[Ocfs2-devel] [PATCH 1/1] Ocfs2: Teach 'coherency=full' O_DIRECT writes to correctly up_read i_alloc_sem.

Tristan Ye tristan.ye at oracle.com
Mon Dec 6 17:47:05 PST 2010


Joel Becker wrote:
> On Mon, Nov 29, 2010 at 05:21:46PM +0800, Tristan Ye wrote:
>> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
>> index f1e962c..857e013 100644
>> --- a/fs/ocfs2/aops.c
>> +++ b/fs/ocfs2/aops.c
>> @@ -568,7 +568,7 @@ static void ocfs2_dio_end_io(struct kiocb *iocb,
>>  			     bool is_async)
>>  {
>>  	struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
>> -	int level;
>> +	int level, sem_locked;
>>  
>>  	/* this io's submitter should not have unlocked this before we could */
>>  	BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
>> @@ -576,7 +576,9 @@ static void ocfs2_dio_end_io(struct kiocb *iocb,
>>  	ocfs2_iocb_clear_rw_locked(iocb);
>>  
>>  	level = ocfs2_iocb_rw_locked_level(iocb);
>> -	if (!level)
>> +	sem_locked = ocfs2_iocb_is_sem_locked(iocb);
>> +	ocfs2_iocb_clear_sem_locked(iocb);
>> +	if (sem_locked)
>>  		up_read(&inode->i_alloc_sem);
>>  	ocfs2_rw_unlock(inode, level);
>
> 	I agree with Tao, we don't need sem_locked.  Just do:
>
> 	if (ocfs2_iocb_is_sem_locked(iocb)) {
> 		up_read(&inode->i_alloc_sem);
> 		ocfs2_iocb_clear_sem_locked(iocb);
> 	}
>
> I think it reads better.
>
>> diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
>> index 76bfdfd..c7a3e5f 100644
>> --- a/fs/ocfs2/aops.h
>> +++ b/fs/ocfs2/aops.h
>> @@ -72,4 +72,10 @@ static inline void ocfs2_iocb_set_rw_locked(struct kiocb *iocb, int level)
>>  	clear_bit(0, (unsigned long *)&iocb->private)
>>  #define ocfs2_iocb_rw_locked_level(iocb) \
>>  	test_bit(1, (unsigned long *)&iocb->private)
>> +#define ocfs2_iocb_set_sem_locked(iocb) \
>> +	set_bit(2, (unsigned long *)&iocb->private)
>> +#define ocfs2_iocb_clear_sem_locked(iocb) \
>> +	clear_bit(2, (unsigned long *)&iocb->private)
>> +#define ocfs2_iocb_is_sem_locked(iocb) \
>> +	test_bit(2, (unsigned long *)&iocb->private)
>
> 	Give these bits an enum.  Anonymous numbers aren't good.  And
> make the comment better.

    Sounds more reasonable;)

>
>>  #endif /* OCFS2_FILE_H */
>> diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
>> index 77b4c04..0e9d729 100644
>> --- a/fs/ocfs2/file.c
>> +++ b/fs/ocfs2/file.c
>> @@ -2246,7 +2246,10 @@ relock:
>>  	if (direct_io) {
>>  		down_read(&inode->i_alloc_sem);
>>  		have_alloc_sem = 1;
>> -	}
>> +		/* communicate with ocfs2_dio_end_io */
>> +		ocfs2_iocb_set_sem_locked(iocb);
>> +	} else
>> +		ocfs2_iocb_clear_sem_locked(iocb);
>
> 	I understand you're trying to avoid bad bits here, but you
> should explicitly clear it at the top of the function.  You know for a
> fact that, at the top (right before relock:), there is no sem held.
> Then set it as you do, and only re-clear it when you up_write().

    Alright.

>
> Joel
>




More information about the Ocfs2-devel mailing list