On 10/29/25 16:09, Yongpeng Yang wrote:
> On 10/29/25 14:31, Chao Yu via Linux-f2fs-devel wrote:
>> It recommends to use i_size_{read,write}() to access and update i_size,
>> otherwise, we may get wrong tearing value due to high 32-bits value
>> and low 32-bits value of i_size field are not updated atomically in
>> 32-bits archicture machine.
>>
>> Signed-off-by: Chao Yu <[email protected]>
>> ---
>> include/trace/events/f2fs.h | 8 ++++----
>> 1 file changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/trace/events/f2fs.h b/include/trace/events/f2fs.h
>> index edbbd869078f..e1fae78d64a5 100644
>> --- a/include/trace/events/f2fs.h
>> +++ b/include/trace/events/f2fs.h
>> @@ -204,7 +204,7 @@ DECLARE_EVENT_CLASS(f2fs__inode,
>> __entry->pino = F2FS_I(inode)->i_pino;
>> __entry->mode = inode->i_mode;
>> __entry->nlink = inode->i_nlink;
>> - __entry->size = inode->i_size;
>> + __entry->size = i_size_read(inode);
>> __entry->blocks = inode->i_blocks;
>> __entry->advise = F2FS_I(inode)->i_advise;
>> ),
>> @@ -353,7 +353,7 @@ TRACE_EVENT(f2fs_unlink_enter,
>> TP_fast_assign(
>> __entry->dev = dir->i_sb->s_dev;
>> __entry->ino = dir->i_ino;
>> - __entry->size = dir->i_size;
>> + __entry->size = i_size_read(dir);
>> __entry->blocks = dir->i_blocks;
>> __assign_str(name);
>> ),
>> @@ -433,7 +433,7 @@ DECLARE_EVENT_CLASS(f2fs__truncate_op,
>> TP_fast_assign(
>> __entry->dev = inode->i_sb->s_dev;
>> __entry->ino = inode->i_ino;
>> - __entry->size = inode->i_size;
>> + __entry->size = i_size_read(inode);
>> __entry->blocks = inode->i_blocks;
>> __entry->from = from;
>> ),
>> @@ -1006,7 +1006,7 @@ TRACE_EVENT(f2fs_fallocate,
>> __entry->mode = mode;
>> __entry->offset = offset;
>> __entry->len = len;
>> - __entry->size = inode->i_size;
>> + __entry->size = i_size_read(inode);
>> __entry->blocks = inode->i_blocks;
>> __entry->ret = ret;
>> ),
>
> inode->i_size usage in fs/f2fs/ also needs to be updated. For example in
> f2fs_need_verity(), which may not protected by inode_lock.
Accessing i_size directly in f2fs_need_verity() should be fine? Because fsverity
actived inode should be readonly.
Do you see any other cases which are not covered by inode_lock()?
Thanks,
>
> Yongpeng,
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
