turbonss/lib/DB.zig

const std = @import("std");
const os = std.os;
const mem = std.mem;
const math = std.math;
const meta = std.meta;
const sort = std.sort;
const assert = std.debug.assert;
const Allocator = std.mem.Allocator;
const ArrayList = std.ArrayList;
const AutoHashMap = std.AutoHashMap;
const BoundedArray = std.BoundedArray;

const Corpus = @import("Corpus.zig");
const pad = @import("padding.zig");
const compress = @import("compress.zig");
const PackedUser = @import("user.zig").PackedUser;
const User = @import("user.zig").User;
const Group = @import("group.zig").Group;
const GroupStored = @import("group.zig").GroupStored;
const PackedGroup = @import("group.zig").PackedGroup;
const ShellSections = @import("shell.zig").ShellWriter.ShellSections;
const ShellReader = @import("shell.zig").ShellReader;
const ShellWriter = @import("shell.zig").ShellWriter;
const Header = @import("header.zig").Header;
const max_shells = @import("shell.zig").max_shells;
const section_length_bits = @import("header.zig").section_length_bits;
const section_length = @import("header.zig").section_length;
const cmph = @import("cmph.zig");
const bdz = @import("bdz.zig");

const zeroes = &[_]u8{0} ** section_length;

const DB = @This();
// All sections, as they end up in the DB. Order is important.
header: Header,
bdz_gid: []const u8,
bdz_groupname: []const u8,
bdz_uid: []const u8,
bdz_username: []const u8,
idx_gid2group: []const u32,
idx_groupname2group: []const u32,
idx_uid2user: []const u32,
idx_name2user: []const u32,
shell_index: []const u16,
shell_blob: []const u8,
groups: []const u8,
users: []const u8,
groupmembers: []const u8,
additional_gids: []const u8,

pub fn fromCorpus(
    allocator: Allocator,
    corpus: *const Corpus,
) error{ OutOfMemory, InvalidRecord, TooMany }!DB {
    const gids = corpus.groups.items(.gid);
    const gnames = corpus.groups.items(.name);
    const uids = corpus.users.items(.uid);
    const unames = corpus.users.items(.name);

    var bdz_gid = try cmph.packU32(allocator, gids);
    errdefer allocator.free(bdz_gid);

    var bdz_groupname = try cmph.packStr(allocator, gnames);
    errdefer allocator.free(bdz_groupname);

    var bdz_uid = try cmph.packU32(allocator, uids);
    errdefer allocator.free(bdz_uid);

    const bdz_username = try cmph.packStr(allocator, unames);
    errdefer allocator.free(bdz_username);

    var shell = try shellSections(allocator, corpus);
    defer shell.deinit();

    var additional_gids = try additionalGids(allocator, corpus);
    errdefer allocator.free(additional_gids.blob);

    var users = try usersSection(allocator, corpus, &additional_gids, &shell);
    allocator.free(additional_gids.idx2offset);
    errdefer allocator.free(users.blob);

    var groupmembers = try groupMembers(allocator, corpus, users.idx2offset);
    errdefer allocator.free(groupmembers.blob);

    var groups = try groupsSection(allocator, corpus, groupmembers.idx2offset);
    allocator.free(groupmembers.idx2offset);
    errdefer allocator.free(groups.blob);

    var idx_gid2group = try bdzIdx(u32, allocator, bdz_gid, gids, groups.idx2offset);
    errdefer allocator.free(idx_gid2group);

    var idx_groupname2group = try bdzIdx([]const u8, allocator, bdz_groupname, gnames, groups.idx2offset);
    allocator.free(groups.idx2offset);
    errdefer allocator.free(idx_groupname2group);

    var idx_uid2user = try bdzIdx(u32, allocator, bdz_uid, uids, users.idx2offset);
    errdefer allocator.free(idx_uid2user);

    var idx_name2user = try bdzIdx([]const u8, allocator, bdz_username, unames, users.idx2offset);
    allocator.free(users.idx2offset);
    errdefer allocator.free(idx_name2user);

    const header = Header{
        .nblocks_shell_blob = nblocks(u8, shell.blob.constSlice()),
        .num_shells = shell.len,
        .num_groups = groups.len,
        .num_users = users.len,
        .nblocks_bdz_gid = nblocks(u32, bdz_gid),
        .nblocks_bdz_groupname = nblocks(u32, bdz_groupname),
        .nblocks_bdz_uid = nblocks(u32, bdz_uid),
        .nblocks_bdz_username = nblocks(u32, bdz_username),
        .nblocks_groups = nblocks(u64, groups.blob),
        .nblocks_users = nblocks(u64, users.blob),
        .nblocks_groupmembers = nblocks(u64, groupmembers.blob),
        .nblocks_additional_gids = nblocks(u64, additional_gids.blob),
    };

    return DB{
        .header = header,
        .bdz_gid = bdz_gid,
        .bdz_groupname = bdz_groupname,
        .bdz_uid = bdz_uid,
        .bdz_username = bdz_username,
        .idx_gid2group = idx_gid2group,
        .idx_groupname2group = idx_groupname2group,
        .idx_uid2user = idx_uid2user,
        .idx_name2user = idx_name2user,
        .shell_index = shell.index.constSlice(),
        .shell_blob = shell.blob.constSlice(),
        .groups = groups.blob,
        .users = users.blob,
        .groupmembers = groupmembers.blob,
        .additional_gids = additional_gids.blob,
    };
}

const DB_fields = meta.fields(DB);
pub fn iov(self: *const DB) BoundedArray(os.iovec_const, DB_fields.len * 2) {
    var result = BoundedArray(os.iovec_const, DB_fields.len * 2).init(0) catch unreachable;
    inline for (DB_fields) |field| {
        comptime assertDefinedLayout(field.field_type);
        const value = @field(self, field.name);
        const bytes: []const u8 = switch (@TypeOf(value)) {
            Header => mem.asBytes(&value),
            else => mem.sliceAsBytes(value),
        };
        result.appendAssumeCapacity(os.iovec_const{
            .iov_base = bytes.ptr,
            .iov_len = bytes.len,
        });
        const padding = pad.until(usize, section_length_bits, bytes.len);
        if (padding != 0)
            result.appendAssumeCapacity(.{
                .iov_base = zeroes,
                .iov_len = padding,
            });
    }

    return result;
}

pub fn fromBytes(buf: []const u8) Header.Invalid!DB {
    const header = try Header.fromBytes(buf);
    const lengths = .{
        .{ "bdz_gid", header.nblocks_bdz_gid },
        .{ "bdz_groupname", header.nblocks_bdz_groupname },
        .{ "bdz_uid", header.nblocks_bdz_uid },
        .{ "bdz_username", header.nblocks_bdz_username },
        .{ "idx_gid2group", nblocks_n(header.num_groups * 4) },
        .{ "idx_groupname2group", nblocks_n(header.num_groups * 4) },
        .{ "idx_uid2user", nblocks_n(header.num_users * 4) },
        .{ "idx_name2user", nblocks_n(header.num_users * 4) },
        .{ "shell_index", nblocks_n(header.num_shells * 2) },
        .{ "shell_blob", header.nblocks_shell_blob },
        .{ "groups", header.nblocks_groups },
        .{ "users", header.nblocks_users },
        .{ "groupmembers", header.nblocks_groupmembers },
        .{ "additional_gids", header.nblocks_additional_gids },
    };

    var result: DB = undefined;
    result.header = header;
    var offset = comptime nblocks_n(usize, @sizeOf(Header));
    comptime assert(DB_fields[0].name == "header");
    inline for (DB_fields[1..]) |field, i| {
        assert(lengths[i][0] == field.name);

        const start = offset << 6;
        const end = (offset + lengths[i][1]) << 6;
        const value = mem.bytesAsValue(field.field_type, buf[start..end]);
        @field(result, field.name) = value;
        offset += lengths[i][1];
    }

    return result;
}

pub fn deinit(self: *DB, allocator: Allocator) void {
    allocator.free(self.bdz_gid);
    allocator.free(self.bdz_groupname);
    allocator.free(self.bdz_uid);
    allocator.free(self.bdz_username);
    allocator.free(self.idx_gid2group);
    allocator.free(self.idx_groupname2group);
    allocator.free(self.idx_uid2user);
    allocator.free(self.idx_name2user);
    allocator.free(self.groups);
    allocator.free(self.users);
    allocator.free(self.groupmembers);
    allocator.free(self.additional_gids);
    self.* = undefined;
}

fn shellSections(
    allocator: Allocator,
    corpus: *const Corpus,
) error{OutOfMemory}!ShellSections {
    var popcon = ShellWriter.init(allocator);
    for (corpus.users.items(.shell)) |shell|
        try popcon.put(shell);
    return popcon.toOwnedSections(max_shells);
}

const AdditionalGids = struct {
    // user index -> offset in blob
    idx2offset: []const u64,
    // compressed user gids blob. A blob contains N <= users.len items,
    // an item is:
    //   len: varint
    //   gid: [varint]varint,
    // ... and the gid list is delta-compressed.
    blob: []const u8,
};

fn additionalGids(
    allocator: Allocator,
    corpus: *const Corpus,
) error{OutOfMemory}!AdditionalGids {
    var blob = ArrayList(u8).init(allocator);
    errdefer blob.deinit();
    var idx2offset = try allocator.alloc(u64, corpus.users.len);
    errdefer allocator.free(idx2offset);

    // zero'th entry is empty, so groupless users can refer to it.
    try compress.appendUvarint(&blob, 0);

    var scratch = try allocator.alloc(u32, 256);
    defer allocator.free(scratch);
    for (corpus.user2groups) |usergroups, user_idx| {
        if (usergroups.len == 0) {
            idx2offset[user_idx] = 0;
            continue;
        }
        idx2offset[user_idx] = blob.items.len;
        scratch = try allocator.realloc(scratch, usergroups.len);
        scratch.len = usergroups.len;
        const corpusGids = corpus.groups.items(.gid);
        for (usergroups) |group_idx, i|
            scratch[i] = corpusGids[group_idx];
        compress.deltaCompress(u32, scratch) catch |err| switch (err) {
            error.NotSorted => unreachable,
        };
        try compress.appendUvarint(&blob, usergroups.len);
        for (scratch) |gid|
            try compress.appendUvarint(&blob, gid);
    }

    return AdditionalGids{
        .idx2offset = idx2offset,
        .blob = blob.toOwnedSlice(),
    };
}

const UsersSection = struct {
    // number of users in this section
    len: u32,
    // user index -> offset in blob
    idx2offset: []const u32,
    blob: []const u8,
};

fn usersSection(
    allocator: Allocator,
    corpus: *const Corpus,
    gids: *const AdditionalGids,
    shells: *const ShellSections,
) error{ OutOfMemory, InvalidRecord, TooMany }!UsersSection {
    var idx2offset = try allocator.alloc(u32, corpus.users.len);
    errdefer allocator.free(idx2offset);
    // as of writing each user takes 12 bytes + blobs + padding, padded to
    // 8 bytes. 24 is an optimistic lower bound for an average record size.
    var blob = try ArrayList(u8).initCapacity(allocator, 24 * corpus.users.len);
    errdefer blob.deinit();
    var i: usize = 0;
    while (i < corpus.users.len) : (i += 1) {
        // TODO: this is inefficient by calling `.slice()` on every iteration
        const user = corpus.users.get(i);
        const user_offset = math.cast(u35, blob.items.len) catch |err| switch (err) {
            error.Overflow => return error.TooMany,
        };
        assert(user_offset & 7 == 0);
        idx2offset[i] = @truncate(u32, user_offset >> 3);
        try PackedUser.packTo(
            &blob,
            user,
            gids.idx2offset[i],
            shells.shell2idx,
        );
        try pad.arrayList(&blob, PackedUser.alignment_bits);
    }
    return UsersSection{
        .len = @intCast(u32, corpus.users.len),
        .idx2offset = idx2offset,
        .blob = blob.toOwnedSlice(),
    };
}

const GroupMembers = struct {
    // group index to it's offset in blob
    idx2offset: []const u64,
    // members are delta-varint encoded byte-offsets to the user struct
    blob: []const u8,
};

fn groupMembers(
    allocator: Allocator,
    corpus: *const Corpus,
    user2offset: []const u32,
) error{OutOfMemory}!GroupMembers {
    var idx2offset = try allocator.alloc(u64, corpus.groups.len);
    errdefer allocator.free(idx2offset);
    var blob = ArrayList(u8).init(allocator);
    errdefer blob.deinit();
    // zero'th entry is empty, so empty groups can refer to it
    try compress.appendUvarint(&blob, 0);

    var scratch = try ArrayList(u32).initCapacity(allocator, 1024);
    defer scratch.deinit();

    for (corpus.group2users) |members, group_idx| {
        if (members.len == 0) {
            idx2offset[group_idx] = 0;
            continue;
        }

        idx2offset[group_idx] = blob.items.len;
        try scratch.ensureTotalCapacity(members.len);
        scratch.items.len = members.len;
        for (members) |user_idx, i|
            scratch.items[i] = user2offset[user_idx];

        compress.deltaCompress(u32, scratch.items) catch |err| switch (err) {
            error.NotSorted => unreachable,
        };
        try compress.appendUvarint(&blob, members.len);
        for (scratch.items) |elem|
            try compress.appendUvarint(&blob, elem);
    }
    return GroupMembers{
        .idx2offset = idx2offset,
        .blob = blob.toOwnedSlice(),
    };
}

const GroupsSection = struct {
    // number of groups in this section
    len: u32,
    // group index -> offset in blob
    idx2offset: []const u32,
    blob: []const u8,
};

fn groupsSection(
    allocator: Allocator,
    corpus: *const Corpus,
    members_offset: []const u64,
) error{ OutOfMemory, InvalidRecord }!GroupsSection {
    var idx2offset = try allocator.alloc(u32, corpus.groups.len);
    errdefer allocator.free(idx2offset);

    var blob = try ArrayList(u8).initCapacity(allocator, 8 * corpus.groups.len);
    errdefer blob.deinit();

    var i: usize = 0;
    while (i < corpus.groups.len) : (i += 1) {
        // TODO: this is inefficient; it's calling `.slice()` on every iteration
        const group = corpus.groups.get(i);
        const group_offset = @intCast(u32, blob.items.len);
        assert(group_offset & 7 == 0);
        idx2offset[i] = @truncate(u32, group_offset >> 3);
        const group_stored = GroupStored{
            .gid = group.gid,
            .name = group.name,
            .members_offset = members_offset[i],
        };
        try PackedGroup.packTo(&blob, group_stored);
        try pad.arrayList(&blob, PackedGroup.alignment_bits);
    }

    return GroupsSection{
        .len = @intCast(u32, corpus.groups.len),
        .idx2offset = idx2offset,
        .blob = blob.toOwnedSlice(),
    };
}

// creates a bdz index using packed_mphf.
//   hash = bdz_search(packed_mphf, keys[i]);
//   result[hash] = idx2offset[i];
fn bdzIdx(
    comptime T: type,
    allocator: Allocator,
    packed_mphf: []const u8,
    keys: []const T,
    idx2offset: []const u32,
) error{OutOfMemory}![]const u32 {
    const search_fn = comptime blk: {
        switch (T) {
            u32 => break :blk bdz.search_u32,
            []const u8 => break :blk bdz.search,
            else => unreachable,
        }
    };
    assert(keys.len <= math.maxInt(u32));
    var result = try allocator.alloc(u32, keys.len);
    for (keys) |key, i|
        result[search_fn(packed_mphf, key)] = idx2offset[i];
    return result;
}

// nblocks_n returns how many blocks a given number of bytes will take
fn nblocks_n(comptime T: type, nbytes: usize) T {
    const B = switch (T) {
        u8 => u14,
        u32 => u38,
        u64 => u70,
        else => @compileError("only u8, u32 and u64 are supported"),
    };
    const upper = pad.roundUp(B, section_length_bits, @intCast(B, nbytes));
    assert(upper & (section_length - 1) == 0);
    return @truncate(T, upper >> 6);
}

// nblocks returns how many blocks a particular slice will take.
fn nblocks(comptime T: type, arr: []const u8) T {
    return nblocks_n(T, arr.len);
}

fn assertDefinedLayout(comptime T: type) void {
    return switch (T) {
        u8, u16, u32, u64 => {},
        else => switch (@typeInfo(T)) {
            .Array, .Pointer => assertDefinedLayout(meta.Elem(T)),
            .Enum => assertDefinedLayout(meta.Tag(T)),
            .Struct => {
                if (meta.containerLayout(T) == .Auto)
                    @compileError("layout of " ++ @typeName(T) ++ " is undefined");
                for (meta.fields(T)) |field|
                    assertDefinedLayout(field.field_type);
            },
            else => @compileError("unexpected type " ++ @typeName(T)),
        },
    };
}

const testing = std.testing;

test "test groups, group members and users" {
    const allocator = testing.allocator;
    var corpus = try Corpus.testCorpus(allocator);
    defer corpus.deinit();

    var db = try DB.fromCorpus(allocator, &corpus);
    defer db.deinit(allocator);

    // TODO: replace with an integration test when high-level
    // reader API is present
    //const blob = sections.groupmembers.blob;
    //var i: usize = 0;
    //while (i < corpus.groups.len) : (i += 1) {
    //const offset = sections.groupmembers.idx2offset[i];
    //var vit = try compress.VarintSliceIterator(blob[offset..]);
    //var it = compress.DeltaDecompressionIterator(&vit);
    //for (corpus.group2users[i]) |user_idx| {
    //    const got_user_offset = (try it.next()).?;
    //    const want_user_offset = sections.users.idx2offset[user_idx];
    //    try testing.expectEqual(got_user_offset, want_user_offset);
    //}
    //try testing.expectEqual(it.next(), null);
    //}

    //var it = PackedUser.iterator(sections.users.blob, sections.shell_reader);
    //i = 0;
    //while (i < corpus.users.len) : (i += 1) {
    //    const got = (try it.next()).?;
    //    const user = corpus.users.get(i);
    //    try testing.expectEqual(user.uid, got.uid());
    //    try testing.expectEqual(user.gid, got.gid());
    //    try testing.expectEqualStrings(user.name, got.name());
    //    try testing.expectEqualStrings(user.gecos, got.gecos());
    //    try testing.expectEqualStrings(user.home, got.home());
    //    try testing.expectEqualStrings(user.shell, got.shell(sections.shell_reader));
    //}

    const fd = try os.memfd_create("test_turbonss_db", 0);
    defer os.close(fd);

    const written = try os.writev(fd, db.iov().constSlice());
    try testing.expect(written != 0);
}

test "additionalGids" {
    const allocator = testing.allocator;
    var corpus = try Corpus.testCorpus(allocator);
    defer corpus.deinit();

    var additional_gids = try additionalGids(allocator, &corpus);
    defer allocator.free(additional_gids.idx2offset);
    defer allocator.free(additional_gids.blob);

    var user_idx: usize = 0;
    while (user_idx < corpus.users.len) : (user_idx += 1) {
        const groups = corpus.user2groups[user_idx];
        const offset = additional_gids.idx2offset[user_idx];
        if (groups.len == 0) {
            try testing.expect(offset == 0);
            continue;
        }
        var vit = try compress.VarintSliceIterator(additional_gids.blob[offset..]);
        var it = compress.DeltaDecompressionIterator(&vit);
        try testing.expectEqual(it.remaining(), groups.len);
        var i: u64 = 0;
        const corpusGids = corpus.groups.items(.gid);
        while (try it.next()) |gid| : (i += 1) {
            try testing.expectEqual(gid, corpusGids[groups[i]]);
        }
        try testing.expectEqual(i, groups.len);
    }
}

test "pack gids" {
    const allocator = testing.allocator;
    var corpus = try Corpus.testCorpus(allocator);
    defer corpus.deinit();

    const cmph_gid = try cmph.packU32(allocator, corpus.groups.items(.gid));
    defer allocator.free(cmph_gid);

    const k1 = bdz.search_u32(cmph_gid, 0);
    const k2 = bdz.search_u32(cmph_gid, 128);
    const k3 = bdz.search_u32(cmph_gid, 9999);
    const k4 = bdz.search_u32(cmph_gid, 100000);
    var hashes = &[_]u32{ k1, k2, k3, k4 };
    sort.sort(u32, hashes, {}, comptime sort.asc(u32));
    for (hashes) |hash, i|
        try testing.expectEqual(i, hash);
}

const hash_offsets = &[_]u32{ 0, 10, 20, 30 };

fn expectUsedHashes(allocator: Allocator, arr: []const u32) !void {
    var used = AutoHashMap(u32, void).init(allocator);
    defer used.deinit();

    for (arr) |elem|
        try used.putNoClobber(elem, {});
    for (hash_offsets) |item|
        try testing.expect(used.get(item) != null);
}

test "bdzIdx on u32" {
    const keys = [_]u32{ 42, 1, 2, 3 };
    const mphf = try cmph.packU32(testing.allocator, keys[0..]);
    defer testing.allocator.free(mphf);
    var result = try bdzIdx(u32, testing.allocator, mphf, keys[0..], hash_offsets);
    defer testing.allocator.free(result);
    try expectUsedHashes(testing.allocator, result);
}

test "bdzIdx on str" {
    const keys = [_][]const u8{ "42", "1", "2", "3" };
    const mphf = try cmph.packStr(testing.allocator, keys[0..]);
    defer testing.allocator.free(mphf);
    var result = try bdzIdx([]const u8, testing.allocator, mphf, keys[0..], hash_offsets);
    defer testing.allocator.free(result);
    try expectUsedHashes(testing.allocator, result);
}

test "nblocks" {
    const tests = .{
        .{ 0, &[_]u8{} },
        .{ 1, &[_]u8{ 1, 2, 42 } },
        .{ 1, &[_]u8{1} ** 63 },
        .{ 1, &[_]u8{1} ** 64 },
        .{ 2, &[_]u8{1} ** 65 },
        .{ 255, &[_]u8{1} ** (255 * 64) },
    };

    inline for (tests) |tt| {
        try testing.expectEqual(nblocks(u8, tt[1]), tt[0]);
        try testing.expectEqual(nblocks(u32, tt[1]), tt[0]);
        try testing.expectEqual(nblocks(u64, tt[1]), tt[0]);
    }
}