[Asus Eee PC 1000] Trouble installing due to I/O errors and package corruption

[Daniel Manrique]

Consistently, on this machine Ubiquity pops up a message indicating the install can't proceed due to an I/O error.

Note that d-i, on the other hand, gives a retry option when an I/O error happens. So on this system it's necessary to use alternate installer with d-i. However, it does pop up a message saying that one or other package is corrupt. It usually takes several tries to get this machine installed, basically until the corrupt package is a non-critical one that doesn't leave the system in an undesirable state.

It's either a networking or a storage problem. I swapped out the hard disk (it's actually an SSD but with an asus-proprietary form factor, physically resembling a memory module) for another one from a compatible Eee PC that we don't currently test. If installation succeeds, this will point to a failing SSD which will probably have to be replaced.

The shipped SSD has 40 GB, split between an 8-GB "partition" and a 32-GB "partition", note that these are presented as independent drives in the BIOS.

[Daniel Manrique (roadmr)]

Ok, the error message from Ubiquity is still present with a different SSD, so it's probably not that. The error reads:

The installer encountered an error copying files to the hard disk:

[Errno 5] Input/Output error

Syslog shows this:

Dec 13 21:18:02 ubuntu : INFO: Simulating trans: /org/debian/apt/transaction/95d6ad71d8414fbbbcd7b91c8cd23a9b
Dec 13 21:18:03 ubuntu : INFO: Upgrade system with safe mode: 1
Dec 13 21:21:05 ubuntu kernel: [ 325.143313] SQUASHFS error: zlib_inflate error, data probably corrupt
Dec 13 21:21:05 ubuntu kernel: [ 325.143326] SQUASHFS error: squashfs_read_data failed to read block 0x27c34a9d
Dec 13 21:21:05 ubuntu kernel: [ 325.143334] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143340] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672
Dec 13 21:21:05 ubuntu kernel: [ 325.143356] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143362] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672
Dec 13 21:21:05 ubuntu kernel: [ 325.143376] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143381] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672
Dec 13 21:21:05 ubuntu kernel: [ 325.143390] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143396] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672
Dec 13 21:21:05 ubuntu kernel: [ 325.143410] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143416] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672
Dec 13 21:21:05 ubuntu kernel: [ 325.143426] SQUASHFS error: Unable to read fragment cache entry [27c34a9d]
Dec 13 21:21:05 ubuntu kernel: [ 325.143432] SQUASHFS error: Unable to read page, block 27c34a9d, size 8672

I have complete logs but it's probably not worth attaching them unless requested, at this point it's either networking or memory corruption. I'll test for both and update further.

[Daniel Manrique (roadmr)]

I did the following two tests:

1- I had the system pull the Ubuntu ISO image and verify its md5 checksum several times, to test for network corruption. It passed every time. The problem here is that TCP pretty much guarantees that data will be consistent, if the link has errors or corruption TCP will retry until it gets things right. So this test may not be conclusive; next step would be to test using UDP, which I still need to research a bit.

2- The system has been running memory test for ~20 hours without a single failure, so I think memory failure can be ruled out.

So far it's looking like network corruption but as mentioned before, I'll still need to do some more testing before reaching a conclusion.

[Daniel Manrique (roadmr)]

In the end, the SSD was the culprit, I swapped it with one from another Eee PC (one we don't test as much) and consistently, the original SSD fails, while the new one installs fine.

I'll leave the working SSD in and mark this fix released.