Re: [PATCH] f2fs: fix atgc bug on issue in 32bits platform

[Chao Yu]

On 2022/11/7 13:04, Zhiguo Niu wrote:
> Dear Chao
>
> Chao Yu <chao@xxxxxxxxxx <mailto:chao@xxxxxxxxxx>> 于2022年11月7日周一 09:25写道：
>
>     On 2022/11/4 15:40, zhiguo.niu wrote:
>      > From: Zhiguo Niu <zhiguo.niu@xxxxxxxxxx <mailto:zhiguo.niu@xxxxxxxxxx>>
>      >
>      > There is bug on issue after atgc feature is enabled in
>      > 32bits platform as the following log:
>      >
>      > F2FS-fs (dm-x): inconsistent rbtree, cur(3470333575168) next(3320009719808)
>      > ------------[ cut here ]------------
>      > kernel BUG at fs/f2fs/gc.c:602!
>      > Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
>      > PC is at get_victim_by_default+0x13c0/0x1498
>      > LR is at f2fs_check_rb_tree_consistence+0xc4/0xd4
>      > ....
>      > [<c04d98b0>] (get_victim_by_default) from [<c04d4f44>] (f2fs_gc+0x220/0x6cc)
>      > [<c04d4f44>] (f2fs_gc) from [<c04d4780>] (gc_thread_func+0x2ac/0x708)
>      > [<c04d4780>] (gc_thread_func) from [<c015c774>] (kthread+0x1a8/0x1b4)
>      > [<c015c774>] (kthread) from [<c01010b4>] (ret_from_fork+0x14/0x20)
>      >
>      > The reason is the 64bits key in struct rb_entry has __packed attibute
>      > but has not in struct victim_entry, so the wrong key value got by
>      > in f2fs_check_rb_tree_consistence, the following are the memory layouts
>      > of struct rb_entry and struct victim_entry in 32bits platform:
>      >
>      > struct rb_entry {
>      >     [0] struct rb_node rb_node;
>      >         union {
>      >             struct {...};
>      >    [12]     unsigned long long key;
>      >         };
>      > }
>      > struct victim_entry {
>      >     [0] struct rb_node rb_node;
>      >         union {
>      >             struct {...};
>      >    [16]     struct victim_info vi;
>      >         };
>      >    [32] struct list_head list;
>      > }
>
>     Shouldn't we add __packed for struct victim_entry?
>
>   I try to keep it consistent with struct rb_entry (__packed is only added to union for 64bits key),
> victim_entry.vi.mtime is used to store the key, and rb_entry.key is used to verify the key,
> so make sure that the memory offsets of the 64 bits key in these two structures are the same.
> Other members of victim_entry, such as list, i think there is no problem because of it storaged and loaded
> are both based on struct victim_entry

I'm not sure, there will be any further update in struct rb_node and
struct list_head later, so, IMO, it will be more safe to add __packed to
align all fields in structure to byte.

Thanks,

>      >
>      > This patch fix this inconsistence layout of 64bits key between
>      > struct rb_entry and struct victim_entry.
>      >
>
>     Fixes: 093749e296e2 ("f2fs: support age threshold based garbage collection")
>
>      > Signed-off-by: Zhiguo Niu <zhiguo.niu@xxxxxxxxxx <mailto:zhiguo.niu@xxxxxxxxxx>>