Re: [f2fs-dev] [PATCH] f2fs: fix to tag FIEMAP_EXTENT_DELALLOC in fiemap() for delay allocated extent

From: Chao Yu
Date: Wed Apr 19 2023 - 11:30:52 EST


On 2023/4/13 17:49, Chao Yu wrote:
On 2023/4/12 1:08, Jaegeuk Kim wrote:
On 04/11, Chao Yu wrote:
On 2023/4/11 16:14, Chao Yu wrote:
On 2023/4/11 1:57, Jaegeuk Kim wrote:
On 04/08, Chao Yu wrote:
On 2023/4/8 4:59, Jaegeuk Kim wrote:
This breaks generic/009?

I guess it is as expected?

Please check description of fiemap ioctl manual from [1]:

FIEMAP_EXTENT_UNKNOWN
The location of this extent is currently unknown. This may
indicate the data is stored on an inaccessible volume or that
no storage has been allocated for the file yet.

FIEMAP_EXTENT_DELALLOC
This will also set FIEMAP_EXTENT_UNKNOWN.

Delayed allocation - while there is data for this extent, its
physical location has not been allocated yet.

FIEMAP_EXTENT_UNWRITTEN
Unwritten extent - the extent is allocated but its data has not
been initialized. This indicates the extent’s data will be all
zero if read through the filesystem but the contents are undefined
if read directly from the device.

[1] https://www.kernel.org/doc/html/latest/filesystems/fiemap.html

According to its description, f2fs only support
FIEMAP_EXTENT_{UNKNOWN, DELALLOC}, but not support
FIEMAP_EXTENT_UNWRITTEN.

No, I don't think so.

Jaegeuk,

Could you please check the detailed description of FIEMAP_EXTENT_UNWRITTEN?
The flag indicates two conditions:
1. on-disk blkaddrs were allocated for extent, and the extent is tagged as
unwritten status.
2. data readed on those blocks will be all zero.

Sorry, I mean:

1. on-disk blkaddrs were allocated for extent;
2. extent is tagged as unwritten status, data readed on those blocks will be
all zero.

I was thinking fallocate/pin cases to give zero data. But, we may need to check
the space discarded securely or disk support?

I thought about similar scheme, we can add F2FS_GET_BLOCK_ZERO for fallocate/pin
case to zero data in valid block address, but the problem is how can we distinguish
zero/valid data from valid block address w/o additional unwritten flag of metadata.

e.g.
1. write page [0,511] w/ valid data;
2. fallocate/pin page [512,1023] w/ zero data;
3. fiemap file --- what output should it be?

Thoughts?

Thanks,


Thanks,



Thanks,


So, let's check f2fs' status:
- fallocate only reserve valid block count and set NEW_ADDR in dnode, so
it does not match condition 1)
- pin & fallocate preallocates blkaddrs and set blkaddrs in dnode, but
content on those blkaddrs may contain zero or random data, so it does not
match  condition 2)

Christoph describes this issue in below patch as well, you can check it.
da8c7fecc9c7 ("f2fs: rename F2FS_MAP_UNWRITTEN to F2FS_MAP_DELALLOC")

Am I missing something?

Thanks,



So 009, 092, 094 .. which expects unwritten status from extent will
fail.

How about disabling those testcase?

Thanks,


On 04/05, Chao Yu wrote:
xfstest generic/614 fails to run due below reason:

generic/614 1s ... [not run] test requires delayed allocation buffered writes

The root cause is f2fs tags wrong fiemap flag for delay allocated
extent.

Quoted from fiemap.h:
FIEMAP_EXTENT_UNKNOWN        0x00000002 /* Data location unknown. */
FIEMAP_EXTENT_DELALLOC        0x00000004 /* Location still pending.
                            * Sets EXTENT_UNKNOWN. */
FIEMAP_EXTENT_UNWRITTEN        0x00000800 /* Space allocated, but
                            * no data (i.e. zero). */

FIEMAP_EXTENT_UNWRITTEN means block address is preallocated, but w/o
been written any data, which status f2fs is not supported now, for all
NEW_ADDR block addresses, it means delay allocated blocks, so let's
tag FIEMAP_EXTENT_DELALLOC instead.

Testcase:
xfs_io -f -c 'pwrite 0 64k' /mnt/f2fs/file;
filefrag -v /mnt/f2fs/file

Output:
- Before
Filesystem type is: f2f52010
Fize of /mnt/f2fs/file is 65536 (16 blocks of 4096 bytes)
   ext:     logical_offset:        physical_offset: length:   expected: flags:
     0:        0..      15:          0..        15:     16:             last,unwritten,merged,eof
/mnt/f2fs/file: 1 extent found

After:
Filesystem type is: f2f52010
File size of /mnt/f2fs/file is 65536 (16 blocks of 4096 bytes)
   ext:     logical_offset:        physical_offset: length:   expected: flags:
     0:        0..      15:          0..         0:      0:             last,unknown_loc,delalloc,eof
/mnt/f2fs/file: 1 extent found

Fixes: 7f63eb77af7b ("f2fs: report unwritten area in f2fs_fiemap")
Signed-off-by: Chao Yu <chao@xxxxxxxxxx>
---
   fs/f2fs/data.c | 7 +++++--
   1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 359de650772e..3afc9764743e 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -1995,7 +1995,10 @@ int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
       }
       if (size) {
-        flags |= FIEMAP_EXTENT_MERGED;
+        if (flags & FIEMAP_EXTENT_DELALLOC)
+            phys = 0;
+        else
+            flags |= FIEMAP_EXTENT_MERGED;
           if (IS_ENCRYPTED(inode))
               flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
@@ -2035,7 +2038,7 @@ int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
                   size += blks_to_bytes(inode, 1);
               }
           } else if (map.m_flags & F2FS_MAP_DELALLOC) {
-            flags = FIEMAP_EXTENT_UNWRITTEN;
+            flags = FIEMAP_EXTENT_DELALLOC;
           }
           start_blk += bytes_to_blks(inode, size);
--
2.36.1


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel