torproject/tor
1
0
Fork 0
mirror of https://gitlab.torproject.org/tpo/core/tor.git synced 2024-12-06 19:41:15 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge branch 'ahf_prop278_21662_squashed'

Browse Source
This commit is contained in:
Nick Mathewson
2017-04-25 08:12:59 -04:00
parent db95a6e171 2210b330e7
commit ba405f86bf
32 changed files with 2474 additions and 421 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/common/Makefile.nmake
+2 -2
View File
@@ -7,8 +7,8 @@ LIBOR_OBJECTS = address.obj backtrace.obj compat.obj container.obj di_ops.obj \
log.obj memarea.obj mempool.obj procmon.obj sandbox.obj util.obj \
util_codedigest.obj
LIBOR_CRYPTO_OBJECTS = aes.obj crypto.obj crypto_format.obj torgzip.obj tortls.obj \
crypto_curve25519.obj curve25519-donna.obj
LIBOR_CRYPTO_OBJECTS = aes.obj crypto.obj crypto_format.obj compress.obj compress_zlib.obj \
tortls.obj crypto_curve25519.obj curve25519-donna.obj
LIBOR_EVENT_OBJECTS = compat_libevent.obj
src/common/compress.c
+377
View File
@@ -0,0 +1,377 @@
/* Copyright (c) 2004, Roger Dingledine.
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress.c
* \brief Common compression API.
**/
#include "orconfig.h"
#include <stdlib.h>
#include <stdio.h>
#include <assert.h>
#include <string.h>
#include "torint.h"
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif
#include "util.h"
#include "torlog.h"
#include "compress.h"
#include "compress_lzma.h"
#include "compress_zlib.h"
#include "compress_zstd.h"
/** @{ */
/* These macros define the maximum allowable compression factor. Anything of
* size greater than CHECK_FOR_COMPRESSION_BOMB_AFTER is not allowed to
* have an uncompression factor (uncompressed size:compressed size ratio) of
* any greater than MAX_UNCOMPRESSION_FACTOR.
*
* Picking a value for MAX_UNCOMPRESSION_FACTOR is a trade-off: we want it to
* be small to limit the attack multiplier, but we also want it to be large
* enough so that no legitimate document --even ones we might invent in the
* future -- ever compresses by a factor of greater than
* MAX_UNCOMPRESSION_FACTOR. Within those parameters, there's a reasonably
* large range of possible values. IMO, anything over 8 is probably safe; IMO
* anything under 50 is probably sufficient.
*/
#define MAX_UNCOMPRESSION_FACTOR 25
#define CHECK_FOR_COMPRESSION_BOMB_AFTER (1024*64)
/** @} */
/** Return true if uncompressing an input of size <b>in_size</b> to an input of
* size at least <b>size_out</b> looks like a compression bomb. */
int
tor_compress_is_compression_bomb(size_t size_in, size_t size_out)
{
if (size_in == 0 || size_out < CHECK_FOR_COMPRESSION_BOMB_AFTER)
return 0;
return (size_out / size_in > MAX_UNCOMPRESSION_FACTOR);
}
/** Given <b>in_len</b> bytes at <b>in</b>, compress them into a newly
* allocated buffer, using the method described in <b>method</b>. Store the
* compressed string in *<b>out</b>, and its length in *<b>out_len</b>.
* Return 0 on success, -1 on failure.
*/
int
tor_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method)
{
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_compress(out, out_len, in, in_len, method);
case LZMA_METHOD:
return tor_lzma_compress(out, out_len, in, in_len, method);
case ZSTD_METHOD:
return tor_zstd_compress(out, out_len, in, in_len, method);
case NO_METHOD:
case UNKNOWN_METHOD:
default:
return -1;
}
}
/** Given zero or more zlib-compressed or gzip-compressed strings of
* total length
* <b>in_len</b> bytes at <b>in</b>, uncompress them into a newly allocated
* buffer, using the method described in <b>method</b>. Store the uncompressed
* string in *<b>out</b>, and its length in *<b>out_len</b>. Return 0 on
* success, -1 on failure.
*
* If <b>complete_only</b> is true, we consider a truncated input as a
* failure; otherwise we decompress as much as we can. Warn about truncated
* or corrupt inputs at <b>protocol_warn_level</b>.
*/
int
tor_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level)
{
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_uncompress(out, out_len, in, in_len,
method,
complete_only,
protocol_warn_level);
case LZMA_METHOD:
return tor_lzma_uncompress(out, out_len, in, in_len,
method,
complete_only,
protocol_warn_level);
case ZSTD_METHOD:
return tor_zstd_uncompress(out, out_len, in, in_len,
method,
complete_only,
protocol_warn_level);
case NO_METHOD:
case UNKNOWN_METHOD:
default:
return -1;
}
}
/** Try to tell whether the <b>in_len</b>-byte string in <b>in</b> is likely
* to be compressed or not. If it is, return the likeliest compression method.
* Otherwise, return UNKNOWN_METHOD.
*/
compress_method_t
detect_compression_method(const char *in, size_t in_len)
{
if (in_len > 2 && fast_memeq(in, "\x1f\x8b", 2)) {
return GZIP_METHOD;
} else if (in_len > 2 && (in[0] & 0x0f) == 8 &&
(ntohs(get_uint16(in)) % 31) == 0) {
return ZLIB_METHOD;
} else if (in_len > 3 &&
fast_memeq(in, "\x5d\x00\x00\x00", 4)) {
return LZMA_METHOD;
} else if (in_len > 3 &&
fast_memeq(in, "\x28\xb5\x2f\xfd", 4)) {
return ZSTD_METHOD;
} else {
return UNKNOWN_METHOD;
}
}
/** Return 1 if a given <b>method</b> is supported; otherwise 0. */
int
tor_compress_supports_method(compress_method_t method)
{
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_method_supported();
case LZMA_METHOD:
return tor_lzma_method_supported();
case ZSTD_METHOD:
return tor_zstd_method_supported();
case NO_METHOD:
case UNKNOWN_METHOD:
default:
return 0;
}
}
/** Return a string representation of the version of the library providing the
* compression method given in <b>method</b>. Returns NULL if <b>method</b> is
* unknown or unsupported. */
const char *
tor_compress_version_str(compress_method_t method)
{
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_get_version_str();
case LZMA_METHOD:
return tor_lzma_get_version_str();
case ZSTD_METHOD:
return tor_zstd_get_version_str();
case NO_METHOD:
case UNKNOWN_METHOD:
default:
return NULL;
}
}
/** Return a string representation of the version of the library, found at
* compile time, providing the compression method given in <b>method</b>.
* Returns NULL if <b>method</b> is unknown or unsupported. */
const char *
tor_compress_header_version_str(compress_method_t method)
{
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_get_header_version_str();
case LZMA_METHOD:
return tor_lzma_get_header_version_str();
case ZSTD_METHOD:
return tor_zstd_get_header_version_str();
case NO_METHOD:
case UNKNOWN_METHOD:
default:
return NULL;
}
}
/** Return the approximate number of bytes allocated for all
* supported compression schemas. */
size_t
tor_compress_get_total_allocation(void)
{
return tor_zlib_get_total_allocation() +
tor_lzma_get_total_allocation() +
tor_zstd_get_total_allocation();
}
/** Internal state for an incremental compression/decompression. The body of
* this struct is not exposed. */
struct tor_compress_state_t {
compress_method_t method; /**< The compression method. */
union {
tor_zlib_compress_state_t *zlib_state;
tor_lzma_compress_state_t *lzma_state;
tor_zstd_compress_state_t *zstd_state;
} u; /**< Compression backend state. */
};
/** Construct and return a tor_compress_state_t object using <b>method</b>. If
* <b>compress</b>, it's for compression; otherwise it's for decompression. */
tor_compress_state_t *
tor_compress_new(int compress, compress_method_t method,
compression_level_t compression_level)
{
tor_compress_state_t *state;
state = tor_malloc_zero(sizeof(tor_compress_state_t));
state->method = method;
switch (method) {
case GZIP_METHOD:
case ZLIB_METHOD: {
tor_zlib_compress_state_t *zlib_state =
tor_zlib_compress_new(compress, method, compression_level);
if (zlib_state == NULL)
goto err;
state->u.zlib_state = zlib_state;
break;
}
case LZMA_METHOD: {
tor_lzma_compress_state_t *lzma_state =
tor_lzma_compress_new(compress, method, compression_level);
if (lzma_state == NULL)
goto err;
state->u.lzma_state = lzma_state;
break;
}
case ZSTD_METHOD: {
tor_zstd_compress_state_t *zstd_state =
tor_zstd_compress_new(compress, method, compression_level);
if (zstd_state == NULL)
goto err;
state->u.zstd_state = zstd_state;
break;
}
case NO_METHOD:
case UNKNOWN_METHOD:
goto err;
}
return state;
err:
tor_free(state);
return NULL;
}
/** Compress/decompress some bytes using <b>state</b>. Read up to
* *<b>in_len</b> bytes from *<b>in</b>, and write up to *<b>out_len</b> bytes
* to *<b>out</b>, adjusting the values as we go. If <b>finish</b> is true,
* we've reached the end of the input.
*
* Return TOR_COMPRESS_DONE if we've finished the entire
* compression/decompression.
* Return TOR_COMPRESS_OK if we're processed everything from the input.
* Return TOR_COMPRESS_BUFFER_FULL if we're out of space on <b>out</b>.
* Return TOR_COMPRESS_ERROR if the stream is corrupt.
*/
tor_compress_output_t
tor_compress_process(tor_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish)
{
tor_assert(state != NULL);
switch (state->method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_compress_process(state->u.zlib_state,
out, out_len, in, in_len,
finish);
case LZMA_METHOD:
return tor_lzma_compress_process(state->u.lzma_state,
out, out_len, in, in_len,
finish);
case ZSTD_METHOD:
return tor_zstd_compress_process(state->u.zstd_state,
out, out_len, in, in_len,
finish);
case NO_METHOD:
case UNKNOWN_METHOD:
goto err;
}
err:
return TOR_COMPRESS_ERROR;
}
/** Deallocate <b>state</b>. */
void
tor_compress_free(tor_compress_state_t *state)
{
if (state == NULL)
return;
switch (state->method) {
case GZIP_METHOD:
case ZLIB_METHOD:
tor_zlib_compress_free(state->u.zlib_state);
break;
case LZMA_METHOD:
tor_lzma_compress_free(state->u.lzma_state);
break;
case ZSTD_METHOD:
tor_zstd_compress_free(state->u.zstd_state);
break;
case NO_METHOD:
case UNKNOWN_METHOD:
break;
}
tor_free(state);
}
/** Return the approximate number of bytes allocated for <b>state</b>. */
size_t
tor_compress_state_size(const tor_compress_state_t *state)
{
tor_assert(state != NULL);
switch (state->method) {
case GZIP_METHOD:
case ZLIB_METHOD:
return tor_zlib_compress_state_size(state->u.zlib_state);
case LZMA_METHOD:
return tor_lzma_compress_state_size(state->u.lzma_state);
case ZSTD_METHOD:
return tor_zstd_compress_state_size(state->u.zstd_state);
case NO_METHOD:
case UNKNOWN_METHOD:
goto err;
}
err:
return 0;
}
src/common/compress.h
+83
View File
@@ -0,0 +1,83 @@
/* Copyright (c) 2003, Roger Dingledine
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress.h
* \brief Headers for compress.c
**/
#ifndef TOR_COMPRESS_H
#define TOR_COMPRESS_H
/** Enumeration of what kind of compression to use. Only ZLIB_METHOD and
* GZIP_METHOD is guaranteed to be supported by the compress/uncompress
* functions here. Call tor_compress_supports_method() to check if a given
* compression schema is supported by Tor. */
typedef enum {
NO_METHOD=0,
GZIP_METHOD=1,
ZLIB_METHOD=2,
LZMA_METHOD=3,
ZSTD_METHOD=4,
UNKNOWN_METHOD=5
} compress_method_t;
/**
* Enumeration to define tradeoffs between memory usage and compression level.
* HIGH_COMPRESSION saves the most bandwidth; LOW_COMPRESSION saves the most
* memory.
**/
typedef enum {
HIGH_COMPRESSION, MEDIUM_COMPRESSION, LOW_COMPRESSION
} compression_level_t;
int tor_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method);
int tor_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level);
compress_method_t detect_compression_method(const char *in, size_t in_len);
int tor_compress_is_compression_bomb(size_t size_in, size_t size_out);
int tor_compress_supports_method(compress_method_t method);
const char *tor_compress_version_str(compress_method_t method);
const char *tor_compress_header_version_str(compress_method_t method);
size_t tor_compress_get_total_allocation(void);
/** Return values from tor_compress_process; see that function's documentation
* for details. */
typedef enum {
TOR_COMPRESS_OK,
TOR_COMPRESS_DONE,
TOR_COMPRESS_BUFFER_FULL,
TOR_COMPRESS_ERROR
} tor_compress_output_t;
/** Internal state for an incremental compression/decompression. */
typedef struct tor_compress_state_t tor_compress_state_t;
tor_compress_state_t *tor_compress_new(int compress,
compress_method_t method,
compression_level_t level);
tor_compress_output_t tor_compress_process(tor_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish);
void tor_compress_free(tor_compress_state_t *state);
size_t tor_compress_state_size(const tor_compress_state_t *state);
#endif // TOR_COMPRESS_H.
src/common/compress_lzma.c
+605
View File
@@ -0,0 +1,605 @@
/* Copyright (c) 2004, Roger Dingledine.
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress_lzma.c
* \brief Compression backend for LZMA.
*
* This module should never be invoked directly. Use the compress module
* instead.
**/
#include "orconfig.h"
#include "util.h"
#include "torlog.h"
#include "compress.h"
#include "compress_lzma.h"
#ifdef HAVE_LZMA
#include <lzma.h>
#endif
/** Total number of bytes allocated for LZMA state. */
static size_t total_lzma_allocation = 0;
#ifdef HAVE_LZMA
/** Given <b>level</b> return the memory level. */
static int
memory_level(compression_level_t level)
{
switch (level) {
default:
case HIGH_COMPRESSION: return 9;
case MEDIUM_COMPRESSION: return 6;
case LOW_COMPRESSION: return 3;
}
}
/** Convert a given <b>error</b> to a human readable error string. */
static const char *
lzma_error_str(lzma_ret error)
{
switch (error) {
case LZMA_OK:
return "Operation completed successfully";
case LZMA_STREAM_END:
return "End of stream";
case LZMA_NO_CHECK:
return "Input stream lacks integrity check";
case LZMA_UNSUPPORTED_CHECK:
return "Unable to calculate integrity check";
case LZMA_GET_CHECK:
return "Integrity check available";
case LZMA_MEM_ERROR:
return "Unable to allocate memory";
case LZMA_MEMLIMIT_ERROR:
return "Memory limit reached";
case LZMA_FORMAT_ERROR:
return "Unknown file format";
case LZMA_OPTIONS_ERROR:
return "Unsupported options";
case LZMA_DATA_ERROR:
return "Corrupt input data";
case LZMA_BUF_ERROR:
return "Unable to progress";
case LZMA_PROG_ERROR:
return "Programming error";
default:
return "Unknown LZMA error";
}
}
#endif // HAVE_LZMA.
/** Return 1 if LZMA compression is supported; otherwise 0. */
int
tor_lzma_method_supported(void)
{
#ifdef HAVE_LZMA
return 1;
#else
return 0;
#endif
}
/** Return a string representation of the version of the currently running
* version of liblzma. Returns NULL if LZMA is unsupported. */
const char *
tor_lzma_get_version_str(void)
{
#ifdef HAVE_LZMA
return lzma_version_string();
#else
return NULL;
#endif
}
/** Return a string representation of the version of liblzma used at
* compilation time. Returns NULL if LZMA is unsupported. */
const char *
tor_lzma_get_header_version_str(void)
{
#ifdef HAVE_LZMA
return LZMA_VERSION_STRING;
#else
return NULL;
#endif
}
/** Given <b>in_len</b> bytes at <b>in</b>, compress them into a newly
* allocated buffer, using the LZMA method. Store the compressed string in
* *<b>out</b>, and its length in *<b>out_len</b>. Return 0 on success, -1 on
* failure.
*/
int
tor_lzma_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method)
{
#ifdef HAVE_LZMA
lzma_stream stream = LZMA_STREAM_INIT;
lzma_options_lzma stream_options;
lzma_ret retval;
lzma_action action;
size_t out_size, old_size;
off_t offset;
tor_assert(out);
tor_assert(out_len);
tor_assert(in);
tor_assert(in_len < UINT_MAX);
tor_assert(method == LZMA_METHOD);
stream.next_in = (unsigned char *)in;
stream.avail_in = in_len;
lzma_lzma_preset(&stream_options,
memory_level(HIGH_COMPRESSION));
retval = lzma_alone_encoder(&stream, &stream_options);
if (retval != LZMA_OK) {
log_warn(LD_GENERAL, "Error from LZMA encoder: %s (%u).",
lzma_error_str(retval), retval);
goto err;
}
out_size = in_len / 2;
if (out_size < 1024)
out_size = 1024;
*out = tor_malloc(out_size);
stream.next_out = (unsigned char *)*out;
stream.avail_out = out_size;
action = LZMA_RUN;
while (1) {
retval = lzma_code(&stream, action);
switch (retval) {
case LZMA_OK:
action = LZMA_FINISH;
break;
case LZMA_STREAM_END:
goto done;
case LZMA_BUF_ERROR:
offset = stream.next_out - ((unsigned char *)*out);
old_size = out_size;
out_size *= 2;
if (out_size < old_size) {
log_warn(LD_GENERAL, "Size overflow in LZMA compression.");
goto err;
}
*out = tor_realloc(*out, out_size);
stream.next_out = (unsigned char *)(*out + offset);
if (out_size - offset > UINT_MAX) {
log_warn(LD_BUG, "Ran over unsigned int limit of LZMA while "
"compressing.");
goto err;
}
stream.avail_out = (unsigned int)(out_size - offset);
break;
// We list all the possible values of `lzma_ret` here to silence the
// `switch-enum` warning and to detect if a new member was added.
case LZMA_NO_CHECK:
case LZMA_UNSUPPORTED_CHECK:
case LZMA_GET_CHECK:
case LZMA_MEM_ERROR:
case LZMA_MEMLIMIT_ERROR:
case LZMA_FORMAT_ERROR:
case LZMA_OPTIONS_ERROR:
case LZMA_DATA_ERROR:
case LZMA_PROG_ERROR:
default:
log_warn(LD_GENERAL, "LZMA compression didn't finish: %s.",
lzma_error_str(retval));
goto err;
}
}
done:
*out_len = stream.total_out;
lzma_end(&stream);
if (tor_compress_is_compression_bomb(*out_len, in_len)) {
log_warn(LD_BUG, "We compressed something and got an insanely high "
"compression factor; other Tor instances would think "
"this is a compression bomb.");
goto err;
}
return 0;
err:
lzma_end(&stream);
tor_free(*out);
return -1;
#else // HAVE_LZMA.
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)method;
return -1;
#endif // HAVE_LZMA.
}
/** Given an LZMA compressed string of total length <b>in_len</b> bytes at
* <b>in</b>, uncompress them into a newly allocated buffer. Store the
* uncompressed string in *<b>out</b>, and its length in *<b>out_len</b>.
* Return 0 on success, -1 on failure.
*
* If <b>complete_only</b> is true, we consider a truncated input as a failure;
* otherwise we decompress as much as we can. Warn about truncated or corrupt
* inputs at <b>protocol_warn_level</b>.
*/
int
tor_lzma_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level)
{
#ifdef HAVE_LZMA
lzma_stream stream = LZMA_STREAM_INIT;
lzma_ret retval;
lzma_action action;
size_t out_size, old_size;
off_t offset;
tor_assert(out);
tor_assert(out_len);
tor_assert(in);
tor_assert(in_len < UINT_MAX);
tor_assert(method == LZMA_METHOD);
stream.next_in = (unsigned char *)in;
stream.avail_in = in_len;
// FIXME(ahf): This should be something more sensible than
// UINT64_MAX: See #21665.
retval = lzma_alone_decoder(&stream, UINT64_MAX);
if (retval != LZMA_OK) {
log_warn(LD_GENERAL, "Error from LZMA decoder: %s (%u).",
lzma_error_str(retval), retval);
goto err;
}
out_size = in_len * 2;
if (out_size < 1024)
out_size = 1024;
if (out_size >= SIZE_T_CEILING || out_size > UINT_MAX)
goto err;
*out = tor_malloc(out_size);
stream.next_out = (unsigned char *)*out;
stream.avail_out = out_size;
// FIXME(ahf): We should figure out how to use LZMA_FULL_FLUSH to
// make the partial string read tests.
// action = complete_only ? LZMA_FINISH : LZMA_SYNC_FLUSH. // To do this,
// it seems like we have to use LZMA using their "xz" encoder instead of just
// regular LZMA.
(void)complete_only;
action = LZMA_FINISH;
while (1) {
retval = lzma_code(&stream, action);
switch (retval) {
case LZMA_STREAM_END:
if (stream.avail_in == 0)
goto done;
// We might have more data here. Reset our stream.
lzma_end(&stream);
retval = lzma_alone_decoder(&stream, UINT64_MAX);
if (retval != LZMA_OK) {
log_warn(LD_GENERAL, "Error from LZMA decoder: %s (%u).",
lzma_error_str(retval), retval);
goto err;
}
break;
case LZMA_OK:
break;
case LZMA_BUF_ERROR:
if (stream.avail_out > 0) {
log_fn(protocol_warn_level, LD_PROTOCOL,
"possible truncated or corrupt LZMA data.");
goto err;
}
offset = stream.next_out - (unsigned char *)*out;
old_size = out_size;
out_size *= 2;
if (out_size < old_size) {
log_warn(LD_GENERAL, "Size overflow in LZMA uncompression.");
goto err;
}
if (tor_compress_is_compression_bomb(in_len, out_size)) {
log_warn(LD_GENERAL, "Input looks like a possible LZMA compression "
"bomb. Not proceeding.");
goto err;
}
if (out_size >= SIZE_T_CEILING) {
log_warn(LD_BUG, "Hit SIZE_T_CEILING limit while uncompressing "
"LZMA data.");
goto err;
}
*out = tor_realloc(*out, out_size);
stream.next_out = (unsigned char *)(*out + offset);
if (out_size - offset > UINT_MAX) {
log_warn(LD_BUG, "Ran over unsigned int limit of LZMA while "
"uncompressing.");
goto err;
}
stream.avail_out = (unsigned int)(out_size - offset);
break;
// We list all the possible values of `lzma_ret` here to silence the
// `switch-enum` warning and to detect if a new member was added.
case LZMA_NO_CHECK:
case LZMA_UNSUPPORTED_CHECK:
case LZMA_GET_CHECK:
case LZMA_MEM_ERROR:
case LZMA_MEMLIMIT_ERROR:
case LZMA_FORMAT_ERROR:
case LZMA_OPTIONS_ERROR:
case LZMA_DATA_ERROR:
case LZMA_PROG_ERROR:
default:
log_warn(LD_GENERAL, "LZMA decompression didn't finish: %s.",
lzma_error_str(retval));
goto err;
}
}
done:
*out_len = stream.next_out - (unsigned char*)*out;
lzma_end(&stream);
// NUL-terminate our output.
if (out_size == *out_len)
*out = tor_realloc(*out, out_size + 1);
(*out)[*out_len] = '\0';
return 0;
err:
lzma_end(&stream);
tor_free(*out);
return -1;
#else // HAVE_LZMA.
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)method;
(void)complete_only;
(void)protocol_warn_level;
return -1;
#endif // HAVE_LZMA.
}
/** Internal LZMA state for incremental compression/decompression.
* The body of this struct is not exposed. */
struct tor_lzma_compress_state_t {
#ifdef HAVE_LZMA
lzma_stream stream; /**< The LZMA stream. */
#endif
int compress; /**< True if we are compressing; false if we are inflating */
/** Number of bytes read so far. Used to detect compression bombs. */
size_t input_so_far;
/** Number of bytes written so far. Used to detect compression bombs. */
size_t output_so_far;
/** Approximate number of bytes allocated for this object. */
size_t allocation;
};
/** Construct and return a tor_lzma_compress_state_t object using
* <b>method</b>. If <b>compress</b>, it's for compression; otherwise it's for
* decompression. */
tor_lzma_compress_state_t *
tor_lzma_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level)
{
tor_assert(method == LZMA_METHOD);
#ifdef HAVE_LZMA
tor_lzma_compress_state_t *result;
lzma_ret retval;
lzma_options_lzma stream_options;
// Note that we do not explicitly initialize the lzma_stream object here,
// since the LZMA_STREAM_INIT "just" initializes all members to 0, which is
// also what `tor_malloc_zero()` does.
result = tor_malloc_zero(sizeof(tor_lzma_compress_state_t));
result->compress = compress;
// FIXME(ahf): We should either try to do the pre-calculation that is done
// with the zlib backend or use a custom allocator here where we pass our
// tor_lzma_compress_state_t as the opaque value.
result->allocation = 0;
if (compress) {
lzma_lzma_preset(&stream_options,
memory_level(compression_level));
retval = lzma_alone_encoder(&result->stream, &stream_options);
if (retval != LZMA_OK) {
log_warn(LD_GENERAL, "Error from LZMA encoder: %s (%u).",
lzma_error_str(retval), retval);
goto err;
}
} else {
// FIXME(ahf): This should be something more sensible than
// UINT64_MAX: See #21665.
retval = lzma_alone_decoder(&result->stream, UINT64_MAX);
if (retval != LZMA_OK) {
log_warn(LD_GENERAL, "Error from LZMA decoder: %s (%u).",
lzma_error_str(retval), retval);
goto err;
}
}
return result;
err:
tor_free(result);
return NULL;
#else // HAVE_LZMA.
(void)compress;
(void)method;
(void)compression_level;
return NULL;
#endif // HAVE_LZMA.
}
/** Compress/decompress some bytes using <b>state</b>. Read up to
* *<b>in_len</b> bytes from *<b>in</b>, and write up to *<b>out_len</b> bytes
* to *<b>out</b>, adjusting the values as we go. If <b>finish</b> is true,
* we've reached the end of the input.
*
* Return TOR_COMPRESS_DONE if we've finished the entire
* compression/decompression.
* Return TOR_COMPRESS_OK if we're processed everything from the input.
* Return TOR_COMPRESS_BUFFER_FULL if we're out of space on <b>out</b>.
* Return TOR_COMPRESS_ERROR if the stream is corrupt.
*/
tor_compress_output_t
tor_lzma_compress_process(tor_lzma_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish)
{
#ifdef HAVE_LZMA
lzma_ret retval;
lzma_action action;
tor_assert(state != NULL);
tor_assert(*in_len <= UINT_MAX);
tor_assert(*out_len <= UINT_MAX);
state->stream.next_in = (unsigned char *)*in;
state->stream.avail_in = *in_len;
state->stream.next_out = (unsigned char *)*out;
state->stream.avail_out = *out_len;
action = finish ? LZMA_FINISH : LZMA_RUN;
retval = lzma_code(&state->stream, action);
state->input_so_far += state->stream.next_in - ((unsigned char *)*in);
state->output_so_far += state->stream.next_out - ((unsigned char *)*out);
*out = (char *)state->stream.next_out;
*out_len = state->stream.avail_out;
*in = (const char *)state->stream.next_in;
*in_len = state->stream.avail_in;
if (! state->compress &&
tor_compress_is_compression_bomb(state->input_so_far,
state->output_so_far)) {
log_warn(LD_DIR, "Possible compression bomb; abandoning stream.");
return TOR_COMPRESS_ERROR;
}
switch (retval) {
case LZMA_OK:
if (state->stream.avail_out == 0 || finish)
return TOR_COMPRESS_BUFFER_FULL;
return TOR_COMPRESS_OK;
case LZMA_BUF_ERROR:
if (state->stream.avail_in == 0 && !finish)
return TOR_COMPRESS_OK;
return TOR_COMPRESS_BUFFER_FULL;
case LZMA_STREAM_END:
return TOR_COMPRESS_DONE;
// We list all the possible values of `lzma_ret` here to silence the
// `switch-enum` warning and to detect if a new member was added.
case LZMA_NO_CHECK:
case LZMA_UNSUPPORTED_CHECK:
case LZMA_GET_CHECK:
case LZMA_MEM_ERROR:
case LZMA_MEMLIMIT_ERROR:
case LZMA_FORMAT_ERROR:
case LZMA_OPTIONS_ERROR:
case LZMA_DATA_ERROR:
case LZMA_PROG_ERROR:
default:
log_warn(LD_GENERAL, "LZMA %s didn't finish: %s.",
state->compress ? "compression" : "decompression",
lzma_error_str(retval));
return TOR_COMPRESS_ERROR;
}
#else // HAVE_LZMA.
(void)state;
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)finish;
return TOR_COMPRESS_ERROR;
#endif // HAVE_LZMA.
}
/** Deallocate <b>state</b>. */
void
tor_lzma_compress_free(tor_lzma_compress_state_t *state)
{
if (state == NULL)
return;
total_lzma_allocation -= state->allocation;
#ifdef HAVE_LZMA
lzma_end(&state->stream);
#endif
tor_free(state);
}
/** Return the approximate number of bytes allocated for <b>state</b>. */
size_t
tor_lzma_compress_state_size(const tor_lzma_compress_state_t *state)
{
tor_assert(state != NULL);
return state->allocation;
}
/** Return the approximate number of bytes allocated for all LZMA states. */
size_t
tor_lzma_get_total_allocation(void)
{
return total_lzma_allocation;
}
src/common/compress_lzma.h
+51
View File
@@ -0,0 +1,51 @@
/* Copyright (c) 2003, Roger Dingledine
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress_lzma.h
* \brief Header for compress_lzma.c
**/
#ifndef TOR_COMPRESS_LZMA_H
#define TOR_COMPRESS_LZMA_H
int tor_lzma_method_supported(void);
const char *tor_lzma_get_version_str(void);
const char *tor_lzma_get_header_version_str(void);
int tor_lzma_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method);
int tor_lzma_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level);
/** Internal state for an incremental LZMA compression/decompression. */
typedef struct tor_lzma_compress_state_t tor_lzma_compress_state_t;
tor_lzma_compress_state_t *
tor_lzma_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level);
tor_compress_output_t
tor_lzma_compress_process(tor_lzma_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish);
void tor_lzma_compress_free(tor_lzma_compress_state_t *state);
size_t tor_lzma_compress_state_size(const tor_lzma_compress_state_t *state);
size_t tor_lzma_get_total_allocation(void);
#endif // TOR_COMPRESS_LZMA_H.
src/common/torgzip.c → src/common/compress_zlib.c
+150 -188
View File
@@ -4,25 +4,19 @@
/* See LICENSE for licensing information */
/**
* \file torgzip.c
* \brief A simple in-memory gzip implementation.
* \file compress_zlib.c
* \brief Compression backend for gzip and zlib.
*
* This module should never be invoked directly. Use the compress module
* instead.
**/
#include "orconfig.h"
#include <stdlib.h>
#include <stdio.h>
#include <assert.h>
#include <string.h>
#include "torint.h"
#ifdef HAVE_NETINET_IN_H
#include <netinet/in.h>
#endif
#include "util.h"
#include "torlog.h"
#include "torgzip.h"
#include "compress.h"
#include "compress_zlib.h"
/* zlib 1.2.4 and 1.2.5 do some "clever" things with macros. Instead of
saying "(defined(FOO) ? FOO : 0)" they like to say "FOO-0", on the theory
@@ -56,6 +50,42 @@ static size_t tor_zlib_state_size_precalc(int inflate,
/** Total number of bytes allocated for zlib state */
static size_t total_zlib_allocation = 0;
/** Given <b>level</b> return the memory level. */
static int
memory_level(compression_level_t level)
{
switch (level) {
default:
case HIGH_COMPRESSION: return 8;
case MEDIUM_COMPRESSION: return 7;
case LOW_COMPRESSION: return 6;
}
}
/** Return the 'bits' value to tell zlib to use <b>method</b>.*/
static inline int
method_bits(compress_method_t method, compression_level_t level)
{
/* Bits+16 means "use gzip" in zlib >= 1.2 */
const int flag = method == GZIP_METHOD ? 16 : 0;
switch (level) {
default:
case HIGH_COMPRESSION: return flag + 15;
case MEDIUM_COMPRESSION: return flag + 13;
case LOW_COMPRESSION: return flag + 11;
}
}
/** Return 1 if zlib/gzip compression is supported; otherwise 0. */
int
tor_zlib_method_supported(void)
{
/* We currently always support zlib/gzip, but we keep this function around in
* case we some day decide to deprecate zlib/gzip support.
*/
return 1;
}
/** Return a string representation of the version of the currently running
* version of zlib. */
const char *
@@ -72,67 +102,13 @@ tor_zlib_get_header_version_str(void)
return ZLIB_VERSION;
}
/** Return the 'bits' value to tell zlib to use <b>method</b>.*/
static inline int
method_bits(compress_method_t method, zlib_compression_level_t level)
{
/* Bits+16 means "use gzip" in zlib >= 1.2 */
const int flag = method == GZIP_METHOD ? 16 : 0;
switch (level) {
default:
case HIGH_COMPRESSION: return flag + 15;
case MEDIUM_COMPRESSION: return flag + 13;
case LOW_COMPRESSION: return flag + 11;
}
}
static inline int
get_memlevel(zlib_compression_level_t level)
{
switch (level) {
default:
case HIGH_COMPRESSION: return 8;
case MEDIUM_COMPRESSION: return 7;
case LOW_COMPRESSION: return 6;
}
}
/** @{ */
/* These macros define the maximum allowable compression factor. Anything of
* size greater than CHECK_FOR_COMPRESSION_BOMB_AFTER is not allowed to
* have an uncompression factor (uncompressed size:compressed size ratio) of
* any greater than MAX_UNCOMPRESSION_FACTOR.
*
* Picking a value for MAX_UNCOMPRESSION_FACTOR is a trade-off: we want it to
* be small to limit the attack multiplier, but we also want it to be large
* enough so that no legitimate document --even ones we might invent in the
* future -- ever compresses by a factor of greater than
* MAX_UNCOMPRESSION_FACTOR. Within those parameters, there's a reasonably
* large range of possible values. IMO, anything over 8 is probably safe; IMO
* anything under 50 is probably sufficient.
*/
#define MAX_UNCOMPRESSION_FACTOR 25
#define CHECK_FOR_COMPRESSION_BOMB_AFTER (1024*64)
/** @} */
/** Return true if uncompressing an input of size <b>in_size</b> to an input
* of size at least <b>size_out</b> looks like a compression bomb. */
static int
is_compression_bomb(size_t size_in, size_t size_out)
{
if (size_in == 0 || size_out < CHECK_FOR_COMPRESSION_BOMB_AFTER)
return 0;
return (size_out / size_in > MAX_UNCOMPRESSION_FACTOR);
}
/** Given <b>in_len</b> bytes at <b>in</b>, compress them into a newly
* allocated buffer, using the method described in <b>method</b>. Store the
* compressed string in *<b>out</b>, and its length in *<b>out_len</b>.
* Return 0 on success, -1 on failure.
* compressed string in *<b>out</b>, and its length in *<b>out_len</b>. Return
* 0 on success, -1 on failure.
*/
int
tor_gzip_compress(char **out, size_t *out_len,
tor_zlib_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method)
{
@@ -156,7 +132,7 @@ tor_gzip_compress(char **out, size_t *out_len,
if (deflateInit2(stream, Z_BEST_COMPRESSION, Z_DEFLATED,
method_bits(method, HIGH_COMPRESSION),
get_memlevel(HIGH_COMPRESSION),
memory_level(HIGH_COMPRESSION),
Z_DEFAULT_STRATEGY) != Z_OK) {
//LCOV_EXCL_START -- we can only provoke failure by giving junk arguments.
log_warn(LD_GENERAL, "Error from deflateInit2: %s",
@@ -224,7 +200,7 @@ tor_gzip_compress(char **out, size_t *out_len,
}
tor_free(stream);
if (is_compression_bomb(*out_len, in_len)) {
if (tor_compress_is_compression_bomb(*out_len, in_len)) {
log_warn(LD_BUG, "We compressed something and got an insanely high "
"compression factor; other Tors would think this was a zlib bomb.");
goto err;
@@ -240,19 +216,17 @@ tor_gzip_compress(char **out, size_t *out_len,
return -1;
}
/** Given zero or more zlib-compressed or gzip-compressed strings of
* total length
* <b>in_len</b> bytes at <b>in</b>, uncompress them into a newly allocated
* buffer, using the method described in <b>method</b>. Store the uncompressed
* string in *<b>out</b>, and its length in *<b>out_len</b>. Return 0 on
* success, -1 on failure.
/** Given an Zlib/Gzip compressed string of total length <b>in_len</b> bytes
* at <b>in</b>, uncompress them into a newly allocated buffer. Store the
* uncompressed string in *<b>out</b>, and its length in *<b>out_len</b>.
* Return 0 on success, -1 on failure.
*
* If <b>complete_only</b> is true, we consider a truncated input as a
* failure; otherwise we decompress as much as we can. Warn about truncated
* or corrupt inputs at <b>protocol_warn_level</b>.
* If <b>complete_only</b> is true, we consider a truncated input as a failure;
* otherwise we decompress as much as we can. Warn about truncated or corrupt
* inputs at <b>protocol_warn_level</b>.
*/
int
tor_gzip_uncompress(char **out, size_t *out_len,
tor_zlib_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
@@ -332,7 +306,7 @@ tor_gzip_uncompress(char **out, size_t *out_len,
log_warn(LD_GENERAL, "Size overflow in uncompression.");
goto err;
}
if (is_compression_bomb(in_len, out_size)) {
if (tor_compress_is_compression_bomb(in_len, out_size)) {
log_warn(LD_GENERAL, "Input looks like a possible zlib bomb; "
"not proceeding.");
goto err;
@@ -382,26 +356,9 @@ tor_gzip_uncompress(char **out, size_t *out_len,
return -1;
}
/** Try to tell whether the <b>in_len</b>-byte string in <b>in</b> is likely
* to be compressed or not. If it is, return the likeliest compression method.
* Otherwise, return UNKNOWN_METHOD.
*/
compress_method_t
detect_compression_method(const char *in, size_t in_len)
{
if (in_len > 2 && fast_memeq(in, "\x1f\x8b", 2)) {
return GZIP_METHOD;
} else if (in_len > 2 && (in[0] & 0x0f) == 8 &&
(ntohs(get_uint16(in)) % 31) == 0) {
return ZLIB_METHOD;
} else {
return UNKNOWN_METHOD;
}
}
/** Internal state for an incremental zlib compression/decompression. The
* body of this struct is not exposed. */
struct tor_zlib_state_t {
/** Internal zlib state for an incremental compression/decompression.
* The body of this struct is not exposed. */
struct tor_zlib_compress_state_t {
struct z_stream_s stream; /**< The zlib stream */
int compress; /**< True if we are compressing; false if we are inflating */
@@ -414,47 +371,79 @@ struct tor_zlib_state_t {
size_t allocation;
};
/** Construct and return a tor_zlib_state_t object using <b>method</b>. If
* <b>compress</b>, it's for compression; otherwise it's for
* decompression. */
tor_zlib_state_t *
tor_zlib_new(int compress_, compress_method_t method,
zlib_compression_level_t compression_level)
/** Return an approximate number of bytes used in RAM to hold a state with
* window bits <b>windowBits</b> and compression level 'memlevel' */
static size_t
tor_zlib_state_size_precalc(int inflate_, int windowbits, int memlevel)
{
tor_zlib_state_t *out;
windowbits &= 15;
#define A_FEW_KILOBYTES 2048
if (inflate_) {
/* From zconf.h:
"The memory requirements for inflate are (in bytes) 1 << windowBits
that is, 32K for windowBits=15 (default value) plus a few kilobytes
for small objects."
*/
return sizeof(tor_zlib_compress_state_t) + sizeof(struct z_stream_s) +
(1 << 15) + A_FEW_KILOBYTES;
} else {
/* Also from zconf.h:
"The memory requirements for deflate are (in bytes):
(1 << (windowBits+2)) + (1 << (memLevel+9))
... plus a few kilobytes for small objects."
*/
return sizeof(tor_zlib_compress_state_t) + sizeof(struct z_stream_s) +
(1 << (windowbits + 2)) + (1 << (memlevel + 9)) + A_FEW_KILOBYTES;
}
#undef A_FEW_KILOBYTES
}
/** Construct and return a tor_zlib_compress_state_t object using
* <b>method</b>. If <b>compress</b>, it's for compression; otherwise it's for
* decompression. */
tor_zlib_compress_state_t *
tor_zlib_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level)
{
tor_zlib_compress_state_t *out;
int bits, memlevel;
if (! compress_) {
/* use this setting for decompression, since we might have the
* max number of window bits */
compression_level = HIGH_COMPRESSION;
}
if (! compress) {
/* use this setting for decompression, since we might have the
* max number of window bits */
compression_level = HIGH_COMPRESSION;
}
out = tor_malloc_zero(sizeof(tor_zlib_state_t));
out->stream.zalloc = Z_NULL;
out->stream.zfree = Z_NULL;
out->stream.opaque = NULL;
out->compress = compress_;
bits = method_bits(method, compression_level);
memlevel = get_memlevel(compression_level);
if (compress_) {
if (deflateInit2(&out->stream, Z_BEST_COMPRESSION, Z_DEFLATED,
bits, memlevel,
Z_DEFAULT_STRATEGY) != Z_OK)
goto err; // LCOV_EXCL_LINE
} else {
if (inflateInit2(&out->stream, bits) != Z_OK)
goto err; // LCOV_EXCL_LINE
}
out->allocation = tor_zlib_state_size_precalc(!compress_, bits, memlevel);
out = tor_malloc_zero(sizeof(tor_zlib_compress_state_t));
out->stream.zalloc = Z_NULL;
out->stream.zfree = Z_NULL;
out->stream.opaque = NULL;
out->compress = compress;
bits = method_bits(method, compression_level);
memlevel = memory_level(compression_level);
if (compress) {
if (deflateInit2(&out->stream, Z_BEST_COMPRESSION, Z_DEFLATED,
bits, memlevel,
Z_DEFAULT_STRATEGY) != Z_OK)
goto err; // LCOV_EXCL_LINE
} else {
if (inflateInit2(&out->stream, bits) != Z_OK)
goto err; // LCOV_EXCL_LINE
}
out->allocation = tor_zlib_state_size_precalc(!compress, bits, memlevel);
total_zlib_allocation += out->allocation;
total_zlib_allocation += out->allocation;
return out;
return out;
err:
tor_free(out);
return NULL;
tor_free(out);
return NULL;
}
/** Compress/decompress some bytes using <b>state</b>. Read up to
@@ -462,18 +451,20 @@ tor_zlib_new(int compress_, compress_method_t method,
* to *<b>out</b>, adjusting the values as we go. If <b>finish</b> is true,
* we've reached the end of the input.
*
* Return TOR_ZLIB_DONE if we've finished the entire compression/decompression.
* Return TOR_ZLIB_OK if we're processed everything from the input.
* Return TOR_ZLIB_BUF_FULL if we're out of space on <b>out</b>.
* Return TOR_ZLIB_ERR if the stream is corrupt.
* Return TOR_COMPRESS_DONE if we've finished the entire
* compression/decompression.
* Return TOR_COMPRESS_OK if we're processed everything from the input.
* Return TOR_COMPRESS_BUFFER_FULL if we're out of space on <b>out</b>.
* Return TOR_COMPRESS_ERROR if the stream is corrupt.
*/
tor_zlib_output_t
tor_zlib_process(tor_zlib_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish)
tor_compress_output_t
tor_zlib_compress_process(tor_zlib_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish)
{
int err;
tor_assert(state != NULL);
tor_assert(*in_len <= UINT_MAX);
tor_assert(*out_len <= UINT_MAX);
state->stream.next_in = (unsigned char*) *in;
@@ -496,38 +487,39 @@ tor_zlib_process(tor_zlib_state_t *state,
*in_len = state->stream.avail_in;
if (! state->compress &&
is_compression_bomb(state->input_so_far, state->output_so_far)) {
tor_compress_is_compression_bomb(state->input_so_far,
state->output_so_far)) {
log_warn(LD_DIR, "Possible zlib bomb; abandoning stream.");
return TOR_ZLIB_ERR;
return TOR_COMPRESS_ERROR;
}
switch (err)
{
case Z_STREAM_END:
return TOR_ZLIB_DONE;
return TOR_COMPRESS_DONE;
case Z_BUF_ERROR:
if (state->stream.avail_in == 0 && !finish)
return TOR_ZLIB_OK;
return TOR_ZLIB_BUF_FULL;
return TOR_COMPRESS_OK;
return TOR_COMPRESS_BUFFER_FULL;
case Z_OK:
if (state->stream.avail_out == 0 || finish)
return TOR_ZLIB_BUF_FULL;
return TOR_ZLIB_OK;
return TOR_COMPRESS_BUFFER_FULL;
return TOR_COMPRESS_OK;
default:
log_warn(LD_GENERAL, "Gzip returned an error: %s",
state->stream.msg ? state->stream.msg : "<no message>");
return TOR_ZLIB_ERR;
return TOR_COMPRESS_ERROR;
}
}
/** Deallocate <b>state</b>. */
void
tor_zlib_free(tor_zlib_state_t *state)
tor_zlib_compress_free(tor_zlib_compress_state_t *state)
{
if (!state)
if (state == NULL)
return;
total_zlib_allocation -= state->allocation;
total_zlib_allocation -= state->allocation;
if (state->compress)
deflateEnd(&state->stream);
@@ -537,41 +529,11 @@ tor_zlib_free(tor_zlib_state_t *state)
tor_free(state);
}
/** Return an approximate number of bytes used in RAM to hold a state with
* window bits <b>windowBits</b> and compression level 'memlevel' */
static size_t
tor_zlib_state_size_precalc(int inflate_, int windowbits, int memlevel)
{
windowbits &= 15;
#define A_FEW_KILOBYTES 2048
if (inflate_) {
/* From zconf.h:
"The memory requirements for inflate are (in bytes) 1 << windowBits
that is, 32K for windowBits=15 (default value) plus a few kilobytes
for small objects."
*/
return sizeof(tor_zlib_state_t) + sizeof(struct z_stream_s) +
(1 << 15) + A_FEW_KILOBYTES;
} else {
/* Also from zconf.h:
"The memory requirements for deflate are (in bytes):
(1 << (windowBits+2)) + (1 << (memLevel+9))
... plus a few kilobytes for small objects."
*/
return sizeof(tor_zlib_state_t) + sizeof(struct z_stream_s) +
(1 << (windowbits + 2)) + (1 << (memlevel + 9)) + A_FEW_KILOBYTES;
}
#undef A_FEW_KILOBYTES
}
/** Return the approximate number of bytes allocated for <b>state</b>. */
size_t
tor_zlib_state_size(const tor_zlib_state_t *state)
tor_zlib_compress_state_size(const tor_zlib_compress_state_t *state)
{
tor_assert(state != NULL);
return state->allocation;
}
src/common/compress_zlib.h
+51
View File
@@ -0,0 +1,51 @@
/* Copyright (c) 2003, Roger Dingledine
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress_zlib.h
* \brief Header for compress_zlib.c
**/
#ifndef TOR_COMPRESS_ZLIB_H
#define TOR_COMPRESS_ZLIB_H
int tor_zlib_method_supported(void);
const char *tor_zlib_get_version_str(void);
const char *tor_zlib_get_header_version_str(void);
int tor_zlib_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method);
int tor_zlib_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level);
/** Internal state for an incremental zlib/gzip compression/decompression. */
typedef struct tor_zlib_compress_state_t tor_zlib_compress_state_t;
tor_zlib_compress_state_t *
tor_zlib_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level);
tor_compress_output_t
tor_zlib_compress_process(tor_zlib_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish);
void tor_zlib_compress_free(tor_zlib_compress_state_t *state);
size_t tor_zlib_compress_state_size(const tor_zlib_compress_state_t *state);
size_t tor_zlib_get_total_allocation(void);
#endif // TOR_COMPRESS_ZLIB_H.
src/common/compress_zstd.c
+609
View File
@@ -0,0 +1,609 @@
/* Copyright (c) 2004, Roger Dingledine.
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress_zstd.c
* \brief Compression backend for Zstandard.
*
* This module should never be invoked directly. Use the compress module
* instead.
**/
#include "orconfig.h"
#include "util.h"
#include "torlog.h"
#include "compress.h"
#include "compress_zstd.h"
#ifdef HAVE_ZSTD
#include <zstd.h>
#include <zstd_errors.h>
#endif
/** Total number of bytes allocated for Zstandard state. */
static size_t total_zstd_allocation = 0;
#ifdef HAVE_ZSTD
/** Given <b>level</b> return the memory level. */
static int
memory_level(compression_level_t level)
{
switch (level) {
default:
case HIGH_COMPRESSION: return 9;
case MEDIUM_COMPRESSION: return 8;
case LOW_COMPRESSION: return 7;
}
}
#endif // HAVE_ZSTD.
/** Return 1 if Zstandard compression is supported; otherwise 0. */
int
tor_zstd_method_supported(void)
{
#ifdef HAVE_ZSTD
return 1;
#else
return 0;
#endif
}
/** Return a string representation of the version of the currently running
* version of libzstd. Returns NULL if Zstandard is unsupported. */
const char *
tor_zstd_get_version_str(void)
{
#ifdef HAVE_ZSTD
static char version_str[16];
size_t version_number;
version_number = ZSTD_versionNumber();
tor_snprintf(version_str, sizeof(version_str),
"%lu.%lu.%lu",
version_number / 10000 % 100,
version_number / 100 % 100,
version_number % 100);
return version_str;
#else
return NULL;
#endif
}
/** Return a string representation of the version of the version of libzstd
* used at compilation time. Returns NULL if Zstandard is unsupported. */
const char *
tor_zstd_get_header_version_str(void)
{
#ifdef HAVE_ZSTD
return ZSTD_VERSION_STRING;
#else
return NULL;
#endif
}
/** Given <b>in_len</b> bytes at <b>in</b>, compress them into a newly
* allocated buffer, using the Zstandard method. Store the compressed string
* in *<b>out</b>, and its length in *<b>out_len</b>. Return 0 on success, -1
* on failure.
*/
int
tor_zstd_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method)
{
#ifdef HAVE_ZSTD
ZSTD_CStream *stream = NULL;
size_t out_size, old_size;
size_t retval;
tor_assert(out);
tor_assert(out_len);
tor_assert(in);
tor_assert(in_len < UINT_MAX);
tor_assert(method == ZSTD_METHOD);
*out = NULL;
stream = ZSTD_createCStream();
if (stream == NULL) {
// Zstandard does not give us any useful error message to why this
// happened. See https://github.com/facebook/zstd/issues/398
log_warn(LD_GENERAL, "Error while creating Zstandard stream");
goto err;
}
retval = ZSTD_initCStream(stream,
memory_level(HIGH_COMPRESSION));
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream initialization error: %s",
ZSTD_getErrorName(retval));
goto err;
}
// Assume 50% compression and update our buffer in case we need to.
out_size = in_len / 2;
if (out_size < 1024)
out_size = 1024;
*out = tor_malloc(out_size);
*out_len = 0;
ZSTD_inBuffer input = { in, in_len, 0 };
ZSTD_outBuffer output = { *out, out_size, 0 };
while (input.pos < input.size) {
retval = ZSTD_compressStream(stream, &output, &input);
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream compression error: %s",
ZSTD_getErrorName(retval));
goto err;
}
if (input.pos < input.size && output.pos == output.size) {
old_size = out_size;
out_size *= 2;
if (out_size < old_size) {
log_warn(LD_GENERAL, "Size overflow in Zstandard compression.");
goto err;
}
if (out_size - output.pos > UINT_MAX) {
log_warn(LD_BUG, "Ran over unsigned int limit of Zstandard while "
"compressing.");
goto err;
}
output.dst = *out = tor_realloc(*out, out_size);
output.size = out_size;
}
}
while (1) {
retval = ZSTD_endStream(stream, &output);
if (retval == 0)
break;
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream error: %s",
ZSTD_getErrorName(retval));
goto err;
}
if (output.pos == output.size) {
old_size = out_size;
out_size *= 2;
if (out_size < old_size) {
log_warn(LD_GENERAL, "Size overflow in Zstandard compression.");
goto err;
}
if (out_size - output.pos > UINT_MAX) {
log_warn(LD_BUG, "Ran over unsigned int limit of Zstandard while "
"compressing.");
goto err;
}
output.dst = *out = tor_realloc(*out, out_size);
output.size = out_size;
}
}
*out_len = output.pos;
if (tor_compress_is_compression_bomb(*out_len, in_len)) {
log_warn(LD_BUG, "We compressed something and got an insanely high "
"compression factor; other Tor instances would think "
"this is a compression bomb.");
goto err;
}
if (stream != NULL) {
ZSTD_freeCStream(stream);
}
return 0;
err:
if (stream != NULL) {
ZSTD_freeCStream(stream);
}
tor_free(*out);
return -1;
#else // HAVE_ZSTD.
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)method;
return -1;
#endif // HAVE_ZSTD.
}
/** Given a Zstandard compressed string of total length <b>in_len</b> bytes at
* <b>in</b>, uncompress them into a newly allocated buffer. Store the
* uncompressed string in *<b>out</b>, and its length in *<b>out_len</b>.
* Return 0 on success, -1 on failure.
*
* If <b>complete_only</b> is true, we consider a truncated input as a failure;
* otherwise we decompress as much as we can. Warn about truncated or corrupt
* inputs at <b>protocol_warn_level</b>.
*/
int
tor_zstd_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level)
{
#ifdef HAVE_ZSTD
ZSTD_DStream *stream = NULL;
size_t retval;
size_t out_size, old_size;
tor_assert(out);
tor_assert(out_len);
tor_assert(in);
tor_assert(in_len < UINT_MAX);
tor_assert(method == ZSTD_METHOD);
// FIXME(ahf): Handle this?
(void)complete_only;
(void)protocol_warn_level;
*out = NULL;
stream = ZSTD_createDStream();
if (stream == NULL) {
// Zstandard does not give us any useful error message to why this
// happened. See https://github.com/facebook/zstd/issues/398
log_warn(LD_GENERAL, "Error while creating Zstandard stream");
goto err;
}
retval = ZSTD_initDStream(stream);
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream initialization error: %s",
ZSTD_getErrorName(retval));
goto err;
}
out_size = in_len * 2;
if (out_size < 1024)
out_size = 1024;
if (out_size >= SIZE_T_CEILING || out_size > UINT_MAX)
goto err;
*out = tor_malloc(out_size);
*out_len = 0;
ZSTD_inBuffer input = { in, in_len, 0 };
ZSTD_outBuffer output = { *out, out_size, 0 };
while (input.pos < input.size) {
retval = ZSTD_decompressStream(stream, &output, &input);
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream decompression error: %s",
ZSTD_getErrorName(retval));
goto err;
}
if (input.pos < input.size && output.pos == output.size) {
old_size = out_size;
out_size *= 2;
if (out_size < old_size) {
log_warn(LD_GENERAL, "Size overflow in Zstandard compression.");
goto err;
}
if (tor_compress_is_compression_bomb(in_len, out_size)) {
log_warn(LD_GENERAL, "Input looks like a possible Zstandard "
"compression bomb. Not proceeding.");
goto err;
}
if (out_size >= SIZE_T_CEILING) {
log_warn(LD_BUG, "Hit SIZE_T_CEILING limit while uncompressing "
"Zstandard data.");
goto err;
}
if (out_size - output.pos > UINT_MAX) {
log_warn(LD_BUG, "Ran over unsigned int limit of Zstandard while "
"decompressing.");
goto err;
}
output.dst = *out = tor_realloc(*out, out_size);
output.size = out_size;
}
}
*out_len = output.pos;
if (stream != NULL) {
ZSTD_freeDStream(stream);
}
// NUL-terminate our output.
if (out_size == *out_len)
*out = tor_realloc(*out, out_size + 1);
(*out)[*out_len] = '\0';
return 0;
err:
if (stream != NULL) {
ZSTD_freeDStream(stream);
}
tor_free(*out);
return -1;
#else // HAVE_ZSTD.
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)method;
(void)complete_only;
(void)protocol_warn_level;
return -1;
#endif // HAVE_ZSTD.
}
/** Internal Zstandard state for incremental compression/decompression.
* The body of this struct is not exposed. */
struct tor_zstd_compress_state_t {
#ifdef HAVE_ZSTD
union {
/** Compression stream. Used when <b>compress</b> is true. */
ZSTD_CStream *compress_stream;
/** Decompression stream. Used when <b>compress</b> is false. */
ZSTD_DStream *decompress_stream;
} u; /**< Zstandard stream objects. */
#endif // HAVE_ZSTD.
int compress; /**< True if we are compressing; false if we are inflating */
/** Number of bytes read so far. Used to detect compression bombs. */
size_t input_so_far;
/** Number of bytes written so far. Used to detect compression bombs. */
size_t output_so_far;
/** Approximate number of bytes allocated for this object. */
size_t allocation;
};
/** Construct and return a tor_zstd_compress_state_t object using
* <b>method</b>. If <b>compress</b>, it's for compression; otherwise it's for
* decompression. */
tor_zstd_compress_state_t *
tor_zstd_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level)
{
tor_assert(method == ZSTD_METHOD);
#ifdef HAVE_ZSTD
tor_zstd_compress_state_t *result;
size_t retval;
result = tor_malloc_zero(sizeof(tor_zstd_compress_state_t));
result->compress = compress;
// FIXME(ahf): We should either try to do the pre-calculation that is done
// with the zlib backend or use a custom allocator here where we pass our
// tor_zstd_compress_state_t as the opaque value.
result->allocation = 0;
if (compress) {
result->u.compress_stream = ZSTD_createCStream();
if (result->u.compress_stream == NULL) {
log_warn(LD_GENERAL, "Error while creating Zstandard stream");
goto err;
}
retval = ZSTD_initCStream(result->u.compress_stream,
memory_level(compression_level));
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream initialization error: %s",
ZSTD_getErrorName(retval));
goto err;
}
} else {
result->u.decompress_stream = ZSTD_createDStream();
if (result->u.decompress_stream == NULL) {
log_warn(LD_GENERAL, "Error while creating Zstandard stream");
goto err;
}
retval = ZSTD_initDStream(result->u.decompress_stream);
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard stream initialization error: %s",
ZSTD_getErrorName(retval));
goto err;
}
}
return result;
err:
if (compress) {
ZSTD_freeCStream(result->u.compress_stream);
} else {
ZSTD_freeDStream(result->u.decompress_stream);
}
tor_free(result);
return NULL;
#else // HAVE_ZSTD.
(void)compress;
(void)method;
(void)compression_level;
return NULL;
#endif // HAVE_ZSTD.
}
/** Compress/decompress some bytes using <b>state</b>. Read up to
* *<b>in_len</b> bytes from *<b>in</b>, and write up to *<b>out_len</b> bytes
* to *<b>out</b>, adjusting the values as we go. If <b>finish</b> is true,
* we've reached the end of the input.
*
* Return TOR_COMPRESS_DONE if we've finished the entire
* compression/decompression.
* Return TOR_COMPRESS_OK if we're processed everything from the input.
* Return TOR_COMPRESS_BUFFER_FULL if we're out of space on <b>out</b>.
* Return TOR_COMPRESS_ERROR if the stream is corrupt.
*/
tor_compress_output_t
tor_zstd_compress_process(tor_zstd_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish)
{
#ifdef HAVE_ZSTD
size_t retval;
tor_assert(state != NULL);
tor_assert(*in_len <= UINT_MAX);
tor_assert(*out_len <= UINT_MAX);
ZSTD_inBuffer input = { *in, *in_len, 0 };
ZSTD_outBuffer output = { *out, *out_len, 0 };
if (state->compress) {
retval = ZSTD_compressStream(state->u.compress_stream,
&output, &input);
} else {
retval = ZSTD_decompressStream(state->u.decompress_stream,
&output, &input);
}
state->input_so_far += input.pos;
state->output_so_far += output.pos;
*out = (char *)output.dst + output.pos;
*out_len = output.size - output.pos;
*in = (char *)input.src + input.pos;
*in_len = input.size - input.pos;
if (! state->compress &&
tor_compress_is_compression_bomb(state->input_so_far,
state->output_so_far)) {
log_warn(LD_DIR, "Possible compression bomb; abandoning stream.");
return TOR_COMPRESS_ERROR;
}
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard %s didn't finish: %s.",
state->compress ? "compression" : "decompression",
ZSTD_getErrorName(retval));
return TOR_COMPRESS_ERROR;
}
if (state->compress && !finish) {
retval = ZSTD_flushStream(state->u.compress_stream, &output);
*out = (char *)output.dst + output.pos;
*out_len = output.size - output.pos;
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard compression unable to flush: %s.",
ZSTD_getErrorName(retval));
return TOR_COMPRESS_ERROR;
}
if (retval > 0)
return TOR_COMPRESS_BUFFER_FULL;
}
if (state->compress && finish) {
retval = ZSTD_endStream(state->u.compress_stream, &output);
*out = (char *)output.dst + output.pos;
*out_len = output.size - output.pos;
if (ZSTD_isError(retval)) {
log_warn(LD_GENERAL, "Zstandard compression unable to write "
"epilogue: %s.",
ZSTD_getErrorName(retval));
return TOR_COMPRESS_ERROR;
}
// endStream returns the number of bytes that is needed to write the
// epilogue.
if (retval > 0)
return TOR_COMPRESS_BUFFER_FULL;
}
return finish ? TOR_COMPRESS_DONE : TOR_COMPRESS_OK;
#else // HAVE_ZSTD.
(void)state;
(void)out;
(void)out_len;
(void)in;
(void)in_len;
(void)finish;
return TOR_COMPRESS_ERROR;
#endif // HAVE_ZSTD.
}
/** Deallocate <b>state</b>. */
void
tor_zstd_compress_free(tor_zstd_compress_state_t *state)
{
if (state == NULL)
return;
total_zstd_allocation -= state->allocation;
#ifdef HAVE_ZSTD
if (state->compress) {
ZSTD_freeCStream(state->u.compress_stream);
} else {
ZSTD_freeDStream(state->u.decompress_stream);
}
#endif // HAVE_ZSTD.
tor_free(state);
}
/** Return the approximate number of bytes allocated for <b>state</b>. */
size_t
tor_zstd_compress_state_size(const tor_zstd_compress_state_t *state)
{
tor_assert(state != NULL);
return state->allocation;
}
/** Return the approximate number of bytes allocated for all Zstandard
* states. */
size_t
tor_zstd_get_total_allocation(void)
{
return total_zstd_allocation;
}
src/common/compress_zstd.h
+51
View File
@@ -0,0 +1,51 @@
/* Copyright (c) 2003, Roger Dingledine
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file compress_zstd.h
* \brief Header for compress_zstd.c
**/
#ifndef TOR_COMPRESS_ZSTD_H
#define TOR_COMPRESS_ZSTD_H
int tor_zstd_method_supported(void);
const char *tor_zstd_get_version_str(void);
const char *tor_zstd_get_header_version_str(void);
int tor_zstd_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method);
int tor_zstd_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level);
/** Internal state for an incremental Zstandard compression/decompression. */
typedef struct tor_zstd_compress_state_t tor_zstd_compress_state_t;
tor_zstd_compress_state_t *
tor_zstd_compress_new(int compress,
compress_method_t method,
compression_level_t compression_level);
tor_compress_output_t
tor_zstd_compress_process(tor_zstd_compress_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish);
void tor_zstd_compress_free(tor_zstd_compress_state_t *state);
size_t tor_zstd_compress_state_size(const tor_zstd_compress_state_t *state);
size_t tor_zstd_get_total_allocation(void);
#endif // TOR_COMPRESS_ZSTD_H.
src/common/include.am
+8 -2
View File
@@ -105,11 +105,14 @@ src/common/src_common_libor_testing_a-log.$(OBJEXT) \
LIBOR_CRYPTO_A_SRC = \
src/common/aes.c \
src/common/compress.c \
src/common/compress_lzma.c \
src/common/compress_zlib.c \
src/common/compress_zstd.c \
src/common/crypto.c \
src/common/crypto_pwbox.c \
src/common/crypto_s2k.c \
src/common/crypto_format.c \
src/common/torgzip.c \
src/common/tortls.c \
src/common/crypto_curve25519.c \
src/common/crypto_ed25519.c
@@ -145,6 +148,10 @@ COMMONHEADERS = \
src/common/compat_openssl.h \
src/common/compat_threads.h \
src/common/compat_time.h \
src/common/compress.h \
src/common/compress_lzma.h \
src/common/compress_zlib.h \
src/common/compress_zstd.h \
src/common/confline.h \
src/common/container.h \
src/common/crypto.h \
@@ -163,7 +170,6 @@ COMMONHEADERS = \
src/common/storagedir.h \
src/common/testsupport.h \
src/common/timers.h \
src/common/torgzip.h \
src/common/torint.h \
src/common/torlog.h \
src/common/tortls.h \
src/common/torgzip.h
-72
View File
@@ -1,72 +0,0 @@
/* Copyright (c) 2003, Roger Dingledine
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2017, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file torgzip.h
* \brief Headers for torgzip.h
**/
#ifndef TOR_TORGZIP_H
#define TOR_TORGZIP_H
/** Enumeration of what kind of compression to use. Only ZLIB_METHOD is
* guaranteed to be supported by the compress/uncompress functions here;
* GZIP_METHOD may be supported if we built against zlib version 1.2 or later
* and is_gzip_supported() returns true. */
typedef enum {
NO_METHOD=0, GZIP_METHOD=1, ZLIB_METHOD=2, UNKNOWN_METHOD=3
} compress_method_t;
/**
* Enumeration to define tradeoffs between memory usage and compression level.
* HIGH_COMPRESSION saves the most bandwidth; LOW_COMPRESSION saves the most
* memory.
**/
typedef enum {
HIGH_COMPRESSION, MEDIUM_COMPRESSION, LOW_COMPRESSION
} zlib_compression_level_t;
int
tor_gzip_compress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method);
int
tor_gzip_uncompress(char **out, size_t *out_len,
const char *in, size_t in_len,
compress_method_t method,
int complete_only,
int protocol_warn_level);
int is_gzip_supported(void);
const char *
tor_zlib_get_version_str(void);
const char *
tor_zlib_get_header_version_str(void);
compress_method_t detect_compression_method(const char *in, size_t in_len);
/** Return values from tor_zlib_process; see that function's documentation for
* details. */
typedef enum {
TOR_ZLIB_OK, TOR_ZLIB_DONE, TOR_ZLIB_BUF_FULL, TOR_ZLIB_ERR
} tor_zlib_output_t;
/** Internal state for an incremental zlib compression/decompression. */
typedef struct tor_zlib_state_t tor_zlib_state_t;
tor_zlib_state_t *tor_zlib_new(int compress, compress_method_t method,
zlib_compression_level_t level);
tor_zlib_output_t tor_zlib_process(tor_zlib_state_t *state,
char **out, size_t *out_len,
const char **in, size_t *in_len,
int finish);
void tor_zlib_free(tor_zlib_state_t *state);
size_t tor_zlib_state_size(const tor_zlib_state_t *state);
size_t tor_zlib_get_total_allocation(void);
#endif