CVE Feed

A vulnerability has been found in itsourcecode Sales and Inventory System 1.0. Impacted is an unknown function of the file /pages/sup_searchfrm.php. The manipulation of the argument ID leads to sql injection. The attack can be initiated remotely. The exploit has been disclosed to the public and may be used.

A flaw has been found in wger-project wger up to 2.6.0-alpha2. This issue affects the function reset_user_password of the file wger/gym/views/gym.py of the component Password Reset. Executing a manipulation can lead to cross-site request forgery. It is possible to launch the attack remotely. This patch is called 3c6ce4b7f3eeafeb35318c6c4e82b1a3fd28b314. It is advisable to implement a patch to correct this issue.

Readest is an open-source e-book reader built on Tauri. In versions prior to 0.11.16, EPUB chapter HTML is sanitized with DOMPurify using a configuration that forbade only the <script> tag (FORBID_TAGS: ['script']) in apps/readest-app/src/services/transformers/sanitizer.ts. DOMPurify does not parse the contents of the srcdoc attribute on <iframe> elements, treating it as an opaque string attribute, so an attacker who can get an <iframe> element to survive sanitization can embed a complete HTML document containing a <script> tag inside srcdoc and have it execute when the browser renders the iframe. The content iframe is configured with sandbox="allow-same-origin allow-scripts", so script executing inside it shares the parent origin and can reach parent.parent.__TAURI_INTERNALS__.invoke(...), giving access to every Tauri IPC command the application is permitted to use, which escalates to arbitrary code execution. The payload can be made invisible (zero-size, transparent iframe) so the reader sees only normal book text. Version 0.11.16 hardened the sanitizer configuration by adding 'iframe', 'object' and 'embed' to FORBID_TAGS and adding 'srcdoc' to FORBID_ATTR.

Keploy versions 3.1.0 through 3.6.25, fixed in 3.6.26, bind the agent control-plane HTTP server to all interfaces without authentication, exposing endpoints that stream TLS session keys and traffic data. Attackers can access the /agent/pcap/keylog endpoint to retrieve NSS keylog lines and decrypt recorded TLS traffic, or invoke /agent/stop and /agent/storemocks to manipulate recording sessions.

browser-use web-ui versions 2.0.0 through 3.0.0 write configured LLM API keys to disk in cleartext without encryption or access restrictions. Attackers with read access to the temporary settings directory can recover provider API keys from predictably-named JSON files.

NextChat versions from 2.15.8 through 2.16.1 contain an improper URL validation vulnerability in the proxy endpoint that allows attackers to obtain the server's OpenAI API key. The x-base-url header is validated using substring matching instead of hostname parsing, allowing any URL containing 'api.openai.com' to pass validation and receive the server's credentials in the Authorization header.

jina-ai reader disables its private-address guard outside Google Cloud deployments, allowing unauthenticated attackers to perform server-side request forgery. Attackers can supply publicly resolvable hostnames mapping to private addresses to retrieve cloud metadata and internal service content.

browser-use web-ui versions 2.0.0 through 3.0.0 fail to validate browser settings paths in run_agent_task, allowing attackers to create directories at arbitrary locations by supplying absolute paths to save_recording_path, save_trace_path, save_agent_history_path, or save_download_path parameters. Attackers can exploit this via the unauthenticated Gradio interface to create directories anywhere the root-running container has write access.

Qubes OS before qubes-core-dom0-linux 4.3.22 allows OS command injection during a qvm-copy-to-vm call from dom0 to an attacker-controlled qube, because the "system" library function is used to process an error message that may have shell metacharacters. This occurs in core-admin-linux/file-copy-vm/qfile-dom0-agent.c.

A vulnerability was detected in vastsa FileCodeBox up to 2.3. This vulnerability affects the function update_file_usage of the file apps/base/views.py of the component Pickup Limit Handler. Performing a manipulation results in race condition. It is possible to initiate the attack remotely. The exploit is now public and may be used. Upgrading to version 2.5.0 is able to resolve this issue. The patch is named 8d7d856c62d73badd0797eb4daec8d2ff10a403a. Upgrading the affected component is recommended.

Frappe Framework development builds contain an authorization flaw in the render_jinja_template endpoint that allows low-privileged users to render arbitrary Jinja templates by supplying raw template strings. Attackers with print permission on any document can execute arbitrary SELECT statements against unrelated tables, including reading password hashes from the __Auth table.

Dolibarr versions 10.0.0 before 24.0.0 fail to perform per-object authorization checks in the Users::getGroups REST API endpoint, allowing authenticated users to retrieve group memberships of other users. Attackers can call GET /users/{id}/groups with arbitrary user identifiers to access group names, entity associations, and private notes across tenant boundaries.

A weakness has been identified in Tenda HG10 300001138. Affected by this issue is the function formIPv6Routing of the file /boaform/admin/formIPv6Routing of the component Boa Web Server. This manipulation of the argument destNet causes buffer overflow. The attack is possible to be carried out remotely. The exploit has been made available to the public and could be used for attacks.

Pake before 3.13.1 joins the JavaScript-supplied filename for the download_file Tauri command onto the user's Downloads directory with no sanitization. A filename containing path traversal sequences (for example ../Library/LaunchAgents/com.evil.plist) or an absolute path resolves outside ~/Downloads. The command then fetches attacker-controlled content from the supplied URL (via Rust HTTP, not the browser) and writes it to that path. A script that can invoke the command can overwrite user-writable files and install persistence (macOS LaunchAgents, Linux autostart, Windows Startup), leading to code execution in the user account. All desktop apps generated from an affected Pake tree expose the same command.

Incorrect Authorization vulnerability in ash-project ash_sql allows a caller to bypass a scoping or authorization filter expressed as exists/2 over a relationship that declares both a limit (or from_many?) and a parent(...)-referencing filter or sort. AshSql.Join.related_query/3 skips the caller-supplied exists predicate for such relationships and delegates it to limit_from_many/5. When the relationship's own filter or sort references parent(...), limit_from_many/5 takes a branch that drops both the limit and the predicate, emitting a bare correlated EXISTS with no predicate. The check then matches any record that has any related row. Most severely, when the expression backs a policy (for example authorize_if expr(exists(memberships, user_id == ^actor(:id)))), the actor-scoping condition disappears and the policy passes for any actor with any related row. This issue affects ash_sql: from 0.4.1 before 0.7.1.

A security flaw has been discovered in itsourcecode Sales and Inventory System 1.0. Affected by this vulnerability is an unknown functionality of the file /pages/sup_edit.php. The manipulation of the argument ID results in sql injection. The attack can be executed remotely. The exploit has been released to the public and may be used for attacks.

reset_password.html parses query string parameters and uses the 'url' parameter as a redirection target (window.location = url) after password reset, optionally delayed by a 'delay' parameter. No validation or allowlisting is performed on url, enabling an attacker to redirect users to an arbitrary external site after completion of the password-reset workflow.

Incorrect Authorization vulnerability in ash-project ash_sql allows a caller to receive an aggregate value computed over rows a more restrictive filter should have excluded, disclosing counts, sums, or lists across an authorization or tenancy boundary. AshSql.Aggregate.different_queries?/2 reports two aggregate queries as different only when their filter and their sort both differ. Aggregate queries rarely carry a sort, so two aggregates that share a name but carry entirely different filters compare as identical. The colliding aggregate keeps its name and is treated as already computed, and select_aggregates returns the first-registered variant's value. The same name reaches the builder twice with different filters when actor or tenant context is stamped into each aggregate's query, so a narrowly filtered aggregate can be served the value of a previously registered broad one. This issue affects ash_sql: from 0.1.0 before 0.7.1.

Incorrect Authorization vulnerability in ash-project ash_sql allows a caller in a schema-based multitenant application to receive aggregate values computed from another tenant's rows. When an aggregate is computed over a distinct query, AshSql.AggregateQuery.add_single_aggs/5 rebuilds the outer query from query.from.source alone, which is only the {table, schema} tuple and does not carry query.prefix or query.from.prefix. For strategy(:context) multitenancy those hold the tenant schema, so the rebuilt outer query reads the repo's default schema while the inner correlated subquery still reads the tenant schema, and the two are joined only on primary key. The aggregate, and any relationship join added off the prefix-less binding, is then computed against the wrong tenant's rows. The neighbouring limit and exists branches instead wrap the query with subquery/1, which preserves the prefix. This issue affects ash_sql: from 0.1.0 before 0.7.1.

Improper Neutralization of Special Elements in Data Query Logic vulnerability in ash-project ash_sql allows a user who supplies a search term to contains/2, string_starts_with/2, or string_ends_with/2 to inject live SQL LIKE wildcards, turning a literal substring search into an attacker-controlled pattern match. The escape helpers in AshSql.Expr prefix % and _ with a backslash but never escape a backslash already present in the input. Because backslash is the default LIKE escape character, the escaping defeats itself: the input \% becomes the pattern fragment \\%, where \\ is a literal backslash and the attacker's % remains a live wildcard. The search value stays parameterized, so this is confined to the LIKE pattern grammar rather than full SQL. An attacker can widen matches to probe values, slip past a negated contains(...) guard, or crash the query with a trailing lone backslash. This issue affects ash_sql: from 0.1.1-rc.10 before 0.7.1.

Incorrect Comparison vulnerability in ash-project ash_sql allows a user to pad a string field with tab, newline, carriage-return, or form-feed characters and pass a trimmed uniqueness or equality check in the database that the same expression would fail in memory (or the reverse). string_trim/1 compiles to REGEXP_REPLACE patterns built from an Elixir string in which \s is the escape for a single space (codepoint 32), not a regex whitespace class. The generated SQL therefore removes only literal spaces and leaves tabs, newlines, carriage returns, and form feeds in place, whereas String.trim/1 in Elixir removes them all. Any Ash filter, validation, or identity that relies on string_trim/1 then behaves differently depending on whether Ash pushes the expression down to SQL or evaluates it in memory, so padded input can register a near-duplicate value or slip past a trimmed comparison. This issue affects ash_sql: from 0.1.0 before 0.7.1.

Uncontrolled Recursion vulnerability in ash-project ash_oban allows a user who can drive a trigger's on_error action to fail on the final attempt to exhaust worker CPU and memory, denying service. The generated worker's atomic handle_error/4 runs the trigger's on_error action on a job's final attempt inside a rescue that, when the action itself raises, calls handle_error/4 again with the same job. The job's attempt still equals max_attempts, so it re-enters the same clause and re-runs the failing action, with no exit. Any deterministic on_error failure (a data-layer outage, a misconfigured action, or a record the action rejects) loops forever; because the recursive call is not in tail position, each iteration retains a formatted stacktrace and the process heap grows without bound while the failing statement is re-issued against the data layer until the runtime kills the worker. This issue affects ash_oban: from 0.8.0-rc.1 before 0.8.14.

Improperly Controlled Modification of Dynamically-Determined Object Attributes vulnerability in ash-project ash_oban allows a user whose input reaches the :args option of AshOban.build_trigger/3 to retarget an update or destroy trigger at another record, including across tenants. build_trigger/3 builds the trusted job arguments with atom keys (:primary_key, :tenant, :action_arguments) and merges the caller's :args underneath so the trusted values win on collision. Because Oban job arguments round-trip through JSON, the caller's keys arrive as strings, so Map.merge sees no collision and both keys survive. When the job is persisted the JSON object is de-duplicated keeping the last (string) key, and the worker reads the caller's value. The documentation describes :args as unable to affect the action, so an application that forwards user input into it for uniqueness scoping is exposed to authorization bypass and tenant isolation breaks. This issue affects ash_oban: from 0.2.5 before 0.8.14.

A vulnerability was identified in itsourcecode Sales and Inventory System 1.0. Affected is an unknown function of the file /pages/cust_searchfrm.php. The manipulation of the argument ID leads to sql injection. Remote exploitation of the attack is possible. The exploit is publicly available and might be used.

A vulnerability was determined in TOTOLINK A720R 4.1.5cu.630_B20250509. This impacts the function setMacFilterRules of the file cstecgi.cgi of the component MAC Filtering. Executing a manipulation of the argument desc can lead to memory corruption. The attack may be launched remotely. The exploit has been publicly disclosed and may be utilized.

A vulnerability was identified in Beetel 450TC3 01.00.00_01. This vulnerability affects unknown code of the component User Management. The manipulation of the argument Username leads to cross site scripting. The attack is possible to be carried out remotely. The exploit is publicly available and might be used. The vendor was contacted early about this disclosure but did not respond in any way.

A vulnerability was determined in Beetel 450TC3 01.00.00_01. This affects an unknown part. Executing a manipulation can lead to weak password recovery. The attack can be executed remotely. The exploit has been publicly disclosed and may be utilized. The vendor was contacted early about this disclosure but did not respond in any way.

A vulnerability was found in SiteServer SSCMS 7.4.0. Affected by this issue is some unknown functionality of the component Agent Installation Workflow. Performing a manipulation of the argument SecurityKey results in improper access controls. Remote exploitation of the attack is possible. The attack is considered to have high complexity. The exploitation is known to be difficult. The project was informed of the problem early through an issue report but has not responded yet.

A vulnerability has been found in itsourcecode Sales and Inventory System 1.0. Affected by this vulnerability is an unknown functionality of the file /pages/pro_edit.php. Such manipulation of the argument ID leads to sql injection. The attack may be launched remotely. The exploit has been disclosed to the public and may be used.

A flaw has been found in itsourcecode Sales and Inventory System 1.0. Affected is an unknown function of the file /pages/emp_searchfrm.php. This manipulation of the argument ID causes sql injection. The attack may be initiated remotely. The exploit has been published and may be used.

A vulnerability was detected in coppermine-gallery Coppermine Photo Gallery up to 1.6.28. This impacts an unknown function of the file db_input.php of the component Hidden Album Update Endpoint. The manipulation results in cross site scripting. The attack can be launched remotely. The exploit is now public and may be used. Upgrading to version 1.6.29 will fix this issue. It is recommended to upgrade the affected component.

A security vulnerability has been detected in coppermine-gallery Coppermine Photo Gallery up to 1.6.28. This affects an unknown function of the file profile.php of the component edit_profile Endpoint. The manipulation of the argument Biography leads to cross site scripting. The attack can be initiated remotely. The exploit has been disclosed publicly and may be used. Upgrading to version 1.6.29 mitigates this issue. Upgrading the affected component is recommended.

The Really Simple Security WordPress plugin before 9.8.0 does not check that the user is allowed to install Really Simple Security WordPress plugin before 9.8.0 before installing one from a user-supplied URL, allowing an administrator of a subsite on a multisite network to install and execute arbitrary code in the network-shared Really Simple Security WordPress plugin before 9.8.0 directory, which WordPress otherwise reserves to the network administrator. Exploitation requires the network administrator to have enabled the Really Simple Security WordPress plugin before 9.8.0 administration menu for subsites, which is not the default.

The Groundhogg — CRM, Newsletters, and Marketing Automation WordPress plugin before 4.5.13 does not validate or escape values submitted to some optional web form fields before storing them and outputting them back in an administrative area, allowing unauthenticated users to perform Stored Cross-Site Scripting attacks against high privilege users.

The MW WP Form WordPress plugin before 5.1.6 does not sanitise and escape some of its form settings before outputting them back in an admin dashboard page, which could allow users with a role as low as Editor to perform Stored Cross-Site Scripting attacks against high privilege users such as admin.

The Customer Reviews for WooCommerce WordPress plugin before 5.118.0 does not sanitise and escape the content of customer reviews received via one of its endpoints, which could allow unauthenticated users to perform Stored Cross-Site Scripting attacks.

The WPvivid — Backup, Migration & Staging WordPress plugin before 0.9.133 does not validate the destination of files extracted from a backup package during restoration, allowing high privilege users such as administrators to write arbitrary files outside the intended restore directory, which can lead to code execution.

The geotargetingwp WordPress plugin before 3.5.6.2 does not sanitise or escape several parameters before reflecting them back in AJAX responses that are served with an HTML content type, allowing unauthenticated attackers to inject arbitrary web scripts that execute when a victim is tricked into submitting a crafted request.

The SOGO Add Script to Individual Pages Header Footer WordPress plugin through 3.9 does not sanitise or escape the custom header/footer script values saved from its post metabox, and does not restrict them to users with the unfiltered_html capability, allowing users with contributor-level access and above to store JavaScript that executes in the browser of any administrator who reviews the post and of any visitor once the post is published.

A security flaw has been discovered in NASA cFS up to 7.0.1. The affected element is the function CFE_SB_GetUserDataLength of the file src/cFS/cfe/modules/sb/fsw/src/cfe_sb_util.c of the component cFE Software Bus. Performing a manipulation of the argument TotalMsgSize/HdrSize results in integer underflow. It is possible to initiate the attack remotely. The vendor was contacted early about this disclosure but did not respond in any way.

A vulnerability was identified in NASA cFS up to 7.0.1. Impacted is the function OS_read of the file modules/protocol/tcp/fsw/src/sbn_tcp_if.c of the component SBN TCP Module. Such manipulation of the argument MsgSz leads to buffer overflow. The attack must be carried out from within the local network. The vendor was contacted early about this disclosure but did not respond in any way.

The MyHome Core plugin for WordPress is vulnerable to Authentication Bypass in all versions up to, and including, 4.4.5. This is due to missing authorization in the send_link() AJAX handler and improper token validation in the activate() function. This makes it possible for unauthenticated attackers to generate an activation token for an unconfirmed user account and obtain a valid authentication cookie for that account, including administrators. Successful exploitation requires the MyHome theme to be configured in legacy/WPBakery mode with frontend registration and confirmation email enabled, and the target account must not already have the myhome_agent_confirmed user meta set.

A vulnerability was determined in NASA Trick 19.6.0. This issue affects the function JSONVariableServerThread::parse_request of the file trick_source/sim_services/JSONVariableServer/JSONVariableServerThread.cpp of the component TCP Socket Handler. This manipulation causes stack-based buffer overflow. The attack is possible to be carried out remotely. The vendor was contacted early about this disclosure but did not respond in any way.

Improper Neutralization of Special Elements in Data Query Logic vulnerability in ash-project ash_sqlite allows an attacker who controls a get_path/2 segment to traverse into nested JSON the application never exposed, disclosing private or sensitive? embedded fields. AshSqlite.SqlImplementation builds the SQLite json_extract path with "$." <> Enum.join(right, "."), so a single segment containing ., [, ], or $ re-interprets the JSON path (for example "private.secret" descends two levels instead of naming one key). The path is bound as a parameter, so this is confined to the JSON-path grammar rather than SQL. Any endpoint that lets user input reach a get_path segment (a common pick-a-field pattern) can read nested values it never meant to expose. This issue affects ash_sqlite: from 0.1.2-rc.0 before 0.2.18.

Improper Verification of Cryptographic Signature vulnerability in erlef oidcc allows an unauthenticated attacker to impersonate an arbitrary user via an encrypted ID token or JARM response carrying no nested signature. OpenID Connect Core 1.0 section 2 requires that an encrypted ID token be signed then encrypted, with the result being a Nested JWT, and JARM processing rule 5 requires the client to check the signature unconditionally. oidcc instead accepted a JWE wrapping unsigned claims as fully validated, so anyone holding the relying party's public encryption key could mint a token with an arbitrary sub, iss, and aud without possessing the provider's signing key. In oidcc_jwt_util:verify_decrypted_token/4, a decrypted payload that is not a signed JWS fell back to parsing the plaintext claims and returning them with no verifying key. oidcc_token:int_validate_jwt/4 then matched on the JOSE structure type rather than on whether a signature had been verified, and returned success. The JARM path in oidcc_token:validate_jarm/3 is reachable through the browser front channel. UserInfo responses are not affected, because OpenID Connect Core 1.0 section 5.3.2 permits them to be encrypted without also being signed. This issue affects oidcc: from 3.2.0-beta.1 before 3.9.0.

Inefficient Algorithmic Complexity vulnerability in ash-project ash_paper_trail allows a user who can submit a large array attribute to a paper-trailed create or update action to cause a denial of service through excessive CPU and memory use. With full-diff change tracking, AshPaperTrail.ChangeBuilders.FullDiff.ListChange pairs each prior array element against the new list by rebuilding the remaining-elements accumulator with acc ++ [tuple] on every step, copying the growing list each time, so the pairing scales cubically in the array length. Nothing bounds the length and the value comes straight from action input, so one request carrying a large accepted {:array, _} attribute forces tens of seconds of CPU and multi-gigabyte allocations. This issue affects ash_paper_trail: from 0.1.1 before 0.7.0.

Cleartext Storage of Sensitive Information vulnerability in ash-project ash_paper_trail allows an attacker with read access to the generated version resource to recover sensitive values nested inside embedded resources, unions, or lists. sensitive_attributes :redact and :ignore only act on the tracked resource's top-level attributes. maybe_redact_changes/3 and the stored-action-input path in AshPaperTrail.Resource.Changes.CreateNewVersion derive the sensitive set from the resource's own attributes and never descend into embedded, union, or list values, so a non-sensitive attribute or action argument that holds an embed with a sensitive? field (for example an accepted credentials embed carrying a token) is written to the version table in cleartext. This issue affects ash_paper_trail: from 0.3.0 before 0.7.0.

Cleartext Storage of Sensitive Information vulnerability in ash-project ash_paper_trail allows an attacker with read access to the generated version resource to recover the plaintext of sensitive? attributes. AshPaperTrail stores the values of tracked sensitive? attributes in the generated version resource's changes map, which is declared public? true and sensitive? false, so the values are returned by the version resource's default read action and printed in logs, inspect output, and error messages instead of being redacted. AshPaperTrail.Resource.Transformers.CreateVersionResource derives the changes map's sensitivity from the ignore_attributes list (the attributes excluded from changes) rather than from the tracked attributes actually stored in it, and ignore_attributes defaults to empty, so the flag is effectively always false. This issue affects ash_paper_trail: from 0.1.1 before 0.7.0.

### Summary When `qs.parse` is called with `comma: true` and `throwOnLimitExceeded: true`, a comma-separated value under a bracket-push key (`a[]=1,2,3,4`) is split into an array without being compared against `arrayLimit`, while the same value under a flat key (`a=1,2,3,4`), an indexed key (`a[0]=`), a nested key (`a[b]=`), or a dotted key (`a.b=` with `allowDots`) throws the documented `RangeError`. A single parameter such as `a[]=1,2,2,...` therefore produces an inner array of arbitrary length even though the caller opted into the hard limit. This is the `[]=` key form that the fix for CVE-2026-2391 (qs 6.14.2) did not cover. ### Details In `lib/parse.js`, a comma-separated value under a `[]=` key is split and then wrapped as a single nested element (`val = [val]`, so that each `a[]=x,y` group counts as one element of the outer array). The `arrayLimit` check that 6.14.2 added for comma values runs after that wrap, so for `[]=` parts it only ever saw the wrapper of length 1. 6.15.3 added a pre-split comma count so that an oversized value throws before it is allocated, but gated it on an `isFlatArrayValue` flag that `parseValues` set to `false` for any part containing `[]=`, and did not pass it for object-valued input, so the gap remained. #### PoC ```js var qs = require('qs'); var options = { comma: true, arrayLimit: 3, throwOnLimitExceeded: true }; qs.parse('a=1,2,3,4', options); // RangeError: Array limit exceeded. Only 3 elements allowed in an array. qs.parse('a[]=1,2,3,4', options); // { a: [ [ '1', '2', '3', '4' ] ] } (no throw) qs.parse('a[]=' + '1,'.repeat(1000000) + '1', { comma: true, arrayLimit: 20, throwOnLimitExceeded: true }); // no throw; a 1,000,001-element inner array is allocated ``` #### Fix `lib/parse.js`, applied in 8859c37 on `main` and released as v6.16.0: the `isFlatArrayValue` gate is removed, so every comma-split value is counted against `arrayLimit` before splitting regardless of key form. An in-limit group under `a[]=` still counts as one element of the outer array, and the default (`throwOnLimitExceeded: false`) path is unchanged. ### Affected versions `>=6.14.2 <6.16.0`, fixed in v6.16.0. v6.14.2 introduced `arrayLimit` enforcement for comma values (the fix for CVE-2026-2391) but only for values not under a `[]=` key, and every release from v6.14.2 through v6.15.3 has the same gap. v6.14.0 and v6.14.1, where `throwOnLimitExceeded` exists but does not apply to any comma form, are covered by CVE-2026-2391 rather than this record. Earlier lines (6.7.x through 6.13.x) have `comma` but no `throwOnLimitExceeded`, so there is no hard cap on any comma path to bypass; releases before 6.7.0 have no `comma` option. ### Impact An unauthenticated attacker who can reach an application that parses untrusted query strings or urlencoded bodies with both `comma: true` and `throwOnLimitExceeded: true` (both non-default) can bypass the configured limit with a single `a[]=` parameter and force the parser to allocate an array proportional to the request size. The cost is strictly linear in the attacker-supplied bytes (about 0.1 microseconds and 6 to 7 retained bytes per input byte; the same out-of-memory threshold as the documented default `throwOnLimitExceeded: false` path), so a transport-layer request or body size limit bounds it completely (and node's default maximum HTTP header size of 16 KB already bounds the request line, so multi-megabyte payloads need a body parser). The impact is that an opt-in hard limit fails open on one key spelling, not unbounded allocation from a small input.

### Summary `qs.stringify` throws a `TypeError` when it serializes an object whose own `constructor` property has a truthy, non-callable `isBuffer` member. `utils.isBuffer` duck-types buffers by calling `obj.constructor.isBuffer(obj)` after checking only that the property is truthy, so a value such as `{ constructor: { isBuffer: "x" } }` makes the call throw `TypeError: obj.constructor.isBuffer is not a function`. ### Details `lib/stringify.js:127` calls `utils.isBuffer` on every non-primitive value it serializes. `utils.isBuffer` (`lib/utils.js:332`) reads `obj.constructor.isBuffer` and invokes it without verifying that it is a function. `constructor` and `isBuffer` are ordinary property names, so any object carrying them as own properties reaches the unchecked call. Such an object can be built from untrusted input. `qs.parse("x[constructor][isBuffer]=y", { plainObjects: true })` or `{ allowPrototypes: true }` keeps the `constructor` key as an own property (the default parse options drop it), and `JSON.parse("{\"a\":{\"constructor\":{\"isBuffer\":\"x\"}}}")` produces the same shape with no qs option involved. Express 4 with its default `query parser` setting and body-parser with `extended: true` both call `qs.parse` with `allowPrototypes: true`, so on those stacks `req.query` and `req.body` can carry the shape directly. #### PoC ```js var qs = require("qs"); qs.stringify(qs.parse("x[constructor][isBuffer]=y", { plainObjects: true })); qs.stringify(JSON.parse("{\"a\":{\"constructor\":{\"isBuffer\":\"x\"}}}")); // TypeError: obj.constructor.isBuffer is not a function // at Object.isBuffer (lib/utils.js:332:78) // at stringify (lib/stringify.js:127:45) ``` #### Fix `lib/utils.js`, applied in e83d321 on `main` and released as v6.16.0: ```diff - return !!(obj.constructor && obj.constructor.isBuffer && obj.constructor.isBuffer(obj)); + return !!(obj.constructor && typeof obj.constructor.isBuffer === "function" && obj.constructor.isBuffer(obj)); ``` Real `Buffer`, `safer-buffer`, and browserify `buffer` polyfill instances serialize exactly as before; only the throw is removed. ### Affected versions `>=2.2.5 <6.16.0`, fixed in v6.16.0. The unguarded duck-type was introduced in 3768a75 and first shipped in v2.2.5 (September 2014). v2.2.4 and earlier used `Buffer.isBuffer` and are not affected. Every release from v2.2.5 through v6.15.3 contains the unguarded call. ### Impact An unauthenticated request can make any code path that re-serializes attacker-influenced data with `qs.stringify` (for example, rebuilding a query string from `req.query` for a redirect or an upstream request, or serializing a parsed JSON body) throw synchronously. In a typical Node.js HTTP framework the throw is caught by the framework error boundary and the affected request returns a 500; the process survives and other requests are unaffected. Where the call runs outside an error boundary, such as an `async` Express 4 handler (where the throw becomes an unhandled promise rejection) or a background job, the process exits, so the impact in that case depends on the application error handling rather than on qs.