for reference; original implementation for nfs; #27329 (and related issue: #26639)

Codecov Report

❗ No coverage uploaded for pull request base (master@12b837e). Click here to learn what that means.
The diff coverage is 33.33%.

@@            Coverage Diff            @@
##             master   #39250   +/-   ##
=========================================
  Coverage          ?   37.06%           
=========================================
  Files             ?      612           
  Lines             ?    45488           
  Branches          ?        0           
=========================================
  Hits              ?    16861           
  Misses            ?    26340           
  Partials          ?     2287

Did this get tested? Does the cifs module really use the same weird addr=:<addr> syntax?

Re: For Testing:

The Docker documentation will have to be updated since there is no current section on CIFS. We're able to create and mount CIFS volumes using the same syntax as NFS.

docker volume create
--driver local
--opt type=cifs
--opt addr=//cifs_volume_name/path
--opt o=username=USERNAMEpassword=PASSWORD
volume_name

From the Docker documentation on NFS (https://docs.docker.com/engine/reference/commandline/volume_create/):

docker volume create --driver local
--opt type=nfs
--opt o=addr=192.168.1.1,rw
--opt device=:/path/to/dir
foo

The "addr:" syntax is part of compose.

Hm, not sure that would work with the current code though; if addr=//cifs_volume_name/path, the current code will try to resolve //cifs_volume_name/path, which is not a valid hostname, so some pre-parsing will have to be done

ipAddr, err := net.ResolveIPAddr("ip", addrValue)

Sorry, copied/pasted an older command (which also works). The same exact syntax as nfs also works for cifs with the path broken out:

docker volume create
--driver local
--opt type=cifs
--opt addr=volume_name
--opt device=:/dir/another_dir
--opt o=username=USERNAME,password=PASSWORD
volume_name

The getAddress function in local.go is just looking for "addr=" for the volume_name, so that should be fine. I'm trying to dive into how CIFS is parsing out the device option for the path and resolving the ":", but it's definitely working on my end.

Hi guys-- What's the status on this? I'm running into this problem and hoping these changes get merged in.

Hi guys-- What's the status on this? I'm running into this problem and hoping these changes get merged in.

Hey Drew. I was going to ask the main contributors what more I can do for testing this. I don't have access to another CIFS environment to verify that the same mounting syntax for NFS works, nor have I had the chance to deep dive into the volume creation to see how it handles CIFS. I'd be happy to add more to the PR if necessary.

ping @cpuguy83 ptal

@shuchow for the documentation; perhaps a block similar to the NFS example on this page; https://docs.docker.com/storage/volumes/#create-a-service-which-creates-an-nfs-volume

(source file for that is in this repository: https://github.com/docker/docker.github.io/blob/master/storage/volumes.md)

And the reference section for volume create; https://docs.docker.com/engine/reference/commandline/volume_create/#driver-specific-options

(source file for that is in the docker/cli repository: https://github.com/docker/cli/blob/master/docs/reference/commandline/volume_create.md#driver-specific-options)

@thaJeztah Great, thanks for the links. I'll send a PR for the documentation in a bit.

Note that these options are passed directly to the kernel module for this fs type.
addr is used by nfs, I was just checking to make sure that cifs does the same thing.

Are we sure this actually works?
Just testing against mount.cifs I see a mount system call like so:

mount("\\\\1.2.3.4\\foo", ".", "cifs", 0, "ip=1.2.3.4,unc=\\\\1.2.3.4\\foo,use"...)

for this command:

mount.cifs -o user=foo,password=bar \\\\1.2.3.4\\foo /tmp

So, based purely on what I'm seeing in the ubuntu mount.cifs we need to be looking at a totally different option for cifs (as compared to nfs).

There's also a prefixpath added if the UNC goes beyond the share name.

eg. //fileserver/backup/daily turns into unc=\\fileserver\backup,prefixpath=daily (and ip=, etc.)

$ sudo ./mount.cifs //fileserver/backup/rolling /backup -o 'user=foo,pass=bar'
mount(//fileserver/backup/rolling, ".", cifs, (null), ip=192.168.86.50,unc=\\fileserver\backup,user=foo,prefixpath=rolling,pass=bar)

when I tried this on Docker 20.10.2, for this syntax

docker volume create
--driver local
--opt type=cifs
--opt addr=volume_name
--opt device=:/dir/another_dir
--opt o=username=USERNAME,password=PASSWORD
volume_name

I get an error during creation of volume: invalid option: "addr"

For the other syntax:

docker volume create --driver local
--opt type=cifs
--opt o=addr=hostname
--opt device=:/path/to/dir
foo

Creation was okay but when I actually mount it I got a different error: invalid argument

From the code changes I can see it is simply replacing addr=hostname to addr=ip, however the errors seemed to suggest addr was never a valid argument to mount.cifs. Could you clarify @shuchow?

@spinosae This still works for me in docker 20.10.2:

docker volume create 
--driver local
--opt type=cifs
--opt device=//core/backup
--opt "o=uid=1000,gid=1000,ip=172.16.20.110,unc=\\core\backup,file_mode=0664,dir_mode=0775,iocharset=utf8,cache=none,mapchars,mfsymlinks,user=bkupusr,pass=${CIFS_PASSWD}"
sync-backup

Most of the extra options can be dropped, but thats a copy of a live volume create from a daily script.

@spinosae I misread your comment, all this PR does is rewrite the addr= from a hostname to an IP.

so this will work:

docker volume create \
--driver local \
--opt type=cifs \
--opt device=//myserver/backup \
--opt "o=addr=myserver.localdomain.com,user=bkupusr,pass=${CIFS_PASSWD}" \
sync-backup

it will find the addr=myserver.localdomain.com look up what the IP address (say 127.0.0.1) is and replace it with that (so addr=127.0.0.1) because the kernel driver won't accept a hostname there, but requires the parameter.

So your second example should have worked if the device was a valid path to a CIFS mount eg. //server/share (but not a subdirectory like //server/share/dir, that needs the prefixpath option I've mentioned previously).

@spinosae I misread your comment, all this PR does is rewrite the addr= from a hostname to an IP.

so this will work:

docker volume create \
--driver local \
--opt type=cifs \
--opt device=//myserver/backup \
--opt "o=addr=myserver.localdomain.com,user=bkupusr,pass=${CIFS_PASSWD}" \
sync-backup

it will find the addr=myserver.localdomain.com look up what the IP address (say 127.0.0.1) is and replace it with that (so addr=127.0.0.1) because the kernel driver won't accept a hostname there, but requires the parameter.

So your second example should have worked if the device was a valid path to a CIFS mount eg. //server/share (but not a subdirectory like //server/share/dir, that needs the prefixpath option I've mentioned previously).

wow, just like that, MAGIC!

so the difference is device option, instead of :/path/to/dir/ it should have been //myserver/share. What finally worked for me is:

docker volume create \
--driver local \
--opt type=cifs \
--opt device=//host.docker.internal/share \
--opt o=addr=host.docker.internal,nounix,noserverino \
source

@spinosae Did you finally manage to connect via domain name?

Hi,

Was this resolved? I'm hitting the CIFS + DNS problem with a couple of containers. I tried adding the 'addr=' option as above, but I get 'invalid argument' when the container tries to start.

root@dock:/home/netadmin# docker volume create --driver local --opt type=cifs \
--opt "o=addr=dougal.mydomain.net" --opt device=//dougal/TransferTemp ttt
root@dock:/home/netadmin# docker volume inspect ttt
[
    {
        "CreatedAt": "2023-08-09T14:13:17Z",
        "Driver": "local",
        "Labels": null,
        "Mountpoint": "/var/lib/docker/volumes/ttt/_data",
        "Name": "ttt",
        "Options": {
            "device": "//dougal/TransferTemp",
            "o": "addr=dougal.mydomain.net",
            "type": "cifs"
        },
        "Scope": "local"
    }
]
root@dock:/home/netadmin# docker run -it --rm -v ttt:/test busybox sh
docker: Error response from daemon: error while mounting volume '/var/lib/docker/volumes/ttt/_data': failed to mount local volume: mount //dougal/TransferTemp:/var/lib/docker/volumes/ttt/_data, data: **addr=192.168.11.22: invalid argument**.
root@dock:/home/netadmin# nslookup dougal.mydomain.net
Server:         192.168.11.1
Address:        192.168.11.1#53

Non-authoritative answer:
Name:   dougal.mydomain.net
Address: 192.168.11.22

root@dock:/home/netadmin# smbclient -L dougal.mydomain.net
Password for [WORKGROUP\root]:
Anonymous login successful

        Sharename       Type      Comment
        ---------       ----      -------
        TransferTemp    Disk
        RemoteBackup    Disk

Have the options changed, or does a FQDN for a CIFS share not work?

@jefft4 could you open a ticket with details? Note that these options are not handled by the docker daemon itself, but passed as options when doing a mount syscall, so options (including the addr= option) depend on support on the daemon's host.

@jefft4 could you open a ticket with details? Note that these options are not handled by the docker daemon itself, but passed as options when doing a mount syscall, so options (including the addr= option) depend on support on the daemon's host.

Sure, will do.

• Unable to mount CIFS using FQDN or 'addr' option - Ubuntu 22.10 host #46180

Does not fix the problem: #46180 (comment)