Skip to contentSkip to conversation

fat: add i_disksize to represent uninitialized size

PATCH Superseded Namjae Jeon 2 participants 8 messages Dec 25, 2013 – Feb 7, 2014

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/6 fat: add i_disksize to represent uninitialized size

From: Namjae Jeon <[email protected]>

Add i_disksize to represent uninitialized allocated size. And mmu_private represent initialized allocated size.

---
 fs/fat/cache.c |    6 +++---
 fs/fat/fat.h   |    3 ++-
 fs/fat/file.c  |    4 +++-
 fs/fat/inode.c |    4 ++++
 4 files changed, 12 insertions(+), 5 deletions(-)
fs/fat/cache.c +3 -3
index 91ad9e1..a132666 100644
@@ -329,10 +329,10 @@ int fat_bmap(struct inode *inode, sector_t sector, sector_t *phys,
 			return 0;
 
 		/*
-		 * ->mmu_private can access on only allocation path.
-		 * (caller must hold ->i_mutex)
+		 * Both ->mmu_private and ->i_disksize can access
+		 * on only allocation path. (caller must hold ->i_mutex)
 		 */
-		last_block = (MSDOS_I(inode)->mmu_private + (blocksize - 1))
+		last_block = (MSDOS_I(inode)->i_disksize + (blocksize - 1))
 			>> blocksize_bits;
 		if (sector >= last_block)
 			return 0;
fs/fat/fat.h +2 -1
index 7c31f4b..7b5851f 100644
@@ -118,7 +118,8 @@ struct msdos_inode_info {
 	unsigned int cache_valid_id;
 
 	/* NOTE: mmu_private is 64bits, so must hold ->i_mutex to access */
-	loff_t mmu_private;	/* physically allocated size */
+	loff_t mmu_private;	/* physically allocated size (initialized) */
+	loff_t i_disksize;	/* physically allocated size (uninitialized) */
 
 	int i_start;		/* first cluster or 0 */
 	int i_logstart;		/* logical first cluster */
fs/fat/file.c +3 -1
index 9b104f5..0bf0d28 100644
@@ -300,8 +300,10 @@ void fat_truncate_blocks(struct inode *inode, loff_t offset)
 	 * This protects against truncating a file bigger than it was then
 	 * trying to write into the hole.
 	 */
-	if (MSDOS_I(inode)->mmu_private > offset)
+	if (MSDOS_I(inode)->i_disksize > offset) {
 		MSDOS_I(inode)->mmu_private = offset;
+		MSDOS_I(inode)->i_disksize = offset;
+	}
 
 	nr_clusters = (offset + (cluster_size - 1)) >> sbi->cluster_bits;
 
fs/fat/inode.c +4 -0
index 854b578..7e5531a 100644
@@ -93,6 +93,7 @@ static inline int __fat_get_block(struct inode *inode, sector_t iblock,
 
 	*max_blocks = min(mapped_blocks, *max_blocks);
 	MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
+	MSDOS_I(inode)->i_disksize = MSDOS_I(inode)->mmu_private;
 
 	err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create);
 	if (err)
@@ -408,6 +409,7 @@ int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
 		if (error < 0)
 			return error;
 		MSDOS_I(inode)->mmu_private = inode->i_size;
+		MSDOS_I(inode)->i_disksize = inode->i_size;
 
 		set_nlink(inode, fat_subdirs(inode));
 	} else { /* not a directory */
@@ -423,6 +425,7 @@ int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
 		inode->i_fop = &fat_file_operations;
 		inode->i_mapping->a_ops = &fat_aops;
 		MSDOS_I(inode)->mmu_private = inode->i_size;
+		MSDOS_I(inode)->i_disksize = inode->i_size;
 	}
 	if (de->attr & ATTR_SYS) {
 		if (sbi->options.sys_immutable)
@@ -1223,6 +1226,7 @@ static int fat_read_root(struct inode *inode)
 			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
 	MSDOS_I(inode)->i_logstart = 0;
 	MSDOS_I(inode)->mmu_private = inode->i_size;
+	MSDOS_I(inode)->i_disksize = inode->i_size;
 
 	fat_save_attrs(inode, ATTR_DIR);
 	inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
Signature · 1 line
1.7.9.5
maintainer
#1 lore raw
Quoted 4 lines · Namjae Jeon

From: Namjae Jeon <[email protected]>

Add i_disksize to represent uninitialized allocated size. And mmu_private represent initialized allocated size.

Don't we need to update ->i_disksize after cont_write_begin()?

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#2 lore raw

2014-02-03, OGAWA Hirofumi <[email protected]>:

Quoted 8 lines
Quoted 4 lines · Namjae Jeon

From: Namjae Jeon <[email protected]>

Add i_disksize to represent uninitialized allocated size. And mmu_private represent initialized allocated size.

Don't we need to update ->i_disksize after cont_write_begin()?

We don't need to update i_disksize after cont_write_begin. It is taken care by the fat_get_block after the allocation. For all write paths we align the mmu_private and i_disksize from fat_fill_inode and fat_get_block.

Thanks.

Quoted 3 lines

--
OGAWA Hirofumi <[email protected]>

maintainer
#3 lore raw
Quoted 5 lines · Namjae Jeon

2014-02-03, OGAWA Hirofumi <[email protected]>:

Quoted 4 lines
Quoted 2 lines · Namjae Jeon

From: Namjae Jeon <[email protected]>

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

Add i_disksize to represent uninitialized allocated size. And mmu_private represent initialized allocated size.

Don't we need to update ->i_disksize after cont_write_begin()?

We don't need to update i_disksize after cont_write_begin. It is taken care by the fat_get_block after the allocation. For all write paths we align the mmu_private and i_disksize from fat_fill_inode and fat_get_block.

fat_fill_inode() just set i_disksize to i_size. So, it is not aligned by cluster size or block size.

E.g. ->mmu_private = 500. Then, cont_write_begin() can set ->mmu_private to 512 on some case. In this case, fat_get_block() will not be called, because no new allocation.

If this is true, it would be possible to have ->mmu_private == 512 and ->i_disksize == 500.

I'm missing something?

Thanks.

Signature · 1 line
OGAWA Hirofumi <[email protected]>
maintainer
#4 lore raw
Quoted 9 lines · OGAWA Hirofumi
Quoted 5 lines
Quoted 1 line

Don't we need to update ->i_disksize after cont_write_begin()?

We don't need to update i_disksize after cont_write_begin. It is taken care by the fat_get_block after the allocation. For all write paths we align the mmu_private and i_disksize from fat_fill_inode and fat_get_block.

fat_fill_inode() just set i_disksize to i_size. So, it is not aligned by cluster size or block size.

Quoted 8 lines

E.g. ->mmu_private = 500. Then, cont_write_begin() can set ->mmu_private to 512 on some case. In this case, fat_get_block() will not be called, because no new allocation.

If this is true, it would be possible to have ->mmu_private == 512 and ->i_disksize == 500.

I'm missing something?

BTW, even if above was right, I'm not checking whether updating ->i_disksize after cont_write_begin() is right fix or not.

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#5 lore raw

2014-02-04, OGAWA Hirofumi <[email protected]>:

Quoted 15 lines
Quoted 13 lines · OGAWA Hirofumi
Quoted 5 lines
Quoted 1 line

Don't we need to update ->i_disksize after cont_write_begin()?

We don't need to update i_disksize after cont_write_begin. It is taken care by the fat_get_block after the allocation. For all write paths we align the mmu_private and i_disksize from fat_fill_inode and fat_get_block.

fat_fill_inode() just set i_disksize to i_size. So, it is not aligned by cluster size or block size.

E.g. ->mmu_private = 500. Then, cont_write_begin() can set ->mmu_private to 512 on some case. In this case, fat_get_block() will not be called, because no new allocation.

Quoted 7 lines
Quoted 4 lines

If this is true, it would be possible to have ->mmu_private == 512 and ->i_disksize == 500.

I'm missing something?

BTW, even if above was right, I'm not checking whether updating ->i_disksize after cont_write_begin() is right fix or not.

I understand your concern. these can be mismatched.
But, when checking your doubt, I can not find any side effect. I think that there is no issue regardless of alignment of two value, in the cont_write_begin.
Could you please share any point I am missing ?
If you suggest checking point or test method, I can check more and share the result.

Thanks OGAWA.

Quoted 3 lines

--
OGAWA Hirofumi <[email protected]>

maintainer
#6 lore raw
Quoted 12 lines · Namjae Jeon
Quoted 12 lines
Quoted 11 lines

fat_fill_inode() just set i_disksize to i_size. So, it is not aligned by cluster size or block size.

E.g. ->mmu_private = 500. Then, cont_write_begin() can set ->mmu_private to 512 on some case. In this case, fat_get_block() will not be called, because no new allocation.

If this is true, it would be possible to have ->mmu_private == 512 and ->i_disksize == 500.

I'm missing something?

Quoted 8 lines
Quoted 2 lines

BTW, even if above was right, I'm not checking whether updating ->i_disksize after cont_write_begin() is right fix or not.

I understand your concern. these can be mismatched. But, when checking your doubt, I can not find any side effect. I think that there is no issue regardless of alignment of two value, in the cont_write_begin. Could you please share any point I am missing ? If you suggest checking point or test method, I can check more and share the result.

I'm not checking whether it is wrong or not. But, like you said, ->mmu_private > ->i_disksize is wrong in theory.

Although, it might have no real problem.

So, how about to set ->i_disksize to aligned by blocksize at first (i.e. when initializing the inode)?

This may change the behavior when ->mmu_private is not aligned to blocksize in current patchset. But, in theory, it is right state (between ->mmu_private and ->i_disksize is uninitialized). I guess, we can do it with small adjustments, and keep state valid in theory too.

This is just a my guess, so it might be wrong though. I guess, worth to try to consider.

Thanks.

Signature · 1 line
OGAWA Hirofumi <[email protected]>
#7 lore raw

2014-02-06, OGAWA Hirofumi <[email protected]>:

Quoted 25 lines
Quoted 14 lines · Namjae Jeon
Quoted 14 lines
Quoted 13 lines

fat_fill_inode() just set i_disksize to i_size. So, it is not aligned by
cluster size or block size.

E.g. ->mmu_private = 500. Then, cont_write_begin() can set ->mmu_private
to 512 on some case. In this case, fat_get_block() will not be called, because no new allocation.

If this is true, it would be possible to have ->mmu_private == 512 and ->i_disksize == 500.

I'm missing something?

Quoted 8 lines
Quoted 2 lines

BTW, even if above was right, I'm not checking whether updating ->i_disksize after cont_write_begin() is right fix or not.

I understand your concern. these can be mismatched. But, when checking your doubt, I can not find any side effect. I think that there is no issue regardless of alignment of two value, in the cont_write_begin. Could you please share any point I am missing ? If you suggest checking point or test method, I can check more and share the result.

Quoted 7 lines

I'm not checking whether it is wrong or not. But, like you said, ->mmu_private > ->i_disksize is wrong in theory.

Although, it might have no real problem.

So, how about to set ->i_disksize to aligned by blocksize at first (i.e. when initializing the inode)?

Yes, It is good idea.

Quoted 5 lines

This may change the behavior when ->mmu_private is not aligned to blocksize in current patchset. But, in theory, it is right state (between ->mmu_private and ->i_disksize is uninitialized). I guess, we can do it with small adjustments, and keep state valid in theory too.

Right.

Quoted 3 lines

This is just a my guess, so it might be wrong though. I guess, worth to try to consider.

Okay, I will include it in next patch-set after checking.

Thanks for your review!

Quoted 5 lines

Thanks.
--
OGAWA Hirofumi <[email protected]>