Ticket #6442 (new defect)
Machine doesn't shut down after filling NAND
| Reported by: | cscott | Owned by: | cscott |
|---|---|---|---|
| Priority: | high | Milestone: | |
| Component: | distro | Version: | Development build as of this date |
| Keywords: | release? joyride-2281:- | Cc: | gnu, jg, mstone, dsd |
| Action Needed: | never set | Verified: | no |
| Deployments affected: | Blocked By: | ||
| Blocking: | #7125 |
Description
From private mail: --- Subject: Does update.1-691 have the olpcrd fix for full filesystems? This fix doesn't have a separate trac bug; it's in the comments in http://dev.laptop.org/ticket/5719.
http://dev.laptop.org/ticket/5317 tracks the real bug, but that trac entry got sidelined into a JFFS2 enforce-free-space patch that didn't make update.1.
The small "readlink" patch that Scott made to olpcrd makes the system bootable when the filesystem is full. I looked in the ramdisk in update.1-691, and it appears to be there. But I want to make 1000% sure, since this is a bug that makes the system very brick-like if not fixed.
I am currently filling my NAND and will report back whether the machine will boot when I'm done. Hmm. I filled it, until it refused to create new files, or even new filenames. Then I ended that shell, which was the only Activity running, and hit "Reboot" in the donut.
It tried to shut down, froze the screen with the UL warnings, and never did shut down. ==> BUG! <== I let it sit that way for at least 10 minutes.
I powered it off manually, powered up manually. There's a long delay after "hello (children of the) world" and the jffs xattr spam message. Several minutes, at least. Probably scanning the whole filesystem, slowly.
It booted all the way up into Sugar. Selecting "About this XO" produces "Build: update.1 691; Firmware: Not available; Serial Number: Not available". This appears to be because /ofw didn't get correctly mounted; df -h shows:
Filesystem ... mtd0 1.0G 1014M 11M 99% / none 1.0G 1014M 11M 99% /ofw tmpfs 117M 44K 117M 1% /dev/shm
But it comes up, gets a terminal, and can become root (which would let the filesystem full problem get corrected). Bravo. ---
This trac bug is for the 'shutdown doesn't complete when NAND is full' issue. I believe I've seen shutdown fail in other circumstances; the summary of this bug may have to change when we get to the root of the problem.
