Hi, I've hit an issue with the free space cache. It looks like we miss writing everything to disk on unmount under rough conditions.
Setup: git head cmason/master, /dev/sdv1 is a 55MB partition on an SSD. Run the following script: -- DEV=/dev/sdv1 MP=/mnt/scratch umount $MP mkfs.btrfs -M $DEV mount -o inode_cache $DEV $MP cat /dev/urandom | head -c 654321 > $MP/1 mkdir $MP/2 mv $MP/1 $MP/2/1 btrfs subvol snap $MP $MP/@1 rm $MP/2/1 umount $MP mount -o inode_cache $DEV $MP cat /dev/urandom | head -c 654321 > $MP/tempname-1 ls -li $MP/@1/2/1 $MP/tempname-1 -- Actual result: > 257 -rw-r--r-- 1 root root 654321 Oct 16 15:04 /mnt/scratch/@1/2/1 > 259 -rw-r--r-- 1 root root 654321 Oct 16 15:04 /mnt/scratch/tempname-1 Expected result: Both files should have inode 257. If I make any of the following changes, I get the expected results: - omit umount/mount from the script - replace SSD partition by spinning disk's partition of the same size - use 266MB partition instead of a 55MB partition (still mixed mode) - dump 554321 bytes into file "1" instead of 654321 - dump 7654321 bytes into file "1" instead of 654321 Despite being that fragile, with the above script the results are 100% reproducible here. I wouldn't be surprised if on a different machine one would need different parameter fine-tuning to trigger the problem. I'm happy to test patches :-) -Jan -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
