[PATCH] f2fs: use crc and cp version to determine roll-forward recovery

From: Jaegeuk Kim
Date: Mon Sep 19 2016 - 20:55:10 EST

Next message: David Rientjes: "Re: [PATCH] mm/mempolicy.c: forbid static or relative flags for local NUMA mode"
Previous message: David Rientjes: "Re: [PATCH V4] mm: Add sysfs interface to dump each node's zonelist information"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Previously, we used cp_version only to detect recoverable dnodes.
In order to avoid same garbage cp_version, we needed to truncate the next
dnode during checkpoint, resulting in additional discard or data write.
If we can distinguish this by using crc in addition to cp_version, we can
remove this overhead.

There is backward compatibility concern where it changes node_footer layout.
But, it only affects the direct nodes written after the last checkpoint.
We simply expect that user would change kernel versions back and forth after
stable checkpoint.

Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
---
fs/f2fs/checkpoint.c | 18 -------------
fs/f2fs/f2fs.h | 1 -
fs/f2fs/node.h | 73 ++++++++++++++++++++++++++++++++--------------------
fs/f2fs/recovery.c | 36 +++++---------------------
fs/f2fs/segment.c | 22 ----------------
fs/f2fs/super.c | 5 +++-
6 files changed, 55 insertions(+), 100 deletions(-)

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index df56a43..6ecc5b8 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -992,7 +992,6 @@ static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
{
struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
- struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
struct f2fs_nm_info *nm_i = NM_I(sbi);
unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
nid_t last_nid = nm_i->next_scan_nid;
@@ -1001,19 +1000,10 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
__u32 crc32 = 0;
int i;
int cp_payload_blks = __cp_payload(sbi);
- block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
- bool invalidate = false;
struct super_block *sb = sbi->sb;
struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
u64 kbytes_written;

- /*
- * This avoids to conduct wrong roll-forward operations and uses
- * metapages, so should be called prior to sync_meta_pages below.
- */
- if (!test_opt(sbi, LFS) && discard_next_dnode(sbi, discard_blk))
- invalidate = true;
-
/* Flush all the NAT/SIT pages */
while (get_pages(sbi, F2FS_DIRTY_META)) {
sync_meta_pages(sbi, META, LONG_MAX);
@@ -1154,14 +1144,6 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
/* wait for previous submitted meta pages writeback */
wait_on_all_pages_writeback(sbi);

- /*
- * invalidate meta page which is used temporarily for zeroing out
- * block at the end of warm node chain.
- */
- if (invalidate)
- invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
- discard_blk);
-
release_ino_entry(sbi, false);

if (unlikely(f2fs_cp_error(sbi)))
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 132756c..a472191 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -2065,7 +2065,6 @@ void refresh_sit_entry(struct f2fs_sb_info *, block_t, block_t);
void f2fs_wait_all_discard_bio(struct f2fs_sb_info *);
void clear_prefree_segments(struct f2fs_sb_info *, struct cp_control *);
void release_discard_addrs(struct f2fs_sb_info *);
-bool discard_next_dnode(struct f2fs_sb_info *, block_t);
int npages_for_summary_flush(struct f2fs_sb_info *, bool);
void allocate_new_segments(struct f2fs_sb_info *);
int f2fs_trim_fs(struct f2fs_sb_info *, struct fstrim_range *);
diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
index fc76845..e8114f9 100644
--- a/fs/f2fs/node.h
+++ b/fs/f2fs/node.h
@@ -229,6 +229,37 @@ static inline void set_to_next_nat(struct f2fs_nm_info *nm_i, nid_t start_nid)
f2fs_change_bit(block_off, nm_i->nat_bitmap);
}

+static inline nid_t ino_of_node(struct page *node_page)
+{
+ struct f2fs_node *rn = F2FS_NODE(node_page);
+ return le32_to_cpu(rn->footer.ino);
+}
+
+static inline nid_t nid_of_node(struct page *node_page)
+{
+ struct f2fs_node *rn = F2FS_NODE(node_page);
+ return le32_to_cpu(rn->footer.nid);
+}
+
+static inline unsigned int ofs_of_node(struct page *node_page)
+{
+ struct f2fs_node *rn = F2FS_NODE(node_page);
+ unsigned flag = le32_to_cpu(rn->footer.flag);
+ return flag >> OFFSET_BIT_SHIFT;
+}
+
+static inline __u64 cpver_of_node(struct page *node_page)
+{
+ struct f2fs_node *rn = F2FS_NODE(node_page);
+ return le64_to_cpu(rn->footer.cp_ver);
+}
+
+static inline block_t next_blkaddr_of_node(struct page *node_page)
+{
+ struct f2fs_node *rn = F2FS_NODE(node_page);
+ return le32_to_cpu(rn->footer.next_blkaddr);
+}
+
static inline void fill_node_footer(struct page *page, nid_t nid,
nid_t ino, unsigned int ofs, bool reset)
{
@@ -259,40 +290,26 @@ static inline void fill_node_footer_blkaddr(struct page *page, block_t blkaddr)
{
struct f2fs_checkpoint *ckpt = F2FS_CKPT(F2FS_P_SB(page));
struct f2fs_node *rn = F2FS_NODE(page);
+ size_t crc_offset = le32_to_cpu(ckpt->checksum_offset);
+ __u64 cp_ver = le64_to_cpu(ckpt->checkpoint_ver);
+ __u64 crc;

- rn->footer.cp_ver = ckpt->checkpoint_ver;
+ crc = le32_to_cpu(*((__le32 *)((unsigned char *)ckpt + crc_offset)));
+ cp_ver |= (crc << 32);
+ rn->footer.cp_ver = cpu_to_le64(cp_ver);
rn->footer.next_blkaddr = cpu_to_le32(blkaddr);
}

-static inline nid_t ino_of_node(struct page *node_page)
-{
- struct f2fs_node *rn = F2FS_NODE(node_page);
- return le32_to_cpu(rn->footer.ino);
-}
-
-static inline nid_t nid_of_node(struct page *node_page)
-{
- struct f2fs_node *rn = F2FS_NODE(node_page);
- return le32_to_cpu(rn->footer.nid);
-}
-
-static inline unsigned int ofs_of_node(struct page *node_page)
-{
- struct f2fs_node *rn = F2FS_NODE(node_page);
- unsigned flag = le32_to_cpu(rn->footer.flag);
- return flag >> OFFSET_BIT_SHIFT;
-}
-
-static inline unsigned long long cpver_of_node(struct page *node_page)
+static inline bool is_recoverable_dnode(struct page *page)
{
- struct f2fs_node *rn = F2FS_NODE(node_page);
- return le64_to_cpu(rn->footer.cp_ver);
-}
+ struct f2fs_checkpoint *ckpt = F2FS_CKPT(F2FS_P_SB(page));
+ size_t crc_offset = le32_to_cpu(ckpt->checksum_offset);
+ __u64 cp_ver = cur_cp_version(ckpt);
+ __u64 crc;

-static inline block_t next_blkaddr_of_node(struct page *node_page)
-{
- struct f2fs_node *rn = F2FS_NODE(node_page);
- return le32_to_cpu(rn->footer.next_blkaddr);
+ crc = le32_to_cpu(*((__le32 *)((unsigned char *)ckpt + crc_offset)));
+ cp_ver |= (crc << 32);
+ return cpu_to_le64(cp_ver) == cpver_of_node(page);
}

/*
diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
index ad748e5..2b8a56d 100644
--- a/fs/f2fs/recovery.c
+++ b/fs/f2fs/recovery.c
@@ -224,7 +224,6 @@ static bool is_same_inode(struct inode *inode, struct page *ipage)

static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
{
- unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
struct curseg_info *curseg;
struct page *page = NULL;
block_t blkaddr;
@@ -242,7 +241,7 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)

page = get_tmp_page(sbi, blkaddr);

- if (cp_ver != cpver_of_node(page))
+ if (!is_recoverable_dnode(page))
break;

if (!is_fsync_dnode(page))
@@ -516,7 +515,6 @@ out:
static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
struct list_head *dir_list)
{
- unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
struct curseg_info *curseg;
struct page *page = NULL;
int err = 0;
@@ -536,7 +534,7 @@ static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,

page = get_tmp_page(sbi, blkaddr);

- if (cp_ver != cpver_of_node(page)) {
+ if (!is_recoverable_dnode(page)) {
f2fs_put_page(page, 1);
break;
}
@@ -628,37 +626,15 @@ out:
}

clear_sbi_flag(sbi, SBI_POR_DOING);
- if (err) {
- bool invalidate = false;
-
- if (test_opt(sbi, LFS)) {
- update_meta_page(sbi, NULL, blkaddr);
- invalidate = true;
- } else if (discard_next_dnode(sbi, blkaddr)) {
- invalidate = true;
- }
-
- f2fs_wait_all_discard_bio(sbi);
-
- /* Flush all the NAT/SIT pages */
- while (get_pages(sbi, F2FS_DIRTY_META))
- sync_meta_pages(sbi, META, LONG_MAX);
-
- /* invalidate temporary meta page */
- if (invalidate)
- invalidate_mapping_pages(META_MAPPING(sbi),
- blkaddr, blkaddr);
-
+ if (err)
set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
- mutex_unlock(&sbi->cp_mutex);
- } else if (need_writecp) {
+ mutex_unlock(&sbi->cp_mutex);
+
+ if (!err && need_writecp) {
struct cp_control cpc = {
.reason = CP_RECOVERY,
};
- mutex_unlock(&sbi->cp_mutex);
err = write_checkpoint(sbi, &cpc);
- } else {
- mutex_unlock(&sbi->cp_mutex);
}

destroy_fsync_dnodes(&dir_list);
diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index c4d0472..2d23d7b 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -669,28 +669,6 @@ static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
}

-bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
-{
- int err = -EOPNOTSUPP;
-
- if (test_opt(sbi, DISCARD)) {
- struct seg_entry *se = get_seg_entry(sbi,
- GET_SEGNO(sbi, blkaddr));
- unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
-
- if (f2fs_test_bit(offset, se->discard_map))
- return false;
-
- err = f2fs_issue_discard(sbi, blkaddr, 1);
- }
-
- if (err) {
- update_meta_page(sbi, NULL, blkaddr);
- return true;
- }
- return false;
-}
-
static void __add_discard_entry(struct f2fs_sb_info *sbi,
struct cp_control *cpc, struct seg_entry *se,
unsigned int start, unsigned int end)
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 29e3cf4..9649b79 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -1827,6 +1827,9 @@ try_onemore:
if (need_fsck)
set_sbi_flag(sbi, SBI_NEED_FSCK);

+ if (!retry)
+ goto skip_recovery;
+
err = recover_fsync_data(sbi, false);
if (err < 0) {
need_fsck = true;
@@ -1844,7 +1847,7 @@ try_onemore:
goto free_kobj;
}
}
-
+skip_recovery:
/* recover_fsync_data() cleared this already */
clear_sbi_flag(sbi, SBI_POR_DOING);

--
2.8.3

Next message: David Rientjes: "Re: [PATCH] mm/mempolicy.c: forbid static or relative flags for local NUMA mode"
Previous message: David Rientjes: "Re: [PATCH V4] mm: Add sysfs interface to dump each node's zonelist information"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]