nucleus-container 0.3.3

Extremely lightweight Docker alternative for agents and production services — isolated execution using cgroups, namespaces, seccomp, Landlock, and gVisor
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
//! Bitmap-based BPF compiler for seccomp filters.
//!
//! Replaces seccompiler's default linear-scan BPF generation with a bitmap
//! lookup that resolves unconditional syscall allows in O(1) – a constant
//! ~10 BPF instructions regardless of allowlist size.
//!
//! Program structure:
//!
//! 1. Arch validation (3 insns)
//! 2. Compute bit_position = nr & 31 → M\[0\], word_index = nr >> 5 (5 insns)
//! 3. Dispatch on word_index → load 32-bit bitmap constant (≤46 insns)
//! 4. Bitmap test: `(bitmap >> bit_position) & 1` → ALLOW if set (5 insns)
//! 5. Arg-filtered syscall chains for conditional rules (variable)
//! 6. Default deny
//!
//! Hot path for unconditional allows: ~15 instructions (vs ~120+ for linear scan).

use crate::error::{NucleusError, Result};
use seccompiler::{sock_filter, BpfProgram, SeccompAction, SeccompRule, TargetArch};
use std::collections::BTreeMap;

// --- cBPF constants (pub(crate) in seccompiler, redefined here) ---

const BPF_LD: u16 = 0x00;
const BPF_LDX: u16 = 0x01;
const BPF_ST: u16 = 0x02;
const BPF_ALU: u16 = 0x04;
const BPF_JMP: u16 = 0x05;
const BPF_RET: u16 = 0x06;

const BPF_W: u16 = 0x00;
const BPF_ABS: u16 = 0x20;
const BPF_IMM: u16 = 0x00;
const BPF_MEM: u16 = 0x60;

const BPF_AND: u16 = 0x50;
const BPF_RSH: u16 = 0x70;

const BPF_JA: u16 = 0x00;
const BPF_JEQ: u16 = 0x10;
const BPF_K: u16 = 0x00;
const BPF_X: u16 = 0x08;

const SECCOMP_DATA_NR_OFFSET: u32 = 0;
const SECCOMP_DATA_ARCH_OFFSET: u32 = 4;

const BPF_MAX_LEN: usize = 4096;

// Architecture audit values from linux/audit.h
const AUDIT_ARCH_X86_64: u32 = 62 | 0x8000_0000 | 0x4000_0000;
const AUDIT_ARCH_AARCH64: u32 = 183 | 0x8000_0000 | 0x4000_0000;
const AUDIT_ARCH_RISCV64: u32 = 243 | 0x8000_0000 | 0x4000_0000;

fn arch_audit_value(arch: TargetArch) -> u32 {
    match arch {
        TargetArch::x86_64 => AUDIT_ARCH_X86_64,
        TargetArch::aarch64 => AUDIT_ARCH_AARCH64,
        TargetArch::riscv64 => AUDIT_ARCH_RISCV64,
    }
}

/// Number of 32-bit bitmap words. Covers syscall numbers 0..NUM_BITMAP_WORDS*32-1.
/// x86_64 syscalls go up to ~467, so 15 words (0..479) is sufficient.
const NUM_BITMAP_WORDS: usize = 15;

/// Scratch memory slot for bit_position.
const M_BIT_POS: u32 = 0;

#[inline(always)]
fn bpf_stmt(code: u16, k: u32) -> sock_filter {
    sock_filter {
        code,
        jt: 0,
        jf: 0,
        k,
    }
}

#[inline(always)]
fn bpf_jump(code: u16, k: u32, jt: u8, jf: u8) -> sock_filter {
    sock_filter { code, jt, jf, k }
}

/// Compile a seccomp BPF program using bitmap lookup for O(1) dispatch.
///
/// Rules with empty `Vec<SeccompRule>` are unconditional allows, encoded as
/// bits in a 15-word bitmap (covering syscall numbers 0..479). Rules with
/// non-empty conditions go through a short linear scan of arg-check chains
/// (typically ~7 syscalls).
///
/// The generated program is semantically equivalent to seccompiler's linear
/// scan but executes in O(1) for the common case of unconditional allows.
pub fn compile_bitmap_bpf(
    rules: BTreeMap<i64, Vec<SeccompRule>>,
    mismatch_action: SeccompAction,
    match_action: SeccompAction,
    target_arch: TargetArch,
) -> Result<BpfProgram> {
    let mismatch_val: u32 = mismatch_action.into();
    let match_val: u32 = match_action.into();
    let audit_arch = arch_audit_value(target_arch);

    // Separate unconditional allows (bitmap) from arg-filtered (linear scan).
    let mut bitmap: [u32; NUM_BITMAP_WORDS] = [0; NUM_BITMAP_WORDS];
    let mut arg_filtered: Vec<(i64, Vec<SeccompRule>)> = Vec::new();

    for (nr, chain) in rules {
        if nr < 0 {
            continue;
        }
        if chain.is_empty() {
            let word_idx = (nr >> 5) as usize;
            if word_idx < NUM_BITMAP_WORDS {
                bitmap[word_idx] |= 1u32 << (nr & 31);
            }
        } else {
            arg_filtered.push((nr, chain));
        }
    }

    // Build arg section first – we need its size for dispatch jump offsets.
    let arg_section = build_arg_section(arg_filtered, mismatch_val, match_val);

    // Only emit dispatch entries for non-zero bitmap words.
    let active_words: Vec<(usize, u32)> = bitmap
        .iter()
        .enumerate()
        .filter(|(_, &w)| w != 0)
        .map(|(i, &w)| (i, w))
        .collect();
    let dispatch_entry_count = active_words.len();
    // Each active word: JEQ + LD IMM + JA = 3 insns. Plus 1 for the out-of-range JA.
    let dispatch_len = dispatch_entry_count * 3 + 1;
    let test_len: usize = 5; // LDX M[0]; RSH X; AND 1; JEQ; RET

    let mut prog: BpfProgram = Vec::with_capacity(64 + arg_section.len());

    // === Section 1: Architecture validation ===
    prog.push(bpf_stmt(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_ARCH_OFFSET));
    prog.push(bpf_jump(BPF_JMP | BPF_JEQ | BPF_K, audit_arch, 1, 0));
    prog.push(bpf_stmt(
        BPF_RET | BPF_K,
        libc::SECCOMP_RET_KILL_PROCESS as u32,
    ));

    // === Section 2: Setup ===
    // Compute bit_position = nr & 31, save in M[0].
    // Compute word_index = nr >> 5 into A.
    prog.push(bpf_stmt(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_NR_OFFSET)); // A = nr
    prog.push(bpf_stmt(BPF_ALU | BPF_AND | BPF_K, 31)); // A = nr & 31
    prog.push(bpf_stmt(BPF_ST, M_BIT_POS)); // M[0] = bit_pos
    prog.push(bpf_stmt(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_NR_OFFSET)); // A = nr
    prog.push(bpf_stmt(BPF_ALU | BPF_RSH | BPF_K, 5)); // A = word_index

    // === Section 3: Dispatch on word_index ===
    //
    // For each non-zero bitmap word i:
    //   JEQ i, jt=0, jf=2   →  if match, fall through to LD; else skip 2
    //   LD IMM bitmap[i]
    //   JA test_offset       →  jump to bitmap test section
    //
    // After all entries:
    //   JA arg_offset        →  word_index not in bitmap; skip to arg section

    let test_start = dispatch_len; // offset from dispatch start to test section
    for (entry_idx, &(word_idx, word_val)) in active_words.iter().enumerate() {
        let insn_pos = entry_idx * 3; // position within dispatch block
        prog.push(bpf_jump(BPF_JMP | BPF_JEQ | BPF_K, word_idx as u32, 0, 2));
        prog.push(bpf_stmt(BPF_LD | BPF_IMM, word_val));
        let ja_offset = test_start - insn_pos - 3; // distance from JA to test start
        prog.push(bpf_stmt(BPF_JMP | BPF_JA, ja_offset as u32));
    }
    // Out-of-range: jump past test section to arg section (or default deny).
    prog.push(bpf_stmt(BPF_JMP | BPF_JA, test_len as u32));

    // === Section 4: Bitmap test ===
    //
    // X = M[0] (bit_position)
    // A = bitmap_word >> X
    // A = A & 1
    // if A == 1: return ALLOW
    // else: fall through to arg section
    prog.push(bpf_stmt(BPF_LDX | BPF_MEM, M_BIT_POS));
    prog.push(bpf_stmt(BPF_ALU | BPF_RSH | BPF_X, 0));
    prog.push(bpf_stmt(BPF_ALU | BPF_AND | BPF_K, 1));
    prog.push(bpf_jump(BPF_JMP | BPF_JEQ | BPF_K, 1, 0, 1)); // bit set → RET
    prog.push(bpf_stmt(BPF_RET | BPF_K, match_val)); // ALLOW

    // === Section 5: Arg-filtered syscall chains ===
    prog.extend(arg_section);

    // === Section 6: Default deny ===
    prog.push(bpf_stmt(BPF_RET | BPF_K, mismatch_val));

    if prog.len() >= BPF_MAX_LEN {
        return Err(NucleusError::SeccompError(format!(
            "BPF program too large: {} instructions (max {})",
            prog.len(),
            BPF_MAX_LEN
        )));
    }

    Ok(prog)
}

/// Build the arg-filtered section: a linear scan of syscalls that need
/// argument-level checks (e.g. clone namespace flags, ioctl request codes).
///
/// Each chain is structured identically to seccompiler's `append_syscall_chain`:
///   JEQ nr → rule chain → RET mismatch
///
/// On JEQ mismatch, the fail-path chains through rule entries to the next
/// syscall's chain (preserving A = nr for the next JEQ).
fn build_arg_section(
    arg_filtered: Vec<(i64, Vec<SeccompRule>)>,
    mismatch_val: u32,
    match_val: u32,
) -> BpfProgram {
    if arg_filtered.is_empty() {
        return Vec::new();
    }

    let mut section = Vec::new();
    // Reload nr – the bitmap test section clobbered A and X.
    section.push(bpf_stmt(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_NR_OFFSET));

    for (syscall_nr, chain) in arg_filtered {
        build_syscall_chain(syscall_nr, chain, mismatch_val, match_val, &mut section);
    }

    section
}

/// Build a single syscall's rule chain, identical to seccompiler's
/// `append_syscall_chain`.
///
/// Uses `SeccompRule::into::<BpfProgram>()` to generate argument-check BPF
/// for each rule, preserving the existing jump-chain semantics.
fn build_syscall_chain(
    syscall_nr: i64,
    chain: Vec<SeccompRule>,
    mismatch_val: u32,
    match_val: u32,
    out: &mut BpfProgram,
) {
    // Convert each rule to BPF and append RET match_action.
    let chain_bpf: Vec<BpfProgram> = chain
        .into_iter()
        .map(|rule| {
            let mut bpf: BpfProgram = rule.into();
            bpf.push(bpf_stmt(BPF_RET | BPF_K, match_val));
            bpf
        })
        .collect();

    // Chain header: check syscall number.
    // jt=0: fall through (matched). jf=1: jump to first rule's fail entry.
    out.push(bpf_jump(BPF_JMP | BPF_JEQ | BPF_K, syscall_nr as u32, 0, 1));

    if chain_bpf.is_empty() {
        // Unconditional allow (shouldn't normally reach here from bitmap path,
        // but handle for correctness with custom profiles).
        out.push(bpf_stmt(BPF_JMP | BPF_JA, 1));
        out.push(bpf_stmt(BPF_JMP | BPF_JA, 2));
        out.push(bpf_stmt(BPF_RET | BPF_K, match_val));
    } else {
        for mut rule_bpf in chain_bpf {
            out.append(&mut rule_bpf);
        }
    }

    // All rules failed for this syscall → deny.
    out.push(bpf_stmt(BPF_RET | BPF_K, mismatch_val));
}

#[cfg(test)]
mod tests {
    use super::*;
    use seccompiler::{SeccompCmpArgLen, SeccompCmpOp, SeccompCondition};

    const AUDIT_ARCH_X86_64: u32 = 62 | 0x8000_0000 | 0x4000_0000;

    /// Minimal software BPF interpreter for testing seccomp programs.
    fn bpf_eval(prog: &[sock_filter], data: &[u8]) -> u32 {
        let mut a: u32 = 0;
        let mut x: u32 = 0;
        let mut mem: [u32; 16] = [0; 16];
        let mut pc: usize = 0;

        for _ in 0..10_000 {
            // Safety limit
            if pc >= prog.len() {
                panic!("BPF: fell off end of program at pc={}", pc);
            }
            let insn = &prog[pc];
            match insn.code {
                0x00 => {
                    // BPF_LD | BPF_IMM: A = k
                    a = insn.k;
                    pc += 1;
                }
                0x20 => {
                    // BPF_LD | BPF_W | BPF_ABS: A = *(u32*)(data + k)
                    let off = insn.k as usize;
                    a = u32::from_ne_bytes(data[off..off + 4].try_into().unwrap());
                    pc += 1;
                }
                0x60 => {
                    // BPF_LD | BPF_MEM: A = M[k]
                    a = mem[insn.k as usize];
                    pc += 1;
                }
                0x61 => {
                    // BPF_LDX | BPF_MEM: X = M[k]
                    x = mem[insn.k as usize];
                    pc += 1;
                }
                0x02 => {
                    // BPF_ST: M[k] = A
                    mem[insn.k as usize] = a;
                    pc += 1;
                }
                0x54 => {
                    // BPF_ALU | BPF_AND | BPF_K: A &= k
                    a &= insn.k;
                    pc += 1;
                }
                0x74 => {
                    // BPF_ALU | BPF_RSH | BPF_K: A >>= k
                    a >>= insn.k;
                    pc += 1;
                }
                0x7c => {
                    // BPF_ALU | BPF_RSH | BPF_X: A >>= X
                    a = a.checked_shr(x).unwrap_or(0);
                    pc += 1;
                }
                0x05 => {
                    // BPF_JMP | BPF_JA: pc += k
                    pc += 1 + insn.k as usize;
                }
                0x15 => {
                    // BPF_JMP | BPF_JEQ | BPF_K
                    if a == insn.k {
                        pc += 1 + insn.jt as usize;
                    } else {
                        pc += 1 + insn.jf as usize;
                    }
                }
                0x25 => {
                    // BPF_JMP | BPF_JGT | BPF_K
                    if a > insn.k {
                        pc += 1 + insn.jt as usize;
                    } else {
                        pc += 1 + insn.jf as usize;
                    }
                }
                0x35 => {
                    // BPF_JMP | BPF_JGE | BPF_K
                    if a >= insn.k {
                        pc += 1 + insn.jt as usize;
                    } else {
                        pc += 1 + insn.jf as usize;
                    }
                }
                0x06 => {
                    // BPF_RET | BPF_K: return k
                    return insn.k;
                }
                0x07 => {
                    // BPF_MISC | BPF_TAX: X = A
                    x = a;
                    pc += 1;
                }
                other => panic!("BPF: unknown opcode 0x{:04x} at pc={}", other, pc),
            }
        }
        panic!("BPF: execution limit exceeded");
    }

    /// Build a seccomp_data byte array for testing.
    fn make_seccomp_data(nr: u32, arch: u32, args: [u64; 6]) -> Vec<u8> {
        let mut data = vec![0u8; 64];
        data[0..4].copy_from_slice(&nr.to_ne_bytes());
        data[4..8].copy_from_slice(&arch.to_ne_bytes());
        // instruction_pointer at offset 8 (zeroed)
        for (i, arg) in args.iter().enumerate() {
            let offset = 16 + i * 8;
            data[offset..offset + 8].copy_from_slice(&arg.to_ne_bytes());
        }
        data
    }

    const RET_ALLOW: u32 = 0x7fff_0000; // SECCOMP_RET_ALLOW
    const RET_KILL: u32 = 0x8000_0000; // SECCOMP_RET_KILL_PROCESS

    #[test]
    fn test_unconditional_allows() {
        let mut rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();
        rules.insert(0, Vec::new()); // read
        rules.insert(1, Vec::new()); // write
        rules.insert(2, Vec::new()); // open
        rules.insert(60, Vec::new()); // exit
        rules.insert(231, Vec::new()); // exit_group

        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // Allowed syscalls should return ALLOW
        for nr in [0, 1, 2, 60, 231] {
            let data = make_seccomp_data(nr, AUDIT_ARCH_X86_64, [0; 6]);
            assert_eq!(
                bpf_eval(&prog, &data),
                RET_ALLOW,
                "syscall {} should be allowed",
                nr
            );
        }

        // Disallowed syscalls should return KILL
        for nr in [3, 4, 59, 100, 300] {
            let data = make_seccomp_data(nr, AUDIT_ARCH_X86_64, [0; 6]);
            assert_eq!(
                bpf_eval(&prog, &data),
                RET_KILL,
                "syscall {} should be killed",
                nr
            );
        }
    }

    #[test]
    fn test_wrong_arch_is_killed() {
        let mut rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();
        rules.insert(0, Vec::new());

        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // Wrong architecture should return KILL_PROCESS
        let data = make_seccomp_data(0, 0xDEADBEEF, [0; 6]);
        assert_eq!(
            bpf_eval(&prog, &data),
            libc::SECCOMP_RET_KILL_PROCESS as u32
        );
    }

    #[test]
    fn test_arg_filtered_syscalls() {
        let mut rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();
        rules.insert(0, Vec::new()); // read: unconditional allow

        // ioctl (16): allow only TCGETS (0x5401)
        let cond =
            SeccompCondition::new(1, SeccompCmpArgLen::Dword, SeccompCmpOp::Eq, 0x5401).unwrap();
        rules.insert(16, vec![SeccompRule::new(vec![cond]).unwrap()]);

        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // read: unconditional allow
        let data = make_seccomp_data(0, AUDIT_ARCH_X86_64, [0; 6]);
        assert_eq!(bpf_eval(&prog, &data), RET_ALLOW);

        // ioctl with TCGETS: allowed
        let data = make_seccomp_data(16, AUDIT_ARCH_X86_64, [0, 0x5401, 0, 0, 0, 0]);
        assert_eq!(bpf_eval(&prog, &data), RET_ALLOW);

        // ioctl with disallowed request: killed
        let data = make_seccomp_data(16, AUDIT_ARCH_X86_64, [0, 0x1234, 0, 0, 0, 0]);
        assert_eq!(bpf_eval(&prog, &data), RET_KILL);

        // unknown syscall: killed
        let data = make_seccomp_data(999, AUDIT_ARCH_X86_64, [0; 6]);
        assert_eq!(bpf_eval(&prog, &data), RET_KILL);
    }

    #[test]
    fn test_multiple_rules_per_syscall() {
        let mut rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();

        // socket (41): allow AF_UNIX (1) and AF_INET (2)
        let cond_unix =
            SeccompCondition::new(0, SeccompCmpArgLen::Dword, SeccompCmpOp::Eq, 1).unwrap();
        let cond_inet =
            SeccompCondition::new(0, SeccompCmpArgLen::Dword, SeccompCmpOp::Eq, 2).unwrap();
        rules.insert(
            41,
            vec![
                SeccompRule::new(vec![cond_unix]).unwrap(),
                SeccompRule::new(vec![cond_inet]).unwrap(),
            ],
        );

        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // AF_UNIX: allowed
        let data = make_seccomp_data(41, AUDIT_ARCH_X86_64, [1, 0, 0, 0, 0, 0]);
        assert_eq!(bpf_eval(&prog, &data), RET_ALLOW);

        // AF_INET: allowed
        let data = make_seccomp_data(41, AUDIT_ARCH_X86_64, [2, 0, 0, 0, 0, 0]);
        assert_eq!(bpf_eval(&prog, &data), RET_ALLOW);

        // AF_NETLINK: killed
        let data = make_seccomp_data(41, AUDIT_ARCH_X86_64, [16, 0, 0, 0, 0, 0]);
        assert_eq!(bpf_eval(&prog, &data), RET_KILL);
    }

    #[test]
    fn test_equivalence_with_linear_scan() {
        // Build the full minimal filter rules and verify bitmap BPF produces
        // the same result as seccompiler's linear-scan BPF for many syscalls.
        use seccompiler::SeccompFilter;

        let rules = crate::security::SeccompManager::minimal_filter_for_test(true, &[]);
        let rules2 = rules.clone();

        // Linear-scan BPF (seccompiler)
        let linear_prog: BpfProgram = SeccompFilter::new(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap()
        .try_into()
        .unwrap();

        // Bitmap BPF (ours)
        let bitmap_prog = compile_bitmap_bpf(
            rules2,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // Test a wide range of syscall numbers with default args
        for nr in 0..500u32 {
            let data = make_seccomp_data(nr, AUDIT_ARCH_X86_64, [0; 6]);
            let linear_result = bpf_eval(&linear_prog, &data);
            let bitmap_result = bpf_eval(&bitmap_prog, &data);
            assert_eq!(
                linear_result, bitmap_result,
                "syscall {} (args=[0;6]): linear=0x{:08x}, bitmap=0x{:08x}",
                nr, linear_result, bitmap_result
            );
        }

        // Test arg-filtered syscalls with specific argument values
        // clone: namespace flags should be denied
        let clone_nr = libc::SYS_clone as u32;
        let data = make_seccomp_data(clone_nr, AUDIT_ARCH_X86_64, [0, 0, 0, 0, 0, 0]); // no ns flags
        assert_eq!(bpf_eval(&linear_prog, &data), bpf_eval(&bitmap_prog, &data));

        // ioctl: TCGETS should be allowed
        let ioctl_nr = libc::SYS_ioctl as u32;
        for req in [0x5401u64, 0x5413, 0x1234] {
            let data = make_seccomp_data(ioctl_nr, AUDIT_ARCH_X86_64, [0, req, 0, 0, 0, 0]);
            assert_eq!(
                bpf_eval(&linear_prog, &data),
                bpf_eval(&bitmap_prog, &data),
                "ioctl with req=0x{:x}",
                req
            );
        }

        // socket: AF_UNIX=1 allowed, AF_NETLINK=16 denied
        let socket_nr = libc::SYS_socket as u32;
        for domain in [1u64, 2, 10, 16] {
            let data = make_seccomp_data(socket_nr, AUDIT_ARCH_X86_64, [domain, 0, 0, 0, 0, 0]);
            assert_eq!(
                bpf_eval(&linear_prog, &data),
                bpf_eval(&bitmap_prog, &data),
                "socket with domain={}",
                domain
            );
        }
    }

    #[test]
    fn test_program_size_is_compact() {
        let rules = crate::security::SeccompManager::minimal_filter_for_test(true, &[]);
        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // Linear scan produces ~1119 instructions. Bitmap: ~246 (78% reduction).
        // The remaining instructions are mostly arg-check chains for ~7 syscalls.
        assert!(
            prog.len() < 400,
            "BPF program should be compact, got {} instructions",
            prog.len()
        );
        assert!(
            prog.len() < BPF_MAX_LEN,
            "BPF program must fit in {} instructions",
            BPF_MAX_LEN
        );
    }

    #[test]
    fn test_empty_rules() {
        let rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();
        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        // All syscalls should be killed
        let data = make_seccomp_data(0, AUDIT_ARCH_X86_64, [0; 6]);
        assert_eq!(bpf_eval(&prog, &data), RET_KILL);
    }

    #[test]
    fn test_high_syscall_numbers() {
        let mut rules: BTreeMap<i64, Vec<SeccompRule>> = BTreeMap::new();
        rules.insert(450, Vec::new()); // high number, still within bitmap range

        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        let data = make_seccomp_data(450, AUDIT_ARCH_X86_64, [0; 6]);
        assert_eq!(bpf_eval(&prog, &data), RET_ALLOW);

        let data = make_seccomp_data(451, AUDIT_ARCH_X86_64, [0; 6]);
        assert_eq!(bpf_eval(&prog, &data), RET_KILL);
    }

    #[test]
    fn test_all_jump_offsets_valid() {
        let rules = crate::security::SeccompManager::minimal_filter_for_test(true, &[]);
        let prog = compile_bitmap_bpf(
            rules,
            SeccompAction::KillProcess,
            SeccompAction::Allow,
            TargetArch::x86_64,
        )
        .unwrap();

        for (pc, insn) in prog.iter().enumerate() {
            match insn.code {
                0x05 => {
                    // BPF_JMP | BPF_JA
                    let target = pc + 1 + insn.k as usize;
                    assert!(
                        target < prog.len(),
                        "JA at pc={} jumps to {} (prog len={})",
                        pc,
                        target,
                        prog.len()
                    );
                }
                0x15 | 0x25 | 0x35 => {
                    // Conditional jumps
                    let target_t = pc + 1 + insn.jt as usize;
                    let target_f = pc + 1 + insn.jf as usize;
                    assert!(
                        target_t < prog.len(),
                        "JEQ/JGT/JGE jt at pc={} jumps to {} (prog len={})",
                        pc,
                        target_t,
                        prog.len()
                    );
                    assert!(
                        target_f < prog.len(),
                        "JEQ/JGT/JGE jf at pc={} jumps to {} (prog len={})",
                        pc,
                        target_f,
                        prog.len()
                    );
                }
                _ => {}
            }
        }
    }
}