File cannot be written to when recovering from crash during nova_extend_inode_log (checksums+parity enabled) about linux-nova

[Andiry]

Thanks for the reporting. However, I am not sure how to use the image:

# echo img > /dev/pmem0 
# mount -t NOVA /dev/pmem0 /mnt/ramdisk/
# dmesg
[ 4337.052594] nova: Arch new instructions support: CLWB NO
[ 4337.052602] nova: Data structure size: inode 120, log_page 4096, file_write_entry 64, dir_entry(max) 304, setattr_entry 56, link_change_entry 40
[ 4351.582107] nova: 16 cpus online
[ 4351.582112] nova: nova_get_nvmm_info: dev pmem0, phys_addr 0x380000000, virt_addr 0xffffa005c0000000, size 2147483648
[ 4351.582113] nova: measure timing 0, metadata checksum 0, wprotect 0, data checksum 0, data parity 0, DRAM checksum 0
[ 4351.582203] nova: Start NOVA snapshot cleaner thread.
[ 4351.582206] nova: Running snapshot cleaner thread
[ 4351.582211] nova: Error in super block: try to repair it with the other copy
[ 4351.582220] nova: nova_init_inode_list_from_inode: 16 inode nodes
[ 4351.582222] nova: Recovered 0 snapshots, latest epoch ID 0
[ 4351.582223] nova: NOVA: Normal shutdown
[ 4351.582228] nova: Current epoch id: 0

Writing to /mnt/ramdisk/myfile and it succeeds.

[pcworld]

echo img > /dev/pmem0

I suppose you meant cat there? (That command would write the literal string "img" onto the device, however mounting should fail after that, so I assume you did it properly).

It seems that NOVA images are not really portable, I have also tried using it on another virtual machine, but when #cpus != 1 (the crash image was produced with #cpus = 1), weird things happen and the file is not even listed. My guess would be that NOVA does not properly recognize when #cpus have changed and the per-CPU journals are not recovered correctly or so. Should I report this as a separate issue?
The mounted image should contain a file called "myfile".
Though even when I set #cpus = 1 in another VM setup (different from the VM from which I extracted the image), I can iirc load the image and the existing file is listed, but I don't observe the problem I originally reported.
So I will need to dig further into how to reproduce this without relying on sharing the image.

[Andiry]

Yes, the change of #cpus is a known issue of NOVA. Fixing that, however, requires a lot of effort I assume.