Skip to content

Commit d65ab7b

Browse files
authored
Merge pull request #1312 from jasnow/puma-advs
Seven new puma advisories
2 parents f34da38 + fbd70d7 commit d65ab7b

7 files changed

Lines changed: 413 additions & 0 deletions

‎gems/puma/GHSA-86xc-wrr9-xmmq.yml‎

Lines changed: 53 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,53 @@
1+
---
2+
gem: puma
3+
ghsa: 86xc-wrr9-xmmq
4+
url: https://github.com/puma/puma/security/advisories/GHSA-86xc-wrr9-xmmq
5+
title: pumactl does not verify the TLS identity of SSL control servers
6+
date: 2026-11-13
7+
description: |
8+
## Impact
9+
10+
An attacker on the network path between pumactl and a Puma control
11+
server reached over ssl:// can steal the control token and use it
12+
to stop, restart, or inspect Puma.
13+
14+
pumactl support for ssl:// control URLs was added in v4.3.0 (#2052).
15+
16+
When pumactl connects to an ssl:// control URL, it accepts any TLS
17+
certificate. It should check that a trusted authority issued the
18+
certificate and that the certificate matches the hostname in the URL.
19+
An attacker who can intercept or redirect the connection can present
20+
their own certificate, and pumactl then sends them the control token.
21+
22+
With the token, the attacker can send authenticated commands to the
23+
real control server, including stop, restart, and stats.
24+
25+
Only deployments that run pumactl against an ssl:// control URL
26+
are affected, for example:
27+
28+
* pumactl --control-url 'ssl://control.example.com:9293'
29+
--control-token "$PUMA_CONTROL_TOKEN" stats
30+
31+
Puma's default configuration is not affected. The control server
32+
is off by default, and unix:// control URLs are not affected.
33+
34+
## Acknowledgements
35+
36+
Thanks to @ahmedehab-nopr for reporting this issue.
37+
cvss_v4: 8.3
38+
unaffected_versions:
39+
- "< 4.3.0"
40+
patched_versions:
41+
- "~> 7.2.2"
42+
- ">= 8.0.3"
43+
related:
44+
url:
45+
- https://rubygems.org/gems/puma/versions/8.0.3
46+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
47+
- https://rubygems.org/gems/puma/versions/7.2.2
48+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
49+
- https://github.com/puma/puma/security/advisories/GHSA-86xc-wrr9-xmmq
50+
notes: |
51+
- No CVE from GHSA URL.
52+
- cvss_v4 from GHSA URL.
53+
- No gap between 7.2.2 and 8.0.0 releases.

‎gems/puma/GHSA-jm4r-w4xx-qxw7.yml‎

Lines changed: 63 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,63 @@
1+
---
2+
gem: puma
3+
ghsa: jm4r-w4xx-qxw7
4+
url: https://github.com/puma/puma/security/advisories/GHSA-jm4r-w4xx-qxw7
5+
title: Graceful shutdown waits indefinitely on trickled request bodies
6+
date: 2026-11-13
7+
description: |
8+
## Impact
9+
10+
Puma servers running in single mode with the default force_shutdown_after
11+
:forever can be prevented from shutting down or restarting by an
12+
unauthenticated client.
13+
14+
When Puma begins a graceful shutdown, it hands any request that is
15+
still being received to the thread pool so that it can finish. The
16+
thread pool then waits for that request with no deadline. Puma's
17+
read timeout restarts every time a byte arrives, so an attacker who
18+
sends a POST with a large Content-Length and then trickles one body
19+
byte every few seconds can keep that request "in progress" for as
20+
long as they like. The same works with chunked bodies and with
21+
trickled headers. Puma stops accepting new connections, but the
22+
process does not exit until the attacker closes the socket.
23+
24+
This delays or blocks deploys, restarts, and operational recovery.
25+
Process supervisors that eventually send SIGKILL will end the
26+
process, but any other in-flight requests are lost when they do.
27+
Many trickling connections can be used together; each holds a
28+
file descriptor but uses very little memory.
29+
30+
Deployments using Puma's default single-mode configuration are affected.
31+
32+
Unaffected:
33+
34+
* Cluster mode (workers N) bounds the delay: the master kills a
35+
stuck worker after worker_shutdown_timeout (30 seconds by default).
36+
37+
* A finite force_shutdown_after bounds the delay.
38+
39+
* Deployments behind a proxy that buffers request bodies before
40+
forwarding them are not affected. For example, nginx's default
41+
proxy_request_buffering on keeps a trickled body from reaching Puma.
42+
43+
## Acknowledgements
44+
45+
Thanks to Zheng Yu of depthfirst (@cla7aye15I4nd) for reporting
46+
this issue and providing a proposed fix.
47+
cvss_v4: 6.9
48+
unaffected_versions:
49+
- "< 5.0.0"
50+
patched_versions:
51+
- "~> 7.2.2"
52+
- ">= 8.0.3"
53+
related:
54+
url:
55+
- https://rubygems.org/gems/puma/versions/8.0.3
56+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
57+
- https://rubygems.org/gems/puma/versions/7.2.2
58+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
59+
- https://github.com/puma/puma/security/advisories/GHSA-jm4r-w4xx-qxw7
60+
notes: |
61+
- No CVE from GHSA URL.
62+
- cvss_v4 from GHSA URL.
63+
- No gap between 7.2.2 and 8.0.0 releases.

‎gems/puma/GHSA-m5xw-3hrw-qcmf.yml‎

Lines changed: 50 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,50 @@
1+
---
2+
gem: puma
3+
ghsa: m5xw-3hrw-qcmf
4+
url: https://github.com/puma/puma/security/advisories/GHSA-m5xw-3hrw-qcmf
5+
title: Reactor timeout queue loses order, letting a slow client
6+
keep other connections open past their timeout
7+
date: 2026-11-13
8+
description: |
9+
## Impact
10+
11+
An unauthenticated client can stop Puma from closing other connections
12+
when their first_data_timeout or persistent_timeout expires.
13+
14+
The reactor keeps waiting connections in a queue sorted by deadline.
15+
Puma sorted the queue only when new connections arrived, but it
16+
extended a connection's deadline every time that connection sent
17+
data. A client that connects and then sends about one byte per
18+
timeout interval, without completing its request, stays at the
19+
front of the queue with a deadline that is always in the future.
20+
Every expired connection behind it stays open for as long as the
21+
client keeps sending, or until a new connection arrives and causes a sort.
22+
23+
Connections held this way cost Puma one file descriptor and a small
24+
amount of memory each. Puma does not crash, and the effect ends
25+
when the slow client stops or new traffic arrives.
26+
This bug was introduced in v5.0.3 by the reactor refactor (#2279).
27+
28+
The default configuration is affected.
29+
30+
## Acknowledgements
31+
32+
Thanks to @0xMigoo for the report and the controlled proof of
33+
concept that isolated the missing sort.
34+
cvss_v4: 6.9
35+
unaffected_versions:
36+
- "< 5.0.3"
37+
patched_versions:
38+
- "~> 7.2.2"
39+
- ">= 8.0.3"
40+
related:
41+
url:
42+
- https://rubygems.org/gems/puma/versions/8.0.3
43+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
44+
- https://rubygems.org/gems/puma/versions/7.2.2
45+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
46+
- https://github.com/puma/puma/security/advisories/GHSA-m5xw-3hrw-qcmf
47+
notes: |
48+
- No CVE from GHSA URL.
49+
- cvss_v4 from GHSA URL.
50+
- No gap between 7.2.2 and 8.0.0 releases.

‎gems/puma/GHSA-r885-m39c-939g.yml‎

Lines changed: 59 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,59 @@
1+
---
2+
gem: puma
3+
ghsa: r885-m39c-939g
4+
url: https://github.com/puma/puma/security/advisories/GHSA-r885-m39c-939g
5+
title: ssl_cipher_filter values containing `+` are decoded
6+
to a wider cipher list
7+
date: 2026-11-13
8+
description: |
9+
## Impact
10+
11+
Puma servers that set an ssl_cipher_filter containing + accept TLS
12+
cipher suites that the filter is meant to exclude. Any client that
13+
can reach the SSL listener can negotiate those suites.
14+
15+
Support for ssl_cipher_filter in ssl:// bind URLs was added in
16+
v3.12.0 (#1478), and in the ssl_bind DSL in v4.0.0 (#1676).
17+
18+
In an OpenSSL cipher string, + means "both": AES+SHA selects suites
19+
that are AES and SHA. Puma decodes the query of a bind URL as HTML
20+
form data, where + means a space, and a space in a cipher string
21+
means "either". So the filter !aNULL:AES+SHA, which is the example
22+
in Puma's README, reaches OpenSSL as !aNULL:AES SHA: every AES
23+
suite plus every SHA suite. On OpenSSL 3.x that is 94 suites in
24+
place of 27, including suites without forward secrecy.
25+
26+
Puma logs nothing, and the startup output shows the filter as written,
27+
so operators cannot see that their filter is not applied. The filter
28+
only gets wider: explicit exclusions such as !aNULL or !3DES still apply.
29+
30+
Deployments that set ssl_cipher_filter with a + are affected, in either form:
31+
32+
* ssl_bind "0.0.0.0", "9292", ssl_cipher_filter: "!aNULL:AES+SHA",
33+
cert: "...", key: "..."
34+
35+
* puma -b 'ssl://0.0.0.0:9292?key=...&cert=...&ssl_cipher_filter=!aNULL:AES+SHA'
36+
37+
The default configuration is not affected. Filters without +,
38+
ssl_ciphersuites (TLS 1.3), and JRuby's ssl_cipher_list are not affected.
39+
40+
## Acknowledgements
41+
42+
Thanks to @b84330808 for reporting this issue.
43+
cvss_v4: 6.3
44+
unaffected_versions:
45+
- "< 3.12.0"
46+
patched_versions:
47+
- "~> 7.2.2"
48+
- ">= 8.0.3"
49+
related:
50+
url:
51+
- https://rubygems.org/gems/puma/versions/8.0.3
52+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
53+
- https://rubygems.org/gems/puma/versions/7.2.2
54+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
55+
- https://github.com/puma/puma/security/advisories/GHSA-r885-m39c-939g
56+
notes: |
57+
- No CVE from GHSA URL.
58+
- cvss_v4 from GHSA URL.
59+
- No gap between 7.2.2 and 8.0.0 releases.

‎gems/puma/GHSA-rf6g-xhmp-gm7v.yml‎

Lines changed: 58 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,58 @@
1+
---
2+
gem: puma
3+
ghsa: rf6g-xhmp-gm7v
4+
url: https://github.com/puma/puma/security/advisories/GHSA-rf6g-xhmp-gm7v
5+
title: Accept loop retries without pausing when out of file
6+
descriptors, flooding the log
7+
date: 2026-11-13
8+
description: |
9+
## Impact
10+
11+
An unauthenticated client can make Puma:
12+
13+
1. write log lines as fast as it can...
14+
2. ...which eventually may fill the disk...
15+
3. ...and when the disk fills, Puma will stop accepting connections.
16+
17+
The client can do this by opening enough idle connections to use up
18+
the process's file descriptors.
19+
20+
When accept fails because the process is out of file descriptors
21+
(EMFILE), the connection stays in the listen backlog, so the listening
22+
socket stays readable. Puma logs the error and calls accept again
23+
at once, with no pause. Each attempt fails the same way and writes
24+
another "Listen loop" error line. An attacker who opens about as
25+
many idle connections as the file descriptor limit, and sends
26+
nothing on them, keeps Puma in this loop. The attacker reconnects
27+
when Puma closes idle connections after first_data_timeout, so
28+
the loop does not end.
29+
30+
If the log destination then raises an error, such as ENOSPC when
31+
the disk is full, that error escapes the listen loop and ends
32+
the thread that accepts connections.
33+
34+
Deployments that accept connections directly from clients are
35+
affected, including the default configuration.
36+
37+
Deployments behind a reverse proxy or load balancer that limits
38+
connections to Puma are not affected, because the client cannot
39+
open enough connections to Puma itself.
40+
41+
## Acknowledgements
42+
43+
Found by Puma's maintainers during an internal security review.
44+
cvss_v4: 8.7
45+
patched_versions:
46+
- "~> 7.2.2"
47+
- ">= 8.0.3"
48+
related:
49+
url:
50+
- https://rubygems.org/gems/puma/versions/8.0.3
51+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
52+
- https://rubygems.org/gems/puma/versions/7.2.2
53+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
54+
- https://github.com/puma/puma/security/advisories/GHSA-rf6g-xhmp-gm7v
55+
notes: |
56+
- No CVE from GHSA URL.
57+
- cvss_v4 from GHSA URL.
58+
- No gap between 7.2.2 and 8.0.0 releases.

‎gems/puma/GHSA-x646-wp7c-h9mg.yml‎

Lines changed: 70 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,70 @@
1+
---
2+
gem: puma
3+
ghsa: x646-wp7c-h9mg
4+
url: https://github.com/puma/puma/security/advisories/GHSA-x646-wp7c-h9mg
5+
title: Client that stops reading a TLS or file response holds
6+
a worker thread indefinitely
7+
date: 2026-11-13
8+
description: |
9+
## Impact
10+
11+
An unauthenticated client can hold a Puma worker thread indefinitely
12+
when Puma sends it a TLS response, or a file response larger than
13+
64 KB. With enough connections, it can take the server offline.
14+
15+
The client sends a request and then stops reading the response.
16+
Puma's write timeout (10 seconds) should then free the worker
17+
thread. In two cases it never fires, and the thread stays blocked
18+
until the client disconnects:
19+
20+
* TLS: MiniSSL::Socket#write_nonblock performs a blocking write.
21+
This has been the case since v3.9.0 (#1274).
22+
23+
* File bodies over 64 KB, on any bind: Puma sends File bodies, and
24+
bodies that respond to to_path, with IO.copy_stream, which has no
25+
write timeout. This has been the case since v6.0.0 (#2896).
26+
Rack::Files, Rails static file serving, and Rails send_file return
27+
these bodies.
28+
29+
An attacker who opens max_threads such connections per worker leaves
30+
no threads to serve requests. The attack needs very little bandwidth
31+
and no CPU on the server.
32+
33+
Two kinds of deployment are affected:
34+
35+
* Deployments where Puma terminates TLS itself, with an ssl://
36+
bind or ssl_bind:
37+
38+
* ssl_bind "0.0.0.0", "9292", cert: "/path/to/cert.pem", key:
39+
"/path/to/key.pem"
40+
41+
* Applications that serve a file larger than 64 KB through Puma, on
42+
any bind. Puma's default configuration is affected in this case.
43+
44+
Deployments behind a proxy that buffers responses are not affected
45+
for files that fit in the proxy's buffer. nginx buffers by default
46+
(proxy_buffering on), up to 1 GB per response in a temporary file.
47+
48+
## Acknowledgements
49+
50+
Thanks to @wtn and @SounLabs, who reported this issue
51+
independently, and to @wtn for the fix.
52+
cvss_v4: 8.7
53+
unaffected_versions:
54+
- "< 3.9.0"
55+
patched_versions:
56+
- "~> 7.2.2"
57+
- ">= 8.0.3"
58+
related:
59+
url:
60+
- https://rubygems.org/gems/puma/versions/8.0.3
61+
- https://github.com/puma/puma/blob/main/History.md#803--2026-11-13
62+
- https://rubygems.org/gems/puma/versions/7.2.2
63+
- https://github.com/puma/puma/blob/main/History.md#722--2026-11-13
64+
- https://github.com/puma/puma/pull/1274
65+
- https://github.com/puma/puma/pull/2896
66+
- https://github.com/puma/puma/security/advisories/GHSA-x646-wp7c-h9mg
67+
notes: |
68+
- No CVE from GHSA URL.
69+
- cvss_v4 from GHSA URL.
70+
- No gap between 7.2.2 and 8.0.0 releases.

0 commit comments

Comments
 (0)