This patch including two changes:
1. We introduce nat journal to cache nat blocks which has less
dirty nat entries, nat entry sets are ranked by dirty nat entry
count if the count less than NAT_JOURNAL_ENTRIES. This method
will break the continuity of nat blocks write back order, and
this patch make a change of that prior to flush continuous nat
blocks.
2. Add read ahead nat blocks in case of a lot of serial synchronized
read as below:
336.785911: f2fs_write_checkpoint: dev = (259,44), checkpoint for Sync, state =
start flush nat
336.785980: flush_nat_entries: flush sum = 278 dirty nat pages
336.786002: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x129f,
oldaddr = 0x129f, newaddr = 0x129f rw = READ_SYNC(MP), type = META
336.791681: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x148e,
oldaddr = 0x148e, newaddr = 0x148e rw = READ_SYNC(MP), type = META
336.797802: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x1482,
oldaddr = 0x1482, newaddr = 0x1482 rw = READ_SYNC(MP), type = META
336.799155: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x1481,
oldaddr = 0x1481, newaddr = 0x1481 rw = READ_SYNC(MP), type = META
336.800603: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x147f,
oldaddr = 0x147f, newaddr = 0x147f rw = READ_SYNC(MP), type = META
336.802102: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x147b,
oldaddr = 0x147b, newaddr = 0x147b rw = READ_SYNC(MP), type = META
336.803528: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x1474,
oldaddr = 0x1474, newaddr = 0x1474 rw = READ_SYNC(MP), type = META
336.805026: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x1455,
oldaddr = 0x1455, newaddr = 0x1455 rw = READ_SYNC(MP), type = META
336.806537: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0xc1e,
oldaddr = 0xc1e, newaddr = 0xc1e rw = READ_SYNC(MP), type = META
336.807864: f2fs_submit_page_bio: dev = (259,44), ino = 2, page_index = 0x1307,
oldaddr = 0x1307, newaddr = 0x1307 rw = READ_SYNC(MP), type = META
336.833539: f2fs_write_checkpoint: dev = (259,44), checkpoint for Sync, state =
end flush nat
Signed-off-by: Yunlei He <heyun...@huawei.com>
Signed-off-by: Wei Fang <fangw...@huawei.com>
---
fs/f2fs/node.c | 60 +++++++++++++++++++++++++++++++++++++++++++++++++++++-----
fs/f2fs/node.h | 1 +
2 files changed, 56 insertions(+), 5 deletions(-)
diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index 833b46b..3b29fcb 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -209,6 +209,7 @@ 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->continue_io = false;
f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
}
@@ -2403,16 +2404,50 @@ static void remove_nats_in_journal(struct f2fs_sb_info
*sbi)
up_write(&curseg->journal_rwsem);
}
-static void __adjust_nat_entry_set(struct nat_entry_set *nes,
- struct list_head *head, int max)
+static bool nat_block_is_continuous(struct f2fs_nm_info *nm_i,
+ struct nat_entry_set *nes)
+{
+ unsigned int nat_index = nes->set;
+ unsigned char bit_cur =
+ f2fs_test_bit(nat_index, nm_i->nat_bitmap) ? 1 : 0;
+
+ if (nat_index && radix_tree_lookup(&nm_i->nat_set_root, nat_index - 1))
{
+ unsigned char bit_pre =
+ f2fs_test_bit(nat_index - 1, nm_i->nat_bitmap) ? 1 : 0;
+
+ if (bit_pre == bit_cur) {
+ nes->continue_io = true;
+ return true;
+ }
+ }
+
+ if (nat_index < nm_i->nat_blocks &&
+ radix_tree_lookup(&nm_i->nat_set_root, nat_index + 1)) {
+ unsigned char bit_next =
+ f2fs_test_bit(nat_index + 1, nm_i->nat_bitmap) ? 1 : 0;
+
+ if (bit_next == bit_cur) {
+ nes->continue_io = true;
+ return true;
+ }
+ }
+
+ return false;
+}
+
+static void __adjust_nat_entry_set(struct f2fs_nm_info *nm_i,
+ struct nat_entry_set *nes, struct list_head *head, int
max)
{
struct nat_entry_set *cur;
if (nes->entry_cnt >= max)
goto add_out;
+ if (nat_block_is_continuous(nm_i, 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->continue_io) {
list_add(&nes->set_list, cur->set_list.prev);
return;
}
@@ -2471,7 +2506,8 @@ static void __flush_nat_entry_set(struct f2fs_sb_info
*sbi,
* #2, flush nat entries to nat page.
*/
if (enabled_nat_bits(sbi, cpc) ||
- !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
+ !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL) ||
+ set->continue_io)
to_journal = false;
if (to_journal) {
@@ -2536,6 +2572,7 @@ void flush_nat_entries(struct f2fs_sb_info *sbi, struct
cp_control *cpc)
struct nat_entry_set *setvec[SETVEC_SIZE];
struct nat_entry_set *set, *tmp;
unsigned int found;
+ unsigned int nats_to_journal = 0;
nid_t set_idx = 0;
LIST_HEAD(sets);
@@ -2558,10 +2595,23 @@ 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(nm_i, setvec[idx], &sets,
MAX_NAT_JENTRIES(journal));
}
+ list_for_each_entry_safe(set, tmp, &sets, set_list) {
+ block_t blk_addr;
+
+ nats_to_journal += set->entry_cnt;
+
+ if (nats_to_journal > NAT_JOURNAL_ENTRIES ||
+ set->continue_io) {
+ blk_addr = current_nat_addr(sbi,
+ set->set * NAT_ENTRY_PER_BLOCK);
+ ra_meta_pages(sbi, blk_addr, 1, META_NAT, true);
+ }
+ }
+
/* flush dirty nats in nat entry set */
list_for_each_entry_safe(set, tmp, &sets, set_list)
__flush_nat_entry_set(sbi, set, cpc);
diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
index 081ef0d..91414de 100644
--- a/fs/f2fs/node.h
+++ b/fs/f2fs/node.h
@@ -149,6 +149,7 @@ 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 */
+ bool continue_io;
};
struct free_nid {
--
1.9.1
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
