Message ID | 20210201185815.382326-1-florian.bezdeka@siemens.com |
---|---|
Headers | show |
Series | wic: warn on usage of Y2038 affected file systems | expand |
01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > Hi ISAR developers, > > this series is the summary of a nice journey through the file system > jungle regarding Y2038 problem. It all began with a warning which is > reported by kernels >= 5.4: > > ext4 filesystem being mounted at (mountpoint) supports timestamps until > 2038 (0x7fffffff) > > I guess that most ISAR layers are using the Debian kernels, so that > warning was not recognized yet or at least not very often. > > When reading this warning I was surprised. Shouldn't a modern file > system like ext4 be Y2038-safe? As it turned out it depends on the > inode size if an ext4 file system is safe or not. So why was the > inode size not sufficient in my case? > > The inode size is chosen during file system generation and depends on > the size of the file system that is going to be created. For details > let's have a look at `man mke2fs`: > > -T usage-type[,...] > Specify how the filesystem is going to be used, so that mke2fs can > choose optimal filesystem parameters for that use. The usage types > that are supported are defined in the configuration file > /etc/mke2fs.conf. The user may specify one or more usage types > using a comma separated list. > > If this option is is not specified, mke2fs will pick a single > default usage type based on the size of the filesystem to be > created. If the filesystem size is less than 3 megabytes, mke2fs > will use the filesystem type floppy. If the filesystem size is > greater than or equal to 3 but less than 512 megabytes, mke2fs(8) > will use the filesystem type small. > > The relevant parts from /etc/mke2fs.conf: > [fs_types] > ... > small = { > blocksize = 1024 > inode_size = 128 > inode_ratio = 4096 > } > ... > > So whenever you create an ext4 file system with less than 512MB in > size you will end up with 128 byte inodes and your file system is > not Y2038-safe. > > The ISAR part: > ext4 may often be used in combination with the expand-on-first-boot > recipe / feature. So whenever creating a small partition (e.g. inside > a wic file) and extending it later may result in a Y2038 affected ext4 > file system. > > That is exactly what happened to me and I would like to make sure that > all other ISAR users are aware of this situation. > > Valid workarounds found so far: > - Tell wic that an partition will grow: > Add `--mkfs-extraopts "-T ext4"` to your wic partition definition > - Set the inode size to 256 (for small ext4 partitions) > Add `--mkfs-extraopts "-I 256"` to your wic partition definition > > The upstream part: > None of the following patches has been sent to any upstream (OE) > mailing lists yet but hopefully that will happen soon. So far: Any > comments welcome! > > Best regards, > Florian > > Florian Bezdeka (2): > wic-img: Forward warnings from wic to bitbake > wic: Warn if an ext filesystem affected by the Y2038 problem is used > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > scripts/lib/wic/partition.py | 38 ++++++++++++++++++++++++++++++++++++ > 2 files changed, 53 insertions(+), 5 deletions(-) > Applied to next, thanks.
Hi all, i never got around to reviewing this. But did we just fork wic? These patches need to go into wic and we later backport them once they are accepted upstream. Maybe they are already ... did not check. When it comes to changing bitbake or wic, we should really not ... We have forks of some files, like the wic plugins and bitbake config, those are fine but should also stay very close to upstream. The recently applied patch from Vijai also violates that. Since the fork of the plugins was not updated with the wic bump and the repair just takes a few bits of what we probably should take. Henning Am Thu, 11 Feb 2021 11:07:52 +0300 schrieb Anton Mikanovich <amikan@ilbers.de>: > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > Hi ISAR developers, > > > > this series is the summary of a nice journey through the file system > > jungle regarding Y2038 problem. It all began with a warning which is > > reported by kernels >= 5.4: > > > > ext4 filesystem being mounted at (mountpoint) supports timestamps > > until 2038 (0x7fffffff) > > > > I guess that most ISAR layers are using the Debian kernels, so that > > warning was not recognized yet or at least not very often. > > > > When reading this warning I was surprised. Shouldn't a modern file > > system like ext4 be Y2038-safe? As it turned out it depends on the > > inode size if an ext4 file system is safe or not. So why was the > > inode size not sufficient in my case? > > > > The inode size is chosen during file system generation and depends > > on the size of the file system that is going to be created. For > > details let's have a look at `man mke2fs`: > > > > -T usage-type[,...] > > Specify how the filesystem is going to be used, so that mke2fs > > can choose optimal filesystem parameters for that use. The usage > > types that are supported are defined in the configuration file > > /etc/mke2fs.conf. The user may specify one or more usage types > > using a comma separated list. > > > > If this option is is not specified, mke2fs will pick a single > > default usage type based on the size of the filesystem to be > > created. If the filesystem size is less than 3 megabytes, > > mke2fs will use the filesystem type floppy. If the filesystem size > > is greater than or equal to 3 but less than 512 megabytes, mke2fs(8) > > will use the filesystem type small. > > > > The relevant parts from /etc/mke2fs.conf: > > [fs_types] > > ... > > small = { > > blocksize = 1024 > > inode_size = 128 > > inode_ratio = 4096 > > } > > ... > > > > So whenever you create an ext4 file system with less than 512MB in > > size you will end up with 128 byte inodes and your file system is > > not Y2038-safe. > > > > The ISAR part: > > ext4 may often be used in combination with the expand-on-first-boot > > recipe / feature. So whenever creating a small partition (e.g. > > inside a wic file) and extending it later may result in a Y2038 > > affected ext4 file system. > > > > That is exactly what happened to me and I would like to make sure > > that all other ISAR users are aware of this situation. > > > > Valid workarounds found so far: > > - Tell wic that an partition will grow: > > Add `--mkfs-extraopts "-T ext4"` to your wic partition > > definition > > - Set the inode size to 256 (for small ext4 partitions) > > Add `--mkfs-extraopts "-I 256"` to your wic partition definition > > > > The upstream part: > > None of the following patches has been sent to any upstream (OE) > > mailing lists yet but hopefully that will happen soon. So far: Any > > comments welcome! > > > > Best regards, > > Florian > > > > Florian Bezdeka (2): > > wic-img: Forward warnings from wic to bitbake > > wic: Warn if an ext filesystem affected by the Y2038 problem is > > used > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > scripts/lib/wic/partition.py | 38 > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > insertions(+), 5 deletions(-) > Applied to next, thanks. >
On 11.02.21 09:23, Henning Schild wrote: > Hi all, > > i never got around to reviewing this. But did we just fork wic? These > patches need to go into wic and we later backport them once they are > accepted upstream. > > Maybe they are already ... did not check. > > When it comes to changing bitbake or wic, we should really not ... We > have forks of some files, like the wic plugins and bitbake config, > those are fine but should also stay very close to upstream. > > The recently applied patch from Vijai also violates that. Since the > fork of the plugins was not updated with the wic bump and the repair > just takes a few bits of what we probably should take. > If you are referring to https://groups.google.com/d/msgid/isar-users/20201126091750.28048-1-Vijaikumar_Kanagarajan%40mentor.com: That one was "only" patching an isar version, though I agree that we should make sure to realign it with the original plugins if we are now imbalanced. This one here is more critical as it changed a formerly vanilla wic file. That should be fixed quickly. Florian, maybe you can propose a similar change to OE upstream? In the meantime, is there a chance to move the changes out of partition.py, to a file that is isar-specific? Jan > Henning > > Am Thu, 11 Feb 2021 11:07:52 +0300 > schrieb Anton Mikanovich <amikan@ilbers.de>: > >> 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: >>> From: Florian Bezdeka <florian.bezdeka@siemens.com> >>> >>> Hi ISAR developers, >>> >>> this series is the summary of a nice journey through the file system >>> jungle regarding Y2038 problem. It all began with a warning which is >>> reported by kernels >= 5.4: >>> >>> ext4 filesystem being mounted at (mountpoint) supports timestamps >>> until 2038 (0x7fffffff) >>> >>> I guess that most ISAR layers are using the Debian kernels, so that >>> warning was not recognized yet or at least not very often. >>> >>> When reading this warning I was surprised. Shouldn't a modern file >>> system like ext4 be Y2038-safe? As it turned out it depends on the >>> inode size if an ext4 file system is safe or not. So why was the >>> inode size not sufficient in my case? >>> >>> The inode size is chosen during file system generation and depends >>> on the size of the file system that is going to be created. For >>> details let's have a look at `man mke2fs`: >>> >>> -T usage-type[,...] >>> Specify how the filesystem is going to be used, so that mke2fs >>> can choose optimal filesystem parameters for that use. The usage >>> types that are supported are defined in the configuration file >>> /etc/mke2fs.conf. The user may specify one or more usage types >>> using a comma separated list. >>> >>> If this option is is not specified, mke2fs will pick a single >>> default usage type based on the size of the filesystem to be >>> created. If the filesystem size is less than 3 megabytes, >>> mke2fs will use the filesystem type floppy. If the filesystem size >>> is greater than or equal to 3 but less than 512 megabytes, mke2fs(8) >>> will use the filesystem type small. >>> >>> The relevant parts from /etc/mke2fs.conf: >>> [fs_types] >>> ... >>> small = { >>> blocksize = 1024 >>> inode_size = 128 >>> inode_ratio = 4096 >>> } >>> ... >>> >>> So whenever you create an ext4 file system with less than 512MB in >>> size you will end up with 128 byte inodes and your file system is >>> not Y2038-safe. >>> >>> The ISAR part: >>> ext4 may often be used in combination with the expand-on-first-boot >>> recipe / feature. So whenever creating a small partition (e.g. >>> inside a wic file) and extending it later may result in a Y2038 >>> affected ext4 file system. >>> >>> That is exactly what happened to me and I would like to make sure >>> that all other ISAR users are aware of this situation. >>> >>> Valid workarounds found so far: >>> - Tell wic that an partition will grow: >>> Add `--mkfs-extraopts "-T ext4"` to your wic partition >>> definition >>> - Set the inode size to 256 (for small ext4 partitions) >>> Add `--mkfs-extraopts "-I 256"` to your wic partition definition >>> >>> The upstream part: >>> None of the following patches has been sent to any upstream (OE) >>> mailing lists yet but hopefully that will happen soon. So far: Any >>> comments welcome! >>> >>> Best regards, >>> Florian >>> >>> Florian Bezdeka (2): >>> wic-img: Forward warnings from wic to bitbake >>> wic: Warn if an ext filesystem affected by the Y2038 problem is >>> used >>> >>> meta/classes/wic-img.bbclass | 20 ++++++++++++++----- >>> scripts/lib/wic/partition.py | 38 >>> ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 >>> insertions(+), 5 deletions(-) >> Applied to next, thanks. >> >
On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > On 11.02.21 09:23, Henning Schild wrote: > > Hi all, > > > > i never got around to reviewing this. But did we just fork wic? These > > patches need to go into wic and we later backport them once they are > > accepted upstream. > > > > Maybe they are already ... did not check. > > > > When it comes to changing bitbake or wic, we should really not ... We > > have forks of some files, like the wic plugins and bitbake config, > > those are fine but should also stay very close to upstream. > > > > The recently applied patch from Vijai also violates that. Since the > > fork of the plugins was not updated with the wic bump and the repair > > just takes a few bits of what we probably should take. > > > > If you are referring to > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fisar-users%2F20201126091750.28048-1-Vijaikumar_Kanagarajan%2540mentor.com&data=04%7C01%7Cflorian.bezdeka%40siemens.com%7Ca5e6b57fc2f34070817c08d8ce6d6dbd%7C38ae3bcd95794fd4addab42e1495d55a%7C1%7C0%7C637486316681424173%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=xERzeBGXiuxTVEC2n4CSuGmlFB9O7h07Hm9ODn33llg%3D&reserved=0: > That one was "only" patching an isar version, though I agree that we > should make sure to realign it with the original plugins if we are now > imbalanced. > > This one here is more critical as it changed a formerly vanilla wic > file. That should be fixed quickly. > > Florian, maybe you can propose a similar change to OE upstream? In the > meantime, is there a chance to move the changes out of partition.py, to > a file that is isar-specific? > I guess the "RFC" tag of this series has been overlooked. It was not intended for merging (yet). Part one (forwarding wic warnings to bitbake) is a pure ISAR change and could be taken as is (if no further comments come up). Sorry for the long description of the series, but if you read closely I already mentioned that the second part should go to OE. I sent it out for feedback collection only. The upstreaming to OE will take some time due to internal clarifications. I never contributed to OE before, so some kind of approval process has to be followed first. At first glance there was no easy way moving the warnings from wic to ISAR. We would have to re-parse the wic template file again and check all the partitions afterwards. wic has all the necessary information at hand so I guess that's way easier. > Jan > > > Henning > > > > Am Thu, 11 Feb 2021 11:07:52 +0300 > > schrieb Anton Mikanovich <amikan@ilbers.de>: > > > > > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > > > > > Hi ISAR developers, > > > > > > > > this series is the summary of a nice journey through the file system > > > > jungle regarding Y2038 problem. It all began with a warning which is > > > > reported by kernels >= 5.4: > > > > > > > > ext4 filesystem being mounted at (mountpoint) supports timestamps > > > > until 2038 (0x7fffffff) > > > > > > > > I guess that most ISAR layers are using the Debian kernels, so that > > > > warning was not recognized yet or at least not very often. > > > > > > > > When reading this warning I was surprised. Shouldn't a modern file > > > > system like ext4 be Y2038-safe? As it turned out it depends on the > > > > inode size if an ext4 file system is safe or not. So why was the > > > > inode size not sufficient in my case? > > > > > > > > The inode size is chosen during file system generation and depends > > > > on the size of the file system that is going to be created. For > > > > details let's have a look at `man mke2fs`: > > > > > > > > -T usage-type[,...] > > > > Specify how the filesystem is going to be used, so that mke2fs > > > > can choose optimal filesystem parameters for that use. The usage > > > > types that are supported are defined in the configuration file > > > > /etc/mke2fs.conf. The user may specify one or more usage types > > > > using a comma separated list. > > > > > > > > If this option is is not specified, mke2fs will pick a single > > > > default usage type based on the size of the filesystem to be > > > > created. If the filesystem size is less than 3 megabytes, > > > > mke2fs will use the filesystem type floppy. If the filesystem size > > > > is greater than or equal to 3 but less than 512 megabytes, mke2fs(8) > > > > will use the filesystem type small. > > > > > > > > The relevant parts from /etc/mke2fs.conf: > > > > [fs_types] > > > > ... > > > > small = { > > > > blocksize = 1024 > > > > inode_size = 128 > > > > inode_ratio = 4096 > > > > } > > > > ... > > > > > > > > So whenever you create an ext4 file system with less than 512MB in > > > > size you will end up with 128 byte inodes and your file system is > > > > not Y2038-safe. > > > > > > > > The ISAR part: > > > > ext4 may often be used in combination with the expand-on-first-boot > > > > recipe / feature. So whenever creating a small partition (e.g. > > > > inside a wic file) and extending it later may result in a Y2038 > > > > affected ext4 file system. > > > > > > > > That is exactly what happened to me and I would like to make sure > > > > that all other ISAR users are aware of this situation. > > > > > > > > Valid workarounds found so far: > > > > - Tell wic that an partition will grow: > > > > Add `--mkfs-extraopts "-T ext4"` to your wic partition > > > > definition > > > > - Set the inode size to 256 (for small ext4 partitions) > > > > Add `--mkfs-extraopts "-I 256"` to your wic partition definition > > > > > > > > The upstream part: > > > > None of the following patches has been sent to any upstream (OE) > > > > mailing lists yet but hopefully that will happen soon. So far: Any > > > > comments welcome! > > > > > > > > Best regards, > > > > Florian > > > > > > > > Florian Bezdeka (2): > > > > wic-img: Forward warnings from wic to bitbake > > > > wic: Warn if an ext filesystem affected by the Y2038 problem is > > > > used > > > > > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > > > scripts/lib/wic/partition.py | 38 > > > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > > > insertions(+), 5 deletions(-) > > > Applied to next, thanks. > > > > > > >
Am Thu, 11 Feb 2021 10:57:31 +0100 schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" <florian.bezdeka@siemens.com>: > On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > > On 11.02.21 09:23, Henning Schild wrote: > > > Hi all, > > > > > > i never got around to reviewing this. But did we just fork wic? > > > These patches need to go into wic and we later backport them once > > > they are accepted upstream. > > > > > > Maybe they are already ... did not check. > > > > > > When it comes to changing bitbake or wic, we should really not > > > ... We have forks of some files, like the wic plugins and bitbake > > > config, those are fine but should also stay very close to > > > upstream. > > > > > > The recently applied patch from Vijai also violates that. Since > > > the fork of the plugins was not updated with the wic bump and the > > > repair just takes a few bits of what we probably should take. > > > > > > > If you are referring to > > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fisar-users%2F20201126091750.28048-1-Vijaikumar_Kanagarajan%2540mentor.com&data=04%7C01%7Cde173c00-e982-4fda-8644-47edf4671d63%40ad011.siemens.com%7Cd67820e7b5d841cf320f08d8ce7372f9%7C38ae3bcd95794fd4addab42e1495d55a%7C1%7C0%7C637486342521796327%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=YQM5jQg9YSx6f9FiuYaduPEccCnNspRle4ZH8ES0nH4%3D&reserved=0: > > That one was "only" patching an isar version, though I agree that we > > should make sure to realign it with the original plugins if we are > > now imbalanced. > > > > This one here is more critical as it changed a formerly vanilla wic > > file. That should be fixed quickly. > > > > Florian, maybe you can propose a similar change to OE upstream? In > > the meantime, is there a chance to move the changes out of > > partition.py, to a file that is isar-specific? > > > > I guess the "RFC" tag of this series has been overlooked. It was not > intended for merging (yet). Part one (forwarding wic warnings to > bitbake) is a pure ISAR change and could be taken as is (if no further > comments come up). I guess that calls for a revert. And for more attention on the maintainers side. > Sorry for the long description of the series, but if you read closely > I already mentioned that the second part should go to OE. I sent it > out for feedback collection only. > > The upstreaming to OE will take some time due to internal > clarifications. I never contributed to OE before, so some kind of > approval process has to be followed first. > > At first glance there was no easy way moving the warnings from wic to > ISAR. We would have to re-parse the wic template file again and check > all the partitions afterwards. wic has all the necessary information > at hand so I guess that's way easier. I guess it can be moved into a task after wic. Here one would need to parse the partition table, which kind of sucks. "losetup" or "kpartx" might help but will not work in kas-container setups because they need root. We once had patches allowing wic to retain all partition images instead of throwing them away after disk assembly. Having a switch for wic to say ... do those partitions ... later do the disk would be generic, allow hooking in this and other things. Isar also has a class that creates ext4 images without, after which such a check should also be done. Is ext4 the only fs we care about? We have some layers doing ubifs, squashfs and all sorts of funny things. Maybe the kernel does warn "on device" so we could have a systemd unit warning for all filesystems ... which would probably best find its place in the kernel and or debian. Henning > > > Jan > > > > > Henning > > > > > > Am Thu, 11 Feb 2021 11:07:52 +0300 > > > schrieb Anton Mikanovich <amikan@ilbers.de>: > > > > > > > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > > > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > > > > > > > Hi ISAR developers, > > > > > > > > > > this series is the summary of a nice journey through the file > > > > > system jungle regarding Y2038 problem. It all began with a > > > > > warning which is reported by kernels >= 5.4: > > > > > > > > > > ext4 filesystem being mounted at (mountpoint) supports > > > > > timestamps until 2038 (0x7fffffff) > > > > > > > > > > I guess that most ISAR layers are using the Debian kernels, > > > > > so that warning was not recognized yet or at least not very > > > > > often. > > > > > > > > > > When reading this warning I was surprised. Shouldn't a modern > > > > > file system like ext4 be Y2038-safe? As it turned out it > > > > > depends on the inode size if an ext4 file system is safe or > > > > > not. So why was the inode size not sufficient in my case? > > > > > > > > > > The inode size is chosen during file system generation and > > > > > depends on the size of the file system that is going to be > > > > > created. For details let's have a look at `man mke2fs`: > > > > > > > > > > -T usage-type[,...] > > > > > Specify how the filesystem is going to be used, so that > > > > > mke2fs can choose optimal filesystem parameters for that use. > > > > > The usage types that are supported are defined in the > > > > > configuration file /etc/mke2fs.conf. The user may specify one > > > > > or more usage types using a comma separated list. > > > > > > > > > > If this option is is not specified, mke2fs will pick a > > > > > single default usage type based on the size of the filesystem > > > > > to be created. If the filesystem size is less than 3 > > > > > megabytes, mke2fs will use the filesystem type floppy. If the > > > > > filesystem size is greater than or equal to 3 but less than > > > > > 512 megabytes, mke2fs(8) will use the filesystem type small. > > > > > > > > > > The relevant parts from /etc/mke2fs.conf: > > > > > [fs_types] > > > > > ... > > > > > small = { > > > > > blocksize = 1024 > > > > > inode_size = 128 > > > > > inode_ratio = 4096 > > > > > } > > > > > ... > > > > > > > > > > So whenever you create an ext4 file system with less than > > > > > 512MB in size you will end up with 128 byte inodes and your > > > > > file system is not Y2038-safe. > > > > > > > > > > The ISAR part: > > > > > ext4 may often be used in combination with the > > > > > expand-on-first-boot recipe / feature. So whenever creating a > > > > > small partition (e.g. inside a wic file) and extending it > > > > > later may result in a Y2038 affected ext4 file system. > > > > > > > > > > That is exactly what happened to me and I would like to make > > > > > sure that all other ISAR users are aware of this situation. > > > > > > > > > > Valid workarounds found so far: > > > > > - Tell wic that an partition will grow: > > > > > Add `--mkfs-extraopts "-T ext4"` to your wic partition > > > > > definition > > > > > - Set the inode size to 256 (for small ext4 partitions) > > > > > Add `--mkfs-extraopts "-I 256"` to your wic partition > > > > > definition > > > > > > > > > > The upstream part: > > > > > None of the following patches has been sent to any upstream > > > > > (OE) mailing lists yet but hopefully that will happen soon. > > > > > So far: Any comments welcome! > > > > > > > > > > Best regards, > > > > > Florian > > > > > > > > > > Florian Bezdeka (2): > > > > > wic-img: Forward warnings from wic to bitbake > > > > > wic: Warn if an ext filesystem affected by the Y2038 > > > > > problem is used > > > > > > > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > > > > scripts/lib/wic/partition.py | 38 > > > > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > > > > insertions(+), 5 deletions(-) > > > > Applied to next, thanks. > > > > > > > > > > > >
On Thu, 2021-02-11 at 11:21 +0100, Henning Schild wrote: > Am Thu, 11 Feb 2021 10:57:31 +0100 > schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > <florian.bezdeka@siemens.com>: > > > On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > > > On 11.02.21 09:23, Henning Schild wrote: > > > > Hi all, > > > > > > > > i never got around to reviewing this. But did we just fork wic? > > > > These patches need to go into wic and we later backport them once > > > > they are accepted upstream. > > > > > > > > Maybe they are already ... did not check. > > > > > > > > When it comes to changing bitbake or wic, we should really not > > > > ... We have forks of some files, like the wic plugins and bitbake > > > > config, those are fine but should also stay very close to > > > > upstream. > > > > > > > > The recently applied patch from Vijai also violates that. Since > > > > the fork of the plugins was not updated with the wic bump and the > > > > repair just takes a few bits of what we probably should take. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > If you are referring to > > > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fisar-users%2F20201126091750.28048-1-Vijaikumar_Kanagarajan%2540mentor.com&data=04%7C01%7Cflorian.bezdeka%40siemens.com%7C48d6471d1d4341e4445d08d8ce778b07%7C38ae3bcd95794fd4addab42e1495d55a%7C1%7C0%7C637486360122035313%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=61t42JDRuSWYJF%2Ff6rE6A7A9o0%2BlDF7zKwN85LVo%2BiU%3D&reserved=0: > > > That one was "only" patching an isar version, though I agree that we > > > should make sure to realign it with the original plugins if we are > > > now imbalanced. > > > > > > This one here is more critical as it changed a formerly vanilla wic > > > file. That should be fixed quickly. > > > > > > Florian, maybe you can propose a similar change to OE upstream? In > > > the meantime, is there a chance to move the changes out of > > > partition.py, to a file that is isar-specific? > > > > > > > > > > > > > > > > > > > > > > > > > > > > I guess the "RFC" tag of this series has been overlooked. It was not > > intended for merging (yet). Part one (forwarding wic warnings to > > bitbake) is a pure ISAR change and could be taken as is (if no further > > comments come up). > > I guess that calls for a revert. And for more attention on the > maintainers side. > > > Sorry for the long description of the series, but if you read closely > > I already mentioned that the second part should go to OE. I sent it > > out for feedback collection only. > > > > The upstreaming to OE will take some time due to internal > > clarifications. I never contributed to OE before, so some kind of > > approval process has to be followed first. > > > > At first glance there was no easy way moving the warnings from wic to > > ISAR. We would have to re-parse the wic template file again and check > > all the partitions afterwards. wic has all the necessary information > > at hand so I guess that's way easier. > > I guess it can be moved into a task after wic. Here one would need to > parse the partition table, which kind of sucks. "losetup" or "kpartx" > might help but will not work in kas-container setups because they need > root. > We once had patches allowing wic to retain all partition images instead > of throwing them away after disk assembly. Having a switch for wic to > say ... do those partitions ... later do the disk would be generic, > allow hooking in this and other things. > > Isar also has a class that creates ext4 images without, after which such > a check should also be done. Yes. But instead of spreading the warnings around it would be nice to have a single place where we could do the Y2038 checks. So maybe it should be a base feature of "image.bbclass"? Or ext4-img.bbclass should call wic instead of the mke2fs utilities directly? BTW: The name ext4-img.bbclass is kind of misleading. You could simply create ext{2,3} file systems by setting MKE2FS_ARGS to something like "-t ext2". > > Is ext4 the only fs we care about? We have some layers doing ubifs, > squashfs and all sorts of funny things. Up to now I cared about the filesystems supported by wic. So ext{2,3,4}, btrfs and squashfs. squashfs will overflow in 2106 (u32) and btrfs will "never" overflow (u64). ubifs is similar to btrfs, so not affected by Y2038. > > Maybe the kernel does warn "on device" so we could have a systemd unit > warning for all filesystems ... which would probably best find its > place in the kernel and or debian. At least for affected ext file systems the kernel will warn (on mount). But I considered that as "too late". > > Henning > > > > > > Jan > > > > > > > > > > > > > > > > > > > > > > > > > > > > Henning > > > > > > > > Am Thu, 11 Feb 2021 11:07:52 +0300 > > > > schrieb Anton Mikanovich <amikan@ilbers.de>: > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > > > > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > > > > > > > > > Hi ISAR developers, > > > > > > > > > > > > this series is the summary of a nice journey through the file > > > > > > system jungle regarding Y2038 problem. It all began with a > > > > > > warning which is reported by kernels >= 5.4: > > > > > > > > > > > > ext4 filesystem being mounted at (mountpoint) supports > > > > > > timestamps until 2038 (0x7fffffff) > > > > > > > > > > > > I guess that most ISAR layers are using the Debian kernels, > > > > > > so that warning was not recognized yet or at least not very > > > > > > often. > > > > > > > > > > > > When reading this warning I was surprised. Shouldn't a modern > > > > > > file system like ext4 be Y2038-safe? As it turned out it > > > > > > depends on the inode size if an ext4 file system is safe or > > > > > > not. So why was the inode size not sufficient in my case? > > > > > > > > > > > > The inode size is chosen during file system generation and > > > > > > depends on the size of the file system that is going to be > > > > > > created. For details let's have a look at `man mke2fs`: > > > > > > > > > > > > -T usage-type[,...] > > > > > > Specify how the filesystem is going to be used, so that > > > > > > mke2fs can choose optimal filesystem parameters for that use. > > > > > > The usage types that are supported are defined in the > > > > > > configuration file /etc/mke2fs.conf. The user may specify one > > > > > > or more usage types using a comma separated list. > > > > > > > > > > > > If this option is is not specified, mke2fs will pick a > > > > > > single default usage type based on the size of the filesystem > > > > > > to be created. If the filesystem size is less than 3 > > > > > > megabytes, mke2fs will use the filesystem type floppy. If the > > > > > > filesystem size is greater than or equal to 3 but less than > > > > > > 512 megabytes, mke2fs(8) will use the filesystem type small. > > > > > > > > > > > > The relevant parts from /etc/mke2fs.conf: > > > > > > [fs_types] > > > > > > ... > > > > > > small = { > > > > > > blocksize = 1024 > > > > > > inode_size = 128 > > > > > > inode_ratio = 4096 > > > > > > } > > > > > > ... > > > > > > > > > > > > So whenever you create an ext4 file system with less than > > > > > > 512MB in size you will end up with 128 byte inodes and your > > > > > > file system is not Y2038-safe. > > > > > > > > > > > > The ISAR part: > > > > > > ext4 may often be used in combination with the > > > > > > expand-on-first-boot recipe / feature. So whenever creating a > > > > > > small partition (e.g. inside a wic file) and extending it > > > > > > later may result in a Y2038 affected ext4 file system. > > > > > > > > > > > > That is exactly what happened to me and I would like to make > > > > > > sure that all other ISAR users are aware of this situation. > > > > > > > > > > > > Valid workarounds found so far: > > > > > > - Tell wic that an partition will grow: > > > > > > Add `--mkfs-extraopts "-T ext4"` to your wic partition > > > > > > definition > > > > > > - Set the inode size to 256 (for small ext4 partitions) > > > > > > Add `--mkfs-extraopts "-I 256"` to your wic partition > > > > > > definition > > > > > > > > > > > > The upstream part: > > > > > > None of the following patches has been sent to any upstream > > > > > > (OE) mailing lists yet but hopefully that will happen soon. > > > > > > So far: Any comments welcome! > > > > > > > > > > > > Best regards, > > > > > > Florian > > > > > > > > > > > > Florian Bezdeka (2): > > > > > > wic-img: Forward warnings from wic to bitbake > > > > > > wic: Warn if an ext filesystem affected by the Y2038 > > > > > > problem is used > > > > > > > > > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > > > > > scripts/lib/wic/partition.py | 38 > > > > > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > > > > > insertions(+), 5 deletions(-) > > > > > Applied to next, thanks. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
On Thu, 2021-02-11 at 12:47 +0000, [ext] florian.bezdeka@siemens.com wrote: > On Thu, 2021-02-11 at 11:21 +0100, Henning Schild wrote: > > Am Thu, 11 Feb 2021 10:57:31 +0100 > > schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > > <florian.bezdeka@siemens.com>: > > > > > On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > > > > On 11.02.21 09:23, Henning Schild wrote: > > > > > Hi all, > > > > > > > > > > i never got around to reviewing this. But did we just fork wic? > > > > > These patches need to go into wic and we later backport them once > > > > > they are accepted upstream. > > > > > > > > > > Maybe they are already ... did not check. > > > > > > > > > > When it comes to changing bitbake or wic, we should really not > > > > > ... We have forks of some files, like the wic plugins and bitbake > > > > > config, those are fine but should also stay very close to > > > > > upstream. > > > > > > > > > > The recently applied patch from Vijai also violates that. Since > > > > > the fork of the plugins was not updated with the wic bump and the > > > > > repair just takes a few bits of what we probably should take. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > If you are referring to > > > > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fisar-users%2F20201126091750.28048-1-Vijaikumar_Kanagarajan%2540mentor.com&data=04%7C01%7Cflorian.bezdeka%40siemens.com%7C070614d51b4b45045bdf08d8ce8b657f%7C38ae3bcd95794fd4addab42e1495d55a%7C1%7C0%7C637486445390373228%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=3H1fH4EhWZMF%2BgUYpPcej9py24HSVAgIjwOwiELlOqs%3D&reserved=0: > > > > That one was "only" patching an isar version, though I agree that we > > > > should make sure to realign it with the original plugins if we are > > > > now imbalanced. > > > > > > > > This one here is more critical as it changed a formerly vanilla wic > > > > file. That should be fixed quickly. > > > > > > > > Florian, maybe you can propose a similar change to OE upstream? In > > > > the meantime, is there a chance to move the changes out of > > > > partition.py, to a file that is isar-specific? > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > I guess the "RFC" tag of this series has been overlooked. It was not > > > intended for merging (yet). Part one (forwarding wic warnings to > > > bitbake) is a pure ISAR change and could be taken as is (if no further > > > comments come up). > > > > I guess that calls for a revert. And for more attention on the > > maintainers side. > > > > > Sorry for the long description of the series, but if you read closely > > > I already mentioned that the second part should go to OE. I sent it > > > out for feedback collection only. > > > > > > The upstreaming to OE will take some time due to internal > > > clarifications. I never contributed to OE before, so some kind of > > > approval process has to be followed first. > > > > > > At first glance there was no easy way moving the warnings from wic to > > > ISAR. We would have to re-parse the wic template file again and check > > > all the partitions afterwards. wic has all the necessary information > > > at hand so I guess that's way easier. > > > > I guess it can be moved into a task after wic. Here one would need to > > parse the partition table, which kind of sucks. "losetup" or "kpartx" > > might help but will not work in kas-container setups because they need > > root. > > We once had patches allowing wic to retain all partition images instead > > of throwing them away after disk assembly. Having a switch for wic to > > say ... do those partitions ... later do the disk would be generic, > > allow hooking in this and other things. > > > > Isar also has a class that creates ext4 images without, after which such > > a check should also be done. > > Yes. But instead of spreading the warnings around it would be nice to > have a single place where we could do the Y2038 checks. So maybe it > should be a base feature of "image.bbclass"? Or ext4-img.bbclass should > call wic instead of the mke2fs utilities directly? > > BTW: The name ext4-img.bbclass is kind of misleading. You could simply > create ext{2,3} file systems by setting MKE2FS_ARGS to something like > "-t ext2". > > > > > Is ext4 the only fs we care about? We have some layers doing ubifs, > > squashfs and all sorts of funny things. > > Up to now I cared about the filesystems supported by wic. So > ext{2,3,4}, btrfs and squashfs. squashfs will overflow in 2106 (u32) > and btrfs will "never" overflow (u64). > > ubifs is similar to btrfs, so not affected by Y2038. > > > > > Maybe the kernel does warn "on device" so we could have a systemd unit > > warning for all filesystems ... which would probably best find its > > place in the kernel and or debian. > > At least for affected ext file systems the kernel will warn (on mount). > But I considered that as "too late". To be more specific: Linux >= 5.4 warns. That's why I guess that many projects did not realize that they are already affected by the Y2038 problem because of older kernel versions. > > > > > Henning > > > > > > > > > Jan > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Henning > > > > > > > > > > Am Thu, 11 Feb 2021 11:07:52 +0300 > > > > > schrieb Anton Mikanovich <amikan@ilbers.de>: > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > > > > > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > > > > > > > > > > > Hi ISAR developers, > > > > > > > > > > > > > > this series is the summary of a nice journey through the file > > > > > > > system jungle regarding Y2038 problem. It all began with a > > > > > > > warning which is reported by kernels >= 5.4: > > > > > > > > > > > > > > ext4 filesystem being mounted at (mountpoint) supports > > > > > > > timestamps until 2038 (0x7fffffff) > > > > > > > > > > > > > > I guess that most ISAR layers are using the Debian kernels, > > > > > > > so that warning was not recognized yet or at least not very > > > > > > > often. > > > > > > > > > > > > > > When reading this warning I was surprised. Shouldn't a modern > > > > > > > file system like ext4 be Y2038-safe? As it turned out it > > > > > > > depends on the inode size if an ext4 file system is safe or > > > > > > > not. So why was the inode size not sufficient in my case? > > > > > > > > > > > > > > The inode size is chosen during file system generation and > > > > > > > depends on the size of the file system that is going to be > > > > > > > created. For details let's have a look at `man mke2fs`: > > > > > > > > > > > > > > -T usage-type[,...] > > > > > > > Specify how the filesystem is going to be used, so that > > > > > > > mke2fs can choose optimal filesystem parameters for that use. > > > > > > > The usage types that are supported are defined in the > > > > > > > configuration file /etc/mke2fs.conf. The user may specify one > > > > > > > or more usage types using a comma separated list. > > > > > > > > > > > > > > If this option is is not specified, mke2fs will pick a > > > > > > > single default usage type based on the size of the filesystem > > > > > > > to be created. If the filesystem size is less than 3 > > > > > > > megabytes, mke2fs will use the filesystem type floppy. If the > > > > > > > filesystem size is greater than or equal to 3 but less than > > > > > > > 512 megabytes, mke2fs(8) will use the filesystem type small. > > > > > > > > > > > > > > The relevant parts from /etc/mke2fs.conf: > > > > > > > [fs_types] > > > > > > > ... > > > > > > > small = { > > > > > > > blocksize = 1024 > > > > > > > inode_size = 128 > > > > > > > inode_ratio = 4096 > > > > > > > } > > > > > > > ... > > > > > > > > > > > > > > So whenever you create an ext4 file system with less than > > > > > > > 512MB in size you will end up with 128 byte inodes and your > > > > > > > file system is not Y2038-safe. > > > > > > > > > > > > > > The ISAR part: > > > > > > > ext4 may often be used in combination with the > > > > > > > expand-on-first-boot recipe / feature. So whenever creating a > > > > > > > small partition (e.g. inside a wic file) and extending it > > > > > > > later may result in a Y2038 affected ext4 file system. > > > > > > > > > > > > > > That is exactly what happened to me and I would like to make > > > > > > > sure that all other ISAR users are aware of this situation. > > > > > > > > > > > > > > Valid workarounds found so far: > > > > > > > - Tell wic that an partition will grow: > > > > > > > Add `--mkfs-extraopts "-T ext4"` to your wic partition > > > > > > > definition > > > > > > > - Set the inode size to 256 (for small ext4 partitions) > > > > > > > Add `--mkfs-extraopts "-I 256"` to your wic partition > > > > > > > definition > > > > > > > > > > > > > > The upstream part: > > > > > > > None of the following patches has been sent to any upstream > > > > > > > (OE) mailing lists yet but hopefully that will happen soon. > > > > > > > So far: Any comments welcome! > > > > > > > > > > > > > > Best regards, > > > > > > > Florian > > > > > > > > > > > > > > Florian Bezdeka (2): > > > > > > > wic-img: Forward warnings from wic to bitbake > > > > > > > wic: Warn if an ext filesystem affected by the Y2038 > > > > > > > problem is used > > > > > > > > > > > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > > > > > > scripts/lib/wic/partition.py | 38 > > > > > > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > > > > > > insertions(+), 5 deletions(-) > > > > > > Applied to next, thanks. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
Am Thu, 11 Feb 2021 14:31:58 +0100 schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" <florian.bezdeka@siemens.com>: > On Thu, 2021-02-11 at 12:47 +0000, [ext] florian.bezdeka@siemens.com > wrote: > > On Thu, 2021-02-11 at 11:21 +0100, Henning Schild wrote: > > > Am Thu, 11 Feb 2021 10:57:31 +0100 > > > schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > > > <florian.bezdeka@siemens.com>: > > > > > > > On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > > > > > On 11.02.21 09:23, Henning Schild wrote: > > > > > > Hi all, > > > > > > > > > > > > i never got around to reviewing this. But did we just fork > > > > > > wic? These patches need to go into wic and we later > > > > > > backport them once they are accepted upstream. > > > > > > > > > > > > Maybe they are already ... did not check. > > > > > > > > > > > > When it comes to changing bitbake or wic, we should really > > > > > > not ... We have forks of some files, like the wic plugins > > > > > > and bitbake config, those are fine but should also stay > > > > > > very close to upstream. > > > > > > > > > > > > The recently applied patch from Vijai also violates that. > > > > > > Since the fork of the plugins was not updated with the wic > > > > > > bump and the repair just takes a few bits of what we > > > > > > probably should take. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > If you are referring to > > > > > https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgroups.google.com%2Fd%2Fmsgid%2Fisar-users%2F20201126091750.28048-1-Vijaikumar_Kanagarajan%2540mentor.com&data=04%7C01%7Cde173c00-e982-4fda-8644-47edf4671d63%40ad011.siemens.com%7Ca81479e099ce4a32a67608d8ce916870%7C38ae3bcd95794fd4addab42e1495d55a%7C1%7C0%7C637486471194236324%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=QvpGJS2QMLxBNb0EQZVilcyZr3CBN%2FZ48rSYlOUVisU%3D&reserved=0: > > > > > That one was "only" patching an isar version, though I agree > > > > > that we should make sure to realign it with the original > > > > > plugins if we are now imbalanced. > > > > > > > > > > This one here is more critical as it changed a formerly > > > > > vanilla wic file. That should be fixed quickly. > > > > > > > > > > Florian, maybe you can propose a similar change to OE > > > > > upstream? In the meantime, is there a chance to move the > > > > > changes out of partition.py, to a file that is isar-specific? > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > I guess the "RFC" tag of this series has been overlooked. It > > > > was not intended for merging (yet). Part one (forwarding wic > > > > warnings to bitbake) is a pure ISAR change and could be taken > > > > as is (if no further comments come up). > > > > > > I guess that calls for a revert. And for more attention on the > > > maintainers side. > > > > > > > Sorry for the long description of the series, but if you read > > > > closely I already mentioned that the second part should go to > > > > OE. I sent it out for feedback collection only. > > > > > > > > The upstreaming to OE will take some time due to internal > > > > clarifications. I never contributed to OE before, so some kind > > > > of approval process has to be followed first. > > > > > > > > At first glance there was no easy way moving the warnings from > > > > wic to ISAR. We would have to re-parse the wic template file > > > > again and check all the partitions afterwards. wic has all the > > > > necessary information at hand so I guess that's way easier. > > > > > > I guess it can be moved into a task after wic. Here one would > > > need to parse the partition table, which kind of sucks. "losetup" > > > or "kpartx" might help but will not work in kas-container setups > > > because they need root. > > > We once had patches allowing wic to retain all partition images > > > instead of throwing them away after disk assembly. Having a > > > switch for wic to say ... do those partitions ... later do the > > > disk would be generic, allow hooking in this and other things. > > > > > > Isar also has a class that creates ext4 images without, after > > > which such a check should also be done. > > > > Yes. But instead of spreading the warnings around it would be nice > > to have a single place where we could do the Y2038 checks. So maybe > > it should be a base feature of "image.bbclass"? Or ext4-img.bbclass > > should call wic instead of the mke2fs utilities directly? > > > > BTW: The name ext4-img.bbclass is kind of misleading. You could > > simply create ext{2,3} file systems by setting MKE2FS_ARGS to > > something like "-t ext2". > > > > > > > > Is ext4 the only fs we care about? We have some layers doing > > > ubifs, squashfs and all sorts of funny things. > > > > Up to now I cared about the filesystems supported by wic. So > > ext{2,3,4}, btrfs and squashfs. squashfs will overflow in 2106 (u32) > > and btrfs will "never" overflow (u64). > > > > ubifs is similar to btrfs, so not affected by Y2038. > > > > > > > > Maybe the kernel does warn "on device" so we could have a systemd > > > unit warning for all filesystems ... which would probably best > > > find its place in the kernel and or debian. > > > > At least for affected ext file systems the kernel will warn (on > > mount). But I considered that as "too late". > > To be more specific: Linux >= 5.4 warns. That's why I guess that many > projects did not realize that they are already affected by the Y2038 > problem because of older kernel versions. Which sounds like that warning needs backporting into the debian10 kernel and maybe cip. Not sure Isar is the best place, but a valid one that could help. Maybe mkfs could warn ... as well. Henning > > > > > > > > Henning > > > > > > > > > > > > Jan > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > Henning > > > > > > > > > > > > Am Thu, 11 Feb 2021 11:07:52 +0300 > > > > > > schrieb Anton Mikanovich <amikan@ilbers.de>: > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > 01.02.2021 21:58, florian.bezdeka@siemens.com wrote: > > > > > > > > From: Florian Bezdeka <florian.bezdeka@siemens.com> > > > > > > > > > > > > > > > > Hi ISAR developers, > > > > > > > > > > > > > > > > this series is the summary of a nice journey through > > > > > > > > the file system jungle regarding Y2038 problem. It all > > > > > > > > began with a warning which is reported by kernels >= > > > > > > > > 5.4: > > > > > > > > > > > > > > > > ext4 filesystem being mounted at (mountpoint) supports > > > > > > > > timestamps until 2038 (0x7fffffff) > > > > > > > > > > > > > > > > I guess that most ISAR layers are using the Debian > > > > > > > > kernels, so that warning was not recognized yet or at > > > > > > > > least not very often. > > > > > > > > > > > > > > > > When reading this warning I was surprised. Shouldn't a > > > > > > > > modern file system like ext4 be Y2038-safe? As it > > > > > > > > turned out it depends on the inode size if an ext4 file > > > > > > > > system is safe or not. So why was the inode size not > > > > > > > > sufficient in my case? > > > > > > > > > > > > > > > > The inode size is chosen during file system generation > > > > > > > > and depends on the size of the file system that is > > > > > > > > going to be created. For details let's have a look at > > > > > > > > `man mke2fs`: > > > > > > > > > > > > > > > > -T usage-type[,...] > > > > > > > > Specify how the filesystem is going to be used, so > > > > > > > > that mke2fs can choose optimal filesystem parameters > > > > > > > > for that use. The usage types that are supported are > > > > > > > > defined in the configuration file /etc/mke2fs.conf. The > > > > > > > > user may specify one or more usage types using a comma > > > > > > > > separated list. > > > > > > > > > > > > > > > > If this option is is not specified, mke2fs will > > > > > > > > pick a single default usage type based on the size of > > > > > > > > the filesystem to be created. If the filesystem size is > > > > > > > > less than 3 megabytes, mke2fs will use the filesystem > > > > > > > > type floppy. If the filesystem size is greater than or > > > > > > > > equal to 3 but less than 512 megabytes, mke2fs(8) will > > > > > > > > use the filesystem type small. > > > > > > > > > > > > > > > > The relevant parts from /etc/mke2fs.conf: > > > > > > > > [fs_types] > > > > > > > > ... > > > > > > > > small = { > > > > > > > > blocksize = 1024 > > > > > > > > inode_size = 128 > > > > > > > > inode_ratio = 4096 > > > > > > > > } > > > > > > > > ... > > > > > > > > > > > > > > > > So whenever you create an ext4 file system with less > > > > > > > > than 512MB in size you will end up with 128 byte inodes > > > > > > > > and your file system is not Y2038-safe. > > > > > > > > > > > > > > > > The ISAR part: > > > > > > > > ext4 may often be used in combination with the > > > > > > > > expand-on-first-boot recipe / feature. So whenever > > > > > > > > creating a small partition (e.g. inside a wic file) and > > > > > > > > extending it later may result in a Y2038 affected ext4 > > > > > > > > file system. > > > > > > > > > > > > > > > > That is exactly what happened to me and I would like to > > > > > > > > make sure that all other ISAR users are aware of this > > > > > > > > situation. > > > > > > > > > > > > > > > > Valid workarounds found so far: > > > > > > > > - Tell wic that an partition will grow: > > > > > > > > Add `--mkfs-extraopts "-T ext4"` to your wic > > > > > > > > partition definition > > > > > > > > - Set the inode size to 256 (for small ext4 > > > > > > > > partitions) Add `--mkfs-extraopts "-I 256"` to your wic > > > > > > > > partition definition > > > > > > > > > > > > > > > > The upstream part: > > > > > > > > None of the following patches has been sent to any > > > > > > > > upstream (OE) mailing lists yet but hopefully that will > > > > > > > > happen soon. So far: Any comments welcome! > > > > > > > > > > > > > > > > Best regards, > > > > > > > > Florian > > > > > > > > > > > > > > > > Florian Bezdeka (2): > > > > > > > > wic-img: Forward warnings from wic to bitbake > > > > > > > > wic: Warn if an ext filesystem affected by the Y2038 > > > > > > > > problem is used > > > > > > > > > > > > > > > > meta/classes/wic-img.bbclass | 20 ++++++++++++++----- > > > > > > > > scripts/lib/wic/partition.py | 38 > > > > > > > > ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 > > > > > > > > insertions(+), 5 deletions(-) > > > > > > > Applied to next, thanks. > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >
On 11.02.21 15:13, Henning Schild wrote: > Am Thu, 11 Feb 2021 14:31:58 +0100 > schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > <florian.bezdeka@siemens.com>: > >> On Thu, 2021-02-11 at 12:47 +0000, [ext] florian.bezdeka@siemens.com >> wrote: >>> On Thu, 2021-02-11 at 11:21 +0100, Henning Schild wrote: >>>> Am Thu, 11 Feb 2021 10:57:31 +0100 >>>> schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" >>>> <florian.bezdeka@siemens.com>: >>>> >>>>> On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: >>>>>> On 11.02.21 09:23, Henning Schild wrote: >>>>>>> Hi all, >>>>>>> >>>>>>> i never got around to reviewing this. But did we just fork >>>>>>> wic? These patches need to go into wic and we later >>>>>>> backport them once they are accepted upstream. >>>>>>> >>>>>>> Maybe they are already ... did not check. >>>>>>> >>>>>>> When it comes to changing bitbake or wic, we should really >>>>>>> not ... We have forks of some files, like the wic plugins >>>>>>> and bitbake config, those are fine but should also stay >>>>>>> very close to upstream. >>>>>>> >>>>>>> The recently applied patch from Vijai also violates that. >>>>>>> Since the fork of the plugins was not updated with the wic >>>>>>> bump and the repair just takes a few bits of what we >>>>>>> probably should take. >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>>> If you are referring to >>>>>> https://groups.google.com/d/msgid/isar-users/20201126091750.28048-1-Vijaikumar_Kanagarajan%40mentor.com >>>>>> That one was "only" patching an isar version, though I agree >>>>>> that we should make sure to realign it with the original >>>>>> plugins if we are now imbalanced. >>>>>> >>>>>> This one here is more critical as it changed a formerly >>>>>> vanilla wic file. That should be fixed quickly. >>>>>> >>>>>> Florian, maybe you can propose a similar change to OE >>>>>> upstream? In the meantime, is there a chance to move the >>>>>> changes out of partition.py, to a file that is isar-specific? >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>> >>>>> I guess the "RFC" tag of this series has been overlooked. It >>>>> was not intended for merging (yet). Part one (forwarding wic >>>>> warnings to bitbake) is a pure ISAR change and could be taken >>>>> as is (if no further comments come up). >>>> >>>> I guess that calls for a revert. And for more attention on the >>>> maintainers side. >>>> >>>>> Sorry for the long description of the series, but if you read >>>>> closely I already mentioned that the second part should go to >>>>> OE. I sent it out for feedback collection only. >>>>> >>>>> The upstreaming to OE will take some time due to internal >>>>> clarifications. I never contributed to OE before, so some kind >>>>> of approval process has to be followed first. >>>>> >>>>> At first glance there was no easy way moving the warnings from >>>>> wic to ISAR. We would have to re-parse the wic template file >>>>> again and check all the partitions afterwards. wic has all the >>>>> necessary information at hand so I guess that's way easier. >>>> >>>> I guess it can be moved into a task after wic. Here one would >>>> need to parse the partition table, which kind of sucks. "losetup" >>>> or "kpartx" might help but will not work in kas-container setups >>>> because they need root. >>>> We once had patches allowing wic to retain all partition images >>>> instead of throwing them away after disk assembly. Having a >>>> switch for wic to say ... do those partitions ... later do the >>>> disk would be generic, allow hooking in this and other things. >>>> >>>> Isar also has a class that creates ext4 images without, after >>>> which such a check should also be done. >>> >>> Yes. But instead of spreading the warnings around it would be nice >>> to have a single place where we could do the Y2038 checks. So maybe >>> it should be a base feature of "image.bbclass"? Or ext4-img.bbclass >>> should call wic instead of the mke2fs utilities directly? >>> >>> BTW: The name ext4-img.bbclass is kind of misleading. You could >>> simply create ext{2,3} file systems by setting MKE2FS_ARGS to >>> something like "-t ext2". >>> >>>> >>>> Is ext4 the only fs we care about? We have some layers doing >>>> ubifs, squashfs and all sorts of funny things. >>> >>> Up to now I cared about the filesystems supported by wic. So >>> ext{2,3,4}, btrfs and squashfs. squashfs will overflow in 2106 (u32) >>> and btrfs will "never" overflow (u64). >>> >>> ubifs is similar to btrfs, so not affected by Y2038. >>> >>>> >>>> Maybe the kernel does warn "on device" so we could have a systemd >>>> unit warning for all filesystems ... which would probably best >>>> find its place in the kernel and or debian. >>> >>> At least for affected ext file systems the kernel will warn (on >>> mount). But I considered that as "too late". >> >> To be more specific: Linux >= 5.4 warns. That's why I guess that many >> projects did not realize that they are already affected by the Y2038 >> problem because of older kernel versions. > > Which sounds like that warning needs backporting into the debian10 > kernel and maybe cip. Not sure Isar is the best place, but a valid one > that could help. > > Maybe mkfs could warn ... as well. > Right, and we want such warnings seen at image *build time*, not only on the target during runtime. That is the key idea behind this instrumentation. Jan
Am Thu, 11 Feb 2021 18:57:24 +0100 schrieb Jan Kiszka <jan.kiszka@siemens.com>: > On 11.02.21 15:13, Henning Schild wrote: > > Am Thu, 11 Feb 2021 14:31:58 +0100 > > schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > > <florian.bezdeka@siemens.com>: > > > >> On Thu, 2021-02-11 at 12:47 +0000, [ext] > >> florian.bezdeka@siemens.com wrote: > >>> On Thu, 2021-02-11 at 11:21 +0100, Henning Schild wrote: > >>>> Am Thu, 11 Feb 2021 10:57:31 +0100 > >>>> schrieb "Bezdeka, Florian (T RDA IOT SES-DE)" > >>>> <florian.bezdeka@siemens.com>: > >>>> > >>>>> On Thu, 2021-02-11 at 10:09 +0100, Jan Kiszka wrote: > >>>>>> On 11.02.21 09:23, Henning Schild wrote: > >>>>>>> Hi all, > >>>>>>> > >>>>>>> i never got around to reviewing this. But did we just fork > >>>>>>> wic? These patches need to go into wic and we later > >>>>>>> backport them once they are accepted upstream. > >>>>>>> > >>>>>>> Maybe they are already ... did not check. > >>>>>>> > >>>>>>> When it comes to changing bitbake or wic, we should really > >>>>>>> not ... We have forks of some files, like the wic plugins > >>>>>>> and bitbake config, those are fine but should also stay > >>>>>>> very close to upstream. > >>>>>>> > >>>>>>> The recently applied patch from Vijai also violates that. > >>>>>>> Since the fork of the plugins was not updated with the wic > >>>>>>> bump and the repair just takes a few bits of what we > >>>>>>> probably should take. > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>> > >>>>>> If you are referring to > >>>>>> https://groups.google.com/d/msgid/isar-users/20201126091750.28048-1-Vijaikumar_Kanagarajan%40mentor.com > >>>>>> That one was "only" patching an isar version, though I agree > >>>>>> that we should make sure to realign it with the original > >>>>>> plugins if we are now imbalanced. > >>>>>> > >>>>>> This one here is more critical as it changed a formerly > >>>>>> vanilla wic file. That should be fixed quickly. > >>>>>> > >>>>>> Florian, maybe you can propose a similar change to OE > >>>>>> upstream? In the meantime, is there a chance to move the > >>>>>> changes out of partition.py, to a file that is isar-specific? > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>>> > >>>>> I guess the "RFC" tag of this series has been overlooked. It > >>>>> was not intended for merging (yet). Part one (forwarding wic > >>>>> warnings to bitbake) is a pure ISAR change and could be taken > >>>>> as is (if no further comments come up). > >>>> > >>>> I guess that calls for a revert. And for more attention on the > >>>> maintainers side. Florian, maybe you send a revert series. Not your fault but maybe your call. Henning > >>>>> Sorry for the long description of the series, but if you read > >>>>> closely I already mentioned that the second part should go to > >>>>> OE. I sent it out for feedback collection only. > >>>>> > >>>>> The upstreaming to OE will take some time due to internal > >>>>> clarifications. I never contributed to OE before, so some kind > >>>>> of approval process has to be followed first. > >>>>> > >>>>> At first glance there was no easy way moving the warnings from > >>>>> wic to ISAR. We would have to re-parse the wic template file > >>>>> again and check all the partitions afterwards. wic has all the > >>>>> necessary information at hand so I guess that's way easier. > >>>> > >>>> I guess it can be moved into a task after wic. Here one would > >>>> need to parse the partition table, which kind of sucks. "losetup" > >>>> or "kpartx" might help but will not work in kas-container setups > >>>> because they need root. > >>>> We once had patches allowing wic to retain all partition images > >>>> instead of throwing them away after disk assembly. Having a > >>>> switch for wic to say ... do those partitions ... later do the > >>>> disk would be generic, allow hooking in this and other things. > >>>> > >>>> Isar also has a class that creates ext4 images without, after > >>>> which such a check should also be done. > >>> > >>> Yes. But instead of spreading the warnings around it would be nice > >>> to have a single place where we could do the Y2038 checks. So > >>> maybe it should be a base feature of "image.bbclass"? Or > >>> ext4-img.bbclass should call wic instead of the mke2fs utilities > >>> directly? > >>> > >>> BTW: The name ext4-img.bbclass is kind of misleading. You could > >>> simply create ext{2,3} file systems by setting MKE2FS_ARGS to > >>> something like "-t ext2". > >>> > >>>> > >>>> Is ext4 the only fs we care about? We have some layers doing > >>>> ubifs, squashfs and all sorts of funny things. > >>> > >>> Up to now I cared about the filesystems supported by wic. So > >>> ext{2,3,4}, btrfs and squashfs. squashfs will overflow in 2106 > >>> (u32) and btrfs will "never" overflow (u64). > >>> > >>> ubifs is similar to btrfs, so not affected by Y2038. > >>> > >>>> > >>>> Maybe the kernel does warn "on device" so we could have a systemd > >>>> unit warning for all filesystems ... which would probably best > >>>> find its place in the kernel and or debian. > >>> > >>> At least for affected ext file systems the kernel will warn (on > >>> mount). But I considered that as "too late". > >> > >> To be more specific: Linux >= 5.4 warns. That's why I guess that > >> many projects did not realize that they are already affected by > >> the Y2038 problem because of older kernel versions. > > > > Which sounds like that warning needs backporting into the debian10 > > kernel and maybe cip. Not sure Isar is the best place, but a valid > > one that could help. > > > > Maybe mkfs could warn ... as well. > > > > Right, and we want such warnings seen at image *build time*, not only > on the target during runtime. That is the key idea behind this > instrumentation. > > Jan >
On Thu, Feb 11, 2021 at 07:01:50PM +0100, Henning Schild wrote: > Florian, maybe you send a revert series. Not your fault but maybe your > call. For that matter, we can discuss reverting. That said, I'd like to understand the situation first. I know that you invested much effort for integrating wic without changes and keeping it unmodified; this prevents maintenance effort. Upstreaming the changes is also good for the same reason; Florian is doing that. If the changes are accepted, we update wic -- everything fine. If not, we still can decide what to do with that -- no doors are closed. Currently, Isar warns users about the problem -- added value. I personally fail to see what value should reverting have in this situation. On the maintainer side, I think we could test the following additions: * Even if the maintainer thinks an RFC patch is good enough as is, it's advised to sync with the list. * If a patch changes upstream copies (bitbake, wic; anything else?), double checking is advised. With kind regards, Baurzhan.
On 01.02.21 19:58, Bezdeka, Florian (T RDA IOT SES-DE) wrote: > Valid workarounds found so far: > - Tell wic that an partition will grow: > Add `--mkfs-extraopts "-T ext4"` to your wic partition definition > - Set the inode size to 256 (for small ext4 partitions) > Add `--mkfs-extraopts "-I 256"` to your wic partition definition Is this check is currently part of Isar (triggering on one of my layers at least) and will eventually come back via OE, concrete help how to resolve the warning should become part of Isar as well. Apparently, --mkfs-extraopts "-T default" is the common pattern now, right? Jan
On 27.03.21 08:20, Jan Kiszka wrote: > On 01.02.21 19:58, Bezdeka, Florian (T RDA IOT SES-DE) wrote: >> Valid workarounds found so far: >> - Tell wic that an partition will grow: >> Add `--mkfs-extraopts "-T ext4"` to your wic partition definition >> - Set the inode size to 256 (for small ext4 partitions) >> Add `--mkfs-extraopts "-I 256"` to your wic partition definition > > Is this check is currently part of Isar (triggering on one of my layers > at least) and will eventually come back via OE, concrete help how to > resolve the warning should become part of Isar as well.> > Apparently, --mkfs-extraopts "-T default" is the common pattern now, right? Right. At least as long as you're not setting up a very tiny or giant FS. > > Jan >
From: Florian Bezdeka <florian.bezdeka@siemens.com> Hi ISAR developers, this series is the summary of a nice journey through the file system jungle regarding Y2038 problem. It all began with a warning which is reported by kernels >= 5.4: ext4 filesystem being mounted at (mountpoint) supports timestamps until 2038 (0x7fffffff) I guess that most ISAR layers are using the Debian kernels, so that warning was not recognized yet or at least not very often. When reading this warning I was surprised. Shouldn't a modern file system like ext4 be Y2038-safe? As it turned out it depends on the inode size if an ext4 file system is safe or not. So why was the inode size not sufficient in my case? The inode size is chosen during file system generation and depends on the size of the file system that is going to be created. For details let's have a look at `man mke2fs`: -T usage-type[,...] Specify how the filesystem is going to be used, so that mke2fs can choose optimal filesystem parameters for that use. The usage types that are supported are defined in the configuration file /etc/mke2fs.conf. The user may specify one or more usage types using a comma separated list. If this option is is not specified, mke2fs will pick a single default usage type based on the size of the filesystem to be created. If the filesystem size is less than 3 megabytes, mke2fs will use the filesystem type floppy. If the filesystem size is greater than or equal to 3 but less than 512 megabytes, mke2fs(8) will use the filesystem type small. The relevant parts from /etc/mke2fs.conf: [fs_types] ... small = { blocksize = 1024 inode_size = 128 inode_ratio = 4096 } ... So whenever you create an ext4 file system with less than 512MB in size you will end up with 128 byte inodes and your file system is not Y2038-safe. The ISAR part: ext4 may often be used in combination with the expand-on-first-boot recipe / feature. So whenever creating a small partition (e.g. inside a wic file) and extending it later may result in a Y2038 affected ext4 file system. That is exactly what happened to me and I would like to make sure that all other ISAR users are aware of this situation. Valid workarounds found so far: - Tell wic that an partition will grow: Add `--mkfs-extraopts "-T ext4"` to your wic partition definition - Set the inode size to 256 (for small ext4 partitions) Add `--mkfs-extraopts "-I 256"` to your wic partition definition The upstream part: None of the following patches has been sent to any upstream (OE) mailing lists yet but hopefully that will happen soon. So far: Any comments welcome! Best regards, Florian Florian Bezdeka (2): wic-img: Forward warnings from wic to bitbake wic: Warn if an ext filesystem affected by the Y2038 problem is used meta/classes/wic-img.bbclass | 20 ++++++++++++++----- scripts/lib/wic/partition.py | 38 ++++++++++++++++++++++++++++++++++++ 2 files changed, 53 insertions(+), 5 deletions(-)