f2fs: keep migration IO order in LFS mode

For non-migration IO, we will keep order of data/node blocks' submitting
as allocation sequence by sorting IOs in per log io_list list, but for
migration IO, it could be out-of-order.

In LFS mode, we should keep all IOs including migration IO be ordered,
so that this patch fixes to add an additional lock to keep submitting
order.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 33c060d..24eab96 100644 (file)
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -1114,6 +1114,8 @@ struct f2fs_sb_info {
        struct f2fs_bio_info *write_io[NR_PAGE_TYPE];   /* for write bios */
        struct mutex wio_mutex[NR_PAGE_TYPE - 1][NR_TEMP_TYPE];
                                                /* bio ordering for NODE/DATA */
+       /* keep migration IO order for LFS mode */
+       struct rw_semaphore io_order_lock;
        mempool_t *write_io_dummy;              /* Dummy pages */
 
        /* for checkpoint */

diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index 9bb2ddb..ec9b157 100644 (file)
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -614,6 +614,7 @@ static void move_data_block(struct inode *inode, block_t bidx,
        struct page *page;
        block_t newaddr;
        int err;
+       bool lfs_mode = test_opt(fio.sbi, LFS);
 
        /* do not read out */
        page = f2fs_grab_cache_page(inode->i_mapping, bidx, false);
@@ -654,6 +655,9 @@ static void move_data_block(struct inode *inode, block_t bidx,
        fio.page = page;
        fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 
+       if (lfs_mode)
+               down_write(&fio.sbi->io_order_lock);
+
        allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
                                        &sum, CURSEG_COLD_DATA, NULL, false);
 
@@ -710,6 +714,8 @@ static void move_data_block(struct inode *inode, block_t bidx,
 put_page_out:
        f2fs_put_page(fio.encrypted_page, 1);
 recover_block:
+       if (lfs_mode)
+               up_write(&fio.sbi->io_order_lock);
        if (err)
                __f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
                                                                true, true);

diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index 8fabf3a..9f27e9b 100644 (file)
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -2750,7 +2750,10 @@ static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
 {
        int type = __get_segment_type(fio);
        int err;
+       bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
 
+       if (keep_order)
+               down_read(&fio->sbi->io_order_lock);
 reallocate:
        allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
                        &fio->new_blkaddr, sum, type, fio, true);
@@ -2763,6 +2766,8 @@ reallocate:
        } else if (!err) {
                update_device_state(fio);
        }
+       if (keep_order)
+               up_read(&fio->sbi->io_order_lock);
 }
 
 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,

diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 51dca39..12282b1 100644 (file)
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -2369,6 +2369,7 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
        for (i = 0; i < NR_PAGE_TYPE - 1; i++)
                for (j = HOT; j < NR_TEMP_TYPE; j++)
                        mutex_init(&sbi->wio_mutex[i][j]);
+       init_rwsem(&sbi->io_order_lock);
        spin_lock_init(&sbi->cp_lock);
 
        sbi->dirty_device = 0;