Merge pull request #4 from CalebQ42/inode_finish

Re-do much of extraction

.zed/debug.json

@@ -0,0 +1,30 @@
+// Project-local debug tasks
+//
+// For more documentation on how to configure debug tasks,
+// see: https://zed.dev/docs/debugger
+[
+	{
+		"label": "Build & Run",
+
+		"adapter": "CodeLLDB",
+		"request": "launch",
+
+		"build": {
+			"command": "zig",
+			"args": [
+				"build",
+				"-Doptimize=Debug",
+				"-Duse_c_libs=true",
+				"-Dvalgrind=true",
+			],
+		},
+
+		"program": "zig-out/bin/unsquashfs",
+		"args": [
+			"--force",
+			"-d",
+			"testing/TestExtractUnsquashfs",
+			"testing/LinuxPATest.sfs",
+		],
+	},
+]

README.md

@@ -30,7 +30,6 @@ Sets the version of `unsquashfs` shown when `--version` is passed.
 
 Most features are present except for the following:
 
-* xattrs are not applied on extraction
 * When using Zig decompression libraries then lzo and lz4 compression types are unavailable. I don't _currently_ plan on spending the time to find and validate a library since neither is popular.
 * When using C decompression libraries, lzo is not supported by default due to [some issues](#build-considerations). If it's needed it's trivial to fix, but it's easiest to just leave it disabled.
 
@@ -38,16 +37,16 @@ Most features are present except for the following:
 
 This is some basic observation's I've made about this library's performance when compared to `unsquashfs`. Unless otherwise stated, most observations were made when extracting my test archive (which is fairly small and uses zstd compression) and with `--release=fast`.
 
-* Under ideal circumstances, my library is ~60% slower (.12s vs .19s).
+* Under ideal circumstances, my library is ~70% slower (.12s vs .20s).
 * Using Zig decompression libraries *significantly* increases decompression time by ~600%. Under ideal circumstances.
 * Performance improvements/regressions will be common. I'm still learning Zig.
 
-Times:
+Example Times:
 
 * *unsquashfs, multi-threaded*: .12s
 * *unsquashfs, single-threaded*: .13s
-* *C-libs, single-threaded*: .56s
-* *C-libs, multi-threaded*: .19s
+* *C-libs, single-threaded*: .45s
+* *C-libs, multi-threaded*: .20s
 * *Zig-libs, single-threaded*: 5.78s
 * *Zig-libs, multi-threaded*: 1.08s
 

build.zig

@@ -21,12 +21,12 @@ pub fn build(b: *std.Build) !void {
     });
     mod.addOptions("config", zig_squashfs_options);
     if (use_c_libs_option == true) {
-        mod.linkSystemLibrary("zlib", .{});
-        mod.linkSystemLibrary("lzma", .{});
+        mod.linkSystemLibrary("zlib-ng", .{ .preferred_link_mode = .static });
+        mod.linkSystemLibrary("lzma", .{ .preferred_link_mode = .static });
         if (allow_lzo == true)
-            mod.linkSystemLibrary("minilzo", .{});
-        mod.linkSystemLibrary("lz4", .{});
-        mod.linkSystemLibrary("zstd", .{});
+            mod.linkSystemLibrary("minilzo", .{ .preferred_link_mode = .static });
+        mod.linkSystemLibrary("lz4", .{ .preferred_link_mode = .static });
+        mod.linkSystemLibrary("zstd", .{ .preferred_link_mode = .static });
     }
 
     var version = version_string_option orelse "0.0.0-testing";
@@ -52,6 +52,7 @@ pub fn build(b: *std.Build) !void {
     const exe = b.addExecutable(.{
         .name = "unsquashfs",
         .root_module = exe_mod,
+        // .use_llvm = true, This can be needed to properly debug
     });
 
     const lib = b.addLibrary(.{

src/archive.zig

@@ -12,7 +12,7 @@ const InodeRef = Inode.Ref;
 const BlockSize = @import("inode_data/file.zig").BlockSize;
 const SfsFile = @import("file.zig");
 const Superblock = @import("super.zig").Superblock;
-const Table = @import("table.zig").Table;
+const Tables = @import("tables.zig");
 const MetadataReader = @import("util/metadata.zig");
 const OffsetFile = @import("util/offset_file.zig");
 const XattrTable = @import("xattr.zig");
@@ -22,14 +22,6 @@ const config = if (builtin.is_test) .{
     .allow_lzo = false,
 } else @import("config");
 
-/// Information about a fragment section. Multiple fragments are contained in the block described by a single FragEntry.
-/// The offset into the block and fragment size is stored in the file's inode.
-pub const FragEntry = packed struct {
-    start: u64,
-    size: BlockSize,
-    _: u32,
-};
-
 const Archive = @This();
 
 alloc: std.mem.Allocator,
@@ -39,10 +31,7 @@ decomp: Decomp.DecompFn,
 
 super: Superblock,
 
-frag_table: Table(FragEntry),
-id_table: Table(u16),
-export_table: Table(InodeRef),
-xattr_table: XattrTable,
+tables: ?Tables = null,
 
 /// Default settings using std.Thread.getCpuCount() threads and the minimum of 4gb or half of system memory for memory usage.
 pub fn init(alloc: std.mem.Allocator, fil: File, offset: u64) !Archive {
@@ -64,19 +53,16 @@ pub fn init(alloc: std.mem.Allocator, fil: File, offset: u64) !Archive {
         .fil = off_fil,
         .decomp = decomp,
         .super = super,
-        .frag_table = try .init(alloc, off_fil, decomp, super.frag_start, super.frag_count),
-        .id_table = try .init(alloc, off_fil, decomp, super.id_start, super.id_count),
-        .export_table = try .init(alloc, off_fil, decomp, super.export_start, super.inode_count),
-        .xattr_table = try .init(alloc, off_fil, decomp, super.xattr_start),
     };
 }
 pub fn deinit(self: *Archive) void {
-    self.frag_table.deinit();
-    self.export_table.deinit();
-    self.id_table.deinit();
+    if (self.tables != null)
+        self.tables.?.deinit();
 }
 
 pub fn inode(self: *Archive, alloc: std.mem.Allocator, num: u32) !Inode {
+    if (self.tables == null)
+        self.tables = try .init(alloc, self);
     const ref = try self.export_table.get(num - 1);
     var rdr = try self.fil.readerAt(ref.block_start + self.super.inode_start, &[0]u8{});
     var meta: MetadataReader = .init(alloc, &rdr.interface, &self.decomp);
@@ -85,6 +71,8 @@ pub fn inode(self: *Archive, alloc: std.mem.Allocator, num: u32) !Inode {
 }
 
 pub fn root(self: *Archive, alloc: std.mem.Allocator) !SfsFile {
+    if (self.tables == null)
+        self.tables = try .init(alloc, self);
     var rdr = try self.fil.readerAt(self.super.root_ref.block_start + self.super.inode_start, &[0]u8{});
     var meta: MetadataReader = .init(alloc, &rdr.interface, self.decomp);
     try meta.interface.discardAll(self.super.root_ref.block_offset);
@@ -93,12 +81,14 @@ pub fn root(self: *Archive, alloc: std.mem.Allocator) !SfsFile {
 }
 
 pub fn open(self: *Archive, alloc: std.mem.Allocator, path: []const u8) !SfsFile {
+    if (self.tables == null)
+        self.tables = try .init(alloc, self);
     var root_fil = try self.root(alloc);
     defer if (!SfsFile.pathIsSelf(path)) root_fil.deinit();
     return root_fil.open(path);
 }
 
-pub fn extract(self: *Archive, alloc: std.mem.Allocator, path: []const u8, options: ExtractionOptions) !void {
+pub fn extract(self: Archive, alloc: std.mem.Allocator, path: []const u8, options: ExtractionOptions) !void {
     var rdr = try self.fil.readerAt(self.super.root_ref.block_start + self.super.inode_start, &[0]u8{});
     var meta: MetadataReader = .init(self.alloc, &rdr.interface, self.decomp);
     try meta.interface.discardAll(self.super.root_ref.block_offset);

src/bin/unsquashfs.zig

@@ -20,6 +20,8 @@ const help_mgs =
     \\  -p <threads>    Specify how many threads to use. If no present or zero, the system's logical cores count is used.
     \\  -v              Verbose
     \\
+    \\  --force         Force extraction. If the destination already exists, it will be deleted.
+    \\
     \\  --help          Display this messages
     \\  --version       Display the version
     \\
@@ -34,6 +36,7 @@ var threads: u32 = 0;
 var verbose: bool = false;
 var ignore_xattrs: bool = false;
 var ignore_permissions: bool = false;
+var force: bool = false;
 
 pub fn main() !void {
     const alloc = std.heap.smp_allocator;
@@ -57,6 +60,8 @@ pub fn main() !void {
         .ignore_xattr = ignore_xattrs,
         .ignore_permissions = ignore_permissions,
     };
+    if (force)
+        try std.fs.cwd().deleteTree(extLoc);
     try arc.extract(alloc, extLoc, options); //TODO: Handle error gracefully.
 }
 
@@ -104,6 +109,9 @@ fn handleArgs(alloc: std.mem.Allocator, out: *Writer) !void {
         } else if (std.mem.eql(u8, arg, "-dp")) {
             ignore_permissions = true;
             continue;
+        } else if (std.mem.eql(u8, arg, "--force")) {
+            force = true;
+            continue;
         } else if (std.mem.eql(u8, arg, "--version")) {
             try out.print("zig-unsquashfs v", .{});
             try config.version.format(out);

src/decomp.zig

@@ -11,7 +11,7 @@ const config = if (builtin.is_test) .{
 } else @import("config");
 
 const c = @cImport({
-    @cInclude("zlib.h");
+    @cInclude("zlib-ng.h");
     @cInclude("lzma.h");
     @cInclude("lz4.h");
     @cInclude("zstd.h");
@@ -43,7 +43,7 @@ fn zigGzip(alloc: std.mem.Allocator, in: []u8, out: []u8) anyerror!usize {
 fn cGzip(alloc: std.mem.Allocator, in: []u8, out: []u8) anyerror!usize {
     _ = alloc;
     var out_len: usize = out.len;
-    const res = c.uncompress(out.ptr, &out_len, in.ptr, in.len);
+    const res = c.zng_uncompress2(out.ptr, &out_len, in.ptr, in.len);
     return switch (res) {
         c.Z_OK => out_len,
         c.Z_MEM_ERROR => error.NotEnoughMemory,
@@ -137,6 +137,7 @@ fn cXz(alloc: std.mem.Allocator, in: []u8, out: []u8) anyerror!usize {
         .avail_in = in.len,
         .next_out = out.ptr,
         .avail_out = out.len,
+        // .allocator = _, TODO: create a custom allocator based on alloc,
     };
     var res = c.lzma_stream_decoder(&stream, in.len * 2, 0);
     switch (res) {

src/inode.zig

@@ -12,10 +12,13 @@ const ExtractionOptions = @import("options.zig");
 const dir = @import("inode_data/dir.zig");
 const file = @import("inode_data/file.zig");
 const misc = @import("inode_data/misc.zig");
+const Tables = @import("tables.zig");
 const DataReader = @import("util/data.zig");
 const ThreadedDataReader = @import("util/data_threaded.zig");
+const InodeExtract = @import("util/extract.zig");
+const InodeFinish = @import("util/inode_finish.zig");
+const FinishUnion = InodeFinish.FinishUnion;
 const MetadataReader = @import("util/metadata.zig");
-const XattrTable = @import("xattr.zig");
 
 pub const Ref = packed struct {
     block_offset: u16,
@@ -94,7 +97,7 @@ pub fn read(alloc: std.mem.Allocator, rdr: *Reader, block_size: u32) !Inode {
         },
     };
 }
-pub fn readFromEntry(alloc: std.mem.Allocator, archive: *Archive, entry: DirEntry) !Inode {
+pub fn readFromEntry(alloc: std.mem.Allocator, archive: Archive, entry: DirEntry) !Inode {
     var rdr = try archive.fil.readerAt(archive.super.inode_start + entry.block_start, &[0]u8{});
     var meta: MetadataReader = .init(alloc, &rdr.interface, archive.decomp);
     try meta.interface.discardAll(entry.block_offset);
@@ -112,31 +115,17 @@ pub fn deinit(self: Inode, alloc: std.mem.Allocator) void {
 }
 
 /// Get the data reader for a file inode.
-pub fn dataReader(self: Inode, alloc: std.mem.Allocator, archive: *Archive) !DataReader {
+pub fn dataReader(self: Inode, alloc: std.mem.Allocator, archive: Archive, tables: *Tables) !DataReader {
     return switch (self.hdr.inode_type) {
-        .file => readerFromData(alloc, archive, self.data.file),
-        .ext_file => readerFromData(alloc, archive, self.data.ext_file),
+        .file => readerFromData(alloc, archive, tables, self.data.file),
+        .ext_file => readerFromData(alloc, archive, tables, self.data.ext_file),
         else => error.NotRegularFile,
     };
 }
-fn readerFromData(alloc: std.mem.Allocator, archive: *Archive, data: anytype) !DataReader {
-    var out: DataReader = .init(alloc, archive.*, data.block_sizes, data.block_start, data.size);
+fn readerFromData(alloc: std.mem.Allocator, archive: Archive, tables: *Tables, data: anytype) !DataReader {
+    var out: DataReader = .init(alloc, archive, data.block_sizes, data.block_start, data.size);
     if (data.frag_idx != 0xFFFFFFFF)
-        out.addFragment(try archive.frag_table.get(data.frag_idx), data.frag_block_offset);
-    return out;
-}
-/// Get a threaded data reader for a file inode.
-pub fn threadedDataReader(self: Inode, alloc: std.mem.Allocator, archive: *Archive) !ThreadedDataReader {
-    return switch (self.hdr.inode_type) {
-        .file => threadedReaderFromData(alloc, archive, self.data.file),
-        .ext_file => threadedReaderFromData(alloc, archive, self.data.ext_file),
-        else => error.NotRegularFile,
-    };
-}
-fn threadedReaderFromData(alloc: std.mem.Allocator, archive: *Archive, data: anytype) !ThreadedDataReader {
-    var out: ThreadedDataReader = .init(alloc, archive.*, data.block_sizes, data.block_start, data.size);
-    if (data.frag_idx != 0xFFFFFFFF)
-        out.addFragment(try archive.frag_table.get(data.frag_idx), data.frag_block_offset);
+        out.addFragment(try tables.frag_table.get(data.frag_idx), data.frag_block_offset);
     return out;
 }
 
@@ -155,7 +144,7 @@ fn entriesFromData(alloc: std.mem.Allocator, archive: Archive, data: anytype) ![
     return DirEntry.readDir(alloc, &meta.interface, data.size);
 }
 
-/// Returns the xattr index for the given inode. If the inode isn't an extended variant or doesn't have any, the returned value is the max u32 value (0xFFFFFFFF)
+/// Returns the xattr index for the given inode. If the inode isn't an extended variant or doesn't have any, the u32 max is returned (0xFFFFFFFF).
 pub fn xattrIdx(self: Inode) u32 {
     return switch (self.data) {
         .ext_dir => |d| d.xattr_id,
@@ -167,17 +156,19 @@ pub fn xattrIdx(self: Inode) u32 {
     };
 }
 
-inline fn setPermissionAndXattr(self: Inode, alloc: std.mem.Allocator, archive: *Archive, fil: std.fs.File, options: ExtractionOptions) !void {
+/// Applies the Inode's metadata to the given File.
+/// Mod time is always set, but permissions and xattrs are set based on the given ExtractionOptions.
+pub fn setMetadata(self: Inode, alloc: std.mem.Allocator, tables: *Tables, fil: std.fs.File, options: ExtractionOptions) !void {
     const time = @as(i128, self.hdr.mod_time) * 1000000000;
     try fil.updateTimes(time, time);
     if (!options.ignore_permissions) {
         try fil.chmod(self.hdr.permissions);
-        try fil.chown(try archive.id_table.get(self.hdr.uid_idx), try archive.id_table.get(self.hdr.gid_idx));
+        try fil.chown(try tables.id_table.get(self.hdr.uid_idx), try tables.id_table.get(self.hdr.gid_idx));
     }
     if (!options.ignore_xattr) {
         const idx = self.xattrIdx();
         if (idx == 0xFFFFFFFF) return;
-        const xattrs = try archive.xattr_table.get(alloc, idx);
+        const xattrs = try tables.xattr_table.get(alloc, idx);
         defer alloc.free(xattrs);
         for (xattrs) |kv| {
             const res = std.os.linux.fsetxattr(fil.handle, @ptrCast(kv.key), @ptrCast(kv.value), kv.value.len, 0);
@@ -194,343 +185,6 @@ inline fn setPermissionAndXattr(self: Inode, alloc: std.mem.Allocator, archive:
 }
 
 /// Extract the inode to the given path.
-pub fn extractTo(self: Inode, alloc: std.mem.Allocator, archive: *Archive, path: []const u8, options: ExtractionOptions) !void {
-    if (options.threads > 1) return self.extractToThreaded(alloc, archive, path, options);
-    switch (self.hdr.inode_type) {
-        .dir, .ext_dir => {
-            // Removing any trailing separators since that's the easiest path forward.
-            if (path[path.len - 1] == '/') return self.extractTo(alloc, archive, path[0 .. path.len - 1], options);
-            std.fs.cwd().makeDir(path) catch |err| {
-                if (err != std.fs.Dir.MakeError.PathAlreadyExists) return err;
-            };
-            const entries = try self.dirEntries(alloc, archive.*);
-            defer {
-                for (entries) |entry| entry.deinit(alloc);
-                alloc.free(entries);
-            }
-            for (entries) |entry| {
-                var new_path = try alloc.alloc(u8, path.len + 1 + entry.name.len);
-                @memcpy(new_path[0..path.len], path);
-                @memcpy(new_path[path.len + 1 ..], entry.name);
-                new_path[path.len] = '/';
-                defer alloc.free(new_path);
-
-                var inode: Inode = try readFromEntry(alloc, archive, entry);
-                defer inode.deinit(alloc);
-                try inode.extractTo(alloc, archive, new_path, options);
-            }
-
-            var fil = try std.fs.cwd().openFile(path, .{});
-            defer fil.close();
-            try self.setPermissionAndXattr(alloc, archive, fil, options);
-        },
-        .file, .ext_file => try self.extractRegFile(alloc, archive, path, options),
-        .symlink, .ext_symlink => try self.extractSymlink(path),
-        else => try self.extractDevice(alloc, archive, path, options),
-    }
-}
-
-const Parent = struct {
-    alloc: std.mem.Allocator,
-
-    inode: Inode,
-    path: []const u8,
-    archive: *Archive,
-    options: ExtractionOptions,
-
-    wg: WaitGroup = .{},
-    mut: Mutex = .{},
-
-    fn create(alloc: std.mem.Allocator, inode: Inode, path: []const u8, archive: *Archive, options: ExtractionOptions, dir_size: usize) !*Parent {
-        const out = try alloc.create(Parent);
-        errdefer alloc.destroy(out);
-        out.* = .{
-            .alloc = alloc,
-
-            .inode = inode,
-            .path = path,
-            .archive = archive,
-            .options = options,
-        };
-        out.wg.startMany(dir_size);
-        return out;
-    }
-
-    fn finish(p: *Parent) !void {
-        p.mut.lock();
-        {
-            defer p.mut.unlock();
-            p.wg.finish();
-            if (!p.wg.isDone()) return;
-        }
-        defer p.alloc.destroy(p);
-        var fil = try std.fs.cwd().openFile(p.path, .{});
-        defer fil.close();
-        try p.inode.setPermissionAndXattr(p.alloc, p.archive, fil, p.options);
-    }
-};
-
-/// Extract the inode to the given path. Multi-threaded.
-/// Functions identically to extractTo on all but regular files and directories.
-fn extractToThreaded(self: Inode, allocator: std.mem.Allocator, archive: *Archive, path: []const u8, options: ExtractionOptions) !void {
-    switch (self.hdr.inode_type) {
-        .dir, .ext_dir => {
-            // Removing any trailing separators since that's the easiest path forward.
-            if (path[path.len - 1] == '/') return self.extractToThreaded(allocator, archive, path[0 .. path.len - 1], options);
-
-            // Arena Allocator
-            var stack_alloc = std.heap.stackFallback(1024 * 1024, allocator);
-            var arena_alloc: std.heap.ArenaAllocator = .init(stack_alloc.get());
-            defer arena_alloc.deinit();
-            var thread_alloc: std.heap.ThreadSafeAllocator = .{ .child_allocator = arena_alloc.allocator() };
-            const alloc = thread_alloc.allocator();
-
-            var wg: WaitGroup = .{};
-            // defer if(!options.ignore_permissions) perms.?.deinit(alloc); We don't need to do this due to ArenaAllocator
-            var pool: Pool = undefined;
-            try pool.init(.{ .allocator = alloc, .n_jobs = options.threads - 1 });
-            defer pool.deinit();
-            var out_err: ?anyerror = null;
-
-            wg.start();
-            self.extractThread(alloc, archive, path, options, &wg, &pool, &out_err, null);
-            pool.waitAndWork(&wg);
-            if (out_err != null) return out_err.?;
-
-            var fil = try std.fs.cwd().openFile(path, .{});
-            defer fil.close();
-            try self.setPermissionAndXattr(alloc, archive, fil, options);
-        },
-        .file, .ext_file => {
-            var pool: Pool = undefined;
-            try pool.init(.{ .allocator = allocator, .n_jobs = options.threads - 1 });
-            defer pool.deinit();
-
-            // Arena Allocator
-            var stack_alloc = std.heap.stackFallback(1024 * 1024, allocator);
-            var arena_alloc: std.heap.ArenaAllocator = .init(stack_alloc.get());
-            defer arena_alloc.deinit();
-            var thread_alloc: std.heap.ThreadSafeAllocator = .{ .child_allocator = arena_alloc.allocator() };
-            const alloc = thread_alloc.allocator();
-
-            try self.extractRegFileThreaded(alloc, archive, path, options, &pool);
-
-            var fil = try std.fs.cwd().openFile(path, .{});
-            defer fil.close();
-            try self.setPermissionAndXattr(alloc, archive, fil, options);
-        },
-        .symlink, .ext_symlink => try self.extractSymlink(path),
-        else => try self.extractDevice(allocator, archive, path, options),
-    }
-}
-
-fn extractThreadEntry(
-    entry: DirEntry,
-    alloc: std.mem.Allocator,
-    archive: *Archive,
-    path: []const u8,
-    options: ExtractionOptions,
-    wg: *WaitGroup,
-    pool: *Pool,
-    out_err: *?anyerror,
-    parent: ?*Parent,
-) void {
-    var new_path = alloc.alloc(u8, path.len + entry.name.len + 1) catch |err| {
-        wg.finish();
-        out_err.* = err;
-        return;
-    };
-    @memcpy(new_path[0..path.len], path);
-    @memcpy(new_path[path.len + 1 ..], entry.name);
-    new_path[path.len] = '/';
-    var inode = readFromEntry(alloc, archive, entry) catch |err| {
-        out_err.* = err;
-        wg.finish();
-        return;
-    };
-    inode.extractThread(alloc, archive, new_path, options, wg, pool, out_err, parent);
-}
-
-/// Extract threadedly the inode to the path.
-fn extractThread(
-    self: Inode,
-    alloc: std.mem.Allocator,
-    archive: *Archive,
-    path: []const u8,
-    options: ExtractionOptions,
-    wg: *WaitGroup,
-    pool: *Pool,
-    out_err: *?anyerror,
-    parent: ?*Parent,
-) void {
-    if (options.verbose)
-        options.verbose_writer.?.print("Extracting inode #{} to {s}\n", .{ self.hdr.num, path }) catch {};
-    defer {
-        if (parent != null) parent.?.finish() catch |err| {
-            if (options.verbose)
-                options.verbose_writer.?.print("Error setting folder permission to {s}: {}\n", .{ path, err }) catch {};
-            out_err.* = err;
-        };
-        wg.finish();
-    }
-    if (out_err.* != null) return;
-    switch (self.hdr.inode_type) {
-        .dir, .ext_dir => {
-            _ = std.fs.cwd().makePathStatus(path) catch |err| {
-                if (options.verbose)
-                    options.verbose_writer.?.print("Error creating {s}: {}\n", .{ path, err }) catch {};
-                out_err.* = err;
-                return;
-            };
-
-            const entries = self.dirEntries(alloc, archive.*) catch |err| {
-                if (options.verbose)
-                    options.verbose_writer.?.print("Error getting directory entries for inode #{} (extracting to {s}): {}\n", .{ self.hdr.num, path, err }) catch {};
-                out_err.* = err;
-                return;
-            };
-            const p = Parent.create(alloc, self, path, archive, options, entries.len) catch |err| {
-                out_err.* = err;
-                return;
-            };
-            wg.startMany(entries.len);
-            // defer files.deinit(alloc); We don't need to do this due to ArenaAllocator
-            for (entries) |entry| {
-                if (entry.inode_type == .dir) {
-                    extractThreadEntry(entry, alloc, archive, path, options, wg, pool, out_err, p);
-                    continue;
-                }
-                pool.spawn(
-                    extractThreadEntry,
-                    .{
-                        entry,
-                        alloc,
-                        archive,
-                        path,
-                        options,
-                        wg,
-                        pool,
-                        out_err,
-                        p,
-                    },
-                ) catch |err| {
-                    wg.finish();
-                    if (options.verbose)
-                        options.verbose_writer.?.print("Error starting extraction thread: {}\n", .{err}) catch {};
-                    out_err.* = err;
-                    continue;
-                };
-            }
-        },
-        .file, .ext_file => {
-            self.extractRegFileThreaded(alloc, archive, path, options, pool) catch |err| {
-                if (options.verbose)
-                    options.verbose_writer.?.print("Error extracting file inode #{} to {s}: {}\n", .{ self.hdr.num, path, err }) catch {};
-                out_err.* = err;
-            };
-        },
-        .symlink, .ext_symlink => {
-            self.extractSymlink(path) catch |err| {
-                if (options.verbose)
-                    options.verbose_writer.?.print("Error extracting symlink inode #{} to {s}: {}\n", .{ self.hdr.num, path, err }) catch {};
-                out_err.* = err;
-            };
-        },
-        else => {
-            self.extractDevice(alloc, archive, path, options) catch |err| {
-                if (options.verbose)
-                    options.verbose_writer.?.print("Error extracting device/IPC inode #{} to {s}: {}\n", .{ self.hdr.num, path, err }) catch {};
-                out_err.* = err;
-            };
-        },
-    }
-}
-/// Creates and writes the inode file contents to the given path.
-/// Optionally set owner & permissions.
-///
-/// Assumes the inode is a file or ext_file type.
-fn extractRegFile(self: Inode, alloc: std.mem.Allocator, archive: *Archive, path: []const u8, options: ExtractionOptions) !void {
-    var fil = try std.fs.cwd().createFile(path, .{ .exclusive = true });
-    defer fil.close();
-    var wrt = fil.writer(&[0]u8{});
-    var dat_rdr = try self.dataReader(alloc, archive);
-    defer dat_rdr.deinit();
-    _ = try dat_rdr.interface.streamRemaining(&wrt.interface);
-    try wrt.interface.flush();
-
-    try self.setPermissionAndXattr(alloc, archive, fil, options);
-}
-/// Extract the inode file contents to the given path threadedly.
-/// pool is used to spawn threads.
-///
-/// Assumes the inode is a file or ext_file type.
-fn extractRegFileThreaded(self: Inode, alloc: std.mem.Allocator, archive: *Archive, path: []const u8, options: ExtractionOptions, pool: *Pool) !void {
-    var fil = try std.fs.cwd().createFile(path, .{});
-    defer fil.close();
-    var data = try self.threadedDataReader(alloc, archive);
-    try data.extractThreaded(fil, pool);
-
-    try self.setPermissionAndXattr(alloc, archive, fil, options);
-}
-/// Creates the symlink described by the inode.
-///
-/// Assumes the inode is a symlink or ext_symlink type.
-fn extractSymlink(self: Inode, path: []const u8) !void {
-    const target = switch (self.data) {
-        .symlink => |s| s.target,
-        .ext_symlink => |s| s.target,
-        else => unreachable,
-    };
-    try std.fs.cwd().symLink(target, path, .{});
-}
-/// Creates the device described by the inode.
-///
-/// Optionally set owner & permissions.
-/// Assumes the inode is a char_dev, block_dev, fifo, socket, or their extended counterparts.
-fn extractDevice(self: Inode, alloc: std.mem.Allocator, archive: *Archive, path: []const u8, options: ExtractionOptions) !void {
-    var mode: u32 = undefined;
-    var dev: u32 = 0;
-    switch (self.data) {
-        .char_dev => |d| {
-            mode = std.posix.S.IFCHR;
-            dev = d.dev;
-        },
-        .ext_char_dev => |d| {
-            mode = std.posix.S.IFCHR;
-            dev = d.dev;
-        },
-        .block_dev => |d| {
-            mode = std.posix.S.IFBLK;
-            dev = d.dev;
-        },
-        .ext_block_dev => |d| {
-            mode = std.posix.S.IFBLK;
-            dev = d.dev;
-        },
-        .fifo, .ext_fifo => mode = std.posix.S.IFIFO,
-        .socket, .ext_socket => mode = std.posix.S.IFSOCK,
-        else => unreachable,
-    }
-    const res: std.os.linux.E = @enumFromInt(std.os.linux.mknod(@ptrCast(path), mode, dev));
-    switch (res) {
-        .SUCCESS => {},
-        .ACCES => return std.fs.Dir.MakeError.AccessDenied,
-        .DQUOT => return std.fs.Dir.MakeError.DiskQuota,
-        .EXIST => return std.fs.Dir.MakeError.PathAlreadyExists,
-        .FAULT, .NOENT => return std.fs.Dir.MakeError.BadPathName,
-        .LOOP => return std.fs.Dir.MakeError.SymLinkLoop,
-        .NAMETOOLONG => return std.fs.Dir.MakeError.NameTooLong,
-        .NOMEM => return std.fs.Dir.MakeError.SystemResources,
-        .NOSPC => return std.fs.Dir.MakeError.NoSpaceLeft,
-        .NOTDIR => return std.fs.Dir.MakeError.NotDir,
-        .PERM => return std.fs.Dir.MakeError.PermissionDenied,
-        .ROFS => return std.fs.Dir.MakeError.ReadOnlyFileSystem,
-        else => return blk: {
-            std.debug.print("unhandled mknod result: {}\n", .{res});
-            break :blk std.fs.Dir.MakeError.Unexpected;
-        },
-    }
-    var fil = try std.fs.cwd().openFile(path, .{});
-    defer fil.close();
-    try self.setPermissionAndXattr(alloc, archive, fil, options);
+pub fn extractTo(self: Inode, alloc: std.mem.Allocator, archive: Archive, path: []const u8, options: ExtractionOptions) !void {
+    return InodeExtract.extractTo(alloc, self, archive, path, options);
 }

src/tables.zig

@@ -1,14 +1,45 @@
 const std = @import("std");
 const Mutex = std.Thread.Mutex;
 
+const Archive = @import("archive.zig");
 const DecompFn = @import("decomp.zig").DecompFn;
+const BlockSize = @import("inode_data/file.zig").BlockSize;
+const InodeRef = @import("inode.zig").Ref;
+const Superblock = @import("super.zig").Superblock;
 const MetadataReader = @import("util/metadata.zig");
 const OffsetFile = @import("util/offset_file.zig");
+const XattrTable = @import("xattr.zig");
 
-const TableError = error{
-    InvalidIndex,
+/// Information about a fragment section. Multiple fragments are contained in the block described by a single FragEntry.
+/// The offset into the block and fragment size is stored in the file's inode.
+pub const FragEntry = packed struct {
+    start: u64,
+    size: BlockSize,
+    _: u32,
 };
 
+const Tables = @This();
+
+frag_table: Table(FragEntry),
+id_table: Table(u16),
+export_table: Table(InodeRef),
+xattr_table: XattrTable,
+
+pub fn init(alloc: std.mem.Allocator, archive: Archive) !Tables {
+    return .{
+        .frag_table = try .init(alloc, archive.fil, archive.decomp, archive.super.frag_start, archive.super.frag_count),
+        .id_table = try .init(alloc, archive.fil, archive.decomp, archive.super.id_start, archive.super.id_count),
+        .export_table = try .init(alloc, archive.fil, archive.decomp, archive.super.export_start, archive.super.inode_count),
+        .xattr_table = try .init(alloc, archive.fil, archive.decomp, archive.super.xattr_start),
+    };
+}
+pub fn deinit(self: *Tables) void {
+    self.frag_table.deinit();
+    self.id_table.deinit();
+    self.export_table.deinit();
+    self.xattr_table.deinit();
+}
+
 /// A two-layer metadata table.
 pub fn Table(T: anytype) type {
     return struct {
@@ -47,7 +78,7 @@ pub fn Table(T: anytype) type {
         }
 
         pub fn get(self: *Self, idx: u32) !T {
-            if (idx >= self.values) return TableError.InvalidIndex;
+            if (idx >= self.values) return error.InvalidIndex;
             const block_num = idx / VALS_PER_BLOCK;
             const idx_offset = idx - (block_num * VALS_PER_BLOCK);
             if (self.tab.contains(block_num)) {

src/util/data.zig

@@ -6,9 +6,9 @@ const Writer = std.Io.Writer;
 const Limit = std.Io.Limit;
 
 const Archive = @import("../archive.zig");
-const FragEntry = Archive.FragEntry;
 const DecompFn = @import("../decomp.zig").DecompFn;
 const BlockSize = @import("../inode_data/file.zig").BlockSize;
+const FragEntry = @import("../tables.zig").FragEntry;
 const OffsetFile = @import("offset_file.zig");
 
 const DataReader = @This();

src/util/data_threaded.zig

@@ -8,9 +8,10 @@ const WaitGroup = std.Thread.WaitGroup;
 const Pool = std.Thread.Pool;
 
 const Archive = @import("../archive.zig");
-const FragEntry = Archive.FragEntry;
 const DecompFn = @import("../decomp.zig").DecompFn;
 const BlockSize = @import("../inode_data/file.zig").BlockSize;
+const FragEntry = @import("../tables.zig").FragEntry;
+const InodeFinish = @import("inode_finish.zig");
 const OffsetFile = @import("offset_file.zig");
 
 const ThreadedDataReader = @This();
@@ -25,6 +26,7 @@ blocks: []BlockSize,
 frag: ?FragEntry = null, // TODO: do something better?
 frag_offset: u32 = 0,
 size: u64,
+num_blocks: usize,
 
 start_offset: u64,
 
@@ -34,8 +36,12 @@ pub fn init(alloc: std.mem.Allocator, archive: Archive, blocks: []BlockSize, sta
         .fil = archive.fil,
         .decomp = archive.decomp,
         .block_size = archive.super.block_size,
+
         .blocks = blocks,
+
         .size = size,
+        .num_blocks = blocks.len,
+
         .start_offset = start,
     };
 }
@@ -43,141 +49,159 @@ pub fn init(alloc: std.mem.Allocator, archive: Archive, blocks: []BlockSize, sta
 pub fn addFragment(self: *ThreadedDataReader, entry: FragEntry, frag_offset: u32) void {
     self.frag = entry;
     self.frag_offset = frag_offset;
-}
-
-fn numBlocks(self: ThreadedDataReader) usize {
-    var res = self.blocks.len;
-    if (self.frag != null) res += 1;
-    return res;
+    self.num_blocks = self.blocks.len + 1;
 }
 
 /// Extract the data to the file threadedly, using pool to spawn threads.
-/// If multiple errors occur, thread spawning errors will have, then the last decompression error that occurs;
-///
-/// The function must be called from an unused DataReader. The DataReader is still usable afterwards.
-/// If only extractThreaded is used, there is no need to call deinit() afterwards.
-///
-/// The file will always be written to starting at 0.
-pub fn extractThreaded(self: ThreadedDataReader, file: std.fs.File, pool: *Pool) !void {
-    var wg: WaitGroup = .{};
-    wg.startMany(self.numBlocks());
-    var out_err: ?anyerror = null;
-
+/// If errors occur, they are set to finish.out_err.
+pub fn extractThreaded(self: ThreadedDataReader, file: std.fs.File, pool: *Pool, finish: *InodeFinish) void {
     var cur_write_offset: u64 = 0;
     var cur_read_offset: u64 = self.start_offset;
     for (0..self.blocks.len) |i| {
-        const cur_block_size = if (i == self.numBlocks() - 1) self.size % self.block_size else self.block_size;
-        try pool.spawn(workThreadBlocks, .{ self, file, cur_write_offset, cur_read_offset, self.blocks[i], cur_block_size, &wg, &out_err });
+        const cur_block_size = if (i == self.num_blocks - 1) self.size % self.block_size else self.block_size;
+        pool.spawn(workThreadBlocks, .{ self, file, cur_write_offset, cur_read_offset, self.blocks[i], cur_block_size, finish }) catch |res_err| {
+            finish.logError("Can't spawn pool task: {}", .{res_err});
+            finish.out_err.* = res_err;
+            finish.finish();
+        };
         cur_write_offset += cur_block_size;
         cur_read_offset += self.blocks[i].size;
     }
-    if (self.frag != null) {
-        try pool.spawn(workThreadFragment, .{ self, file, cur_write_offset, &wg, &out_err });
-    }
-    pool.waitAndWork(&wg);
-    if (out_err != null) return out_err.?;
+    if (self.frag != null)
+        pool.spawn(workThreadFragment, .{ self, file, cur_write_offset, finish }) catch |res_err| {
+            finish.logError("Can't spawn pool task: {}", .{res_err});
+            finish.out_err.* = res_err;
+            finish.finish();
+        };
 }
 
-fn workThreadBlocks(self: ThreadedDataReader, fil: std.fs.File, write_offset: u64, read_offset: u64, block: BlockSize, cur_block_size: u64, wg: *WaitGroup, out_err: *?anyerror) void {
-    defer wg.finish();
+fn workThreadBlocks(
+    self: ThreadedDataReader,
+    fil: std.fs.File,
+    write_offset: u64,
+    read_offset: u64,
+    block: BlockSize,
+    cur_block_size: u64,
+    finish: *InodeFinish,
+) void {
+    defer finish.finish();
     var wrt = fil.writer(&[0]u8{});
     wrt.seekTo(write_offset) catch |err| {
-        out_err.* = err;
+        finish.logError("Error seeking file writer: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer wrt.interface.flush() catch |err| {
-        out_err.* = err;
+        finish.logError("Error flushing file writer: {}", .{err});
+        finish.out_err.* = err;
     };
     if (block.size == 0) {
         wrt.interface.splatByteAll(0, cur_block_size) catch |err| {
-            out_err.* = err;
+            finish.logError("Error writing zeroes: {}", .{err});
+            finish.out_err.* = err;
             return;
         };
         return;
     }
     var rdr = self.fil.readerAt(read_offset, &[0]u8{}) catch |err| {
-        out_err.* = err;
+        finish.logError("Error creating file reader: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     if (block.uncompressed) {
         rdr.interface.streamExact(&wrt.interface, block.size) catch |err| {
-            out_err.* = err;
+            finish.logError("Error streaming data: {}", .{err});
+            finish.out_err.* = err;
             return;
         };
         return;
     }
     // TODO: shared buffers
     const read_buf = self.alloc.alloc(u8, block.size) catch |err| {
-        out_err.* = err;
+        finish.logError("Error creating reader buffer: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer self.alloc.free(read_buf);
     rdr.interface.readSliceAll(read_buf) catch |err| {
-        out_err.* = err;
+        finish.logError("Error reading data into reader buffer: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     // TODO: shared buffers
     const res_buf = self.alloc.alloc(u8, cur_block_size) catch |err| {
-        out_err.* = err;
+        finish.logError("Error creating result buffer: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer self.alloc.free(res_buf);
     _ = self.decomp(self.alloc, read_buf, res_buf) catch |err| {
-        out_err.* = err;
+        finish.logError("Error decompressing data block: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     wrt.interface.writeAll(res_buf) catch |err| {
-        out_err.* = err;
+        finish.logError("Error writing to file: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
 }
-fn workThreadFragment(self: ThreadedDataReader, fil: std.fs.File, write_offset: u64, wg: *WaitGroup, out_err: *?anyerror) void {
-    defer wg.finish();
+fn workThreadFragment(self: ThreadedDataReader, fil: std.fs.File, write_offset: u64, finish: *InodeFinish) void {
+    defer finish.finish();
 
     var wrt = fil.writer(&[0]u8{});
     wrt.seekTo(write_offset) catch |err| {
-        out_err.* = err;
+        finish.logError("Error seeking file writer for file fragment: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer wrt.interface.flush() catch |err| {
-        out_err.* = err;
+        finish.out_err.* = err;
     };
 
     var rdr = self.fil.readerAt(self.frag.?.start, &[0]u8{}) catch |err| {
-        out_err.* = err;
+        finish.logError("Error creating file reader for file fragment: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     if (self.frag.?.size.uncompressed) {
         rdr.interface.discardAll(self.frag_offset) catch |err| {
-            out_err.* = err;
+            finish.logError("Error discarding useless fragment data: {}", .{err});
+            finish.out_err.* = err;
             return;
         };
         rdr.interface.streamExact(&wrt.interface, self.size % self.block_size) catch |err| {
-            out_err.* = err;
+            finish.logError("Error streaming fragment data: {}", .{err});
+            finish.out_err.* = err;
             return;
         };
         return;
     }
     const tmp_buf = self.alloc.alloc(u8, self.frag.?.size.size) catch |err| {
-        out_err.* = err;
+        finish.logError("Error creating a temporary buffer for a file fragment: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer self.alloc.free(tmp_buf);
     rdr.interface.readSliceAll(tmp_buf) catch |err| {
-        out_err.* = err;
+        finish.logError("Error reading data into fragment buffer: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     const needed_block = self.alloc.alloc(u8, self.block_size) catch |err| {
-        out_err.* = err;
+        finish.logError("Error allocating fragment decompression results: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     defer self.alloc.free(needed_block);
     _ = self.decomp(self.alloc, tmp_buf, needed_block) catch |err| {
-        out_err.* = err;
+        finish.logError("Error decompressing fragment: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
     wrt.interface.writeAll(needed_block[self.frag_offset .. self.frag_offset + (self.size % self.block_size)]) catch |err| {
-        out_err.* = err;
+        finish.logError("Error writing fragment: {}", .{err});
+        finish.out_err.* = err;
         return;
     };
 }

src/util/extract.zig

@@ -0,0 +1,353 @@
+const std = @import("std");
+const Allocator = std.mem.Allocator;
+const Pool = std.Thread.Pool;
+const WaitGroup = std.Thread.WaitGroup;
+
+const Archive = @import("../archive.zig");
+const DirEntry = @import("../dir_entry.zig");
+const Inode = @import("../inode.zig");
+const ExtractionOptions = @import("../options.zig");
+const Tables = @import("../tables.zig");
+const InodeFinish = @import("inode_finish.zig");
+const FinishUnion = InodeFinish.FinishUnion;
+const ThreadedDataReader = @import("data_threaded.zig");
+
+// 1 MB
+const STACK_ALLOC_SIZE = 1024 * 1024;
+
+pub fn extractTo(
+    allocator: Allocator,
+    inode: Inode,
+    archive: Archive,
+    path: []const u8,
+    options: ExtractionOptions,
+) !void {
+    if (path[path.len - 1] == '/')
+        return extractTo(allocator, inode, archive, path[0 .. path.len - 2], options);
+
+    var stack_alloc = std.heap.stackFallback(STACK_ALLOC_SIZE, allocator);
+    var arena: std.heap.ArenaAllocator = .init(stack_alloc.get());
+    defer arena.deinit();
+    if (options.threads <= 1) {
+        const alloc = arena.allocator();
+        var tables: Tables = try .init(alloc, archive);
+        return extractSingleThread(arena.allocator(), inode, archive, &tables, path, options);
+    }
+
+    var thread_alloc = std.heap.ThreadSafeAllocator{ .child_allocator = arena.allocator() };
+    const alloc = thread_alloc.allocator();
+    var tables: Tables = try .init(alloc, archive);
+
+    var pool_alloc = std.heap.stackFallback(10 * 1024, alloc);
+    var pool: Pool = undefined;
+    try pool.init(.{ .allocator = pool_alloc.get(), .n_jobs = options.threads - 1 });
+
+    var wg: WaitGroup = .{};
+    var err: ?anyerror = null;
+    wg.start();
+    try pool.spawn(extractMultiThread, .{
+        alloc,
+        inode,
+        archive,
+        &tables,
+        path,
+        options,
+        &pool,
+        FinishUnion{ .wg = &wg },
+        &err,
+    });
+    pool.waitAndWork(&wg);
+    if (err != null) return err.?;
+}
+
+fn extractSingleThread(
+    alloc: Allocator,
+    inode: Inode,
+    archive: Archive,
+    tables: *Tables,
+    path: []const u8,
+    options: ExtractionOptions,
+) !void {
+    switch (inode.hdr.inode_type) {
+        .dir, .ext_dir => {
+            _ = std.fs.cwd().makeDir(path) catch |err| switch (err) {
+                std.fs.Dir.MakeError.PathAlreadyExists => {},
+                else => return err,
+            };
+
+            // Currently we are ignoring any deinit or free calls since we know we are under an ArenaAllocator.
+            // Possibly in the future, do some simple math to see if it would be safe to ONLY deinit via Arena,
+            // otherwise be more conscientious about freeing memory.
+            // For now, this is good enough.
+
+            const entries = try inode.dirEntries(alloc, archive);
+            for (entries) |ent| {
+                const sub_inode: Inode = try .readFromEntry(alloc, archive, ent);
+                const new_path = try std.mem.concat(alloc, u8, &[_][]const u8{ path, "/", ent.name });
+                try extractSingleThread(alloc, sub_inode, archive, tables, new_path, options);
+            }
+
+            const fil = try std.fs.cwd().openFile(path, .{});
+            defer fil.close();
+            try inode.setMetadata(alloc, tables, fil, options);
+        },
+        .file, .ext_file => {
+            var fil = try std.fs.cwd().createFile(path, .{ .exclusive = true });
+            defer fil.close();
+            var wrt = fil.writer(&[0]u8{});
+            var dat_rdr = try inode.dataReader(alloc, archive, tables);
+            defer dat_rdr.deinit();
+            _ = try dat_rdr.interface.streamRemaining(&wrt.interface);
+            try wrt.interface.flush();
+
+            try inode.setMetadata(alloc, tables, fil, options);
+        },
+        .symlink, .ext_symlink => return extractSymlink(inode, path),
+        else => return extractDeviceAndIPC(inode, alloc, tables, path, options),
+    }
+}
+
+fn extractMultiThread(
+    alloc: Allocator,
+    inode: Inode,
+    archive: Archive,
+    tables: *Tables,
+    path: []const u8,
+    options: ExtractionOptions,
+    pool: *Pool,
+    fin: FinishUnion,
+    err: *?anyerror,
+) void {
+    if (err.* != null) {
+        fin.finish();
+        return;
+    }
+    switch (inode.hdr.inode_type) {
+        .dir, .ext_dir => {
+            _ = std.fs.cwd().makeDir(path) catch |res_err| switch (res_err) {
+                std.fs.Dir.MakeError.PathAlreadyExists => {},
+                else => {
+                    err.* = res_err;
+                    fin.finish();
+                    return;
+                },
+            };
+
+            // Currently we are ignoring any deinit or free calls since we know we are under an ArenaAllocator.
+            // Possibly in the future, do some simple math to see if it would be safe to ONLY deinit via Arena,
+            // otherwise be more conscientious about freeing memory.
+            // For now, this is good enough.
+
+            const entries = inode.dirEntries(alloc, archive) catch |res_err| {
+                err.* = res_err;
+                fin.finish();
+                return;
+            };
+
+            if (entries.len == 0) {
+                fin.finish();
+                return;
+            }
+
+            var dir_fin = InodeFinish.create(
+                alloc,
+                inode,
+                path,
+                tables,
+                options,
+                fin,
+                err,
+                null,
+                entries.len,
+            ) catch |res_err| {
+                err.* = res_err;
+                fin.finish();
+                return;
+            };
+
+            for (entries) |ent| {
+                if (ent.inode_type == .dir) {
+                    extractEntry(
+                        alloc,
+                        ent,
+                        archive,
+                        tables,
+                        path,
+                        options,
+                        pool,
+                        .{ .fin = dir_fin },
+                        err,
+                    );
+                    continue;
+                }
+
+                pool.spawn(
+                    extractEntry,
+                    .{ alloc, ent, archive, tables, path, options, pool, FinishUnion{ .fin = dir_fin }, err },
+                ) catch |res_err| {
+                    err.* = res_err;
+                    dir_fin.finish();
+                    return;
+                };
+            }
+        },
+        .file, .ext_file => {
+            const fil = std.fs.cwd().createFile(path, .{ .exclusive = true }) catch |res_err| {
+                if (options.verbose)
+                    options.verbose_writer.?.print("Can't create file at {s}: {}\n", .{ path, res_err }) catch {};
+                err.* = res_err;
+                fin.finish();
+                return;
+            };
+
+            var data_rdr = threadedDataReader(inode, alloc, archive, tables) catch |res_err| {
+                if (options.verbose)
+                    options.verbose_writer.?.print("Can't create data reader for inode #{} (extracting to {s}): {}\n", .{ inode.hdr.num, path, res_err }) catch {};
+                err.* = res_err;
+                fin.finish();
+                return;
+            };
+            if (data_rdr == null) {
+                inode.setMetadata(alloc, tables, fil, options) catch |res_err| {
+                    if (options.verbose)
+                        options.verbose_writer.?.print("Can't set metadata to {s}: {}\n", .{ path, res_err }) catch {};
+                    err.* = res_err;
+                };
+                fin.finish();
+                return;
+            }
+            const file_fin = InodeFinish.create(
+                alloc,
+                inode,
+                path,
+                tables,
+                options,
+                fin,
+                err,
+                fil,
+                data_rdr.?.num_blocks,
+            ) catch |res_err| {
+                if (options.verbose)
+                    options.verbose_writer.?.print("Can't create callback for inode #{} (extracting to {s}): {}\n", .{ inode.hdr.num, path, res_err }) catch {};
+                err.* = res_err;
+                fin.finish();
+                return;
+            };
+
+            data_rdr.?.extractThreaded(fil, pool, file_fin);
+        },
+        .symlink, .ext_symlink => {
+            extractSymlink(inode, path) catch |res_err| {
+                err.* = res_err;
+            };
+            fin.finish();
+        },
+        else => {
+            extractDeviceAndIPC(inode, alloc, tables, path, options) catch |res_err| {
+                err.* = res_err;
+            };
+            fin.finish();
+        },
+    }
+}
+
+fn extractEntry(
+    alloc: Allocator,
+    ent: DirEntry,
+    archive: Archive,
+    tables: *Tables,
+    path: []const u8,
+    options: ExtractionOptions,
+    pool: *Pool,
+    fin: FinishUnion,
+    err: *?anyerror,
+) void {
+    const new_path = std.mem.concat(alloc, u8, &[_][]const u8{ path, "/", ent.name }) catch |res_err| {
+        err.* = res_err;
+        fin.finish();
+        return;
+    };
+
+    const inode = Inode.readFromEntry(alloc, archive, ent) catch |res_err| {
+        err.* = res_err;
+        fin.finish();
+        return;
+    };
+    extractMultiThread(alloc, inode, archive, tables, new_path, options, pool, fin, err);
+}
+
+/// Get a threaded data reader for a file inode.
+fn threadedDataReader(self: Inode, alloc: std.mem.Allocator, archive: Archive, tables: *Tables) !?ThreadedDataReader {
+    return switch (self.hdr.inode_type) {
+        .file => threadedReaderFromData(alloc, archive, tables, self.data.file),
+        .ext_file => threadedReaderFromData(alloc, archive, tables, self.data.ext_file),
+        else => error.NotRegularFile,
+    };
+}
+fn threadedReaderFromData(alloc: std.mem.Allocator, archive: Archive, tables: *Tables, data: anytype) !?ThreadedDataReader {
+    if (data.block_sizes.len == 0 and data.frag_idx == 0xFFFFFFFF) return null;
+    var out: ThreadedDataReader = .init(alloc, archive, data.block_sizes, data.block_start, data.size);
+    if (data.frag_idx != 0xFFFFFFFF)
+        out.addFragment(try tables.frag_table.get(data.frag_idx), data.frag_block_offset);
+    return out;
+}
+
+/// Creates the symlink described by the inode.
+/// Sets metadata.
+fn extractSymlink(self: Inode, path: []const u8) !void {
+    const target = switch (self.data) {
+        .symlink => |s| s.target,
+        .ext_symlink => |s| s.target,
+        else => unreachable,
+    };
+    try std.fs.cwd().symLink(target, path, .{});
+}
+/// Creates the device described by the inode.
+/// Sets metadata.
+fn extractDeviceAndIPC(self: Inode, alloc: std.mem.Allocator, tables: *Tables, path: []const u8, options: ExtractionOptions) !void {
+    var mode: u32 = undefined;
+    var dev: u32 = 0;
+    switch (self.data) {
+        .char_dev => |d| {
+            mode = std.posix.S.IFCHR;
+            dev = d.dev;
+        },
+        .ext_char_dev => |d| {
+            mode = std.posix.S.IFCHR;
+            dev = d.dev;
+        },
+        .block_dev => |d| {
+            mode = std.posix.S.IFBLK;
+            dev = d.dev;
+        },
+        .ext_block_dev => |d| {
+            mode = std.posix.S.IFBLK;
+            dev = d.dev;
+        },
+        .fifo, .ext_fifo => mode = std.posix.S.IFIFO,
+        .socket, .ext_socket => mode = std.posix.S.IFSOCK,
+        else => unreachable,
+    }
+    const res: std.os.linux.E = @enumFromInt(std.os.linux.mknod(@ptrCast(path), mode, dev));
+    switch (res) {
+        .SUCCESS => {},
+        .ACCES => return std.fs.Dir.MakeError.AccessDenied,
+        .DQUOT => return std.fs.Dir.MakeError.DiskQuota,
+        .EXIST => return std.fs.Dir.MakeError.PathAlreadyExists,
+        .FAULT, .NOENT => return std.fs.Dir.MakeError.BadPathName,
+        .LOOP => return std.fs.Dir.MakeError.SymLinkLoop,
+        .NAMETOOLONG => return std.fs.Dir.MakeError.NameTooLong,
+        .NOMEM => return std.fs.Dir.MakeError.SystemResources,
+        .NOSPC => return std.fs.Dir.MakeError.NoSpaceLeft,
+        .NOTDIR => return std.fs.Dir.MakeError.NotDir,
+        .PERM => return std.fs.Dir.MakeError.PermissionDenied,
+        .ROFS => return std.fs.Dir.MakeError.ReadOnlyFileSystem,
+        else => return blk: {
+            std.debug.print("unhandled mknod result: {}\n", .{res});
+            break :blk std.fs.Dir.MakeError.Unexpected;
+        },
+    }
+    var fil = try std.fs.cwd().openFile(path, .{});
+    defer fil.close();
+    try self.setMetadata(alloc, tables, fil, options);
+}

src/util/inode_finish.zig

@@ -0,0 +1,101 @@
+const std = @import("std");
+const WaitGroup = std.Thread.WaitGroup;
+const Mutex = std.Thread.Mutex;
+
+const Archive = @import("../archive.zig");
+const Inode = @import("../inode.zig");
+const ExtractionOptions = @import("../options.zig");
+const Tables = @import("../tables.zig");
+
+const InodeFinish = @This();
+
+const FinishEnum = enum {
+    wg,
+    fin,
+};
+pub const FinishUnion = union(FinishEnum) {
+    wg: *WaitGroup,
+    fin: *InodeFinish,
+
+    pub fn finish(self: FinishUnion) void {
+        switch (self) {
+            .wg => |wg| wg.finish(),
+            .fin => |fin| fin.finish(),
+        }
+    }
+};
+
+alloc: std.mem.Allocator,
+
+inode: Inode,
+path: []const u8,
+tables: *Tables,
+options: ExtractionOptions,
+parent_finish: FinishUnion,
+fil: ?std.fs.File,
+out_err: *?anyerror,
+
+wg: WaitGroup = .{},
+mut: Mutex = .{},
+
+pub fn create(
+    alloc: std.mem.Allocator,
+    inode: Inode,
+    path: []const u8,
+    tables: *Tables,
+    options: ExtractionOptions,
+    parent_finish: FinishUnion,
+    out_err: *?anyerror,
+    fil: ?std.fs.File,
+    work_size: usize,
+) !*InodeFinish {
+    if (work_size == 0)
+        return error.InvalidWorkSize;
+    const out = try alloc.create(InodeFinish);
+    errdefer alloc.destroy(out);
+    out.* = .{
+        .alloc = alloc,
+
+        .inode = inode,
+        .path = path,
+        .tables = tables,
+        .options = options,
+        .parent_finish = parent_finish,
+        .out_err = out_err,
+        .fil = fil,
+    };
+    out.wg.startMany(work_size);
+    return out;
+}
+
+pub fn logError(self: *InodeFinish, comptime fmt: []const u8, args: anytype) void {
+    if (self.options.verbose)
+        self.options.verbose_writer.?.print(fmt, args) catch {};
+}
+
+pub fn finish(self: *InodeFinish) void {
+    self.mut.lock();
+    {
+        defer self.mut.unlock();
+        self.wg.finish();
+        if (!self.wg.isDone()) return;
+    }
+    defer {
+        self.parent_finish.finish();
+        self.alloc.destroy(self);
+    }
+    if (self.fil == null)
+        self.fil = std.fs.cwd().openFile(self.path, .{}) catch |err| {
+            if (self.options.verbose)
+                self.options.verbose_writer.?.print("Error opening {s} to set metadata: {}\n", .{ self.path, err }) catch {};
+            self.out_err.* = err;
+            return;
+        };
+    defer self.fil.?.close();
+    self.inode.setMetadata(self.alloc, self.tables, self.fil.?, self.options) catch |err| {
+        if (self.options.verbose)
+            self.options.verbose_writer.?.print("Error setting metadata to {s}: {}\n", .{ self.path, err }) catch {};
+        self.out_err.* = err;
+        return;
+    };
+}

src/xattr.zig

@@ -1,7 +1,7 @@
 const std = @import("std");
 
 const DecompFn = @import("decomp.zig").DecompFn;
-const Table = @import("table.zig").Table;
+const Table = @import("tables.zig").Table;
 const MetadataReader = @import("util/metadata.zig");
 const OffsetFile = @import("util/offset_file.zig");
 
@@ -30,7 +30,7 @@ const KeyRaw = packed struct {
 };
 
 pub const KeyValue = struct {
-    key: []u8,
+    key: [:0]u8,
     value: []u8,
 };
 
@@ -62,7 +62,7 @@ pub fn init(alloc: std.mem.Allocator, fil: OffsetFile, decomp: DecompFn, table_s
         .table = try .init(alloc, fil, decomp, table_start + 16, info.count),
     };
 }
-pub fn deinit(self: XattrTable) void {
+pub fn deinit(self: *XattrTable) void {
     self.table.deinit();
 }
 
@@ -80,19 +80,22 @@ pub fn get(self: *XattrTable, alloc: std.mem.Allocator, idx: u32) ![]KeyValue {
 
         switch (key_raw.type.prefix) {
             .user => {
-                kv.key = try alloc.alloc(u8, key_raw.name_size + 5);
+                kv.key = @ptrCast(try alloc.alloc(u8, key_raw.name_size + 5 + 1));
                 @memcpy(kv.key[0..5], "user.");
-                try meta.interface.readSliceAll(kv.key[5..]);
+                try meta.interface.readSliceAll(kv.key[5 .. kv.key.len - 1]);
+                kv.key[kv.key.len - 1] = 0;
             },
             .security => {
-                kv.key = try alloc.alloc(u8, key_raw.name_size + 9);
+                kv.key = @ptrCast(try alloc.alloc(u8, key_raw.name_size + 9 + 1));
                 @memcpy(kv.key[0..9], "security.");
-                try meta.interface.readSliceAll(kv.key[9..]);
+                try meta.interface.readSliceAll(kv.key[9 .. kv.key.len - 1]);
+                kv.key[kv.key.len - 1] = 0;
             },
             .trusted => {
-                kv.key = try alloc.alloc(u8, key_raw.name_size + 8);
+                kv.key = @ptrCast(try alloc.alloc(u8, key_raw.name_size + 8 + 1));
                 @memcpy(kv.key[0..8], "trusted.");
-                try meta.interface.readSliceAll(kv.key[8..]);
+                try meta.interface.readSliceAll(kv.key[8 .. kv.key.len - 1]);
+                kv.key[kv.key.len - 1] = 0;
             },
         }
         if (key_raw.type.out_of_line) {