Root Access

allquixotic

16:01

This answer is so very correct, but when I first read it I didn't know what I was talking about. It took over a year to realize that a drive CANNOT "change" from 512e to 4kn just by some configuration setting; this is baked into the drive. Guys please give this answer the recognition it deserves. This is the clearest explanation I've seen yet of the 512n/512e/4kn thing. This will become popular once 4kn is common in the consumer space ;p — allquixotic 12 secs ago

qasdfdsaq

Which answer?

Ah comment, nvm

allquixotic

remind me to reward that bounty tomorrow

4

A: Upgrading RAID controller to 4kn - do I have to reinitialize the array to use 4kn?

Your disks are either 512e (512 sectors on SAS/SATA interface) or 4k native (4k sectors on SAS/SATA interface), and unfortunately there is no way to change that via software or jumpers etc. You select the transfer mode when you buy the disks. Buy 4k native disk if you have adapter that supports 4...

qasdfdsaq

I thought I remember coming across a drive once where you could switch between 512e and 4kn in firmware, although it was a niche enterprise drive and doing so would lose all your data.

I might be mistaken

allquixotic

ah - yeah, I guess 99% of the drives out there wouldn't bother with such an esoteric feature in the fw

that sounds like a firmware programmer hunting for a bonus :P

the rest just say "eh, no one will notice it's a bit slower"

qasdfdsaq

Thankfully I didn't even notice when I started using 4K native drives, everything of mine just supports it :D

DragonLord

16:12

5

Q: What's the point of hard drives reporting their physical sector size?

I have an SSD that can be configured to report its physical sector size to an OS in two different ways: Option 1: Logical = 512 Bytes, Physical = 512 Bytes Option 2: Logical = 512 Bytes, Physical = 4096 Bytes (4K) What benefit does an OS gain by being aware of the 4K physical sector size, cons...

qasdfdsaq

8TB for $300 is basically unbeatable

@DragonLord Hmm, maybe I was being confused by something like that.

@allquixotic Probably. I'm sure they'd much rather sell 8 different versions of the same drive...

I really dislike 4K drives with ZFS though

The space wastage lost as a result of the 4K sectors has exceeded the size of an entire drive (4TB).

allquixotic

@DragonLord Man, brilliant answer!

Bob

allquixotic

@qasdfdsaq Unfortunately, I'm not quite up to that level of wealth, yet. I'm somewhere in the middle between the "I can't afford even a 1 TB drive" and the "I have a storage cabinet with 64GB of usable space" crowds.

Hence the 4 x 4 TB array with maxCache and a pretty beastly RAID card.

Bob

@allquixotic 64GB? :P

allquixotic

16:21

tb

Bob

@qasdfdsaq ?

Space wastage?

How big would an array have to be to lose that much space?

allquixotic

@Bob he has 64TB of usable space -- USABLE -- in his array

Bob

o.O

allquixotic

I think the problem is that a file will take up at least 1 sector, and there can't be multiple files' data within a sector, on most filesystems.

if you have 4097 bytes of data, it'll take up two 4096-byte sectors.

Bob

eh, the default for typical consumer use is 4k anyway (on NTFS)

allquixotic

16:24

plus whatever metadata, checksums, etc

I believe Reiserfs4 supported packing files efficiently into sectors to ameliorate some of that

Bob

Most people with mass data storage wouldn't have a great number of tiny files

allquixotic

but the storage space savings there would have a performance hit, since you'd end up doing a RMW (Read/Modify/Write) cycle on update

Bob

And for most purposes where you might want to create so many small files there's probably a more space efficient format than a general disk FS.

allquixotic

and the read/modify would go all the way to system RAM, not just the DRAM of the disk

@Bob InnoDB? :P (for example)

that said, UNIX-like operating systems do love them some high file counts on some of the builtin packages

Bob

@allquixotic Just about any of the compression/archive formats, some of the nosql stuff, etc..

@allquixotic Nowhere near 1TB worth, let alone 64TB worth.

allquixotic

16:28

The metadata overhead must be why most games ship their assets in a few package files of some various file format (some are just disguised ZIP, others are truly custom)

Bob

To lose a significant amount of space out of 1TB you'd literally need hundreds of millions of tiny files.

allquixotic

@Bob an order of magnitude less with ZFS because the checksums take up space, and the metadata for the checksums, and...

so, maybe tens of millions

Bob

@allquixotic That's more because many FSes (especially the ubiquitous NTFS) deal with metadata (and therefore large file count) very slowly.

Not so much due to space savings, I'd guess, since most of those assets are still several times 4kB.

DragonLord

@Bob NTFS stores small files within the MFT itself.

Bob

@allquixotic Do we know if those are packed efficiently?

Packed with the data, packed with other metadata?

allquixotic

16:30

@Bob also he probably has RAID striping overhead, possibly parity, redundancy, etc

DragonLord

That can quickly fill up the MFT Zone, though.

Bob

@DragonLord I'm aware.

@DragonLord Nope.

That's not how it works.

DragonLord

The MFT can extend beyond the reserved area, right?

Bob

Basically each MFT entry is a minimum of 4kB. If the file fits within 4kB - metadata, the rest is packed into the entry.

ext FSes do something similar, packing into inodes.

@DragonLord Yea, given enough free space.

@allquixotic Parity wouldn't be affected by 512 vs 4k

qasdfdsaq

@allquixotic In the case of this array, it's a 4KB sector on a 16-wide stripe, so the minimum allocation is 64KB

so any file, file fragment, metadata block, or anything else takes a minimum of 64KB allocation + 12KB parity.

Bob

16:32

@DragonLord More files, period, can create many entries. But a small file does not use up any more entry space than a large file, with the exception of a heavily fragmented file.

xCare

@Bob, hello!

Bob

hi

xCare

got your messages, in a more advanced stage now..

Bob

@qasdfdsaq well, that's more on you for choosing large stripes :P

xCare

here's the details:

0

Q: How does one recover from an internal Hard Drive I/O error?

This is the latest stage in a long running chain of events that has seen me cross the seven circles of HDD hell only to arrive at these depths of virtual inferno. My Hard Drive is a 750GB Seagate Barracuda. At the point it is now, Hard drive is still recognized in BIOS and is able to be recogniz...

windows-7 hard-drive hard-drive-recovery io

qasdfdsaq

16:34

@Bob Well, given 4TB were the biggest drives available at the time, it wasn't much of a choice

allquixotic

A 64kb stripe is not actually that big.

I use 256kb stripes.

DavidPostill

"Having said all that it wouldn't surprise me if a build would fail. The source is from 2005 ...
Let me try it ..."

Bob

@xCare ...I honestly think you might've gotten to the point where the something has failed so badly nothing will talk to it anymore.

qasdfdsaq

There's a difference between minimum allocation unit and stripe though. A stripe can be partially filled, the O/S has no awareness of it.

DavidPostill

I had to install gcc but it built perfectly.

allquixotic

16:36

@qasdfdsaq Ah, right. I was just explaining that aloud to myself as I typed it.

Bob

@xCare The drive is there. It responds enough to be seen. But it doesn't respond at all to SMART, nor ID requests.

allquixotic

Yeah, you can pack multiple logical sectors of data into a stripe. Of course.

Bob

It's a very odd situation at best, and I'm not sure if recovery is even possible at this point.

qasdfdsaq

There's a basic writeup about it here: louwrentius.com/…

And a somewhat more convoluted discussion here: github.com/zfsonlinux/zfs/issues/548

Bob

If you wanted to pay a lot of money you could probably get someone to try a board swap with another drive from the same batch (if one can be found). But that's expensive, risky and very involved.

xCare

16:37

@Bob, it's still recognized in the bios, it recognizes it in disk management, it even can tell the firmware of it as you can see from the pics... if the problem is software or firmware caused, there's got to be a way to recover from it.

allquixotic

I'm at that point where I understand more about modern storage and I/O than 98% of the population, but I'm still in a completely different (lower, more ignorant) league than the remaining 2% :P

xCare

I'm going to run through those tests that @JourneymanGeek outlined, see what that says

Bob

@xCare You say "recognized in the bios" -- but it's not!

I've seen your previous screenshots.

xCare

btw, how do you get pictures to show in posts, like journeyman did?

Bob

Your motherboard itself reports a SMART error.

DragonLord

16:39

@xCare Adding images inline. Just a moment...

Bob

As I said -- somehow your drive is in a state where the system can detect that there's something connected, but can't get any really meaningful info from it.

xCare

yes, it does, but it still shows up after it's been allowed to rest a while

Bob

@xCare Often, firmware failure is synonymous with a complete irrecoverable failure.

qasdfdsaq

@Bob That doesn't actually mean much

Bob

Do not assume firmware == software.

qasdfdsaq

16:40

You could remove the heads and the platters from the disk and it would show these symptoms

Bob

There's a reason it has a different name.

xCare

hotplugging it might have caused a firmware problem to arise, it's been known to occur in Seagate Barracuda's

Bob

Firmware is at a far lower level than your typical software.

qasdfdsaq

i.e. the firmware would respond to a probe/ident request yet any attempt to access it would bork

Bob

Failed firmware can be nigh impossible to flash, if it was corrupted somehow.

xCare

16:40

hm, good point

Bob

Failed firmware could cause hardware failure. It could also be a symptom of hardware failure.

qasdfdsaq

hackaday.com/2012/07/30/…

xCare

well, perhaps but I've seen articles talking about just that

Bob

Firmware is usually very closely tied to the hardware.

xCare

^

qasdfdsaq

16:41

Dunno if your issue is at all related to the notorius Seagate bug above

Wait it's aHitachi

xCare

not sure, but if it is, there might still be hope

qasdfdsaq

No that's someone else's screenshot

xCare

no, that's Journeyman's, not mine

allquixotic

everything is a "something -gate" these days, so would that make that the Seagate-gate?

2

Mokubai

Awesome. It's official. Computers suck balls.

2

DragonLord

16:42

Edited.

Bob

@qasdfdsaq @xCare I mean, sure, if you want to try that.

qasdfdsaq

Gate-ception?

Bob

It could work. Just keep in mind that it's a rather risky procedure.

xCare

haha @allquixotic ... quote of the day lol

Bob

Don't expect too much from it.

Mokubai

16:42

Which moron decided to not allow windows to use gpt on a bios based system?

qasdfdsaq

I swapped the heads out of a hard-drive myself once

xCare

hi @Mokubai

qasdfdsaq

@Mokubai: Ermm, the people who invented BIOS?

Isn't the whole issue that BIOSes cannot boot from GPT?

Mokubai

@qasdfdsaq well then those guys can go die in a fire.

qasdfdsaq

...

allquixotic

16:44

@qasdfdsaq do you remember the days when significant bits of hardware sold to consumers would be breakable due to certain user inputs? like, the hardware would actually break, right then and there if you executed certain code?

Bob

@qasdfdsaq The hack around it is the "hybrid MBR". Avoid at all costs - non-standard and poorly supported at best.

allquixotic

CRTs that you could ruin by changing resolution...

xCare

@DragonLord thanks!

qasdfdsaq

I overclocked my CRTs

And yeah, though these days everyone expects everything to be completely impervious to any user input

Looks like the drive in question is a 7200.12 so the firmware bug above wouldn't apply

allquixotic

I seem to remember it being possible to break certain GPUs with certain user input. And of course many types of hardware that have field-flashable firmware can be bricked (even today, with smartphones) due to a bad flash.

bricked is not quite the same as actually damaging electro/mechanical parts though

qasdfdsaq

16:46

I can still break my GPU, just set the voltage and clockspeed sliders to the far end of the scale, and tell my environmental controller to cut power to the water cooling pump.

allquixotic

lol

Bob

@allquixotic Eh, usually recoverable as long as the bootloader is intact.

It's when you start flashing bootloaders that everything gets all scary.

qasdfdsaq

I've already got a modded BIOS on there that removes the stock power/TDP limitations

allquixotic

Then there were higher "levels" of breakage that have since been closed off. For instance, on XP, the graphics drivers were GIGO, so if you passed an invalid command to your GPU, guess what -- GPU reset, kernel panic/BSOD, reboot.

qasdfdsaq

@Bob: Indeed. That said I quite like Samsung phones for the reason they are nigh-on unbrickable because of the excellent firmware recovery mode

allquixotic

16:47

Now they have a command stream verifier that avoids sending garbage to the GPU, and if the GPU does crash, they pull the power on the PCIe slot for a split-second and let it reinit :P

Bob

@qasdfdsaq I stopped liking them quite so much around the time they implemented an efuse.

qasdfdsaq

I liked them less, but still.

allquixotic

I haven't messed around with rooting/bootloader-unlocking smartphones since my Droid Razr Maxx HD

haven't had a use case to do it

Bob

@allquixotic Funny how we just accept that the GPU will crash, but the CPU must never, under any circumstances, do anything odd or the whole system will go down.

qasdfdsaq

I've rooted a lot but never touched the bootloader

allquixotic

16:48

and now I'm in a walled garden and let me tell you, the honeysuckle is in full bloom and it smells wonderful

@Bob heh, yeah, it's like how 32-bit "dedicated servers" for games hosted on Windows have to have some kind of a crash/network timeout detection and auto restart feature, but nginx won't crash on you unless you really torture it

qasdfdsaq

scontent-lhr3-1.xx.fbcdn.net/hphotos-xfp1/v/t1.0-9/…

Bob

@allquixotic Speaking of which -- had the phone a few weeks now, still no stuttering with BT audio.

qasdfdsaq

Your efuse comment reminded me of that...

Bob

None at all.

allquixotic

@Bob maybe just a US thing? I have nothing :S

Bob

16:50

So perhaps 5.1 fixed it? Or LG fixed it? Or LG was never susceptible? Or something to do with the 808?

allquixotic

anyway, same cans, same locations, iPhone screen off, no stuttering - MY problem was solved by going to Apple; others who don't have the problem are either lucky or just not seeing the same problem as I saw across 4 phones and 4 headphones for whatever reason

I'm pretty sure people were reporting issues with LG, but I never tested an LG phone personally

just Sammy and Moto

or could be the 808, or QCOM fixed a bug in the BT... I have no idea

qasdfdsaq

embed goddamn it

Bob

@qasdfdsaq ha

allquixotic

Knox is a good idea in principle - if it didn't have so many holes that defeat the purpose of having it

qasdfdsaq

@allquixotic If you keep up to date on some of the KVM/Xen virtualization hypervisor development mailing lists you'll realise some of those issues actually still exist today

allquixotic

16:52

using something like SELinux to lock down Android phones against all the malware and crap slung at them? yes, good, go, do it right

qasdfdsaq

As in they keep finding security holes where a virtual machine can set a certain bit of a certain register to some value and bam, they have DMA to a PCI device which in turn has DMA to the host memory

allquixotic

@qasdfdsaq tasty

qasdfdsaq

Flip a certain different bit and the dom0 hypervisor kernel panics

Write something weird in some place and you have R/W access to the CPU MSR

allquixotic

since my current "multi-tenant" dedi is only adminned by people I trust 100%, I'm not terribly worried, but I'm sure lxd is full of many such holes

it's mostly there to protect against accidental toe-stepping at this point

qasdfdsaq

Yeah, my server (which you pinged I think) hasn't been patched in a year.

allquixotic

16:54

ouchies

qasdfdsaq

But the hypervisor is pretty locked down so meh.

allquixotic

I at least do an occasional aptitude update; aptitude full-upgrade on my phys and each container

Bob

Heh. My old one running linux-vserver (still not decomissioned) hasn't been updated in... a while

qasdfdsaq

It's a Xen bare-metal hypervisor

Also, er... I accidentally did a chmod <something> / on it a year ago and have been afraid to restart it since.

9

allquixotic

ah. I only use hypervision (is that a word? it is now!) for Windoze on Linux

Bob

16:55

root@gaia:~# uptime
 17:05:07 up 116 days, 20:55,  1 user,  load average: 0.08, 0.03, 0.05

Hm... was expecting it to be higher.

allquixotic

those were some fast stars, must have caused a few chuckles

Bob

Then again, this is a pretty unreliable host... lots of power outages.

qasdfdsaq

One day I'll compare the file mode permissions to another Xenserver I'm running and patch them up, until then, yeah. I don't want to restart and patch.

Bob

root@debian:~# uptime
 17:03:03 up 99 days,  4:55,  2 users,  load average: 0.04, 0.09, 0.07

The newer server has been up almost as long.

allquixotic

Oh, @qasdfdsaq is it just me or do you notice an unseen hand eating a CPU thread 24/7 with no trace of it in top when you're using ZFS?

qasdfdsaq

16:56

Oh I got starred? Crap.

Bob

@allquixotic O_O

qasdfdsaq

@allquixotic: Nope. Though my ZFS kernel threads do show up in top

Bob

load average: 0.02, 0.08, 0.06

DavidPostill

@qasdfdsaq You don't like stars? ;)

Bob

That's the system with ZFS.

qasdfdsaq

16:57

I just don't get notified I don't think

@allquixotic: P.S. Dunno if I ever mentioned this but one huge advantage of Online.net over OVH: You get access to the BMC

Bob

You have heavy disk I/O coming from somewhere?

qasdfdsaq

@Bob Hah, my mobile phone can nearly beat that

Bob

@qasdfdsaq Eh... having experienced both I have no real preference one way or another.

allquixotic

@qasdfdsaq my ZOL kernel threads show up in top, too, but not as taking any measurable percentage of CPU beyond 0.1 - 0.2%

Bob

@qasdfdsaq It's a relatively new server. Been up pretty much since I finished installing the host.

qasdfdsaq

16:58

1931 hours uptime on my phone

Bob

@qasdfdsaq Old server: I have access to some godawful old version of iDRAC.

Also, BMC's failed once and they had to replace the mobo.

New server: I have access to the OVH control panel. A lot less info, but more or less the same functionality exposed to me.

qasdfdsaq

Meh, better than nothing. OVH hasn't even got the option. SYS (a subsidiary) I believe charges £20 PER DAY

Bob

They managed to knock something out a day after I got it and had to replace an "electric connector".

:S

qasdfdsaq

I dunno, when I was with OVH you couldn't install your own O/S, at least without a lot of severe fudgery

Bob

@qasdfdsaq ? I still get KVM-over-IP access.

Transcript for