[v2] ocfs2: retry on ENOSPC if sufficient space in truncate log
diff mbox

Message ID 1467802651-12461-1-git-send-email-zren@suse.com
State New
Headers show

Commit Message

Zhen Ren July 6, 2016, 10:57 a.m. UTC
The testcase "mmaptruncate" in ocfs2 test suite always fails with
ENOSPC error on small volume (say less than 10G). This testcase
repeatedly performs "extend" and "truncate" on a file. Continuously,
it truncates the file to 1/2 of the size, and then extends to 100% of
the size. The main bitmap will quickly run out of space because the
"truncate" code prevent truncate log from being flushed by
ocfs2_schedule_truncate_log_flush(osb, 1), while truncate log may
have cached lots of clusters.

So retry to allocate after flushing truncate log when ENOSPC is
returned. And we cannot reuse the deleted blocks before the transaction
committed. Fortunately, we already have a function to do this -
ocfs2_try_to_free_truncate_log(). Just need to remove the "static"
modifier and put it into the right place.

The "unlock"/"lock" code isn't elegant, but looks no better option.

v2:
1. Lock allocator inode again if ocfs2_schedule_truncate_log_flush()
fails. -- spotted by Joseph Qi <joseph.qi@huawei.com>

Signed-off-by: Eric Ren <zren@suse.com>
---
 fs/ocfs2/alloc.c    | 37 +++++++++++++++++++++++++++++++++++++
 fs/ocfs2/alloc.h    |  2 ++
 fs/ocfs2/aops.c     | 37 -------------------------------------
 fs/ocfs2/suballoc.c | 20 +++++++++++++++++++-
 4 files changed, 58 insertions(+), 38 deletions(-)

Comments

Joseph Qi July 7, 2016, 1 a.m. UTC | #1
Hi Eric,

On 2016/7/6 18:57, Eric Ren wrote:
> The testcase "mmaptruncate" in ocfs2 test suite always fails with
> ENOSPC error on small volume (say less than 10G). This testcase
> repeatedly performs "extend" and "truncate" on a file. Continuously,
> it truncates the file to 1/2 of the size, and then extends to 100% of
> the size. The main bitmap will quickly run out of space because the
> "truncate" code prevent truncate log from being flushed by
> ocfs2_schedule_truncate_log_flush(osb, 1), while truncate log may
> have cached lots of clusters.
> 
> So retry to allocate after flushing truncate log when ENOSPC is
> returned. And we cannot reuse the deleted blocks before the transaction
> committed. Fortunately, we already have a function to do this -
> ocfs2_try_to_free_truncate_log(). Just need to remove the "static"
> modifier and put it into the right place.
> 
> The "unlock"/"lock" code isn't elegant, but looks no better option.
> 
> v2:
> 1. Lock allocator inode again if ocfs2_schedule_truncate_log_flush()
> fails. -- spotted by Joseph Qi <joseph.qi@huawei.com>
> 
> Signed-off-by: Eric Ren <zren@suse.com>
> ---
>  fs/ocfs2/alloc.c    | 37 +++++++++++++++++++++++++++++++++++++
>  fs/ocfs2/alloc.h    |  2 ++
>  fs/ocfs2/aops.c     | 37 -------------------------------------
>  fs/ocfs2/suballoc.c | 20 +++++++++++++++++++-
>  4 files changed, 58 insertions(+), 38 deletions(-)
> 
> diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
> index 460c0ce..7dabbc3 100644
> --- a/fs/ocfs2/alloc.c
> +++ b/fs/ocfs2/alloc.c
> @@ -6106,6 +6106,43 @@ void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
>  	}
>  }
>  
> +/*
> + * Try to flush truncate logs if we can free enough clusters from it.
> + * As for return value, "< 0" means error, "0" no space and "1" means
> + * we have freed enough spaces and let the caller try to allocate again.
> + */
> +int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
> +					unsigned int needed)
> +{
> +	tid_t target;
> +	int ret = 0;
> +	unsigned int truncated_clusters;
> +
> +	inode_lock(osb->osb_tl_inode);
> +	truncated_clusters = osb->truncated_clusters;
> +	inode_unlock(osb->osb_tl_inode);
> +
> +	/*
> +	 * Check whether we can succeed in allocating if we free
> +	 * the truncate log.
> +	 */
> +	if (truncated_clusters < needed)
> +		goto out;
> +
> +	ret = ocfs2_flush_truncate_log(osb);
> +	if (ret) {
> +		mlog_errno(ret);
> +		goto out;
> +	}
> +
> +	if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
> +		jbd2_log_wait_commit(osb->journal->j_journal, target);
> +		ret = 1;
> +	}
> +out:
> +	return ret;
> +}
> +
>  static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
>  				       int slot_num,
>  				       struct inode **tl_inode,
> diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
> index f3dc1b0..4a5152e 100644
> --- a/fs/ocfs2/alloc.h
> +++ b/fs/ocfs2/alloc.h
> @@ -188,6 +188,8 @@ int ocfs2_truncate_log_append(struct ocfs2_super *osb,
>  			      u64 start_blk,
>  			      unsigned int num_clusters);
>  int __ocfs2_flush_truncate_log(struct ocfs2_super *osb);
> +int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
> +				   unsigned int needed);
>  
>  /*
>   * Process local structure which describes the block unlinks done
> diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
> index c034edf..1802aef 100644
> --- a/fs/ocfs2/aops.c
> +++ b/fs/ocfs2/aops.c
> @@ -1645,43 +1645,6 @@ static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
>  	return ret;
>  }
>  
> -/*
> - * Try to flush truncate logs if we can free enough clusters from it.
> - * As for return value, "< 0" means error, "0" no space and "1" means
> - * we have freed enough spaces and let the caller try to allocate again.
> - */
> -static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
> -					  unsigned int needed)
> -{
> -	tid_t target;
> -	int ret = 0;
> -	unsigned int truncated_clusters;
> -
> -	inode_lock(osb->osb_tl_inode);
> -	truncated_clusters = osb->truncated_clusters;
> -	inode_unlock(osb->osb_tl_inode);
> -
> -	/*
> -	 * Check whether we can succeed in allocating if we free
> -	 * the truncate log.
> -	 */
> -	if (truncated_clusters < needed)
> -		goto out;
> -
> -	ret = ocfs2_flush_truncate_log(osb);
> -	if (ret) {
> -		mlog_errno(ret);
> -		goto out;
> -	}
> -
> -	if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
> -		jbd2_log_wait_commit(osb->journal->j_journal, target);
> -		ret = 1;
> -	}
> -out:
> -	return ret;
> -}
> -
>  int ocfs2_write_begin_nolock(struct address_space *mapping,
>  			     loff_t pos, unsigned len, ocfs2_write_type_t type,
>  			     struct page **pagep, void **fsdata,
> diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
> index 2f19aee..1458989 100644
> --- a/fs/ocfs2/suballoc.c
> +++ b/fs/ocfs2/suballoc.c
> @@ -1164,7 +1164,8 @@ static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
>  					     int flags,
>  					     struct ocfs2_alloc_context **ac)
>  {
> -	int status;
> +	int status, ret = 0;
> +	int retried = 0;
>  
>  	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
>  	if (!(*ac)) {
> @@ -1189,7 +1190,24 @@ static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
>  	}
>  
>  	if (status == -ENOSPC) {
> +retry:
>  		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
> +		/* Retry if there is sufficient space cached in truncate log */
> +		if (status == -ENOSPC && !retried) {
> +			retried = 1;
> +			ocfs2_inode_unlock((*ac)->ac_inode, 1);
> +			inode_unlock((*ac)->ac_inode);
> +
> +			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
> +			if (ret == 1)
> +				goto retry;
> +
> +			if (ret < 0) {
> +				mlog_errno(ret);
> +				inode_lock((*ac)->ac_inode);
> +				ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
> +			}
Here not only the case error occurs in ocfs2_try_to_free_truncate_log,
but also have tried to free truncate log but still no space, you have
already unlocked inode and it will still call ocfs2_free_alloc_context
and unlock again.
I suggest here we consider a bit code refactoring but not just lock
again.

Thanks,
Joseph

> +		}
>  		if (status < 0) {
>  			if (status != -ENOSPC)
>  				mlog_errno(status);
>
Zhen Ren July 7, 2016, 9:59 a.m. UTC | #2
Hi Joseph,

On 07/07/2016 09:00 AM, Joseph Qi wrote:
>> @@ -1164,7 +1164,8 @@ static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
>>   					     int flags,
>>   					     struct ocfs2_alloc_context **ac)
>>   {
>> -	int status;
>> +	int status, ret = 0;
>> +	int retried = 0;
>>
>>   	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
>>   	if (!(*ac)) {
>> @@ -1189,7 +1190,24 @@ static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
>>   	}
>>
>>   	if (status == -ENOSPC) {
>> +retry:
>>   		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
>> +		/* Retry if there is sufficient space cached in truncate log */
>> +		if (status == -ENOSPC && !retried) {
>> +			retried = 1;
>> +			ocfs2_inode_unlock((*ac)->ac_inode, 1);
>> +			inode_unlock((*ac)->ac_inode);
>> +
>> +			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
>> +			if (ret == 1)
>> +				goto retry;
>> +
>> +			if (ret < 0) {
>> +				mlog_errno(ret);
>> +				inode_lock((*ac)->ac_inode);
>> +				ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
>> +			}
> Here not only the case error occurs in ocfs2_try_to_free_truncate_log,
> but also have tried to free truncate log but still no space, you have
> already unlocked inode and it will still call ocfs2_free_alloc_context
> and unlock again.
> I suggest here we consider a bit code refactoring but not just lock
> again.

Oh, yes! "< 0" means error and "0" means no space. I also considered the 
possibility of refactoring somehow. But it needs plenty of changes
because of DLM glue layer doesn't keep track of the lock holder at 
process level. I know it looks ugly and inefficient, but fortunately
the chance of hitting here is very little, so I think it doesn't worth
a refactor. Any suggestions?

Eric

>
> Thanks,
> Joseph
>
>> +		}
>>   		if (status < 0) {
>>   			if (status != -ENOSPC)
>>   				mlog_errno(status);

Patch
diff mbox

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index 460c0ce..7dabbc3 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -6106,6 +6106,43 @@  void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
 	}
 }
 
+/*
+ * Try to flush truncate logs if we can free enough clusters from it.
+ * As for return value, "< 0" means error, "0" no space and "1" means
+ * we have freed enough spaces and let the caller try to allocate again.
+ */
+int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
+					unsigned int needed)
+{
+	tid_t target;
+	int ret = 0;
+	unsigned int truncated_clusters;
+
+	inode_lock(osb->osb_tl_inode);
+	truncated_clusters = osb->truncated_clusters;
+	inode_unlock(osb->osb_tl_inode);
+
+	/*
+	 * Check whether we can succeed in allocating if we free
+	 * the truncate log.
+	 */
+	if (truncated_clusters < needed)
+		goto out;
+
+	ret = ocfs2_flush_truncate_log(osb);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
+		jbd2_log_wait_commit(osb->journal->j_journal, target);
+		ret = 1;
+	}
+out:
+	return ret;
+}
+
 static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
 				       int slot_num,
 				       struct inode **tl_inode,
diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
index f3dc1b0..4a5152e 100644
--- a/fs/ocfs2/alloc.h
+++ b/fs/ocfs2/alloc.h
@@ -188,6 +188,8 @@  int ocfs2_truncate_log_append(struct ocfs2_super *osb,
 			      u64 start_blk,
 			      unsigned int num_clusters);
 int __ocfs2_flush_truncate_log(struct ocfs2_super *osb);
+int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
+				   unsigned int needed);
 
 /*
  * Process local structure which describes the block unlinks done
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index c034edf..1802aef 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -1645,43 +1645,6 @@  static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
 	return ret;
 }
 
-/*
- * Try to flush truncate logs if we can free enough clusters from it.
- * As for return value, "< 0" means error, "0" no space and "1" means
- * we have freed enough spaces and let the caller try to allocate again.
- */
-static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
-					  unsigned int needed)
-{
-	tid_t target;
-	int ret = 0;
-	unsigned int truncated_clusters;
-
-	inode_lock(osb->osb_tl_inode);
-	truncated_clusters = osb->truncated_clusters;
-	inode_unlock(osb->osb_tl_inode);
-
-	/*
-	 * Check whether we can succeed in allocating if we free
-	 * the truncate log.
-	 */
-	if (truncated_clusters < needed)
-		goto out;
-
-	ret = ocfs2_flush_truncate_log(osb);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
-
-	if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
-		jbd2_log_wait_commit(osb->journal->j_journal, target);
-		ret = 1;
-	}
-out:
-	return ret;
-}
-
 int ocfs2_write_begin_nolock(struct address_space *mapping,
 			     loff_t pos, unsigned len, ocfs2_write_type_t type,
 			     struct page **pagep, void **fsdata,
diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
index 2f19aee..1458989 100644
--- a/fs/ocfs2/suballoc.c
+++ b/fs/ocfs2/suballoc.c
@@ -1164,7 +1164,8 @@  static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 					     int flags,
 					     struct ocfs2_alloc_context **ac)
 {
-	int status;
+	int status, ret = 0;
+	int retried = 0;
 
 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 	if (!(*ac)) {
@@ -1189,7 +1190,24 @@  static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 	}
 
 	if (status == -ENOSPC) {
+retry:
 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
+		/* Retry if there is sufficient space cached in truncate log */
+		if (status == -ENOSPC && !retried) {
+			retried = 1;
+			ocfs2_inode_unlock((*ac)->ac_inode, 1);
+			inode_unlock((*ac)->ac_inode);
+
+			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
+			if (ret == 1)
+				goto retry;
+
+			if (ret < 0) {
+				mlog_errno(ret);
+				inode_lock((*ac)->ac_inode);
+				ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
+			}
+		}
 		if (status < 0) {
 			if (status != -ENOSPC)
 				mlog_errno(status);