CVE-2026-25800

Summary

The Assembler component that assembles unordered stream fragments into consecutive chunks of the stream incurs some overhead for non-contiguous fragments. Readers that read from a RecvStream in order (through an AsyncRead impl for example) will be sensitive to peers that send fragments while leaving out early parts of the stream, and in particular, fragments with many gaps (because these cannot be defragmented). In such a scenario, the receiving connection suffers from high buffer overhead, enabling memory exhaustion.

The Assembler component that assembles unordered stream fragments into consecutive chunks of the stream incurs some overhead for non-contiguous fragments. Readers that read from a RecvStream in order (through an AsyncRead impl for example) will be sensitive to peers that send fragments while leaving out early parts of the stream, and in particular, fragments with many gaps (because these cannot be defragmented). In such a scenario, the receiving connection suffers from high buffer overhead, enabling memory exhaustion.

Quinn is a pure-Rust, async-compatible implementation of the IETF QUIC transport protocol. Starting in version 0.1.0 and prior to version 0.11.15, the Assembler component that assembles unordered stream fragments into consecutive chunks of the stream incurs some overhead for non-contiguous fragments. Readers that read from a RecvStream in order (through an AsyncRead impl for example) will be sensitive to peers that send fragments while leaving out early parts of the stream, and in particular, fragments with many gaps (because these cannot be defragmented). In such a scenario, the receiving connection suffers from high buffer overhead, enabling memory exhaustion. Version 0.11.15 fixes the issue.

These are all security issues fixed in the python313-uv-0.12.7-1.1 package on the GA media of openSUSE Tumbleweed.

These are all security issues fixed in the openai-codex-0.152.0-1.1 package on the GA media of openSUSE Tumbleweed.

This update for openai-codex fixes the following issues:

Changes in openai-codex:

Update to version 0.154.0:

• Experimental worktree support for isolated checkouts of new or forked sessions, with browse and resume
• Answer questions inline while Codex keeps working, keeping the main draft
• Vim R replace mode with undo and dot-repeat; /copy keeps formatting and covers status and session fields
• Plugin tools refresh after external upgrades, MCP OAuth refresh is coordinated, helpers no longer run before trust is established; resume and fork keep saved permissions
• Drop the deprecated codex mcp-server entry point, now codex mcp

• Regenerate vendor.tar.zst: rmcp 3.1.3 -> 3.2.0 plus routine bumps; webrtc/gstreamer/cpal stay out of the codex-cli link graph, so the License tag is unchanged
• Legal-Review-Notice: linked-crate count 876/878 -> 880/882 on aarch64/x86_64; the vendored dependency licence set is unchanged
• cargo-audit reports the same five advisories as 0.151.0, none new and none fixable by re-vendoring

Update to version 0.153.4:

• GPT-6-Astra is listed in the bundled model picker and is the bundled default when no model is configured
• Astra's async-question guidance now applies only when request_user_input_async is available in the session

Update to version 0.153.3:

• GPT-6-Astra is available in the Amazon Bedrock model picker for Mantle and Runtime global/US routes
• Astra's async-question guidance now names request_user_input_async and notes that it accepts text only

• Keep vendor.tar.zst: 0.153.4 has the same Cargo.lock as 0.153.2, so the linked crate set, the License tag and the Legal-Review-Notice counts are unchanged

• Update to version 0.153.2:

  • Correct the GPT-6-Astra Fast tier description to "2x speed" instead of "1.5x"; display text only, request handling is unchanged

• Update to version 0.153.1:

  • GPT-6-Astra can be configured through the API without changing the default model or listing it in the model picker

• Build against the system bzip2, libzstd, oniguruma and sqlite instead of the copies bzip2-sys, zstd-sys, onig_sys and libsqlite3-sys otherwise compile in; drop their bundled() Provides and add pkgconfig(bzip2), pkgconfig(libzstd), pkgconfig(oniguruma) >= 6.9.3 and pkgconfig(sqlite3) >= 3.34.1

  • oniguruma and sqlite have no usable manifest switch, so %build and %check export RUSTONIG_SYSTEM_LIBONIG and LIBSQLITE3_SYS_USE_PKG_CONFIG instead, and %install now fails the build unless all four are DT_NEEDED of codex
  • aws-lc stays bundled: aws-lc-sys has no system-library mode

• Keep vendor.tar.zst: 0.153.2 has the same Cargo.lock as 0.153.0, so the linked crate set, the License tag and the Legal-Review-Notice counts are unchanged

• Update to version 0.153.0:

  • Vim mode gains undo (u) and redo (Ctrl+R), preserving whole drafts including pasted content and attachments
  • The plugin CLI lists, installs and removes plugins from remote marketplaces
  • tui.auto_recap = false turns off automatic recaps while keeping /recap available
  • TUI history shows complete patches, input sent to background terminals and individual completed commands
  • TUI sessions reconnect after an app-server connection drops, keeping drafts and holding queued submissions for review
  • Guardian review history survives compaction, restarts and forks; Full Access skips Guardian review for confirmation-only actions
  • Remembered MCP tool approvals are scoped to the selected app account
  • tui.disable_paste_burst replaces the top-level setting, which stays supported as a fallback
  • New disabled-by-default features.context_management.experimental_mode

• Regenerate vendor.tar.zst: the 14 crates 0.153.0 adds to the lockfile are all Windows-only, so the linked third-party set, the License tag and the bundled() versions are unchanged; the …