| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <55FC575C.40205@harvyl.se>
Date: Fri, 18 Sep 2015 20:26:36 +0200
From: Johan Harvyl <johan@...vyl.se>
To: Andreas Dilger <adilger@...ger.ca>, Theodore Ts'o <tytso@....edu>
Cc: "linux-ext4@...r.kernel.org" <linux-ext4@...r.kernel.org>
Subject: Re: resize2fs: Should never happen: resize inode corrupt! - lost key
inodes
Hi,
I should have thought of that, but unfortunately it will not allow me to
do so.
# MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf
/root/elatest/out/sbin/mkfs.ext4 /dev/md0 -m 0 -b 1024 -O 64bit 3906887168k
mke2fs 1.43-WIP (18-May-2015)
Warning: specified blocksize 1024 is less than device physical
sectorsize 4096
/dev/md0: Cannot create filesystem with requested number of inodes while
setting up superblock
#
Instead, I stuck to the 1k blocks and divided by four again...
# MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf
/root/elatest/out/sbin/mkfs.ext4 /dev/md0 -m 0 -b 1024 -i 262144 -O
64bit 976721792k
mke2fs 1.43-WIP (18-May-2015)
Warning: specified blocksize 1024 is less than device physical
sectorsize 4096
Creating filesystem with 976721792 1k blocks and 3815328 inodes
Filesystem UUID: 2626eb2a-0691-48b2-a64c-2f4802437166
Superblock backups stored on blocks:
8193, 24577, 40961, 57345, 73729, 204801, 221185, 401409, 663553,
1024001, 1990657, 2809857, 5120001, 5971969, 17915905, 19668993,
25600001, 53747713, 128000001, 137682945, 161243137, 483729409,
640000001, 963780609
Allocating group tables: done
Writing inode tables: done
Creating journal (32768 blocks): done
Writing superblocks and filesystem accounting information: done
...e2fsck is ok here...
Now for a proportional resize, i.e. + 25 %:
# MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf
/root/elatest/out/sbin/resize2fs -p /dev/md0 1220902240k
resize2fs 1.43-WIP (18-May-2015)
Resizing the filesystem on /dev/md0 to 1220902240 (1k) blocks.
Begin pass 2 (max = 14)
Relocating blocks XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Begin pass 3 (max = 119229)
Scanning inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Begin pass 5 (max = 16)
Moving inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
The filesystem on /dev/md0 is now 1220902240 (1k) blocks long.
Already after the first resize the fs seems much more corrupted and in a
different way than my original report.
Below are a few of the errors, there are many many pages of them.
This appears to be completely reproducible. I'll try to shrink things
further. Using 4k blocks instead of 1k it does not reproduce.
-johan
# MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf
/root/elatest/out/sbin/e2fsck -fnv /dev/md0 2>&1 |less
e2fsck 1.43-WIP (18-May-2015)
Pass 1: Checking inodes, blocks, and sizes
Inode 13 passes checks, but checksum does not match inode. Fix? no
Deleted inode 14 has zero dtime. Fix? no
...
Inode 1024 passes checks, but checksum does not match inode. Fix? no
Inode 1437 seems to contain garbage. Clear? no
Inode 1437 is in use, but has dtime set. Fix? no
Inode 1437 has a extra size (1656) which is invalid
Fix? no
Inode 1437 has INDEX_FL flag set but is not a directory.
Clear HTree index? no
...
Illegal block #11 (2674298790) in inode 1442. IGNORED.
Illegal block number passed to ext2fs_test_block_bitmap #1906002301 for
metadata block map
Too many illegal blocks in inode 1442.
Clear inode? no
Suppress messages? no
Illegal indirect block (1906002301) in inode 1442. IGNORED.
Illegal block number passed to ext2fs_test_block_bitmap #3316469983 for
metadata block map
On 2015-09-17 03:21, Andreas Dilger wrote:
> If you add "-b 1024" to the mke2fs command line to use 1KB instead of 4KB blocks, and reduce the sizes by a factor of 4 does the problem still happen? That would make it easier for someone else to test, since it would only need a 4-5TB disk instead of a 19Tb array.
>
> Cheers, Andreas
>
>> On Sep 15, 2015, at 11:55, Johan Harvyl <johan@...vyl.se> wrote:
>>
>> I have now been able to reproduce the issue that resize2fs corrupts at least the root, resize and journal
>> inodes with versions 1.42.13 and the more recent commit 956b0f1 of e2fsprogs.
>>
>> Note that older versions of e2fsprogs need *not* be involved, 1.42.13 and newer also have issues.
>>
>> Please advice on things I can try to narrow down the root cause of what has to be an e2fsprogs bug. In
>> particular it would be very useful to reproduce it faster, running through the mkfs and two resize steps
>> takes around ten minutes so iterative testing is a slow and I do not really have much of clue what steps
>> would be more likely to overwrite the inodes.
>>
>> At some point I would like to return this array to service but I am not really comfortable creating a
>> new ext4 filesystem on it without first understanding how it can become corrupted without even
>> mounting the file system.
>>
>> For 1.42.13:
>> # mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit 15627548672k
>> # resize2fs -p /dev/md0 19534435840k
>> # resize2fs -p /dev/md0
>> # e2fsck -fn /dev/md0
>> e2fsck 1.42.13 (17-May-2015)
>> ext2fs_check_desc: Corrupt group descriptor: bad block for inode table
>> e2fsck: Group descriptors look bad... trying backup blocks...
>> Superblock has an invalid journal (inode 8).
>> Clear? no
>>
>> e2fsck: Illegal inode number while checking ext3 journal for /dev/md0
>>
>> /dev/md0: ********** WARNING: Filesystem still has errors **********
>>
>>
>> or for 956b0f1:
>> # MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf /root/elatest/out/sbin/mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit 15627548672k
>> # MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf /root/elatest/out/sbin/resize2fs -p /dev/md0 19534435840k
>> # MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf /root/elatest/out/sbin/resize2fs -p /dev/md0
>> # MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf /root/elatest/out/sbin/e2fsck -fn /dev/md0
>> e2fsck 1.43-WIP (18-May-2015)
>> ext2fs_open2: Superblock checksum does not match superblock
>> /root/elatest/out/sbin/e2fsck: Superblock invalid, trying backup blocks...
>> Superblock has an invalid journal (inode 8).
>> Clear? no
>>
>> /root/elatest/out/sbin/e2fsck: Illegal inode number while checking ext3 journal for /dev/md0
>>
>> /dev/md0: ********** WARNING: Filesystem still has errors **********
>>
>> # MKE2FS_CONFIG=/root/elatest/out/etc/mke2fs.conf /root/elatest/out/sbin/debugfs -c /dev/md0
>> debugfs 1.43-WIP (18-May-2015)
>> /dev/md0: Superblock checksum does not match superblock while opening filesystem
>> debugfs: stat <2>
>> stat: Filesystem not open
>>
>> # debugfs -c /dev/md0
>> debugfs 1.42.13 (17-May-2015)
>> /dev/md0: catastrophic mode - not reading inode or group bitmaps
>> debugfs: stat <2>
>> Inode: 2 Type: bad type Mode: 0004 Flags: 0x1
>> Generation: 1 Version: 0x00000001
>> User: 9440 Group: 0 Size: 618659860
>> File ACL: 1 Directory ACL: 0
>> Links: 0 Blockcount: 724107776
>> Fragment: Address: 0 Number: 0 Size: 0
>> ctime: 0x02008000 -- Sun Jan 24 18:46:40 1971
>> atime: 0x24e000a0 -- Wed Aug 9 12:00:00 1989
>> mtime: 0x00030000 -- Sat Jan 3 07:36:48 1970
>> Size of extra inode fields: 6
>> BLOCKS:
>> (0):1, (6):618659845 .... and it goes on...
>>
>>> On 2015-09-14 23:35, Johan Harvyl wrote:
>>> In an attempt to further isolate what versions of e2fsprogs, at a commit level, that are
>>> needed to reproduce the bad behavior I tried my own step-by-step, initially with a much
>>> higher -i 16777216 to mkfs.ext4 in the hope that fewer inodes would make all the
>>> operations run faster.
>>>
>>> When I was unable to reproduce with -i 16777216 instead, I switched back to exactly
>>> what I reproduced with the first time, and I *still* did not get the "Should never happen:
>>> resize inode corrupt!".
>>>
>>> The only reasonable explanation I can come up with to this is that something is not being
>>> initialized properly that resize2fs expects to be initialized. I have no indications of any
>>> issues with any hardware or the underlying md block.
>>>
>>> What I did however notice is that I can have the same kind of filesystem corruption
>>> *without* seeing the "Should never happen: resize inode corrupt!" message using the
>>> following sequence, and this *is* reproducible one time after another:
>>>
>>> # MKE2FS_CONFIG=/root/e10/out/etc/mke2fs.conf /root/e10/out/sbin/mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit 15627548672k
>>> # e2fsck -fy /dev/md0 (using 1.42.13)
>>> # resize2fs -p /dev/md0 19534435840k (using 1.42.13)
>>> # resize2fs -p /dev/md0 (using 1.42.13)
>>> # e2fsck -fn /dev/md0
>>> e2fsck 1.42.13 (17-May-2015)
>>> ext2fs_check_desc: Corrupt group descriptor: bad block for inode table
>>> e2fsck: Group descriptors look bad... trying backup blocks...
>>> Superblock has an invalid journal (inode 8).
>>> Clear? no
>>>
>>> e2fsck: Illegal inode number while checking ext3 journal for /dev/md0
>>>
>>> At this point the root inode is also bad and this fails:
>>> # mount /dev/md0 /mnt/loop -o ro,noload
>>> mount: mount /dev/md0 on /mnt/loop failed: Stale file handle
>>> [3766493.732188] EXT4-fs (md0): get root inode failed
>>> [3766493.732190] EXT4-fs (md0): mount failed
>>>
>>> Note that only versions 1.42.10 and 1.42.13 are involved now, 1.42.12 is not needed.
>>>
>>> Kernel is the debian:
>>> ii linux-image-4.0.0-2-amd64 4.0.8-2 amd64 Linux 4.0 for 64-bit PCs
>>>
>>> For the record I also tried a more recent e2fsprogs for the resize (instead of 1.42.13),
>>> locally built from:
>>> 956b0f1 Merge branch 'maint' into next
>>> and I could still reproduce it on the first attempt.
>>>
>>> More verbose logs follows.
>>>
>>> Does anyone else have some kind of testbed to test the same sequence of commands?
>>>
>>> ===
>>>
>>> # MKE2FS_CONFIG=/root/e10/out/etc/mke2fs.conf /root/e10/out/sbin/mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit 15627548672k
>>> mke2fs 1.42.10 (18-May-2014)
>>> /dev/md0 contains a ext4 file system
>>> last mounted on Sun Sep 13 22:19:28 2015
>>> Proceed anyway? (y,n) y
>>> Creating filesystem with 3906887168 4k blocks and 61045248 inodes
>>> Filesystem UUID: e263356e-4fe4-4e9b-bd0c-8edc2c411735
>>> Superblock backups stored on blocks:
>>> 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
>>> 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
>>> 102400000, 214990848, 512000000, 550731776, 644972544, 1934917632,
>>> 2560000000, 3855122432
>>>
>>> Allocating group tables: done
>>> Writing inode tables: done
>>> Creating journal (32768 blocks): done
>>> Writing superblocks and filesystem accounting information: done
>>>
>>> # e2fsck -fy /dev/md0
>>> e2fsck 1.42.13 (17-May-2015)
>>> Pass 1: Checking inodes, blocks, and sizes
>>> Pass 2: Checking directory structure
>>> Pass 3: Checking directory connectivity
>>> Pass 4: Checking reference counts
>>> Pass 5: Checking group summary information
>>> Free blocks count wrong (512088558484167, counted=3902749383).
>>> Fix? yes
>>>
>>>
>>> /dev/md0: ***** FILE SYSTEM WAS MODIFIED *****
>>> /dev/md0: 11/61045248 files (0.0% non-contiguous), 4137785/3906887168 blocks
>>>
>>> # resize2fs -p /dev/md0 19534435840k
>>> resize2fs 1.42.13 (17-May-2015)
>>> Resizing the filesystem on /dev/md0 to 4883608960 (4k) blocks.
>>> Begin pass 2 (max = 6)
>>> Relocating blocks XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> Begin pass 3 (max = 119229)
>>> Scanning inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> Begin pass 5 (max = 8)
>>> Moving inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> The filesystem on /dev/md0 is now 4883608960 (4k) blocks long.
>>>
>>> # resize2fs -p /dev/md0
>>> resize2fs 1.42.13 (17-May-2015)
>>> Resizing the filesystem on /dev/md0 to 5860330752 (4k) blocks.
>>> Begin pass 2 (max = 6)
>>> Relocating blocks XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> Begin pass 3 (max = 149036)
>>> Scanning inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> Begin pass 5 (max = 14)
>>> Moving inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>> The filesystem on /dev/md0 is now 5860330752 (4k) blocks long.
>>>
>>> # e2fsck -fn /dev/md0
>>> e2fsck 1.42.13 (17-May-2015)
>>> ext2fs_check_desc: Corrupt group descriptor: bad block for inode table
>>> e2fsck: Group descriptors look bad... trying backup blocks...
>>> Superblock has an invalid journal (inode 8).
>>> Clear? no
>>>
>>> e2fsck: Illegal inode number while checking ext3 journal for /dev/md0
>>>
>>>> On 2015-09-12 12:27, Johan Harvyl wrote:
>>>> Hi,
>>>>
>>>> I have now evacuated the data on the filesystem and I *did* manage to recreate the
>>>> "Should never happen: resize inode corrupt!" using the versions of e2fsprogs I believe I was using at the time.
>>>>
>>>> The vast majority of the data that I was able to checksum was ok.
>>>>
>>>> For me I guess the way forward should be to recreate the fs with 1.42.13 and stick to online resize
>>>> from now on, correct?
>>>>
>>>> Are there any feature flags that I should not use when expanding file systems or any that I must use?
>>>>
>>>> -johan
>>>>
>>>>
>>>> Here is a step by step of what I did to reproduce
>>>>
>>>> I have built the following two versions of e2fsprogs (configure, make, make install, nothing else):
>>>> 421d693 (HEAD) libext2fs: fix potential buffer overflow in closefs()
>>>> 6a3741a (tag: v1.42.12) Update release notes, etc. for final 1.42.12 release
>>>>
>>>> 9779e29 (HEAD, tag: v1.42.10) Update release notes, etc. for final 1.42.10 release
>>>>
>>>> ===
>>>>
>>>> First build the fs with 1.42.10 with the exact number of blocks I originally had.
>>>>
>>>> # MKE2FS_CONFIG=/root/e10/out/etc/mke2fs.conf /root/e10/out/sbin/mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit 15627548672k
>>>> mke2fs 1.42.10 (18-May-2014)
>>>> /dev/md0 contains a ext4 file system
>>>> created on Sat Sep 12 11:23:02 2015
>>>> Proceed anyway? (y,n) y
>>>> Creating filesystem with 3906887168 4k blocks and 61045248 inodes
>>>> Filesystem UUID: d00e9e59-3756-4e59-9539-bc00fe2446b5
>>>> Superblock backups stored on blocks:
>>>> 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
>>>> 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
>>>> 102400000, 214990848, 512000000, 550731776, 644972544, 1934917632,
>>>> 2560000000, 3855122432
>>>>
>>>> Allocating group tables: done
>>>> Writing inode tables: done
>>>> Creating journal (32768 blocks): done
>>>> Writing superblocks and filesystem accounting information: done
>>>>
>>>> From dumpe2fs I observe:
>>>> 1) the fs features match what I had on my broken fs
>>>> 2) the number of free blocks is 512088558484167 which is clearly wrong.
>>>>
>>>> # e2fsck -fnv /dev/md0
>>>> e2fsck 1.42.13 (17-May-2015)
>>>> Pass 1: Checking inodes, blocks, and sizes
>>>> Pass 2: Checking directory structure
>>>> Pass 3: Checking directory connectivity
>>>> Pass 4: Checking reference counts
>>>> Pass 5: Checking group summary information
>>>> Free blocks count wrong (512088558484167, counted=3902749383).
>>>> Fix? no
>>>>
>>>> So the initial fs created by 1.42.10 appear to be bad.
>>>>
>>>> Filesystem volume name: <none>
>>>> Last mounted on: <not available>
>>>> Filesystem UUID: d00e9e59-3756-4e59-9539-bc00fe2446b5
>>>> Filesystem magic number: 0xEF53
>>>> Filesystem revision #: 1 (dynamic)
>>>> Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent 64bit flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize
>>>> Filesystem flags: signed_directory_hash
>>>> Default mount options: user_xattr acl
>>>> Filesystem state: clean
>>>> Errors behavior: Continue
>>>> Filesystem OS type: Linux
>>>> Inode count: 61045248
>>>> Block count: 3906887168
>>>> Reserved block count: 0
>>>> Free blocks: 512088558484167
>>>> Free inodes: 61045237
>>>> First block: 0
>>>> Block size: 4096
>>>> Fragment size: 4096
>>>> Group descriptor size: 64
>>>> Reserved GDT blocks: 185
>>>> Blocks per group: 32768
>>>> Fragments per group: 32768
>>>> Inodes per group: 512
>>>> Inode blocks per group: 32
>>>> Flex block group size: 16
>>>> Filesystem created: Sat Sep 12 11:27:55 2015
>>>> Last mount time: n/a
>>>> Last write time: Sat Sep 12 11:27:55 2015
>>>> Mount count: 0
>>>> Maximum mount count: -1
>>>> Last checked: Sat Sep 12 11:27:55 2015
>>>> Check interval: 0 (<none>)
>>>> Lifetime writes: 158 MB
>>>> Reserved blocks uid: 0 (user root)
>>>> Reserved blocks gid: 0 (group root)
>>>> First inode: 11
>>>> Inode size: 256
>>>> Required extra isize: 28
>>>> Desired extra isize: 28
>>>> Journal inode: 8
>>>> Default directory hash: half_md4
>>>> Directory Hash Seed: f252a723-7016-43d1-97f8-579062a215e1
>>>> Journal backup: inode blocks
>>>> Journal features: (none)
>>>> Journal size: 128M
>>>> Journal length: 32768
>>>> Journal sequence: 0x00000001
>>>> Journal start: 0
>>>>
>>>>
>>>>
>>>> The next step is resizing + 4 TB with 1.42.12.
>>>> # MKE2FS_CONFIG=/root/e12/out/etc/mke2fs.conf /root/e12/out/sbin/resize2fs -p /dev/md0 19534435840k
>>>> resize2fs 1.42.12 (29-Aug-2014)
>>>> <and nothing more>
>>>> It did *not* print the "Resizing the filesystem on /dev/md0 to 4883608960 (4k) blocks." that it should have.
>>>>
>>>> I let it run for 90+ minutes sampling CPU and IO usage with iotop from time to time. It was using more or less 100% CPU and no visible io.
>>>>
>>>> So, I let e2fsck fix the free block count and re-did the resize:
>>>> # e2fsck -f /dev/md0
>>>> e2fsck 1.42.13 (17-May-2015)
>>>> Pass 1: Checking inodes, blocks, and sizes
>>>> Pass 2: Checking directory structure
>>>> Pass 3: Checking directory connectivity
>>>> Pass 4: Checking reference counts
>>>> Pass 5: Checking group summary information
>>>> Free blocks count wrong (512088558484167, counted=3902749383).
>>>> Fix<y>? yes
>>>>
>>>> /dev/md0: ***** FILE SYSTEM WAS MODIFIED *****
>>>> /dev/md0: 11/61045248 files (0.0% non-contiguous), 4137785/3906887168 blocks
>>>>
>>>> # MKE2FS_CONFIG=/root/e12/out/etc/mke2fs.conf /root/e12/out/sbin/resize2fs -p /dev/md0 19534435840k
>>>> resize2fs 1.42.12 (29-Aug-2014)
>>>> Resizing the filesystem on /dev/md0 to 4883608960 (4k) blocks.
>>>> Begin pass 2 (max = 6)
>>>> Relocating blocks XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> Begin pass 3 (max = 119229)
>>>> Scanning inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> Begin pass 5 (max = 8)
>>>> Moving inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> The filesystem on /dev/md0 is now 4883608960 (4k) blocks long.
>>>>
>>>> dumpe2fs 1.42.13 (17-May-2015)
>>>> Filesystem volume name: <none>
>>>> Last mounted on: <not available>
>>>> Filesystem UUID: 159d3929-1842-4f8d-907f-7509c16f06df
>>>> Filesystem magic number: 0xEF53
>>>> Filesystem revision #: 1 (dynamic)
>>>> Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent 64bit flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize
>>>> Filesystem flags: signed_directory_hash
>>>> Default mount options: user_xattr acl
>>>> Filesystem state: clean
>>>> Errors behavior: Continue
>>>> Filesystem OS type: Linux
>>>> Inode count: 76306432
>>>> Block count: 4883608960
>>>> Reserved block count: 0
>>>> Free blocks: 4878450712
>>>> Free inodes: 76306421
>>>> First block: 0
>>>> Block size: 4096
>>>> Fragment size: 4096
>>>> Group descriptor size: 64
>>>> Blocks per group: 32768
>>>> Fragments per group: 32768
>>>> Inodes per group: 512
>>>> Inode blocks per group: 32
>>>> RAID stride: 32752
>>>> Flex block group size: 16
>>>> Filesystem created: Sat Sep 12 11:41:10 2015
>>>> Last mount time: n/a
>>>> Last write time: Sat Sep 12 11:56:20 2015
>>>> Mount count: 0
>>>> Maximum mount count: -1
>>>> Last checked: Sat Sep 12 11:49:28 2015
>>>> Check interval: 0 (<none>)
>>>> Lifetime writes: 279 MB
>>>> Reserved blocks uid: 0 (user root)
>>>> Reserved blocks gid: 0 (group root)
>>>> First inode: 11
>>>> Inode size: 256
>>>> Required extra isize: 28
>>>> Desired extra isize: 28
>>>> Journal inode: 8
>>>> Default directory hash: half_md4
>>>> Directory Hash Seed: feeea566-bb38-44c6-a4d5-f97aa78001d4
>>>> Journal backup: inode blocks
>>>> Journal features: (none)
>>>> Journal size: 128M
>>>> Journal length: 32768
>>>> Journal sequence: 0x00000001
>>>> Journal start: 0
>>>>
>>>> Looking good so far, and now for the final resize to 24 TB using 1.42.13:
>>>> # resize2fs -p /dev/md0
>>>> resize2fs 1.42.13 (17-May-2015)
>>>> Resizing the filesystem on /dev/md0 to 5860330752 (4k) blocks.
>>>> Begin pass 2 (max = 6)
>>>> Relocating blocks XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> Begin pass 3 (max = 149036)
>>>> Scanning inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> Begin pass 5 (max = 14)
>>>> Moving inode table XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
>>>> Should never happen: resize inode corrupt!
>>>>
>>>> # dumpe2fs -h /dev/md0
>>>> dumpe2fs 1.42.13 (17-May-2015)
>>>> Filesystem volume name: <none>
>>>> Last mounted on: <not available>
>>>> Filesystem UUID: 159d3929-1842-4f8d-907f-7509c16f06df
>>>> Filesystem magic number: 0xEF53
>>>> Filesystem revision #: 1 (dynamic)
>>>> Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent 64bit flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize
>>>> Filesystem flags: signed_directory_hash
>>>> Default mount options: user_xattr acl
>>>> Filesystem state: clean with errors
>>>> Errors behavior: Continue
>>>> Filesystem OS type: Linux
>>>> Inode count: 91568128
>>>> Block count: 5860330752
>>>> Reserved block count: 0
>>>> Free blocks: 5853069550
>>>> Free inodes: 91568117
>>>> First block: 0
>>>> Block size: 4096
>>>> Fragment size: 4096
>>>> Group descriptor size: 64
>>>> Blocks per group: 32768
>>>> Fragments per group: 32768
>>>> Inodes per group: 512
>>>> Inode blocks per group: 32
>>>> RAID stride: 32752
>>>> Flex block group size: 16
>>>> Filesystem created: Sat Sep 12 11:41:10 2015
>>>> Last mount time: n/a
>>>> Last write time: Sat Sep 12 12:03:55 2015
>>>> Mount count: 0
>>>> Maximum mount count: -1
>>>> Last checked: Sat Sep 12 11:49:28 2015
>>>> Check interval: 0 (<none>)
>>>> Lifetime writes: 279 MB
>>>> Reserved blocks uid: 0 (user root)
>>>> Reserved blocks gid: 0 (group root)
>>>> First inode: 11
>>>> Inode size: 256
>>>> Required extra isize: 28
>>>> Desired extra isize: 28
>>>> Journal inode: 8
>>>> Default directory hash: half_md4
>>>> Directory Hash Seed: feeea566-bb38-44c6-a4d5-f97aa78001d4
>>>> Journal backup: inode blocks
>>>> Journal superblock magic number invalid!
>>>>
>>>>
>>>>> On 2015-09-04 00:16, Johan Harvyl wrote:
>>>>> Hello again,
>>>>>
>>>>> I finally got around to dig some more into this and made what I consider some good progress as I am now able to mount the filesystem read-only so I thought I would update this thread a bit.
>>>>>
>>>>> Short one sentence recap since it's been a while since the original post: I am trying to recover a filesystem that was quite badly damaged by an offline resize2fs of a fairly modern ext4fs from 20 TB to 24 TB.
>>>>>
>>>>> I spent a lot of time trying to get something meaningful out of e2fsck/debugfs and learned quite a bit in the process and I would like to briefly share some observations.
>>>>>
>>>>> 1) The first hurdle running e2fsck -fnv is that the "Superblock has an invalid journal (inode 8)" is considered fatal and cannot be fixed, at least not in r/o mode so e2fsck just stops, this check needed to go away.
>>>>>
>>>>> 2) e2fsck gets utterly confused by the "bad block inode" that incorrectly gets identified as having something worth looking at and spends days iterating through blocks (before I cancelled it). Removing handling if ino == EXT2_BAD_INO in pass1 and pass1b made things a bit better.
>>>>>
>>>>> 3) e2fsck using a backup superblock
>>>>> ext2fs_check_desc: Corrupt group descriptor: bad block for inode table
>>>>> e2fsck: Group descriptors look bad... trying backup blocks...
>>>>> This is bad, as it means using a superblock that has not been updated with the +4TB. Consequently it gets the location of the first block group wrong, or at the very least the first inode table that houses the root inode.
>>>>> Forcing it to use the master superblock again makes things a bit better.
>>>>>
>>>>> I have some logs from various e2fsck runs with various amounts of hacks applied if they are of any interest to developers? I will also likely have the filesystem in this state for a week or two more if any other information I can extract is of interest to figure out what made resize2fs screw things up.
>>>>>
>>>>>
>>>>>
>>>>> In the end, the only actual change I have made to the filesystem to make it mountable is that I borrowed a root inode from a different filesystem and updated the i_block pointer to point to the extent tree corresponding to the root inode of my broken filesystem which was quite easy to find by just looking for the string "lost+found".
>>>>>
>>>>> # mount -o ro,noload /dev/md0 /mnt/loop
>>>>> [2815465.034803] EXT4-fs (md0): mounted filesystem without journal. Opts: noload
>>>>>
>>>>> # df -h /dev/md0
>>>>> Filesystem Size Used Avail Use% Mounted on
>>>>> /dev/md0 22T -382T 404T - /mnt/loop
>>>>>
>>>>> Uh oh, does not look to good.. But hey, doing some checks on the data contents and so far results are very promising. An "ls /" looks good and so does a lot of the data that I can verify checksums on, checks are still running...
>>>>>
>>>>> I really do not know how to move on with trying to repair the filesystem with e2fsck. I do not feel brave enough to let it run r/w on the given how many hacks that I consider very dirty were required to even get it this far. At this point letting it make changes to the filesystem may actually make it worse so I see no other way forward than extracting all the contents and recreating the filesystem from scratch.
>>>>>
>>>>> Question is though, what is the recommended way to create the filesystem? 64bit is clearly necessary, but what about the other feature flags like flex_bg/meta_bg/resize_inode...? I do not care much about slight gains in performance, robustness is more important, and that it can be resized in the future.
>>>>>
>>>>> Only online resize from now on, never offlline, I learned that lesson...
>>>>>
>>>>> Will it be possible to expand from 24 TB to 28 TB online?
>>>>>
>>>>> thanks,
>>>>> -johan
>>>>>
>>>>>
>>>>>> On 2015-08-13 20:12, Johan Harvyl wrote:
>>>>>>> On 2015-08-13 15:27, Theodore Ts'o wrote:
>>>>>>> On Thu, Aug 13, 2015 at 12:00:50AM +0200, Johan Harvyl wrote:
>>>>>>>
>>>>>>>>> I'm not aware of any offline resize with 1.42.13, but it sounds like
>>>>>>>>> you were originally using mke2fs and resize2fs 1.42.10, which did have
>>>>>>>>> some bugs, and so the question is what sort of might it might have
>>>>>>>>> left things.
>>>>>>>> What kind of bugs are we talking about, mke2fs? resize2fs? e2fsck? Any
>>>>>>>> specific commits of interest?
>>>>>>> I suspect it was caused by a bug in resize2fs 1.42.10. The problem is
>>>>>>> that off-line resize2fs is much more powerful; it can handle moving
>>>>>>> file system metadata blocks around, so it can grow file systems in
>>>>>>> cases which aren't supported by online resize --- and it can shrink
>>>>>>> file systems when online resize doesn't support any kind of file
>>>>>>> system shrink. As such, the code is a lot more complicated, whereas
>>>>>>> the online resize code is much simpler, and ultimately, much more
>>>>>>> robust.
>>>>>> Understood, so would it have been possible to move from my 20 TB -> 24 TB fs with
>>>>>> online resize? I am confused by the threads I see on the net with regards to this.
>>>>>>>> Can you think of why it would zero out the first thousands of
>>>>>>>> inodes, like the root inode, lost+found and so on? I am thinking
>>>>>>>> that would help me assess the potential damage to the files. Could I
>>>>>>>> perhaps expect the same kind of zeroed out blocks at regular
>>>>>>>> intervals all over the device?
>>>>>>> I didn't realize that the first thousands of inodes had been zeroed;
>>>>>>> either you didn't mention this earier or I had missed that from your
>>>>>>> e-mail. I suspect the resize inode before the resize was pretty
>>>>>>> terribly corrupted, but in a way that e2fsck didn't complain.
>>>>>> Hi,
>>>>>>
>>>>>> I may not have been clear on that it was not just the first handful of inodes.
>>>>>>
>>>>>> When I manually sampled some inodes with debugfs and a disk editor, the first group
>>>>>> I found valid inodes in was:
>>>>>> Group 48: block bitmap at 1572864, inode bitmap at 1572880, inode table at 1572896
>>>>>>
>>>>>> With 512 inodes per group that would mean at least some 24k inodes are blanked out,
>>>>>> but I did not check them all, I just sampled groups manually so there could be some
>>>>>> valid in some of the groups below group 48 or a lot more invalid afterwards.
>>>>>>
>>>>>>> I'll have to try to reproduce the problem based how you originally
>>>>>>> created and grew the file system and see if I can somehow reproduce
>>>>>>> the problem. Obviously e2fsck and resize2fs should be changed to make
>>>>>>> this operation much more robust. If you can tell me the exact
>>>>>>> original size (just under 16TB is probably good enough, but if you
>>>>>>> know the exact starting size, that might be helpful), and then steps
>>>>>>> by which the file system was grown, and which version of e2fsprogs was
>>>>>>> installed at the time, that would be quite helpful.
>>>>>>>
>>>>>>> Thanks,
>>>>>>>
>>>>>>> - Ted
>>>>>> Cool, I will try to go through its history in some detail below.
>>>>>>
>>>>>> If you have ideas on what I could look for, like ideas on if there is a particular periodicity
>>>>>> to the corruption I can write some python to explore such theories.
>>>>>>
>>>>>>
>>>>>> The filesystem was originally created with e2fsprogs 1.42.10-1 and most likely linux-image
>>>>>> 3.14 from Debian.
>>>>>>
>>>>>> # mkfs.ext4 /dev/md0 -i 262144 -m 0 -O 64bit
>>>>>> mke2fs 1.42.10 (18-May-2014)
>>>>>> Creating filesystem with 3906887168 4k blocks and 61045248 inodes
>>>>>> Filesystem UUID: 13c2eb37-e951-4ad1-b194-21f0880556db
>>>>>> Superblock backups stored on blocks:
>>>>>> 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
>>>>>> 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
>>>>>> 102400000, 214990848, 512000000, 550731776, 644972544, 1934917632,
>>>>>> 2560000000, 3855122432
>>>>>>
>>>>>> Allocating group tables: done
>>>>>> Writing inode tables: done
>>>>>> Creating journal (32768 blocks): done
>>>>>> Writing superblocks and filesystem accounting information: done
>>>>>> #
>>>>>>
>>>>>> It was expanded with 4 TB (another 976721792 4k blocks). Best I can tell from my logs this
>>>>>> was done with either e2fsprogs:amd64 1.42.12-1 or 1.42.12-1.1 (debian packages) and
>>>>>> Linux 3.16. Everything was running fine after this.
>>>>>> NOTE #1: It does *not* look like this filesystem was ever touched by resize2fs 1.42.10.
>>>>>> NOTE #2: The diff between debian packages 1.42.12-1 and 1.42.12-1.1 appear to be this:
>>>>>> 49d0fe2 libext2fs: fix potential buffer overflow in closefs()
>>>>>>
>>>>>> Then for the final 4 TB for a total of 5860330752 4k blocks which was done with
>>>>>> e2fsprogs:amd64 1.42.13-1 and Linux 4.0. This is where the:
>>>>>> "Should never happen: resize inode corrupt"
>>>>>> was seen.
>>>>>>
>>>>>> In both cases the same offline resize was done, with no exotic options:
>>>>>> # umount /dev/md0
>>>>>> # fsck.ext4 -f /dev/md0
>>>>>> # resize2fs /dev/md0
>>>>>>
>>>>>> thanks,
>>>>>> -johan
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
>> the body of a message to majordomo@...r.kernel.org
>> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists