honey_version.cc

Go to the documentation of this file.

 
/* Copyright (C) 2006,2007,2008,2009,2010,2013,2014,2015,2016,2017,2018 Olly Betts
 * Copyright (C) 2011 Dan Colish
 *
 * 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
 * <https://www.gnu.org/licenses/>.
 */
 
#include <config.h>
 
#include "honey_version.h"
 
#include "debuglog.h"
#include "fd.h"
#include "honey_defs.h"
#include "io_utils.h"
#include "min_non_zero.h"
#include "omassert.h"
#include "pack.h"
#include "posixy_wrapper.h"
#include "stringutils.h" // For STRINGIZE() and CONST_STRLEN().
 
#include <cerrno>
#include <cstring> // For memcmp().
#include <string>
#include <sys/types.h>
#include "safesysstat.h"
#include "safefcntl.h"
#include "safeunistd.h"
#include "str.h"
#include "stringutils.h"
 
#include "backends/uuids.h"
 
#include "xapian/constants.h"
#include "xapian/error.h"
 
using namespace std;
 
#define HONEY_FORMAT_VERSION DATE_TO_VERSION(2018,4,3)
// 2018,4,3   2.0.0 outlaw mixed-wdf terms
// 2018,3,28        don't special case first entry in SSTable
// 2018,3,27        new key format for value stats, value chunks, doclen chunks
// 2018,3,26        use known suffix from spelling B and T keys
// 2018,3,25        use known prefix from spelling B and H keys
// 2018,3,15        avoid storing flat wdf
// 2018,3,14        store per term wdf_max
// 2018,3,12        binary chop index
// 2018,3,11        spelling key encoding changed
// 2018,2,22        index valuestream chunks by last docid in chunk
// 2018,2,21        index doclen chunks by last docid in chunk
// 2018,2,20        implement array index
// 2018,2,19        allow 1,2,3 as well as 4 byte doc length width
// 2018,2,2         special case tf=2; first_wdf = floor(collfreq/2)
// 2018,2,1         pack_uint for postlist data
// 2018,1,31        Special case postlist when termfreq==2
// 2018,1,30        More compact postlist chunk headers
// 2018,1,23        Elide last-first for single occurrence terms
// 2018,1,4         Merge values used and terms used
// 2018,1,3         Table start offset in RootInfo
// 2017,12,30       Value stats key changes
// 2017,12,29       User metadata key changes
// 2017,12,5        New Honey backend
 
#define DATE_TO_VERSION(Y,M,D) \
        ((unsigned(Y) - 2014) << 9 | unsigned(M) << 5 | unsigned(D))
#define VERSION_TO_YEAR(V) ((unsigned(V) >> 9) + 2014)
#define VERSION_TO_MONTH(V) ((unsigned(V) >> 5) & 0x0f)
#define VERSION_TO_DAY(V) (unsigned(V) & 0x1f)
 
#define HONEY_VERSION_MAGIC_LEN 14
#define HONEY_VERSION_MAGIC_AND_VERSION_LEN 16
 
static const char HONEY_VERSION_MAGIC[HONEY_VERSION_MAGIC_AND_VERSION_LEN] = {
    '\x0f', '\x0d', 'X', 'a', 'p', 'i', 'a', 'n', ' ', 'H', 'o', 'n', 'e', 'y',
    char((HONEY_FORMAT_VERSION >> 8) & 0xff), char(HONEY_FORMAT_VERSION & 0xff)
};
 
HoneyVersion::HoneyVersion(int fd_)
    : fd(fd_), db_dir()
{
    offset = lseek(fd, 0, SEEK_CUR);
    if (rare(offset < 0)) {
        string msg = "lseek failed on file descriptor ";
        msg += str(fd);
        throw Xapian::DatabaseOpeningError(msg, errno);
    }
}
 
HoneyVersion::~HoneyVersion()
{
    // Either this is a single-file database, or this fd is from opening a new
    // version file in write(), but sync() was never called.
    if (fd != -1)
        (void)::close(fd);
}
 
void
HoneyVersion::read()
{
    LOGCALL_VOID(DB, "HoneyVersion::read", NO_ARGS);
    FD close_fd(-1);
    int fd_in;
    if (single_file()) {
        if (rare(lseek(fd, offset, SEEK_SET) < 0)) {
            string msg = "Failed to rewind file descriptor ";
            msg += str(fd);
            throw Xapian::DatabaseOpeningError(msg, errno);
        }
        fd_in = fd;
    } else {
        string filename = db_dir;
        filename += "/iamhoney";
        fd_in = posixy_open(filename.c_str(), O_RDONLY|O_BINARY);
        if (rare(fd_in < 0)) {
            string msg = filename;
            msg += ": Failed to open honey revision file for reading";
            if (errno == ENOENT || errno == ENOTDIR) {
                throw Xapian::DatabaseNotFoundError(msg, errno);
            }
            throw Xapian::DatabaseOpeningError(msg, errno);
        }
        close_fd = fd_in;
    }
 
    char buf[256];
 
    const char* p = buf;
    const char* end = p + io_read(fd_in, buf, sizeof(buf), 33);
 
    if (memcmp(buf, HONEY_VERSION_MAGIC, HONEY_VERSION_MAGIC_LEN) != 0)
        throw Xapian::DatabaseCorruptError("Rev file magic incorrect");
 
    unsigned version;
    version = static_cast<unsigned char>(buf[HONEY_VERSION_MAGIC_LEN]);
    version <<= 8;
    version |= static_cast<unsigned char>(buf[HONEY_VERSION_MAGIC_LEN + 1]);
    if (version != HONEY_FORMAT_VERSION) {
        string msg;
        if (!single_file()) {
            msg = db_dir;
            msg += ": ";
        }
        msg += "Database is format version ";
        msg += str(VERSION_TO_YEAR(version) * 10000 +
                   VERSION_TO_MONTH(version) * 100 +
                   VERSION_TO_DAY(version));
        msg += " but I only understand ";
        msg += str(VERSION_TO_YEAR(HONEY_FORMAT_VERSION) * 10000 +
                   VERSION_TO_MONTH(HONEY_FORMAT_VERSION) * 100 +
                   VERSION_TO_DAY(HONEY_FORMAT_VERSION));
        throw Xapian::DatabaseVersionError(msg);
    }
 
    p += HONEY_VERSION_MAGIC_AND_VERSION_LEN;
    uuid.assign(p);
    p += uuid.BINARY_SIZE;
 
    if (!unpack_uint(&p, end, &rev)) {
        throw Xapian::DatabaseCorruptError("Rev file failed to decode "
                                           "revision");
    }
 
    for (unsigned table_no = 0; table_no < Honey::MAX_; ++table_no) {
        if (!root[table_no].unserialise(&p, end)) {
            throw Xapian::DatabaseCorruptError("Rev file root_info missing");
        }
        old_root[table_no] = root[table_no];
    }
 
    // For a single-file database, this will assign extra data.  We read
    // sizeof(buf) above, then skip HONEY_VERSION_MAGIC_AND_VERSION_LEN,
    // then 16, then the size of the serialised root info.
    serialised_stats.assign(p, end);
    unserialise_stats();
}
 
void
HoneyVersion::serialise_stats()
{
    serialised_stats.resize(0);
    pack_uint(serialised_stats, doccount);
    // last_docid must always be >= doccount.
    pack_uint(serialised_stats, last_docid - doccount);
    pack_uint(serialised_stats, doclen_lbound);
    pack_uint(serialised_stats, wdf_ubound);
    // doclen_ubound should always be >= wdf_ubound, so we store the
    // difference as it may encode smaller.  wdf_ubound is likely to
    // be larger than doclen_lbound.
    pack_uint(serialised_stats, doclen_ubound - wdf_ubound);
    pack_uint(serialised_stats, oldest_changeset);
    pack_uint(serialised_stats, total_doclen);
    pack_uint(serialised_stats, spelling_wordfreq_ubound);
    // If total_doclen == 0 then unique_terms is always zero (or there are no
    // documents at all) so storing these just complicates things because
    // uniq_terms_lbound could legitimately be zero.
    if (total_doclen != 0) {
        // We rely on uniq_terms_lbound being non-zero to detect if it's present
        // for a single file DB.
        Assert(uniq_terms_lbound != 0);
        pack_uint(serialised_stats, uniq_terms_lbound);
        pack_uint(serialised_stats, uniq_terms_ubound);
    }
}
 
void
HoneyVersion::unserialise_stats()
{
    const char* p = serialised_stats.data();
    const char* end = p + serialised_stats.size();
    if (p == end) {
        doccount = 0;
        total_doclen = 0;
        last_docid = 0;
        doclen_lbound = 0;
        doclen_ubound = 0;
        wdf_ubound = 0;
        oldest_changeset = 0;
        spelling_wordfreq_ubound = 0;
        uniq_terms_lbound = 0;
        uniq_terms_ubound = 0;
        return;
    }
 
    if (!unpack_uint(&p, end, &doccount) ||
        !unpack_uint(&p, end, &last_docid) ||
        !unpack_uint(&p, end, &doclen_lbound) ||
        !unpack_uint(&p, end, &wdf_ubound) ||
        !unpack_uint(&p, end, &doclen_ubound) ||
        !unpack_uint(&p, end, &oldest_changeset) ||
        !unpack_uint(&p, end, &total_doclen) ||
        !unpack_uint(&p, end, &spelling_wordfreq_ubound)) {
        const char* m = p ?
            "Bad serialised DB stats (overflowed)" :
            "Bad serialised DB stats (out of data)";
        throw Xapian::DatabaseCorruptError(m);
    }
 
    // last_docid must always be >= doccount.
    last_docid += doccount;
    // doclen_ubound should always be >= wdf_ubound, so we store the
    // difference as it may encode smaller.  wdf_ubound is likely to
    // be larger than doclen_lbound.
    doclen_ubound += wdf_ubound;
 
    // We don't check if there's undecoded data between p and end - in the
    // single-file DB case there will be extra zero bytes in serialised_stats,
    // and more generally it's useful to be able to add new stats when it is
    // safe for old versions to just ignore them and there are sensible values
    // to use when a new version reads an old database.
 
    // Read bounds on unique_terms if stored.  This test relies on the first
    // byte of pack_uint(x) being zero if and only if x is zero, and on
    // uniq_terms_lbound being non-zero.
    if (p == end || *p == '\0') {
        // No bounds stored so use weak bounds based on other stats.
        if (total_doclen == 0) {
            uniq_terms_lbound = uniq_terms_ubound = 0;
        } else {
            Assert(doclen_lbound != 0);
            Assert(wdf_ubound != 0);
            uniq_terms_lbound = (doclen_lbound - 1) / wdf_ubound + 1;
            uniq_terms_ubound = doclen_ubound;
        }
    } else if (!unpack_uint(&p, end, &uniq_terms_lbound) ||
               !unpack_uint(&p, end, &uniq_terms_ubound)) {
        const char* m = p ?
            "Bad serialised unique_terms bounds (overflowed)" :
            "Bad serialised unique_terms bounds (out of data)";
        throw Xapian::DatabaseCorruptError(m);
    }
}
 
void
HoneyVersion::merge_stats(const HoneyVersion& o)
{
    doccount += o.get_doccount();
    if (doccount < o.get_doccount()) {
        throw Xapian::DatabaseError("doccount overflowed!");
    }
 
    doclen_lbound = min_non_zero(doclen_lbound, o.get_doclength_lower_bound());
    doclen_ubound = max(doclen_ubound, o.get_doclength_upper_bound());
    wdf_ubound = max(wdf_ubound, o.get_wdf_upper_bound());
    total_doclen += o.get_total_doclen();
    if (total_doclen < o.get_total_doclen()) {
        throw Xapian::DatabaseError("Total document length overflowed!");
    }
 
    // The upper bounds might be on the same word, so we must sum them.
    spelling_wordfreq_ubound += o.get_spelling_wordfreq_upper_bound();
 
    uniq_terms_lbound = min_non_zero(uniq_terms_lbound,
                                     o.get_unique_terms_lower_bound());
    uniq_terms_ubound = max(uniq_terms_ubound,
                            o.get_unique_terms_upper_bound());
}
 
void
HoneyVersion::merge_stats(Xapian::doccount o_doccount,
                          Xapian::termcount o_doclen_lbound,
                          Xapian::termcount o_doclen_ubound,
                          Xapian::termcount o_wdf_ubound,
                          Xapian::totallength o_total_doclen,
                          Xapian::termcount o_spelling_wordfreq_ubound,
                          Xapian::termcount o_uniq_terms_lbound,
                          Xapian::termcount o_uniq_terms_ubound)
{
    doccount += o_doccount;
    if (doccount < o_doccount) {
        throw Xapian::DatabaseError("doccount overflowed!");
    }
 
    doclen_lbound = min_non_zero(doclen_lbound, o_doclen_lbound);
    doclen_ubound = max(doclen_ubound, o_doclen_ubound);
    wdf_ubound = max(wdf_ubound, o_wdf_ubound);
    total_doclen += o_total_doclen;
    if (total_doclen < o_total_doclen) {
        throw Xapian::DatabaseError("Total document length overflowed!");
    }
 
    // The upper bounds might be on the same word, so we must sum them.
    spelling_wordfreq_ubound += o_spelling_wordfreq_ubound;
 
    uniq_terms_lbound = min_non_zero(uniq_terms_lbound, o_uniq_terms_lbound);
    uniq_terms_ubound = max(uniq_terms_ubound, o_uniq_terms_ubound);
}
 
void
HoneyVersion::cancel()
{
    LOGCALL_VOID(DB, "HoneyVersion::cancel", NO_ARGS);
    for (unsigned table_no = 0; table_no < Honey::MAX_; ++table_no) {
        root[table_no] = old_root[table_no];
    }
    unserialise_stats();
}
 
const string
HoneyVersion::write(honey_revision_number_t new_rev, int flags)
{
    LOGCALL(DB, const string, "HoneyVersion::write", new_rev|flags);
 
    string s(HONEY_VERSION_MAGIC, HONEY_VERSION_MAGIC_AND_VERSION_LEN);
    s.append(uuid.data(), uuid.BINARY_SIZE);
 
    pack_uint(s, new_rev);
 
    for (unsigned table_no = 0; table_no < Honey::MAX_; ++table_no) {
        root[table_no].serialise(s);
    }
 
    // Serialise database statistics.
    serialise_stats();
    s += serialised_stats;
 
    string tmpfile;
    if (!single_file()) {
        tmpfile = db_dir;
        // In dangerous mode, just write the new version file in place.
        if (flags & Xapian::DB_DANGEROUS)
            tmpfile += "/iamhoney";
        else
            tmpfile += "/v.tmp";
 
        int open_flags = O_CREAT|O_TRUNC|O_WRONLY|O_BINARY;
        fd = posixy_open(tmpfile.c_str(), open_flags, 0666);
        if (rare(fd < 0)) {
            string msg = "Couldn't write new rev file: ";
            msg += tmpfile;
            throw Xapian::DatabaseOpeningError(msg, errno);
        }
 
        if (flags & Xapian::DB_DANGEROUS)
            tmpfile = string();
    }
 
    try {
        io_write(fd, s.data(), s.size());
    } catch (...) {
        if (!single_file())
            (void)close(fd);
        throw;
    }
 
    RETURN(tmpfile);
}
 
bool
HoneyVersion::sync(const string& tmpfile,
                   honey_revision_number_t new_rev, int flags)
{
    Assert(new_rev > rev || rev == 0);
 
    if (single_file()) {
        if ((flags & Xapian::DB_NO_SYNC) == 0 &&
            ((flags & Xapian::DB_FULL_SYNC) ?
              !io_full_sync(fd) :
              !io_sync(fd))) {
            // FIXME what to do?
        }
    } else {
        int fd_to_close = fd;
        fd = -1;
        if ((flags & Xapian::DB_NO_SYNC) == 0 &&
            ((flags & Xapian::DB_FULL_SYNC) ?
              !io_full_sync(fd_to_close) :
              !io_sync(fd_to_close))) {
            int save_errno = errno;
            (void)close(fd_to_close);
            if (!tmpfile.empty())
                (void)unlink(tmpfile.c_str());
            errno = save_errno;
            return false;
        }
 
        if (close(fd_to_close) != 0) {
            if (!tmpfile.empty()) {
                int save_errno = errno;
                (void)unlink(tmpfile.c_str());
                errno = save_errno;
            }
            return false;
        }
 
        if (!tmpfile.empty()) {
            if (!io_tmp_rename(tmpfile, db_dir + "/iamhoney")) {
                return false;
            }
        }
    }
 
    for (unsigned table_no = 0; table_no < Honey::MAX_; ++table_no) {
        old_root[table_no] = root[table_no];
    }
 
    rev = new_rev;
    return true;
}
 
/* Only try to compress tags strictly longer than this many bytes.
 *
 * This can theoretically usefully be set as low as 4, but in practical terms
 * zlib can't compress in very many cases for short inputs and even when it can
 * the savings are small, so we default to a higher threshold to save CPU time
 * for marginal size reductions.
 */
const size_t COMPRESS_MIN = 18;
 
static const uint4 compress_min_tab[] = {
    0, // POSTLIST
    COMPRESS_MIN, // DOCDATA
    COMPRESS_MIN, // TERMLIST
    0, // POSITION
    COMPRESS_MIN, // SPELLING
    COMPRESS_MIN  // SYNONYM
};
 
void
HoneyVersion::create()
{
    uuid.generate();
    for (unsigned table_no = 0; table_no < Honey::MAX_; ++table_no) {
        root[table_no].init(compress_min_tab[table_no]);
    }
}
 
namespace Honey {
 
void
RootInfo::init(uint4 compress_min_)
{
    offset = 0;
    root = 0;
    num_entries = 0;
    compress_min = compress_min_;
    fl_serialised.resize(0);
}
 
void
RootInfo::serialise(string& s) const
{
    AssertRel(offset, >=, 0);
    std::make_unsigned_t<off_t> uoffset = offset;
    AssertRel(root, >=, offset);
    pack_uint(s, uoffset);
    pack_uint(s, root - uoffset);
    pack_uint(s, 0u);
    pack_uint(s, num_entries);
    pack_uint(s, 2048u >> 11);
    pack_uint(s, compress_min);
    pack_string(s, fl_serialised);
}
 
bool
RootInfo::unserialise(const char** p, const char* end)
{
    std::make_unsigned_t<off_t> uoffset, uroot;
    unsigned dummy_val;
    unsigned dummy_blocksize;
    if (!unpack_uint(p, end, &uoffset) ||
        !unpack_uint(p, end, &uroot) ||
        !unpack_uint(p, end, &dummy_val) ||
        !unpack_uint(p, end, &num_entries) ||
        !unpack_uint(p, end, &dummy_blocksize) ||
        !unpack_uint(p, end, &compress_min) ||
        !unpack_string(p, end, fl_serialised)) return false;
    offset = uoffset;
    root = uoffset + uroot;
    // Not meaningful, but still there so that existing honey databases
    // continue to work.
    (void)dummy_val;
    (void)dummy_blocksize;
    // Map old default to new default.
    if (compress_min == 4) {
        compress_min = COMPRESS_MIN;
    }
    return true;
}
 
}