There was further discussion in #ubuntu-devel and #ubuntu-release on this review. See irclogs.ubuntu.com for details.
I have reviewed the patch carefully. As far as I can tell, it does what it is supposed to do, and I don't see any bugs. However, this code, particularly the udev_event_apply_format function, seems to have grown organically and could be significantly improved IMHO. This patch makes the state of the code worse, and increasingly fragile. It took hours for me to review the patches for correctness as a consequence, and I think that this and future changes and reviews are far more prone to error than they need to be. Some specifics: 1) The mechanism employed of redirecting s and l (destination buffer and length remaining count) means that the code inside the redirection is incredibly fragile. The code inside the switch should be pulled out into a separate function. Then the redirection would become unnecessary. On IRC, Dan acknowledged that this may require a large number of parameters to be provided to the factored out function, but this is in part my point. Unintended interactions could happen through any of those parameters. Isolating them would be helpful, since that would identify and reduce the interactions that need to be checked. 2) IMHO, the patch should use the sizeof(x)/sizeof(x[0]) idiom to get the buffer lengths, rather than assuming that they are defined by the same macro. I think likely that someone will attempt to change the array length by assuming it is only necessary to change it in one place. Dan suggested on IRC that the opposite point is that changing it from a buffer to a pointer would cause the sizeof idiom to break, but I think it's far more likely that a developer will check how a variable is used before changing its type. 3) The upstream PR refers to tests, but does not add a test for this bug being fixed. Given the fragility of this fix, I think we should do that. We have to maintain this code for another four years in Xenial. If we have to touch this function again in that time, I think we'll need to spend even more time fixing and reviewing it. I realise that the fix has already been accepted upstream. But as we (in Ubuntu) originated the patch, I think we have a responsibility to make it cleaner. Please refactor it and send that upstream. I understand that this SRU is necessary to fix an important regression in Ubuntu, so I don't think it's appropriate to block this given that I can't find a bug in it. However, I've written this up to register my unhappiness with this patch. I appreciate the previous state of the code made it difficult to fix in a better way; IMHO, this is a reason to refactor it first when sending the development fix upstream. Perhaps the SRU would then need to have been a different, more minimal and hacky patch, but that could be focused on minimising regression risk and fixing just the problem at hand rather than the general case. If you add a test for this bug, please bundle it with the next SRU of this package in Xenial. I note that Zesty is not Fix Released. However, since this is a regression fix, I don't think it would be appropriate to wait. Please get the fix landed in Zesty as soon as possible; I certainly expect the Zesty fix to be able to land well before we are ready to release this in to Xenial/Yakkety -updates. Given that this is a regression fix, we might expect to cut short the usual 7-day aging period. However, given the fragility of the code involved and because this is not a straight revert, I don't think that is warranted in this case. ** Changed in: systemd (Ubuntu Xenial) Status: Confirmed => Fix Committed ** Tags added: verification-needed ** Tags removed: verification-needed ** Tags added: regression-update -- You received this bug notification because you are a member of Ubuntu Touch seeded packages, which is subscribed to systemd in Ubuntu. https://bugs.launchpad.net/bugs/1647485 Title: NVMe symlinks broken by devices with spaces in model or serial strings Status in systemd: New Status in systemd package in Ubuntu: Confirmed Status in systemd source package in Trusty: Confirmed Status in systemd source package in Xenial: Fix Committed Status in systemd source package in Yakkety: Fix Committed Status in systemd source package in Zesty: Confirmed Status in systemd package in Debian: New Bug description: [Impact] After including the patch from bug 1642903, NVMe devices that include spaces in their model or serial strings result in incorrect symlinks, e.g. if the model string is "XYZ Corp NVMe drive" then instead of creating: /dev/disk/by-id/nvme-XYZ Corp NVMe drive_SERIAL -> ../../nvme0n1 it creates: /dev/disk/by-id/nvme-XYZ -> ../../nvme0n1 /dev/Corp -> nvme0n1 /dev/NVMe -> nvme0n1 /dev/drive_SERIAL -> nvme0n1 This is because of the way udev handles the SYMLINK value strings; by default, it does not do any whitespace replacement. To enable whitespace replacement of a symlink value, the rule must also include OPTIONS+="string_escape=replace". This is done for 'md' and 'dm' devices in their rules. However, there are no rules that actually want to specify multiple symlinks, and defaulting to not replacing whitespace makes no sense; instead, the default should be to replace all whitespace in each symlink value, unless the rule explicitly specifies OPTIONS+="string_escape=none". [Test Case] This assumes using udev with the patch from bug 1642903. Without this patch, when using a NVMe drive that contains spaces in its model and/or serial strings, check the /dev/disk/by-id/ directory. It should contain a partially-correct symlink to the NVMe drive, with the name up to the first space. All following space-separated parts of the mode/serial string should have symlinks in the /dev/ directory. This is the incorrect behavior. With this patch, check the /dev/disk/by-id/ directory. It should contain a fully-correct symlink to the NVMe drive, and no part of the drive's model/serial number string should be a link in the /dev directory. An example of the correct/incorrect naming is in the Impact section. There should be no other changes to any of the symlinks under /dev before and after this patch. Typical locations for symlinks are /dev/, /dev/disk/by-name/, /dev/disk/by-id/, /dev/disk/by-uuid/, /dev/disk/by-label/ [Regression Potential] Errors in udev rules can lead to an unbootable or otherwise completely broken system if they unintentionally break or clobber existing /dev/disks/ symlinks. [Other Info] This is also tracked with upstream systemd (udev) bug 4833: https://github.com/systemd/systemd/issues/4833 Also note, this can be worked around in individual rules ONLY (i.e. not fixed for all rules) by appending OPTIONS+="string_escape=replace" to each of the NVMe rules with SYMLINK+="..." assignment, e.g.: KERNEL=="nvme*[0-9]n*[0-9]", ENV{DEVTYPE}=="disk", ATTRS{model}=="?*", ENV{ID_SERIAL_SHORT}=="?*", ENV{ID_SERIAL}="$attr{model}_$env{ID_SERIAL_SHORT}", SYMLINK+="disk /by-id/nvme-$env{ID_SERIAL}", OPTIONS+="string_escape=replace" Related bugs: * bug 1642903: introduce disk/by-id (model_serial) symlinks for NVMe drives * bug 1651602: NVMe driver regression for non-smp/1-cpu systems * bug 1649635: export nvme drive model/serial strings via sysfs (trusty) To manage notifications about this bug go to: https://bugs.launchpad.net/systemd/+bug/1647485/+subscriptions -- Mailing list: https://launchpad.net/~touch-packages Post to : touch-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~touch-packages More help : https://help.launchpad.net/ListHelp