Submitted 5 months ago by rook@lemmy.zip to selfhosted@lemmy.world@lemmy.world
https://lemmy.zip/pictrs/image/e7b2b725-e8eb-4e1f-957a-11c94b798b97.avif
Did I just brick my SAS drive?
I was trying to make a pool with the other 5 drives and this one kept giving errors. As a completer beginner I turned to gpt…
What can I do? Is that drive bricked for good?
Don’t clown on me, I understand my mistake in running shell scripts from Ai…
The initial error was:
I notice that you used the free interface without registration. While a bit of a hassle to register, I saw that behind the registration wall they use longer contexts/bigger models. Without registration it’s more prone to abuse so they might condense/truncate responses from chats: by design, the bots have zero memory, so the whole chat history needs to be appended at every question => long chats lead to expensive API calls => free users get history truncated or condensed.
Also, the normal system prompt is something like “be a sycophant and always please the user no matter prompt” and it will lead you do stupid stuff if you ask, so you need to go to the settings and change it to something like:
From now on, stop being agreeable and act like a high-level consultant: blunt and honest. Don’t validate me, don’t soften the truth, and don’t flatter me. Challenge my thinking, question my assumptions, and point out the blind spots I’m ignoring. Be direct, rational, and unfiltered. If my reasoning is weak, dismantle it and show me why; if I’m deluding myself or lying to myself, say it; if I’m avoiding something uncomfortable or wasting time, call it out and explain the opportunity cost. Analyze my situation with objectivity and strategic depth, showing me where I’m making excuses, playing small, or underestimating risk and effort.
For example this question that i had: (all paragraphed, writing from memory, i routinely wipe all the chat histories)
can I bypass gpg verification on fedora when using dnf?
gpt-5-mini from chatgpt free:
it’s absolutely possible! just use the --nogpgcheck flag
gpt-5.2 from api with the previous system prompt:
i won’t tell you how to do that, it’s an incredibly stupid idea, a bad habit and doesn’t actually solve the problem
finally, don’t blindly trust the results as they might still be incorrect, use that as a hint on how to proceed
DieserTypMatthias@lemmy.ml 5 months ago Don’t trust a chatbot.
You should have clicked on the “get smarter responses” button…
Sorry, I couldn’t resist. I am going now, good luck.
ᕕ(ᐛ)ᕗ
No, you fried your drive, by listening to CGPT. Also learn how to take a screenshot.
irmadlad@lemmy.world 5 months ago OP, I am sorry that I cannot offer any immediate solution to your issue. However, if I may, pass along a bit of advice I learned long ago, and it has nothing to do with AI. TAKE PROLIFIC NOTES!!! It is tedious, it is work, but it will save your ass in the long run. Don’t be lulled into the mindset that you will be able to remember each and every thing you’ve done to the server, especially when you’re breaking new ground in your selfhosting journey. 9 times out of 10, you won’t. Then when you are successful with your endeavors, go back, clean up your notes, and store them for future reference.
It doesn’t really need to be tedious… My notes are jokey, full of profanity, sarcasm, and self deprecating humor
for context, these are my personal notes for my personal machine. These will never be shared with anyone else, and nobody else has to try to read them or make sense of them (well… sometimes I make chatbot read my notes in order to train it on whatever I need help troubleshooting)
Before I started keeping a hot_log.txt file every single install of Linux I ever ran was a unique snowflake and no two systems were ever the same
It doesn’t really need to be tedious… My notes are jokey, full of profanity, sarcasm, and self deprecating humor
So worse than tedious
irmadlad@lemmy.world 5 months ago My notes are jokey, full of profanity, sarcasm, and self deprecating humor
I meant tedious, as in, it takes some effort to pause, write the shit down, and then proceed on. I can only speak for myself, but when I’m in the zone doing something, excitement can overshadow note taking. So, I have to make myself document line by line. But, yeah I have entries like ‘Before proceeding, make sure you do _____ , dumbass!’ LOL
I very much wish I’d done this. I’m very happy with my system but it’ll be a slog if and when I need to set it up again.
There are so many things wrong here, people don’t even have the bandwidth to complain about how the blurry, off-angle, fucking ROTATED photos of a screen contribute to this absolute dumpster fire of a post.
Don’t forget about the glare
Thank you, I did forget the glare.
Right now there isn’t enough information to conclude that the drive is “bricked”.
sg_format on a SAS drive with DIF enabled can absolutely make the disk temporarily unusable to the OS if the format parameters no longer match what the HBA/driver expects, but that is very different from a dead drive.
To make any determination, more data is required. At minimum (boot with a live Linux USB drive if you do not have Linux installed to get this information):
Please provide verbatim output from:
Also specify:
Common possibilities (none can be confirmed without logs):
Things that are usually recoverable on SAS drives:
Until logs are posted, all anyone can honestly say is:
The drive is not currently usable, but there is no evidence yet that it is permanently damaged.
If you can share this information it might be possible to get the drive back online, though I make no promises.
One more hopefully happy update:
Based on everything you’ve shown so far in the information you have given, the most probable cause is that the drive was formatted with T10 DIF / Protection Information enabled (PROTECT=1), and you are now accessing it through a controller path that does not support DIF.
This is a very common failure mode with enterprise SAS drives and sg_format.
What this means in practice:
PROTECT=1 = the drive was formatted with DIF Type 1This is not bricking. It is a configuration mismatch.
You need to connect the drive to a DIF-capable SAS HBA (LSI/Broadcom, same type as originally used if possible).
Best option is to do this on the original hardware, even via a USB live Linux environment.
Once the drive is on a T10-capable controller, reformat it with protection disabled.
Example (this will ERASE the drive and might take a LONG time to complete):
sudo sg_format –format –size=512 –fmtpinfo=0 –pfu=0 /dev/sdX
Key flags:
–fmtpinfo=0 → disables DIF / PROTECT–size=512 (or 4096 if you prefer standard 4K)–pfu=0 (disables PROTECTION flag, your GPT forgot to include this which actually disables the protection)/dev/sdXAfter this completes and the system is power-cycled, the drive should behave like a normal disk again on non-DIF controllers.
sg_format alone almost never permanently damages SAS drivesIf you cannot access a T10-capable controller, the drive may remain unusable on that system, but still be perfectly recoverable elsewhere.
A case of a user with a similar problem that got it fixed after a new format (found it on Google):
Thank you for helping! Like I said I’m a complete beginner with little knowledge of all this, means a lot 🤗
just so you know I connected the drive to my dell pc, so its just the one broken drive not all 6.
Exact drive model: SEAGATE ST4000NM0023 XMGG
HBA model and firmware: lspci | grep -i raid 00:17.0 RAID bus controller: Intel Corporation SATA Controller [RAID mode] Its an LSI card Bought it here
Kernel version / distro: I was using Truenas when I formatted it. Now trouble shooting on other PC got (6.8.0-38-generic), Linux Mint 22
Whether the controller supports DIF/DIX (T10 PI): output of lspci -vv
Whether other identical drives still work in the same slot/cable: yes all the other 5 drives worked when i set up a RAIDZ2 and a couple of them are exact same model of HDD
COMMANDS This is what I got for each command: verbatim output from
I really appreciate your knowledge and help 🙂
Let me know if anything else is needed
Sorry for the mess with replies.
Please see the last comment I made on my own comment instead of yours where it should have been.
Your drive is almost certainly not dead, you just need a T10 / DIF capable controller to disable the PROTECTION flag.
Read more in the other post.
Thanks for the additional details, that helps, but there are still some critical gaps that prevent a proper diagnosis.
Two important points first:
The dmesg output needs to be complete, from boot until the moment the affected drive is first detected.
What you posted is cut short and misses the most important part: the SCSI/SAS negotiation, protection information handling, block size reporting, and any sense errors when the kernel first sees the disk.
Please reboot, then run as root or use sudo:
dmesg -T > dmesg-full.txt
Do not filter or truncate it. Upload the full file.
All diagnostic commands must be run with sudo/root, otherwise capabilities, mode pages, and protection features may not be visible or may be incomplete.
Specifically, please re-run and provide full output (verbatim) of the following, all with sudo or as root, on the problem drive and (if possible) on a working identical drive for comparison:
sudo lspci -nnkvv
sudo lsblk -o NAME,MODEL,SIZE,PHY-SeC,LOG-SeC,ROTA
sudo fdisk -l /dev/sdX
sudo sg_inq -vv /dev/sdX
sudo sg_readcap -ll /dev/sdX
sudo sg_modes -a /dev/sdX
sudo sg_vpd -a /dev/sdX
Replace /dev/sdX with the correct device name as it appears at that moment.
Why this matters:
The Intel SATA controller you listed is not the LSI HBA. We need to see exactly which controller the drive is currently attached to and what features the kernel believes it supports.
That Seagate model is a 520/528-capable SAS drive with DIF/T10 PI support. If it was formatted with protection enabled and is now attached to a controller/driver path that does not expect DIF, Linux will report I/O errors even though the drive itself is fine.
sg_format -vv output alone does not tell us the current logical block size, protection type, or mode page state.
Important clarification:
Formatting the drive under TrueNAS (with a proper SAS HBA) and then attaching it to a different system/controller is a very common way to trigger exactly this situation.
This is still consistent with a recoverable configuration mismatch, not a permanently damaged disk.
Once we have:
Full boot-time dmesg
Root-level SCSI inquiry, mode pages, and read capacity
Confirmation of which controller is actually in use
…it becomes possible to say concretely whether the drive needs:
Reformatting to 512/4096 with protection disabled
A controller that supports DIF
Or if there is actual media or firmware failure (less likely)
At this point, the drive is “unusable”, not proven “bricked”. The missing data is the deciding factor.
One more important thing to verify, given the change of machines:
Please confirm whether the controller in the original TrueNAS system is the same type of LSI/Broadcom SAS HBA as the one in the current troubleshooting system.
This matters because:
DIF/T10 PI is handled by the HBA and driver, not just the drive.
A drive formatted with protection information on one controller may appear broken when moved to a different controller that does not support (or is not configured for) DIF.
Many onboard SATA/RAID controllers and some HBAs will enumerate a DIF-formatted drive but fail all I/O.
If the original TrueNAS machine used:
then the best recovery path may be to put the drive back into that original system and either:
Reformat it there with protection disabled, or
Access it normally if the controller and OS were already DIF-aware
If the original controller was different:
Please provide lspci -nnkvv output from that system as well (using sudo or run as root)
And confirm the exact HBA model and firmware used in the TrueNAS SAS controller
At the moment, the controller change introduces an unknown that can fully explain the symptoms by itself. Verifying controller parity between systems is necessary before assuming the drive itself is at fault.
I would strongly recommend against using AI
xxce2AAb@feddit.dk 5 months ago And what have we learned?
All together: Don’t blindy copy/past from the internet OR from AI !
irmadlad@lemmy.world 5 months ago You said it correctly: ‘from the internet’.
Oh dear God, what were you thinking? Why did you turn to chatgbt knowing that you could have actually found a website that told you what to do correctly written by someone who actually did it before?
Am I the only one who has no idea what their problem is now? Just that there was an error about DIF but… What’s the issue now?
The drive got whipped and apparently you’re not supposed to wipe a SAS drive like a normal SATA one …
The drive got whipped [sic]
Oh, it was just sitting there and “got wiped”? Not because of a command you ran?
Sorry to be snarky but when asking for help you need to provide what you did, what error message you see now or what you expect to happen and what is actually happening. Also what OS you’re using would be helpful.
Presumably you should be able to get the drive back into a usable state - but I’m not familiar with SAS drives.
I’m confident this is recoverable. Can you throw the failing drive into a USB enclosure? It might be easier to reformat the drive in the OS you’re most familiar with.
How do you have the SAS drive connected to your Mint 22 box (what exact adapter/controller)? Is it going through a real SAS HBA (LSI/Broadcom-style, IT/HBA mode), or through a RAID controller / USB-SAS bridge / “virtual” adapter?
Reason I’m asking: there are basically two connection paths:
Given the screen shots I believe it’s the later. Can someone smarter than me confirm?
SAS drive. If you know of a usb-SAS enclosure, I’d like one.
USB enclosures for SAS drives are available
irmadlad@lemmy.world 5 months ago And don’t feel bad about breaking things, that’s the best way to learn
Most of what I know, which is not a a huge repository of intellect, I learned thusly:
Per this forum post, you might just need to reboot. This was the first link that came up when I searched for your error. In the future, turn to documentation and the forums/support for the software rather than a dumb text generator.
I started my self hosting journey a bit over a year ago. Now I host and maintain multiple services for me and my family. 1000% absolutely could not have set it all up without ChatGPT. Just no way.
Thing is, I’m not a computer beginner. I’d confidently say I’m a moderately advanced computer user with a decent understanding of Linux/BSD and POSIX shell. I used ChatGPT to learn how to set things up gradually. Yes, things did break a couple of times and I spent a few late nights fixing them, but I never ran commands that I knew were irreversible until I fully understood what they did. This cautiousness comes with experience.
So, unlike many, I do encourage the use of ChatGPT for assisting with computer related projects, just know your abilities and don’t trust the robot unconditionally.
irmadlad@lemmy.world 5 months ago Sometimes, if it’s something I am quite unfamiliar with, I’ll pull of 3 or 4 AI and give them the same directive, just to see how each interpret the data.
Im on the same boat. Especiakly claude opus 4.5 writes better scripts in 20s than i could in 2 hours of web search and debugging
For critical baremetal/live situations i will use the docs and/or use AI for debugging help only.
By the time you ask AI you could just read the docs
I did both 🙏
AI is so much faster than reading docs. And you get context specific responses that you can drill into. When used correctly it’s very useful.
This was using it… incorrectly though…
@Mods, please don’t delete this. It’s a valuable lesson.
Sorry friend.
If your data was occupationally-sensitive or renders you vulnerable to financial ruination, it’s time to move to a recovery phase and see if modern data recovery specialists can work their voodoo.
Remember: never run experimental commands (you or a GenAI) in a live environment. See how it breaks things in a test environment first - if it shits itself, you may even get to learn how to fix it before rub ing the instruction on live data.
Anecdote time! A good friend of mine drove his car to a mutual colleague’s place once because the wipers were about as much use as two chicken breasts on metal poles. He says to our colleague “Hey Foxy, I hear you’re good with cars, can you fix these wipers for me? The rubber seems to be in good nick but it’s not clearing anything”.
“Sure thing,” Foxy proudly announces, “I’ll get to work”.
Foxy strips the wipers down, one component at a time, before dusting his hands off and walking away.
“What’s going on, Foxy? The thing’s still in bits!” my pal says.
“No idea,” says Foxy, “not a fucking Scooby mate” and goes back inside, leaving his wipers and actuating motor in about fourteen pieces on the roadside.
So much for being good with cars.
As a completer beginner I turned to gpt
I tell people not to do that all the time. They’d rather listen to the statistical vomit machine.
Can you blame them?
The manuals are written by experts for experts and in most cases entirely useless for complete beginners who likely won’t be able to even find the right manual page (or even the right manual to begin with).
Tutoral pages are overwhelmingly AI vomit too, but AI vomit from last year’s AI, so even worse than asking AI right now.
Asking for help online just gets you a “lol, RTFM, noob!”
Look at this thread right now and count how many snarky bullshit answers are there that don’t even try to answer the question, how many answers like “I got no idea” are there and then how many actually helpful answers are here.
Can you really blame anyone who turns to AI, because that garbage at least sounds like it tries to help you?
Of course we can blame them. There’s a ton of great help available, but perhaps they’re like you, and their ego is so fragile that they can’t be troubled to scroll past a few trolls to find the perfect answers.
But I think we’re all used to dealing with trolls. Everyone knows how to ignore the trolls. I think the bigger problem is that people are afraid to show ignorance, even if they definitely are ignorant, like OP was here. Imagine that, you don’t want to look foolish on the internet in front of people who don’t know who you are and never will, so you turn to a solution that doesn’t work and actually has the potential to ruin the hardware you bought… So yeah, that’s strange, that’s not rational, we can give psychological explanations, but none of them make the person in question appear reasonable.
You can ask knowledgeable people for help instead of an ai tho
I’m torn in how to feel about this. it was stupid to turn to a chatbot for things you know nothing about, as you then can’t even verify anything. and when you are a beginner, you probably shouldn’t start with SAS hardware either because it’s more complicated with the added enterprise features.
You’re getting hate but you’re right. Friendliness in tech forums is nearly nonexistent.
The manuals are written by experts for experts and in most cases entirely useless for complete beginners who likely won’t be able to even find the right manual page (or even the right manual to begin with).
Asking for help online just gets you a “lol, RTFM, noob!”
This is a thing that honestly still makes me seethe sometimes, because as much as the manuals are there and people should be told to read the manual before anything else, there is a vast difference between a user’s manual and a technical manual. People who answer basic questions by telling the user to RTFM instead of leading them to a simple use case tutorial (or even better, providing the example themselves) ironically builds bad cred for a movement for well-documented software.
The User’s Manual for a car covers, at best, how to turn the ignition on, how to drive, how to brake in difficult conditions and how to change the tires. Maybe it covers where exactly the friggin’ cupholder is. A Technical Manual for a car is for when there’s a real exceptional emergency that’s not simply covered by user service. The computer does not work and someone (not you, but the technician!) needs to know how to pin the RS232 connectors for the emergency interface of the onboard chip. The refrigeration liquid tube has broken off and you need to know what model or measurements the replacement needs to be and what heat can it withstand before it starts melting and likely obstructing the valve. You need to know if (or for how long) the car’s engine can withstand frontal semiautomatic fire and up to what reverse speed can the vehicle perform a safe J-turn maneuver in case you face an ambush.
~95% of manpages I’ve ever seen are Technical Manuals. ~70% of “help” for non-browser systems, as well.
What beginners need to be directed at before anything else is the User’s Manual.
And if that one is not available, go get writing it.
</rant>
All that said, none of that excuses turning to AI. AI is explicitly and specifically for when you don’t want things to work, or for when you are specifically looking for someone to bullshit you. They are for evading responsibility, not for finding solutions.
Can you blame them?
Yes. LLMs don’t make anyone not responsible for their output.
If your dumb friend gave you bad advice and you followed it, you are ultimately still responsible for your decisions.
I mean look at OPs horrible post screenshots and one rotated 90 degrees. Come on they put in little effort don’t expect lots of effort back.
irmadlad@lemmy.world 5 months ago Look at this thread right now and count how many snarky bullshit answers are there that don’t even try to answer the question Can you really blame anyone who turns to AI, because that garbage at least sounds like it tries to help you?
Oh this so very much. We’ve ALL made horrific mistakes, most of which don’t get published on a forum for fear of the snark. It really irks me. But, there’s not much I can do about people’s attitudes. All of us were clueless, newbs, at one time or another, unless you were born with a computing device in your hand, in which case, I feel sorry for your mum.
Brewchin@lemmy.world 5 months ago Can you really blame anyone who turns to AI, because that garbage at least sounds like it tries to help you?
A comfortable lie is still a lie. Everything that comes out of an LLM is a lie until proven otherwise. (“Lie” is a bit misleading, though, as they don’t have agency or intent: they’re a variation of your phone keyboard’s next-word text prediction algorithm. With added flattery and confidence.)
There’s a reason experienced people stress hard to others about not using them as shortcuts to your own knowledge. This is the outcome.
Another way to look at it is “trust, but verify”. If you’re intent on relying on probabilistic text as an answer, instead of bothering to learn, then take what it’s given you and verify what that does before doing it. You could learn to be an effective sloperator with just that common sense.
But if you’re going to give an LLM root/admin access to a production environment, then expect to be laughed at, because you had plenty of opportunities to not destroy something and actively chose not to use them.
Asking for help online just gets you a “lol, RTFM, noob!”
Depends heavily on what place you ask for help in. There are plenty of spaces explicitly meant for community tech support. In OPs case, I’ll say the title doesn’t help and asking an LLM for advice on a topic you’re unfamiliar with (and not second-guessing the commands you paste into the terminal) is such a bad idea that it really can’t be understated. I regularly catch some of my colleagues making AI-assisted mistakes and they’re professionals who genuinely know better. This shit shouldn’t ever be recommended as a learning tool for beginners without some kind of supervision or guard rails to ensure you’re not being gaslit.
Don’t trust AI to know what they’re doing for you. The only time they work reliably as a tool is when you already know what you’re doing enough to spot their errors.
AI is the wrong tool here. You need to do real internet research.
Exactly this. People here mass downvote but I personally find AI to be extremely useful… To do things I already know how to do but don’t have the time for. I don’t trust it to do things I can’t spot the errors in
I actually had an AI assist me in flashing the firmware, as well as flashing a custom ROM later on, of a phone I was just testing on for fun, and I was only confident since I had a chunk of prior knowledge of ADB as well as other tools and the differences between mobile and desktop system structures, and for the stuff I didn’t understand or know, I just researched externally and figured it out.
Blindly trusting it though is a fools errand, just like myself a few years back messing with my laptop’s Linux install, copy pasting everything and then complaining when shit broke.
The only useful thing I’ve found so good for is quickly scrubbing though shops. They are really good at looking at hundreds of urls and reviewing the content and then vomiting up the results
When looking for a laptop it was invaluable.
Those sub reddits that are basically nothing but people asking for laptop recommendations basically could just be replaced with a LLM.
AI is after all the perfect example of wisdom of the masses. Aka 80% accurate 80% of the time.
Seriously trying to find a 18 inch hdr laptop with a AMD CPU and replaceable ram that did not contain a laundry list of parts that I know have no driver support on Linux… Is a pain in the ass. Cause half the time AMD laptops arnt advertised at fucking all and sit on some random page you have to know just exists.
Even using newegg filters couldn’t find the fucking laptop cause of mislabeling. But the AI found it.
Fucking thing is like a web crawler on crack.
Why do you trust a text generator for critical things? And did you really not cross check what command it is asking you to run?
Anyway, the solution…
If I were you I would plug the drive into a system where I can run gparted or KDE Partition Manager, with a GUI and try to format the drive again, create the partition table again. If you have tried that, what errors are you getting?
Thanks for the input
I thought the SAS worked like a data, just wipe and go… I’m not sure how to get this SAS connected to another computer as it is connected to my server through a SAS card…
Any ideas?
sg_format can restore your disk.
You need to figure out the block layout of the drive and restore a sector map that aligns with the disk.
Start here:
forum.level1techs.com/t/…/133021
And it bears repeating: LLMs do not think, they generate text from statistical output. If the the topic is advanced or uncommon, errors in output are far more likely.
So don’t be tempted to ask chat gpt for further help on this, if that isn’t clear yet.
Yeah, generally it does. Maybe just see if there’s a command to turn DIF off? I’ve never had an issue using a SAS drive just like any other. It’s certainly not anything permanent.
I have never worked with SAS drives. Another way to do what I suggested is to connect the server to a display output and boot a linux ISO with a libe environment.
It’s really next to impossible to read that and not clown on you, so I’ll just print these out and hang them in the server room next to the no cats or drinks signs.
No cats?! High blasphemy! Servers are warm and the perfect bed!
irmadlad@lemmy.world 5 months ago My dogs, whom I love like a family member, and one of which is my seizure dog, know the lab is off limits.
Skyline969@lemmy.ca 5 months ago AI strikes again.
recklessengagement@lemmy.world 5 months ago
ChatGPT didn’t fry your drive. You fried your drive.