const std = @import("std"); const fmt = std.fmt; const mem = std.mem; const math = std.math; const sort = std.sort; const unicode = std.unicode; const Allocator = std.mem.Allocator; const ArrayListUnmanaged = std.ArrayListUnmanaged; const ArrayList = std.ArrayList; const MultiArrayList = std.MultiArrayList; const StringHashMap = std.StringHashMap; const AutoHashMap = std.AutoHashMap; const BufSet = std.BufSet; const pad = @import("padding.zig"); const compress = @import("compress.zig"); const shellImport = @import("shell.zig"); const userImport = @import("user.zig"); const groupImport = @import("group.zig"); const cmph = @import("cmph.zig"); const bdz = @import("bdz.zig"); const User = userImport.User; const Group = groupImport.Group; const ShellSections = shellImport.ShellWriter.ShellSections; const Corpus = struct { arena: std.heap.ArenaAllocator, // sorted by name, by unicode codepoint users: []User, // sorted by gid groups: []Group, // columnar users and groups of the above usersMulti: MultiArrayList(User), groupsMulti: MultiArrayList(Group), name2user: StringHashMap(u32), name2group: StringHashMap(u32), group2users: []const []const u32, user2groups: []const []const u32, pub fn init( baseAllocator: Allocator, usersConst: []const User, groupsConst: []const Group, ) error{ OutOfMemory, InvalidUtf8, Duplicate, NotFound }!Corpus { var arena = std.heap.ArenaAllocator.init(baseAllocator); var allocator = arena.allocator(); errdefer arena.deinit(); var users = try allocator.alloc(User, usersConst.len); var groups = try allocator.alloc(Group, groupsConst.len); for (usersConst) |*user, i| users[i] = try user.clone(allocator); for (groupsConst) |*group, i| groups[i] = try group.clone(allocator); sort.sort(User, users, {}, cmpUser); sort.sort(Group, groups, {}, cmpGroup); var usersMulti = MultiArrayList(User){}; try usersMulti.ensureTotalCapacity(allocator, users.len); for (users) |user| usersMulti.appendAssumeCapacity(user); var groupsMulti = MultiArrayList(Group){}; try groupsMulti.ensureTotalCapacity(allocator, groups.len); for (groups) |group| groupsMulti.appendAssumeCapacity(group); var name2user = StringHashMap(u32).init(allocator); var name2group = StringHashMap(u32).init(allocator); for (usersMulti.items(.name)) |name, i| { var res1 = try name2user.getOrPut(name); if (res1.found_existing) return error.Duplicate; res1.value_ptr.* = @intCast(u32, i); } for (groupsMulti.items(.name)) |name, i| { var res1 = try name2group.getOrPut(name); if (res1.found_existing) return error.Duplicate; res1.value_ptr.* = @intCast(u32, i); } var group2users = try allocator.alloc([]u32, groups.len); // uses baseAllocator, because it will be freed before // returning from this function. This keeps the arena clean. var user2groups = try baseAllocator.alloc(ArrayListUnmanaged(u32), users.len); defer baseAllocator.free(user2groups); mem.set(ArrayListUnmanaged(u32), user2groups, ArrayListUnmanaged(u32){}); for (groupsMulti.items(.members)) |group_members, i| { var members = try allocator.alloc(u32, group_members.count()); members.len = 0; var it = group_members.iterator(); while (it.next()) |memberName| { if (name2user.get(memberName.*)) |user_idx| { members.len += 1; members[members.len - 1] = user_idx; try user2groups[user_idx].append(allocator, @intCast(u32, i)); } else { return error.NotFound; } } group2users[i] = members; } for (group2users) |*groupusers| { sort.sort(u32, groupusers.*, {}, comptime sort.asc(u32)); } var user2groups_final = try allocator.alloc([]const u32, users.len); user2groups_final.len = users.len; for (user2groups) |*usergroups, i| { sort.sort(u32, usergroups.items, {}, comptime sort.asc(u32)); user2groups_final[i] = usergroups.toOwnedSlice(allocator); } return Corpus{ .arena = arena, .users = users, .groups = groups, .usersMulti = usersMulti, .groupsMulti = groupsMulti, .name2user = name2user, .name2group = name2group, .group2users = group2users, .user2groups = user2groups_final, }; } pub fn deinit(self: *Corpus) void { self.arena.deinit(); self.* = undefined; } }; pub fn bdzGid(allocator: Allocator, corpus: *const Corpus) cmph.Error![]const u8 { return try cmph.pack_u32(allocator, corpus.groupsMulti.items(.gid)); } pub fn bdzGroupname(allocator: Allocator, corpus: *const Corpus) cmph.Error![]const u8 { return try cmph.pack_str(allocator, corpus.groupsMulti.items(.name)); } pub fn bdzUid(allocator: Allocator, corpus: *const Corpus) cmph.Error![]const u8 { return try cmph.pack_u32(allocator, corpus.usersMulti.items(.uid)); } pub fn bdzUsername(allocator: Allocator, corpus: *const Corpus) cmph.Error![]const u8 { return try cmph.pack_str(allocator, corpus.usersMulti.items(.name)); } // TODO(motiejus) there are a few problems: // - memory management for shell sections is a mess. Make it easier by ... // - shell module should accept a list of shells and spit out two slices // (allocated with a given allocator). There is too much dancing around // here. pub fn shellSections( allocator: Allocator, corpus: *const Corpus, ) error{ OutOfMemory, Overflow }!ShellSections { var popcon = shellImport.ShellWriter.init(allocator); for (corpus.usersMulti.items(.shell)) |shell| try popcon.put(shell); return popcon.toOwnedSections(shellImport.max_shells); } pub const UserGids = 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, pub fn deinit(self: *UserGids, allocator: Allocator) void { allocator.free(self.idx2offset); allocator.free(self.blob); self.* = undefined; } }; pub fn userGids( allocator: Allocator, corpus: *const Corpus, ) error{ OutOfMemory, Overflow }!UserGids { 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; for (usergroups) |group_idx, i| scratch[i] = corpus.groups[group_idx].gid; 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 UserGids{ .idx2offset = idx2offset, .blob = blob.toOwnedSlice(), }; } pub const UsersSection = struct { // user index -> offset in blob idx2offset: []const u32, blob: []const u8, pub fn deinit(self: *UsersSection, allocator: Allocator) void { allocator.free(self.idx2offset); allocator.free(self.blob); self.* = undefined; } }; pub fn usersSection( allocator: Allocator, corpus: *const Corpus, gids: *const UserGids, shells: *const ShellSections, ) error{ OutOfMemory, Overflow, InvalidRecord }!UsersSection { var idx2offset = try allocator.alloc(u32, corpus.users.len); errdefer allocator.free(idx2offset); // as of writing each user takes 15 bytes + strings + 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(); for (corpus.users) |user, i| { const userOffset = try math.cast(u32, blob.items.len); std.debug.assert(userOffset & 7 == 0); idx2offset[i] = userOffset; try userImport.PackedUserHash.packTo( &blob, user, gids.idx2offset[i], shells.indices, ); } return UsersSection{ .idx2offset = idx2offset, .blob = blob.toOwnedSlice(), }; } pub const GroupMembers = struct { // group index to it's offset in blob idx2offset: []const u32, blob: []const u8, pub fn deinit(self: *GroupMembers, allocator: Allocator) void { allocator.free(self.idx2offset); allocator.free(self.blob); self.* = undefined; } }; pub fn groupMembers( allocator: Allocator, corpus: *const Corpus, user2offset: []const u32, ) error{OutOfMemory}!GroupMembers { var idx2offset = try allocator.alloc(u32, 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); for (corpus.groupsMulti.items(.group)) |group_users, i| { if (group_users.len == 0) { idx2offset[i] = 0; continue; } idx2offset[i] = blob.len; compress.appendUvarint(&blob, group_users.len); for (group_users) |userIdx| compress.appendUvarint(&blob, user2offset[userIdx]); } return GroupMembers{ .idx2offset = idx2offset, .blob = blob.toOwnedSlice(), }; } // cmpUser compares two users for sorting. By username's utf8 codepoints, ascending. fn cmpUser(_: void, a: User, b: User) bool { var utf8_a = (unicode.Utf8View.init(a.name) catch unreachable).iterator(); var utf8_b = (unicode.Utf8View.init(b.name) catch unreachable).iterator(); while (utf8_a.nextCodepoint()) |codepoint_a| { if (utf8_b.nextCodepoint()) |codepoint_b| { if (codepoint_a == codepoint_b) { continue; } else { return codepoint_a < codepoint_b; } } // a is a prefix of b. It is thus shorter. return false; } // b is a prefix of a return true; } fn cmpGroup(_: void, a: Group, b: Group) bool { return a.gid < b.gid; } pub const AllSections = struct { allocator: Allocator, bdz_gid: []const u8, bdz_groupname: []const u8, bdz_uid: []const u8, bdz_username: []const u8, users: UsersSection, shell_sections: ShellSections, shell_index: []const u8, shell_blob: []const u8, user_gids: UserGids, user_gids_b: []const u8, pub fn init( allocator: Allocator, corpus: *const Corpus, ) error{ Overflow, OutOfMemory, InvalidRecord }!AllSections { const bdz_gid = try bdzGid(allocator, corpus); const bdz_groupname = try bdzGroupname(allocator, corpus); const bdz_uid = try bdzUid(allocator, corpus); const bdz_username = try bdzUsername(allocator, corpus); const shell_sections = try shellSections(allocator, corpus); const shell_index = shell_sections.index; const shell_blob = shell_sections.blob; const user_gids = try userGids(allocator, corpus); const users = try usersSection( allocator, corpus, &user_gids, &shell_sections, ); return AllSections{ .allocator = allocator, .bdz_gid = bdz_gid, .bdz_groupname = bdz_groupname, .bdz_uid = bdz_uid, .bdz_username = bdz_username, .shell_sections = shell_sections, .shell_index = mem.sliceAsBytes(shell_index.constSlice()), .shell_blob = mem.sliceAsBytes(shell_blob.constSlice()), .user_gids = user_gids, .user_gids_b = user_gids.blob, .users = users, }; } pub fn deinit(self: *AllSections) void { self.allocator.free(self.bdz_gid); self.allocator.free(self.bdz_groupname); self.allocator.free(self.bdz_uid); self.allocator.free(self.bdz_username); self.shell_sections.deinit(); self.user_gids.deinit(self.allocator); self.users.deinit(self.allocator); self.* = undefined; } }; const testing = std.testing; fn testCorpus(allocator: Allocator) !Corpus { const users = [_]User{ User{ .uid = 128, .gid = 128, .name = "vidmantas", .gecos = "Vidmantas Kaminskas", .home = "/home/vidmantas", .shell = "/bin/bash", }, User{ .uid = 0, .gid = math.maxInt(u32), .name = "Name" ** 8, .gecos = "Gecos" ** 51, .home = "Home" ** 16, .shell = "She.LllL" ** 8, }, User{ .uid = 1002, .gid = 1002, .name = "svc-bar", .gecos = "", .home = "/", .shell = "/", }, User{ .uid = 65534, .gid = 65534, .name = "nobody", .gecos = "nobody", .home = "/nonexistent", .shell = "/usr/sbin/nologin", } }; var members1 = try groupImport.someMembers( allocator, &[_][]const u8{"vidmantas"}, ); defer members1.deinit(); var members2 = try groupImport.someMembers( allocator, &[_][]const u8{ "svc-bar", "vidmantas" }, ); defer members2.deinit(); var members3 = try groupImport.someMembers( allocator, &[_][]const u8{ "svc-bar", "Name" ** 8, "vidmantas" }, ); defer members3.deinit(); const groups = [_]Group{ Group{ .gid = 128, .name = "vidmantas", .members = members1, }, Group{ .gid = 9999, .name = "all", .members = members3, }, Group{ .gid = 0, .name = "service-account", .members = members2, } }; return try Corpus.init(allocator, users[0..], groups[0..]); } test "test corpus" { var corpus = try testCorpus(testing.allocator); defer corpus.deinit(); const name_name = 0; const nobody = 1; const svc_bar = 2; const vidmantas = 3; try testing.expectEqualStrings(corpus.users[name_name].name, "Name" ** 8); try testing.expectEqualStrings(corpus.users[nobody].name, "nobody"); try testing.expectEqualStrings(corpus.users[svc_bar].name, "svc-bar"); try testing.expectEqualStrings(corpus.users[vidmantas].name, "vidmantas"); const g_service_account = 0; const g_vidmantas = 1; const g_all = 2; try testing.expectEqualStrings(corpus.groups[g_service_account].name, "service-account"); try testing.expectEqualStrings(corpus.groups[g_vidmantas].name, "vidmantas"); try testing.expectEqualStrings(corpus.groups[g_all].name, "all"); try testing.expectEqual(corpus.name2user.get("404"), null); try testing.expectEqual(corpus.name2user.get("vidmantas").?, vidmantas); try testing.expectEqual(corpus.name2group.get("404"), null); try testing.expectEqual(corpus.name2group.get("vidmantas").?, g_vidmantas); const membersOfAll = corpus.group2users[g_all]; try testing.expectEqual(membersOfAll[0], name_name); try testing.expectEqual(membersOfAll[1], svc_bar); try testing.expectEqual(membersOfAll[2], vidmantas); const groupsOfVidmantas = corpus.user2groups[vidmantas]; try testing.expectEqual(groupsOfVidmantas[0], g_service_account); try testing.expectEqual(groupsOfVidmantas[1], g_vidmantas); try testing.expectEqual(groupsOfVidmantas[2], g_all); } test "test sections" { const allocator = testing.allocator; var corpus = try testCorpus(allocator); defer corpus.deinit(); var all = try AllSections.init(allocator, &corpus); defer all.deinit(); } test "userGids" { const allocator = testing.allocator; var corpus = try testCorpus(allocator); defer corpus.deinit(); var user_gids = try userGids(allocator, &corpus); defer user_gids.deinit(allocator); for (corpus.users) |_, userIdx| { const groups = corpus.user2groups[userIdx]; const offset = user_gids.idx2offset[userIdx]; if (groups.len == 0) { try testing.expect(offset == 0); continue; } var vit = try compress.VarintSliceIterator(user_gids.blob[offset..]); var it = compress.DeltaDecompressionIterator(&vit); try testing.expectEqual(it.remaining(), groups.len); var i: u64 = 0; while (try it.next()) |gid| : (i += 1) { try testing.expectEqual(gid, corpus.groups[groups[i]].gid); } } } test "pack gids" { const allocator = testing.allocator; var corpus = try testCorpus(allocator); defer corpus.deinit(); const cmph_gid = try cmph.pack_u32(allocator, corpus.groupsMulti.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); var hashes = &[_]u32{ k1, k2, k3 }; sort.sort(u32, hashes, {}, comptime sort.asc(u32)); for (hashes) |hash, i| try testing.expectEqual(i, hash); } fn testUser(name: []const u8) User { var result = std.mem.zeroes(User); result.name = name; return result; } test "users compare function" { const a = testUser("a"); const b = testUser("b"); const bb = testUser("bb"); try testing.expect(cmpUser({}, a, b)); try testing.expect(!cmpUser({}, b, a)); try testing.expect(cmpUser({}, a, bb)); try testing.expect(!cmpUser({}, bb, a)); try testing.expect(cmpUser({}, b, bb)); try testing.expect(!cmpUser({}, bb, b)); }