Message ID | CAKVti-BRav=mmfcQBEz_qpezmKfR_fFW9uihJb+QXiJazugTmA@mail.gmail.com |
---|---|
State | New |
Headers | show |
Series | linux: fix ntp_gettime abi break | expand |
Could you please file a bug at https://sourceware.org/bugzilla/?
> > Could you please file a bug at https://sourceware.org/bugzilla/? > > -- > Andreas Schwab, SUSE Labs, schwab@suse.de > GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7 > "And now for something completely different." https://sourceware.org/bugzilla/show_bug.cgi?id=30156
On 2/20/23 23:04, Kacper Piwiński via Libc-alpha wrote: > Between versions v2.11 and v2.12 struct ntptimeval got new fields. > That wasn't a problem because new function ntp_gettimex was created > (and made default) to support new struct. Old ntp_gettime was not > using new fields so it was safe to call with old struct > definition. Then between versions v2.31 and v2.32 when working on > y2038 bugs, ntp_gettime start setting new fields. > > Sets fields manually to maintain compatibility with v2.11 struct definition. The patch fails to apply. This fails pre-commit CI: https://patchwork.sourceware.org/project/glibc/patch/CAKVti-BRav=mmfcQBEz_qpezmKfR_fFW9uihJb+QXiJazugTmA@mail.gmail.com/ Please have a look at: https://sourceware.org/glibc/wiki/Contribution%20checklist Did you generate the patch with `git format-patch`? > > Author: Kacper Piwiński <vfjpl1@gmail.com> > --- > sysdeps/unix/sysv/linux/ntp_gettime.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/sysdeps/unix/sysv/linux/ntp_gettime.c > b/sysdeps/unix/sysv/linux/ntp_gettime.c > index 4879573530..3844b2e487 100644 > --- a/sysdeps/unix/sysv/linux/ntp_gettime.c > +++ b/sysdeps/unix/sysv/linux/ntp_gettime.c > @@ -54,7 +54,11 @@ __ntp_gettime (struct ntptimeval *ntv) > int result; > > result = __ntp_gettime64 (&ntv64); > - *ntv = valid_ntptimeval64_to_ntptimeval (ntv64); > + //manually update fields to maintain abi with v2.11 struct definition > + struct ntptimeval tntv = valid_ntptimeval64_to_ntptimeval (ntv64); > + ntv->time = tntv.time; > + ntv->maxerror = tntv.maxerror; > + ntv->esterror = tntv.esterror; > > return result; > } > -- > 2.38.1.windows.1 >
czw., 2 mar 2023 o 18:47 Carlos O'Donell <carlos@redhat.com> napisał(a): > > On 2/20/23 23:04, Kacper Piwiński via Libc-alpha wrote: > > Between versions v2.11 and v2.12 struct ntptimeval got new fields. > > That wasn't a problem because new function ntp_gettimex was created > > (and made default) to support new struct. Old ntp_gettime was not > > using new fields so it was safe to call with old struct > > definition. Then between versions v2.31 and v2.32 when working on > > y2038 bugs, ntp_gettime start setting new fields. > > > > Sets fields manually to maintain compatibility with v2.11 struct definition. > > The patch fails to apply. > > This fails pre-commit CI: > https://patchwork.sourceware.org/project/glibc/patch/CAKVti-BRav=mmfcQBEz_qpezmKfR_fFW9uihJb+QXiJazugTmA@mail.gmail.com/ > > Please have a look at: > https://sourceware.org/glibc/wiki/Contribution%20checklist > > Did you generate the patch with `git format-patch`? > Hi, yes I did but I didn't copy the whole thing. Anyway this patch is not correct. I attatched new patch here https://sourceware.org/bugzilla/show_bug.cgi?id=30156 > > > > > Author: Kacper Piwiński <vfjpl1@gmail.com> > > --- > > sysdeps/unix/sysv/linux/ntp_gettime.c | 6 +++++- > > 1 file changed, 5 insertions(+), 1 deletion(-) > > > > diff --git a/sysdeps/unix/sysv/linux/ntp_gettime.c > > b/sysdeps/unix/sysv/linux/ntp_gettime.c > > index 4879573530..3844b2e487 100644 > > --- a/sysdeps/unix/sysv/linux/ntp_gettime.c > > +++ b/sysdeps/unix/sysv/linux/ntp_gettime.c > > @@ -54,7 +54,11 @@ __ntp_gettime (struct ntptimeval *ntv) > > int result; > > > > result = __ntp_gettime64 (&ntv64); > > - *ntv = valid_ntptimeval64_to_ntptimeval (ntv64); > > + //manually update fields to maintain abi with v2.11 struct definition > > + struct ntptimeval tntv = valid_ntptimeval64_to_ntptimeval (ntv64); > > + ntv->time = tntv.time; > > + ntv->maxerror = tntv.maxerror; > > + ntv->esterror = tntv.esterror; > > > > return result; > > } > > -- > > 2.38.1.windows.1 > > > > -- > Cheers, > Carlos. >
On 03/03/23 05:47, Kacper Piwiński via Libc-alpha wrote: > czw., 2 mar 2023 o 18:47 Carlos O'Donell <carlos@redhat.com> napisał(a): >> >> On 2/20/23 23:04, Kacper Piwiński via Libc-alpha wrote: >>> Between versions v2.11 and v2.12 struct ntptimeval got new fields. >>> That wasn't a problem because new function ntp_gettimex was created >>> (and made default) to support new struct. Old ntp_gettime was not >>> using new fields so it was safe to call with old struct >>> definition. Then between versions v2.31 and v2.32 when working on >>> y2038 bugs, ntp_gettime start setting new fields. >>> >>> Sets fields manually to maintain compatibility with v2.11 struct definition. >> >> The patch fails to apply. >> >> This fails pre-commit CI: >> https://patchwork.sourceware.org/project/glibc/patch/CAKVti-BRav=mmfcQBEz_qpezmKfR_fFW9uihJb+QXiJazugTmA@mail.gmail.com/ >> >> Please have a look at: >> https://sourceware.org/glibc/wiki/Contribution%20checklist >> >> Did you generate the patch with `git format-patch`? >> > > Hi, yes I did but I didn't copy the whole thing. Anyway this patch is > not correct. I attatched new patch here > https://sourceware.org/bugzilla/show_bug.cgi?id=30156 Could you please resend as v2? Patch submission and discussion are done through the maillist. PS: maybe we should be more clear on wiki that patches are done discussed on bugzilla.
From 64bbb7cc698c8e2c8d8c94247e0bc47ac3b18f77 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Kacper=20Piwi=C5=84ski?= <vfjpl1@gmail.com>
Date: Wed, 22 Feb 2023 18:58:38 +0100
Subject: [PATCH] linux: fix ntp_gettime abi break
---
sysdeps/unix/sysv/linux/ntp_gettime.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/sysdeps/unix/sysv/linux/ntp_gettime.c
b/sysdeps/unix/sysv/linux/ntp_gettime.c
index 4879573530..a577314636 100644
--- a/sysdeps/unix/sysv/linux/ntp_gettime.c
+++ b/sysdeps/unix/sysv/linux/ntp_gettime.c
@@ -40,7 +40,6 @@ __ntp_gettime64 (struct __ntptimeval64 *ntv)
ntv->time = tntx.time;
ntv->maxerror = tntx.maxerror;
ntv->esterror = tntx.esterror;
- ntv->tai = tntx.tai;
return result;
}
@@ -54,7 +53,10 @@ __ntp_gettime (struct ntptimeval *ntv)
int result;
result = __ntp_gettime64 (&ntv64);
- *ntv = valid_ntptimeval64_to_ntptimeval (ntv64);
+ //manually update fields to maintain abi with v2.11 struct definition
+ ntv->time = valid_timeval64_to_timeval (ntv64.time);
+ ntv->maxerror = ntv64.maxerror;
+ ntv->esterror = ntv64.esterror;
return result;
}
On 03/03/23 13:43, Kacper Piwiński wrote: > From 64bbb7cc698c8e2c8d8c94247e0bc47ac3b18f77 Mon Sep 17 00:00:00 2001 > From: =?UTF-8?q?Kacper=20Piwi=C5=84ski?= <vfjpl1@gmail.com> > Date: Wed, 22 Feb 2023 18:58:38 +0100 > Subject: [PATCH] linux: fix ntp_gettime abi break > > --- > sysdeps/unix/sysv/linux/ntp_gettime.c | 6 ++++-- > 1 file changed, 4 insertions(+), 2 deletions(-) > > diff --git a/sysdeps/unix/sysv/linux/ntp_gettime.c > b/sysdeps/unix/sysv/linux/ntp_gettime.c > index 4879573530..a577314636 100644 > --- a/sysdeps/unix/sysv/linux/ntp_gettime.c > +++ b/sysdeps/unix/sysv/linux/ntp_gettime.c > @@ -40,7 +40,6 @@ __ntp_gettime64 (struct __ntptimeval64 *ntv) > ntv->time = tntx.time; > ntv->maxerror = tntx.maxerror; > ntv->esterror = tntx.esterror; > - ntv->tai = tntx.tai; > return result; > } > > @@ -54,7 +53,10 @@ __ntp_gettime (struct ntptimeval *ntv) > int result; > > result = __ntp_gettime64 (&ntv64); > - *ntv = valid_ntptimeval64_to_ntptimeval (ntv64); > + //manually update fields to maintain abi with v2.11 struct definition Sigh, so we decided to not add symbol versioning, and add ntp_gettimex instead. Comments should be in the form of '/* ... */', with first uppercase and with final period. But I think we should make it clear that this function should not work on 'struct ntptimeval', even though users might call glibc 2.12 version: /* glibc 2.12 added the 'tai' field to follow along the kernel, but it did not add a compatibility symbol, instead it added __ntp_gettimex. However ntptimeval was still used in both cases, so to keep compatibility it can not set all the new field. */ struct old_ntptimeval { struct timeval time; long int maxerror; long int esterror; }; int __old_ntp_gettime (struct old_ntptimeval *ntv) { struct __ntptimeval64 ntv64; int result; result = __ntp_gettime64 (&ntv64); ntv->time = valid_timeval64_to_timeval (ntv64.time); ntv->maxerror = ntv64.maxerror; ntv->esterror = ntv64.esterror; return result; } strong_alias (__old_ntp_gettime, __ntp_gettime) > + ntv->time = valid_timeval64_to_timeval (ntv64.time); > + ntv->maxerror = ntv64.maxerror; > + ntv->esterror = ntv64.esterror; > > return result; > }
diff --git a/sysdeps/unix/sysv/linux/ntp_gettime.c b/sysdeps/unix/sysv/linux/ntp_gettime.c index 4879573530..3844b2e487 100644 --- a/sysdeps/unix/sysv/linux/ntp_gettime.c +++ b/sysdeps/unix/sysv/linux/ntp_gettime.c @@ -54,7 +54,11 @@ __ntp_gettime (struct ntptimeval *ntv) int result; result = __ntp_gettime64 (&ntv64); - *ntv = valid_ntptimeval64_to_ntptimeval (ntv64); + //manually update fields to maintain abi with v2.11 struct definition + struct ntptimeval tntv = valid_ntptimeval64_to_ntptimeval (ntv64); + ntv->time = tntv.time; + ntv->maxerror = tntv.maxerror; + ntv->esterror = tntv.esterror; return result; }