Package: pm-utils Version: 1.4.1-15 Severity: normal resuming from a hibernation swap partition after the root disk was mounted inbetween can result in file system corruption.
this doesn't happen when following the debian-installer's recommended setup and not dual-booting, but still can happen in practical setups. steps to reproduce ------------------ as far as i can tell, this resuming with anything happening on the disks is not *expected* to go right, but i'll show that this can happen inadvertedly in some setups: * run debian 8.2 installer on a virtual machine (no gui required): create a /boot partition, a luks encrypted partition for swap, and a luks encrypted partition for / with btrfs on top. * boot up (as expected, you're asked for the root and swap passwords) * install pm-utils * pm-hibernate * (this is a good point in time to snapshot the vm) * boot up, but mistype the swap password three times in a row. * when systemd asks for a new swap password, wait for it to timeout. * log in, maybe touch a new file in /, and shut down. * boot up again, now entering the passwords correctly. * wait for the file system corruption that has ensued to bite you, or provoke it like this: > $ ls / > (no error, but a file touched during the boot without resuming doesn't show > up) > # btrfs scrub start / > scrub started on /, fsid ... (pid=...) > [ ... ] BTRFS: bdev /dev/mapper/sda3_crypt errrs: wr 0, rd 0, flush 0, > corrupt 0, gen 1 > [ ... ] BTRFS: bdev /dev/mapper/sda3_crypt errrs: wr 0, rd 0, flush 0, > corrupt 0, gen 1 > [ ... ] BTRFS: unable to fixup (regular) error at logical 13090816 on dev > /dev/mapper/sda3_crypt > # sudo btrfs scrub status / > scrub status for ... > scrub started at ..., running for 10 seconds > total bytes scrubbed: 59.10MiB with 3 errors > error details: super=2 csum=1 > corrected errors: 0, uncorrectable errors: 1, unverified errors: 0 > ERROR: There are uncorrectable errors. (that was my reproducing setup. i've observed the same issues on my sid production machine with kernel 4.1 as well). other situations affected ------------------------- apart from encrypted volumes, such a situation is assumed to occur with other setups as well: 1) swap on external device with bad connection (eg. small fanless system with partitions on usb sticks or desktop with esata hard disks) 2) picking a wrong kernel on resuming that doesn't mount the swap partition 3) booting another os inbetween (that's an obvious then-don't-do-that for people who know what suspend-to-disk does, approaches ---------- * hide bootloader after hibernating[1]. this only solves 2) and 3), provided that grub is the actually employed means of selecting the os. * don't allow the prompt for a new password to expire (doesn't solve any of 1)/2)/3), and requires /etc/crypttab to contain the swap option, which might not be present on older systems) [1] https://bugs.launchpad.net/ubuntu/+source/pm-utils/+bug/42376 how do you think such problems can be avoided? can pm-utils solve this, or is help from the kernel / file system drivers required? best regards chrysn
