nfs-utils: musl upgrade to 3.1.1 and prerequisite systemd fixes
cover nfs-utils: musl upgrade to 3.1.1 and prerequisite systemd fixes
This series upgrades nfs-utils to 3.1.1 and fixes the two systemd musl issues that block it.
On musl, nfs-utils cannot be built until systemd builds, because nfs-utils depends on systemd (it installs systemd units). systemd 261.3 currently fails on musl for two reasons, both addressed here first:
1. systemd's src/include/musl/limits.h override does not expose musl's POSIX limits (PATH_MAX, NAME_MAX, SSIZE_MAX, ...), so do_compile fails. Fixed by defining them in the override (submitted upstream, see
https://github.com/systemd/systemd/pull/43824).
2. do_package emits spurious "cannot find any provider" warnings for optional (recommended/suggested) dlopen dependencies that legitimately have no provider on musl, e.g. libintl.so.8 (musl keeps the gettext symbols in libc). Downgraded to a note for non-required dependencies.
With those in place, nfs-utils is upgraded to 3.1.1: two local patches that were merged upstream are dropped and the remaining ones refreshed.
The patches are ordered so the tree builds cleanly on musl at every step (systemd fixes first, then the nfs-utils upgrade that relies on them). glibc builds are unaffected; the systemd limits fix is libc-musl only.
Jaipaul Cheernam (3):
systemd: fix musl build by providing missing POSIX limits
systemd: dlopen-deps: only warn for missing required providers
nfs-utils: upgrade 2.9.2 -> 3.1.1
...tport-fix-missing-stddef.h-inclusion.patch | 27 ------
.../0004-Use-nogroup-for-nobody-group.patch | 4 +-
.../0005-find-OE-provided-Kerberos.patch | 2 +-
...e-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch | 47 ----------
...{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} | 4 +-
meta/recipes-core/systemd/dlopen-deps.inc | 12 ++-
...X-limits-missing-from-limits.h-overr.patch | 92 +++++++++++++++++++
meta/recipes-core/systemd/systemd_261.3.bb | 6 ++
8 files changed, 113 insertions(+), 81 deletions(-)
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch
rename meta/recipes-connectivity/nfs-utils/{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} (96%)
create mode 100644 meta/recipes-core/systemd/systemd/0001-musl-define-POSIX-limits-missing-from-limits.h-overr.patchPATCH 1/3 systemd: fix musl build by providing missing POSIX limits
systemd's src/include/musl/limits.h override sits first on the include path and relies on '#include_next <limits.h>' to pull in the C library's definitions. Because <limits.h> is a compiler-provided fixed header (with a two-pass '#include_next' guarded by _GCC_LIMITS_H_), the override overshoots musl's <limits.h> and its feature-guarded POSIX limits (PATH_MAX, NAME_MAX, HOST_NAME_MAX, IOV_MAX, LINE_MAX, PIPE_BUF, PTHREAD_STACK_MIN, SSIZE_MAX, ...) are never defined, so every musl build (-Dlibc=musl) fails to compile.
Add a patch to the override defining the POSIX limits systemd references with musl's canonical values, each #ifndef-guarded so glibc is unaffected. Applied for libc-musl only.
Upstream-Status: Submitted [https://github.com/systemd/systemd/pull/43824]
--- ...X-limits-missing-from-limits.h-overr.patch | 92 +++++++++++++++++++ meta/recipes-core/systemd/systemd_261.3.bb | 6 ++ 2 files changed, 98 insertions(+) create mode 100644 meta/recipes-core/systemd/systemd/0001-musl-define-POSIX-limits-missing-from-limits.h-overr.patch
meta/recipes-core/systemd/systemd/0001-musl-define-POSIX-limits-missing-from-limits.h-overr.patch new file +92 -0
@@ -0,0 +1,92 @@ +From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 +From: Jaipaul Cheernam <[email protected]> +Date: Mon, 21 Sep 2026 10:24:40 +0000 +Subject: [PATCH] musl: define POSIX limits missing from limits.h override + +src/include/musl/limits.h sits first on the include path (-isystem +src/include/musl) and, like the sibling overrides, relies on +'#include_next <limits.h>' to pull in the C library's definitions. This +works for <stdlib.h>, <string.h>, etc., but not for <limits.h>: the compiler +ships its own <limits.h>, which reaches the C library header via a two-pass +'#include_next' guarded by _GCC_LIMITS_H_. With this override in the first +search-path slot, that chain overshoots musl's <limits.h> and never includes +it, so musl's feature-guarded POSIX limits are missing and every musl build +(-Dlibc=musl) fails, e.g. 'NAME_MAX'/'PATH_MAX'/'SSIZE_MAX' undeclared. + +Reordering the -isystem paths does not help, since <limits.h> is a +compiler-provided fixed header. Define the POSIX limits systemd references, +using musl's canonical values from its include/limits.h, each #ifndef-guarded +so glibc and toolchains that do reach musl's <limits.h> are unaffected. + +Upstream-Status: Submitted [https://github.com/systemd/systemd/pull/43824] +Signed-off-by: Jaipaul Cheernam <[email protected]> +--- + src/include/musl/limits.h | 52 +++++++++++++++++++++++++++++++++++++-- + 1 file changed, 50 insertions(+), 2 deletions(-) + +diff --git a/src/include/musl/limits.h b/src/include/musl/limits.h +index 9620a3b0acc..48fe0c68e89 100644 +--- a/src/include/musl/limits.h ++++ b/src/include/musl/limits.h +@@ -6,11 +6,59 @@ + #include <assert.h> + #include <sys/types.h> + ++/* When this override is first on the include search path (-isystem ++ * src/include/musl), '#include_next <limits.h>' reaches the compiler's own ++ * <limits.h>, which uses a two-pass '#include_next' sequence (guarded by ++ * _GCC_LIMITS_H_ via <syslimits.h>) to pull in the C library's <limits.h>. ++ * With this override occupying the first search-path slot, its own ++ * '#include_next' shifts the position the compiler's second-pass ++ * '#include_next' resolves from, so the chain overshoots musl's <limits.h> ++ * and never includes it. The POSIX limits musl defines in its ++ * feature-test-macro guarded section are therefore missing. This differs ++ * from the sibling overrides in this directory (e.g. <stdlib.h>, <string.h>) ++ * because the compiler ships its own <limits.h> but not those; reordering ++ * the -isystem paths does not help. ++ * ++ * Provide the POSIX limits systemd relies on, using musl's canonical values ++ * (see musl's include/limits.h). Each is #ifndef-guarded so glibc, and any ++ * toolchain whose '#include_next' does reach musl's <limits.h>, are ++ * unaffected. */ ++#ifndef NAME_MAX ++# define NAME_MAX 255 ++#endif ++#ifndef PATH_MAX ++# define PATH_MAX 4096 ++#endif ++#ifndef HOST_NAME_MAX ++# define HOST_NAME_MAX 255 ++#endif ++#ifndef LINE_MAX ++# define LINE_MAX 4096 ++#endif ++#ifndef IOV_MAX ++# define IOV_MAX 1024 ++#endif ++#ifndef PIPE_BUF ++# define PIPE_BUF 4096 ++#endif ++#ifndef PTHREAD_STACK_MIN ++# define PTHREAD_STACK_MIN 2048 ++#endif ++#ifndef _POSIX_PATH_MAX ++# define _POSIX_PATH_MAX 256 ++#endif ++#ifndef _POSIX_TZNAME_MAX ++# define _POSIX_TZNAME_MAX 6 ++#endif ++ + /* musl defines SSIZE_MAX as LONG_MAX, so its type is always long. However, on 32-bit architectures, musl + * defines ssize_t as int. Strictly speaking, this is not a bug in musl. POSIX only requires SSIZE_MAX to + * evaluate to the maximum value representable by ssize_t; it does not require SSIZE_MAX itself to have type + * ssize_t. However, our code assumes that SSIZE_MAX has type ssize_t, as is the case with glibc. Cast the +- * value explicitly so that SSIZE_MAX has type ssize_t. */ ++ * value explicitly so that SSIZE_MAX has type ssize_t. The assert is guarded because SSIZE_MAX is only ++ * visible when <limits.h> actually supplied it (see above). */ ++#ifdef SSIZE_MAX + static_assert(SSIZE_MAX == LONG_MAX, ""); +-#undef SSIZE_MAX ++# undef SSIZE_MAX ++#endif + #define SSIZE_MAX ((ssize_t) LONG_MAX)
meta/recipes-core/systemd/systemd_261.3.bb +6 -0
@@ -38,6 +38,12 @@ SRC_URI += "file://touchscreen.rules \ file://0003-Do-not-create-var-log-README.patch \ " +# systemd's src/include/musl/limits.h override sits first on the include path +# and relies on '#include_next <limits.h>', but the compiler's own two-pass +# limits.h consumes that chain so musl's POSIX limits (PATH_MAX, NAME_MAX, +# HOST_NAME_MAX, SSIZE_MAX, ...) are never defined, breaking every musl build. +SRC_URI:append:libc-musl = " file://0001-musl-define-POSIX-limits-missing-from-limits.h-overr.patch" + PAM_PLUGINS = " \ pam-plugin-unix \ pam-plugin-loginuid \
PATCH 2/3 systemd: dlopen-deps: only warn for missing required providers
In reply to Jaipaul Cheernampackage_generate_dlopen_deps() warns whenever it cannot find a provider for a soname listed in a .note.dlopen segment, regardless of the dependency's priority. "recommended" and "suggested" entries are optional (they map to RRECOMMENDS) and legitimately have no provider in some configurations - for example systemd suggests libintl.so.8, which does not exist on musl, where the gettext symbols live in libc. This produces spurious warnings on every musl build.
Only warn for "required" dlopen dependencies; downgrade the missing-provider message for optional ones to a note.
--- meta/recipes-core/systemd/dlopen-deps.inc | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
meta/recipes-core/systemd/dlopen-deps.inc +11 -1
@@ -77,7 +77,17 @@ python package_generate_dlopen_deps() { found = True break if not found: - bb.warn(f"cannot find any provider for dlopen dependency: {dep['soname']}") + # A missing provider only matters for "required" dlopen + # dependencies. "recommended"/"suggested" ones are optional + # (they map to RRECOMMENDS) and legitimately have no provider + # on some configurations - e.g. systemd suggests libintl.so.8, + # which does not exist on musl where the gettext symbols live + # in libc. Only warn for required deps; note the rest. + msg = f"cannot find any provider for dlopen dependency: {dep['soname']}" + if dep["priority"] == "required": + bb.warn(msg) + else: + bb.note(msg) except oe.qa.NotELFFileError as e: bb.note(f"Cannot extract ELF notes: {e}") pass
PATCH 3/3 nfs-utils: upgrade 2.9.2 -> 3.1.1
In reply to Jaipaul CheernamChangeLog: https://www.kernel.org/pub/linux/utils/nfs-utils/3.1.1/3.1.1-Changelog
Drop two local patches now merged upstream:
- 0001-getport-fix-missing-stddef.h-inclusion.patch
- 0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch (upstream now gates HAVE_NFSD_NETLINK on --enable-nfsdctl)
Refresh the remaining local patches (0004, 0005) against the new source.
---
...tport-fix-missing-stddef.h-inclusion.patch | 27 -----------
.../0004-Use-nogroup-for-nobody-group.patch | 4 +-
.../0005-find-OE-provided-Kerberos.patch | 2 +-
...e-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch | 47 -------------------
...{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} | 4 +-
5 files changed, 4 insertions(+), 80 deletions(-)
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch
rename meta/recipes-connectivity/nfs-utils/{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} (96%)meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch deleted +0 -27
@@ -1,27 +0,0 @@ -From: Jaipaul Cheernam <[email protected]> -Date: Thu, 6 Aug 2026 10:55:00 +0000 -Subject: [PATCH] getport: fix missing stddef.h inclusion - -getport.c uses offsetof() but does not include <stddef.h>. On glibc -this works because offsetof is pulled in transitively, but on musl it -is not available without the explicit include, causing: - - getport.c:459:41: error: implicit declaration of function 'offsetof' - -Upstream-Status: Submitted [https://lore.kernel.org/linux-nfs/[email protected]/] -Signed-off-by: Jaipaul Cheernam <[email protected]> ---- - support/nfs/getport.c | 1 + - 1 file changed, 1 insertion(+) - -diff --git a/support/nfs/getport.c b/support/nfs/getport.c ---- a/support/nfs/getport.c -+++ b/support/nfs/getport.c -@@ -27,6 +27,7 @@ - #include <config.h> - #endif - -+#include <stddef.h> - #include <sys/types.h> - #include <sys/time.h> - #include <unistd.h>
meta/recipes-connectivity/nfs-utils/nfs-utils/0004-Use-nogroup-for-nobody-group.patch +2 -2
@@ -1,4 +1,4 @@ -From 5f4739475f3c01145174734ed8a3b31b3da99b77 Mon Sep 17 00:00:00 2001 +From f8407c33b26641c08ad76422ba2ff78214364b19 Mon Sep 17 00:00:00 2001 From: Daniel McGregor <[email protected]> Date: Tue, 6 Jun 2023 16:07:53 -0600 Subject: [PATCH] Use "nogroup" for nobody group
@@ -24,7 +24,7 @@ index 2a2f79a..e6f3724 100644 [Translation] diff --git a/utils/idmapd/idmapd.c b/utils/idmapd/idmapd.c -index 5231f56..e1aa8a5 100644 +index c41ae1f..7339a88 100644 --- a/utils/idmapd/idmapd.c +++ b/utils/idmapd/idmapd.c @@ -89,7 +89,7 @@
meta/recipes-connectivity/nfs-utils/nfs-utils/0005-find-OE-provided-Kerberos.patch +1 -1
@@ -1,4 +1,4 @@ -From a1759d9079a436096be5ce44622a3a5a85ec9faa Mon Sep 17 00:00:00 2001 +From 43a992e7bb491fd6e7e50861c9eaf0c549e98293 Mon Sep 17 00:00:00 2001 From: Daniel McGregor <[email protected]> Date: Wed, 8 Nov 2023 16:24:20 -0600 Subject: [PATCH] find OE provided Kerberos
meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch deleted +0 -47
47 lines not on this page · raw message
meta/recipes-connectivity/nfs-utils/nfs-utils_3.1.1.bb renamed from meta/recipes-connectivity/nfs-utils/nfs-utils_2.9.2.bb +1 -3
@@ -23,11 +23,9 @@ SRC_URI = "${KERNELORG_MIRROR}/linux/utils/nfs-utils/${PV}/nfs-utils-${PV}.tar.x file://nfscommon \ file://0004-Use-nogroup-for-nobody-group.patch \ file://0005-find-OE-provided-Kerberos.patch \ - file://0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch \ - file://0001-getport-fix-missing-stddef.h-inclusion.patch \ " -SRC_URI[sha256sum] = "e1dd8a9c95af15492065942cc3b52b1339ffd586baa2280ed86c9d3dc4097e8c" +SRC_URI[sha256sum] = "7dbc05eb8b32828b322872b759a62cb3979cb4118f344dd8616813d1390e84e0" # Only kernel-module-nfsd is required here (but can be built-in) - the nfsd module will # pull in the remainder of the dependencies.
v2 cover systemd musl build fixes and nfs-utils upgrade
In reply to Jaipaul CheernamTwo systemd fixes needed for musl builds, plus the nfs-utils 3.1.1 upgrade.
Changes in v2:
- 0001: use the upstream fix from systemd PR #43831 (more accurate).
- 0002: reword the commit message and shorten the inline comment.
- 0003: no change since v1.
Jaipaul Cheernam (3):
systemd: fix musl build by fixing limits.h override
systemd: dlopen-deps: only warn for missing required providers
nfs-utils: upgrade 2.9.2 -> 3.1.1
...tport-fix-missing-stddef.h-inclusion.patch | 27 -----------
.../0004-Use-nogroup-for-nobody-group.patch | 4 +-
.../0005-find-OE-provided-Kerberos.patch | 2 +-
...e-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch | 47 -------------------
...{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} | 4 +-
meta/recipes-core/systemd/dlopen-deps.inc | 8 +++-
.../0001-include-fix-limits.h-override.patch | 47 +++++++++++++++++++
meta/recipes-core/systemd/systemd_261.3.bb | 2 +
8 files changed, 60 insertions(+), 81 deletions(-)
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch
rename meta/recipes-connectivity/nfs-utils/{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} (96%)
create mode 100644 meta/recipes-core/systemd/systemd/0001-include-fix-limits.h-override.patchPATCH v2 1/3 systemd: fix musl build by fixing limits.h override
Backport the upstream fix for the limits.h override, which breaks GCC's <limits.h> include cycle (notably GCC 16.x) and fails musl builds. musl only.
Upstream fix: https://github.com/systemd/systemd/pull/43831
(cherry picked from commit 4e08fdf59ffcafef6ca2df6342101e7e82b6183b)
---
Changes in v2:
- Use the upstream fix from systemd PR #43831 (more accurate).
.../0001-include-fix-limits.h-override.patch | 47 +++++++++++++++++++ meta/recipes-core/systemd/systemd_261.3.bb | 2 + 2 files changed, 49 insertions(+) create mode 100644 meta/recipes-core/systemd/systemd/0001-include-fix-limits.h-override.patch
meta/recipes-core/systemd/systemd/0001-include-fix-limits.h-override.patch new file +47 -0
@@ -0,0 +1,47 @@ +From 4e08fdf59ffcafef6ca2df6342101e7e82b6183b Mon Sep 17 00:00:00 2001 +From: Yu Watanabe <[email protected]> +Date: Tue, 22 Sep 2026 01:55:01 +0900 +Subject: [PATCH] include: fix limits.h override + +This is especially necessary since GCC 16.x. + +Replaces #43824. + +(cherry picked from commit 4e08fdf59ffcafef6ca2df6342101e7e82b6183b) + +Upstream-Status: Backport [https://github.com/systemd/systemd/pull/43831] +Signed-off-by: Jaipaul Cheernam <[email protected]> +--- + src/include/musl/limits.h | 19 +++++++++++++++++++ + 1 file changed, 19 insertions(+) + +diff --git a/src/include/musl/limits.h b/src/include/musl/limits.h +index 9620a3b0acc..cffb8ddd274 100644 +--- a/src/include/musl/limits.h ++++ b/src/include/musl/limits.h +@@ -1,6 +1,25 @@ + /* SPDX-License-Identifier: LGPL-2.1-or-later */ + #pragma once + ++/* GCC introduces a spurious, tricky inclusion cycle: ++ * GCC's limits.h ++ * -> GCC's syslimits.h ++ * -> GCC's limits.h (again!!!) ++ * -> glibc/musl's limits.h ++ * -> define POSIX defines ++ * -> define ISO C defines ++ * This works only when GCC's limits.h is included first, but a user override breaks the cycle. Hence, we ++ * need to manually achieve the cycle here. If GCC is not used (e.g., Clang), then including the compiler's ++ * limits.h twice should be redundant but harmless. */ ++ ++/* First, get the POSIX defines from glibc/musl's limits.h. When the two macros below are defined, GCC's ++ * limits.h includes the next limits.h, that is, one from glibc/musl. */ ++#define _GCC_LIMITS_H_ ++#define _GCC_NEXT_LIMITS_H ++#include_next <limits.h> /* IWYU pragma: export */ ++ ++/* Next, get the ISO C defines from GCC's limits.h. */ ++#undef _GCC_LIMITS_H_ + #include_next <limits.h> /* IWYU pragma: export */ + + #include <assert.h>
meta/recipes-core/systemd/systemd_261.3.bb +2 -0
@@ -38,6 +38,8 @@ SRC_URI += "file://touchscreen.rules \ file://0003-Do-not-create-var-log-README.patch \ " +SRC_URI:append:libc-musl = " file://0001-include-fix-limits.h-override.patch" + PAM_PLUGINS = " \ pam-plugin-unix \ pam-plugin-loginuid \
Re: [OE-core] [PATCH v2 1/3] systemd: fix musl build by fixing limits.h override
Quoted 5 lines · Jaipaul Cheernam via lists.openembedded.org
Backport the upstream fix for the
limits.hoverride, which breaks GCC's <limits.h> include cycle (notably GCC 16.x) and fails musl builds. musl only.+SRC_URI:append:libc-musl = " file://0001-include-fix-limits.h-override.patch"
Using the musl override like this might seem safer, but this is an upstream fix that can be applied everwhere. Also when upgrading to a version that includes this patch, it’s highly unlikely that the contributor will test the musl case, which will likely then mean the patch fails to apply.
I’ll post a v3 with this changed to expedite it merging, but if you are able to send your own with your authorship then please do!
Ross
PATCH v2 2/3 systemd: dlopen-deps: only warn for missing required providers
In reply to Jaipaul CheernamOptional recommended/suggested dlopen deps often have no provider, so musl builds print warnings like:
WARNING: cannot find any provider for dlopen dependency: libintl.so.8
Only warn for required deps, note the rest.
---
Changes in v2:
- Reword the commit message and shorten the inline comment.
meta/recipes-core/systemd/dlopen-deps.inc | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
meta/recipes-core/systemd/dlopen-deps.inc +7 -1
@@ -77,7 +77,13 @@ python package_generate_dlopen_deps() { found = True break if not found: - bb.warn(f"cannot find any provider for dlopen dependency: {dep['soname']}") + # Optional (recommended/suggested) dlopen deps may have + # no provider on some configs, so only warn for required ones. + msg = f"cannot find any provider for dlopen dependency: {dep['soname']}" + if dep["priority"] == "required": + bb.warn(msg) + else: + bb.note(msg) except oe.qa.NotELFFileError as e: bb.note(f"Cannot extract ELF notes: {e}") pass
Re: [OE-core] [PATCH v2 2/3] systemd: dlopen-deps: only warn for missing required providers
Quoted 7 lines · Jaipaul Cheernam via lists.openembedded.org
Optional recommended/suggested dlopen deps often have no provider, so musl builds print warnings like:
WARNING: cannot find any provider for dlopen dependency: libintl.so.8
Only warn for required deps, note the rest.
If optional dependencies no longer cause a warning then we’d never know that we should be either explicitly disabling them, or adding build dependencies.
The actual problem here is that systemd dlopen’s libintl, and under musl that doesn’t exist. Luckily, this has been fixed upstream:
https://github.com/systemd/systemd/commit/25714364a981e8e0ca3ac44ff09819a22a46b442
It’s late to backport this for blacksail, but the 262 release includes this change.
Ross
Re: [OE-core] [PATCH v2 2/3] systemd: dlopen-deps: only warn for missing required providers
Quoted 15 lines · Ross Burton via lists.openembedded.org
Quoted 2 lines · Jaipaul Cheernam via lists.openembedded.org
Optional recommended/suggested dlopen deps often have no provider, so
musl
Quoted 5 lines
builds print warnings like:
WARNING: cannot find any provider for dlopen dependency: libintl.so.8
Only warn for required deps, note the rest.
If optional dependencies no longer cause a warning then we’d never know that we should be either explicitly disabling them, or adding build dependencies.
Quoted 9 lines
The actual problem here is that systemd dlopen’s libintl, and under musl that doesn’t exist. Luckily, this has been fixed upstream:
https://github.com/systemd/systemd/commit/25714364a981e8e0ca3ac44ff09819a22a46b442
It’s late to backport this for blacksail, but the 262 release includes this change.
I guess you mean its late for first release of blacksail but I assume its ok to backport it to point releases later.
Quoted 13 lines
Ross
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#246579):
https://lists.openembedded.org/g/openembedded-core/message/246579
Mute This Topic: https://lists.openembedded.org/mt/121364963/1997914
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
PATCH v2 3/3 nfs-utils: upgrade 2.9.2 -> 3.1.1
In reply to Jaipaul CheernamChangeLog: https://www.kernel.org/pub/linux/utils/nfs-utils/3.1.1/3.1.1-Changelog
Drop two local patches now merged upstream:
- 0001-getport-fix-missing-stddef.h-inclusion.patch
- 0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch (upstream now gates HAVE_NFSD_NETLINK on --enable-nfsdctl)
Refresh the remaining local patches (0004, 0005) against the new source.
---
Changes in v2:
- No change since v1.
...tport-fix-missing-stddef.h-inclusion.patch | 27 -----------
.../0004-Use-nogroup-for-nobody-group.patch | 4 +-
.../0005-find-OE-provided-Kerberos.patch | 2 +-
...e-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch | 47 -------------------
...{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} | 4 +-
5 files changed, 4 insertions(+), 80 deletions(-)
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch
delete mode 100644 meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch
rename meta/recipes-connectivity/nfs-utils/{nfs-utils_2.9.2.bb => nfs-utils_3.1.1.bb} (96%)meta/recipes-connectivity/nfs-utils/nfs-utils/0001-getport-fix-missing-stddef.h-inclusion.patch deleted +0 -27
@@ -1,27 +0,0 @@ -From: Jaipaul Cheernam <[email protected]> -Date: Thu, 6 Aug 2026 10:55:00 +0000 -Subject: [PATCH] getport: fix missing stddef.h inclusion - -getport.c uses offsetof() but does not include <stddef.h>. On glibc -this works because offsetof is pulled in transitively, but on musl it -is not available without the explicit include, causing: - - getport.c:459:41: error: implicit declaration of function 'offsetof' - -Upstream-Status: Submitted [https://lore.kernel.org/linux-nfs/[email protected]/] -Signed-off-by: Jaipaul Cheernam <[email protected]> ---- - support/nfs/getport.c | 1 + - 1 file changed, 1 insertion(+) - -diff --git a/support/nfs/getport.c b/support/nfs/getport.c ---- a/support/nfs/getport.c -+++ b/support/nfs/getport.c -@@ -27,6 +27,7 @@ - #include <config.h> - #endif - -+#include <stddef.h> - #include <sys/types.h> - #include <sys/time.h> - #include <unistd.h>
meta/recipes-connectivity/nfs-utils/nfs-utils/0004-Use-nogroup-for-nobody-group.patch +2 -2
@@ -1,4 +1,4 @@ -From 5f4739475f3c01145174734ed8a3b31b3da99b77 Mon Sep 17 00:00:00 2001 +From f8407c33b26641c08ad76422ba2ff78214364b19 Mon Sep 17 00:00:00 2001 From: Daniel McGregor <[email protected]> Date: Tue, 6 Jun 2023 16:07:53 -0600 Subject: [PATCH] Use "nogroup" for nobody group
@@ -24,7 +24,7 @@ index 2a2f79a..e6f3724 100644 [Translation] diff --git a/utils/idmapd/idmapd.c b/utils/idmapd/idmapd.c -index 5231f56..e1aa8a5 100644 +index c41ae1f..7339a88 100644 --- a/utils/idmapd/idmapd.c +++ b/utils/idmapd/idmapd.c @@ -89,7 +89,7 @@
meta/recipes-connectivity/nfs-utils/nfs-utils/0005-find-OE-provided-Kerberos.patch +1 -1
@@ -1,4 +1,4 @@ -From a1759d9079a436096be5ce44622a3a5a85ec9faa Mon Sep 17 00:00:00 2001 +From 43a992e7bb491fd6e7e50861c9eaf0c549e98293 Mon Sep 17 00:00:00 2001 From: Daniel McGregor <[email protected]> Date: Wed, 8 Nov 2023 16:24:20 -0600 Subject: [PATCH] find OE provided Kerberos
meta/recipes-connectivity/nfs-utils/nfs-utils/0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch deleted +0 -47
47 lines not on this page · raw message
meta/recipes-connectivity/nfs-utils/nfs-utils_3.1.1.bb renamed from meta/recipes-connectivity/nfs-utils/nfs-utils_2.9.2.bb +1 -3
@@ -23,11 +23,9 @@ SRC_URI = "${KERNELORG_MIRROR}/linux/utils/nfs-utils/${PV}/nfs-utils-${PV}.tar.x file://nfscommon \ file://0004-Use-nogroup-for-nobody-group.patch \ file://0005-find-OE-provided-Kerberos.patch \ - file://0006-configure.ac-gate-HAVE_NFSD_NETLINK-on-enable-nfsdct.patch \ - file://0001-getport-fix-missing-stddef.h-inclusion.patch \ " -SRC_URI[sha256sum] = "e1dd8a9c95af15492065942cc3b52b1339ffd586baa2280ed86c9d3dc4097e8c" +SRC_URI[sha256sum] = "7dbc05eb8b32828b322872b759a62cb3979cb4118f344dd8616813d1390e84e0" # Only kernel-module-nfsd is required here (but can be built-in) - the nfsd module will # pull in the remainder of the dependencies.