Skip to contentSkip to conversation

ext4: Do not allow retry alloc loop under open transaction

PATCH In review Dmitry Monakhov 2 participants 3 messages Feb 3, 2010 – Feb 4, 2010

Subsystems (2)
VFS ext4 file system
lore
Share & tools
lore

Dmitry Monakhov

author
#0 lore raw

PATCH 1/2 ext4: Do not allow retry alloc loop under open transaction

Some times we call ->write_begin() with opened journal but write_begin() may internally call ext4_should_retry_alloc() in case of ENOSPC, which result in deadlock.
This patch introduce new AOP_FLAG which should be tested on retry alloc path.

Signed-off-by Dmitry Monakhov <[email protected]>
---
 fs/ext4/ext4.h        |    5 +++++
 fs/ext4/inode.c       |    6 ++++--
 fs/ext4/move_extent.c |    2 +-
 include/linux/fs.h    |    1 +
 4 files changed, 11 insertions(+), 3 deletions(-)
fs/ext4/ext4.h +5 -0
index af7b626..0efb224 100644
@@ -792,6 +792,11 @@ struct ext4_inode_info {
 #define EXT4_DFL_CHECKINTERVAL		0	/* Don't use interval check */
 
 /*
+ * Address space flags.
+ */
+#define EXT4_AOP_FLAG_NORETRY	AOP_FLAG_LAST	/* Do not allow to fail in to
+						   retry alloc loop */
+/*
  * Behaviour when detecting errors
  */
 #define EXT4_ERRORS_CONTINUE		1	/* Continue execution */
fs/ext4/inode.c +4 -2
index c818972..2d3fe4d 100644
@@ -1593,7 +1593,8 @@ retry:
 		}
 	}
 
-	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
+	if (!(flags & EXT4_AOP_FLAG_NORETRY) && ret == -ENOSPC &&
+		ext4_should_retry_alloc(inode->i_sb, &retries))
 		goto retry;
 out:
 	return ret;

@@ -3089,7 +3090,8 @@ retry:
 			ext4_truncate_failed_write(inode);
 	}
 
-	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
+	if (!(flags & EXT4_AOP_FLAG_NORETRY) &&
+		ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 		goto retry;
 out:
 	return ret;
fs/ext4/move_extent.c +1 -1
index 82c415b..f894382 100644
@@ -796,7 +796,7 @@ move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
 	ext4_lblk_t orig_blk_offset;
 	long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
 	unsigned long blocksize = orig_inode->i_sb->s_blocksize;
-	unsigned int w_flags = 0;
+	unsigned int w_flags = EXT4_AOP_FLAG_NORETRY;
 	unsigned int tmp_data_size, data_size, replaced_size;
 	void *fsdata;
 	int i, jblocks;
include/linux/fs.h +1 -0
index b1bcb27..6f68707 100644
@@ -504,6 +504,7 @@ enum positive_aop_returns {
 #define AOP_FLAG_NOFS			0x0004 /* used by filesystem to direct
 						* helper code (eg buffer layer)
 						* to clear GFP_FS from alloc */
+#define AOP_FLAG_LAST			0x0008 /* First unused private flag */
 
 /*
  * oh the beauties of C type declarations.
Signature · 1 line
1.6.3.3
#1 lore raw
Quoted 7 lines · Dmitry Monakhov

Some times we call ->write_begin() with opened journal but write_begin() may internally call ext4_should_retry_alloc() in case of ENOSPC, which result in deadlock.
This patch introduce new AOP_FLAG which should be tested on retry alloc path.

Can you explain this further. We do a journal_stop before ext4_should_retry_alloc. So not sure about the deadlock.

-aneesh

author
#2 lore raw
Quoted 11 lines · Aneesh Kumar K. V
Quoted 7 lines · Dmitry Monakhov

Some times we call ->write_begin() with opened journal but write_begin() may internally call ext4_should_retry_alloc() in case of ENOSPC, which result in deadlock.
This patch introduce new AOP_FLAG which should be tested on retry alloc path.

Can you explain this further. We do a journal_stop before ext4_should_retry_alloc. So not sure about the deadlock.

move_extent_per_page
ext4_journal_start -> current->journal_info != NULL ->write_begin

    ext4_journal_start
    ext4_journal_stop
    ext4_should_retry_alloc : here current->journal_info != NULL

Sorry i've missed the fact that ext4_should_retry_alloc() check for current->journal_info and skip running transaction in that case. Seems what this check was added long time ago, but i miss it. So this patch is useless.

Quoted 2 lines

-aneesh