Snapshot errors when storage backed by Windows file share

### Bug report

__System info:__ InfluxDB 1.3 ("influxdb:latest" Docker image), Linux Docker host

__Steps to reproduce:__

1. Mount a Windows fileserver's SMB export into a folder on the Docker host
2. Run the InfluxDB Docker image with /var/lib/influxdb bind mounted to the mounted samba folder
3. Wait a while

__Expected behavior:__ No adverse affects

__Actual behavior:__ Errors writing snapshots, unbounded disk consumption

__Additional info:__ [Include gist of relevant config, logs, etc.]

I'm experiencing this bug with InfluxDB 1.3, via the `influxdb:latest` Docker image.

After some time, I get these error messages:

```
[I] 2017-11-02T14:19:38Z Snapshot for path /var/lib/influxdb/data/telegraf/autogen/2 written in 19.637876ms engine=tsm1
[I] 2017-11-02T14:19:38Z error writing snapshot: sync /var/lib/influxdb/data/telegraf/autogen/2: invalid argument engine=tsm1
[I] 2017-11-02T14:19:39Z error adding new TSM files from snapshot: sync /var/lib/influxdb/data/telegraf/autogen/2: invalid argument engine=tsm1
```

My disk use grew to 36GB consumed in 12 hours, from only 2 Telegraf agents reporting in.

I don't have any problems if I bind mount `/var/lib/influxdb` to a regular folder on the Docker host.

Start with the TSM1 snapshot path and the bind-mounted /var/lib/influxdb/data/telegraf/autogen/2 directory described in the report. Reproduce using a Docker bind mount backed by a Windows SMB share, then inspect the snapshot and sync handling around the reported “invalid argument” errors. Done means snapshots sync without errors and disk usage no longer grows without bound.

Written by the indexing model from the issue text.