Extended attributes error on AFP share

[nas2160]

Problem:

While running my backup task using Carbon Copy Cloner on OS X copying data from my local machine to my FreeNAS box via AFP I get an error saying:

“The “task name” task has encountered physical read errors"
This task is still running, and CCC will copy as many files as possible.
Physical read errors may cause the backup task to take longer than normal. You can stop the backup task at any time. CCC will pick up where it left off the next time you run the task. Open CCC and click on the source volume in the sidebar for read error statistics.”​

The detailed error message is:

Code:

Error codes: 1016 : 1900 : 5
Error: rsync: [generator]get_xattr_names: llistxattr("/Volumes/.../PNLF SB.txt",1024) failed: Input/output error (5)
Advice: Delete this file from the destination, then try the backup task again.

CCC uses rsync in the background to perform backup operations.

I have contacted CCC support and they have said:

"these errors are occurring because the destination filesystem is failing to return a list of extended attributes for a particular file upon request. This may not be an issue with the ZFS filesystem, in fact it's more likely an issue with the file sharing service on the NAS. These errors don't occur with a Finder copy because nothing exists at the destination when Finder is trying to copy a file there. Indeed, "I have also tried deleting the file from the NAS and running the task again as requested by CCC" -- the error only occurs when CCC tries to get a list of extended attributes on an existing file so it can determine whether they need to be updated.

I really like CCC so it would be great to get to the bottom of the error and fix anything that needs fixing etc so it works as required.

There's a simple workaround -- instruct CCC to not copy any extended attributes. This section of CCC's documentation explains the workaround (in a slightly different context):

https://bombich.com/kb/ccc4/trouble...en-copying-files-or-from-synology-diskstation
"
As the response above states the error occurs: "because the destination filesystem is failing to return a list of extended attributes for a particular file upon request".

The problem happens every time I run the backup task sometimes a fair while into the backup task and at other times almost immediately.

My question is why is FreeNAS/AFP not returning the extended attributes for a given file, why so some work and others not?

Is it an AFP setup issue?

Has anyone had similar issues?

I have searched the forum and the internet for a few hours but decided its time to post in the forum for help.

Instructing CCC to not copy the extended attributes solves the issue as it leaves them out, however that isn't really a solution only a work around.

Ideally I would like to solve the issue so that extended attributes can be copied and checked during my backup tasks etc.

Carbon Copy Cloner version:

• Version 4.1.2 (4131)

OS X version:

• OS X Yosemite 10.10.3 (14D136)

System Configuration Information:

FreeNAS version:

• FreeNAS-9.3-STABLE-201505130355

Hardware:

• Motherboard: SuperMicro X10SL7-F
• CPU: Intel Xeon E3-1231 v3
• RAM: 16 GB (2x8GB) ECC SAMSUNG 1.35v 1600 MHz (M391B1G73QH0-YK0)
• Storage: 2x 4 TB WD RED + 2x 4TB SEAGATE NAS HDD setup as a GELI encrypted RAIDZ2.
• Boot: Sandisk Extreme 32 GB USB3.0 (SDCZ80-032G-GAM46)

 

[Robert Trevellyan]

Is the share set up with Mac as the Permission Type?

If truly accurate Mac backups are important to you, there's no substitute for Arq Backup. No affiliation...

 

[nas2160]

Yes the share is setup with the Mac permission type. Being that the ZFS dataset the share points to has the Mac permission type set. I assume that is what you meant?

Okay, I will have a look at Arq Backup

However I think the issue is with FreeNAS and AFP and not the program used to make the backups. We will see though!

 

[Robert Trevellyan]

Yes the share is setup with the Mac permission type. Being that the ZFS dataset the share points to has the Mac permission type set. I assume that is what you meant?

Yes, that's what I meant. Unfortunately I don't have much insight into the details of your problem.

However I think the issue is with FreeNAS and AFP and not the program used to make the backups. We will see though!

Maybe so, but backup programs like Arq don't rely on the destination filesystem to support the features of the source.

CCC is a cloner, not a backup program, so it does require the destination to support the features of the source.

I'm not criticizing CCC, which I like and use, just pointing out that its purpose is to create bootable clones, not accurate backups on arbitrary storage destinations.

 

[nas2160]

Okay cool.

Ah I see, so how does Arq store my files on my NAS? a proprietary file format?

Yes, it does. Well what I'm trying to achieve by cloning my files from my Mac to FreeNAS is to have a backup all my files directly on the ZFS filesystem, so as stated above the workaround to backup to a disk image on my NAS is not a solution for me as I am then reliant on using OS X to open and work with the files.

Can you provide a link to how Arq would store my data on my NAS were I to use it?

Okay, part of my backup strategy will be to periodically clone my OS install to a disk image on my NAS.

 

[jlpellet]

Does the share point to the "root" dataset on the pool, which is created when the pool is created, or a subsidiary? I asked because 9.3 has trouble with Windows extended attribute on the default shell but the issue does not occur on a subsequent share. That is, if the pool is, for example Z2, a default pool names Z2 is created & shares to that pool have the issue but if a lower level dataset, such a Z2a is made, all is OK. I have no idea if the same issue exists for Mac shares. I opened a bug report but, since the problem is very narrow and only on the root pool the resolution is not to be fixed, which I understand. FYI, bug is 7189. Hope this helps.

 

[nas2160]

Thanks for the response. Thats interesting, though my share doesn't point to the root dataset, but to a data set that is under the root data set in the form: name_of_pool->root_dataset_with_same_name_as_pool -> my_dataset_shared_via_afp

 

[Robert Trevellyan]

Can you provide a link to how Arq would store my data on my NAS were I to use it?

Arq uses an encrypted content-addressable data structure, so it can use an untrusted destination, which means you can't just browse the files.

 

[Savage]

I and my department peons have used CCC and similar mac tools to archive / backup osx boot volumes for years. Usually 10-20 different 10-200+gb bootable drives per week.

A we -NEVER- use such tools without creating/specifying the destination as some flavor of apple sparse image file, especially when on a remote SMB or AFP volume and the backup content is a bootable OSX install.

Most of those gui tools remember their target DMG files for future or repeat backup scheduling anyways. Arbitrarily dumping half a million files directly into your netatalk server should be avoided when possible anyways.

 

[nas2160]

Arq uses an encrypted content-addressable data structure, so it can use an untrusted destination, which means you can't just browse the files.

Thanks for pointing out Arq Robert

I have been running Arq for a few weeks backing up to FreeNAS via SFTP and Google Drive as a test, and I really like it, so it will definitely be a part of my backup strategy.
The next step is to learn how to use the command line recovery tool for the Arq backup data structure so that if the developer of Arq ever disappears I can still recover my data from the Arq data structure if necessary.

 

[nas2160]

I and my department peons have used CCC and similar mac tools to archive / backup osx boot volumes for years. Usually 10-20 different 10-200+gb bootable drives per week.

A we -NEVER- use such tools without creating/specifying the destination as some flavor of apple sparse image file, especially when on a remote SMB or AFP volume and the backup content is a bootable OSX install.

Most of those gui tools remember their target DMG files for future or repeat backup scheduling anyways. Arbitrarily dumping half a million files directly into your netatalk server should be avoided when possible anyways.

Okay, yes I understand that it is a requirement for backing up OS X boot volumes.

I have done testing backup runs with CCC to disk images.

However for the backup of my personal files I want to have them stored on a modern file system as I don't like or trust HFS+.
So a disk image is not an option for these files as all Apple disk image formats use HFS+ as their filesystem.

Are which was mentioned above is a great little program and it saves and restores all extended attributes and file meta data for each file etc.

At the moment I am also using CCC with the option to ignore extended attributes to backup my files to FreeNAS directly.

 

[Robert Trevellyan]

The next step is to learn how to use the command line recovery tool for the Arq backup data structure so that if the developer of Arq ever disappears I can still recover my data from the Arq data structure if necessary.

As far as I know, the author has not updated that tool to work with SFTP destinations.

 

[nas2160]

I have checked and the description mentions sftp: http://sreitshamer.github.io/arq_restore/
So I think it is supported :) I will try it out and report back.
However it looks like Dropbox and Onedrive are not supported by the restore tool.
Do you use Arq to backup to s3/glacier or your FreeNAS server via SFTP?

 

[Robert Trevellyan]

I have checked and the description mentions sftp: http://sreitshamer.github.io/arq_restore/

That's great (it's been a while since I checked).

Do you use Arq to backup to s3/glacier or your FreeNAS server via SFTP?

Currently we have 3 Macs backing up to FreeNAS with Arq via SFTP, and one of them also backing up to Google Nearline Storage with Arq. Our other cloud backups are with CrashPlan. A year from now, when our existing CrashPlan subscription is close to expiring , I'll be looking for an alternative and Arq/Google/Amazon will be a strong candidate.

 

[Savage]

Okay, yes I understand that it is a requirement for backing up OS X boot volumes.

However for the backup of my personal files I want to have them stored on a modern file system as I don't like or trust HFS+.
So a disk image is not an option for these files as all Apple disk image formats use HFS+ as their filesystem.

Ah; I agree on that point. When I saw CCC mentioned I thought it was somewhat of a different use case. We tend to use those tools for scenarios where digging out a bootable copy of system files + data quickly is the emphasis over reliability (like when sending macs in for apple service or swapping some machine in the fleet) . . .not personal data backup outside of some bootable 1:1 thing on an attached drive.

On my own personal business data backups I just run a sync of /Users/ to a freenas dataset (with snapshots) and have a quota constrained time machine pointed elsewhere (different server/nas), configured to exclude everything but /Users/ as well. Every once in a while I drop a bootable clone dmg of /Applications /Library /System in storage. And there is some crash plan in the mix there too :p.