/* * upb - a minimalist implementation of protocol buffers. * * Copyright (c) 2014 Google Inc. See LICENSE for details. * Author: Josh Haberman * * Since we are implementing pure handlers (ie. without any out-of-band access * to pre-computed lengths), we have to buffer all submessages before we can * emit even their first byte. * * Not knowing the size of submessages also means we can't write a perfect * zero-copy implementation, even with buffering. Lengths are stored as * varints, which means that we don't know how many bytes to reserve for the * length until we know what the length is. * * This leaves us with three main choices: * * 1. buffer all submessage data in a temporary buffer, then copy it exactly * once into the output buffer. * * 2. attempt to buffer data directly into the output buffer, estimating how * many bytes each length will take. When our guesses are wrong, use * memmove() to grow or shrink the allotted space. * * 3. buffer directly into the output buffer, allocating a max length * ahead-of-time for each submessage length. If we overallocated, we waste * space, but no memcpy() or memmove() is required. This approach requires * defining a maximum size for submessages and rejecting submessages that * exceed that size. * * (2) and (3) have the potential to have better performance, but they are more * complicated and subtle to implement: * * (3) requires making an arbitrary choice of the maximum message size; it * wastes space when submessages are shorter than this and fails * completely when they are longer. This makes it more finicky and * requires configuration based on the input. It also makes it impossible * to perfectly match the output of reference encoders that always use the * optimal amount of space for each length. * * (2) requires guessing the the size upfront, and if multiple lengths are * guessed wrong the minimum required number of memmove() operations may * be complicated to compute correctly. Implemented properly, it may have * a useful amortized or average cost, but more investigation is required * to determine this and what the optimal algorithm is to achieve it. * * (1) makes you always pay for exactly one copy, but its implementation is * the simplest and its performance is predictable. * * So for now, we implement (1) only. If we wish to optimize later, we should * be able to do it without affecting users. * * The strategy is to buffer the segments of data that do *not* depend on * unknown lengths in one buffer, and keep a separate buffer of segment pointers * and lengths. When the top-level submessage ends, we can go beginning to end, * alternating the writing of lengths with memcpy() of the rest of the data. * At the top level though, no buffering is required. */ #include "upb/pb/encoder.h" #include "upb/pb/varint.int.h" #include // The output buffer is divided into segments; a segment is a string of data // that is "ready to go" -- it does not need any varint lengths inserted into // the middle. The seams between segments are where varints will be inserted // once they are known. // // We also use the concept of a "run", which is a range of encoded bytes that // occur at a single submessage level. Every segment contains one or more runs. // // A segment can span messages. Consider: // // .--Submessage lengths---------. // | | | // | V V // V | |--------------- | |----------------- // Submessages: | |----------------------------------------------- // Top-level msg: ------------------------------------------------------------ // // Segments: ----- ------------------- ----------------- // Runs: *---- *--------------*--- *---------------- // (* marks the start) // // Note that the top-level menssage is not in any segment because it does not // have any length preceding it. // // A segment is only interrupted when another length needs to be inserted. So // observe how the second segment spans both the inner submessage and part of // the next enclosing message. typedef struct { uint32_t msglen; // The length to varint-encode before this segment. uint32_t seglen; // Length of the segment. } upb_pb_encoder_segment; struct upb_pb_encoder { upb_env *env; // Our input and output. upb_sink input_; upb_bytessink *output_; // The "subclosure" -- used as the inner closure as part of the bytessink // protocol. void *subc; // The output buffer and limit, and our current write position. "buf" // initially points to "initbuf", but is dynamically allocated if we need to // grow beyond the initial size. char *buf, *ptr, *limit; // The beginning of the current run, or undefined if we are at the top level. char *runbegin; // The list of segments we are accumulating. upb_pb_encoder_segment *segbuf, *segptr, *seglimit; // The stack of enclosing submessages. Each entry in the stack points to the // segment where this submessage's length is being accumulated. int *stack, *top, *stacklimit; // Depth of startmsg/endmsg calls. int depth; }; /* low-level buffering ********************************************************/ // Low-level functions for interacting with the output buffer. // TODO(haberman): handle pushback static void putbuf(upb_pb_encoder *e, const char *buf, size_t len) { size_t n = upb_bytessink_putbuf(e->output_, e->subc, buf, len, NULL); UPB_ASSERT_VAR(n, n == len); } static upb_pb_encoder_segment *top(upb_pb_encoder *e) { return &e->segbuf[*e->top]; } // Call to ensure that at least "bytes" bytes are available for writing at // e->ptr. Returns false if the bytes could not be allocated. static bool reserve(upb_pb_encoder *e, size_t bytes) { if ((e->limit - e->ptr) < bytes) { // Grow buffer. size_t needed = bytes + (e->ptr - e->buf); size_t old_size = e->limit - e->buf; size_t new_size = old_size; while (new_size < needed) { new_size *= 2; } char *new_buf = upb_env_realloc(e->env, e->buf, old_size, new_size); if (new_buf == NULL) { return false; } e->ptr = new_buf + (e->ptr - e->buf); e->runbegin = new_buf + (e->runbegin - e->buf); e->limit = new_buf + new_size; e->buf = new_buf; } return true; } // Call when "bytes" bytes have been writte at e->ptr. The caller *must* have // previously called reserve() with at least this many bytes. static void encoder_advance(upb_pb_encoder *e, size_t bytes) { assert((e->limit - e->ptr) >= bytes); e->ptr += bytes; } // Call when all of the bytes for a handler have been written. Flushes the // bytes if possible and necessary, returning false if this failed. static bool commit(upb_pb_encoder *e) { if (!e->top) { // We aren't inside a delimited region. Flush our accumulated bytes to // the output. // // TODO(haberman): in the future we may want to delay flushing for // efficiency reasons. putbuf(e, e->buf, e->ptr - e->buf); e->ptr = e->buf; } return true; } // Writes the given bytes to the buffer, handling reserve/advance. static bool encode_bytes(upb_pb_encoder *e, const void *data, size_t len) { if (!reserve(e, len)) { return false; } memcpy(e->ptr, data, len); encoder_advance(e, len); return true; } // Finish the current run by adding the run totals to the segment and message // length. static void accumulate(upb_pb_encoder *e) { assert(e->ptr >= e->runbegin); size_t run_len = e->ptr - e->runbegin; e->segptr->seglen += run_len; top(e)->msglen += run_len; e->runbegin = e->ptr; } // Call to indicate the start of delimited region for which the full length is // not yet known. All data will be buffered until the length is known. // Delimited regions may be nested; their lengths will all be tracked properly. static bool start_delim(upb_pb_encoder *e) { if (e->top) { // We are already buffering, advance to the next segment and push it on the // stack. accumulate(e); if (++e->top == e->stacklimit) { // TODO(haberman): grow stack? return false; } if (++e->segptr == e->seglimit) { // Grow segment buffer. size_t old_size = (e->seglimit - e->segbuf) * sizeof(upb_pb_encoder_segment); size_t new_size = old_size * 2; upb_pb_encoder_segment *new_buf = upb_env_realloc(e->env, e->segbuf, old_size, new_size); if (new_buf == NULL) { return false; } e->segptr = new_buf + (e->segptr - e->segbuf); e->seglimit = new_buf + (new_size / sizeof(upb_pb_encoder_segment)); e->segbuf = new_buf; } } else { // We were previously at the top level, start buffering. e->segptr = e->segbuf; e->top = e->stack; e->runbegin = e->ptr; } *e->top = e->segptr - e->segbuf; e->segptr->seglen = 0; e->segptr->msglen = 0; return true; } // Call to indicate the end of a delimited region. We now know the length of // the delimited region. If we are not nested inside any other delimited // regions, we can now emit all of the buffered data we accumulated. static bool end_delim(upb_pb_encoder *e) { accumulate(e); size_t msglen = top(e)->msglen; if (e->top == e->stack) { // All lengths are now available, emit all buffered data. char buf[UPB_PB_VARINT_MAX_LEN]; upb_pb_encoder_segment *s; const char *ptr = e->buf; for (s = e->segbuf; s <= e->segptr; s++) { size_t lenbytes = upb_vencode64(s->msglen, buf); putbuf(e, buf, lenbytes); putbuf(e, ptr, s->seglen); ptr += s->seglen; } e->ptr = e->buf; e->top = NULL; } else { // Need to keep buffering; propagate length info into enclosing submessages. --e->top; top(e)->msglen += msglen + upb_varint_size(msglen); } return true; } /* tag_t **********************************************************************/ // A precomputed (pre-encoded) tag and length. typedef struct { uint8_t bytes; char tag[7]; } tag_t; // Allocates a new tag for this field, and sets it in these handlerattr. static void new_tag(upb_handlers *h, const upb_fielddef *f, upb_wiretype_t wt, upb_handlerattr *attr) { uint32_t n = upb_fielddef_number(f); tag_t *tag = malloc(sizeof(tag_t)); tag->bytes = upb_vencode64((n << 3) | wt, tag->tag); upb_handlerattr_init(attr); upb_handlerattr_sethandlerdata(attr, tag); upb_handlers_addcleanup(h, tag, free); } static bool encode_tag(upb_pb_encoder *e, const tag_t *tag) { return encode_bytes(e, tag->tag, tag->bytes); } /* encoding of wire types *****************************************************/ static bool encode_fixed64(upb_pb_encoder *e, uint64_t val) { // TODO(haberman): byte-swap for big endian. return encode_bytes(e, &val, sizeof(uint64_t)); } static bool encode_fixed32(upb_pb_encoder *e, uint32_t val) { // TODO(haberman): byte-swap for big endian. return encode_bytes(e, &val, sizeof(uint32_t)); } static bool encode_varint(upb_pb_encoder *e, uint64_t val) { if (!reserve(e, UPB_PB_VARINT_MAX_LEN)) { return false; } encoder_advance(e, upb_vencode64(val, e->ptr)); return true; } static uint64_t dbl2uint64(double d) { uint64_t ret; memcpy(&ret, &d, sizeof(uint64_t)); return ret; } static uint32_t flt2uint32(float d) { uint32_t ret; memcpy(&ret, &d, sizeof(uint32_t)); return ret; } /* encoding of proto types ****************************************************/ static bool startmsg(void *c, const void *hd) { upb_pb_encoder *e = c; UPB_UNUSED(hd); if (e->depth++ == 0) { upb_bytessink_start(e->output_, 0, &e->subc); } return true; } static bool endmsg(void *c, const void *hd, upb_status *status) { upb_pb_encoder *e = c; UPB_UNUSED(hd); UPB_UNUSED(status); if (--e->depth == 0) { upb_bytessink_end(e->output_); } return true; } static void *encode_startdelimfield(void *c, const void *hd) { bool ok = encode_tag(c, hd) && commit(c) && start_delim(c); return ok ? c : UPB_BREAK; } static bool encode_enddelimfield(void *c, const void *hd) { UPB_UNUSED(hd); return end_delim(c); } static void *encode_startgroup(void *c, const void *hd) { return (encode_tag(c, hd) && commit(c)) ? c : UPB_BREAK; } static bool encode_endgroup(void *c, const void *hd) { return encode_tag(c, hd) && commit(c); } static void *encode_startstr(void *c, const void *hd, size_t size_hint) { UPB_UNUSED(size_hint); return encode_startdelimfield(c, hd); } static size_t encode_strbuf(void *c, const void *hd, const char *buf, size_t len, const upb_bufhandle *h) { UPB_UNUSED(hd); UPB_UNUSED(h); return encode_bytes(c, buf, len) ? len : 0; } #define T(type, ctype, convert, encode) \ static bool encode_scalar_##type(void *e, const void *hd, ctype val) { \ return encode_tag(e, hd) && encode(e, (convert)(val)) && commit(e); \ } \ static bool encode_packed_##type(void *e, const void *hd, ctype val) { \ UPB_UNUSED(hd); \ return encode(e, (convert)(val)); \ } T(double, double, dbl2uint64, encode_fixed64) T(float, float, flt2uint32, encode_fixed32); T(int64, int64_t, uint64_t, encode_varint); T(int32, int32_t, uint32_t, encode_varint); T(fixed64, uint64_t, uint64_t, encode_fixed64); T(fixed32, uint32_t, uint32_t, encode_fixed32); T(bool, bool, bool, encode_varint); T(uint32, uint32_t, uint32_t, encode_varint); T(uint64, uint64_t, uint64_t, encode_varint); T(enum, int32_t, uint32_t, encode_varint); T(sfixed32, int32_t, uint32_t, encode_fixed32); T(sfixed64, int64_t, uint64_t, encode_fixed64); T(sint32, int32_t, upb_zzenc_32, encode_varint); T(sint64, int64_t, upb_zzenc_64, encode_varint); #undef T /* code to build the handlers *************************************************/ static void newhandlers_callback(const void *closure, upb_handlers *h) { UPB_UNUSED(closure); upb_handlers_setstartmsg(h, startmsg, NULL); upb_handlers_setendmsg(h, endmsg, NULL); const upb_msgdef *m = upb_handlers_msgdef(h); upb_msg_field_iter i; for(upb_msg_field_begin(&i, m); !upb_msg_field_done(&i); upb_msg_field_next(&i)) { const upb_fielddef *f = upb_msg_iter_field(&i); bool packed = upb_fielddef_isseq(f) && upb_fielddef_isprimitive(f) && upb_fielddef_packed(f); upb_handlerattr attr; upb_wiretype_t wt = packed ? UPB_WIRE_TYPE_DELIMITED : upb_pb_native_wire_types[upb_fielddef_descriptortype(f)]; // Pre-encode the tag for this field. new_tag(h, f, wt, &attr); if (packed) { upb_handlers_setstartseq(h, f, encode_startdelimfield, &attr); upb_handlers_setendseq(h, f, encode_enddelimfield, &attr); } #define T(upper, lower, upbtype) \ case UPB_DESCRIPTOR_TYPE_##upper: \ if (packed) { \ upb_handlers_set##upbtype(h, f, encode_packed_##lower, &attr); \ } else { \ upb_handlers_set##upbtype(h, f, encode_scalar_##lower, &attr); \ } \ break; switch (upb_fielddef_descriptortype(f)) { T(DOUBLE, double, double); T(FLOAT, float, float); T(INT64, int64, int64); T(INT32, int32, int32); T(FIXED64, fixed64, uint64); T(FIXED32, fixed32, uint32); T(BOOL, bool, bool); T(UINT32, uint32, uint32); T(UINT64, uint64, uint64); T(ENUM, enum, int32); T(SFIXED32, sfixed32, int32); T(SFIXED64, sfixed64, int64); T(SINT32, sint32, int32); T(SINT64, sint64, int64); case UPB_DESCRIPTOR_TYPE_STRING: case UPB_DESCRIPTOR_TYPE_BYTES: upb_handlers_setstartstr(h, f, encode_startstr, &attr); upb_handlers_setendstr(h, f, encode_enddelimfield, &attr); upb_handlers_setstring(h, f, encode_strbuf, &attr); break; case UPB_DESCRIPTOR_TYPE_MESSAGE: upb_handlers_setstartsubmsg(h, f, encode_startdelimfield, &attr); upb_handlers_setendsubmsg(h, f, encode_enddelimfield, &attr); break; case UPB_DESCRIPTOR_TYPE_GROUP: { // Endgroup takes a different tag (wire_type = END_GROUP). upb_handlerattr attr2; new_tag(h, f, UPB_WIRE_TYPE_END_GROUP, &attr2); upb_handlers_setstartsubmsg(h, f, encode_startgroup, &attr); upb_handlers_setendsubmsg(h, f, encode_endgroup, &attr2); upb_handlerattr_uninit(&attr2); break; } } #undef T upb_handlerattr_uninit(&attr); } } void upb_pb_encoder_reset(upb_pb_encoder *e) { e->segptr = NULL; e->top = NULL; e->depth = 0; } /* public API *****************************************************************/ const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m, const void *owner) { return upb_handlers_newfrozen(m, owner, newhandlers_callback, NULL); } upb_pb_encoder *upb_pb_encoder_create(upb_env *env, const upb_handlers *h, upb_bytessink *output) { const size_t initial_bufsize = 256; const size_t initial_segbufsize = 16; // TODO(haberman): make this configurable. const size_t stack_size = 64; #ifndef NDEBUG const size_t size_before = upb_env_bytesallocated(env); #endif upb_pb_encoder *e = upb_env_malloc(env, sizeof(upb_pb_encoder)); if (!e) return NULL; e->buf = upb_env_malloc(env, initial_bufsize); e->segbuf = upb_env_malloc(env, initial_segbufsize * sizeof(*e->segbuf)); e->stack = upb_env_malloc(env, stack_size * sizeof(*e->stack)); if (!e->buf || !e->segbuf || !e->stack) { return NULL; } e->limit = e->buf + initial_bufsize; e->seglimit = e->segbuf + initial_segbufsize; e->stacklimit = e->stack + stack_size; upb_pb_encoder_reset(e); upb_sink_reset(&e->input_, h, e); e->env = env; e->output_ = output; e->subc = output->closure; e->ptr = e->buf; // If this fails, increase the value in encoder.h. assert(upb_env_bytesallocated(env) - size_before <= UPB_PB_ENCODER_SIZE); return e; } upb_sink *upb_pb_encoder_input(upb_pb_encoder *e) { return &e->input_; }