summaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
authorGabriel Schneider <[email protected]>2026-10-29 14:27:42 -0300
committerGabriel Schneider <[email protected]>2026-10-29 14:27:42 -0300
commit46c3c5db5a9a32ec6c43ad23ba041126ead5bd07 (patch)
treec498cd865ee0bc15df4d32ef35e5bad4009a8a8f /src
parente9328a417ef82d72a73f5e3e7d182591616137de (diff)
downloadpardes-46c3c5db5a9a32ec6c43ad23ba041126ead5bd07.tar.gz
pardes-46c3c5db5a9a32ec6c43ad23ba041126ead5bd07.zip
9p_io.zig: the /os table is not swept while a request is out in a syscall, which made a walk of a directory just made answer ENOENT
The tty frame's tick sweeps the table with the turn, but the turn comes back to the editor after a step of its own with a 9P request still out in a stat: the entry that request's walk had just made, which no fid named yet, was freed under it, and the next element looked up nothing. That was fs.py's one-in-a-few `/os` listing failure; walks of a fresh directory against file churn in the session's directory failed 144 in 2799, and none in 4440 after. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Diffstat (limited to 'src')
-rw-r--r--src/9p_io.zig15
1 files changed, 15 insertions, 0 deletions
diff --git a/src/9p_io.zig b/src/9p_io.zig
index 5f22b326..9bcb2827 100644
--- a/src/9p_io.zig
+++ b/src/9p_io.zig
@@ -724,6 +724,12 @@ pub const Listener = struct {
/// and only between the editor's steps: a step out in a syscall may be
/// reading one of these entries.
fn collectOs(l: *Listener) void {
+ // A request out in a host syscall may be using an entry's path, or
+ // made one that no fid names yet (a walk's next element): swept at
+ // the next quiet instead. Taken back after a step of its own, the
+ // editor's frame (tty pollFrame) came here with one out, and a walk
+ // of a directory just made was answered ENOENT.
+ if (pardes.turn.out != 0) return;
const core = l.core;
var i: usize = 0;
while (i < core.fs.os_paths.items.len) {
@@ -1760,6 +1766,15 @@ test "a listening editor posts itself into the 9P registry and unposts on stop"
const n = libc.readlink(entry, &link, link.len);
try testing.expect(n > 0);
try testing.expectEqualStrings(sock, link[0..@intCast(n)]);
+ // A host path no fid names is swept, but never while a request is
+ // out in a syscall: it may be that request's, mid-walk.
+ try p.fs.os_paths.append(gpa, .{ .node = @import("fs.zig").os_node | 1, .path = try gpa.dupe(u8, "/tmp") });
+ pardes.turn.out = 1;
+ l.collectOs();
+ pardes.turn.out = 0;
+ try testing.expectEqual(@as(usize, 1), p.fs.os_paths.items.len);
+ l.collectOs();
+ try testing.expectEqual(@as(usize, 0), p.fs.os_paths.items.len);
}
// Stopping takes the name back out, so the next editor of that name
// is not refused by its own corpse; the lock goes with the socket.