Re: [PATCH] Btrfs: fix passing wrong arg gfp_t to decide the correct allocation mode
Hello Jan,
> On Tue, May 07, 2013 at 08:20 (+0200), Wang Shilong wrote:
>> If you look the code carefully, you will see all the tree_mod_alloc()
>> has to use GFP_ATOMIC. However, the original code pass the wrong arg
>> gfp_t in some places, this dosen't cause any problems, because in the
>> tree_mod_alloc(), it ignores arg gfp_t and just use GFP_ATOMIC directly,
>> this is not good.
>>
>> However, i think we should try best not to allocate with GFP_ATOMIC, so
>> i keep the gfp_t there in the hope we can change allocation mode in the
>> future.
>
> NAK.
>
> The code as it is now is prepared to get rid of at least some GFP_ATOMIC
> allocations. You won't get rid of all of them, as there are a lot of spin lock
> situations where we need to add to the tree mod lock anyway.
>
> As a preparation we currently pass the "best" flags (least restrictive) we can
> instead of always passing GFP_ATOMIC. I pointed you to this comment already:
>
> 557 /*
> 558 * once we switch from spin locks to something different, we
> should
> 559 * honor the flags parameter here.
> 560 */
> 561 tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
>
> So, if you want less atomic allocations, find something more suitable than an
> rwlock for fs_info->tree_mod_log_lock an you can in fact replace "GFP_ATOMIC"
> with "flags" in the kzalloc().
>
> The good thing is, because everything is already prepared you don't have to
> think about all the callers again an pass the correct flags.
Anyway, your original code looks messy about passing arg gfp_t..isn't it?
And you pass GFP_NOFS to a function, but in fact this function is surrounded
by rw locks.
I make it clear to the caller what kind of gfp_t we should pass(although now we
always
come to GFP_ATOMIC)...
>
> -Jan
>
>
>> Signed-off-by: Wang Shilong
>> ---
>> fs/btrfs/ctree.c | 37 ++---
>> 1 files changed, 18 insertions(+), 19 deletions(-)
>>
>> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
>> index de6de8e..33c9061 100644
>> --- a/fs/btrfs/ctree.c
>> +++ b/fs/btrfs/ctree.c
>> @@ -553,7 +553,7 @@ static inline int tree_mod_alloc(struct btrfs_fs_info
>> *fs_info, gfp_t flags,
>> * once we switch from spin locks to something different, we should
>> * honor the flags parameter here.
>> */
>> -tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
>> +tm = *tm_ret = kzalloc(sizeof(*tm), flags);
>> if (!tm)
>> return -ENOMEM;
>>
>> @@ -591,14 +591,14 @@ __tree_mod_log_insert_key(struct btrfs_fs_info
>> *fs_info,
>> static noinline int
>> tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
>> struct extent_buffer *eb, int slot,
>> - enum mod_log_op op, gfp_t flags)
>> + enum mod_log_op op)
>> {
>> int ret;
>>
>> if (tree_mod_dont_log(fs_info, eb))
>> return 0;
>>
>> -ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
>> +ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
>>
>> tree_mod_log_write_unlock(fs_info);
>> return ret;
>> @@ -608,7 +608,7 @@ static noinline int
>> tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer
>> *eb,
>> int slot, enum mod_log_op op)
>> {
>> -return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
>> +return tree_mod_log_insert_key_mask(fs_info, eb, slot, op);
>> }
>>
>> static noinline int
>> @@ -616,13 +616,13 @@ tree_mod_log_insert_key_locked(struct btrfs_fs_info
>> *fs_info,
>> struct extent_buffer *eb, int slot,
>> enum mod_log_op op)
>> {
>> -return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
>> +return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
>> }
>>
>> static noinline int
>> tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
>> struct extent_buffer *eb, int dst_slot, int src_slot,
>> - int nr_items, gfp_t flags)
>> + int nr_items)
>> {
>> struct tree_mod_elem *tm;
>> int ret;
>> @@ -642,7 +642,7 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
>> BUG_ON(ret < 0);
>> }
>>
>> -ret = tree_mod_alloc(fs_info, flags, &tm);
>> +ret = tree_mod_alloc(fs_info, GFP_ATOMIC, &tm);
>> if (ret < 0)
>> goto out;
>>
>> @@ -679,7 +679,7 @@ __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
>> struct extent_buffer *eb)
>> static noinline int
>> tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
>> struct extent_buffer *old_root,
>> - struct extent_buffer *new_root, gfp_t flags,
>> + struct extent_buffer *new_root,
>> int log_removal)
>> {
>> struct tree_mod_elem *tm;
>> @@ -
Re: [PATCH] Btrfs: fix passing wrong arg gfp_t to decide the correct allocation mode
On Tue, May 07, 2013 at 08:20 (+0200), Wang Shilong wrote:
> If you look the code carefully, you will see all the tree_mod_alloc()
> has to use GFP_ATOMIC. However, the original code pass the wrong arg
> gfp_t in some places, this dosen't cause any problems, because in the
> tree_mod_alloc(), it ignores arg gfp_t and just use GFP_ATOMIC directly,
> this is not good.
>
> However, i think we should try best not to allocate with GFP_ATOMIC, so
> i keep the gfp_t there in the hope we can change allocation mode in the
> future.
NAK.
The code as it is now is prepared to get rid of at least some GFP_ATOMIC
allocations. You won't get rid of all of them, as there are a lot of spin lock
situations where we need to add to the tree mod lock anyway.
As a preparation we currently pass the "best" flags (least restrictive) we can
instead of always passing GFP_ATOMIC. I pointed you to this comment already:
557 /*
558 * once we switch from spin locks to something different, we should
559 * honor the flags parameter here.
560 */
561 tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
So, if you want less atomic allocations, find something more suitable than an
rwlock for fs_info->tree_mod_log_lock an you can in fact replace "GFP_ATOMIC"
with "flags" in the kzalloc().
The good thing is, because everything is already prepared you don't have to
think about all the callers again an pass the correct flags.
-Jan
> Signed-off-by: Wang Shilong
> ---
> fs/btrfs/ctree.c | 37 ++---
> 1 files changed, 18 insertions(+), 19 deletions(-)
>
> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
> index de6de8e..33c9061 100644
> --- a/fs/btrfs/ctree.c
> +++ b/fs/btrfs/ctree.c
> @@ -553,7 +553,7 @@ static inline int tree_mod_alloc(struct btrfs_fs_info
> *fs_info, gfp_t flags,
>* once we switch from spin locks to something different, we should
>* honor the flags parameter here.
>*/
> - tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
> + tm = *tm_ret = kzalloc(sizeof(*tm), flags);
> if (!tm)
> return -ENOMEM;
>
> @@ -591,14 +591,14 @@ __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
> static noinline int
> tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
>struct extent_buffer *eb, int slot,
> - enum mod_log_op op, gfp_t flags)
> + enum mod_log_op op)
> {
> int ret;
>
> if (tree_mod_dont_log(fs_info, eb))
> return 0;
>
> - ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
> + ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
>
> tree_mod_log_write_unlock(fs_info);
> return ret;
> @@ -608,7 +608,7 @@ static noinline int
> tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer
> *eb,
> int slot, enum mod_log_op op)
> {
> - return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
> + return tree_mod_log_insert_key_mask(fs_info, eb, slot, op);
> }
>
> static noinline int
> @@ -616,13 +616,13 @@ tree_mod_log_insert_key_locked(struct btrfs_fs_info
> *fs_info,
>struct extent_buffer *eb, int slot,
>enum mod_log_op op)
> {
> - return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
> + return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
> }
>
> static noinline int
> tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
>struct extent_buffer *eb, int dst_slot, int src_slot,
> - int nr_items, gfp_t flags)
> + int nr_items)
> {
> struct tree_mod_elem *tm;
> int ret;
> @@ -642,7 +642,7 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
> BUG_ON(ret < 0);
> }
>
> - ret = tree_mod_alloc(fs_info, flags, &tm);
> + ret = tree_mod_alloc(fs_info, GFP_ATOMIC, &tm);
> if (ret < 0)
> goto out;
>
> @@ -679,7 +679,7 @@ __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
> struct extent_buffer *eb)
> static noinline int
> tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
>struct extent_buffer *old_root,
> - struct extent_buffer *new_root, gfp_t flags,
> + struct extent_buffer *new_root,
>int log_removal)
> {
> struct tree_mod_elem *tm;
> @@ -691,7 +691,7 @@ tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
> if (log_removal)
> __tree_mod_log_free_eb(fs_info, old_root);
>
> - ret = tree_mod_alloc(fs_info, flags, &tm);
> + ret = tree_mod_alloc(fs_info, GFP_ATOMIC, &tm);
> if (ret < 0)
> goto out;
>
> @@ -809,19 +809,18 @@ tree_mod_log_eb_move(struct btrfs_fs_info *fs_info,
> struct
[PATCH] Btrfs: fix passing wrong arg gfp_t to decide the correct allocation mode
If you look the code carefully, you will see all the tree_mod_alloc()
has to use GFP_ATOMIC. However, the original code pass the wrong arg
gfp_t in some places, this dosen't cause any problems, because in the
tree_mod_alloc(), it ignores arg gfp_t and just use GFP_ATOMIC directly,
this is not good.
However, i think we should try best not to allocate with GFP_ATOMIC, so
i keep the gfp_t there in the hope we can change allocation mode in the
future.
Signed-off-by: Wang Shilong
---
fs/btrfs/ctree.c | 37 ++---
1 files changed, 18 insertions(+), 19 deletions(-)
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index de6de8e..33c9061 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -553,7 +553,7 @@ static inline int tree_mod_alloc(struct btrfs_fs_info
*fs_info, gfp_t flags,
* once we switch from spin locks to something different, we should
* honor the flags parameter here.
*/
- tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
+ tm = *tm_ret = kzalloc(sizeof(*tm), flags);
if (!tm)
return -ENOMEM;
@@ -591,14 +591,14 @@ __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
static noinline int
tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
struct extent_buffer *eb, int slot,
-enum mod_log_op op, gfp_t flags)
+enum mod_log_op op)
{
int ret;
if (tree_mod_dont_log(fs_info, eb))
return 0;
- ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
+ ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
tree_mod_log_write_unlock(fs_info);
return ret;
@@ -608,7 +608,7 @@ static noinline int
tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer
*eb,
int slot, enum mod_log_op op)
{
- return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
+ return tree_mod_log_insert_key_mask(fs_info, eb, slot, op);
}
static noinline int
@@ -616,13 +616,13 @@ tree_mod_log_insert_key_locked(struct btrfs_fs_info
*fs_info,
struct extent_buffer *eb, int slot,
enum mod_log_op op)
{
- return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
+ return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_ATOMIC);
}
static noinline int
tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
struct extent_buffer *eb, int dst_slot, int src_slot,
-int nr_items, gfp_t flags)
+int nr_items)
{
struct tree_mod_elem *tm;
int ret;
@@ -642,7 +642,7 @@ tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
BUG_ON(ret < 0);
}
- ret = tree_mod_alloc(fs_info, flags, &tm);
+ ret = tree_mod_alloc(fs_info, GFP_ATOMIC, &tm);
if (ret < 0)
goto out;
@@ -679,7 +679,7 @@ __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
struct extent_buffer *eb)
static noinline int
tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
struct extent_buffer *old_root,
-struct extent_buffer *new_root, gfp_t flags,
+struct extent_buffer *new_root,
int log_removal)
{
struct tree_mod_elem *tm;
@@ -691,7 +691,7 @@ tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
if (log_removal)
__tree_mod_log_free_eb(fs_info, old_root);
- ret = tree_mod_alloc(fs_info, flags, &tm);
+ ret = tree_mod_alloc(fs_info, GFP_ATOMIC, &tm);
if (ret < 0)
goto out;
@@ -809,19 +809,18 @@ tree_mod_log_eb_move(struct btrfs_fs_info *fs_info,
struct extent_buffer *dst,
{
int ret;
ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
- nr_items, GFP_NOFS);
+ nr_items);
BUG_ON(ret < 0);
}
static noinline void
tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
- struct extent_buffer *eb, int slot, int atomic)
+ struct extent_buffer *eb, int slot)
{
int ret;
ret = tree_mod_log_insert_key_mask(fs_info, eb, slot,
- MOD_LOG_KEY_REPLACE,
- atomic ? GFP_ATOMIC : GFP_NOFS);
+ MOD_LOG_KEY_REPLACE);
BUG_ON(ret < 0);
}
@@ -843,7 +842,7 @@ tree_mod_log_set_root_pointer(struct btrfs_root *root,
{
int ret;
ret = tree_mod_log_insert_root(root->fs_info, root->node,
- new_root_node, GFP_NOFS, log_removal);
+ new_root_node, log_removal);
