{
  "schema_version": "1.6.1",
  "id": "RPI-CVE-2023-54303",
  "modified": "2026-10-09T11:39:44Z",
  "upstream": [
    "CVE-2023-54303"
  ],
  "published": "2025-12-30T12:24:11Z",
  "summary": "bpf: Disable preemption in bpf_perf_event_output",
  "details": "In the Linux kernel, the following vulnerability has been resolved:\n\nbpf: Disable preemption in bpf_perf_event_output\n\nThe nesting protection in bpf_perf_event_output relies on disabled\npreemption, which is guaranteed for kprobes and tracepoints.\n\nHowever bpf_perf_event_output can be also called from uprobes context\nthrough bpf_prog_run_array_sleepable function which disables migration,\nbut keeps preemption enabled.\n\nThis can cause task to be preempted by another one inside the nesting\nprotection and lead eventually to two tasks using same perf_sample_data\nbuffer and cause crashes like:\n\n  kernel tried to execute NX-protected page - exploit attempt? (uid: 0)\n  BUG: unable to handle page fault for address: ffffffff82be3eea\n  ...\n  Call Trace:\n   ? __die+0x1f/0x70\n   ? page_fault_oops+0x176/0x4d0\n   ? exc_page_fault+0x132/0x230\n   ? asm_exc_page_fault+0x22/0x30\n   ? perf_output_sample+0x12b/0x910\n   ? perf_event_output+0xd0/0x1d0\n   ? bpf_perf_event_output+0x162/0x1d0\n   ? bpf_prog_c6271286d9a4c938_krava1+0x76/0x87\n   ? __uprobe_perf_func+0x12b/0x540\n   ? uprobe_dispatcher+0x2c4/0x430\n   ? uprobe_notify_resume+0x2da/0xce0\n   ? atomic_notifier_call_chain+0x7b/0x110\n   ? exit_to_user_mode_prepare+0x13e/0x290\n   ? irqentry_exit_to_user_mode+0x5/0x30\n   ? asm_exc_int3+0x35/0x40\n\nFixing this by disabling preemption in bpf_perf_event_output.",
  "severity": [
    {
      "type": "CVSS_V3",
      "score": "CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H"
    }
  ],
  "affected": [
    {
      "package": {
        "ecosystem": "Raspberry Pi OS:12",
        "name": "linux"
      },
      "versions": [
        "1:6.1.21-1+rpt1"
      ],
      "ecosystem_specific": {
        "binaries": [
          {
            "binary_name": "linux-image-6.1.0-rpi1-rpi-v6",
            "binary_version": "1:6.1.21-1+rpt1"
          },
          {
            "binary_name": "linux-image-6.1.0-rpi1-rpi-v7",
            "binary_version": "1:6.1.21-1+rpt1"
          },
          {
            "binary_name": "linux-image-6.1.0-rpi1-rpi-v7l",
            "binary_version": "1:6.1.21-1+rpt1"
          },
          {
            "binary_name": "linux-image-6.1.0-rpi1-rpi-v8",
            "binary_version": "1:6.1.21-1+rpt1"
          }
        ]
      },
      "database_specific": {
        "per_version": {
          "1:6.1.21-1+rpt1": {
            "linux_commit": "0afb5e98488aed7017b9bf321b575d0177feb7ed",
            "upstream_version": "6.1.21",
            "assessment": "commit ancestry (strak rule)",
            "upstream_fixed": "6.1.45"
          }
        }
      }
    }
  ],
  "references": [
    {
      "type": "ADVISORY",
      "url": "https://www.cve.org/CVERecord?id=CVE-2023-54303"
    },
    {
      "type": "FIX",
      "url": "https://git.kernel.org/stable/c/3654ed5daf492463c3faa434c7000d45c2da2ace"
    },
    {
      "type": "FIX",
      "url": "https://git.kernel.org/stable/c/a0ac32cf61e5a76e2429e486925a52ee41dd75e3"
    },
    {
      "type": "FIX",
      "url": "https://git.kernel.org/stable/c/f2c67a3e60d1071b65848efaa8c3b66c363dd025"
    }
  ],
  "database_specific": {
    "source": "https://git.kernel.org/pub/scm/linux/security/vulns.git",
    "upstream_modified": "2026-08-05T11:24:39Z",
    "severity": {
      "source": "kernel CNA",
      "cvss_v3_score": 7.8,
      "cvss_v3_severity": "High",
      "scenario": "AV:L - The vulnerable code is BPF/uprobe tracing infrastructure reached only by executing code on the machine (hitting an instrumented userspace function via local process execution). There is no remote or adjacent-network entry point.\nAC:L - The attacker controls both racing sides \u2014 they run all the threads that hit the uprobe \u2014 and can pin them to a single CPU with sched_setaffinity, widen the window with large bpf_perf_event_output payloads, and rely on full preemption being the default on PREEMPT_DYNAMIC distro and Android kernels. Nothing required is outside attacker influence.\nPR:L - In the realistic worst case \u2014 a host running a system-wide uprobe/USDT observability agent (bpftrace, BCC, continuous profiler), whose programs use perf_submit()/bpf_perf_event_output by default \u2014 any unprivileged local user triggers the corruption merely by executing the instrumented binary concurrently. No capability of the attacker's own is needed.\nUI:N - The attacker's own threads hit the uprobe and generate the preemption; no victim action or interaction is required.\nS:U - The corrupted per-CPU perf_sample_data and the resulting code execution are entirely within the kernel's own security authority, with no crossing into a hypervisor, IOMMU, or other authority.\nC:H - The shared slot leaves data->raw pointing at another task's freed/reused kernel stack, so __output_copy() copies frag->size bytes from an arbitrary frag->data pointer into the mmapped perf ring buffer, yielding arbitrary kernel memory disclosure.\nI:H - __output_custom() performs an indirect call through frag->copy read from a reused kernel stack \u2014 confirmed by the \"kernel tried to execute NX-protected page - exploit attempt?\" oops \u2014 giving an attacker-sprayable control-flow hijack primitive on top of concurrent writes to a shared perf_sample_data.\nA:H - The commit message documents a reproducible kernel page fault / oops in perf_output_sample, and the race can be re-triggered at will to repeatedly panic the system."
    }
  }
}