Debian Patches

Status for rails/2:7.2.4+dfsg-3

Patch Description Author Forwarded Bugs Origin Last update
0001-Be-careful-with-that-bundler.patch Be careful with that bundler Antonio Terceiro <terceiro@debian.org> no 2025-01-08
use-system-yarnpkg.patch Use system yarnpkg instead of yarn In Debian, yarn is packaged as "yarnpkg".
.
This patch will replace all the "bin/yarn" usages to "bin/yarnpkg".
Jongmin Kim <jmkim@pukyong.ac.kr> no 2019-06-21
CVE-2026-126188.patch Restrict the formats ffprobe and ffmpeg may open. Active Storage let ffmpeg and ffprobe choose any demuxer they were
built with, though previews and analysis need only common audio and
video containers.
.
Pass `-format_whitelist` immediately before the input at every ffmpeg
and ffprobe call site, from the new
`config.active_storage.ffmpeg_allowed_formats`. Log a warning when
ffprobe exits unsuccessfully, so a refused file's missing metadata has
a cause in the log.
.
Files in other formats, such as MXF, DV and AC3, now fail to preview
and get no metadata. Applications that need them must add their
demuxers to the list, or set it to `nil`; the CHANGELOG names the
refused formats.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/69aff80d115bbb9d1e82fd6c6f2fd3ddeafb958b 2026-11-20
CVE-2026-126073.patch Vary on every header that content negotiation reads. Also vary XHR responses without `Accept` on `Content-Type`, and append
these fields to an existing `Vary` value unless it is `*`.
Jeremy Daer <jeremy@37signals.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/b58c107abe9c1f10c1d6a4941402c60e4dc4742c 2026-11-20
CVE-2026-126072.patch Close a live stream when its buffer is already closed. Jeremy Daer <jeremy@37signals.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/27ad18ed5900dee28ba20e9c6fc718ff751fa215 2026-11-20
relax-dependencies.patch Relax dependencies
This patch will relax the dependencies version, remove useless
dependencies, or replace the gems to which is already in Debian.

Relax dependencies version:
* selenium-webdriver
* redis-nameserver (use packaged version)
* mysql2
* selenium-webdriver
* webpacker

Remove/replace dependencies:
* minitest-bisect
+ Not packaged in Debian
+ Not used (minitest is used)
* minitest-retry
+ Not packaged in Debian
+ Used only when running on Buildkite CI
* webdrivers
+ Not packaged in Debian
+ Could be replaced by chromedriver-helper (ruby-chromedriver-helper)
Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-08-24
remove-ignored-dependencies.patch Remove ignored dependencies Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-06-21
skip-test-internet-access.patch Skip the tests which need Internet access .. due to Debian policy 4.9. Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-07-17
skip-test-unpackaged-dependencies.patch Skip the test due to unpackaged dependencies * websocket-client-simple
+ actioncable/client_test.rb
* sneakers
+ actionjob/Rakefile
+ actionjob/test/cases/exceptions_test.rb
* que
+ actionjob/Rakefile
* queue_classic
+ actionjob/Rakefile
* resque
+ actionjob/Rakefile
* sucker_punch
+ actionjob/Rakefile
* backburner
+ actionjob/Rakefile
* minitest-retry
+ railties/test/isolation/abstract_unit.rb
Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-06-21
skip-test-railties-postgresql.patch Skip the railties test which needs postgresql instance Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-06-24
ignore-test-stuck.patch Ignore test which goes stuck Ignore the test case which makes test stuck. Jongmin Kim <jmkim@pukyong.ac.kr> not-needed 2019-06-21
adapt-to-babel7.patch Adapt to babel7 no
relax-irb.patch relax version dependency on irb gem to be dropped when ruby3.3 is the default Cédric Boutillier not-needed 2025-01-27
remove-sprockets-rails.patch no
fix-line-filtering-minitest-6.patch Fix line_filtering test with Minitest 6 zzak <zzak@hey.com> no https://github.com/rails/rails/pull/56436 2026-02-18
drop-minitest-upper-limit.patch Remove the upper limit on minitest to make it installable Simon Quigley <tsimonq2@debian.org> not-needed vendor 2026-04-12
drop-connection-pool-upperlimit.patch Drop the connection_pool upperlimit Simon Quigley <tsimonq2@debian.org> not-needed vendor 2026-10-02
allow-marcel-2.patch Allow marcel 2 in the Active Storage gemspec Simon Quigley <tsimonq2@debian.org> not-needed vendor 2026-10-15
CVE-2026-126186.patch Keep an extension only when it matches the media type. `Blob#open` named its tempfile after the uploaded filename, extension
included. FFmpeg picks its AviSynth demuxer on a `.avs` extension
alone, so a script named `attack.avs` and declared `video/mp4` was run by
ffprobe and ffmpeg.
.
Keep the extension only when Marcel maps it to the blob's content type.
.
ref: https://hackerone.com/reports/3900470
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/8106938fb0a595f11006dd1497cd409e27d7d636 2026-11-20
CVE-2026-126084.patch Sign GCS direct uploads with V2 when cache_control is set. Since e3fe39285f, a GCS service with `cache_control` signed direct-upload
URLs with V4, which did not bind the Content-Type or Content-MD5
headers. The uploader could store any Content-Type on the object.
.
Always sign with V2, which binds both. `headers_for_direct_upload` still
returns Cache-Control, which the client now sends unsigned, so the
uploader can change its own object's Cache-Control. Direct-upload URLs
with `cache_control` are no longer limited to V4's 7-day expiry.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/ddc08ba61a8001e7402864c62e466cf732fc7614 2026-11-20
CVE-2026-126085.patch Bind the declared byte size on a GCS direct upload. GCS direct-upload URLs did not bind the declared content length, so a
client could store a body of any size under a blob whose `byte_size`
said otherwise.
.
Sign `x-goog-content-length-range` for the declared length and return
it from `headers_for_direct_upload`.
.
Browsers send the new header, so a bucket's CORS configuration must
allow `x-goog-content-length-range`; the guide's example now does.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/7fe85a72a789eae5ee070ce3ee1dad10a8a5f70c 2026-11-20
CVE-2026-126081.patch Return the parameters object from each_key. `ActionController::Parameters#each_key` was delegated to the internal
hash, so with a block it returned that `HashWithIndifferentAccess`
instead of the `Parameters`, and mass assignment of the return value
skipped the `permitted?` check.
.
Define `each_key` to return `self`, as `each_pair` and `each_value` do.
Document that `as_json` returns every parameter, permitted or not, in a
plain hash that mass assignment does not check.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/2fc8838300133931e19eed3ea6cee76a7e83f19c 2026-11-20
CVE-2026-126080.patch Match safe_constantize load errors as a literal string. `safe_constantize` built a `Regexp` from the constant name, unescaped,
to check a `LoadError` message, so a caller passing untrusted input,
such as an Action Cable subscribe command, chose the pattern the
server compiled.
.
Match the name's first segment as a literal string instead, and add a
CHANGELOG entry.
Jeremy Daer <jeremy@37signals.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/210ac22d446c8e2625b091e79256806b836eee20 2026-11-20
CVE-2026-126079.patch Require rails-html-sanitizer 1.7.2. With the HTML5 sanitizer, Nokogiri before 1.19.5 left `<` and `>`
unescaped in attribute values, so `highlight` and `simple_format` could
turn markup inside an attribute value into live elements.
.
Require rails-html-sanitizer 1.7.2, which requires Loofah 2.25.3, which
requires Nokogiri 1.19.5. Add tests for both helpers with the HTML5
sanitizer, and update the Action Text tests that expected the old
serialization.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/009ea68342d05d02087c04dae7faaa4e4c1def11 2026-11-20
CVE-2026-126078.patch Escape each match substituted into highlight's :highlighter. `highlight` copied each match into the `:highlighter` unescaped and
returned the result `html_safe`, so with `\1` inside a quoted attribute,
as in the documented `'<a href="search?q=\1">\1</a>'`, a match
containing a quote added attributes such as `onfocus`. Escaping the text
first does not help, because `sanitize` turns `&quot;` back into `"`.
.
Expand `\0` to `\9` directly and pass each group through
`ERB::Util.html_escape_once`. Never let a match start on the `>` that
closes the preceding tag, which the escaped highlighter would otherwise
replace with `&gt;`, making the following text attributes of that tag.
Use a `title` attribute in the documented example, since a URL needs
escaping of its own.
Jeremy Daer <jeremy@37signals.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/c796f6198837479ada13290eb8fa767b5219a58a 2026-11-20
CVE-2026-126077.patch Keep a proxy from resending an upgraded session cookie. Reading a signed or encrypted cookie that needed an upgrade, from a
rotated secret or a `:hybrid` serializer migration, wrote it back to the
response even when `session_options[:skip]` was set. A response that
skips the session, such as the Active Storage proxy's public, cacheable
response, could then carry the session cookie in `Set-Cookie`.
.
Track upgrade writes on the cookie jar and leave them out of a response
that skips the session. The check happens when the jar is written, so
it holds whether the cookie was read before or after `skip` was set.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/0e6cf8faa89caf4c64311d35ca2bc88c1458d4e3 2026-11-20
CVE-2026-126076.patch Do not write back a cache session deleted mid-request. A request that loaded a session from `CacheStore` wrote it back when it
finished, even if another request had deleted it in the meantime, as
`reset_session` does on logout. The deleted session came back under its
old ID.
.
Skip the write when the cache entry of the session this request loaded
is gone.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/5bb5ec66ed8faaf590ffe1222d7a5eb6e9d050f5 2026-11-20
CVE-2026-126074.patch Take a filename base and extension from the sanitized name. Jeremy Daer <jeremy@37signals.com> not-needed debian upstream backport, https://github.com/rails/rails/commit/9b0d4bef2211a2c08cab2e0ea17fa1d0efd5d47a 2026-11-20
CVE-2026-126070.patch Encode quotation marks in strip_tags output. `strip_tags` returns an `html_safe` string, but was not encoding
quotation marks. The method now encodes all five HTML-significant
characters (`&`, `<`, `>`, `"`, `'`), consistent with the existing
`html_safe` convention.
.
Callers that use the output as plain text, such as text templates or
JSON, now see `&quot;` and `&#39;` where they saw `"` and `'`.
.
Also document that `sanitize` output is safe only as element content.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/c3772a10f405f9644125cbb484adfd2aa7a25a59 2026-11-20
CVE-2026-126069.patch Skip a tag attribute name that is blank as a string. The guard 63f5ad83 added to `tag_options` tested `blank?` on the raw key,
but `ERB::Util.xml_name_escape` escapes the key converted to a String.
`Symbol#blank?` is `Symbol#empty?`, so `:" "` passed the guard, escaped to
the empty string, and produced an attribute with no name, which the HTML5
tokenizer reads as a real one. `image_tag`, `video_tag`, `audio_tag` and
`picture_tag` symbolize their keys, so a String key reached it too.
.
Keep `key.blank?`, then convert the key once and pass the converted name
to `append_tag_option`, `append_prefix_tag_option` and
`append_boolean_tag_option`, so the name tested is the name emitted.
.
ref: https://hackerone.com/reports/3904103
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/f47452533e1f28d847c63181b9d5f8b7243ef9a4 2026-11-20
CVE-2026-126068.patch Reject a Mailgun timestamp more than two minutes ahead. The Mailgun ingress rejected a timestamp more than 2 minutes old but
accepted any timestamp ahead of local time.
.
Allow 2 minutes of drift in both directions.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/4aa2ec66822bc4eb8b341a0608f1c90a3e371a34 2026-11-20
CVE-2026-126067.patch Apply the SameSite policy on every session store. Only `CookieStore` defaulted `:same_site` to
`config.action_dispatch.cookies_same_site_protection`. `CacheStore`,
`MemCacheStore` and third-party stores built on `AbstractStore` or the
session mixins sent session cookies with no `SameSite` attribute.
.
Move the default into `Session::Compatibility`, which every session
store includes. `CookieStore` still forces `:cookie_only`; the other
stores keep their own setting.
.
ref: https://hackerone.com/reports/2637805
Jean Boussier <jean.boussier@gmail.com> not-needed debian upstream upstream, https://github.com/rails/rails/commit/f2392996be723708271a8d30bfcf0edc221e53ef 2026-11-20
CVE-2026-122006.patch Apply an ImageMagick policy to every MiniMagick process. CVE-2026-122006. Active Storage now prepends its own policy.xml to
MAGICK_CONFIGURE_PATH through MiniMagick.cli_env; it denies MSL, the
Ghostscript coders and gs delegate, and @filename indirection.
Mike Dalessio <mike.dalessio@gmail.com> not-needed debian upstream backport, https://github.com/rails/rails/commit/ea8c40121ce53a3d0255cd4eb6d13bfe3bc01776 2026-11-20

All known versions for source package 'rails'

Links