - struct list_head *head, int max)
> +static void __adjust_nat_entry_set(struct f2fs_sb_info *sbi,
> + struct nat_entry_set *nes, struct list_head *head, int max)
> {
> struct nat_entry_set *cur;
>
> @@ -2410,7 +2418,9 @@ static void __adjust_nat_entry_set(struct nat_entry_set *nes,
> goto add_out;
>
> list_for_each_entry(cur, head, set_list) {
> - if (cur->entry_cnt >= nes->entry_cnt) {
> + if (cur->entry_cnt > nes->entry_cnt ||
> + (cur->entry_cnt == nes->entry_cnt &&
> + cur->cp_ver < nes->cp_ver)) {
> list_add(&nes->set_list, cur->set_list.prev);
> return;
> }
> @@ -2458,7 +2468,6 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
> struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
> struct f2fs_journal *journal = curseg->journal;
> nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
> - bool to_journal = true;
> struct f2fs_nat_block *nat_blk;
> struct nat_entry *ne, *cur;
> struct page *page = NULL;
> @@ -2468,11 +2477,14 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
> * #1, flush nat entries to journal in current hot data summary block.
> * #2, flush nat entries to nat page.
> */
> +
> + set->to_journal = true;
> +
> if (enabled_nat_bits(sbi, cpc) ||
> !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
> - to_journal = false;
> + set->to_journal = false;
>
> - if (to_journal) {
> + if (set->to_journal) {
> down_write(&curseg->journal_rwsem);
> } else {
> page = get_next_nat_page(sbi, start_nid);
> @@ -2488,7 +2500,7 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
>
> f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
>
> - if (to_journal) {
> + if (set->to_journal) {
> offset = lookup_journal_in_cursum(journal,
> NAT_JOURNAL, nid, 1);
> f2fs_bug_on(sbi, offset < 0);
> @@ -2509,7 +2521,7 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
> }
> }
>
> - if (to_journal) {
> + if (set->to_journal) {
> up_write(&curseg->journal_rwsem);
> } else {
> __update_nat_bits(sbi, start_nid, page);
> @@ -2517,7 +2529,7 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
> }
>
> /* Allow dirty nats by node block allocation in write_begin */
> - if (!set->entry_cnt) {
> + if (!set->to_journal || cpc->reason & CP_UMOUNT) {
if (!set->to_journal && !set->entry_cnt) ?
For CP_UMOUNT case, we will call destroy_node_manager to release nat set cache.
> radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
> kmem_cache_free(nat_entry_set_slab, set);
> }
> @@ -2556,7 +2568,7 @@ void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
> unsigned idx;
> set_idx = setvec[found - 1]->set + 1;
> for (idx = 0; idx < found; idx++)
> - __adjust_nat_entry_set(setvec[idx], &sets,
> + __adjust_nat_entry_set(sbi, setvec[idx], &sets,
> MAX_NAT_JENTRIES(journal));
> }
>
> diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
> index 081ef0d..76afb24 100644
> --- a/fs/f2fs/node.h
> +++ b/fs/f2fs/node.h
> @@ -149,6 +149,8 @@ struct nat_entry_set {
> struct list_head entry_list; /* link with dirty nat entries */
> nid_t set; /* set number*/
> unsigned int entry_cnt; /* the # of nat entries in set */
> + __u64 cp_ver; /* cp version of this set modify */
> + bool to_journal; /* set flush to journal */
How about restructuring as below to reduce memory cost:
nid_t set;
unsigned int entry_cnt:9
unsigned int to_journal:1
__u64 cp_ver;
Thanks,
> };
>
> struct free_nid {
>
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
prev parent reply other threads:[~2018-03-09 14:30 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-03-09 9:53 [PATCH v2] f2fs: introduce a method to make nat journal more fresh Yunlei He
2018-03-09 14:30 ` Chao Yu [this message]
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=9d48f1e3-dcb9-fce0-7306-c04c3e9865c1@kernel.org \
--to=chao@kernel.org \
--cc=heyunlei@huawei.com \
--cc=jaegeuk@kernel.org \
--cc=linux-f2fs-devel@lists.sourceforge.net \
--cc=yuchao0@huawei.com \
--cc=zhangdianfang@huawei.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).