There is some concern that these iput()'s could be the final iputs and could induce lockups on people waiting on writeback. This would happen in the rare case that we don't create ordered extents because of an error, but it is theoretically possible and we already have a mechanism to deal with this so just make them delayed iputs to negate any worry.
Unified · Split
| @@ -987,7 +987,7 @@ static noinline void async_cow_start(struct btrfs_work *work) · 986 unmodified lines | ||
| 987 | 987 | async_cow->start, async_cow->end, async_cow, |
| 988 | 988 | &num_added); |
| 989 | 989 | if (num_added == 0) { |
| 990 | − iput(async_cow->inode); | |
| 990 | + btrfs_add_delayed_iput(async_cow->inode); | |
| 991 | 991 | async_cow->inode = NULL; |
| 992 | 992 | } |
| 993 | 993 | } |
| @@ -1023,7 +1023,7 @@ static noinline void async_cow_free(struct btrfs_work *work) · 29 unmodified lines | ||
| 1023 | 1023 | struct async_cow *async_cow; |
| 1024 | 1024 | async_cow = container_of(work, struct async_cow, work); |
| 1025 | 1025 | if (async_cow->inode) |
| 1026 | − iput(async_cow->inode); | |
| 1026 | + btrfs_add_delayed_iput(async_cow->inode); | |
| 1027 | 1027 | kfree(async_cow); |
| 1028 | 1028 | } |
| 1029 | 1029 | |