Re: [PATCH] f2fs: merge f2fs_show_injection_info() into time_to_inject()

[Chao Yu]

On 2022/12/16 11:48, Yangtao Li wrote:
> > After moving f2fs_show_injection_info() core functionality into time_to_inject(),
> > __builtin_return_address(0) result changes from return address of caller of
> > f2fs_show_injection_info() to return address of time_to_inject().
>
> I tried the __builtin_return_address(1) parameter just now, and no error
> was reported when compiling, but a null pointer problem will be triggered
> when the kernel is running.
>
> I thought about it, and the print address didn't seem clear enough.
> Let's just print the line number of the caller?

That will cause a regression when searching last injection call paths after
bug occurs, since there are many similar callers of time_to_inject, but the
caller's call paths are different. So, IMO, it's useful to keep
__builtin_return_address in the log to distinguish the real call path of
fault injection.

> #define time_to_inject(sbi, type) __time_to_inject(sbi, type, __func__, __LINE__)

Any way to pass __builtin_return_address(0) from parameter in __time_to_inject(...)?

Thanks,

> static inline bool __time_to_inject(struct f2fs_sb_info *sbi, int type,
>                                      const char *func_name, unsigned int line)
> {
>      struct f2fs_fault_info *ffi = &F2FS_OPTION(sbi).fault_info;
>
>      if (!ffi->inject_rate)
>          return false;
>
>      if (!IS_FAULT_SET(ffi, type))
>          return false;
>
>      atomic_inc(&ffi->inject_ops);
>      if (atomic_read(&ffi->inject_ops) >= ffi->inject_rate) {
>          atomic_set(&ffi->inject_ops, 0);
>          printk_ratelimited("%sF2FS-fs (%s) : inject %s in [%s] %d\n",
>              KERN_INFO, sbi->sb->s_id, f2fs_fault_name[type],
>              func_name, line);
>          return true;
>      }
>      return false;
> }
>
> Thx,
> Yangtao