On 4.01.20 г. 15:56 ч., Qu Wenruo wrote:
> [BUG]
> There are several different KASAN reports for balance + snapshot
> workloads.
> Involved call paths include:
>
> should_ignore_root+0x54/0xb0 [btrfs]
> build_backref_tree+0x11af/0x2280 [btrfs]
> relocate_tree_blocks+0x391/0xb80 [btrfs]
> relocate_block_group+0x3e5/0xa00 [btrfs]
> btrfs_relocate_block_group+0x240/0x4d0 [btrfs]
> btrfs_relocate_chunk+0x53/0xf0 [btrfs]
> btrfs_balance+0xc91/0x1840 [btrfs]
> btrfs_ioctl_balance+0x416/0x4e0 [btrfs]
> btrfs_ioctl+0x8af/0x3e60 [btrfs]
> do_vfs_ioctl+0x831/0xb10
> ksys_ioctl+0x67/0x90
> __x64_sys_ioctl+0x43/0x50
> do_syscall_64+0x79/0xe0
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> create_reloc_root+0x9f/0x460 [btrfs]
> btrfs_reloc_post_snapshot+0xff/0x6c0 [btrfs]
> create_pending_snapshot+0xa9b/0x15f0 [btrfs]
> create_pending_snapshots+0x111/0x140 [btrfs]
> btrfs_commit_transaction+0x7a6/0x1360 [btrfs]
> btrfs_mksubvol+0x915/0x960 [btrfs]
> btrfs_ioctl_snap_create_transid+0x1d5/0x1e0 [btrfs]
> btrfs_ioctl_snap_create_v2+0x1d3/0x270 [btrfs]
> btrfs_ioctl+0x241b/0x3e60 [btrfs]
> do_vfs_ioctl+0x831/0xb10
> ksys_ioctl+0x67/0x90
> __x64_sys_ioctl+0x43/0x50
> do_syscall_64+0x79/0xe0
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> btrfs_reloc_pre_snapshot+0x85/0xc0 [btrfs]
> create_pending_snapshot+0x209/0x15f0 [btrfs]
> create_pending_snapshots+0x111/0x140 [btrfs]
> btrfs_commit_transaction+0x7a6/0x1360 [btrfs]
> btrfs_mksubvol+0x915/0x960 [btrfs]
> btrfs_ioctl_snap_create_transid+0x1d5/0x1e0 [btrfs]
> btrfs_ioctl_snap_create_v2+0x1d3/0x270 [btrfs]
> btrfs_ioctl+0x241b/0x3e60 [btrfs]
> do_vfs_ioctl+0x831/0xb10
> ksys_ioctl+0x67/0x90
> __x64_sys_ioctl+0x43/0x50
> do_syscall_64+0x79/0xe0
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> [CAUSE]
> All these call sites are only relying on root->reloc_root, which can
> undergo btrfs_drop_snapshot(), and since we don't have real refcount
> based protection to reloc roots, we can reach already dropped reloc
> root, triggering KASAN.
>
> [FIX]
> To avoid such access to unstable root->reloc_root, we should check
> BTRFS_ROOT_DEAD_RELOC_TREE bit first.
>
> This patch introduces a new wrapper, have_reloc_root(), to do the proper
> check for most callers who don't distinguish merged reloc tree and no
> reloc tree.
>
> The only exception is should_ignore_root(), as merged reloc tree can be
> ignored, while no reloc tree shouldn't.
>
> Also, set_bit()/clear_bit()/test_bit() doesn't imply a memory barrier,
> and BTRFS_ROOT_DEAD_RELOC_TREE is the only indicator, also add extra
> memory barrier for that bit.
>
> Reported-by: Zygo Blaxell <ce3g8jdj@xxxxxxxxxxxxxxxxxxxxx>
> Fixes: d2311e698578 ("btrfs: relocation: Delay reloc tree deletion after merge_reloc_roots")
> Singed-off-by: David Sterba <dsterba@xxxxxxxx>
> Signed-off-by: Qu Wenruo <wqu@xxxxxxxx>
> ---
> Difference between this and David's diff:
> - Use proper smp_mb__after_atomic() for clear_bit()
> - Use test_bit() only check for should_ignore_root()
> That call site is an except, can't go regular have_reloc_root() check
> - Add extra comment for have_reloc_root()
> ---
> fs/btrfs/relocation.c | 32 ++++++++++++++++++++++++++++----
> 1 file changed, 28 insertions(+), 4 deletions(-)
>
> diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
> index d897a8e5e430..586f045bb6dc 100644
> --- a/fs/btrfs/relocation.c
> +++ b/fs/btrfs/relocation.c
> @@ -517,6 +517,23 @@ static int update_backref_cache(struct btrfs_trans_handle *trans,
> return 1;
> }
>
> +/*
> + * Check if this subvolume tree has valid reloc(*) tree.
> + *
> + * *: Reloc tree after swap is considered dead, thus not considered as valid.
> + * This is enough for most callers, as they don't distinguish dead reloc
> + * root from no reloc root.
> + * But should_ignore_root() below is a special case.
> + */
> +static bool have_reloc_root(struct btrfs_root *root)
> +{
> + smp_mb__before_atomic();
> + if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
> + return false;
> + if (!root->reloc_root)
> + return false;
> + return true;
> +}
>
> static int should_ignore_root(struct btrfs_root *root)
> {
> @@ -525,6 +542,11 @@ static int should_ignore_root(struct btrfs_root *root)
> if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
> return 0;
>
> + /* This root has been merged with its reloc tree, we can ignore it */
> + smp_mb__before_atomic();
Haven't analyzed the patch deeply but if you add memory barriers you
*must* add comments explaining the ordering guarantees those barriers
provide.
<snip>
