diff options
| author | Onur Gündüz <ogunduz@gmail.com> | 2026-06-04 18:46:46 +0200 |
|---|---|---|
| committer | Onur Gündüz <ogunduz@gmail.com> | 2026-06-14 08:27:03 +0200 |
| commit | d08f68523ea9a39af8176703e48fd47a987893b5 (patch) | |
| tree | 8f9b9d05c6b3fcb3d59c424e9e7b888f70e71b77 | |
| parent | 8fc09516a8028a754e8cd4aecefa0677791379cf (diff) | |
| download | ffwd-d08f68523ea9a39af8176703e48fd47a987893b5.tar.gz ffwd-d08f68523ea9a39af8176703e48fd47a987893b5.zip | |
chore: address static analyzer warnings
| -rw-r--r-- | main.c | 9 | ||||
| -rw-r--r-- | pplx_embed.c | 7 | ||||
| -rw-r--r-- | pplx_embed_mlx.c | 18 | ||||
| -rw-r--r-- | pplx_server.c | 34 | ||||
| -rw-r--r-- | qwen_asr_tokenizer.c | 51 |
5 files changed, 92 insertions, 27 deletions
@@ -527,7 +527,6 @@ int main(int argc, char *argv[]) int mlx_quantize_bits = 0; int mlx_quantize_group_size = 64; int dist_activation_bits = 32; - const char *backend = "cpu"; const char *host = "127.0.0.1"; const char *api_key = NULL; int port = 8000; @@ -559,10 +558,13 @@ int main(int argc, char *argv[]) } } else if (!strcmp(f, "--model")) { - if (append_model_spec(&model_specs, argv[++arg_start]) != 0) return 1; + if (append_model_spec(&model_specs, argv[++arg_start]) != 0) { + free_model_specs(&model_specs); + return 1; + } } else if (!strcmp(f, "--backend")) { - backend = argv[++arg_start]; + const char *backend = argv[++arg_start]; if (!strcmp(backend, "mlx")) { #ifdef USE_MLX use_mlx = 1; @@ -627,7 +629,6 @@ int main(int argc, char *argv[]) else if (!strcmp(f, "--mlx")) { #ifdef USE_MLX use_mlx = 1; - backend = "mlx"; #else fprintf(stderr, "--mlx not available (build with: make mlx)\n"); free_model_specs(&model_specs); diff --git a/pplx_embed.c b/pplx_embed.c index 35bd739..2c0ade6 100644 --- a/pplx_embed.c +++ b/pplx_embed.c @@ -388,8 +388,11 @@ static int tensor_has_supported_shape(const safetensors_file_t *sf, } size_t bytes; - if (mul_size(numel, elem_size, &bytes) != 0 || - t->data_size != bytes) { + if (mul_size(numel, elem_size, &bytes) != 0) { + fprintf(stderr, "pplx: tensor data too large for %s\n", name); + return 0; + } + if (t->data_size != bytes) { fprintf(stderr, "pplx: bad data size for %s: got %zu, expected %zu\n", name, t->data_size, bytes); return 0; diff --git a/pplx_embed_mlx.c b/pplx_embed_mlx.c index ca98c5f..6e5476a 100644 --- a/pplx_embed_mlx.c +++ b/pplx_embed_mlx.c @@ -118,8 +118,11 @@ static int mlx_tensor_has_supported_shape(const safetensors_file_t *sf, } size_t bytes; - if (mlx_mul_size(numel, elem_size, &bytes) != 0 || - t->data_size != bytes) { + if (mlx_mul_size(numel, elem_size, &bytes) != 0) { + fprintf(stderr, "mlx: tensor data too large for %s\n", name); + return 0; + } + if (t->data_size != bytes) { fprintf(stderr, "mlx: bad data size for %s: got %zu, expected %zu\n", name, t->data_size, bytes); return 0; @@ -1484,14 +1487,19 @@ int pplx_mlx_forward_slice_batch(pplx_mlx_ctx_t *ctx, if (inputs[b].n_tokens != max_seq) has_padding = 1; } - size_t rows, state_values, ids_bytes, mask_bytes, state_bytes, row_bytes; + size_t rows, ids_bytes, mask_bytes, state_bytes, row_bytes; if (mlx_mul_size((size_t)batch, (size_t)max_seq, &rows) != 0 || - mlx_mul_size(rows, (size_t)hidden, &state_values) != 0 || mlx_mul_size(rows, sizeof(int), &ids_bytes) != 0 || mlx_mul_size(rows, sizeof(float), &mask_bytes) != 0 || - mlx_mul_size(state_values, sizeof(float), &state_bytes) != 0 || mlx_mul_size((size_t)hidden, sizeof(float), &row_bytes) != 0) return -1; + if (rows == 0 || row_bytes == 0 || (has_padding && mask_bytes == 0)) + return -1; + if (layer_start != 0 && + mlx_mul_size(rows, (size_t)hidden, &state_bytes) != 0) + return -1; + if (layer_start == 0) + state_bytes = 0; int *padded_ids = layer_start == 0 ? (int *)malloc(ids_bytes) : NULL; diff --git a/pplx_server.c b/pplx_server.c index 63c363d..85df0e5 100644 --- a/pplx_server.c +++ b/pplx_server.c @@ -371,6 +371,16 @@ static void client_decref(client *c) { if (free_now) client_free(c); } +static void client_decref_n(client *c, int n) { + int free_now = 0; + if (n <= 0) return; + pthread_mutex_lock(&c->srv->mu); + c->refcount -= n; + if (c->refcount <= 0) free_now = 1; + pthread_mutex_unlock(&c->srv->mu); + if (free_now) client_free(c); +} + static void client_unlink(client *c) { http_server *s = c->srv; if (!c->linked) return; @@ -382,8 +392,8 @@ static void client_unlink(client *c) { if (s->n_clients > 0) s->n_clients--; } -static void close_client(client *c) { - if (!c || c->cancelled) return; +static int close_client_unlink(client *c) { + if (!c || c->cancelled) return 0; c->cancelled = 1; if (c->fd >= 0) { aeDeleteFileEvent(c->srv->loop, c->fd, AE_READABLE); @@ -392,6 +402,11 @@ static void close_client(client *c) { c->fd = -1; } client_unlink(c); + return 1; +} + +static void close_client(client *c) { + if (!close_client_unlink(c)) return; client_decref(c); } @@ -734,15 +749,16 @@ static void completion_cb(aeEventLoop *loop, int fd, void *clientData, int mask) job *j; while ((j = pop_done(s)) != NULL) { client *c = j->c; + int refs_to_drop = 1; /* job reference */ if (!c->cancelled) { append_http_response_ex(c, j->status, j->content_type, j->extra_headers, j->response ? j->response : "", j->response ? j->response_len : 0); if (queue_write(c) != AE_OK) - close_client(c); + refs_to_drop += close_client_unlink(c); } - client_decref(c); + client_decref_n(c, refs_to_drop); job_free(j); } } @@ -1594,13 +1610,13 @@ static void prepare_embedding_request(job *j, cJSON *root, http_server *s, return; } - if (cJSON_GetArraySize(detail) == 0 && m) { + if (cJSON_GetArraySize(detail) == 0 && m && input) { out->n_inputs = n_inputs; out->tokens = xcalloc((size_t)n_inputs, sizeof(*out->tokens)); out->inputs = xmalloc((size_t)n_inputs * sizeof(*out->inputs)); int idx = 0; - if (cJSON_IsString(input)) { + if (input && cJSON_IsString(input) && input->valuestring) { if (tokenize_one(m, j, input->valuestring, &out->tokens[0]) != 0) { ve_add(detail, "[\"body\",\"input\"]", "tokenization failed", "value_error.tokenization"); @@ -1618,6 +1634,12 @@ static void prepare_embedding_request(job *j, cJSON *root, http_server *s, cJSON_ArrayForEach(item, input) { char loc[64]; snprintf(loc, sizeof(loc), "[\"body\",\"input\",%d]", idx); + if (!cJSON_IsString(item) || !item->valuestring) { + ve_add(detail, loc, "input item must be a string", + "type_error.string"); + idx++; + continue; + } if (tokenize_one(m, j, item->valuestring, &out->tokens[idx]) != 0) { ve_add(detail, loc, "tokenization failed", diff --git a/qwen_asr_tokenizer.c b/qwen_asr_tokenizer.c index 8c86f22..fa85017 100644 --- a/qwen_asr_tokenizer.c +++ b/qwen_asr_tokenizer.c @@ -82,8 +82,8 @@ static char *decode_gpt2_token(const char *token_str) { const unsigned char *end = p + len; while (p < end) { - int cp; - int nbytes; + int cp = 0; + int nbytes = 1; if ((*p & 0x80) == 0) { cp = *p; @@ -256,6 +256,28 @@ static int map_insert(str_int_entry_t *map, int cap, char *key, int value) { return -1; } +static int map_insert_owned(str_int_entry_t *map, int cap, char **keyp, int value) { + if (!keyp) return -1; + char *key = *keyp; + if (!map || cap <= 0 || !key) return -1; + int mask = cap - 1; + int pos = (int)(fnv1a_hash(key) & (uint64_t)mask); + for (int i = 0; i < cap; i++) { + int idx = (pos + i) & mask; + if (!map[idx].key) { + map[idx].key = key; + map[idx].value = value; + *keyp = NULL; + return 0; + } + if (strcmp(map[idx].key, key) == 0) { + map[idx].value = value; + return 0; + } + } + return -1; +} + static int map_get(const str_int_entry_t *map, int cap, const char *key) { if (!map || cap <= 0 || !key) return -1; int mask = cap - 1; @@ -1244,7 +1266,10 @@ static int load_merges_map(qwen_tokenizer_t *tok, const char *merges_path) { return -1; } - rewind(f); + if (fseek(f, 0, SEEK_SET) != 0) { + fclose(f); + return -1; + } int rank = 0; while (fgets(line, sizeof(line), f)) { char *a = NULL, *b = NULL; @@ -1258,9 +1283,9 @@ static int load_merges_map(qwen_tokenizer_t *tok, const char *merges_path) { memcpy(key + la + 1, b, lb); key[la + 1 + lb] = '\0'; - if (map_insert((str_int_entry_t *)tok->merge_map, tok->merge_map_cap, key, rank) != 0) { - free(key); - } + (void)map_insert_owned((str_int_entry_t *)tok->merge_map, + tok->merge_map_cap, &key, rank); + free(key); rank++; } fclose(f); @@ -1278,9 +1303,15 @@ qwen_tokenizer_t *qwen_tokenizer_load(const char *vocab_json_path) { return NULL; } - fseek(f, 0, SEEK_END); + if (fseek(f, 0, SEEK_END) != 0) { + fclose(f); + return NULL; + } long file_size = ftell(f); - fseek(f, 0, SEEK_SET); + if (file_size < 0 || fseek(f, 0, SEEK_SET) != 0) { + fclose(f); + return NULL; + } char *json = (char *)malloc((size_t)file_size + 1); if (!json || fread(json, 1, (size_t)file_size, f) != (size_t)file_size) { @@ -1306,7 +1337,7 @@ qwen_tokenizer_t *qwen_tokenizer_load(const char *vocab_json_path) { if (*p == ',') { p++; continue; } if (*p == '}') break; - char key[4096]; + char key[4096] = {0}; if (parse_json_string(&p, key, sizeof(key)) != 0) break; skip_ws(&p); if (*p != ':') break; @@ -1336,7 +1367,7 @@ qwen_tokenizer_t *qwen_tokenizer_load(const char *vocab_json_path) { if (*p == ',') { p++; continue; } if (*p == '}') break; - char key[4096]; + char key[4096] = {0}; if (parse_json_string(&p, key, sizeof(key)) != 0) break; skip_ws(&p); if (*p != ':') break; |
