Pomerium is an open source identity-aware access proxy. Envoy, which Pomerium is based on, incorrectly handles resetting of HTTP/2 streams with excessive complexity. This can lead to high CPU utilization when a large number of streams are reset. This can result in a DoS condition. Pomerium versions 0.14.8 and 0.15.1 contain an upgraded envoy binary with this vulnerability patched.

Pomerium is an open source identity-aware access proxy. Envoy, which Pomerium is based on, can abnormally terminate if an H/2 GOAWAY and SETTINGS frame are received in the same IO event. This can lead to a DoS in the presence of untrusted *upstream* servers. 0.15.1 contains an upgraded envoy binary with this vulnerability patched. If only trusted upstreams are configured, there is not substantial risk of this condition being triggered.

Envoy is a cloud-native edge/middle/service proxy. Envoy does not decode escaped slash sequences `%2F` and `%5C` in HTTP URL paths in versions 1.18.2 and before. A remote attacker may craft a path with escaped slashes, e.g. `/something%2F..%2Fadmin`, to bypass access control, e.g. a block on `/admin`. A backend server could then decode slash sequences and normalize path and provide an attacker access beyond the scope provided for by the access control policy. ### Impact Escalation of Privileges when using RBAC or JWT filters with enforcement based on URL path. Users with back end servers that interpret `%2F` and `/` and `%5C` and `\` interchangeably are impacted. ### Attack Vector URL paths containing escaped slash characters delivered by untrusted client. Patches in versions 1.18.3, 1.17.3, 1.16.4, 1.15.5 contain new path normalization option to decode escaped slash characters. As a workaround, if back end servers treat `%2F` and `/` and `%5C` and `\` interchangeably and a URL path based access control is configured, one may reconfigure the back end server to not treat `%2F` and `/` and `%5C` and `\` interchangeably.

Envoy before 1.16.1 mishandles dropped and truncated datagrams, as demonstrated by a segmentation fault for a UDP packet size larger than 1500.

Envoy before 1.16.1 logs an incorrect downstream address because it considers only the directly connected peer, not the information in the proxy protocol header. This affects situations with tcp-proxy as the network filter (not HTTP filters).

Envoy through 1.15.0 only considers the first value when multiple header values are present for some HTTP headers. Envoy’s setCopy() header map API does not replace all existing occurences of a non-inline header.

In Envoy before versions 1.12.6, 1.13.4, 1.14.4, and 1.15.0 when validating TLS certificates, Envoy would incorrectly allow a wildcard DNS Subject Alternative Name apply to multiple subdomains. For example, with a SAN of *.example.com, Envoy would incorrectly allow nested.subdomain.example.com, when it should only allow subdomain.example.com. This defect applies to both validating a client TLS certificate in mTLS, and validating a server TLS certificate for upstream connections. This vulnerability is only applicable to situations where an untrusted entity can obtain a signed wildcard TLS certificate for a domain of which you only intend to trust a subdomain of. For example, if you intend to trust api.mysubdomain.example.com, and an untrusted actor can obtain a signed TLS certificate for *.example.com or *.com. Configurations are vulnerable if they use verify_subject_alt_name in any Envoy version, or if they use match_subject_alt_names in version 1.14 or later. This issue has been fixed in Envoy versions 1.12.6, 1.13.4, 1.14.4, 1.15.0.

Envoy version 1.14.2, 1.13.2, 1.12.4 or earlier may exhaust file descriptors and/or memory when accepting too many connections.

Envoy version 1.14.2, 1.13.2, 1.12.4 or earlier may consume excessive amounts of memory when processing HTTP/1.1 headers with long field names or requests with long URLs.

Envoy version 1.14.2, 1.13.2, 1.12.4 or earlier is susceptible to increased memory usage in the case where an HTTP/2 client requests a large payload but does not send enough window updates to consume the entire stream and does not reset the stream.

Envoy version 1.14.2, 1.13.2, 1.12.4 or earlier may consume excessive amounts of memory when proxying HTTP/2 requests or responses with many small (i.e. 1 byte) data frames.

Istio through 1.5.1 and Envoy through 1.14.1 have a data-leak issue. If there is a TCP connection (negotiated with SNI over HTTPS) to *.example.com, a request for a domain concurrently configured explicitly (e.g., abc.example.com) is sent to the server(s) listening behind *.example.com. The outcome should instead be 421 Misdirected Request. Imagine a shared caching forward proxy re-using an HTTP/2 connection for a large subnet with many users. If a victim is interacting with abc.example.com, and a server (for abc.example.com) recycles the TCP connection to the forward proxy, the victim's browser may suddenly start sending sensitive data to a *.example.com server. This occurs because the forward proxy between the victim and the origin server reuses connections (which obeys the specification), but neither Istio nor Envoy corrects this by sending a 421 error. Similarly, this behavior voids the security model browsers have put in place between domains.

CNCF Envoy through 1.13.0 TLS inspector bypass. TLS inspector could have been bypassed (not recognized as a TLS client) by a client using only TLS 1.3. Because TLS extensions (SNI, ALPN) were not inspected, those connections might have been matched to a wrong filter chain, possibly bypassing some security restrictions in the process.

An issue was discovered in Envoy 1.12.0. Upon receipt of a malformed HTTP request without a Host header, it sends an internally generated "Invalid request" response. This internally generated response is dispatched through the configured encoder filter chain before being sent to the client. An encoder filter that invokes route manager APIs that access a request's Host header causes a NULL pointer dereference, resulting in abnormal termination of the Envoy process.

An issue was discovered in Envoy 1.12.0. An untrusted remote client may send an HTTP header (such as Host) with whitespace after the header content. Envoy will treat "header-value " as a different string from "header-value" so for example with the Host header "example.com " one could bypass "example.com" matchers.

An issue was discovered in Envoy 1.12.0. An untrusted remote client may send HTTP/2 requests that write to the heap outside of the request buffers when the upstream is HTTP/1. This may be used to corrupt nearby heap contents (leading to a query-of-death scenario) or may be used to bypass Envoy's access control mechanisms such as path based routing. An attacker can also modify requests from other users that happen to be proximal temporally and spatially.

Upon receiving each incoming request header data, Envoy will iterate over existing request headers to verify that the total size of the headers stays below a maximum limit. The implementation in versions 1.10.0 through 1.11.1 for HTTP/1.x traffic and all versions of Envoy for HTTP/2 traffic had O(n^2) performance characteristics. A remote attacker may craft a request that stays below the maximum request header size but consists of many thousands of small headers to consume CPU and result in a denial-of-service attack.

In Envoy through 1.11.1, users may configure a route to match incoming path headers via the libstdc++ regular expression implementation. A remote attacker may send a request with a very long URI to result in a denial of service (memory consumption). This is a related issue to CVE-2019-14993.

Envoy 1.9.0 and before does not normalize HTTP URL paths. A remote attacker may craft a relative path, e.g., something/../admin, to bypass access control, e.g., a block on /admin. A backend server could then interpret the non-normalized path and provide an attacker access beyond the scope provided for by the access control policy.

When parsing HTTP/1.x header values, Envoy 1.9.0 and before does not reject embedded zero characters (NUL, ASCII 0x0). This allows remote attackers crafting header values containing embedded NUL characters to potentially bypass header matching rules, gaining access to unauthorized resources.