kernel/ext2: Implement directory traversal

kernel/debug.cmake

@@ -7,5 +7,6 @@ target_compile_definitions(moon PRIVATE EXEC_DEBUG)
 target_compile_definitions(moon PRIVATE OPEN_DEBUG)
 target_compile_definitions(moon PRIVATE REAP_DEBUG)
 target_compile_definitions(moon PRIVATE PCI_DEBUG)
+target_compile_definitions(moon PRIVATE EXT2_DEBUG)
 target_compile_definitions(moon PRIVATE DEVICE_REGISTRY_DEBUG)
 target_compile_options(moon PRIVATE -fsanitize=undefined)

kernel/src/fs/ext2/FileSystem.cpp

@@ -25,7 +25,7 @@ namespace Ext2
     {
     }
 
-    Result<SharedPtr<VFS::Inode>> FileSystem::find_inode_by_number(ino_t inum)
+    Result<SharedPtr<VFS::Inode>> FileSystem::find_inode_by_number(ino_t inum, bool initialize_dir_now)
     {
         check(inum < m_superblock.nr_inodes);
 
@@ -37,23 +37,29 @@ namespace Ext2
         const auto* block_group_descriptor = TRY(find_block_group_descriptor(block_group));
         check(block_group_descriptor);
 
-        // FIXME: This is determined by a field in the Superblock if the Ext2 revision >= 1.0.
+        // FIXME: Even if the inode size is bigger (Ext2::FileSystem::m_inode_size), we only read this amount. Enlarge
+        // the Inode structure to fit this case.
         static constexpr usize INODE_SIZE = 128;
 
         const u64 index = (inum - 1) % m_superblock.inodes_per_block_group;
 
-        const u64 inode_address = (block_group_descriptor->inode_table_start * m_block_size) + (index * INODE_SIZE);
+        const u64 inode_address = (block_group_descriptor->inode_table_start * m_block_size) + (index * m_inode_size);
 
         auto inode = TRY(adopt_shared_if_nonnull(new (std::nothrow) Ext2::Inode({}, this)));
         TRY(m_host_device->read((u8*)&inode->m_raw_inode, inode_address, INODE_SIZE));
         inode->m_type = vfs_type_from_ext2_type(inode->m_raw_inode.mode);
         inode->m_inum = inum;
 
+#ifdef EXT2_DEBUG
         kdbgln("ext2: Read inode %lu with mode %#x (%#x + %#o), size %lu", inum, inode->m_raw_inode.mode,
                inode->m_raw_inode.mode & 0xf000, inode->mode(), inode->size());
+#endif
 
-        // TODO: Locate the inode's block group descriptor and find it in the block group's inode table.
-        return err(ENOENT);
+        m_inode_cache.try_set(inum, inode);
+
+        if (initialize_dir_now && inode->m_type == VFS::InodeType::Directory) TRY(inode->lazy_initialize_dir());
+
+        return (SharedPtr<VFS::Inode>)inode;
     }
 
     Result<const BlockGroupDescriptor*> FileSystem::find_block_group_descriptor(u32 index)
@@ -80,20 +86,43 @@ namespace Ext2
         if (nread != 1024) return err(EINVAL);                            // Source had an invalid superblock.
         if (fs->m_superblock.signature != EXT2_MAGIC) return err(EINVAL); // Source had an invalid superblock.
 
+        if (fs->m_superblock.major_version >= 1)
+        {
+            auto required = fs->m_superblock.ext_superblock.required_features;
+            if (required & EXT2_REQUIRED_COMPAT_D_TYPE) fs->m_dirs_have_type_field = true;
+
+            required &= ~EXT2_REQUIRED_COMPAT_D_TYPE;
+
+            if (required > 0)
+            {
+                kwarnln("ext2: File system has required features not supported by the implementation, cannot mount");
+                return err(EINVAL);
+            }
+
+            fs->m_uses_extended_size = true;
+            fs->m_inode_size = fs->m_superblock.ext_superblock.inode_size;
+        }
+
         fs->m_host_device = host_device;
 
         fs->m_block_size = 1024 << fs->m_superblock.log_block_size;
         fs->m_block_groups = get_blocks_from_size(fs->m_superblock.nr_blocks, fs->m_superblock.blocks_per_block_group);
 
+#ifdef EXT2_DEBUG
         kdbgln("ext2: Mounting new Ext2 file system, block size=%lu, blocks=%u, inodes=%u, block group=(%u blocks, %u "
                "inodes), %lu block groups",
                fs->m_block_size, fs->m_superblock.nr_blocks, fs->m_superblock.nr_inodes,
                fs->m_superblock.blocks_per_block_group, fs->m_superblock.inodes_per_block_group, fs->m_block_groups);
+#endif
 
         // Lookup the root inode.
-        fs->find_inode_by_number(2);
+        fs->m_root_inode = TRY(fs->find_inode_by_number(2, true));
 
-        // TODO: Implement basic Ext2 reading, enough to be able to mount a volume.
-        return err(ENOTSUP);
+        return (SharedPtr<VFS::FileSystem>)fs;
+    }
+
+    Result<void> FileSystem::set_mount_dir(SharedPtr<VFS::Inode> inode)
+    {
+        return m_root_inode->replace_entry(inode, "..");
     }
 }

kernel/src/fs/ext2/FileSystem.h

@@ -5,6 +5,8 @@
 
 #define EXT2_MAGIC 0xef53
 
+#define EXT2_REQUIRED_COMPAT_D_TYPE 0x0002
+
 namespace Ext2
 {
     struct [[gnu::packed]] Superblock
@@ -37,8 +39,27 @@ namespace Ext2
 
         u16 reserved_block_uid;
         u16 reserved_block_gid;
-        // TODO: Add extended superblock fields.
-        u8 padding[1024 - 84];
+        struct
+        {
+            u32 first_non_reserved_inode;
+            u16 inode_size;
+            u16 this_block_group;
+            u32 optional_features;
+            u32 required_features;
+            u32 ro_features;
+            u8 fsid[16];
+            u8 name[16];
+            u8 last_mountpoint[64];
+            u32 compression_algs;
+            u8 nr_preallocated_blocks_for_files;
+            u8 nr_preallocated_blocks_for_dirs;
+            u16 unused;
+            u8 journal_id[16];
+            u32 journal_inode;
+            u32 journal_device;
+            u32 orphan_inode_head;
+        } ext_superblock;
+        u8 padding[1024 - 236];
     };
 
     struct [[gnu::packed]] BlockGroupDescriptor
@@ -77,6 +98,21 @@ namespace Ext2
         u32 os_specific_2[3];
     };
 
+    struct [[gnu::packed]] RawDirectoryEntry
+    {
+        u32 inum;
+        u16 size;
+        union {
+            u16 name_length;
+            struct
+            {
+                u8 name_length_low;
+                u8 type;
+            };
+        };
+        char name[4]; // Names should be padded to a multiple of 4 bytes.
+    };
+
     static_assert(sizeof(Superblock) == 1024);
     static_assert(sizeof(BlockGroupDescriptor) == 32);
     static_assert(sizeof(RawInode) == 128);
@@ -111,10 +147,7 @@ namespace Ext2
             return err(EROFS);
         }
 
-        Result<void> set_mount_dir(SharedPtr<VFS::Inode>) override
-        {
-            return {};
-        }
+        Result<void> set_mount_dir(SharedPtr<VFS::Inode>) override;
 
         bool is_readonly() const override
         {
@@ -128,7 +161,7 @@ namespace Ext2
             return m_host_device_id;
         }
 
-        Result<SharedPtr<VFS::Inode>> find_inode_by_number(ino_t inode);
+        Result<SharedPtr<VFS::Inode>> find_inode_by_number(ino_t inode, bool initialize_dir_now = false);
         Result<const BlockGroupDescriptor*> find_block_group_descriptor(u32 index);
 
         virtual ~FileSystem() = default;
@@ -146,6 +179,11 @@ namespace Ext2
         u64 m_block_size;
         u64 m_block_groups;
 
+        u32 m_inode_size { 128 };
+
+        bool m_dirs_have_type_field { false };
+        bool m_uses_extended_size { false };
+
         // FIXME: This inode cache will keep all inodes in it alive despite having no other references to it, but we're
         // not worrying about that as for now the filesystem implementation is read-only.
         HashMap<ino_t, SharedPtr<VFS::Inode>> m_inode_cache;

kernel/src/fs/ext2/Inode.cpp

@@ -1,4 +1,5 @@
 #include "fs/ext2/Inode.h"
+#include <luna/String.h>
 
 namespace Ext2
 {
@@ -6,6 +7,13 @@ namespace Ext2
     {
     }
 
+    usize Inode::size() const
+    {
+        return (m_fs->m_uses_extended_size && (m_type == VFS::InodeType::RegularFile))
+                   ? ((u64)m_raw_inode.size_high << 32) | (u64)m_raw_inode.size_low
+                   : m_raw_inode.size_low;
+    }
+
     Result<usize> Inode::read(u8* buf, usize offset, usize length) const
     {
         if (length == 0) return 0;
@@ -22,6 +30,7 @@ namespace Ext2
             usize block_offset = (offset % block_size);
             usize block = find_block(offset / block_size);
             usize size_to_read = block_size - block_offset;
+            if (size_to_read > to_read) size_to_read = to_read;
 
             usize host_offset = (block * block_size) + block_offset;
 
@@ -66,4 +75,91 @@ namespace Ext2
 
         return m_raw_inode.direct_pointers[index];
     }
+
+    Result<void> Inode::lazy_initialize_dir() const
+    {
+        check(m_type == VFS::InodeType::Directory);
+
+        const usize inode_size = size();
+        const usize block_size = m_fs->m_block_size;
+
+        u8* const buf = TRY(make_array<u8>(block_size));
+        auto guard = make_scope_guard([buf] { delete[] buf; });
+
+        m_entries.clear();
+
+        for (usize offset = 0; offset < inode_size; offset += block_size)
+        {
+            TRY(read(buf, offset, block_size));
+
+            usize dir_offset = 0;
+            while (dir_offset < block_size)
+            {
+                auto& entry = *(Ext2::RawDirectoryEntry*)&buf[dir_offset];
+
+                if (entry.inum != 0)
+                {
+                    auto inode = TRY(m_fs->find_inode_by_number(entry.inum));
+
+                    VFS::DirectoryEntry vfs_entry { inode, "" };
+                    vfs_entry.name.adopt(entry.name,
+                                         m_fs->m_dirs_have_type_field ? entry.name_length_low : entry.name_length);
+
+#ifdef EXT2_DEBUG
+                    kdbgln("ext2: Read new directory entry: inum=%u, name=%s, namelen=%lu", entry.inum,
+                           vfs_entry.name.chars(), vfs_entry.name.length());
+#endif
+
+                    TRY(m_entries.try_append(move(vfs_entry)));
+                }
+
+                dir_offset += entry.size;
+            }
+        }
+
+        m_dir_already_lazily_initialized = true;
+
+        return {};
+    }
+
+    Result<void> Inode::replace_entry(SharedPtr<VFS::Inode> inode, const char* name)
+    {
+        if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+        if (!m_dir_already_lazily_initialized) TRY(lazy_initialize_dir());
+
+        for (auto& entry : m_entries)
+        {
+            if (!strcmp(name, entry.name.chars()))
+            {
+                entry.inode = inode;
+                return {};
+            }
+        }
+
+        return err(ENOENT);
+    }
+
+    Result<SharedPtr<VFS::Inode>> Inode::find(const char* name) const
+    {
+        if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+        if (!m_dir_already_lazily_initialized) TRY(lazy_initialize_dir());
+
+        for (const auto& entry : m_entries)
+        {
+            if (!strcmp(name, entry.name.chars())) return entry.inode;
+        }
+
+        return err(ENOENT);
+    }
+
+    Option<VFS::DirectoryEntry> Inode::get(usize index) const
+    {
+        if (m_type != VFS::InodeType::Directory) return {};
+        if (!m_dir_already_lazily_initialized)
+            if (lazy_initialize_dir().has_error()) return {};
+
+        if (index >= m_entries.size()) return {};
+
+        return m_entries[index];
+    }
 }

kernel/src/fs/ext2/Inode.h

@@ -11,7 +11,7 @@
 
 namespace Ext2
 {
-    class Inode : public VFS::FileInode
+    class Inode : public VFS::Inode
     {
       public:
         VFS::InodeType type() const override
@@ -19,11 +19,7 @@ namespace Ext2
             return m_type;
         }
 
-        usize size() const override
-        {
-            // FIXME: If EXT2_REVISION >= 1.0, use size_high as well.
-            return m_raw_inode.size_low;
-        }
+        usize size() const override;
 
         mode_t mode() const override
         {
@@ -85,6 +81,54 @@ namespace Ext2
             return err(EROFS);
         }
 
+        Result<SharedPtr<VFS::Inode>> find(const char*) const override;
+
+        Option<VFS::DirectoryEntry> get(usize) const override;
+
+        Result<SharedPtr<VFS::Inode>> create_file(const char*) override
+        {
+            if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+
+            return err(EROFS);
+        }
+
+        Result<SharedPtr<VFS::Inode>> create_subdirectory(const char*) override
+        {
+            if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+
+            return err(EROFS);
+        }
+
+        Result<void> add_entry(SharedPtr<VFS::Inode>, const char*) override
+        {
+            if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+
+            return err(EROFS);
+        }
+
+        Result<void> replace_entry(SharedPtr<VFS::Inode>, const char*) override;
+
+        Result<void> remove_entry(const char*) override
+        {
+            if (m_type != VFS::InodeType::Directory) return err(ENOTDIR);
+
+            return err(EROFS);
+        }
+
+        usize entries() const override
+        {
+            return m_entries.size();
+        }
+
+        bool blocking() const override
+        {
+            return false;
+        }
+
+        Result<void> lazy_initialize_dir() const;
+
+        // FIXME: Implement readlink() and device numbers.
+
         Inode(Badge<FileSystem>, FileSystem* fs);
         virtual ~Inode() = default;
 
@@ -94,6 +138,9 @@ namespace Ext2
         FileSystem* m_fs;
         ino_t m_inum;
 
+        mutable Vector<VFS::DirectoryEntry> m_entries;
+        mutable bool m_dir_already_lazily_initialized { false };
+
         usize find_block(usize index) const;
 
         friend class FileSystem;

libluna/include/luna/StaticString.h

@@ -25,6 +25,16 @@ template <usize Size> class StaticString
             m_length = length;
     }
 
+    void adopt(const char* string, usize length)
+    {
+        if (length > Size) length = Size;
+
+        memcpy(m_buffer, string, length);
+
+        m_buffer[length] = 0;
+        m_length = length;
+    }
+
     void adopt(StringView string)
     {
         usize length = strlcpy(m_buffer, string.chars(),