From bbde23be1b45b6cdb29dead9d3214adac921a5d6 Mon Sep 17 00:00:00 2001 From: Josh Liebow-Feeser's Agent Date: Mon, 7 Sep 2026 16:18:56 +0000 Subject: [PATCH] Fix `TrailingSliceLayout` and remove `Split::via_unaligned` Represent slice-DST sizes as C + round_up(r + nE, A), encoding A and r in one nonzero word. Carry that four-word layout formula through composition, padding, metadata inference, casts, projections, allocation, and split overlap checks. This is a smaller state-model alternative to #3630. Exhaustive reduced-width model checking matches #3630's core formula transforms and representability behavior. Add regression coverage for compiler-accepted generic packed/aligned nesting, parsing, allocation, zeroing, projection, metadata plateaus, and dynamic padding. Add modular Kani proofs for the normal form, transforms, inverse validator, affine cast arithmetic, alignment, containment, the conservative dynamic-padding predicate, and both modern and MSRV unchecked-arithmetic implementations. The arithmetic proofs are conditional on the documented compiler-layout premise; regression and Miri coverage exercise the actual compiler layout and pointer paths. Refresh dynamic-size and dynamic-padding codegen expectations for the pinned nightly. Report both expected and actual assembly on failure, and join every benchmark worker so all mismatches are reported. Closes #3617 *Authored by an AI agent acting on Josh Liebow-Feeser's and Jack Wrenn's behalf.* gherrit-pr-id: Ger2cetjqc3naucboffw4r4hal423gaa5 Agent-authored-by: AI agent acting on joshlf's behalf --- zerocopy/AGENTS.md | 7 + .../ref_from_bytes_dynamic_padding.x86-64 | 31 +- .../ref_from_bytes_dynamic_padding.x86-64.mca | 70 +- .../ref_from_bytes_dynamic_size.x86-64 | 21 +- .../ref_from_bytes_dynamic_size.x86-64.mca | 58 +- .../ref_from_prefix_dynamic_padding.x86-64 | 14 +- ...ref_from_prefix_dynamic_padding.x86-64.mca | 48 +- .../ref_from_suffix_dynamic_padding.x86-64 | 29 +- ...ref_from_suffix_dynamic_padding.x86-64.mca | 68 +- ...t_via_runtime_check_dynamic_padding.x86-64 | 5 +- ...a_runtime_check_dynamic_padding.x86-64.mca | 30 +- .../try_ref_from_bytes_dynamic_padding.x86-64 | 34 +- ..._ref_from_bytes_dynamic_padding.x86-64.mca | 76 +- .../try_ref_from_bytes_dynamic_size.x86-64 | 31 +- ...try_ref_from_bytes_dynamic_size.x86-64.mca | 70 +- ...try_ref_from_prefix_dynamic_padding.x86-64 | 14 +- ...ref_from_prefix_dynamic_padding.x86-64.mca | 58 +- ...try_ref_from_suffix_dynamic_padding.x86-64 | 29 +- ...ref_from_suffix_dynamic_padding.x86-64.mca | 74 +- zerocopy/src/layout.rs | 3727 +++++++++++++++-- zerocopy/src/lib.rs | 195 +- zerocopy/src/macros.rs | 32 + zerocopy/src/pointer/inner.rs | 35 +- zerocopy/src/pointer/mod.rs | 28 +- zerocopy/src/split_at.rs | 710 +++- zerocopy/src/util/mod.rs | 30 +- zerocopy/tests/codegen.rs | 15 +- zerocopy/zerocopy-derive/src/derive/mod.rs | 32 +- .../split_at_repr_c_packed.expected.rs | 22 + .../zerocopy-derive/src/output_tests/mod.rs | 18 +- .../tests/ui/struct.msrv.stderr | 4 +- .../tests/ui/struct.nightly.stderr | 7 +- .../tests/ui/struct.stable.stderr | 7 +- 33 files changed, 4576 insertions(+), 1053 deletions(-) create mode 100644 zerocopy/zerocopy-derive/src/output_tests/expected/split_at_repr_c_packed.expected.rs diff --git a/zerocopy/AGENTS.md b/zerocopy/AGENTS.md index 2469b82bd6..07090a2e6e 100644 --- a/zerocopy/AGENTS.md +++ b/zerocopy/AGENTS.md @@ -41,6 +41,13 @@ These instructions apply to work in the `zerocopy/` subtree. In Markdown, indent wrapped bullet continuation lines by two spaces and put a blank line after each section heading. +- **Derived traits in tests:** For traits whose documentation requires + derive-only implementation, test fixture types **MUST** use Zerocopy's + derives. Do not bypass a derive rejection with a manual implementation; + instead, test the lower-level internal primitive or extend the derive. + Generated derive-output snapshots and crate-owned built-in implementations + are exempt. + ## Task-Specific Skills Use the applicable skills in addition to these standing rules. Skills compose; diff --git a/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64 b/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64 index e844a4608f..7f695f942e 100644 --- a/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64 +++ b/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64 @@ -1,22 +1,21 @@ bench_ref_from_bytes_dynamic_padding: - test dil, 3 - jne .LBB5_3 - movabs rax, 9223372036854775804 - and rax, rsi - cmp rax, 9 - jb .LBB5_3 - add rax, -9 + cmp rsi, 12 + setae al + mov ecx, esi + or ecx, edi + test cl, 3 + sete cl + and cl, al + cmp cl, 1 + je .LBB5_2 + xor edi, edi + mov rax, rdi + ret +.LBB5_2: + add rsi, -9 movabs rcx, -6148914691236517205 + mov rax, rsi mul rcx shr rdx - lea rax, [rdx + 2*rdx] - or rax, 3 - add rax, 9 - cmp rsi, rax - je .LBB5_4 -.LBB5_3: - xor edi, edi - mov rdx, rsi -.LBB5_4: mov rax, rdi ret diff --git a/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64.mca b/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64.mca index 423ed38ba2..5eb4fd1a11 100644 --- a/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/ref_from_bytes_dynamic_padding.x86-64.mca @@ -1,11 +1,11 @@ Iterations: 100 Instructions: 1900 -Total Cycles: 645 +Total Cycles: 641 Total uOps: 2000 Dispatch Width: 4 -uOps Per Cycle: 3.10 -IPC: 2.95 +uOps Per Cycle: 3.12 +IPC: 2.96 Block RThroughput: 5.0 @@ -18,23 +18,23 @@ Instruction Info: [6]: HasSideEffects (U) [1] [2] [3] [4] [5] [6] Instructions: - 1 1 0.33 test dil, 3 - 1 1 1.00 jne .LBB5_3 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rax, rsi - 1 1 0.33 cmp rax, 9 - 1 1 1.00 jb .LBB5_3 - 1 1 0.33 add rax, -9 + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setae al + 1 1 0.33 mov ecx, esi + 1 1 0.33 or ecx, edi + 1 1 0.33 test cl, 3 + 1 1 0.50 sete cl + 1 1 0.33 and cl, al + 1 1 0.33 cmp cl, 1 + 1 1 1.00 je .LBB5_2 + 1 0 0.25 xor edi, edi + 1 1 0.33 mov rax, rdi + 1 1 1.00 U ret + 1 1 0.33 add rsi, -9 1 1 0.33 movabs rcx, -6148914691236517205 + 1 1 0.33 mov rax, rsi 2 4 1.00 mul rcx 1 1 0.50 shr rdx - 1 1 0.50 lea rax, [rdx + 2*rdx] - 1 1 0.33 or rax, 3 - 1 1 0.33 add rax, 9 - 1 1 0.33 cmp rsi, rax - 1 1 1.00 je .LBB5_4 - 1 0 0.25 xor edi, edi - 1 1 0.33 mov rdx, rsi 1 1 0.33 mov rax, rdi 1 1 1.00 U ret @@ -52,26 +52,26 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 6.32 6.33 - 6.35 - - + - - 6.33 6.33 - 6.34 - - Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.64 0.35 - 0.01 - - test dil, 3 - - - - - - 1.00 - - jne .LBB5_3 - - - 0.34 0.65 - 0.01 - - movabs rax, 9223372036854775804 - - - 0.35 0.65 - - - - and rax, rsi - - - 0.33 0.34 - 0.33 - - cmp rax, 9 - - - - - - 1.00 - - jb .LBB5_3 - - - 0.35 - - 0.65 - - add rax, -9 - - - 0.97 0.01 - 0.02 - - movabs rcx, -6148914691236517205 - - - 1.00 1.00 - - - - mul rcx - - - 0.99 - - 0.01 - - shr rdx - - - 0.33 0.67 - - - - lea rax, [rdx + 2*rdx] - - - 0.34 0.66 - - - - or rax, 3 - - - 0.33 0.66 - 0.01 - - add rax, 9 - - - 0.01 0.99 - - - - cmp rsi, rax - - - - - - 1.00 - - je .LBB5_4 + - - 0.99 - - 0.01 - - cmp rsi, 12 + - - 1.00 - - - - - setae al + - - - 0.34 - 0.66 - - mov ecx, esi + - - - 0.99 - 0.01 - - or ecx, edi + - - - 1.00 - - - - test cl, 3 + - - 0.67 - - 0.33 - - sete cl + - - - 1.00 - - - - and cl, al + - - - 1.00 - - - - cmp cl, 1 + - - - - - 1.00 - - je .LBB5_2 - - - - - - - - xor edi, edi - - - 0.32 0.01 - 0.67 - - mov rdx, rsi - - - 0.02 0.34 - 0.64 - - mov rax, rdi + - - 0.66 - - 0.34 - - mov rax, rdi + - - - - - 1.00 - - ret + - - 0.35 0.01 - 0.64 - - add rsi, -9 + - - 0.65 0.33 - 0.02 - - movabs rcx, -6148914691236517205 + - - 0.01 0.33 - 0.66 - - mov rax, rsi + - - 1.00 1.00 - - - - mul rcx + - - 1.00 - - - - - shr rdx + - - - 0.33 - 0.67 - - mov rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64 b/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64 index cc905b76c0..43cadf3527 100644 --- a/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64 +++ b/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64 @@ -1,20 +1,11 @@ bench_ref_from_bytes_dynamic_size: mov rdx, rsi - cmp rsi, 4 - setb al - or al, dil - test al, 1 - je .LBB5_2 + mov ecx, edx + or ecx, edi xor eax, eax - ret -.LBB5_2: - lea rcx, [rdx - 4] - mov rsi, rcx - and rsi, -2 - add rsi, 4 - shr rcx - xor eax, eax - cmp rdx, rsi - cmove rdx, rcx + sub rdx, 4 + cmovb rdi, rax + shr rdx + test cl, 1 cmove rax, rdi ret diff --git a/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64.mca b/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64.mca index 68aea583e4..444f4fad4f 100644 --- a/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64.mca +++ b/zerocopy/benches/ref_from_bytes_dynamic_size.x86-64.mca @@ -1,12 +1,12 @@ Iterations: 100 -Instructions: 1800 -Total Cycles: 704 -Total uOps: 2000 +Instructions: 1000 +Total Cycles: 373 +Total uOps: 1200 Dispatch Width: 4 -uOps Per Cycle: 2.84 -IPC: 2.56 -Block RThroughput: 5.0 +uOps Per Cycle: 3.22 +IPC: 2.68 +Block RThroughput: 3.0 Instruction Info: @@ -19,21 +19,13 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.33 mov rdx, rsi - 1 1 0.33 cmp rsi, 4 - 1 1 0.50 setb al - 1 1 0.33 or al, dil - 1 1 0.33 test al, 1 - 1 1 1.00 je .LBB5_2 + 1 1 0.33 mov ecx, edx + 1 1 0.33 or ecx, edi 1 0 0.25 xor eax, eax - 1 1 1.00 U ret - 1 1 0.50 lea rcx, [rdx - 4] - 1 1 0.33 mov rsi, rcx - 1 1 0.33 and rsi, -2 - 1 1 0.33 add rsi, 4 - 1 1 0.50 shr rcx - 1 0 0.25 xor eax, eax - 1 1 0.33 cmp rdx, rsi - 2 2 0.67 cmove rdx, rcx + 1 1 0.33 sub rdx, 4 + 2 2 0.67 cmovb rdi, rax + 1 1 0.50 shr rdx + 1 1 0.33 test cl, 1 2 2 0.67 cmove rax, rdi 1 1 1.00 U ret @@ -51,25 +43,17 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 5.97 5.98 - 6.05 - - + - - 3.66 3.67 - 3.67 - - Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.97 0.01 - 0.02 - - mov rdx, rsi - - - 0.01 0.02 - 0.97 - - cmp rsi, 4 - - - 0.03 - - 0.97 - - setb al - - - 0.01 0.02 - 0.97 - - or al, dil - - - - 0.98 - 0.02 - - test al, 1 - - - - - - 1.00 - - je .LBB5_2 - - - - - - - - - xor eax, eax - - - - - - 1.00 - - ret - - - 0.98 0.02 - - - - lea rcx, [rdx - 4] - - - 0.01 0.99 - - - - mov rsi, rcx - - - - 0.98 - 0.02 - - and rsi, -2 - - - 0.98 0.01 - 0.01 - - add rsi, 4 - - - 0.99 - - 0.01 - - shr rcx + - - 0.32 0.01 - 0.67 - - mov rdx, rsi + - - 0.33 0.35 - 0.32 - - mov ecx, edx + - - 0.01 0.34 - 0.65 - - or ecx, edi - - - - - - - - xor eax, eax - - - 0.02 0.97 - 0.01 - - cmp rdx, rsi - - - 0.99 0.99 - 0.02 - - cmove rdx, rcx - - - 0.98 0.99 - 0.03 - - cmove rax, rdi + - - 0.67 0.33 - - - - sub rdx, 4 + - - 0.99 0.98 - 0.03 - - cmovb rdi, rax + - - 0.03 - - 0.97 - - shr rdx + - - 0.32 0.67 - 0.01 - - test cl, 1 + - - 0.99 0.99 - 0.02 - - cmove rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64 b/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64 index f380fee6c2..e7f7820a69 100644 --- a/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64 +++ b/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64 @@ -1,19 +1,19 @@ bench_ref_from_prefix_dynamic_padding: test dil, 3 + sete al + cmp rsi, 12 + setae cl + test al, cl jne .LBB5_2 - movabs rax, 9223372036854775804 - and rsi, rax - cmp rsi, 8 - ja .LBB5_3 -.LBB5_2: xor edx, edx test dil, 3 sete dl xor edi, edi mov rax, rdi ret -.LBB5_3: - add rsi, -9 +.LBB5_2: + or rsi, 3 + add rsi, -12 movabs rcx, -6148914691236517205 mov rax, rsi mul rcx diff --git a/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64.mca b/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64.mca index 499c58d94b..a3d68f37a8 100644 --- a/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/ref_from_prefix_dynamic_padding.x86-64.mca @@ -1,12 +1,12 @@ Iterations: 100 -Instructions: 1900 -Total Cycles: 607 -Total uOps: 2000 +Instructions: 2000 +Total Cycles: 640 +Total uOps: 2100 Dispatch Width: 4 -uOps Per Cycle: 3.29 +uOps Per Cycle: 3.28 IPC: 3.13 -Block RThroughput: 5.0 +Block RThroughput: 5.3 Instruction Info: @@ -19,18 +19,19 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.33 test dil, 3 + 1 1 0.50 sete al + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setae cl + 1 1 0.33 test al, cl 1 1 1.00 jne .LBB5_2 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rsi, rax - 1 1 0.33 cmp rsi, 8 - 1 1 1.00 ja .LBB5_3 1 0 0.25 xor edx, edx 1 1 0.33 test dil, 3 1 1 0.50 sete dl 1 0 0.25 xor edi, edi 1 1 0.33 mov rax, rdi 1 1 1.00 U ret - 1 1 0.33 add rsi, -9 + 1 1 0.33 or rsi, 3 + 1 1 0.33 add rsi, -12 1 1 0.33 movabs rcx, -6148914691236517205 1 1 0.33 mov rax, rsi 2 4 1.00 mul rcx @@ -52,26 +53,27 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 6.00 5.99 - 6.01 - - + - - 6.33 6.33 - 6.34 - - Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.01 0.98 - 0.01 - - test dil, 3 + - - 0.65 0.34 - 0.01 - - test dil, 3 + - - 0.99 - - 0.01 - - sete al + - - 0.01 0.33 - 0.66 - - cmp rsi, 12 + - - 0.34 - - 0.66 - - setae cl + - - 1.00 - - - - - test al, cl - - - - - 1.00 - - jne .LBB5_2 - - - 0.98 0.01 - 0.01 - - movabs rax, 9223372036854775804 - - - 0.01 0.99 - - - - and rsi, rax - - - 0.01 0.99 - - - - cmp rsi, 8 - - - - - - 1.00 - - ja .LBB5_3 - - - - - - - - xor edx, edx - - - 0.99 0.01 - - - - test dil, 3 - - - 0.99 - - 0.01 - - sete dl + - - - 0.99 - 0.01 - - test dil, 3 + - - - - - 1.00 - - sete dl - - - - - - - - xor edi, edi - - - - 0.01 - 0.99 - - mov rax, rdi + - - - 1.00 - - - - mov rax, rdi - - - - - 1.00 - - ret - - - 0.01 0.99 - - - - add rsi, -9 - - - - 1.00 - - - - movabs rcx, -6148914691236517205 - - - 1.00 - - - - - mov rax, rsi + - - 0.01 0.99 - - - - or rsi, 3 + - - 0.33 0.67 - - - - add rsi, -12 + - - 0.66 0.34 - - - - movabs rcx, -6148914691236517205 + - - 0.34 0.33 - 0.33 - - mov rax, rsi - - 1.00 1.00 - - - - mul rcx - - 1.00 - - - - - shr rdx - - - - 0.01 - 0.99 - - mov rax, rdi + - - - 0.34 - 0.66 - - mov rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64 b/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64 index 3e05f6023f..68f21fe801 100644 --- a/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64 +++ b/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64 @@ -1,23 +1,22 @@ bench_ref_from_suffix_dynamic_padding: lea eax, [rsi + rdi] test al, 3 - jne .LBB5_1 - movabs rax, 9223372036854775804 - and rax, rsi - cmp rax, 9 - jae .LBB5_3 -.LBB5_1: - xor eax, eax + setne al + cmp rsi, 12 + setb cl + or cl, al + je .LBB5_2 + xor edi, edi + mov rax, rdi ret -.LBB5_3: - add rax, -9 +.LBB5_2: + mov rax, rsi + or rax, 3 + add rax, -12 movabs rcx, -6148914691236517205 mul rcx shr rdx - lea rax, [rdx + 2*rdx] - sub rsi, rax - or rax, -4 - add rsi, rdi - add rax, rsi - add rax, -8 + and esi, 3 + add rdi, rsi + mov rax, rdi ret diff --git a/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64.mca b/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64.mca index 73599d5b6a..c31e098aed 100644 --- a/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/ref_from_suffix_dynamic_padding.x86-64.mca @@ -1,11 +1,11 @@ Iterations: 100 Instructions: 2000 -Total Cycles: 682 +Total Cycles: 673 Total uOps: 2100 Dispatch Width: 4 -uOps Per Cycle: 3.08 -IPC: 2.93 +uOps Per Cycle: 3.12 +IPC: 2.97 Block RThroughput: 5.3 @@ -20,23 +20,23 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.50 lea eax, [rsi + rdi] 1 1 0.33 test al, 3 - 1 1 1.00 jne .LBB5_1 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rax, rsi - 1 1 0.33 cmp rax, 9 - 1 1 1.00 jae .LBB5_3 - 1 0 0.25 xor eax, eax + 1 1 0.50 setne al + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setb cl + 1 1 0.33 or cl, al + 1 1 1.00 je .LBB5_2 + 1 0 0.25 xor edi, edi + 1 1 0.33 mov rax, rdi 1 1 1.00 U ret - 1 1 0.33 add rax, -9 + 1 1 0.33 mov rax, rsi + 1 1 0.33 or rax, 3 + 1 1 0.33 add rax, -12 1 1 0.33 movabs rcx, -6148914691236517205 2 4 1.00 mul rcx 1 1 0.50 shr rdx - 1 1 0.50 lea rax, [rdx + 2*rdx] - 1 1 0.33 sub rsi, rax - 1 1 0.33 or rax, -4 - 1 1 0.33 add rsi, rdi - 1 1 0.33 add rax, rsi - 1 1 0.33 add rax, -8 + 1 1 0.33 and esi, 3 + 1 1 0.33 add rdi, rsi + 1 1 0.33 mov rax, rdi 1 1 1.00 U ret @@ -53,27 +53,27 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 6.65 6.67 - 6.68 - - + - - 6.67 6.66 - 6.67 - - Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.90 0.10 - - - - lea eax, [rsi + rdi] - - - 0.93 - - 0.07 - - test al, 3 - - - - - - 1.00 - - jne .LBB5_1 - - - 0.51 0.47 - 0.02 - - movabs rax, 9223372036854775804 - - - - - - 1.00 - - and rax, rsi - - - - 0.09 - 0.91 - - cmp rax, 9 - - - - - - 1.00 - - jae .LBB5_3 - - - - - - - - - xor eax, eax + - - 0.35 0.65 - - - - lea eax, [rsi + rdi] + - - 0.01 0.99 - - - - test al, 3 + - - 1.00 - - - - - setne al + - - 0.97 - - 0.03 - - cmp rsi, 12 + - - 0.66 - - 0.34 - - setb cl + - - - 1.00 - - - - or cl, al + - - - - - 1.00 - - je .LBB5_2 + - - - - - - - - xor edi, edi + - - 0.01 0.01 - 0.98 - - mov rax, rdi - - - - - 1.00 - - ret - - - 0.43 0.47 - 0.10 - - add rax, -9 - - - 0.42 0.39 - 0.19 - - movabs rcx, -6148914691236517205 + - - 0.01 0.99 - - - - mov rax, rsi + - - - 0.34 - 0.66 - - or rax, 3 + - - - 1.00 - - - - add rax, -12 + - - 0.66 0.31 - 0.03 - - movabs rcx, -6148914691236517205 - - 1.00 1.00 - - - - mul rcx - - - 0.69 - - 0.31 - - shr rdx - - - 0.54 0.46 - - - - lea rax, [rdx + 2*rdx] - - - 0.07 0.91 - 0.02 - - sub rsi, rax - - - 0.91 0.05 - 0.04 - - or rax, -4 - - - 0.08 0.90 - 0.02 - - add rsi, rdi - - - 0.09 0.91 - - - - add rax, rsi - - - 0.08 0.92 - - - - add rax, -8 + - - 1.00 - - - - - shr rdx + - - 0.67 0.01 - 0.32 - - and esi, 3 + - - 0.32 0.01 - 0.67 - - add rdi, rsi + - - 0.01 0.35 - 0.64 - - mov rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64 b/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64 index 03684cb09f..0a8a4f501a 100644 --- a/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64 +++ b/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64 @@ -1,9 +1,8 @@ bench_split_via_runtime_check_dynamic_padding: mov rax, rdi mov rdx, qword ptr [rsi + 16] - mov ecx, edx - and ecx, 3 - cmp ecx, 1 + lea ecx, [rdx - 1] + test cl, 3 jne .LBB5_1 mov rcx, qword ptr [rsi] mov rsi, qword ptr [rsi + 8] diff --git a/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64.mca b/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64.mca index 5034ab0583..7140b06e88 100644 --- a/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/split_via_runtime_check_dynamic_padding.x86-64.mca @@ -1,11 +1,11 @@ Iterations: 100 -Instructions: 2000 +Instructions: 1900 Total Cycles: 708 -Total uOps: 2000 +Total uOps: 1900 Dispatch Width: 4 -uOps Per Cycle: 2.82 -IPC: 2.82 +uOps Per Cycle: 2.68 +IPC: 2.68 Block RThroughput: 5.0 @@ -20,9 +20,8 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.33 mov rax, rdi 1 5 0.50 * mov rdx, qword ptr [rsi + 16] - 1 1 0.33 mov ecx, edx - 1 1 0.33 and ecx, 3 - 1 1 0.33 cmp ecx, 1 + 1 1 0.50 lea ecx, [rdx - 1] + 1 1 0.33 test cl, 3 1 1 1.00 jne .LBB5_1 1 5 0.50 * mov rcx, qword ptr [rsi] 1 5 0.50 * mov rsi, qword ptr [rsi + 8] @@ -53,22 +52,21 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 3.00 3.02 5.00 4.98 4.00 4.00 + - - 2.99 3.00 5.00 4.01 4.00 4.00 Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - - 0.99 - 0.01 - - mov rax, rdi + - - 0.99 - - 0.01 - - mov rax, rdi - - - - - - - 1.00 mov rdx, qword ptr [rsi + 16] - - - 0.99 0.01 - - - - mov ecx, edx - - - 0.99 0.01 - - - - and ecx, 3 - - - 0.97 0.03 - - - - cmp ecx, 1 + - - - 1.00 - - - - lea ecx, [rdx - 1] + - - - 0.98 - 0.02 - - test cl, 3 - - - - - 1.00 - - jne .LBB5_1 - - - - - - 1.00 - mov rcx, qword ptr [rsi] - - - - - - 0.99 0.01 mov rsi, qword ptr [rsi + 8] - - - 0.01 0.99 - - - - lea rdi, [rdx + 2*rdx] - - - - 0.96 - 0.04 - - add rdi, rcx - - - 0.03 - - 0.97 - - add rdi, 9 - - - 0.01 0.03 - 0.96 - - sub rsi, rdx + - - 1.00 - - - - - lea rdi, [rdx + 2*rdx] + - - 0.01 0.01 - 0.98 - - add rdi, rcx + - - - 1.00 - - - - add rdi, 9 + - - 0.99 0.01 - - - - sub rsi, rdx - - - - 1.00 - 1.00 - mov qword ptr [rax + 8], rdx - - - - 1.00 - - 1.00 mov qword ptr [rax + 16], rdi - - - - 1.00 - 0.01 0.99 mov qword ptr [rax + 24], rsi diff --git a/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64 b/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64 index 217c5fc617..5584e90541 100644 --- a/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64 +++ b/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64 @@ -1,24 +1,22 @@ bench_try_ref_from_bytes_dynamic_padding: - test dil, 3 - jne .LBB5_4 - movabs rax, 9223372036854775804 - and rax, rsi - cmp rax, 9 - jb .LBB5_4 - add rax, -9 + cmp rsi, 12 + setae al + mov ecx, esi + or ecx, edi + test cl, 3 + sete cl + and cl, al + cmp cl, 1 + je .LBB5_2 + xor eax, eax + ret +.LBB5_2: + add rsi, -9 movabs rcx, -6148914691236517205 + mov rax, rsi mul rcx shr rdx - lea rax, [rdx + 2*rdx] - or rax, 3 - add rax, 9 - cmp rsi, rax - jne .LBB5_4 + xor eax, eax cmp word ptr [rdi], -16192 - je .LBB5_5 -.LBB5_4: - xor edi, edi - mov rdx, rsi -.LBB5_5: - mov rax, rdi + cmove rax, rdi ret diff --git a/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64.mca b/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64.mca index 95b993c7e0..ccd610c7ef 100644 --- a/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/try_ref_from_bytes_dynamic_padding.x86-64.mca @@ -1,10 +1,10 @@ Iterations: 100 -Instructions: 2100 -Total Cycles: 709 +Instructions: 2000 +Total Cycles: 675 Total uOps: 2300 Dispatch Width: 4 -uOps Per Cycle: 3.24 +uOps Per Cycle: 3.41 IPC: 2.96 Block RThroughput: 5.8 @@ -18,26 +18,25 @@ Instruction Info: [6]: HasSideEffects (U) [1] [2] [3] [4] [5] [6] Instructions: - 1 1 0.33 test dil, 3 - 1 1 1.00 jne .LBB5_4 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rax, rsi - 1 1 0.33 cmp rax, 9 - 1 1 1.00 jb .LBB5_4 - 1 1 0.33 add rax, -9 + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setae al + 1 1 0.33 mov ecx, esi + 1 1 0.33 or ecx, edi + 1 1 0.33 test cl, 3 + 1 1 0.50 sete cl + 1 1 0.33 and cl, al + 1 1 0.33 cmp cl, 1 + 1 1 1.00 je .LBB5_2 + 1 0 0.25 xor eax, eax + 1 1 1.00 U ret + 1 1 0.33 add rsi, -9 1 1 0.33 movabs rcx, -6148914691236517205 + 1 1 0.33 mov rax, rsi 2 4 1.00 mul rcx 1 1 0.50 shr rdx - 1 1 0.50 lea rax, [rdx + 2*rdx] - 1 1 0.33 or rax, 3 - 1 1 0.33 add rax, 9 - 1 1 0.33 cmp rsi, rax - 1 1 1.00 jne .LBB5_4 + 1 0 0.25 xor eax, eax 2 6 0.50 * cmp word ptr [rdi], -16192 - 1 1 1.00 je .LBB5_5 - 1 0 0.25 xor edi, edi - 1 1 0.33 mov rdx, rsi - 1 1 0.33 mov rax, rdi + 2 2 0.67 cmove rax, rdi 1 1 1.00 U ret @@ -54,28 +53,27 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 6.98 6.99 - 7.03 0.50 0.50 + - - 6.67 6.66 - 6.67 0.50 0.50 Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.48 0.51 - 0.01 - - test dil, 3 - - - - - - 1.00 - - jne .LBB5_4 - - - 0.51 0.49 - - - - movabs rax, 9223372036854775804 - - - 0.01 0.99 - - - - and rax, rsi - - - 0.51 0.49 - - - - cmp rax, 9 - - - - - - 1.00 - - jb .LBB5_4 - - - 0.98 - - 0.02 - - add rax, -9 - - - 0.98 0.02 - - - - movabs rcx, -6148914691236517205 + - - 0.33 0.66 - 0.01 - - cmp rsi, 12 + - - 1.00 - - - - - setae al + - - 0.66 0.34 - - - - mov ecx, esi + - - - 0.99 - 0.01 - - or ecx, edi + - - - 0.67 - 0.33 - - test cl, 3 + - - 0.67 - - 0.33 - - sete cl + - - 0.33 0.34 - 0.33 - - and cl, al + - - 0.33 - - 0.67 - - cmp cl, 1 + - - - - - 1.00 - - je .LBB5_2 + - - - - - - - - xor eax, eax + - - - - - 1.00 - - ret + - - 0.67 0.33 - - - - add rsi, -9 + - - 0.99 0.01 - - - - movabs rcx, -6148914691236517205 + - - - 0.99 - 0.01 - - mov rax, rsi - - 1.00 1.00 - - - - mul rcx - - - 0.99 - - 0.01 - - shr rdx - - - - 1.00 - - - - lea rax, [rdx + 2*rdx] - - - - 0.51 - 0.49 - - or rax, 3 - - - 0.01 0.49 - 0.50 - - add rax, 9 - - - - 0.02 - 0.98 - - cmp rsi, rax - - - - - - 1.00 - - jne .LBB5_4 - - - 0.51 0.49 - - 0.50 0.50 cmp word ptr [rdi], -16192 - - - - - - 1.00 - - je .LBB5_5 - - - - - - - - - xor edi, edi - - - 0.50 0.50 - - - - mov rdx, rsi - - - 0.50 0.48 - 0.02 - - mov rax, rdi + - - 0.01 - - 0.99 - - shr rdx + - - - - - - - - xor eax, eax + - - 0.67 0.33 - - 0.50 0.50 cmp word ptr [rdi], -16192 + - - 0.01 1.00 - 0.99 - - cmove rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64 b/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64 index cf67afd31c..39b12e14ee 100644 --- a/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64 +++ b/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64 @@ -1,22 +1,21 @@ bench_try_ref_from_bytes_dynamic_size: - mov rdx, rsi - mov rax, rdi cmp rsi, 4 - setb cl - or cl, al + setae al + mov ecx, esi + or ecx, edi test cl, 1 - jne .LBB5_4 - lea rcx, [rdx - 4] - mov rsi, rcx - and rsi, -2 - add rsi, 4 - cmp rdx, rsi - jne .LBB5_4 - cmp word ptr [rax], -16192 - jne .LBB5_4 - shr rcx - mov rdx, rcx + sete cl + and cl, al + cmp cl, 1 + je .LBB5_2 + xor eax, eax ret -.LBB5_4: +.LBB5_2: + lea rdx, [rsi - 4] + shr rdx + movzx ecx, word ptr [rdi] xor eax, eax + cmp ecx, 49344 + cmovne rdx, rsi + cmove rax, rdi ret diff --git a/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64.mca b/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64.mca index ecd7a18f6d..eecf3a1554 100644 --- a/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64.mca +++ b/zerocopy/benches/try_ref_from_bytes_dynamic_size.x86-64.mca @@ -1,10 +1,10 @@ Iterations: 100 -Instructions: 2000 -Total Cycles: 639 +Instructions: 1900 +Total Cycles: 607 Total uOps: 2100 Dispatch Width: 4 -uOps Per Cycle: 3.29 +uOps Per Cycle: 3.46 IPC: 3.13 Block RThroughput: 5.3 @@ -18,25 +18,24 @@ Instruction Info: [6]: HasSideEffects (U) [1] [2] [3] [4] [5] [6] Instructions: - 1 1 0.33 mov rdx, rsi - 1 1 0.33 mov rax, rdi 1 1 0.33 cmp rsi, 4 - 1 1 0.50 setb cl - 1 1 0.33 or cl, al + 1 1 0.50 setae al + 1 1 0.33 mov ecx, esi + 1 1 0.33 or ecx, edi 1 1 0.33 test cl, 1 - 1 1 1.00 jne .LBB5_4 - 1 1 0.50 lea rcx, [rdx - 4] - 1 1 0.33 mov rsi, rcx - 1 1 0.33 and rsi, -2 - 1 1 0.33 add rsi, 4 - 1 1 0.33 cmp rdx, rsi - 1 1 1.00 jne .LBB5_4 - 2 6 0.50 * cmp word ptr [rax], -16192 - 1 1 1.00 jne .LBB5_4 - 1 1 0.50 shr rcx - 1 1 0.33 mov rdx, rcx + 1 1 0.50 sete cl + 1 1 0.33 and cl, al + 1 1 0.33 cmp cl, 1 + 1 1 1.00 je .LBB5_2 + 1 0 0.25 xor eax, eax 1 1 1.00 U ret + 1 1 0.50 lea rdx, [rsi - 4] + 1 1 0.50 shr rdx + 1 5 0.50 * movzx ecx, word ptr [rdi] 1 0 0.25 xor eax, eax + 1 1 0.33 cmp ecx, 49344 + 2 2 0.67 cmovne rdx, rsi + 2 2 0.67 cmove rax, rdi 1 1 1.00 U ret @@ -53,27 +52,26 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 6.32 6.32 - 6.36 0.50 0.50 + - - 5.99 5.99 - 6.02 0.50 0.50 Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.33 0.66 - 0.01 - - mov rdx, rsi - - - 0.66 0.34 - - - - mov rax, rdi - - - 0.34 0.66 - - - - cmp rsi, 4 - - - 0.99 - - 0.01 - - setb cl - - - 0.01 0.99 - - - - or cl, al - - - - 1.00 - - - - test cl, 1 - - - - - - 1.00 - - jne .LBB5_4 - - - 0.66 0.34 - - - - lea rcx, [rdx - 4] - - - 0.33 0.66 - 0.01 - - mov rsi, rcx - - - 1.00 - - - - - and rsi, -2 - - - 0.66 0.34 - - - - add rsi, 4 - - - - 1.00 - - - - cmp rdx, rsi - - - - - - 1.00 - - jne .LBB5_4 - - - - - - 1.00 0.50 0.50 cmp word ptr [rax], -16192 - - - - - - 1.00 - - jne .LBB5_4 - - - 0.67 - - 0.33 - - shr rcx - - - 0.67 0.33 - - - - mov rdx, rcx + - - 0.02 0.97 - 0.01 - - cmp rsi, 4 + - - 1.00 - - - - - setae al + - - - 1.00 - - - - mov ecx, esi + - - 0.93 0.05 - 0.02 - - or ecx, edi + - - 0.94 0.06 - - - - test cl, 1 + - - 0.99 - - 0.01 - - sete cl + - - 0.98 0.02 - - - - and cl, al + - - 0.98 0.02 - - - - cmp cl, 1 + - - - - - 1.00 - - je .LBB5_2 + - - - - - - - - xor eax, eax - - - - - 1.00 - - ret + - - 0.01 0.99 - - - - lea rdx, [rsi - 4] + - - - - - 1.00 - - shr rdx + - - - - - - 0.50 0.50 movzx ecx, word ptr [rdi] - - - - - - - - xor eax, eax + - - 0.02 0.96 - 0.02 - - cmp ecx, 49344 + - - 0.05 0.97 - 0.98 - - cmovne rdx, rsi + - - 0.07 0.95 - 0.98 - - cmove rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64 b/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64 index aa4ddaaca7..ff64bb6595 100644 --- a/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64 +++ b/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64 @@ -1,18 +1,18 @@ bench_try_ref_from_prefix_dynamic_padding: test dil, 3 + sete al + cmp rsi, 12 + setae cl + test al, cl jne .LBB5_2 - movabs rax, 9223372036854775804 - and rsi, rax - cmp rsi, 8 - ja .LBB5_3 -.LBB5_2: xor edx, edx test dil, 3 sete dl xor eax, eax ret -.LBB5_3: - add rsi, -9 +.LBB5_2: + or rsi, 3 + add rsi, -12 movabs rcx, -6148914691236517205 mov rax, rsi mul rcx diff --git a/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64.mca b/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64.mca index f595d499cf..29225404a2 100644 --- a/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/try_ref_from_prefix_dynamic_padding.x86-64.mca @@ -1,12 +1,12 @@ Iterations: 100 -Instructions: 2500 -Total Cycles: 808 -Total uOps: 2800 +Instructions: 2600 +Total Cycles: 843 +Total uOps: 2900 Dispatch Width: 4 -uOps Per Cycle: 3.47 -IPC: 3.09 -Block RThroughput: 7.0 +uOps Per Cycle: 3.44 +IPC: 3.08 +Block RThroughput: 7.3 Instruction Info: @@ -19,17 +19,18 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.33 test dil, 3 + 1 1 0.50 sete al + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setae cl + 1 1 0.33 test al, cl 1 1 1.00 jne .LBB5_2 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rsi, rax - 1 1 0.33 cmp rsi, 8 - 1 1 1.00 ja .LBB5_3 1 0 0.25 xor edx, edx 1 1 0.33 test dil, 3 1 1 0.50 sete dl 1 0 0.25 xor eax, eax 1 1 1.00 U ret - 1 1 0.33 add rsi, -9 + 1 1 0.33 or rsi, 3 + 1 1 0.33 add rsi, -12 1 1 0.33 movabs rcx, -6148914691236517205 1 1 0.33 mov rax, rsi 2 4 1.00 mul rcx @@ -58,32 +59,33 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 8.00 8.00 - 8.00 0.50 0.50 + - - 8.33 8.33 - 8.34 0.50 0.50 Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.97 0.02 - 0.01 - - test dil, 3 + - - 0.01 0.98 - 0.01 - - test dil, 3 + - - 0.98 - - 0.02 - - sete al + - - - 1.00 - - - - cmp rsi, 12 + - - - - - 1.00 - - setae cl + - - 0.01 0.03 - 0.96 - - test al, cl - - - - - 1.00 - - jne .LBB5_2 - - - 0.99 0.01 - - - - movabs rax, 9223372036854775804 - - - 0.02 0.98 - - - - and rsi, rax - - - 0.03 0.97 - - - - cmp rsi, 8 - - - - - - 1.00 - - ja .LBB5_3 - - - - - - - - xor edx, edx - - - 0.98 0.02 - - - - test dil, 3 + - - - 1.00 - - - - test dil, 3 - - 0.99 - - 0.01 - - sete dl - - - - - - - - xor eax, eax - - - - - 1.00 - - ret - - - 0.98 0.02 - - - - add rsi, -9 - - - 0.02 0.98 - - - - movabs rcx, -6148914691236517205 - - - 0.98 0.02 - - - - mov rax, rsi + - - - 1.00 - - - - or rsi, 3 + - - 0.01 0.99 - - - - add rsi, -12 + - - 0.01 0.99 - - - - movabs rcx, -6148914691236517205 + - - 0.34 0.66 - - - - mov rax, rsi - - 1.00 1.00 - - - - mul rcx - - - - 0.01 - 0.99 - - mov rax, rdx - - - 0.01 - - 0.99 - - shr rax + - - 0.99 - - 0.01 - - mov rax, rdx + - - 1.00 - - - - - shr rax - - - - - - 0.50 0.50 movzx ecx, word ptr [rdi] - - - - 0.99 - 0.01 - - cmp cx, -16192 - - - 0.99 0.01 - - - - mov edx, 2 - - - 0.01 1.00 - 0.99 - - cmove rdx, rax + - - 0.34 0.32 - 0.34 - - cmp cx, -16192 + - - 0.33 0.01 - 0.66 - - mov edx, 2 + - - 0.99 0.01 - 1.00 - - cmove rdx, rax - - - - - - - - xor eax, eax - - - 0.02 0.98 - - - - cmp ecx, 49344 - - - 0.01 0.99 - 1.00 - - cmove rax, rdi + - - 0.33 0.33 - 0.34 - - cmp ecx, 49344 + - - 1.00 0.01 - 0.99 - - cmove rax, rdi - - - - - 1.00 - - ret diff --git a/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64 b/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64 index b3e9244428..cee9a31790 100644 --- a/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64 +++ b/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64 @@ -1,26 +1,23 @@ bench_try_ref_from_suffix_dynamic_padding: lea eax, [rsi + rdi] test al, 3 - jne .LBB5_1 - movabs rax, 9223372036854775804 - and rax, rsi - cmp rax, 9 - jae .LBB5_3 -.LBB5_1: + setne al + cmp rsi, 12 + setb cl + or cl, al + je .LBB5_2 xor eax, eax ret -.LBB5_3: - add rax, -9 +.LBB5_2: + mov rax, rsi + or rax, 3 + add rax, -12 movabs rcx, -6148914691236517205 mul rcx shr rdx - lea rcx, [rdx + 2*rdx] - sub rsi, rcx - or rcx, -4 - add rsi, rdi - lea rdi, [rcx + rsi] - add rdi, -8 + and esi, 3 + lea rcx, [rdi + rsi] xor eax, eax - cmp word ptr [rcx + rsi - 8], -16192 - cmove rax, rdi + cmp word ptr [rdi + rsi], -16192 + cmove rax, rcx ret diff --git a/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64.mca b/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64.mca index d56ae56d85..92df4484b8 100644 --- a/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64.mca +++ b/zerocopy/benches/try_ref_from_suffix_dynamic_padding.x86-64.mca @@ -1,12 +1,12 @@ Iterations: 100 -Instructions: 2300 -Total Cycles: 791 -Total uOps: 2600 +Instructions: 2100 +Total Cycles: 710 +Total uOps: 2400 Dispatch Width: 4 -uOps Per Cycle: 3.29 -IPC: 2.91 -Block RThroughput: 6.5 +uOps Per Cycle: 3.38 +IPC: 2.96 +Block RThroughput: 6.0 Instruction Info: @@ -20,26 +20,24 @@ Instruction Info: [1] [2] [3] [4] [5] [6] Instructions: 1 1 0.50 lea eax, [rsi + rdi] 1 1 0.33 test al, 3 - 1 1 1.00 jne .LBB5_1 - 1 1 0.33 movabs rax, 9223372036854775804 - 1 1 0.33 and rax, rsi - 1 1 0.33 cmp rax, 9 - 1 1 1.00 jae .LBB5_3 + 1 1 0.50 setne al + 1 1 0.33 cmp rsi, 12 + 1 1 0.50 setb cl + 1 1 0.33 or cl, al + 1 1 1.00 je .LBB5_2 1 0 0.25 xor eax, eax 1 1 1.00 U ret - 1 1 0.33 add rax, -9 + 1 1 0.33 mov rax, rsi + 1 1 0.33 or rax, 3 + 1 1 0.33 add rax, -12 1 1 0.33 movabs rcx, -6148914691236517205 2 4 1.00 mul rcx 1 1 0.50 shr rdx - 1 1 0.50 lea rcx, [rdx + 2*rdx] - 1 1 0.33 sub rsi, rcx - 1 1 0.33 or rcx, -4 - 1 1 0.33 add rsi, rdi - 1 1 0.50 lea rdi, [rcx + rsi] - 1 1 0.33 add rdi, -8 + 1 1 0.33 and esi, 3 + 1 1 0.50 lea rcx, [rdi + rsi] 1 0 0.25 xor eax, eax - 2 6 0.50 * cmp word ptr [rcx + rsi - 8], -16192 - 2 2 0.67 cmove rax, rdi + 2 6 0.50 * cmp word ptr [rdi + rsi], -16192 + 2 2 0.67 cmove rax, rcx 1 1 1.00 U ret @@ -56,30 +54,28 @@ Resources: Resource pressure per iteration: [0] [1] [2] [3] [4] [5] [6.0] [6.1] - - - 7.70 7.58 - 7.72 0.50 0.50 + - - 7.00 7.00 - 7.00 0.50 0.50 Resource pressure by instruction: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: - - - 0.26 0.74 - - - - lea eax, [rsi + rdi] - - - 0.19 0.28 - 0.53 - - test al, 3 - - - - - - 1.00 - - jne .LBB5_1 - - - 0.93 0.06 - 0.01 - - movabs rax, 9223372036854775804 - - - 0.81 0.14 - 0.05 - - and rax, rsi - - - 0.55 0.43 - 0.02 - - cmp rax, 9 - - - - - - 1.00 - - jae .LBB5_3 + - - 0.99 0.01 - - - - lea eax, [rsi + rdi] + - - 0.01 0.99 - - - - test al, 3 + - - 0.98 - - 0.02 - - setne al + - - 0.99 - - 0.01 - - cmp rsi, 12 + - - 0.02 - - 0.98 - - setb cl + - - - 0.03 - 0.97 - - or cl, al + - - - - - 1.00 - - je .LBB5_2 - - - - - - - - xor eax, eax - - - - - 1.00 - - ret - - - 0.42 0.56 - 0.02 - - add rax, -9 - - - 0.67 0.30 - 0.03 - - movabs rcx, -6148914691236517205 + - - - 1.00 - - - - mov rax, rsi + - - 0.03 0.97 - - - - or rax, 3 + - - - 1.00 - - - - add rax, -12 + - - 0.99 - - 0.01 - - movabs rcx, -6148914691236517205 - - 1.00 1.00 - - - - mul rcx - - - 0.71 - - 0.29 - - shr rdx - - - 0.32 0.68 - - - - lea rcx, [rdx + 2*rdx] - - - 0.57 0.04 - 0.39 - - sub rsi, rcx - - - 0.28 0.67 - 0.05 - - or rcx, -4 - - - 0.29 0.29 - 0.42 - - add rsi, rdi - - - 0.02 0.98 - - - - lea rdi, [rcx + rsi] - - - 0.02 0.41 - 0.57 - - add rdi, -8 + - - 0.01 - - 0.99 - - shr rdx + - - 1.00 - - - - - and esi, 3 + - - 0.97 0.03 - - - - lea rcx, [rdi + rsi] - - - - - - - - xor eax, eax - - - 0.57 0.01 - 0.42 0.50 0.50 cmp word ptr [rcx + rsi - 8], -16192 - - - 0.09 0.99 - 0.92 - - cmove rax, rdi + - - - 0.97 - 0.03 0.50 0.50 cmp word ptr [rdi + rsi], -16192 + - - 0.01 1.00 - 0.99 - - cmove rax, rcx - - - - - 1.00 - - ret diff --git a/zerocopy/src/layout.rs b/zerocopy/src/layout.rs index d58786d637..35af28f517 100644 --- a/zerocopy/src/layout.rs +++ b/zerocopy/src/layout.rs @@ -30,6 +30,7 @@ const POINTER_WIDTH_BITS: usize = mem::size_of::() * 8; #[allow(missing_debug_implementations, missing_copy_implementations)] #[cfg_attr(any(kani, test), derive(Debug, PartialEq, Eq))] #[derive(Copy, Clone)] +#[cfg_attr(kani, derive(kani::Arbitrary))] pub struct DstLayout { pub(crate) align: NonZeroUsize, pub(crate) size_info: SizeInfo, @@ -42,29 +43,680 @@ pub struct DstLayout { #[cfg_attr(any(kani, test), derive(Debug, PartialEq, Eq))] #[derive(Copy, Clone)] +#[cfg_attr(kani, derive(kani::Arbitrary))] pub(crate) enum SizeInfo { Sized { size: usize }, SliceDst(TrailingSliceLayout), } +/// The alignment and phase of a rounded size calculation. +/// +/// If `encoded` is the stored non-zero integer, then its highest set bit is a +/// power-of-two alignment `align`, and its remaining lower bits are a `phase` +/// in `0..align`. Thus every non-zero bit pattern is a valid encoding, and: +/// +/// ```text +/// encoded = align + phase +/// ``` +#[cfg_attr(any(kani, test), derive(Debug, PartialEq, Eq))] +#[repr(transparent)] +#[derive(Copy, Clone)] +#[cfg_attr(kani, derive(kani::Arbitrary))] +pub(crate) struct RoundingAlignAndPhase(NonZeroUsize); + +impl RoundingAlignAndPhase { + /// Encodes `align` and `phase`. + /// + /// # Panics + /// + /// Panics if `align` is not a power of two and `phase` is not less than + /// `align`. + #[inline(always)] + #[cfg_attr(kani, kani::requires(align.is_power_of_two() && phase < align.get()))] + #[cfg_attr(kani, kani::ensures(|result| result.0.get() == align.get() + phase))] + pub(crate) const fn new(align: NonZeroUsize, phase: usize) -> Self { + #[cfg(kani)] + #[kani::proof_for_contract(RoundingAlignAndPhase::new)] + #[kani::solver(kissat)] + fn proof() { + RoundingAlignAndPhase::new(kani::any(), kani::any()); + } + + const_assert!(align.get().is_power_of_two()); + const_assert!(phase < align.get()); + + // Since `align` is a power of two and `phase < align`, their set bits + // are disjoint. The result is therefore non-zero and losslessly stores + // both components. + let encoded = align.get() | phase; + match NonZeroUsize::new(encoded) { + Some(encoded) => Self(encoded), + None => const_unreachable!(), + } + } + + /// Decodes the alignment and phase. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&(align, phase)| { + align.is_power_of_two() && phase < align.get() + && align.get().checked_add(phase) == Some(self.0.get()) + }))] + pub(crate) const fn components(self) -> (NonZeroUsize, usize) { + #[cfg(kani)] + #[kani::proof_for_contract(RoundingAlignAndPhase::components)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().components(); + } + // `leading_zeros <= POINTER_WIDTH_BITS - 1` because the encoded value + // is non-zero. Converting `leading_zeros` to `usize` cannot truncate + // because it is at most the number of bits in a `usize`. + #[allow( + clippy::arithmetic_side_effects, + clippy::as_conversions, + clippy::needless_nonzero_get + )] + let shift = (POINTER_WIDTH_BITS - 1) - (self.0.get().leading_zeros() as usize); + #[allow(clippy::arithmetic_side_effects)] + let align = 1usize << shift; + let align = match NonZeroUsize::new(align) { + Some(align) => align, + None => const_unreachable!(), + }; + // `align` is the highest set bit, so XOR removes exactly that bit and + // leaves a value in `0..align`. + let phase = self.0.get() ^ align.get(); + (align, phase) + } + + /// Decodes the alignment, which is the highest set bit of `self`. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|align| { + align.is_power_of_two() && align.get() <= self.0.get() + && self.0.get() - align.get() < align.get() + }))] + pub(crate) const fn align(self) -> NonZeroUsize { + #[cfg(kani)] + #[kani::proof_for_contract(RoundingAlignAndPhase::align)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().align(); + } + + self.components().0 + } +} + #[cfg_attr(any(kani, test), derive(Debug, PartialEq, Eq))] #[derive(Copy, Clone)] +#[cfg_attr(kani, derive(kani::Arbitrary))] pub(crate) struct TrailingSliceLayout { - // The offset of the first byte of the trailing slice field. Note that this - // is NOT the same as the minimum size of the type. For example, consider - // the following type: - // - // struct Foo { - // a: u16, - // b: u8, - // c: [u8], - // } - // - // In `Foo`, `c` is at byte offset 3. When `c.len() == 0`, `c` is followed - // by a padding byte. + /// The offset of the first byte of the trailing slice field. Note that this + /// is NOT the same as the minimum size of the type. For example, consider + /// the following type: + /// + /// ```ignore + /// #[repr(C, align(2))] + /// struct Foo { + /// header: u16, + /// tag: u8, + /// tail: [u8], + /// } + /// ``` + /// + /// In `Foo`, `tail` is at byte offset 3. When `tail.len() == 0`, `tail` is + /// followed by a padding byte. pub(crate) offset: usize, - // The size of the element type of the trailing slice field. + /// The size of the element type of the trailing slice field. pub(crate) elem_size: E, + /// The trailing slice's byte offset plus fixed trailing padding, minus the + /// rounding phase. + /// + /// Fixed trailing padding excludes padding added by the final alignment + /// rounding. The rounding alignment and phase are stored in + /// `size_rounding_align_and_phase`. This value is independent of the slice + /// length and can be smaller than, equal to, or larger than `offset`. + /// + /// For a completed layout, this value is a multiple of `DstLayout::align`. + /// The minimum object size equals this value when the phase is zero; + /// otherwise, it exceeds this value by one rounding-alignment unit. + pub(crate) size_base: usize, + /// The alignment and byte phase of the rounding applied to the trailing + /// slice's contribution to object size. + /// + /// The alignment `size_align` is a nonzero power of two. The phase + /// `size_phase` is the fixed byte contribution inside that rounding, in + /// `0..size_align`. Both are independent of the trailing slice's length. + /// For `elems` trailing elements, the enclosing object's size is: + /// + /// ```text + /// size_base + round_up(size_phase + elems * elem_size, size_align) + /// ``` + /// + /// For a completed layout, `size_align` is at least `DstLayout::align`. + /// Packing can reduce the enclosing type's alignment and the alignment + /// used to place its trailing field without removing padding required by + /// a nested DST's own layout. For example, a `repr(packed(2))` wrapper + /// around a DST with alignment 4 can have an object alignment of 2 and a + /// rounding alignment of 4. + /// + /// The rounding parameters account for padding preserved within nested + /// fields. Their rounding origin can differ from the physical start of + /// the trailing slice, so the phase need not equal `offset % size_align`. + pub(crate) size_rounding_align_and_phase: RoundingAlignAndPhase, +} + +impl TrailingSliceLayout { + /// Produces `self.size_base` rounded down to a multiple of `size_align`, + /// plus `size_phase`, where both components come from + /// `self.size_rounding_align_and_phase`. + /// + /// This value need not equal `self.offset`, the trailing slice's byte + /// offset. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&offset| { + let (align, phase) = self.size_rounding_align_and_phase.components(); + offset == util::round_down_to_next_multiple_of_alignment(self.size_base, align) + phase + }))] + const fn size_offset(&self) -> usize { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::::size_offset)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().size_offset(); + } + + let (size_align, size_phase) = self.size_rounding_align_and_phase.components(); + let aligned_base = + util::round_down_to_next_multiple_of_alignment(self.size_base, size_align); + // The aligned base and phase have disjoint set bits. + aligned_base | size_phase + } + + /// Produces the largest trailing-slice byte capacity whose complete, padded + /// object fits in `available_bytes`, or `None` if the object does not fit + /// even with an empty trailing slice. + /// + /// This capacity is independent of `elem_size`: it need not represent a + /// whole number of elements, even for a valid Rust layout. For example, a + /// slice of `[u8; 3]` elements has three-byte elements and alignment 1. + /// With `available_bytes = 4`, this method produces `Some(4)`, although + /// only one whole element, occupying three bytes, fits. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&result| { + let (align, phase) = self.size_rounding_align_and_phase.components(); + result == available_bytes.checked_sub(self.size_base) + .map(|bytes| util::round_down_to_next_multiple_of_alignment(bytes, align)) + .and_then(|rounded| rounded.checked_sub(phase)) + }))] + const fn max_trailing_bytes(&self, available_bytes: usize) -> Option { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::::max_trailing_bytes)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().max_trailing_bytes(kani::any()); + } + + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::::max_trailing_bytes)] + #[kani::solver(kissat)] + fn proof_nonzero() { + kani::any::>().max_trailing_bytes(kani::any()); + } + + let (size_align, size_phase) = self.size_rounding_align_and_phase.components(); + + // Interpret the normalized size formula over the nonnegative integers: + // + // object_size(trailing_bytes) + // = self.size_base + // + round_up(size_phase + trailing_bytes, size_align). + // + // We seek the largest byte count with + // `object_size(trailing_bytes) <= available_bytes`. + // + // At zero trailing bytes, the input to rounding is just `size_phase`. + // Since `0 <= size_phase < size_align`, the least alignment multiple at + // or above it is zero for a zero phase, and `size_align` otherwise. + // Thus `rounded_phase = round_up(size_phase, size_align)` is exactly + // the rounded contribution of an empty tail. + let rounded_phase = if size_phase == 0 { 0 } else { size_align.get() }; + + // Adding the fixed base gives `minimum_size = object_size(0)`. The size + // formula is nondecreasing in the trailing byte count, so this is the + // minimum object size. If this sum exceeds `usize::MAX`, it also + // exceeds `available_bytes`, and no tail can fit. + let minimum_size = match self.size_base.checked_add(rounded_phase) { + Some(size) => size, + None => return None, + }; + + // Subtracting the minimum size gives the budget for increasing the + // rounded contribution. Failure means even the minimum size exceeds the + // available space. Success establishes `minimum_size + extra_bytes = + // available_bytes`. + let extra_bytes = match available_bytes.checked_sub(minimum_size) { + Some(bytes) => bytes, + None => return None, + }; + + // The rounded contribution can increase only by multiples of + // `size_align`. Its largest permitted increase is therefore + // `aligned_bytes = round_down(extra_bytes, size_align)`, with: + // + // aligned_bytes <= extra_bytes < aligned_bytes + size_align. + // + // Hence `rounded_phase + aligned_bytes` is the largest alignment + // multiple that fits after the base: adding the base gives at most + // `minimum_size + extra_bytes = available_bytes`, while the next + // alignment multiple would exceed that budget. + let aligned_bytes = util::round_down_to_next_multiple_of_alignment(extra_bytes, size_align); + + // Rounding an empty tail adds `rounded_phase - size_phase` bytes. This + // is `initial_padding`: the tail can use these bytes without increasing + // the rounded contribution. The two cases for `rounded_phase` give `0 + // <= initial_padding < size_align`, so the subtraction is + // representable. + #[allow(clippy::arithmetic_side_effects)] + let initial_padding = rounded_phase - size_phase; + + // For any nonnegative candidate byte count, the aligned limit above + // gives the following equivalences: + // + // object_size(candidate_bytes) <= available_bytes + // iff round_up(size_phase + candidate_bytes, size_align) + // <= rounded_phase + aligned_bytes + // iff size_phase + candidate_bytes + // <= rounded_phase + aligned_bytes + // iff candidate_bytes <= aligned_bytes + initial_padding. + // + // Removing `round_up` is valid because its upper bound is an alignment + // multiple. Thus `aligned_bytes + initial_padding` is exactly the + // largest tail that fits. Since `aligned_bytes` is a multiple of the + // power-of-two `size_align` and `initial_padding < size_align`, their + // set bits are disjoint. Bitwise OR therefore equals their sum, with no + // carry or overflow. + let trailing_bytes = aligned_bytes | initial_padding; + Some(trailing_bytes) + } +} + +impl TrailingSliceLayout { + /// Returns the difference between the complete object size and the + /// trailing-slice end for `elems` elements, modulo `usize::MAX + 1`. + /// + /// If the object size fits in `usize` and contains the trailing slice, + /// this is the exact trailing padding. Other inputs need not describe a + /// Rust type; their result is still the modular difference. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&padding| { + let (align, phase) = self.size_rounding_align_and_phase.components(); + let trailing_bytes = elems.wrapping_mul(self.elem_size); + let mask = align.get() - 1; + let rounded = phase.wrapping_add(trailing_bytes).wrapping_add(mask) & !mask; + let object_size = self.size_base.wrapping_add(rounded); + padding == object_size.wrapping_sub(self.offset.wrapping_add(trailing_bytes)) + }))] + pub(crate) const fn padding_for_elems(self, elems: usize) -> usize { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::padding_for_elems)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().padding_for_elems(kani::any()); + } + + let (size_align, size_phase) = self.size_rounding_align_and_phase.components(); + + // Since `size_align` is a nonzero power of two, subtracting one gives + // a mask selecting precisely the remainder modulo `size_align`. + #[allow(clippy::arithmetic_side_effects)] + let size_mask = size_align.get() - 1; + + // Removing whole alignment multiples from each element's size + // preserves the trailing-byte count modulo `size_align`. If elements + // occupy whole alignment units, this is zero, making the padding's + // independence from the element count explicit to the optimizer. + let elem_remainder = self.elem_size & size_mask; + + // The product is congruent to `elems * self.elem_size` modulo + // `size_align`. Wrapping preserves that congruence because every + // power-of-two `size_align` divides `usize::MAX + 1`. Masking then + // gives the exact trailing-byte remainder in `0..size_align`. + let trailing_remainder = elems.wrapping_mul(elem_remainder) & size_mask; + + // Adding `size_phase` gives the same remainder as the mathematical + // `size_phase + elems * self.elem_size`. Reducing before adding the + // phase also exposes this periodicity to the optimizer. + let rounding_input = size_phase.wrapping_add(trailing_remainder); + + // `rounding_padding` is the unique value in `0..size_align` that + // rounds `rounding_input` up to an alignment multiple. Since only + // the remainder determines this value, it is also the padding added + // by `round_up(size_phase + elems * self.elem_size, size_align)`. + let rounding_padding = util::padding_needed_for(rounding_input, size_align); + + // Write `trailing_bytes = elems * self.elem_size` over mathematical + // integers. Subtracting the trailing-slice end from the complete + // object size cancels `trailing_bytes`: + // + // size_base + round_up(size_phase + trailing_bytes, size_align) + // - (offset + trailing_bytes) + // = size_base + size_phase - offset + rounding_padding. + // + // Wrapping computes this difference modulo `usize::MAX + 1`, even + // if an intermediate value or the difference is unrepresentable. + // If the object size fits in `usize` and contains the trailing slice, + // the difference is in `0..=usize::MAX`; its residue is then the + // exact number of padding bytes after the trailing slice. + self.size_base + .wrapping_add(size_phase) + .wrapping_sub(self.offset) + .wrapping_add(rounding_padding) + } + + /// Computes the size described by this layout for `elems` trailing slice + /// elements. + /// + /// Returns `None` if the element-byte count or complete rounded size + /// cannot be represented as a `usize`. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&size| { + size == elems.checked_mul(self.elem_size) + .and_then(|bytes| proofs::slice_dst_size_for_trailing_bytes(self, bytes)) + }))] + pub(crate) const fn size_for_elems(self, elems: usize) -> Option { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::size_for_elems)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().size_for_elems(kani::any()); + } + + // Let `size_align` and `size_phase` denote the components stored in + // `size_rounding_align_and_phase`. The normalized layout formula, + // evaluated over the nonnegative integers, is: + // + // object_size(trailing_bytes) + // = self.size_base + // + round_up(size_phase + trailing_bytes, size_align) + // + // Here, `round_up` gives the least multiple of `size_align` greater + // than or equal to its input. Thus `object_size` is nondecreasing. + // By the capacity helper's contract, `max_trailing_bytes` is the + // largest byte count for which `object_size` fits in a `usize`. + // Monotonicity then gives the equivalence: + // + // object_size(trailing_bytes) <= usize::MAX + // iff trailing_bytes <= max_trailing_bytes. + // + // If the helper returns `None`, even `object_size(0)` exceeds the + // limit, so no element count can produce a representable size. + let max_trailing_bytes = match self.max_trailing_bytes(usize::MAX) { + Some(bytes) => bytes, + None => return None, + }; + + // Each of the `elems` elements contributes `self.elem_size` bytes, + // so substitution into the formula requires their product. Checked + // multiplication yields that exact product, including zero for + // zero-sized elements. If it overflows, the complete size cannot fit + // either: `object_size(trailing_bytes) >= trailing_bytes` because + // the base, phase, and rounding increment are all nonnegative. + let trailing_size = match self.elem_size.checked_mul(elems) { + Some(bytes) => bytes, + None => return None, + }; + + // The equivalence above rejects exactly the remaining cases whose + // complete size exceeds `usize::MAX`. After this check we have + // `object_size(trailing_size) <= usize::MAX`. + if trailing_size > max_trailing_bytes { + return None; + } + + // The trailing slice starts at `self.offset` and occupies + // `trailing_size` bytes. Thus `trailing_end` is its end offset modulo + // `usize::MAX + 1`. This sum need not fit: a raw layout need not + // describe a Rust type, so its trailing slice need not lie within + // the complete object size bounded above. + let trailing_end = self.offset.wrapping_add(trailing_size); + + // `padding_for_elems` returns `object_size(trailing_size)` minus the + // trailing-slice end, modulo `usize::MAX + 1`. Adding the end back + // therefore gives `object_size(trailing_size)` modulo the same + // modulus. The bound above establishes that this size fits in + // `usize`, so its residue is the exact size specified by the layout. + let size = trailing_end.wrapping_add(self.padding_for_elems(elems)); + Some(size) + } + + /// Returns `true` only when `self` and `other` describe the same size for + /// every trailing slice length. + /// + /// This recognizes sufficient conditions for equality. A `false` result + /// also includes equivalent size sequences that this test cannot recognize. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&same| { + let (self_align, self_phase) = self.size_rounding_align_and_phase.components(); + let (other_align, other_phase) = other.size_rounding_align_and_phase.components(); + let initial_size = proofs::slice_dst_size_for_trailing_bytes(self, 0); + same == (self.elem_size == other.elem_size + && initial_size.is_some() + && initial_size == proofs::slice_dst_size_for_trailing_bytes(other, 0) + && ((self.elem_size % self_align.get() == 0 + && self.elem_size % other_align.get() == 0) + || (self_align == other_align && self_phase == other_phase))) + }))] + pub(crate) const fn has_same_size_sequence(self, other: Self) -> bool { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::has_same_size_sequence)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().has_same_size_sequence(kani::any()); + } + + // For either layout, interpret its normalized size formula over the + // nonnegative integers: + // + // object_size(elems) + // = size_base + // + round_up(size_phase + elems * elem_size, size_align). + // + // In two cases below, we establish equality of these mathematical sizes + // for every `elems`. Equal sizes also have the same `usize` + // representability bound. + // + // Both cases require a common byte contribution at each element count. + // Equal element sizes give that shared term, `elems * elem_size`. + if self.elem_size != other.elem_size { + return false; + } + + // At `elems == 0`, the element contribution vanishes. Require equal, + // representable initial sizes; each case below then shows that + // increasing `elems` adds the same amount to both initial sizes. + match (self.size_for_elems(0), other.size_for_elems(0)) { + (Some(self_size), Some(other_size)) if self_size == other_size => {} + _ => return false, + } + + let (self_align, self_phase) = self.size_rounding_align_and_phase.components(); + let (other_align, other_phase) = other.size_rounding_align_and_phase.components(); + let self_align = self_align.get(); + let other_align = other_align.get(); + + // Case 1: The element size is a multiple of both alignments. + // + // For two power-of-two alignments, the larger is divisible by the + // smaller. Thus `max_align` is their least common multiple, and a byte + // count is divisible by both alignments exactly when it is divisible by + // `max_align`. + let max_align = if self_align > other_align { self_align } else { other_align }; + + // If the element size is divisible by both alignments, so is `elems * + // elem_size`. Adding an alignment multiple commutes with rounding, + // giving the following identity for each layout: + // + // round_up(size_phase + elems * elem_size, size_align) + // = round_up(size_phase, size_align) + elems * elem_size. + // + // Consequently, `object_size(elems) = object_size(0) + elems * + // elem_size`. The initial sizes and element sizes already agree, so the + // complete size sequences agree. This also covers zero-sized elements, + // whose contribution is zero for every element count. + #[allow(clippy::arithmetic_side_effects)] + if self.elem_size % max_align == 0 { + return true; + } + + // Case 2: The rounding alignments and phases match. + // + // Require a common rounding alignment before comparing phases. If the + // alignments differ, neither case establishes equality. + if self_align != other_align { + return false; + } + + // If the phases also agree, both `round_up` operations have the same + // input and alignment at every element count. In particular, their + // rounded terms agree at zero. Subtracting that common term from the + // equal initial sizes proves `self.size_base == other.size_base`. Both + // the fixed and rounded contributions therefore agree for every + // `elems`, which establishes equality of the complete size formulas. + self_phase == other_phase + } + + /// Advances the size formula by `bytes` while replacing its element size. + /// + /// Let `(size_align, size_phase)` be decoded from + /// `self.size_rounding_align_and_phase`. For a trailing slice length + /// `elems`, the returned layout describes: + /// + /// ```text + /// self.size_base + /// + round_up(size_phase + bytes + elems * elem_size, size_align) + /// ``` + /// + /// Here, `elem_size` is the replacement element size passed to this + /// method. The returned layout preserves the physical slice offset + /// `self.offset` and the rounding alignment; advancing changes the size + /// calculation. + /// + /// Returns `None` if advancing cannot produce representable normalized + /// components. Even when those components fit, evaluating the rounded + /// object size may overflow for some or all trailing slice lengths. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|result| { + let expected_offset = self.size_offset().checked_add(bytes); + match result { + Some(advanced) => { + let align = self.size_rounding_align_and_phase.align(); + expected_offset == Some(advanced.size_offset()) + && advanced.offset == self.offset + && advanced.elem_size == elem_size + && advanced.size_rounding_align_and_phase.align() == align + && (advanced.size_base & (align.get() - 1)) == (self.size_base & (align.get() - 1)) + } + None => expected_offset.is_none(), + } + }))] + const fn advance(self, bytes: usize, elem_size: usize) -> Option { + #[cfg(kani)] + #[kani::proof_for_contract(TrailingSliceLayout::advance)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().advance(kani::any(), kani::any()); + } + + let (size_align, size_phase) = self.size_rounding_align_and_phase.components(); + + // Advancing replaces the phase inside `round_up` with `size_phase + + // bytes`. To normalize this sum, separate it into whole alignment units + // and a remainder. The whole units can move outside `round_up` into the + // base; the remainder becomes the new phase. + // + // Since `size_align` is a nonzero power of two, `align_mask` selects + // exactly the bits representing a remainder modulo `size_align`. + // Nonzero `size_align` also makes this subtraction representable. + #[allow(clippy::arithmetic_side_effects)] + let align_mask = size_align.get() - 1; + + // The base can increase by at most `usize::MAX - self.size_base`. Since + // only whole alignment units move into the base, the largest + // transferable contribution is: + // + // whole_capacity + // = round_down(usize::MAX - self.size_base, size_align). + // + // A remainder up to `align_mask` stays in the phase and consumes no + // base capacity. Thus the largest advanced phase that can be split into + // representable normalized components is `phase_capacity = + // whole_capacity + align_mask`. Setting the low bits of the remaining + // base capacity computes exactly that sum. Its two terms have disjoint + // set bits, so it cannot overflow. Any larger sum would transfer more + // whole units than the base can hold. + #[allow(clippy::arithmetic_side_effects)] + let phase_capacity = (usize::MAX - self.size_base) | align_mask; + + // The existing phase already occupies `size_phase` bytes of this + // capacity. Subtracting it gives the greatest permitted increment: + // + // bytes <= max_advance + // iff size_phase + bytes <= phase_capacity. + // + // Since `phase_capacity >= align_mask >= size_phase`, the subtraction + // cannot underflow. + #[allow(clippy::arithmetic_side_effects)] + let max_advance = phase_capacity - size_phase; + if bytes > max_advance { + return None; + } + + // This is the fixed term inside rounding after the requested + // advancement, before normalization. The check above proves `size_phase + // + bytes <= phase_capacity <= usize::MAX`, so the addition is + // representable. + #[allow(clippy::arithmetic_side_effects)] + let advanced_phase = size_phase + bytes; + + // Taking the remainder retains precisely the part that cannot move + // outside rounding as whole alignment units. In particular, `0 <= + // normalized_phase < size_align`, as required for the new rounding + // parameters. + let normalized_phase = advanced_phase & align_mask; + + // Rounding down removes precisely the remainder retained in the + // phase, so `whole_bytes = advanced_phase - normalized_phase`. + let whole_bytes = + util::round_down_to_next_multiple_of_alignment(advanced_phase, size_align); + + // Because `advanced_phase <= phase_capacity`, we have + // `whole_bytes <= whole_capacity <= usize::MAX - self.size_base`. + // Adding this transferred contribution to the base therefore fits. + #[allow(clippy::arithmetic_side_effects)] + let size_base = self.size_base + whole_bytes; + + // To verify the returned formula, let `trailing_size = elems * + // elem_size` and evaluate the following over the nonnegative integers. + // Since `whole_bytes` is a multiple of `size_align`, it can move back + // inside `round_up` without changing the result: + // + // size_base + round_up(normalized_phase + trailing_size, size_align) + // = self.size_base + // + round_up(whole_bytes + normalized_phase + trailing_size, size_align) + // = self.size_base + // + round_up(size_phase + bytes + trailing_size, size_align). + // + // This is the requested advanced formula. The replacement element size + // supplies its new coefficient of `elems`; `offset` retains the + // caller's field-placement description. + Some(Self { + offset: self.offset, + elem_size, + size_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(size_align, normalized_phase), + }) + } } impl SizeInfo { @@ -73,12 +725,42 @@ impl SizeInfo { #[allow(unused)] #[cfg_attr(not(zerocopy_inline_always), inline)] #[cfg_attr(zerocopy_inline_always, inline(always))] + #[cfg_attr(kani, kani::ensures(|result| match (*self, *result) { + (SizeInfo::Sized { size }, Some(SizeInfo::Sized { size: converted })) => { + size == converted + } + (SizeInfo::SliceDst(original), Some(SizeInfo::SliceDst(converted))) => { + converted.elem_size.get() == original.elem_size + && converted.offset == original.offset + && converted.size_base == original.size_base + && converted.size_rounding_align_and_phase == original.size_rounding_align_and_phase + } + (SizeInfo::SliceDst(original), None) => original.elem_size == 0, + _ => false, + }))] const fn try_to_nonzero_elem_size(&self) -> Option> { + #[cfg(kani)] + #[kani::proof_for_contract(SizeInfo::try_to_nonzero_elem_size)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().try_to_nonzero_elem_size(); + } + Some(match *self { SizeInfo::Sized { size } => SizeInfo::Sized { size }, - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) => { + SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base, + size_rounding_align_and_phase, + }) => { if let Some(elem_size) = NonZeroUsize::new(elem_size) { - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) + SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base, + size_rounding_align_and_phase, + }) } else { return None; } @@ -87,6 +769,40 @@ impl SizeInfo { } } +/// Returns the largest number of `elem_size`-byte elements which fit in +/// `bytes`, along with the number of bytes those elements occupy. +#[cfg_attr( + kani, + kani::ensures(|&(elems, used)| { + (match elems.checked_mul(elem_size.get()) { + Some(product) => { + product == used && used <= bytes && bytes - used < elem_size.get() + } + None => false, + }) && used + .checked_add(elem_size.get()) + .map(|next_used| next_used > bytes) + .unwrap_or(true) + }) +)] +#[inline(always)] +const fn max_elems_for_bytes(bytes: usize, elem_size: NonZeroUsize) -> (usize, usize) { + #[cfg(kani)] + #[kani::proof_for_contract(max_elems_for_bytes)] + #[kani::solver(kissat)] + fn proof() { + max_elems_for_bytes(kani::any(), kani::any()); + } + + #[allow(clippy::arithmetic_side_effects)] + let elems = bytes / elem_size.get(); + let used = match elems.checked_mul(elem_size.get()) { + Some(used) => used, + None => const_unreachable!(), + }; + (elems, used) +} + #[doc(hidden)] #[derive(Copy, Clone)] #[cfg_attr(test, derive(Debug))] @@ -97,6 +813,7 @@ pub enum CastType { } #[cfg_attr(test, derive(Debug))] +#[cfg_attr(kani, derive(kani::Arbitrary))] pub(crate) enum MetadataCastError { Alignment, Size, @@ -128,7 +845,7 @@ impl DstLayout { /// 229. #[cfg(not(kani))] #[cfg(not(target_pointer_width = "16"))] - pub(crate) const CURRENT_MAX_ALIGN: NonZeroUsize = match NonZeroUsize::new(1 << 28) { + pub(crate) const CURRENT_MAX_ALIGN: NonZeroUsize = match NonZeroUsize::new(1 << 29) { Some(max_align) => max_align, None => const_unreachable!(), }; @@ -164,7 +881,18 @@ impl DstLayout { /// shallow padding, unsafe code may assume that the result of this method /// accurately reflects the size, alignment, and lack of static shallow /// padding of that type. + #[cfg_attr(kani, kani::ensures(|result| { + result.align == self.align && result.size_info == self.size_info + && result.statically_shallow_unpadded + }))] const fn assume_shallow_unpadded(self) -> Self { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::assume_shallow_unpadded)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().assume_shallow_unpadded(); + } + Self { statically_shallow_unpadded: true, ..self } } @@ -182,7 +910,20 @@ impl DstLayout { #[doc(hidden)] #[must_use] #[inline] + #[cfg_attr(kani, kani::requires(repr_align.map_or(true, |align| align.is_power_of_two())))] + #[cfg_attr(kani, kani::ensures(|result| { + result.align == repr_align.unwrap_or(Self::MIN_ALIGN) + && result.size_info == SizeInfo::Sized { size: 0 } + && result.statically_shallow_unpadded + }))] pub const fn new_zst(repr_align: Option) -> DstLayout { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::new_zst)] + #[kani::solver(kissat)] + fn proof() { + let _ = DstLayout::new_zst(kani::any()); + } + let align = match repr_align { Some(align) => align, None => Self::MIN_ALIGN, @@ -206,7 +947,24 @@ impl DstLayout { #[doc(hidden)] #[must_use] #[inline] + #[cfg_attr(kani, kani::ensures(|result| { + result.align.get() == mem::align_of::() + && result.size_info == SizeInfo::Sized { size: mem::size_of::() } + && result.statically_shallow_unpadded == false + }))] pub const fn for_type() -> DstLayout { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_type::<()>)] + fn proof_zst() { + let _ = DstLayout::for_type::<()>(); + } + + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_type::<(u8, u64)>)] + fn proof_padded() { + let _ = DstLayout::for_type::<(u8, u64)>(); + } + // SAFETY: `align` is correct by construction. `T: Sized`, and so it is // sound to initialize `size_info` to `SizeInfo::Sized { size }`; the // `size` field is also correct by construction. `unpadded` can safely @@ -221,16 +979,39 @@ impl DstLayout { } } - /// Constructs a `DstLayout` which describes a `T` that does not contain - /// padding. + /// Produces `T`'s size and alignment with static shallow padding assumed + /// absent. + /// + /// This method records the assumption without checking `T` for padding. + /// It can be used to ignore padding within a field when computing padding + /// introduced by its enclosing struct. /// /// # Safety /// - /// Unsafe code may assume that `DstLayout` is the correct layout for `T`. + /// Unsafe code may rely on the result's size and alignment being correct + /// for `T`. Relying on its recorded absence of static shallow padding + /// requires an independent justification that `T` lacks such padding. #[doc(hidden)] #[must_use] #[inline] + #[cfg_attr(kani, kani::ensures(|result| { + result.align.get() == mem::align_of::() + && result.size_info == SizeInfo::Sized { size: mem::size_of::() } + && result.statically_shallow_unpadded == true + }))] pub const fn for_unpadded_type() -> DstLayout { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_unpadded_type::<()>)] + fn proof_zst() { + let _ = DstLayout::for_unpadded_type::<()>(); + } + + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_unpadded_type::<(u8, u64)>)] + fn proof_padded() { + let _ = DstLayout::for_unpadded_type::<(u8, u64)>(); + } + Self::for_type::().assume_shallow_unpadded() } @@ -239,7 +1020,31 @@ impl DstLayout { /// # Safety /// /// Unsafe code may assume that `DstLayout` is the correct layout for `[T]`. + #[cfg_attr(kani, kani::ensures(|result| { + result.align.get() == mem::align_of::() && result.statically_shallow_unpadded + && matches!(result.size_info, SizeInfo::SliceDst(trailing) + if trailing.offset == 0 && trailing.size_base == 0 + && trailing.elem_size == mem::size_of::() + && trailing.size_rounding_align_and_phase.components() == (result.align, 0)) + }))] pub(crate) const fn for_slice() -> DstLayout { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_slice::<()>)] + fn proof_zst() { + let _ = DstLayout::for_slice::<()>(); + } + + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_slice::<(u8, u64)>)] + fn proof_padded() { + let _ = DstLayout::for_slice::<(u8, u64)>(); + } + + let align = match NonZeroUsize::new(mem::align_of::()) { + Some(align) => align, + None => const_unreachable!(), + }; + // SAFETY: The alignment of a slice is equal to the alignment of its // element type, and so `align` is initialized correctly. // @@ -249,13 +1054,12 @@ impl DstLayout { // construction. Since `[T]` is a (degenerate case of a) slice DST, it // is correct to initialize `size_info` to `SizeInfo::SliceDst`. DstLayout { - align: match NonZeroUsize::new(mem::align_of::()) { - Some(align) => align, - None => const_unreachable!(), - }, + align, size_info: SizeInfo::SliceDst(TrailingSliceLayout { offset: 0, elem_size: mem::size_of::(), + size_base: 0, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(align, 0), }), statically_shallow_unpadded: true, } @@ -273,15 +1077,42 @@ impl DstLayout { /// alignment modifiers and fields corresponding to that definition, the /// resulting `DstLayout` will correctly encode the layout of that struct. /// + /// This guarantee uses one explicit compatibility premise for generic + /// indirection: while Rust accepts an instantiation which transitively + /// places a `repr(align)` type inside a `repr(packed)` type, that + /// instantiation follows the otherwise-applicable `repr(C)`, + /// `repr(packed)`, and inner-type layout rules. The Reference prohibits the + /// direct form, and Rust's continued acceptance and layout behavior are + /// outside the algebra proved by this implementation. Re-audit if the + /// compiler rejects such an instantiation or assigns it a different + /// layout. + /// /// We make no guarantees to the behavior of this method when it is invoked /// with arguments that cannot correspond to a valid `repr(C)` struct. #[must_use] #[inline] + #[cfg_attr(kani, kani::requires(proofs::repr_c_layout(repr_align, repr_packed, fields).is_some()))] + #[cfg_attr(kani, kani::ensures(|&result| { + Some(result) == proofs::repr_c_layout(repr_align, repr_packed, fields) + }))] pub const fn for_repr_c_struct( repr_align: Option, repr_packed: Option, fields: &[DstLayout], ) -> DstLayout { + // This harness covers up to three fields, including an optional final + // DST. Arbitrary-length composition is not established by this bound. + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::for_repr_c_struct)] + #[kani::solver(kissat)] + #[kani::unwind(4)] + fn proof() { + let fields: [DstLayout; 3] = kani::any(); + let len: usize = kani::any(); + kani::assume(len <= fields.len()); + let _ = DstLayout::for_repr_c_struct(kani::any(), kani::any(), &fields[..len]); + } + let mut layout = DstLayout::new_zst(repr_align); let mut i = 0; @@ -338,7 +1169,18 @@ impl DstLayout { #[doc(hidden)] #[must_use] #[inline] + #[cfg_attr(kani, kani::requires(proofs::extended_layout(self, field, repr_packed).is_some()))] + #[cfg_attr(kani, kani::ensures(|&result| { + Some(result) == proofs::extended_layout(self, field, repr_packed) + }))] pub const fn extend(self, field: DstLayout, repr_packed: Option) -> Self { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::extend)] + #[kani::solver(kissat)] + fn proof() { + let _ = kani::any::().extend(kani::any(), kani::any()); + } + use util::{max, min, padding_needed_for}; // If `repr_packed` is `None`, there are no alignment constraints, and @@ -427,6 +1269,8 @@ impl DstLayout { SizeInfo::SliceDst(TrailingSliceLayout { offset: trailing_offset, elem_size, + size_base, + size_rounding_align_and_phase, }) => { // If the trailing field is dynamically sized, so too // will the resulting layout. The offset of the trailing @@ -440,11 +1284,30 @@ impl DstLayout { // panic otherwise (e.g., combining or aligning the // components would create a size exceeding // `usize::MAX`). - let offset = match offset.checked_add(trailing_offset) { + let trailing_offset = match offset.checked_add(trailing_offset) { Some(offset) => offset, None => const_panic!("`field` cannot be appended without the total size overflowing `usize`"), }; - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) + + // Before normalization, the composite's size is: + // + // offset + size_base + // + round_up(size_phase + elems * elem_size, size_align) + // + // `offset` does not participate in the rounding, + // so it can be added directly to the normalized + // base. + let size_base = match offset.checked_add(size_base) { + Some(base) => base, + None => const_panic!("`field` cannot be appended without the total size overflowing `usize`"), + }; + + SizeInfo::SliceDst(TrailingSliceLayout { + offset: trailing_offset, + elem_size, + size_base, + size_rounding_align_and_phase, + }) } }, ) @@ -459,18 +1322,11 @@ impl DstLayout { } /// Like `Layout::pad_to_align`, this routine rounds the size of this layout - /// up to the nearest multiple of this type's alignment or `repr_packed` - /// (whichever is less). This method leaves DST layouts unchanged, since the - /// trailing padding of DSTs is computed at runtime. - /// - /// The accompanying boolean is `true` if the resulting composition of - /// fields necessitated static (as opposed to dynamic) padding; otherwise - /// `false`. + /// up to the nearest multiple of this type's alignment. For DST layouts, + /// this updates the runtime size formula to account for trailing padding. /// /// In order to match the layout of a `#[repr(C)]` struct, this method - /// should be invoked after the invocations of [`DstLayout::extend`]. If - /// `self` corresponds to a type marked with `repr(packed(N))`, then - /// `repr_packed` should be set to `Some(N)`, otherwise `None`. + /// should be invoked after the invocations of [`DstLayout::extend`]. /// /// This method cannot be used to match the layout of a record with the /// default representation, as that representation is mostly unspecified. @@ -480,7 +1336,7 @@ impl DstLayout { /// If a (potentially hypothetical) valid `repr(C)` type begins with fields /// whose layout are `self` followed only by zero or more bytes of trailing /// padding (not included in `self`), then unsafe code may rely on - /// `self.pad_to_align(repr_packed)` producing a layout that correctly + /// `self.pad_to_align()` producing a layout that correctly /// encapsulates the layout of that type. /// /// We make no guarantees to the behavior of this method if `self` cannot @@ -489,7 +1345,16 @@ impl DstLayout { #[doc(hidden)] #[must_use] #[inline] + #[cfg_attr(kani, kani::requires(proofs::padded_layout(self).is_some()))] + #[cfg_attr(kani, kani::ensures(|&result| Some(result) == proofs::padded_layout(self)))] pub const fn pad_to_align(self) -> Self { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::pad_to_align)] + #[kani::solver(kissat)] + fn proof() { + let _ = kani::any::().pad_to_align(); + } + use util::padding_needed_for; let (static_padding, size_info) = match self.size_info { @@ -504,10 +1369,69 @@ impl DstLayout { (padding, SizeInfo::Sized { size }) } // For DST layouts, trailing padding depends on the length of the - // trailing DST and is computed at runtime. This does not alter the - // offset or element size of the layout, so we leave `size_info` - // unchanged. - size_info @ SizeInfo::SliceDst(_) => (0, size_info), + // trailing DST and is computed at runtime. Normalize the composed + // rounding operation back into the representation documented on + // `TrailingSliceLayout`. + SizeInfo::SliceDst(mut trailing) => { + let (size_align, size_phase) = trailing.size_rounding_align_and_phase.components(); + if size_align.get() < self.align.get() { + // For a trailing-slice byte count `trailing_size`, the + // input layout describes: + // + // trailing.size_base + // + round_up(size_phase + trailing_size, size_align) + // + // Here, `size_align < self.align`. Both alignments are + // powers of two, so rounding the complete input size to + // `self.align` is equivalent to: + // + // rounded_base = round_up(trailing.size_base, size_align) + // fixed_contribution = rounded_base + size_phase + // new_phase = fixed_contribution % self.align + // new_base = fixed_contribution - new_phase + // new_base + round_up(new_phase + trailing_size, self.align) + // + // `prove_dst_layout_pad_to_align_dst_outer_alignment_transition` + // checks the exact field transition and preserved + // invariants. Its `_formula` companion proves this + // identity for every representable trailing byte count. + let base_padding = padding_needed_for(trailing.size_base, size_align); + let rounded_base = match trailing.size_base.checked_add(base_padding) { + Some(base) => base, + None => const_panic!("Adding padding caused size to overflow `usize`."), + }; + let fixed_contribution = match rounded_base.checked_add(size_phase) { + Some(bytes) => bytes, + None => const_panic!("Adding padding caused size to overflow `usize`."), + }; + #[allow(clippy::arithmetic_side_effects)] + let phase = fixed_contribution & (self.align.get() - 1); + // Rounding down removes `phase`, the fixed contribution + // modulo `self.align`, leaving the new base. + let size_base = util::round_down_to_next_multiple_of_alignment( + fixed_contribution, + self.align, + ); + trailing.size_base = size_base; + trailing.size_rounding_align_and_phase = + RoundingAlignAndPhase::new(self.align, phase); + } else { + // `size_align` is a multiple of `self.align`, so + // the inner rounded size is already aligned to + // `self.align`. Only `size_base` needs to be rounded. + // `prove_dst_layout_pad_to_align_dst_inner_rounding_transition` + // checks the exact field transition and preserved + // invariants. Its `_formula` companion proves size + // equivalence for every representable trailing byte count. + let padding = padding_needed_for(trailing.size_base, self.align); + trailing.size_base = match trailing.size_base.checked_add(padding) { + Some(base) => base, + None => const_panic!("Adding padding caused size to overflow `usize`."), + }; + } + + (0, SizeInfo::SliceDst(trailing)) + } }; let statically_shallow_unpadded = self.statically_shallow_unpadded && static_padding == 0; @@ -515,48 +1439,127 @@ impl DstLayout { DstLayout { align: self.align, size_info, statically_shallow_unpadded } } - /// Produces `true` if `self` requires static padding; otherwise `false`. + /// Produces `false` when this layout is marked as having no static shallow + /// padding, and `true` otherwise. + /// + /// A `true` result is conservative: it does not establish that the type + /// contains padding. For example, this method produces `true` for + /// `DstLayout::for_type::()`, since `for_type` does not record whether + /// static shallow padding is absent. #[must_use] #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&padding| padding == !self.statically_shallow_unpadded))] pub const fn requires_static_padding(self) -> bool { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::requires_static_padding)] + #[kani::solver(kissat)] + fn proof() { + let _ = kani::any::().requires_static_padding(); + } + !self.statically_shallow_unpadded } - /// Produces `true` if there exists any metadata for which a type of layout - /// `self` would require dynamic trailing padding; otherwise `false`. + /// Produces `false` only if every valid metadata describes an instance + /// which requires no dynamic trailing padding. + /// + /// A `true` result conservatively selects the dynamic-padding path; it does + /// not guarantee that any valid metadata actually requires padding. #[must_use] #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&padding| match self.size_info { + SizeInfo::Sized { .. } => !padding, + SizeInfo::SliceDst(trailing) => padding == ( + proofs::slice_dst_size_for_trailing_bytes(trailing, 0) != Some(trailing.offset) + || trailing.elem_size % trailing.size_rounding_align_and_phase.align().get() != 0 + ), + }))] pub const fn requires_dynamic_padding(self) -> bool { - // A `% self.align.get()` cannot panic, since `align` is non-zero. - #[allow(clippy::arithmetic_side_effects)] + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::requires_dynamic_padding)] + #[kani::solver(kissat)] + fn proof() { + let _ = kani::any::().requires_dynamic_padding(); + } + match self.size_info { SizeInfo::Sized { .. } => false, SizeInfo::SliceDst(trailing_slice_layout) => { - // SAFETY: This predicate is formally proved sound by - // `proofs::prove_requires_dynamic_padding`. - trailing_slice_layout.offset % self.align.get() != 0 - || trailing_slice_layout.elem_size % self.align.get() != 0 + // SAFETY: `proofs::prove_requires_dynamic_padding` formally + // proves the safety-critical direction: when this predicate + // returns false, no valid metadata requires dynamic padding. + match trailing_slice_layout.size_for_elems(0) { + Some(initial_size) => { + #[allow(clippy::arithmetic_side_effects)] + let has_static_stride = trailing_slice_layout.elem_size + % trailing_slice_layout.size_rounding_align_and_phase.align().get() + == 0; + initial_size != trailing_slice_layout.offset || !has_static_stride + } + None => true, + } + } + } + } + + /// Returns the largest trailing slice length whose object size is exactly + /// `size`. + /// + /// Returns `None` if there is no such length, if this describes a sized + /// type, or if the trailing slice element is zero-sized. + #[inline(always)] + #[cfg_attr(kani, kani::ensures(|&result| { + result == match self.size_info { + SizeInfo::SliceDst(trailing) if trailing.elem_size != 0 => { + proofs::max_trailing_bytes(trailing, size) + .map(|bytes| bytes / trailing.elem_size) + .filter(|&elems| trailing.size_for_elems(elems) == Some(size)) + } + _ => None, + } + }))] + const fn metadata_for_exact_size(&self, size: usize) -> Option { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::metadata_for_exact_size)] + #[kani::solver(kissat)] + fn proof() { + kani::any::().metadata_for_exact_size(kani::any()); + } + + match self.size_info { + SizeInfo::Sized { .. } + | SizeInfo::SliceDst(TrailingSliceLayout { elem_size: 0, .. }) => None, + SizeInfo::SliceDst(_) => { + // Address zero is aligned to every supported alignment, and + // `0 + size` cannot overflow. The validator returns the + // largest metadata whose object fits in `size`; accepting it + // only when the resulting object consumes all `size` bytes + // turns that bounded query into an exact-size query. + match self.validate_cast_and_convert_metadata(0, size, CastType::Prefix) { + Ok((elems, object_size)) if object_size == size => Some(elems), + _ => None, + } } } } - /// Validates that a cast is sound from a layout perspective. + /// Given a completed type layout, produces metadata and a split index for + /// a prefix or suffix cast that satisfies the type's size and alignment + /// requirements. /// - /// Validates that the size and alignment requirements of a type with the - /// layout described in `self` would not be violated by performing a - /// `cast_type` cast from a pointer with address `addr` which refers to a - /// memory region of size `bytes_len`. + /// `addr` and `bytes_len` describe the address and size of the source + /// memory region, and `cast_type` selects its prefix or suffix. /// /// If the cast is valid, `validate_cast_and_convert_metadata` returns - /// `(elems, split_at)`. If `self` describes a dynamically-sized type, then - /// `elems` is the maximum number of trailing slice elements for which a - /// cast would be valid (for sized types, `elem` is meaningless and should - /// be ignored). `split_at` is the index at which to split the memory region - /// in order for the prefix (suffix) to contain the result of the cast, and - /// in order for the remaining suffix (prefix) to contain the leftover - /// bytes. + /// `Ok((elems, split_at))`. For a dynamically-sized type, `elems` is the + /// maximum number of trailing slice elements for which the cast is valid. + /// For sized types, `elems` is meaningless and should be ignored. + /// `split_at` is the index at which to split the memory region so that the + /// prefix (suffix) contains the result of the cast and the remaining + /// suffix (prefix) contains the leftover bytes. /// /// There are three conditions under which a cast can fail: + /// /// - The smallest possible value for the type is larger than the provided /// memory region /// - A prefix cast is requested, and `addr` does not satisfy `self`'s @@ -568,10 +1571,10 @@ impl DstLayout { /// /// # Safety /// - /// The caller may assume that this implementation is correct, and may rely - /// on that assumption for the soundness of their code. In particular, the - /// caller may assume that, if `validate_cast_and_convert_metadata` returns - /// `Some((elems, split_at))`, then: + /// When `self` accurately describes a completed type layout, including + /// trailing padding, and `addr + bytes_len` fits in `usize`, callers may + /// rely on the following guarantees for `Ok((elems, split_at))`: + /// /// - A pointer to the type (for dynamically sized types, this includes /// `elems` as its pointer metadata) describes an object of size `size <= /// bytes_len` @@ -582,6 +1585,11 @@ impl DstLayout { /// - `split_at == bytes_len - size` /// - `addr + split_at` satisfies `self`'s alignment /// + /// These guarantees require the completed-layout premise; `DstLayout` + /// also permits unfinished layout fragments and arbitrary field values. + /// In particular, suffix alignment relies on every object size being a + /// multiple of the type's alignment. + /// /// Note that this method does *not* ensure that a pointer constructed from /// its return values will be a valid pointer. In particular, this method /// does not reason about `isize` overflow, which is a requirement of many @@ -606,12 +1614,57 @@ impl DstLayout { /// condition, even if `debug_assertions` are enabled. #[allow(unused)] #[inline(always)] + #[cfg_attr(kani, kani::requires(addr.checked_add(bytes_len).is_some()))] + #[cfg_attr(kani, kani::requires(!matches!(self.size_info, + SizeInfo::SliceDst(TrailingSliceLayout { elem_size: 0, .. }))))] + #[cfg_attr(kani, kani::ensures(|result| { + let endpoint = match cast_type { + CastType::Prefix => addr, + CastType::Suffix => addr + bytes_len, + }; + if endpoint % self.align.get() != 0 { + return matches!(result, Err(MetadataCastError::Alignment)); + } + let candidate = match self.size_info { + SizeInfo::Sized { size } => { + if size <= bytes_len { Some((0, size)) } else { None } + } + SizeInfo::SliceDst(trailing) => { + proofs::max_trailing_bytes(trailing, bytes_len).and_then(|bytes| { + let elems = bytes / trailing.elem_size; + trailing.size_for_elems(elems).map(|size| (elems, size)) + }) + } + }; + match (result, candidate) { + (Ok((elems, split_at)), Some((expected_elems, size))) => { + *elems == expected_elems && *split_at == match cast_type { + CastType::Prefix => size, + CastType::Suffix => bytes_len - size, + } + } + (Err(MetadataCastError::Size), None) => true, + _ => false, + } + }))] pub(crate) const fn validate_cast_and_convert_metadata( &self, addr: usize, bytes_len: usize, cast_type: CastType, ) -> Result<(usize, usize), MetadataCastError> { + #[cfg(kani)] + #[kani::proof_for_contract(DstLayout::validate_cast_and_convert_metadata)] + #[kani::solver(kissat)] + fn proof() { + let cast_type = if kani::any() { CastType::Prefix } else { CastType::Suffix }; + let _ = kani::any::().validate_cast_and_convert_metadata( + kani::any(), + kani::any(), + cast_type, + ); + } + // `debug_assert!`, but with `#[allow(clippy::arithmetic_side_effects)]`. macro_rules! __const_debug_assert { ($e:expr $(, $msg:expr)?) => { @@ -651,9 +1704,8 @@ impl DstLayout { // of this check is trivial - `addr` is the address the object will // live at. // - // For a suffix cast, we know that all valid sizes for the type are - // a multiple of the alignment (and by safety precondition, we know - // `DstLayout` may only describe valid Rust types). Thus, a + // Under the documented completed-layout premise, all valid sizes + // for the type are multiples of its alignment. Thus, a // validly-sized instance which lives at a validly-aligned address // must also end at a validly-aligned address. Thus, if the end // address for a suffix cast (`addr + bytes_len`) is not aligned, @@ -666,7 +1718,7 @@ impl DstLayout { // Addition is guaranteed not to overflow because `offset <= // bytes_len`, and `addr + bytes_len <= usize::MAX` is a // precondition of this method. Modulus is guaranteed not to divide - // by 0 because `align` is non-zero. + // by zero because `self.align` is non-zero. #[allow(clippy::arithmetic_side_effects)] if (addr + offset) % self.align.get() != 0 { return Err(MetadataCastError::Alignment); @@ -680,59 +1732,50 @@ impl DstLayout { } (0, size) } - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) => { - // Calculate the maximum number of bytes that could be consumed - // - any number of bytes larger than this will either not be a - // multiple of the alignment, or will be larger than - // `bytes_len`. - let max_total_bytes = - util::round_down_to_next_multiple_of_alignment(bytes_len, self.align); - // Calculate the maximum number of bytes that could be consumed - // by the trailing slice. - // - // FIXME(#67): Once our MSRV is 1.65, use let-else: - // https://blog.rust-lang.org/2022/11/03/Rust-1.65.0.html#let-else-statements - let max_slice_and_padding_bytes = match max_total_bytes.checked_sub(offset) { - Some(max) => max, - // `bytes_len` too small even for 0 trailing slice elements. + SizeInfo::SliceDst(trailing) => { + let max_slice_bytes = match trailing.max_trailing_bytes(bytes_len) { + Some(bytes) => bytes, None => return Err(MetadataCastError::Size), }; - // Calculate the number of elements that fit in - // `max_slice_and_padding_bytes`; any remaining bytes will be - // considered padding. + // `max_slice_bytes`. Any remaining input bytes stay outside + // the selected object unless the object's own rounding + // operation consumes them as padding. // // Guaranteed not to divide by zero: `elem_size` is non-zero. - #[allow(clippy::arithmetic_side_effects)] - let elems = max_slice_and_padding_bytes / elem_size.get(); - // Guaranteed not to overflow on multiplication: `usize::MAX >= - // max_slice_and_padding_bytes >= (max_slice_and_padding_bytes / - // elem_size) * elem_size`. + let (elems, _) = max_elems_for_bytes(max_slice_bytes, trailing.elem_size); + // The remainder is `max_slice_bytes - elems * elem_size`. + // Writing it as a remainder exposes its bound to the + // optimizer: when `elem_size <= size_align`, rounding the + // remainder down to `size_align` always gives zero. + let size_align = trailing.size_rounding_align_and_phase.align(); + // `size_phase + max_slice_bytes == max_rounded_bytes`, an + // alignment multiple. The selected tail removes `unused_bytes` + // from this boundary; rounding back up removes only whole + // alignment units. Thus its normalized rounded size is: // - // Guaranteed not to overflow on addition: - // - max_slice_and_padding_bytes == max_total_bytes - offset - // - elems * elem_size <= max_slice_and_padding_bytes == max_total_bytes - offset - // - elems * elem_size + offset <= max_total_bytes <= usize::MAX - #[allow(clippy::arithmetic_side_effects)] - let without_padding = offset + elems * elem_size.get(); - // `self_bytes` is equal to the offset bytes plus the bytes - // consumed by the trailing slice plus any padding bytes - // required to satisfy the alignment. Note that we have computed - // the maximum number of trailing slice elements that could fit - // in `self_bytes`, so any padding is guaranteed to be less than - // the size of an extra element. + // round_up(size_phase + elems * elem_size, size_align) + // = max_rounded_bytes - round_down(unused_bytes, size_align) // - // Guaranteed not to overflow: - // - By previous comment: without_padding == elems * elem_size + - // offset <= max_total_bytes - // - By construction, `max_total_bytes` is a multiple of - // `self.align`. - // - At most, adding padding needed to round `without_padding` - // up to the next multiple of the alignment will bring - // `self_bytes` up to `max_total_bytes`. + // `prove_slice_dst_validator_selected_size_fits` checks this + // identity and its intermediate bounds, using the relations + // in `prove_max_trailing_bytes_components`. The capacity is + // exact by `prove_max_trailing_bytes`; the element helper's + // contract establishes the largest whole-element count. #[allow(clippy::arithmetic_side_effects)] - let self_bytes = - without_padding + util::padding_needed_for(without_padding, self.align); + let unused_bytes = max_slice_bytes % trailing.elem_size.get(); + let unused_aligned_bytes = + util::round_down_to_next_multiple_of_alignment(unused_bytes, size_align); + #[allow(clippy::arithmetic_side_effects)] + let max_rounded_bytes = util::round_down_to_next_multiple_of_alignment( + bytes_len - trailing.size_base, + size_align, + ); + // `unused_aligned_bytes <= unused_bytes <= max_slice_bytes + // <= max_rounded_bytes`, and `size_base + max_rounded_bytes + // <= bytes_len`, so neither operation can overflow. + #[allow(clippy::arithmetic_side_effects)] + let self_bytes = trailing.size_base + (max_rounded_bytes - unused_aligned_bytes); (elems, self_bytes) } }; @@ -745,7 +1788,7 @@ impl DstLayout { // - In the `Sized` branch, only returns `size` if `size <= // bytes_len`. // - In the `SliceDst` branch, calculates `self_bytes <= - // max_toatl_bytes`, which is upper-bounded by `bytes_len`. + // bytes_len`. #[allow(clippy::arithmetic_side_effects)] CastType::Suffix => bytes_len - self_bytes, }; @@ -754,6 +1797,95 @@ impl DstLayout { } } +/// Computes the affine destination-metadata map. +/// +/// # Safety +/// +/// `base + metadata * multiple`, evaluated over the mathematical nonnegative +/// integers, must not exceed `usize::MAX`. +#[inline(always)] +#[allow(unstable_name_collisions)] +#[cfg_attr(kani, kani::requires(metadata.checked_mul(multiple) + .and_then(|scaled| base.checked_add(scaled)).is_some()))] +#[cfg_attr(kani, kani::ensures(|&result| { + metadata.checked_mul(multiple).and_then(|scaled| base.checked_add(scaled)) == Some(result) +}))] +unsafe fn add_scaled_metadata(base: usize, metadata: usize, multiple: usize) -> usize { + // The call-site safety argument establishes that both operations are + // representable for valid source metadata. This harness verifies the + // production implementation selected by Kani's modern rustc against + // checked arithmetic under precisely that precondition. The distributive + // natural-number argument which establishes the precondition is kept + // beside the unsafe pointer projection where it can be reviewed with the + // layout premises that it consumes. + #[cfg(kani)] + #[kani::proof_for_contract(add_scaled_metadata)] + #[kani::solver(kissat)] + fn proof() { + let base: usize = kani::any(); + let metadata: usize = kani::any(); + let multiple: usize = kani::any(); + let Some(scaled) = metadata.checked_mul(multiple) else { + kani::assume(false); + loop {} + }; + let Some(expected) = base.checked_add(scaled) else { + kani::assume(false); + loop {} + }; + + // SAFETY: The checked operations above returned `Some`, so the + // operation is representable in `usize`. The standard-library + // contracts say that this rules out undefined behavior for both + // unchecked operations [1][2]. + // + // [1] Per https://doc.rust-lang.org/1.93.1/std/primitive.usize.html#method.unchecked_mul: + // + // This results in undefined behavior [..] when `checked_mul` would + // return `None`. + // + // [2] Per https://doc.rust-lang.org/1.93.1/std/primitive.usize.html#method.unchecked_add: + // + // This results in undefined behavior [..] when `checked_add` would + // return `None`. + assert_eq!(unsafe { add_scaled_metadata(base, metadata, multiple) }, expected); + } + + #[allow(unused_imports)] + use crate::util::polyfills::*; + + // SAFETY: The caller promises that `base + metadata * multiple` is + // representable over the nonnegative integers. Thus `metadata * multiple` + // is representable too. On toolchains with inherent unchecked arithmetic, + // this is exactly the condition under which `usize::unchecked_mul` does + // not have undefined behavior [1]. On older supported toolchains, method + // resolution selects `polyfills::NumExt::unchecked_mul`, whose safety + // contract requires the same no-overflow condition. Kani checks the two + // implementations in separate harnesses. + // + // [1] Per https://doc.rust-lang.org/1.93.1/std/primitive.usize.html#method.unchecked_mul: + // + // This results in undefined behavior when `self * rhs > usize::MAX` or + // `self * rhs < usize::MIN`, i.e. when `checked_mul` would return `None`. + // + let scaled = unsafe { metadata.unchecked_mul(multiple) }; + + // SAFETY: The caller promises that `base + metadata * multiple` is + // representable in `usize`, and `scaled == metadata * multiple`. On + // toolchains with inherent unchecked arithmetic, that is exactly the + // condition under which `usize::unchecked_add` does not have undefined + // behavior [1]. On older supported toolchains, method resolution selects + // `polyfills::NumExt::unchecked_add`, whose safety contract requires the + // same no-overflow condition. Kani checks the two implementations in + // separate harnesses. + // + // [1] Per https://doc.rust-lang.org/1.93.1/std/primitive.usize.html#method.unchecked_add: + // + // This results in undefined behavior when `self + rhs > usize::MAX` or + // `self + rhs < usize::MIN`, i.e. when `checked_add` would return `None`. + unsafe { base.unchecked_add(scaled) } +} + pub(crate) use cast_from::CastFrom; mod cast_from { use crate::*; @@ -823,52 +1955,19 @@ mod cast_from { // - If this is possible, any information necessary to perform // the `Src`->`Dst` metadata conversion at runtime. // - // Assume that `Src` and `Dst` are slice DSTs, and define: - // - `S_OFF = Src::LAYOUT.size_info.offset` - // - `S_ELEM = Src::LAYOUT.size_info.elem_size` - // - `D_OFF = Dst::LAYOUT.size_info.offset` - // - `D_ELEM = Dst::LAYOUT.size_info.elem_size` - // - // We are trying to solve the following equation: - // - // D_OFF + d_meta * D_ELEM = S_OFF + s_meta * S_ELEM - // - // At runtime, we will be attempting to compute `d_meta`, given - // `s_meta` (a runtime value) and all other parameters (which - // are compile-time values). We can solve like so: - // - // D_OFF + d_meta * D_ELEM = S_OFF + s_meta * S_ELEM - // - // d_meta * D_ELEM = S_OFF - D_OFF + s_meta * S_ELEM - // - // d_meta = (S_OFF - D_OFF + s_meta * S_ELEM)/D_ELEM - // - // Since `d_meta` will be a `usize`, we need the right-hand side - // to be an integer, and this needs to hold for *any* value of - // `s_meta` (in order for our conversion to be infallible - ie, - // to not have to reject certain values of `s_meta` at runtime). - // This means that: - // - // - `s_meta * S_ELEM` must be a multiple of `D_ELEM` - // - Since this must hold for any value of `s_meta`, `S_ELEM` - // must be a multiple of `D_ELEM` - // - `S_OFF - D_OFF` must be a multiple of `D_ELEM` - // - // Thus, let `OFFSET_DELTA_ELEMS = (S_OFF - D_OFF)/D_ELEM` and - // `ELEM_MULTIPLE = S_ELEM/D_ELEM`. We can rewrite the above - // expression as: - // - // d_meta = (S_OFF - D_OFF + s_meta * S_ELEM)/D_ELEM + // For slice DSTs, destination metadata is an affine function + // of source metadata: // - // d_meta = OFFSET_DELTA_ELEMS + s_meta * ELEM_MULTIPLE + // dst_meta = offset_delta_elems + src_meta * elem_multiple // - // Thus, we just need to compute the following and confirm that - // they have integer solutions in order to both a) determine - // whether infallible `Src` -> `Dst` casts are possible and, b) - // pre-compute the parameters necessary to perform those casts - // at runtime. These parameters are encapsulated in - // `CastParams`, which acts as a witness that such infallible - // casts are possible. + // `elem_multiple` scales the destination's trailing element + // size to the source's. `offset_delta_elems` selects a + // destination size equal to the source's zero-element size. + // We then prove that the two complete rounded size formulas + // produce the same sequence for all metadata. This is + // necessary because a packed outer DST can retain a nested + // field's rounding operation; comparing only physical + // trailing-slice offsets is not sufficient. /// The parameters required in order to perform an /// unsized-to-unsized pointer cast from `Src` to `Dst` as /// described above. @@ -910,15 +2009,60 @@ mod cast_from { } impl CastParams { + /// Given a nonzero `dst.elem_size` that exactly divides + /// `src.elem_size`, produces `true` only when the following + /// metadata map preserves the complete object size for every + /// source element count `src_meta`: + /// + /// ```text + /// dst_meta = dst_base + src_meta * (src.elem_size / dst.elem_size) + /// ``` + /// + /// This recognizes sufficient conditions for equality. A + /// `false` result also includes equivalent size sequences + /// that this test cannot recognize, or an adjustment whose + /// normalized components cannot be represented in `usize`. + const fn size_sequences_match( + src: TrailingSliceLayout, + dst: TrailingSliceLayout, + dst_base: usize, + ) -> bool { + let base_bytes = match dst_base.checked_mul(dst.elem_size) { + Some(bytes) => bytes, + None => return false, + }; + let shifted_dst = match dst.advance(base_bytes, src.elem_size) { + Some(layout) => layout, + None => return false, + }; + src.has_same_size_sequence(shifted_dst) + } + + /// Given the complete layouts of `Src` and `Dst`, produces + /// parameters for a cast that preserves object size. + /// + /// A `Some` result establishes that `Src`'s alignment is at + /// least `Dst`'s and that the metadata map preserves size for + /// every valid source metadata value. + /// + /// Supports casts between sized types of equal size, from a + /// sized type to a slice DST, and between slice DSTs. A slice + /// DST destination must have nonzero-sized elements; between + /// slice DSTs, its element size must exactly divide the + /// source's nonzero element size. + /// + /// Produces `None` for unsupported casts or when this method + /// cannot establish a size-preserving metadata map. Rejection + /// does not establish that no such map exists. const fn try_compute( - src: &DstLayout, - dst: &DstLayout, + src_layout: &DstLayout, + dst_layout: &DstLayout, ) -> Option> { - if src.align.get() < dst.align.get() { + if src_layout.align.get() < dst_layout.align.get() { return None; } - let inner = match (src.size_info, dst.size_info) { + let inner = match (src_layout.size_info, dst_layout.size_info) { ( SizeInfo::Sized { size: src_size }, SizeInfo::Sized { size: dst_size }, @@ -931,82 +2075,79 @@ mod cast_from { // dst_size`. CastParamsInner::SizedToSized } - (SizeInfo::Sized { size: src_size }, SizeInfo::SliceDst(dst)) => { - let offset_delta = if let Some(od) = src_size.checked_sub(dst.offset) { - od - } else { - return None; - }; - - let dst_elem_size = if let Some(e) = NonZeroUsize::new(dst.elem_size) { - e - } else { - return None; + (SizeInfo::Sized { size: src_size }, SizeInfo::SliceDst(_)) => { + let dst_meta = match dst_layout.metadata_for_exact_size(src_size) { + Some(meta) => meta, + None => return None, }; - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let delta_mod_other_elem = offset_delta % dst_elem_size.get(); - - if delta_mod_other_elem != 0 { - return None; - } - - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let dst_meta = offset_delta / dst_elem_size.get(); - // SAFETY: The preceding math ensures that a `Dst` // with `dst_meta` addresses `src_size` bytes. CastParamsInner::SizedToUnsized { dst_meta } } (SizeInfo::SliceDst(src), SizeInfo::SliceDst(dst)) => { - let offset_delta = if let Some(od) = src.offset.checked_sub(dst.offset) - { - od - } else { - return None; - }; - let dst_elem_size = if let Some(e) = NonZeroUsize::new(dst.elem_size) { e } else { return None; }; - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let delta_mod_other_elem = offset_delta % dst_elem_size.get(); - - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let elem_remainder = src.elem_size % dst_elem_size.get(); - - if delta_mod_other_elem != 0 - || src.elem_size < dst.elem_size - || elem_remainder != 0 - { + if src.elem_size < dst.elem_size { return None; } - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let offset_delta_elems = offset_delta / dst_elem_size.get(); + let (elem_multiple, described_src_elem_size) = + super::max_elems_for_bytes(src.elem_size, dst_elem_size); + if described_src_elem_size != src.elem_size { + return None; + } - // PANICS: `dst_elem_size: NonZeroUsize`, so this won't - // divide by zero. - #[allow(clippy::arithmetic_side_effects)] - let elem_multiple = src.elem_size / dst_elem_size.get(); + // Prefer an exact destination-element adjustment + // between `src.size_offset()` and + // `dst.size_offset()`. This preserves offset-based + // cast compatibility and metadata selection when + // several lengths have the same rounded size. It + // also avoids choosing a length whose subsequent + // padding sequence differs from the source's. + let exact_offset_delta_elems = + match src.size_offset().checked_sub(dst.size_offset()) { + Some(delta) => { + let (elems, described_delta) = + super::max_elems_for_bytes(delta, dst_elem_size); + if described_delta == delta { + Some(elems) + } else { + None + } + } + None => None, + }; + + let offset_delta_elems = match exact_offset_delta_elems { + Some(elems) if Self::size_sequences_match(src, dst, elems) => elems, + _ => { + let src_zero_size = match src.size_for_elems(0) { + Some(size) => size, + None => return None, + }; + let elems = + match dst_layout.metadata_for_exact_size(src_zero_size) { + Some(elems) => elems, + None => return None, + }; + if !Self::size_sequences_match(src, dst, elems) { + return None; + } + elems + } + }; CastParamsInner::UnsizedToUnsized { - // SAFETY: We checked above that this is an exact ratio. + // SAFETY: `size_sequences_match` proves that + // this affine metadata map preserves size. offset_delta_elems, - // SAFETY: We checked above that this is an exact ratio. + // SAFETY: We checked above that this is an exact + // ratio of source to destination element size. elem_multiple, } } @@ -1019,40 +2160,77 @@ mod cast_from { } impl CastParams { + /// Produces `Dst` metadata describing an object of the same + /// size as the `Src` described by `src_meta`. + /// /// # Safety /// /// `src_meta` describes a `Src` whose size is no larger than /// `isize::MAX`. - /// - /// The returned metadata describes a `Dst` of the same size as - /// the original `Src`. #[inline(always)] unsafe fn cast_metadata( self, src_meta: Src::PointerMetadata, ) -> Dst::PointerMetadata { - #[allow(unused)] - use crate::util::polyfills::*; - let dst_meta = match self.inner { CastParamsInner::UnsizedToUnsized { offset_delta_elems, elem_multiple } => { let src_meta = src_meta.to_elem_count(); - #[allow( - unstable_name_collisions, - clippy::multiple_unsafe_ops_per_block - )] - // SAFETY: `self` is a witness that the following - // equation holds: + // SAFETY: `self` witnesses that this affine map + // makes `Src` and `Dst`'s complete rounded size + // formulas equal for every source metadata value. + // Interpret the arithmetic in this proof over the + // mathematical nonnegative integers. Since the + // caller promises that `src_meta` is valid `Src` + // metadata, the source object size `src_size` it + // describes is at most `isize::MAX`. Let + // `src_elem_size` and `dst_elem_size` be the + // respective trailing-slice element sizes. Then: + // + // src_meta * src_elem_size + // <= src_size <= isize::MAX. + // + // Since `elem_multiple` is the exact ratio + // `src_elem_size / dst_elem_size` and + // `dst_elem_size >= 1`, // - // D_OFF + d_meta * D_ELEM = S_OFF + s_meta * S_ELEM + // src_meta * elem_multiple + // <= src_meta * src_elem_size + // <= isize::MAX <= usize::MAX. // - // Since the caller promises that `src_meta` is - // valid `Src` metadata, this math will not - // overflow, and the returned value will describe a - // `Dst` of the same size. + // Thus `scaled_elems = src_meta * elem_multiple` + // is representable, and its byte contribution is + // `scaled_elems * dst_elem_size = src_meta * + // src_elem_size`. Let `base_bytes = + // offset_delta_elems * dst_elem_size`. + // `size_sequences_match` compares the source + // formula with the destination formula advanced by + // `base_bytes`. The latter's unrounded term + // contains `base_bytes + src_meta * src_elem_size`, + // and its complete size is at least that large + // (formally checked by Kani in + // `prove_size_formula_bounds_size_offset`). + // Sequence equality therefore gives, over the + // mathematical nonnegative integers, + // + // base_bytes + src_meta * src_elem_size + // = (offset_delta_elems + scaled_elems) + // * dst_elem_size + // <= src_size. + // + // Since `dst_elem_size >= 1`, the metadata sum + // `offset_delta_elems + scaled_elems` is also at + // most `src_size`, and thus at most `isize::MAX <= + // usize::MAX`. The sum is therefore representable, + // and the returned metadata describes a `Dst` of + // the same size. + // SAFETY: The bounds above establish the helper's + // complete no-overflow precondition. unsafe { - offset_delta_elems - .unchecked_add(src_meta.unchecked_mul(elem_multiple)) + super::add_scaled_metadata( + offset_delta_elems, + src_meta, + elem_multiple, + ) } } CastParamsInner::SizedToUnsized { dst_meta } => dst_meta, @@ -1095,6 +2273,55 @@ mod cast_from { } } +#[cfg(any(test, kani))] +mod padding_testutil { + use crate::KnownLayout; + + #[derive(KnownLayout)] + #[repr(C, align(4))] + pub(super) struct Aligned { + pub(super) prefix: Prefix, + pub(super) tail: Tail, + } + + /// Compare direct padding with the complete size minus the slice end. + /// Counts whose complete object size exceeds `usize::MAX` are omitted. + pub(super) fn check_padding_for_elems(elems: usize) -> Option + where + Target: ?Sized + KnownLayout, + { + let layout = crate::trailing_slice_layout::(); + let trailing_size = elems.checked_mul(layout.elem_size)?; + + // Evaluate the complete rounded size independently of both helpers: + // `size_for_elems` delegates to the padding method being tested. + let (alignment, phase) = layout.size_rounding_align_and_phase.components(); + #[allow(clippy::arithmetic_side_effects)] + let mask = alignment.get() - 1; + let unrounded = phase.checked_add(trailing_size)?; + let rounded = unrounded.checked_add(mask)? & !mask; + let object_size = layout.size_base.checked_add(rounded)?; + let trailing_end = layout.offset.checked_add(trailing_size).unwrap(); + let expected = object_size.checked_sub(trailing_end).unwrap(); + + let padding = layout.padding_for_elems(elems); + assert_eq!(padding, expected); + Some(padding) + } + + pub(super) fn check_layouts(elems: usize) { + // Slices have no trailing padding. The aligned fixtures cover + // zero and nonzero phases, a three-byte element whose padding + // cycles with the element count, and zero-sized elements. + let _ = check_padding_for_elems::<[u8]>(elems); + let _ = check_padding_for_elems::<[()]>(elems); + let _ = check_padding_for_elems::>(elems); + let _ = check_padding_for_elems::>(elems); + let _ = check_padding_for_elems::>(elems); + let _ = check_padding_for_elems::>(elems); + } +} + // FIXME(#67): For some reason, on our MSRV toolchain, this `allow` isn't // enforced despite having `#![allow(unknown_lints)]` at the crate root, but // putting it here works. Once our MSRV is high enough that this bug has been @@ -1103,12 +2330,560 @@ mod cast_from { #[cfg(test)] mod tests { use super::*; + use crate::PointerMetadata as _; + + /// Models the size of nested `repr(C)` structs ending in a slice. + /// + /// `leading` lists structs from outermost to innermost. Each tuple gives + /// the packing factor, minimum alignment, and prefix size before padding + /// for the final field. `trailing_elem_size` and `trailing_alignment` give + /// the slice element size and alignment. An empty `leading` describes the + /// slice itself. + /// + /// The returned closure maps a slice length to the complete size, including + /// padding at every nesting level. It returns `None` only on `usize` + /// overflow; sizes above `isize::MAX` are permitted by this arithmetic model. + /// Packing affects field placement, preserving each field's internal + /// padding, even for descriptors that Rust would reject as types. + /// + /// # Panics + /// + /// Panics if any alignment or packing factor is not a power of two, a + /// minimum alignment exceeds its packing factor, or the element size is + /// not a multiple of the slice alignment. + fn size_for_metadata_model( + leading: &[(NonZeroUsize, NonZeroUsize, usize)], + trailing_elem_size: usize, + trailing_alignment: NonZeroUsize, + ) -> impl Fn(usize) -> Option + '_ { + // Rust requires power-of-two alignments and sizes divisible by their + // alignment [1]; alignment modifiers also take powers of two [2]. + // + // [1] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#size-and-alignment + // [2] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#the-alignment-modifiers + assert!(trailing_alignment.get().is_power_of_two()); + assert_eq!(trailing_elem_size % trailing_alignment, 0); + for &(packed, align, _) in leading { + assert!(packed.get().is_power_of_two()); + assert!(align.get().is_power_of_two()); + // The descriptor requires its minimum to fit under the packing + // cap. It also models combinations that Rust forbids, including + // simultaneous `align` and `packed` modifiers [2]. + assert!(align <= packed); + } + + move |elems| { + // A slice shares its array section's layout [3]. Array size is + // element size times length, with the element's alignment [4]. + // + // [3] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#slice-layout + // [4] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#array-layout + let mut size = trailing_elem_size.checked_mul(elems)?; + let mut alignment = trailing_alignment; + + // Work outward with each field's complete size, including its + // trailing padding: an enclosing representation "does not change + // the layout of the fields themselves" [5]. Thus packing a nested + // field preserves the padding already included in `size`. + // + // [5] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#representations + for &(packed, align, prefix) in leading.iter().rev() { + // Packing caps the field's alignment for placement [2]. + let field_align = alignment.min(packed); + + // `repr(C)` places the next field at an aligned offset [6]; + // packing requires the least sufficient inter-field padding + // [2]. Here `prefix` is the end of the preceding fields. + // + // [6] https://doc.rust-lang.org/1.93.0/reference/type-layout.html#reprc-structs + let offset = prefix.checked_add(util::padding_needed_for(prefix, field_align))?; + + // `repr(C)` uses the greatest field alignment [6], and an + // explicit alignment can only raise it [2]. Here `align` + // summarizes the minimum from the prefix and any explicit + // alignment. Both operands are <= `packed`, so their maximum + // also respects the packing cap. + alignment = align.max(field_align); + + // Advance past the complete field, then round to the struct's + // alignment, following `repr(C)`'s final sizing steps [6]. + size = offset.checked_add(size)?; + size = size.checked_add(util::padding_needed_for(size, alignment))?; + } + // Each enclosing size is at least its field's size. Thus any + // checked multiplication or addition that overflowed above would + // also make the final size unrepresentable in `usize`. + Some(size) + } + } + + /// Computes [`size_for_metadata_model`]'s size function using `DstLayout`. + /// + /// Layout construction uses [`DstLayout::for_repr_c_struct`], and the + /// returned closure uses [`crate::PointerMetadata::size_for_metadata`]. + /// + /// # Panics + /// + /// Panics for invalid descriptors as in [`size_for_metadata_model`], or if + /// `DstLayout` cannot construct the layout. This includes static size + /// overflow and, in debug builds, alignments or packing factors above + /// [`DstLayout::CURRENT_MAX_ALIGN`]. + fn size_for_metadata_via_dst_layout( + leading: &[(NonZeroUsize, NonZeroUsize, usize)], + trailing_elem_size: usize, + trailing_alignment: NonZeroUsize, + ) -> impl Fn(usize) -> Option + '_ { + assert_eq!(trailing_elem_size % trailing_alignment, 0); + let mut layout = DstLayout { + align: trailing_alignment, + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 0, + elem_size: trailing_elem_size, + size_base: 0, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(trailing_alignment, 0), + }), + statically_shallow_unpadded: true, + }; + for &(packed, align, prefix) in leading.iter().rev() { + assert!(align <= packed); + let prefix = DstLayout { + align: DstLayout::MIN_ALIGN, + size_info: SizeInfo::Sized { size: prefix }, + statically_shallow_unpadded: true, + }; + layout = DstLayout::for_repr_c_struct(Some(align), Some(packed), &[prefix, layout]); + } + move |elems: usize| elems.size_for_metadata(layout) + } + + #[test] + fn test_model_via_dst_layout() { + let nz = |n| NonZeroUsize::new(n).unwrap(); + let cases: &[&[_]] = &[ + &[], + &[(nz(8), nz(1), 3)], + &[(nz(8), nz(8), 3)], + &[(nz(2), nz(1), 1), (nz(8), nz(8), 3)], + &[(nz(8), nz(8), 1), (nz(2), nz(1), 3), (nz(8), nz(8), 5)], + &[(nz(1), nz(1), 1), (nz(2), nz(2), 3), (nz(8), nz(8), 5)], + ]; + for &leading in cases { + for &(elem_size, alignment) in &[(0, 1), (0, 8), (1, 1), (2, 2), (3, 1), (8, 8)] { + let alignment = nz(alignment); + let expected = size_for_metadata_model(leading, elem_size, alignment); + let actual = size_for_metadata_via_dst_layout(leading, elem_size, alignment); + let max_elems = usize::MAX / elem_size.max(1); + for elems in (0..32).chain([ + max_elems - 1, + max_elems, + max_elems.saturating_add(1), + usize::MAX, + ]) { + assert_eq!( + actual(elems), + expected(elems), + "leading: {:?}, trailing: {:?}, elems: {}", + leading, + (elem_size, alignment), + elems, + ); + } + } + } + + for prefix in [usize::MAX - 1, usize::MAX] { + let leading = [(nz(1), nz(1), prefix)]; + let expected = size_for_metadata_model(&leading, 1, nz(1)); + let actual = size_for_metadata_via_dst_layout(&leading, 1, nz(1)); + for elems in [0, 1, 2] { + assert_eq!(actual(elems), expected(elems)); + } + } + } + + // This exercises arithmetic rather than unsafe operations; keep the + // smaller comparison test enabled under Miri instead of this large search. + #[test] + #[cfg_attr(miri, ignore)] + fn test_model_via_dst_layout_combinations() { + use rand::{rngs::SmallRng, Rng as _, SeedableRng as _}; + + fn check( + leading: &[(NonZeroUsize, NonZeroUsize, usize)], + trailing_elem_size: usize, + trailing_alignment: NonZeroUsize, + ) { + let expected = size_for_metadata_model(leading, trailing_elem_size, trailing_alignment); + assert!( + expected(0).is_some(), + "unrepresentable test layout: {:?}, {:?}", + leading, + (trailing_elem_size, trailing_alignment) + ); + let actual = std::panic::catch_unwind(|| { + size_for_metadata_via_dst_layout(leading, trailing_elem_size, trailing_alignment) + }) + .unwrap_or_else(|_| { + panic!( + "DstLayout rejected leading: {:?}, trailing: {:?}", + leading, + (trailing_elem_size, trailing_alignment) + ) + }); + let compare = |elems| { + let expected = expected(elems); + assert_eq!( + actual(elems), + expected, + "leading: {:?}, trailing: {:?}, elems: {}", + leading, + (trailing_elem_size, trailing_alignment), + elems, + ); + expected + }; + + let first_sixteen = 0..=16; + let pointer_width = (0..POINTER_WIDTH_BITS).step_by(4).flat_map(|shift| { + let elems = 1usize << shift; + [elems - 1, elems, elems + 1] + }); + let last_two = (usize::MAX - 1)..=usize::MAX; + let elemses = first_sixteen.chain(pointer_width).chain(last_two); + for elems in elemses { + let _ = compare(elems); + } + + // The modeled size is nondecreasing. Find the last fitting length, + // comparing every probe, then check both sides of that boundary. + // Zero-sized elements can remain representable at usize::MAX. + let (mut low, mut high) = (0, usize::MAX); + while low < high { + let distance = high - low; + let mid = low + distance / 2 + distance % 2; + if compare(mid).is_some() { + low = mid; + } else { + high = mid - 1; + } + } + assert!(compare(low).is_some()); + if let Some(previous) = low.checked_sub(1) { + assert!(compare(previous).is_some()); + } + if let Some(next) = low.checked_add(1) { + assert_eq!(compare(next), None); + } + } + + let nz = |n| NonZeroUsize::new(n).unwrap(); + let check_small_tails = |leading: &[_]| { + for alignment in [1, 2, 4, 8] { + for multiple in [0, 1, 2, 3, 7] { + check(leading, multiple * alignment, nz(alignment)); + } + } + }; + let mut fragments = Vec::new(); + for packed in [1, 2, 4] { + for align in [1, 2, 4].iter().copied().filter(|&align| align <= packed) { + for prefix in 0..=4 { + fragments.push((nz(packed), nz(align), prefix)); + } + } + } + + // Exhaust every combination in this small domain at depths 0, 1, 2. + check_small_tails(&[]); + for &outer in &fragments { + check_small_tails(&[outer]); + for &inner in &fragments { + check_small_tails(&[outer, inner]); + } + } + + let mut rng = SmallRng::seed_from_u64(0); + for depth in (0usize..=128).chain([256, 512, 1024]) { + // A small prefix adds at most 4 * max_align bytes per layer. + // Reserve half of usize::MAX for a large outermost prefix so + // construction remains representable even at the largest depth. + let align_limit = + DstLayout::CURRENT_MAX_ALIGN.get().min(usize::MAX / (8 * depth.max(1))); + let alignments: Vec<_> = (0..POINTER_WIDTH_BITS) + .map(|shift| 1usize << shift) + .take_while(|&align| align <= align_limit) + .collect(); + let last = alignments.len() - 1; + let max_align = alignments[last]; + for case in 0..40 { + let mut leading = Vec::with_capacity(depth); + for level in 0..depth { + // Mix arbitrary modifiers, increasing and decreasing + // alignments, alternating packing, and fully packed trees. + let (packed, align) = match case / 8 { + 0 => { + let packed = rng.gen_range(0..=last); + (packed, rng.gen_range(0..=packed)) + } + 1 => (last, level % alignments.len()), + 2 => (last, last - level % alignments.len()), + 3 => { + let index = if level % 2 == 0 { 0 } else { last }; + (index, index) + } + _ => (0, 0), + }; + let (packed, align) = (alignments[packed], alignments[align]); + let prefix = if level == 0 && case % 4 == 3 { + usize::MAX / 2 + rng.gen_range(0..=2) - 1 + } else { + match rng.gen_range(0..8) { + 0 => 0, + 1 => 1, + 2 => align - 1, + 3 => align, + 4 => align + 1, + 5 => packed - 1, + 6 => packed + 1, + _ => rng.gen_range(0..=2 * max_align), + } + }; + leading.push((nz(packed), nz(align), prefix)); + } + let alignment = alignments[rng.gen_range(0..=last)]; + let elem_size = match case % 8 { + 0 => 0, + 1 => alignment, + 2 => 2 * alignment, + 3 => 3 * alignment, + 4 => 7 * alignment, + 5 => 15 * alignment, + 6 => (usize::MAX / 2 / alignment) * alignment, + _ => (usize::MAX / alignment) * alignment, + }; + check(&leading, elem_size, nz(alignment)); + } + } + } + + const TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_ALIGN: NonZeroUsize = match NonZeroUsize::new(8) { + Some(align) => align, + None => const_unreachable!(), + }; + const TEST_SIZE_ROUNDING_ALIGN_AND_PHASE: RoundingAlignAndPhase = + RoundingAlignAndPhase::new(TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_ALIGN, 3); + const TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_DECODED_ALIGN: usize = + TEST_SIZE_ROUNDING_ALIGN_AND_PHASE.align().get(); + const TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_DECODED_PHASE: usize = + TEST_SIZE_ROUNDING_ALIGN_AND_PHASE.components().1; + + #[test] + #[allow(clippy::arithmetic_side_effects)] + fn test_size_rounding_align_and_phase_encoding() { + // These constants ensure that encoding and decoding remain evaluable + // on every supported compiler, including the MSRV. + assert_eq!(TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_DECODED_ALIGN, 8); + assert_eq!(TEST_SIZE_ROUNDING_ALIGN_AND_PHASE_DECODED_PHASE, 3); + + for shift in 0..POINTER_WIDTH_BITS { + #[allow(clippy::arithmetic_side_effects)] + let align = NonZeroUsize::new(1usize << shift).unwrap(); + for phase in [0, align.get() / 2, align.get() - 1] { + let rounding = RoundingAlignAndPhase::new(align, phase); + assert_eq!(rounding.align(), align); + assert_eq!(rounding.components().1, phase); + assert_eq!(rounding.0.get(), align.get() | phase); + } + } + + let max_align = DstLayout::THEORETICAL_MAX_ALIGN; + let max_phase = max_align.get() - 1; + let max = RoundingAlignAndPhase::new(max_align, max_phase); + assert_eq!(max.0.get(), usize::MAX); + assert_eq!(max.align(), max_align); + assert_eq!(max.components().1, max_phase); + + assert_eq!(mem::size_of::(), 4 * mem::size_of::()); + } + + #[test] + #[allow(clippy::arithmetic_side_effects)] + fn test_trailing_slice_layout_normal_form() { + // Compare the stored base-and-phase formula with + // `prefix + round_up(offset + elems * elem_size, align)`, where + // `prefix < align`. Move whole alignment-sized chunks of `offset` + // into `size_base` and retain the remainder as the phase. Also verify + // that `size_offset()` recovers the exact `offset` used by casts to + // choose between metadata values with the same rounded object size. + for align in [1, 2, 4, 8, 16] { + let align = NonZeroUsize::new(align).unwrap(); + for prefix in 0..align.get() { + for offset in 0..32 { + let phase = offset % align; + let size_base = (offset - phase) | prefix; + let layout = TrailingSliceLayout { + offset: 0, + elem_size: 3, + size_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(align, phase), + }; + assert_eq!(layout.size_offset(), offset); + + for elems in 0..32 { + let unrounded = offset + elems * layout.elem_size; + let old_size = + prefix + unrounded + util::padding_needed_for(unrounded, align); + assert_eq!(layout.size_for_elems(elems), Some(old_size)); + } + } + } + } + } + + #[test] + #[allow(clippy::arithmetic_side_effects)] + fn test_normalized_size_bounds() { + // Use checked evaluation as an independent oracle, including layouts + // whose empty object is already too large to represent. + let checked_size = |layout: TrailingSliceLayout, trailing_bytes: usize| { + let (alignment, phase) = layout.size_rounding_align_and_phase.components(); + let unrounded = phase.checked_add(trailing_bytes)?; + let rounded = unrounded.checked_add(util::padding_needed_for(unrounded, alignment))?; + layout.size_base.checked_add(rounded) + }; + for exponent in 0..POINTER_WIDTH_BITS { + let alignment = NonZeroUsize::new(1usize << exponent).unwrap(); + for phase in [0, alignment.get() / 2, alignment.get() - 1] { + for base in [0, 1, alignment.get() - 1, alignment.get(), usize::MAX - 1, usize::MAX] + { + let mut layout = TrailingSliceLayout { + offset: 0, + elem_size: 1, + size_base: base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(alignment, phase), + }; + for available in [0, 1, alignment.get() - 1, alignment.get(), usize::MAX] { + match layout.max_trailing_bytes(available) { + Some(max_bytes) => { + assert!(checked_size(layout, max_bytes).unwrap() <= available); + if let Some(next_bytes) = max_bytes.checked_add(1) { + assert!(checked_size(layout, next_bytes) + .map(|size| size > available) + .unwrap_or(true)); + } + } + None => { + assert!(checked_size(layout, 0) + .map(|size| size > available) + .unwrap_or(true)); + } + } + } + + for elem_size in [0, 1, 3, alignment.get(), usize::MAX] { + layout.elem_size = elem_size; + let max_elems = layout + .max_trailing_bytes(usize::MAX) + .and_then(|bytes| bytes.checked_div(elem_size)) + .unwrap_or(usize::MAX); + for elems in [ + 0, + 1, + max_elems.saturating_sub(1), + max_elems, + max_elems.saturating_add(1), + usize::MAX, + ] { + let expected = elems + .checked_mul(elem_size) + .and_then(|bytes| checked_size(layout, bytes)); + // The physical slice offset does not affect + // the size formula. Raw layouts may place the + // slice beyond the object, and adding its byte + // count to the offset may overflow. + for offset in [0, usize::MAX] { + layout.offset = offset; + assert_eq!(layout.size_for_elems(elems), expected); + } + } + } + } + } + } + } + + #[test] + fn test_padding_for_elems_for_rust_layouts() { + use super::padding_testutil::{check_layouts, check_padding_for_elems, Aligned}; + + for elems in 0..256 { + check_layouts(elems); + } + let size_limit = usize::MAX; + for remaining in 0..16 { + check_layouts(size_limit - remaining); + check_layouts(size_limit / 3 - remaining); + } + check_layouts(usize::MAX); + + for (elems, expected) in [(0, 3), (1, 0), (2, 1), (3, 2)] { + assert_eq!( + check_padding_for_elems::>(elems), + Some(expected) + ); + } + // Every count is valid for zero-sized trailing elements. + assert_eq!(check_padding_for_elems::>(usize::MAX), Some(3)); + } + + #[test] + fn test_padding_for_elems() { + for exponent in 0..POINTER_WIDTH_BITS { + let alignment = NonZeroUsize::new(1usize << exponent).unwrap(); + let mask = alignment.get() - 1; + for phase in [0, alignment.get() / 2, mask] { + // Include zero-sized elements, a negative fixed term, fixed + // padding larger than the alignment, and overflowing sums + // and products. Raw layouts need not describe Rust types. + for (offset, elem_size, size_base) in [ + (0, 0, 0), + (0, 1, 1), + (9, 3, 8), + (3, 3, 0), + (9, 3, 12), + (usize::MAX, usize::MAX, usize::MAX), + (usize::MAX, 0, usize::MAX), + ] { + let layout = TrailingSliceLayout { + offset, + elem_size, + size_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(alignment, phase), + }; + for elems in [0, 1, 2, 3, alignment.get(), usize::MAX] { + // Independently reconstruct the complete rounded + // size and slice end, then subtract. Round up by + // adding the mask and clearing the low bits. + let trailing_bytes = elems.wrapping_mul(elem_size); + let unrounded = phase.wrapping_add(trailing_bytes); + let rounded = unrounded.wrapping_add(mask) & !mask; + let object_size = size_base.wrapping_add(rounded); + let trailing_end = offset.wrapping_add(trailing_bytes); + assert_eq!( + layout.padding_for_elems(elems), + object_size.wrapping_sub(trailing_end), + ); + } + } + } + } + } #[test] fn test_dst_layout_for_slice() { let layout = DstLayout::for_slice::(); match layout.size_info { - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) => { + SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size, .. }) => { assert_eq!(offset, 0); assert_eq!(elem_size, 4); } @@ -1121,17 +2896,18 @@ mod tests { #[allow(clippy::decimal_literal_representation)] #[test] fn test_dst_layout_extend_sized_with_sized() { - // This macro constructs a layout corresponding to a `u8` and extends it - // with a zero-sized trailing field of given alignment `n`. The macro - // tests that the resulting layout has both size and alignment `min(n, - // P)` for all valid values of `repr(packed(P))`. + // This macro constructs a layout corresponding to a `u8` and extends + // it with a zero-sized trailing field of the specified alignment. The + // resulting size and alignment must both be the field's alignment + // capped by the packing limit. Exercise every supported packing limit + // as well as the absence of a packing limit. macro_rules! test_align_is_size { ($n:expr) => { let base = DstLayout::for_type::(); let trailing_field = DstLayout::for_type::>(); - let packs = - core::iter::once(None).chain((0..29).map(|p| NonZeroUsize::new(2usize.pow(p)))); + let packs = core::iter::once(None) + .chain((0..=29).map(|p| NonZeroUsize::new(2usize.pow(p)))); for pack in packs { let composite = base.extend(trailing_field, pack); @@ -1179,6 +2955,7 @@ mod tests { test_align_is_size!(33554432); test_align_is_size!(134217728); test_align_is_size!(268435456); + test_align_is_size!(536870912); } /// Tests of when a sized `DstLayout` is extended with a DST field. @@ -1192,15 +2969,29 @@ mod tests { let aligns = (0..29).map(|p| NonZeroUsize::new(2usize.pow(p)).unwrap()); let packs = core::iter::once(None).chain(aligns.clone().map(Some)); - for align in aligns { + for field_type_align in aligns { for pack in packs.clone() { let base = DstLayout::for_type::(); let elem_size = 42; let trailing_field_offset = 11; + #[allow(clippy::arithmetic_side_effects)] + let trailing_phase = trailing_field_offset % field_type_align; + let trailing_base = util::round_down_to_next_multiple_of_alignment( + trailing_field_offset, + field_type_align, + ); let trailing_field = DstLayout { - align, - size_info: SizeInfo::SliceDst(TrailingSliceLayout { elem_size, offset: 11 }), + align: field_type_align, + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + elem_size, + offset: trailing_field_offset, + size_base: trailing_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + field_type_align, + trailing_phase, + ), + }), statically_shallow_unpadded: false, }; @@ -1208,7 +2999,9 @@ mod tests { let max_align = pack.unwrap_or(DstLayout::CURRENT_MAX_ALIGN).get(); - let align = align.get().min(max_align); + let align = field_type_align.get().min(max_align); + let field_offset = align; + let size_base = field_offset + trailing_base; assert_eq!( composite, @@ -1216,7 +3009,12 @@ mod tests { align: NonZeroUsize::new(align).unwrap(), size_info: SizeInfo::SliceDst(TrailingSliceLayout { elem_size, - offset: align + trailing_field_offset, + offset: field_offset + trailing_field_offset, + size_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + field_type_align, + trailing_phase, + ), }), statically_shallow_unpadded: false, } @@ -1292,7 +3090,8 @@ mod tests { => padded { size: current_max_align * 2, align: current_max_align }); } - /// Tests that calling `pad_to_align` on a DST `DstLayout` is a no-op. + /// Tests that calling `pad_to_align` on a DST `DstLayout` incorporates the + /// type's trailing-padding operation into its runtime size formula. #[test] fn test_dst_layout_pad_to_align_with_dst() { for align in (0..29).map(|p| NonZeroUsize::new(2usize.pow(p)).unwrap()) { @@ -1300,42 +3099,183 @@ mod tests { for elem_size in 0..10 { let layout = DstLayout { align, - size_info: SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base: offset, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + DstLayout::MIN_ALIGN, + 0, + ), + }), statically_shallow_unpadded: false, }; - assert_eq!(layout.pad_to_align(), layout); + assert_eq!( + layout.pad_to_align(), + DstLayout { + align, + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base: offset - offset % align, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + align, + offset % align + ), + }), + statically_shallow_unpadded: false, + } + ); } } } - } - - // This test takes a long time when running under Miri, so we skip it in - // that case. This is acceptable because this is a logic test that doesn't - // attempt to expose UB. - #[test] - #[cfg_attr(miri, ignore)] - fn test_validate_cast_and_convert_metadata() { - #[allow(non_local_definitions)] - impl From for SizeInfo { - fn from(size: usize) -> SizeInfo { - SizeInfo::Sized { size } - } - } - - #[allow(non_local_definitions)] - impl From<(usize, usize)> for SizeInfo { - fn from((offset, elem_size): (usize, usize)) -> SizeInfo { - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) - } - } - fn layout>(s: S, align: usize) -> DstLayout { + // Exercise the branch where the existing inner rounding alignment is + // at least the enclosing alignment. The inner rounded term is already + // aligned to two, so padding only advances `size_base` from three to + // four. + let inner_rounding_dominates = DstLayout { + align: NonZeroUsize::new(2).unwrap(), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 3, + elem_size: 1, + size_base: 3, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 1, + ), + }), + statically_shallow_unpadded: false, + }; + assert_eq!( + inner_rounding_dominates.pad_to_align(), DstLayout { - size_info: s.into(), - align: NonZeroUsize::new(align).unwrap(), + align: NonZeroUsize::new(2).unwrap(), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 3, + elem_size: 1, + size_base: 4, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 1 + ), + }), statically_shallow_unpadded: false, } - } + ); + + // Exercise the branch where the enclosing alignment is larger. For + // a trailing-slice byte count `trailing_size`, rounding + // `6 + round_up(1 + trailing_size, 4)` to eight gives the equivalent + // formula `8 + round_up(1 + trailing_size, 8)`. + let outer_rounding_dominates = DstLayout { + align: NonZeroUsize::new(8).unwrap(), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 7, + elem_size: 1, + size_base: 6, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 1, + ), + }), + statically_shallow_unpadded: false, + }; + assert_eq!( + outer_rounding_dominates.pad_to_align(), + DstLayout { + align: NonZeroUsize::new(8).unwrap(), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 7, + elem_size: 1, + size_base: 8, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(8).unwrap(), + 1 + ), + }), + statically_shallow_unpadded: false, + } + ); + } + + #[test] + fn test_trailing_slice_layout_size_sequence_equivalence() { + // These are the two normalized descriptions produced by a + // `repr(C, packed(2))` wrapper around slice DSTs whose element types + // both have size four but alignments four and two respectively. Their + // formulas differ structurally, but both simplify to `2 + 4 * elems`. + let align4 = TrailingSliceLayout { + offset: 2, + elem_size: 4, + size_base: 2, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 0, + ), + }; + let align2 = TrailingSliceLayout { + offset: 2, + elem_size: 4, + size_base: 2, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(2).unwrap(), + 0, + ), + }; + assert!(align4.has_same_size_sequence(align2)); + for elems in 0..8 { + assert_eq!(align4.size_for_elems(elems), Some(2 + 4 * elems)); + assert_eq!(align4.size_for_elems(elems), align2.size_for_elems(elems)); + } + + let dynamically_padded = TrailingSliceLayout { elem_size: 1, ..align4 }; + assert!(!dynamically_padded.has_same_size_sequence(align2)); + } + + // This test takes a long time when running under Miri, so we skip it in + // that case. This is acceptable because this is a logic test that doesn't + // attempt to expose UB. + #[test] + #[cfg_attr(miri, ignore)] + fn test_validate_cast_and_convert_metadata() { + #[allow(non_local_definitions)] + impl From for SizeInfo { + fn from(size: usize) -> SizeInfo { + SizeInfo::Sized { size } + } + } + + #[allow(non_local_definitions)] + impl From<(usize, usize)> for SizeInfo { + fn from((offset, elem_size): (usize, usize)) -> SizeInfo { + SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base: offset, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + DstLayout::MIN_ALIGN, + 0, + ), + }) + } + } + + fn layout>(s: S, align: usize) -> DstLayout { + let align = NonZeroUsize::new(align).unwrap(); + let size_info = match s.into() { + SizeInfo::SliceDst(mut trailing) => { + let phase = trailing.size_base % align; + trailing.size_base = + util::round_down_to_next_multiple_of_alignment(trailing.size_base, align); + trailing.size_rounding_align_and_phase = + RoundingAlignAndPhase::new(align, phase); + SizeInfo::SliceDst(trailing) + } + size_info => size_info, + }; + DstLayout { size_info, align, statically_shallow_unpadded: false } + } /// This macro accepts arguments in the form of: /// @@ -1531,10 +3471,18 @@ mod tests { let resulting_size = match layout.size_info { SizeInfo::Sized { size } => size, - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) => { + SizeInfo::SliceDst(TrailingSliceLayout { + elem_size, + size_base, + size_rounding_align_and_phase, + .. + }) => { + let (size_align, size_phase) = size_rounding_align_and_phase.components(); let padded_size = |elems| { - let without_padding = offset + elems * elem_size; - without_padding + util::padding_needed_for(without_padding, align) + let without_padding = size_phase + elems * elem_size; + size_base + + without_padding + + util::padding_needed_for(without_padding, size_align) }; let resulting_size = padded_size(elems); @@ -1562,8 +3510,13 @@ mod tests { } else { let min_size = match layout.size_info { SizeInfo::Sized { size } => size, - SizeInfo::SliceDst(TrailingSliceLayout { offset, .. }) => { - offset + util::padding_needed_for(offset, layout.align) + SizeInfo::SliceDst(TrailingSliceLayout { + size_base, + size_rounding_align_and_phase, + .. + }) => { + let (size_align, size_phase) = size_rounding_align_and_phase.components(); + size_base + size_phase + util::padding_needed_for(size_phase, size_align) } }; @@ -1592,6 +3545,35 @@ mod tests { .map(|(size_info, align)| layout(size_info, align)); itertools::iproduct!(layouts, 0..8, 0..8, [CastType::Prefix, CastType::Suffix]) .for_each(validate_behavior); + + // Exercise a normalized formula which cannot be represented as + // `round_up(offset + elems * elem_size, align)`: the physical trailing + // slice begins at offset 7, while an inner aligned DST contributes the + // size formula `2 + round_up(5 + elems, 4)` to a packed outer type. + let nested_packed = DstLayout { + align: NonZeroUsize::new(2).unwrap(), + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset: 7, + elem_size: 1, + size_base: 6, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 1, + ), + }), + statically_shallow_unpadded: false, + }; + itertools::iproduct!([nested_packed], 0..8, 0..24, [CastType::Prefix, CastType::Suffix]) + .for_each(validate_behavior); + + assert!(matches!( + nested_packed.validate_cast_and_convert_metadata(0, 8, CastType::Prefix), + Err(MetadataCastError::Size) + )); + assert!(matches!( + nested_packed.validate_cast_and_convert_metadata(0, 10, CastType::Prefix), + Ok((3, 10)) + )); } #[test] @@ -1631,9 +3613,18 @@ mod tests { let dst = args.elem_size.is_some(); let layout = { let size_info = match args.elem_size { - Some(elem_size) => { - SizeInfo::SliceDst(TrailingSliceLayout { offset: args.offset, elem_size }) - } + Some(elem_size) => SizeInfo::SliceDst(TrailingSliceLayout { + offset: args.offset, + elem_size, + size_base: util::round_down_to_next_multiple_of_alignment( + args.offset, + args.align, + ), + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + args.align, + args.offset % args.align, + ), + }), None => SizeInfo::Sized { // Rust only supports types whose sizes are a multiple // of their alignment. If the macro created a type like @@ -1951,70 +3942,750 @@ mod tests { #[cfg(kani)] mod proofs { + //! Arithmetic models and input generators for the layout proofs. + //! + //! The contract models specify expected results without calling the + //! operation being checked. The composition models (`extended_layout`, + //! `padded_layout`, and `repr_c_layout`) also define the contracts' input + //! domains: `Some` supplies an expected layout, while `None` excludes an + //! input from the corresponding non-panicking contract. These models check + //! representability in `usize`, without imposing Rust's `isize::MAX` bound. + use core::alloc::Layout; use super::*; - impl kani::Arbitrary for DstLayout { - fn any() -> Self { - let align: NonZeroUsize = kani::any(); - let size_info: SizeInfo = kani::any(); - - kani::assume(align.is_power_of_two()); - kani::assume(align < DstLayout::THEORETICAL_MAX_ALIGN); - - // For testing purposes, we most care about instantiations of - // `DstLayout` that can correspond to actual Rust types. We use - // `Layout` to verify that our `DstLayout` satisfies the validity - // conditions of Rust layouts. - kani::assume( - match size_info { - SizeInfo::Sized { size } => Layout::from_size_align(size, align.get()), - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size: _ }) => { - // `SliceDst` cannot encode an exact size, but we know - // it is at least `offset` bytes. - Layout::from_size_align(offset, align.get()) + /// Evaluates the normalized object size for a trailing slice of + /// `trailing_size` bytes: + /// + /// ```text + /// size_base + round_up(size_phase + trailing_size, size_align) + /// ``` + /// + /// Here, `size_align` and `size_phase` come from `layout.size_rounding_align_and_phase`. + /// `round_up` rounds its input up to a multiple of its alignment. Returns + /// `None` if the size is not representable in `usize`. + /// + /// Using byte counts separates rounding from element-count multiplication + /// and permits proofs over byte counts that are not whole elements. + pub(super) fn slice_dst_size_for_trailing_bytes( + layout: TrailingSliceLayout, + trailing_size: usize, + ) -> Option { + let without_padding = + layout.size_rounding_align_and_phase.components().1.checked_add(trailing_size)?; + let padding = + util::padding_needed_for(without_padding, layout.size_rounding_align_and_phase.align()); + let rounded = without_padding.checked_add(padding)?; + layout.size_base.checked_add(rounded) + } + + /// Returns the least multiple of `align` greater than or equal to `bytes`, + /// or `None` if that multiple exceeds `usize::MAX`. + /// + /// Requires a power-of-two `align`. Composition models use this checked + /// operation to identify inputs for which adding padding would overflow. + fn round_up(bytes: usize, align: NonZeroUsize) -> Option { + bytes.checked_add(util::padding_needed_for(bytes, align)) + } + + /// Models the largest trailing-slice byte count whose object size fits in + /// `available` bytes. Returns `None` if even an empty trailing slice cannot + /// fit. The result need not be a multiple of the element size. + /// + /// Subtracting `size_base` leaves the budget for the rounded contribution. + /// Rounding that budget down to `size_align` and subtracting `size_phase` + /// gives the largest byte count whose `slice_dst_size_for_trailing_bytes` + /// result is at most `available`. Metadata and casting contracts use + /// this bound without calling the production capacity calculation. + pub(super) fn max_trailing_bytes( + trailing: TrailingSliceLayout, + available: usize, + ) -> Option { + let (align, phase) = trailing.size_rounding_align_and_phase.components(); + let capacity = available.checked_sub(trailing.size_base)?; + util::round_down_to_next_multiple_of_alignment(capacity, align).checked_sub(phase) + } + + /// Models appending `field` to a sized `prefix` for `DstLayout::extend`. + /// + /// Places the field at the first offset after the prefix that satisfies the + /// field's alignment, capped by `packed` when present. The resulting + /// alignment is the maximum of the prefix's alignment and this effective + /// field alignment. A DST field retains its inner rounding operation, with + /// its slice offset and size base shifted by the field's placement. + /// Trailing padding for the enclosing layout is left to `padded_layout`. + /// + /// Returns `None` for a non-power-of-two alignment, an unsized prefix, or + /// arithmetic overflow. The `statically_shallow_unpadded` flag is true only + /// if both inputs have it set and field placement introduces no gap. + pub(super) fn extended_layout( + prefix: DstLayout, + field: DstLayout, + packed: Option, + ) -> Option { + if !prefix.align.is_power_of_two() + || !field.align.is_power_of_two() + || packed.map_or(false, |align| !align.is_power_of_two()) + { + return None; + } + let SizeInfo::Sized { size: prefix_size } = prefix.size_info else { return None }; + let field_align = field.align.min(packed.unwrap_or(DstLayout::THEORETICAL_MAX_ALIGN)); + let field_offset = round_up(prefix_size, field_align)?; + let size_info = match field.size_info { + SizeInfo::Sized { size } => SizeInfo::Sized { size: field_offset.checked_add(size)? }, + SizeInfo::SliceDst(trailing) => SizeInfo::SliceDst(TrailingSliceLayout { + offset: field_offset.checked_add(trailing.offset)?, + size_base: field_offset.checked_add(trailing.size_base)?, + ..trailing + }), + }; + Some(DstLayout { + align: prefix.align.max(field_align), + size_info, + statically_shallow_unpadded: prefix.statically_shallow_unpadded + && field.statically_shallow_unpadded + && field_offset == prefix_size, + }) + } + + /// Models adding trailing padding for `DstLayout::pad_to_align`. + /// + /// A sized layout rounds its size up to a multiple of `layout.align`. A DST + /// instead composes that rounding with its existing size expression: + /// + /// ```text + /// round_up(size_base + round_up(size_phase + trailing_size, size_align), + /// layout.align) + /// ``` + /// + /// Here, `size_align` and `size_phase` come from the DST's + /// `size_rounding_align_and_phase`. The result encodes this expression with + /// one base, phase, and rounding alignment, preserving the physical slice + /// offset and element size. The `statically_shallow_unpadded` flag is + /// cleared only when a sized layout gains padding; DST padding is accounted + /// for by the size expression. + /// + /// Returns `None` if `layout.align` is not a power of two or a checked + /// intermediate overflows. The contract compares this expected encoding + /// with the production result; the separate padding formula proofs check + /// that the encoding preserves the rounded size. + pub(super) fn padded_layout(layout: DstLayout) -> Option { + if !layout.align.is_power_of_two() { + return None; + } + let (size_info, no_static_padding) = match layout.size_info { + SizeInfo::Sized { size } => { + let rounded = round_up(size, layout.align)?; + (SizeInfo::Sized { size: rounded }, rounded == size) + } + SizeInfo::SliceDst(trailing) => { + let (inner_align, phase) = trailing.size_rounding_align_and_phase.components(); + let (size_base, size_rounding_align_and_phase) = if inner_align >= layout.align { + ( + round_up(trailing.size_base, layout.align)?, + trailing.size_rounding_align_and_phase, + ) + } else { + let fixed = round_up(trailing.size_base, inner_align)?.checked_add(phase)?; + let new_phase = fixed & (layout.align.get() - 1); + (fixed - new_phase, RoundingAlignAndPhase::new(layout.align, new_phase)) + }; + ( + SizeInfo::SliceDst(TrailingSliceLayout { + size_base, + size_rounding_align_and_phase, + ..trailing + }), + true, + ) + } + }; + Some(DstLayout { + size_info, + statically_shallow_unpadded: layout.statically_shallow_unpadded && no_static_padding, + ..layout + }) + } + + /// Models a complete layout for `DstLayout::for_repr_c_struct` by composing + /// the field-placement and trailing-padding specifications. + /// + /// Starts with an empty prefix of alignment `align`, or one if absent. + /// Appends fields in order, with `packed` as an optional cap on field + /// alignment; the final layout is padded to its resulting alignment. Only + /// the last field may be unsized, since field extension requires a sized + /// prefix. + /// + /// Returns `None` if the initial alignment is not a power of two or any + /// field extension or final padding step returns `None`. + pub(super) fn repr_c_layout( + align: Option, + packed: Option, + fields: &[DstLayout], + ) -> Option { + let align = align.unwrap_or(DstLayout::MIN_ALIGN); + if !align.is_power_of_two() { + return None; + } + let mut layout = DstLayout { + align, + size_info: SizeInfo::Sized { size: 0 }, + statically_shallow_unpadded: true, + }; + for &field in fields { + layout = extended_layout(layout, field, packed)?; + } + padded_layout(layout) + } + + // The inline contract harnesses use unrestricted `Arbitrary` values. + // These generators retain the narrower domains of the Rust-layout proofs. + + /// Generates any power-of-two alignment representable in `usize`, including + /// `DstLayout::THEORETICAL_MAX_ALIGN`. This restricts the alignment alone; + /// callers must impose any constraints relating it to a layout's size. + fn any_layout_align() -> NonZeroUsize { + let exponent: u8 = kani::any(); + kani::assume(usize::from(exponent) < POINTER_WIDTH_BITS); + + // `exponent < POINTER_WIDTH_BITS`, so the shift is in range and its + // result is non-zero. This construction is surjective over every + // power-of-two alignment up to and including + // `THEORETICAL_MAX_ALIGN`. `Layout` admits that maximum alignment for + // zero-sized layouts, so the validity assumptions below—not this + // generator—filter combinations that cannot describe Rust types. + NonZeroUsize::new(1usize << exponent).unwrap() + } + + /// Generates layouts satisfying the arithmetic restrictions used by the + /// standalone Rust-layout proofs. + /// + /// The fixed size, or the zero-element DST size, must be accepted by + /// `Layout::from_size_align`. For a DST, `size_base + size_phase` must also + /// be representable and at least the physical slice offset. Component + /// bounds come from `any_bounded_size_info`; `statically_shallow_unpadded` + /// remains arbitrary. These conditions do not establish that a Rust type + /// with this layout exists. + fn any_valid_dst_layout() -> DstLayout { + let align = any_layout_align(); + let size_info = any_bounded_size_info(); + + kani::assume( + match size_info { + SizeInfo::Sized { size } => Layout::from_size_align(size, align.get()), + SizeInfo::SliceDst(trailing) => { + // `SliceDst` cannot encode one exact size. Validate its + // minimum size and the invariants required of the size + // formula for layouts of real Rust types. + let min_size = trailing.size_for_elems(0); + let unrounded_min_size = trailing + .size_base + .checked_add(trailing.size_rounding_align_and_phase.components().1); + + kani::assume(matches!( + unrounded_min_size, + Some(size) if size >= trailing.offset + )); + + match min_size { + Some(min_size) => Layout::from_size_align(min_size, align.get()), + None => Layout::from_size_align(usize::MAX, align.get()), } } - .is_ok(), - ); + } + .is_ok(), + ); - Self { align: align, size_info: size_info, statically_shallow_unpadded: kani::any() } + DstLayout { align: align, size_info: size_info, statically_shallow_unpadded: kani::any() } + } + + /// Generates either a fixed size at most `DstLayout::MAX_SIZE`, or a DST + /// description from `any_bounded_trailing_layout`. + /// + /// Supplies bounded size components to `any_valid_dst_layout`, which adds + /// constraints involving the enclosing layout's alignment. + fn any_bounded_size_info() -> SizeInfo { + let is_sized: bool = kani::any(); + + match is_sized { + true => { + let size: usize = kani::any(); + + kani::assume(size <= DstLayout::MAX_SIZE); + + SizeInfo::Sized { size } + } + false => SizeInfo::SliceDst(any_bounded_trailing_layout()), } } - impl kani::Arbitrary for SizeInfo { - fn any() -> Self { - let is_sized: bool = kani::any(); + /// Generates a trailing-slice description whose element size, slice offset, + /// and size base are each at most `DstLayout::MAX_SIZE` (`isize::MAX`). + /// + /// The rounding alignment spans all representable powers of two, and the + /// phase spans all values below that alignment. No further relationships + /// between these components are imposed: their combined size may overflow + /// or fail to contain the slice. Standalone proofs add the relationships + /// they require; contract harnesses use unrestricted `Arbitrary` values + /// instead. + fn any_bounded_trailing_layout() -> TrailingSliceLayout { + let elem_size: usize = kani::any(); + let offset: usize = kani::any(); + let size_base: usize = kani::any(); + let size_align = any_layout_align(); + let raw_size_phase: usize = kani::any(); + // Since `size_align` is a power of two, masking is surjective over + // precisely the values in `0..size_align`. + #[allow(clippy::arithmetic_side_effects)] + let size_phase = raw_size_phase & (size_align.get() - 1); + + kani::assume(elem_size <= DstLayout::MAX_SIZE); + kani::assume(offset <= DstLayout::MAX_SIZE); + kani::assume(size_base <= DstLayout::MAX_SIZE); - match is_sized { - true => { - let size: usize = kani::any(); + TrailingSliceLayout { + elem_size, + offset, + size_base, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(size_align, size_phase), + } + } - kani::assume(size <= DstLayout::MAX_SIZE); + #[kani::proof] + fn prove_padding_for_elems_for_rust_layouts() { + // For each fixture, check every element count whose complete size + // fits in `usize`, including all counts for zero-sized elements. + padding_testutil::check_layouts(kani::any()); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_padding_for_elems() { + // Check the modular result for every field value and element count, + // including layouts that cannot describe a Rust type. + let layout = TrailingSliceLayout { + offset: kani::any(), + elem_size: kani::any(), + size_base: kani::any(), + size_rounding_align_and_phase: RoundingAlignAndPhase(kani::any()), + }; + let elems: usize = kani::any(); + let (size_align, size_phase) = layout.size_rounding_align_and_phase.components(); + let mask = size_align.get() - 1; + + // Evaluate the complete size and trailing-slice end separately in + // modular arithmetic, without reducing the trailing bytes first or + // canceling them from the difference. + let trailing_bytes = elems.wrapping_mul(layout.elem_size); + let unrounded = size_phase.wrapping_add(trailing_bytes); + let rounded = unrounded.wrapping_add(mask) & !mask; + let object_size = layout.size_base.wrapping_add(rounded); + let trailing_end = layout.offset.wrapping_add(trailing_bytes); + assert_eq!(layout.padding_for_elems(elems), object_size.wrapping_sub(trailing_end)); + } + + #[kani::proof] + fn prove_size_rounding_align_and_phase_encoding() { + let encoded: usize = kani::any(); + kani::assume(encoded != 0); + let encoded = NonZeroUsize::new(encoded).unwrap(); + let rounding = RoundingAlignAndPhase(encoded); + + let align = rounding.align(); + let phase = rounding.components().1; + assert!(align.is_power_of_two()); + assert!(phase < align.get()); + assert_eq!(RoundingAlignAndPhase::new(align, phase), rounding); + } + + #[kani::proof] + fn prove_size_rounding_align_and_phase_decoding() { + let align: NonZeroUsize = kani::any(); + let phase: usize = kani::any(); + kani::assume(align.is_power_of_two()); + kani::assume(phase < align.get()); + + let rounding = RoundingAlignAndPhase::new(align, phase); + assert_eq!(rounding.components(), (align, phase)); + } + + #[kani::proof] + fn prove_trailing_slice_layout_normal_form() { + let rounding_word: usize = kani::any(); + kani::assume(rounding_word != 0); + let align = RoundingAlignAndPhase(NonZeroUsize::new(rounding_word).unwrap()).align(); + + let raw_prefix: usize = kani::any(); + let offset: usize = kani::any(); + let trailing_size: usize = kani::any(); + #[allow(clippy::arithmetic_side_effects)] + let prefix = raw_prefix & (align.get() - 1); + #[allow(clippy::arithmetic_side_effects)] + let phase = offset & (align.get() - 1); + #[allow(clippy::arithmetic_side_effects)] + let aligned_offset = offset - phase; + let layout = TrailingSliceLayout { + offset: 0, + elem_size: 1, + size_base: aligned_offset | prefix, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(align, phase), + }; + + assert_eq!(layout.size_offset(), offset); + + let old_size = offset + .checked_add(trailing_size) + .and_then(|without_padding| { + without_padding.checked_add(util::padding_needed_for(without_padding, align)) + }) + .and_then(|rounded| prefix.checked_add(rounded)); + assert_eq!(layout.size_for_elems(trailing_size), old_size); + assert_eq!(slice_dst_size_for_trailing_bytes(layout, trailing_size), old_size); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_size_for_elems_uses_element_product() { + let layout: TrailingSliceLayout = any_bounded_trailing_layout(); + let elems: usize = kani::any(); + let expected = layout + .elem_size + .checked_mul(elems) + .and_then(|bytes| slice_dst_size_for_trailing_bytes(layout, bytes)); + assert_eq!(layout.size_for_elems(elems), expected); + } - SizeInfo::Sized { size } + #[kani::proof] + #[kani::solver(kissat)] + fn prove_element_product_preserves_alignment() { + let elem_size: usize = kani::any(); + let elems: usize = kani::any(); + let align = any_layout_align(); + #[allow(clippy::arithmetic_side_effects)] + let align_mask = align.get() - 1; + + kani::assume(elem_size & align_mask == 0); + let Some(bytes) = elem_size.checked_mul(elems) else { + kani::assume(false); + loop {} + }; + assert_eq!(bytes & align_mask, 0); + } + + #[kani::proof] + fn prove_trailing_slice_layout_advance() { + let layout: TrailingSliceLayout = any_bounded_trailing_layout(); + let bytes: usize = kani::any(); + let elem_size: usize = kani::any(); + let trailing_size: usize = kani::any(); + + let size_offset = layout.size_offset(); + let advanced_offset = size_offset.checked_add(bytes); + let advanced = layout.advance(bytes, elem_size); + assert_eq!(advanced.is_some(), advanced_offset.is_some()); + let Some(advanced_offset) = advanced_offset else { + kani::assume(false); + loop {} + }; + let Some(advanced) = advanced else { unreachable!() }; + + assert_eq!(advanced.offset, layout.offset); + assert_eq!(advanced.size_offset(), advanced_offset); + assert_eq!(advanced.elem_size, elem_size); + assert_eq!( + advanced.size_rounding_align_and_phase.align(), + layout.size_rounding_align_and_phase.align() + ); + #[allow(clippy::arithmetic_side_effects)] + let size_align_mask = advanced.size_rounding_align_and_phase.align().get() - 1; + assert_eq!(advanced.size_base & size_align_mask, layout.size_base & size_align_mask); + + let size_align = layout.size_rounding_align_and_phase.align(); + #[allow(clippy::arithmetic_side_effects)] + let size_prefix = layout.size_base & (size_align.get() - 1); + let expected_size = advanced_offset + .checked_add(trailing_size) + .and_then(|without_padding| { + without_padding.checked_add(util::padding_needed_for(without_padding, size_align)) + }) + .and_then(|rounded| size_prefix.checked_add(rounded)); + assert_eq!(slice_dst_size_for_trailing_bytes(advanced, trailing_size), expected_size); + } + + #[kani::proof] + fn prove_has_same_size_sequence() { + let left: TrailingSliceLayout = any_bounded_trailing_layout(); + let right: TrailingSliceLayout = any_bounded_trailing_layout(); + let trailing_size: usize = kani::any(); + + kani::assume(left.has_same_size_sequence(right)); + let left_align = left.size_rounding_align_and_phase.align().get(); + let right_align = right.size_rounding_align_and_phase.align().get(); + let max_align = if left_align > right_align { left_align } else { right_align }; + #[allow(clippy::arithmetic_side_effects)] + let max_align_mask = max_align - 1; + if left.elem_size & max_align_mask == 0 { + // In this branch `has_same_size_sequence` relies on every actual + // trailing byte count being a multiple of both alignments. Prove + // the formula for that entire byte class, which includes every + // representable `elems * elem_size`, without introducing a + // nonlinear symbolic multiplication into the harness. + kani::assume(trailing_size & max_align_mask == 0); + } + assert_eq!( + slice_dst_size_for_trailing_bytes(left, trailing_size), + slice_dst_size_for_trailing_bytes(right, trailing_size) + ); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_max_trailing_bytes() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let available_bytes: usize = kani::any(); + let trailing_bytes: usize = kani::any(); + let fits = slice_dst_size_for_trailing_bytes(trailing, trailing_bytes) + .map(|size| size <= available_bytes) + .unwrap_or(false); + match trailing.max_trailing_bytes(available_bytes) { + Some(max_bytes) => assert_eq!(trailing_bytes <= max_bytes, fits), + None => assert!(!fits), + } + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_max_trailing_bytes_components() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let bytes_len: usize = kani::any(); + let Some(max_slice_bytes) = trailing.max_trailing_bytes(bytes_len) else { + kani::assume(false); + loop {} + }; + let (size_align, size_phase) = trailing.size_rounding_align_and_phase.components(); + let Some(bytes_after_base) = bytes_len.checked_sub(trailing.size_base) else { + unreachable!() + }; + let max_rounded_bytes = + util::round_down_to_next_multiple_of_alignment(bytes_after_base, size_align); + assert_eq!(size_phase.checked_add(max_slice_bytes), Some(max_rounded_bytes)); + assert!(trailing.size_base.checked_add(max_rounded_bytes).unwrap() <= bytes_len); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_slice_dst_validator_selected_size_fits() { + let size_align = any_layout_align(); + #[allow(clippy::arithmetic_side_effects)] + let align_mask = size_align.get() - 1; + let raw_phase: usize = kani::any(); + let size_phase = raw_phase & align_mask; + let raw_capacity: usize = kani::any(); + let max_rounded_bytes = raw_capacity & !align_mask; + let size_base: usize = kani::any(); + let bytes_len: usize = kani::any(); + let trailing_size: usize = kani::any(); + + // These are exactly the relations established for every successful + // capacity query by `prove_max_trailing_bytes_components`. Isolate + // them from the helper's implementation to keep this arithmetic + // identity independent of how that helper computes its bound. + let Some(max_slice_bytes) = max_rounded_bytes.checked_sub(size_phase) else { + kani::assume(false); + loop {} + }; + let Some(max_object_size) = size_base.checked_add(max_rounded_bytes) else { + kani::assume(false); + loop {} + }; + kani::assume(max_object_size <= bytes_len); + kani::assume(trailing_size <= max_slice_bytes); + + // For a whole-element count selected by the validator, this unused + // byte count is also `max_slice_bytes % elem_size`. The identity here + // holds for every byte count satisfying the capacity bound. + #[allow(clippy::arithmetic_side_effects)] + let unused_bytes = max_slice_bytes - trailing_size; + let unused_aligned_bytes = + util::round_down_to_next_multiple_of_alignment(unused_bytes, size_align); + #[allow(clippy::arithmetic_side_effects)] + let object_size = size_base + (max_rounded_bytes - unused_aligned_bytes); + + let unrounded = size_phase.checked_add(trailing_size).unwrap(); + let rounded = + unrounded.checked_add(util::padding_needed_for(unrounded, size_align)).unwrap(); + assert_eq!(Some(object_size), size_base.checked_add(rounded)); + assert!(object_size <= bytes_len); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_slice_dst_validator_rejects_larger_trailing_size() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let bytes_len: usize = kani::any(); + let larger_trailing_size: usize = kani::any(); + let Some(max_slice_bytes) = trailing.max_trailing_bytes(bytes_len) else { + kani::assume(false); + loop {} + }; + kani::assume(larger_trailing_size > max_slice_bytes); + if let Some(object_size) = slice_dst_size_for_trailing_bytes(trailing, larger_trailing_size) + { + assert!(object_size > bytes_len); + } + } + + #[kani::proof] + fn prove_slice_formula_contains_trailing_slice() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let trailing_size: usize = kani::any(); + let (_, size_phase) = trailing.size_rounding_align_and_phase.components(); + + let Some(unrounded_min) = trailing.size_base.checked_add(size_phase) else { + kani::assume(false); + loop {} + }; + kani::assume(unrounded_min >= trailing.offset); + let Some(object_size) = slice_dst_size_for_trailing_bytes(trailing, trailing_size) else { + kani::assume(false); + loop {} + }; + + let Some(trailing_end) = trailing.offset.checked_add(trailing_size) else { unreachable!() }; + assert!(trailing_end <= object_size); + } + + #[kani::proof] + fn prove_finalized_slice_formula_is_aligned() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let outer_align = any_layout_align(); + let trailing_size: usize = kani::any(); + kani::assume(trailing.size_rounding_align_and_phase.align() >= outer_align); + #[allow(clippy::arithmetic_side_effects)] + let outer_align_mask = outer_align.get() - 1; + kani::assume(trailing.size_base & outer_align_mask == 0); + let Some(object_size) = slice_dst_size_for_trailing_bytes(trailing, trailing_size) else { + kani::assume(false); + loop {} + }; + + assert_eq!(object_size & outer_align_mask, 0); + } + + #[kani::proof] + fn prove_validator_checks_selected_endpoint_alignment() { + let align = any_layout_align(); + let addr: usize = kani::any(); + let bytes_len: usize = kani::any(); + let Some(end) = addr.checked_add(bytes_len) else { + kani::assume(false); + loop {} + }; + let suffix: bool = kani::any(); + let cast_type = if suffix { CastType::Suffix } else { CastType::Prefix }; + let selected_endpoint = if suffix { end } else { addr }; + #[allow(clippy::arithmetic_side_effects)] + let align_mask = align.get() - 1; + + // The endpoint check precedes and is independent of the size-info + // branch. A zero-sized layout makes every aligned case succeed and + // isolates that control flow, including prefix/suffix split selection. + let layout = DstLayout { + align, + size_info: SizeInfo::Sized { size: 0 }, + statically_shallow_unpadded: false, + }; + match layout.validate_cast_and_convert_metadata(addr, bytes_len, cast_type) { + Ok((elems, split_at)) => { + assert_eq!(selected_endpoint & align_mask, 0); + assert_eq!(elems, 0); + if suffix { + assert_eq!(split_at, bytes_len); + } else { + assert_eq!(split_at, 0); } - false => SizeInfo::SliceDst(kani::any()), } + Err(MetadataCastError::Alignment) => { + assert_ne!(selected_endpoint & align_mask, 0); + } + Err(MetadataCastError::Size) => unreachable!(), } } - impl kani::Arbitrary for TrailingSliceLayout { - fn any() -> Self { - let elem_size: usize = kani::any(); - let offset: usize = kani::any(); + #[kani::proof] + fn prove_aligned_suffix_end_and_size_imply_aligned_start() { + let align = any_layout_align(); + #[allow(clippy::arithmetic_side_effects)] + let align_mask = align.get() - 1; + let addr: usize = kani::any(); + let bytes_len: usize = kani::any(); + let object_size: usize = kani::any(); + let Some(end) = addr.checked_add(bytes_len) else { + kani::assume(false); + loop {} + }; + kani::assume(end & align_mask == 0); + kani::assume(object_size & align_mask == 0); + kani::assume(object_size <= bytes_len); - kani::assume(elem_size < DstLayout::MAX_SIZE); - kani::assume(offset < DstLayout::MAX_SIZE); + #[allow(clippy::arithmetic_side_effects)] + let split_at = bytes_len - object_size; + let Some(object_start) = addr.checked_add(split_at) else { unreachable!() }; + assert_eq!(object_start & align_mask, 0); + } - TrailingSliceLayout { elem_size, offset } - } + #[kani::proof] + #[kani::solver(kissat)] + fn prove_size_formula_bounds_size_offset() { + let trailing: TrailingSliceLayout = any_bounded_trailing_layout(); + let elems: usize = kani::any(); + let Some(object_size) = trailing.size_for_elems(elems) else { + kani::assume(false); + loop {} + }; + let Some(trailing_size) = trailing.elem_size.checked_mul(elems) else { unreachable!() }; + let size_offset = trailing.size_offset(); + let Some(without_padding) = size_offset.checked_add(trailing_size) else { unreachable!() }; + + assert!(without_padding <= object_size); + } + + // Older supported compilers resolve the unchecked operation syntax in + // `add_scaled_metadata` to `polyfills::NumExt`; newer compilers prefer the + // inherent methods. Use UFCS to verify the fallback implementation which + // Kani's modern rustc cannot select in the production helper. + #[kani::proof] + #[kani::solver(kissat)] + fn prove_num_ext_affine_fallback_matches_checked_arithmetic() { + let base: usize = kani::any(); + let metadata: usize = kani::any(); + let multiple: usize = kani::any(); + let Some(scaled) = metadata.checked_mul(multiple) else { + kani::assume(false); + loop {} + }; + let Some(expected) = base.checked_add(scaled) else { + kani::assume(false); + loop {} + }; + + // SAFETY: The checked multiplication returned `Some`, satisfying + // `NumExt::unchecked_mul`'s no-overflow contract. + let scaled = + unsafe { ::unchecked_mul(metadata, multiple) }; + // SAFETY: The checked addition returned `Some`, and `scaled` is the + // checked product, satisfying `NumExt::unchecked_add`'s no-overflow + // contract. + let actual = + unsafe { ::unchecked_add(base, scaled) }; + assert_eq!(actual, expected); } #[kani::proof] fn prove_requires_dynamic_padding() { - let layout: DstLayout = kani::any(); + let layout: DstLayout = any_valid_dst_layout(); let SizeInfo::SliceDst(size_info) = layout.size_info else { kani::assume(false); @@ -2029,22 +4700,25 @@ mod proofs { loop {} }; - let Some(unpadded_size) = size_info.offset.checked_add(trailing_slice_size) else { - // The `unpadded_size` exceeds `usize::MAX`; `meta`` is invalid. + let Some(trailing_slice_end) = size_info.offset.checked_add(trailing_slice_size) else { + // The trailing slice end exceeds `usize::MAX`; `meta` is invalid. kani::assume(false); loop {} }; - if unpadded_size >= DstLayout::MAX_SIZE { - // The `unpadded_size` exceeds `isize::MAX`; `meta` is invalid. + let Some(size) = slice_dst_size_for_trailing_bytes(size_info, trailing_slice_size) else { kani::assume(false); loop {} - } + }; - let trailing_padding = util::padding_needed_for(unpadded_size, layout.align); + if size > DstLayout::MAX_SIZE || trailing_slice_end > size { + // This metadata cannot describe a valid instance. + kani::assume(false); + loop {} + } if !layout.requires_dynamic_padding() { - assert!(trailing_padding == 0); + assert_eq!(trailing_slice_end, size); } } @@ -2052,8 +4726,8 @@ mod proofs { fn prove_dst_layout_extend() { use crate::util::{max, min, padding_needed_for}; - let base: DstLayout = kani::any(); - let field: DstLayout = kani::any(); + let base: DstLayout = any_valid_dst_layout(); + let field: DstLayout = any_valid_dst_layout(); let packed: Option = kani::any(); if let Some(max_align) = packed { @@ -2069,9 +4743,6 @@ mod proofs { unreachable!(); }; - // Under the above conditions, `DstLayout::extend` will not panic. - let composite = base.extend(field, packed); - // The field's alignment is clamped by `max_align` (i.e., the // `packed` attribute, if any) [1]. // @@ -2082,10 +4753,6 @@ mod proofs { // alignment of the field's type. let field_align = min(field.align, packed.unwrap_or(DstLayout::THEORETICAL_MAX_ALIGN)); - // The struct's alignment is the maximum of its previous alignment and - // `field_align`. - assert_eq!(composite.align, max(base.align, field_align)); - // Compute the minimum amount of inter-field padding needed to // satisfy the field's alignment, and offset of the trailing field. // [1] @@ -2095,7 +4762,38 @@ mod proofs { // Inter-field padding is guaranteed to be the minimum required in // order to satisfy each field's (possibly altered) alignment. let padding = padding_needed_for(base_size, field_align); - let offset = base_size + padding; + let offset = match base_size.checked_add(padding) { + Some(offset) => offset, + None => { + kani::assume(false); + loop {} + } + }; + + // Restrict the proof to extensions whose zero-metadata composite size + // is representable. This semantic condition implies the + // representability of each implementation intermediate below; prove + // those implications rather than assuming the intermediates. + match field.size_info { + SizeInfo::Sized { size } => kani::assume(offset.checked_add(size).is_some()), + SizeInfo::SliceDst(trailing) => { + let Some(field_min) = trailing.size_for_elems(0) else { + kani::assume(false); + loop {} + }; + kani::assume(offset.checked_add(field_min).is_some()); + + assert!(offset.checked_add(trailing.offset).is_some()); + assert!(offset.checked_add(trailing.size_base).is_some()); + } + } + + // Under the above conditions, `DstLayout::extend` will not panic. + let composite = base.extend(field, packed); + + // The struct's alignment is the maximum of its previous alignment and + // `field_align`. + assert_eq!(composite.align, max(base.align, field_align)); // For testing purposes, we'll also construct `alloc::Layout` // stand-ins for `DstLayout`, and show that `extend` behaves @@ -2132,29 +4830,64 @@ mod proofs { panic!("The composite of two sized layouts must be sized.") } } - SizeInfo::SliceDst(TrailingSliceLayout { - offset: field_offset, - elem_size: field_elem_size, - }) => { - if let SizeInfo::SliceDst(TrailingSliceLayout { - offset: composite_offset, - elem_size: composite_elem_size, - }) = composite.size_info - { + SizeInfo::SliceDst(field_trailing) => { + if let SizeInfo::SliceDst(composite_trailing) = composite.size_info { // The offset of the trailing slice component is the sum // of the offset of the trailing field and the trailing // slice offset within that field. - assert_eq!(composite_offset, offset + field_offset); + assert_eq!(composite_trailing.offset, offset + field_trailing.offset); // The elem size is unchanged. - assert_eq!(composite_elem_size, field_elem_size); + assert_eq!(composite_trailing.elem_size, field_trailing.elem_size); + // Extension preserves the invariant that the physical + // trailing-slice field fits within the unrounded minimum + // size. This is one of the premises consumed by the + // validator soundness proof. + let composite_phase = + composite_trailing.size_rounding_align_and_phase.components().1; + let Some(composite_unrounded_min) = + composite_trailing.size_base.checked_add(composite_phase) + else { + unreachable!() + }; + assert!(composite_unrounded_min >= composite_trailing.offset); + + // The normalized formula must continue to describe the + // size of the field at every trailing byte contribution + // for which the checked size computation succeeds. This + // is stronger than considering only products of a slice + // length and `elem_size`, and avoids a nonlinear + // multiplication in the verifier. + let trailing_size: usize = kani::any(); + let field_size = + match slice_dst_size_for_trailing_bytes(field_trailing, trailing_size) { + Some(size) => size, + None => { + kani::assume(false); + loop {} + } + }; + let expected_size = match offset.checked_add(field_size) { + Some(size) => size, + None => { + kani::assume(false); + loop {} + } + }; + assert_eq!( + slice_dst_size_for_trailing_bytes(composite_trailing, trailing_size), + Some(expected_size) + ); + // `alloc::Layout` permits sizes which are not a multiple + // of alignment, so this represents the unpadded prefix of + // the DST through the start of its trailing slice. let field_analog = - Layout::from_size_align(field_offset, field_align.get()).unwrap(); + Layout::from_size_align(field_trailing.offset, field_align.get()).unwrap(); if let Ok((actual_composite, actual_offset)) = base_analog.extend(field_analog) { assert_eq!(actual_offset, offset); - assert_eq!(actual_composite.size(), composite_offset); + assert_eq!(actual_composite.size(), composite_trailing.offset); assert_eq!(actual_composite.align(), composite.align.get()); } else { // An error here reflects that composite of `base` @@ -2175,8 +4908,8 @@ mod proofs { #[kani::proof] #[kani::should_panic] fn prove_dst_layout_extend_dst_panics() { - let base: DstLayout = kani::any(); - let field: DstLayout = kani::any(); + let base: DstLayout = any_valid_dst_layout(); + let field: DstLayout = any_valid_dst_layout(); let packed: Option = kani::any(); if let Some(max_align) = packed { @@ -2189,39 +4922,263 @@ mod proofs { let _ = base.extend(field, packed); } + fn prove_dst_layout_pad_to_align_dst_invariants( + layout: DstLayout, + unpadded_trailing: TrailingSliceLayout, + padded: DstLayout, + ) { + // Calling `pad_to_align` does not alter the `DstLayout`'s alignment. + assert_eq!(padded.align, layout.align); + assert_eq!(padded.statically_shallow_unpadded, layout.statically_shallow_unpadded); + + let SizeInfo::SliceDst(padded_trailing) = padded.size_info else { + panic!("The padding of a DST layout must result in a DST layout.") + }; + assert_eq!(padded_trailing.offset, unpadded_trailing.offset); + assert_eq!(padded_trailing.elem_size, unpadded_trailing.elem_size); + + // A finalized DST formula is visibly aligned: the rounding alignment + // is at least the outer alignment, and the fixed base is a multiple of + // the outer alignment. This is the invariant which lets suffix + // validation infer alignment of the object's start from alignment of + // its end. + assert!(padded_trailing.size_rounding_align_and_phase.align() >= padded.align); + #[allow(clippy::arithmetic_side_effects)] + let padded_align_mask = padded.align.get() - 1; + assert_eq!(padded_trailing.size_base & padded_align_mask, 0); + assert_eq!(padded.pad_to_align(), padded); + let padded_phase = padded_trailing.size_rounding_align_and_phase.components().1; + let Some(padded_unrounded_min) = padded_trailing.size_base.checked_add(padded_phase) else { + unreachable!() + }; + assert!(padded_unrounded_min >= padded_trailing.offset); + } + #[kani::proof] - fn prove_dst_layout_pad_to_align() { + fn prove_dst_layout_pad_to_align_sized() { use crate::util::padding_needed_for; - let layout: DstLayout = kani::any(); + let layout: DstLayout = any_valid_dst_layout(); + let SizeInfo::Sized { size: unpadded_size } = layout.size_info else { + kani::assume(false); + loop {} + }; let padded = layout.pad_to_align(); - - // Calling `pad_to_align` does not alter the `DstLayout`'s alignment. assert_eq!(padded.align, layout.align); + let SizeInfo::Sized { size: padded_size } = padded.size_info else { + panic!("The padding of a sized layout must result in a sized layout.") + }; - if let SizeInfo::Sized { size: unpadded_size } = layout.size_info { - if let SizeInfo::Sized { size: padded_size } = padded.size_info { - // If the layout is sized, it will remain sized after padding is - // added. Its sum will be its unpadded size and the size of the - // trailing padding needed to satisfy its alignment - // requirements. - let padding = padding_needed_for(unpadded_size, layout.align); - assert_eq!(padded_size, unpadded_size + padding); - - // Prove that calling `DstLayout::pad_to_align` behaves - // identically to `Layout::pad_to_align`. - let layout_analog = - Layout::from_size_align(unpadded_size, layout.align.get()).unwrap(); - let padded_analog = layout_analog.pad_to_align(); - assert_eq!(padded_analog.align(), layout.align.get()); - assert_eq!(padded_analog.size(), padded_size); - } else { - panic!("The padding of a sized layout must result in a sized layout.") - } - } else { - // If the layout is a DST, padding cannot be statically added. - assert_eq!(padded.size_info, layout.size_info); - } + // If the layout is sized, it will remain sized after padding is added. + // Its sum will be its unpadded size and the size of the trailing + // padding needed to satisfy its alignment requirements. + let padding = padding_needed_for(unpadded_size, layout.align); + assert_eq!(padded_size, unpadded_size + padding); + assert_eq!( + padded.statically_shallow_unpadded, + layout.statically_shallow_unpadded && padding == 0 + ); + + // Prove that calling `DstLayout::pad_to_align` behaves identically to + // `Layout::pad_to_align`. + let layout_analog = Layout::from_size_align(unpadded_size, layout.align.get()).unwrap(); + let padded_analog = layout_analog.pad_to_align(); + assert_eq!(padded_analog.align(), layout.align.get()); + assert_eq!(padded_analog.size(), padded_size); + } + + #[kani::proof] + fn prove_dst_layout_pad_to_align_dst_inner_rounding_transition() { + use crate::util::padding_needed_for; + + let layout: DstLayout = any_valid_dst_layout(); + let SizeInfo::SliceDst(trailing) = layout.size_info else { + kani::assume(false); + loop {} + }; + kani::assume(trailing.size_rounding_align_and_phase.align() >= layout.align); + + // `any_valid_dst_layout` requires the zero-metadata layout to be valid. + // Prove that this semantic condition bounds both the padded minimum + // and the implementation intermediate; do not assume the intermediate + // merely because the implementation uses it. + let Some(min_size) = trailing.size_for_elems(0) else { unreachable!() }; + let min_padding = padding_needed_for(min_size, layout.align); + let Some(padded_min_size) = min_size.checked_add(min_padding) else { unreachable!() }; + assert!(padded_min_size <= DstLayout::MAX_SIZE); + + let padding = padding_needed_for(trailing.size_base, layout.align); + let Some(expected_base) = trailing.size_base.checked_add(padding) else { unreachable!() }; + + let padded = layout.pad_to_align(); + let SizeInfo::SliceDst(padded_trailing) = padded.size_info else { unreachable!() }; + assert_eq!(padded_trailing.size_base, expected_base); + assert_eq!( + padded_trailing.size_rounding_align_and_phase, + trailing.size_rounding_align_and_phase + ); + prove_dst_layout_pad_to_align_dst_invariants(layout, trailing, padded); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_dst_layout_pad_to_align_dst_inner_rounding_formula() { + use crate::util::padding_needed_for; + + let size_align = any_layout_align(); + let outer_align = any_layout_align(); + kani::assume(size_align >= outer_align); + let size_base: usize = kani::any(); + let raw_size_phase: usize = kani::any(); + #[allow(clippy::arithmetic_side_effects)] + let size_phase = raw_size_phase & (size_align.get() - 1); + let trailing_size: usize = kani::any(); + + // Evaluate `size_base + round_up(size_phase + trailing_size, + // size_align)`, then round that complete size to `outer_align`. + // Compare with the equivalent expression obtained by rounding only + // `size_base` to `outer_align`, since `size_align` is a multiple of + // `outer_align`. Valid Rust objects have representable sizes; discard + // only trailing byte counts whose input or padded object size is not + // representable. + let Some(unrounded_inner) = size_phase.checked_add(trailing_size) else { + kani::assume(false); + loop {} + }; + let Some(rounded_inner) = + unrounded_inner.checked_add(padding_needed_for(unrounded_inner, size_align)) + else { + kani::assume(false); + loop {} + }; + let Some(unpadded_size) = size_base.checked_add(rounded_inner) else { + kani::assume(false); + loop {} + }; + let Some(expected_size) = + unpadded_size.checked_add(padding_needed_for(unpadded_size, outer_align)) + else { + kani::assume(false); + loop {} + }; + + let Some(padded_base) = size_base.checked_add(padding_needed_for(size_base, outer_align)) + else { + unreachable!() + }; + let Some(actual_size) = padded_base.checked_add(rounded_inner) else { unreachable!() }; + assert_eq!(actual_size, expected_size); + } + + #[kani::proof] + fn prove_dst_layout_pad_to_align_dst_outer_alignment_transition() { + use crate::util::padding_needed_for; + + let layout: DstLayout = any_valid_dst_layout(); + let SizeInfo::SliceDst(trailing) = layout.size_info else { + kani::assume(false); + loop {} + }; + let size_align = trailing.size_rounding_align_and_phase.align(); + kani::assume(size_align < layout.align); + + // `any_valid_dst_layout` requires the zero-metadata layout to be valid. + // Prove that this semantic condition bounds both the padded minimum + // and every implementation intermediate; do not assume the + // intermediates merely because the implementation uses them. + let Some(min_size) = trailing.size_for_elems(0) else { unreachable!() }; + let min_padding = padding_needed_for(min_size, layout.align); + let Some(padded_min_size) = min_size.checked_add(min_padding) else { unreachable!() }; + assert!(padded_min_size <= DstLayout::MAX_SIZE); + + let base_padding = padding_needed_for(trailing.size_base, size_align); + let Some(rounded_base) = trailing.size_base.checked_add(base_padding) else { + unreachable!() + }; + let Some(t) = + rounded_base.checked_add(trailing.size_rounding_align_and_phase.components().1) + else { + unreachable!() + }; + #[allow(clippy::arithmetic_side_effects)] + let expected_phase = t & (layout.align.get() - 1); + #[allow(clippy::arithmetic_side_effects)] + let expected_base = t - expected_phase; + + let padded = layout.pad_to_align(); + let SizeInfo::SliceDst(padded_trailing) = padded.size_info else { unreachable!() }; + assert_eq!(padded_trailing.size_base, expected_base); + assert_eq!( + padded_trailing.size_rounding_align_and_phase, + RoundingAlignAndPhase::new(layout.align, expected_phase) + ); + prove_dst_layout_pad_to_align_dst_invariants(layout, trailing, padded); + } + + #[kani::proof] + #[kani::solver(kissat)] + fn prove_dst_layout_pad_to_align_dst_outer_alignment_formula() { + use crate::util::padding_needed_for; + + let size_align = any_layout_align(); + let outer_align = any_layout_align(); + kani::assume(size_align < outer_align); + let size_base: usize = kani::any(); + let raw_size_phase: usize = kani::any(); + #[allow(clippy::arithmetic_side_effects)] + let size_phase = raw_size_phase & (size_align.get() - 1); + let trailing_size: usize = kani::any(); + + // Evaluate `size_base + round_up(size_phase + trailing_size, + // size_align)`, then round that complete size to `outer_align`. + let Some(unrounded_inner) = size_phase.checked_add(trailing_size) else { + kani::assume(false); + loop {} + }; + let Some(rounded_inner) = + unrounded_inner.checked_add(padding_needed_for(unrounded_inner, size_align)) + else { + kani::assume(false); + loop {} + }; + let Some(unpadded_size) = size_base.checked_add(rounded_inner) else { + kani::assume(false); + loop {} + }; + let Some(expected_size) = + unpadded_size.checked_add(padding_needed_for(unpadded_size, outer_align)) + else { + kani::assume(false); + loop {} + }; + + // Since `outer_align` is a larger multiple of `size_align`, absorb + // the inner rounding into the outer rounding and split the fixed + // contribution into a base and phase: + // + // fixed_contribution = round_up(size_base, size_align) + size_phase + // new_phase = fixed_contribution % outer_align + // new_base = fixed_contribution - new_phase + // new_base + round_up(new_phase + trailing_size, outer_align) + let Some(rounded_base) = size_base.checked_add(padding_needed_for(size_base, size_align)) + else { + unreachable!() + }; + let Some(t) = rounded_base.checked_add(size_phase) else { unreachable!() }; + #[allow(clippy::arithmetic_side_effects)] + let new_phase = t & (outer_align.get() - 1); + #[allow(clippy::arithmetic_side_effects)] + let new_base = t - new_phase; + let Some(new_unrounded_inner) = new_phase.checked_add(trailing_size) else { + unreachable!() + }; + let Some(new_rounded_inner) = + new_unrounded_inner.checked_add(padding_needed_for(new_unrounded_inner, outer_align)) + else { + unreachable!() + }; + let Some(actual_size) = new_base.checked_add(new_rounded_inner) else { unreachable!() }; + assert_eq!(actual_size, expected_size); } } diff --git a/zerocopy/src/lib.rs b/zerocopy/src/lib.rs index 63bb4d5d5b..6040d6e0dd 100644 --- a/zerocopy/src/lib.rs +++ b/zerocopy/src/lib.rs @@ -778,10 +778,16 @@ pub unsafe trait KnownLayout { /// where `size == size_of::()` /// - If `Self` is a slice DST, then `LAYOUT.size_info == /// SizeInfo::SliceDst(slice_layout)` where: + /// - `slice_layout.size_rounding_align_and_phase` decodes to an alignment + /// that is a power of two and a phase which is less than that alignment /// - The size, `size`, of an instance of `Self` with `elems` trailing - /// slice elements is equal to `slice_layout.offset + - /// slice_layout.elem_size * elems` rounded up to the nearest multiple - /// of `LAYOUT.align` + /// slice elements is equal to `slice_layout.size_base + + /// round_up(size_phase + slice_layout.elem_size * elems, size_align)`, + /// where `(size_align, size_phase)` is decoded from + /// `slice_layout.size_rounding_align_and_phase` + /// - The trailing slice begins at byte offset `slice_layout.offset` + /// - `slice_layout.offset + slice_layout.elem_size * elems` does not + /// overflow and is no greater than `size` /// - For such an instance, any bytes in the range `[slice_layout.offset + /// slice_layout.elem_size * elems, size)` are padding and must not be /// assumed to be initialized @@ -996,11 +1002,7 @@ impl PointerMetadata for usize { #[inline] fn size_for_metadata(self, layout: DstLayout) -> Option { match layout.size_info { - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) => { - let slice_len = elem_size.checked_mul(self)?; - let without_padding = offset.checked_add(slice_len)?; - without_padding.checked_add(util::padding_needed_for(without_padding, layout.align)) - } + SizeInfo::SliceDst(trailing) => trailing.size_for_elems(self), // NOTE: This branch is unreachable, but we return `None` rather // than `unreachable!()` to avoid generating panic paths. SizeInfo::Sized { .. } => None, @@ -6740,6 +6742,39 @@ mod tests { } } + // These types exercise a nested slice DST whose inner trailing-padding + // operation cannot be flattened into the alignment of the packed outer + // type. For `NestedPackedDstOuter<[u8]>`, the trailing slice begins at + // offset 7, but the size for `elems` elements is + // `2 + round_up(5 + elems, 4)`. + // + // The Reference rejects directly placing an alignment-4 field in a + // `repr(packed(2))` type. Rust currently accepts this construction through + // generic indirection. So long as Rust accepts it, this test deliberately + // treats its layout as following the otherwise-applicable `repr(C)` and + // `repr(packed)` rules. + #[derive(FromBytes, Immutable, KnownLayout)] + #[repr(C, align(4))] + struct NestedPackedDstWord([u8; 4]); + + #[derive(FromBytes, Immutable, KnownLayout)] + #[repr(C)] + struct NestedPackedDstInner { + a: NestedPackedDstWord, + b: u8, + tail: T, + } + + #[derive(FromBytes, Immutable, KnownLayout)] + #[repr(C, packed(2))] + struct NestedPackedDstOuter { + z: u8, + inner: ManuallyDrop>, + } + + #[repr(C, align(2))] + struct Align2(T); + #[test] fn test_known_layout() { // Test that `$ty` and `ManuallyDrop<$ty>` have the expected layout. @@ -6754,17 +6789,25 @@ mod tests { }; } - let layout = - |offset, align, trailing_slice_elem_size, statically_shallow_unpadded| DstLayout { - align: NonZeroUsize::new(align).unwrap(), + let layout = |offset, align, trailing_slice_elem_size, statically_shallow_unpadded| { + let align = NonZeroUsize::new(align).unwrap(); + let size_phase = offset % align; + DstLayout { + align, size_info: match trailing_slice_elem_size { None => SizeInfo::Sized { size: offset }, - Some(elem_size) => { - SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }) - } + Some(elem_size) => SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base: offset - size_phase, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + align, size_phase, + ), + }), }, statically_shallow_unpadded, - }; + } + }; test!((), layout(0, 1, None, false)); test!(u8, layout(1, 1, None, false)); @@ -6832,6 +6875,101 @@ mod tests { } } + #[test] + fn test_nested_packed_dst_layout() { + fn assert_native_size(expected: usize) { + let value = NestedPackedDstOuter { + z: 0, + inner: ManuallyDrop::new(NestedPackedDstInner { + a: NestedPackedDstWord([0; 4]), + b: 0, + tail: [0; N], + }), + }; + let value: &NestedPackedDstOuter<[u8]> = &value; + assert_eq!(mem::size_of_val(value), expected); + assert_eq!(NestedPackedDstOuter::<[u8]>::size_for_metadata(N), Some(expected)); + } + + assert_native_size::<0>(10); + assert_native_size::<1>(10); + assert_native_size::<2>(10); + assert_native_size::<3>(10); + assert_native_size::<4>(14); + assert_native_size::<7>(14); + assert_native_size::<8>(18); + + let layout = NestedPackedDstOuter::<[u8]>::LAYOUT; + assert_eq!(layout.align.get(), 2); + assert!(layout.requires_dynamic_padding()); + assert_eq!( + layout.size_info, + SizeInfo::SliceDst(TrailingSliceLayout { + offset: 7, + elem_size: 1, + size_base: 6, + size_rounding_align_and_phase: RoundingAlignAndPhase::new( + NonZeroUsize::new(4).unwrap(), + 1 + ), + }) + ); + + // Eight bytes used to be incorrectly accepted for a zero-element + // value. The actual Rust layout requires ten bytes. + let short = Align2([0u8; 8]); + assert!(NestedPackedDstOuter::<[u8]>::ref_from_bytes_with_elems(&short.0, 0).is_err()); + assert!(pointer::Ptr::from_ref(&short.0[..]) + .try_cast_into::, pointer::BecauseImmutable>( + CastType::Prefix, + Some(0), + ) + .is_err()); + + let full = Align2([0, 0, 0, 0, 0, 0, 0, 0, b'S', b'!']); + let value = NestedPackedDstOuter::<[u8]>::ref_from_bytes_with_elems(&full.0, 0).unwrap(); + assert_eq!(mem::size_of_val(value), 10); + + // When metadata is inferred, lengths 0 through 3 all produce a + // ten-byte object. Parsing chooses the largest trailing-slice length + // which fits exactly. + let inferred = NestedPackedDstOuter::<[u8]>::ref_from_bytes(&full.0).unwrap(); + assert_eq!(pointer::Ptr::from_ref(inferred).len(), 3); + + // Trailing padding includes the inner type's padding, even when the + // outer type's alignment is reduced by packing. + let trailing_layout = trailing_slice_layout::>(); + for (elems, expected) in [(0, 3), (1, 2), (2, 1), (3, 0), (4, 3)] { + assert_eq!(trailing_layout.padding_for_elems(elems), expected); + } + } + + // Before Rust 1.76, rust-lang/rust#118537 caused `addr_of_mut!` to + // miscompute the offset of an unsized field in a packed struct. The fix is + // rust-lang/rust#118540. Keep the layout and parsing regression above on + // all supported compilers, and gate only the projection regression which + // necessarily exercises that independent compiler bug. + #[rustversion::since(1.76.0)] + #[test] + fn test_nested_packed_dst_projection() { + let full = Align2([0, 0, 0, 0, 0, 0, 0, 0, b'S', b'!']); + + // The projected inner value occupies the final eight bytes of the full + // ten-byte input, including the two sentinel bytes. + let (value, rest) = pointer::Ptr::from_ref(&full.0[..]) + .try_cast_into::, pointer::BecauseImmutable>( + CastType::Prefix, + Some(0), + ) + .unwrap(); + assert!(rest.as_ref().is_empty()); + let inner = value.project::().unwrap(); + assert_eq!( + inner.as_bytes::().as_ref(), + &[0, 0, 0, 0, 0, 0, b'S', b'!'] + ); + } + #[cfg(feature = "derive")] #[test] fn test_known_layout_derive() { @@ -6887,10 +7025,19 @@ mod tests { statically_shallow_unpadded: false, }; - let unsized_layout = |align, elem_size, offset, statically_shallow_unpadded| DstLayout { - align: NonZeroUsize::new(align).unwrap(), - size_info: SizeInfo::SliceDst(TrailingSliceLayout { offset, elem_size }), - statically_shallow_unpadded, + let unsized_layout = |align, elem_size, offset, statically_shallow_unpadded| { + let align = NonZeroUsize::new(align).unwrap(); + let size_phase = offset % align; + DstLayout { + align, + size_info: SizeInfo::SliceDst(TrailingSliceLayout { + offset, + elem_size, + size_base: offset - size_phase, + size_rounding_align_and_phase: RoundingAlignAndPhase::new(align, size_phase), + }), + statically_shallow_unpadded, + } }; // | `repr(C)`? | generic? | `KnownLayout`? | `Sized`? | Type Name | @@ -7907,6 +8054,16 @@ mod tests { assert_eq!(&*s, &[0, 3, 0]); } + #[test] + fn test_new_box_zeroed_with_elems_nested_packed_dst() { + let mut value = NestedPackedDstOuter::<[u8]>::new_box_zeroed_with_elems(0).unwrap(); + assert_eq!(mem::size_of_val(&*value), 10); + + // Exercise the whole-object write that exposed the former + // under-allocation under Miri and AddressSanitizer. + value.zero(); + } + #[test] fn test_new_box_zeroed_with_elems_empty() { let s: Box<[u64]> = <[u64]>::new_box_zeroed_with_elems(0).unwrap(); diff --git a/zerocopy/src/macros.rs b/zerocopy/src/macros.rs index 5f7479aa1e..9e908a0c3a 100644 --- a/zerocopy/src/macros.rs +++ b/zerocopy/src/macros.rs @@ -1411,6 +1411,38 @@ mod tests { let x: &SliceDst = transmute_ref!(slice_dst_big); assert_eq!(x, slice_dst_small); + // A packed outer DST can preserve the trailing field's stronger + // rounding alignment. These two types consequently have different + // normalized layout formulas even though both have size `2 + 4 * n` + // for every trailing slice length. Ensure the semantic size-sequence + // comparison accepts this representation-preserving transmutation. + #[derive(FromBytes, Immutable, IntoBytes, KnownLayout)] + #[repr(C, align(2))] + struct TransmuteAlign2([u8; 4]); + + #[derive(FromBytes, Immutable, IntoBytes, KnownLayout)] + #[repr(C, packed(2))] + struct TransmutePacked4 { + head: u16, + tail: [u32], + } + + #[derive(FromBytes, Immutable, IntoBytes, KnownLayout)] + #[repr(C, packed(2))] + struct TransmutePacked2 { + head: u16, + tail: [TransmuteAlign2], + } + + #[repr(C, align(2))] + struct TransmuteBytes([u8; 10]); + + let bytes = TransmuteBytes([0; 10]); + let src = TransmutePacked4::ref_from_bytes(&bytes.0).unwrap(); + let dst: &TransmutePacked2 = transmute_ref!(src); + assert_eq!(mem::size_of_val(src), 10); + assert_eq!(mem::size_of_val(dst), 10); + // Test that it's legal to transmute a reference while shrinking the // lifetime (note that `X` has the lifetime `'static`). let x: &[u8; 8] = transmute_ref!(X); diff --git a/zerocopy/src/pointer/inner.rs b/zerocopy/src/pointer/inner.rs index 023df07d04..d926e1e520 100644 --- a/zerocopy/src/pointer/inner.rs +++ b/zerocopy/src/pointer/inner.rs @@ -289,10 +289,10 @@ where /// /// Given `let (left, right) = ptr.split_at(l_len)`, it is guaranteed that /// `left` and `right` are contiguous and non-overlapping if - /// `l_len.padding_needed_for() == 0`. This is true for all `[T]`. + /// the left part has no trailing padding. This is true for all `[T]`. /// - /// If `l_len.padding_needed_for() != 0`, then the left pointer will overlap - /// the right pointer to satisfy `T`'s padding requirements. + /// Any trailing padding in the left part begins at the right part's + /// starting address and may overlap its referent. #[inline] #[must_use] pub unsafe fn split_at_unchecked( @@ -313,23 +313,19 @@ where // Trivially, `slf.meta() <= slf.meta()`. let right = unsafe { right.slice_unchecked(l_len..self.meta().get()) }; - // SAFETY: If `l_len.padding_needed_for() == 0`, then `left` and `right` - // are non-overlapping. Proof: `left` is constructed `slf` with `l_len` - // as its (exclusive) upper bound. If `l_len.padding_needed_for() == 0`, - // then `left` requires no trailing padding following its final element. - // Since `right` is constructed from `slf`'s trailing slice with `l_len` - // as its (inclusive) lower bound, no byte is referred to by both - // pointers. + // SAFETY: If the left part has no trailing padding, then `left` and + // `right` are non-overlapping. Proof: `left` has `l_len` trailing + // elements, and `right` starts at element `l_len` of the original + // trailing slice. Without padding after `left`'s final element, no + // byte is referred to by both pointers. // - // Conversely, `l_len.padding_needed_for() == N`, where `N - // > 0`, `left` requires `N` bytes of trailing padding following its - // final element. Since `right` is constructed from the trailing slice - // of `slf` with `l_len` as its (inclusive) lower bound, the first `N` - // bytes of `right` are aliased by `left`. + // Any trailing padding in `left` begins at `right`'s starting + // address. If `right` is nonzero sized, the referents overlap. (left, right) } - /// Produces the trailing slice of `self`. + /// Produces the trailing slice of `self` using `T::Elem` as its element + /// representation. #[inline] #[must_use] pub fn trailing_slice(self) -> PtrInner<'a, [T::Elem]> @@ -368,9 +364,10 @@ where // 1. If `ptr`'s referent is not zero sized, then `ptr` has valid // provenance for `A` because `raw` is derived from the same // allocated object as `self` via provenance-preserving operations. - // 2. If `ptr`'s referent is not zero sized, then `ptr` addresses a byte - // range which is entirely contained in `A`, by previous safety proof - // on `bytes`. + // 2. `T: SplitAt` guarantees that `T::Elem` has the size of the actual + // trailing element. The new slice has the same length and start as + // the original trailing slice, so it covers the same byte range. + // If nonzero sized, that range is entirely contained in `A`. // 3. `ptr` addresses a byte range whose length fits in an `isize`, by // consequence of #2. // 4. `ptr` addresses a byte range which does not wrap around the diff --git a/zerocopy/src/pointer/mod.rs b/zerocopy/src/pointer/mod.rs index 03b17bf078..52ba00fd95 100644 --- a/zerocopy/src/pointer/mod.rs +++ b/zerocopy/src/pointer/mod.rs @@ -47,10 +47,7 @@ where pub mod cast { use core::{marker::PhantomData, mem}; - use crate::{ - layout::{SizeInfo, TrailingSliceLayout}, - HasField, KnownLayout, PtrInner, - }; + use crate::{layout::SizeInfo, HasField, KnownLayout, PtrInner}; /// A pointer cast or projection. /// @@ -182,11 +179,9 @@ pub mod cast { // SAFETY: By the `static_assert!`, `Src` and `Dst` are either: // - Both sized and equal in size - // - Both slice DSTs with the same trailing slice offset and element size - // and with align_of::() == align_of::(). These ensure that any - // given pointer metadata encodes the same size for both `Src` and `Dst` - // (note that the alignment is required as it affects the amount of - // trailing padding). Thus, `project` preserves the set of referent bytes. + // - Both slice DSTs with the same trailing-slice offset, the same object + // size for every metadata value, and with align_of::() == + // align_of::(). Thus, `project` preserves the set of referent bytes. unsafe impl Project for CastUnsized where Src: ?Sized + KnownLayout, @@ -201,10 +196,11 @@ pub mod cast { let dst = ::LAYOUT; match (src.size_info, dst.size_info) { (SizeInfo::Sized { size: src_size }, SizeInfo::Sized { size: dst_size }) => src_size == dst_size, - ( - SizeInfo::SliceDst(TrailingSliceLayout { offset: src_offset, elem_size: src_elem_size }), - SizeInfo::SliceDst(TrailingSliceLayout { offset: dst_offset, elem_size: dst_elem_size }) - ) => src.align.get() == dst.align.get() && src_offset == dst_offset && src_elem_size == dst_elem_size, + (SizeInfo::SliceDst(src_trailing), SizeInfo::SliceDst(dst_trailing)) => { + src.align.get() == dst.align.get() + && src_trailing.offset == dst_trailing.offset + && src_trailing.has_same_size_sequence(dst_trailing) + }, _ => false, } }); @@ -225,10 +221,8 @@ pub mod cast { // SAFETY: By the `static_assert!` in `Project::project`, `Src` and `Dst` // are either: // - Both sized and equal in size - // - Both slice DSTs with the same alignment, trailing slice offset, and - // element size. These ensure that any given pointer metadata encodes the - // same size for both `Src` and `Dst` (note that the alignment is required - // as it affects the amount of trailing padding). + // - Both slice DSTs with the same alignment and trailing-slice offset, and + // with equal sizes for every metadata value. unsafe impl CastExact for CastUnsized where Src: ?Sized + KnownLayout, diff --git a/zerocopy/src/split_at.rs b/zerocopy/src/split_at.rs index 720ba740c8..0a98c130d0 100644 --- a/zerocopy/src/split_at.rs +++ b/zerocopy/src/split_at.rs @@ -17,6 +17,12 @@ use crate::pointer::invariant::{Aligned, Exclusive, Invariants, Safe, Shared}; /// This trait generalizes Rust's existing support for splitting slices to /// support slices and slice-based dynamically-sized types ("slice DSTs"). /// +/// For a `repr(C, packed)` or `repr(C, packed(N))` struct, the derive wraps +/// the trailing field's [`Elem`](Self::Elem) in [`Unalign`]. The returned +/// slice can therefore be accessed even when packing misaligns the original +/// elements. Each packed struct in the trailing-field chain adds one wrapper; +/// unpacked structs and [`ManuallyDrop`] forward their trailing type's `Elem`. +/// /// # Implementation /// /// **Do not implement this trait yourself!** Instead, use @@ -38,6 +44,28 @@ use crate::pointer::invariant::{Aligned, Exclusive, Invariants, Safe, Shared}; /// This derive performs a sophisticated, compile-time safety analysis to /// determine whether a type is `SplitAt`. /// +/// A packed struct exposes its trailing elements through `Unalign`: +/// +/// ``` +/// use zerocopy::{FromBytes, SplitAt, Unalign}; +/// # use zerocopy_derive::*; +/// +/// #[derive(FromBytes, IntoBytes, KnownLayout, SplitAt)] +/// #[repr(C, packed)] +/// struct Packet { +/// tag: u8, +/// words: [u16], +/// } +/// +/// let mut bytes = [0u8; 5]; +/// let packet = Packet::mut_from_bytes(&mut bytes[..]).unwrap(); +/// let (left, right): (&mut Packet, &mut [Unalign]) = +/// packet.split_at_mut(1).unwrap().via_into_bytes(); +/// left.tag = 1; +/// right[0] = Unalign::new(0x0202); +/// assert_eq!(bytes, [1, 0, 0, 2, 2]); +/// ``` +/// /// # Safety /// /// This trait does not convey any safety guarantees to code outside this crate. @@ -58,10 +86,20 @@ use crate::pointer::invariant::{Aligned, Exclusive, Invariants, Safe, Shared}; )] // # Safety // -// The trailing slice is well-aligned for its element type. `Self` is `[T]`, or -// a `repr(C)` or `repr(transparent)` slice DST. +// `Self` is a slice, a `repr(C)` (possibly packed) or `repr(transparent)` slice +// DST, or `ManuallyDrop` for some `T: SplitAt`. +// +// `Self::Elem` has the same size, bit validity, and `UnsafeCell` coverage as +// the actual trailing element type. Access through either representation +// preserves the other's validity, including writes through mutable references +// and interior mutation through shared references. For every aligned `Self`, +// its trailing slice's address is aligned for `Self::Elem`. pub unsafe trait SplitAt: KnownLayout { - /// The element type of the trailing slice. + /// The element type exposed by the split's returned slice. + /// + /// For `[T]`, this is `T`. For a struct, this is the trailing field's `Elem`, + /// wrapped in [`Unalign`] if the struct is packed. Nested packed structs + /// produce nested `Unalign` wrappers. type Elem; #[doc(hidden)] @@ -188,10 +226,7 @@ pub unsafe trait SplitAt: KnownLayout { /// Attempts to split `self` in two. /// /// Returns `None` if `l_len` is greater than the length of `self`'s - /// trailing slice, or if the given `l_len` would result in [the trailing - /// padding](KnownLayout#slice-dst-layout) of the left portion overlapping - /// the right portion. - /// + /// trailing slice. /// /// # Examples /// @@ -249,7 +284,8 @@ pub unsafe trait SplitAt: KnownLayout { } } -// SAFETY: `[T]`'s trailing slice is `[T]`, which is trivially aligned. +// SAFETY: `[T]` exposes its own element type, preserving size, validity, and +// `UnsafeCell` coverage. An aligned slice is aligned for its elements. unsafe impl SplitAt for [T] { type Elem = T; @@ -262,6 +298,29 @@ unsafe impl SplitAt for [T] { } } +// SAFETY: `ManuallyDrop` has the same layout and bit validity as `T` [1]. +// Its `KnownLayout` implementation preserves `T`'s metadata and trailing-slice +// layout. Forwarding `T::Elem` therefore preserves its size, validity, and +// alignment guarantees. Its `UnsafeCell` coverage also matches `T`'s, as +// established for `ManuallyDrop` in `impls.rs`; accesses through the forwarded +// element representation preserve validity in both directions. +// +// [1] Per https://doc.rust-lang.org/1.93.0/std/mem/struct.ManuallyDrop.html: +// +// `ManuallyDrop` is guaranteed to have the same layout and bit validity as +// `T` +unsafe impl SplitAt for ManuallyDrop { + type Elem = T::Elem; + + #[inline] + #[allow(dead_code)] + fn only_derive_is_allowed_to_implement_this_trait() + where + Self: Sized, + { + } +} + /// A `T` that has been split into two possibly-overlapping parts. /// /// For some dynamically sized types, the padding that appears after the @@ -270,10 +329,11 @@ unsafe impl SplitAt for [T] { /// requires trailing padding, the trailing padding of the left part of the /// split `T` will overlap the right part. If `T` is a mutable reference or /// permits interior mutation, you must ensure that the left and right parts do -/// not overlap. You can do this at zero-cost using using -/// [`Self::via_immutable`], [`Self::via_into_bytes`], or -/// [`Self::via_unaligned`], or with a dynamic check by using -/// [`Self::via_runtime_check`]. +/// not overlap. Use [`Self::via_into_bytes`] or +/// [`Self::via_no_dynamic_padding`] to establish this without a runtime check, +/// or [`Self::via_runtime_check`] to check a particular split. Shared references +/// to an [`Immutable`] type may overlap; use [`Self::via_immutable`] in that +/// case. #[derive(Debug)] pub struct Split { /// A pointer to the source slice DST. @@ -423,59 +483,100 @@ where (l.as_ref(), r.as_ref()) } - /// Produces the split parts of `self`, using [`Unaligned`] to ensure that - /// it is sound to have concurrent references to both parts. + /// Produces the split parts of `self` using [`Self::via_no_dynamic_padding`]. + #[deprecated(note = "use `Split::via_no_dynamic_padding` instead")] + #[doc(hidden)] + #[must_use = "has no side effects"] + #[inline(always)] + pub fn via_unaligned(self) -> (&'a T, &'a [T::Elem]) + where + T: Unaligned, + { + self.via_no_dynamic_padding() + } + + /// Produces the split parts of `self` using a compile-time assertion to + /// ensure that it is unconditionally sound to have concurrent references to + /// both parts. + /// + /// When possible, you should prefer using [`Self::via_into_bytes`], which + /// ensures this safety property through regular trait bounds. /// /// # Examples /// + /// In the below example, a single byte of padding exists (on most + /// platforms) between `Packet<[u16]>`'s `length` and `body` fields, + /// precluding the use of [`Self::via_into_bytes`]. + /// /// ``` - /// use zerocopy::{SplitAt, FromBytes}; + /// use zerocopy::SplitAt; /// # use zerocopy_derive::*; /// - /// #[derive(SplitAt, FromBytes, KnownLayout, Immutable, Unaligned)] + /// #[derive(SplitAt, KnownLayout)] /// #[repr(C)] - /// struct Packet { + /// struct Packet { /// length: u8, - /// body: [u8], + /// body: B, /// } /// - /// // These bytes encode a `Packet`. - /// let bytes = &[4, 1, 2, 3, 4, 5, 6, 7, 8, 9][..]; - /// - /// let packet = Packet::ref_from_bytes(bytes).unwrap(); + /// let packet: &Packet<[u16]> = &Packet { length: 4, body: [1u16, 2, 3, 4, 5, 6, 7, 8, 9] }; /// /// assert_eq!(packet.length, 4); /// assert_eq!(packet.body, [1, 2, 3, 4, 5, 6, 7, 8, 9]); /// - /// // Attempt to split `packet` at `length`. - /// let split = packet.split_at(packet.length as usize).unwrap(); + /// if let Some(split) = packet.split_at(packet.length as usize) { + /// // Every body length requires no trailing padding. + /// let (packet, rest) = split.via_no_dynamic_padding(); + /// assert_eq!(packet.length, 4); + /// assert_eq!(packet.body, [1, 2, 3, 4]); + /// assert_eq!(rest, [5, 6, 7, 8, 9]); + /// } else { + /// unreachable!("The packet's length field is within its body"); + /// } + /// ``` /// - /// // Use the `Unaligned` bound on `Packet` to prove that it's okay to - /// // return concurrent references to `packet` and `rest`. - /// let (packet, rest) = split.via_unaligned(); + /// # Compile-Time Assertions /// - /// assert_eq!(packet.length, 4); - /// assert_eq!(packet.body, [1, 2, 3, 4]); - /// assert_eq!(rest, [5, 6, 7, 8, 9]); + /// This method rejects at compile-time any type that admits dynamic + /// trailing padding. In the below example, some `split_at` indices would + /// produce left parts whose trailing padding overlaps with the right part: + /// + /// ```compile_fail,E0080 + /// use zerocopy::SplitAt; + /// # use zerocopy_derive::*; + /// + /// #[derive(SplitAt, KnownLayout)] + /// #[repr(C, align(2))] + /// struct Packet { + /// length: u16, + /// body: B, + /// } + /// + /// let packet: &Packet<[u8]> = &Packet { length: 4, body: [1u8, 2, 3, 4, 5, 6, 7, 8, 9] }; + /// + /// if let Some(split) = packet.split_at(packet.length as usize) { + /// let _ = split.via_no_dynamic_padding(); // ⚠ Compile Error! + /// } else { + /// unreachable!("The packet's length field is within its body"); + /// } /// ``` /// - #[doc = codegen_header!("h5", "split_via_unaligned")] + /// If you need to split such types, use [`Self::via_runtime_check`]. + /// + #[doc = codegen_header!("h5", "split_via_no_dynamic_padding")] /// /// See [`Split::via_immutable`](#method.split_via_immutable.codegen). #[must_use = "has no side effects"] #[inline(always)] - pub fn via_unaligned(self) -> (&'a T, &'a [T::Elem]) - where - T: Unaligned, - { - let (l, r) = self.into_ptr().via_unaligned(); + pub fn via_no_dynamic_padding(self) -> (&'a T, &'a [T::Elem]) { + let (l, r) = self.into_ptr().via_no_dynamic_padding(); (l.as_ref(), r.as_ref()) } /// Produces the split parts of `self`, using a dynamic check to ensure that /// it is sound to have concurrent references to both parts. You should /// prefer using [`Self::via_immutable`], [`Self::via_into_bytes`], or - /// [`Self::via_unaligned`], which have no runtime cost. + /// [`Self::via_no_dynamic_padding`], which perform no runtime check. /// /// Note that this check is overly conservative if `T` is [`Immutable`]; for /// some types, this check will reject some splits which @@ -484,52 +585,40 @@ where /// # Examples /// /// ``` - /// use zerocopy::{SplitAt, FromBytes, IntoBytes, network_endian::U16}; + /// use zerocopy::SplitAt; /// # use zerocopy_derive::*; /// - /// #[derive(SplitAt, FromBytes, KnownLayout, Immutable, Debug)] + /// #[derive(SplitAt, KnownLayout)] /// #[repr(C, align(2))] - /// struct Packet { - /// length: U16, - /// body: [u8], + /// struct Packet { + /// length: u16, + /// body: B, /// } /// - /// // These bytes encode a `Packet`. - /// let bytes = [ - /// 4u16.to_be(), - /// 1u16.to_be(), - /// 2u16.to_be(), - /// 3u16.to_be(), - /// 4u16.to_be() - /// ]; - /// - /// let packet = Packet::ref_from_bytes(bytes.as_bytes()).unwrap(); + /// let packet: &Packet<[u8]> = &Packet { length: 4, body: [1u8, 2, 3, 4, 5, 6, 7, 8, 9] }; /// /// assert_eq!(packet.length, 4); - /// assert_eq!(packet.body, [0, 1, 0, 2, 0, 3, 0, 4]); - /// - /// // Attempt to split `packet` at `length`. - /// let split = packet.split_at(packet.length.into()).unwrap(); + /// assert_eq!(packet.body, [1, 2, 3, 4, 5, 6, 7, 8, 9]); /// - /// // Use a dynamic check to prove that it's okay to return concurrent - /// // references to `packet` and `rest`. - /// let (packet, rest) = split.via_runtime_check().unwrap(); + /// // The two-byte prefix and four-byte body need no trailing padding. + /// if let Some(split) = packet.split_at(packet.length as usize) { + /// if let Ok((packet, rest)) = split.via_runtime_check() { + /// assert_eq!(packet.length, 4); + /// assert_eq!(packet.body, [1, 2, 3, 4]); + /// assert_eq!(rest, [5, 6, 7, 8, 9]); + /// } else { + /// unreachable!("A four-byte body requires no trailing padding"); + /// } + /// } else { + /// unreachable!("The packet's length field is within its body"); + /// } /// - /// assert_eq!(packet.length, 4); - /// assert_eq!(packet.body, [0, 1, 0, 2]); - /// assert_eq!(rest, [0, 3, 0, 4]); - /// - /// // Attempt to split `packet` at `length - 1`. - /// let idx = packet.length.get() - 1; - /// let split = packet.split_at(idx as usize).unwrap(); - /// - /// // Attempt (and fail) to use a dynamic check to prove that it's okay - /// // to return concurrent references to `packet` and `rest`. Note that - /// // this is a case of `via_runtime_check` being overly conservative. - /// // Although the left and right parts indeed overlap, the `Immutable` - /// // bound ensures that concurrently referencing these overlapping - /// // parts is sound. - /// assert!(split.via_runtime_check().is_err()); + /// // A three-byte body needs a padding byte that would overlap `rest`. + /// if let Some(split) = packet.split_at((packet.length - 1) as usize) { + /// assert!(split.via_runtime_check().is_err()); + /// } else { + /// unreachable!("One less than the packet's length is within its body"); + /// } /// ``` /// #[doc = codegen_section!( @@ -665,101 +754,135 @@ where (l.as_mut(), r.as_mut()) } - /// Produces the split parts of `self`, using [`Unaligned`] to ensure that - /// it is sound to have concurrent references to both parts. + /// Produces the split parts of `self` using [`Self::via_no_dynamic_padding`]. + #[deprecated(note = "use `Split::via_no_dynamic_padding` instead")] + #[doc(hidden)] + #[must_use = "has no side effects"] + #[inline(always)] + pub fn via_unaligned(self) -> (&'a mut T, &'a mut [T::Elem]) + where + T: Unaligned, + { + self.via_no_dynamic_padding() + } + + /// Produces the split parts of `self` using a compile-time assertion to + /// ensure that it is unconditionally sound to have concurrent references to + /// both parts. + /// + /// When possible, you should prefer using [`Self::via_into_bytes`], which + /// ensures this safety property through regular trait bounds. /// /// # Examples /// + /// In the below example, a single byte of padding exists (on most + /// platforms) between `Packet<[u16]>`'s `length` and `body` fields, + /// precluding the use of [`Self::via_into_bytes`]. + /// /// ``` - /// use zerocopy::{SplitAt, FromBytes}; + /// use zerocopy::SplitAt; /// # use zerocopy_derive::*; /// - /// #[derive(SplitAt, FromBytes, KnownLayout, IntoBytes, Unaligned)] + /// #[derive(SplitAt, KnownLayout)] /// #[repr(C)] /// struct Packet { /// length: u8, /// body: B, /// } /// - /// // These bytes encode a `Packet`. - /// let mut bytes = &mut [4, 1, 2, 3, 4, 5, 6, 7, 8, 9][..]; - /// - /// let packet = Packet::<[u8]>::mut_from_bytes(bytes).unwrap(); + /// let packet: &mut Packet<[u16]> = &mut Packet { length: 4, body: [1u16, 2, 3, 4, 5, 6, 7, 8, 9] }; /// /// assert_eq!(packet.length, 4); /// assert_eq!(packet.body, [1, 2, 3, 4, 5, 6, 7, 8, 9]); /// - /// { - /// // Attempt to split `packet` at `length`. - /// let split = packet.split_at_mut(packet.length as usize).unwrap(); - /// - /// // Use the `Unaligned` bound on `Packet` to prove that it's okay to - /// // return concurrent references to `packet` and `rest`. - /// let (packet, rest) = split.via_unaligned(); - /// + /// // Every body length requires no trailing padding. + /// if let Some(split) = packet.split_at_mut(packet.length as usize) { + /// let (packet, rest) = split.via_no_dynamic_padding(); /// assert_eq!(packet.length, 4); /// assert_eq!(packet.body, [1, 2, 3, 4]); /// assert_eq!(rest, [5, 6, 7, 8, 9]); - /// /// rest.fill(0); + /// } else { + /// unreachable!("The packet's length field is within its body"); /// } /// /// assert_eq!(packet.length, 4); /// assert_eq!(packet.body, [1, 2, 3, 4, 0, 0, 0, 0, 0]); /// ``` /// + /// # Compile-Time Assertions + /// + /// This method rejects at compile-time any type that admits dynamic + /// trailing padding. In the below example, some `split_at` indices would + /// produce left parts whose trailing padding overlaps with the right part: + /// + /// ```compile_fail,E0080 + /// use zerocopy::SplitAt; + /// # use zerocopy_derive::*; + /// + /// #[derive(SplitAt, KnownLayout)] + /// #[repr(C, align(2))] + /// struct Packet { + /// length: u16, + /// body: B, + /// } + /// + /// let packet: &mut Packet<[u8]> = &mut Packet { length: 4, body: [1u8, 2, 3, 4, 5, 6, 7, 8, 9] }; + /// + /// if let Some(split) = packet.split_at_mut(packet.length as usize) { + /// let _ = split.via_no_dynamic_padding(); // ⚠ Compile Error! + /// } else { + /// unreachable!("The packet's length field is within its body"); + /// } + /// ``` + /// + /// If you need to split such types, use [`Self::via_runtime_check`]. + /// /// # Code Generation /// /// See [`Split::via_immutable`](#method.split_via_immutable.codegen). #[must_use = "has no side effects"] #[inline(always)] - pub fn via_unaligned(self) -> (&'a mut T, &'a mut [T::Elem]) - where - T: Unaligned, - { - let (l, r) = self.into_ptr().via_unaligned(); + pub fn via_no_dynamic_padding(self) -> (&'a mut T, &'a mut [T::Elem]) { + let (l, r) = self.into_ptr().via_no_dynamic_padding(); (l.as_mut(), r.as_mut()) } /// Produces the split parts of `self`, using a dynamic check to ensure that /// it is sound to have concurrent references to both parts. You should - /// prefer using [`Self::via_into_bytes`] or [`Self::via_unaligned`], which - /// have no runtime cost. + /// prefer using [`Self::via_into_bytes`] or + /// [`Self::via_no_dynamic_padding`], which perform no runtime check. /// /// # Examples /// /// ``` - /// use zerocopy::{SplitAt, FromBytes}; + /// use zerocopy::SplitAt; /// # use zerocopy_derive::*; /// - /// #[derive(SplitAt, FromBytes, KnownLayout, IntoBytes, Debug)] - /// #[repr(C)] + /// #[derive(SplitAt, KnownLayout)] + /// #[repr(C, align(2))] /// struct Packet { - /// length: u8, + /// length: u16, /// body: B, /// } /// - /// // These bytes encode a `Packet`. - /// let mut bytes = &mut [4, 1, 2, 3, 4, 5, 6, 7, 8, 9][..]; - /// - /// let packet = Packet::<[u8]>::mut_from_bytes(bytes).unwrap(); + /// let packet: &mut Packet<[u8]> = &mut Packet { length: 4, body: [1u8, 2, 3, 4, 5, 6, 7, 8, 9] }; /// /// assert_eq!(packet.length, 4); /// assert_eq!(packet.body, [1, 2, 3, 4, 5, 6, 7, 8, 9]); /// - /// { - /// // Attempt to split `packet` at `length`. - /// let split = packet.split_at_mut(packet.length as usize).unwrap(); - /// - /// // Use a dynamic check to prove that it's okay to return concurrent - /// // references to `packet` and `rest`. - /// let (packet, rest) = split.via_runtime_check().unwrap(); - /// - /// assert_eq!(packet.length, 4); - /// assert_eq!(packet.body, [1, 2, 3, 4]); - /// assert_eq!(rest, [5, 6, 7, 8, 9]); - /// - /// rest.fill(0); + /// // The two-byte prefix and four-byte body need no trailing padding. + /// if let Some(split) = packet.split_at_mut(packet.length as usize) { + /// if let Ok((packet, rest)) = split.via_runtime_check() { + /// assert_eq!(packet.length, 4); + /// assert_eq!(packet.body, [1, 2, 3, 4]); + /// assert_eq!(rest, [5, 6, 7, 8, 9]); + /// rest.fill(0); + /// } else { + /// unreachable!("A four-byte body requires no trailing padding"); + /// } + /// } else { + /// unreachable!("The packet's length field is within its body"); /// } /// /// assert_eq!(packet.length, 4); @@ -860,37 +983,57 @@ where unsafe { self.via_unchecked() } } - /// Produces the split parts of `self`, using [`Unaligned`] to ensure that - /// it is sound to have concurrent references to both parts. + /// Produces the split parts of `self` after the conservative layout + /// predicate establishes that `T` never requires dynamic trailing padding. #[inline(always)] - fn via_unaligned(self) -> (Ptr<'a, T, I>, Ptr<'a, [T::Elem], I>) - where - T: Unaligned, - { - // SAFETY: By `T: SplitAt + Unaligned`, `T` is either a slice or a - // `repr(C)` or `repr(transparent)` slice DST that is well-aligned at - // any address and length. If `T` is a slice DST with alignment 1, - // `repr(C)` or `repr(transparent)` ensures that no padding is placed - // after the final element of the trailing slice. Consequently, `T` can - // be split into strictly non-overlapping parts any any index. + fn via_no_dynamic_padding(self) -> (Ptr<'a, T, I>, Ptr<'a, [T::Elem], I>) { + static_assert!( + T: ?Sized + KnownLayout => !T::LAYOUT.requires_dynamic_padding(), + "`Split::via_no_dynamic_padding` cannot be used with a type whose layout may require dynamic trailing padding; use `Split::via_runtime_check` instead" + ); + + // SAFETY: The assertion establishes that `requires_dynamic_padding()` + // is false. By that method's guarantee, every valid metadata—including + // `self.l_len()`, which is valid by `Split`'s invariant—describes an + // object with no trailing padding. By the postcondition of + // `PtrInner::split_at_unchecked`, the two returned referents are then + // contiguous and non-overlapping. Since the source already conforms to + // `I::Aliasing`, splitting it into disjoint byte ranges preserves that + // invariant: under `Exclusive`, neither result aliases or accesses the + // other's bytes; under `Shared`, interior mutation through one result + // cannot mutate bytes pointed to by the other. This satisfies Rust's + // reference aliasing rules [1]. + // + // [1] Per https://doc.rust-lang.org/1.93.1/std/ptr/index.html#pointer-to-reference-conversion: + // + // When creating a mutable reference, then while this reference + // exists, the memory it points to must not get accessed (read or + // written) through any other pointer or reference not derived from + // this reference. + // + // When creating a shared reference, then while this reference + // exists, the memory it points to must not get mutated (except inside + // `UnsafeCell`). unsafe { self.via_unchecked() } } /// Produces the split parts of `self`, using a dynamic check to ensure that /// it is sound to have concurrent references to both parts. You should /// prefer using [`Self::via_immutable`], [`Self::via_into_bytes`], or - /// [`Self::via_unaligned`], which have no runtime cost. + /// [`Self::via_no_dynamic_padding`], which perform no runtime check. #[inline(always)] fn via_runtime_check(self) -> Result<(Ptr<'a, T, I>, Ptr<'a, [T::Elem], I>), Self> { let l_len = self.l_len(); + // `l_len` witnesses an object size at most `isize::MAX`, and + // `KnownLayout::LAYOUT` guarantees that it contains the trailing + // slice. Thus `padding_for_elems` returns the exact trailing padding. + let trailing_padding = crate::trailing_slice_layout::().padding_for_elems(l_len.get()); // FIXME(#1290): Once we require `KnownLayout` on all fields, add an // `IS_IMMUTABLE` associated const, and add `T::IS_IMMUTABLE ||` to the // below check. - if l_len.padding_needed_for() == 0 { - // SAFETY: By `T: SplitAt`, `T` is either `[T]`, or a `repr(C)` or - // `repr(transparent)` slice DST, for which the trailing padding - // needed to accommodate `l_len` trailing elements is - // `l_len.padding_needed_for()`. If no trailing padding is required, + if trailing_padding == 0 { + // SAFETY: As established above, `trailing_padding` is the exact + // padding after the left part's trailing slice. If it is zero, // the left and right parts are strictly non-overlapping. Ok(unsafe { self.via_unchecked() }) } else { @@ -903,7 +1046,7 @@ where /// # Safety /// /// The caller promises that if `I::Aliasing` is [`Exclusive`] or `T` - /// permits interior mutation, then `l_len.padding_needed_for() == 0`. + /// permits interior mutation, then the left part has no trailing padding. #[inline(always)] unsafe fn via_unchecked(self) -> (Ptr<'a, T, I>, Ptr<'a, [T::Elem], I>) { let l_len = self.l_len(); @@ -915,11 +1058,11 @@ where // Lemma 0: `left` and `right` conform to the aliasing invariant // `I::Aliasing`. Proof: If `I::Aliasing` is `Exclusive` or `T` permits - // interior mutation, the caller promises that `l_len.padding_needed_for() - // == 0`. Consequently, by post-condition on `PtrInner::split_at_unchecked`, - // there is no trailing padding after `left`'s final element that would - // overlap into `right`. If `I::Aliasing` is shared and `T` forbids interior - // mutation, then overlap between their referents is permissible. + // interior mutation, the caller promises that the left part has no + // trailing padding. Consequently, by the postcondition of + // `PtrInner::split_at_unchecked`, `left` and `right` do not overlap. + // If `I::Aliasing` is shared and `T` forbids interior mutation, then + // overlap between their referents is permissible. // SAFETY: // 0. `left` conforms to the aliasing invariant of `I::Aliasing`, by Lemma 0. @@ -933,17 +1076,19 @@ where // SAFETY: // 0. `right` conforms to the aliasing invariant of `I::Aliasing`, by Lemma // 0. - // 1. `right` conforms to the alignment invariant of `I::Alignment, because - // if `ptr` with `I::Alignment = Aligned`, then by invariant on `T: - // SplitAt`, the trailing slice of `ptr` (from which `right` is derived) - // will also be well-aligned. + // 1. `right` conforms to the alignment invariant of `I::Alignment`: + // if the source is aligned, `T: SplitAt` guarantees that its + // trailing slice's address is aligned for `T::Elem`. Advancing by + // whole elements preserves that alignment. // 2. `right` conforms to the validity invariant of `I::Validity`, - // because `right: [T::Elem]` is derived from the trailing slice of - // `ptr`, which, by contract on `T: SplitAt::Elem`, has type - // `[T::Elem]`. The `left` part cannot be used to invalidate `right`, - // because the caller promises that if `I::Aliasing` is `Exclusive` - // or `T` permits interior mutation, then `l_len.padding_needed_for() - // == 0` and thus the parts will be non-overlapping. + // because `T: SplitAt` guarantees that `T::Elem` has the size and + // bit validity of the original trailing element. It also guarantees + // compatible shared and mutable access, so writes through `right` + // preserve the original elements' validity and interior mutation + // respects the original `UnsafeCell` coverage. The left part + // cannot invalidate `right`: when the references are exclusive or + // permit interior mutation, the caller guarantees no trailing + // padding and hence no overlap. let right = unsafe { Ptr::from_inner(right) }; (left, right) @@ -952,6 +1097,185 @@ where #[cfg(test)] mod tests { + use core::{cell::Cell, mem::ManuallyDrop}; + + use crate::{Immutable, KnownLayout, SplitAt, Unalign, Unaligned}; + + #[derive(KnownLayout, SplitAt, Immutable, Unaligned)] + #[repr(C, packed)] + struct Packed { + prefix: u8, + tail: ManuallyDrop, + } + + #[derive(KnownLayout, SplitAt, Immutable)] + #[repr(C, packed(2))] + struct Packed2 { + prefix: u8, + tail: ManuallyDrop, + } + + #[derive(KnownLayout, SplitAt, Immutable)] + #[repr(C)] + struct Inner { + prefix: u32, + tail: T, + } + + #[test] + fn test_split_at_packed() { + use crate::{FromBytes, IntoBytes}; + + #[derive(FromBytes, IntoBytes, KnownLayout, SplitAt, Immutable, Unaligned)] + #[repr(C, packed)] + struct Packet { + prefix: u8, + tail: [u16], + } + + // The tail starts at an odd address, exercising unaligned reads and + // writes on targets where `u16` has alignment 2. + #[repr(align(8))] + struct Bytes([u8; 9]); + + for i in 0..=4 { + let mut bytes = Bytes([1; 9]); + let packet = Packet::ref_from_bytes(&bytes.0).unwrap(); + let (left, right): (&Packet, &[Unalign]) = + packet.split_at(i).unwrap().via_immutable(); + assert_eq!(core::mem::size_of_val(left), 1 + 2 * i); + assert_eq!(right.len(), 4 - i); + assert!(right.iter().all(|elem| elem.get() == 0x0101)); + assert!(left.split_at(i).is_some()); + assert!(left.split_at(i + 1).is_none()); + assert!(packet.split_at(5).is_none()); + + let (_, right) = packet.split_at(i).unwrap().via_into_bytes(); + assert_eq!(right.len(), 4 - i); + let (_, right) = packet.split_at(i).unwrap().via_no_dynamic_padding(); + assert_eq!(right.len(), 4 - i); + + let packet = Packet::mut_from_bytes(&mut bytes.0).unwrap(); + let (left, right): (&mut Packet, &mut [Unalign]) = + packet.split_at_mut(i).unwrap().via_runtime_check().ok().unwrap(); + left.prefix = 2; + for elem in right { + *elem = Unalign::new(0x0202); + } + assert_eq!(bytes.0[0], 2); + assert!(bytes.0[1..1 + 2 * i].iter().all(|&byte| byte == 1)); + assert!(bytes.0[1 + 2 * i..].iter().all(|&byte| byte == 2)); + } + } + + #[test] + fn test_split_at_nested_packed() { + // A packing factor greater than 1 still exposes `Unalign` elements, + // including when the factor is below the original element alignment. + let mut words = Packed2 { prefix: 0, tail: ManuallyDrop::new([1u32, 2, 3]) }; + let dst: &mut Packed2<[u32]> = &mut words; + let (left, right): (&mut _, &mut [Unalign]) = + dst.split_at_mut(1).unwrap().via_no_dynamic_padding(); + left.prefix = 4; + right[0] = Unalign::new(5); + let dst: &Packed2<[u32]> = &words; + let (left, right) = dst.split_at(0).unwrap().via_immutable(); + assert_eq!(left.prefix, 4); + assert_eq!(right[0].get(), 1); + assert_eq!(right[1].get(), 5); + assert_eq!(right[2].get(), 3); + + let mut packet = Inner { + prefix: 0, + tail: Packed2 { + prefix: 1, + tail: ManuallyDrop::new(Packed { + prefix: 2, + tail: ManuallyDrop::new([3u16, 4, 5]), + }), + }, + }; + let dst: &Inner>> = &packet; + let (left, right): (&_, &[Unalign>]) = + dst.split_at(1).unwrap().via_immutable(); + assert_eq!(left.prefix, 0); + assert_eq!(right.len(), 2); + assert_eq!(right[0].get().get(), 4); + assert_eq!(right[1].get().get(), 5); + + // Split the packed field separately: the unpacked outer struct's + // padding does not constrain this split. + let dst: &mut Packed2> = &mut packet.tail; + let (left, right): (&mut _, &mut [Unalign>]) = + dst.split_at_mut(1).unwrap().via_no_dynamic_padding(); + left.prefix = 6; + right[0] = Unalign::new(Unalign::new(7)); + let dst: &Inner>> = &packet; + let (_, right) = dst.split_at(0).unwrap().via_immutable(); + assert_eq!(right[0].get().get(), 3); + assert_eq!(right[1].get().get(), 7); + assert_eq!(right[2].get().get(), 5); + } + + #[test] + fn test_split_at_packed_preserves_inner_padding() { + for i in 0..=4 { + let mut packet = Packed { + prefix: 1, + tail: ManuallyDrop::new(Inner { prefix: 0, tail: [2u8, 3, 4, 5] }), + }; + let dst: &Packed> = &packet; + let (left, right): (&_, &[Unalign]) = dst.split_at(i).unwrap().via_immutable(); + assert_eq!(right.len(), 4 - i); + assert_eq!(right.first().map(Unalign::get), [2, 3, 4, 5].get(i).copied()); + + // The inner `u32` prefix makes the tail start at byte 5 of the + // packed wrapper. Its size may still include inner padding. + let has_padding = core::mem::size_of_val(left) != 5 + i; + assert_eq!(dst.split_at(i).unwrap().via_runtime_check().is_err(), has_padding); + + let dst: &mut Packed> = &mut packet; + let split = dst.split_at_mut(i).unwrap().via_runtime_check(); + assert_eq!(split.is_err(), has_padding); + if let Ok((left, right)) = split { + left.prefix = 6; + for elem in right { + *elem = Unalign::new(7); + } + let dst: &Packed> = &packet; + let (left, right) = dst.split_at(i).unwrap().via_immutable(); + assert_eq!(left.prefix, 6); + assert!(right.iter().all(|elem| elem.get() == 7)); + } + } + } + + #[test] + fn test_split_at_packed_interior_mutation_and_zsts() { + let packet = Packed { + prefix: 0, + tail: ManuallyDrop::new([Cell::new(1u8), Cell::new(2), Cell::new(3)]), + }; + let dst: &Packed<[Cell]> = &packet; + let (left, right): (&_, &[Unalign>]) = + dst.split_at(1).unwrap().via_no_dynamic_padding(); + left.tail[0].set(4); + right[0].try_deref().unwrap().set(5); + assert_eq!(dst.tail[0].get(), 4); + assert_eq!(dst.tail[1].get(), 5); + assert_eq!(dst.tail[2].get(), 3); + + let mut packet = Packed { prefix: 0, tail: ManuallyDrop::new([(); 3]) }; + for i in 0..=3 { + let dst: &mut Packed<[()]> = &mut packet; + let (left, right): (&mut _, &mut [Unalign<()>]) = + dst.split_at_mut(i).unwrap().via_no_dynamic_padding(); + assert_eq!(left.tail.len(), i); + assert_eq!(right.len(), 3 - i); + assert_eq!(core::mem::size_of_val(left), 1); + } + } + #[cfg(feature = "derive")] #[test] fn test_split_at() { @@ -1011,6 +1335,8 @@ mod tests { trailing: [u8], } + assert!(SliceDst::LAYOUT.requires_dynamic_padding()); + const N: usize = 16; let arr = [1u16; N]; @@ -1070,21 +1396,79 @@ mod tests { assert_eq!(r, &[4]); } #[test] + #[allow(deprecated)] fn test_split_at_via_unaligned() { - use crate::{FromBytes, Immutable, IntoBytes, KnownLayout, SplitAt, Unaligned}; - #[derive(FromBytes, KnownLayout, SplitAt, IntoBytes, Immutable, Unaligned)] + use crate::{Immutable, KnownLayout, Split, SplitAt, Unaligned}; + + fn via_unaligned<'a, T>(split: Split<&'a T>) -> (&'a T, &'a [T::Elem]) + where + T: ?Sized + SplitAt + Unaligned, + { + split.via_unaligned() + } + + fn via_unaligned_mut<'a, T>(split: Split<&'a mut T>) -> (&'a mut T, &'a mut [T::Elem]) + where + T: ?Sized + SplitAt + Unaligned, + { + split.via_unaligned() + } + + #[derive(KnownLayout, SplitAt, Immutable, Unaligned)] #[repr(C)] - struct Packet { - length: u8, - body: [u8], + struct Packet { + prefix: [u8; 2], + body: B, } - let arr = [1, 2, 3, 4]; - let packet = Packet::ref_from_bytes(&arr[..]).unwrap(); + // Exercise generic callers using the original `T: Unaligned` bound. + assert!(!Packet::<[[u8; 2]]>::LAYOUT.requires_dynamic_padding()); + let packet = Packet { prefix: [0, 1], body: [[2, 3], [4, 5], [6, 7]] }; + let packet: &Packet<[[u8; 2]]> = &packet; let split = packet.split_at(2).unwrap(); - let (l, r) = split.via_unaligned(); - assert_eq!(l.length, 1); - assert_eq!(r, &[4]); + let (l, r) = via_unaligned(split); + assert_eq!(l.body, [[2, 3], [4, 5]]); + assert_eq!(r, &[[6, 7]]); + + let mut packet = Packet { prefix: [0, 1], body: [[2, 3], [4, 5], [6, 7]] }; + { + let packet: &mut Packet<[[u8; 2]]> = &mut packet; + let split = packet.split_at_mut(2).unwrap(); + let (l, r) = via_unaligned_mut(split); + l.body[0] = [8, 9]; + r[0] = [10, 11]; + } + assert_eq!(packet.body, [[8, 9], [4, 5], [10, 11]]); + } + + #[test] + fn test_split_at_via_no_dynamic_padding() { + use core::cell::Cell; + + use crate::SplitAt; + + // `u16` does not implement `Unaligned`, but `[u16]` never requires + // dynamic trailing padding. + let words = [1u16, 2, 3, 4]; + let split = SplitAt::split_at(&words[..], 2).unwrap(); + let (left, right) = split.via_no_dynamic_padding(); + assert_eq!(left, [1, 2]); + assert_eq!(right, [3, 4]); + + let mut words = [1u16, 2, 3, 4]; + let split = SplitAt::split_at_mut(&mut words[..], 2).unwrap(); + let (left, right) = split.via_no_dynamic_padding(); + left[0] = 5; + right[0] = 6; + assert_eq!(words, [5, 2, 6, 4]); + + // Exercise the safety-sensitive shared route with interior mutation. + let cells = [Cell::new(1u16), Cell::new(2), Cell::new(3)]; + let split = SplitAt::split_at(&cells[..], 2).unwrap(); + let (left, right) = split.via_no_dynamic_padding(); + left[0].set(4); + right[0].set(5); + assert_eq!([cells[0].get(), cells[1].get(), cells[2].get()], [4, 2, 5]); } } diff --git a/zerocopy/src/util/mod.rs b/zerocopy/src/util/mod.rs index a8a56a3b0e..a54980170a 100644 --- a/zerocopy/src/util/mod.rs +++ b/zerocopy/src/util/mod.rs @@ -144,9 +144,10 @@ pub(crate) fn validate_aligned_to(t: T) -> Result<(), Alignment /// on the answer it gives if this is not the case. #[cfg_attr( kani, - kani::requires(len <= DstLayout::MAX_SIZE), kani::requires(align.is_power_of_two()), - kani::ensures(|&p| (len + p) % align.get() == 0), + // A power-of-two alignment divides the `usize` modulus, so wrapping + // preserves congruence even when the next aligned value exceeds `usize`. + kani::ensures(|&p| len.wrapping_add(p) % align.get() == 0), // Ensures that we add the minimum required padding. kani::ensures(|&p| p < align.get()), )] @@ -562,31 +563,6 @@ mod len_of { self.meta } - #[inline] - pub(crate) fn padding_needed_for(&self) -> usize - where - T: KnownLayout, - { - let trailing_slice_layout = crate::trailing_slice_layout::(); - - // FIXME(#67): Remove this allow. See NumExt for more details. - #[allow( - unstable_name_collisions, - clippy::incompatible_msrv, - clippy::multiple_unsafe_ops_per_block - )] - // SAFETY: By invariant on `self`, a `&T` with metadata `self.meta` - // describes an object of size `<= isize::MAX`. This computes the - // size of such a `&T` without any trailing padding, and so neither - // the multiplication nor the addition will overflow. - let unpadded_size = unsafe { - let trailing_size = self.meta.unchecked_mul(trailing_slice_layout.elem_size); - trailing_size.unchecked_add(trailing_slice_layout.offset) - }; - - util::padding_needed_for(unpadded_size, T::LAYOUT.align) - } - #[inline(always)] pub(crate) fn validate_cast_and_convert_metadata( addr: usize, diff --git a/zerocopy/tests/codegen.rs b/zerocopy/tests/codegen.rs index cc3b0eb34e..92dc92846f 100644 --- a/zerocopy/tests/codegen.rs +++ b/zerocopy/tests/codegen.rs @@ -123,10 +123,15 @@ fn run_codegen_test(bench_name: &str, target_cpu: &str, bless: bool) { if bless { std::fs::write(expected_file_path, actual_result.as_bytes()).unwrap(); } else { - let expected_result = std::fs::read(expected_file_path).unwrap_or_default(); + let expected_result = std::fs::read(&expected_file_path).unwrap_or_default(); if actual_result.as_bytes() != expected_result { let expected = String::from_utf8_lossy(&expected_result[..]); - panic!("Bless codegen tests with BLESS=1\nGot unexpected output:\n{}", expected); + panic!( + "Bless codegen tests with BLESS=1\nSnapshot: {}\nExpected:\n{}\nActual:\n{}", + expected_file_path.display(), + expected, + actual_result + ); } } }; @@ -153,7 +158,11 @@ fn codegen() { }) .collect(); - let failed = handles.into_iter().any(|handle| handle.join().unwrap().is_err()); + // Join every worker so an early failure cannot hide later diagnostics. + let mut failed = false; + for handle in handles { + failed |= handle.join().unwrap().is_err(); + } if failed { panic!("One or more codegen tests failed. See thread panics above for details."); diff --git a/zerocopy/zerocopy-derive/src/derive/mod.rs b/zerocopy/zerocopy-derive/src/derive/mod.rs index 174434561e..6823157bc4 100644 --- a/zerocopy/zerocopy-derive/src/derive/mod.rs +++ b/zerocopy/zerocopy-derive/src/derive/mod.rs @@ -103,13 +103,6 @@ pub(crate) fn derive_split_at(ctx: &Ctx, _top_level: Trait) -> Result Result::Elem); + let elem = + if repr.get_packed().is_some() { quote!(#zerocopy_crate::Unalign<#elem>) } else { elem }; + + // SAFETY: The checks above require `repr(C)` or `repr(transparent)`, and + // `FieldBounds::TRAILING_SELF` requires the trailing field to be `SplitAt`. + // Its `Elem` therefore has the size, bit validity, and `UnsafeCell` + // coverage of the actual trailing element. For an unpacked struct, the + // trailing field is aligned, so its `SplitAt` alignment guarantee applies. + // For a packed struct, wrapping its `Elem` in `Unalign` preserves those + // representation properties and gives alignment 1. Thus the trailing + // slice can be accessed as `[#elem]` even if packing misaligns its field. + // Packing does not change the field's internal layout [1], so its element + // size and byte offsets are preserved in either case. + // + // [1] Per https://doc.rust-lang.org/1.93.0/reference/type-layout.html#layout.repr.alignment.intro: + // + // `packed` may also alter the padding between fields (although it will + // not alter the padding inside of any field). Ok(ImplBlockBuilder::new(ctx, &ctx.ast.data, Trait::SplitAt, FieldBounds::TRAILING_SELF) .inner_extras(quote! { - type Elem = <#trailing_field as #zerocopy_crate::SplitAt>::Elem; + type Elem = #elem; }) .build()) } diff --git a/zerocopy/zerocopy-derive/src/output_tests/expected/split_at_repr_c_packed.expected.rs b/zerocopy/zerocopy-derive/src/output_tests/expected/split_at_repr_c_packed.expected.rs new file mode 100644 index 0000000000..e6c330bf5e --- /dev/null +++ b/zerocopy/zerocopy-derive/src/output_tests/expected/split_at_repr_c_packed.expected.rs @@ -0,0 +1,22 @@ +#[allow( + deprecated, + private_bounds, + non_local_definitions, + non_camel_case_types, + non_upper_case_globals, + non_snake_case, + non_ascii_idents, + clippy::missing_inline_in_public_items, +)] +#[deny(ambiguous_associated_items)] +#[automatically_derived] +const _: () = { + unsafe impl ::zerocopy::SplitAt for Foo + where + Self: Copy, + T: ::zerocopy::SplitAt, + { + fn only_derive_is_allowed_to_implement_this_trait() {} + type Elem = ::zerocopy::Unalign<::Elem>; + } +}; diff --git a/zerocopy/zerocopy-derive/src/output_tests/mod.rs b/zerocopy/zerocopy-derive/src/output_tests/mod.rs index ad4eb29e04..c788845d4a 100644 --- a/zerocopy/zerocopy-derive/src/output_tests/mod.rs +++ b/zerocopy/zerocopy-derive/src/output_tests/mod.rs @@ -1004,13 +1004,27 @@ fn test_split_at() { } expands to "expected/split_at_repr_transparent.expected.rs" } + test! { + SplitAt { + #[repr(C, packed)] + struct Foo(T) where Self: Copy; + } expands to "expected/split_at_repr_c_packed.expected.rs" + } + + test! { + SplitAt { + #[repr(C, packed(2))] + struct Foo(T) where Self: Copy; + } expands to "expected/split_at_repr_c_packed.expected.rs" + } + test! { SplitAt { #[repr(packed)] struct Foo(T) where Self: Copy; } expands to { ::core::compile_error! { - "must not have #[repr(packed)] attribute" + "must have #[repr(C)] or #[repr(transparent)] in order to guarantee this type's layout is splitable" } } no_build } @@ -1021,7 +1035,7 @@ fn test_split_at() { struct Foo(T) where Self: Copy; } expands to { ::core::compile_error! { - "must not have #[repr(packed)] attribute" + "must have #[repr(C)] or #[repr(transparent)] in order to guarantee this type's layout is splitable" } } no_build } diff --git a/zerocopy/zerocopy-derive/tests/ui/struct.msrv.stderr b/zerocopy/zerocopy-derive/tests/ui/struct.msrv.stderr index 6358819d63..ad2db50c39 100644 --- a/zerocopy/zerocopy-derive/tests/ui/struct.msrv.stderr +++ b/zerocopy/zerocopy-derive/tests/ui/struct.msrv.stderr @@ -317,9 +317,9 @@ error[E0277]: the trait bound `SplitAtNotKnownLayout: zerocopy_renamed::KnownLay | ^^^^^^^ the trait `zerocopy_renamed::KnownLayout` is not implemented for `SplitAtNotKnownLayout` | note: required by a bound in `SplitAt` - --> $WORKSPACE/src/split_at.rs:63:27 + --> $WORKSPACE/src/split_at.rs:97:27 | -63 | pub unsafe trait SplitAt: KnownLayout { +97 | pub unsafe trait SplitAt: KnownLayout { | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `SplitAt` = note: this error originates in the derive macro `SplitAt` (in Nightly builds, run with -Z macro-backtrace for more info) diff --git a/zerocopy/zerocopy-derive/tests/ui/struct.nightly.stderr b/zerocopy/zerocopy-derive/tests/ui/struct.nightly.stderr index 49b44a751b..07e443a31b 100644 --- a/zerocopy/zerocopy-derive/tests/ui/struct.nightly.stderr +++ b/zerocopy/zerocopy-derive/tests/ui/struct.nightly.stderr @@ -484,7 +484,7 @@ help: the trait `zerocopy_renamed::KnownLayout` is not implemented for `SplitAtN $IMPLEMENTATION_SAMPLES and $OTHER_TYPES others note: required by a bound in `SplitAt` - --> src/split_at.rs:63:0 + --> src/split_at.rs:97:0 = note: this error originates in the derive macro `SplitAt` (in Nightly builds, run with -Z macro-backtrace for more info) error[E0277]: the trait bound `u8: SplitAt` is not satisfied @@ -502,9 +502,12 @@ help: the following other types implement trait `SplitAt` ... 363 | #[derive(SplitAt, KnownLayout)] | ^^^^^^^ `SplitAtSized` - --> src/split_at.rs:253:0 + --> src/split_at.rs:289:0 | = note: `[T]` + ::: src/split_at.rs:312:0 + | + = note: `ManuallyDrop` = note: this error originates in the derive macro `SplitAt` (in Nightly builds, run with -Z macro-backtrace for more info) help: add `#![feature(trivial_bounds)]` to the crate attributes to enable | diff --git a/zerocopy/zerocopy-derive/tests/ui/struct.stable.stderr b/zerocopy/zerocopy-derive/tests/ui/struct.stable.stderr index 27038d9b49..1ae10d808c 100644 --- a/zerocopy/zerocopy-derive/tests/ui/struct.stable.stderr +++ b/zerocopy/zerocopy-derive/tests/ui/struct.stable.stderr @@ -428,7 +428,7 @@ help: the trait `zerocopy_renamed::KnownLayout` is not implemented for `SplitAtN $IMPLEMENTATION_SAMPLES and $OTHER_TYPES others note: required by a bound in `SplitAt` - --> src/split_at.rs:63:0 + --> src/split_at.rs:97:0 = note: this error originates in the derive macro `SplitAt` (in Nightly builds, run with -Z macro-backtrace for more info) error[E0277]: the trait bound `u8: SplitAt` is not satisfied @@ -446,9 +446,12 @@ help: the following other types implement trait `SplitAt` ... 363 | #[derive(SplitAt, KnownLayout)] | ^^^^^^^ `SplitAtSized` - --> src/split_at.rs:253:0 + --> src/split_at.rs:289:0 | = note: `[T]` + ::: src/split_at.rs:312:0 + | + = note: `ManuallyDrop` = note: this error originates in the derive macro `SplitAt` (in Nightly builds, run with -Z macro-backtrace for more info) error[E0277]: the trait bound `AU16: zerocopy_renamed::Unaligned` is not satisfied