Skip to contentSkip to conversation

classes/sbom-cve-check: fall back to the stable SBOM symlink

PATCH In review [email protected] 3 participants 13 messages Aug 4 – Sep 3

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 classes/sbom-cve-check: fall back to the stable SBOM symlink

do_sbom_cve_check builds its input path from ${IMAGE_NAME}, which carries ${IMAGE_VERSION_SUFFIX} and so ${DATETIME}. That value is excluded from task hashes but changes on every bitbake invocation, so the path is only correct when do_create_image_sbom_spdx ran in the same invocation.

It does not when the class is newly enabled on an existing build tree: the image SBOM is already deployed and stamped from an earlier build, do_sbom_cve_check has never run, so bitbake executes only the latter and it looks for a file whose timestamp belongs to the current invocation. The same happens on any forced re-run of just this task:

  $ bitbake -f -c sbom_cve_check core-image-minimal
  ERROR: core-image-minimal-1.0-r0 do_sbom_cve_check: sbom-cve-check failed: [...]
  sbom-cve-check: error: [Errno 2] No such file or directory:
    '.../core-image-minimal-qemux86-64.rootfs-20260804101106.spdx.json'
  $ ls tmp/deploy/images/qemux86-64/*.spdx.json
  core-image-minimal-qemux86-64.rootfs-20260804095834.spdx.json
  core-image-minimal-qemux86-64.rootfs.spdx.json -> ...-20260804095834.spdx.json

The file is there under ${IMAGE_LINK_NAME}, the symlink do_create_image_sbom_spdx maintains. Keep preferring the timestamped name, so behaviour is unchanged whenever it exists, and fall back to the symlink rather than failing. Guarded on link_name being set, since IMAGE_LINK_NAME can be empty.

---
 meta/classes-recipe/sbom-cve-check.bbclass | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)
meta/classes-recipe/sbom-cve-check.bbclass +12 -1
index 451595f..b184c12 100644
@@ -14,9 +14,20 @@ python do_sbom_cve_check() {
     """
     Task: Run sbom-cve-check analysis on SBOM.
     """
-    sbom_path = d.expand("${DEPLOY_DIR_IMAGE}/${IMAGE_NAME}.spdx.json")
+    import os
+
     image_name = d.getVar("IMAGE_NAME")
     link_name = d.getVar("IMAGE_LINK_NAME")
+    deploy_dir = d.getVar("DEPLOY_DIR_IMAGE")
+
+    # IMAGE_NAME carries DATETIME, which changes on every invocation while being
+    # excluded from task hashes, so this path is only valid when
+    # do_create_image_sbom_spdx ran in the same invocation. Fall back to the
+    # symlink it maintains, which is stable across builds.
+    sbom_path = os.path.join(deploy_dir, "%s.spdx.json" % image_name)
+    if not os.path.exists(sbom_path) and link_name:
+        sbom_path = os.path.join(deploy_dir, "%s.spdx.json" % link_name)
+
     run_sbom_cve_check(d, sbom_path, image_name, link_name)
 }
 
Signature · 1 line
2.55.0
#1 lore raw
Quoted 20 lines · On Tue, 2026-08-04 at 13:24 +0200, [email protected] wrote:

do_sbom_cve_check builds its input path from ${IMAGE_NAME}, which carries ${IMAGE_VERSION_SUFFIX} and so ${DATETIME}. That value is excluded from task hashes but changes on every bitbake invocation, so the path is only correct when do_create_image_sbom_spdx ran in the same invocation.

It does not when the class is newly enabled on an existing build tree: the image SBOM is already deployed and stamped from an earlier build, do_sbom_cve_check has never run, so bitbake executes only the latter and it looks for a file whose timestamp belongs to the current invocation. The same happens on any forced re-run of just this task:

  $ bitbake -f -c sbom_cve_check core-image-minimal
  ERROR: core-image-minimal-1.0-r0 do_sbom_cve_check: sbom-cve-check failed: [...]
  sbom-cve-check: error: [Errno 2] No such file or directory:
    '.../core-image-minimal-qemux86-64.rootfs-20260804101106.spdx.json'
  $ ls tmp/deploy/images/qemux86-64/*.spdx.json
  core-image-minimal-qemux86-64.rootfs-20260804095834.spdx.json
  core-image-minimal-qemux86-64.rootfs.spdx.json -> ...-20260804095834.spdx.json
Quoted 6 lines

The file is there under ${IMAGE_LINK_NAME}, the symlink do_create_image_sbom_spdx maintains. Keep preferring the timestamped name, so behaviour is unchanged whenever it exists, and fall back to the symlink rather than failing. Guarded on link_name being set, since IMAGE_LINK_NAME can be empty.

Hi Thomas,

This code has already changed recently in commit 2a80840a4fe0 ("sbom-cve-check: Fix breakage with empty IMAGE_LINK_NAME"). Perhaps we should have just added a fallback to IMAGE_NAME if IMAGE_LINK_NAME is unset.

A quick scan of the code shows that there are other places where it is assumed that IMAGE_LINK_NAME is valid, such as vex.bbclass.

Best regards,

Signature · 1 line
Paul Barker
#2 lore raw

v2 cover Handle an empty IMAGE_LINK_NAME consistently

v1 fixed the sbom-cve-check input path. Paul noted that 2a80840a4fe0 ("sbom-cve-check: Fix breakage with empty IMAGE_LINK_NAME") had already touched this class, and that other classes make the same assumption. That commit guarded the export symlinks; the input path is separate and still built from IMAGE_NAME. Patch 1 now uses the shape Paul suggested, and patches 2-4 fix the other sites found by auditing meta/classes*, meta/lib/oe and meta/lib/oeqa.

Read and write sites need opposite handling of an empty value: read sites fall back to IMAGE_NAME, write sites skip the link. A central default (pin IMAGE_LINK_NAME to IMAGE_NAME when empty) does not work, as create_symlinks and multiubi_mkfs have no link != target check and IMAGE_NAME is itself derived from IMAGE_LINK_NAME.

Tested on qemux86-64, each fix confirmed in both directions. Patch 5 covers patches 2-4 in one build; patch 1 was checked by hand, as a selftest would need the sbom-cve-check databases (~8.7G of git).

AI-Generated: Uses Claude Opus 5

Thomas Roos (5):
sbom-cve-check: read the image SBOM through the stable symlink
vex: skip the VEX symlink when IMAGE_LINK_NAME is empty
testexport: fall back to IMAGE_NAME when IMAGE_LINK_NAME is empty
image_types: skip multiubi symlinks when IMAGE_LINK_NAME is empty oeqa/selftest/imagefeatures: test an empty IMAGE_LINK_NAME

 meta/classes-recipe/image_types.bbclass       |  2 +-
 meta/classes-recipe/sbom-cve-check.bbclass    |  3 +-
 meta/classes-recipe/testexport.bbclass        |  2 +-
 meta/classes/vex.bbclass                      |  3 +-
 meta/lib/oeqa/selftest/cases/imagefeatures.py | 42 +++++++++++++++++++
 5 files changed, 48 insertions(+), 4 deletions(-)
Signature · 1 line
2.55.0
#3 lore raw

PATCH v2 1/5 sbom-cve-check: read the image SBOM through the stable symlink

The input path was built from IMAGE_NAME, which carries DATETIME and so only resolves within the invocation that wrote the SBOM: enabling the class on an existing build tree, or "bitbake -f -c sbom_cve_check", failed with ENOENT. Read IMAGE_LINK_NAME instead, falling back to IMAGE_NAME when it is unset.

AI-Generated: Uses Claude Opus 5

---
 meta/classes-recipe/sbom-cve-check.bbclass | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
meta/classes-recipe/sbom-cve-check.bbclass +2 -1
index 451595f..ca47834 100644
@@ -14,9 +14,10 @@ python do_sbom_cve_check() {
     """
     Task: Run sbom-cve-check analysis on SBOM.
     """
-    sbom_path = d.expand("${DEPLOY_DIR_IMAGE}/${IMAGE_NAME}.spdx.json")
     image_name = d.getVar("IMAGE_NAME")
     link_name = d.getVar("IMAGE_LINK_NAME")
+    sbom_path = os.path.join(d.getVar("DEPLOY_DIR_IMAGE"),
+                             "%s.spdx.json" % (link_name or image_name))
     run_sbom_cve_check(d, sbom_path, image_name, link_name)
 }
 
Signature · 1 line
2.55.0
#4 lore raw

PATCH v2 2/5 vex: skip the VEX symlink when IMAGE_LINK_NAME is empty

In reply to Thomas Roos

An empty IMAGE_LINK_NAME left a bare ".vex.json" symlink in the deploy directory. Skip the link, as image.bbclass and rootfs-postcommands do.

AI-Generated: Uses Claude Opus 5

---
 meta/classes/vex.bbclass | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
meta/classes/vex.bbclass +2 -1
index b3c2aca..eee3af6 100644
@@ -204,7 +204,8 @@ python vex_write_rootfs_manifest () {
     with open(manifest_name, "w") as f:
         json.dump(json_data, f, indent=2)
 
-    update_symlinks(manifest_name, link_path)
+    if link_name:
+        update_symlinks(manifest_name, link_path)
     bb.plain("Image VEX JSON report stored in: %s" % manifest_name)
 }
 
Signature · 1 line
2.55.0
#5 lore raw

Re: [PATCH v2 2/5] vex: skip the VEX symlink when IMAGE_LINK_NAME is empty

Quoted 14 lines · Thomas Roos

An empty IMAGE_LINK_NAME left a bare ".vex.json" symlink in the deploy directory. Skip the link, as image.bbclass and rootfs-postcommands do.

AI-Generated: Uses Claude Opus 5

---
 meta/classes/vex.bbclass | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
meta/classes/vex.bbclass +0 -0
index b3c2aca..eee3af6 100644

meta/classes/vex.bbclass python vex_write_rootfs_manifest () { · line 208

Quoted 6 lines
meta/classes/vex.bbclass +2 -1
@@ -204,7 +204,8 @@ python vex_write_rootfs_manifest () {
     with open(manifest_name, "w") as f:
         json.dump(json_data, f, indent=2)
 
-    update_symlinks(manifest_name, link_path)
+    if link_name:
+        update_symlinks(manifest_name, link_path)

Using link_name in the conditional and link_path in the function argument is confusing. Perhaps we should move the link_path assignment inside this conditional? It's not used anywhere else.

if link_name:
link_path = os.path.join(deploy_dir, "%s.vex.json" % link_name) update_symlinks(manifest_name, link_path)

Best regards,

Signature · 1 line
Paul Barker
#6 lore raw

Re: [PATCH v2 2/5] vex: skip the VEX symlink when IMAGE_LINK_NAME is empty

Quoted 26 lines · Paul Barker
Quoted 14 lines · Thomas Roos

An empty IMAGE_LINK_NAME left a bare ".vex.json" symlink in the deploy directory. Skip the link, as image.bbclass and rootfs-postcommands do.

AI-Generated: Uses Claude Opus 5

---
 meta/classes/vex.bbclass | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
meta/classes/vex.bbclass +0 -0
index b3c2aca..eee3af6 100644

meta/classes/vex.bbclass python vex_write_rootfs_manifest () { · line 208

Quoted 6 lines
meta/classes/vex.bbclass +2 -1
@@ -204,7 +204,8 @@ python vex_write_rootfs_manifest () {
     with open(manifest_name, "w") as f:
         json.dump(json_data, f, indent=2)
 
-    update_symlinks(manifest_name, link_path)
+    if link_name:
+        update_symlinks(manifest_name, link_path)

Using link_name in the conditional and link_path in the function argument is confusing. Perhaps we should move the link_path assignment inside this conditional? It's not used anywhere else.

Quoted 3 lines

if link_name:
link_path = os.path.join(deploy_dir, "%s.vex.json" % link_name) update_symlinks(manifest_name, link_path)

Actually, ignore this, I'm going in the wrong direction here.... I'll reply to the cover letter about the whole series.

Sorry for the noise!

Signature · 1 line
Paul Barker
#7 lore raw

PATCH v2 3/5 testexport: fall back to IMAGE_NAME when IMAGE_LINK_NAME is empty

In reply to Thomas Roos

copy_needed_files() built the manifest and testdata paths from IMAGE_LINK_NAME alone, so an empty value made it copy ".manifest" and fail. testexport_main() in the same file already falls back to IMAGE_NAME.

AI-Generated: Uses Claude Opus 5

---
 meta/classes-recipe/testexport.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes-recipe/testexport.bbclass +1 -1
index 12b7e17..a0be7cd 100644
@@ -134,7 +134,7 @@ def copy_needed_files(d, tc):
 
     # Copy test data
     image_name = ("%s/%s" % (d.getVar('DEPLOY_DIR_IMAGE'),
-                            d.getVar('IMAGE_LINK_NAME')))
+                            d.getVar('IMAGE_LINK_NAME') or d.getVar('IMAGE_NAME')))
     image_manifest = "%s.manifest" % image_name
     tdname = "%s.testdata.json" % image_name
     test_data_path = os.path.join(export_path, 'data')
Signature · 1 line
2.55.0
#8 lore raw

PATCH v2 4/5 image_types: skip multiubi symlinks when IMAGE_LINK_NAME is empty

In reply to Thomas Roos

An empty IMAGE_LINK_NAME made multiubi_mkfs link the named volumes from the suffix alone, e.g. "_volume.ubifs". Guard the block.

AI-Generated: Uses Claude Opus 5

---
 meta/classes-recipe/image_types.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes-recipe/image_types.bbclass +1 -1
index ca13729..6c620a5 100644
@@ -217,7 +217,7 @@ multiubi_mkfs() {
 	mv ubinize${vname}-${IMAGE_NAME}.cfg ${IMGDEPLOYDIR}/
 
 	# Create own symlinks for 'named' volumes
-	if [ -n "$vname" ]; then
+	if [ -n "$vname" ] && [ -n "${IMAGE_LINK_NAME}" ]; then
 		cd ${IMGDEPLOYDIR}
 		if [ -e ${IMAGE_NAME}${vname}.ubifs ]; then
 			ln -sf ${IMAGE_NAME}${vname}.ubifs \
Signature · 1 line
2.55.0
#9 lore raw

PATCH v2 5/5 oeqa/selftest/imagefeatures: test an empty IMAGE_LINK_NAME

In reply to Thomas Roos

Nothing covered this configuration, which is why the vex, multiubi and testexport sites went unnoticed. One core-image-minimal build with IMAGE_LINK_NAME cleared covers all three; the test fails without them.

AI-Generated: Uses Claude Opus 5

---
 meta/lib/oeqa/selftest/cases/imagefeatures.py | 42 +++++++++++++++++++
 1 file changed, 42 insertions(+)
meta/lib/oeqa/selftest/cases/imagefeatures.py +42 -0
index 87c3da2..fac1d47 100644
@@ -332,3 +332,45 @@ CORE_IMAGE_EXTRA_INSTALL = "man-pages"
             status, output = qemu.run_serial("man --pager=cat intro")
             self.assertEqual(status, 1, 'Failed to run man: %s' % (output))
             self.assertIn("introduction to user commands", output)
+
+    def test_empty_image_link_name(self):
+        """
+        Test that an empty IMAGE_LINK_NAME suppresses the artifact symlinks
+        and that the classes reading an artifact back still locate it.
+        """
+        image = 'core-image-minimal'
+        vname = 'mtd_4_256'
+        config = """
+INHERIT += "vex"
+IMAGE_CLASSES += "testexport"
+IMAGE_LINK_NAME = ""
+IMAGE_NAME = "${IMAGE_BASENAME}${IMAGE_MACHINE_SUFFIX}${IMAGE_NAME_SUFFIX}"
+IMAGE_FSTYPES += "multiubi"
+MULTIUBI_BUILD = "%s"
+MKUBIFS_ARGS_%s ?= "-m 4096 -e 253952 -c 4096"
+UBINIZE_ARGS_%s ?= "-m 4096 -p 256KiB"
+TEST_TARGET_IP = "192.168.7.2"
+TEST_SERVER_IP = "192.168.7.1"
+""" % (vname, vname, vname)
+        self.write_config(config)
+
+        bitbake(image)
+        bb_vars = get_bb_vars(['DEPLOY_DIR_IMAGE', 'IMAGE_NAME', 'TEST_EXPORT_DIR'], image)
+        deploy_dir = bb_vars['DEPLOY_DIR_IMAGE']
+
+        # Artifacts are still named after IMAGE_NAME
+        for name in ("%s.vex.json" % bb_vars['IMAGE_NAME'],
+                     "%s_%s.ubifs" % (bb_vars['IMAGE_NAME'], vname)):
+            self.assertTrue(os.path.exists(os.path.join(deploy_dir, name)),
+                            "%s is missing from %s" % (name, deploy_dir))
+
+        # ...and nothing is linked from the suffix alone
+        for name in (".vex.json", "_%s.ubifs" % vname, "_%s.ubi" % vname):
+            self.assertFalse(os.path.lexists(os.path.join(deploy_dir, name)),
+                             "%s was created from an empty IMAGE_LINK_NAME" % name)
+
+        # testexport copies these back out of DEPLOY_DIR_IMAGE
+        bitbake("-c testexport %s" % image)
+        for name in ('manifest', 'testdata.json'):
+            path = os.path.join(bb_vars['TEST_EXPORT_DIR'], 'data', name)
+            self.assertTrue(os.path.exists(path), "%s was not exported" % path)
Signature · 1 line
2.55.0
#10 lore raw

Re: [PATCH v2 0/5] Handle an empty IMAGE_LINK_NAME consistently

In reply to Thomas Roos
Quoted 14 lines · Thomas Roos

v1 fixed the sbom-cve-check input path. Paul noted that 2a80840a4fe0 ("sbom-cve-check: Fix breakage with empty IMAGE_LINK_NAME") had already touched this class, and that other classes make the same assumption. That commit guarded the export symlinks; the input path is separate and still built from IMAGE_NAME. Patch 1 now uses the shape Paul suggested, and patches 2-4 fix the other sites found by auditing meta/classes*, meta/lib/oe and meta/lib/oeqa.

Read and write sites need opposite handling of an empty value: read sites fall back to IMAGE_NAME, write sites skip the link. A central default (pin IMAGE_LINK_NAME to IMAGE_NAME when empty) does not work, as create_symlinks and multiubi_mkfs have no link != target check and IMAGE_NAME is itself derived from IMAGE_LINK_NAME.

Quoted 5 lines

Tested on qemux86-64, each fix confirmed in both directions. Patch 5 covers patches 2-4 in one build; patch 1 was checked by hand, as a selftest would need the sbom-cve-check databases (~8.7G of git).

AI-Generated: Uses Claude Opus 5

Hi Thomas,

I discussed this with Richard, we want to avoid increasing the complexity of the code unnecessarily.

IMAGE_LINK_NAME is intended to always be valid, we need the link because IMAGE_NAME is dependent on the date and time that an image was built. Do we really need to support the case where IMAGE_LINK_NAME is unset?

If we don't need to support IMAGE_LINK_NAME being unset then perhaps we should just print a better error message in that case and give up.

Best regards,

Signature · 1 line
Paul Barker
#11 lore raw

Re: [PATCH v2 0/5] Handle an empty IMAGE_LINK_NAME consistently

Hi Paul,

Retested the original do_sbom_cve_check bug.
If you enable the sbom_cve_check class after a successful build it fails as do_create_image_sbom_spdx stays cached from the earlier build and its output keeps the old build's DATETIME. Result:

sbom-cve-check: error: [Errno 2] No such file or directory:

    '.../core-image-minimal-qemux86-64.rootfs-20260902135131.spdx.json'

Before: path built as ${DEPLOY_DIR_IMAGE}/${IMAGE_NAME}.spdx.json as IMAGE_NAME carries DATETIME, so it only matches the SBOM written in the same bitbake run.

After: build path from IMAGE_LINK_NAME instead (link_name or image_name, falls back when link name empty). IMAGE_LINK_NAME is the stable, non-timestamped symlink do_create_image_sbom_spdx maintains pointing at the current SBOM. create-spdx-image-3.0.bbclass keeps it updated to the latest build's file. Reading through it means do_sbom_cve_check finds the SBOM regardless of which invocation wrote it, since the symlink always points at the current one, not just this run's timestamp.

bitbake core-image-minimal # build once, class
not yet enabled
bitbake-config-build --set-fragment sbom-cve-check # or add fragment
manually
git checkout 1a32e30~1 -- meta/classes-recipe/sbom-cve-check.bbclass #
pre-fix code
bitbake core-image-minimal # ENOENT on
do_sbom_cve_check
git checkout HEAD -- meta/classes-recipe/sbom-cve-check.bbclass #
restore fix
bitbake -c sbom_cve_check core-image-minimal -f #
succeeds

The pattern is pre-existing, not part of this series, it was already established:

- meta/classes-recipe/testexport.bbclass:52 (testexport_main()) —
d.getVar('IMAGE_LINK_NAME') or d.getVar('IMAGE_NAME')
- meta/classes-recipe/testimage.bbclass:228 — same IMAGE_LINK_NAME or
IMAGE_NAME fallback, reading testdata.json back

Plus the empty-guard pattern (skip symlink creation rather than fail):

- meta/classes-recipe/image.bbclass:610 (create_symlinks) — if not
link_name: return
- meta/classes-recipe/rootfs-postcommands.bbclass:402,479
(write_image_manifest) — if ... and link_name
- meta/classes-recipe/qemuboot.bbclass:147 — if
d.getVar('IMAGE_LINK_NAME'): before writing the symlink
- meta/classes-recipe/kernel.bbclass:804 — if [ -n
"${KERNEL_IMAGE_LINK_NAME}" ] before the kernel symlink
- meta/classes-recipe/license_image.bbclass:300 — same link_name guard
pattern
- meta/classes-recipe/baremetal-image.bbclass:73 — same

I'm not dying if these series are not merged, I just think this fix is useful and there are already places using the same pattern! Please correct me if I'm wrong, also agree on not adding more code complexity. May a central place is better, but I do not know one!

Cheers, Thomas

Am Do., 27. Aug. 2026 um 17:37 Uhr schrieb Paul Barker <[email protected]>:

Quoted 43 lines · Thomas Roos
Quoted 10 lines

v1 fixed the sbom-cve-check input path. Paul noted that 2a80840a4fe0 ("sbom-cve-check: Fix breakage with empty IMAGE_LINK_NAME") had already touched this class, and that other classes make the same assumption. That commit guarded the export symlinks; the input path is separate and still built from IMAGE_NAME. Patch 1 now uses the shape Paul suggested, and patches 2-4 fix the other sites found by auditing meta/classes*, meta/lib/oe and meta/lib/oeqa.

Read and write sites need opposite handling of an empty value: read sites fall back to IMAGE_NAME, write sites skip the link. A central default

(pin

Quoted 1 line

IMAGE_LINK_NAME to IMAGE_NAME when empty) does not work, as

create_symlinks

Quoted 4 lines

and multiubi_mkfs have no link != target check and IMAGE_NAME is itself derived from IMAGE_LINK_NAME.

Tested on qemux86-64, each fix confirmed in both directions. Patch 5

covers

Quoted 1 line

patches 2-4 in one build; patch 1 was checked by hand, as a selftest

would

Quoted 3 lines

need the sbom-cve-check databases (~8.7G of git).

AI-Generated: Uses Claude Opus 5

Hi Thomas,

I discussed this with Richard, we want to avoid increasing the complexity of the code unnecessarily.

IMAGE_LINK_NAME is intended to always be valid, we need the link because IMAGE_NAME is dependent on the date and time that an image was built. Do we really need to support the case where IMAGE_LINK_NAME is unset?

If we don't need to support IMAGE_LINK_NAME being unset then perhaps we should just print a better error message in that case and give up.

Best regards,

--
Paul Barker

#12 lore raw
In reply to Paul Barker

Hi Paul,
updated patch and added also a fix for the other places. Together with a test case. Let me know if such patches are not welcome, acc. to the discussion of yesterdays engineering sync. I do not want to loose my reputation and end up in hell ;)
I just want to contribute a bit back to the greatest project in the world!
So please tell me!

Cheers, Thomas