Files
tty7/.github/scripts/check-host-boundary.sh
l0ng-aiandl0ng-ai bf9c57dec7 fix(ssh): let a rejected stored credential ask again (#519)
* fix(ssh): let a rejected stored passphrase ask again (#486)

Saving the wrong passphrase for an encrypted key locked that key out
permanently. `passphrase_submit` wrote `SetKeyPassphrase` on the
"remember" checkbox alone — before the daemon had tried the secret, since
`apply_keychain_write` runs ahead of `respond_active` — and
`try_identity_file` treated a stored passphrase as final: a decrypt
failure with one went straight to "could not decrypt identity file", with
no prompt and nothing in the UI that could let go of it.

The daemon now says so. `AuthPromptKind::KeyPassphrase` grows a
`rejected` flag, and a stored passphrase that does not open the file
falls through to the interactive prompt carrying it, so the typed answer
still gets its attempt. A passphrase the user typed this time keeps the
hard failure — that is a wrong answer, not stale state. The sheet renders
the warning line the password sheet already had, and a rejected prompt
answered without "remember" now emits `DeleteKeyPassphrase`, mirroring
the password idiom exactly.

The flag is a `#[serde(default)]` field on a struct variant of an
externally tagged enum, which is compatible in both directions: an older
peer never sets it and serde ignores fields it does not know. So
`PROTOCOL_VERSION` deliberately does not move — the remote-server
handshake gates on it, and a bump would turn away older servers over a
field they can safely ignore. `protocol.rs`'s compat test pins both
directions.

Also: deleting an SSH profile now drops the key-passphrase entries no
other profile still references, which is what `delete_profile_confirmed`'s
own comment already claimed to do but only ever did for the password.

* fix(ssh): stop replaying a stale password at keyboard-interactive (#487)

`try_keyboard_interactive` answered a password-shaped round from the
keychain, marked the stored password spent whether or not it had been
used, and returned on the first `Failure` — so the `MAX_ROUNDS` loop
never got a second pass with the stored password withheld. The same dead
secret went out on every reconnect and the user was never once asked to
type a different one; `ki_submit` always emitted `KeychainWrite::None`,
so nothing could clear it either.

`collect_ki_answers` now reports where its answers came from, and only a
round that actually sent the stored password spends it — which also fixes
an OTP-then-password flow that was refusing the stored password for no
reason, its first round having burned the allowance on a code. On a
rejection whose last round came from the keychain, and where the server
still offers the method, the request is started over with the stored
password withheld, so the next round reaches the prompt. That retry is
bounded twice over: the restart spends the stored password, so no second
restart can qualify, and the round counter it shares with the
info-request loop caps the method either way. The failure text now says
which of the two was turned down.

Scope, honestly: the only live scenario is auth mode Auto against a
server offering keyboard-interactive but not password, with a stored
password for that endpoint — a profile pinned to KeyboardInteractive gets
`password: None` and always prompts, and Password never tries KI. Whether
the symptom shows also depends on the server: OpenSSH ends a rejected
kbdint request with USERAUTH_FAILURE (symptom holds), while a device that
re-issues an InfoRequest in the same request already reached the prompt.

`AuthPromptKind::KeyboardInteractive` grows a `#[serde(default)]`
`stored_rejected`, same both-directions compatibility as `KeyPassphrase`'s
`rejected` and the same reason `PROTOCOL_VERSION` stays put. The sheet
shows the warning line and, on submit, forgets the rejected password.

That needed an endpoint the KI prompt does not carry, which also fixed a
bug next door: `raise_routed_auth` called `from_prompt(.., None, false)`,
so every routed password write was keyed to port 22 regardless of the real
port and the rejected self-heal could never fire there. `PendingAuth` now
carries the endpoint and the auto-supplied flag, read straight off the
route's `NativeSshSpec`.

---------

Co-authored-by: l0ng-ai <24760907+l0ng-ai@users.noreply.github.com>
2026-08-11 21:40:49 +08:00

181 lines
7.3 KiB
Bash
Executable File

#!/usr/bin/env bash
#
# The GUI must not touch the filesystem or git directly.
#
# Once a workspace can live on a remote machine, a path held by `ui::` or
# `terminal::` is not necessarily a path on *this* box, and `std::path`'s
# fs-backed APIs quietly answer for the wrong machine:
# `canonicalize` walks the local filesystem, `is_absolute` says `false` for
# `/home/me` on Windows, `read_dir` lists the client's disk. Everything that may
# be looking at a workspace path has to go through `ui::host_ops` / the `Host`
# trait, which routes to the local disk or the far side as appropriate.
#
# This script enforces that. It is deliberately *not* the raw grep from the
# contract: run bare, that grep reports 42 hits on a clean tree — not one of them
# git, all of them modules whose paths are local by construction — and a guard
# that always fails is a guard everyone learns to ignore. Two things fix it:
#
# 1. Test bodies are cut off properly. The contract's `grep -v '#\[cfg(test)\]'`
# only removes the attribute line itself, leaving the whole test module
# behind it in scope; here the scan stops at the `#[cfg(test)] mod …` that
# opens the trailing test region.
# 2. An explicit allowlist, keyed on (file, pattern) rather than whole files, so
# a module exempted for its `.is_absolute()` still trips on a new
# `std::fs::`. Every entry carries the reason its paths cannot be remote.
#
# Adding an entry is a deliberate act: if the path could ever be a workspace
# path, the answer is `Host`, not an allowlist line.
#
# Usage: bash .github/scripts/check-host-boundary.sh
# Exit 0 = clean, 1 = violations (printed to stderr).
set -euo pipefail
cd "$(dirname "$0")/../.."
ROOTS=(src/ui src/terminal)
# The forbidden constructs, as extended-regex alternatives.
PATTERN='std::fs::|Command::new\("git"\)|\.canonicalize\(\)|\.is_absolute\(\)'
# Allowlist entries are `<path>|<literal pattern>`, one per line. `<path>|*`
# exempts a file wholesale (used only for the boundary module itself). The
# pattern side is matched as a plain substring of the offending line.
ALLOW=$(
cat <<'EOF'
# The host boundary itself: every routed filesystem call lands here by design.
src/ui/host_ops.rs|*
# Themes and presets are app-owned files under the local config dir
# (`themes_dir()`), never a workspace path — a remote workspace does not carry
# the user's color schemes with it.
src/ui/presets.rs|std::fs::
src/ui/presets.rs|.is_absolute()
src/ui/app.rs|std::fs::create_dir_all
# Reading a private key off *this* machine to hash it into a keychain account
# (`core::keychain`). The key is the client's credential; the far side never
# sees the file, only the resulting auth.
src/ui/ssh_prompt.rs|std::fs::read
src/ui/ssh_connect.rs|std::fs::read
src/ui/settings.rs|std::fs::read
# Shell history lives in the local user's home (`~/.zsh_history` &co.) and backs
# this app's own history search. A remote pane's history is the remote shell's
# business, read over the wire, not through here.
src/terminal/history.rs|std::fs::
# Completion is already remote-aware: a remote pane is signalled by `cwd: None`,
# which disables exactly these local-filesystem candidate sources in favour of
# `remote_path_request` / `remote_path_candidates`. The `$PATH` scan is likewise
# this machine's `$PATH`, deliberately withheld from remote panes.
src/terminal/completion.rs|std::fs::read_dir
src/terminal/completion.rs|.is_absolute()
# Bundled completion specs shipped in `assets/completions`, resolved from the
# app bundle / dev manifest dir. Program data, not user or workspace data.
src/terminal/signature.rs|std::fs::read_to_string
# Opening a path scraped out of terminal output resolves it against the local
# cwd — a local-pane affordance; `resolve_existing_path` declines when there is
# no local cwd.
src/terminal/search.rs|.is_absolute()
# Clipboard-image paste stages the bytes into the OS temp dir so an agent TUI
# can be handed a path. Always `std::env::temp_dir()` on this machine.
src/terminal/view.rs|std::fs::create_dir_all
src/terminal/view.rs|std::fs::write
# The source side of a file drop. What the desktop hands over is by
# construction a path on the desktop's own machine, so reading it is a local
# read even when the tree being dropped on is remote — the destination side of
# that copy goes through `Host`, and the one `std::fs::copy` that touches a
# destination sits inside a branch already gated on `host.id().is_local()`.
src/ui/file_copy.rs|std::fs::
EOF
)
# Where the trailing `#[cfg(test)] mod …` starts, if any. Line numbers are
# preserved because only the tail is dropped.
body_of() {
local file=$1 cut
# `attr` starts unset, which awk reads as 0 — so a file whose *first* line
# is `mod something` used to match `attr == NR - 1` and cut the body at
# line 0, leaving `head -n -1` to error out and the file to be scanned as
# empty. Two files opened that way, and the guard was blind to both.
cut=$(awk '
BEGIN { attr = -1 }
/^#\[cfg\(test\)\]$/ { attr = NR }
/^mod [A-Za-z_]/ { if (attr == NR - 1) { print NR - 1; exit } }
' "$file")
if [ -n "$cut" ]; then
head -n "$((cut - 1))" "$file"
else
cat "$file"
fi
}
allowed() {
local file=$1 line=$2 entry path pat
while IFS= read -r entry; do
case "$entry" in '' | '#'*) continue ;; esac
path=${entry%%|*}
pat=${entry#*|}
[ "$path" = "$file" ] || continue
if [ "$pat" = '*' ] || [ "${line#*"$pat"}" != "$line" ]; then
return 0
fi
done <<<"$ALLOW"
return 1
}
# A guard that scans nothing reports success, which is the one failure mode worse
# than a noisy guard. Fail loudly if a root has moved out from under us.
for root in "${ROOTS[@]}"; do
if [ ! -d "$root" ]; then
echo "check-host-boundary: scan root '$root' does not exist (moved? renamed?)" >&2
exit 2
fi
done
violations=0
scanned=0
while IFS= read -r file; do
scanned=$((scanned + 1))
while IFS= read -r hit; do
lineno=${hit%%:*}
text=${hit#*:}
# Prose, not code.
case "$(printf '%s' "$text" | sed 's/^[[:space:]]*//')" in '//'*) continue ;; esac
if allowed "$file" "$text"; then
continue
fi
printf '%s:%s:%s\n' "$file" "$lineno" "$text" >&2
violations=$((violations + 1))
done < <(body_of "$file" | grep -nE "$PATTERN" || true)
done < <(find "${ROOTS[@]}" -name '*.rs' | sort)
if [ "$scanned" -lt 10 ]; then
echo "check-host-boundary: only $scanned files scanned — the roots look wrong" >&2
exit 2
fi
if [ "$violations" -ne 0 ]; then
cat >&2 <<'EOF'
--------------------------------------------------------------------------------
The GUI reached the filesystem/git directly.
A path in `ui::` or `terminal::` may belong to a remote workspace, where these
calls answer for the wrong machine. Route it through `ui::host_ops` / `Host`
instead (`Host::read_dir`, `join`, `is_absolute`, `canonicalize`, …).
If the path genuinely cannot be remote — app config, bundled assets, the local
temp dir — add it to the allowlist in this script with the reason why.
--------------------------------------------------------------------------------
EOF
exit 1
fi
echo "host boundary clean: $scanned files in ${ROOTS[*]}, no direct fs/git calls"