f2fs: remove the own bi_private allocation
authorJaegeuk Kim <jaegeuk.kim@samsung.com>
Sat, 30 Nov 2013 01:10:31 +0000 (10:10 +0900)
committerJaegeuk Kim <jaegeuk.kim@samsung.com>
Mon, 23 Dec 2013 01:18:05 +0000 (10:18 +0900)
Previously f2fs allocates its own bi_private data structure all the time even
though we don't use it. But, can we remove this bi_private allocation?

This patch removes such the additional bi_private allocation.

1. Retrieve f2fs_sb_info from its page->mapping->host->i_sb.
 - This removes the usecases of bi_private in end_io.

2. Use bi_private only when we really need it.
 - The bi_private is used only when the checkpoint procedure is conducted.
 - When conducting the checkpoint, f2fs submits a META_FLUSH bio to wait its bio
completion.
 - Since we have no dependancies to remove bi_private now, let's just use
 bi_private pointer as the completion pointer.

Reviewed-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
fs/f2fs/segment.c
fs/f2fs/segment.h

index 03878634a0fe6a544ad2327573ca0c7b3637c06d..0db40271f0d862f90cb831454898ba185bd688ee 100644 (file)
@@ -791,7 +791,7 @@ static void f2fs_end_io_write(struct bio *bio, int err)
 {
        const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
        struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
-       struct bio_private *p = bio->bi_private;
+       struct f2fs_sb_info *sbi = F2FS_SB(bvec->bv_page->mapping->host->i_sb);
 
        do {
                struct page *page = bvec->bv_page;
@@ -802,21 +802,21 @@ static void f2fs_end_io_write(struct bio *bio, int err)
                        SetPageError(page);
                        if (page->mapping)
                                set_bit(AS_EIO, &page->mapping->flags);
-                       set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
-                       p->sbi->sb->s_flags |= MS_RDONLY;
+
+                       set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
+                       sbi->sb->s_flags |= MS_RDONLY;
                }
                end_page_writeback(page);
-               dec_page_count(p->sbi, F2FS_WRITEBACK);
+               dec_page_count(sbi, F2FS_WRITEBACK);
        } while (bvec >= bio->bi_io_vec);
 
-       if (p->is_sync)
-               complete(p->wait);
+       if (bio->bi_private)
+               complete(bio->bi_private);
 
-       if (!get_pages(p->sbi, F2FS_WRITEBACK) &&
-                       !list_empty(&p->sbi->cp_wait.task_list))
-               wake_up(&p->sbi->cp_wait);
+       if (!get_pages(sbi, F2FS_WRITEBACK) &&
+                       !list_empty(&sbi->cp_wait.task_list))
+               wake_up(&sbi->cp_wait);
 
-       kfree(p);
        bio_put(bio);
 }
 
@@ -838,7 +838,6 @@ static void do_submit_bio(struct f2fs_sb_info *sbi,
        int rw = sync ? WRITE_SYNC : WRITE;
        enum page_type btype = PAGE_TYPE_OF_BIO(type);
        struct f2fs_bio_info *io = &sbi->write_io[btype];
-       struct bio_private *p;
 
        if (!io->bio)
                return;
@@ -851,18 +850,16 @@ static void do_submit_bio(struct f2fs_sb_info *sbi,
 
        trace_f2fs_submit_write_bio(sbi->sb, rw, btype, io->bio);
 
-       p = io->bio->bi_private;
-       p->sbi = sbi;
-       io->bio->bi_end_io = f2fs_end_io_write;
-
+       /*
+        * META_FLUSH is only from the checkpoint procedure, and we should wait
+        * this metadata bio for FS consistency.
+        */
        if (type == META_FLUSH) {
                DECLARE_COMPLETION_ONSTACK(wait);
-               p->is_sync = true;
-               p->wait = &wait;
+               io->bio->bi_private = &wait;
                submit_bio(rw, io->bio);
                wait_for_completion(&wait);
        } else {
-               p->is_sync = false;
                submit_bio(rw, io->bio);
        }
        io->bio = NULL;
@@ -897,18 +894,10 @@ static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
                do_submit_bio(sbi, type, false);
 alloc_new:
        if (io->bio == NULL) {
-               struct bio_private *priv;
-retry:
-               priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
-               if (!priv) {
-                       cond_resched();
-                       goto retry;
-               }
-
                bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
                io->bio = f2fs_bio_alloc(bdev, bio_blocks);
                io->bio->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
-               io->bio->bi_private = priv;
+               io->bio->bi_end_io = f2fs_end_io_write;
                /*
                 * The end_io will be assigned at the sumbission phase.
                 * Until then, let bio_add_page() merge consecutive IOs as much
index 7fea2ee8a5da0a1ae8fa4f3c968d9eef10b2f4a7..26812fc0fa12b7cf458fd32bb5bbbdd65777f8b2 100644 (file)
 #define MAX_BIO_BLOCKS(max_hw_blocks)                                  \
        (min((int)max_hw_blocks, BIO_MAX_PAGES))
 
-/* during checkpoint, bio_private is used to synchronize the last bio */
-struct bio_private {
-       struct f2fs_sb_info *sbi;
-       bool is_sync;
-       struct completion *wait;
-};
-
 /*
  * indicate a block allocation direction: RIGHT and LEFT.
  * RIGHT means allocating new sections towards the end of volume.