Skip to contentSkip to conversation

scripts/gen-btf.sh: fix shebang for NixOS

PATCH In review Gary Guo 5 participants 10 messages Jan 20 – Jan 21 BPF

lore
Share & tools
lore

Gary Guo

author
#0 lore raw

PATCH 1/1 scripts/gen-btf.sh: fix shebang for NixOS

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Signature · 1 line
2.51.2
reviewer
#1 lore raw
Quoted 27 lines · Gary Guo

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #
Quoted 1 line

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Hi Gary, thanks for the patch.

I'm guessing this came up because gen-btf.sh is part of the vmlinux build, and link-vmlinux.sh has #!/bin/sh (not bash).

But I wonder if all of these need the same change then:

$ git log --oneline -n1
24d479d26b25 (HEAD, tag: v6.19-rc6) Linux 6.19-rc6
$ grep -r '#!/bin/bash' ./scripts/
./scripts/check-uapi.sh:#!/bin/bash
./scripts/decodecode:#!/bin/bash
./scripts/extract-fwblobs:#!/bin/bash
./scripts/gfp-translate:#!/bin/bash
./scripts/git-resolve.sh:#!/bin/bash
./scripts/mkuboot.sh:#!/bin/bash
./scripts/objdiff:#!/bin/bash
./scripts/objdump-func:#!/bin/bash
./scripts/prune-kernel:#!/bin/bash
./scripts/tags.sh:#!/bin/bash
./scripts/tracing/ftrace-bisect.sh:#!/bin/bash
./scripts/livepatch/klp-build:#!/bin/bash
#2 lore raw
Quoted 28 lines · Ihor Solodrai
Quoted 27 lines · Gary Guo

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #
Quoted 6 lines
Quoted 1 line

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Hi Gary, thanks for the patch.

I'm guessing this came up because gen-btf.sh is part of the vmlinux build, and link-vmlinux.sh has #!/bin/sh (not bash).

My typical experience is that when I see an error about /bin/bash not found, I just use "bash <script path>" instead without putting much think into it.

gen-btf.sh is part of build so I don't get to invoke the script directly, hence the patch.

Best,
Gary

Quoted 18 lines

But I wonder if all of these need the same change then:

$ git log --oneline -n1
24d479d26b25 (HEAD, tag: v6.19-rc6) Linux 6.19-rc6
$ grep -r '#!/bin/bash' ./scripts/
./scripts/check-uapi.sh:#!/bin/bash
./scripts/decodecode:#!/bin/bash
./scripts/extract-fwblobs:#!/bin/bash
./scripts/gfp-translate:#!/bin/bash
./scripts/git-resolve.sh:#!/bin/bash
./scripts/mkuboot.sh:#!/bin/bash
./scripts/objdiff:#!/bin/bash
./scripts/objdump-func:#!/bin/bash
./scripts/prune-kernel:#!/bin/bash
./scripts/tags.sh:#!/bin/bash
./scripts/tracing/ftrace-bisect.sh:#!/bin/bash
./scripts/livepatch/klp-build:#!/bin/bash
#3 lore raw
Quoted 33 lines · Gary Guo
Quoted 32 lines · Ihor Solodrai
Quoted 27 lines · Gary Guo

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #
Quoted 1 line

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Hi Gary, thanks for the patch.

Quoted 8 lines
Quoted 2 lines

I'm guessing this came up because gen-btf.sh is part of the vmlinux build, and link-vmlinux.sh has #!/bin/sh (not bash).

My typical experience is that when I see an error about /bin/bash not found, I just use "bash <script path>" instead without putting much think into it.

gen-btf.sh is part of build so I don't get to invoke the script directly, hence the patch.

For the build we have $(CONFIG_SHELL) to stick before the script. It is only 'sh', but it is the well-known variable users know to override. IMO all script invocations should use this mechanism.

Thomas

reviewer
#4 lore raw
Quoted 40 lines · Thomas Weißschuh
Quoted 39 lines · Gary Guo
Quoted 28 lines · Ihor Solodrai
Quoted 27 lines · Gary Guo

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #
Quoted 6 lines
Quoted 1 line

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Hi Gary, thanks for the patch.

I'm guessing this came up because gen-btf.sh is part of the vmlinux build, and link-vmlinux.sh has #!/bin/sh (not bash).

My typical experience is that when I see an error about /bin/bash not found, I just use "bash <script path>" instead without putting much think into it.

Quoted 6 lines
Quoted 2 lines

gen-btf.sh is part of build so I don't get to invoke the script directly, hence the patch.

For the build we have $(CONFIG_SHELL) to stick before the script. It is only 'sh', but it is the well-known variable users know to override. IMO all script invocations should use this mechanism.

There are two places where gen-btf.sh is invoked.

In link-vmlinux.sh:

if ! ${srctree}/scripts/gen-btf.sh .tmp_vmlinux1; then
echo >&2 "Failed to generate BTF for vmlinux" echo >&2 "Try to disable CONFIG_DEBUG_INFO_BTF" exit 1
fi

And in Makefile.modfinal:

quiet_cmd_btf_ko = BTF [M] $@

      cmd_btf_ko = 							\
	if [ ! -f $(objtree)/vmlinux ]; then				\
		printf "Skipping BTF generation for %s due to unavailability of vmlinux\n" $@ 1>&2; \
	else	\
		$(srctree)/scripts/gen-btf.sh --btf_base $(objtree)/vmlinux $@; \
	fi;

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Quoted 3 lines

Thomas

#5 lore raw
Quoted 68 lines · Ihor Solodrai
Quoted 40 lines · Thomas Weißschuh
Quoted 39 lines · Gary Guo
Quoted 28 lines · Ihor Solodrai
Quoted 27 lines · Gary Guo

From: Gary Guo <[email protected]>

NixOS only puts /usr/bin/env and /bin/sh at the standard location as required by POSIX, but not other shells. Other program that kernel build depends on is supplied via PATH, so shebang needs to use /usr/bin/env to find them.

This has been done to a few other scripts already, e.g. gen_test_kallsyms.sh or decode_stacktrace.sh.

Fixes 522397d05e7d ("resolve_btfids: Change in-place update with raw binary output")
---
 scripts/gen-btf.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
scripts/gen-btf.sh +1 -1
index be21ccee3487..4cd3159f2ddb 100755
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash
 # SPDX-License-Identifier: GPL-2.0
 # Copyright (c) 2025 Meta Platforms, Inc. and affiliates.
 #
Quoted 6 lines
Quoted 1 line

base-commit: 053966c344dbd346e71305f530e91ea77916189f

Hi Gary, thanks for the patch.

I'm guessing this came up because gen-btf.sh is part of the vmlinux build, and link-vmlinux.sh has #!/bin/sh (not bash).

My typical experience is that when I see an error about /bin/bash not found, I just use "bash <script path>" instead without putting much think into it.

Quoted 6 lines
Quoted 2 lines

gen-btf.sh is part of build so I don't get to invoke the script directly, hence the patch.

For the build we have $(CONFIG_SHELL) to stick before the script. It is only 'sh', but it is the well-known variable users know to override. IMO all script invocations should use this mechanism.

There are two places where gen-btf.sh is invoked.

In link-vmlinux.sh:

if ! ${srctree}/scripts/gen-btf.sh .tmp_vmlinux1; then
echo >&2 "Failed to generate BTF for vmlinux" echo >&2 "Try to disable CONFIG_DEBUG_INFO_BTF" exit 1
fi

And in Makefile.modfinal:

quiet_cmd_btf_ko = BTF [M] $@

      cmd_btf_ko = 							\
	if [ ! -f $(objtree)/vmlinux ]; then				\
		printf "Skipping BTF generation for %s due to unavailability of vmlinux\n" $@ 1>&2; \
	else	\
		$(srctree)/scripts/gen-btf.sh --btf_base $(objtree)/vmlinux $@; \
	fi;
Quoted 2 lines

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Documentation/kbuild/makefiles.rst says:

Make rules may invoke scripts to build the kernel. The rules shall always provide the appropriate interpreter to execute the script. They shall not rely on the execute bits being set, and shall not invoke the script directly. For the convenience of manual script invocation, such as invoking ./scripts/checkpatch.pl, it is recommended to set execute bits on the scripts nonetheless.

Kbuild provides variables $(CONFIG_SHELL), $(AWK), $(PERL), and $(PYTHON3) to refer to interpreters for the respective scripts.

So I suppose it would need to migate to POSIX shell...

Best,
Gary

reviewer
#6 lore raw
Quoted 14 lines · Gary Guo
Quoted 4 lines

[...]

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Documentation/kbuild/makefiles.rst says:

Make rules may invoke scripts to build the kernel. The rules shall always provide the appropriate interpreter to execute the script. They shall not rely on the execute bits being set, and shall not invoke the script directly. For the convenience of manual script invocation, such as invoking ./scripts/checkpatch.pl, it is recommended to set execute bits on the scripts nonetheless.

Quoted 5 lines

Kbuild provides variables $(CONFIG_SHELL), $(AWK), $(PERL), and $(PYTHON3) to refer to interpreters for the respective scripts.

So I suppose it would need to migate to POSIX shell...

That's my blunder then.

I'll send a patch tomorrow to make gen-btf.sh runnable with /bin/sh

Quoted 3 lines

Best,
Gary

#7 lore raw
Quoted 22 lines · Ihor Solodrai
Quoted 18 lines · Gary Guo
Quoted 4 lines

[...]

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Documentation/kbuild/makefiles.rst says:

Make rules may invoke scripts to build the kernel. The rules shall always provide the appropriate interpreter to execute the script. They shall not rely on the execute bits being set, and shall not invoke the script directly. For the convenience of manual script invocation, such as invoking ./scripts/checkpatch.pl, it is recommended to set execute bits on the scripts nonetheless.

Kbuild provides variables $(CONFIG_SHELL), $(AWK), $(PERL), and $(PYTHON3) to refer to interpreters for the respective scripts.

Quoted 5 lines
Quoted 1 line

So I suppose it would need to migate to POSIX shell...

That's my blunder then.

I'll send a patch tomorrow to make gen-btf.sh runnable with /bin/sh

Isn't the current patch enough?
-#!/bin/bash
+#!/usr/bin/env bash

I don't think all kernel build scripts are pure 'sh'.

reviewer
#8 lore raw
Quoted 33 lines · Alexei Starovoitov
Quoted 22 lines · Ihor Solodrai
Quoted 18 lines · Gary Guo
Quoted 4 lines

[...]

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Documentation/kbuild/makefiles.rst says:

Make rules may invoke scripts to build the kernel. The rules shall always provide the appropriate interpreter to execute the script. They shall not rely on the execute bits being set, and shall not invoke the script directly. For the convenience of manual script invocation, such as invoking ./scripts/checkpatch.pl, it is recommended to set execute bits on the scripts nonetheless.

Kbuild provides variables $(CONFIG_SHELL), $(AWK), $(PERL), and $(PYTHON3) to refer to interpreters for the respective scripts.

Quoted 5 lines
Quoted 1 line

So I suppose it would need to migate to POSIX shell...

That's my blunder then.

I'll send a patch tomorrow to make gen-btf.sh runnable with /bin/sh

Isn't the current patch enough?

(quoted diff) +1 -1
-#!/bin/bash
+#!/usr/bin/env bash
Quoted 1 line

I don't think all kernel build scripts are pure 'sh'.

Well, if we want to follow the makefiles.rst (and Thomas') recommendation, which is executing the script via $(CONFIG_SHELL), then we have to migrate to #!/bin/sh, because that's the default.

This concerns only scripts in the kernel build pipeline AFAIU.

But you may be right that this is not a strict rule, idk. It did come up quite late on the list.

#9 lore raw
Quoted 43 lines · Ihor Solodrai
Quoted 33 lines · Alexei Starovoitov
Quoted 22 lines · Ihor Solodrai
Quoted 18 lines · Gary Guo
Quoted 4 lines

[...]

So it's trivial to put $(CONFIG_SHELL) in front of it. But then it must be migrated to #!/bin/sh, right?

Documentation/kbuild/makefiles.rst says:

Make rules may invoke scripts to build the kernel. The rules shall always provide the appropriate interpreter to execute the script. They shall not rely on the execute bits being set, and shall not invoke the script directly. For the convenience of manual script invocation, such as invoking ./scripts/checkpatch.pl, it is recommended to set execute bits on the scripts nonetheless.

Kbuild provides variables $(CONFIG_SHELL), $(AWK), $(PERL), and $(PYTHON3) to refer to interpreters for the respective scripts.

Quoted 5 lines
Quoted 1 line

So I suppose it would need to migate to POSIX shell...

That's my blunder then.

I'll send a patch tomorrow to make gen-btf.sh runnable with /bin/sh

Isn't the current patch enough?

(quoted diff) +1 -1
-#!/bin/bash
+#!/usr/bin/env bash
Quoted 1 line

I don't think all kernel build scripts are pure 'sh'.

Well, if we want to follow the makefiles.rst (and Thomas') recommendation, which is executing the script via $(CONFIG_SHELL), then we have to migrate to #!/bin/sh, because that's the default.

Quoted 4 lines

This concerns only scripts in the kernel build pipeline AFAIU.

But you may be right that this is not a strict rule, idk. It did come up quite late on the list.

git grep CONFIG_SHELL|wc -l
68

ok, let's convert to pure sh then.