/* Plzip - Massively parallel implementation of lzip
Copyright (C) 2009 Laszlo Ersek.
Copyright (C) 2009-2024 Antonio Diaz Diaz.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see .
*/
#define _FILE_OFFSET_BITS 64
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include "lzip.h"
/* When a problem is detected by any thread:
- the thread sets shared_retval to 1 or 2.
- the splitter sets eof and returns.
- the courier discards new packets received or collected.
- the workers drain the queue and return.
- the muxer drains the queue and returns.
(Draining seems to be faster than cleaning up later). */
namespace {
enum { max_packet_size = 1 << 20 };
unsigned long long in_size = 0;
unsigned long long out_size = 0;
struct Packet // data block
{
uint8_t * data; // data may be null if size == 0
int size; // number of bytes in data (if any)
bool eom; // end of member
Packet() : data( 0 ), size( 0 ), eom( true ) {}
Packet( uint8_t * const d, const int s, const bool e )
: data( d ), size( s ), eom ( e ) {}
~Packet() { if( data ) delete[] data; }
};
class Packet_courier // moves packets around
{
public:
unsigned icheck_counter;
unsigned iwait_counter;
unsigned ocheck_counter;
unsigned owait_counter;
private:
int receive_worker_id; // worker queue currently receiving packets
int deliver_worker_id; // worker queue currently delivering packets
Slot_tally slot_tally; // limits the number of input packets
std::vector< std::queue< const Packet * > > ipacket_queues;
std::vector< std::queue< const Packet * > > opacket_queues;
int num_working; // number of workers still running
const int num_workers; // number of workers
const unsigned out_slots; // max output packets per queue
pthread_mutex_t imutex;
pthread_cond_t iav_or_eof; // input packet available or splitter done
pthread_mutex_t omutex;
pthread_cond_t oav_or_exit; // output packet available or all workers exited
std::vector< pthread_cond_t > slot_av; // output slot available
const Shared_retval & shared_retval; // discard new packets on error
bool eof; // splitter done
bool trailing_data_found_; // a worker found trailing data
Packet_courier( const Packet_courier & ); // declared as private
void operator=( const Packet_courier & ); // declared as private
public:
Packet_courier( const Shared_retval & sh_ret, const int workers,
const int in_slots, const int oslots )
: icheck_counter( 0 ), iwait_counter( 0 ),
ocheck_counter( 0 ), owait_counter( 0 ),
receive_worker_id( 0 ), deliver_worker_id( 0 ),
slot_tally( in_slots ), ipacket_queues( workers ),
opacket_queues( workers ), num_working( workers ),
num_workers( workers ), out_slots( oslots ), slot_av( workers ),
shared_retval( sh_ret ), eof( false ), trailing_data_found_( false )
{
xinit_mutex( &imutex ); xinit_cond( &iav_or_eof );
xinit_mutex( &omutex ); xinit_cond( &oav_or_exit );
for( unsigned i = 0; i < slot_av.size(); ++i ) xinit_cond( &slot_av[i] );
}
~Packet_courier()
{
if( shared_retval() ) // cleanup to avoid memory leaks
for( int i = 0; i < num_workers; ++i )
{
while( !ipacket_queues[i].empty() )
{ delete ipacket_queues[i].front(); ipacket_queues[i].pop(); }
while( !opacket_queues[i].empty() )
{ delete opacket_queues[i].front(); opacket_queues[i].pop(); }
}
for( unsigned i = 0; i < slot_av.size(); ++i ) xdestroy_cond( &slot_av[i] );
xdestroy_cond( &oav_or_exit ); xdestroy_mutex( &omutex );
xdestroy_cond( &iav_or_eof ); xdestroy_mutex( &imutex );
}
/* Make a packet with data received from splitter.
If eom == true (end of member), move to next queue. */
void receive_packet( uint8_t * const data, const int size, const bool eom )
{
if( shared_retval() ) { delete[] data; return; } // discard packet on error
const Packet * const ipacket = new Packet( data, size, eom );
slot_tally.get_slot(); // wait for a free slot
xlock( &imutex );
ipacket_queues[receive_worker_id].push( ipacket );
xbroadcast( &iav_or_eof );
xunlock( &imutex );
if( eom && ++receive_worker_id >= num_workers ) receive_worker_id = 0;
}
// distribute a packet to a worker
const Packet * distribute_packet( const int worker_id )
{
const Packet * ipacket = 0;
xlock( &imutex );
++icheck_counter;
while( ipacket_queues[worker_id].empty() && !eof )
{
++iwait_counter;
xwait( &iav_or_eof, &imutex );
}
if( !ipacket_queues[worker_id].empty() )
{
ipacket = ipacket_queues[worker_id].front();
ipacket_queues[worker_id].pop();
}
xunlock( &imutex );
if( ipacket ) slot_tally.leave_slot();
else // no more packets
{
xlock( &omutex ); // notify muxer when last worker exits
if( --num_working == 0 ) xsignal( &oav_or_exit );
xunlock( &omutex );
}
return ipacket;
}
// collect a packet from a worker, discard packet on error
void collect_packet( const Packet * const opacket, const int worker_id )
{
xlock( &omutex );
if( opacket->data )
while( opacket_queues[worker_id].size() >= out_slots )
{
if( shared_retval() ) { delete opacket; goto done; }
xwait( &slot_av[worker_id], &omutex );
}
opacket_queues[worker_id].push( opacket );
if( worker_id == deliver_worker_id ) xsignal( &oav_or_exit );
done:
xunlock( &omutex );
}
/* deliver a packet to muxer
if packet->eom, move to next queue
if packet data == 0, wait again */
const Packet * deliver_packet()
{
const Packet * opacket = 0;
xlock( &omutex );
++ocheck_counter;
while( true )
{
while( opacket_queues[deliver_worker_id].empty() && num_working > 0 )
{
++owait_counter;
xwait( &oav_or_exit, &omutex );
}
if( opacket_queues[deliver_worker_id].empty() ) break;
opacket = opacket_queues[deliver_worker_id].front();
opacket_queues[deliver_worker_id].pop();
if( opacket_queues[deliver_worker_id].size() + 1 == out_slots )
xsignal( &slot_av[deliver_worker_id] );
if( opacket->eom && ++deliver_worker_id >= num_workers )
deliver_worker_id = 0;
if( opacket->data ) break;
delete opacket; opacket = 0;
}
xunlock( &omutex );
return opacket;
}
void add_sizes( const unsigned long long partial_in_size,
const unsigned long long partial_out_size )
{
xlock( &imutex );
in_size += partial_in_size;
out_size += partial_out_size;
xunlock( &imutex );
}
void set_trailing_flag() { trailing_data_found_ = true; }
bool trailing_data_found() { return trailing_data_found_; }
void finish( const int workers_started )
{
xlock( &imutex ); // splitter has no more packets to send
eof = true;
xbroadcast( &iav_or_eof );
xunlock( &imutex );
xlock( &omutex ); // notify muxer if all workers have exited
num_working -= num_workers - workers_started; // workers spared
if( num_working <= 0 ) xsignal( &oav_or_exit );
xunlock( &omutex );
}
bool finished() // all packets delivered to muxer
{
if( !slot_tally.all_free() || !eof || num_working != 0 ) return false;
for( int i = 0; i < num_workers; ++i )
if( !ipacket_queues[i].empty() ) return false;
for( int i = 0; i < num_workers; ++i )
if( !opacket_queues[i].empty() ) return false;
return true;
}
};
struct Worker_arg
{
Packet_courier * courier;
const Pretty_print * pp;
Shared_retval * shared_retval;
int worker_id;
bool ignore_trailing;
bool loose_trailing;
bool testing;
bool nocopy; // avoid copying decompressed data when testing
};
struct Splitter_arg
{
struct Worker_arg worker_arg;
Worker_arg * worker_args;
pthread_t * worker_threads;
unsigned long long cfile_size;
int infd;
unsigned dictionary_size; // returned by splitter to main thread
int num_workers; // returned by splitter to main thread
};
/* Consume packets from courier, decompress their contents and, if not
testing, give to courier the packets produced.
*/
extern "C" void * dworker_s( void * arg )
{
const Worker_arg & tmp = *(const Worker_arg *)arg;
Packet_courier & courier = *tmp.courier;
const Pretty_print & pp = *tmp.pp;
Shared_retval & shared_retval = *tmp.shared_retval;
const int worker_id = tmp.worker_id;
const bool ignore_trailing = tmp.ignore_trailing;
const bool loose_trailing = tmp.loose_trailing;
const bool testing = tmp.testing;
const bool nocopy = tmp.nocopy;
unsigned long long partial_in_size = 0, partial_out_size = 0;
int new_pos = 0;
bool draining = false; // either trailing data or an error were found
uint8_t * new_data = 0;
LZ_Decoder * const decoder = LZ_decompress_open();
if( !decoder || LZ_decompress_errno( decoder ) != LZ_ok )
{ draining = true; if( shared_retval.set_value( 1 ) ) pp( mem_msg ); }
while( true )
{
const Packet * const ipacket = courier.distribute_packet( worker_id );
if( !ipacket ) break; // no more packets to process
int written = 0;
while( !draining ) // else discard trailing data or drain queue
{
if( LZ_decompress_write_size( decoder ) > 0 && written < ipacket->size )
{
const int wr = LZ_decompress_write( decoder, ipacket->data + written,
ipacket->size - written );
if( wr < 0 ) internal_error( "library error (LZ_decompress_write)." );
written += wr;
if( written > ipacket->size )
internal_error( "ipacket size exceeded in worker." );
}
if( ipacket->eom && written == ipacket->size )
LZ_decompress_finish( decoder );
unsigned long long total_in = 0; // detect empty member + corrupt header
while( !draining ) // read and pack decompressed data
{
if( !nocopy && !new_data &&
!( new_data = new( std::nothrow ) uint8_t[max_packet_size] ) )
{ draining = true; if( shared_retval.set_value( 1 ) ) pp( mem_msg );
break; }
const int rd = LZ_decompress_read( decoder,
nocopy ? 0 : new_data + new_pos,
max_packet_size - new_pos );
if( rd < 0 ) // trailing data or decoder error
{
draining = true;
const enum LZ_Errno lz_errno = LZ_decompress_errno( decoder );
if( lz_errno == LZ_header_error )
{
courier.set_trailing_flag();
if( !ignore_trailing )
{ if( shared_retval.set_value( 2 ) ) pp( trailing_msg ); }
}
else if( lz_errno == LZ_data_error &&
LZ_decompress_member_position( decoder ) == 0 )
{
courier.set_trailing_flag();
if( !loose_trailing )
{ if( shared_retval.set_value( 2 ) ) pp( corrupt_mm_msg ); }
else if( !ignore_trailing )
{ if( shared_retval.set_value( 2 ) ) pp( trailing_msg ); }
}
else
decompress_error( decoder, pp, shared_retval, worker_id );
}
else new_pos += rd;
if( new_pos > max_packet_size )
internal_error( "opacket size exceeded in worker." );
if( LZ_decompress_member_finished( decoder ) == 1 )
{
partial_in_size += LZ_decompress_member_position( decoder );
partial_out_size += LZ_decompress_data_position( decoder );
}
const bool eom = draining || LZ_decompress_finished( decoder ) == 1;
if( new_pos == max_packet_size || eom )
{
if( !testing ) // make data packet
{
const Packet * const opacket =
new Packet( ( new_pos > 0 ) ? new_data : 0, new_pos, eom );
courier.collect_packet( opacket, worker_id );
if( new_pos > 0 ) new_data = 0;
}
new_pos = 0;
if( eom )
{ LZ_decompress_reset( decoder ); // prepare for new member
break; }
}
if( rd == 0 )
{
const unsigned long long size = LZ_decompress_total_in_size( decoder );
if( total_in == size ) break; else total_in = size;
}
}
if( !ipacket->data || written == ipacket->size ) break;
}
delete ipacket;
}
if( new_data ) delete[] new_data;
courier.add_sizes( partial_in_size, partial_out_size );
if( LZ_decompress_member_position( decoder ) != 0 &&
shared_retval.set_value( 1 ) )
pp( "Error, some data remains in decoder." );
if( LZ_decompress_close( decoder ) < 0 && shared_retval.set_value( 1 ) )
pp( "LZ_decompress_close failed." );
return 0;
}
bool start_worker( const Worker_arg & worker_arg,
Worker_arg * const worker_args,
pthread_t * const worker_threads, const int worker_id,
Shared_retval & shared_retval )
{
worker_args[worker_id] = worker_arg;
worker_args[worker_id].worker_id = worker_id;
const int errcode = pthread_create( &worker_threads[worker_id], 0,
dworker_s, &worker_args[worker_id] );
if( errcode && shared_retval.set_value( 1 ) )
show_error( "Can't create worker threads", errcode );
return errcode == 0;
}
/* Split data from input file into chunks and pass them to courier for
packaging and distribution to workers.
Start a worker per member up to a maximum of num_workers.
*/
extern "C" void * dsplitter_s( void * arg )
{
Splitter_arg & tmp = *(Splitter_arg *)arg;
const Worker_arg & worker_arg = tmp.worker_arg;
Packet_courier & courier = *worker_arg.courier;
const Pretty_print & pp = *worker_arg.pp;
Shared_retval & shared_retval = *worker_arg.shared_retval;
Worker_arg * const worker_args = tmp.worker_args;
pthread_t * const worker_threads = tmp.worker_threads;
const int infd = tmp.infd;
int worker_id = 0; // number of workers started
const int hsize = Lzip_header::size;
const int tsize = Lzip_trailer::size;
const int buffer_size = max_packet_size;
// buffer with room for trailer, header, data, and sentinel "LZIP"
const int base_buffer_size = tsize + hsize + buffer_size + 4;
uint8_t * const base_buffer = new( std::nothrow ) uint8_t[base_buffer_size];
if( !base_buffer )
{
mem_fail:
if( shared_retval.set_value( 1 ) ) pp( mem_msg );
fail:
delete[] base_buffer;
courier.finish( worker_id ); // no more packets to send
tmp.num_workers = worker_id;
return 0;
}
uint8_t * const buffer = base_buffer + tsize;
int size = readblock( infd, buffer, buffer_size + hsize ) - hsize;
bool at_stream_end = ( size < buffer_size );
if( size != buffer_size && errno )
{ if( shared_retval.set_value( 1 ) )
{ pp(); show_error( "Read error", errno ); } goto fail; }
if( size + hsize < min_member_size )
{ if( shared_retval.set_value( 2 ) ) show_file_error( pp.name(),
( size <= 0 ) ? "File ends unexpectedly at member header." :
"Input file is too short." ); goto fail; }
const Lzip_header & header = *(const Lzip_header *)buffer;
if( !header.check_magic() )
{ if( shared_retval.set_value( 2 ) )
{ show_file_error( pp.name(), bad_magic_msg ); } goto fail; }
if( !header.check_version() )
{ if( shared_retval.set_value( 2 ) )
{ pp( bad_version( header.version() ) ); } goto fail; }
tmp.dictionary_size = header.dictionary_size();
if( !isvalid_ds( tmp.dictionary_size ) )
{ if( shared_retval.set_value( 2 ) ) { pp( bad_dict_msg ); } goto fail; }
if( verbosity >= 1 ) pp();
show_progress( 0, tmp.cfile_size, &pp ); // init
unsigned long long partial_member_size = 0;
bool worker_pending = true; // start 1 worker per first packet of member
while( true )
{
if( shared_retval() ) break; // stop sending packets on error
int pos = 0; // current searching position
std::memcpy( buffer + hsize + size, lzip_magic, 4 ); // sentinel
for( int newpos = 1; newpos <= size; ++newpos )
{
while( buffer[newpos] != lzip_magic[0] ||
buffer[newpos+1] != lzip_magic[1] ||
buffer[newpos+2] != lzip_magic[2] ||
buffer[newpos+3] != lzip_magic[3] ) ++newpos;
if( newpos <= size )
{
const Lzip_trailer & trailer =
*(const Lzip_trailer *)(buffer + newpos - tsize);
const unsigned long long member_size = trailer.member_size();
if( partial_member_size + newpos - pos == member_size &&
trailer.check_consistency() )
{ // header found
const Lzip_header & header = *(const Lzip_header *)(buffer + newpos);
if( !header.check_version() )
{ if( shared_retval.set_value( 2 ) )
{ pp( bad_version( header.version() ) ); } goto fail; }
const unsigned dictionary_size = header.dictionary_size();
if( !isvalid_ds( dictionary_size ) )
{ if( shared_retval.set_value( 2 ) ) pp( bad_dict_msg );
goto fail; }
if( tmp.dictionary_size < dictionary_size )
tmp.dictionary_size = dictionary_size;
uint8_t * const data = new( std::nothrow ) uint8_t[newpos - pos];
if( !data ) goto mem_fail;
std::memcpy( data, buffer + pos, newpos - pos );
courier.receive_packet( data, newpos - pos, true ); // eom
partial_member_size = 0;
pos = newpos;
if( worker_pending )
{ if( !start_worker( worker_arg, worker_args, worker_threads,
worker_id, shared_retval ) ) goto fail;
++worker_id; }
worker_pending = worker_id < tmp.num_workers;
show_progress( member_size );
}
}
}
if( at_stream_end )
{
uint8_t * data = new( std::nothrow ) uint8_t[size + hsize - pos];
if( !data ) goto mem_fail;
std::memcpy( data, buffer + pos, size + hsize - pos );
courier.receive_packet( data, size + hsize - pos, true ); // eom
if( worker_pending &&
start_worker( worker_arg, worker_args, worker_threads,
worker_id, shared_retval ) ) ++worker_id;
break;
}
if( pos < buffer_size )
{
partial_member_size += buffer_size - pos;
uint8_t * data = new( std::nothrow ) uint8_t[buffer_size - pos];
if( !data ) goto mem_fail;
std::memcpy( data, buffer + pos, buffer_size - pos );
courier.receive_packet( data, buffer_size - pos, false );
if( worker_pending )
{ if( !start_worker( worker_arg, worker_args, worker_threads,
worker_id, shared_retval ) ) break;
++worker_id; worker_pending = false; }
}
if( courier.trailing_data_found() ) break;
std::memcpy( base_buffer, base_buffer + buffer_size, tsize + hsize );
size = readblock( infd, buffer + hsize, buffer_size );
at_stream_end = ( size < buffer_size );
if( size != buffer_size && errno )
{ if( shared_retval.set_value( 1 ) )
{ pp(); show_error( "Read error", errno ); } break; }
}
delete[] base_buffer;
courier.finish( worker_id ); // no more packets to send
tmp.num_workers = worker_id;
return 0;
}
/* Get from courier the processed and sorted packets, and write their
contents to the output file. Drain queue on error.
*/
void muxer( Packet_courier & courier, const Pretty_print & pp,
Shared_retval & shared_retval, const int outfd )
{
while( true )
{
const Packet * const opacket = courier.deliver_packet();
if( !opacket ) break; // queue is empty. all workers exited
if( shared_retval() == 0 &&
writeblock( outfd, opacket->data, opacket->size ) != opacket->size &&
shared_retval.set_value( 1 ) )
{ pp(); show_error( "Write error", errno ); }
delete opacket;
}
}
} // end namespace
/* Init the courier, then start the splitter and the workers and, if not
testing, call the muxer.
*/
int dec_stream( const unsigned long long cfile_size, const int num_workers,
const int infd, const int outfd, const Cl_options & cl_opts,
const Pretty_print & pp, const int debug_level,
const int in_slots, const int out_slots )
{
const int total_in_slots = ( INT_MAX / num_workers >= in_slots ) ?
num_workers * in_slots : INT_MAX;
in_size = 0;
out_size = 0;
Shared_retval shared_retval;
Packet_courier courier( shared_retval, num_workers, total_in_slots, out_slots );
if( debug_level & 2 ) std::fputs( "decompress stream.\n", stderr );
Worker_arg * worker_args = new( std::nothrow ) Worker_arg[num_workers];
pthread_t * worker_threads = new( std::nothrow ) pthread_t[num_workers];
if( !worker_args || !worker_threads )
{ pp( mem_msg ); delete[] worker_threads; delete[] worker_args; return 1; }
#if defined LZ_API_VERSION && LZ_API_VERSION >= 1012
const bool nocopy = ( outfd < 0 && LZ_api_version() >= 1012 );
#else
const bool nocopy = false;
#endif
Splitter_arg splitter_arg;
splitter_arg.worker_arg.courier = &courier;
splitter_arg.worker_arg.pp = &pp;
splitter_arg.worker_arg.shared_retval = &shared_retval;
splitter_arg.worker_arg.worker_id = 0;
splitter_arg.worker_arg.ignore_trailing = cl_opts.ignore_trailing;
splitter_arg.worker_arg.loose_trailing = cl_opts.loose_trailing;
splitter_arg.worker_arg.testing = ( outfd < 0 );
splitter_arg.worker_arg.nocopy = nocopy;
splitter_arg.worker_args = worker_args;
splitter_arg.worker_threads = worker_threads;
splitter_arg.cfile_size = cfile_size;
splitter_arg.infd = infd;
splitter_arg.num_workers = num_workers;
pthread_t splitter_thread;
int errcode = pthread_create( &splitter_thread, 0, dsplitter_s, &splitter_arg );
if( errcode )
{ show_error( "Can't create splitter thread", errcode );
delete[] worker_threads; delete[] worker_args; return 1; }
if( outfd >= 0 ) muxer( courier, pp, shared_retval, outfd );
errcode = pthread_join( splitter_thread, 0 );
if( errcode && shared_retval.set_value( 1 ) )
show_error( "Can't join splitter thread", errcode );
for( int i = splitter_arg.num_workers; --i >= 0; )
{ // join only the workers started
errcode = pthread_join( worker_threads[i], 0 );
if( errcode && shared_retval.set_value( 1 ) )
show_error( "Can't join worker threads", errcode );
}
delete[] worker_threads;
delete[] worker_args;
if( shared_retval() ) return shared_retval(); // some thread found a problem
show_results( in_size, out_size, splitter_arg.dictionary_size, outfd < 0 );
if( debug_level & 1 )
{
std::fprintf( stderr,
"workers started %8u\n"
"any worker tried to consume from splitter %8u times\n"
"any worker had to wait %8u times\n",
splitter_arg.num_workers,
courier.icheck_counter, courier.iwait_counter );
if( outfd >= 0 )
std::fprintf( stderr,
"muxer tried to consume from workers %8u times\n"
"muxer had to wait %8u times\n",
courier.ocheck_counter, courier.owait_counter );
}
if( !courier.finished() ) internal_error( "courier not finished." );
return 0;
}