/* Copyright (c) 2003-2012, Arvid Norberg All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. * Neither the name of the author nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "libtorrent/pch.hpp" #include #include #include #include #include #include #include "libtorrent/piece_picker.hpp" #include "libtorrent/bitfield.hpp" #include "libtorrent/random.hpp" #if defined TORRENT_DEBUG || TORRENT_RELEASE_ASSERTS #include "libtorrent/peer_connection.hpp" #include "libtorrent/torrent.hpp" #include "libtorrent/policy.hpp" // for policy::peer #endif #include "libtorrent/invariant_check.hpp" #define TORRENT_PIECE_PICKER_INVARIANT_CHECK INVARIANT_CHECK //#define TORRENT_NO_EXPENSIVE_INVARIANT_CHECK //#define TORRENT_PIECE_PICKER_INVARIANT_CHECK //#define TORRENT_PICKER_LOG namespace libtorrent { const piece_block piece_block::invalid(0x7FFFF, 0x1FFF); piece_picker::piece_picker() : m_seeds(0) , m_priority_boundries(1, int(m_pieces.size())) , m_blocks_per_piece(0) , m_blocks_in_last_piece(0) , m_num_filtered(0) , m_num_have_filtered(0) , m_num_have(0) , m_cursor(0) , m_reverse_cursor(0) , m_sparse_regions(1) , m_dirty(false) { #ifdef TORRENT_PICKER_LOG std::cerr << "new piece_picker" << std::endl; #endif #if defined TORRENT_DEBUG && !defined TORRENT_DISABLE_INVARIANT_CHECKS check_invariant(); #endif } void piece_picker::init(int blocks_per_piece, int blocks_in_last_piece, int total_num_pieces) { TORRENT_ASSERT(blocks_per_piece > 0); TORRENT_ASSERT(total_num_pieces > 0); #ifdef TORRENT_PICKER_LOG std::cerr << "piece_picker::init()" << std::endl; #endif // allocate the piece_map to cover all pieces // and make them invalid (as if we don't have a single piece) m_piece_map.resize(total_num_pieces, piece_pos(0, 0)); m_reverse_cursor = int(m_piece_map.size()); m_cursor = 0; m_downloads.clear(); m_block_info.clear(); m_num_filtered += m_num_have_filtered; m_num_have_filtered = 0; m_num_have = 0; m_dirty = true; for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { i->peer_count = 0; i->downloading = 0; i->index = 0; #ifdef TORRENT_DEBUG_REFCOUNTS i->have_peers.clear(); #endif } for (std::vector::iterator i = m_piece_map.begin() + m_cursor , end(m_piece_map.end()); i != end && (i->have() || i->filtered()); ++i, ++m_cursor); for (std::vector::reverse_iterator i = m_piece_map.rend() - m_reverse_cursor; m_reverse_cursor > 0 && (i->have() || i->filtered()); ++i, --m_reverse_cursor); // the piece index is stored in 20 bits, which limits the allowed // number of pieces somewhat TORRENT_ASSERT(m_piece_map.size() < piece_pos::we_have_index); m_blocks_per_piece = blocks_per_piece; m_blocks_in_last_piece = blocks_in_last_piece; if (m_blocks_in_last_piece == 0) m_blocks_in_last_piece = blocks_per_piece; TORRENT_ASSERT(m_blocks_in_last_piece <= m_blocks_per_piece); } void piece_picker::increase_num_pieces(int total_num_pieces) { #ifdef TORRENT_PICKER_LOG std::cerr << "piece_picker::increase_num_pieces(" << total_num_pieces << ")" << std::endl; #endif int old_num_pieces = m_piece_map.size(); // allocate the piece_map to cover all pieces // and make them invalid m_piece_map.resize(total_num_pieces, piece_pos(0, 0)); m_reverse_cursor = int(m_piece_map.size()); //[MF] check this //m_downloads.clear(); //m_block_info.clear(); m_dirty = true; for (std::vector::iterator i = m_piece_map.begin() + old_num_pieces , end(m_piece_map.end()); i != end; ++i) { i->peer_count = 0; i->downloading = 0; i->index = 0; #ifdef TORRENT_DEBUG_REFCOUNTS i->have_peers.clear(); #endif } for (std::vector::reverse_iterator i = m_piece_map.rend() - m_reverse_cursor; m_reverse_cursor > 0 && (i->have() || i->filtered()); ++i, --m_reverse_cursor); // the piece index is stored in 20 bits, which limits the allowed // number of pieces somewhat TORRENT_ASSERT(m_piece_map.size() < piece_pos::we_have_index); update_pieces(); } void piece_picker::piece_info(int index, piece_picker::downloading_piece& st) const { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < int(m_piece_map.size())); if (m_piece_map[index].downloading) { std::vector::const_iterator piece = find_dl_piece(index); TORRENT_ASSERT(piece != m_downloads.end()); st = *piece; st.info = 0; return; } st.info = 0; st.index = index; st.writing = 0; st.requested = 0; if (m_piece_map[index].have()) { st.finished = blocks_in_piece(index); return; } st.finished = 0; } piece_picker::downloading_piece& piece_picker::add_download_piece(int piece) { int num_downloads = m_downloads.size(); int block_index = num_downloads * m_blocks_per_piece; if (int(m_block_info.size()) < block_index + m_blocks_per_piece) { block_info* base = 0; if (!m_block_info.empty()) base = &m_block_info[0]; m_block_info.resize(block_index + m_blocks_per_piece); if (!m_downloads.empty() && &m_block_info[0] != base) { // this means the memory was reallocated, update the pointers for (int i = 0; i < int(m_downloads.size()); ++i) m_downloads[i].info = &m_block_info[m_downloads[i].info - base]; } } downloading_piece cmp; cmp.index = piece; std::vector::iterator i = std::lower_bound(m_downloads.begin() , m_downloads.end(), cmp); TORRENT_ASSERT(i == m_downloads.end() || i->index != piece); i = m_downloads.insert(i, downloading_piece()); downloading_piece& ret = *i; ret.index = piece; ret.info = &m_block_info[block_index]; for (int i = 0; i < m_blocks_per_piece; ++i) { ret.info[i].num_peers = 0; ret.info[i].state = block_info::state_none; ret.info[i].peer = 0; #if defined TORRENT_DEBUG || TORRENT_RELEASE_ASSERTS ret.info[i].piece_index = piece; #endif } return ret; } void piece_picker::erase_download_piece(std::vector::iterator i) { std::vector::iterator other = std::find_if( m_downloads.begin(), m_downloads.end() , boost::bind(&downloading_piece::info, _1) == &m_block_info[(m_downloads.size() - 1) * m_blocks_per_piece]); TORRENT_ASSERT(other != m_downloads.end()); if (i != other) { std::copy(other->info, other->info + m_blocks_per_piece, i->info); other->info = i->info; } m_piece_map[i->index].downloading = false; m_downloads.erase(i); } #ifdef TORRENT_DEBUG void piece_picker::verify_pick(std::vector const& picked , bitfield const& bits) const { TORRENT_ASSERT(bits.size() == m_piece_map.size()); for (std::vector::const_iterator i = picked.begin() , end(picked.end()); i != end; ++i) { TORRENT_ASSERT(i->piece_index >= 0); TORRENT_ASSERT(i->piece_index < bits.size()); TORRENT_ASSERT(bits[i->piece_index]); TORRENT_ASSERT(!m_piece_map[i->piece_index].have()); TORRENT_ASSERT(!m_piece_map[i->piece_index].filtered()); } } void piece_picker::verify_priority(int range_start, int range_end, int prio) const { TORRENT_ASSERT(range_start <= range_end); TORRENT_ASSERT(range_end <= int(m_pieces.size())); for (std::vector::const_iterator i = m_pieces.begin() + range_start , end(m_pieces.begin() + range_end); i != end; ++i) { int index = *i; TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < int(m_piece_map.size())); int p = m_piece_map[index].priority(this); TORRENT_ASSERT(p == prio); } } #if defined TORRENT_PICKER_LOG void piece_picker::print_pieces() const { for (std::vector::const_iterator i = m_priority_boundries.begin() , end(m_priority_boundries.end()); i != end; ++i) { std::cerr << *i << " "; } std::cout << std::endl; int index = 0; std::vector::const_iterator j = m_priority_boundries.begin(); for (std::vector::const_iterator i = m_pieces.begin() , end(m_pieces.end()); i != end; ++i, ++index) { if (*i == -1) break; while (j != m_priority_boundries.end() && *j <= index) { std::cerr << "| "; ++j; } std::cerr << *i << "(" << m_piece_map[*i].index << ") "; } std::cerr << std::endl; } #endif // TORRENT_PIECE_PICKER #endif // TORRENT_DEBUG #if defined TORRENT_DEBUG && !defined TORRENT_DISABLE_INVARIANT_CHECKS void piece_picker::check_invariant(const torrent* t) const { #ifndef TORRENT_DEBUG_REFCOUNTS #if TORRENT_COMPACT_PICKER TORRENT_ASSERT(sizeof(piece_pos) == 4); #else TORRENT_ASSERT(sizeof(piece_pos) == 8); #endif #endif TORRENT_ASSERT(m_num_have >= 0); TORRENT_ASSERT(m_num_have_filtered >= 0); TORRENT_ASSERT(m_num_filtered >= 0); TORRENT_ASSERT(m_seeds >= 0); if (!m_downloads.empty()) { for (std::vector::const_iterator i = m_downloads.begin(); i != m_downloads.end() - 1; ++i) { downloading_piece const& dp = *i; downloading_piece const& next = *(i + 1); // TORRENT_ASSERT(dp.finished + dp.writing >= next.finished + next.writing); TORRENT_ASSERT(dp.index < next.index); } } if (t != 0) TORRENT_ASSERT((int)m_piece_map.size() == t->torrent_file().num_pieces()); for (std::vector::const_iterator i = m_downloads.begin() , end(m_downloads.end()); i != end; ++i) { bool blocks_requested = false; int num_blocks = blocks_in_piece(i->index); int num_requested = 0; int num_finished = 0; int num_writing = 0; for (int k = 0; k < num_blocks; ++k) { TORRENT_ASSERT(i->info[k].piece_index == i->index); TORRENT_ASSERT(i->info[k].peer == 0 || static_cast(i->info[k].peer)->in_use); if (i->info[k].state == block_info::state_finished) { ++num_finished; TORRENT_ASSERT(i->info[k].num_peers == 0); } else if (i->info[k].state == block_info::state_requested) { ++num_requested; blocks_requested = true; TORRENT_ASSERT(i->info[k].num_peers > 0); } else if (i->info[k].state == block_info::state_writing) { ++num_writing; TORRENT_ASSERT(i->info[k].num_peers == 0); } } TORRENT_ASSERT(blocks_requested == (i->state != none)); TORRENT_ASSERT(num_requested == i->requested); TORRENT_ASSERT(num_writing == i->writing); TORRENT_ASSERT(num_finished == i->finished); if (m_piece_map[i->index].full) TORRENT_ASSERT(num_finished + num_writing + num_requested == num_blocks); } int num_pieces = int(m_piece_map.size()); TORRENT_ASSERT(m_cursor >= 0); TORRENT_ASSERT(m_cursor <= num_pieces); TORRENT_ASSERT(m_reverse_cursor <= num_pieces); TORRENT_ASSERT(m_reverse_cursor >= 0); TORRENT_ASSERT(m_reverse_cursor > m_cursor || (m_cursor == num_pieces && m_reverse_cursor == 0)); #ifdef TORRENT_NO_EXPENSIVE_INVARIANT_CHECK return; #endif if (!m_dirty) { TORRENT_ASSERT(!m_priority_boundries.empty()); int prio = 0; int start = 0; for (std::vector::const_iterator i = m_priority_boundries.begin() , end(m_priority_boundries.end()); i != end; ++i) { verify_priority(start, *i, prio); ++prio; start = *i; } TORRENT_ASSERT(m_priority_boundries.back() == int(m_pieces.size())); } int index = 0; for (std::vector::const_iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end && (i->have() || i->filtered()); ++i, ++index); TORRENT_ASSERT(m_cursor == index); index = num_pieces; if (num_pieces > 0) { for (std::vector::reverse_iterator i = m_piece_map.rend() - index; index > 0 && (i->have() || i->filtered()); ++i, --index); TORRENT_ASSERT(index == num_pieces || m_piece_map[index].have() || m_piece_map[index].filtered()); TORRENT_ASSERT(m_reverse_cursor == index); } else { TORRENT_ASSERT(m_reverse_cursor == 0); } int num_filtered = 0; int num_have_filtered = 0; int num_have = 0; for (std::vector::const_iterator i = m_piece_map.begin(); i != m_piece_map.end(); ++i) { int index = static_cast(i - m_piece_map.begin()); piece_pos const& p = *i; if (p.filtered()) { if (p.index != piece_pos::we_have_index) ++num_filtered; else ++num_have_filtered; } #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(p.have_peers.size() == p.peer_count + m_seeds); #endif if (p.index == piece_pos::we_have_index) ++num_have; #if 0 if (t != 0) { int actual_peer_count = 0; for (torrent::const_peer_iterator peer = t->begin(); peer != t->end(); ++peer) { if (peer->second->has_piece(index)) actual_peer_count++; } TORRENT_ASSERT((int)i->peer_count == actual_peer_count); /* int num_downloaders = 0; for (std::vector::const_iterator peer = t->begin(); peer != t->end(); ++peer) { const std::vector& queue = (*peer)->download_queue(); if (std::find_if(queue.begin(), queue.end(), has_index(index)) == queue.end()) continue; ++num_downloaders; } if (i->downloading) { TORRENT_ASSERT(num_downloaders == 1); } else { TORRENT_ASSERT(num_downloaders == 0); } */ } #endif if (p.index == piece_pos::we_have_index) { TORRENT_ASSERT(t == 0 || t->have_piece(index)); TORRENT_ASSERT(p.downloading == 0); } if (t != 0) TORRENT_ASSERT(!t->have_piece(index)); int prio = p.priority(this); TORRENT_ASSERT(prio == -1 || p.downloading == (prio % piece_picker::prio_factor == 0)); if (!m_dirty) { TORRENT_ASSERT(prio < int(m_priority_boundries.size()) || m_dirty); if (prio >= 0) { TORRENT_ASSERT(p.index < m_pieces.size()); TORRENT_ASSERT(m_pieces[p.index] == index); } else { TORRENT_ASSERT(prio == -1); // make sure there's no entry // with this index. (there shouldn't // be since the priority is -1) TORRENT_ASSERT(std::find(m_pieces.begin(), m_pieces.end(), index) == m_pieces.end()); } } int count = std::count_if(m_downloads.begin(), m_downloads.end() , has_index(index)); if (i->downloading == 1) { TORRENT_ASSERT(count == 1); } else { TORRENT_ASSERT(count == 0); } } TORRENT_ASSERT(num_have == m_num_have); TORRENT_ASSERT(num_filtered == m_num_filtered); TORRENT_ASSERT(num_have_filtered == m_num_have_filtered); if (!m_dirty) { for (std::vector::const_iterator i = m_pieces.begin() , end(m_pieces.end()); i != end; ++i) { TORRENT_ASSERT(m_piece_map[*i].priority(this) >= 0); } } } #endif std::pair piece_picker::distributed_copies() const { TORRENT_ASSERT(m_seeds >= 0); const int num_pieces = m_piece_map.size(); if (num_pieces == 0) return std::make_pair(1, 0); int min_availability = piece_pos::max_peer_count; // find the lowest availability count // count the number of pieces that have that availability // and also the number of pieces that have more than that. int integer_part = 0; int fraction_part = 0; for (std::vector::const_iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { int peer_count = int(i->peer_count); // take ourself into account if (i->have()) ++peer_count; if (min_availability > peer_count) { min_availability = peer_count; fraction_part += integer_part; integer_part = 1; } else if (peer_count == min_availability) { ++integer_part; } else { TORRENT_ASSERT(peer_count > min_availability); ++fraction_part; } } TORRENT_ASSERT(integer_part + fraction_part == num_pieces); return std::make_pair(min_availability + m_seeds, fraction_part * 1000 / num_pieces); } void piece_picker::priority_range(int prio, int* start, int* end) { TORRENT_ASSERT(prio >= 0); TORRENT_ASSERT(prio < int(m_priority_boundries.size()) || m_dirty); if (prio == 0) *start = 0; else *start = m_priority_boundries[prio - 1]; *end = m_priority_boundries[prio]; TORRENT_ASSERT(*start <= *end); } void piece_picker::add(int index) { TORRENT_ASSERT(!m_dirty); TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < int(m_piece_map.size())); piece_pos& p = m_piece_map[index]; TORRENT_ASSERT(!p.filtered()); TORRENT_ASSERT(!p.have()); int priority = p.priority(this); TORRENT_ASSERT(priority >= 0); if (int(m_priority_boundries.size()) <= priority) m_priority_boundries.resize(priority + 1, m_pieces.size()); TORRENT_ASSERT(int(m_priority_boundries.size()) >= priority); int range_start, range_end; priority_range(priority, &range_start, &range_end); int new_index; if (range_end == range_start) new_index = range_start; else new_index = random() % (range_end - range_start + 1) + range_start; #ifdef TORRENT_PICKER_LOG std::cerr << "add " << index << " (" << priority << ")" << std::endl; print_pieces(); #endif m_pieces.push_back(-1); for (;;) { TORRENT_ASSERT(new_index < int(m_pieces.size())); int temp = m_pieces[new_index]; m_pieces[new_index] = index; m_piece_map[index].index = new_index; index = temp; do { temp = m_priority_boundries[priority]++; ++priority; } while (temp == new_index && priority < int(m_priority_boundries.size())); new_index = temp; #ifdef TORRENT_PICKER_LOG print_pieces(); std::cerr << " index: " << index << " prio: " << priority << " new_index: " << new_index << std::endl; #endif if (priority >= int(m_priority_boundries.size())) break; TORRENT_ASSERT(temp >= 0); } if (index != -1) { TORRENT_ASSERT(new_index == int(m_pieces.size() - 1)); m_pieces[new_index] = index; m_piece_map[index].index = new_index; #ifdef TORRENT_PICKER_LOG print_pieces(); #endif // shuffle(priority, new_index); #ifdef TORRENT_PICKER_LOG // print_pieces(); #endif } } void piece_picker::remove(int priority, int elem_index) { TORRENT_ASSERT(!m_dirty); TORRENT_ASSERT(priority >= 0); TORRENT_ASSERT(elem_index < int(m_pieces.size())); TORRENT_ASSERT(elem_index >= 0); #ifdef TORRENT_PICKER_LOG std::cerr << "remove " << m_pieces[elem_index] << " (" << priority << ")" << std::endl; #endif int next_index = elem_index; TORRENT_ASSERT(m_piece_map[m_pieces[elem_index]].priority(this) == -1); for (;;) { #ifdef TORRENT_PICKER_LOG print_pieces(); #endif TORRENT_ASSERT(elem_index < int(m_pieces.size())); int temp; do { temp = --m_priority_boundries[priority]; ++priority; } while (next_index == temp && priority < int(m_priority_boundries.size())); if (next_index == temp) break; next_index = temp; int piece = m_pieces[next_index]; m_pieces[elem_index] = piece; m_piece_map[piece].index = elem_index; TORRENT_ASSERT(m_piece_map[piece].priority(this) == priority - 1); TORRENT_ASSERT(elem_index < int(m_pieces.size() - 1)); elem_index = next_index; if (priority == int(m_priority_boundries.size())) break; } m_pieces.pop_back(); TORRENT_ASSERT(next_index == int(m_pieces.size())); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif } // will update the piece with the given properties (priority, elem_index) // to place it at the correct position void piece_picker::update(int priority, int elem_index) { TORRENT_ASSERT(!m_dirty); TORRENT_ASSERT(priority >= 0); TORRENT_ASSERT(elem_index >= 0); TORRENT_ASSERT(int(m_priority_boundries.size()) > priority); int index = m_pieces[elem_index]; // update the piece_map piece_pos& p = m_piece_map[index]; TORRENT_ASSERT(int(p.index) == elem_index || p.have()); int new_priority = p.priority(this); if (new_priority == priority) return; if (new_priority == -1) { remove(priority, elem_index); return; } if (int(m_priority_boundries.size()) <= new_priority) m_priority_boundries.resize(new_priority + 1, m_pieces.size()); #ifdef TORRENT_PICKER_LOG std::cerr << "update " << index << " (" << priority << "->" << new_priority << ")" << std::endl; #endif if (priority > new_priority) { int new_index; int temp = index; for (;;) { #ifdef TORRENT_PICKER_LOG print_pieces(); #endif --priority; new_index = m_priority_boundries[priority]++; TORRENT_ASSERT(new_index < int(m_pieces.size())); if (temp != m_pieces[new_index]) { temp = m_pieces[new_index]; m_pieces[elem_index] = temp; m_piece_map[temp].index = elem_index; TORRENT_ASSERT(elem_index < int(m_pieces.size())); } elem_index = new_index; if (priority == new_priority) break; } #ifdef TORRENT_PICKER_LOG print_pieces(); #endif m_pieces[elem_index] = index; m_piece_map[index].index = elem_index; TORRENT_ASSERT(elem_index < int(m_pieces.size())); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif shuffle(priority, elem_index); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif TORRENT_ASSERT(m_piece_map[index].priority(this) == priority); } else { int new_index; int temp = index; for (;;) { #ifdef TORRENT_PICKER_LOG print_pieces(); #endif new_index = --m_priority_boundries[priority]; TORRENT_ASSERT(new_index < int(m_pieces.size())); if (temp != m_pieces[new_index]) { temp = m_pieces[new_index]; m_pieces[elem_index] = temp; m_piece_map[temp].index = elem_index; TORRENT_ASSERT(elem_index < int(m_pieces.size())); } elem_index = new_index; ++priority; if (priority == new_priority) break; } #ifdef TORRENT_PICKER_LOG print_pieces(); #endif m_pieces[elem_index] = index; m_piece_map[index].index = elem_index; TORRENT_ASSERT(elem_index < int(m_pieces.size())); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif shuffle(priority, elem_index); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif TORRENT_ASSERT(m_piece_map[index].priority(this) == priority); } } void piece_picker::shuffle(int priority, int elem_index) { #ifdef TORRENT_PICKER_LOG std::cerr << "shuffle()" << std::endl; #endif TORRENT_ASSERT(!m_dirty); TORRENT_ASSERT(priority >= 0); TORRENT_ASSERT(elem_index >= 0); TORRENT_ASSERT(elem_index < int(m_pieces.size())); TORRENT_ASSERT(m_piece_map[m_pieces[elem_index]].priority(this) == priority); int range_start, range_end; priority_range(priority, &range_start, &range_end); TORRENT_ASSERT(range_start < range_end); int other_index = random() % (range_end - range_start) + range_start; if (other_index == elem_index) return; // swap other_index with elem_index piece_pos& p1 = m_piece_map[m_pieces[other_index]]; piece_pos& p2 = m_piece_map[m_pieces[elem_index]]; int temp = p1.index; p1.index = p2.index; p2.index = temp; std::swap(m_pieces[other_index], m_pieces[elem_index]); } /* void piece_picker::sort_piece(std::vector::iterator dp) { TORRENT_ASSERT(m_piece_map[dp->index].downloading); int complete = dp->writing + dp->finished; if (dp != m_downloads.begin()) { for (std::vector::iterator j(dp-1); dp != m_downloads.begin(); --dp, --j) { TORRENT_ASSERT(j >= m_downloads.begin()); if (j->finished + j->writing >= complete) break; using std::swap; swap(*j, *dp); if (j == m_downloads.begin()) return; } } TORRENT_ASSERT(dp != m_downloads.end()); for (std::vector::iterator j(dp+1); dp != m_downloads.end() - 1; ++dp, ++j) { TORRENT_ASSERT(j < m_downloads.end()); if (j->finished + j->writing <= complete) break; using std::swap; swap(*j, *dp); if (j == m_downloads.end() - 1) return; } } */ void piece_picker::restore_piece(int index) { TORRENT_PIECE_PICKER_INVARIANT_CHECK; TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size()); TORRENT_ASSERT(m_piece_map[index].downloading == 1); std::vector::iterator i = find_dl_piece(index); TORRENT_ASSERT(i != m_downloads.end()); #ifdef TORRENT_DEBUG int num_blocks = blocks_in_piece(i->index); for (int k = 0; k < num_blocks; ++k) { TORRENT_ASSERT(i->info[k].piece_index == index); TORRENT_ASSERT(i->info[k].state == block_info::state_finished); TORRENT_ASSERT(i->info[k].num_peers == 0); } #endif piece_pos& p = m_piece_map[index]; int prev_priority = p.priority(this); erase_download_piece(i); int new_priority = p.priority(this); if (new_priority == prev_priority) return; if (m_dirty) return; if (prev_priority == -1) add(index); else update(prev_priority, p.index); } void piece_picker::inc_refcount_all(const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif ++m_seeds; if (m_seeds == 1) { // when m_seeds is increased from 0 to 1 // we may have to add pieces that previously // didn't have any peers m_dirty = true; } #ifdef TORRENT_DEBUG_REFCOUNTS for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { TORRENT_ASSERT(i->have_peers.count(peer) == 0); i->have_peers.insert(peer); } #endif } void piece_picker::dec_refcount_all(const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif if (m_seeds > 0) { --m_seeds; if (m_seeds == 0) { // when m_seeds is decreased from 1 to 0 // we may have to remove pieces that previously // didn't have any peers m_dirty = true; } #ifdef TORRENT_DEBUG_REFCOUNTS for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { TORRENT_ASSERT(i->have_peers.count(peer) == 1); i->have_peers.erase(peer); } #endif return; } TORRENT_ASSERT(m_seeds == 0); for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(i->have_peers.count(peer) == 1); i->have_peers.erase(peer); #endif TORRENT_ASSERT(i->peer_count > 0); --i->peer_count; } m_dirty = true; } void piece_picker::inc_refcount(int index, const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif piece_pos& p = m_piece_map[index]; #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(p.have_peers.count(peer) == 0); p.have_peers.insert(peer); #endif int prev_priority = p.priority(this); ++p.peer_count; if (m_dirty) return; int new_priority = p.priority(this); if (prev_priority == new_priority) return; if (prev_priority == -1) add(index); else update(prev_priority, p.index); } // this function decrements the m_seeds counter // and increments the peer counter on every piece // instead. Sometimes of we connect to a seed that // later sends us a dont-have message, we'll need to // turn that m_seed into counts on the pieces since // they can't be negative void piece_picker::break_one_seed() { TORRENT_PIECE_PICKER_INVARIANT_CHECK; TORRENT_ASSERT(m_seeds > 0); --m_seeds; for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { ++i->peer_count; } m_dirty = true; } void piece_picker::dec_refcount(int index, const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif piece_pos& p = m_piece_map[index]; #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(p.have_peers.count(peer) == 1); p.have_peers.erase(peer); #endif if (p.peer_count == 0) { TORRENT_ASSERT(m_seeds > 0); // this is the case where we have one or more // seeds, and one of them saying: I don't have this // piece anymore. we need to break up one of the seed // counters into actual peer counters on the pieces break_one_seed(); } int prev_priority = p.priority(this); TORRENT_ASSERT(p.peer_count > 0); --p.peer_count; if (m_dirty) return; if (prev_priority >= 0) update(prev_priority, p.index); } void piece_picker::inc_refcount(bitfield const& bitmask, const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(bitmask.size() == m_piece_map.size()); int index = 0; bool updated = false; for (bitfield::const_iterator i = bitmask.begin() , end(bitmask.end()); i != end; ++i, ++index) { if (*i) { #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(m_piece_map[index].have_peers.count(peer) == 0); m_piece_map[index].have_peers.insert(peer); #endif ++m_piece_map[index].peer_count; updated = true; } } if (updated) m_dirty = true; } void piece_picker::dec_refcount(bitfield const& bitmask, const void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(bitmask.size() <= m_piece_map.size()); int index = 0; bool updated = false; #if defined TORRENT_DEBUG || TORRENT_RELEASE_ASSERTS bool seed_broken = false; #endif for (bitfield::const_iterator i = bitmask.begin() , end(bitmask.end()); i != end; ++i, ++index) { if (*i) { #ifdef TORRENT_DEBUG_REFCOUNTS TORRENT_ASSERT(m_piece_map[index].have_peers.count(peer) == 1); m_piece_map[index].have_peers.erase(peer); #endif piece_pos& p = m_piece_map[index]; if (p.peer_count == 0) { TORRENT_ASSERT(!seed_broken); TORRENT_ASSERT(m_seeds > 0); // this is the case where we have one or more // seeds, and one of them saying: I don't have this // piece anymore. we need to break up one of the seed // counters into actual peer counters on the pieces break_one_seed(); #if defined TORRENT_DEBUG || TORRENT_RELEASE_ASSERTS seed_broken = true; #endif } --p.peer_count; updated = true; } } if (updated) m_dirty = true; } void piece_picker::update_pieces() const { TORRENT_ASSERT(m_dirty); if (m_priority_boundries.empty()) m_priority_boundries.resize(1, 0); #ifdef TORRENT_PICKER_LOG std::cerr << "update_pieces" << std::endl; #endif std::fill(m_priority_boundries.begin(), m_priority_boundries.end(), 0); for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i) { int prio = i->priority(this); if (prio == -1) continue; if (prio >= int(m_priority_boundries.size())) m_priority_boundries.resize(prio + 1, 0); i->index = m_priority_boundries[prio]; ++m_priority_boundries[prio]; } #ifdef TORRENT_PICKER_LOG print_pieces(); #endif int index = 0; for (std::vector::iterator i = m_priority_boundries.begin() , end(m_priority_boundries.end()); i != end; ++i) { *i += index; index = *i; } m_pieces.resize(index, 0); #ifdef TORRENT_PICKER_LOG print_pieces(); #endif index = 0; for (std::vector::iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i, ++index) { piece_pos& p = *i; int prio = p.priority(this); if (prio == -1) continue; int new_index = (prio == 0 ? 0 : m_priority_boundries[prio - 1]) + p.index; m_pieces[new_index] = index; } int start = 0; for (std::vector::iterator i = m_priority_boundries.begin() , end(m_priority_boundries.end()); i != end; ++i) { if (start == *i) continue; std::random_shuffle(&m_pieces[0] + start, &m_pieces[0] + *i); start = *i; } index = 0; for (std::vector::const_iterator i = m_pieces.begin() , end(m_pieces.end()); i != end; ++i, ++index) { TORRENT_ASSERT(*i >= 0 && *i < int(m_piece_map.size())); m_piece_map[*i].index = index; } m_dirty = false; #ifdef TORRENT_PICKER_LOG print_pieces(); #endif } void piece_picker::we_dont_have(int index) { TORRENT_PIECE_PICKER_INVARIANT_CHECK; TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size()); piece_pos& p = m_piece_map[index]; TORRENT_ASSERT(p.downloading == 0); #ifdef TORRENT_PICKER_LOG std::cerr << "piece_picker::we_dont_have(" << index << ")" << std::endl; #endif if (!p.have()) return; if (p.filtered()) { ++m_num_filtered; --m_num_have_filtered; } else { // update cursors if (index < m_cursor) m_cursor = index; if (index >= m_reverse_cursor) m_reverse_cursor = index + 1; if (m_reverse_cursor == m_cursor) { m_reverse_cursor = 0; m_cursor = num_pieces(); } } --m_num_have; p.set_not_have(); if (m_dirty) return; if (p.priority(this) >= 0) add(index); } // this is used to indicate that we succesfully have // downloaded a piece, and that no further attempts // to pick that piece should be made. The piece will // be removed from the available piece list. void piece_picker::we_have(int index) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size()); #ifdef TORRENT_PICKER_LOG std::cerr << "piece_picker::we_have(" << index << ")" << std::endl; #endif piece_pos& p = m_piece_map[index]; int info_index = p.index; int priority = p.priority(this); TORRENT_ASSERT(priority < int(m_priority_boundries.size()) || m_dirty); if (p.downloading) { std::vector::iterator i = find_dl_piece(index); TORRENT_ASSERT(i != m_downloads.end()); erase_download_piece(i); } TORRENT_ASSERT(find_dl_piece(index) == m_downloads.end()); if (p.have()) return; // maintain sparse_regions if (index == 0) { if (index == int(m_piece_map.size()) - 1 || m_piece_map[index + 1].have()) --m_sparse_regions; } else if (index == int(m_piece_map.size() - 1)) { if (index == 0 || m_piece_map[index - 1].have()) --m_sparse_regions; } else { bool have_before = m_piece_map[index-1].have(); bool have_after = m_piece_map[index+1].have(); if (have_after && have_before) --m_sparse_regions; else if (!have_after && !have_before) ++m_sparse_regions; } if (p.filtered()) { --m_num_filtered; ++m_num_have_filtered; } ++m_num_have; p.set_have(); if (m_cursor == m_reverse_cursor - 1 && m_cursor == index) { m_cursor = int(m_piece_map.size()); m_reverse_cursor = 0; TORRENT_ASSERT(num_pieces() > 0); } else if (m_cursor == index) { ++m_cursor; for (std::vector::const_iterator i = m_piece_map.begin() + m_cursor , end(m_piece_map.end()); i != end && (i->have() || i->filtered()); ++i, ++m_cursor); } else if (m_reverse_cursor - 1 == index) { --m_reverse_cursor; TORRENT_ASSERT(m_piece_map[m_reverse_cursor].have() || m_piece_map[m_reverse_cursor].filtered()); for (std::vector::const_iterator i = m_piece_map.begin() + m_reverse_cursor - 1; m_reverse_cursor > 0 && (i->have() || i->filtered()); --i, --m_reverse_cursor); TORRENT_ASSERT(m_piece_map[m_reverse_cursor].have() || m_piece_map[m_reverse_cursor].filtered()); } TORRENT_ASSERT(m_reverse_cursor > m_cursor || (m_cursor == num_pieces() && m_reverse_cursor == 0)); if (priority == -1) return; if (m_dirty) return; remove(priority, info_index); TORRENT_ASSERT(p.priority(this) == -1); } bool piece_picker::set_piece_priority(int index, int new_piece_priority) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(new_piece_priority >= 0); TORRENT_ASSERT(new_piece_priority <= 7); TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size()); piece_pos& p = m_piece_map[index]; // if the priority isn't changed, don't do anything if (new_piece_priority == int(p.piece_priority)) return false; int prev_priority = p.priority(this); TORRENT_ASSERT(m_dirty || prev_priority < int(m_priority_boundries.size())); bool ret = false; if (new_piece_priority == piece_pos::filter_priority && p.piece_priority != piece_pos::filter_priority) { // the piece just got filtered if (p.have()) { ++m_num_have_filtered; } else { ++m_num_filtered; // update m_cursor if (m_cursor == m_reverse_cursor - 1 && m_cursor == index) { m_cursor = int(m_piece_map.size()); m_reverse_cursor = 0; } else if (m_cursor == index) { ++m_cursor; while (m_cursor < int(m_piece_map.size()) && (m_piece_map[m_cursor].have() || m_piece_map[m_cursor].filtered())) ++m_cursor; } else if (m_reverse_cursor == index + 1) { --m_reverse_cursor; while (m_reverse_cursor > 0 && (m_piece_map[m_reverse_cursor-1].have() || m_piece_map[m_reverse_cursor-1].filtered())) --m_reverse_cursor; } } ret = true; } else if (new_piece_priority != piece_pos::filter_priority && p.piece_priority == piece_pos::filter_priority) { // the piece just got unfiltered if (p.have()) { --m_num_have_filtered; } else { --m_num_filtered; // update cursors if (index < m_cursor) m_cursor = index; if (index >= m_reverse_cursor) m_reverse_cursor = index + 1; if (m_reverse_cursor == m_cursor) { m_reverse_cursor = 0; m_cursor = num_pieces(); } } ret = true; } TORRENT_ASSERT(m_num_filtered >= 0); TORRENT_ASSERT(m_num_have_filtered >= 0); p.piece_priority = new_piece_priority; int new_priority = p.priority(this); if (prev_priority == new_priority) return ret; if (m_dirty) return ret; if (prev_priority == -1) { add(index); } else { update(prev_priority, p.index); } return ret; } int piece_picker::piece_priority(int index) const { TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size()); return m_piece_map[index].piece_priority; } void piece_picker::piece_priorities(std::vector& pieces) const { pieces.resize(m_piece_map.size()); std::vector::iterator j = pieces.begin(); for (std::vector::const_iterator i = m_piece_map.begin(), end(m_piece_map.end()); i != end; ++i, ++j) { *j = i->piece_priority; } } // ============ start deprecation ============== void piece_picker::filtered_pieces(std::vector& mask) const { mask.resize(m_piece_map.size()); std::vector::iterator j = mask.begin(); for (std::vector::const_iterator i = m_piece_map.begin(), end(m_piece_map.end()); i != end; ++i, ++j) { *j = i->filtered(); } } // ============ end deprecation ============== namespace { int append_blocks(std::vector& dst, std::vector& src , int num_blocks) { if (src.empty()) return num_blocks; int to_copy; // if (prefer_whole_pieces == 0) to_copy = (std::min)(int(src.size()), num_blocks); // else // to_copy = int(src.size()); dst.insert(dst.end() , src.begin(), src.begin() + to_copy); src.clear(); return num_blocks - to_copy; } } // pieces describes which pieces the peer we're requesting from // has. // interesting_blocks is an out parameter, and will be filled // with (up to) num_blocks of interesting blocks that the peer has. // prefer_whole_pieces can be set if this peer should download // whole pieces rather than trying to download blocks from the // same piece as other peers. // the void* is the pointer to the policy::peer of the peer we're // picking pieces from. This is used when downloading whole pieces, // to only pick from the same piece the same peer is downloading // from. state is supposed to be set to fast if the peer is downloading // relatively fast, by some notion. Slow peers will prefer not // to pick blocks from the same pieces as fast peers, and vice // versa. Downloading pieces are marked as being fast, medium // or slow once they're started. // options are: // * rarest_first // pick the rarest pieces first // * reverse // reverse the piece picking. Pick the most common // pieces first or the last pieces (if picking sequential) // * sequential // download pieces in-order // * on_parole // the peer is on parole, only pick whole pieces which // has only been downloaded and requested from the same // peer // * prioritize_partials // pick blocks from downloading pieces first // * speed_affinity // have an affinity to pick pieces in the same speed // category. // * ignore_whole_pieces // ignores the prefer_whole_pieces parameter (as if // it was 0) // only one of rarest_first, sequential can be set void piece_picker::pick_pieces(bitfield const& pieces , std::vector& interesting_blocks, int num_blocks , int prefer_whole_pieces, void* peer, piece_state_t speed , int options, std::vector const& suggested_pieces , int num_peers) const { TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); // prevent the number of partial pieces to grow indefinitely // make this scale by the number of peers we have. For large // scale clients, we would have more peers, and allow a higher // threshold for the number of partials // TODO: 2 m_downloads size will be > 0 just by having pad-files // in the torrent. That should be taken into account here. if (m_downloads.size() > num_peers * 3 / 2) options |= prioritize_partials; if (options & ignore_whole_pieces) prefer_whole_pieces = 0; // only one of rarest_first and sequential can be set. TORRENT_ASSERT(((options & rarest_first) ? 1 : 0) + ((options & sequential) ? 1 : 0) <= 1); #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(num_blocks > 0); TORRENT_ASSERT(pieces.size() == m_piece_map.size()); TORRENT_ASSERT(!m_priority_boundries.empty() || m_dirty); // this will be filled with blocks that we should not request // unless we can't find num_blocks among the other ones. // blocks that belong to pieces with a mismatching speed // category for instance, or if we prefer whole pieces, // blocks belonging to a piece that others have // downloaded to std::vector backup_blocks; std::vector backup_blocks2; const std::vector empty_vector; // When prefer_whole_pieces is set (usually set when downloading from // fast peers) the partial pieces will not be prioritized, but actually // ignored as long as possible. All blocks found in downloading // pieces are regarded as backup blocks if (options & prioritize_partials) { for (std::vector::const_iterator i = m_downloads.begin() , end(m_downloads.end()); i != end; ++i) { if (!is_piece_free(i->index, pieces)) continue; if (m_piece_map[i->index].full && backup_blocks.size() >= num_blocks && backup_blocks2.size() >= num_blocks) continue; num_blocks = add_blocks_downloading(*i, pieces , interesting_blocks, backup_blocks, backup_blocks2 , num_blocks, prefer_whole_pieces, peer, speed, options); if (num_blocks <= 0) return; } num_blocks = append_blocks(interesting_blocks, backup_blocks , num_blocks); if (num_blocks <= 0) return; num_blocks = append_blocks(interesting_blocks, backup_blocks2 , num_blocks); if (num_blocks <= 0) return; } if (!suggested_pieces.empty()) { for (std::vector::const_iterator i = suggested_pieces.begin(); i != suggested_pieces.end(); ++i) { if (!is_piece_free(*i, pieces)) continue; num_blocks = add_blocks(*i, pieces , interesting_blocks, backup_blocks , backup_blocks2, num_blocks , prefer_whole_pieces, peer, empty_vector , speed, options); if (num_blocks <= 0) return; } } if (options & sequential) { if (options & reverse) { for (int i = m_reverse_cursor - 1; i >= m_cursor; --i) { if (!is_piece_free(i, pieces)) continue; num_blocks = add_blocks(i, pieces , interesting_blocks, backup_blocks , backup_blocks2, num_blocks , prefer_whole_pieces, peer, suggested_pieces , speed, options); if (num_blocks <= 0) return; } } else { for (int i = m_cursor; i < m_reverse_cursor; ++i) { if (!is_piece_free(i, pieces)) continue; num_blocks = add_blocks(i, pieces , interesting_blocks, backup_blocks , backup_blocks2, num_blocks , prefer_whole_pieces, peer, suggested_pieces , speed, options); if (num_blocks <= 0) return; } } } else if (options & rarest_first) { if (m_dirty) update_pieces(); TORRENT_ASSERT(!m_dirty); if (options & reverse) { // it's a bit complicated in order to always prioritize // partial pieces, and respect priorities. Every chunk // of 4 priority levels are traversed in forward order, but otherwise // they are traversed in reverse order // round up to an even 4 priority boundry, to make it simpler // to do the akward reverse traversing #define div_round_up(n, d) (((n) + (d) - 1) / (d)) m_priority_boundries.resize(div_round_up(m_priority_boundries.size() , prio_factor) * prio_factor, m_priority_boundries.back()); for (int i = m_priority_boundries.size() - 1; i >= 0; --i) { int prio = (i / prio_factor) * prio_factor + prio_factor - 1 - (i % prio_factor); TORRENT_ASSERT(prio >= 0); TORRENT_ASSERT(prio < int(m_priority_boundries.size())); int start = prio == 0 ? 0 : m_priority_boundries[prio - 1]; for (int p = start; p < m_priority_boundries[prio]; ++p) { if (!is_piece_free(m_pieces[p], pieces)) continue; num_blocks = add_blocks(m_pieces[p], pieces , interesting_blocks, backup_blocks , backup_blocks2, num_blocks , prefer_whole_pieces, peer, suggested_pieces , speed, options); if (num_blocks <= 0) return; } } #undef div_round_up } else { for (std::vector::const_iterator i = m_pieces.begin(); i != m_pieces.end(); ++i) { if (!is_piece_free(*i, pieces)) continue; num_blocks = add_blocks(*i, pieces , interesting_blocks, backup_blocks , backup_blocks2, num_blocks , prefer_whole_pieces, peer, suggested_pieces , speed, options); if (num_blocks <= 0) return; } } } else { // we're not using rarest first (only for the first // bucket, since that's where the currently downloading // pieces are) int start_piece = random() % m_piece_map.size(); int piece = start_piece; while (num_blocks > 0) { bool done = false; // skip pieces we can't pick, and suggested pieces // since we've already picked those while (!can_pick(piece, pieces) || std::find(suggested_pieces.begin() , suggested_pieces.end(), piece) != suggested_pieces.end()) { ++piece; if (piece == int(m_piece_map.size())) piece = 0; // could not find any more pieces if (piece == start_piece) { done = true; break; } } if (done) break; TORRENT_ASSERT(can_pick(piece, pieces)); TORRENT_ASSERT(m_piece_map[piece].downloading == false); int start, end; boost::tie(start, end) = expand_piece(piece, prefer_whole_pieces, pieces); for (int k = start; k < end; ++k) { TORRENT_ASSERT(m_piece_map[k].downloading == false); TORRENT_ASSERT(m_piece_map[k].priority(this) >= 0); int num_blocks_in_piece = blocks_in_piece(k); if (prefer_whole_pieces == 0 && num_blocks_in_piece > num_blocks) num_blocks_in_piece = num_blocks; for (int j = 0; j < num_blocks_in_piece; ++j) { TORRENT_ASSERT(is_piece_free(k, pieces)); interesting_blocks.push_back(piece_block(k, j)); --num_blocks; } } piece = end; if (piece == int(m_piece_map.size())) piece = 0; // could not find any more pieces if (piece == start_piece) break; } } if (num_blocks <= 0) return; // we might have to re-pick some backup blocks // from full pieces, since we skipped those the // first pass over for (std::vector::const_iterator i = m_downloads.begin() , end(m_downloads.end()); i != end; ++i) { if (!pieces[i->index]) continue; // we've already considered the non-full pieces if (!m_piece_map[i->index].full) continue; std::vector temp; add_blocks_downloading(*i, pieces , temp, backup_blocks, backup_blocks2 , num_blocks, prefer_whole_pieces, peer, speed, options); } #ifdef TORRENT_DEBUG verify_pick(interesting_blocks, pieces); verify_pick(backup_blocks, pieces); verify_pick(backup_blocks2, pieces); #endif std::vector temp; for (std::vector::const_iterator i = m_downloads.begin() , end(m_downloads.end()); i != end; ++i) { if (!pieces[i->index]) continue; if (piece_priority(i->index) == 0) continue; int num_blocks_in_piece = blocks_in_piece(i->index); // fill in with blocks requested from other peers // as backups bool done = false; for (int j = 0; j < num_blocks_in_piece; ++j) { block_info const& info = i->info[j]; TORRENT_ASSERT(info.peer == 0 || static_cast(info.peer)->in_use); TORRENT_ASSERT(info.piece_index == i->index); if (info.state != block_info::state_requested || info.peer == peer) continue; temp.push_back(piece_block(i->index, j)); done = true; } if (done) break; } num_blocks = append_blocks(interesting_blocks, backup_blocks , num_blocks); if (num_blocks <= 0) return; num_blocks = append_blocks(interesting_blocks, backup_blocks2, num_blocks); if (num_blocks <= 0) return; // don't double-pick anything if the peer is on parole if (options & on_parole) return; // pick one random block from the first busy piece we encountered // none of these blocks have more than one request to them if (!temp.empty()) interesting_blocks.push_back(temp[random() % temp.size()]); #ifdef TORRENT_DEBUG // make sure that we at this point have added requests to all unrequested blocks // in all downloading pieces for (std::vector::const_iterator i = m_downloads.begin() , end(m_downloads.end()); i != end; ++i) { if (!pieces[i->index]) continue; if (piece_priority(i->index) == 0) continue; int num_blocks_in_piece = blocks_in_piece(i->index); for (int j = 0; j < num_blocks_in_piece; ++j) { block_info const& info = i->info[j]; TORRENT_ASSERT(info.piece_index == i->index); if (info.state != block_info::state_none) continue; std::vector::iterator k = std::find( interesting_blocks.begin(), interesting_blocks.end() , piece_block(i->index, j)); if (k != interesting_blocks.end()) continue; fprintf(stderr, "interesting blocks:\n"); for (k = interesting_blocks.begin(); k != interesting_blocks.end(); ++k) fprintf(stderr, "(%d, %d)", k->piece_index, k->block_index); fprintf(stderr, "\nnum_blocks: %d\n", num_blocks); for (std::vector::const_iterator l = m_downloads.begin() , end(m_downloads.end()); l != end; ++l) { fprintf(stderr, "%d : ", l->index); int num_blocks_in_piece = blocks_in_piece(l->index); for (int m = 0; m < num_blocks_in_piece; ++m) fprintf(stderr, "%d", l->info[m].state); fprintf(stderr, "\n"); } TORRENT_ASSERT(false); } } if (interesting_blocks.empty()) { // print_pieces(); for (int i = 0; i < num_pieces(); ++i) { if (!pieces[i]) continue; if (m_piece_map[i].priority(this) <= 0) continue; if (have_piece(i)) continue; std::vector::const_iterator k = find_dl_piece(i); TORRENT_ASSERT(k != m_downloads.end()); if (k == m_downloads.end()) continue; // this assert is not valid for web_seeds /* int num_blocks_in_piece = blocks_in_piece(k->index); for (int j = 0; j < num_blocks_in_piece; ++j) { block_info const& info = k->info[j]; TORRENT_ASSERT(info.piece_index == k->index); if (info.state == block_info::state_finished) continue; TORRENT_ASSERT(info.peer != 0); } */ } } #endif } int piece_picker::blocks_in_piece(int index) const { TORRENT_ASSERT(index >= 0); TORRENT_ASSERT(index < (int)m_piece_map.size() || m_piece_map.empty()); /* [MF] if (index+1 == (int)m_piece_map.size()) return m_blocks_in_last_piece; else */ return m_blocks_per_piece; } bool piece_picker::is_piece_free(int piece, bitfield const& bitmask) const { TORRENT_ASSERT(piece >= 0 && piece < int(m_piece_map.size())); return bitmask[piece] && !m_piece_map[piece].have() && !m_piece_map[piece].filtered(); } bool piece_picker::can_pick(int piece, bitfield const& bitmask) const { TORRENT_ASSERT(piece >= 0 && piece < int(m_piece_map.size())); return bitmask[piece] && !m_piece_map[piece].have() && !m_piece_map[piece].downloading && !m_piece_map[piece].filtered(); } void piece_picker::clear_peer(void* peer) { for (std::vector::iterator i = m_block_info.begin() , end(m_block_info.end()); i != end; ++i) { TORRENT_ASSERT(i->peer == 0 || static_cast(i->peer)->in_use); if (i->peer == peer) i->peer = 0; } } namespace { // the first bool is true if this is the only peer that has requested and downloaded // blocks from this piece. // the second bool is true if this is the only active peer that is requesting // and downloading blocks from this piece. Active means having a connection. boost::tuple requested_from(piece_picker::downloading_piece const& p , int num_blocks_in_piece, void* peer) { bool exclusive = true; bool exclusive_active = true; for (int j = 0; j < num_blocks_in_piece; ++j) { piece_picker::block_info const& info = p.info[j]; TORRENT_ASSERT(info.peer == 0 || static_cast(info.peer)->in_use); TORRENT_ASSERT(info.piece_index == p.index); if (info.state != piece_picker::block_info::state_none && info.peer != peer) { exclusive = false; if (info.state == piece_picker::block_info::state_requested && info.peer != 0) { exclusive_active = false; return boost::make_tuple(exclusive, exclusive_active); } } } return boost::make_tuple(exclusive, exclusive_active); } } int piece_picker::add_blocks(int piece , bitfield const& pieces , std::vector& interesting_blocks , std::vector& backup_blocks , std::vector& backup_blocks2 , int num_blocks, int prefer_whole_pieces , void* peer, std::vector const& ignore , piece_state_t speed, int options) const { TORRENT_ASSERT(piece >= 0); TORRENT_ASSERT(piece < (int)m_piece_map.size()); TORRENT_ASSERT(is_piece_free(piece, pieces)); // std::cout << "add_blocks(" << piece << ")" << std::endl; // std::cout << " num_blocks " << num_blocks << std::endl; // ignore pieces found in the ignore list if (std::find(ignore.begin(), ignore.end(), piece) != ignore.end()) return num_blocks; TORRENT_ASSERT(m_piece_map[piece].priority(this) >= 0); if (m_piece_map[piece].downloading) { if (m_piece_map[piece].full) return num_blocks; // if we're prioritizing partials, we've already // looked through the downloading pieces if (options & prioritize_partials) return num_blocks; std::vector::const_iterator i = find_dl_piece(piece); TORRENT_ASSERT(i != m_downloads.end()); // std::cout << "add_blocks_downloading(" << piece << ")" << std::endl; return add_blocks_downloading(*i, pieces , interesting_blocks, backup_blocks, backup_blocks2 , num_blocks, prefer_whole_pieces, peer, speed, options); } int num_blocks_in_piece = blocks_in_piece(piece); // pick a new piece if (prefer_whole_pieces == 0) { if (num_blocks_in_piece > num_blocks) num_blocks_in_piece = num_blocks; TORRENT_ASSERT(is_piece_free(piece, pieces)); for (int j = 0; j < num_blocks_in_piece; ++j) interesting_blocks.push_back(piece_block(piece, j)); num_blocks -= num_blocks_in_piece; } else { int start, end; boost::tie(start, end) = expand_piece(piece, prefer_whole_pieces, pieces); for (int k = start; k < end; ++k) { TORRENT_ASSERT(m_piece_map[k].priority(this) > 0); num_blocks_in_piece = blocks_in_piece(k); TORRENT_ASSERT(is_piece_free(k, pieces)); for (int j = 0; j < num_blocks_in_piece; ++j) { interesting_blocks.push_back(piece_block(k, j)); --num_blocks; } } } #ifdef TORRENT_DEBUG verify_pick(interesting_blocks, pieces); #endif if (num_blocks <= 0) return 0; return num_blocks; } int piece_picker::add_blocks_downloading(downloading_piece const& dp , bitfield const& pieces , std::vector& interesting_blocks , std::vector& backup_blocks , std::vector& backup_blocks2 , int num_blocks, int prefer_whole_pieces , void* peer, piece_state_t speed, int options) const { if (!pieces[dp.index]) return num_blocks; if (m_piece_map[dp.index].filtered()) return num_blocks; int num_blocks_in_piece = blocks_in_piece(dp.index); // if all blocks have been requested (and we don't need any backup // blocks), we might as well return immediately /* if (int(backup_blocks2.size()) >= num_blocks && int(backup_blocks.size()) >= num_blocks && dp.requested + dp.writing + dp.finished == num_blocks_in_piece) return num_blocks; */ // is true if all the other pieces that are currently // requested from this piece are from the same // peer as 'peer'. bool exclusive; bool exclusive_active; boost::tie(exclusive, exclusive_active) = requested_from(dp, num_blocks_in_piece, peer); // peers on parole are only allowed to pick blocks from // pieces that only they have downloaded/requested from if ((options & on_parole) && !exclusive) return num_blocks; // we prefer whole blocks, but there are other peers // downloading from this piece, add it as backups if (prefer_whole_pieces > 0 && !exclusive_active) { if (int(backup_blocks2.size()) >= num_blocks) return num_blocks; for (int j = 0; j < num_blocks_in_piece; ++j) { // ignore completed blocks and already requested blocks block_info const& info = dp.info[j]; TORRENT_ASSERT(info.piece_index == dp.index); if (info.state != block_info::state_none) continue; backup_blocks2.push_back(piece_block(dp.index, j)); } return num_blocks; } for (int j = 0; j < num_blocks_in_piece; ++j) { // ignore completed blocks and already requested blocks block_info const& info = dp.info[j]; TORRENT_ASSERT(info.piece_index == dp.index); if (info.state != block_info::state_none) continue; // if the piece is fast and the peer is slow, or vice versa, // add the block as a backup. // override this behavior if all the other blocks // have been requested from the same peer or // if the state of the piece is none (the // piece will in that case change state). if (dp.state != none && dp.state != speed && !exclusive_active && (options & speed_affinity)) { if (abs(dp.state - speed) == 1) { // don't pick too many back-up blocks if (int(backup_blocks.size()) >= num_blocks) return num_blocks; backup_blocks.push_back(piece_block(dp.index, j)); } else { // don't pick too many back-up blocks if (int(backup_blocks2.size()) >= num_blocks) return num_blocks; backup_blocks2.push_back(piece_block(dp.index, j)); } continue; } // this block is interesting (we don't have it // yet). interesting_blocks.push_back(piece_block(dp.index, j)); // we have found a block that's free to download num_blocks--; // if we prefer whole pieces, continue picking from this // piece even though we have num_blocks if (prefer_whole_pieces > 0) continue; TORRENT_ASSERT(num_blocks >= 0); if (num_blocks <= 0) return num_blocks; } TORRENT_ASSERT(num_blocks >= 0 || prefer_whole_pieces > 0); if (num_blocks <= 0) return 0; if (options & on_parole) return num_blocks; if (int(backup_blocks.size()) >= num_blocks) return num_blocks; #ifdef TORRENT_DEBUG verify_pick(backup_blocks, pieces); #endif return num_blocks; } std::pair piece_picker::expand_piece(int piece, int whole_pieces , bitfield const& have) const { if (whole_pieces == 0) return std::make_pair(piece, piece + 1); int start = piece - 1; int lower_limit = piece - whole_pieces; if (lower_limit < -1) lower_limit = -1; while (start > lower_limit && can_pick(start, have)) --start; ++start; TORRENT_ASSERT(start >= 0); int end = piece + 1; int upper_limit = start + whole_pieces; if (upper_limit > int(m_piece_map.size())) upper_limit = int(m_piece_map.size()); while (end < upper_limit && can_pick(end, have)) ++end; return std::make_pair(start, end); } bool piece_picker::is_piece_finished(int index) const { TORRENT_ASSERT(index < (int)m_piece_map.size()); TORRENT_ASSERT(index >= 0); if (m_piece_map[index].downloading == 0) { TORRENT_ASSERT(find_dl_piece(index) == m_downloads.end()); return false; } std::vector::const_iterator i = find_dl_piece(index); TORRENT_ASSERT(i != m_downloads.end()); TORRENT_ASSERT((int)i->finished <= m_blocks_per_piece); int max_blocks = blocks_in_piece(index); if (int(i->finished) + int(i->writing) < max_blocks) return false; TORRENT_ASSERT(int(i->finished) + int(i->writing) == max_blocks); #ifdef TORRENT_DEBUG for (int k = 0; k < max_blocks; ++k) { TORRENT_ASSERT(i->info[k].piece_index == index); TORRENT_ASSERT(i->info[k].state == block_info::state_finished || i->info[k].state == block_info::state_writing); } #endif return true; } std::vector::iterator piece_picker::find_dl_piece(int index) { // return std::find_if(m_downloads.begin(), m_downloads.end(), has_index(index)); downloading_piece cmp; cmp.index = index; std::vector::iterator i = std::lower_bound( m_downloads.begin(), m_downloads.end(), cmp); if (i == m_downloads.end()) return i; if (i->index == index) return i; return m_downloads.end(); } std::vector::const_iterator piece_picker::find_dl_piece(int index) const { // return std::find_if(m_downloads.begin(), m_downloads.end(), has_index(index)); downloading_piece cmp; cmp.index = index; std::vector::const_iterator i = std::lower_bound( m_downloads.begin(), m_downloads.end(), cmp); if (i == m_downloads.end()) return i; if (i->index == index) return i; return m_downloads.end(); } void piece_picker::update_full(downloading_piece& dp) { int num_blocks = blocks_in_piece(dp.index); m_piece_map[dp.index].full = dp.requested + dp.finished + dp.writing == num_blocks; } bool piece_picker::is_requested(piece_block block) const { TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); if (m_piece_map[block.piece_index].downloading == 0) return false; std::vector::const_iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); TORRENT_ASSERT(i->info[block.block_index].piece_index == block.piece_index); return i->info[block.block_index].state == block_info::state_requested; } bool piece_picker::is_downloaded(piece_block block) const { TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); if (m_piece_map[block.piece_index].index == piece_pos::we_have_index) return true; if (m_piece_map[block.piece_index].downloading == 0) return false; std::vector::const_iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); TORRENT_ASSERT(i->info[block.block_index].piece_index == block.piece_index); return i->info[block.block_index].state == block_info::state_finished || i->info[block.block_index].state == block_info::state_writing; } bool piece_picker::is_finished(piece_block block) const { TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); if (m_piece_map[block.piece_index].index == piece_pos::we_have_index) return true; if (m_piece_map[block.piece_index].downloading == 0) return false; std::vector::const_iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); TORRENT_ASSERT(i->info[block.block_index].piece_index == block.piece_index); return i->info[block.block_index].state == block_info::state_finished; } bool piece_picker::mark_as_downloading(piece_block block , void* peer, piece_state_t state) { TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); TORRENT_ASSERT(state != piece_picker::none); TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); TORRENT_ASSERT(!m_piece_map[block.piece_index].have()); piece_pos& p = m_piece_map[block.piece_index]; if (p.downloading == 0) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif int prio = p.priority(this); TORRENT_ASSERT(prio < int(m_priority_boundries.size()) || m_dirty); p.downloading = 1; if (prio >= 0 && !m_dirty) update(prio, p.index); downloading_piece& dp = add_download_piece(block.piece_index); dp.state = state; block_info& info = dp.info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); info.state = block_info::state_requested; info.peer = peer; info.num_peers = 1; ++dp.requested; update_full(dp); } else { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif std::vector::iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); block_info& info = i->info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); if (info.state == block_info::state_writing || info.state == block_info::state_finished) return false; TORRENT_ASSERT(info.state == block_info::state_none || (info.state == block_info::state_requested && (info.num_peers > 0))); info.peer = peer; if (info.state != block_info::state_requested) { info.state = block_info::state_requested; ++i->requested; update_full(*i); } ++info.num_peers; if (i->state == none) i->state = state; } return true; } int piece_picker::num_peers(piece_block block) const { TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); piece_pos const& p = m_piece_map[block.piece_index]; if (!p.downloading) return 0; std::vector::const_iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); block_info const& info = i->info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); return info.num_peers; } void piece_picker::get_availability(std::vector& avail) const { TORRENT_ASSERT(m_seeds >= 0); TORRENT_PIECE_PICKER_INVARIANT_CHECK; avail.resize(m_piece_map.size()); std::vector::iterator j = avail.begin(); for (std::vector::const_iterator i = m_piece_map.begin() , end(m_piece_map.end()); i != end; ++i, ++j) *j = i->peer_count + m_seeds; } bool piece_picker::mark_as_writing(piece_block block, void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); piece_pos& p = m_piece_map[block.piece_index]; if (p.downloading == 0) { // if we already have this piece, just ignore this if (have_piece(block.piece_index)) return false; #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif int prio = p.priority(this); TORRENT_ASSERT(prio < int(m_priority_boundries.size()) || m_dirty); p.downloading = 1; // prio being -1 can happen if a block is requested before // the piece priority was set to 0 if (prio >= 0 && !m_dirty) update(prio, p.index); downloading_piece& dp = add_download_piece(block.piece_index); dp.state = none; block_info& info = dp.info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); info.state = block_info::state_writing; info.peer = peer; info.num_peers = 0; dp.writing = 1; update_full(dp); // sort_piece(m_downloads.end()-1); } else { std::vector::iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); block_info& info = i->info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); info.peer = peer; if (info.state == block_info::state_requested) --i->requested; TORRENT_ASSERT(i->requested >= 0); if (info.state == block_info::state_writing || info.state == block_info::state_finished) return false; ++i->writing; info.state = block_info::state_writing; TORRENT_ASSERT(info.piece_index == block.piece_index); // all other requests for this block should have been // cancelled now info.num_peers = 0; if (i->requested == 0) { // there are no blocks requested in this piece. // remove the fast/slow state from it i->state = none; } // sort_piece(i); } return true; } void piece_picker::write_failed(piece_block block) { TORRENT_PIECE_PICKER_INVARIANT_CHECK; std::vector::iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); if (i == m_downloads.end()) return; block_info& info = i->info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); TORRENT_ASSERT(info.state == block_info::state_writing); TORRENT_ASSERT(info.num_peers == 0); TORRENT_ASSERT(i->writing > 0); TORRENT_ASSERT(info.state == block_info::state_writing); if (info.state == block_info::state_finished) return; if (info.state == block_info::state_writing) --i->writing; info.peer = 0; info.state = block_info::state_none; update_full(*i); if (i->finished + i->writing + i->requested == 0) { piece_pos& p = m_piece_map[block.piece_index]; int prev_priority = p.priority(this); erase_download_piece(i); int new_priority = p.priority(this); if (m_dirty) return; if (new_priority == prev_priority) return; if (prev_priority == -1) add(block.piece_index); else update(prev_priority, p.index); } else { // sort_piece(i); } } void piece_picker::mark_as_finished(piece_block block, void* peer) { TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); TORRENT_ASSERT(block.piece_index >= 0); TORRENT_ASSERT(block.block_index >= 0); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); piece_pos& p = m_piece_map[block.piece_index]; if (p.downloading == 0) { // if we already have this piece, just ignore this if (have_piece(block.piece_index)) return; #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(peer == 0); int prio = p.priority(this); TORRENT_ASSERT(prio < int(m_priority_boundries.size()) || m_dirty); p.downloading = 1; if (prio >= 0 && !m_dirty) update(prio, p.index); downloading_piece& dp = add_download_piece(block.piece_index); dp.state = none; block_info& info = dp.info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); info.peer = peer; TORRENT_ASSERT(info.state == block_info::state_none); TORRENT_ASSERT(info.num_peers == 0); if (info.state != block_info::state_finished) { ++dp.finished; // sort_piece(m_downloads.end() - 1); } info.state = block_info::state_finished; } else { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif std::vector::iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); block_info& info = i->info[block.block_index]; TORRENT_ASSERT(info.piece_index == block.piece_index); if (info.state == block_info::state_finished) return; TORRENT_ASSERT(info.num_peers == 0); info.peer = peer; TORRENT_ASSERT(info.state == block_info::state_writing || peer == 0); TORRENT_ASSERT(i->writing >= 0); ++i->finished; if (info.state == block_info::state_writing) { --i->writing; info.state = block_info::state_finished; } else { TORRENT_ASSERT(info.state == block_info::state_none); info.state = block_info::state_finished; // sort_piece(i); } } } void piece_picker::get_downloaders(std::vector& d, int index) const { TORRENT_ASSERT(index >= 0 && index <= (int)m_piece_map.size()); std::vector::const_iterator i = find_dl_piece(index); TORRENT_ASSERT(i != m_downloads.end()); d.clear(); for (int j = 0, end(blocks_in_piece(index)); j != end; ++j) { TORRENT_ASSERT(i->info[j].peer == 0 || static_cast(i->info[j].peer)->in_use); d.push_back(i->info[j].peer); } } void* piece_picker::get_downloader(piece_block block) const { std::vector::const_iterator i = find_dl_piece(block.piece_index); if (i == m_downloads.end()) return 0; TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(i->info[block.block_index].piece_index == block.piece_index); if (i->info[block.block_index].state == block_info::state_none) return 0; void* peer = i->info[block.block_index].peer; TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); return peer; } // this is called when a request is rejected or when // a peer disconnects. The piece might be in any state void piece_picker::abort_download(piece_block block, void* peer) { #ifdef TORRENT_EXPENSIVE_INVARIANT_CHECKS TORRENT_PIECE_PICKER_INVARIANT_CHECK; #endif TORRENT_ASSERT(peer == 0 || static_cast(peer)->in_use); TORRENT_ASSERT(block.block_index != piece_block::invalid.block_index); TORRENT_ASSERT(block.piece_index != piece_block::invalid.piece_index); TORRENT_ASSERT(block.piece_index < m_piece_map.size()); TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); if (m_piece_map[block.piece_index].downloading == 0) { TORRENT_ASSERT(find_dl_piece(block.piece_index) == m_downloads.end()); return; } std::vector::iterator i = find_dl_piece(block.piece_index); TORRENT_ASSERT(i != m_downloads.end()); block_info& info = i->info[block.block_index]; TORRENT_ASSERT(info.peer == 0 || static_cast(info.peer)->in_use); TORRENT_ASSERT(info.piece_index == block.piece_index); TORRENT_ASSERT(info.state != block_info::state_none); if (info.state == block_info::state_finished || info.state == block_info::state_none || info.state == block_info::state_writing) return; if (info.state == block_info::state_requested) { TORRENT_ASSERT(info.num_peers > 0); if (info.num_peers > 0) --info.num_peers; if (info.peer == peer) info.peer = 0; TORRENT_ASSERT(int(block.block_index) < blocks_in_piece(block.piece_index)); // if there are other peers, leave the block requested if (info.num_peers > 0) return; // clear the downloader of this block info.peer = 0; // clear this block as being downloaded info.state = block_info::state_none; --i->requested; update_full(*i); } // if there are no other blocks in this piece // that's being downloaded, remove it from the list if (i->requested + i->finished + i->writing == 0) { piece_pos& p = m_piece_map[block.piece_index]; int prev_prio = p.priority(this); TORRENT_ASSERT(prev_prio < int(m_priority_boundries.size()) || m_dirty); erase_download_piece(i); if (!m_dirty) { int prio = p.priority(this); if (prev_prio == -1 && prio >= 0) add(block.piece_index); else if (prev_prio >= 0) update(prev_prio, p.index); } TORRENT_ASSERT(find_dl_piece(block.piece_index) == m_downloads.end()); } else if (i->requested == 0) { // there are no blocks requested in this piece. // remove the fast/slow state from it i->state = none; } } int piece_picker::unverified_blocks() const { int counter = 0; for (std::vector::const_iterator i = m_downloads.begin(); i != m_downloads.end(); ++i) { counter += (int)i->finished; } return counter; } }