/* * * Copyright 2016 gRPC authors. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * */ #include #include "src/core/lib/slice/slice_internal.h" #include #include #include #include #include "src/core/lib/gpr/murmur_hash.h" #include "src/core/lib/iomgr/iomgr_internal.h" /* for iomgr_abort_on_leaks() */ #include "src/core/lib/profiling/timers.h" #include "src/core/lib/slice/slice_string_helpers.h" #include "src/core/lib/transport/static_metadata.h" #define LOG2_SHARD_COUNT 5 #define SHARD_COUNT (1 << LOG2_SHARD_COUNT) #define INITIAL_SHARD_CAPACITY 8 #define TABLE_IDX(hash, capacity) (((hash) >> LOG2_SHARD_COUNT) % (capacity)) #define SHARD_IDX(hash) ((hash) & ((1 << LOG2_SHARD_COUNT) - 1)) typedef struct interned_slice_refcount { grpc_slice_refcount base; grpc_slice_refcount sub; size_t length; gpr_atm refcnt; uint32_t hash; struct interned_slice_refcount* bucket_next; } interned_slice_refcount; typedef struct slice_shard { gpr_mu mu; interned_slice_refcount** strs; size_t count; size_t capacity; } slice_shard; /* hash seed: decided at initialization time */ static uint32_t g_hash_seed; static int g_forced_hash_seed = 0; static slice_shard g_shards[SHARD_COUNT]; typedef struct { uint32_t hash; uint32_t idx; } static_metadata_hash_ent; static static_metadata_hash_ent static_metadata_hash[4 * GRPC_STATIC_MDSTR_COUNT]; static uint32_t max_static_metadata_hash_probe; static uint32_t static_metadata_hash_values[GRPC_STATIC_MDSTR_COUNT]; static void interned_slice_ref(void* p) { interned_slice_refcount* s = static_cast(p); GPR_ASSERT(gpr_atm_no_barrier_fetch_add(&s->refcnt, 1) > 0); } static void interned_slice_destroy(interned_slice_refcount* s) { slice_shard* shard = &g_shards[SHARD_IDX(s->hash)]; gpr_mu_lock(&shard->mu); GPR_ASSERT(0 == gpr_atm_no_barrier_load(&s->refcnt)); interned_slice_refcount** prev_next; interned_slice_refcount* cur; for (prev_next = &shard->strs[TABLE_IDX(s->hash, shard->capacity)], cur = *prev_next; cur != s; prev_next = &cur->bucket_next, cur = cur->bucket_next) ; *prev_next = cur->bucket_next; shard->count--; gpr_free(s); gpr_mu_unlock(&shard->mu); } static void interned_slice_unref(void* p) { interned_slice_refcount* s = static_cast(p); if (1 == gpr_atm_full_fetch_add(&s->refcnt, -1)) { interned_slice_destroy(s); } } static void interned_slice_sub_ref(void* p) { interned_slice_ref((static_cast(p)) - offsetof(interned_slice_refcount, sub)); } static void interned_slice_sub_unref(void* p) { interned_slice_unref((static_cast(p)) - offsetof(interned_slice_refcount, sub)); } static uint32_t interned_slice_hash(grpc_slice slice) { interned_slice_refcount* s = reinterpret_cast(slice.refcount); return s->hash; } static int interned_slice_eq(grpc_slice a, grpc_slice b) { return a.refcount == b.refcount; } static const grpc_slice_refcount_vtable interned_slice_vtable = { interned_slice_ref, interned_slice_unref, interned_slice_eq, interned_slice_hash}; static const grpc_slice_refcount_vtable interned_slice_sub_vtable = { interned_slice_sub_ref, interned_slice_sub_unref, grpc_slice_default_eq_impl, grpc_slice_default_hash_impl}; static void grow_shard(slice_shard* shard) { GPR_TIMER_SCOPE("grow_strtab", 0); size_t capacity = shard->capacity * 2; size_t i; interned_slice_refcount** strtab; interned_slice_refcount *s, *next; strtab = static_cast( gpr_zalloc(sizeof(interned_slice_refcount*) * capacity)); for (i = 0; i < shard->capacity; i++) { for (s = shard->strs[i]; s; s = next) { size_t idx = TABLE_IDX(s->hash, capacity); next = s->bucket_next; s->bucket_next = strtab[idx]; strtab[idx] = s; } } gpr_free(shard->strs); shard->strs = strtab; shard->capacity = capacity; } static grpc_slice materialize(interned_slice_refcount* s) { grpc_slice slice; slice.refcount = &s->base; slice.data.refcounted.bytes = reinterpret_cast(s + 1); slice.data.refcounted.length = s->length; return slice; } uint32_t grpc_slice_default_hash_impl(grpc_slice s) { return gpr_murmur_hash3(GRPC_SLICE_START_PTR(s), GRPC_SLICE_LENGTH(s), g_hash_seed); } uint32_t grpc_static_slice_hash(grpc_slice s) { return static_metadata_hash_values[GRPC_STATIC_METADATA_INDEX(s)]; } int grpc_static_slice_eq(grpc_slice a, grpc_slice b) { return GRPC_STATIC_METADATA_INDEX(a) == GRPC_STATIC_METADATA_INDEX(b); } uint32_t grpc_slice_hash(grpc_slice s) { return s.refcount == nullptr ? grpc_slice_default_hash_impl(s) : s.refcount->vtable->hash(s); } grpc_slice grpc_slice_maybe_static_intern(grpc_slice slice, bool* returned_slice_is_different) { if (GRPC_IS_STATIC_METADATA_STRING(slice)) { return slice; } uint32_t hash = grpc_slice_hash(slice); for (uint32_t i = 0; i <= max_static_metadata_hash_probe; i++) { static_metadata_hash_ent ent = static_metadata_hash[(hash + i) % GPR_ARRAY_SIZE(static_metadata_hash)]; if (ent.hash == hash && ent.idx < GRPC_STATIC_MDSTR_COUNT && grpc_slice_eq(grpc_static_slice_table[ent.idx], slice)) { *returned_slice_is_different = true; return grpc_static_slice_table[ent.idx]; } } return slice; } bool grpc_slice_is_interned(grpc_slice slice) { return (slice.refcount && slice.refcount->vtable == &interned_slice_vtable) || GRPC_IS_STATIC_METADATA_STRING(slice); } grpc_slice grpc_slice_intern(grpc_slice slice) { GPR_TIMER_SCOPE("grpc_slice_intern", 0); if (GRPC_IS_STATIC_METADATA_STRING(slice)) { return slice; } uint32_t hash = grpc_slice_hash(slice); for (uint32_t i = 0; i <= max_static_metadata_hash_probe; i++) { static_metadata_hash_ent ent = static_metadata_hash[(hash + i) % GPR_ARRAY_SIZE(static_metadata_hash)]; if (ent.hash == hash && ent.idx < GRPC_STATIC_MDSTR_COUNT && grpc_slice_eq(grpc_static_slice_table[ent.idx], slice)) { return grpc_static_slice_table[ent.idx]; } } interned_slice_refcount* s; slice_shard* shard = &g_shards[SHARD_IDX(hash)]; gpr_mu_lock(&shard->mu); /* search for an existing string */ size_t idx = TABLE_IDX(hash, shard->capacity); for (s = shard->strs[idx]; s; s = s->bucket_next) { if (s->hash == hash && grpc_slice_eq(slice, materialize(s))) { if (gpr_atm_no_barrier_fetch_add(&s->refcnt, 1) == 0) { /* If we get here, we've added a ref to something that was about to * die - drop it immediately. * The *only* possible path here (given the shard mutex) should be to * drop from one ref back to zero - assert that with a CAS */ GPR_ASSERT(gpr_atm_rel_cas(&s->refcnt, 1, 0)); /* and treat this as if we were never here... sshhh */ } else { gpr_mu_unlock(&shard->mu); return materialize(s); } } } /* not found: create a new string */ /* string data goes after the internal_string header */ s = static_cast( gpr_malloc(sizeof(*s) + GRPC_SLICE_LENGTH(slice))); gpr_atm_rel_store(&s->refcnt, 1); s->length = GRPC_SLICE_LENGTH(slice); s->hash = hash; s->base.vtable = &interned_slice_vtable; s->base.sub_refcount = &s->sub; s->sub.vtable = &interned_slice_sub_vtable; s->sub.sub_refcount = &s->sub; s->bucket_next = shard->strs[idx]; shard->strs[idx] = s; memcpy(s + 1, GRPC_SLICE_START_PTR(slice), GRPC_SLICE_LENGTH(slice)); shard->count++; if (shard->count > shard->capacity * 2) { grow_shard(shard); } gpr_mu_unlock(&shard->mu); return materialize(s); } void grpc_test_only_set_slice_hash_seed(uint32_t seed) { g_hash_seed = seed; g_forced_hash_seed = 1; } void grpc_slice_intern_init(void) { if (!g_forced_hash_seed) { g_hash_seed = static_cast(gpr_now(GPR_CLOCK_REALTIME).tv_nsec); } for (size_t i = 0; i < SHARD_COUNT; i++) { slice_shard* shard = &g_shards[i]; gpr_mu_init(&shard->mu); shard->count = 0; shard->capacity = INITIAL_SHARD_CAPACITY; shard->strs = static_cast( gpr_zalloc(sizeof(*shard->strs) * shard->capacity)); } for (size_t i = 0; i < GPR_ARRAY_SIZE(static_metadata_hash); i++) { static_metadata_hash[i].hash = 0; static_metadata_hash[i].idx = GRPC_STATIC_MDSTR_COUNT; } max_static_metadata_hash_probe = 0; for (size_t i = 0; i < GRPC_STATIC_MDSTR_COUNT; i++) { static_metadata_hash_values[i] = grpc_slice_default_hash_impl(grpc_static_slice_table[i]); for (size_t j = 0; j < GPR_ARRAY_SIZE(static_metadata_hash); j++) { size_t slot = (static_metadata_hash_values[i] + j) % GPR_ARRAY_SIZE(static_metadata_hash); if (static_metadata_hash[slot].idx == GRPC_STATIC_MDSTR_COUNT) { static_metadata_hash[slot].hash = static_metadata_hash_values[i]; static_metadata_hash[slot].idx = static_cast(i); if (j > max_static_metadata_hash_probe) { max_static_metadata_hash_probe = static_cast(j); } break; } } } } void grpc_slice_intern_shutdown(void) { for (size_t i = 0; i < SHARD_COUNT; i++) { slice_shard* shard = &g_shards[i]; gpr_mu_destroy(&shard->mu); /* TODO(ctiller): GPR_ASSERT(shard->count == 0); */ if (shard->count != 0) { gpr_log(GPR_DEBUG, "WARNING: %" PRIuPTR " metadata strings were leaked", shard->count); for (size_t j = 0; j < shard->capacity; j++) { for (interned_slice_refcount* s = shard->strs[j]; s; s = s->bucket_next) { char* text = grpc_dump_slice(materialize(s), GPR_DUMP_HEX | GPR_DUMP_ASCII); gpr_log(GPR_DEBUG, "LEAKED: %s", text); gpr_free(text); } } if (grpc_iomgr_abort_on_leaks()) { abort(); } } gpr_free(shard->strs); } }