Skip to contentSkip to conversation

t/gpg: simplify test for unknown key

PATCH Superseded Fabian Stelzer 5 participants 10 messages Jan 7, 2022 – Jan 14, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 t/gpg: simplify test for unknown key

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to an empty directory instead, therefore making all used keys unknown for this single verify-commit call.

---
This was reported by Ævar in <[email protected]>. Just using an empty keyring / gpg homedir should achieve the same effect and keeps the stress of generating a gpg key out of the CI.

 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +2 -20
index 9882b69ae2..2d38580847 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	GNUPGHOME=./empty_home test_must_fail git verify-commit initial 2>actual &&
 	! grep "Good signature from" actual &&
 	! grep "BAD signature from" actual &&
 	grep -q -F -e "No public key" -e "public key not found" actual
Signature · 1 line
2.33.1
#1 lore raw
Quoted 7 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to an empty directory instead, therefore making all used keys unknown for this single verify-commit call.

Quoted 6 lines

---
This was reported by Ævar in <[email protected]>. Just using an empty keyring / gpg homedir should achieve the same effect and keeps the stress of generating a gpg key out of the CI.

Clever. Losing lines of code and gaining more stability in CI is a great thing.

Quoted 45 lines

 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +2 -20
index 9882b69ae2..2d38580847 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	GNUPGHOME=./empty_home test_must_fail git verify-commit initial 2>actual &&
 	! grep "Good signature from" actual &&
 	! grep "BAD signature from" actual &&
 	grep -q -F -e "No public key" -e "public key not found" actual
#2 lore raw
In reply to Fabian Stelzer
Quoted 7 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to an empty directory instead, therefore making all used keys unknown for this single verify-commit call.

Quoted 6 lines

---
This was reported by Ævar in <[email protected]>. Just using an empty keyring / gpg homedir should achieve the same effect and keeps the stress of generating a gpg key out of the CI.

Thanks, it would be great to have this in and before v2.35.0. I've run into several boxes (on the GCC farm) that hang without this patch.

Quoted 35 lines
 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +1 -19
index 9882b69ae2..2d38580847 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '

t/t7510-signed-commit.sh test_expect_success GPG 'verify and show signatures' ' · line 114

Quoted 5 lines
t/t7510-signed-commit.sh +1 -1
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	GNUPGHOME=./empty_home test_must_fail git verify-commit initial 2>actual &&

Before I noticed this thread (I looked at
https://lore.kernel.org/git/20211230111038.jtoqytdhkilv2732@fs/ first, and the In-Reply-To chain wasn't connected) I was about to submit exactly this patch for you but with:

- test_must_fail git verify-commit thirteenth-signed 2>actual &&
+ test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&

Both of those are probably a good thing to do here. I.e.:

1. Didn't we have portability issues with "ENV_VAR=VALUE shell_function ..." ?
2. You're pointing to a nonexisting ./empty_home, but shouldn't we use $GNUPGHOME_NOT_USED? The existing "show unknown signature with custom format" test in the same file does that.

#3 lore raw
Quoted 57 lines · Ævar Arnfjörð Bjarmason
Quoted 7 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to an empty directory instead, therefore making all used keys unknown for this single verify-commit call.

Quoted 6 lines

---
This was reported by Ævar in <[email protected]>. Just using an empty keyring / gpg homedir should achieve the same effect and keeps the stress of generating a gpg key out of the CI.

Thanks, it would be great to have this in and before v2.35.0. I've run into several boxes (on the GCC farm) that hang without this patch.

Quoted 37 lines
 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +1 -19
index 9882b69ae2..2d38580847 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
Quoted 7 lines

t/t7510-signed-commit.sh test_expect_success GPG 'verify and show signatures' ' · line 114

Quoted 3 lines
t/t7510-signed-commit.sh +1 -1
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	GNUPGHOME=./empty_home test_must_fail git verify-commit initial 2>actual &&

Before I noticed this thread (I looked at
https://lore.kernel.org/git/20211230111038.jtoqytdhkilv2732@fs/ first, and the In-Reply-To chain wasn't connected)

Yeah, sorry about that. I forgot to add the in-reply-to :/

Quoted 9 lines

I was about to submit
exactly this patch for you but with:

- test_must_fail git verify-commit thirteenth-signed 2>actual &&
+ test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&

Both of those are probably a good thing to do here. I.e.:

1. Didn't we have portability issues with "ENV_VAR=VALUE shell_function ..." ?

I'm not good with portability stuff and trust your judgment on this.

Quoted 3 lines

2. You're pointing to a nonexisting ./empty_home, but shouldn't we use $GNUPGHOME_NOT_USED? The existing "show unknown signature with custom format" test in the same file does that.

I was not aware of $GNUPGHOME_NOT_USED but it is used in a similar fashion. However it is set to the old value of $GNUPGHOME before we change it in lib-gpg.sh which seems wrong to me. Wouldn't it then just pick up the gpg homedir of whatever the test environment has? Using the variable is good, but i would set it to a known empty directory or?

#4 lore raw
Quoted 11 lines · Fabian Stelzer
Quoted 9 lines

I was about to submit
exactly this patch for you but with:

- test_must_fail git verify-commit thirteenth-signed 2>actual &&
+ test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&

Both of those are probably a good thing to do here. I.e.:

1. Didn't we have portability issues with "ENV_VAR=VALUE shell_function ..." ?

I'm not good with portability stuff and trust your judgment on this.

See [1] and the ensuing discussion for a good summary. Re-reading that thread and comparing it with what we see with git grep test_must_fail env -- t confirms that that

test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit ...

is the right thing to do here.

Quoted 10 lines
Quoted 3 lines

2. You're pointing to a nonexisting ./empty_home, but shouldn't we use $GNUPGHOME_NOT_USED? The existing "show unknown signature with custom format" test in the same file does that.

I was not aware of $GNUPGHOME_NOT_USED but it is used in a similar fashion. However it is set to the old value of $GNUPGHOME before we change it in lib-gpg.sh which seems wrong to me. Wouldn't it then just pick up the gpg homedir of whatever the test environment has?
Using the variable is good, but i would set it to a known empty directory or?

Yeah, t7510 captures the value of $GNUPGHOME as $GNUPGHOME_NOT_USED before sourcing t/lib-gpg.sh. So long as nobody else has tampered with $GNUPGHOME, they should get $TRASH_DIRECTORY/gnupg-home-not-used.

But I'm less certain that there isn't somebody accidentally ignoring the "not-used" portion of the test $GNUPGHOME ;).

And I don't think that reasoning through it all is that worthwhile, so I'm fine with what a much more direct ./empty_home here.

Thanks,
Taylor

[1]: https://lore.kernel.org/git/[email protected]/

#5 lore raw
Quoted 13 lines · Taylor Blau
Quoted 11 lines · Fabian Stelzer
Quoted 9 lines

I was about to submit
exactly this patch for you but with:

- test_must_fail git verify-commit thirteenth-signed 2>actual &&
+ test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&

Both of those are probably a good thing to do here. I.e.:

1. Didn't we have portability issues with "ENV_VAR=VALUE shell_function ..." ?

I'm not good with portability stuff and trust your judgment on this.

Quoted 7 lines

See [1] and the ensuing discussion for a good summary. Re-reading that thread and comparing it with what we see with git grep test_must_fail env -- t confirms that that

test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit ...

is the right thing to do here.

Thanks. Interesting

Quoted 16 lines
Quoted 11 lines
Quoted 4 lines

2. You're pointing to a nonexisting ./empty_home, but shouldn't we use $GNUPGHOME_NOT_USED? The existing "show unknown signature with custom format" test in the same file does that.

I was not aware of $GNUPGHOME_NOT_USED but it is used in a similar fashion. However it is set to the old value of $GNUPGHOME before we change it in lib-gpg.sh which seems wrong to me. Wouldn't it then just pick up the gpg homedir of whatever the test environment has?
Using the variable is good, but i would set it to a known empty directory or?

Yeah, t7510 captures the value of $GNUPGHOME as $GNUPGHOME_NOT_USED before sourcing t/lib-gpg.sh. So long as nobody else has tampered with $GNUPGHOME, they should get $TRASH_DIRECTORY/gnupg-home-not-used.

Quoted 6 lines

But I'm less certain that there isn't somebody accidentally ignoring the "not-used" portion of the test $GNUPGHOME ;).

And I don't think that reasoning through it all is that worthwhile, so I'm fine with what a much more direct ./empty_home here.

Yeah. I checked the other uses of GNUPGHOME and the NOT_USED seems fine (and more consistent with existing tests). So let's use it. Even if some other (non gpg related) test accidentally pollutes this gpghome it would surely not do so with the actual signing key used in the test suite.

Thanks, I'll send a new patch in a second

Quoted 4 lines

Thanks,
Taylor

[1]: https://lore.kernel.org/git/[email protected]/

#6 lore raw

PATCH v2 1/1 t/gpg: simplify test for unknown key

In reply to Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to the empty $GNUPGHOME_NOT_USED instead, therefore making all used keys unknown for this single verify-commit call.

---
 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +2 -20
index 9882b69ae2..8593b7e3cb 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&
 	! grep "Good signature from" actual &&
 	! grep "BAD signature from" actual &&
 	grep -q -F -e "No public key" -e "public key not found" actual
Signature · 1 line
2.34.1
Reviewed-by on v2 1/1
#7 lore raw
Quoted 9 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to the empty $GNUPGHOME_NOT_USED instead, therefore making all used keys unknown for this single verify-commit call.

#8 lore raw
In reply to Fabian Stelzer
Quoted 10 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to the empty $GNUPGHOME_NOT_USED instead, therefore making all used keys unknown for this single verify-commit call.

---

The original one is already in 'next' so I'll revert the merge and queue this one instead.

We could turn this into incremental and apply it on the original one, which may give us a chance to highlight this common mistake of using a single-shot environment with shell functions, but let's not bother.

Thanks.

Quoted 43 lines
 t/t7510-signed-commit.sh | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)
t/t7510-signed-commit.sh +2 -20
index 9882b69ae2..8593b7e3cb 100755
@@ -71,25 +71,7 @@ test_expect_success GPG 'create signed commits' '
 	git tag eleventh-signed $(cat oid) &&
 	echo 12 | git commit-tree --gpg-sign=B7227189 HEAD^{tree} >oid &&
 	test_line_count = 1 oid &&
-	git tag twelfth-signed-alt $(cat oid) &&
-
-	cat >keydetails <<-\EOF &&
-	Key-Type: RSA
-	Key-Length: 2048
-	Subkey-Type: RSA
-	Subkey-Length: 2048
-	Name-Real: Unknown User
-	Name-Email: [email protected]
-	Expire-Date: 0
-	%no-ask-passphrase
-	%no-protection
-	EOF
-	gpg --batch --gen-key keydetails &&
-	echo 13 >file && git commit -a -S"[email protected]" -m thirteenth &&
-	git tag thirteenth-signed &&
-	DELETE_FINGERPRINT=$(gpg -K --with-colons --fingerprint --batch [email protected] | grep "^fpr" | head -n 1 | awk -F ":" "{print \$10;}") &&
-	gpg --batch --yes --delete-secret-keys $DELETE_FINGERPRINT &&
-	gpg --batch --yes --delete-keys [email protected]
+	git tag twelfth-signed-alt $(cat oid)
 '
 
 test_expect_success GPG 'verify and show signatures' '
@@ -129,7 +111,7 @@ test_expect_success GPG 'verify and show signatures' '
 '
 
 test_expect_success GPG 'verify-commit exits failure on unknown signature' '
-	test_must_fail git verify-commit thirteenth-signed 2>actual &&
+	test_must_fail env GNUPGHOME="$GNUPGHOME_NOT_USED" git verify-commit initial 2>actual &&
 	! grep "Good signature from" actual &&
 	! grep "BAD signature from" actual &&
 	grep -q -F -e "No public key" -e "public key not found" actual
Reviewed-by on v1 1/1
#9 lore raw
In reply to Fabian Stelzer
Quoted 7 lines · Fabian Stelzer

To test for a key that is completely unknown to the keyring we need one to sign the commit with. This was done by generating a new key and not add it into the keyring. To avoid the key generation overhead and problems where GPG did hang in CI during it, switch GNUPGHOME to an empty directory instead, therefore making all used keys unknown for this single verify-commit call.

Quoted 6 lines

---
This was reported by Ævar in <[email protected]>. Just using an empty keyring / gpg homedir should achieve the same effect and keeps the stress of generating a gpg key out of the CI.

Looks good to me.