[PATCH 1/2] f2fs: fix iostat lock protection

[PATCH 1/2] f2fs: fix iostat lock protection

[Bo Ye]

From: Qilin Tan <qilin.tan@mediatek.com>

Made iostat lock irq safe to avoid potentinal deadlock.

Deadlock scenario:
f2fs_attr_store
  -> f2fs_sbi_store
  -> _sbi_store
  -> spin_lock(sbi->iostat_lock)
    <interrupt request>
    -> scsi_end_request
    -> bio_endio
    -> f2fs_dio_read_end_io
    -> f2fs_update_iostat
    -> spin_lock_irqsave(sbi->iostat_lock)  ===> Dead lock here

Signed-off-by: Qilin Tan <qilin.tan@mediatek.com>
---
 fs/f2fs/sysfs.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
index 0b19163c90d4..fd238a68017e 100644
--- a/fs/f2fs/sysfs.c
+++ b/fs/f2fs/sysfs.c
@@ -575,9 +575,9 @@ static ssize_t __sbi_store(struct f2fs_attr *a,
 	if (!strcmp(a->attr.name, "iostat_period_ms")) {
 		if (t < MIN_IOSTAT_PERIOD_MS || t > MAX_IOSTAT_PERIOD_MS)
 			return -EINVAL;
-		spin_lock(&sbi->iostat_lock);
+		spin_lock_irq(&sbi->iostat_lock);
 		sbi->iostat_period_ms = (unsigned int)t;
-		spin_unlock(&sbi->iostat_lock);
+		spin_unlock_irq(&sbi->iostat_lock);
 		return count;
 	}
 #endif
-- 
2.17.0

[PATCH 2/2] f2fs: add __pack attribute for extent_info

[Bo Ye]

From: Qilin Tan <qilin.tan@mediatek.com>

Need add __pack for struct extent_info to align to memory
layout of struct rb_entry.

struct rb_entry {
    struct rb_node rb_node;     /* rb node located in rb-tree */
    union {
        struct {
            unsigned int ofs;   /* start offset of the entry */
            unsigned int len;   /* length of the entry */
        };
        unsigned long long key;     /* 64-bits key */
    } __packed;
};

struct extent_info {
    unsigned int fofs;      /* start offset in a file */
    unsigned int len;       /* length of the extent */
    union {
        ...
        /* block age extent_cache */
        struct {
            /* block age of the extent */
            unsigned long long age;
            /* last total blocks allocated */
            unsigned long long last_blocks;
        };
};

The new fields(age, last_blocks) are u64 in change 71644dff4811, it
cause the memory alignment based on 8 bytes in some complier. So the
field fofs and len are alloced with 8 bytes and using the last 4 byts.
Its memory is not aligned with struct rb_entry. the ofs of rb_entry
pointer to a invalid value and cause writing file failed.

struct extent_info's offset should be 12 rather than 16 from the
beginning of struct rb_entry.

The offset dump for the bad case as:
kworker/u16:6: [name:f2fs&]f2fs_lookup_rb_tree_for_insert:MTK_DEBUG: ino=1629 re=0xc675dc08 ofs=0 re->ofs=0, re->len=68 ei=0xc675dc18 ei.fofs=68 ei.len=277

Fixes: 71644dff4811 ("f2fs: add block_age-based extent cache")
Signed-off-by: Bo Ye <bo.ye@mediatek.com>
Signed-off-by: Qilin Tan <qilin.tan@mediatek.com>
---
 fs/f2fs/f2fs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index b0ab2062038a..7c690667a42f 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -660,7 +660,7 @@ struct extent_info {
 			unsigned long long last_blocks;
 		};
 	};
-};
+} __packed;
 
 struct extent_node {
 	struct rb_node rb_node;		/* rb node located in rb-tree */
-- 
2.17.0

Re: [PATCH 2/2] f2fs: add __pack attribute for extent_info

[Chao Yu]

On 2023/3/31 17:26, Bo Ye wrote:
> From: Qilin Tan <qilin.tan@mediatek.com>
> 
> Need add __pack for struct extent_info to align to memory
> layout of struct rb_entry.

Jaegeuk has fixed this bug w/ below patchset, please check it:

https://lore.kernel.org/linux-f2fs-devel/20230313201216.924234-1-jaegeuk@kernel.org/

Thanks,

> 
> struct rb_entry {
>      struct rb_node rb_node;     /* rb node located in rb-tree */
>      union {
>          struct {
>              unsigned int ofs;   /* start offset of the entry */
>              unsigned int len;   /* length of the entry */
>          };
>          unsigned long long key;     /* 64-bits key */
>      } __packed;
> };
> 
> struct extent_info {
>      unsigned int fofs;      /* start offset in a file */
>      unsigned int len;       /* length of the extent */
>      union {
>          ...
>          /* block age extent_cache */
>          struct {
>              /* block age of the extent */
>              unsigned long long age;
>              /* last total blocks allocated */
>              unsigned long long last_blocks;
>          };
> };
> 
> The new fields(age, last_blocks) are u64 in change 71644dff4811, it
> cause the memory alignment based on 8 bytes in some complier. So the
> field fofs and len are alloced with 8 bytes and using the last 4 byts.
> Its memory is not aligned with struct rb_entry. the ofs of rb_entry
> pointer to a invalid value and cause writing file failed.
> 
> struct extent_info's offset should be 12 rather than 16 from the
> beginning of struct rb_entry.
> 
> The offset dump for the bad case as:
> kworker/u16:6: [name:f2fs&]f2fs_lookup_rb_tree_for_insert:MTK_DEBUG: ino=1629 re=0xc675dc08 ofs=0 re->ofs=0, re->len=68 ei=0xc675dc18 ei.fofs=68 ei.len=277
> 
> Fixes: 71644dff4811 ("f2fs: add block_age-based extent cache")
> Signed-off-by: Bo Ye <bo.ye@mediatek.com>
> Signed-off-by: Qilin Tan <qilin.tan@mediatek.com>
> ---
>   fs/f2fs/f2fs.h | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index b0ab2062038a..7c690667a42f 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -660,7 +660,7 @@ struct extent_info {
>   			unsigned long long last_blocks;
>   		};
>   	};
> -};
> +} __packed;
>   
>   struct extent_node {
>   	struct rb_node rb_node;		/* rb node located in rb-tree */

Re: [PATCH 2/2] f2fs: add __pack attribute for extent_info

[Bo Ye (叶波)]

On Sat, 2023-04-01 at 09:00 +0800, Chao Yu wrote:
> External email : Please do not click links or open attachments until
> you have verified the sender or the content.
> 
> 
> On 2023/3/31 17:26, Bo Ye wrote:
> > From: Qilin Tan <qilin.tan@mediatek.com>
> > 
> > Need add __pack for struct extent_info to align to memory
> > layout of struct rb_entry.
> 
> Jaegeuk has fixed this bug w/ below patchset, please check it:
> 
> 
https://lore.kernel.org/linux-f2fs-devel/20230313201216.924234-1-jaegeuk@kernel.org/
> 
> Thanks,
> 
  OK, under verifying, I will reply with test result. 
  Thanks a lot!
> > 
> > struct rb_entry {
> >      struct rb_node rb_node;     /* rb node located in rb-tree */
> >      union {
> >          struct {
> >              unsigned int ofs;   /* start offset of the entry */
> >              unsigned int len;   /* length of the entry */
> >          };
> >          unsigned long long key;     /* 64-bits key */
> >      } __packed;
> > };
> > 
> > struct extent_info {
> >      unsigned int fofs;      /* start offset in a file */
> >      unsigned int len;       /* length of the extent */
> >      union {
> >          ...
> >          /* block age extent_cache */
> >          struct {
> >              /* block age of the extent */
> >              unsigned long long age;
> >              /* last total blocks allocated */
> >              unsigned long long last_blocks;
> >          };
> > };
> > 
> > The new fields(age, last_blocks) are u64 in change 71644dff4811, it
> > cause the memory alignment based on 8 bytes in some complier. So
> > the
> > field fofs and len are alloced with 8 bytes and using the last 4
> > byts.
> > Its memory is not aligned with struct rb_entry. the ofs of rb_entry
> > pointer to a invalid value and cause writing file failed.
> > 
> > struct extent_info's offset should be 12 rather than 16 from the
> > beginning of struct rb_entry.
> > 
> > The offset dump for the bad case as:
> > kworker/u16:6:
> > [name:f2fs&]f2fs_lookup_rb_tree_for_insert:MTK_DEBUG: ino=1629
> > re=0xc675dc08 ofs=0 re->ofs=0, re->len=68 ei=0xc675dc18 ei.fofs=68
> > ei.len=277
> > 
> > Fixes: 71644dff4811 ("f2fs: add block_age-based extent cache")
> > Signed-off-by: Bo Ye <bo.ye@mediatek.com>
> > Signed-off-by: Qilin Tan <qilin.tan@mediatek.com>
> > ---
> >   fs/f2fs/f2fs.h | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index b0ab2062038a..7c690667a42f 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -660,7 +660,7 @@ struct extent_info {
> >                       unsigned long long last_blocks;
> >               };
> >       };
> > -};
> > +} __packed;
> > 
> >   struct extent_node {
> >       struct rb_node rb_node;         /* rb node located in rb-tree 
> > */

Re: [PATCH 1/2] f2fs: fix iostat lock protection

[Chao Yu]

On 2023/3/31 17:26, Bo Ye wrote:
> From: Qilin Tan <qilin.tan@mediatek.com>
> 
> Made iostat lock irq safe to avoid potentinal deadlock.
> 
> Deadlock scenario:
> f2fs_attr_store
>    -> f2fs_sbi_store
>    -> _sbi_store
>    -> spin_lock(sbi->iostat_lock)
>      <interrupt request>
>      -> scsi_end_request
>      -> bio_endio
>      -> f2fs_dio_read_end_io
>      -> f2fs_update_iostat
>      -> spin_lock_irqsave(sbi->iostat_lock)  ===> Dead lock here
> 

Fixes: 61803e984307 ("f2fs: fix iostat related lock protection")
Fixes: a1e09b03e6f5 ("f2fs: use iomap for direct I/O")

> Signed-off-by: Qilin Tan <qilin.tan@mediatek.com>

Reviewed-by: Chao Yu <chao@kernel.org>

Thanks,

> ---
>   fs/f2fs/sysfs.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
> index 0b19163c90d4..fd238a68017e 100644
> --- a/fs/f2fs/sysfs.c
> +++ b/fs/f2fs/sysfs.c
> @@ -575,9 +575,9 @@ static ssize_t __sbi_store(struct f2fs_attr *a,
>   	if (!strcmp(a->attr.name, "iostat_period_ms")) {
>   		if (t < MIN_IOSTAT_PERIOD_MS || t > MAX_IOSTAT_PERIOD_MS)
>   			return -EINVAL;
> -		spin_lock(&sbi->iostat_lock);
> +		spin_lock_irq(&sbi->iostat_lock);
>   		sbi->iostat_period_ms = (unsigned int)t;
> -		spin_unlock(&sbi->iostat_lock);
> +		spin_unlock_irq(&sbi->iostat_lock);
>   		return count;
>   	}
>   #endif

Re: [f2fs-dev] [PATCH 1/2] f2fs: fix iostat lock protection

[patchwork-bot+f2fs]

Hello:

This series was applied to jaegeuk/f2fs.git (dev)
by Jaegeuk Kim <jaegeuk@kernel.org>:

On Fri, 31 Mar 2023 17:26:56 +0800 you wrote:
> From: Qilin Tan <qilin.tan@mediatek.com>
> 
> Made iostat lock irq safe to avoid potentinal deadlock.
> 
> Deadlock scenario:
> f2fs_attr_store
>   -> f2fs_sbi_store
>   -> _sbi_store
>   -> spin_lock(sbi->iostat_lock)
>     <interrupt request>
>     -> scsi_end_request
>     -> bio_endio
>     -> f2fs_dio_read_end_io
>     -> f2fs_update_iostat
>     -> spin_lock_irqsave(sbi->iostat_lock)  ===> Dead lock here
> 
> [...]

Here is the summary with links:
  - [f2fs-dev,1/2] f2fs: fix iostat lock protection
    https://git.kernel.org/jaegeuk/f2fs/c/daa080db4e7e
  - [f2fs-dev,2/2] f2fs: add __pack attribute for extent_info
    (no matching commit)

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html