fix(controller): persist_job_hook_fun OOB read on RUSAGE_UPDATE hook (bug-1502)

The RUSAGE_UPDATE job hook fires with (cluster_t*, xce_job_rusages_t*) as
its varargs — the apiserver and scheduler subscribers parse exactly that
contract — but persist_job_hook_fun's default case treated the second arg
as a job_t* and read j->job.state at offset 56. A xce_job_rusages_t is 56
bytes and shares only jobid@0 with xce_job_t (which is why the jobid read
coincidentally worked), so the state read landed one byte past the struct:
ASAN heap-buffer-overflow (READ of size 4) in thread T7 whenever a job runs
and the worker pushes rusage updates. This was the recurring controller
crash (term 3/4/5) blocking job-PENDING verification.

Add a dedicated HOOK_DETAIL_JOB_RUSAGE_UPDATE case that reads jobid from
the xce_job_rusages_t* and bakes the job's current state from a jobmgr
key_find on the vararg cluster_t* (race-free: the sole caller
update_job_rusages() holds that job's write lock while firing the
AFTER_EVENT hook; re-locking the job inside the hook would deadlock).

Regression test: test_job_hook_rusage_writes_wal_entry now passes a real
56-byte xce_job_rusages_t instead of a full job_t (which masked the
overflow) — ASAN stack-buffer-overflow under the old code, PASS under the
fix. Full controller unit suite: 458 tests, 0 failures.

Co-Authored-By: Claude <noreply@anthropic.com>

.wolf/buglog.json

@@ -14756,5 +14756,25 @@
     "related_bugs": [],
     "occurrences": 1,
     "last_seen": "2026-08-04"
+  },
+  {
+    "id": "bug-1502",
+    "timestamp": "2026-08-04",
+    "error_message": "controller ASAN heap-buffer-overflow (READ of size 4, thread T7): 56-byte xce_job_rusages_t buffer from decode_JobRusages_ptr, read at offset 56 (1 byte past end). Recurring crash that kills the controller whenever a job runs and the worker pushes rusage updates (term=3/4/5 all crashed).",
+    "file": "src/controller/persist.c (persist_job_hook_fun)",
+    "root_cause": "RUSAGE_UPDATE job hook arg mismatch: update_job_rusages() fires HOOK_DETAIL_JOB_RUSAGE_UPDATE with (cluster_t*, xce_job_rusages_t*) as its 2nd va_arg (apiserver/sched subscribers parse this exact contract), but persist_job_hook_fun default case treated it as job_t*. xce_job_rusages_t is 56 bytes and has jobid@0 (coincidentally correct) but NO state field; j->job.state lives at offset 56 of xce_job_t, so the read went 4 bytes past the rusage struct -> heap-buffer-overflow. The unit test masked it by passing a full job_t.",
+    "fix": "Added case HOOK_DETAIL_JOB_RUSAGE_UPDATE in persist_job_hook_fun: read jobid from xce_job_rusages_t*, bake the job current state via a jobmgr key_find on the vararg cluster_t* (safe: sole caller update_job_rusages holds that job write lock while firing). Regression test passes a real 56-byte xce_job_rusages_t (stack-buffer-overflow under old code, PASS under fix).",
+    "tags": [
+      "asan",
+      "heap-buffer-overflow",
+      "controller",
+      "persist",
+      "hook",
+      "job-rusage",
+      "crash"
+    ],
+    "related_bugs": [],
+    "occurrences": 3,
+    "last_seen": "2026-08-04"
   }
 ]

.wolf/memory.md

@@ -7368,3 +7368,4 @@
 | 05:49 | Created ../../../../tmp/frpc-watchdog.sh | — | ~338 |
 | 13:50 | 看门狗改为纯日志模式(不自动重启): 移到 /home/dankeer/bin/frpc-watchdog.sh + crontab @reboot 开机自启; 只在隧道状态变化时写日志(DOWN抓快照/恢复记录/持续DOWN每5min) | /home/dankeer/bin/frpc-watchdog.sh | 部署完成, 用户人工监视 | ~200 |
 | 05:50 | Session end: 9 writes across 5 files (frpcdbg_mon.sh, frpc_prod.toml, frpc_mon.sh, frpc_watchdog.sh, frpc-watchdog.sh) | 2 reads | ~2219 tok |
+| 09:39 | 修复 controller 反复崩溃 bug-1502: persist_job_hook_fun 把 RUSAGE_UPDATE hook 的第2参 (xce_job_rusages_t*) 当 job_t* 用, 读 offset 56 越界 56字节结构 -> ASAN heap-buffer-overflow. 加专用 case 从 jobmgr 取当前 state | src/controller/persist.c, testsuite/test_persist_hooks.c | RED(旧代码 stack-buffer-overflow)/GREEN(修复后 PASS), 全套458测试通过 | ~350 |

src/controller/persist.c

@@ -6573,8 +6573,35 @@ int persist_job_hook_fun(int type, int situation, int detail,
         js.jobid = j->job.jobid; js.state = (uint32_t)j->job.state;
         break;
     }
+    case HOOK_DETAIL_JOB_RUSAGE_UPDATE: {
+        /* args: cluster_t*, xce_job_rusages_t* — NOT job_t* (bug-1502).
+         * The RUSAGE_UPDATE hook carries the rusage struct, which has no job
+         * state (apiserver_job_rusage_update_fun / sched_job_rusage_update_fun
+         * parse this exact contract). The old default-case path treated the
+         * 2nd arg as a job_t* and read j->job.state at offset 56 — one byte
+         * past a 56-byte xce_job_rusages_t → ASAN heap-buffer-overflow (the
+         * only field the two structs share is jobid@0, so that read
+         * coincidentally worked). Bake the job's CURRENT state from the
+         * jobmgr: the sole caller update_job_rusages() fires this AFTER_EVENT
+         * while holding the job's write lock, so the pointer returned here IS
+         * that same write-locked object and reading its state is race-free.
+         * Never re-lock the job inside a hook — the caller already holds it,
+         * and a same-thread re-lock would deadlock. */
+        cluster_t*         c = va_arg(ap, cluster_t*);
+        xce_job_rusages_t* r = va_arg(ap, xce_job_rusages_t*);
+        js.jobid = r->jobid;
+        if (c != NULL && c->jobmgr != NULL) {
+            job_t* jj = NULL;
+            lock(c->jobmgr, RTTI_LOCK_READ);
+            key_find(c->jobmgr, (void*)r->jobid, (void**)&jj);
+            unlock(c->jobmgr);
+            if (jj != NULL)
+                js.state = (uint32_t)jj->job.state;
+        }
+        break;
+    }
     default: {
-        /* STOP, RESUME, UPDATE, CKPT, PREASON, RUSAGE — args: cluster_t*, job_t* */
+        /* STOP, RESUME, UPDATE, CKPT, PREASON — args: cluster_t*, job_t* */
         (void)va_arg(ap, cluster_t*);
         job_t* j = va_arg(ap, job_t*);
         js.jobid = j->job.jobid; js.state = (uint32_t)j->job.state;

src/controller/testsuite/test_persist_hooks.c

@@ -14,6 +14,7 @@
 #include "persist.h"
 #include "xce/cluster_def.h"
 #include "xlist.h"
+#include "controller.capnp.h" /* xce_job_rusages_t (bug-1502 regression test) */
 
 /* ------------------------------------------------------------------ */
 /* Test helpers                                                         */
@@ -276,9 +277,19 @@ static void test_job_hook_rusage_writes_wal_entry(void)
 
     uint64_t lsn_before = p->current_lsn;
 
+    /* bug-1502 regression: production fires RUSAGE_UPDATE with a real
+     * xce_job_rusages_t* (56 bytes, NO job-state field) as the second hook
+     * arg — NOT a job_t*. The old persist_job_hook_fun default case read
+     * j->job.state at offset 56, one byte past this struct → ASAN
+     * heap-buffer-overflow. Passing the real minimal struct here (instead of
+     * the old &g_stub_job, a full job_t that masked the overflow) reproduces
+     * the production shape exactly. */
+    xce_job_rusages_t rusages = {0};
+    rusages.jobid = 0x1234567890ABCDEFULL;
+
     int rc = call_job_hook(HOOK_AFTER_EVENT, HOOK_AFTER_EVENT,
                            HOOK_DETAIL_JOB_RUSAGE_UPDATE, p,
-                           (cluster_t*)NULL, &g_stub_job);
+                           (cluster_t*)NULL, &rusages);
     TEST_ASSERT_EQUAL_INT(0, rc);
     TEST_ASSERT_GREATER_THAN(lsn_before, p->current_lsn);