Ed Nisley's Blog: Shop notes, electronics, firmware, machinery, 3D printing, laser cuttery, and curiosities. Contents: 100% human thinking, 0% AI slop.
The backup scripts running on Mary’s folks’ PC kvetched on each backup:
WARNING: Some files and/or directories in /home/ only transferred partially during rsync operation
WARNING: /usr/bin/rsnapshot daily: completed, but with some warnings
Poking around showed that the problem came from the .gvfs “directory” tucked into each home directory, which produced this unenlightening ls -al result:
d????????? ? ? ? ? ? .gvfs
Come to find out that it’s an old problem with mysterious causes that should be fixed by now; evidently I triggered it by installing basic Ubuntu-with-Unity and then installing the Xbubuntu desktop. Or something like that.
Anyhow, the solution workaround involves an rsnapshot configuration entry that bypasses that directory:
I did five minutes of standup comedy at yesterday’s MHV Lug meeting, pointing out some of the more interesting ways to compromise a PC when you have an infinite budget for development and consumables.
You don’t get my patter with the PDF (unless you had access to the room’s bugging hardware), but the links may come in handy in the unlikely event you haven’t been following the story closely.
If you have a security clearance or are in line for one, you probably shouldn’t click on the link, because it contains copies of pages from the leaked NSA catalog:
This resembles the 32 GB Micro SD card checkout, with the exception that, for some unknown reason, the available space doesn’t match up with the actual space occupied by the file. It also turns out that rsync deletes the incomplete file, rather than leaving a stub, which makes perfect sense, but was still a bit disappointing after two hours.
I had two identical Sandisk Cruzer Fit Flash Drives, one of which appears here:
32 GB Sandisk USB Flash Drive
Those squares are an inch on a side, so it’s a bit larger than the Micro SD card. Adding a lanyard loop on the plastic cap or a string between cap and drive seems like a great idea, because that little thing is certain to get lost.
The snippets here represent a compendium of Things Done that happened over the course of two days; I didn’t save all the logs. The process started with the same 32 GB file of entropy I used for the Micro SD card:
df -B1 /mnt/part2
Filesystem 1B-blocks Used Available Use% Mounted on
/dev/sdc1 31512350720 180424704 31331926016 1% /mnt/part2
-----------------------
time rsync --progress /mnt/part/Testdata/Testdata.bin /mnt/part2
Testdata.bin
31298191360 99% 14.18kB/s 0:39:38
rsync: writefd_unbuffered failed to write 4 bytes to socket [sender]: Broken pipe (32)
rsync: write failed on "/mnt/part2/Testdata.bin": No space left on device (28)
rsync error: error in file IO (code 11) at receiver.c(322) [receiver=3.0.9]
rsync: connection unexpectedly closed (28 bytes received so far) [sender]
rsync error: error in rsync protocol data stream (code 12) at io.c(605) [sender=3.0.9]
real 126m20.505s
user 3m6.393s
sys 2m17.492s
-----------------------
time dd bs=8K count=20000000 if=/mnt/part/Testdata/Testdata.bin of=/mnt/part2/Test1.bin
dd: writing ‘/mnt/part2/Test1.bin’: No space left on device
3820963+0 records in
3820962+0 records out
31301320704 bytes (31 GB) copied, 7455.97 s, 4.2 MB/s
real 124m15.970s
user 0m1.607s
sys 1m17.546s
-----------------------
truncate -s 31301320704 /mnt/part/Testdata/Testdata.bin
-----------------------
ll /mnt/part/Testdata/Testdata.bin
-rw-r--r-- 1 ed ed 31301320704 Dec 24 18:13 /mnt/part/Testdata/Testdata.bin
-----------------------
time diff /mnt/part/Testdata/Testdata.bin /mnt/part3/Test1.bin
real 26m37.081s
user 0m4.400s
sys 0m52.723s
Notice that the write speed runs around 4 MB/s, which is a lot slower than you might expect from a USB 2.0 device; as with a hard drive, the interface doesn’t limit the throughput! The read speed, on the other paw, trots along at about 20 MB/s.
One of these will go to Mary’s folks as an online daily backup device; their PC will soon run a version of the rsnapshot scripts that back up our basement file server. It’s not off-site backup and it’s not proof against catastrophic hardware failure, but it should be good enough.
Mary’s compadres sometimes send her pictures of garden vegetables and quilting projects. Those pictures usually pass through Microsoft Outlook (or its ilk) and emerge in winmail.dat files that aren’t particularly useful in a Linux context. That page gives a good overview of the problem and how to resolve it; I’m just documenting the process here, so I can find it again.
Start by installing both tnef and convmv. I think the latter isn’t needed in our situation, because most folks use flat ASCII file names that come through just fine.
Save the attachment in, say /tmp and unleash tnef on it:
cd /tmp
tnef --file=winmail.dat
That unpacks all the attachments into /tmp, where one may have one’s way with them.
It’s not worth my effort to bolt that into the email programs and then maintain that mess across updates, so we’ll do it by hand as needed.
Microsoft certainly had a good reason for inventing Yet Another Encapsulation Format, although I wonder why good old ZIP wouldn’t have worked nearly as well…
The new-to-me Optiplex 980 has a tool-free clamp securing the PCI card brackets to the chassis, with a nice plastic dress cover that really finishes off that side of the case. Alas, it’s secured by five small heat-staked plastic pegs that I managed to shear off as part of a finger fumble that you’ll recognize when it happens to you and which I need not further discuss:
Optiplex 980 PCI Clamp Cover – disassembled
So I drilled two slightly undersized holes for the tiniest screws in the Little Box o’ Tiny Screws:
Optiplex 980 PCI Clamp Cover – drilling
The two end plates sticking up are the only square parts of the cover, so that thing is actually clamped by the right-side plate and sheer will power. I ran the drill down 3 mm from the top of the post at the slowest manual jog speed from the Joggy Thing and I did not break through the top and did not hit that lathe bit under the cover.
The screw threads and a dab of epoxy hold them in place:
Optiplex 980 PCI Clamp Cover – tiny screws
I’d like to say the finished repair looked like this:
Optiplex 980 PCI Clamp Cover – in place
But, alas, the eagle-eyed reader will note that the screws are gone, replaced by two dabs of clear acrylic caulk; those faint threads and epoxy were no match for the snap of that latching lever and the slight distortion caused by the spring fingers applying force to the brackets.
Quick summary: the current Linux startup machinery Runs All The Things! in parallel, leaving you to figure out all the interdependencies and update all the script files to match your requirements. Mostly, the distro maintainers figure all that, but if you have essential files mounted as NFS shares, then you can will reach a login screen before the mount process completes.
Having wrestled with this problem for a while, I think I’ve doped out the right way to coerce the Upstart Pachinko Machine to converge on a workable login.
The solution is to fire off a unique signal after the NFS mount command, then force the display manager to wait until it receives that signal, rather than depend on happenstance as I did before. The mounts occur in /etc/init/local.conf, which now looks like this:
description "Stuff that should be in /etc/rc.local"
author "Ed Nisley - KE4ZNU"
start on (local-filesystems and net-device-up IFACE=em1)
stop on shutdown
emits nfs-mounted
script
logger Starting local init...
logger Mounting NFS filesystems
mount /mnt/bulkdata
mount /mnt/userfiles
mount /mnt/diskimages
mount /mnt/music
initctl emit nfs-mounted
logger Ending local init
end script
The start condition ensures that this code won’t run until the wired LAN is up; note that what was once eth0 is now em1. Then, after the mounts happen, initctl fires the nfs-mounted signal.
The modification to /etc/init/lightdm.conf script consists of one additional line to wait for that signal:
start on ((filesystem
and runlevel [!06]
and started dbus
and plymouth-ready
and nfs-mounted)
or runlevel PREVLEVEL=S)
stop on runlevel [016]
emits login-session-start
emits desktop-session-start
emits desktop-shutdown
I’m not convinced lightdm.conf is the right spot to jam a stick in the gears, but it seems to be the least-awful alternative. The login-session-start signal doesn’t appear in any file in that subdirectory and I have no idea where else to look.
Anyhow, the greeter screen now shows a desktop background from the NFS mount, which I regard as A Good Sign:
The Smell of Molten Projects in the Morning 