From cca4818eb7769d6e776bdc30516a5f871f1d6393 Mon Sep 17 00:00:00 2001 From: Joshua Haberman Date: Sat, 31 Mar 2012 12:17:32 -0700 Subject: Sync from internal Google development. --- upb/stdc/README | 15 +++++ upb/stdc/error.c | 44 ++++++++++++++ upb/stdc/error.h | 27 +++++++++ upb/stdc/io.c | 175 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ upb/stdc/io.h | 73 +++++++++++++++++++++++ 5 files changed, 334 insertions(+) create mode 100644 upb/stdc/README create mode 100644 upb/stdc/error.c create mode 100644 upb/stdc/error.h create mode 100644 upb/stdc/io.c create mode 100644 upb/stdc/io.h (limited to 'upb/stdc') diff --git a/upb/stdc/README b/upb/stdc/README new file mode 100644 index 0000000..1815af4 --- /dev/null +++ b/upb/stdc/README @@ -0,0 +1,15 @@ +This directory contains code that is ANSI C but uses parts of the +standard library that are not available to very limited environments +like Linux Kernel modules. The standard calls environments like this +"freestanding implementations." + +This does *not* imply that the upb core can be compiled directly on a +freestanding implementation. Even the core uses library functions +that are not directly available on freestanding implementations +(notably malloc()/free(), vsnprintf(), and assert()). So compiling on +freestanding implementations may require implementing compatibility +versions of functions like malloc(). + +Also, Linux is not technically a freestanding implementation either, +since it does not accept functions that return float or double on +x86-64 (these use SSE registers which are disabled in kernel mode). diff --git a/upb/stdc/error.c b/upb/stdc/error.c new file mode 100644 index 0000000..313866c --- /dev/null +++ b/upb/stdc/error.c @@ -0,0 +1,44 @@ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Handling of errno. + */ + +#include "upb/stdc/error.h" + +#include +#include + +void upb_status_fromerrno(upb_status *status, int code) { + if (code != 0 && !upb_errno_is_wouldblock(code)) { + status->error = true; + upb_status_setcode(status, &upb_stdc_errorspace, code); + } +} + +bool upb_errno_is_wouldblock(int code) { + return +#ifdef EAGAIN + code == EAGAIN || +#endif +#ifdef EWOULDBLOCK + code == EWOULDBLOCK || +#endif + false; +} + +bool upb_stdc_codetostr(int code, char *buf, size_t len) { + // strerror() may use static buffers and is not guaranteed to be thread-safe, + // but it appears that it is not subject to buffer overflows in practice, and + // it used by other portable and high-quality software like Lua. For more + // discussion see: http://thread.gmane.org/gmane.comp.lang.lua.general/89506 + char *err = strerror(code); + if (strlen(err) >= len) return false; + strcpy(buf, err); + return true; +} + +upb_errorspace upb_stdc_errorspace = {"stdc", &upb_stdc_codetostr}; diff --git a/upb/stdc/error.h b/upb/stdc/error.h new file mode 100644 index 0000000..9802097 --- /dev/null +++ b/upb/stdc/error.h @@ -0,0 +1,27 @@ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Handling of errno. + */ + +#include "upb/upb.h" + +#ifndef UPB_STDC_ERROR_H_ +#define UPB_STDC_ERROR_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +extern upb_errorspace upb_stdc_errorspace; +void upb_status_fromerrno(upb_status *status, int code); +bool upb_errno_is_wouldblock(int code); + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_STDC_ERROR_H_ */ diff --git a/upb/stdc/io.c b/upb/stdc/io.c new file mode 100644 index 0000000..1abed32 --- /dev/null +++ b/upb/stdc/io.c @@ -0,0 +1,175 @@ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + +#include "upb/stdc/io.h" + +#include "upb/stdc/error.h" + +// We can make this configurable if necessary. +#define BUF_SIZE 32768 + +/* upb_stdio ******************************************************************/ + +int upb_stdio_cmpbuf(const void *_key, const void *_elem) { + const uint64_t *ofs = _key; + const upb_stdio_buf *buf = _elem; + return (*ofs / BUF_SIZE) - (buf->ofs / BUF_SIZE); +} + +static upb_stdio_buf *upb_stdio_findbuf(const upb_stdio *s, uint64_t ofs) { + // TODO: it is probably faster to linear search short lists, and to + // special-case the last one or two bufs. + return bsearch(&ofs, s->bufs, s->nbuf, sizeof(*s->bufs), &upb_stdio_cmpbuf); +} + +static upb_stdio_buf *upb_stdio_rotatebufs(upb_stdio *s) { + upb_stdio_buf **reuse = NULL; // XXX + int num_reused = 0, num_inuse = 0; + + // Could sweep only a subset of bufs if this was a hotspot. + for (int i = 0; i < s->nbuf; i++) { + upb_stdio_buf *buf = s->bufs[i]; + if (buf->refcount > 0) { + s->bufs[num_inuse++] = buf; + } else { + reuse[num_reused++] = buf; + } + } + assert(num_reused + num_inuse == s->nbuf); + memcpy(s->bufs + num_inuse, reuse, num_reused * sizeof(upb_stdio_buf*)); + if (num_reused == 0) { + ++s->nbuf; + s->bufs = realloc(s->bufs, s->nbuf * sizeof(*s->bufs)); + s->bufs[s->nbuf-1] = malloc(sizeof(upb_stdio_buf) + BUF_SIZE); + return s->bufs[s->nbuf-1]; + } + return s->bufs[s->nbuf-num_reused]; +} + +void upb_stdio_discard(void *src, uint64_t ofs) { + (void)src; + (void)ofs; +} + +upb_bytesuccess_t upb_stdio_fetch(void *src, uint64_t ofs, size_t *bytes_read) { + (void)ofs; + upb_stdio *stdio = (upb_stdio*)src; + upb_stdio_buf *buf = upb_stdio_rotatebufs(stdio); +retry: + *bytes_read = fread(&buf->data, 1, BUF_SIZE, stdio->file); + buf->len = *bytes_read; + if (*bytes_read < (size_t)BUF_SIZE) { + // Error or EOF. + if (feof(stdio->file)) { + upb_status_seteof(&stdio->src.status); + return UPB_BYTE_EOF; + } + if (ferror(stdio->file)) { +#ifdef EINTR + // If we encounter a client who doesn't want to retry EINTR, we can easily + // add a boolean property of the stdio that controls this behavior. + if (errno == EINTR) { + clearerr(stdio->file); + goto retry; + } +#endif + upb_status_fromerrno(&stdio->src.status, errno); + return upb_errno_is_wouldblock(errno) ? + UPB_BYTE_WOULDBLOCK : UPB_BYTE_ERROR; + } + assert(false); + } + return UPB_BYTE_OK; +} + +void upb_stdio_copy(const void *src, uint64_t ofs, size_t len, char *dst) { + upb_stdio_buf *buf = upb_stdio_findbuf(src, ofs); + ofs -= buf->ofs; + memcpy(dst, buf->data + ofs, BUF_SIZE - ofs); + len -= (BUF_SIZE - ofs); + dst += (BUF_SIZE - ofs); + while (len > 0) { + ++buf; + size_t bytes = UPB_MIN(len, BUF_SIZE); + memcpy(dst, buf->data, bytes); + len -= bytes; + dst += bytes; + } +} + +const char *upb_stdio_getptr(const void *src, uint64_t ofs, size_t *len) { + upb_stdio_buf *buf = upb_stdio_findbuf(src, ofs); + ofs -= buf->ofs; + *len = BUF_SIZE - ofs; + return &buf->data[ofs]; +} + +#if 0 +upb_strlen_t upb_stdio_putstr(upb_bytesink *sink, upb_string *str, upb_status *status) { + upb_stdio *stdio = (upb_stdio*)((char*)sink - offsetof(upb_stdio, sink)); + upb_strlen_t len = upb_string_len(str); + upb_strlen_t written = fwrite(upb_string_getrobuf(str), 1, len, stdio->file); + if (written < len) { + upb_status_setf(status, UPB_ERROR, "Error writing to stdio stream."); + return -1; + } + return written; +} + +uint32_t upb_stdio_vprintf(upb_bytesink *sink, upb_status *status, + const char *fmt, va_list args) { + upb_stdio *stdio = (upb_stdio*)((char*)sink - offsetof(upb_stdio, sink)); + int written = vfprintf(stdio->file, fmt, args); + if (written < 0) { + upb_status_seterrf(status, "Error writing to stdio stream."); + return -1; + } + return written; +} +#endif + +void upb_stdio_init(upb_stdio *stdio) { + static upb_bytesrc_vtbl bytesrc_vtbl = { + &upb_stdio_fetch, + &upb_stdio_discard, + &upb_stdio_copy, + &upb_stdio_getptr, + }; + upb_bytesrc_init(&stdio->src, &bytesrc_vtbl); + + //static upb_bytesink_vtbl bytesink_vtbl = { + // upb_stdio_putstr, + // upb_stdio_vprintf + //}; + //upb_bytesink_init(&stdio->bytesink, &bytesink_vtbl); +} + +void upb_stdio_reset(upb_stdio* stdio, FILE *file) { + stdio->file = file; + stdio->should_close = false; +} + +void upb_stdio_open(upb_stdio *stdio, const char *filename, const char *mode, + upb_status *s) { + FILE *f = fopen(filename, mode); + if (!f) { + upb_status_fromerrno(s, errno); + return; + } + setvbuf(stdio->file, NULL, _IONBF, 0); // Disable buffering; we do our own. + upb_stdio_reset(stdio, f); + stdio->should_close = true; +} + +void upb_stdio_uninit(upb_stdio *stdio) { + // Can't report status; caller should flush() to ensure data is written. + if (stdio->should_close) fclose(stdio->file); + stdio->file = NULL; +} + +upb_bytesrc* upb_stdio_bytesrc(upb_stdio *stdio) { return &stdio->src; } +upb_bytesink* upb_stdio_bytesink(upb_stdio *stdio) { return &stdio->sink; } diff --git a/upb/stdc/io.h b/upb/stdc/io.h new file mode 100644 index 0000000..fd19bef --- /dev/null +++ b/upb/stdc/io.h @@ -0,0 +1,73 @@ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * ANSI C file I/O. + */ + +#ifndef UPB_STDC_IO_H_ +#define UPB_STDC_IO_H_ + +#include +#include "upb/bytestream.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/* upb_stdio ******************************************************************/ + +// bytesrc/bytesink for ANSI C stdio, which is less efficient than posixfd, but +// more portable. +// +// Specifically, stdio functions acquire locks on every operation (unless you +// use the f{read,write,...}_unlocked variants, which are not standard) and +// performs redundant buffering (unless you disable it with setvbuf(), but we +// can only do this on newly-opened filehandles). + +typedef struct { + uint64_t ofs; + size_t len; + uint32_t refcount; + char data[]; +} upb_stdio_buf; + +// We use a single object for both bytesrc and bytesink for simplicity. +// The object is still not thread-safe, and may only be used by one reader +// and one writer at a time. +typedef struct { + upb_bytesrc src; + upb_bytesink sink; + FILE *file; + bool should_close; + upb_stdio_buf **bufs; + int nbuf; + uint32_t szbuf; +} upb_stdio; + +void upb_stdio_init(upb_stdio *stdio); +// Caller should call upb_stdio_flush prior to calling this to ensure that +// all data is flushed, otherwise data can be silently dropped if an error +// occurs flushing the remaining buffers. +void upb_stdio_uninit(upb_stdio *stdio); + +// Resets the object to read/write to the given "file." The caller is +// responsible for closing the file, which must outlive this object. +void upb_stdio_reset(upb_stdio *stdio, FILE *file); + +// As an alternative to upb_stdio_reset(), initializes the object by opening a +// file, and will handle closing it. This may result in more efficient I/O +// than the previous since we can call setvbuf() to disable buffering. +void upb_stdio_open(upb_stdio *stdio, const char *filename, const char *mode, + upb_status *s); + +upb_bytesrc *upb_stdio_bytesrc(upb_stdio *stdio); +upb_bytesink *upb_stdio_bytesink(upb_stdio *stdio); + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_STDC_IO_H_ */ -- cgit v1.2.3