Regression: Commit "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" breaks ruleset loading in linux-stable
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Thanks and kind regards,
Timo
#regzbot introduced: 0ebc1064e487 #regzbot monitor: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1051592
Hi Timo,
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine.
See patches enclosed to this email.
On 12.09.23 00:57, Pablo Neira Ayuso wrote:
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine. [...]
Hmmmm. Well, this sounds like a kernel regression to me that normally should be dealt with on the kernel level, as users after updating the kernel should never have to update any userspace stuff to continue what they have been doing before the kernel update.
Can't the kernel somehow detect the incorrect bytecode and do the right thing(tm) somehow?
But yes, don't worry, I know that reality is not black and white and that it's crucial that things like package filtering do exactly what the user expect it to do; that's why this might be one of those rare situations where "user has to update userspace components to support newer kernels" might be the better of two bad choices. But I had to ask to ensure it's something like that.
Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) -- Everything you wanna know about Linux kernel regression tracking: https://linux-regtracking.leemhuis.info/about/#tldr If I did something stupid, please tell me, as explained on that page.
Linux regression tracking (Thorsten Leemhuis) regressions@leemhuis.info wrote:
On 12.09.23 00:57, Pablo Neira Ayuso wrote:
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine. [...]
Hmmmm. Well, this sounds like a kernel regression to me that normally should be dealt with on the kernel level, as users after updating the kernel should never have to update any userspace stuff to continue what they have been doing before the kernel update.
This is a combo of a userspace bug and this new sanity check that rejects the incorrect ordering (adding rules to the already-bound anonymous chain).
nf_tables uses a transaction allor-nothing model, this means that any error that occurs during a transaction has to be reverse/undo all the pending changes. This has caused a myriad of bugs already.
So while this can be theoretically fixed in the kernel I don't see a sane way to do it. Error unwinding / recovery from deeply nested errors is already too complex for my taste.
Can't the kernel somehow detect the incorrect bytecode and do the right thing(tm) somehow?
Theoretically yes, but I don't feel competent enough to do it, just look at all the UaF bugs of the past month.
Hi,
Florian Westphal schrieb am 12.09.2023 12:27 (GMT +02:00):
Linux regression tracking (Thorsten Leemhuis) regressions@leemhuis.info wrote:
On 12.09.23 00:57, Pablo Neira Ayuso wrote:
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine. [...]
Hmmmm. Well, this sounds like a kernel regression to me that normally should be dealt with on the kernel level, as users after updating the kernel should never have to update any userspace stuff to continue what they have been doing before the kernel update.
This is a combo of a userspace bug and this new sanity check that rejects the incorrect ordering (adding rules to the already-bound anonymous chain).
Out of curiosity, did the incorrect ordering or bytecode from the older userspace components actually lead to a wrong representation of the rules in the kernel or did the rules still work despite all that?
Thanks,
Timo
Timo Sigurdsson public_timo.s@silentcreek.de wrote:
Linux regression tracking (Thorsten Leemhuis) regressions@leemhuis.info wrote:
On 12.09.23 00:57, Pablo Neira Ayuso wrote:
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine. [...]
Hmmmm. Well, this sounds like a kernel regression to me that normally should be dealt with on the kernel level, as users after updating the kernel should never have to update any userspace stuff to continue what they have been doing before the kernel update.
This is a combo of a userspace bug and this new sanity check that rejects the incorrect ordering (adding rules to the already-bound anonymous chain).
Out of curiosity, did the incorrect ordering or bytecode from the older userspace components actually lead to a wrong representation of the rules in the kernel or did the rules still work despite all that?
It works, but without the stricter behaviour userspace can trigger memory corruption in the kernel. nftables userland will not trigger this.
On 12.09.23 12:27, Florian Westphal wrote:
Linux regression tracking (Thorsten Leemhuis) regressions@leemhuis.info wrote:
On 12.09.23 00:57, Pablo Neira Ayuso wrote:
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine. [...]
Hmmmm. Well, this sounds like a kernel regression to me that normally should be dealt with on the kernel level, as users after updating the kernel should never have to update any userspace stuff to continue what they have been doing before the kernel update.
This is a combo of a userspace bug and this new sanity check that rejects the incorrect ordering (adding rules to the already-bound anonymous chain).
nf_tables uses a transaction allor-nothing model, this means that any error that occurs during a transaction has to be reverse/undo all the pending changes. This has caused a myriad of bugs already.
So while this can be theoretically fixed in the kernel I don't see a sane way to do it. Error unwinding / recovery from deeply nested errors is already too complex for my taste.
Can't the kernel somehow detect the incorrect bytecode and do the right thing(tm) somehow?
Theoretically yes, but I don't feel competent enough to do it, just look at all the UaF bugs of the past month.
Thx for the answer. FWIW, as this was a judgement call I mentioned this in my last regression report to Linus; he didn't reply, so I guess it is -- and will remove this issue from my tracking:
#regzbot resolve: can be solved by a nftables userspace update; not nice, but likely best solution in this case #regzbot ignore-activity
Ciao, Thorsten (wearing his 'the Linux kernel's regression tracker' hat) -- Everything you wanna know about Linux kernel regression tracking: https://linux-regtracking.leemhuis.info/about/#tldr If I did something stupid, please tell me, as explained on that page.
Hi Pablo,
Pablo Neira Ayuso schrieb am 12.09.2023 00:57 (GMT +02:00):
Hi Timo,
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine.
hmm, that doesn't explain why Salvatore didn't observe this with more recent kernels.
Salvatore, did you use newer userspace components when you tested your 6.4.13 and 6.5.2 builds?
As for the regression and how it be dealt with: Personally, I don't really care whether the regression is solved in the kernel or userspace. If everybody agrees that this is the best or only viable option and Debian decides to push a nftables update to fix this, that works for me. But I do feel the burden to justify this should be high. A kernel change that leaves users without a working packet filter after upgrading their machines is serious, if you ask me. And since it affects several stable/longterm trees, I would assume this will hit other stable (non-rolling) distributions as well, since they will also use older userspace components (unless this is behavior specific to nftables 1.0.6 but not older versions). They probably should get a heads up then.
Regards,
Timo
On Tue, Sep 12, 2023 at 01:39:59PM +0200, Timo Sigurdsson wrote:
Hi Pablo,
Pablo Neira Ayuso schrieb am 12.09.2023 00:57 (GMT +02:00):
Hi Timo,
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine.
hmm, that doesn't explain why Salvatore didn't observe this with more recent kernels.
Salvatore, did you use newer userspace components when you tested your 6.4.13 and 6.5.2 builds?
As for the regression and how it be dealt with: Personally, I don't really care whether the regression is solved in the kernel or userspace. If everybody agrees that this is the best or only viable option and Debian decides to push a nftables update to fix this, that works for me. But I do feel the burden to justify this should be high. A kernel change that leaves users without a working packet filter after upgrading their machines is serious, if you ask me. And since it affects several stable/longterm trees, I would assume this will hit other stable (non-rolling) distributions as well, since they will also use older userspace components (unless this is behavior specific to nftables 1.0.6 but not older versions). They probably should get a heads up then.
There is coverage for the chain binding feature in our tests infrastructure, but unfortunately the bug did not trigger with newer nftables versions. In the future, I will keep an eye with running our tests on older userspace nftables versions when working on stable trees.
Hi Timo,
On Tue, Sep 12, 2023 at 01:39:59PM +0200, Timo Sigurdsson wrote:
Hi Pablo,
Pablo Neira Ayuso schrieb am 12.09.2023 00:57 (GMT +02:00):
Hi Timo,
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine.
hmm, that doesn't explain why Salvatore didn't observe this with more recent kernels.
Salvatore, did you use newer userspace components when you tested your 6.4.13 and 6.5.2 builds?
It does explain now because understanding the issue better. While one while experinting should only change each one constraint for the 6.4.13 and 6.5.2 testing I indeed switched to a Debian unstable system, which has newer userpace nftables and so not triggering the issue. This was missleading for the report.
As for the regression and how it be dealt with: Personally, I don't really care whether the regression is solved in the kernel or userspace. If everybody agrees that this is the best or only viable option and Debian decides to push a nftables update to fix this, that works for me. But I do feel the burden to justify this should be high. A kernel change that leaves users without a working packet filter after upgrading their machines is serious, if you ask me. And since it affects several stable/longterm trees, I would assume this will hit other stable (non-rolling) distributions as well, since they will also use older userspace components (unless this is behavior specific to nftables 1.0.6 but not older versions). They probably should get a heads up then.
So if it is generally believed on kernel side there should not happen any further changes to work with older userland, I guess in Debian we will need to patch nftables. I'm CC'ing Arturo Borrero Gonzalez arturo@debian.org, maintainer for the package. The update should go ideally in the next point releases from October (and maybe released earlier as well trough the stable-updates mechanism).
FWIW: In Debian bullseye we have 0.9.8 based nftables, in bookworm 1.0.6, so both will need those fixes.
As 0ebc1064e487 is to address CVE-2023-4147 other distros picking the fix will likely encounter the problem at some point. It looks Red Hat has taken it (some RHSA's were released), I assume Ubuntu will shortly as well release USN's containing a fix.
Regards, Salvatore
Hi,
Salvatore Bonaccorso schrieb am 12.09.2023 21:13 (GMT +02:00):
Hi Timo,
On Tue, Sep 12, 2023 at 01:39:59PM +0200, Timo Sigurdsson wrote:
Hi Pablo,
Pablo Neira Ayuso schrieb am 12.09.2023 00:57 (GMT +02:00):
Hi Timo,
On Mon, Sep 11, 2023 at 11:37:50PM +0200, Timo Sigurdsson wrote:
Hi,
recently, Debian updated their stable kernel from 6.1.38 to 6.1.52 which broke nftables ruleset loading on one of my machines with lots of "Operation not supported" errors. I've reported this to the Debian project (see link below) and Salvatore Bonaccorso and I identified "netfilter: nf_tables: disallow rule addition to bound chain via NFTA_RULE_CHAIN_ID" (0ebc1064e487) as the offending commit that introduced the regression. Salvatore also found that this issue affects the 5.10 stable tree as well (observed in 5.10.191), but he cannot reproduce it on 6.4.13 and 6.5.2.
The issue only occurs with some rulesets. While I can't trigger it with simple/minimal rulesets that I use on some machines, it does occur with a more complex ruleset that has been in use for months (if not years, for large parts of it). I'm attaching a somewhat stripped down version of the ruleset from the machine I originally observed this issue on. It's still not a small or simple ruleset, but I'll try to reduce it further when I have more time.
The error messages shown when trying to load the ruleset don't seem to be helpful. Just two simple examples: Just to give two simple examples from the log when nftables fails to start: /etc/nftables.conf:99:4-44: Error: Could not process rule: Operation not supported tcp option maxseg size 1-500 counter drop ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ /etc/nftables.conf:308:4-27: Error: Could not process rule: Operation not supported tcp dport sip-tls accept ^^^^^^^^^^^^^^^^^^^^^^^^
I can reproduce this issue with 5.10.191 and 6.1.52 and nftables v1.0.6, this is not reproducible with v1.0.7 and v1.0.8.
Since the issue only affects some stable trees, Salvatore thought it might be an incomplete backport that causes this.
If you need further information, please let me know.
Userspace nftables v1.0.6 generates incorrect bytecode that hits a new kernel check that rejects adding rules to bound chains. The incorrect bytecode adds the chain binding, attach it to the rule and it adds the rules to the chain binding. I have cherry-picked these three patches for nftables v1.0.6 userspace and your ruleset restores fine.
hmm, that doesn't explain why Salvatore didn't observe this with more recent kernels.
Salvatore, did you use newer userspace components when you tested your 6.4.13 and 6.5.2 builds?
It does explain now because understanding the issue better. While one while experinting should only change each one constraint for the 6.4.13 and 6.5.2 testing I indeed switched to a Debian unstable system, which has newer userpace nftables and so not triggering the issue. This was missleading for the report.
As for the regression and how it be dealt with: Personally, I don't really care whether the regression is solved in the kernel or userspace. If everybody agrees that this is the best or only viable option and Debian decides to push a nftables update to fix this, that works for me. But I do feel the burden to justify this should be high. A kernel change that leaves users without a working packet filter after upgrading their machines is serious, if you ask me. And since it affects several stable/longterm trees, I would assume this will hit other stable (non-rolling) distributions as well, since they will also use older userspace components (unless this is behavior specific to nftables 1.0.6 but not older versions). They probably should get a heads up then.
So if it is generally believed on kernel side there should not happen any further changes to work with older userland, I guess in Debian we will need to patch nftables. I'm CC'ing Arturo Borrero Gonzalez arturo@debian.org, maintainer for the package. The update should go ideally in the next point releases from October (and maybe released earlier as well trough the stable-updates mechanism).
So, I built nftables 1.0.6-2+deb12u1 with the three cherry-picked patches from Pablo and can confirm that they resolve the issue for me on bookworm. I can now run linux 6.1.52-1 and load my original nftables ruleset again.
FWIW: In Debian bullseye we have 0.9.8 based nftables, in bookworm 1.0.6, so both will need those fixes.
As 0ebc1064e487 is to address CVE-2023-4147 other distros picking the fix will likely encounter the problem at some point. It looks Red Hat has taken it (some RHSA's were released), I assume Ubuntu will shortly as well release USN's containing a fix.
SUSE has also picked this patch for SLES/SLED. I hope maintainers follow the mailing lists cc'ed here or that someone gives them a heads up before this hits more production systems.
Thanks and regards,
Timo
linux-stable-mirror@lists.linaro.org
-
Florian Westphal -
Linux regression tracking (Thorsten Leemhuis) -
Pablo Neira Ayuso -
Salvatore Bonaccorso -
Timo Sigurdsson