AIOHTTP versions 3.9.1 up to but not including 4.0.0 contain a vulnerability where multipart parts encoded with base64 can bypass the client_max_size limit. This occurs because the BodyPartReader.read_chunk() method reads until the chunk length is a multiple of 4 without enforcing a byte limit, allowing an attacker to cause excessive memory consumption or event loop blocking by sending specially crafted multipart requests. The issue is fixed in version 4.0.0.
AIOHTTP versions from 3.9.1 up to but not including 4.0.0 have a vulnerability where the Request.post() method does not limit the number of URL-encoded form fields. This can cause the event loop to be blocked due to quadratic time complexity when processing many repeated keys in a MultiDict. The issue is fixed in version 4.0.0.
AIOHTTP's client CookieJar component improperly parsed Set-Cookie headers, creating extra cookies from unrecognized name=value attributes without limiting the number of stored cookies. This allowed a malicious server to cause excessive memory and CPU usage on the client by making it store a very large number of cookies, resulting in oversized Cookie headers on subsequent requests. The issue affects versions from 3.9.1 up to but not including 4.0.0 and is fixed in version 4.0.0.
AIOHTTP's Application.add_domain() function had a performance issue where using two or more wildcards in a domain mask caused polynomial-time processing of long Host headers, leading to event loop stalls. This could allow an attacker to stall the server by sending requests with long Host headers against such domain masks. The issue affects versions from 3.9.1 up to but not including 4.0.0 and is fixed in version 4.0.0.
The NLTK library's dot2img function executes the Graphviz 'dot' binary without applying the trusted-execution security checks that NLTK uses elsewhere. This allows a local unprivileged attacker on a POSIX system to place a malicious 'dot' executable in a writable directory on the victim's PATH, leading to arbitrary code execution when dot2img is invoked. The vulnerability arises because dot2img only performs a basic binary search without ownership or environment scrubbing, unlike other NLTK wrappers. This flaw enables local code execution in the victim's user context without requiring control over NLTK inputs.
The NLTK library's sql_query() function in the chat80 module bypasses the pathsec sandbox intended to restrict SQLite database access to specific data roots. It does so by following symbolic links inside trusted data roots that point to SQLite databases outside those roots, allowing read access to out-of-sandbox databases. This occurs because sql_query() uses nltk.data.find() which only performs lexical checks and does not resolve symlinks, unlike other functions that enforce path containment. An attacker with write access to a trusted NLTK data root can exploit this to read sensitive SQLite databases elsewhere on the filesystem.
NLTK's save_maxent_params() and load_maxent_params() functions contain a path traversal vulnerability that allows arbitrary file read and write outside the intended directories. This occurs because these functions construct file paths from a user-controlled argument and open files directly without enforcing NLTK's pathsec security mechanism. The vulnerability bypasses the filesystem sandboxing intended by pathsec, potentially allowing attackers to read or overwrite files outside permitted directories. The issue has been fixed in NLTK version 3.8.1_3 and later.
AIOHTTP versions 0.6.0 up to but not including 0.7.0 and versions 0.8.0 are affected by a vulnerability in the WebSocket receive queue handling. The issue allows the receive queue limits to be bypassed by sending many small or compressed messages, especially when permessage-deflate compression is used. This can cause excessive memory consumption on the server or client side. A fix is available in version 0.7.0.
AIOHTTP versions >=0.6.0 <0.7.0 and >=0.8.0 contain a vulnerability where decompressing concatenated compressed streams (gzip, deflate, or zstd) causes excessive CPU usage due to quadratic time complexity. This can block the event loop of servers or clients when processing specially crafted compressed payloads. The issue is fixed in version 0.7.0.
A vulnerability in AIOHTTP's pure-Python HTTP parser causes the request buffer to grow excessively when handling declined upgrade requests, leading to repeated processing of pipelined requests and excessive memory use. This affects users relying on the pure-Python parser, such as those setting AIOHTTP_NO_EXTENSIONS or not using the standard C extension wheel. A patch is available in version 0.7.0 that fixes this issue.
AIOHTTP versions >=0.6.0 <0.7.0 and >=0.8.0 have a vulnerability where data received during an upgrade request (such as a WebSocket handshake) is buffered without limit until the handler completes. This can lead to unbounded memory usage if an attacker streams data behind the upgrade request to a handler that does not promptly accept or reject the upgrade. A fix is available in version 0.7.0.
AIOHTTP versions >=0.6.0 <0.7.0 and >=0.8.0 contain a vulnerability in the WebSocket max_msg_size check due to an integer overflow in the C extension. This flaw allows an attacker to bypass the max_msg_size limit by sending a crafted WebSocket message with a declared length near 2^63, causing unbounded memory usage. The issue is fixed in version 0.7.0. A workaround is to disable C extensions by setting AIOHTTP_NO_EXTENSIONS=1 to use the unaffected Python reader.
In nanobot, the WebSocket gateway allows a standard client holding a valid client token to select an existing directory outside the operator-configured workspace as the root of a new conversation. The agent can then read and expose files in that directory that are readable by the nanobot service account.
This issue affects nanobot (nanobot-ai) from 0.3.5 before 0.4.1.
In nanobot, when sandboxed exec is enabled, the shell executable used to run commands can be selected by the caller and is launched before the sandbox is applied. An attacker who can induce the agent to create an executable in the workspace and select it as the shell can cause that executable to run outside the sandbox (e.g. Linux bubblewrap), with the operating-system permissions of the nanobot process.
This issue affects nanobot (nanobot-ai) from 0.2.1 before 0.4.1.
In the Matter door lock applications of
the Nordic Semiconductor nRF Door Lock and Access Control Add-on v1.0.0
through v1.1.1, AccessManager::ValidateCredential() checks only credential
occupancy and the secret bytes, and never reads the owning user's UserStatus.
It also returns success when no owning user can be resolved. A user suspended
by setting their status to OccupiedDisabled can continue to unlock the door
with their previously assigned credential, and an orphaned but occupied
credential is also accepted. Fixed in v1.2.0.
In
the Matter and Aliro Door Lock Application of the Nordic Semiconductor nRF Door
Lock and Access Control Add-on v1.0.0 through v1.1.1 with
CONFIG_DOOR_LOCK_STEP_UP_PHASE enabled, deleting a Credential Issuer passes the
1-based Matter credential index to ClearValidityIterations(), which expects a
0-based storage index. The intended issuer's ValidityIteration state is left in
place, and the state of the adjacent, still-provisioned issuer is cleared. This
resets that issuer's rollback protection. An attacker holding an older but correctly
signed Access Document from the adjacent issuer, whose validity iteration had
been superseded, can then potentially obtain physical access. Fixed in v1.2.0.