From 44e18ef93995f3040660750b527e5becf85899d0 Mon Sep 17 00:00:00 2001 From: Jeff Bolz Date: Sat, 18 Jan 2025 02:26:50 -0600 Subject: [PATCH 01/26] vulkan: fix coopmat2 flash attention for non-contiguous inputs (#11281) Add code similar to mul_mm_cm2 to force alignment of strides, to avoid a performance regression. Add noncontiguous FA tests in test-backend-ops. Fixes #11268. --- ggml/src/ggml-vulkan/ggml-vulkan.cpp | 43 ++++++++++++++++--- .../vulkan-shaders/flash_attn_cm2.comp | 20 +++++++++ tests/test-backend-ops.cpp | 31 ++++++++++--- 3 files changed, 82 insertions(+), 12 deletions(-) diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp index 8e3e91495..437e9cdcc 100644 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp @@ -386,10 +386,13 @@ struct vk_flash_attn_push_constants { uint32_t nev3; uint32_t nem1; + uint32_t nb01; uint32_t nb02; uint32_t nb03; + uint32_t nb11; uint32_t nb12; uint32_t nb13; + uint32_t nb21; uint32_t nb22; uint32_t nb23; uint32_t nb31; @@ -4809,7 +4812,14 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx } assert(pipelines); - bool aligned = (KV % pipelines[1]->align) == 0; + const uint32_t q_stride = (uint32_t)(nbq1 / ggml_type_size(q->type)); + const uint32_t k_stride = (uint32_t)(nbk1 / ggml_type_size(k->type)); + const uint32_t v_stride = (uint32_t)(nbv1 / ggml_type_size(v->type)); + + bool aligned = (KV % pipelines[1]->align) == 0 && + // the "aligned" shader variant will forcibly align strides, for performance + (q_stride & 7) == 0 && (k_stride & 7) == 0 && (v_stride & 7) == 0; + vk_pipeline pipeline = pipelines[aligned]; assert(pipeline); @@ -4845,15 +4855,15 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx if (ctx->device->uma) { ggml_vk_host_get(ctx->device, q->data, d_Q, q_buf_offset); - ggml_vk_host_get(ctx->device, k->data, d_K, q_buf_offset); - ggml_vk_host_get(ctx->device, v->data, d_V, q_buf_offset); - ggml_vk_host_get(ctx->device, dst->data, d_D, q_buf_offset); + ggml_vk_host_get(ctx->device, k->data, d_K, k_buf_offset); + ggml_vk_host_get(ctx->device, v->data, d_V, v_buf_offset); + ggml_vk_host_get(ctx->device, dst->data, d_D, d_buf_offset); Q_uma = d_Q != nullptr; K_uma = d_K != nullptr; V_uma = d_V != nullptr; D_uma = d_D != nullptr; if (mask) { - ggml_vk_host_get(ctx->device, mask->data, d_M, q_buf_offset); + ggml_vk_host_get(ctx->device, mask->data, d_M, m_buf_offset); M_uma = d_M != nullptr; } } @@ -4891,7 +4901,18 @@ static void ggml_vk_flash_attn(ggml_backend_vk_context * ctx, vk_context& subctx } } - const vk_flash_attn_push_constants pc = { N, KV, (uint32_t)ne1, (uint32_t)ne2, (uint32_t)ne3, (uint32_t)neq2, (uint32_t)neq3, (uint32_t)nek2, (uint32_t)nek3, (uint32_t)nev2, (uint32_t)nev3, nem1, (uint32_t)nbq2, (uint32_t)nbq3, (uint32_t)nbk2, (uint32_t)nbk3, (uint32_t)nbv2, (uint32_t)nbv3, nbm1, scale, max_bias, logit_softcap, mask != nullptr, n_head_log2, m0, m1 }; + const vk_flash_attn_push_constants pc = { N, KV, + (uint32_t)ne1, (uint32_t)ne2, (uint32_t)ne3, + (uint32_t)neq2, (uint32_t)neq3, + (uint32_t)nek2, (uint32_t)nek3, + (uint32_t)nev2, (uint32_t)nev3, + nem1, + q_stride, (uint32_t)nbq2, (uint32_t)nbq3, + k_stride, (uint32_t)nbk2, (uint32_t)nbk3, + v_stride, (uint32_t)nbv2, (uint32_t)nbv3, + nbm1, + scale, max_bias, logit_softcap, + mask != nullptr, n_head_log2, m0, m1 }; ggml_vk_dispatch_pipeline(ctx, subctx, pipeline, { vk_subbuffer{d_Q, q_buf_offset, VK_WHOLE_SIZE}, @@ -8668,6 +8689,7 @@ static void ggml_vk_check_results_1(ggml_tensor * tensor) { ggml_tensor * src0 = tensor->src[0]; ggml_tensor * src1 = tensor->src[1]; ggml_tensor * src2 = tensor->src[2]; + ggml_tensor * src3 = tensor->src[3]; void * tensor_data = tensor->data; @@ -8730,6 +8752,9 @@ static void ggml_vk_check_results_1(ggml_tensor * tensor) { if (src2 != nullptr) { std::cerr << "src2=" << src2 << " src2->name=" << src2->name << " op=" << ggml_op_name(src2->op) << " type=" << ggml_type_name(src2->type) << " ne0=" << src2->ne[0] << " nb0=" << src2->nb[0] << " ne1=" << src2->ne[1] << " nb1=" << src2->nb[1] << " ne2=" << src2->ne[2] << " nb2=" << src2->nb[2] << " ne3=" << src2->ne[3] << " nb3=" << src2->nb[3] << " offset=" << src2->view_offs << std::endl; } + if (src3 != nullptr) { + std::cerr << "src3=" << src3 << " src3->name=" << src3->name << " op=" << ggml_op_name(src3->op) << " type=" << ggml_type_name(src3->type) << " ne0=" << src3->ne[0] << " nb0=" << src3->nb[0] << " ne1=" << src3->ne[1] << " nb1=" << src3->nb[1] << " ne2=" << src3->ne[2] << " nb2=" << src3->nb[2] << " ne3=" << src3->ne[3] << " nb3=" << src3->nb[3] << " offset=" << src3->view_offs << std::endl; + } std::cerr << "First error: result=" << first_error_result << " correct=" << first_error_correct << " i3=" << first_error[3] << " i2=" << first_error[2] << " i1=" << first_error[1] << " i0=" << first_error[0] << std::endl; std::cerr << std::endl << "Result:" << std::endl; ggml_vk_print_tensor_area(tensor, tensor_data, i0, i1, i2, i3); @@ -8774,6 +8799,9 @@ static void ggml_vk_check_results_1(ggml_tensor * tensor) { if (src2 != nullptr) { std::cerr << "src2=" << src2 << " op=" << ggml_op_name(src2->op) << " type=" << ggml_type_name(src2->type) << " ne0=" << src2->ne[0] << " nb0=" << src2->nb[0] << " ne1=" << src2->ne[1] << " nb1=" << src2->nb[1] << " ne2=" << src2->ne[2] << " nb2=" << src2->nb[2] << " ne3=" << src2->ne[3] << " nb3=" << src2->nb[3] << " offset=" << src2->view_offs << std::endl; } + if (src3 != nullptr) { + std::cerr << "src3=" << src3 << " op=" << ggml_op_name(src3->op) << " type=" << ggml_type_name(src3->type) << " ne0=" << src3->ne[0] << " nb0=" << src3->nb[0] << " ne1=" << src3->ne[1] << " nb1=" << src3->nb[1] << " ne2=" << src3->ne[2] << " nb2=" << src3->nb[2] << " ne3=" << src3->ne[3] << " nb3=" << src3->nb[3] << " offset=" << src3->view_offs << std::endl; + } std::cerr << "First error: result=" << first_error_result << " correct=" << first_error_correct << " i3=" << first_error[3] << " i2=" << first_error[2] << " i1=" << first_error[1] << " i0=" << first_error[0] << std::endl; std::cerr << std::endl << "Result:" << std::endl; ggml_vk_print_tensor_area(tensor, tensor_data, 5, 5, 0, 0); @@ -8796,6 +8824,9 @@ static void ggml_vk_check_results_1(ggml_tensor * tensor) { if (src2 != nullptr) { std::cerr << "src2=" << src2 << " op=" << ggml_op_name(src2->op) << " type=" << ggml_type_name(src2->type) << " ne0=" << src2->ne[0] << " nb0=" << src2->nb[0] << " ne1=" << src2->ne[1] << " nb1=" << src2->nb[1] << " ne2=" << src2->ne[2] << " nb2=" << src2->nb[2] << " ne3=" << src2->ne[3] << " nb3=" << src2->nb[3] << " offset=" << src2->view_offs << std::endl; } + if (src3 != nullptr) { + std::cerr << "src3=" << src3 << " op=" << ggml_op_name(src3->op) << " type=" << ggml_type_name(src3->type) << " ne0=" << src3->ne[0] << " nb0=" << src3->nb[0] << " ne1=" << src3->ne[1] << " nb1=" << src3->nb[1] << " ne2=" << src3->ne[2] << " nb2=" << src3->nb[2] << " ne3=" << src3->ne[3] << " nb3=" << src3->nb[3] << " offset=" << src3->view_offs << std::endl; + } std::cerr << "First error: result=" << first_error_result << " correct=" << first_error_correct << " i3=" << first_error[3] << " i2=" << first_error[2] << " i1=" << first_error[1] << " i0=" << first_error[0] << std::endl; std::cerr << std::endl << "Result:" << std::endl; ggml_vk_print_tensor_area(tensor, tensor_data, first_error[0], first_error[1], first_error[2], first_error[3]); diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp index c5be8131b..ca3a59b8f 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp @@ -42,10 +42,13 @@ layout (push_constant) uniform parameter { uint32_t nev3; uint32_t nem1; + uint32_t nb01; uint32_t nb02; uint32_t nb03; + uint32_t nb11; uint32_t nb12; uint32_t nb13; + uint32_t nb21; uint32_t nb22; uint32_t nb23; uint32_t nb31; @@ -146,6 +149,23 @@ void main() { tensorLayoutK = setTensorLayoutDimensionNV(tensorLayoutK, KV, D); tensorLayoutV = setTensorLayoutDimensionNV(tensorLayoutV, KV, D); + // nb?1 are already divided by the type size and are in units of elements + uint32_t q_stride = p.nb01; + uint32_t k_stride = p.nb11; + uint32_t v_stride = p.nb21; + // hint to the compiler that strides are aligned for the aligned variant of the shader + if (Clamp != gl_CooperativeMatrixClampModeConstantNV) + { + q_stride &= ~7; +#if !defined(BLOCK_SIZE) + k_stride &= ~7; + v_stride &= ~7; +#endif + } + tensorLayoutQ = setTensorLayoutStrideNV(tensorLayoutQ, q_stride, 1); + tensorLayoutK = setTensorLayoutStrideNV(tensorLayoutK, k_stride, 1); + tensorLayoutV = setTensorLayoutStrideNV(tensorLayoutV, v_stride, 1); + coopmat Q; coopmat Qf16; diff --git a/tests/test-backend-ops.cpp b/tests/test-backend-ops.cpp index 5cde8289f..74d1bee39 100644 --- a/tests/test-backend-ops.cpp +++ b/tests/test-backend-ops.cpp @@ -3046,9 +3046,10 @@ struct test_flash_attn_ext : public test_case { const float logit_softcap; // Gemma 2 const ggml_type type_KV; + std::array permute; std::string vars() override { - return VARS_TO_STR8(hs, nh, kv, nb, mask, max_bias, logit_softcap, type_KV); + return VARS_TO_STR9(hs, nh, kv, nb, mask, max_bias, logit_softcap, type_KV, permute); } double max_nmse_err() override { @@ -3063,19 +3064,33 @@ struct test_flash_attn_ext : public test_case { } test_flash_attn_ext(int64_t hs = 128, int64_t nh = 32, int64_t kv = 96, int64_t nb = 8, - bool mask = true, float max_bias = 0.0f, float logit_softcap = 0.0f, ggml_type type_KV = GGML_TYPE_F16) - : hs(hs), nh(nh), kv(kv), nb(nb), mask(mask), max_bias(max_bias), logit_softcap(logit_softcap), type_KV(type_KV) {} + bool mask = true, float max_bias = 0.0f, float logit_softcap = 0.0f, ggml_type type_KV = GGML_TYPE_F16, + std::array permute = {0, 1, 2, 3}) + : hs(hs), nh(nh), kv(kv), nb(nb), mask(mask), max_bias(max_bias), logit_softcap(logit_softcap), type_KV(type_KV), permute(permute) {} ggml_tensor * build_graph(ggml_context * ctx) override { const int64_t hs_padded = GGML_PAD(hs, ggml_blck_size(type_KV)); - ggml_tensor * q = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, hs_padded, nb, nh, 1); + auto const &create_permuted = [&](ggml_type type, int64_t ne0, int64_t ne1, int64_t ne2, int64_t ne3) -> ggml_tensor * { + int64_t ne[4] = {ne0, ne1, ne2, ne3}; + int64_t ne_perm[4]; + for (int i = 0; i < 4; ++i) { + ne_perm[permute[i]] = ne[i]; + } + ggml_tensor * t = ggml_new_tensor_4d(ctx, type, ne_perm[0], ne_perm[1], ne_perm[2], ne_perm[3]); + if (permute != std::array{0, 1, 2, 3}) { + t = ggml_permute(ctx, t, permute[0], permute[1], permute[2], permute[3]); + } + return t; + }; + + ggml_tensor * q = create_permuted(GGML_TYPE_F32, hs_padded, nb, nh, 1); ggml_set_name(q, "q"); - ggml_tensor * k = ggml_new_tensor_4d(ctx, type_KV, hs_padded, kv, nh, 1); + ggml_tensor * k = create_permuted(type_KV, hs_padded, kv, nh, 1); ggml_set_name(k, "k"); - ggml_tensor * v = ggml_new_tensor_4d(ctx, type_KV, hs_padded, kv, nh, 1); + ggml_tensor * v = create_permuted(type_KV, hs_padded, kv, nh, 1); ggml_set_name(v, "v"); ggml_tensor * m = nullptr; @@ -4167,6 +4182,10 @@ static std::vector> make_test_cases_eval() { for (int nb : { 1, 3, 32, 35, }) { for (ggml_type type_KV : {GGML_TYPE_F16, GGML_TYPE_BF16, GGML_TYPE_Q8_0, GGML_TYPE_Q4_0}) { test_cases.emplace_back(new test_flash_attn_ext(hs, nh, kv, nb, mask, max_bias, logit_softcap, type_KV)); + // run fewer test cases permuted + if (mask == true && max_bias == 0.0f && logit_softcap == 0 && kv == 512) { + test_cases.emplace_back(new test_flash_attn_ext(hs, nh, kv, nb, mask, max_bias, logit_softcap, type_KV, {0, 2, 1, 3})); + } } } } From 6390a998bfc63241fde8509022b0768a71bf20bb Mon Sep 17 00:00:00 2001 From: LostRuins Concedo <39025047+LostRuins@users.noreply.github.com> Date: Sat, 18 Jan 2025 18:20:57 +0800 Subject: [PATCH 02/26] tts : add guide tokens support (#11186) * Added the ability to use guide tokens for OuteTTS, greatly improving TTS recitation accuracy over long input sequences. * applied linting suggestions, updated to latest llama_vocab changes, added a safety check, added newline to guide token start --- common/arg.cpp | 7 +++++++ common/common.h | 2 ++ examples/tts/tts.cpp | 45 +++++++++++++++++++++++++++++++++++++++++++- 3 files changed, 53 insertions(+), 1 deletion(-) diff --git a/common/arg.cpp b/common/arg.cpp index 9069950eb..dede335fb 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -2254,6 +2254,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex params.vocoder.model = value; } ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER})); + add_opt(common_arg( + {"--tts-use-guide-tokens"}, + "Use guide tokens to improve TTS word recall", + [](common_params & params) { + params.vocoder.use_guide_tokens = true; + } + ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER})); // model-specific add_opt(common_arg( diff --git a/common/common.h b/common/common.h index 691141d6b..3bcc637cc 100644 --- a/common/common.h +++ b/common/common.h @@ -184,6 +184,8 @@ struct common_params_vocoder { std::string model = ""; // model path // NOLINT std::string model_url = ""; // model url to download // NOLINT + + bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy // NOLINT }; struct common_params { diff --git a/examples/tts/tts.cpp b/examples/tts/tts.cpp index 5a9161181..f78f76303 100644 --- a/examples/tts/tts.cpp +++ b/examples/tts/tts.cpp @@ -425,6 +425,33 @@ static void prompt_init(llama_tokens & prompt, const llama_vocab * vocab) { prompt_add(prompt, vocab, "<|im_start|>\n", true, true); } +static std::vector prepare_guide_tokens(const llama_vocab * vocab, const std::string & str) { + const std::string& delimiter = "<|text_sep|>"; + + std::vector result; + size_t start = 0; + size_t end = str.find(delimiter); + + //first token is always a newline, as it was not previously added + result.push_back(common_tokenize(vocab, "\n", false, true)[0]); + + while (end != std::string::npos) { + std::string current_word = str.substr(start, end - start); + auto tmp = common_tokenize(vocab, current_word, false, true); + result.push_back(tmp[0]); + start = end + delimiter.length(); + end = str.find(delimiter, start); + } + + // Add the last part + std::string current_word = str.substr(start); + auto tmp = common_tokenize(vocab, current_word, false, true); + if (tmp.size() > 0) { + result.push_back(tmp[0]); + } + return result; +} + int main(int argc, char ** argv) { common_params params; @@ -494,6 +521,7 @@ int main(int argc, char ** argv) { const auto t_main_start = ggml_time_us(); std::vector codes; + std::vector guide_tokens; // process prompt and generate voice codes { @@ -508,6 +536,9 @@ int main(int argc, char ** argv) { // convert the input text into the necessary format expected by OuteTTS { std::string prompt_clean = process_text(params.prompt); + if (params.vocoder.use_guide_tokens) { + guide_tokens = prepare_guide_tokens(vocab, prompt_clean); + } LOG_INF("%s: prompt: '%s'\n", __func__, prompt_clean.c_str()); @@ -717,6 +748,8 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14 int n_past = batch.n_tokens; int n_decode = 0; + bool next_token_uses_guide_token = true; + while (n_decode <= n_predict) { // prepare the next batch common_batch_clear(batch); @@ -728,7 +761,17 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14 continue; } - const llama_token new_token_id = common_sampler_sample(smpl[i], ctx_ttc, i_batch[i]); + llama_token new_token_id = common_sampler_sample(smpl[i], ctx_ttc, i_batch[i]); + + //guide tokens help prevent hallucinations by forcing the TTS to use the correct word + if (!guide_tokens.empty() && next_token_uses_guide_token && !llama_vocab_is_control(vocab, new_token_id) && !llama_vocab_is_eog(vocab, new_token_id)) { + llama_token guide_token = guide_tokens[0]; + guide_tokens.erase(guide_tokens.begin()); + new_token_id = guide_token; //ensure correct word fragment is used + } + + //this is the token id that always precedes a new word + next_token_uses_guide_token = (new_token_id == 198); common_sampler_accept(smpl[i], new_token_id, true); From f26c87417999209e7f4576b4f3ecf7a5b9c66a29 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sat, 18 Jan 2025 13:18:32 +0200 Subject: [PATCH 03/26] scripts : restore hf.sh (#11288) ggml-ci --- scripts/hf.sh | 112 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 112 insertions(+) create mode 100755 scripts/hf.sh diff --git a/scripts/hf.sh b/scripts/hf.sh new file mode 100755 index 000000000..b251925fa --- /dev/null +++ b/scripts/hf.sh @@ -0,0 +1,112 @@ +#!/bin/bash +# +# Shortcut for downloading HF models +# +# Usage: +# ./llama-cli -m $(./scripts/hf.sh https://huggingface.co/TheBloke/Mixtral-8x7B-v0.1-GGUF/resolve/main/mixtral-8x7b-v0.1.Q4_K_M.gguf) +# ./llama-cli -m $(./scripts/hf.sh --url https://huggingface.co/TheBloke/Mixtral-8x7B-v0.1-GGUF/blob/main/mixtral-8x7b-v0.1.Q4_K_M.gguf) +# ./llama-cli -m $(./scripts/hf.sh --repo TheBloke/Mixtral-8x7B-v0.1-GGUF --file mixtral-8x7b-v0.1.Q4_K_M.gguf) +# + +# all logs go to stderr +function log { + echo "$@" 1>&2 +} + +function usage { + log "Usage: $0 [[--url] ] [--repo ] [--file ] [--outdir [-h|--help]" + exit 1 +} + +# check for curl or wget +function has_cmd { + if ! [ -x "$(command -v $1)" ]; then + return 1 + fi +} + +if has_cmd wget; then + cmd="wget -q -c -O %s/%s %s" +elif has_cmd curl; then + cmd="curl -C - -f --output-dir %s -o %s -L %s" +else + log "[E] curl or wget not found" + exit 1 +fi + +url="" +repo="" +file="" +outdir="." + +# parse args +while [[ $# -gt 0 ]]; do + case "$1" in + --url) + url="$2" + shift 2 + ;; + --repo) + repo="$2" + shift 2 + ;; + --file) + file="$2" + shift 2 + ;; + --outdir) + outdir="$2" + shift 2 + ;; + -h|--help) + usage + ;; + *) + url="$1" + shift + ;; + esac +done + +if [ -n "$repo" ] && [ -n "$file" ]; then + url="https://huggingface.co/$repo/resolve/main/$file" +fi + +if [ -z "$url" ]; then + log "[E] missing --url" + usage +fi + +# check if the URL is a HuggingFace model, and if so, try to download it +is_url=false + +if [[ ${#url} -gt 22 ]]; then + if [[ ${url:0:22} == "https://huggingface.co" ]]; then + is_url=true + fi +fi + +if [ "$is_url" = false ]; then + log "[E] invalid URL, must start with https://huggingface.co" + exit 0 +fi + +# replace "blob/main" with "resolve/main" +url=${url/blob\/main/resolve\/main} + +basename=$(basename $url) + +log "[+] attempting to download $basename" + +if [ -n "$cmd" ]; then + cmd=$(printf "$cmd" "$outdir" "$basename" "$url") + log "[+] $cmd" + if $cmd; then + echo $outdir/$basename + exit 0 + fi +fi + +log "[-] failed to download" + +exit 1 From f30f099228f774209aa3010b78dfbe5d262e69aa Mon Sep 17 00:00:00 2001 From: Xuan Son Nguyen Date: Sat, 18 Jan 2025 14:12:05 +0100 Subject: [PATCH 04/26] server : implement cancellable request (#11285) * server : implement cancellable request * fix typo * httplib 0.18.5 * fix i underflow --- examples/server/httplib.h | 1726 +++++++++++++---- examples/server/server.cpp | 73 +- examples/server/tests/unit/test_completion.py | 21 + examples/server/tests/utils.py | 7 +- 4 files changed, 1396 insertions(+), 431 deletions(-) diff --git a/examples/server/httplib.h b/examples/server/httplib.h index f360bd93e..c2f12dd2a 100644 --- a/examples/server/httplib.h +++ b/examples/server/httplib.h @@ -8,7 +8,7 @@ #ifndef CPPHTTPLIB_HTTPLIB_H #define CPPHTTPLIB_HTTPLIB_H -#define CPPHTTPLIB_VERSION "0.15.3" +#define CPPHTTPLIB_VERSION "0.18.5" /* * Configuration @@ -18,8 +18,12 @@ #define CPPHTTPLIB_KEEPALIVE_TIMEOUT_SECOND 5 #endif +#ifndef CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND +#define CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND 10000 +#endif + #ifndef CPPHTTPLIB_KEEPALIVE_MAX_COUNT -#define CPPHTTPLIB_KEEPALIVE_MAX_COUNT 5 +#define CPPHTTPLIB_KEEPALIVE_MAX_COUNT 100 #endif #ifndef CPPHTTPLIB_CONNECTION_TIMEOUT_SECOND @@ -30,20 +34,36 @@ #define CPPHTTPLIB_CONNECTION_TIMEOUT_USECOND 0 #endif -#ifndef CPPHTTPLIB_READ_TIMEOUT_SECOND -#define CPPHTTPLIB_READ_TIMEOUT_SECOND 5 +#ifndef CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND +#define CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND 5 #endif -#ifndef CPPHTTPLIB_READ_TIMEOUT_USECOND -#define CPPHTTPLIB_READ_TIMEOUT_USECOND 0 +#ifndef CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND +#define CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND 0 #endif -#ifndef CPPHTTPLIB_WRITE_TIMEOUT_SECOND -#define CPPHTTPLIB_WRITE_TIMEOUT_SECOND 5 +#ifndef CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND +#define CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND 5 #endif -#ifndef CPPHTTPLIB_WRITE_TIMEOUT_USECOND -#define CPPHTTPLIB_WRITE_TIMEOUT_USECOND 0 +#ifndef CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND +#define CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND 0 +#endif + +#ifndef CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND +#define CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND 300 +#endif + +#ifndef CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND +#define CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND 0 +#endif + +#ifndef CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND +#define CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND 5 +#endif + +#ifndef CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND +#define CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND 0 #endif #ifndef CPPHTTPLIB_IDLE_INTERVAL_SECOND @@ -90,8 +110,12 @@ #define CPPHTTPLIB_TCP_NODELAY false #endif +#ifndef CPPHTTPLIB_IPV6_V6ONLY +#define CPPHTTPLIB_IPV6_V6ONLY false +#endif + #ifndef CPPHTTPLIB_RECV_BUFSIZ -#define CPPHTTPLIB_RECV_BUFSIZ size_t(4096u) +#define CPPHTTPLIB_RECV_BUFSIZ size_t(16384u) #endif #ifndef CPPHTTPLIB_COMPRESSION_BUFSIZ @@ -145,11 +169,11 @@ using ssize_t = long; #endif // _MSC_VER #ifndef S_ISREG -#define S_ISREG(m) (((m)&S_IFREG) == S_IFREG) +#define S_ISREG(m) (((m) & S_IFREG) == S_IFREG) #endif // S_ISREG #ifndef S_ISDIR -#define S_ISDIR(m) (((m)&S_IFDIR) == S_IFDIR) +#define S_ISDIR(m) (((m) & S_IFDIR) == S_IFDIR) #endif // S_ISDIR #ifndef NOMINMAX @@ -269,7 +293,12 @@ using socket_t = int; #include #include -#if OPENSSL_VERSION_NUMBER < 0x30000000L +#if defined(OPENSSL_IS_BORINGSSL) || defined(LIBRESSL_VERSION_NUMBER) +#if OPENSSL_VERSION_NUMBER < 0x1010107f +#error Please use OpenSSL or a current version of BoringSSL +#endif +#define SSL_get1_peer_certificate SSL_get_peer_certificate +#elif OPENSSL_VERSION_NUMBER < 0x30000000L #error Sorry, OpenSSL versions prior to 3.0.0 are not supported #endif @@ -312,16 +341,63 @@ make_unique(std::size_t n) { return std::unique_ptr(new RT[n]); } -struct ci { - bool operator()(const std::string &s1, const std::string &s2) const { - return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), - s2.end(), - [](unsigned char c1, unsigned char c2) { - return ::tolower(c1) < ::tolower(c2); - }); +namespace case_ignore { + +inline unsigned char to_lower(int c) { + const static unsigned char table[256] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, + 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, + 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, + 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, + 60, 61, 62, 63, 64, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, + 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, + 122, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, + 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, + 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, + 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, + 150, 151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, + 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, + 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 224, 225, 226, + 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241, + 242, 243, 244, 245, 246, 215, 248, 249, 250, 251, 252, 253, 254, 223, 224, + 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, + 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, + 255, + }; + return table[(unsigned char)(char)c]; +} + +inline bool equal(const std::string &a, const std::string &b) { + return a.size() == b.size() && + std::equal(a.begin(), a.end(), b.begin(), [](char ca, char cb) { + return to_lower(ca) == to_lower(cb); + }); +} + +struct equal_to { + bool operator()(const std::string &a, const std::string &b) const { + return equal(a, b); } }; +struct hash { + size_t operator()(const std::string &key) const { + return hash_core(key.data(), key.size(), 0); + } + + size_t hash_core(const char *s, size_t l, size_t h) const { + return (l == 0) ? h + : hash_core(s + 1, l - 1, + // Unsets the 6 high bits of h, therefore no + // overflow happens + (((std::numeric_limits::max)() >> 6) & + h * 33) ^ + static_cast(to_lower(*s))); + } +}; + +} // namespace case_ignore + // This is based on // "http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4189". @@ -427,7 +503,9 @@ enum StatusCode { NetworkAuthenticationRequired_511 = 511, }; -using Headers = std::multimap; +using Headers = + std::unordered_multimap; using Params = std::multimap; using Match = std::smatch; @@ -534,6 +612,7 @@ using Ranges = std::vector; struct Request { std::string method; std::string path; + Params params; Headers headers; std::string body; @@ -545,11 +624,11 @@ struct Request { // for server std::string version; std::string target; - Params params; MultipartFormDataMap files; Ranges ranges; Match matches; std::unordered_map path_params; + std::function is_connection_closed = []() { return true; }; // for client ResponseHandler response_handler; @@ -560,8 +639,10 @@ struct Request { #endif bool has_header(const std::string &key) const; - std::string get_header_value(const std::string &key, size_t id = 0) const; - uint64_t get_header_value_u64(const std::string &key, size_t id = 0) const; + std::string get_header_value(const std::string &key, const char *def = "", + size_t id = 0) const; + uint64_t get_header_value_u64(const std::string &key, uint64_t def = 0, + size_t id = 0) const; size_t get_header_value_count(const std::string &key) const; void set_header(const std::string &key, const std::string &val); @@ -592,8 +673,10 @@ struct Response { std::string location; // Redirect location bool has_header(const std::string &key) const; - std::string get_header_value(const std::string &key, size_t id = 0) const; - uint64_t get_header_value_u64(const std::string &key, size_t id = 0) const; + std::string get_header_value(const std::string &key, const char *def = "", + size_t id = 0) const; + uint64_t get_header_value_u64(const std::string &key, uint64_t def = 0, + size_t id = 0) const; size_t get_header_value_count(const std::string &key) const; void set_header(const std::string &key, const std::string &val); @@ -614,6 +697,10 @@ struct Response { const std::string &content_type, ContentProviderWithoutLength provider, ContentProviderResourceReleaser resource_releaser = nullptr); + void set_file_content(const std::string &path, + const std::string &content_type); + void set_file_content(const std::string &path); + Response() = default; Response(const Response &) = default; Response &operator=(const Response &) = default; @@ -631,6 +718,8 @@ struct Response { ContentProviderResourceReleaser content_provider_resource_releaser_; bool is_chunked_content_provider_ = false; bool content_provider_success_ = false; + std::string file_content_path_; + std::string file_content_content_type_; }; class Stream { @@ -646,8 +735,6 @@ public: virtual void get_local_ip_and_port(std::string &ip, int &port) const = 0; virtual socket_t socket() const = 0; - template - ssize_t write_format(const char *fmt, const Args &...args); ssize_t write(const char *ptr); ssize_t write(const std::string &s); }; @@ -719,13 +806,18 @@ private: if (pool_.shutdown_ && pool_.jobs_.empty()) { break; } - fn = std::move(pool_.jobs_.front()); + fn = pool_.jobs_.front(); pool_.jobs_.pop_front(); } assert(true == static_cast(fn)); fn(); } + +#if defined(CPPHTTPLIB_OPENSSL_SUPPORT) && !defined(OPENSSL_IS_BORINGSSL) && \ + !defined(LIBRESSL_VERSION_NUMBER) + OPENSSL_thread_stop(); +#endif } ThreadPool &pool_; @@ -787,7 +879,6 @@ public: bool match(Request &request) const override; private: - static constexpr char marker = ':'; // Treat segment separators as the end of path parameter capture // Does not need to handle query parameters as they are parsed before path // matching @@ -871,8 +962,13 @@ public: Server &set_default_file_mimetype(const std::string &mime); Server &set_file_request_handler(Handler handler); - Server &set_error_handler(HandlerWithResponse handler); - Server &set_error_handler(Handler handler); + template + Server &set_error_handler(ErrorHandlerFunc &&handler) { + return set_error_handler_core( + std::forward(handler), + std::is_convertible{}); + } + Server &set_exception_handler(ExceptionHandler handler); Server &set_pre_routing_handler(HandlerWithResponse handler); Server &set_post_routing_handler(Handler handler); @@ -882,6 +978,7 @@ public: Server &set_address_family(int family); Server &set_tcp_nodelay(bool on); + Server &set_ipv6_v6only(bool on); Server &set_socket_options(SocketOptions socket_options); Server &set_default_headers(Headers headers); @@ -914,21 +1011,24 @@ public: bool is_running() const; void wait_until_ready() const; void stop(); + void decommission(); std::function new_task_queue; protected: - bool process_request(Stream &strm, bool close_connection, + bool process_request(Stream &strm, const std::string &remote_addr, + int remote_port, const std::string &local_addr, + int local_port, bool close_connection, bool &connection_closed, const std::function &setup_request); std::atomic svr_sock_{INVALID_SOCKET}; size_t keep_alive_max_count_ = CPPHTTPLIB_KEEPALIVE_MAX_COUNT; time_t keep_alive_timeout_sec_ = CPPHTTPLIB_KEEPALIVE_TIMEOUT_SECOND; - time_t read_timeout_sec_ = CPPHTTPLIB_READ_TIMEOUT_SECOND; - time_t read_timeout_usec_ = CPPHTTPLIB_READ_TIMEOUT_USECOND; - time_t write_timeout_sec_ = CPPHTTPLIB_WRITE_TIMEOUT_SECOND; - time_t write_timeout_usec_ = CPPHTTPLIB_WRITE_TIMEOUT_USECOND; + time_t read_timeout_sec_ = CPPHTTPLIB_SERVER_READ_TIMEOUT_SECOND; + time_t read_timeout_usec_ = CPPHTTPLIB_SERVER_READ_TIMEOUT_USECOND; + time_t write_timeout_sec_ = CPPHTTPLIB_SERVER_WRITE_TIMEOUT_SECOND; + time_t write_timeout_usec_ = CPPHTTPLIB_SERVER_WRITE_TIMEOUT_USECOND; time_t idle_interval_sec_ = CPPHTTPLIB_IDLE_INTERVAL_SECOND; time_t idle_interval_usec_ = CPPHTTPLIB_IDLE_INTERVAL_USECOND; size_t payload_max_length_ = CPPHTTPLIB_PAYLOAD_MAX_LENGTH; @@ -943,6 +1043,9 @@ private: static std::unique_ptr make_matcher(const std::string &pattern); + Server &set_error_handler_core(HandlerWithResponse handler, std::true_type); + Server &set_error_handler_core(Handler handler, std::false_type); + socket_t create_server_socket(const std::string &host, int port, int socket_flags, SocketOptions socket_options) const; @@ -985,7 +1088,7 @@ private: virtual bool process_and_close_socket(socket_t sock); std::atomic is_running_{false}; - std::atomic done_{false}; + std::atomic is_decommisioned{false}; struct MountPointEntry { std::string mount_point; @@ -1018,6 +1121,7 @@ private: int address_family_ = AF_UNSPEC; bool tcp_nodelay_ = CPPHTTPLIB_TCP_NODELAY; + bool ipv6_v6only_ = CPPHTTPLIB_IPV6_V6ONLY; SocketOptions socket_options_ = default_socket_options; Headers default_headers_; @@ -1037,6 +1141,7 @@ enum class Error { SSLConnection, SSLLoadingCerts, SSLServerVerification, + SSLServerHostnameVerification, UnsupportedMultipartBoundaryChars, Compression, ConnectionTimeout, @@ -1074,9 +1179,10 @@ public: // Request Headers bool has_request_header(const std::string &key) const; std::string get_request_header_value(const std::string &key, + const char *def = "", size_t id = 0) const; uint64_t get_request_header_value_u64(const std::string &key, - size_t id = 0) const; + uint64_t def = 0, size_t id = 0) const; size_t get_request_header_value_count(const std::string &key) const; private: @@ -1140,10 +1246,18 @@ public: const std::string &content_type); Result Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Post(const std::string &path, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1159,6 +1273,8 @@ public: Result Post(const std::string &path, const Params ¶ms); Result Post(const std::string &path, const Headers &headers, const Params ¶ms); + Result Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Post(const std::string &path, const MultipartFormDataItems &items); Result Post(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1173,10 +1289,18 @@ public: const std::string &content_type); Result Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Put(const std::string &path, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); Result Put(const std::string &path, @@ -1191,6 +1315,8 @@ public: Result Put(const std::string &path, const Params ¶ms); Result Put(const std::string &path, const Headers &headers, const Params ¶ms); + Result Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Put(const std::string &path, const MultipartFormDataItems &items); Result Put(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1203,13 +1329,23 @@ public: Result Patch(const std::string &path); Result Patch(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1227,13 +1363,24 @@ public: Result Delete(const std::string &path, const Headers &headers); Result Delete(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Options(const std::string &path); Result Options(const std::string &path, const Headers &headers); @@ -1258,6 +1405,7 @@ public: void set_address_family(int family); void set_tcp_nodelay(bool on); + void set_ipv6_v6only(bool on); void set_socket_options(SocketOptions socket_options); void set_connection_timeout(time_t sec, time_t usec = 0); @@ -1309,6 +1457,8 @@ public: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT void enable_server_certificate_verification(bool enabled); + void enable_server_hostname_verification(bool enabled); + void set_server_certificate_verifier(std::function verifier); #endif void set_logger(Logger logger); @@ -1375,10 +1525,10 @@ protected: time_t connection_timeout_sec_ = CPPHTTPLIB_CONNECTION_TIMEOUT_SECOND; time_t connection_timeout_usec_ = CPPHTTPLIB_CONNECTION_TIMEOUT_USECOND; - time_t read_timeout_sec_ = CPPHTTPLIB_READ_TIMEOUT_SECOND; - time_t read_timeout_usec_ = CPPHTTPLIB_READ_TIMEOUT_USECOND; - time_t write_timeout_sec_ = CPPHTTPLIB_WRITE_TIMEOUT_SECOND; - time_t write_timeout_usec_ = CPPHTTPLIB_WRITE_TIMEOUT_USECOND; + time_t read_timeout_sec_ = CPPHTTPLIB_CLIENT_READ_TIMEOUT_SECOND; + time_t read_timeout_usec_ = CPPHTTPLIB_CLIENT_READ_TIMEOUT_USECOND; + time_t write_timeout_sec_ = CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_SECOND; + time_t write_timeout_usec_ = CPPHTTPLIB_CLIENT_WRITE_TIMEOUT_USECOND; std::string basic_auth_username_; std::string basic_auth_password_; @@ -1395,6 +1545,7 @@ protected: int address_family_ = AF_UNSPEC; bool tcp_nodelay_ = CPPHTTPLIB_TCP_NODELAY; + bool ipv6_v6only_ = CPPHTTPLIB_IPV6_V6ONLY; SocketOptions socket_options_ = nullptr; bool compress_ = false; @@ -1422,6 +1573,8 @@ protected: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT bool server_certificate_verification_ = true; + bool server_hostname_verification_ = true; + std::function server_certificate_verifier_; #endif Logger logger_; @@ -1430,6 +1583,9 @@ private: bool send_(Request &req, Response &res, Error &error); Result send_(Request &&req); +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT + bool is_ssl_peer_could_be_closed(SSL *ssl) const; +#endif socket_t create_client_socket(Error &error) const; bool read_response_line(Stream &strm, const Request &req, Response &res) const; @@ -1448,7 +1604,7 @@ private: const Headers &headers, const char *body, size_t content_length, ContentProvider content_provider, ContentProviderWithoutLength content_provider_without_length, - const std::string &content_type); + const std::string &content_type, Progress progress); ContentProviderWithoutLength get_multipart_content_provider( const std::string &boundary, const MultipartFormDataItems &items, const MultipartFormDataProviderItems &provider_items) const; @@ -1477,6 +1633,7 @@ public: const std::string &client_key_path); Client(Client &&) = default; + Client &operator=(Client &&) = default; ~Client(); @@ -1523,10 +1680,18 @@ public: const std::string &content_type); Result Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Post(const std::string &path, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Post(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1542,6 +1707,8 @@ public: Result Post(const std::string &path, const Params ¶ms); Result Post(const std::string &path, const Headers &headers, const Params ¶ms); + Result Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Post(const std::string &path, const MultipartFormDataItems &items); Result Post(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1556,10 +1723,18 @@ public: const std::string &content_type); Result Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Put(const std::string &path, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Put(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); Result Put(const std::string &path, @@ -1574,6 +1749,8 @@ public: Result Put(const std::string &path, const Params ¶ms); Result Put(const std::string &path, const Headers &headers, const Params ¶ms); + Result Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress); Result Put(const std::string &path, const MultipartFormDataItems &items); Result Put(const std::string &path, const Headers &headers, const MultipartFormDataItems &items); @@ -1586,13 +1763,23 @@ public: Result Patch(const std::string &path); Result Patch(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Patch(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type); @@ -1610,13 +1797,24 @@ public: Result Delete(const std::string &path, const Headers &headers); Result Delete(const std::string &path, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const char *body, + size_t content_length, const std::string &content_type, + Progress progress); Result Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress); Result Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type); + Result Delete(const std::string &path, const Headers &headers, + const std::string &body, const std::string &content_type, + Progress progress); Result Options(const std::string &path); Result Options(const std::string &path, const Headers &headers); @@ -1685,6 +1883,8 @@ public: #ifdef CPPHTTPLIB_OPENSSL_SUPPORT void enable_server_certificate_verification(bool enabled); + void enable_server_hostname_verification(bool enabled); + void set_server_certificate_verifier(std::function verifier); #endif void set_logger(Logger logger); @@ -1730,6 +1930,9 @@ public: SSL_CTX *ssl_context() const; + void update_certs(X509 *cert, EVP_PKEY *private_key, + X509_STORE *client_ca_cert_store = nullptr); + private: bool process_and_close_socket(socket_t sock) override; @@ -1810,68 +2013,58 @@ inline void duration_to_sec_and_usec(const T &duration, U callback) { callback(static_cast(sec), static_cast(usec)); } +inline bool is_numeric(const std::string &str) { + return !str.empty() && std::all_of(str.begin(), str.end(), ::isdigit); +} + inline uint64_t get_header_value_u64(const Headers &headers, - const std::string &key, size_t id, - uint64_t def) { + const std::string &key, uint64_t def, + size_t id, bool &is_invalid_value) { + is_invalid_value = false; auto rng = headers.equal_range(key); auto it = rng.first; std::advance(it, static_cast(id)); if (it != rng.second) { - return std::strtoull(it->second.data(), nullptr, 10); + if (is_numeric(it->second)) { + return std::strtoull(it->second.data(), nullptr, 10); + } else { + is_invalid_value = true; + } } return def; } +inline uint64_t get_header_value_u64(const Headers &headers, + const std::string &key, uint64_t def, + size_t id) { + bool dummy = false; + return get_header_value_u64(headers, key, def, id, dummy); +} + } // namespace detail inline uint64_t Request::get_header_value_u64(const std::string &key, - size_t id) const { - return detail::get_header_value_u64(headers, key, id, 0); + uint64_t def, size_t id) const { + return detail::get_header_value_u64(headers, key, def, id); } inline uint64_t Response::get_header_value_u64(const std::string &key, - size_t id) const { - return detail::get_header_value_u64(headers, key, id, 0); -} - -template -inline ssize_t Stream::write_format(const char *fmt, const Args &...args) { - const auto bufsiz = 2048; - std::array buf{}; - - auto sn = snprintf(buf.data(), buf.size() - 1, fmt, args...); - if (sn <= 0) { return sn; } - - auto n = static_cast(sn); - - if (n >= buf.size() - 1) { - std::vector glowable_buf(buf.size()); - - while (n >= glowable_buf.size() - 1) { - glowable_buf.resize(glowable_buf.size() * 2); - n = static_cast( - snprintf(&glowable_buf[0], glowable_buf.size() - 1, fmt, args...)); - } - return write(&glowable_buf[0], n); - } else { - return write(buf.data(), n); - } + uint64_t def, size_t id) const { + return detail::get_header_value_u64(headers, key, def, id); } inline void default_socket_options(socket_t sock) { - int yes = 1; + int opt = 1; #ifdef _WIN32 setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, - reinterpret_cast(&yes), sizeof(yes)); - setsockopt(sock, SOL_SOCKET, SO_EXCLUSIVEADDRUSE, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else #ifdef SO_REUSEPORT setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #endif #endif } @@ -1997,6 +2190,8 @@ inline std::string to_string(const Error error) { case Error::SSLConnection: return "SSL connection failed"; case Error::SSLLoadingCerts: return "SSL certificate loading failed"; case Error::SSLServerVerification: return "SSL server verification failed"; + case Error::SSLServerHostnameVerification: + return "SSL server hostname verification failed"; case Error::UnsupportedMultipartBoundaryChars: return "Unsupported HTTP multipart boundary characters"; case Error::Compression: return "Compression failed"; @@ -2016,8 +2211,9 @@ inline std::ostream &operator<<(std::ostream &os, const Error &obj) { } inline uint64_t Result::get_request_header_value_u64(const std::string &key, + uint64_t def, size_t id) const { - return detail::get_header_value_u64(request_headers_, key, id, 0); + return detail::get_header_value_u64(request_headers_, key, def, id); } template @@ -2080,6 +2276,36 @@ make_basic_authentication_header(const std::string &username, namespace detail { +#if defined(_WIN32) +inline std::wstring u8string_to_wstring(const char *s) { + std::wstring ws; + auto len = static_cast(strlen(s)); + auto wlen = ::MultiByteToWideChar(CP_UTF8, 0, s, len, nullptr, 0); + if (wlen > 0) { + ws.resize(wlen); + wlen = ::MultiByteToWideChar( + CP_UTF8, 0, s, len, + const_cast(reinterpret_cast(ws.data())), wlen); + if (wlen != static_cast(ws.size())) { ws.clear(); } + } + return ws; +} +#endif + +struct FileStat { + FileStat(const std::string &path); + bool is_file() const; + bool is_dir() const; + +private: +#if defined(_WIN32) + struct _stat st_; +#else + struct stat st_; +#endif + int ret_ = -1; +}; + std::string encode_query_param(const std::string &value); std::string decode_url(const std::string &s, bool convert_plus_to_space); @@ -2088,6 +2314,16 @@ void read_file(const std::string &path, std::string &out); std::string trim_copy(const std::string &s); +void divide( + const char *data, std::size_t size, char d, + std::function + fn); + +void divide( + const std::string &str, char d, + std::function + fn); + void split(const char *b, const char *e, char d, std::function fn); @@ -2099,18 +2335,23 @@ bool process_client_socket(socket_t sock, time_t read_timeout_sec, time_t write_timeout_usec, std::function callback); -socket_t create_client_socket( - const std::string &host, const std::string &ip, int port, - int address_family, bool tcp_nodelay, SocketOptions socket_options, - time_t connection_timeout_sec, time_t connection_timeout_usec, - time_t read_timeout_sec, time_t read_timeout_usec, time_t write_timeout_sec, - time_t write_timeout_usec, const std::string &intf, Error &error); +socket_t create_client_socket(const std::string &host, const std::string &ip, + int port, int address_family, bool tcp_nodelay, + bool ipv6_v6only, SocketOptions socket_options, + time_t connection_timeout_sec, + time_t connection_timeout_usec, + time_t read_timeout_sec, time_t read_timeout_usec, + time_t write_timeout_sec, + time_t write_timeout_usec, + const std::string &intf, Error &error); const char *get_header_value(const Headers &headers, const std::string &key, - size_t id = 0, const char *def = nullptr); + const char *def, size_t id); std::string params_to_query_str(const Params ¶ms); +void parse_query_text(const char *data, std::size_t size, Params ¶ms); + void parse_query_text(const std::string &s, Params ¶ms); bool parse_multipart_boundary(const std::string &content_type, @@ -2270,15 +2511,70 @@ public: private: #if defined(_WIN32) - HANDLE hFile_; - HANDLE hMapping_; + HANDLE hFile_ = NULL; + HANDLE hMapping_ = NULL; #else - int fd_; + int fd_ = -1; #endif - size_t size_; - void *addr_; + size_t size_ = 0; + void *addr_ = nullptr; + bool is_open_empty_file = false; }; +// NOTE: https://www.rfc-editor.org/rfc/rfc9110#section-5 +namespace fields { + +inline bool is_token_char(char c) { + return std::isalnum(c) || c == '!' || c == '#' || c == '$' || c == '%' || + c == '&' || c == '\'' || c == '*' || c == '+' || c == '-' || + c == '.' || c == '^' || c == '_' || c == '`' || c == '|' || c == '~'; +} + +inline bool is_token(const std::string &s) { + if (s.empty()) { return false; } + for (auto c : s) { + if (!is_token_char(c)) { return false; } + } + return true; +} + +inline bool is_field_name(const std::string &s) { return is_token(s); } + +inline bool is_vchar(char c) { return c >= 33 && c <= 126; } + +inline bool is_obs_text(char c) { return 128 <= static_cast(c); } + +inline bool is_field_vchar(char c) { return is_vchar(c) || is_obs_text(c); } + +inline bool is_field_content(const std::string &s) { + if (s.empty()) { return false; } + + if (s.size() == 1) { + return is_field_vchar(s[0]); + } else if (s.size() == 2) { + return is_field_vchar(s[0]) && is_field_vchar(s[1]); + } else { + size_t i = 0; + + if (!is_field_vchar(s[i])) { return false; } + i++; + + while (i < s.size() - 1) { + auto c = s[i++]; + if (c == ' ' || c == '\t' || is_field_vchar(c)) { + } else { + return false; + } + } + + return is_field_vchar(s[i]); + } +} + +inline bool is_field_value(const std::string &s) { return is_field_content(s); } + +} // namespace fields + } // namespace detail // ---------------------------------------------------------------------------- @@ -2392,20 +2688,6 @@ inline std::string base64_encode(const std::string &in) { return out; } -inline bool is_file(const std::string &path) { -#ifdef _WIN32 - return _access_s(path.c_str(), 0) == 0; -#else - struct stat st; - return stat(path.c_str(), &st) >= 0 && S_ISREG(st.st_mode); -#endif -} - -inline bool is_dir(const std::string &path) { - struct stat st; - return stat(path.c_str(), &st) >= 0 && S_ISDIR(st.st_mode); -} - inline bool is_valid_path(const std::string &path) { size_t level = 0; size_t i = 0; @@ -2448,6 +2730,21 @@ inline bool is_valid_path(const std::string &path) { return true; } +inline FileStat::FileStat(const std::string &path) { +#if defined(_WIN32) + auto wpath = u8string_to_wstring(path.c_str()); + ret_ = _wstat(wpath.c_str(), &st_); +#else + ret_ = stat(path.c_str(), &st_); +#endif +} +inline bool FileStat::is_file() const { + return ret_ >= 0 && S_ISREG(st_.st_mode); +} +inline bool FileStat::is_dir() const { + return ret_ >= 0 && S_ISDIR(st_.st_mode); +} + inline std::string encode_query_param(const std::string &value) { std::ostringstream escaped; escaped.fill('0'); @@ -2579,6 +2876,27 @@ inline std::string trim_double_quotes_copy(const std::string &s) { return s; } +inline void +divide(const char *data, std::size_t size, char d, + std::function + fn) { + const auto it = std::find(data, data + size, d); + const auto found = static_cast(it != data + size); + const auto lhs_data = data; + const auto lhs_size = static_cast(it - data); + const auto rhs_data = it + found; + const auto rhs_size = size - lhs_size - found; + + fn(lhs_data, lhs_size, rhs_data, rhs_size); +} + +inline void +divide(const std::string &str, char d, + std::function + fn) { + divide(str.data(), str.size(), d, std::move(fn)); +} + inline void split(const char *b, const char *e, char d, std::function fn) { return split(b, e, d, (std::numeric_limits::max)(), std::move(fn)); @@ -2636,6 +2954,10 @@ inline bool stream_line_reader::getline() { fixed_buffer_used_size_ = 0; glowable_buffer_.clear(); +#ifndef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR + char prev_byte = 0; +#endif + for (size_t i = 0;; i++) { char byte; auto n = strm_.read(&byte, 1); @@ -2652,7 +2974,12 @@ inline bool stream_line_reader::getline() { append(byte); +#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR if (byte == '\n') { break; } +#else + if (prev_byte == '\r' && byte == '\n') { break; } + prev_byte = byte; +#endif } return true; @@ -2671,16 +2998,7 @@ inline void stream_line_reader::append(char c) { } } -inline mmap::mmap(const char *path) -#if defined(_WIN32) - : hFile_(NULL), hMapping_(NULL) -#else - : fd_(-1) -#endif - , - size_(0), addr_(nullptr) { - open(path); -} +inline mmap::mmap(const char *path) { open(path); } inline mmap::~mmap() { close(); } @@ -2688,29 +3006,60 @@ inline bool mmap::open(const char *path) { close(); #if defined(_WIN32) - std::wstring wpath; - for (size_t i = 0; i < strlen(path); i++) { - wpath += path[i]; - } + auto wpath = u8string_to_wstring(path); + if (wpath.empty()) { return false; } +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 hFile_ = ::CreateFile2(wpath.c_str(), GENERIC_READ, FILE_SHARE_READ, OPEN_EXISTING, NULL); +#else + hFile_ = ::CreateFileW(wpath.c_str(), GENERIC_READ, FILE_SHARE_READ, NULL, + OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL); +#endif if (hFile_ == INVALID_HANDLE_VALUE) { return false; } LARGE_INTEGER size{}; if (!::GetFileSizeEx(hFile_, &size)) { return false; } + // If the following line doesn't compile due to QuadPart, update Windows SDK. + // See: + // https://github.com/yhirose/cpp-httplib/issues/1903#issuecomment-2316520721 + if (static_cast(size.QuadPart) > + (std::numeric_limits::max)()) { + // `size_t` might be 32-bits, on 32-bits Windows. + return false; + } size_ = static_cast(size.QuadPart); +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 hMapping_ = ::CreateFileMappingFromApp(hFile_, NULL, PAGE_READONLY, size_, NULL); +#else + hMapping_ = ::CreateFileMappingW(hFile_, NULL, PAGE_READONLY, 0, 0, NULL); +#endif + + // Special treatment for an empty file... + if (hMapping_ == NULL && size_ == 0) { + close(); + is_open_empty_file = true; + return true; + } if (hMapping_ == NULL) { close(); return false; } +#if _WIN32_WINNT >= _WIN32_WINNT_WIN8 addr_ = ::MapViewOfFileFromApp(hMapping_, FILE_MAP_READ, 0, 0); +#else + addr_ = ::MapViewOfFile(hMapping_, FILE_MAP_READ, 0, 0, 0); +#endif + + if (addr_ == nullptr) { + close(); + return false; + } #else fd_ = ::open(path, O_RDONLY); if (fd_ == -1) { return false; } @@ -2723,22 +3072,26 @@ inline bool mmap::open(const char *path) { size_ = static_cast(sb.st_size); addr_ = ::mmap(NULL, size_, PROT_READ, MAP_PRIVATE, fd_, 0); -#endif - if (addr_ == nullptr) { + // Special treatment for an empty file... + if (addr_ == MAP_FAILED && size_ == 0) { close(); + is_open_empty_file = true; return false; } +#endif return true; } -inline bool mmap::is_open() const { return addr_ != nullptr; } +inline bool mmap::is_open() const { + return is_open_empty_file ? true : addr_ != nullptr; +} inline size_t mmap::size() const { return size_; } inline const char *mmap::data() const { - return static_cast(addr_); + return is_open_empty_file ? "" : static_cast(addr_); } inline void mmap::close() { @@ -2757,6 +3110,8 @@ inline void mmap::close() { ::CloseHandle(hFile_); hFile_ = INVALID_HANDLE_VALUE; } + + is_open_empty_file = false; #else if (addr_ != nullptr) { munmap(addr_, size_); @@ -2782,7 +3137,10 @@ template inline ssize_t handle_EINTR(T fn) { ssize_t res = 0; while (true) { res = fn(); - if (res < 0 && errno == EINTR) { continue; } + if (res < 0 && errno == EINTR) { + std::this_thread::sleep_for(std::chrono::microseconds{1}); + continue; + } break; } return res; @@ -2991,23 +3349,37 @@ private: }; #endif -inline bool keep_alive(socket_t sock, time_t keep_alive_timeout_sec) { +inline bool keep_alive(const std::atomic &svr_sock, socket_t sock, + time_t keep_alive_timeout_sec) { using namespace std::chrono; - auto start = steady_clock::now(); + + const auto interval_usec = + CPPHTTPLIB_KEEPALIVE_TIMEOUT_CHECK_INTERVAL_USECOND; + + // Avoid expensive `steady_clock::now()` call for the first time + if (select_read(sock, 0, interval_usec) > 0) { return true; } + + const auto start = steady_clock::now() - microseconds{interval_usec}; + const auto timeout = seconds{keep_alive_timeout_sec}; + while (true) { - auto val = select_read(sock, 0, 10000); + if (svr_sock == INVALID_SOCKET) { + break; // Server socket is closed + } + + auto val = select_read(sock, 0, interval_usec); if (val < 0) { - return false; + break; // Ssocket error } else if (val == 0) { - auto current = steady_clock::now(); - auto duration = duration_cast(current - start); - auto timeout = keep_alive_timeout_sec * 1000; - if (duration.count() > timeout) { return false; } - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + if (steady_clock::now() - start > timeout) { + break; // Timeout + } } else { - return true; + return true; // Ready for read } } + + return false; } template @@ -3018,8 +3390,7 @@ process_server_socket_core(const std::atomic &svr_sock, socket_t sock, assert(keep_alive_max_count > 0); auto ret = false; auto count = keep_alive_max_count; - while (svr_sock != INVALID_SOCKET && count > 0 && - keep_alive(sock, keep_alive_timeout_sec)) { + while (count > 0 && keep_alive(svr_sock, sock, keep_alive_timeout_sec)) { auto close_connection = count == 1; auto connection_closed = false; ret = callback(close_connection, connection_closed); @@ -3063,10 +3434,29 @@ inline int shutdown_socket(socket_t sock) { #endif } +inline std::string escape_abstract_namespace_unix_domain(const std::string &s) { + if (s.size() > 1 && s[0] == '\0') { + auto ret = s; + ret[0] = '@'; + return ret; + } + return s; +} + +inline std::string +unescape_abstract_namespace_unix_domain(const std::string &s) { + if (s.size() > 1 && s[0] == '@') { + auto ret = s; + ret[0] = '\0'; + return ret; + } + return s; +} + template socket_t create_socket(const std::string &host, const std::string &ip, int port, int address_family, int socket_flags, bool tcp_nodelay, - SocketOptions socket_options, + bool ipv6_v6only, SocketOptions socket_options, BindOrConnect bind_or_connect) { // Get address info const char *node = nullptr; @@ -3075,7 +3465,7 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, memset(&hints, 0, sizeof(struct addrinfo)); hints.ai_socktype = SOCK_STREAM; - hints.ai_protocol = 0; + hints.ai_protocol = IPPROTO_IP; if (!ip.empty()) { node = ip.c_str(); @@ -3093,20 +3483,32 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, const auto addrlen = host.length(); if (addrlen > sizeof(sockaddr_un::sun_path)) { return INVALID_SOCKET; } +#ifdef SOCK_CLOEXEC + auto sock = socket(hints.ai_family, hints.ai_socktype | SOCK_CLOEXEC, + hints.ai_protocol); +#else auto sock = socket(hints.ai_family, hints.ai_socktype, hints.ai_protocol); +#endif + if (sock != INVALID_SOCKET) { sockaddr_un addr{}; addr.sun_family = AF_UNIX; - std::copy(host.begin(), host.end(), addr.sun_path); + + auto unescaped_host = unescape_abstract_namespace_unix_domain(host); + std::copy(unescaped_host.begin(), unescaped_host.end(), addr.sun_path); hints.ai_addr = reinterpret_cast(&addr); hints.ai_addrlen = static_cast( sizeof(addr) - sizeof(addr.sun_path) + addrlen); +#ifndef SOCK_CLOEXEC fcntl(sock, F_SETFD, FD_CLOEXEC); +#endif + if (socket_options) { socket_options(sock); } - if (!bind_or_connect(sock, hints)) { + bool dummy; + if (!bind_or_connect(sock, hints, dummy)) { close_socket(sock); sock = INVALID_SOCKET; } @@ -3123,6 +3525,7 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, #endif return INVALID_SOCKET; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); for (auto rp = result; rp; rp = rp->ai_next) { // Create a socket @@ -3148,11 +3551,18 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, sock = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol); } #else + +#ifdef SOCK_CLOEXEC + auto sock = + socket(rp->ai_family, rp->ai_socktype | SOCK_CLOEXEC, rp->ai_protocol); +#else auto sock = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol); +#endif + #endif if (sock == INVALID_SOCKET) { continue; } -#ifndef _WIN32 +#if !defined _WIN32 && !defined SOCK_CLOEXEC if (fcntl(sock, F_SETFD, FD_CLOEXEC) == -1) { close_socket(sock); continue; @@ -3160,39 +3570,38 @@ socket_t create_socket(const std::string &host, const std::string &ip, int port, #endif if (tcp_nodelay) { - auto yes = 1; + auto opt = 1; #ifdef _WIN32 setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); #else setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, - reinterpret_cast(&yes), sizeof(yes)); + reinterpret_cast(&opt), sizeof(opt)); +#endif + } + + if (rp->ai_family == AF_INET6) { + auto opt = ipv6_v6only ? 1 : 0; +#ifdef _WIN32 + setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, + reinterpret_cast(&opt), sizeof(opt)); +#else + setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, + reinterpret_cast(&opt), sizeof(opt)); #endif } if (socket_options) { socket_options(sock); } - if (rp->ai_family == AF_INET6) { - auto no = 0; -#ifdef _WIN32 - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, - reinterpret_cast(&no), sizeof(no)); -#else - setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, - reinterpret_cast(&no), sizeof(no)); -#endif - } - // bind or connect - if (bind_or_connect(sock, *rp)) { - freeaddrinfo(result); - return sock; - } + auto quit = false; + if (bind_or_connect(sock, *rp, quit)) { return sock; } close_socket(sock); + + if (quit) { break; } } - freeaddrinfo(result); return INVALID_SOCKET; } @@ -3225,6 +3634,7 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { hints.ai_protocol = 0; if (getaddrinfo(host.c_str(), "0", &hints, &result)) { return false; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); auto ret = false; for (auto rp = result; rp; rp = rp->ai_next) { @@ -3235,7 +3645,6 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { } } - freeaddrinfo(result); return ret; } @@ -3247,6 +3656,8 @@ inline bool bind_ip_address(socket_t sock, const std::string &host) { inline std::string if2ip(int address_family, const std::string &ifn) { struct ifaddrs *ifap; getifaddrs(&ifap); + auto se = detail::scope_exit([&] { freeifaddrs(ifap); }); + std::string addr_candidate; for (auto ifa = ifap; ifa; ifa = ifa->ifa_next) { if (ifa->ifa_addr && ifn == ifa->ifa_name && @@ -3256,7 +3667,6 @@ inline std::string if2ip(int address_family, const std::string &ifn) { auto sa = reinterpret_cast(ifa->ifa_addr); char buf[INET_ADDRSTRLEN]; if (inet_ntop(AF_INET, &sa->sin_addr, buf, INET_ADDRSTRLEN)) { - freeifaddrs(ifap); return std::string(buf, INET_ADDRSTRLEN); } } else if (ifa->ifa_addr->sa_family == AF_INET6) { @@ -3269,7 +3679,6 @@ inline std::string if2ip(int address_family, const std::string &ifn) { if (s6_addr_head == 0xfc || s6_addr_head == 0xfd) { addr_candidate = std::string(buf, INET6_ADDRSTRLEN); } else { - freeifaddrs(ifap); return std::string(buf, INET6_ADDRSTRLEN); } } @@ -3277,20 +3686,21 @@ inline std::string if2ip(int address_family, const std::string &ifn) { } } } - freeifaddrs(ifap); return addr_candidate; } #endif inline socket_t create_client_socket( const std::string &host, const std::string &ip, int port, - int address_family, bool tcp_nodelay, SocketOptions socket_options, - time_t connection_timeout_sec, time_t connection_timeout_usec, - time_t read_timeout_sec, time_t read_timeout_usec, time_t write_timeout_sec, + int address_family, bool tcp_nodelay, bool ipv6_v6only, + SocketOptions socket_options, time_t connection_timeout_sec, + time_t connection_timeout_usec, time_t read_timeout_sec, + time_t read_timeout_usec, time_t write_timeout_sec, time_t write_timeout_usec, const std::string &intf, Error &error) { auto sock = create_socket( - host, ip, port, address_family, 0, tcp_nodelay, std::move(socket_options), - [&](socket_t sock2, struct addrinfo &ai) -> bool { + host, ip, port, address_family, 0, tcp_nodelay, ipv6_v6only, + std::move(socket_options), + [&](socket_t sock2, struct addrinfo &ai, bool &quit) -> bool { if (!intf.empty()) { #ifdef USE_IF2IP auto ip_from_if = if2ip(address_family, intf); @@ -3314,7 +3724,10 @@ inline socket_t create_client_socket( } error = wait_until_socket_is_ready(sock2, connection_timeout_sec, connection_timeout_usec); - if (error != Error::Success) { return false; } + if (error != Error::Success) { + if (error == Error::ConnectionTimeout) { quit = true; } + return false; + } } set_nonblocking(sock2, false); @@ -3439,7 +3852,7 @@ inline unsigned int str2tag(const std::string &s) { namespace udl { -inline constexpr unsigned int operator"" _t(const char *s, size_t l) { +inline constexpr unsigned int operator""_t(const char *s, size_t l) { return str2tag_core(s, l, 0); } @@ -3524,8 +3937,9 @@ inline bool can_compress_content_type(const std::string &content_type) { case "application/protobuf"_t: case "application/xhtml+xml"_t: return true; - default: - return !content_type.rfind("text/", 0) && tag != "text/event-stream"_t; + case "text/event-stream"_t: return false; + + default: return !content_type.rfind("text/", 0); } } @@ -3762,8 +4176,8 @@ inline bool has_header(const Headers &headers, const std::string &key) { } inline const char *get_header_value(const Headers &headers, - const std::string &key, size_t id, - const char *def) { + const std::string &key, const char *def, + size_t id) { auto rng = headers.equal_range(key); auto it = rng.first; std::advance(it, static_cast(id)); @@ -3771,14 +4185,6 @@ inline const char *get_header_value(const Headers &headers, return def; } -inline bool compare_case_ignore(const std::string &a, const std::string &b) { - if (a.size() != b.size()) { return false; } - for (size_t i = 0; i < b.size(); i++) { - if (::tolower(a[i]) != ::tolower(b[i])) { return false; } - } - return true; -} - template inline bool parse_header(const char *beg, const char *end, T fn) { // Skip trailing spaces and tabs. @@ -3801,15 +4207,27 @@ inline bool parse_header(const char *beg, const char *end, T fn) { p++; } - if (p < end) { + if (p <= end) { auto key_len = key_end - beg; if (!key_len) { return false; } auto key = std::string(beg, key_end); - auto val = compare_case_ignore(key, "Location") + auto val = case_ignore::equal(key, "Location") ? std::string(p, end) : decode_url(std::string(p, end), false); - fn(std::move(key), std::move(val)); + + // NOTE: From RFC 9110: + // Field values containing CR, LF, or NUL characters are + // invalid and dangerous, due to the varying ways that + // implementations might parse and interpret those + // characters; a recipient of CR, LF, or NUL within a field + // value MUST either reject the message or replace each of + // those characters with SP before further processing or + // forwarding of that message. + static const std::string CR_LF_NUL("\r\n\0", 3); + if (val.find_first_of(CR_LF_NUL) != std::string::npos) { return false; } + + fn(key, val); return true; } @@ -3829,27 +4247,27 @@ inline bool read_headers(Stream &strm, Headers &headers) { if (line_reader.end_with_crlf()) { // Blank line indicates end of headers. if (line_reader.size() == 2) { break; } -#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR } else { +#ifdef CPPHTTPLIB_ALLOW_LF_AS_LINE_TERMINATOR // Blank line indicates end of headers. if (line_reader.size() == 1) { break; } line_terminator_len = 1; - } #else - } else { continue; // Skip invalid line. - } #endif + } if (line_reader.size() > CPPHTTPLIB_HEADER_MAX_LENGTH) { return false; } // Exclude line terminator auto end = line_reader.ptr() + line_reader.size() - line_terminator_len; - parse_header(line_reader.ptr(), end, - [&](std::string &&key, std::string &&val) { - headers.emplace(std::move(key), std::move(val)); - }); + if (!parse_header(line_reader.ptr(), end, + [&](const std::string &key, std::string &val) { + headers.emplace(key, val); + })) { + return false; + } } return true; @@ -3937,8 +4355,19 @@ inline bool read_content_chunked(Stream &strm, T &x, assert(chunk_len == 0); - // Trailer - if (!line_reader.getline()) { return false; } + // NOTE: In RFC 9112, '7.1 Chunked Transfer Coding' mentiones "The chunked + // transfer coding is complete when a chunk with a chunk-size of zero is + // received, possibly followed by a trailer section, and finally terminated by + // an empty line". https://www.rfc-editor.org/rfc/rfc9112.html#section-7.1 + // + // In '7.1.3. Decoding Chunked', however, the pseudo-code in the section + // does't care for the existence of the final CRLF. In other words, it seems + // to be ok whether the final CRLF exists or not in the chunked data. + // https://www.rfc-editor.org/rfc/rfc9112.html#section-7.1.3 + // + // According to the reference code in RFC 9112, cpp-htpplib now allows + // chuncked transfer coding data without the final CRLF. + if (!line_reader.getline()) { return true; } while (strcmp(line_reader.ptr(), "\r\n") != 0) { if (line_reader.size() > CPPHTTPLIB_HEADER_MAX_LENGTH) { return false; } @@ -3948,8 +4377,8 @@ inline bool read_content_chunked(Stream &strm, T &x, auto end = line_reader.ptr() + line_reader.size() - line_terminator_len; parse_header(line_reader.ptr(), end, - [&](std::string &&key, std::string &&val) { - x.headers.emplace(std::move(key), std::move(val)); + [&](const std::string &key, const std::string &val) { + x.headers.emplace(key, val); }); if (!line_reader.getline()) { return false; } @@ -3959,8 +4388,8 @@ inline bool read_content_chunked(Stream &strm, T &x, } inline bool is_chunked_transfer_encoding(const Headers &headers) { - return compare_case_ignore( - get_header_value(headers, "Transfer-Encoding", 0, ""), "chunked"); + return case_ignore::equal( + get_header_value(headers, "Transfer-Encoding", "", 0), "chunked"); } template @@ -4026,8 +4455,14 @@ bool read_content(Stream &strm, T &x, size_t payload_max_length, int &status, } else if (!has_header(x.headers, "Content-Length")) { ret = read_content_without_length(strm, out); } else { - auto len = get_header_value_u64(x.headers, "Content-Length", 0, 0); - if (len > payload_max_length) { + auto is_invalid_value = false; + auto len = get_header_value_u64(x.headers, "Content-Length", + std::numeric_limits::max(), + 0, is_invalid_value); + + if (is_invalid_value) { + ret = false; + } else if (len > payload_max_length) { exceed_payload_max_length = true; skip_content_with_length(strm, len); ret = false; @@ -4042,13 +4477,36 @@ bool read_content(Stream &strm, T &x, size_t payload_max_length, int &status, } return ret; }); -} // namespace detail +} + +inline ssize_t write_request_line(Stream &strm, const std::string &method, + const std::string &path) { + std::string s = method; + s += " "; + s += path; + s += " HTTP/1.1\r\n"; + return strm.write(s.data(), s.size()); +} + +inline ssize_t write_response_line(Stream &strm, int status) { + std::string s = "HTTP/1.1 "; + s += std::to_string(status); + s += " "; + s += httplib::status_message(status); + s += "\r\n"; + return strm.write(s.data(), s.size()); +} inline ssize_t write_headers(Stream &strm, const Headers &headers) { ssize_t write_len = 0; for (const auto &x : headers) { - auto len = - strm.write_format("%s: %s\r\n", x.first.c_str(), x.second.c_str()); + std::string s; + s = x.first; + s += ": "; + s += x.second; + s += "\r\n"; + + auto len = strm.write(s.data(), s.size()); if (len < 0) { return len; } write_len += len; } @@ -4302,22 +4760,22 @@ inline std::string params_to_query_str(const Params ¶ms) { return query; } -inline void parse_query_text(const std::string &s, Params ¶ms) { +inline void parse_query_text(const char *data, std::size_t size, + Params ¶ms) { std::set cache; - split(s.data(), s.data() + s.size(), '&', [&](const char *b, const char *e) { + split(data, data + size, '&', [&](const char *b, const char *e) { std::string kv(b, e); if (cache.find(kv) != cache.end()) { return; } - cache.insert(kv); + cache.insert(std::move(kv)); std::string key; std::string val; - split(b, e, '=', [&](const char *b2, const char *e2) { - if (key.empty()) { - key.assign(b2, e2); - } else { - val.assign(b2, e2); - } - }); + divide(b, static_cast(e - b), '=', + [&](const char *lhs_data, std::size_t lhs_size, const char *rhs_data, + std::size_t rhs_size) { + key.assign(lhs_data, lhs_size); + val.assign(rhs_data, rhs_size); + }); if (!key.empty()) { params.emplace(decode_url(key, true), decode_url(val, true)); @@ -4325,6 +4783,10 @@ inline void parse_query_text(const std::string &s, Params ¶ms) { }); } +inline void parse_query_text(const std::string &s, Params ¶ms) { + parse_query_text(s.data(), s.size(), params); +} + inline bool parse_multipart_boundary(const std::string &content_type, std::string &boundary) { auto boundary_keyword = "boundary="; @@ -4365,35 +4827,44 @@ inline bool parse_range_header(const std::string &s, Ranges &ranges) { #else inline bool parse_range_header(const std::string &s, Ranges &ranges) try { #endif - static auto re_first_range = std::regex(R"(bytes=(\d*-\d*(?:,\s*\d*-\d*)*))"); - std::smatch m; - if (std::regex_match(s, m, re_first_range)) { - auto pos = static_cast(m.position(1)); - auto len = static_cast(m.length(1)); + auto is_valid = [](const std::string &str) { + return std::all_of(str.cbegin(), str.cend(), + [](unsigned char c) { return std::isdigit(c); }); + }; + + if (s.size() > 7 && s.compare(0, 6, "bytes=") == 0) { + const auto pos = static_cast(6); + const auto len = static_cast(s.size() - 6); auto all_valid_ranges = true; split(&s[pos], &s[pos + len], ',', [&](const char *b, const char *e) { if (!all_valid_ranges) { return; } - static auto re_another_range = std::regex(R"(\s*(\d*)-(\d*))"); - std::cmatch cm; - if (std::regex_match(b, e, cm, re_another_range)) { - ssize_t first = -1; - if (!cm.str(1).empty()) { - first = static_cast(std::stoll(cm.str(1))); - } - ssize_t last = -1; - if (!cm.str(2).empty()) { - last = static_cast(std::stoll(cm.str(2))); - } - - if (first != -1 && last != -1 && first > last) { - all_valid_ranges = false; - return; - } - ranges.emplace_back(std::make_pair(first, last)); + const auto it = std::find(b, e, '-'); + if (it == e) { + all_valid_ranges = false; + return; } + + const auto lhs = std::string(b, it); + const auto rhs = std::string(it + 1, e); + if (!is_valid(lhs) || !is_valid(rhs)) { + all_valid_ranges = false; + return; + } + + const auto first = + static_cast(lhs.empty() ? -1 : std::stoll(lhs)); + const auto last = + static_cast(rhs.empty() ? -1 : std::stoll(rhs)); + if ((first == -1 && last == -1) || + (first != -1 && last != -1 && first > last)) { + all_valid_ranges = false; + return; + } + + ranges.emplace_back(first, last); }); - return all_valid_ranges; + return all_valid_ranges && !ranges.empty(); } return false; #ifdef CPPHTTPLIB_NO_EXCEPTIONS @@ -4452,7 +4923,7 @@ public: const auto header = buf_head(pos); if (!parse_header(header.data(), header.data() + header.size(), - [&](std::string &&, std::string &&) {})) { + [&](const std::string &, const std::string &) {})) { is_valid_ = false; return false; } @@ -4562,7 +5033,9 @@ private: const std::string &b) const { if (a.size() < b.size()) { return false; } for (size_t i = 0; i < b.size(); i++) { - if (::tolower(a[i]) != ::tolower(b[i])) { return false; } + if (case_ignore::to_lower(a[i]) != case_ignore::to_lower(b[i])) { + return false; + } } return true; } @@ -4645,16 +5118,6 @@ private: size_t buf_epos_ = 0; }; -inline std::string to_lower(const char *beg, const char *end) { - std::string out; - auto it = beg; - while (it != end) { - out += static_cast(::tolower(*it)); - it++; - } - return out; -} - inline std::string random_string(size_t length) { static const char data[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; @@ -4768,7 +5231,18 @@ inline bool range_error(Request &req, Response &res) { last_pos = contant_len - 1; } - if (last_pos == -1) { last_pos = contant_len - 1; } + // NOTE: RFC-9110 '14.1.2. Byte Ranges': + // A client can limit the number of bytes requested without knowing the + // size of the selected representation. If the last-pos value is absent, + // or if the value is greater than or equal to the current length of the + // representation data, the byte range is interpreted as the remainder of + // the representation (i.e., the server replaces the value of last-pos + // with a value that is one less than the current length of the selected + // representation). + // https://www.rfc-editor.org/rfc/rfc9110.html#section-14.1.2-6 + if (last_pos == -1 || last_pos >= contant_len) { + last_pos = contant_len - 1; + } // Range must be within content length if (!(0 <= first_pos && first_pos <= last_pos && @@ -4795,12 +5269,11 @@ inline bool range_error(Request &req, Response &res) { inline std::pair get_range_offset_and_length(Range r, size_t content_length) { - (void)(content_length); // patch to get rid of "unused parameter" on release build assert(r.first != -1 && r.second != -1); assert(0 <= r.first && r.first < static_cast(content_length)); assert(r.first <= r.second && r.second < static_cast(content_length)); - + (void)(content_length); return std::make_pair(r.first, static_cast(r.second - r.first) + 1); } @@ -5230,6 +5703,7 @@ inline void hosted_at(const std::string &hostname, #endif return; } + auto se = detail::scope_exit([&] { freeaddrinfo(result); }); for (auto rp = result; rp; rp = rp->ai_next) { const auto &addr = @@ -5241,8 +5715,6 @@ inline void hosted_at(const std::string &hostname, addrs.push_back(ip); } } - - freeaddrinfo(result); } inline std::string append_query_params(const std::string &path, @@ -5291,8 +5763,8 @@ inline bool Request::has_header(const std::string &key) const { } inline std::string Request::get_header_value(const std::string &key, - size_t id) const { - return detail::get_header_value(headers, key, id, ""); + const char *def, size_t id) const { + return detail::get_header_value(headers, key, def, id); } inline size_t Request::get_header_value_count(const std::string &key) const { @@ -5302,7 +5774,8 @@ inline size_t Request::get_header_value_count(const std::string &key) const { inline void Request::set_header(const std::string &key, const std::string &val) { - if (!detail::has_crlf(key) && !detail::has_crlf(val)) { + if (detail::fields::is_field_name(key) && + detail::fields::is_field_value(val)) { headers.emplace(key, val); } } @@ -5356,8 +5829,9 @@ inline bool Response::has_header(const std::string &key) const { } inline std::string Response::get_header_value(const std::string &key, + const char *def, size_t id) const { - return detail::get_header_value(headers, key, id, ""); + return detail::get_header_value(headers, key, def, id); } inline size_t Response::get_header_value_count(const std::string &key) const { @@ -5367,13 +5841,14 @@ inline size_t Response::get_header_value_count(const std::string &key) const { inline void Response::set_header(const std::string &key, const std::string &val) { - if (!detail::has_crlf(key) && !detail::has_crlf(val)) { + if (detail::fields::is_field_name(key) && + detail::fields::is_field_value(val)) { headers.emplace(key, val); } } inline void Response::set_redirect(const std::string &url, int stat) { - if (!detail::has_crlf(url)) { + if (detail::fields::is_field_value(url)) { set_header("Location", url); if (300 <= stat && stat < 400) { this->status = stat; @@ -5436,14 +5911,25 @@ inline void Response::set_chunked_content_provider( is_chunked_content_provider_ = true; } +inline void Response::set_file_content(const std::string &path, + const std::string &content_type) { + file_content_path_ = path; + file_content_content_type_ = content_type; +} + +inline void Response::set_file_content(const std::string &path) { + file_content_path_ = path; +} + // Result implementation inline bool Result::has_request_header(const std::string &key) const { return request_headers_.find(key) != request_headers_.end(); } inline std::string Result::get_request_header_value(const std::string &key, + const char *def, size_t id) const { - return detail::get_header_value(request_headers_, key, id, ""); + return detail::get_header_value(request_headers_, key, def, id); } inline size_t @@ -5584,6 +6070,8 @@ inline socket_t BufferStream::socket() const { return 0; } inline const std::string &BufferStream::get_buffer() const { return buffer; } inline PathParamsMatcher::PathParamsMatcher(const std::string &pattern) { + static constexpr char marker[] = "/:"; + // One past the last ending position of a path param substring std::size_t last_param_end = 0; @@ -5596,13 +6084,14 @@ inline PathParamsMatcher::PathParamsMatcher(const std::string &pattern) { #endif while (true) { - const auto marker_pos = pattern.find(marker, last_param_end); + const auto marker_pos = pattern.find( + marker, last_param_end == 0 ? last_param_end : last_param_end - 1); if (marker_pos == std::string::npos) { break; } static_fragments_.push_back( - pattern.substr(last_param_end, marker_pos - last_param_end)); + pattern.substr(last_param_end, marker_pos - last_param_end + 1)); - const auto param_name_start = marker_pos + 1; + const auto param_name_start = marker_pos + 2; auto sep_pos = pattern.find(separator, param_name_start); if (sep_pos == std::string::npos) { sep_pos = pattern.length(); } @@ -5664,7 +6153,7 @@ inline bool PathParamsMatcher::match(Request &request) const { request.path_params.emplace( param_name, request.path.substr(starting_pos, sep_pos - starting_pos)); - // Mark everythin up to '/' as matched + // Mark everything up to '/' as matched starting_pos = sep_pos + 1; } // Returns false if the path is longer than the pattern @@ -5763,7 +6252,8 @@ inline bool Server::set_base_dir(const std::string &dir, inline bool Server::set_mount_point(const std::string &mount_point, const std::string &dir, Headers headers) { - if (detail::is_dir(dir)) { + detail::FileStat stat(dir); + if (stat.is_dir()) { std::string mnt = !mount_point.empty() ? mount_point : "/"; if (!mnt.empty() && mnt[0] == '/') { base_dirs_.push_back({mnt, dir, std::move(headers)}); @@ -5800,12 +6290,14 @@ inline Server &Server::set_file_request_handler(Handler handler) { return *this; } -inline Server &Server::set_error_handler(HandlerWithResponse handler) { +inline Server &Server::set_error_handler_core(HandlerWithResponse handler, + std::true_type) { error_handler_ = std::move(handler); return *this; } -inline Server &Server::set_error_handler(Handler handler) { +inline Server &Server::set_error_handler_core(Handler handler, + std::false_type) { error_handler_ = [handler](const Request &req, Response &res) { handler(req, res); return HandlerResponse::Handled; @@ -5849,6 +6341,11 @@ inline Server &Server::set_tcp_nodelay(bool on) { return *this; } +inline Server &Server::set_ipv6_v6only(bool on) { + ipv6_v6only_ = on; + return *this; +} + inline Server &Server::set_socket_options(SocketOptions socket_options) { socket_options_ = std::move(socket_options); return *this; @@ -5900,27 +6397,27 @@ inline Server &Server::set_payload_max_length(size_t length) { inline bool Server::bind_to_port(const std::string &host, int port, int socket_flags) { - return bind_internal(host, port, socket_flags) >= 0; + auto ret = bind_internal(host, port, socket_flags); + if (ret == -1) { is_decommisioned = true; } + return ret >= 0; } inline int Server::bind_to_any_port(const std::string &host, int socket_flags) { - return bind_internal(host, 0, socket_flags); + auto ret = bind_internal(host, 0, socket_flags); + if (ret == -1) { is_decommisioned = true; } + return ret; } -inline bool Server::listen_after_bind() { - auto se = detail::scope_exit([&]() { done_ = true; }); - return listen_internal(); -} +inline bool Server::listen_after_bind() { return listen_internal(); } inline bool Server::listen(const std::string &host, int port, int socket_flags) { - auto se = detail::scope_exit([&]() { done_ = true; }); return bind_to_port(host, port, socket_flags) && listen_internal(); } inline bool Server::is_running() const { return is_running_; } inline void Server::wait_until_ready() const { - while (!is_running() && !done_) { + while (!is_running_ && !is_decommisioned) { std::this_thread::sleep_for(std::chrono::milliseconds{1}); } } @@ -5932,8 +6429,11 @@ inline void Server::stop() { detail::shutdown_socket(sock); detail::close_socket(sock); } + is_decommisioned = false; } +inline void Server::decommission() { is_decommisioned = true; } + inline bool Server::parse_request_line(const char *s, Request &req) const { auto len = strlen(s); if (len < 2 || s[len - 2] != '\r' || s[len - 1] != '\n') { return false; } @@ -5972,26 +6472,13 @@ inline bool Server::parse_request_line(const char *s, Request &req) const { } } - size_t count = 0; - - detail::split(req.target.data(), req.target.data() + req.target.size(), '?', - 2, [&](const char *b, const char *e) { - switch (count) { - case 0: - req.path = detail::decode_url(std::string(b, e), false); - break; - case 1: { - if (e - b > 0) { - detail::parse_query_text(std::string(b, e), req.params); - } - break; - } - default: break; - } - count++; - }); - - if (count > 2) { return false; } + detail::divide(req.target, '?', + [&](const char *lhs_data, std::size_t lhs_size, + const char *rhs_data, std::size_t rhs_size) { + req.path = detail::decode_url( + std::string(lhs_data, lhs_size), false); + detail::parse_query_text(rhs_data, rhs_size, req.params); + }); } return true; @@ -6030,23 +6517,24 @@ inline bool Server::write_response_core(Stream &strm, bool close_connection, if (close_connection || req.get_header_value("Connection") == "close") { res.set_header("Connection", "close"); } else { - std::stringstream ss; - ss << "timeout=" << keep_alive_timeout_sec_ - << ", max=" << keep_alive_max_count_; - res.set_header("Keep-Alive", ss.str()); + std::string s = "timeout="; + s += std::to_string(keep_alive_timeout_sec_); + s += ", max="; + s += std::to_string(keep_alive_max_count_); + res.set_header("Keep-Alive", s); } - if (!res.has_header("Content-Type") && - (!res.body.empty() || res.content_length_ > 0 || res.content_provider_)) { + if ((!res.body.empty() || res.content_length_ > 0 || res.content_provider_) && + !res.has_header("Content-Type")) { res.set_header("Content-Type", "text/plain"); } - if (!res.has_header("Content-Length") && res.body.empty() && - !res.content_length_ && !res.content_provider_) { + if (res.body.empty() && !res.content_length_ && !res.content_provider_ && + !res.has_header("Content-Length")) { res.set_header("Content-Length", "0"); } - if (!res.has_header("Accept-Ranges") && req.method == "HEAD") { + if (req.method == "HEAD" && !res.has_header("Accept-Ranges")) { res.set_header("Accept-Ranges", "bytes"); } @@ -6055,12 +6543,7 @@ inline bool Server::write_response_core(Stream &strm, bool close_connection, // Response line and headers { detail::BufferStream bstrm; - - if (!bstrm.write_format("HTTP/1.1 %d %s\r\n", res.status, - status_message(res.status))) { - return false; - } - + if (!detail::write_response_line(bstrm, res.status)) { return false; } if (!header_writer_(bstrm, res.headers)) { return false; } // Flush buffer @@ -6254,7 +6737,14 @@ inline bool Server::handle_file_request(const Request &req, Response &res, auto path = entry.base_dir + sub_path; if (path.back() == '/') { path += "index.html"; } - if (detail::is_file(path)) { + detail::FileStat stat(path); + + if (stat.is_dir()) { + res.set_redirect(sub_path + "/", StatusCode::MovedPermanently_301); + return true; + } + + if (stat.is_file()) { for (const auto &kv : entry.headers) { res.set_header(kv.first, kv.second); } @@ -6289,8 +6779,8 @@ Server::create_server_socket(const std::string &host, int port, SocketOptions socket_options) const { return detail::create_socket( host, std::string(), port, address_family_, socket_flags, tcp_nodelay_, - std::move(socket_options), - [](socket_t sock, struct addrinfo &ai) -> bool { + ipv6_v6only_, std::move(socket_options), + [](socket_t sock, struct addrinfo &ai, bool & /*quit*/) -> bool { if (::bind(sock, ai.ai_addr, static_cast(ai.ai_addrlen))) { return false; } @@ -6301,6 +6791,8 @@ Server::create_server_socket(const std::string &host, int port, inline int Server::bind_internal(const std::string &host, int port, int socket_flags) { + if (is_decommisioned) { return -1; } + if (!is_valid()) { return -1; } svr_sock_ = create_server_socket(host, port, socket_flags, socket_options_); @@ -6326,6 +6818,8 @@ inline int Server::bind_internal(const std::string &host, int port, } inline bool Server::listen_internal() { + if (is_decommisioned) { return false; } + auto ret = true; is_running_ = true; auto se = detail::scope_exit([&]() { is_running_ = false; }); @@ -6346,13 +6840,22 @@ inline bool Server::listen_internal() { #ifndef _WIN32 } #endif + +#if defined _WIN32 + // sockets conneced via WASAccept inherit flags NO_HANDLE_INHERIT, + // OVERLAPPED + socket_t sock = WSAAccept(svr_sock_, nullptr, nullptr, nullptr, 0); +#elif defined SOCK_CLOEXEC + socket_t sock = accept4(svr_sock_, nullptr, nullptr, SOCK_CLOEXEC); +#else socket_t sock = accept(svr_sock_, nullptr, nullptr); +#endif if (sock == INVALID_SOCKET) { if (errno == EMFILE) { // The per-process limit of open file descriptors has been reached. // Try to accept new connections after a short sleep. - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{1}); continue; } else if (errno == EINTR || errno == EAGAIN) { continue; @@ -6406,6 +6909,7 @@ inline bool Server::listen_internal() { task_queue->shutdown(); } + is_decommisioned = !ret; return ret; } @@ -6503,7 +7007,7 @@ inline bool Server::dispatch_request(Request &req, Response &res, inline void Server::apply_ranges(const Request &req, Response &res, std::string &content_type, std::string &boundary) const { - if (req.ranges.size() > 1) { + if (req.ranges.size() > 1 && res.status == StatusCode::PartialContent_206) { auto it = res.headers.find("Content-Type"); if (it != res.headers.end()) { content_type = it->second; @@ -6521,7 +7025,7 @@ inline void Server::apply_ranges(const Request &req, Response &res, if (res.body.empty()) { if (res.content_length_ > 0) { size_t length = 0; - if (req.ranges.empty()) { + if (req.ranges.empty() || res.status != StatusCode::PartialContent_206) { length = res.content_length_; } else if (req.ranges.size() == 1) { auto offset_and_length = detail::get_range_offset_and_length( @@ -6550,7 +7054,7 @@ inline void Server::apply_ranges(const Request &req, Response &res, } } } else { - if (req.ranges.empty()) { + if (req.ranges.empty() || res.status != StatusCode::PartialContent_206) { ; } else if (req.ranges.size() == 1) { auto offset_and_length = @@ -6621,7 +7125,9 @@ inline bool Server::dispatch_request_for_content_reader( } inline bool -Server::process_request(Stream &strm, bool close_connection, +Server::process_request(Stream &strm, const std::string &remote_addr, + int remote_port, const std::string &local_addr, + int local_port, bool close_connection, bool &connection_closed, const std::function &setup_request) { std::array buf{}; @@ -6675,11 +7181,13 @@ Server::process_request(Stream &strm, bool close_connection, connection_closed = true; } - strm.get_remote_ip_and_port(req.remote_addr, req.remote_port); + req.remote_addr = remote_addr; + req.remote_port = remote_port; req.set_header("REMOTE_ADDR", req.remote_addr); req.set_header("REMOTE_PORT", std::to_string(req.remote_port)); - strm.get_local_ip_and_port(req.local_addr, req.local_port); + req.local_addr = local_addr; + req.local_port = local_port; req.set_header("LOCAL_ADDR", req.local_addr); req.set_header("LOCAL_PORT", std::to_string(req.local_port)); @@ -6701,13 +7209,20 @@ Server::process_request(Stream &strm, bool close_connection, switch (status) { case StatusCode::Continue_100: case StatusCode::ExpectationFailed_417: - strm.write_format("HTTP/1.1 %d %s\r\n\r\n", status, - status_message(status)); + detail::write_response_line(strm, status); + strm.write("\r\n"); break; - default: return write_response(strm, close_connection, req, res); + default: + connection_closed = true; + return write_response(strm, true, req, res); } } + // Setup `is_connection_closed` method + req.is_connection_closed = [&]() { + return !detail::is_socket_alive(strm.socket()); + }; + // Routing auto routed = false; #ifdef CPPHTTPLIB_NO_EXCEPTIONS @@ -6750,6 +7265,32 @@ Server::process_request(Stream &strm, bool close_connection, : StatusCode::PartialContent_206; } + // Serve file content by using a content provider + if (!res.file_content_path_.empty()) { + const auto &path = res.file_content_path_; + auto mm = std::make_shared(path.c_str()); + if (!mm->is_open()) { + res.body.clear(); + res.content_length_ = 0; + res.content_provider_ = nullptr; + res.status = StatusCode::NotFound_404; + return write_response(strm, close_connection, req, res); + } + + auto content_type = res.file_content_content_type_; + if (content_type.empty()) { + content_type = detail::find_content_type( + path, file_extension_and_mimetype_map_, default_file_mimetype_); + } + + res.set_content_provider( + mm->size(), content_type, + [mm](size_t offset, size_t length, DataSink &sink) -> bool { + sink.write(mm->data() + offset, length); + return true; + }); + } + if (detail::range_error(req, res)) { res.body.clear(); res.content_length_ = 0; @@ -6769,12 +7310,21 @@ Server::process_request(Stream &strm, bool close_connection, inline bool Server::is_valid() const { return true; } inline bool Server::process_and_close_socket(socket_t sock) { + std::string remote_addr; + int remote_port = 0; + detail::get_remote_ip_and_port(sock, remote_addr, remote_port); + + std::string local_addr; + int local_port = 0; + detail::get_local_ip_and_port(sock, local_addr, local_port); + auto ret = detail::process_server_socket( svr_sock_, sock, keep_alive_max_count_, keep_alive_timeout_sec_, read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, - [this](Stream &strm, bool close_connection, bool &connection_closed) { - return process_request(strm, close_connection, connection_closed, + [&](Stream &strm, bool close_connection, bool &connection_closed) { + return process_request(strm, remote_addr, remote_port, local_addr, + local_port, close_connection, connection_closed, nullptr); }); @@ -6793,8 +7343,8 @@ inline ClientImpl::ClientImpl(const std::string &host, int port) inline ClientImpl::ClientImpl(const std::string &host, int port, const std::string &client_cert_path, const std::string &client_key_path) - : host_(host), port_(port), - host_and_port_(adjust_host_string(host) + ":" + std::to_string(port)), + : host_(detail::escape_abstract_namespace_unix_domain(host)), port_(port), + host_and_port_(adjust_host_string(host_) + ":" + std::to_string(port)), client_cert_path_(client_cert_path), client_key_path_(client_key_path) {} inline ClientImpl::~ClientImpl() { @@ -6825,6 +7375,7 @@ inline void ClientImpl::copy_settings(const ClientImpl &rhs) { url_encode_ = rhs.url_encode_; address_family_ = rhs.address_family_; tcp_nodelay_ = rhs.tcp_nodelay_; + ipv6_v6only_ = rhs.ipv6_v6only_; socket_options_ = rhs.socket_options_; compress_ = rhs.compress_; decompress_ = rhs.decompress_; @@ -6845,6 +7396,8 @@ inline void ClientImpl::copy_settings(const ClientImpl &rhs) { #endif #ifdef CPPHTTPLIB_OPENSSL_SUPPORT server_certificate_verification_ = rhs.server_certificate_verification_; + server_hostname_verification_ = rhs.server_hostname_verification_; + server_certificate_verifier_ = rhs.server_certificate_verifier_; #endif logger_ = rhs.logger_; } @@ -6853,9 +7406,9 @@ inline socket_t ClientImpl::create_client_socket(Error &error) const { if (!proxy_host_.empty() && proxy_port_ != -1) { return detail::create_client_socket( proxy_host_, std::string(), proxy_port_, address_family_, tcp_nodelay_, - socket_options_, connection_timeout_sec_, connection_timeout_usec_, - read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, - write_timeout_usec_, interface_, error); + ipv6_v6only_, socket_options_, connection_timeout_sec_, + connection_timeout_usec_, read_timeout_sec_, read_timeout_usec_, + write_timeout_sec_, write_timeout_usec_, interface_, error); } // Check is custom IP specified for host_ @@ -6864,10 +7417,10 @@ inline socket_t ClientImpl::create_client_socket(Error &error) const { if (it != addr_map_.end()) { ip = it->second; } return detail::create_client_socket( - host_, ip, port_, address_family_, tcp_nodelay_, socket_options_, - connection_timeout_sec_, connection_timeout_usec_, read_timeout_sec_, - read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, interface_, - error); + host_, ip, port_, address_family_, tcp_nodelay_, ipv6_v6only_, + socket_options_, connection_timeout_sec_, connection_timeout_usec_, + read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, + write_timeout_usec_, interface_, error); } inline bool ClientImpl::create_and_connect_socket(Socket &socket, @@ -6956,6 +7509,18 @@ inline bool ClientImpl::send(Request &req, Response &res, Error &error) { return ret; } +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT +inline bool ClientImpl::is_ssl_peer_could_be_closed(SSL *ssl) const { + detail::set_nonblocking(socket_.sock, true); + auto se = detail::scope_exit( + [&]() { detail::set_nonblocking(socket_.sock, false); }); + + char buf[1]; + return !SSL_peek(ssl, buf, 1) && + SSL_get_error(ssl, 0) == SSL_ERROR_ZERO_RETURN; +} +#endif + inline bool ClientImpl::send_(Request &req, Response &res, Error &error) { { std::lock_guard guard(socket_mutex_); @@ -6967,6 +7532,13 @@ inline bool ClientImpl::send_(Request &req, Response &res, Error &error) { auto is_alive = false; if (socket_.is_open()) { is_alive = detail::is_socket_alive(socket_.sock); + +#ifdef CPPHTTPLIB_OPENSSL_SUPPORT + if (is_alive && is_ssl()) { + if (is_ssl_peer_could_be_closed(socket_.ssl)) { is_alive = false; } + } +#endif + if (!is_alive) { // Attempt to avoid sigpipe by shutting down nongracefully if it seems // like the other side has already closed the connection Also, there @@ -7144,7 +7716,7 @@ inline bool ClientImpl::redirect(Request &req, Response &res, Error &error) { if (location.empty()) { return false; } const static std::regex re( - R"((?:(https?):)?(?://(?:\[([\d:]+)\]|([^:/?#]+))(?::(\d+))?)?([^?#]*)(\?[^#]*)?(?:#.*)?)"); + R"((?:(https?):)?(?://(?:\[([a-fA-F\d:]+)\]|([^:/?#]+))(?::(\d+))?)?([^?#]*)(\?[^#]*)?(?:#.*)?)"); std::smatch m; if (!std::regex_match(location, m, re)) { return false; } @@ -7243,12 +7815,26 @@ inline bool ClientImpl::write_request(Stream &strm, Request &req, if (!req.has_header("Accept")) { req.set_header("Accept", "*/*"); } -#ifndef CPPHTTPLIB_NO_DEFAULT_USER_AGENT - if (!req.has_header("User-Agent")) { - auto agent = std::string("cpp-httplib/") + CPPHTTPLIB_VERSION; - req.set_header("User-Agent", agent); - } + if (!req.content_receiver) { + if (!req.has_header("Accept-Encoding")) { + std::string accept_encoding; +#ifdef CPPHTTPLIB_BROTLI_SUPPORT + accept_encoding = "br"; #endif +#ifdef CPPHTTPLIB_ZLIB_SUPPORT + if (!accept_encoding.empty()) { accept_encoding += ", "; } + accept_encoding += "gzip, deflate"; +#endif + req.set_header("Accept-Encoding", accept_encoding); + } + +#ifndef CPPHTTPLIB_NO_DEFAULT_USER_AGENT + if (!req.has_header("User-Agent")) { + auto agent = std::string("cpp-httplib/") + CPPHTTPLIB_VERSION; + req.set_header("User-Agent", agent); + } +#endif + }; if (req.body.empty()) { if (req.content_provider_) { @@ -7308,8 +7894,14 @@ inline bool ClientImpl::write_request(Stream &strm, Request &req, { detail::BufferStream bstrm; - const auto &path = url_encode_ ? detail::encode_url(req.path) : req.path; - bstrm.write_format("%s %s HTTP/1.1\r\n", req.method.c_str(), path.c_str()); + const auto &path_with_query = + req.params.empty() ? req.path + : append_query_params(req.path, req.params); + + const auto &path = + url_encode_ ? detail::encode_url(path_with_query) : path_with_query; + + detail::write_request_line(bstrm, req.method, path); header_writer_(bstrm, req.headers); @@ -7417,11 +8009,12 @@ inline Result ClientImpl::send_with_content_provider( const std::string &method, const std::string &path, const Headers &headers, const char *body, size_t content_length, ContentProvider content_provider, ContentProviderWithoutLength content_provider_without_length, - const std::string &content_type) { + const std::string &content_type, Progress progress) { Request req; req.method = method; req.headers = headers; req.path = path; + req.progress = progress; auto error = Error::Success; @@ -7448,9 +8041,7 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, if (is_ssl()) { auto is_proxy_enabled = !proxy_host_.empty() && proxy_port_ != -1; if (!is_proxy_enabled) { - char buf[1]; - if (SSL_peek(socket_.ssl, buf, 1) == 0 && - SSL_get_error(socket_.ssl, 0) == SSL_ERROR_ZERO_RETURN) { + if (is_ssl_peer_could_be_closed(socket_.ssl)) { error = Error::SSLPeerCouldBeClosed_; return false; } @@ -7468,7 +8059,9 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, // Body if ((res.status != StatusCode::NoContent_204) && req.method != "HEAD" && req.method != "CONNECT") { - auto redirect = 300 < res.status && res.status < 400 && follow_location_; + auto redirect = 300 < res.status && res.status < 400 && + res.status != StatusCode::NotModified_304 && + follow_location_; if (req.response_handler && !redirect) { if (!req.response_handler(res)) { @@ -7489,9 +8082,7 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, : static_cast( [&](const char *buf, size_t n, uint64_t /*off*/, uint64_t /*len*/) { - if (res.body.size() + n > res.body.max_size()) { - return false; - } + assert(res.body.size() + n <= res.body.max_size()); res.body.append(buf, n); return true; }); @@ -7503,12 +8094,25 @@ inline bool ClientImpl::process_request(Stream &strm, Request &req, return ret; }; - int dummy_status; - if (!detail::read_content(strm, res, (std::numeric_limits::max)(), - dummy_status, std::move(progress), std::move(out), - decompress_)) { - if (error != Error::Canceled) { error = Error::Read; } - return false; + if (res.has_header("Content-Length")) { + if (!req.content_receiver) { + auto len = res.get_header_value_u64("Content-Length"); + if (len > res.body.max_size()) { + error = Error::Read; + return false; + } + res.body.reserve(static_cast(len)); + } + } + + if (res.status != StatusCode::NotModified_304) { + int dummy_status; + if (!detail::read_content(strm, res, (std::numeric_limits::max)(), + dummy_status, std::move(progress), + std::move(out), decompress_)) { + if (error != Error::Canceled) { error = Error::Read; } + return false; + } } } @@ -7717,14 +8321,22 @@ inline Result ClientImpl::Post(const std::string &path, inline Result ClientImpl::Post(const std::string &path, const char *body, size_t content_length, const std::string &content_type) { - return Post(path, Headers(), body, content_length, content_type); + return Post(path, Headers(), body, content_length, content_type, nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return send_with_content_provider("POST", path, headers, body, content_length, - nullptr, nullptr, content_type); + nullptr, nullptr, content_type, nullptr); +} + +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("POST", path, headers, body, content_length, + nullptr, nullptr, content_type, progress); } inline Result ClientImpl::Post(const std::string &path, const std::string &body, @@ -7732,12 +8344,27 @@ inline Result ClientImpl::Post(const std::string &path, const std::string &body, return Post(path, Headers(), body, content_type); } +inline Result ClientImpl::Post(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return Post(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return send_with_content_provider("POST", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + nullptr); +} + +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("POST", path, headers, body.data(), + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Post(const std::string &path, const Params ¶ms) { @@ -7763,14 +8390,15 @@ inline Result ClientImpl::Post(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("POST", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("POST", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Post(const std::string &path, const Headers &headers, @@ -7779,6 +8407,13 @@ inline Result ClientImpl::Post(const std::string &path, const Headers &headers, return Post(path, headers, query, "application/x-www-form-urlencoded"); } +inline Result ClientImpl::Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + auto query = detail::params_to_query_str(params); + return Post(path, headers, query, "application/x-www-form-urlencoded", + progress); +} + inline Result ClientImpl::Post(const std::string &path, const MultipartFormDataItems &items) { return Post(path, Headers(), items); @@ -7816,7 +8451,7 @@ ClientImpl::Post(const std::string &path, const Headers &headers, return send_with_content_provider( "POST", path, headers, nullptr, 0, nullptr, get_multipart_content_provider(boundary, items, provider_items), - content_type); + content_type, nullptr); } inline Result ClientImpl::Put(const std::string &path) { @@ -7833,7 +8468,15 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, body, content_length, - nullptr, nullptr, content_type); + nullptr, nullptr, content_type, nullptr); +} + +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("PUT", path, headers, body, content_length, + nullptr, nullptr, content_type, progress); } inline Result ClientImpl::Put(const std::string &path, const std::string &body, @@ -7841,12 +8484,27 @@ inline Result ClientImpl::Put(const std::string &path, const std::string &body, return Put(path, Headers(), body, content_type); } +inline Result ClientImpl::Put(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return Put(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + nullptr); +} + +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return send_with_content_provider("PUT", path, headers, body.data(), + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Put(const std::string &path, size_t content_length, @@ -7868,14 +8526,15 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Put(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("PUT", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Put(const std::string &path, const Params ¶ms) { @@ -7888,6 +8547,13 @@ inline Result ClientImpl::Put(const std::string &path, const Headers &headers, return Put(path, headers, query, "application/x-www-form-urlencoded"); } +inline Result ClientImpl::Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + auto query = detail::params_to_query_str(params); + return Put(path, headers, query, "application/x-www-form-urlencoded", + progress); +} + inline Result ClientImpl::Put(const std::string &path, const MultipartFormDataItems &items) { return Put(path, Headers(), items); @@ -7925,7 +8591,7 @@ ClientImpl::Put(const std::string &path, const Headers &headers, return send_with_content_provider( "PUT", path, headers, nullptr, 0, nullptr, get_multipart_content_provider(boundary, items, provider_items), - content_type); + content_type, nullptr); } inline Result ClientImpl::Patch(const std::string &path) { return Patch(path, std::string(), std::string()); @@ -7937,12 +8603,26 @@ inline Result ClientImpl::Patch(const std::string &path, const char *body, return Patch(path, Headers(), body, content_length, content_type); } +inline Result ClientImpl::Patch(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return Patch(path, Headers(), body, content_length, content_type, progress); +} + inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { + return Patch(path, headers, body, content_length, content_type, nullptr); +} + +inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { return send_with_content_provider("PATCH", path, headers, body, content_length, nullptr, nullptr, - content_type); + content_type, progress); } inline Result ClientImpl::Patch(const std::string &path, @@ -7951,12 +8631,26 @@ inline Result ClientImpl::Patch(const std::string &path, return Patch(path, Headers(), body, content_type); } +inline Result ClientImpl::Patch(const std::string &path, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Patch(path, Headers(), body, content_type, progress); +} + inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { + return Patch(path, headers, body, content_type, nullptr); +} + +inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { return send_with_content_provider("PATCH", path, headers, body.data(), - body.size(), nullptr, nullptr, - content_type); + body.size(), nullptr, nullptr, content_type, + progress); } inline Result ClientImpl::Patch(const std::string &path, size_t content_length, @@ -7978,14 +8672,15 @@ inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, const std::string &content_type) { return send_with_content_provider("PATCH", path, headers, nullptr, content_length, std::move(content_provider), - nullptr, content_type); + nullptr, content_type, nullptr); } inline Result ClientImpl::Patch(const std::string &path, const Headers &headers, ContentProviderWithoutLength content_provider, const std::string &content_type) { return send_with_content_provider("PATCH", path, headers, nullptr, 0, nullptr, - std::move(content_provider), content_type); + std::move(content_provider), content_type, + nullptr); } inline Result ClientImpl::Delete(const std::string &path) { @@ -8003,14 +8698,30 @@ inline Result ClientImpl::Delete(const std::string &path, const char *body, return Delete(path, Headers(), body, content_length, content_type); } +inline Result ClientImpl::Delete(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return Delete(path, Headers(), body, content_length, content_type, progress); +} + inline Result ClientImpl::Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { + return Delete(path, headers, body, content_length, content_type, nullptr); +} + +inline Result ClientImpl::Delete(const std::string &path, + const Headers &headers, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { Request req; req.method = "DELETE"; req.headers = headers; req.path = path; + req.progress = progress; if (!content_type.empty()) { req.set_header("Content-Type", content_type); } req.body.assign(body, content_length); @@ -8024,6 +8735,14 @@ inline Result ClientImpl::Delete(const std::string &path, return Delete(path, Headers(), body.data(), body.size(), content_type); } +inline Result ClientImpl::Delete(const std::string &path, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Delete(path, Headers(), body.data(), body.size(), content_type, + progress); +} + inline Result ClientImpl::Delete(const std::string &path, const Headers &headers, const std::string &body, @@ -8031,6 +8750,15 @@ inline Result ClientImpl::Delete(const std::string &path, return Delete(path, headers, body.data(), body.size(), content_type); } +inline Result ClientImpl::Delete(const std::string &path, + const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return Delete(path, headers, body.data(), body.size(), content_type, + progress); +} + inline Result ClientImpl::Options(const std::string &path) { return Options(path, Headers()); } @@ -8138,6 +8866,8 @@ inline void ClientImpl::set_address_family(int family) { inline void ClientImpl::set_tcp_nodelay(bool on) { tcp_nodelay_ = on; } +inline void ClientImpl::set_ipv6_v6only(bool on) { ipv6_v6only_ = on; } + inline void ClientImpl::set_socket_options(SocketOptions socket_options) { socket_options_ = std::move(socket_options); } @@ -8187,13 +8917,11 @@ inline void ClientImpl::set_ca_cert_store(X509_STORE *ca_cert_store) { inline X509_STORE *ClientImpl::create_ca_cert_store(const char *ca_cert, std::size_t size) const { auto mem = BIO_new_mem_buf(ca_cert, static_cast(size)); + auto se = detail::scope_exit([&] { BIO_free_all(mem); }); if (!mem) { return nullptr; } auto inf = PEM_X509_INFO_read_bio(mem, nullptr, nullptr, nullptr); - if (!inf) { - BIO_free_all(mem); - return nullptr; - } + if (!inf) { return nullptr; } auto cts = X509_STORE_new(); if (cts) { @@ -8207,13 +8935,21 @@ inline X509_STORE *ClientImpl::create_ca_cert_store(const char *ca_cert, } sk_X509_INFO_pop_free(inf, X509_INFO_free); - BIO_free_all(mem); return cts; } inline void ClientImpl::enable_server_certificate_verification(bool enabled) { server_certificate_verification_ = enabled; } + +inline void ClientImpl::enable_server_hostname_verification(bool enabled) { + server_hostname_verification_ = enabled; +} + +inline void ClientImpl::set_server_certificate_verifier( + std::function verifier) { + server_certificate_verifier_ = verifier; +} #endif inline void ClientImpl::set_logger(Logger logger) { @@ -8257,13 +8993,30 @@ inline SSL *ssl_new(socket_t sock, SSL_CTX *ctx, std::mutex &ctx_mutex, return ssl; } -inline void ssl_delete(std::mutex &ctx_mutex, SSL *ssl, +inline void ssl_delete(std::mutex &ctx_mutex, SSL *ssl, socket_t sock, bool shutdown_gracefully) { // sometimes we may want to skip this to try to avoid SIGPIPE if we know // the remote has closed the network connection // Note that it is not always possible to avoid SIGPIPE, this is merely a // best-efforts. - if (shutdown_gracefully) { SSL_shutdown(ssl); } + if (shutdown_gracefully) { +#ifdef _WIN32 + (void)(sock); + SSL_shutdown(ssl); +#else + timeval tv; + tv.tv_sec = 1; + tv.tv_usec = 0; + setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO, + reinterpret_cast(&tv), sizeof(tv)); + + auto ret = SSL_shutdown(ssl); + while (ret == 0) { + std::this_thread::sleep_for(std::chrono::milliseconds{100}); + ret = SSL_shutdown(ssl); + } +#endif + } std::lock_guard guard(ctx_mutex); SSL_free(ssl); @@ -8366,7 +9119,7 @@ inline ssize_t SSLSocketStream::read(char *ptr, size_t size) { if (SSL_pending(ssl_) > 0) { return SSL_read(ssl_, ptr, static_cast(size)); } else if (is_readable()) { - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{10}); ret = SSL_read(ssl_, ptr, static_cast(size)); if (ret >= 0) { return ret; } err = SSL_get_error(ssl_, ret); @@ -8397,7 +9150,7 @@ inline ssize_t SSLSocketStream::write(const char *ptr, size_t size) { while (--n >= 0 && err == SSL_ERROR_WANT_WRITE) { #endif if (is_writable()) { - std::this_thread::sleep_for(std::chrono::milliseconds(1)); + std::this_thread::sleep_for(std::chrono::microseconds{10}); ret = SSL_write(ssl_, ptr, static_cast(handle_size)); if (ret >= 0) { return ret; } err = SSL_get_error(ssl_, ret); @@ -8439,7 +9192,7 @@ inline SSLServer::SSLServer(const char *cert_path, const char *private_key_path, SSL_OP_NO_COMPRESSION | SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION); - SSL_CTX_set_min_proto_version(ctx_, TLS1_1_VERSION); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); if (private_key_password != nullptr && (private_key_password[0] != '\0')) { SSL_CTX_set_default_passwd_cb_userdata( @@ -8449,7 +9202,8 @@ inline SSLServer::SSLServer(const char *cert_path, const char *private_key_path, if (SSL_CTX_use_certificate_chain_file(ctx_, cert_path) != 1 || SSL_CTX_use_PrivateKey_file(ctx_, private_key_path, SSL_FILETYPE_PEM) != - 1) { + 1 || + SSL_CTX_check_private_key(ctx_) != 1) { SSL_CTX_free(ctx_); ctx_ = nullptr; } else if (client_ca_cert_file_path || client_ca_cert_dir_path) { @@ -8471,7 +9225,7 @@ inline SSLServer::SSLServer(X509 *cert, EVP_PKEY *private_key, SSL_OP_NO_COMPRESSION | SSL_OP_NO_SESSION_RESUMPTION_ON_RENEGOTIATION); - SSL_CTX_set_min_proto_version(ctx_, TLS1_1_VERSION); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); if (SSL_CTX_use_certificate(ctx_, cert) != 1 || SSL_CTX_use_PrivateKey(ctx_, private_key) != 1) { @@ -8505,6 +9259,19 @@ inline bool SSLServer::is_valid() const { return ctx_; } inline SSL_CTX *SSLServer::ssl_context() const { return ctx_; } +inline void SSLServer::update_certs(X509 *cert, EVP_PKEY *private_key, + X509_STORE *client_ca_cert_store) { + + std::lock_guard guard(ctx_mutex_); + + SSL_CTX_use_certificate(ctx_, cert); + SSL_CTX_use_PrivateKey(ctx_, private_key); + + if (client_ca_cert_store != nullptr) { + SSL_CTX_set_cert_store(ctx_, client_ca_cert_store); + } +} + inline bool SSLServer::process_and_close_socket(socket_t sock) { auto ssl = detail::ssl_new( sock, ctx_, ctx_mutex_, @@ -8516,20 +9283,29 @@ inline bool SSLServer::process_and_close_socket(socket_t sock) { auto ret = false; if (ssl) { + std::string remote_addr; + int remote_port = 0; + detail::get_remote_ip_and_port(sock, remote_addr, remote_port); + + std::string local_addr; + int local_port = 0; + detail::get_local_ip_and_port(sock, local_addr, local_port); + ret = detail::process_server_socket_ssl( svr_sock_, ssl, sock, keep_alive_max_count_, keep_alive_timeout_sec_, read_timeout_sec_, read_timeout_usec_, write_timeout_sec_, write_timeout_usec_, - [this, ssl](Stream &strm, bool close_connection, - bool &connection_closed) { - return process_request(strm, close_connection, connection_closed, + [&](Stream &strm, bool close_connection, bool &connection_closed) { + return process_request(strm, remote_addr, remote_port, local_addr, + local_port, close_connection, + connection_closed, [&](Request &req) { req.ssl = ssl; }); }); // Shutdown gracefully if the result seemed successful, non-gracefully if // the connection appeared to be closed. const bool shutdown_gracefully = ret; - detail::ssl_delete(ctx_mutex_, ssl, shutdown_gracefully); + detail::ssl_delete(ctx_mutex_, ssl, sock, shutdown_gracefully); } detail::shutdown_socket(sock); @@ -8551,6 +9327,8 @@ inline SSLClient::SSLClient(const std::string &host, int port, : ClientImpl(host, port, client_cert_path, client_key_path) { ctx_ = SSL_CTX_new(TLS_client_method()); + SSL_CTX_set_min_proto_version(ctx_, TLS1_2_VERSION); + detail::split(&host_[0], &host_[host_.size()], '.', [&](const char *b, const char *e) { host_components_.emplace_back(b, e); @@ -8758,36 +9536,47 @@ inline bool SSLClient::initialize_ssl(Socket &socket, Error &error) { } if (server_certificate_verification_) { - verify_result_ = SSL_get_verify_result(ssl2); + if (server_certificate_verifier_) { + if (!server_certificate_verifier_(ssl2)) { + error = Error::SSLServerVerification; + return false; + } + } else { + verify_result_ = SSL_get_verify_result(ssl2); - if (verify_result_ != X509_V_OK) { - error = Error::SSLServerVerification; - return false; + if (verify_result_ != X509_V_OK) { + error = Error::SSLServerVerification; + return false; + } + + auto server_cert = SSL_get1_peer_certificate(ssl2); + auto se = detail::scope_exit([&] { X509_free(server_cert); }); + + if (server_cert == nullptr) { + error = Error::SSLServerVerification; + return false; + } + + if (server_hostname_verification_) { + if (!verify_host(server_cert)) { + error = Error::SSLServerHostnameVerification; + return false; + } + } } - - auto server_cert = SSL_get1_peer_certificate(ssl2); - - if (server_cert == nullptr) { - error = Error::SSLServerVerification; - return false; - } - - if (!verify_host(server_cert)) { - X509_free(server_cert); - error = Error::SSLServerVerification; - return false; - } - X509_free(server_cert); } return true; }, [&](SSL *ssl2) { +#if defined(OPENSSL_IS_BORINGSSL) + SSL_set_tlsext_host_name(ssl2, host_.c_str()); +#else // NOTE: Direct call instead of using the OpenSSL macro to suppress // -Wold-style-cast warning - // SSL_set_tlsext_host_name(ssl2, host_.c_str()); SSL_ctrl(ssl2, SSL_CTRL_SET_TLSEXT_HOSTNAME, TLSEXT_NAMETYPE_host_name, static_cast(const_cast(host_.c_str()))); +#endif return true; }); @@ -8812,7 +9601,8 @@ inline void SSLClient::shutdown_ssl_impl(Socket &socket, return; } if (socket.ssl) { - detail::ssl_delete(ctx_mutex_, socket.ssl, shutdown_gracefully); + detail::ssl_delete(ctx_mutex_, socket.ssl, socket.sock, + shutdown_gracefully); socket.ssl = nullptr; } assert(socket.ssl == nullptr); @@ -8861,8 +9651,8 @@ SSLClient::verify_host_with_subject_alt_name(X509 *server_cert) const { auto type = GEN_DNS; - struct in6_addr addr6 {}; - struct in_addr addr {}; + struct in6_addr addr6{}; + struct in_addr addr{}; size_t addr_len = 0; #ifndef __MINGW32__ @@ -8965,7 +9755,7 @@ inline Client::Client(const std::string &scheme_host_port, const std::string &client_cert_path, const std::string &client_key_path) { const static std::regex re( - R"((?:([a-z]+):\/\/)?(?:\[([\d:]+)\]|([^:/?#]+))(?::(\d+))?)"); + R"((?:([a-z]+):\/\/)?(?:\[([a-fA-F\d:]+)\]|([^:/?#]+))(?::(\d+))?)"); std::smatch m; if (std::regex_match(scheme_host_port, m, re)) { @@ -9002,10 +9792,12 @@ inline Client::Client(const std::string &scheme_host_port, client_key_path); } } else { + // NOTE: Update TEST(UniversalClientImplTest, Ipv6LiteralAddress) + // if port param below changes. cli_ = detail::make_unique(scheme_host_port, 80, client_cert_path, client_key_path); } -} +} // namespace detail inline Client::Client(const std::string &host, int port) : cli_(detail::make_unique(host, port)) {} @@ -9111,15 +9903,30 @@ inline Result Client::Post(const std::string &path, const Headers &headers, const std::string &content_type) { return cli_->Post(path, headers, body, content_length, content_type); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress) { + return cli_->Post(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Post(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Post(path, body, content_type); } +inline Result Client::Post(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Post(path, body, content_type, progress); +} inline Result Client::Post(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Post(path, headers, body, content_type); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Post(path, headers, body, content_type, progress); +} inline Result Client::Post(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9150,6 +9957,10 @@ inline Result Client::Post(const std::string &path, const Headers &headers, const Params ¶ms) { return cli_->Post(path, headers, params); } +inline Result Client::Post(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + return cli_->Post(path, headers, params, progress); +} inline Result Client::Post(const std::string &path, const MultipartFormDataItems &items) { return cli_->Post(path, items); @@ -9180,15 +9991,29 @@ inline Result Client::Put(const std::string &path, const Headers &headers, const std::string &content_type) { return cli_->Put(path, headers, body, content_length, content_type); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, Progress progress) { + return cli_->Put(path, headers, body, content_length, content_type, progress); +} inline Result Client::Put(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Put(path, body, content_type); } +inline Result Client::Put(const std::string &path, const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Put(path, body, content_type, progress); +} inline Result Client::Put(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Put(path, headers, body, content_type); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, Progress progress) { + return cli_->Put(path, headers, body, content_type, progress); +} inline Result Client::Put(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9219,6 +10044,10 @@ inline Result Client::Put(const std::string &path, const Headers &headers, const Params ¶ms) { return cli_->Put(path, headers, params); } +inline Result Client::Put(const std::string &path, const Headers &headers, + const Params ¶ms, Progress progress) { + return cli_->Put(path, headers, params, progress); +} inline Result Client::Put(const std::string &path, const MultipartFormDataItems &items) { return cli_->Put(path, items); @@ -9246,20 +10075,44 @@ inline Result Client::Patch(const std::string &path, const char *body, const std::string &content_type) { return cli_->Patch(path, body, content_length, content_type); } +inline Result Client::Patch(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, body, content_length, content_type, progress); +} inline Result Client::Patch(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return cli_->Patch(path, headers, body, content_length, content_type); } +inline Result Client::Patch(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Patch(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Patch(path, body, content_type); } +inline Result Client::Patch(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, body, content_type, progress); +} inline Result Client::Patch(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Patch(path, headers, body, content_type); } +inline Result Client::Patch(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Patch(path, headers, body, content_type, progress); +} inline Result Client::Patch(const std::string &path, size_t content_length, ContentProvider content_provider, const std::string &content_type) { @@ -9294,20 +10147,44 @@ inline Result Client::Delete(const std::string &path, const char *body, const std::string &content_type) { return cli_->Delete(path, body, content_length, content_type); } +inline Result Client::Delete(const std::string &path, const char *body, + size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, body, content_length, content_type, progress); +} inline Result Client::Delete(const std::string &path, const Headers &headers, const char *body, size_t content_length, const std::string &content_type) { return cli_->Delete(path, headers, body, content_length, content_type); } +inline Result Client::Delete(const std::string &path, const Headers &headers, + const char *body, size_t content_length, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, headers, body, content_length, content_type, + progress); +} inline Result Client::Delete(const std::string &path, const std::string &body, const std::string &content_type) { return cli_->Delete(path, body, content_type); } +inline Result Client::Delete(const std::string &path, const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, body, content_type, progress); +} inline Result Client::Delete(const std::string &path, const Headers &headers, const std::string &body, const std::string &content_type) { return cli_->Delete(path, headers, body, content_type); } +inline Result Client::Delete(const std::string &path, const Headers &headers, + const std::string &body, + const std::string &content_type, + Progress progress) { + return cli_->Delete(path, headers, body, content_type, progress); +} inline Result Client::Options(const std::string &path) { return cli_->Options(path); } @@ -9417,6 +10294,15 @@ inline void Client::set_proxy_digest_auth(const std::string &username, inline void Client::enable_server_certificate_verification(bool enabled) { cli_->enable_server_certificate_verification(enabled); } + +inline void Client::enable_server_hostname_verification(bool enabled) { + cli_->enable_server_hostname_verification(enabled); +} + +inline void Client::set_server_certificate_verifier( + std::function verifier) { + cli_->set_server_certificate_verifier(verifier); +} #endif inline void Client::set_logger(Logger logger) { diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 64c0c4ef6..d1e8ee829 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -19,6 +19,7 @@ #include "loading.html.hpp" #include +#include #include #include #include @@ -32,6 +33,8 @@ using json = nlohmann::ordered_json; +constexpr int HTTP_POLLING_SECONDS = 1; + enum stop_type { STOP_TYPE_NONE, STOP_TYPE_EOS, @@ -1602,6 +1605,30 @@ struct server_response { // should never reach here } + // same as recv(), but have timeout in seconds + // if timeout is reached, nullptr is returned + server_task_result_ptr recv_with_timeout(const std::unordered_set & id_tasks, int timeout) { + while (true) { + std::unique_lock lock(mutex_results); + bool cr_res = condition_results.wait_for(lock, std::chrono::seconds(timeout), [&]{ + return !queue_results.empty(); + }); + if (!cr_res) { + return nullptr; + } + + for (int i = 0; i < (int) queue_results.size(); i++) { + if (id_tasks.find(queue_results[i]->id) != id_tasks.end()) { + server_task_result_ptr res = std::move(queue_results[i]); + queue_results.erase(queue_results.begin() + i); + return res; + } + } + } + + // should never reach here + } + // single-task version of recv() server_task_result_ptr recv(int id_task) { std::unordered_set id_tasks = {id_task}; @@ -2322,10 +2349,21 @@ struct server_context { void receive_multi_results( const std::unordered_set & id_tasks, const std::function&)> & result_handler, - const std::function & error_handler) { + const std::function & error_handler, + const std::function & is_connection_closed) { std::vector results(id_tasks.size()); - for (size_t i = 0; i < id_tasks.size(); i++) { - server_task_result_ptr result = queue_results.recv(id_tasks); + for (int i = 0; i < (int)id_tasks.size(); i++) { + server_task_result_ptr result = queue_results.recv_with_timeout(id_tasks, HTTP_POLLING_SECONDS); + + if (is_connection_closed()) { + cancel_tasks(id_tasks); + return; + } + + if (result == nullptr) { + i--; // retry + continue; + } if (result->is_error()) { error_handler(result->to_json()); @@ -2349,10 +2387,20 @@ struct server_context { void receive_cmpl_results_stream( const std::unordered_set & id_tasks, const std::function & result_handler, - const std::function & error_handler) { + const std::function & error_handler, + const std::function & is_connection_closed) { size_t n_finished = 0; while (true) { - server_task_result_ptr result = queue_results.recv(id_tasks); + server_task_result_ptr result = queue_results.recv_with_timeout(id_tasks, HTTP_POLLING_SECONDS); + + if (is_connection_closed()) { + cancel_tasks(id_tasks); + return; + } + + if (result == nullptr) { + continue; // retry + } if (result->is_error()) { error_handler(result->to_json()); @@ -3633,6 +3681,7 @@ int main(int argc, char ** argv) { const auto handle_completions_impl = [&ctx_server, &res_error, &res_ok]( server_task_type type, json & data, + std::function is_connection_closed, httplib::Response & res, oaicompat_type oaicompat) { GGML_ASSERT(type == SERVER_TASK_TYPE_COMPLETION || type == SERVER_TASK_TYPE_INFILL); @@ -3694,7 +3743,7 @@ int main(int argc, char ** argv) { } }, [&](const json & error_data) { res_error(res, error_data); - }); + }, is_connection_closed); ctx_server.queue_results.remove_waiting_task_ids(task_ids); } else { @@ -3704,6 +3753,7 @@ int main(int argc, char ** argv) { if (res_json.is_array()) { for (const auto & res : res_json) { if (!server_sent_event(sink, "data", res)) { + // sending failed (HTTP connection closed), cancel the generation return false; } } @@ -3713,6 +3763,9 @@ int main(int argc, char ** argv) { } }, [&](const json & error_data) { server_sent_event(sink, "error", error_data); + }, [&sink]() { + // note: do not use req.is_connection_closed here because req is already destroyed + return !sink.is_writable(); }); if (oaicompat != OAICOMPAT_TYPE_NONE) { static const std::string ev_done = "data: [DONE]\n\n"; @@ -3735,6 +3788,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_NONE); }; @@ -3744,6 +3798,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_COMPLETION); }; @@ -3820,6 +3875,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_INFILL, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_NONE); // infill is not OAI compatible }; @@ -3834,6 +3890,7 @@ int main(int argc, char ** argv) { return handle_completions_impl( SERVER_TASK_TYPE_COMPLETION, data, + req.is_connection_closed, res, OAICOMPAT_TYPE_CHAT); }; @@ -3980,7 +4037,7 @@ int main(int argc, char ** argv) { }, [&](const json & error_data) { res_error(res, error_data); error = true; - }); + }, req.is_connection_closed); ctx_server.queue_results.remove_waiting_task_ids(task_ids); } @@ -4070,7 +4127,7 @@ int main(int argc, char ** argv) { }, [&](const json & error_data) { res_error(res, error_data); error = true; - }); + }, req.is_connection_closed); } if (error) { diff --git a/examples/server/tests/unit/test_completion.py b/examples/server/tests/unit/test_completion.py index e5e3b6077..c1fc12462 100644 --- a/examples/server/tests/unit/test_completion.py +++ b/examples/server/tests/unit/test_completion.py @@ -1,4 +1,5 @@ import pytest +import requests import time from openai import OpenAI from utils import * @@ -405,3 +406,23 @@ def test_n_probs_post_sampling(): assert "bytes" in prob and type(prob["bytes"]) == list # because the test model usually output token with either 100% or 0% probability, we need to check all the top_probs assert any(prob["prob"] == 1.0 for prob in tok["top_probs"]) + + +def test_cancel_request(): + global server + server.n_ctx = 4096 + server.n_predict = -1 + server.n_slots = 1 + server.server_slots = True + server.start() + # send a request that will take a long time, but cancel it before it finishes + try: + server.make_request("POST", "/completion", data={ + "prompt": "I believe the meaning of life is", + }, timeout=0.1) + except requests.exceptions.ReadTimeout: + pass # expected + # make sure the slot is free + time.sleep(1) # wait for HTTP_POLLING_SECONDS + res = server.make_request("GET", "/slots") + assert res.body[0]["is_processing"] == False diff --git a/examples/server/tests/utils.py b/examples/server/tests/utils.py index a1a94d0f1..73be4c92f 100644 --- a/examples/server/tests/utils.py +++ b/examples/server/tests/utils.py @@ -219,17 +219,18 @@ class ServerProcess: path: str, data: dict | Any | None = None, headers: dict | None = None, + timeout: float | None = None, ) -> ServerResponse: url = f"http://{self.server_host}:{self.server_port}{path}" parse_body = False if method == "GET": - response = requests.get(url, headers=headers) + response = requests.get(url, headers=headers, timeout=timeout) parse_body = True elif method == "POST": - response = requests.post(url, headers=headers, json=data) + response = requests.post(url, headers=headers, json=data, timeout=timeout) parse_body = True elif method == "OPTIONS": - response = requests.options(url, headers=headers) + response = requests.options(url, headers=headers, timeout=timeout) else: raise ValueError(f"Unimplemented method: {method}") result = ServerResponse() From 4dd34ff83165a483ebff7bd43621b28490fa1fd6 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sat, 18 Jan 2025 16:18:15 +0200 Subject: [PATCH 05/26] cmake : add sanitizer flags for llama.cpp (#11279) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * cmake : add sanitizer flags for llama.cpp ggml-ci * tests : fix compile warnings ggml-ci * cmake : move sanitizer flags to llama_add_compile_flags ggml-ci * cmake : move llama.cpp compile flags to top level lists ggml-ci * cmake : apply only sanitizer flags at top level ggml-ci * tests : fix gguf context use in same_tensor_data * gguf-test: tensor data comparison * dummy : trigger ggml-ci * unicode : silence gcc warnings ggml-ci * ci : use sanitizer builds only in Debug mode ggml-ci * cmake : add status messages [no ci] --------- Co-authored-by: Johannes Gäßler --- .github/workflows/build.yml | 2 +- CMakeLists.txt | 73 +++++++++++++++++++++++++------------ ggml/src/gguf.cpp | 4 ++ src/unicode.cpp | 5 +-- tests/CMakeLists.txt | 2 + tests/test-gguf.cpp | 28 ++++++++------ tests/test-sampling.cpp | 1 - 7 files changed, 74 insertions(+), 41 deletions(-) diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml index c85999b89..9e0c4a675 100644 --- a/.github/workflows/build.yml +++ b/.github/workflows/build.yml @@ -234,7 +234,7 @@ jobs: strategy: matrix: sanitizer: [ADDRESS, THREAD, UNDEFINED] - build_type: [Debug, Release] + build_type: [Debug] steps: - name: Clone diff --git a/CMakeLists.txt b/CMakeLists.txt index a717a508f..42caed486 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -83,11 +83,8 @@ include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake) include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake) # override ggml options -set(GGML_SANITIZE_THREAD ${LLAMA_SANITIZE_THREAD}) -set(GGML_SANITIZE_ADDRESS ${LLAMA_SANITIZE_ADDRESS}) -set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED}) -set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS}) -set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS}) +set(GGML_ALL_WARNINGS ${LLAMA_ALL_WARNINGS}) +set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS}) # change the default for these ggml options if (NOT DEFINED GGML_LLAMAFILE) @@ -117,16 +114,62 @@ llama_option_depr(WARNING LLAMA_SYCL GGML_SYCL) llama_option_depr(WARNING LLAMA_SYCL_F16 GGML_SYCL_F16) llama_option_depr(WARNING LLAMA_CANN GGML_CANN) +if (NOT MSVC) + if (LLAMA_SANITIZE_THREAD) + message(STATUS "Using -fsanitize=thread") + + add_compile_options(-fsanitize=thread) + link_libraries (-fsanitize=thread) + endif() + + if (LLAMA_SANITIZE_ADDRESS) + message(STATUS "Using -fsanitize=address") + + add_compile_options(-fsanitize=address -fno-omit-frame-pointer) + link_libraries (-fsanitize=address) + endif() + + if (LLAMA_SANITIZE_UNDEFINED) + message(STATUS "Using -fsanitize=undefined") + + add_compile_options(-fsanitize=undefined) + link_libraries (-fsanitize=undefined) + endif() +endif() + # -# build the library +# 3rd-party # if (NOT TARGET ggml) add_subdirectory(ggml) # ... otherwise assume ggml is added by a parent CMakeLists.txt endif() + +# +# build the library +# + add_subdirectory(src) +# +# utils, programs, examples and tests +# + +if (LLAMA_BUILD_COMMON) + add_subdirectory(common) +endif() + +if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION) + include(CTest) + add_subdirectory(tests) +endif() + +if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES) + add_subdirectory(examples) + add_subdirectory(pocs) +endif() + # # install # @@ -200,21 +243,3 @@ configure_file(cmake/llama.pc.in install(FILES "${CMAKE_CURRENT_BINARY_DIR}/llama.pc" DESTINATION lib/pkgconfig) - -# -# utils, programs, examples and tests -# - -if (LLAMA_BUILD_COMMON) - add_subdirectory(common) -endif() - -if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION) - include(CTest) - add_subdirectory(tests) -endif() - -if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES) - add_subdirectory(examples) - add_subdirectory(pocs) -endif() diff --git a/ggml/src/gguf.cpp b/ggml/src/gguf.cpp index 655ed600a..ab13669c5 100644 --- a/ggml/src/gguf.cpp +++ b/ggml/src/gguf.cpp @@ -648,6 +648,10 @@ struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_par ok = ok && data != nullptr; + if (ok) { + ggml_set_name(data, "GGUF tensor data binary blob"); + } + // read the binary blob with the tensor data ok = ok && gr.read(data->data, ctx->size); diff --git a/src/unicode.cpp b/src/unicode.cpp index 7aca6544b..89180da41 100644 --- a/src/unicode.cpp +++ b/src/unicode.cpp @@ -7,18 +7,17 @@ #include #include +#include #include #include +#include #include #include #include #include #include -#include #include #include -#include -#include size_t unicode_len_utf8(char src) { const size_t lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 }; diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt index 2b5e5fd4a..3fa43c295 100644 --- a/tests/CMakeLists.txt +++ b/tests/CMakeLists.txt @@ -1,3 +1,5 @@ +llama_add_compile_flags() + function(llama_test target) include(CMakeParseArguments) set(options) diff --git a/tests/test-gguf.cpp b/tests/test-gguf.cpp index 611957ac0..6ed696328 100644 --- a/tests/test-gguf.cpp +++ b/tests/test-gguf.cpp @@ -48,7 +48,7 @@ enum handcrafted_file_type { HANDCRAFTED_DATA_CUSTOM_ALIGN = 810 + offset_has_data, }; -std::string handcrafted_file_type_name(const enum handcrafted_file_type hft) { +static std::string handcrafted_file_type_name(const enum handcrafted_file_type hft) { switch (hft) { case HANDCRAFTED_HEADER_BAD_MAGIC: return "HEADER_BAD_MAGIC"; case HANDCRAFTED_HEADER_BAD_VERSION_1: return "HEADER_BAD_VERSION_1"; @@ -99,7 +99,7 @@ static bool expect_context_not_null(const enum handcrafted_file_type hft) { typedef std::pair> tensor_config_t; -std::vector get_tensor_configs(std::mt19937 & rng) { +static std::vector get_tensor_configs(std::mt19937 & rng) { std::vector tensor_configs; tensor_configs.reserve(100); @@ -122,7 +122,7 @@ std::vector get_tensor_configs(std::mt19937 & rng) { return tensor_configs; } -std::vector> get_kv_types(std::mt19937 rng) { +static std::vector> get_kv_types(std::mt19937 rng) { std::vector> kv_types; kv_types.reserve(100); @@ -626,8 +626,6 @@ static bool handcrafted_check_tensor_data(const gguf_context * gguf_ctx, const u bool ok = true; - const uint32_t alignment = GGUF_DEFAULT_ALIGNMENT; - for (int i = 0; i < int(tensor_configs.size()); ++i) { const ggml_type type = tensor_configs[i].first; const std::array shape = tensor_configs[i].second; @@ -866,13 +864,13 @@ static struct random_gguf_context_result get_random_gguf_context(ggml_backend_t case GGUF_TYPE_COUNT: default: { GGML_ABORT("fatal error"); - } break; + } } } break; case GGUF_TYPE_COUNT: default: { GGML_ABORT("fatal error"); - } break; + } } } @@ -938,7 +936,7 @@ static bool all_kv_in_other(const gguf_context * ctx, const gguf_context * other } if (type == GGUF_TYPE_ARRAY) { - const int arr_n = gguf_get_arr_n(ctx, id); + const size_t arr_n = gguf_get_arr_n(ctx, id); if (arr_n != gguf_get_arr_n(other, idx_other)) { ok = false; continue; @@ -953,7 +951,7 @@ static bool all_kv_in_other(const gguf_context * ctx, const gguf_context * other if (type_arr == GGUF_TYPE_BOOL) { const int8_t * data = reinterpret_cast(gguf_get_arr_data(ctx, id)); const int8_t * data_other = reinterpret_cast(gguf_get_arr_data(other, idx_other)); - for (int arr_i = 0; arr_i < arr_n; ++arr_i) { + for (size_t arr_i = 0; arr_i < arr_n; ++arr_i) { if (bool(data[arr_i]) != bool(data_other[arr_i])) { ok = false; } @@ -962,7 +960,7 @@ static bool all_kv_in_other(const gguf_context * ctx, const gguf_context * other } if (type_arr == GGUF_TYPE_STRING) { - for (int arr_i = 0; arr_i < arr_n; ++arr_i) { + for (size_t arr_i = 0; arr_i < arr_n; ++arr_i) { const std::string str = gguf_get_arr_str(ctx, id, arr_i); const std::string str_other = gguf_get_arr_str(other, idx_other, arr_i); if (str != str_other) { @@ -1033,6 +1031,12 @@ static bool same_tensor_data(const struct ggml_context * orig, const struct ggml struct ggml_tensor * t_orig = ggml_get_first_tensor(orig); struct ggml_tensor * t_read = ggml_get_first_tensor(read); + + if (std::string(t_read->name) != "GGUF tensor data binary blob") { + return false; + } + t_read = ggml_get_next_tensor(read, t_read); + while (t_orig) { if (!t_read) { ok = false; @@ -1051,13 +1055,13 @@ static bool same_tensor_data(const struct ggml_context * orig, const struct ggml } t_orig = ggml_get_next_tensor(orig, t_orig); - t_read = ggml_get_next_tensor(orig, t_read); + t_read = ggml_get_next_tensor(read, t_read); } if (t_read) { ok = false; } - return true; + return ok; } static std::pair test_roundtrip(ggml_backend_dev_t dev, const unsigned int seed, const bool only_meta) { diff --git a/tests/test-sampling.cpp b/tests/test-sampling.cpp index c0dcb4848..61bd67850 100644 --- a/tests/test-sampling.cpp +++ b/tests/test-sampling.cpp @@ -144,7 +144,6 @@ static void test_penalties( sampler_tester tester(probs, probs_expected); - const size_t n_vocab = probs.size(); auto * sampler = llama_sampler_init_penalties(last_tokens.size(), repeat_penalty, alpha_frequency, alpha_presence); for (size_t i = 0; i < last_tokens.size(); i++) { From a1649cc13f89946322358f92ea268ae1b7b5096c Mon Sep 17 00:00:00 2001 From: Eric Curtin Date: Sat, 18 Jan 2025 14:42:31 +0000 Subject: [PATCH 06/26] Adding linenoise.cpp to llama-run (#11252) This is a fork of linenoise that is C++17 compatible. I intend on adding it to llama-run so we can do things like traverse prompt history via the up and down arrows: https://github.com/ericcurtin/linenoise.cpp Signed-off-by: Eric Curtin --- .github/workflows/build.yml | 4 + examples/run/CMakeLists.txt | 2 +- examples/run/linenoise.cpp/LICENSE | 26 + examples/run/linenoise.cpp/linenoise.cpp | 1351 ++++++++++++++++++++++ examples/run/linenoise.cpp/linenoise.h | 114 ++ examples/run/run.cpp | 38 +- 6 files changed, 1524 insertions(+), 11 deletions(-) create mode 100644 examples/run/linenoise.cpp/LICENSE create mode 100644 examples/run/linenoise.cpp/linenoise.cpp create mode 100644 examples/run/linenoise.cpp/linenoise.h diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml index 9e0c4a675..fe3b2cdfa 100644 --- a/.github/workflows/build.yml +++ b/.github/workflows/build.yml @@ -87,6 +87,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.zip ./build/bin/* - name: Upload artifacts @@ -149,6 +150,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip ./build/bin/* - name: Upload artifacts @@ -217,6 +219,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | cp LICENSE ./build/bin/ + cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.zip ./build/bin/* - name: Upload artifacts @@ -796,6 +799,7 @@ jobs: if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }} run: | Copy-Item LICENSE .\build\bin\Release\llama.cpp.txt + Copy-Item .\examples\run\linenoise.cpp\LICENSE .\build\bin\Release\linenoise.cpp.txt 7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}.zip .\build\bin\Release\* - name: Upload artifacts diff --git a/examples/run/CMakeLists.txt b/examples/run/CMakeLists.txt index 0686d6305..cd6b0520e 100644 --- a/examples/run/CMakeLists.txt +++ b/examples/run/CMakeLists.txt @@ -1,5 +1,5 @@ set(TARGET llama-run) -add_executable(${TARGET} run.cpp) +add_executable(${TARGET} run.cpp linenoise.cpp/linenoise.cpp) install(TARGETS ${TARGET} RUNTIME) target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT}) target_compile_features(${TARGET} PRIVATE cxx_std_17) diff --git a/examples/run/linenoise.cpp/LICENSE b/examples/run/linenoise.cpp/LICENSE new file mode 100644 index 000000000..b006b3b24 --- /dev/null +++ b/examples/run/linenoise.cpp/LICENSE @@ -0,0 +1,26 @@ +Copyright (c) 2010-2014, Salvatore Sanfilippo +Copyright (c) 2010-2013, Pieter Noordhuis +Copyright (c) 2025, Eric Curtin + +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +* Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + +* Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR +ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON +ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/examples/run/linenoise.cpp/linenoise.cpp b/examples/run/linenoise.cpp/linenoise.cpp new file mode 100644 index 000000000..050c23012 --- /dev/null +++ b/examples/run/linenoise.cpp/linenoise.cpp @@ -0,0 +1,1351 @@ +#ifndef _WIN32 +/* + * You can find the latest source code at: + * + * http://github.com/ericcurtin/linenoise.cpp + * + * Does a number of crazy assumptions that happen to be true in 99.9999% of + * the 2010 UNIX computers around. + * + * ------------------------------------------------------------------------ + * + * Copyright (c) 2010-2023, Salvatore Sanfilippo + * Copyright (c) 2010-2013, Pieter Noordhuis + * Copyright (c) 2025, Eric Curtin + * + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * ------------------------------------------------------------------------ + * + * References: + * - http://invisible-island.net/xterm/ctlseqs/ctlseqs.html + * - http://www.3waylabs.com/nw/WWW/products/wizcon/vt220.html + * + * Todo list: + * - Filter bogus Ctrl+ combinations. + * - Win32 support + * + * Bloat: + * - History search like Ctrl+r in readline? + * + * List of escape sequences used by this program, we do everything just + * with three sequences. In order to be so cheap we may have some + * flickering effect with some slow terminal, but the lesser sequences + * the more compatible. + * + * EL (Erase Line) + * Sequence: ESC [ n K + * Effect: if n is 0 or missing, clear from cursor to end of line + * Effect: if n is 1, clear from beginning of line to cursor + * Effect: if n is 2, clear entire line + * + * CUF (CUrsor Forward) + * Sequence: ESC [ n C + * Effect: moves cursor forward n chars + * + * CUB (CUrsor Backward) + * Sequence: ESC [ n D + * Effect: moves cursor backward n chars + * + * The following is used to get the terminal width if getting + * the width with the TIOCGWINSZ ioctl fails + * + * DSR (Device Status Report) + * Sequence: ESC [ 6 n + * Effect: reports the current cusor position as ESC [ n ; m R + * where n is the row and m is the column + * + * When multi line mode is enabled, we also use an additional escape + * sequence. However multi line editing is disabled by default. + * + * CUU (Cursor Up) + * Sequence: ESC [ n A + * Effect: moves cursor up of n chars. + * + * CUD (Cursor Down) + * Sequence: ESC [ n B + * Effect: moves cursor down of n chars. + * + * When linenoiseClearScreen() is called, two additional escape sequences + * are used in order to clear the screen and position the cursor at home + * position. + * + * CUP (Cursor position) + * Sequence: ESC [ H + * Effect: moves the cursor to upper left corner + * + * ED (Erase display) + * Sequence: ESC [ 2 J + * Effect: clear the whole screen + * + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "linenoise.h" + +#define LINENOISE_DEFAULT_HISTORY_MAX_LEN 100 +#define LINENOISE_MAX_LINE 4096 +static std::vector unsupported_term = {"dumb","cons25","emacs",nullptr}; +static linenoiseCompletionCallback *completionCallback = NULL; +static linenoiseHintsCallback *hintsCallback = NULL; +static linenoiseFreeHintsCallback *freeHintsCallback = NULL; +static char *linenoiseNoTTY(void); +static void refreshLineWithCompletion(struct linenoiseState *ls, linenoiseCompletions *lc, int flags); +static void refreshLineWithFlags(struct linenoiseState *l, int flags); + +static struct termios orig_termios; /* In order to restore at exit.*/ +static int maskmode = 0; /* Show "***" instead of input. For passwords. */ +static int rawmode = 0; /* For atexit() function to check if restore is needed*/ +static int mlmode = 0; /* Multi line mode. Default is single line. */ +static int atexit_registered = 0; /* Register atexit just 1 time. */ +static int history_max_len = LINENOISE_DEFAULT_HISTORY_MAX_LEN; +static int history_len = 0; +static char **history = NULL; + +enum KEY_ACTION{ + KEY_NULL = 0, /* NULL */ + CTRL_A = 1, /* Ctrl+a */ + CTRL_B = 2, /* Ctrl-b */ + CTRL_C = 3, /* Ctrl-c */ + CTRL_D = 4, /* Ctrl-d */ + CTRL_E = 5, /* Ctrl-e */ + CTRL_F = 6, /* Ctrl-f */ + CTRL_H = 8, /* Ctrl-h */ + TAB = 9, /* Tab */ + CTRL_K = 11, /* Ctrl+k */ + CTRL_L = 12, /* Ctrl+l */ + ENTER = 13, /* Enter */ + CTRL_N = 14, /* Ctrl-n */ + CTRL_P = 16, /* Ctrl-p */ + CTRL_T = 20, /* Ctrl-t */ + CTRL_U = 21, /* Ctrl+u */ + CTRL_W = 23, /* Ctrl+w */ + ESC = 27, /* Escape */ + BACKSPACE = 127 /* Backspace */ +}; + +static void linenoiseAtExit(void); +int linenoiseHistoryAdd(const char *line); +#define REFRESH_CLEAN (1<<0) // Clean the old prompt from the screen +#define REFRESH_WRITE (1<<1) // Rewrite the prompt on the screen. +#define REFRESH_ALL (REFRESH_CLEAN|REFRESH_WRITE) // Do both. +static void refreshLine(struct linenoiseState *l); + +__attribute__((format(printf, 1, 2))) +/* Debugging function. */ +#if 0 +static void lndebug(const char *fmt, ...) { + static FILE *lndebug_fp = NULL; + if (lndebug_fp == NULL) { + lndebug_fp = fopen("/tmp/lndebug.txt", "a"); + } + + if (lndebug_fp != NULL) { + va_list args; + va_start(args, fmt); + vfprintf(lndebug_fp, fmt, args); + va_end(args); + fflush(lndebug_fp); + } +} +#else +static void lndebug(const char *, ...) { +} +#endif + +/* ======================= Low level terminal handling ====================== */ + +/* Enable "mask mode". When it is enabled, instead of the input that + * the user is typing, the terminal will just display a corresponding + * number of asterisks, like "****". This is useful for passwords and other + * secrets that should not be displayed. */ +void linenoiseMaskModeEnable(void) { + maskmode = 1; +} + +/* Disable mask mode. */ +void linenoiseMaskModeDisable(void) { + maskmode = 0; +} + +/* Set if to use or not the multi line mode. */ +void linenoiseSetMultiLine(int ml) { + mlmode = ml; +} + +/* Return true if the terminal name is in the list of terminals we know are + * not able to understand basic escape sequences. */ +static int isUnsupportedTerm(void) { + char *term = getenv("TERM"); + if (term == NULL) return 0; + for (int j = 0; unsupported_term[j]; ++j) + if (!strcasecmp(term, unsupported_term[j])) return 1; + return 0; +} + +/* Raw mode: 1960 magic shit. */ +static int enableRawMode(int fd) { + struct termios raw; + + if (!isatty(STDIN_FILENO)) goto fatal; + if (!atexit_registered) { + atexit(linenoiseAtExit); + atexit_registered = 1; + } + if (tcgetattr(fd,&orig_termios) == -1) goto fatal; + + raw = orig_termios; /* modify the original mode */ + /* input modes: no break, no CR to NL, no parity check, no strip char, + * no start/stop output control. */ + raw.c_iflag &= ~(BRKINT | ICRNL | INPCK | ISTRIP | IXON); + /* output modes - disable post processing */ + raw.c_oflag &= ~(OPOST); + /* control modes - set 8 bit chars */ + raw.c_cflag |= (CS8); + /* local modes - choing off, canonical off, no extended functions, + * no signal chars (^Z,^C) */ + raw.c_lflag &= ~(ECHO | ICANON | IEXTEN | ISIG); + /* control chars - set return condition: min number of bytes and timer. + * We want read to return every single byte, without timeout. */ + raw.c_cc[VMIN] = 1; raw.c_cc[VTIME] = 0; /* 1 byte, no timer */ + + /* put terminal in raw mode after flushing */ + if (tcsetattr(fd,TCSAFLUSH,&raw) < 0) goto fatal; + rawmode = 1; + return 0; + +fatal: + errno = ENOTTY; + return -1; +} + +static void disableRawMode(int fd) { + /* Don't even check the return value as it's too late. */ + if (rawmode && tcsetattr(fd,TCSAFLUSH,&orig_termios) != -1) + rawmode = 0; +} + +/* Use the ESC [6n escape sequence to query the horizontal cursor position + * and return it. On error -1 is returned, on success the position of the + * cursor. */ +static int getCursorPosition(int ifd, int ofd) { + char buf[32]; + int cols, rows; + unsigned int i = 0; + + /* Report cursor location */ + if (write(ofd, "\x1b[6n", 4) != 4) return -1; + + /* Read the response: ESC [ rows ; cols R */ + while (i < sizeof(buf)-1) { + if (read(ifd,buf+i,1) != 1) break; + if (buf[i] == 'R') break; + i++; + } + buf[i] = '\0'; + + /* Parse it. */ + if (buf[0] != ESC || buf[1] != '[') return -1; + if (sscanf(buf+2,"%d;%d",&rows,&cols) != 2) return -1; + return cols; +} + +/* Try to get the number of columns in the current terminal, or assume 80 + * if it fails. */ +static int getColumns(int ifd, int ofd) { + struct winsize ws; + + if (ioctl(1, TIOCGWINSZ, &ws) == -1 || ws.ws_col == 0) { + /* ioctl() failed. Try to query the terminal itself. */ + int start, cols; + + /* Get the initial position so we can restore it later. */ + start = getCursorPosition(ifd,ofd); + if (start == -1) goto failed; + + /* Go to right margin and get position. */ + if (write(ofd,"\x1b[999C",6) != 6) goto failed; + cols = getCursorPosition(ifd,ofd); + if (cols == -1) goto failed; + + /* Restore position. */ + if (cols > start) { + char seq[32]; + snprintf(seq,32,"\x1b[%dD",cols-start); + if (write(ofd,seq,strlen(seq)) == -1) { + /* Can't recover... */ + } + } + return cols; + } else { + return ws.ws_col; + } + +failed: + return 80; +} + +/* Clear the screen. Used to handle ctrl+l */ +void linenoiseClearScreen(void) { + if (write(STDOUT_FILENO,"\x1b[H\x1b[2J",7) <= 0) { + /* nothing to do, just to avoid warning. */ + } +} + +/* Beep, used for completion when there is nothing to complete or when all + * the choices were already shown. */ +static void linenoiseBeep(void) { + fprintf(stderr, "\x7"); + fflush(stderr); +} + +/* ============================== Completion ================================ */ + +/* Free a list of completion option populated by linenoiseAddCompletion(). */ +static void freeCompletions(linenoiseCompletions *lc) { + size_t i; + for (i = 0; i < lc->len; i++) + free(lc->cvec[i]); + if (lc->cvec != NULL) + free(lc->cvec); +} + +/* Called by completeLine() and linenoiseShow() to render the current + * edited line with the proposed completion. If the current completion table + * is already available, it is passed as second argument, otherwise the + * function will use the callback to obtain it. + * + * Flags are the same as refreshLine*(), that is REFRESH_* macros. */ +static void refreshLineWithCompletion(struct linenoiseState *ls, linenoiseCompletions *lc, int flags) { + /* Obtain the table of completions if the caller didn't provide one. */ + linenoiseCompletions ctable = { 0, NULL }; + if (lc == NULL) { + completionCallback(ls->buf,&ctable); + lc = &ctable; + } + + /* Show the edited line with completion if possible, or just refresh. */ + if (ls->completion_idx < lc->len) { + struct linenoiseState saved = *ls; + ls->len = ls->pos = strlen(lc->cvec[ls->completion_idx]); + ls->buf = lc->cvec[ls->completion_idx]; + refreshLineWithFlags(ls,flags); + ls->len = saved.len; + ls->pos = saved.pos; + ls->buf = saved.buf; + } else { + refreshLineWithFlags(ls,flags); + } + + /* Free the completions table if needed. */ + if (lc != &ctable) freeCompletions(&ctable); +} + +/* This is an helper function for linenoiseEdit*() and is called when the + * user types the key in order to complete the string currently in the + * input. + * + * The state of the editing is encapsulated into the pointed linenoiseState + * structure as described in the structure definition. + * + * If the function returns non-zero, the caller should handle the + * returned value as a byte read from the standard input, and process + * it as usually: this basically means that the function may return a byte + * read from the termianl but not processed. Otherwise, if zero is returned, + * the input was consumed by the completeLine() function to navigate the + * possible completions, and the caller should read for the next characters + * from stdin. */ +static int completeLine(struct linenoiseState *ls, int keypressed) { + linenoiseCompletions lc = { 0, NULL }; + int nwritten; + char c = keypressed; + + completionCallback(ls->buf,&lc); + if (lc.len == 0) { + linenoiseBeep(); + ls->in_completion = 0; + } else { + switch(c) { + case 9: /* tab */ + if (ls->in_completion == 0) { + ls->in_completion = 1; + ls->completion_idx = 0; + } else { + ls->completion_idx = (ls->completion_idx+1) % (lc.len+1); + if (ls->completion_idx == lc.len) linenoiseBeep(); + } + c = 0; + break; + case 27: /* escape */ + /* Re-show original buffer */ + if (ls->completion_idx < lc.len) refreshLine(ls); + ls->in_completion = 0; + c = 0; + break; + default: + /* Update buffer and return */ + if (ls->completion_idx < lc.len) { + nwritten = snprintf(ls->buf,ls->buflen,"%s", + lc.cvec[ls->completion_idx]); + ls->len = ls->pos = nwritten; + } + ls->in_completion = 0; + break; + } + + /* Show completion or original buffer */ + if (ls->in_completion && ls->completion_idx < lc.len) { + refreshLineWithCompletion(ls,&lc,REFRESH_ALL); + } else { + refreshLine(ls); + } + } + + freeCompletions(&lc); + return c; /* Return last read character */ +} + +/* Register a callback function to be called for tab-completion. */ +void linenoiseSetCompletionCallback(linenoiseCompletionCallback *fn) { + completionCallback = fn; +} + +/* Register a hits function to be called to show hits to the user at the + * right of the prompt. */ +void linenoiseSetHintsCallback(linenoiseHintsCallback *fn) { + hintsCallback = fn; +} + +/* Register a function to free the hints returned by the hints callback + * registered with linenoiseSetHintsCallback(). */ +void linenoiseSetFreeHintsCallback(linenoiseFreeHintsCallback *fn) { + freeHintsCallback = fn; +} + +/* This function is used by the callback function registered by the user + * in order to add completion options given the input string when the + * user typed . See the example.c source code for a very easy to + * understand example. */ +void linenoiseAddCompletion(linenoiseCompletions *lc, const char *str) { + size_t len = strlen(str); + char *copy, **cvec; + + copy = (char*) malloc(len + 1); + if (copy == NULL) return; + memcpy(copy,str,len+1); + cvec = (char**) realloc(lc->cvec,sizeof(char*)*(lc->len+1)); + if (cvec == NULL) { + free(copy); + return; + } + lc->cvec = cvec; + lc->cvec[lc->len++] = copy; +} + +/* =========================== Line editing ================================= */ + +/* We define a very simple "append buffer" structure, that is an heap + * allocated string where we can append to. This is useful in order to + * write all the escape sequences in a buffer and flush them to the standard + * output in a single call, to avoid flickering effects. */ +struct abuf { + char *b; + int len; +}; + +static void abInit(struct abuf *ab) { + ab->b = NULL; + ab->len = 0; +} + +static void abAppend(struct abuf *ab, const char *s, int len) { + char *new_ptr = (char*) realloc(ab->b,ab->len+len); + + if (new_ptr == NULL) return; + memcpy(new_ptr+ab->len,s,len); + ab->b = new_ptr; + ab->len += len; +} + +static void abFree(struct abuf *ab) { + free(ab->b); +} + +/* Helper of refreshSingleLine() and refreshMultiLine() to show hints + * to the right of the prompt. */ +static void refreshShowHints(struct abuf * ab, struct linenoiseState * l, int plen) { + char seq[64]; + if (hintsCallback && plen+l->len < l->cols) { + int color = -1, bold = 0; + const char *hint = hintsCallback(l->buf,&color,&bold); + if (hint) { + int hintlen = strlen(hint); + int hintmaxlen = l->cols-(plen+l->len); + if (hintlen > hintmaxlen) hintlen = hintmaxlen; + if (bold == 1 && color == -1) color = 37; + if (color != -1 || bold != 0) + snprintf(seq,64,"\033[%d;%d;49m",bold,color); + else + seq[0] = '\0'; + abAppend(ab,seq,strlen(seq)); + abAppend(ab,hint,hintlen); + if (color != -1 || bold != 0) + abAppend(ab,"\033[0m",4); + /* Call the function to free the hint returned. */ + if (freeHintsCallback) freeHintsCallback(hint); + } + } +} + +/* Single line low level line refresh. + * + * Rewrite the currently edited line accordingly to the buffer content, + * cursor position, and number of columns of the terminal. + * + * Flags is REFRESH_* macros. The function can just remove the old + * prompt, just write it, or both. */ +static void refreshSingleLine(struct linenoiseState *l, int flags) { + char seq[64]; + size_t plen = strlen(l->prompt); + int fd = l->ofd; + char *buf = l->buf; + size_t len = l->len; + size_t pos = l->pos; + struct abuf ab; + + while((plen+pos) >= l->cols) { + buf++; + len--; + pos--; + } + while (plen+len > l->cols) { + len--; + } + + abInit(&ab); + /* Cursor to left edge */ + snprintf(seq,sizeof(seq),"\r"); + abAppend(&ab,seq,strlen(seq)); + + if (flags & REFRESH_WRITE) { + /* Write the prompt and the current buffer content */ + abAppend(&ab,l->prompt,strlen(l->prompt)); + if (maskmode == 1) { + while (len--) abAppend(&ab,"*",1); + } else { + abAppend(&ab,buf,len); + } + /* Show hits if any. */ + refreshShowHints(&ab,l,plen); + } + + /* Erase to right */ + snprintf(seq,sizeof(seq),"\x1b[0K"); + abAppend(&ab,seq,strlen(seq)); + + if (flags & REFRESH_WRITE) { + /* Move cursor to original position. */ + snprintf(seq,sizeof(seq),"\r\x1b[%dC", (int)(pos+plen)); + abAppend(&ab,seq,strlen(seq)); + } + + if (write(fd,ab.b,ab.len) == -1) {} /* Can't recover from write error. */ + abFree(&ab); +} + +/* Multi line low level line refresh. + * + * Rewrite the currently edited line accordingly to the buffer content, + * cursor position, and number of columns of the terminal. + * + * Flags is REFRESH_* macros. The function can just remove the old + * prompt, just write it, or both. */ +static void refreshMultiLine(struct linenoiseState *l, int flags) { + char seq[64]; + int plen = strlen(l->prompt); + int rows = (plen+l->len+l->cols-1)/l->cols; /* rows used by current buf. */ + int rpos = (plen+l->oldpos+l->cols)/l->cols; /* cursor relative row. */ + int rpos2; /* rpos after refresh. */ + int col; /* colum position, zero-based. */ + int old_rows = l->oldrows; + int fd = l->ofd, j; + struct abuf ab; + + l->oldrows = rows; + + /* First step: clear all the lines used before. To do so start by + * going to the last row. */ + abInit(&ab); + + if (flags & REFRESH_CLEAN) { + if (old_rows-rpos > 0) { + lndebug("go down %d", old_rows-rpos); + snprintf(seq,64,"\x1b[%dB", old_rows-rpos); + abAppend(&ab,seq,strlen(seq)); + } + + /* Now for every row clear it, go up. */ + for (j = 0; j < old_rows-1; j++) { + lndebug("clear+up"); + snprintf(seq,64,"\r\x1b[0K\x1b[1A"); + abAppend(&ab,seq,strlen(seq)); + } + } + + if (flags & REFRESH_ALL) { + /* Clean the top line. */ + lndebug("clear"); + snprintf(seq,64,"\r\x1b[0K"); + abAppend(&ab,seq,strlen(seq)); + } + + if (flags & REFRESH_WRITE) { + /* Write the prompt and the current buffer content */ + abAppend(&ab,l->prompt,strlen(l->prompt)); + if (maskmode == 1) { + unsigned int i; + for (i = 0; i < l->len; i++) abAppend(&ab,"*",1); + } else { + abAppend(&ab,l->buf,l->len); + } + + /* Show hits if any. */ + refreshShowHints(&ab,l,plen); + + /* If we are at the very end of the screen with our prompt, we need to + * emit a newline and move the prompt to the first column. */ + if (l->pos && + l->pos == l->len && + (l->pos+plen) % l->cols == 0) + { + lndebug(""); + abAppend(&ab,"\n",1); + snprintf(seq,64,"\r"); + abAppend(&ab,seq,strlen(seq)); + rows++; + if (rows > (int)l->oldrows) l->oldrows = rows; + } + + /* Move cursor to right position. */ + rpos2 = (plen+l->pos+l->cols)/l->cols; /* Current cursor relative row */ + lndebug("rpos2 %d", rpos2); + + /* Go up till we reach the expected positon. */ + if (rows-rpos2 > 0) { + lndebug("go-up %d", rows-rpos2); + snprintf(seq,64,"\x1b[%dA", rows-rpos2); + abAppend(&ab,seq,strlen(seq)); + } + + /* Set column. */ + col = (plen+(int)l->pos) % (int)l->cols; + lndebug("set col %d", 1+col); + if (col) + snprintf(seq,64,"\r\x1b[%dC", col); + else + snprintf(seq,64,"\r"); + abAppend(&ab,seq,strlen(seq)); + } + + lndebug("\n"); + l->oldpos = l->pos; + + if (write(fd,ab.b,ab.len) == -1) {} /* Can't recover from write error. */ + abFree(&ab); +} + +/* Calls the two low level functions refreshSingleLine() or + * refreshMultiLine() according to the selected mode. */ +static void refreshLineWithFlags(struct linenoiseState *l, int flags) { + if (mlmode) + refreshMultiLine(l,flags); + else + refreshSingleLine(l,flags); +} + +/* Utility function to avoid specifying REFRESH_ALL all the times. */ +static void refreshLine(struct linenoiseState *l) { + refreshLineWithFlags(l,REFRESH_ALL); +} + +/* Hide the current line, when using the multiplexing API. */ +void linenoiseHide(struct linenoiseState *l) { + if (mlmode) + refreshMultiLine(l,REFRESH_CLEAN); + else + refreshSingleLine(l,REFRESH_CLEAN); +} + +/* Show the current line, when using the multiplexing API. */ +void linenoiseShow(struct linenoiseState *l) { + if (l->in_completion) { + refreshLineWithCompletion(l,NULL,REFRESH_WRITE); + } else { + refreshLineWithFlags(l,REFRESH_WRITE); + } +} + +/* Insert the character 'c' at cursor current position. + * + * On error writing to the terminal -1 is returned, otherwise 0. */ +static int linenoiseEditInsert(struct linenoiseState * l, char c) { + if (l->len < l->buflen) { + if (l->len == l->pos) { + l->buf[l->pos] = c; + l->pos++; + l->len++; + l->buf[l->len] = '\0'; + if ((!mlmode && l->plen+l->len < l->cols && !hintsCallback)) { + /* Avoid a full update of the line in the + * trivial case. */ + char d = (maskmode==1) ? '*' : c; + if (write(l->ofd,&d,1) == -1) return -1; + } else { + refreshLine(l); + } + } else { + memmove(l->buf+l->pos+1,l->buf+l->pos,l->len-l->pos); + l->buf[l->pos] = c; + l->len++; + l->pos++; + l->buf[l->len] = '\0'; + refreshLine(l); + } + } + return 0; +} + +/* Move cursor on the left. */ +static void linenoiseEditMoveLeft(struct linenoiseState * l) { + if (l->pos > 0) { + l->pos--; + refreshLine(l); + } +} + +/* Move cursor on the right. */ +static void linenoiseEditMoveRight(struct linenoiseState * l) { + if (l->pos != l->len) { + l->pos++; + refreshLine(l); + } +} + +/* Move cursor to the start of the line. */ +static void linenoiseEditMoveHome(struct linenoiseState * l) { + if (l->pos != 0) { + l->pos = 0; + refreshLine(l); + } +} + +/* Move cursor to the end of the line. */ +static void linenoiseEditMoveEnd(struct linenoiseState * l) { + if (l->pos != l->len) { + l->pos = l->len; + refreshLine(l); + } +} + +/* Substitute the currently edited line with the next or previous history + * entry as specified by 'dir'. */ +#define LINENOISE_HISTORY_NEXT 0 +#define LINENOISE_HISTORY_PREV 1 + +static void linenoiseEditHistoryNext(struct linenoiseState * l, int dir) { + if (history_len > 1) { + /* Update the current history entry before to + * overwrite it with the next one. */ + free(history[history_len - 1 - l->history_index]); + history[history_len - 1 - l->history_index] = strdup(l->buf); + /* Show the new entry */ + l->history_index += (dir == LINENOISE_HISTORY_PREV) ? 1 : -1; + if (l->history_index < 0) { + l->history_index = 0; + return; + } else if (l->history_index >= history_len) { + l->history_index = history_len-1; + return; + } + strncpy(l->buf,history[history_len - 1 - l->history_index],l->buflen); + l->buf[l->buflen-1] = '\0'; + l->len = l->pos = strlen(l->buf); + refreshLine(l); + } +} + +/* Delete the character at the right of the cursor without altering the cursor + * position. Basically this is what happens with the "Delete" keyboard key. */ +static void linenoiseEditDelete(struct linenoiseState * l) { + if (l->len > 0 && l->pos < l->len) { + memmove(l->buf+l->pos,l->buf+l->pos+1,l->len-l->pos-1); + l->len--; + l->buf[l->len] = '\0'; + refreshLine(l); + } +} + +/* Backspace implementation. */ +static void linenoiseEditBackspace(struct linenoiseState * l) { + if (l->pos > 0 && l->len > 0) { + memmove(l->buf+l->pos-1,l->buf+l->pos,l->len-l->pos); + l->pos--; + l->len--; + l->buf[l->len] = '\0'; + refreshLine(l); + } +} + +/* Delete the previosu word, maintaining the cursor at the start of the + * current word. */ +static void linenoiseEditDeletePrevWord(struct linenoiseState * l) { + size_t old_pos = l->pos; + size_t diff; + + while (l->pos > 0 && l->buf[l->pos-1] == ' ') + l->pos--; + while (l->pos > 0 && l->buf[l->pos-1] != ' ') + l->pos--; + diff = old_pos - l->pos; + memmove(l->buf+l->pos,l->buf+old_pos,l->len-old_pos+1); + l->len -= diff; + refreshLine(l); +} + +/* This function is part of the multiplexed API of Linenoise, that is used + * in order to implement the blocking variant of the API but can also be + * called by the user directly in an event driven program. It will: + * + * 1. Initialize the linenoise state passed by the user. + * 2. Put the terminal in RAW mode. + * 3. Show the prompt. + * 4. Return control to the user, that will have to call linenoiseEditFeed() + * each time there is some data arriving in the standard input. + * + * The user can also call linenoiseEditHide() and linenoiseEditShow() if it + * is required to show some input arriving asyncronously, without mixing + * it with the currently edited line. + * + * When linenoiseEditFeed() returns non-NULL, the user finished with the + * line editing session (pressed enter CTRL-D/C): in this case the caller + * needs to call linenoiseEditStop() to put back the terminal in normal + * mode. This will not destroy the buffer, as long as the linenoiseState + * is still valid in the context of the caller. + * + * The function returns 0 on success, or -1 if writing to standard output + * fails. If stdin_fd or stdout_fd are set to -1, the default is to use + * STDIN_FILENO and STDOUT_FILENO. + */ +int linenoiseEditStart(struct linenoiseState *l, int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt) { + /* Populate the linenoise state that we pass to functions implementing + * specific editing functionalities. */ + l->in_completion = 0; + l->ifd = stdin_fd != -1 ? stdin_fd : STDIN_FILENO; + l->ofd = stdout_fd != -1 ? stdout_fd : STDOUT_FILENO; + l->buf = buf; + l->buflen = buflen; + l->prompt = prompt; + l->plen = strlen(prompt); + l->oldpos = l->pos = 0; + l->len = 0; + + /* Enter raw mode. */ + if (enableRawMode(l->ifd) == -1) return -1; + + l->cols = getColumns(stdin_fd, stdout_fd); + l->oldrows = 0; + l->history_index = 0; + + /* Buffer starts empty. */ + l->buf[0] = '\0'; + l->buflen--; /* Make sure there is always space for the nulterm */ + + /* If stdin is not a tty, stop here with the initialization. We + * will actually just read a line from standard input in blocking + * mode later, in linenoiseEditFeed(). */ + if (!isatty(l->ifd)) return 0; + + /* The latest history entry is always our current buffer, that + * initially is just an empty string. */ + linenoiseHistoryAdd(""); + + if (write(l->ofd,prompt,l->plen) == -1) return -1; + return 0; +} + +const char* linenoiseEditMore = "If you see this, you are misusing the API: when linenoiseEditFeed() is called, if it returns linenoiseEditMore the user is yet editing the line. See the README file for more information."; + +/* This function is part of the multiplexed API of linenoise, see the top + * comment on linenoiseEditStart() for more information. Call this function + * each time there is some data to read from the standard input file + * descriptor. In the case of blocking operations, this function can just be + * called in a loop, and block. + * + * The function returns linenoiseEditMore to signal that line editing is still + * in progress, that is, the user didn't yet pressed enter / CTRL-D. Otherwise + * the function returns the pointer to the heap-allocated buffer with the + * edited line, that the user should free with linenoiseFree(). + * + * On special conditions, NULL is returned and errno is populated: + * + * EAGAIN if the user pressed Ctrl-C + * ENOENT if the user pressed Ctrl-D + * + * Some other errno: I/O error. + */ +const char *linenoiseEditFeed(struct linenoiseState *l) { + /* Not a TTY, pass control to line reading without character + * count limits. */ + if (!isatty(l->ifd)) return linenoiseNoTTY(); + + char c; + int nread; + char seq[3]; + + nread = read(l->ifd,&c,1); + if (nread <= 0) return NULL; + + /* Only autocomplete when the callback is set. It returns < 0 when + * there was an error reading from fd. Otherwise it will return the + * character that should be handled next. */ + if ((l->in_completion || c == 9) && completionCallback != NULL) { + c = completeLine(l,c); + /* Read next character when 0 */ + if (c == 0) return linenoiseEditMore; + } + + switch(c) { + case ENTER: /* enter */ + history_len--; + free(history[history_len]); + if (mlmode) linenoiseEditMoveEnd(l); + if (hintsCallback) { + /* Force a refresh without hints to leave the previous + * line as the user typed it after a newline. */ + linenoiseHintsCallback *hc = hintsCallback; + hintsCallback = NULL; + refreshLine(l); + hintsCallback = hc; + } + return strdup(l->buf); + case CTRL_C: /* ctrl-c */ + errno = EAGAIN; + return NULL; + case BACKSPACE: /* backspace */ + case 8: /* ctrl-h */ + linenoiseEditBackspace(l); + break; + case CTRL_D: /* ctrl-d, remove char at right of cursor, or if the + line is empty, act as end-of-file. */ + if (l->len > 0) { + linenoiseEditDelete(l); + } else { + history_len--; + free(history[history_len]); + errno = ENOENT; + return NULL; + } + break; + case CTRL_T: /* ctrl-t, swaps current character with previous. */ + if (l->pos > 0 && l->pos < l->len) { + int aux = l->buf[l->pos-1]; + l->buf[l->pos-1] = l->buf[l->pos]; + l->buf[l->pos] = aux; + if (l->pos != l->len-1) l->pos++; + refreshLine(l); + } + break; + case CTRL_B: /* ctrl-b */ + linenoiseEditMoveLeft(l); + break; + case CTRL_F: /* ctrl-f */ + linenoiseEditMoveRight(l); + break; + case CTRL_P: /* ctrl-p */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_PREV); + break; + case CTRL_N: /* ctrl-n */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_NEXT); + break; + case ESC: /* escape sequence */ + /* Read the next two bytes representing the escape sequence. + * Use two calls to handle slow terminals returning the two + * chars at different times. */ + if (read(l->ifd,seq,1) == -1) break; + if (read(l->ifd,seq+1,1) == -1) break; + + /* ESC [ sequences. */ + if (seq[0] == '[') { + if (seq[1] >= '0' && seq[1] <= '9') { + /* Extended escape, read additional byte. */ + if (read(l->ifd,seq+2,1) == -1) break; + if (seq[2] == '~') { + switch(seq[1]) { + case '3': /* Delete key. */ + linenoiseEditDelete(l); + break; + } + } + } else { + switch(seq[1]) { + case 'A': /* Up */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_PREV); + break; + case 'B': /* Down */ + linenoiseEditHistoryNext(l, LINENOISE_HISTORY_NEXT); + break; + case 'C': /* Right */ + linenoiseEditMoveRight(l); + break; + case 'D': /* Left */ + linenoiseEditMoveLeft(l); + break; + case 'H': /* Home */ + linenoiseEditMoveHome(l); + break; + case 'F': /* End*/ + linenoiseEditMoveEnd(l); + break; + } + } + } + + /* ESC O sequences. */ + else if (seq[0] == 'O') { + switch(seq[1]) { + case 'H': /* Home */ + linenoiseEditMoveHome(l); + break; + case 'F': /* End*/ + linenoiseEditMoveEnd(l); + break; + } + } + break; + default: + if (linenoiseEditInsert(l,c)) return NULL; + break; + case CTRL_U: /* Ctrl+u, delete the whole line. */ + l->buf[0] = '\0'; + l->pos = l->len = 0; + refreshLine(l); + break; + case CTRL_K: /* Ctrl+k, delete from current to end of line. */ + l->buf[l->pos] = '\0'; + l->len = l->pos; + refreshLine(l); + break; + case CTRL_A: /* Ctrl+a, go to the start of the line */ + linenoiseEditMoveHome(l); + break; + case CTRL_E: /* ctrl+e, go to the end of the line */ + linenoiseEditMoveEnd(l); + break; + case CTRL_L: /* ctrl+l, clear screen */ + linenoiseClearScreen(); + refreshLine(l); + break; + case CTRL_W: /* ctrl+w, delete previous word */ + linenoiseEditDeletePrevWord(l); + break; + } + return linenoiseEditMore; +} + +/* This is part of the multiplexed linenoise API. See linenoiseEditStart() + * for more information. This function is called when linenoiseEditFeed() + * returns something different than NULL. At this point the user input + * is in the buffer, and we can restore the terminal in normal mode. */ +void linenoiseEditStop(struct linenoiseState *l) { + if (!isatty(l->ifd)) return; + disableRawMode(l->ifd); + printf("\n"); +} + +/* This just implements a blocking loop for the multiplexed API. + * In many applications that are not event-drivern, we can just call + * the blocking linenoise API, wait for the user to complete the editing + * and return the buffer. */ +static const char *linenoiseBlockingEdit(int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt) +{ + struct linenoiseState l; + + /* Editing without a buffer is invalid. */ + if (buflen == 0) { + errno = EINVAL; + return NULL; + } + + linenoiseEditStart(&l,stdin_fd,stdout_fd,buf,buflen,prompt); + const char *res; + while((res = linenoiseEditFeed(&l)) == linenoiseEditMore); + linenoiseEditStop(&l); + return res; +} + +/* This special mode is used by linenoise in order to print scan codes + * on screen for debugging / development purposes. It is implemented + * by the linenoise_example program using the --keycodes option. */ +void linenoisePrintKeyCodes(void) { + char quit[4]; + + printf("Linenoise key codes debugging mode.\n" + "Press keys to see scan codes. Type 'quit' at any time to exit.\n"); + if (enableRawMode(STDIN_FILENO) == -1) return; + memset(quit,' ',4); + while(1) { + char c; + int nread; + + nread = read(STDIN_FILENO,&c,1); + if (nread <= 0) continue; + memmove(quit,quit+1,sizeof(quit)-1); /* shift string to left. */ + quit[sizeof(quit)-1] = c; /* Insert current char on the right. */ + if (memcmp(quit,"quit",sizeof(quit)) == 0) break; + + printf("'%c' %02x (%d) (type quit to exit)\n", + isprint(c) ? c : '?', (int)c, (int)c); + printf("\r"); /* Go left edge manually, we are in raw mode. */ + fflush(stdout); + } + disableRawMode(STDIN_FILENO); +} + +/* This function is called when linenoise() is called with the standard + * input file descriptor not attached to a TTY. So for example when the + * program using linenoise is called in pipe or with a file redirected + * to its standard input. In this case, we want to be able to return the + * line regardless of its length (by default we are limited to 4k). */ +static char *linenoiseNoTTY(void) { + char *line = NULL; + size_t len = 0, maxlen = 0; + + while(1) { + if (len == maxlen) { + if (maxlen == 0) maxlen = 16; + maxlen *= 2; + char *oldval = line; + line = (char*) realloc(line,maxlen); + if (line == NULL) { + if (oldval) free(oldval); + return NULL; + } + } + int c = fgetc(stdin); + if (c == EOF || c == '\n') { + if (c == EOF && len == 0) { + free(line); + return NULL; + } else { + line[len] = '\0'; + return line; + } + } else { + line[len] = c; + len++; + } + } +} + +/* The high level function that is the main API of the linenoise library. + * This function checks if the terminal has basic capabilities, just checking + * for a blacklist of stupid terminals, and later either calls the line + * editing function or uses dummy fgets() so that you will be able to type + * something even in the most desperate of the conditions. */ +const char *linenoise(const char *prompt) { + char buf[LINENOISE_MAX_LINE]; + + if (!isatty(STDIN_FILENO)) { + /* Not a tty: read from file / pipe. In this mode we don't want any + * limit to the line size, so we call a function to handle that. */ + return linenoiseNoTTY(); + } else if (isUnsupportedTerm()) { + size_t len; + + printf("%s",prompt); + fflush(stdout); + if (fgets(buf,LINENOISE_MAX_LINE,stdin) == NULL) return NULL; + len = strlen(buf); + while(len && (buf[len-1] == '\n' || buf[len-1] == '\r')) { + len--; + buf[len] = '\0'; + } + return strdup(buf); + } else { + const char *retval = linenoiseBlockingEdit(STDIN_FILENO,STDOUT_FILENO,buf,LINENOISE_MAX_LINE,prompt); + return retval; + } +} + +/* This is just a wrapper the user may want to call in order to make sure + * the linenoise returned buffer is freed with the same allocator it was + * created with. Useful when the main program is using an alternative + * allocator. */ +void linenoiseFree(void *ptr) { + if (ptr == linenoiseEditMore) return; // Protect from API misuse. + free(ptr); +} + +/* ================================ History ================================= */ + +/* Free the history, but does not reset it. Only used when we have to + * exit() to avoid memory leaks are reported by valgrind & co. */ +static void freeHistory(void) { + if (history) { + int j; + + for (j = 0; j < history_len; j++) + free(history[j]); + free(history); + } +} + +/* At exit we'll try to fix the terminal to the initial conditions. */ +static void linenoiseAtExit(void) { + disableRawMode(STDIN_FILENO); + freeHistory(); +} + +/* This is the API call to add a new entry in the linenoise history. + * It uses a fixed array of char pointers that are shifted (memmoved) + * when the history max length is reached in order to remove the older + * entry and make room for the new one, so it is not exactly suitable for huge + * histories, but will work well for a few hundred of entries. + * + * Using a circular buffer is smarter, but a bit more complex to handle. */ +int linenoiseHistoryAdd(const char *line) { + char *linecopy; + + if (history_max_len == 0) return 0; + + /* Initialization on first call. */ + if (history == NULL) { + history = (char**) malloc(sizeof(char*)*history_max_len); + if (history == NULL) return 0; + memset(history,0,(sizeof(char*)*history_max_len)); + } + + /* Don't add duplicated lines. */ + if (history_len && !strcmp(history[history_len-1], line)) return 0; + + /* Add an heap allocated copy of the line in the history. + * If we reached the max length, remove the older line. */ + linecopy = strdup(line); + if (!linecopy) return 0; + if (history_len == history_max_len) { + free(history[0]); + memmove(history,history+1,sizeof(char*)*(history_max_len-1)); + history_len--; + } + history[history_len] = linecopy; + history_len++; + return 1; +} + +/* Set the maximum length for the history. This function can be called even + * if there is already some history, the function will make sure to retain + * just the latest 'len' elements if the new history length value is smaller + * than the amount of items already inside the history. */ +int linenoiseHistorySetMaxLen(int len) { + char **new_ptr; + + if (len < 1) return 0; + if (history) { + int tocopy = history_len; + + new_ptr = (char**) malloc(sizeof(char*)*len); + if (new_ptr == NULL) return 0; + + /* If we can't copy everything, free the elements we'll not use. */ + if (len < tocopy) { + int j; + + for (j = 0; j < tocopy-len; j++) free(history[j]); + tocopy = len; + } + memset(new_ptr,0,sizeof(char*)*len); + memcpy(new_ptr,history+(history_len-tocopy), sizeof(char*)*tocopy); + free(history); + history = new_ptr; + } + history_max_len = len; + if (history_len > history_max_len) + history_len = history_max_len; + return 1; +} + +/* Save the history in the specified file. On success 0 is returned + * otherwise -1 is returned. */ +int linenoiseHistorySave(const char *filename) { + mode_t old_umask = umask(S_IXUSR|S_IRWXG|S_IRWXO); + FILE *fp; + int j; + + fp = fopen(filename,"w"); + umask(old_umask); + if (fp == NULL) return -1; + chmod(filename,S_IRUSR|S_IWUSR); + for (j = 0; j < history_len; j++) + fprintf(fp,"%s\n",history[j]); + fclose(fp); + return 0; +} + +/* Load the history from the specified file. If the file does not exist + * zero is returned and no operation is performed. + * + * If the file exists and the operation succeeded 0 is returned, otherwise + * on error -1 is returned. */ +int linenoiseHistoryLoad(const char *filename) { + FILE *fp = fopen(filename,"r"); + char buf[LINENOISE_MAX_LINE]; + + if (fp == NULL) return -1; + + while (fgets(buf,LINENOISE_MAX_LINE,fp) != NULL) { + char *p; + + p = strchr(buf,'\r'); + if (!p) p = strchr(buf,'\n'); + if (p) *p = '\0'; + linenoiseHistoryAdd(buf); + } + fclose(fp); + return 0; +} +#endif diff --git a/examples/run/linenoise.cpp/linenoise.h b/examples/run/linenoise.cpp/linenoise.h new file mode 100644 index 000000000..3e25f4de3 --- /dev/null +++ b/examples/run/linenoise.cpp/linenoise.h @@ -0,0 +1,114 @@ +/* linenoise.h -- VERSION 1.0 + * + * Guerrilla line editing library against the idea that a line editing lib + * needs to be 20,000 lines of C++ code. + * + * See linenoise.cpp for more information. + * + * ------------------------------------------------------------------------ + * + * Copyright (c) 2010-2023, Salvatore Sanfilippo + * Copyright (c) 2010-2013, Pieter Noordhuis + * Copyright (c) 2025, Eric Curtin + * + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are + * met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __LINENOISE_H +#define __LINENOISE_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include /* For size_t. */ + +extern const char *linenoiseEditMore; + +/* The linenoiseState structure represents the state during line editing. + * We pass this state to functions implementing specific editing + * functionalities. */ +struct linenoiseState { + int in_completion; /* The user pressed TAB and we are now in completion + * mode, so input is handled by completeLine(). */ + size_t completion_idx; /* Index of next completion to propose. */ + int ifd; /* Terminal stdin file descriptor. */ + int ofd; /* Terminal stdout file descriptor. */ + char *buf; /* Edited line buffer. */ + size_t buflen; /* Edited line buffer size. */ + const char *prompt; /* Prompt to display. */ + size_t plen; /* Prompt length. */ + size_t pos; /* Current cursor position. */ + size_t oldpos; /* Previous refresh cursor position. */ + size_t len; /* Current edited line length. */ + size_t cols; /* Number of columns in terminal. */ + size_t oldrows; /* Rows used by last refrehsed line (multiline mode) */ + int history_index; /* The history index we are currently editing. */ +}; + +typedef struct linenoiseCompletions { + size_t len; + char **cvec; +} linenoiseCompletions; + +/* Non blocking API. */ +int linenoiseEditStart(struct linenoiseState *l, int stdin_fd, int stdout_fd, char *buf, size_t buflen, const char *prompt); +const char *linenoiseEditFeed(struct linenoiseState *l); +void linenoiseEditStop(struct linenoiseState *l); +void linenoiseHide(struct linenoiseState *l); +void linenoiseShow(struct linenoiseState *l); + +/* Blocking API. */ +const char *linenoise(const char *prompt); +void linenoiseFree(void *ptr); + +/* Completion API. */ +typedef void(linenoiseCompletionCallback)(const char *, linenoiseCompletions *); +typedef const char*(linenoiseHintsCallback)(const char *, int *color, int *bold); +typedef void(linenoiseFreeHintsCallback)(const char *); +void linenoiseSetCompletionCallback(linenoiseCompletionCallback *); +void linenoiseSetHintsCallback(linenoiseHintsCallback *); +void linenoiseSetFreeHintsCallback(linenoiseFreeHintsCallback *); +void linenoiseAddCompletion(linenoiseCompletions *, const char *); + +/* History API. */ +int linenoiseHistoryAdd(const char *line); +int linenoiseHistorySetMaxLen(int len); +int linenoiseHistorySave(const char *filename); +int linenoiseHistoryLoad(const char *filename); + +/* Other utilities. */ +void linenoiseClearScreen(void); +void linenoiseSetMultiLine(int ml); +void linenoisePrintKeyCodes(void); +void linenoiseMaskModeEnable(void); +void linenoiseMaskModeDisable(void); + +#ifdef __cplusplus +} +#endif + +#endif /* __LINENOISE_H */ diff --git a/examples/run/run.cpp b/examples/run/run.cpp index 0ad8bb15b..dd9ea79e8 100644 --- a/examples/run/run.cpp +++ b/examples/run/run.cpp @@ -19,12 +19,14 @@ #include #include #include +#include #include #include #include #include "common.h" #include "json.hpp" +#include "linenoise.cpp/linenoise.h" #include "llama-cpp.h" #if defined(__unix__) || (defined(__APPLE__) && defined(__MACH__)) || defined(_WIN32) @@ -536,7 +538,7 @@ class LlamaData { llama_sampler_ptr sampler; llama_context_ptr context; std::vector messages; - std::vector msg_strs; + std::list msg_strs; std::vector fmtted; int init(Opt & opt) { @@ -807,24 +809,44 @@ static int generate(LlamaData & llama_data, const std::string & prompt, std::str batch = llama_batch_get_one(&new_token_id, 1); } + printf("\033[0m"); return 0; } -static int read_user_input(std::string & user) { - std::getline(std::cin, user); +static int read_user_input(std::string & user_input) { + static const char * prompt_prefix = "> "; +#ifdef WIN32 + printf( + "\r%*s" + "\r\033[0m%s", + get_terminal_width(), " ", prompt_prefix); + + std::getline(std::cin, user_input); if (std::cin.eof()) { printf("\n"); return 1; } - - if (user == "/bye") { +#else + std::unique_ptr line(const_cast(linenoise(prompt_prefix)), free); + if (!line) { return 1; } - if (user.empty()) { + user_input = line.get(); +#endif + + if (user_input == "/bye") { + return 1; + } + + if (user_input.empty()) { return 2; } +#ifndef WIN32 + linenoiseHistoryAdd(line.get()); +#endif + return 0; // Should have data in happy path } @@ -865,10 +887,6 @@ static int handle_user_input(std::string & user_input, const std::string & user) return 0; // No need for interactive input } - printf( - "\r%*s" - "\r\033[32m> \033[0m", - get_terminal_width(), " "); return read_user_input(user_input); // Returns true if input ends the loop } From 99487b57d47e14dc342b7b89d238ca11c0345241 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Nicol=C3=B2=20Scipione?= Date: Sun, 19 Jan 2025 14:33:34 +0100 Subject: [PATCH 07/26] SYCL: Introducing memory host pool (#11251) * Implement host pool for matrix_info Creating a new memory pool on the host to store memory location for matrix_info needed to launch gemm_batch from oneMKL/oneMath. Removing complex support in gemm_batch since it is not used in llama.cpp * Remove unnecessary headers and cast * Reorder member variable to avoid warning on initialization * Formatting * Remove unused variable * Address PR review feedback - remove warning --------- Signed-off-by: nscipione --- ggml/src/ggml-sycl/common.hpp | 13 +++ ggml/src/ggml-sycl/dpct/helper.hpp | 135 +++++++++-------------------- ggml/src/ggml-sycl/ggml-sycl.cpp | 92 ++++++++++++++++++-- 3 files changed, 137 insertions(+), 103 deletions(-) diff --git a/ggml/src/ggml-sycl/common.hpp b/ggml/src/ggml-sycl/common.hpp index e9500f3a1..abad847ca 100644 --- a/ggml/src/ggml-sycl/common.hpp +++ b/ggml/src/ggml-sycl/common.hpp @@ -333,8 +333,12 @@ struct ggml_backend_sycl_context { // pool std::unique_ptr pools[GGML_SYCL_MAX_DEVICES]; + std::unique_ptr host_pools[GGML_SYCL_MAX_DEVICES]; + static std::unique_ptr new_pool_for_device(queue_ptr qptr, int device); + static std::unique_ptr new_pool_for_host(queue_ptr qptr, int device); + ggml_sycl_pool & pool(int device) { if (pools[device] == nullptr) { pools[device] = new_pool_for_device(stream(device,0), device); @@ -345,6 +349,15 @@ struct ggml_backend_sycl_context { ggml_sycl_pool & pool() { return pool(device); } + + ggml_sycl_pool & host_pool(int device) { + if (host_pools[device] == nullptr) { + host_pools[device] = new_pool_for_host(stream(device, 0), device); + } + return *host_pools[device]; + } + + ggml_sycl_pool & host_pool() { return host_pool(device); } }; // common device functions diff --git a/ggml/src/ggml-sycl/dpct/helper.hpp b/ggml/src/ggml-sycl/dpct/helper.hpp index e167948e7..c96395be6 100644 --- a/ggml/src/ggml-sycl/dpct/helper.hpp +++ b/ggml/src/ggml-sycl/dpct/helper.hpp @@ -82,6 +82,14 @@ inline std::string get_device_backend_and_type(const sycl::device &device) { return device_type.str(); } +template struct matrix_info_t { + oneapi::mkl::transpose transpose_info[2]; + Ts value_info[2]; + std::int64_t size_info[3]; + std::int64_t ld_info[3]; + std::int64_t groupsize_info; +}; + namespace dpct { typedef sycl::queue *queue_ptr; @@ -1727,26 +1735,13 @@ namespace dpct }; template - inline void gemm_batch_impl(sycl::queue &q, oneapi::mkl::transpose a_trans, - oneapi::mkl::transpose b_trans, int m, int n, int k, - const void *alpha, const void **a, int lda, - const void **b, int ldb, const void *beta, void **c, - int ldc, int batch_size) - { - struct matrix_info_t - { - oneapi::mkl::transpose transpose_info[2]; - Ts value_info[2]; - std::int64_t size_info[3]; - std::int64_t ld_info[3]; - std::int64_t groupsize_info; - }; - + inline void gemm_batch_impl(sycl::queue & q, oneapi::mkl::transpose a_trans, oneapi::mkl::transpose b_trans, + int m, int n, int k, const void * alpha, const void ** a, int lda, const void ** b, + int ldb, const void * beta, void ** c, int ldc, int batch_size, + matrix_info_t * matrix_info) { Ts alpha_value = dpct::get_value(reinterpret_cast(alpha), q); Ts beta_value = dpct::get_value(reinterpret_cast(beta), q); - matrix_info_t *matrix_info = - (matrix_info_t *)std::malloc(sizeof(matrix_info_t)); matrix_info->transpose_info[0] = a_trans; matrix_info->transpose_info[1] = b_trans; matrix_info->value_info[0] = alpha_value; @@ -1763,23 +1758,18 @@ namespace dpct sycl::event e = oneapi::mkl::blas::column_major::gemm_batch( oneapi::mkl::backend_selector{ q }, matrix_info->transpose_info, matrix_info->transpose_info + 1, matrix_info->size_info, matrix_info->size_info + 1, - matrix_info->size_info + 2, matrix_info->value_info, reinterpret_cast(a), - matrix_info->ld_info, reinterpret_cast(b), matrix_info->ld_info + 1, - matrix_info->value_info + 1, reinterpret_cast(c), matrix_info->ld_info + 2, 1, - &(matrix_info->groupsize_info)); + matrix_info->size_info + 2, reinterpret_cast(matrix_info->value_info), + reinterpret_cast(a), matrix_info->ld_info, reinterpret_cast(b), + matrix_info->ld_info + 1, reinterpret_cast(matrix_info->value_info + 1), + reinterpret_cast(c), matrix_info->ld_info + 2, 1, &(matrix_info->groupsize_info)); #else sycl::event e = oneapi::mkl::blas::column_major::gemm_batch( q, matrix_info->transpose_info, matrix_info->transpose_info + 1, matrix_info->size_info, - matrix_info->size_info + 1, matrix_info->size_info + 2, matrix_info->value_info, + matrix_info->size_info + 1, matrix_info->size_info + 2, reinterpret_cast(matrix_info->value_info), reinterpret_cast(a), matrix_info->ld_info, reinterpret_cast(b), - matrix_info->ld_info + 1, matrix_info->value_info + 1, reinterpret_cast(c), - matrix_info->ld_info + 2, 1, &(matrix_info->groupsize_info)); + matrix_info->ld_info + 1, reinterpret_cast(matrix_info->value_info + 1), + reinterpret_cast(c), matrix_info->ld_info + 2, 1, &(matrix_info->groupsize_info)); #endif - - q.submit([&](sycl::handler &cgh) - { - cgh.depends_on(e); - cgh.host_task([=] { std::free(matrix_info); }); }); } template @@ -2422,25 +2412,11 @@ namespace dpct /// \param [in] ldc Leading dimension of C. /// \param [in] batch_size Specifies the number of matrix multiply operations to perform. /// \param [in] scaling_type Data type of the scaling factors. - inline void gemm_batch(sycl::queue &q, oneapi::mkl::transpose a_trans, - oneapi::mkl::transpose b_trans, int m, int n, int k, - const void *alpha, const void *a[], - library_data_t a_type, int lda, const void *b[], - library_data_t b_type, int ldb, const void *beta, - void *c[], library_data_t c_type, int ldc, - int batch_size, library_data_t scaling_type) - { - if (scaling_type == library_data_t::real_float && - c_type == library_data_t::complex_float) - { - scaling_type = library_data_t::complex_float; - } - else if (scaling_type == library_data_t::real_double && - c_type == library_data_t::complex_double) - { - scaling_type = library_data_t::complex_double; - } - + inline void gemm_batch(sycl::queue & q, oneapi::mkl::transpose a_trans, oneapi::mkl::transpose b_trans, int m, + int n, int k, const void * alpha, const void * a[], library_data_t a_type, int lda, + const void * b[], library_data_t b_type, int ldb, const void * beta, void * c[], + library_data_t c_type, int ldc, int batch_size, library_data_t scaling_type, + matrix_info_t * matrix_info) { std::uint64_t key = detail::get_type_combination_id(a_type, b_type, c_type, scaling_type); switch (key) @@ -2449,48 +2425,24 @@ namespace dpct library_data_t::real_float, library_data_t::real_float, library_data_t::real_float, library_data_t::real_float): { - detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); + detail::gemm_batch_impl(q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, + beta, c, ldc, batch_size, matrix_info); break; } case detail::get_type_combination_id( library_data_t::real_double, library_data_t::real_double, library_data_t::real_double, library_data_t::real_double): { - detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); - break; - } - case detail::get_type_combination_id( - library_data_t::complex_float, library_data_t::complex_float, - library_data_t::complex_float, library_data_t::complex_float): - { - detail::gemm_batch_impl, std::complex, - std::complex, std::complex>( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); - break; - } - case detail::get_type_combination_id( - library_data_t::complex_double, library_data_t::complex_double, - library_data_t::complex_double, library_data_t::complex_double): - { - detail::gemm_batch_impl, std::complex, - std::complex, std::complex>( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); + detail::gemm_batch_impl(q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, + beta, c, ldc, batch_size, matrix_info); break; } case detail::get_type_combination_id( library_data_t::real_half, library_data_t::real_half, library_data_t::real_half, library_data_t::real_half): { - detail::gemm_batch_impl(q, a_trans, b_trans, m, n, k, alpha, - a, lda, b, ldb, beta, c, ldc, - batch_size); + detail::gemm_batch_impl( + q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, batch_size, matrix_info); break; } #ifdef __INTEL_MKL__ @@ -2498,19 +2450,16 @@ namespace dpct library_data_t::real_bfloat16, library_data_t::real_bfloat16, library_data_t::real_bfloat16, library_data_t::real_float): { - detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); + detail::gemm_batch_impl( + q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, batch_size, matrix_info); break; } case detail::get_type_combination_id( library_data_t::real_bfloat16, library_data_t::real_bfloat16, library_data_t::real_float, library_data_t::real_float): { - detail::gemm_batch_impl(q, a_trans, b_trans, m, n, k, alpha, a, lda, - b, ldb, beta, c, ldc, batch_size); + detail::gemm_batch_impl( + q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, batch_size, matrix_info); break; } #endif @@ -2522,10 +2471,9 @@ namespace dpct dpct::get_value(reinterpret_cast(alpha), q); float beta_float = dpct::get_value(reinterpret_cast(beta), q); - detail::gemm_batch_impl(q, a_trans, b_trans, m, n, k, &alpha_float, - a, lda, b, ldb, &beta_float, c, ldc, - batch_size); + detail::gemm_batch_impl( + q, a_trans, b_trans, m, n, k, &alpha_float, a, lda, b, ldb, &beta_float, c, ldc, batch_size, + matrix_info); break; } case detail::get_type_combination_id( @@ -2533,8 +2481,7 @@ namespace dpct library_data_t::real_float, library_data_t::real_float): { detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); + q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, batch_size, matrix_info); break; } case detail::get_type_combination_id( @@ -2542,8 +2489,7 @@ namespace dpct library_data_t::real_float, library_data_t::real_float): { detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, - batch_size); + q, a_trans, b_trans, m, n, k, alpha, a, lda, b, ldb, beta, c, ldc, batch_size, matrix_info); break; } case detail::get_type_combination_id( @@ -2557,8 +2503,7 @@ namespace dpct sycl::half alpha_half(alpha_value); sycl::half beta_half(beta_value); detail::gemm_batch_impl( - q, a_trans, b_trans, m, n, k, &alpha_half, a, lda, b, ldb, &beta_half, c, ldc, - batch_size); + q, a_trans, b_trans, m, n, k, &alpha_half, a, lda, b, ldb, &beta_half, c, ldc, batch_size, matrix_info); break; } default: diff --git a/ggml/src/ggml-sycl/ggml-sycl.cpp b/ggml/src/ggml-sycl/ggml-sycl.cpp index 5272ca454..ed4d8bb8b 100644 --- a/ggml/src/ggml-sycl/ggml-sycl.cpp +++ b/ggml/src/ggml-sycl/ggml-sycl.cpp @@ -1173,6 +1173,85 @@ struct ggml_sycl_pool_leg : public ggml_sycl_pool { } }; +struct ggml_sycl_pool_host : public ggml_sycl_pool { + queue_ptr qptr; + int device; + + inline static int counter{ 0 }; + + struct ggml_sycl_buffer { + void * ptr = nullptr; + size_t size = 0; + }; + + // Set arbitrarly to 64 + static constexpr int MAX_POOL_SIZE{ 64 }; + std::vector buffer_pool = std::vector(MAX_POOL_SIZE); + size_t pool_size = 0; + + explicit ggml_sycl_pool_host(queue_ptr qptr_, int device_) : qptr(qptr_), device(device_) {} + + ~ggml_sycl_pool_host() { + for (int i = 0; i < MAX_POOL_SIZE; ++i) { + ggml_sycl_buffer & b = buffer_pool[i]; + if (b.ptr != nullptr) { + SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(b.ptr, *qptr))); + b.ptr = nullptr; + pool_size -= b.size; + b.size = 0; + } + } + counter = 0; + } + + void * alloc(size_t size, size_t * actual_size) override { + if (counter == MAX_POOL_SIZE) { + ggml_sycl_buffer b = buffer_pool[0]; + void * ptr = b.ptr; + *actual_size = b.size; + counter = 1; + return ptr; + } + ggml_sycl_buffer & b = buffer_pool[counter]; + + if (b.ptr == nullptr) { + void * ptr; + + SYCL_CHECK(CHECK_TRY_ERROR(ptr = (void *) sycl::malloc_host(size, *qptr))); + if (!ptr) { + GGML_LOG_ERROR("%s: can't allocate %lu Bytes of memory on host\n", __func__, size); + return nullptr; + } + pool_size += size; + *actual_size = size; + counter = counter + 1; + return ptr; + } else { + ++counter; + b.size = size; + return b.ptr; + } + } + + void free(void * ptr, size_t size) override { + // if the pool is not completed add the pointer to it in place of the first nullptr found. + // Otherwise do nothing, pointers will be freed once the pool is deallocated. + for (int i = 0; i < MAX_POOL_SIZE; ++i) { + ggml_sycl_buffer & b = buffer_pool[i]; + if (b.ptr == nullptr) { + b.ptr = ptr; + b.size = size; + return; + } + } + } +}; + +std::unique_ptr ggml_backend_sycl_context::new_pool_for_host(queue_ptr qptr, int device) { + // return pool for the host to speed up memory management + return std::unique_ptr(new ggml_sycl_pool_host(qptr, device)); +} + std::unique_ptr ggml_backend_sycl_context::new_pool_for_device(queue_ptr qptr, int device) { // TBD: NO VMM support // if (ggml_sycl_info().devices[device].vmm) { @@ -3363,6 +3442,7 @@ static void ggml_sycl_mul_mat_batched_sycl(ggml_backend_sycl_context & ctx, ggml_sycl_pool_alloc ptrs_src(ctx.pool(), 2*ne23); ggml_sycl_pool_alloc< void *> ptrs_dst(ctx.pool(), 1*ne23); + ggml_sycl_pool_alloc> matrix_info(ctx.host_pool(), 1); sycl::range<3> block_dims(1, ne12, ne13); /* @@ -3391,14 +3471,10 @@ static void ggml_sycl_mul_mat_batched_sycl(ggml_backend_sycl_context & ctx, }); } SYCL_CHECK(CHECK_TRY_ERROR(dpct::gemm_batch( - *main_stream, oneapi::mkl::transpose::trans, - oneapi::mkl::transpose::nontrans, ne01, ne11, ne10, alpha, - (const void **)(ptrs_src.get() + 0 * ne23), - dpct::library_data_t::real_half, nb01 / nb00, - (const void **)(ptrs_src.get() + 1 * ne23), - dpct::library_data_t::real_half, nb11 / nb10, beta, - (void **)(ptrs_dst.get() + 0 * ne23), cu_data_type, ne01, ne23, - cu_compute_type))); + *main_stream, oneapi::mkl::transpose::trans, oneapi::mkl::transpose::nontrans, ne01, ne11, ne10, alpha, + (const void **) (ptrs_src.get() + 0 * ne23), dpct::library_data_t::real_half, nb01 / nb00, + (const void **) (ptrs_src.get() + 1 * ne23), dpct::library_data_t::real_half, nb11 / nb10, beta, + (void **) (ptrs_dst.get() + 0 * ne23), cu_data_type, ne01, ne23, cu_compute_type, matrix_info.get()))); } } catch (sycl::exception const &exc) { From b9daaffe02d6a77d85f0420bce5dfe0e00daeff6 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sun, 19 Jan 2025 18:12:09 +0200 Subject: [PATCH 08/26] simple-chat : fix BOS being added to each message (#11278) --- examples/simple-chat/simple-chat.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/examples/simple-chat/simple-chat.cpp b/examples/simple-chat/simple-chat.cpp index e8eda9c22..26422601d 100644 --- a/examples/simple-chat/simple-chat.cpp +++ b/examples/simple-chat/simple-chat.cpp @@ -95,11 +95,11 @@ int main(int argc, char ** argv) { llama_sampler_chain_add(smpl, llama_sampler_init_dist(LLAMA_DEFAULT_SEED)); // helper function to evaluate a prompt and generate a response - auto generate = [&](const std::string & prompt) { + auto generate = [&](const std::string & prompt, bool is_first) { std::string response; // tokenize the prompt - const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, true, true); + const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true); std::vector prompt_tokens(n_prompt_tokens); if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), llama_get_kv_cache_used_cells(ctx) == 0, true) < 0) { GGML_ABORT("failed to tokenize the prompt\n"); @@ -180,7 +180,7 @@ int main(int argc, char ** argv) { // generate a response printf("\033[33m"); - std::string response = generate(prompt); + std::string response = generate(prompt, prev_len == 0); printf("\n\033[0m"); // add the response to the messages From 92bc493917d43b83e592349e138b54c90b1c3ea7 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sun, 19 Jan 2025 20:22:30 +0200 Subject: [PATCH 09/26] tests : increase timeout when sanitizers are enabled (#11300) * tests : increase timeout when sanitizers are enabled * tests : add DEFAULT_HTTP_TIMEOUT --- .github/workflows/server.yml | 25 ++++++++++++++++++++++--- examples/server/tests/utils.py | 5 ++++- 2 files changed, 26 insertions(+), 4 deletions(-) diff --git a/.github/workflows/server.yml b/.github/workflows/server.yml index 671fe595c..ed1c357a5 100644 --- a/.github/workflows/server.yml +++ b/.github/workflows/server.yml @@ -112,9 +112,9 @@ jobs: -DGGML_OPENMP=OFF ; cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server - - name: Build - id: cmake_build - if: ${{ matrix.sanitizer != 'THREAD' }} + - name: Build (sanitizers) + id: cmake_build_sanitizers + if: ${{ matrix.sanitizer != '' && matrix.sanitizer != 'THREAD' }} run: | cmake -B build \ -DGGML_NATIVE=OFF \ @@ -124,12 +124,31 @@ jobs: -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ; cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server + - name: Build (sanitizers) + id: cmake_build + if: ${{ matrix.sanitizer == '' }} + run: | + cmake -B build \ + -DGGML_NATIVE=OFF \ + -DLLAMA_BUILD_SERVER=ON \ + -DLLAMA_CURL=ON \ + -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ; + cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server + - name: Tests id: server_integration_tests + if: ${{ matrix.sanitizer == '' }} run: | cd examples/server/tests ./tests.sh + - name: Tests (sanitizers) + id: server_integration_tests_sanitizers + if: ${{ matrix.sanitizer != '' }} + run: | + cd examples/server/tests + LLAMA_SANITIZE=1 ./tests.sh + - name: Slow tests id: server_integration_tests_slow if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }} diff --git a/examples/server/tests/utils.py b/examples/server/tests/utils.py index 73be4c92f..9d1a7a5b0 100644 --- a/examples/server/tests/utils.py +++ b/examples/server/tests/utils.py @@ -26,6 +26,9 @@ from re import RegexFlag import wget +DEFAULT_HTTP_TIMEOUT = 10 if "LLAMA_SANITIZE" not in os.environ else 30 + + class ServerResponse: headers: dict status_code: int @@ -88,7 +91,7 @@ class ServerProcess: if "PORT" in os.environ: self.server_port = int(os.environ["PORT"]) - def start(self, timeout_seconds: int = 10) -> None: + def start(self, timeout_seconds: int | None = DEFAULT_HTTP_TIMEOUT) -> None: if "LLAMA_SERVER_BIN_PATH" in os.environ: server_path = os.environ["LLAMA_SERVER_BIN_PATH"] elif os.name == "nt": From ae3c1db2f9c4beec0737c6a82d1f3791fd6fcdb2 Mon Sep 17 00:00:00 2001 From: Kyle Bruene Date: Mon, 20 Jan 2025 01:21:01 -0600 Subject: [PATCH 10/26] llama : re-add LLM_ARCH_PHIMOE (#11305) Phi 3.5 MoE was partially removed during a refactor. The code was originally in llama.cpp and should be in llama-model.cpp after the refactor. --- src/llama-model.cpp | 44 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 44 insertions(+) diff --git a/src/llama-model.cpp b/src/llama-model.cpp index c2d23a8d3..6dfcd5f59 100644 --- a/src/llama-model.cpp +++ b/src/llama-model.cpp @@ -2203,6 +2203,50 @@ bool llama_model::load_tensors(llama_model_loader & ml) { layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0)); } } break; + case LLM_ARCH_PHIMOE: + { + const int64_t n_embd_head = n_embd / n_head; + + tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0); + + // output + output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0); + output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); + output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab }, 0); + output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), { n_vocab }, 0); + + for (int i = 0; i < n_layer; ++i) { + auto & layer = layers[i]; + + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0); + layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), { n_embd }, 0); + + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, llama_model_loader::TENSOR_NOT_REQUIRED); + if (layer.wqkv == nullptr) { + layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); + layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); + + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); + + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); + layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); + } + layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd, n_embd }, 0); + layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), { n_embd }, 0); + + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), { n_embd }, 0); + layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), { n_embd }, 0); + + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); + layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff, n_embd, n_expert}, 0); + layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); + + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), { n_embd_head/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0)); + layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0)); + } + } break; case LLM_ARCH_PLAMO: { tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0); From ef6dada60ca710f4edfbb6fd9e1258685d8ea49d Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Mon, 20 Jan 2025 09:29:32 +0200 Subject: [PATCH 11/26] cont : fix whitespaces (#11305) --- src/llama-model.cpp | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/src/llama-model.cpp b/src/llama-model.cpp index 6dfcd5f59..590386e64 100644 --- a/src/llama-model.cpp +++ b/src/llama-model.cpp @@ -2206,43 +2206,43 @@ bool llama_model::load_tensors(llama_model_loader & ml) { case LLM_ARCH_PHIMOE: { const int64_t n_embd_head = n_embd / n_head; - + tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, 0); - + // output output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), { n_embd }, 0); output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, 0); output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), { n_embd, n_vocab }, 0); output_b = create_tensor(tn(LLM_TENSOR_OUTPUT, "bias"), { n_vocab }, 0); - + for (int i = 0; i < n_layer; ++i) { auto & layer = layers[i]; - + layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0); layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), { n_embd }, 0); - + layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, llama_model_loader::TENSOR_NOT_REQUIRED); if (layer.wqkv == nullptr) { layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0); layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0); - + layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_gqa}, 0); layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, 0); - + layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_gqa}, 0); layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, 0); } layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd, n_embd }, 0); layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), { n_embd }, 0); - + layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), { n_embd }, 0); layer.ffn_norm_b = create_tensor(tn(LLM_TENSOR_FFN_NORM, "bias", i), { n_embd }, 0); - + layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0); layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff, n_embd, n_expert}, 0); layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), {n_embd, n_ff, n_expert}, 0); - + layer.rope_long = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG, "weight", i), { n_embd_head/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0)); layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), { n_embd_head/2 }, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0)); } From ec7f3ac9ab33e46b136eb5ab6a76c4d81f57c7f1 Mon Sep 17 00:00:00 2001 From: Xuan Son Nguyen Date: Mon, 20 Jan 2025 14:35:07 +0100 Subject: [PATCH 12/26] llama : add support for Deepseek-R1-Qwen distill model (#11310) * llama : add support for Deepseek-R1-Qwen distill model * coding style --- convert_hf_to_gguf.py | 3 + convert_hf_to_gguf_update.py | 87 +++++++-------- models/ggml-vocab-deepseek-r1-qwen.gguf.inp | 112 ++++++++++++++++++++ models/ggml-vocab-deepseek-r1-qwen.gguf.out | 46 ++++++++ src/llama-chat.cpp | 2 +- src/llama-vocab.cpp | 3 +- 6 files changed, 208 insertions(+), 45 deletions(-) create mode 100644 models/ggml-vocab-deepseek-r1-qwen.gguf.inp create mode 100644 models/ggml-vocab-deepseek-r1-qwen.gguf.out diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py index 95f112043..63b54a9cf 100755 --- a/convert_hf_to_gguf.py +++ b/convert_hf_to_gguf.py @@ -696,6 +696,9 @@ class Model: if chkhsh == "877081d19cf6996e2c4ff0e1236341e9b7bde288f5311a56a937f0afbbb3aeb5": # ref: https://huggingface.co/deepseek-ai/DeepSeek-V3 res = "deepseek-v3" + if chkhsh == "b3f499bb4255f8ca19fccd664443283318f2fd2414d5e0b040fbdd0cc195d6c5": + # ref: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B + res = "deepseek-r1-qwen" if res is None: logger.warning("\n") diff --git a/convert_hf_to_gguf_update.py b/convert_hf_to_gguf_update.py index 56edc64a7..cea34413f 100755 --- a/convert_hf_to_gguf_update.py +++ b/convert_hf_to_gguf_update.py @@ -65,49 +65,50 @@ else: # TODO: add models here, base models preferred models = [ - {"name": "llama-spm", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", }, - {"name": "llama-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", }, - {"name": "phi-3", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", }, - {"name": "deepseek-llm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", }, - {"name": "deepseek-coder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", }, - {"name": "falcon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", }, - {"name": "bert-bge", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", }, - {"name": "falcon3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", }, - {"name": "bert-bge-large", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", }, - {"name": "mpt", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", }, - {"name": "starcoder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", }, - {"name": "gpt-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", }, - {"name": "stablelm2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", }, - {"name": "refact", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", }, - {"name": "command-r", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", }, - {"name": "qwen2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", }, - {"name": "olmo", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", }, - {"name": "dbrx", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", }, - {"name": "jina-v1-en", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", }, - {"name": "jina-v2-en", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM! - {"name": "jina-v2-es", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", }, - {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", }, - {"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", }, - {"name": "poro-chat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", }, - {"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", }, - {"name": "viking", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B - {"name": "gemma", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2b", }, - {"name": "gemma-2", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", }, - {"name": "jais", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", }, - {"name": "t5", "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", }, - {"name": "codeshell", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", }, - {"name": "tekken", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", }, - {"name": "smollm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", }, - {'name': "bloom", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", }, - {'name': "gpt3-finnish", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", }, - {"name": "exaone", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", }, - {"name": "phi-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", }, - {"name": "chameleon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", }, - {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", }, - {"name": "roberta-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"}, - {"name": "gigachat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"}, - {"name": "megrez", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"}, - {"name": "deepseek-v3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"}, + {"name": "llama-spm", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", }, + {"name": "llama-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", }, + {"name": "phi-3", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", }, + {"name": "deepseek-llm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", }, + {"name": "deepseek-coder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", }, + {"name": "falcon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", }, + {"name": "bert-bge", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", }, + {"name": "falcon3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", }, + {"name": "bert-bge-large", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", }, + {"name": "mpt", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", }, + {"name": "starcoder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", }, + {"name": "gpt-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", }, + {"name": "stablelm2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", }, + {"name": "refact", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", }, + {"name": "command-r", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", }, + {"name": "qwen2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", }, + {"name": "olmo", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", }, + {"name": "dbrx", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", }, + {"name": "jina-v1-en", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", }, + {"name": "jina-v2-en", "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM! + {"name": "jina-v2-es", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", }, + {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", }, + {"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", }, + {"name": "poro-chat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", }, + {"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", }, + {"name": "viking", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B + {"name": "gemma", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2b", }, + {"name": "gemma-2", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", }, + {"name": "jais", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", }, + {"name": "t5", "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", }, + {"name": "codeshell", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", }, + {"name": "tekken", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", }, + {"name": "smollm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", }, + {'name': "bloom", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", }, + {'name': "gpt3-finnish", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", }, + {"name": "exaone", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", }, + {"name": "phi-2", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", }, + {"name": "chameleon", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", }, + {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", }, + {"name": "roberta-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"}, + {"name": "gigachat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"}, + {"name": "megrez", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"}, + {"name": "deepseek-v3", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"}, + {"name": "deepseek-r1-qwen", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"}, ] diff --git a/models/ggml-vocab-deepseek-r1-qwen.gguf.inp b/models/ggml-vocab-deepseek-r1-qwen.gguf.inp new file mode 100644 index 000000000..9baf7d77a --- /dev/null +++ b/models/ggml-vocab-deepseek-r1-qwen.gguf.inp @@ -0,0 +1,112 @@ +ied 4 ½ months +__ggml_vocab_test__ +Führer +__ggml_vocab_test__ + +__ggml_vocab_test__ + +__ggml_vocab_test__ + +__ggml_vocab_test__ + +__ggml_vocab_test__ + +__ggml_vocab_test__ + + +__ggml_vocab_test__ + + + +__ggml_vocab_test__ + + + + +__ggml_vocab_test__ + + +__ggml_vocab_test__ +Hello world +__ggml_vocab_test__ + Hello world +__ggml_vocab_test__ +Hello World +__ggml_vocab_test__ + Hello World +__ggml_vocab_test__ + Hello World! +__ggml_vocab_test__ +Hello, world! +__ggml_vocab_test__ + Hello, world! +__ggml_vocab_test__ + this is 🦙.cpp +__ggml_vocab_test__ +w048 7tuijk dsdfhu +__ggml_vocab_test__ +нещо на Български +__ggml_vocab_test__ +កាន់តែពិសេសអាចខលចេញ +__ggml_vocab_test__ +🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ (only emoji that has its own token) +__ggml_vocab_test__ +Hello +__ggml_vocab_test__ + Hello +__ggml_vocab_test__ + Hello +__ggml_vocab_test__ + Hello +__ggml_vocab_test__ + Hello +__ggml_vocab_test__ + Hello + Hello +__ggml_vocab_test__ + ( +__ggml_vocab_test__ + + = +__ggml_vocab_test__ +' era +__ggml_vocab_test__ +Hello, y'all! How are you 😁 ?我想在apple工作1314151天～ +__ggml_vocab_test__ +!!!!!! +__ggml_vocab_test__ +3 +__ggml_vocab_test__ +33 +__ggml_vocab_test__ +333 +__ggml_vocab_test__ +3333 +__ggml_vocab_test__ +33333 +__ggml_vocab_test__ +333333 +__ggml_vocab_test__ +3333333 +__ggml_vocab_test__ +33333333 +__ggml_vocab_test__ +333333333 +__ggml_vocab_test__ +Cửa Việt +__ggml_vocab_test__ + discards +__ggml_vocab_test__ + + + + + + + + + + + +🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ 🦙🦙 3 33 333 3333 33333 333333 3333333 33333333 3.3 3..3 3...3 កាន់តែពិសេសអាច😁 ?我想在apple工作1314151天～ ------======= нещо на Български ''''''```````""""......!!!!!!?????? I've been 'told he's there, 'RE you sure? 'M not sure I'll make it, 'D you like some tea? We'Ve a'lL +__ggml_vocab_test__ diff --git a/models/ggml-vocab-deepseek-r1-qwen.gguf.out b/models/ggml-vocab-deepseek-r1-qwen.gguf.out new file mode 100644 index 000000000..18b4b45cd --- /dev/null +++ b/models/ggml-vocab-deepseek-r1-qwen.gguf.out @@ -0,0 +1,46 @@ + 1122 220 19 220 26062 3951 + 37 50753 261 + + 220 + 256 + 262 + 197 + 198 + 271 + 1406 + 1572 + 9707 1879 + 21927 1879 + 9707 4337 + 21927 4337 + 21927 4337 0 + 9707 11 1879 0 + 21927 11 1879 0 + 419 374 11162 99 247 13 10821 + 86 15 19 23 220 22 83 1963 41808 11472 2940 16739 + 78762 14144 1456 13073 63471 33594 3038 133178 79012 + 146394 97529 241 44258 233 146568 44258 224 147603 20879 115 146280 44258 223 146280 147272 97529 227 147805 148301 147270 44258 223 146848 + 145836 320 8252 8 26525 114 378 235 149921 30543 320 35673 99066 97534 8 25521 227 320 3243 42365 429 702 1181 1828 3950 8 + 9707 + 21927 + 220 21927 + 256 21927 + 262 21927 + 262 21927 198 262 21927 + 320 + 198 284 + 6 11385 + 9707 11 379 64848 0 2585 525 498 26525 223 937 104100 18493 22377 99257 16 18 16 19 16 20 16 35727 21216 + 17085 2928 + 18 + 18 18 + 18 18 18 + 18 18 18 18 + 18 18 18 18 18 + 18 18 18 18 18 18 + 18 18 18 18 18 18 18 + 18 18 18 18 18 18 18 18 + 18 18 18 18 18 18 18 18 18 + 34 90063 128324 + 2560 2347 + 198 4710 14731 65497 7847 1572 2303 78672 10947 145836 320 8252 8 26525 114 378 235 149921 30543 320 35673 99066 97534 8 25521 227 11162 99 247 149955 220 18 220 18 18 220 18 18 18 220 18 18 18 18 220 18 18 18 18 18 220 18 18 18 18 18 18 220 18 18 18 18 18 18 18 220 18 18 18 18 18 18 18 18 220 18 13 18 220 18 496 18 220 18 1112 18 220 146394 97529 241 44258 233 146568 44258 224 147603 20879 115 146280 44258 223 146280 147272 97529 227 144534 937 104100 18493 22377 99257 16 18 16 19 16 20 16 35727 21216 55460 53237 18658 14144 1456 13073 63471 33594 3038 133178 79012 3355 4605 4605 13874 13874 73594 3014 3014 28149 17085 2928 26610 7646 358 3003 1012 364 83 813 566 594 1052 11 364 787 498 2704 30 364 44 537 2704 358 3278 1281 432 11 364 35 498 1075 1045 15243 30 1205 6 42612 264 63866 43 diff --git a/src/llama-chat.cpp b/src/llama-chat.cpp index 1347ec156..5c19bab24 100644 --- a/src/llama-chat.cpp +++ b/src/llama-chat.cpp @@ -152,7 +152,7 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) { return LLM_CHAT_TEMPLATE_MINICPM; } else if (tmpl_contains("'Assistant: ' + message['content'] + eos_token")) { return LLM_CHAT_TEMPLATE_DEEPSEEK_2; - } else if (tmpl_contains(LU8("'<｜Assistant｜>' + message['content'] + '<｜end▁of▁sentence｜>'"))) { + } else if (tmpl_contains(LU8("<｜Assistant｜>")) && tmpl_contains(LU8("<｜User｜>")) && tmpl_contains(LU8("<｜end▁of▁sentence｜>"))) { return LLM_CHAT_TEMPLATE_DEEPSEEK_3; } else if (tmpl_contains("[|system|]") && tmpl_contains("[|assistant|]") && tmpl_contains("[|endofturn|]")) { // ref: https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/discussions/8#66bae61b1893d14ee8ed85bb diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp index 9a680aed4..0782d3a41 100644 --- a/src/llama-vocab.cpp +++ b/src/llama-vocab.cpp @@ -1523,7 +1523,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) { pre_type = LLAMA_VOCAB_PRE_TYPE_COMMAND_R; clean_spaces = false; } else if ( - tokenizer_pre == "qwen2") { + tokenizer_pre == "qwen2" || + tokenizer_pre == "deepseek-r1-qwen") { pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN2; clean_spaces = false; } else if ( From a4251edd6fc16d168f517a7e4b0936212b296603 Mon Sep 17 00:00:00 2001 From: Michael Podvitskiy Date: Mon, 20 Jan 2025 15:02:15 +0100 Subject: [PATCH 13/26] cmake: fix shell command quoting in build-info script (#11309) --- cmake/build-info.cmake | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/cmake/build-info.cmake b/cmake/build-info.cmake index ea3dc55c8..c1a456e17 100644 --- a/cmake/build-info.cmake +++ b/cmake/build-info.cmake @@ -44,7 +44,7 @@ if(MSVC) set(BUILD_TARGET ${CMAKE_VS_PLATFORM_NAME}) else() execute_process( - COMMAND sh -c "$@ --version | head -1" _ ${CMAKE_C_COMPILER} + COMMAND sh -c "\"$@\" --version | head -1" _ ${CMAKE_C_COMPILER} OUTPUT_VARIABLE OUT OUTPUT_STRIP_TRAILING_WHITESPACE ) From 90d987b105db6016cb395f673b4d21e02129721e Mon Sep 17 00:00:00 2001 From: Christopher Nielsen <62156882+mascguy@users.noreply.github.com> Date: Mon, 20 Jan 2025 09:02:43 -0500 Subject: [PATCH 14/26] mmap: add include for cerrno (#11296) ggml-ci Co-authored-by: Xuan Son Nguyen --- src/llama-mmap.cpp | 1 + 1 file changed, 1 insertion(+) diff --git a/src/llama-mmap.cpp b/src/llama-mmap.cpp index 57c6e4f51..b716630a8 100644 --- a/src/llama-mmap.cpp +++ b/src/llama-mmap.cpp @@ -7,6 +7,7 @@ #include #include #include +#include #ifdef __has_include #if __has_include() From 9f7add1cde670ff744b791f5ed6649e86a0c586c Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Mon, 20 Jan 2025 16:36:08 +0200 Subject: [PATCH 15/26] examples : fix add_special conditions (#11311) --- examples/run/run.cpp | 10 ++++++---- examples/simple-chat/simple-chat.cpp | 8 +++++--- 2 files changed, 11 insertions(+), 7 deletions(-) diff --git a/examples/run/run.cpp b/examples/run/run.cpp index dd9ea79e8..d04108e71 100644 --- a/examples/run/run.cpp +++ b/examples/run/run.cpp @@ -729,10 +729,12 @@ static int apply_chat_template(LlamaData & llama_data, const bool append) { // Function to tokenize the prompt static int tokenize_prompt(const llama_vocab * vocab, const std::string & prompt, - std::vector & prompt_tokens) { - const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, true, true); + std::vector & prompt_tokens, const LlamaData & llama_data) { + const bool is_first = llama_get_kv_cache_used_cells(llama_data.context.get()) == 0; + + const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true); prompt_tokens.resize(n_prompt_tokens); - if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), true, + if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), is_first, true) < 0) { printe("failed to tokenize the prompt\n"); return -1; @@ -778,7 +780,7 @@ static int generate(LlamaData & llama_data, const std::string & prompt, std::str const llama_vocab * vocab = llama_model_get_vocab(llama_data.model.get()); std::vector tokens; - if (tokenize_prompt(vocab, prompt, tokens) < 0) { + if (tokenize_prompt(vocab, prompt, tokens, llama_data) < 0) { return 1; } diff --git a/examples/simple-chat/simple-chat.cpp b/examples/simple-chat/simple-chat.cpp index 26422601d..212b3fd79 100644 --- a/examples/simple-chat/simple-chat.cpp +++ b/examples/simple-chat/simple-chat.cpp @@ -95,13 +95,15 @@ int main(int argc, char ** argv) { llama_sampler_chain_add(smpl, llama_sampler_init_dist(LLAMA_DEFAULT_SEED)); // helper function to evaluate a prompt and generate a response - auto generate = [&](const std::string & prompt, bool is_first) { + auto generate = [&](const std::string & prompt) { std::string response; + const bool is_first = llama_get_kv_cache_used_cells(ctx) == 0; + // tokenize the prompt const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true); std::vector prompt_tokens(n_prompt_tokens); - if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), llama_get_kv_cache_used_cells(ctx) == 0, true) < 0) { + if (llama_tokenize(vocab, prompt.c_str(), prompt.size(), prompt_tokens.data(), prompt_tokens.size(), is_first, true) < 0) { GGML_ABORT("failed to tokenize the prompt\n"); } @@ -180,7 +182,7 @@ int main(int argc, char ** argv) { // generate a response printf("\033[33m"); - std::string response = generate(prompt, prev_len == 0); + std::string response = generate(prompt); printf("\n\033[0m"); // add the response to the messages From aea8ddd5165d525a449e2fc3839db77a71f4a318 Mon Sep 17 00:00:00 2001 From: Jeff Bolz Date: Mon, 20 Jan 2025 10:38:32 -0600 Subject: [PATCH 16/26] vulkan: fix coopmat2 validation failures (#11284) mul mat and flash attention shaders were loading f32 types directly into A/B matrices, which happens to work but is technically invalid usage. For FA, we can load it as an Accumulator matrix and convert and this is not in the inner loop and is cheap enough. For mul mat, it's more efficient to do this conversion in a separate pass and have the input(s) be f16. coopmat2 requires SPIR-V 1.6 (related using to LocalSizeId). LocalSizeId requires maintenance4 be enabled, and SPIR-V 1.6 requires Vulkan 1.3. --- ggml/src/ggml-vulkan/ggml-vulkan.cpp | 65 +++++++++++-------- .../vulkan-shaders/flash_attn_cm2.comp | 2 +- .../vulkan-shaders/mul_mm_cm2.comp | 35 +++------- .../vulkan-shaders/vulkan-shaders-gen.cpp | 7 +- 4 files changed, 53 insertions(+), 56 deletions(-) diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp index 437e9cdcc..78c2f5c45 100644 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp @@ -29,8 +29,6 @@ #include "ggml-vulkan-shaders.hpp" -#define VK_API_VERSION VK_API_VERSION_1_2 - #define CEIL_DIV(M, N) (((M) + (N)-1) / (N)) #define VK_VENDOR_ID_AMD 0x1002 @@ -1614,11 +1612,7 @@ static void ggml_vk_load_shaders(vk_device& device) { CREATE_MM(PIPELINE_NAME . f16acc, NAMELC, _f16acc, WG_DENOMS, WARPTILE, PUSHCONST, PARAMCOUNT) \ CREATE_MM(PIPELINE_NAME . f32acc, NAMELC, , WG_DENOMS, WARPTILE, PUSHCONST, PARAMCOUNT) \ - CREATE_MM(pipeline_matmul_f32, matmul_f32_f32, , wg_denoms, warptile, vk_mat_mat_push_constants, 3) - CREATE_MM(pipeline_matmul_f32_f16, matmul_f32_f16, , wg_denoms, warptile, vk_mat_mat_push_constants, 3) - CREATE_MM2(pipeline_matmul_f16, matmul_f16, wg_denoms, warptile, vk_mat_mat_push_constants, 3) - CREATE_MM2(pipeline_matmul_f16_f32, matmul_f16_f32, wg_denoms, warptile, vk_mat_mat_push_constants, 3) CREATE_MM(pipeline_dequant_mul_mat_mat_f16[GGML_TYPE_Q4_0].f16acc, matmul_q4_0_f16, _f16acc, mmq_wg_denoms, warptile_mmq, vk_mat_mat_push_constants, 3) CREATE_MM(pipeline_dequant_mul_mat_mat_f16[GGML_TYPE_Q4_1].f16acc, matmul_q4_1_f16, _f16acc, mmq_wg_denoms, warptile_mmq, vk_mat_mat_push_constants, 3) CREATE_MM(pipeline_dequant_mul_mat_mat_f16[GGML_TYPE_Q5_0].f16acc, matmul_q5_0_f16, _f16acc, mmq_wg_denoms, warptile_mmq, vk_mat_mat_push_constants, 3) @@ -1631,21 +1625,18 @@ static void ggml_vk_load_shaders(vk_device& device) { CREATE_MM(pipeline_dequant_mul_mat_mat_f16[GGML_TYPE_Q6_K].f16acc, matmul_q6_k_f16, _f16acc, mmq_wg_denoms_k, warptile_mmq_k, vk_mat_mat_push_constants, 3) CREATE_MM(pipeline_dequant_mul_mat_mat_f16[GGML_TYPE_IQ4_NL].f16acc, matmul_iq4_nl_f16, _f16acc, mmq_wg_denoms, warptile_mmq, vk_mat_mat_push_constants, 3) - CREATE_MM(pipeline_matmul_id_f32, matmul_id_f32_f32, , wg_denoms, warptile, vk_mat_mat_id_push_constants, 4) CREATE_MM2(pipeline_matmul_id_f16, matmul_id_f16, wg_denoms, warptile, vk_mat_mat_id_push_constants, 4) - CREATE_MM2(pipeline_matmul_id_f16_f32, matmul_id_f16_f32, wg_denoms, warptile, vk_mat_mat_id_push_constants, 4) - - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_0].f16acc, matmul_id_q4_0_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_1].f16acc, matmul_id_q4_1_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_0].f16acc, matmul_id_q5_0_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_1].f16acc, matmul_id_q5_1_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q8_0].f16acc, matmul_id_q8_0_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q2_K].f16acc, matmul_id_q2_k_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q3_K].f16acc, matmul_id_q3_k_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_K].f16acc, matmul_id_q4_k_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_K].f16acc, matmul_id_q5_k_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q6_K].f16acc, matmul_id_q6_k_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) - CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_IQ4_NL].f16acc, matmul_id_iq4_nl_f32, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_0].f16acc, matmul_id_q4_0_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_1].f16acc, matmul_id_q4_1_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_0].f16acc, matmul_id_q5_0_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_1].f16acc, matmul_id_q5_1_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q8_0].f16acc, matmul_id_q8_0_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q2_K].f16acc, matmul_id_q2_k_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q3_K].f16acc, matmul_id_q3_k_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q4_K].f16acc, matmul_id_q4_k_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q5_K].f16acc, matmul_id_q5_k_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_Q6_K].f16acc, matmul_id_q6_k_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) + CREATE_MM(pipeline_dequant_mul_mat_mat_id[GGML_TYPE_IQ4_NL].f16acc, matmul_id_iq4_nl_f16, , mmqid_wg_denoms, warptile_mmqid, vk_mat_mat_id_push_constants, 4) #undef CREATE_MM #undef CREATE_MM2 } else @@ -2287,6 +2278,14 @@ static vk_device ggml_vk_get_device(size_t idx) { } #endif + VkPhysicalDeviceMaintenance4Features maint4_features {}; + maint4_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_4_FEATURES; + if (maintenance4_support) { + last_struct->pNext = (VkBaseOutStructure *)&maint4_features; + last_struct = (VkBaseOutStructure *)&maint4_features; + device_extensions.push_back("VK_KHR_maintenance4"); + } + vkGetPhysicalDeviceFeatures2(device->physical_device, &device_features2); device->fp16 = device->fp16 && vk12_features.shaderFloat16; @@ -2662,7 +2661,14 @@ void ggml_vk_instance_init() { vk_instance_initialized = true; - vk::ApplicationInfo app_info{ "ggml-vulkan", 1, nullptr, 0, VK_API_VERSION }; + uint32_t api_version = vk::enumerateInstanceVersion(); + + if (api_version < VK_API_VERSION_1_2) { + std::cerr << "ggml_vulkan: Error: Vulkan 1.2 required." << std::endl; + GGML_ABORT("fatal error"); + } + + vk::ApplicationInfo app_info{ "ggml-vulkan", 1, nullptr, 0, api_version }; const std::vector instance_extensions = vk::enumerateInstanceExtensionProperties(); const bool validation_ext = ggml_vk_instance_validation_ext_available(instance_extensions); @@ -2972,7 +2978,7 @@ static vk_matmul_pipeline ggml_vk_get_mul_mat_mat_id_pipeline(ggml_backend_vk_co } } - GGML_ASSERT(src1_type == GGML_TYPE_F32); + GGML_ASSERT(src1_type == GGML_TYPE_F32 || (ctx->device->coopmat2 && src1_type == GGML_TYPE_F16)); switch (src0_type) { case GGML_TYPE_Q4_0: @@ -3812,8 +3818,9 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub src1_uma = d_Qy != nullptr; } - const bool x_non_contig = !ggml_vk_dim01_contiguous(src0); - // Reformat and convert to fp16 if src1 is non-contiguous, or for coopmat2 for better perf + // Reformat and convert to fp16 if non-contiguous, or for coopmat2 for better perf + const bool x_non_contig = (ctx->device->coopmat2 && src0->type == GGML_TYPE_F32) || + !ggml_vk_dim01_contiguous(src0); const bool y_non_contig = (ctx->device->coopmat2 && src1->type == GGML_TYPE_F32) || !ggml_vk_dim01_contiguous(src1); @@ -4393,8 +4400,11 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context& ids_uma = d_ids != nullptr; } - const bool x_non_contig = !ggml_vk_dim01_contiguous(src0); - const bool y_non_contig = !ggml_vk_dim01_contiguous(src1); + // Reformat and convert to fp16 if non-contiguous, or for coopmat2 for better perf + const bool x_non_contig = (ctx->device->coopmat2 && src0->type == GGML_TYPE_F32) || + !ggml_vk_dim01_contiguous(src0); + const bool y_non_contig = (ctx->device->coopmat2 && src1->type == GGML_TYPE_F32) || + !ggml_vk_dim01_contiguous(src1); const bool y_f32_kernel = src1->type == GGML_TYPE_F32 && !y_non_contig; @@ -4404,7 +4414,8 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context& const bool qy_needs_dequant = (src1->type != GGML_TYPE_F16 && !y_f32_kernel) || y_non_contig; if (qx_needs_dequant) { - GGML_ABORT("fatal error"); + // Fall back to dequant + f16 mulmat + mmp = ggml_vk_get_mul_mat_mat_id_pipeline(ctx, GGML_TYPE_F16, y_f32_kernel ? GGML_TYPE_F32 : GGML_TYPE_F16, (ggml_prec)dst->op_params[0]); } // Not implemented diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp index ca3a59b8f..3735d0dbb 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp @@ -166,7 +166,7 @@ void main() { tensorLayoutK = setTensorLayoutStrideNV(tensorLayoutK, k_stride, 1); tensorLayoutV = setTensorLayoutStrideNV(tensorLayoutV, v_stride, 1); - coopmat Q; + coopmat Q; coopmat Qf16; uint32_t q_offset = iq2*p.nb02+iq3*p.nb03; diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp index cbfa5dce1..57f9e7245 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp @@ -57,17 +57,13 @@ layout (binding = 2) writeonly buffer D {D_TYPE data_d[];}; #if QUANT_K > 1 #define DECODEFUNCA , dequantFuncA -#define MAT_A_TYPE float16_t #include "dequant_funcs_cm2.comp" #else #define DECODEFUNCA -#define MAT_A_TYPE A_TYPE #endif -#define MAT_B_TYPE B_TYPE - #ifdef MUL_MAT_ID layout (binding = 3) readonly buffer IDS {int data_ids[];}; @@ -236,16 +232,13 @@ void main() { for (uint block_k = start_k, i = 0; i < k_iters; block_k += BK, ++i) { - coopmat mat_a; - coopmat mat_b; + coopmat mat_a; + coopmat mat_b; coopMatLoadTensorNV(mat_a, data_a, pos_a, sliceTensorLayoutNV(tensorLayoutA, ir * BM, BM, block_k, BK) DECODEFUNCA); - coopmat mat_a_ft = coopmat(mat_a); - coopMatLoadTensorNV(mat_b, data_b, pos_b, sliceTensorLayoutNV(tensorLayoutB, ic * BN, BN, block_k, BK), tensorViewTranspose); - coopmat mat_b_ft = coopmat(mat_b); - sum = coopMatMulAdd(mat_a_ft, mat_b_ft, sum); + sum = coopMatMulAdd(mat_a, mat_b, sum); } } else #endif // !defined(MUL_MAT_ID) @@ -261,10 +254,8 @@ void main() { [[dont_unroll]] for (uint block_k = start_k; block_k < end_k; block_k += BK) { - coopmat mat_a; - coopmat mat_b; - coopmat mat_a_ft; - coopmat mat_b_ft; + coopmat mat_a; + coopmat mat_b; // Clamping is expensive, so detect different code paths for each combination // of A and B needing clamping. @@ -281,16 +272,12 @@ void main() { #else coopMatLoadTensorNV(mat_b, data_b, pos_b, sliceTensorLayoutNV(tensorLayoutB, ic * BN, BN, (block_k & ~7), BK), tensorViewTranspose); #endif - mat_a_ft = coopmat(mat_a); - mat_b_ft = coopmat(mat_b); - sum = coopMatMulAdd(mat_a_ft, mat_b_ft, sum); + sum = coopMatMulAdd(mat_a, mat_b, sum); } else if (unclampedA && !unclampedB) { coopMatLoadTensorNV(mat_a, data_a, pos_a, sliceTensorLayoutNV(tensorLayoutA, ir * BM, BM, (block_k & ~7), BK) DECODEFUNCA); coopMatLoadTensorNV(mat_b, data_b, pos_b, sliceTensorLayoutNV(tensorLayoutBClamp, ic * BN, BN, block_k, BK), tensorViewTranspose); - mat_a_ft = coopmat(mat_a); - mat_b_ft = coopmat(mat_b); - sum = coopMatMulAdd(mat_a_ft, mat_b_ft, sum); + sum = coopMatMulAdd(mat_a, mat_b, sum); } else if (!unclampedA && unclampedB) { coopMatLoadTensorNV(mat_a, data_a, pos_a, sliceTensorLayoutNV(tensorLayoutAClamp, ir * BM, BM, block_k, BK) DECODEFUNCA); #ifdef MUL_MAT_ID @@ -298,16 +285,12 @@ void main() { #else coopMatLoadTensorNV(mat_b, data_b, pos_b, sliceTensorLayoutNV(tensorLayoutB, ic * BN, BN, (block_k & ~7), BK), tensorViewTranspose); #endif - mat_a_ft = coopmat(mat_a); - mat_b_ft = coopmat(mat_b); - sum = coopMatMulAdd(mat_a_ft, mat_b_ft, sum); + sum = coopMatMulAdd(mat_a, mat_b, sum); } else if (!unclampedA && !unclampedB) { coopMatLoadTensorNV(mat_a, data_a, pos_a, sliceTensorLayoutNV(tensorLayoutAClamp, ir * BM, BM, block_k, BK) DECODEFUNCA); coopMatLoadTensorNV(mat_b, data_b, pos_b, sliceTensorLayoutNV(tensorLayoutBClamp, ic * BN, BN, block_k, BK), tensorViewTranspose); - mat_a_ft = coopmat(mat_a); - mat_b_ft = coopmat(mat_b); - sum = coopMatMulAdd(mat_a_ft, mat_b_ft, sum); + sum = coopMatMulAdd(mat_a, mat_b, sum); } } } diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp b/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp index b7890ef1e..8bcb64101 100644 --- a/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +++ b/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp @@ -316,8 +316,11 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool // For aligned matmul loads std::string load_vec_a = (coopmat2 || tname == "f32" || tname == "f16") ? load_vec : "2"; - string_to_spv(shader_name + "_" + tname + "_f32", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc); - string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc); + // don't generate f32 variants for coopmat2 + if (!coopmat2) { + string_to_spv(shader_name + "_" + tname + "_f32", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc); + string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc); + } if (tname != "f16" && tname != "f32") { string_to_spv(shader_name + "_" + tname + "_f16", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc); From 80d0d6b4b7abca342fc990399e78af8d213eabaf Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Mon, 20 Jan 2025 22:29:43 +0200 Subject: [PATCH 17/26] common : add -hfd option for the draft model (#11318) * common : add -hfd option for the draft model * cont : fix env var * cont : more fixes --- common/arg.cpp | 17 +++++++++++++---- common/common.h | 8 +++++++- examples/server/server.cpp | 5 ++++- 3 files changed, 24 insertions(+), 6 deletions(-) diff --git a/common/arg.cpp b/common/arg.cpp index dede335fb..126970950 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -133,7 +133,8 @@ static void common_params_handle_model_default( const std::string & model_url, std::string & hf_repo, std::string & hf_file, - const std::string & hf_token) { + const std::string & hf_token, + const std::string & model_default) { if (!hf_repo.empty()) { // short-hand to avoid specifying --hf-file -> default it to --model if (hf_file.empty()) { @@ -163,7 +164,7 @@ static void common_params_handle_model_default( model = fs_get_cache_file(string_split(f, '/').back()); } } else if (model.empty()) { - model = DEFAULT_MODEL_PATH; + model = model_default; } } @@ -299,8 +300,9 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context } // TODO: refactor model params in a common struct - common_params_handle_model_default(params.model, params.model_url, params.hf_repo, params.hf_file, params.hf_token); - common_params_handle_model_default(params.vocoder.model, params.vocoder.model_url, params.vocoder.hf_repo, params.vocoder.hf_file, params.hf_token); + common_params_handle_model_default(params.model, params.model_url, params.hf_repo, params.hf_file, params.hf_token, DEFAULT_MODEL_PATH); + common_params_handle_model_default(params.speculative.model, params.speculative.model_url, params.speculative.hf_repo, params.speculative.hf_file, params.hf_token, ""); + common_params_handle_model_default(params.vocoder.model, params.vocoder.model_url, params.vocoder.hf_repo, params.vocoder.hf_file, params.hf_token, ""); if (params.escape) { string_process_escapes(params.prompt); @@ -1629,6 +1631,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex params.hf_repo = value; } ).set_env("LLAMA_ARG_HF_REPO")); + add_opt(common_arg( + {"-hfd", "-hfrd", "--hf-repo-draft"}, "/[:quant]", + "Same as --hf-repo, but for the draft model (default: unused)", + [](common_params & params, const std::string & value) { + params.speculative.hf_repo = value; + } + ).set_env("LLAMA_ARG_HFD_REPO")); add_opt(common_arg( {"-hff", "--hf-file"}, "FILE", "Hugging Face model file. If specified, it will override the quant in --hf-repo (default: unused)", diff --git a/common/common.h b/common/common.h index 3bcc637cc..b2709c044 100644 --- a/common/common.h +++ b/common/common.h @@ -175,7 +175,11 @@ struct common_params_speculative { struct cpu_params cpuparams; struct cpu_params cpuparams_batch; - std::string model = ""; // draft model for speculative decoding // NOLINT + std::string hf_repo = ""; // HF repo // NOLINT + std::string hf_file = ""; // HF file // NOLINT + + std::string model = ""; // draft model for speculative decoding // NOLINT + std::string model_url = ""; // model url to download // NOLINT }; struct common_params_vocoder { @@ -508,12 +512,14 @@ struct llama_model * common_load_model_from_url( const std::string & local_path, const std::string & hf_token, const struct llama_model_params & params); + struct llama_model * common_load_model_from_hf( const std::string & repo, const std::string & remote_path, const std::string & local_path, const std::string & hf_token, const struct llama_model_params & params); + std::pair common_get_hf_file( const std::string & hf_repo_with_tag, const std::string & hf_token); diff --git a/examples/server/server.cpp b/examples/server/server.cpp index d1e8ee829..f35206d7b 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -1728,13 +1728,16 @@ struct server_context { add_bos_token = llama_vocab_get_add_bos(vocab); has_eos_token = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL; - if (!params_base.speculative.model.empty()) { + if (!params_base.speculative.model.empty() || !params_base.speculative.hf_repo.empty()) { SRV_INF("loading draft model '%s'\n", params_base.speculative.model.c_str()); auto params_dft = params_base; params_dft.devices = params_base.speculative.devices; + params_dft.hf_file = params_base.speculative.hf_file; + params_dft.hf_repo = params_base.speculative.hf_repo; params_dft.model = params_base.speculative.model; + params_dft.model_url = params_base.speculative.model_url; params_dft.n_ctx = params_base.speculative.n_ctx == 0 ? params_base.n_ctx / params_base.n_parallel : params_base.speculative.n_ctx; params_dft.n_gpu_layers = params_base.speculative.n_gpu_layers; params_dft.n_parallel = 1; From 153e8524113621d3ca90d146e6dc5d42a5c42160 Mon Sep 17 00:00:00 2001 From: Olivier Chafik Date: Mon, 20 Jan 2025 20:55:52 +0000 Subject: [PATCH 18/26] Apply suggestions from code review Co-authored-by: Xuan Son Nguyen Co-authored-by: Georgi Gerganov --- common/common.cpp | 6 +++--- common/common.h | 4 ++-- include/llama.h | 1 + 3 files changed, 6 insertions(+), 5 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index 9c535a176..ce023fc2b 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1821,11 +1821,11 @@ std::string common_chat_format_example(const llama_chat_template & tmpl, bool us return common_chat_apply_template(tmpl, msgs, true, use_jinja); } -llama_chat_templates llama_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) +llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) { auto vocab = llama_model_get_vocab(model); - auto bos_token = common_token_to_piece(vocab, llama_vocab_bos(vocab), true); - auto eos_token = common_token_to_piece(vocab, llama_vocab_eos(vocab), true); + auto token_bos = common_token_to_piece(vocab, llama_vocab_bos(vocab), true); + auto token_eos = common_token_to_piece(vocab, llama_vocab_eos(vocab), true); std::string default_template_src = chat_template_override; std::string tool_use_template_src = chat_template_override; bool has_explicit_template = !chat_template_override.empty(); diff --git a/common/common.h b/common/common.h index a96a99531..352cbb0fa 100644 --- a/common/common.h +++ b/common/common.h @@ -607,8 +607,8 @@ typedef minja::chat_template llama_chat_template; struct llama_chat_templates { bool has_explicit_template; // Model had builtin template or template overridde was specified. - std::unique_ptr default_template; // always set (defaults to chatml) - std::unique_ptr tool_use_template; + std::unique_ptr template_default; // always set (defaults to chatml) + std::unique_ptr template_tool_use; }; // CPP wrapper for llama_chat_apply_template diff --git a/include/llama.h b/include/llama.h index dca9314aa..3b75e7607 100644 --- a/include/llama.h +++ b/include/llama.h @@ -510,6 +510,7 @@ extern "C" { LLAMA_API uint64_t llama_model_size(const struct llama_model * model); // Get the default chat template. Returns nullptr if not available + // If name is NULL, returns the default chat template LLAMA_API const char * llama_model_chat_template(const struct llama_model * model, const char * name); // Returns the total number of parameters in the model From db9dd0c1acc497766f5b0957f4d5d32c883d7904 Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 21:06:18 +0000 Subject: [PATCH 19/26] Finish suggested renamings --- common/common.cpp | 14 +++++++------- common/common.h | 2 +- examples/main/main.cpp | 8 ++++---- examples/run/run.cpp | 8 ++++---- examples/server/server.cpp | 26 +++++++++++++------------- 5 files changed, 29 insertions(+), 29 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index ce023fc2b..2c0558b5b 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1827,7 +1827,7 @@ llama_chat_templates common_chat_templates_from_model(const struct llama_model * auto token_bos = common_token_to_piece(vocab, llama_vocab_bos(vocab), true); auto token_eos = common_token_to_piece(vocab, llama_vocab_eos(vocab), true); std::string default_template_src = chat_template_override; - std::string tool_use_template_src = chat_template_override; + std::string template_tool_use_src = chat_template_override; bool has_explicit_template = !chat_template_override.empty(); if (chat_template_override.empty()) { auto str = llama_model_chat_template(model, /* name */ nullptr); @@ -1837,13 +1837,13 @@ llama_chat_templates common_chat_templates_from_model(const struct llama_model * } str = llama_model_chat_template(model, /* name */ "tool_use"); if (str) { - tool_use_template_src = str; + template_tool_use_src = str; has_explicit_template = true; } } if (default_template_src.empty() || default_template_src == "chatml") { - if (!tool_use_template_src.empty()) { - default_template_src = tool_use_template_src; + if (!template_tool_use_src.empty()) { + default_template_src = template_tool_use_src; } else { default_template_src = R"( {%- for message in messages -%} @@ -1857,10 +1857,10 @@ llama_chat_templates common_chat_templates_from_model(const struct llama_model * } return { has_explicit_template, - std::make_unique(default_template_src, bos_token, eos_token), - tool_use_template_src.empty() + std::make_unique(default_template_src, token_bos, token_eos), + template_tool_use_src.empty() ? nullptr - : std::make_unique(tool_use_template_src, bos_token, eos_token) + : std::make_unique(template_tool_use_src, token_bos, token_eos) }; } diff --git a/common/common.h b/common/common.h index 352cbb0fa..7b50c82d2 100644 --- a/common/common.h +++ b/common/common.h @@ -632,7 +632,7 @@ std::string common_chat_format_single( std::string common_chat_format_example( const llama_chat_template & tmpl, bool use_jinja); -llama_chat_templates llama_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override); +llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override); // // KV cache utils diff --git a/examples/main/main.cpp b/examples/main/main.cpp index 903a92faf..da2a03ab9 100644 --- a/examples/main/main.cpp +++ b/examples/main/main.cpp @@ -158,7 +158,7 @@ int main(int argc, char ** argv) { } const llama_vocab * vocab = llama_model_get_vocab(model); - auto chat_templates = llama_chat_templates_from_model(model, params.chat_template); + auto chat_templates = common_chat_templates_from_model(model, params.chat_template); LOG_INF("%s: llama threadpool init, n_threads = %d\n", __func__, (int) params.cpuparams.n_threads); @@ -201,7 +201,7 @@ int main(int argc, char ** argv) { } // auto enable conversation mode if chat template is available - const bool has_chat_template = chat_templates.has_explicit_template && chat_templates.default_template; + const bool has_chat_template = chat_templates.has_explicit_template && chat_templates.template_default; if (params.conversation_mode == COMMON_CONVERSATION_MODE_AUTO) { if (has_chat_template) { LOG_INF("%s: chat template is available, enabling conversation mode (disable it with -no-cnv)\n", __func__); @@ -219,7 +219,7 @@ int main(int argc, char ** argv) { // print chat template example in conversation mode if (params.conversation_mode) { if (params.enable_chat_template) { - LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(*chat_templates.default_template, params.use_jinja).c_str()); + LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(*chat_templates.template_default, params.use_jinja).c_str()); } else { LOG_INF("%s: in-suffix/prefix is specified, chat template will be disabled\n", __func__); } @@ -265,7 +265,7 @@ int main(int argc, char ** argv) { auto chat_add_and_format = [&chat_msgs, &chat_templates](const std::string & role, const std::string & content) { common_chat_msg new_msg{role, content}; - auto formatted = common_chat_format_single(*chat_templates.default_template, chat_msgs, new_msg, role == "user", g_params->use_jinja); + auto formatted = common_chat_format_single(*chat_templates.template_default, chat_msgs, new_msg, role == "user", g_params->use_jinja); chat_msgs.push_back({role, content}); LOG_DBG("formatted: '%s'\n", formatted.c_str()); return formatted; diff --git a/examples/run/run.cpp b/examples/run/run.cpp index 46a945347..408bd7181 100644 --- a/examples/run/run.cpp +++ b/examples/run/run.cpp @@ -936,8 +936,8 @@ static int get_user_input(std::string & user_input, const std::string & user) { static int chat_loop(LlamaData & llama_data, const std::string & user, bool use_jinja) { int prev_len = 0; llama_data.fmtted.resize(llama_n_ctx(llama_data.context.get())); - auto chat_templates = llama_chat_templates_from_model(llama_data.model.get(), ""); - GGML_ASSERT(chat_templates.default_template); + auto chat_templates = common_chat_templates_from_model(llama_data.model.get(), ""); + GGML_ASSERT(chat_templates.template_default); static const bool stdout_a_terminal = is_stdout_a_terminal(); while (true) { // Get user input @@ -948,7 +948,7 @@ static int chat_loop(LlamaData & llama_data, const std::string & user, bool use_ add_message("user", user.empty() ? user_input : user, llama_data); int new_len; - if (apply_chat_template_with_error_handling(*chat_templates.default_template, llama_data, true, new_len, use_jinja) < 0) { + if (apply_chat_template_with_error_handling(*chat_templates.template_default, llama_data, true, new_len, use_jinja) < 0) { return 1; } @@ -963,7 +963,7 @@ static int chat_loop(LlamaData & llama_data, const std::string & user, bool use_ } add_message("assistant", response, llama_data); - if (apply_chat_template_with_error_handling(*chat_templates.default_template, llama_data, false, prev_len, use_jinja) < 0) { + if (apply_chat_template_with_error_handling(*chat_templates.template_default, llama_data, false, prev_len, use_jinja) < 0) { return 1; } } diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 189290df9..6717198c5 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -1745,15 +1745,15 @@ struct server_context { llama_chat_message chat[] = {{"user", "test"}}; if (use_jinja) { - auto templates = llama_chat_templates_from_model(model, ""); - GGML_ASSERT(templates.default_template); + auto templates = common_chat_templates_from_model(model, ""); + GGML_ASSERT(templates.template_default); try { - templates.default_template->apply({{ + templates.template_default->apply({{ {"role", "user"}, {"content", "test"}, }}, json(), true); - if (templates.tool_use_template) { - templates.tool_use_template->apply({{ + if (templates.template_tool_use) { + templates.template_tool_use->apply({{ {"role", "user"}, {"content", "test"}, }}, json(), true); @@ -3631,8 +3631,8 @@ int main(int argc, char ** argv) { auto get_chat_templates = [&ctx_server, &chat_templates_mutex, &chat_templates]() -> const llama_chat_templates & { std::lock_guard lock(chat_templates_mutex); if (!chat_templates) { - chat_templates = llama_chat_templates_from_model(ctx_server.model, ctx_server.params_base.chat_template); - GGML_ASSERT(chat_templates->default_template); + chat_templates = common_chat_templates_from_model(ctx_server.model, ctx_server.params_base.chat_template); + GGML_ASSERT(chat_templates->template_default); } return *chat_templates; }; @@ -3644,11 +3644,11 @@ int main(int argc, char ** argv) { { "default_generation_settings", ctx_server.default_generation_settings_for_props }, { "total_slots", ctx_server.params_base.n_parallel }, { "model_path", ctx_server.params_base.model }, - { "chat_template", templates.default_template->source() }, + { "chat_template", templates.template_default->source() }, { "build_info", build_info }, }; - if (ctx_server.params_base.use_jinja && templates.tool_use_template) { - data["chat_template_tool_use"] = templates.tool_use_template->source(); + if (ctx_server.params_base.use_jinja && templates.template_tool_use) { + data["chat_template_tool_use"] = templates.template_tool_use->source(); } res_ok(res, data); @@ -3871,7 +3871,7 @@ int main(int argc, char ** argv) { auto body = json::parse(req.body); const auto & templates = get_chat_templates(); - const auto & chat_template = body.contains("tools") && templates.tool_use_template ? *templates.tool_use_template : *templates.default_template; + const auto & chat_template = body.contains("tools") && templates.template_tool_use ? *templates.template_tool_use : *templates.template_default; json data = oaicompat_completion_params_parse(body, chat_template, params.use_jinja); return handle_completions_impl( @@ -4290,8 +4290,8 @@ int main(int argc, char ** argv) { // print sample chat example to make it clear which template is used LOG_INF("%s: chat template, chat_template: %s, example_format: '%s'\n", __func__, - get_chat_templates().default_template->source().c_str(), - common_chat_format_example(*get_chat_templates().default_template, ctx_server.params_base.use_jinja).c_str()); + get_chat_templates().template_default->source().c_str(), + common_chat_format_example(*get_chat_templates().template_default, ctx_server.params_base.use_jinja).c_str()); ctx_server.queue_tasks.on_new_task(std::bind( &server_context::process_single_task, &ctx_server, std::placeholders::_1)); From c9e8fdd70e576c1c71635db645227a3d5738423a Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 21:25:18 +0000 Subject: [PATCH 20/26] Move chat_templates inside server_context + remove mutex --- examples/server/server.cpp | 34 ++++++++++++---------------------- 1 file changed, 12 insertions(+), 22 deletions(-) diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 6717198c5..eabbf7940 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -1662,6 +1662,8 @@ struct server_context { // Necessary similarity of prompt for slot selection float slot_prompt_similarity = 0.0f; + llama_chat_templates chat_templates; + ~server_context() { // Clear any sampling context for (server_slot & slot : slots) { @@ -1738,6 +1740,8 @@ struct server_context { cparams_dft.type_v = GGML_TYPE_F16; } + chat_templates = common_chat_templates_from_model(model, params_base.chat_template); + return true; } @@ -3625,30 +3629,17 @@ int main(int argc, char ** argv) { } }; - std::mutex chat_templates_mutex; - std::optional chat_templates; - - auto get_chat_templates = [&ctx_server, &chat_templates_mutex, &chat_templates]() -> const llama_chat_templates & { - std::lock_guard lock(chat_templates_mutex); - if (!chat_templates) { - chat_templates = common_chat_templates_from_model(ctx_server.model, ctx_server.params_base.chat_template); - GGML_ASSERT(chat_templates->template_default); - } - return *chat_templates; - }; - - const auto handle_props = [&ctx_server, &res_ok, &get_chat_templates](const httplib::Request &, httplib::Response & res) { + const auto handle_props = [&ctx_server, &res_ok](const httplib::Request &, httplib::Response & res) { // this endpoint is publicly available, please only return what is safe to be exposed - const auto & templates = get_chat_templates(); json data = { { "default_generation_settings", ctx_server.default_generation_settings_for_props }, { "total_slots", ctx_server.params_base.n_parallel }, { "model_path", ctx_server.params_base.model }, - { "chat_template", templates.template_default->source() }, + { "chat_template", ctx_server.chat_templates.template_default->source() }, { "build_info", build_info }, }; - if (ctx_server.params_base.use_jinja && templates.template_tool_use) { - data["chat_template_tool_use"] = templates.template_tool_use->source(); + if (ctx_server.params_base.use_jinja && ctx_server.chat_templates.template_tool_use) { + data["chat_template_tool_use"] = ctx_server.chat_templates.template_tool_use->source(); } res_ok(res, data); @@ -3863,15 +3854,14 @@ int main(int argc, char ** argv) { OAICOMPAT_TYPE_NONE); // infill is not OAI compatible }; - const auto handle_chat_completions = [&ctx_server, ¶ms, &res_error, &handle_completions_impl, &get_chat_templates](const httplib::Request & req, httplib::Response & res) { + const auto handle_chat_completions = [&ctx_server, ¶ms, &res_error, &handle_completions_impl](const httplib::Request & req, httplib::Response & res) { if (ctx_server.params_base.embedding) { res_error(res, format_error_response("This server does not support completions. Start it without `--embeddings`", ERROR_TYPE_NOT_SUPPORTED)); return; } auto body = json::parse(req.body); - const auto & templates = get_chat_templates(); - const auto & chat_template = body.contains("tools") && templates.template_tool_use ? *templates.template_tool_use : *templates.template_default; + const auto & chat_template = body.contains("tools") && ctx_server.chat_templates.template_tool_use ? *ctx_server.chat_templates.template_tool_use : *ctx_server.chat_templates.template_default; json data = oaicompat_completion_params_parse(body, chat_template, params.use_jinja); return handle_completions_impl( @@ -4290,8 +4280,8 @@ int main(int argc, char ** argv) { // print sample chat example to make it clear which template is used LOG_INF("%s: chat template, chat_template: %s, example_format: '%s'\n", __func__, - get_chat_templates().template_default->source().c_str(), - common_chat_format_example(*get_chat_templates().template_default, ctx_server.params_base.use_jinja).c_str()); + ctx_server.chat_templates.template_default->source().c_str(), + common_chat_format_example(*ctx_server.chat_templates.template_default, ctx_server.params_base.use_jinja).c_str()); ctx_server.queue_tasks.on_new_task(std::bind( &server_context::process_single_task, &ctx_server, std::placeholders::_1)); From 8c84aefd4d8609def3127cc37f091648a1af8820 Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 21:48:31 +0000 Subject: [PATCH 21/26] Update --chat-template-file w/ recent change to --chat-template --- common/arg.cpp | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/common/arg.cpp b/common/arg.cpp index b46f205f6..53bd32e3a 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -1966,10 +1966,12 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE")); add_opt(common_arg( {"--chat-template-file"}, "JINJA_TEMPLATE_FILE", - "set custom jinja chat template file (default: template taken from model's metadata)\n" - "if suffix/prefix are specified, template will be disabled\n" - "only commonly used templates are accepted (unless --jinja is set before this flag):\n" - "https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template", + string_format( + "set custom jinja chat template file (default: template taken from model's metadata)\n" + "if suffix/prefix are specified, template will be disabled\n" + "only commonly used templates are accepted (unless --jinja is set before this flag):\n" + "list of built-in templates:\n%s", list_builtin_chat_templates().c_str() + ), [](common_params & params, const std::string & value) { std::ifstream file(value); if (!file) { From 154bfaaa390d537b4e84a9cc5f9c539bcb93bf2c Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 21:54:34 +0000 Subject: [PATCH 22/26] Refactor chat template validation --- common/arg.cpp | 27 +++++++++------------------ 1 file changed, 9 insertions(+), 18 deletions(-) diff --git a/common/arg.cpp b/common/arg.cpp index 53bd32e3a..5799d7832 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -323,6 +323,14 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context throw std::invalid_argument("error: either --embedding or --reranking can be specified, but not both"); } + if (!params.chat_template.empty() && !common_chat_verify_template(params.chat_template, params.use_jinja)) { + throw std::runtime_error(string_format( + "error: the supplied chat template is not supported: %s%s\n", + params.chat_template.c_str(), + params.use_jinja ? "" : "\nnote: llama.cpp was started without --jinja, we only support commonly used templates" + )); + } + return true; } @@ -1954,13 +1962,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex "list of built-in templates:\n%s", list_builtin_chat_templates().c_str() ), [](common_params & params, const std::string & value) { - if (!common_chat_verify_template(value, params.use_jinja)) { - throw std::runtime_error(string_format( - "error: the supplied chat template is not supported: %s%s\n", - value.c_str(), - params.use_jinja ? "" : "\nnote: llama.cpp does not use jinja parser, we only support commonly used templates" - )); - } params.chat_template = value; } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE")); @@ -1977,20 +1978,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex if (!file) { throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } - std::string chat_template; std::copy( std::istreambuf_iterator(file), std::istreambuf_iterator(), - std::back_inserter(chat_template) - ); - if (!common_chat_verify_template(chat_template, params.use_jinja)) { - throw std::runtime_error(string_format( - "error: the supplied chat template is not supported: %s%s\n", - value.c_str(), - params.use_jinja ? "" : "\nnote: llama.cpp does not use jinja parser, we only support commonly used templates" - )); - } - params.chat_template = chat_template; + std::back_inserter(params.chat_template)); } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE_FILE")); add_opt(common_arg( From 54a669e09e8c565bb8b1b14bc6340da685632529 Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 22:50:08 +0000 Subject: [PATCH 23/26] Guard against missing eos/bos tokens (null token otherwise throws in llama_vocab::impl::token_get_attr) --- common/common.cpp | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index 2c0558b5b..58529b63d 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1824,8 +1824,9 @@ std::string common_chat_format_example(const llama_chat_template & tmpl, bool us llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) { auto vocab = llama_model_get_vocab(model); - auto token_bos = common_token_to_piece(vocab, llama_vocab_bos(vocab), true); - auto token_eos = common_token_to_piece(vocab, llama_vocab_eos(vocab), true); + // TODO: consider detecting if the template needs bos / eos tokens and warn / error when missing. + auto token_bos = llama_vocab_bos(vocab) == LLAMA_TOKEN_NULL ? "" : common_token_to_piece(vocab, llama_vocab_bos(vocab), true); + auto token_eos = llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL ? "" : common_token_to_piece(vocab, llama_vocab_eos(vocab), true); std::string default_template_src = chat_template_override; std::string template_tool_use_src = chat_template_override; bool has_explicit_template = !chat_template_override.empty(); From 8348c605acc017fe46dd5fd2e460d7d69758a231 Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 23:00:47 +0000 Subject: [PATCH 24/26] Warn against missing eos / bos tokens when jinja template references them --- common/common.cpp | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index 58529b63d..161e2aa35 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1824,9 +1824,6 @@ std::string common_chat_format_example(const llama_chat_template & tmpl, bool us llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) { auto vocab = llama_model_get_vocab(model); - // TODO: consider detecting if the template needs bos / eos tokens and warn / error when missing. - auto token_bos = llama_vocab_bos(vocab) == LLAMA_TOKEN_NULL ? "" : common_token_to_piece(vocab, llama_vocab_bos(vocab), true); - auto token_eos = llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL ? "" : common_token_to_piece(vocab, llama_vocab_eos(vocab), true); std::string default_template_src = chat_template_override; std::string template_tool_use_src = chat_template_override; bool has_explicit_template = !chat_template_override.empty(); @@ -1856,6 +1853,19 @@ llama_chat_templates common_chat_templates_from_model(const struct llama_model * )"; } } + const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) { + if (token == LLAMA_TOKEN_NULL) { + if (default_template_src.find(jinja_variable_name) != std::string::npos + || template_tool_use_src.find(jinja_variable_name) != std::string::npos) { + LOG_WRN("%s: warning: vocab does not have a %s token, jinja template won't work as intended.\n", __func__, name); + } + return std::string(); + } else { + return common_token_to_piece(vocab, token, true); + } + }; + auto token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token"); + auto token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token"); return { has_explicit_template, std::make_unique(default_template_src, token_bos, token_eos), From ee475d2f513b15956db8a18f5507fedeb04f171e Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 23:42:07 +0000 Subject: [PATCH 25/26] rename: common_chat_template[s] --- common/common.cpp | 8 ++++---- common/common.h | 16 ++++++++-------- examples/run/run.cpp | 4 ++-- examples/server/server.cpp | 2 +- examples/server/utils.hpp | 4 ++-- 5 files changed, 17 insertions(+), 17 deletions(-) diff --git a/common/common.cpp b/common/common.cpp index 161e2aa35..727ab0a10 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1749,7 +1749,7 @@ bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) { } std::string common_chat_apply_template( - const llama_chat_template & tmpl, + const common_chat_template & tmpl, const std::vector & msgs, bool add_ass, bool use_jinja) { @@ -1791,7 +1791,7 @@ std::string common_chat_apply_template( } std::string common_chat_format_single( - const llama_chat_template & tmpl, + const common_chat_template & tmpl, const std::vector & past_msg, const common_chat_msg & new_msg, bool add_ass, @@ -1811,7 +1811,7 @@ std::string common_chat_format_single( return ss.str(); } -std::string common_chat_format_example(const llama_chat_template & tmpl, bool use_jinja) { +std::string common_chat_format_example(const common_chat_template & tmpl, bool use_jinja) { std::vector msgs = { {"system", "You are a helpful assistant"}, {"user", "Hello"}, @@ -1821,7 +1821,7 @@ std::string common_chat_format_example(const llama_chat_template & tmpl, bool us return common_chat_apply_template(tmpl, msgs, true, use_jinja); } -llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) +common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override) { auto vocab = llama_model_get_vocab(model); std::string default_template_src = chat_template_override; diff --git a/common/common.h b/common/common.h index ac25a6f65..7c9d73ce1 100644 --- a/common/common.h +++ b/common/common.h @@ -611,26 +611,26 @@ namespace minja { class chat_template; } -typedef minja::chat_template llama_chat_template; +typedef minja::chat_template common_chat_template; -struct llama_chat_templates { +struct common_chat_templates { bool has_explicit_template; // Model had builtin template or template overridde was specified. - std::unique_ptr template_default; // always set (defaults to chatml) - std::unique_ptr template_tool_use; + std::unique_ptr template_default; // always set (defaults to chatml) + std::unique_ptr template_tool_use; }; // CPP wrapper for llama_chat_apply_template // If the built-in template is not supported, we default to chatml // If the custom "tmpl" is not supported, we throw an error std::string common_chat_apply_template( - const llama_chat_template & tmpl, + const common_chat_template & tmpl, const std::vector & chat, bool add_ass, bool use_jinja); // Format single message, while taking into account the position of that message in chat history std::string common_chat_format_single( - const llama_chat_template & tmpl, + const common_chat_template & tmpl, const std::vector & past_msg, const common_chat_msg & new_msg, bool add_ass, @@ -638,9 +638,9 @@ std::string common_chat_format_single( // Returns an example of formatted chat std::string common_chat_format_example( - const llama_chat_template & tmpl, bool use_jinja); + const common_chat_template & tmpl, bool use_jinja); -llama_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override); +common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override); // // KV cache utils diff --git a/examples/run/run.cpp b/examples/run/run.cpp index 4c72f22f9..e567ad716 100644 --- a/examples/run/run.cpp +++ b/examples/run/run.cpp @@ -717,7 +717,7 @@ static void add_message(const char * role, const std::string & text, LlamaData & } // Function to apply the chat template and resize `formatted` if needed -static int apply_chat_template(const llama_chat_template & tmpl, LlamaData & llama_data, const bool append, bool use_jinja) { +static int apply_chat_template(const common_chat_template & tmpl, LlamaData & llama_data, const bool append, bool use_jinja) { if (use_jinja) { json messages = json::array(); for (const auto & msg : llama_data.messages) { @@ -893,7 +893,7 @@ static int generate_response(LlamaData & llama_data, const std::string & prompt, } // Helper function to apply the chat template and handle errors -static int apply_chat_template_with_error_handling(const llama_chat_template & tmpl, LlamaData & llama_data, const bool append, int & output_length, bool use_jinja) { +static int apply_chat_template_with_error_handling(const common_chat_template & tmpl, LlamaData & llama_data, const bool append, int & output_length, bool use_jinja) { const int new_len = apply_chat_template(tmpl, llama_data, append, use_jinja); if (new_len < 0) { printe("failed to apply the chat template\n"); diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 408e50e39..798b7facc 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -1689,7 +1689,7 @@ struct server_context { // Necessary similarity of prompt for slot selection float slot_prompt_similarity = 0.0f; - llama_chat_templates chat_templates; + common_chat_templates chat_templates; ~server_context() { // Clear any sampling context diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp index b6cec0eb8..c5987250c 100644 --- a/examples/server/utils.hpp +++ b/examples/server/utils.hpp @@ -351,7 +351,7 @@ static llama_tokens format_infill( } // Format given chat. If tmpl is empty, we take the template from model metadata -inline std::string format_chat(const llama_chat_template & tmpl, const std::vector & messages) { +inline std::string format_chat(const common_chat_template & tmpl, const std::vector & messages) { std::vector chat; for (size_t i = 0; i < messages.size(); ++i) { @@ -580,7 +580,7 @@ static json oaicompat_completion_params_parse(const json & body) { static json oaicompat_completion_params_parse( const json & body, /* openai api json semantics */ - const llama_chat_template & tmpl, + const common_chat_template & tmpl, bool use_jinja) { json llama_params; From 8a7c89e60c90be8c04f58335cd11ab5c91ae1ac7 Mon Sep 17 00:00:00 2001 From: ochafik Date: Mon, 20 Jan 2025 23:44:42 +0000 Subject: [PATCH 26/26] reinstate assert on chat_templates.template_default --- examples/server/server.cpp | 1 + 1 file changed, 1 insertion(+) diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 798b7facc..865be4d8d 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -1771,6 +1771,7 @@ struct server_context { } chat_templates = common_chat_templates_from_model(model, params_base.chat_template); + GGML_ASSERT(chat_templates.template_default.get() != nullptr); return true; }