Ship the Files tab container-side only
Four successive audits found the same thing: host filesystem paths crossing
IPC is where the criticals in this work live. The most recent one found the
`link(2)` upload reservation returning success against a *directory* (linking
into it, leaving permanent stray files, and via a symlink-to-directory writing
outside the validated write root), failing every upload permanently on any
filesystem without hard links, and the post-resolution credential check
weakened from a general rule to an eleven-name denylist.
Rather than fix that a fifth time, the Files tab ships as what it is good at:
a browser, viewer and renamer that never touches the host.
Removed: `upload_file_to_container`, `download_container_file`, and everything
that existed only for them — the whole reservation (`UPLOAD_RESERVATION_SCRIPT`,
`reserve_upload_destination`, the placeholder rollback, `exec_oneshot_as_within`
which had no other caller), `stream_container_file_to_host`, `ChannelReader`,
`save_to_host`, the download ceiling, and the collision marker with its
frontend contract. On the frontend: the upload button, the pane's
`onDragDropEvent` handler, both "Save to host…" affordances, `uploadPaths` /
`downloadFile` / the overwrite prompt, and `OverwriteConfirmModal`.
`lib/uploadErrors.ts` is now `lib/refusalText.ts` and keeps only the half that
turns any backend refusal into the sentence a person reads.
Kept, and not weakened: `upload_host_file_to_terminal` and
`download_container_backup`. They predate this work, their hardening is a real
improvement over main, and they are now the whole answer to "how do I get a
file in or out" — drop it on the Terminal, or Back up container. The drop gate
(`lib/dropTarget.ts`, `PaneVisibility`) is untouched.
`resolve_host_path` gets the general hidden-component rule back. Round 3
replaced it with `HOST_CREDENTIAL_DIRS`, which is allow-by-omission for the
rest of `$HOME`: `~/.local/bin` (write there and you own the user's next shell
command), `~/.password-store`, browser profiles and `~/.pki/nssdb` were all
reachable through a planted symlink with a visible name — verified against a
real home directory, and all five refused now. It over-catches `.pnpm` and
`~/.cache`; for two occasional callers that is the cheaper mistake, and the
refusal says which folder it resolved through.
Two defects fixed while in here:
* A symlinked directory listed as empty. `find` defaults to `-P`, which does
not follow a symlink even as the starting point, so `-mindepth 1` discarded
the only match and a real directory rendered as "Empty directory" — a
first-order defect now that browsing *is* the feature. `-H` follows the
starting point and nothing else, so a loop is `ELOOP` rather than a walk
that does not end; verified against a live container for a symlinked
directory, a broken link and a loop. `find`'s errno for the loop case is
now a sentence.
* `finish_download`'s replace path fired on *any* rename failure with a
destination present — a vanished partial, a permission error, a directory
at the destination — and deleted the user's file to complete a move that
could not complete. It is now fenced to Windows (where a rename onto an
existing path genuinely fails) and to a partial that still exists.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GBq2rGum6GX7xXgsas1fDc
This commit is contained in:
@@ -0,0 +1,124 @@
|
||||
/**
|
||||
* Turning a backend refusal into the sentence a person reads.
|
||||
*
|
||||
* Tauri command errors cross the IPC boundary as whatever `serde` made of them:
|
||||
* a bare string from `Err(String)`, an object from a `#[derive(Serialize)]`
|
||||
* error enum, or an `Error` if a JS layer wrapped it on the way through. All
|
||||
* three are the same refusal, and the UI must not read differently depending on
|
||||
* which one a future refactor produces — so everything here is tolerant about
|
||||
* the *shape* of an error and picks the most human string out of it.
|
||||
*/
|
||||
|
||||
/**
|
||||
* Field names a serialised Rust error realistically uses for its discriminant
|
||||
* and for its human text. `error` is listed as a discriminant field and yet
|
||||
* routinely carries a whole sentence, which is why a kind string is read as
|
||||
* prose too.
|
||||
*/
|
||||
const KIND_FIELDS = ["kind", "code", "type", "error", "reason"] as const;
|
||||
const MESSAGE_FIELDS = ["message", "msg", "detail", "description"] as const;
|
||||
|
||||
function asRecord(e: unknown): Record<string, unknown> | null {
|
||||
return typeof e === "object" && e !== null ? (e as Record<string, unknown>) : null;
|
||||
}
|
||||
|
||||
/**
|
||||
* Every string an error carries, flattened: the error itself if it is one, its
|
||||
* message-ish fields, and its kind-ish fields. Nesting is followed one level
|
||||
* because a wrapped error (`{ error: { message: … } }`) is the same refusal.
|
||||
*/
|
||||
function stringsIn(e: unknown, depth = 0): string[] {
|
||||
if (typeof e === "string") return [e];
|
||||
if (e instanceof Error) return [e.message, e.name];
|
||||
const record = asRecord(e);
|
||||
if (!record || depth > 1) return [];
|
||||
const out: string[] = [];
|
||||
const walk = (value: unknown) => {
|
||||
if (typeof value === "string") out.push(value);
|
||||
else if (value !== undefined) out.push(...stringsIn(value, depth + 1));
|
||||
};
|
||||
for (const field of KIND_FIELDS) walk(record[field]);
|
||||
for (const field of MESSAGE_FIELDS) walk(record[field]);
|
||||
return out;
|
||||
}
|
||||
|
||||
/**
|
||||
* Fragments that identify a refusal the backend already wrote **for a person**.
|
||||
*
|
||||
* The file commands guard two policies that a user can trip over by accident,
|
||||
* and both answer with a finished sentence that names the offending path and
|
||||
* says what to do instead:
|
||||
*
|
||||
* the path goes through ".ssh", a hidden folder — Triple-C will not save …
|
||||
* Folder path is outside the folders this panel can change (/workspace, /home/claude, /tmp): /etc
|
||||
*
|
||||
* Those sentences were being used as the *detail* of a generic toast, and
|
||||
* `ToastHost` renders a detail as collapsed monospace behind a "Details"
|
||||
* button — so the one part of the message that explained anything was the part
|
||||
* nobody saw. Matching them here lets the caller promote the sentence to the
|
||||
* toast's headline.
|
||||
*
|
||||
* Matched on a stable fragment rather than the whole string, because the path
|
||||
* and the verb ("save"/"read", "file"/"folder") vary per call. Deliberately a
|
||||
* short list: an error that is *not* recognised still reports exactly as it
|
||||
* did before, so a wrong guess here can only fail to promote, never mangle.
|
||||
*/
|
||||
const REFUSAL_MARKERS = [
|
||||
// `validate_host_path` — hidden host component, and system locations.
|
||||
"Triple-C will not",
|
||||
// `validate_container_write_path` — outside /workspace, /home/claude, /tmp.
|
||||
"outside the folders this panel can change",
|
||||
] as const;
|
||||
|
||||
/**
|
||||
* `Error: …`, `TypeError: …`, `invoke failed: …` — wrappers a JS layer may have
|
||||
* put in front of the backend's sentence on the way through. Stripped so the
|
||||
* prose starts where the backend started it; applied twice at most, because a
|
||||
* doubly-wrapped error is the realistic worst case and looping on user text is
|
||||
* not.
|
||||
*/
|
||||
const WRAPPER_PREFIX = /^(?:uncaught\s*(?:\(in promise\)\s*)?)?(?:[a-z]*error|invoke(?:\s+failed)?)\s*:\s*/i;
|
||||
|
||||
function stripWrapper(text: string): string {
|
||||
let out = text.trim();
|
||||
for (let i = 0; i < 2; i++) {
|
||||
const next = out.replace(WRAPPER_PREFIX, "").trim();
|
||||
if (next === out) break;
|
||||
out = next;
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
/**
|
||||
* The backend's own user-facing sentence, when this failure is one — otherwise
|
||||
* `null`, and the caller reports it however it reported everything else.
|
||||
*/
|
||||
export function readableRefusal(e: unknown): string | null {
|
||||
for (const s of stringsIn(e)) {
|
||||
const text = stripWrapper(s);
|
||||
if (REFUSAL_MARKERS.some((marker) => text.includes(marker))) return text;
|
||||
}
|
||||
return null;
|
||||
}
|
||||
|
||||
/**
|
||||
* The most human form of any failure, for the places that show one verbatim.
|
||||
*
|
||||
* `String(e)` is what these used to be, which turns a serialised error object
|
||||
* into `[object Object]` and leaves a JS wrapper prefix on a sentence that
|
||||
* reads perfectly well without it.
|
||||
*/
|
||||
export function errorText(e: unknown): string {
|
||||
const readable = readableRefusal(e);
|
||||
if (readable) return readable;
|
||||
if (typeof e === "string") return stripWrapper(e);
|
||||
if (e instanceof Error) return stripWrapper(e.message);
|
||||
const record = asRecord(e);
|
||||
if (record) {
|
||||
for (const field of [...MESSAGE_FIELDS, ...KIND_FIELDS]) {
|
||||
const value = record[field];
|
||||
if (typeof value === "string" && value.trim().length > 0) return stripWrapper(value);
|
||||
}
|
||||
}
|
||||
return String(e);
|
||||
}
|
||||
Reference in New Issue
Block a user