2020-05-27 05:08:57

by Sahitya Tummala

[permalink] [raw]
Subject: [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

In case a compressed file is getting overwritten, the current retry
logic doesn't include the current page to be retried now as it sets
the new start index as 0 and new end index as writeback_index - 1.
This causes the corresponding cluster to be uncompressed and written
as normal pages without compression. Fix this by allowing writeback to
be retried for the current page as well (in case of compressed page
getting retried due to index mismatch with cluster index). So that
this cluster can be written compressed in case of overwrite.

Signed-off-by: Sahitya Tummala <[email protected]>
---
fs/f2fs/data.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 4af5fcd..bfd1df4 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
if ((!cycled && !done) || retry) {
cycled = 1;
index = 0;
- end = writeback_index - 1;
+ end = retry ? -1 : writeback_index - 1;
goto retry;
}
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
--
Qualcomm India Private Limited, on behalf of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.


2020-05-28 02:48:21

by Chao Yu

[permalink] [raw]
Subject: Re: [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

On 2020/5/27 10:20, Sahitya Tummala wrote:
> In case a compressed file is getting overwritten, the current retry
> logic doesn't include the current page to be retried now as it sets
> the new start index as 0 and new end index as writeback_index - 1.
> This causes the corresponding cluster to be uncompressed and written
> as normal pages without compression. Fix this by allowing writeback to
> be retried for the current page as well (in case of compressed page
> getting retried due to index mismatch with cluster index). So that
> this cluster can be written compressed in case of overwrite.
>
> Signed-off-by: Sahitya Tummala <[email protected]>
> ---
> fs/f2fs/data.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 4af5fcd..bfd1df4 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> if ((!cycled && !done) || retry) {

IMO, we add retry logic in wrong place, you can see that cycled value is
zero only if wbc->range_cyclic is true, in that case writeback_index is valid.

However if retry is true and wbc->range_cyclic is false, then writeback_index
would be uninitialized variable.

Thoughts?

Thanks,

> cycled = 1;
> index = 0;
> - end = writeback_index - 1;
> + end = retry ? -1 : writeback_index - 1;
> goto retry;
> }
> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
>

2020-05-28 02:58:36

by Chao Yu

[permalink] [raw]
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

On 2020/5/28 10:45, Chao Yu wrote:
> On 2020/5/27 10:20, Sahitya Tummala wrote:
>> In case a compressed file is getting overwritten, the current retry
>> logic doesn't include the current page to be retried now as it sets
>> the new start index as 0 and new end index as writeback_index - 1.
>> This causes the corresponding cluster to be uncompressed and written
>> as normal pages without compression. Fix this by allowing writeback to
>> be retried for the current page as well (in case of compressed page
>> getting retried due to index mismatch with cluster index). So that
>> this cluster can be written compressed in case of overwrite.
>>
>> Signed-off-by: Sahitya Tummala <[email protected]>
>> ---
>> fs/f2fs/data.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>> index 4af5fcd..bfd1df4 100644
>> --- a/fs/f2fs/data.c
>> +++ b/fs/f2fs/data.c
>> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
>> if ((!cycled && !done) || retry) {
>
> IMO, we add retry logic in wrong place, you can see that cycled value is
> zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
>
> However if retry is true and wbc->range_cyclic is false, then writeback_index
> would be uninitialized variable.
>
> Thoughts?
>
> Thanks,
>
>> cycled = 1;
>> index = 0;
>> - end = writeback_index - 1;

BTW, I notice that range_cyclic writeback flow was refactored in below commit,
and skeleton of f2fs.writepages was copied from mm/page-writeback.c::write_cache_pages(),
I guess we need follow that change.

64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock")

Thanks,

>> + end = retry ? -1 : writeback_index - 1;
>> goto retry;
>> }
>> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
>>
>
>
> _______________________________________________
> Linux-f2fs-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> .
>

2020-05-28 19:20:58

by Jaegeuk Kim

[permalink] [raw]
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

On 05/28, Chao Yu wrote:
> On 2020/5/28 10:45, Chao Yu wrote:
> > On 2020/5/27 10:20, Sahitya Tummala wrote:
> >> In case a compressed file is getting overwritten, the current retry
> >> logic doesn't include the current page to be retried now as it sets
> >> the new start index as 0 and new end index as writeback_index - 1.
> >> This causes the corresponding cluster to be uncompressed and written
> >> as normal pages without compression. Fix this by allowing writeback to
> >> be retried for the current page as well (in case of compressed page
> >> getting retried due to index mismatch with cluster index). So that
> >> this cluster can be written compressed in case of overwrite.
> >>
> >> Signed-off-by: Sahitya Tummala <[email protected]>
> >> ---
> >> fs/f2fs/data.c | 2 +-
> >> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> >> index 4af5fcd..bfd1df4 100644
> >> --- a/fs/f2fs/data.c
> >> +++ b/fs/f2fs/data.c
> >> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> >> if ((!cycled && !done) || retry) {
> >
> > IMO, we add retry logic in wrong place, you can see that cycled value is
> > zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
> >
> > However if retry is true and wbc->range_cyclic is false, then writeback_index
> > would be uninitialized variable.
> >
> > Thoughts?
> >
> > Thanks,
> >
> >> cycled = 1;
> >> index = 0;
> >> - end = writeback_index - 1;
>
> BTW, I notice that range_cyclic writeback flow was refactored in below commit,
> and skeleton of f2fs.writepages was copied from mm/page-writeback.c::write_cache_pages(),
> I guess we need follow that change.
>
> 64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock")

Is that something like this?

---
fs/f2fs/data.c | 11 ++---------
1 file changed, 2 insertions(+), 9 deletions(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 48a622b95b76e..28fcdf0d4dcb9 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2861,7 +2861,6 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
pgoff_t index;
pgoff_t end; /* Inclusive */
pgoff_t done_index;
- int cycled;
int range_whole = 0;
xa_mark_t tag;
int nwritten = 0;
@@ -2879,17 +2878,12 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
if (wbc->range_cyclic) {
writeback_index = mapping->writeback_index; /* prev offset */
index = writeback_index;
- if (index == 0)
- cycled = 1;
- else
- cycled = 0;
end = -1;
} else {
index = wbc->range_start >> PAGE_SHIFT;
end = wbc->range_end >> PAGE_SHIFT;
if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
range_whole = 1;
- cycled = 1; /* ignore range_cyclic tests */
}
if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
tag = PAGECACHE_TAG_TOWRITE;
@@ -3054,10 +3048,9 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
}
}
#endif
- if ((!cycled && !done) || retry) {
- cycled = 1;
+ if (retry) {
index = 0;
- end = writeback_index - 1;
+ end = -1;
goto retry;
}
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
--
2.27.0.rc0.183.gde8f92d652-goog

2020-06-01 09:19:06

by Sahitya Tummala

[permalink] [raw]
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

Hi Chao,

Can you please help review below diff given by Jaegeuk?
If it looks good, I can send a v2.

Thanks,

On Thu, May 28, 2020 at 12:18:39PM -0700, Jaegeuk Kim wrote:
> On 05/28, Chao Yu wrote:
> > On 2020/5/28 10:45, Chao Yu wrote:
> > > On 2020/5/27 10:20, Sahitya Tummala wrote:
> > >> In case a compressed file is getting overwritten, the current retry
> > >> logic doesn't include the current page to be retried now as it sets
> > >> the new start index as 0 and new end index as writeback_index - 1.
> > >> This causes the corresponding cluster to be uncompressed and written
> > >> as normal pages without compression. Fix this by allowing writeback to
> > >> be retried for the current page as well (in case of compressed page
> > >> getting retried due to index mismatch with cluster index). So that
> > >> this cluster can be written compressed in case of overwrite.
> > >>
> > >> Signed-off-by: Sahitya Tummala <[email protected]>
> > >> ---
> > >> fs/f2fs/data.c | 2 +-
> > >> 1 file changed, 1 insertion(+), 1 deletion(-)
> > >>
> > >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > >> index 4af5fcd..bfd1df4 100644
> > >> --- a/fs/f2fs/data.c
> > >> +++ b/fs/f2fs/data.c
> > >> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> > >> if ((!cycled && !done) || retry) {
> > >
> > > IMO, we add retry logic in wrong place, you can see that cycled value is
> > > zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
> > >
> > > However if retry is true and wbc->range_cyclic is false, then writeback_index
> > > would be uninitialized variable.
> > >
> > > Thoughts?
> > >
> > > Thanks,
> > >
> > >> cycled = 1;
> > >> index = 0;
> > >> - end = writeback_index - 1;
> >
> > BTW, I notice that range_cyclic writeback flow was refactored in below commit,
> > and skeleton of f2fs.writepages was copied from mm/page-writeback.c::write_cache_pages(),
> > I guess we need follow that change.
> >
> > 64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock")
>
> Is that something like this?
>
> ---
> fs/f2fs/data.c | 11 ++---------
> 1 file changed, 2 insertions(+), 9 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 48a622b95b76e..28fcdf0d4dcb9 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -2861,7 +2861,6 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> pgoff_t index;
> pgoff_t end; /* Inclusive */
> pgoff_t done_index;
> - int cycled;
> int range_whole = 0;
> xa_mark_t tag;
> int nwritten = 0;
> @@ -2879,17 +2878,12 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> if (wbc->range_cyclic) {
> writeback_index = mapping->writeback_index; /* prev offset */
> index = writeback_index;
> - if (index == 0)
> - cycled = 1;
> - else
> - cycled = 0;
> end = -1;
> } else {
> index = wbc->range_start >> PAGE_SHIFT;
> end = wbc->range_end >> PAGE_SHIFT;
> if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
> range_whole = 1;
> - cycled = 1; /* ignore range_cyclic tests */
> }
> if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
> tag = PAGECACHE_TAG_TOWRITE;
> @@ -3054,10 +3048,9 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> }
> }
> #endif
> - if ((!cycled && !done) || retry) {
> - cycled = 1;
> + if (retry) {
> index = 0;
> - end = writeback_index - 1;
> + end = -1;
> goto retry;
> }
> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
> --
> 2.27.0.rc0.183.gde8f92d652-goog
>

--
--
Sent by a consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

2020-06-01 09:23:37

by Sahitya Tummala

[permalink] [raw]
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

Hi Chao,

Can you please help review the diff given by Jaegeuk below?
If it looks good, I can post a v2.

Thanks,

On Thu, May 28, 2020 at 12:18:39PM -0700, Jaegeuk Kim wrote:
> On 05/28, Chao Yu wrote:
> > On 2020/5/28 10:45, Chao Yu wrote:
> > > On 2020/5/27 10:20, Sahitya Tummala wrote:
> > >> In case a compressed file is getting overwritten, the current retry
> > >> logic doesn't include the current page to be retried now as it sets
> > >> the new start index as 0 and new end index as writeback_index - 1.
> > >> This causes the corresponding cluster to be uncompressed and written
> > >> as normal pages without compression. Fix this by allowing writeback to
> > >> be retried for the current page as well (in case of compressed page
> > >> getting retried due to index mismatch with cluster index). So that
> > >> this cluster can be written compressed in case of overwrite.
> > >>
> > >> Signed-off-by: Sahitya Tummala <[email protected]>
> > >> ---
> > >> fs/f2fs/data.c | 2 +-
> > >> 1 file changed, 1 insertion(+), 1 deletion(-)
> > >>
> > >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > >> index 4af5fcd..bfd1df4 100644
> > >> --- a/fs/f2fs/data.c
> > >> +++ b/fs/f2fs/data.c
> > >> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> > >> if ((!cycled && !done) || retry) {
> > >
> > > IMO, we add retry logic in wrong place, you can see that cycled value is
> > > zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
> > >
> > > However if retry is true and wbc->range_cyclic is false, then writeback_index
> > > would be uninitialized variable.
> > >
> > > Thoughts?
> > >
> > > Thanks,
> > >
> > >> cycled = 1;
> > >> index = 0;
> > >> - end = writeback_index - 1;
> >
> > BTW, I notice that range_cyclic writeback flow was refactored in below commit,
> > and skeleton of f2fs.writepages was copied from mm/page-writeback.c::write_cache_pages(),
> > I guess we need follow that change.
> >
> > 64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock")
>
> Is that something like this?
>
> ---
> fs/f2fs/data.c | 11 ++---------
> 1 file changed, 2 insertions(+), 9 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 48a622b95b76e..28fcdf0d4dcb9 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -2861,7 +2861,6 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> pgoff_t index;
> pgoff_t end; /* Inclusive */
> pgoff_t done_index;
> - int cycled;
> int range_whole = 0;
> xa_mark_t tag;
> int nwritten = 0;
> @@ -2879,17 +2878,12 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> if (wbc->range_cyclic) {
> writeback_index = mapping->writeback_index; /* prev offset */
> index = writeback_index;
> - if (index == 0)
> - cycled = 1;
> - else
> - cycled = 0;
> end = -1;
> } else {
> index = wbc->range_start >> PAGE_SHIFT;
> end = wbc->range_end >> PAGE_SHIFT;
> if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
> range_whole = 1;
> - cycled = 1; /* ignore range_cyclic tests */
> }
> if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
> tag = PAGECACHE_TAG_TOWRITE;
> @@ -3054,10 +3048,9 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
> }
> }
> #endif
> - if ((!cycled && !done) || retry) {
> - cycled = 1;
> + if (retry) {
> index = 0;
> - end = writeback_index - 1;
> + end = -1;
> goto retry;
> }
> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
> --
> 2.27.0.rc0.183.gde8f92d652-goog
>

--
--
Sent by a consultant of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

2020-06-01 11:46:08

by Chao Yu

[permalink] [raw]
Subject: Re: [f2fs-dev] [PATCH] f2fs: fix retry logic in f2fs_write_cache_pages()

On 2020/6/1 17:20, Sahitya Tummala wrote:
> Hi Chao,
>
> Can you please help review the diff given by Jaegeuk below?
> If it looks good, I can post a v2.

Oops, I missed to review that patch...

>
> Thanks,
>
> On Thu, May 28, 2020 at 12:18:39PM -0700, Jaegeuk Kim wrote:
>> On 05/28, Chao Yu wrote:
>>> On 2020/5/28 10:45, Chao Yu wrote:
>>>> On 2020/5/27 10:20, Sahitya Tummala wrote:
>>>>> In case a compressed file is getting overwritten, the current retry
>>>>> logic doesn't include the current page to be retried now as it sets
>>>>> the new start index as 0 and new end index as writeback_index - 1.
>>>>> This causes the corresponding cluster to be uncompressed and written
>>>>> as normal pages without compression. Fix this by allowing writeback to
>>>>> be retried for the current page as well (in case of compressed page
>>>>> getting retried due to index mismatch with cluster index). So that
>>>>> this cluster can be written compressed in case of overwrite.
>>>>>
>>>>> Signed-off-by: Sahitya Tummala <[email protected]>
>>>>> ---
>>>>> fs/f2fs/data.c | 2 +-
>>>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>>>>> index 4af5fcd..bfd1df4 100644
>>>>> --- a/fs/f2fs/data.c
>>>>> +++ b/fs/f2fs/data.c
>>>>> @@ -3024,7 +3024,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
>>>>> if ((!cycled && !done) || retry) {
>>>>
>>>> IMO, we add retry logic in wrong place, you can see that cycled value is
>>>> zero only if wbc->range_cyclic is true, in that case writeback_index is valid.
>>>>
>>>> However if retry is true and wbc->range_cyclic is false, then writeback_index
>>>> would be uninitialized variable.
>>>>
>>>> Thoughts?
>>>>
>>>> Thanks,
>>>>
>>>>> cycled = 1;
>>>>> index = 0;
>>>>> - end = writeback_index - 1;
>>>
>>> BTW, I notice that range_cyclic writeback flow was refactored in below commit,
>>> and skeleton of f2fs.writepages was copied from mm/page-writeback.c::write_cache_pages(),
>>> I guess we need follow that change.
>>>
>>> 64081362e8ff ("mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock")
>>
>> Is that something like this?
>>
>> ---
>> fs/f2fs/data.c | 11 ++---------
>> 1 file changed, 2 insertions(+), 9 deletions(-)
>>
>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>> index 48a622b95b76e..28fcdf0d4dcb9 100644
>> --- a/fs/f2fs/data.c
>> +++ b/fs/f2fs/data.c
>> @@ -2861,7 +2861,6 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
>> pgoff_t index;
>> pgoff_t end; /* Inclusive */
>> pgoff_t done_index;
>> - int cycled;
>> int range_whole = 0;
>> xa_mark_t tag;
>> int nwritten = 0;
>> @@ -2879,17 +2878,12 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
>> if (wbc->range_cyclic) {
>> writeback_index = mapping->writeback_index; /* prev offset */
>> index = writeback_index;
>> - if (index == 0)
>> - cycled = 1;
>> - else
>> - cycled = 0;
>> end = -1;
>> } else {
>> index = wbc->range_start >> PAGE_SHIFT;
>> end = wbc->range_end >> PAGE_SHIFT;
>> if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
>> range_whole = 1;
>> - cycled = 1; /* ignore range_cyclic tests */
>> }
>> if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
>> tag = PAGECACHE_TAG_TOWRITE;
>> @@ -3054,10 +3048,9 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
>> }
>> }
>> #endif
>> - if ((!cycled && !done) || retry) {
>> - cycled = 1;
>> + if (retry) {
>> index = 0;
>> - end = writeback_index - 1;
>> + end = -1;
>> goto retry;
>> }

+ if (wbc->range_cyclic && !done)
+ done_index = 0;

Thanks,

>> if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
>> --
>> 2.27.0.rc0.183.gde8f92d652-goog
>>
>