On 2021/6/18 下午12:16, Anand Jain wrote:

On 17/06/2021 13:14, Qu Wenruo wrote:

[...]

quoted

+            atomic_inc(&cb->pending_bios);
              ret = btrfs_bio_wq_end_io(fs_info, comp_bio,
                            BTRFS_WQ_ENDIO_DATA);
              BUG_ON(ret); /* -ENOMEM */
-            /*
-             * inc the count before we submit the bio so
-             * we know the end IO handler won't happen before
-             * we inc the count.  Otherwise, the cb might get
-             * freed before we're done setting it up
-             */
-            refcount_inc(&cb->pending_bios);
-

Looks good so far.
I understand pending_bios will go away in favour of pending_sectors.
But here, is there any purpose  atomic_inc(&cb->pending_bios) is called 
before btrfs_bio_wq_end_io() that might fail.

The pending_bios is increased before we may want to submit the bio.

There is no special reason, but we will later merge refcount_inc();
btrfs_bio_wq_end_io(), btrfs_map_bio() into one call, thus I just choose
btrfs_bio_wq_end_io() as a marker.

Thanks,
Qu

-Anand

quoted

              ret = btrfs_lookup_bio_sums(inode, comp_bio, sums);
              BUG_ON(ret); /* -ENOMEM */

@@ -805,6 +816,7 @@ blk_status_t btrfs_submit_compressed_read(struct 

inode *inode, struct bio *bio,
          cur_disk_byte += pg_len;
      }
+    atomic_inc(&cb->pending_bios);
      ret = btrfs_bio_wq_end_io(fs_info, comp_bio, BTRFS_WQ_ENDIO_DATA);
      BUG_ON(ret); /* -ENOMEM */

diff --git a/fs/btrfs/compression.h b/fs/btrfs/compression.h
index c359f20920d0..8940e9e9fed3 100644
--- a/fs/btrfs/compression.h
+++ b/fs/btrfs/compression.h

@@ -29,7 +29,10 @@ struct btrfs_inode;

  struct compressed_bio {
      /* number of bios pending for this compressed extent */
-    refcount_t pending_bios;
+    atomic_t pending_bios;
+
+    /* Number of sectors with unfinished IO (unsubmitted or 
unfinished) */
+    refcount_t pending_sectors;
      /* Number of compressed pages in the array */
      unsigned int nr_pages;