Ollama: permissions issue when using volume storage

[ntnsndr]

I've got OpenWebUI running successfully, but I can't get the local Ollama server running right. I have been able to connect the Anthropic API, so the problem is not with the interface in general.

In OpenWebUI's Connections settings, when I try to add a model via Ollama (http://127.0.0.1:11434), I get "Failed to fetch models" and "Server Connection Error."

Here's some output I'm getting on the server:

# ps -af | grep -w "ollama"
root         242     108  0 03:21 pts/1    00:00:00 grep --color=auto -w ollama
# ollama ls
Error: could not connect to ollama app, is it running?
# ollama serve
Couldn't find '/root/.ollama/id_ed25519'. Generating new private key.
Error: could not create directory mkdir /root/.ollama: read-only file system

There seems to be a filesystem permissions issue here. The problem does not occur under the default app settings. But it does when I use the recommended path of creating a separate volume for the models.

As instructed here, I created a volume in the Filesystem at /media/ollama-vol. It is set up under the Storage settings for the app with "Read and Write" permissions. I have tried both chmod 775 and chmod 777 on that directory. In either case, when I start up the app, a models/ directory is created in there. As instructed, I have added to env.sh this line:

export OLLAMA_MODELS=/media/ollama-vol/models

Have other succeeded in getting this working by following these steps?

Thanks in advance for any help you can offer.

[BrutalBirdie]

What are you using for the Volume?
I am using a sub-account of my Hetzner Storage box with CIFS and I know that some apps do not like this at all.
Some just work and some refuse to work.
If I remember correctly I also had issues with permissions in my case.
But that is some time ago.

[joseph]

You have to run ollama server as cloudron user. Like gosu cloudron ollama serve

[ntnsndr]

@BrutalBirdie Thanks for sharing your experience—I'm using just the Filesystem mode in hopes that it would be most compatible.

@joseph Shouldn't the app automatically run ollama rather than requiring me to do it? And why wouldn't it work on the filesystem volume even when it is set up on chmod 777?

Thanks all!

[joseph]

@ntnsndr the supervisor config file in the package sets additional env vars like HOME and NTLK_DATA . Have you tried restarting the app (which ends up restarting supervisor) ?

[ntnsndr]

@joseph Yep, many times! Any other ideas?

[joseph]

@ntnsndr think we are missing something.

• why are you trying to run ollama serve manually on web terminal?
• when you start via supervisor, you still get same error as readonlyfs in /root/.ollama ?

[ntnsndr]

@joseph I was only try to run ollama manually to understand why it is not starting on its own when the app boots.

So, when I reboot the app with the /media/ollama-vol directory at 777 permissions, here is what I get:

Apr 15 08:31:05 2025-04-15 14:31:05,744 CRIT Supervisor is running as root. Privileges were not dropped because no user is specified in the config file. If you intend to run as root, you can set user=root in the config file to avoid this message.
Apr 15 08:31:05 2025-04-15 14:31:05,744 INFO Included extra file "/etc/supervisor/conf.d/ollama.conf" during parsing
Apr 15 08:31:05 2025-04-15 14:31:05,744 INFO Included extra file "/etc/supervisor/conf.d/openwebui.conf" during parsing
Apr 15 08:31:05 2025-04-15 14:31:05,748 INFO RPC interface 'supervisor' initialized
Apr 15 08:31:05 2025-04-15 14:31:05,748 CRIT Server 'unix_http_server' running without any HTTP authentication checking
Apr 15 08:31:05 2025-04-15 14:31:05,749 INFO supervisord started with pid 1
Apr 15 08:31:06 2025-04-15 14:31:06,752 INFO spawned: 'ollama' with pid 25
Apr 15 08:31:06 2025-04-15 14:31:06,755 INFO spawned: 'openwebui' with pid 26
Apr 15 08:31:06 2025/04/15 14:31:06 routes.go:1231: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:2048 OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/media/ollama-vol/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
Apr 15 08:31:06 Error: mkdir /media/ollama-vol/models/blobs: permission denied
Apr 15 08:31:06 2025-04-15 14:31:06,790 WARN exited: ollama (exit status 1; not expected)
Apr 15 08:31:07 2025-04-15 14:31:07,792 INFO spawned: 'ollama' with pid 33
Apr 15 08:31:07 2025-04-15 14:31:07,793 INFO success: openwebui entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
Apr 15 08:31:07 2025/04/15 14:31:07 routes.go:1231: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:2048 OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/media/ollama-vol/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
Apr 15 08:31:07 Error: mkdir /media/ollama-vol/models/blobs: permission denied
Apr 15 08:31:07 2025-04-15 14:31:07,823 WARN exited: ollama (exit status 1; not expected)
Apr 15 08:31:09 2025-04-15 14:31:09,828 INFO spawned: 'ollama' with pid 43
Apr 15 08:31:09 2025/04/15 14:31:09 routes.go:1231: INFO server config env="map[CUDA_VISIBLE_DEVICES: GPU_DEVICE_ORDINAL: HIP_VISIBLE_DEVICES: HSA_OVERRIDE_GFX_VERSION: HTTPS_PROXY: HTTP_PROXY: NO_PROXY: OLLAMA_CONTEXT_LENGTH:2048 OLLAMA_DEBUG:false OLLAMA_FLASH_ATTENTION:false OLLAMA_GPU_OVERHEAD:0 OLLAMA_HOST:http://127.0.0.1:11434 OLLAMA_INTEL_GPU:false OLLAMA_KEEP_ALIVE:5m0s OLLAMA_KV_CACHE_TYPE: OLLAMA_LLM_LIBRARY: OLLAMA_LOAD_TIMEOUT:5m0s OLLAMA_MAX_LOADED_MODELS:0 OLLAMA_MAX_QUEUE:512 OLLAMA_MODELS:/media/ollama-vol/models OLLAMA_MULTIUSER_CACHE:false OLLAMA_NEW_ENGINE:false OLLAMA_NOHISTORY:false OLLAMA_NOPRUNE:false OLLAMA_NUM_PARALLEL:0 OLLAMA_ORIGINS:[http://localhost https://localhost http://localhost:* https://localhost:* http://127.0.0.1 https://127.0.0.1 http://127.0.0.1:* https://127.0.0.1:* http://0.0.0.0 https://0.0.0.0 http://0.0.0.0:* https://0.0.0.0:* app://* file://* tauri://* vscode-webview://* vscode-file://*] OLLAMA_SCHED_SPREAD:false ROCR_VISIBLE_DEVICES: http_proxy: https_proxy: no_proxy:]"
Apr 15 08:31:09 Error: mkdir /media/ollama-vol/models/blobs: permission denied
Apr 15 08:31:09 2025-04-15 14:31:09,862 WARN exited: ollama (exit status 1; not expected)
Apr 15 08:31:10 => Healtheck error: Error: connect ECONNREFUSED 172.18.16.191:8080

[ntnsndr]

Relatedly (I am trying to try again with another volume), does anyone see if I'm doing something wrong here?

To create the /media/ollama-vol directory before I had to mkdir it manually before adding the volume.

Could there be something screwy with my filesystem permissions in general?

[BrutalBirdie]

Did you ever try without the volume if it is working at all?
If no, that should be done first to minimize the error potential.

[ntnsndr]

@BrutalBirdie yes, as noted above: "The problem does not occur under the default app settings. But it does when I use the recommended path of creating a separate volume for the models."

[BrutalBirdie]

Huh
Then there must be something really strange going on.
Either perimissions on the server itself or some other ghost in the shell.

[joseph]

@ntnsndr the Local Directory is something like /opt/ollama-models . You have to create this directory on the server manually . This prevents admins from mounting random paths (the idea being that only a person with SSH access can "create" paths).

In your initial post, what kind of volume are you using? I wonder if it's CIFS and it doesn't support file permissions ?

[ntnsndr]

As I said above, I'm using the Filesystem mode for the volume. Should I use another?

[ntnsndr]

Question: What should be the correct directory ownership and permissions setting for the ollama-home folder on a mounted Volume?