diff --git a/build.sh b/build.sh
index 2548d40..c5a3da7 100755
--- a/build.sh
+++ b/build.sh
@@ -21,3 +21,4 @@ mcopy -i build/os.img@@"$((partition_offset * sector_size))" build/shell ::bin/s
 mcopy -i build/os.img@@"$((partition_offset * sector_size))" build/echo ::bin/echo
 mcopy -i build/os.img@@"$((partition_offset * sector_size))" build/ls ::bin/ls
 mcopy -i build/os.img@@"$((partition_offset * sector_size))" build/cat ::bin/cat
+mcopy -i build/os.img@@"$((partition_offset * sector_size))" build/cp ::bin/cp
diff --git a/meson.build b/meson.build
index 104ea0b..ba9a79c 100644
--- a/meson.build
+++ b/meson.build
@@ -310,3 +310,42 @@ custom_target(
   command: ['objcopy', '-O', 'binary', '@INPUT@', '@OUTPUT@'],
   build_by_default: true,
 )
+
+cp_start = custom_target(
+  'cp_start',
+  input:  'src/user/start.asm',
+  output: 'cp_start.o',
+  command: [nasm, '-f', 'elf64', '-o', '@OUTPUT@', '@INPUT@'],
+)
+
+cp_syscall = custom_target(
+  'cp_syscall',
+  input:  'src/user/syscall.asm',
+  output: 'cp_syscall.o',
+  command: [nasm, '-f', 'elf64', '-o', '@OUTPUT@', '@INPUT@'],
+)
+
+cp_elf = executable(
+  'cp.elf',
+  sources: [cp_start, cp_syscall, lib_sources, 'src/user/app/cp/cp.c'],
+  c_args: [
+    '-ffreestanding',
+    '-nostdlib',
+    '-fno-stack-protector',
+    '-mcmodel=large',
+    '-DVERSION="' + meson.project_version() + '"',
+  ],
+  link_args: [
+    '-T', meson.project_source_root() / 'src/user/linker.ld',
+    '-nostdlib',
+  ],
+  link_depends: 'src/user/linker.ld',
+)
+
+custom_target(
+  'cp',
+  input:  cp_elf,
+  output: 'cp',
+  command: ['objcopy', '-O', 'binary', '@INPUT@', '@OUTPUT@'],
+  build_by_default: true,
+)
diff --git a/src/kernel/fat16.c b/src/kernel/fat16.c
index 23d816e..8dbae2b 100644
--- a/src/kernel/fat16.c
+++ b/src/kernel/fat16.c
@@ -33,15 +33,96 @@ typedef struct __attribute__((packed)) {
   uint32_t size;
 } fat16_dir_entry_t;
 
+#define MAX_INODES 256
+
+typedef struct {
+  uint32_t refs;
+  uint16_t first_cluster;
+  uint32_t size;
+  uint16_t dir_cluster;
+  uint16_t dir_index;
+  uint8_t valid;
+} fat16_inode_t;
+
+static fat16_inode_t inodes[MAX_INODES];
+
+static fat16_inode_t *inode_create(uint16_t first_cluster, uint32_t size, uint16_t dir_cluster, uint16_t dir_index) {
+  for (uint64_t i = 0; i < MAX_INODES; i++) {
+    if (!inodes[i].valid) {
+      inodes[i] = (fat16_inode_t){
+          .refs = 1,
+          .first_cluster = first_cluster,
+          .size = size,
+          .dir_cluster = dir_cluster,
+          .dir_index = dir_index,
+          .valid = 1,
+      };
+      return &inodes[i];
+    }
+  }
+  return NUL;
+}
+
+static fat16_inode_t *inode_get(uint32_t dir_cluster, uint16_t dir_index) {
+  for (uint64_t i = 0; i < MAX_INODES; i++) {
+    if (inodes[i].valid && inodes[i].dir_cluster == dir_cluster && inodes[i].dir_index == dir_index) {
+      return &inodes[i];
+    }
+  }
+  return NUL;
+}
+
+static fat16_inode_t *inode_use(uint16_t first_cluster, uint32_t size, uint16_t dir_cluster, uint16_t dir_index) {
+  fat16_inode_t *result = inode_get(dir_cluster, dir_index);
+  if (!result) {
+    result = inode_create(first_cluster, size, dir_cluster, dir_index);
+  } else {
+    result->refs++;
+  }
+  return result;
+}
+
+static void inode_free(fat16_inode_t *inode) {
+  inode->refs--;
+  if (!inode->refs) {
+    for (uint64_t i = 0; i < MAX_INODES; i++) {
+      if (&inodes[i] == inode) {
+        inodes[i].valid = 0;
+        break;
+      }
+    }
+  }
+}
+
 typedef struct {
   fs_node_t base;
-  fat16_dir_entry_t entry;
+  fat16_inode_t *inode;
 } fat16_node_t;
 
 static fat16_bpb_t bpb; // Assuming one partition.
+static uint32_t data_start;
 static fs_node_t *fs = NUL;
 static uint16_t *fat = NUL;
 
+fs_node_t *fat16_mount() { // Assuming one partition.
+  uint8_t sector[512];
+  ata_read_sectors(FIRST_PARTITION_SECTOR, 1, &sector);
+  memory_copy(sector + 11, sizeof(fat16_bpb_t), &bpb);
+  data_start = FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.sectors_per_fat * bpb.fats_count +
+               (bpb.root_entry_count * sizeof(fat16_dir_entry_t) + SECTOR_SIZE - 1) / SECTOR_SIZE;
+  fat16_node_t *node = memory_allocate(sizeof(fat16_node_t));
+  node->base.type = FAT16;
+  node->base.name[0] = '/';
+  node->base.size = sizeof(fat16_dir_entry_t) * bpb.root_entry_count;
+  node->base.is_dir = 1;
+  node->inode = inode_use(0, node->base.size, 0, UINT16_MAX);
+  if (!node->inode) {
+    memory_free(node);
+    return NUL;
+  }
+  return fs = (fs_node_t *)node;
+}
+
 static void to_8_3(const char *name, char *output) {
   memory_set(' ', 11, output);
   output[11] = '\0';
@@ -75,18 +156,6 @@ static void from_8_3(const char *name, const char *extension, char *output) {
   }
 }
 
-fs_node_t *fat16_mount() { // Assuming one partition.
-  uint8_t sector[512];
-  ata_read_sectors(FIRST_PARTITION_SECTOR, 1, &sector);
-  memory_copy(sector + 11, sizeof(fat16_bpb_t), &bpb);
-  fat16_node_t *node = memory_allocate(sizeof(fat16_node_t));
-  node->base.type = FAT16;
-  node->base.name[0] = node->entry.name[0] = '/';
-  node->base.size = node->entry.size = sizeof(fat16_dir_entry_t) * bpb.root_entry_count;
-  node->base.is_dir = 1;
-  return fs = (fs_node_t *)node;
-}
-
 static void ensure_fat() {
   ASSERT(bpb.sectors_per_fat < 256, "ensure_fat: big FAT not implemented")
   if (!fat) {
@@ -95,17 +164,32 @@ static void ensure_fat() {
   }
 }
 
-static fat16_dir_entry_t *load_directory(fat16_node_t *directory) {
+static uint16_t allocate_cluster() {
+  ensure_fat();
+  for (uint16_t i = 2; i < bpb.sectors_per_fat * SECTOR_SIZE / 2; i++) {
+    if (fat[i] == 0x0000) {
+      fat[i] = 0xFFFF;
+      return i;
+    }
+  }
+  return 0;
+}
+
+static void flush_fat() {
+  ata_write_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors, (uint8_t)bpb.sectors_per_fat, fat);
+}
+
+static fat16_dir_entry_t *read_directory(uint16_t dir_cluster) {
   fat16_dir_entry_t *entries;
 
-  if (!directory->entry.first_cluster) {
-    uint8_t sectors = (uint8_t)((directory->base.size + SECTOR_SIZE - 1) / SECTOR_SIZE);
+  if (!dir_cluster) {
+    uint8_t sectors = (uint8_t)((sizeof(fat16_dir_entry_t) * bpb.root_entry_count + SECTOR_SIZE - 1) / SECTOR_SIZE);
     entries = memory_allocate(sectors * SECTOR_SIZE + sizeof(fat16_dir_entry_t)); // One extra as null terminator.
     ata_read_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.fats_count * bpb.sectors_per_fat, sectors, entries);
   } else {
     ensure_fat();
 
-    uint16_t next_cluster = directory->entry.first_cluster;
+    uint16_t next_cluster = dir_cluster;
     uint32_t cluster_count = 0;
     while (next_cluster < 0xFFF8) {
       cluster_count++;
@@ -115,12 +199,9 @@ static fat16_dir_entry_t *load_directory(fat16_node_t *directory) {
         memory_allocate(cluster_count * bpb.sectors_per_cluster * SECTOR_SIZE + sizeof(fat16_dir_entry_t)); // One extra as null terminator.
 
     uint8_t *chunk = (uint8_t *)entries;
-    next_cluster = directory->entry.first_cluster;
+    next_cluster = dir_cluster;
     while (next_cluster < 0xFFF8) {
-      ata_read_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.sectors_per_fat * bpb.fats_count +
-                           (bpb.root_entry_count * sizeof(fat16_dir_entry_t) + SECTOR_SIZE - 1) / SECTOR_SIZE +
-                           bpb.sectors_per_cluster * (next_cluster - 2),
-                       bpb.sectors_per_cluster, chunk);
+      ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, chunk);
       chunk += bpb.sectors_per_cluster * SECTOR_SIZE;
       next_cluster = fat[next_cluster];
     }
@@ -129,50 +210,80 @@ static fat16_dir_entry_t *load_directory(fat16_node_t *directory) {
   return entries;
 }
 
-static fs_node_t *open_entry(const char *name, const fat16_dir_entry_t *entry) {
+static fs_node_t *open_entry(uint16_t dir_cluster, const fat16_dir_entry_t *entries, uint16_t index) {
+  const fat16_dir_entry_t *entry = &entries[index];
+
   if (!entry->name[0]) {
     return NUL;
   }
 
   fat16_node_t *result = memory_allocate(sizeof(fat16_node_t));
-  memory_copy((char *)name, string_length(name) + 1, &(result->base.name));
   result->base.type = FAT16;
+  from_8_3(entry->name, entry->ext, result->base.name);
+
   result->base.size = entry->size;
   result->base.is_dir = entry->attributes & ATTRIBUTE_SUBDIRECTORY;
-  memory_copy((fat16_dir_entry_t *)entry, sizeof(fat16_dir_entry_t), (uint8_t *)result + sizeof(fs_node_t));
+  result->inode = inode_use(entry->first_cluster, entry->size, dir_cluster, index);
+
+  if (!result->inode) {
+    memory_free(result);
+    return NUL;
+  }
 
   return (fs_node_t *)result;
 }
 
+static void write_directory(uint16_t dir_cluster, const fat16_dir_entry_t *entries) {
+  if (!dir_cluster) {
+    ata_write_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.fats_count * bpb.sectors_per_fat,
+                      (uint8_t)((sizeof(fat16_dir_entry_t) * bpb.root_entry_count + SECTOR_SIZE - 1) / SECTOR_SIZE), entries);
+  } else {
+    ensure_fat();
+
+    // Assuming no entries were added or removed.
+    uint8_t *chunk = (uint8_t *)entries;
+    uint16_t next_cluster = dir_cluster;
+    while (next_cluster < 0xFFF8) {
+      ata_write_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, chunk);
+      chunk += bpb.sectors_per_cluster * SECTOR_SIZE;
+      next_cluster = fat[next_cluster];
+    }
+  }
+}
+
 fs_node_t *fat16_open_by(const fs_node_t *directory, const char *name) {
-  ASSERT(directory->type == FAT16, "fat16_open_by: directory is not FAT16");
-  ASSERT(directory->is_dir, "fat16_open_by: directory is not a directory");
+  if (directory->type != FAT16 || !directory->is_dir) {
+    return NUL;
+  }
 
   char name_8_3[12];
   to_8_3(name, name_8_3);
 
-  fat16_dir_entry_t *entries = load_directory((fat16_node_t *)directory);
+  fat16_dir_entry_t *entries = read_directory(((fat16_node_t *)directory)->inode->first_cluster);
 
   fat16_dir_entry_t *entry = entries;
+  uint16_t index = 0;
   while (entry->name[0]) {
     if ((uint8_t)entry->name[0] != 0xE5 && (uint8_t)entry->attributes != 0x0F && bytes_equal(name_8_3, (char *)entry, 11)) {
       break;
     }
     entry++;
+    index++;
   }
 
-  fs_node_t *result = open_entry(name, entry);
+  fs_node_t *result = open_entry(((fat16_node_t *)directory)->inode->first_cluster, entries, index);
   memory_free(entries);
   return result;
 }
 
 fs_node_t *fat16_open_at(const fs_node_t *directory, uint64_t index) {
-  ASSERT(directory->type == FAT16, "fat16_open_at: directory is not FAT16");
-  ASSERT(directory->is_dir, "fat16_open_at: directory is not a directory");
+  if (directory->type != FAT16 || !directory->is_dir) {
+    return NUL;
+  }
 
-  fat16_dir_entry_t *entries = load_directory((fat16_node_t *)directory);
+  fat16_dir_entry_t *entries = read_directory(((fat16_node_t *)directory)->inode->first_cluster);
 
-  uint64_t ei = 0, vi = 0;
+  uint16_t ei = 0, vi = 0;
   while (entries[ei].name[0]) {
     if ((uint8_t)entries[ei].name[0] != 0xE5 && (uint8_t)entries[ei].attributes != 0x0F) {
       if (vi == index) {
@@ -183,14 +294,7 @@ fs_node_t *fat16_open_at(const fs_node_t *directory, uint64_t index) {
     ei++;
   }
 
-  if (!entries[ei].name[0]) {
-    return NUL;
-  }
-
-  char name[13];
-  from_8_3(entries[ei].name, entries[ei].ext, name);
-
-  fs_node_t *result = open_entry(name, entries + ei);
+  fs_node_t *result = open_entry(((fat16_node_t *)directory)->inode->first_cluster, entries, ei);
   memory_free(entries);
   return result;
 }
@@ -198,30 +302,228 @@ fs_node_t *fat16_open_at(const fs_node_t *directory, uint64_t index) {
 fs_node_t *fat16_open_again(const fs_node_t *source) {
   fs_node_t *result = memory_allocate(sizeof(fat16_node_t));
   memory_copy(source, sizeof(fat16_node_t), result);
+  ((fat16_node_t *)result)->inode->refs++;
   return result;
 }
 
+uint64_t fat16_read(const fs_node_t *file, uint32_t offset, uint32_t bytes, void *to) {
+  fat16_node_t *fat_file = (fat16_node_t *)file;
+
+  if (fat_file->base.type != FAT16 || fat_file->base.is_dir) {
+    return (uint64_t)-1;
+  }
+
+  if (offset >= fat_file->inode->size) {
+    return 0;
+  }
+
+  if (offset + bytes >= fat_file->inode->size) {
+    bytes = fat_file->inode->size - offset;
+  }
+
+  ensure_fat();
+
+  uint32_t cluster_size = bpb.sectors_per_cluster * SECTOR_SIZE;
+  uint32_t next_cluster = fat_file->inode->first_cluster;
+
+  while (offset >= cluster_size) {
+    next_cluster = fat[next_cluster];
+    offset -= cluster_size;
+  }
+
+  uint32_t readden = 0;
+
+  uint8_t *cursor = to;
+
+  uint8_t *tmp = memory_allocate(cluster_size);
+  ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+  uint32_t prefix_size = bytes <= cluster_size - offset ? bytes : cluster_size - offset;
+  memory_copy(tmp + offset, prefix_size, cursor);
+  bytes -= prefix_size;
+  readden += prefix_size;
+  cursor += prefix_size;
+  next_cluster = fat[next_cluster];
+
+  while (bytes >= cluster_size) {
+    ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, cursor);
+    bytes -= cluster_size;
+    readden += cluster_size;
+    cursor += cluster_size;
+    next_cluster = fat[next_cluster];
+  }
+
+  if (bytes) {
+    ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+    memory_copy(tmp, bytes, cursor);
+    readden += bytes;
+  }
+
+  memory_free(tmp);
+
+  return readden;
+}
+
+uint64_t fat16_write(fs_node_t *file, uint32_t offset, const void *from, uint32_t bytes) {
+  fat16_node_t *fat_file = (fat16_node_t *)file;
+
+  if (fat_file->base.type != FAT16 || fat_file->base.is_dir) {
+    return (uint64_t)-1;
+  }
+
+  ensure_fat();
+
+  if (!fat_file->inode->first_cluster) {
+    fat_file->inode->first_cluster = allocate_cluster();
+  }
+
+  uint32_t cluster_size = bpb.sectors_per_cluster * SECTOR_SIZE;
+  uint32_t next_cluster = fat_file->inode->first_cluster;
+
+  uint32_t allocated = cluster_size;
+  while (allocated < offset + bytes) {
+    if (fat[next_cluster] >= 0xFFF8) {
+      fat[next_cluster] = allocate_cluster();
+    }
+    next_cluster = fat[next_cluster];
+    allocated += cluster_size;
+  }
+
+  flush_fat();
+
+  cluster_size = bpb.sectors_per_cluster * SECTOR_SIZE;
+  next_cluster = fat_file->inode->first_cluster;
+
+  while (offset >= cluster_size) {
+    next_cluster = fat[next_cluster];
+    offset -= cluster_size;
+  }
+
+  uint32_t written = 0;
+
+  const uint8_t *cursor = from;
+
+  uint8_t *tmp = memory_allocate(cluster_size);
+  ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+  uint32_t prefix_size = bytes <= cluster_size - offset ? bytes : cluster_size - offset;
+  memory_copy(cursor, prefix_size, tmp + offset);
+  ata_write_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+  bytes -= prefix_size;
+  written += prefix_size;
+  cursor += prefix_size;
+  next_cluster = fat[next_cluster];
+
+  while (bytes >= cluster_size) {
+    ata_write_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, cursor);
+    bytes -= cluster_size;
+    written += cluster_size;
+    cursor += cluster_size;
+    next_cluster = fat[next_cluster];
+  }
+
+  if (bytes) {
+    ata_read_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+    memory_copy(cursor, bytes, tmp);
+    ata_write_sectors(data_start + bpb.sectors_per_cluster * (next_cluster - 2), bpb.sectors_per_cluster, tmp);
+    written += bytes;
+  }
+
+  memory_free(tmp);
+
+  if (offset + written > fat_file->inode->size) {
+    fat_file->inode->size = offset + written;
+
+    fat16_dir_entry_t *entries = read_directory(fat_file->inode->dir_cluster);
+    entries[fat_file->inode->dir_index].first_cluster = fat_file->inode->first_cluster;
+    entries[fat_file->inode->dir_index].size = fat_file->inode->size;
+    write_directory(fat_file->inode->dir_cluster, entries);
+    memory_free(entries);
+  }
+
+  return written;
+}
+
+uint64_t fat16_truncate(fs_node_t *file, uint32_t size) {
+  fat16_node_t *fat_file = (fat16_node_t *)file;
+
+  if (fat_file->base.type != FAT16 || fat_file->base.is_dir) {
+    return (uint64_t)-1;
+  }
+
+  if (file->size == size) {
+    return size;
+  } else if (file->size < size) {
+    uint32_t diff = size - file->size;
+    uint8_t *tmp = memory_allocate(diff);
+    uint64_t written = fat16_write(file, file->size, tmp, diff);
+    memory_free(tmp);
+    return written == (uint64_t)-1 ? (uint64_t)-1 : file->size + written;
+  } else if (file->size > size) {
+    uint32_t cluster_size = bpb.sectors_per_cluster * SECTOR_SIZE;
+    uint32_t prev_cluster = 0;
+    uint32_t next_cluster = fat_file->inode->first_cluster;
+
+    ensure_fat();
+    uint32_t allocated = 0;
+    while (allocated < size) {
+      prev_cluster = next_cluster;
+      next_cluster = fat[next_cluster];
+      allocated += cluster_size;
+    }
+    if (prev_cluster) {
+      fat[prev_cluster] = 0xFFFF;
+    } else {
+      fat_file->inode->first_cluster = 0;
+    }
+    do {
+      uint32_t swap = fat[next_cluster];
+      fat[next_cluster] = 0;
+      next_cluster = swap;
+    } while (next_cluster < 0xFFF8);
+    flush_fat();
+
+    fat_file->inode->size = size;
+    fat16_dir_entry_t *entries = read_directory(fat_file->inode->dir_cluster);
+    entries[fat_file->inode->dir_index].first_cluster = fat_file->inode->first_cluster;
+    entries[fat_file->inode->dir_index].size = fat_file->inode->size;
+    write_directory(fat_file->inode->dir_cluster, entries);
+    memory_free(entries);
+
+    return size;
+  }
+  return (uint64_t)-1;
+}
+
 typedef struct {
   stream_t stream;
   fs_node_t *node;
-  uint64_t offset;
+  uint32_t offset;
 } fat16_file_stream_t;
 
-static uint64_t file_stream_write(__attribute__((unused)) const stream_t *self, __attribute__((unused)) const char *from,
-                                  __attribute__((unused)) uint64_t bytes) {
-  return (uint64_t)-1;
+static uint64_t file_stream_write(stream_t *self, const char *from, uint64_t bytes) {
+  fat16_file_stream_t *ffs = (fat16_file_stream_t *)self;
+  uint64_t written = fat16_write(ffs->node, ffs->offset, from, bytes > UINT32_MAX ? UINT32_MAX : (uint32_t)bytes);
+  if (written != (uint64_t)-1) {
+    ffs->offset += written;
+  }
+  return written;
 }
 
-static uint64_t file_stream_read(const stream_t *self, uint64_t max, char *to) {
+static uint64_t file_stream_read(stream_t *self, uint64_t max, char *to) {
   fat16_file_stream_t *ffs = (fat16_file_stream_t *)self;
-  if (ffs->offset >= ffs->node->size) {
-    return 0;
+  uint64_t readden = fat16_read(ffs->node, ffs->offset, max > UINT32_MAX ? UINT32_MAX : (uint32_t)max, to);
+  if (readden != (uint64_t)-1) {
+    ffs->offset += readden;
   }
-  uint64_t remaining = ffs->node->size - ffs->offset;
-  uint64_t to_read = remaining > max ? max : remaining;
-  fat16_read(ffs->node, ffs->offset, to_read, to);
-  ffs->offset += to_read;
-  return to_read;
+  return readden;
+}
+
+static uint64_t file_stream_truncate(stream_t *self, uint64_t size) {
+  fat16_file_stream_t *ffs = (fat16_file_stream_t *)self;
+  uint64_t resized = fat16_truncate(ffs->node, size > UINT32_MAX ? UINT32_MAX : (uint32_t)size);
+  if (resized != (uint64_t)-1 && resized < ffs->offset) {
+    ffs->offset = (uint32_t)resized;
+  }
+  return resized;
 }
 
 static void file_stream_close(stream_t *self) {
@@ -233,15 +535,15 @@ static void file_stream_close(stream_t *self) {
 typedef struct {
   stream_t stream;
   fs_node_t *node;
-  uint64_t index;
+  uint32_t index;
 } fat16_directory_stream_t;
 
-static uint64_t directory_stream_write(__attribute__((unused)) const stream_t *self, __attribute__((unused)) const char *from,
+static uint64_t directory_stream_write(__attribute__((unused)) stream_t *self, __attribute__((unused)) const char *from,
                                        __attribute__((unused)) uint64_t bytes) {
   return (uint64_t)-1;
 }
 
-static uint64_t directory_stream_read(const stream_t *self, uint64_t max, char *to) {
+static uint64_t directory_stream_read(stream_t *self, uint64_t max, char *to) {
   fat16_directory_stream_t *fds = (fat16_directory_stream_t *)self;
   fs_node_t *node = fat16_open_at(fds->node, fds->index++);
   if (!node) {
@@ -250,11 +552,14 @@ static uint64_t directory_stream_read(const stream_t *self, uint64_t max, char *
   uint64_t length = string_length(node->name);
   uint64_t to_read = length + 1 > max ? max : length + 1;
   memory_copy(node->name, to_read, to);
-  to[max - 1] = '\0';
-
+  to[to_read - 1] = '\0';
   return to_read;
 }
 
+static uint64_t directory_stream_truncate(__attribute__((unused)) stream_t *self, __attribute__((unused)) uint64_t size) {
+  return (uint64_t)-1;
+}
+
 static void directory_stream_close(stream_t *self) {
   fat16_directory_stream_t *fds = (fat16_directory_stream_t *)self;
   fat16_close(fds->node);
@@ -266,6 +571,7 @@ stream_t *fat16_open_stream(const fs_node_t *source) {
     fat16_file_stream_t *result = memory_allocate(sizeof(fat16_file_stream_t));
     result->stream.read = file_stream_read;
     result->stream.write = file_stream_write;
+    result->stream.truncate = file_stream_truncate;
     result->stream.close = file_stream_close;
     result->node = fat16_open_again(source);
     result->offset = 0;
@@ -274,6 +580,7 @@ stream_t *fat16_open_stream(const fs_node_t *source) {
     fat16_directory_stream_t *result = memory_allocate(sizeof(fat16_directory_stream_t));
     result->stream.read = directory_stream_read;
     result->stream.write = directory_stream_write;
+    result->stream.truncate = directory_stream_truncate;
     result->stream.close = directory_stream_close;
     result->node = fat16_open_again(source);
     result->index = 0;
@@ -281,68 +588,25 @@ stream_t *fat16_open_stream(const fs_node_t *source) {
   }
 }
 
-void fat16_read(const fs_node_t *file, uint64_t offset, uint64_t size, void *to) {
-  ASSERT(file->type == FAT16, "fat16_read: file is not FAT16");
-  ASSERT(!file->is_dir, "fat16_read: can not read directory");
-  ASSERT(file->size >= offset + size, "fat16_read: offset/size are out of bounds");
-
-  ensure_fat();
-
-  fat16_node_t *fat_file = (fat16_node_t *)file;
-
-  uint32_t cluster_size = bpb.sectors_per_cluster * SECTOR_SIZE;
-  uint32_t next_cluster = fat_file->entry.first_cluster;
-
-  // Assuming filesystem is correct. TODO Check for real.
-
-  while (offset >= cluster_size) {
-    next_cluster = fat[next_cluster];
-    offset -= cluster_size;
-  }
-
-  uint8_t *cursor = to;
-
-  uint8_t *tmp = memory_allocate(cluster_size);
-  ata_read_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.sectors_per_fat * bpb.fats_count +
-                       (bpb.root_entry_count * sizeof(fat16_dir_entry_t) + SECTOR_SIZE - 1) / SECTOR_SIZE +
-                       bpb.sectors_per_cluster * (next_cluster - 2),
-                   bpb.sectors_per_cluster, tmp);
-  uint64_t prefix_size = size <= cluster_size - offset ? size : cluster_size - offset;
-  memory_copy(tmp + offset, prefix_size, cursor);
-  size -= prefix_size;
-  cursor += prefix_size;
-  next_cluster = fat[next_cluster];
-
-  while (size >= cluster_size) {
-    ata_read_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.sectors_per_fat * bpb.fats_count +
-                         (bpb.root_entry_count * sizeof(fat16_dir_entry_t) + SECTOR_SIZE - 1) / SECTOR_SIZE +
-                         bpb.sectors_per_cluster * (next_cluster - 2),
-                     bpb.sectors_per_cluster, cursor);
-    size -= cluster_size;
-    cursor += cluster_size;
-    next_cluster = fat[next_cluster];
-  }
-
-  if (size) {
-    ata_read_sectors(FIRST_PARTITION_SECTOR + bpb.reserved_sectors + bpb.sectors_per_fat * bpb.fats_count +
-                         (bpb.root_entry_count * sizeof(fat16_dir_entry_t) + SECTOR_SIZE - 1) / SECTOR_SIZE +
-                         bpb.sectors_per_cluster * (next_cluster - 2),
-                     bpb.sectors_per_cluster, tmp);
-    memory_copy(tmp, size, cursor);
-  }
-
-  memory_free(tmp);
-}
-
 void fat16_close(fs_node_t *node) {
-  ASSERT(node->type == FAT16, "fat16_close: node is not FAT16");
-  ASSERT(node != fs, "fat16_close: can not unmount FS");
+  if (node->type != FAT16 || node == fs) {
+    return;
+  }
+  inode_free(((fat16_node_t *)node)->inode);
   memory_free(node);
 }
 
 void fat16_unmount(fs_node_t *node) {
-  ASSERT(node->type == FAT16, "fat16_unmount: node is not FAT16");
-  ASSERT(node == fs, "fat16_unmount: node is not filesystem");
+  if (node->type != FAT16 || node != fs) {
+    return;
+  }
+  inode_free(((fat16_node_t *)node)->inode);
   memory_free(node);
+
+  if (fat) {
+    memory_free(fat);
+    fat = NUL;
+  }
+
   fs = NUL;
 }
diff --git a/src/kernel/fat16.h b/src/kernel/fat16.h
index 45616fc..f1c4e1f 100644
--- a/src/kernel/fat16.h
+++ b/src/kernel/fat16.h
@@ -15,7 +15,11 @@ fs_node_t *fat16_open_again(const fs_node_t *source);
 
 stream_t *fat16_open_stream(const fs_node_t *source);
 
-void fat16_read(const fs_node_t *file, uint64_t offset, uint64_t size, void *to);
+uint64_t fat16_read(const fs_node_t *file, uint32_t offset, uint32_t bytes, void *to);
+
+uint64_t fat16_write(fs_node_t *file, uint32_t offset, const void *from, uint32_t bytes);
+
+uint64_t fat16_truncate(fs_node_t *file, uint32_t size);
 
 void fat16_close(fs_node_t *node);
 
diff --git a/src/kernel/fs.c b/src/kernel/fs.c
index 43f5390..e551fe2 100644
--- a/src/kernel/fs.c
+++ b/src/kernel/fs.c
@@ -21,8 +21,16 @@ stream_t *fs_open_stream(const fs_node_t *source) {
   return fat16_open_stream(source);
 }
 
-void fs_read(const fs_node_t *file, uint64_t offset, uint64_t size, void *to) {
-  fat16_read(file, offset, size, to);
+void fs_read(const fs_node_t *file, uint32_t offset, uint32_t bytes, void *to) {
+  fat16_read(file, offset, bytes, to);
+}
+
+void fs_write(fs_node_t *file, uint32_t offset, const void *from, uint32_t bytes) {
+  fat16_write(file, offset, from, bytes);
+}
+
+void fs_truncate(fs_node_t *file, uint32_t size) {
+  fat16_truncate(file, size);
 }
 
 void fs_close(fs_node_t *node) {
diff --git a/src/kernel/fs.h b/src/kernel/fs.h
index e9b02cd..a6b0b6d 100644
--- a/src/kernel/fs.h
+++ b/src/kernel/fs.h
@@ -22,7 +22,11 @@ fs_node_t *fs_open_again(const fs_node_t *source);
 
 stream_t *fs_open_stream(const fs_node_t *source);
 
-void fs_read(const fs_node_t *file, uint64_t offset, uint64_t size, void *to);
+void fs_read(const fs_node_t *file, uint32_t offset, uint32_t bytes, void *to);
+
+void fs_write(fs_node_t *file, uint32_t offset, const void *from, uint32_t bytes);
+
+void fs_truncate(fs_node_t *file, uint32_t size);
 
 void fs_close(fs_node_t *node);
 
diff --git a/src/kernel/keyboard.c b/src/kernel/keyboard.c
index 5a20671..6d6f1ac 100644
--- a/src/kernel/keyboard.c
+++ b/src/kernel/keyboard.c
@@ -51,12 +51,12 @@ static void append_sequence(const char *s) {
   }
 }
 
-static uint64_t stream_write(__attribute__((unused)) const stream_t *self, __attribute__((unused)) const char *from,
+static uint64_t stream_write(__attribute__((unused)) stream_t *self, __attribute__((unused)) const char *from,
                              __attribute__((unused)) uint64_t bytes) {
   return 0;
 }
 
-static uint64_t stream_read(__attribute__((unused)) const stream_t *self, uint64_t max, char *to) {
+static uint64_t stream_read(__attribute__((unused)) stream_t *self, uint64_t max, char *to) {
   while (!buffer_length) {
     __asm__ volatile("sti");
     process_next();
@@ -79,10 +79,14 @@ static uint64_t stream_read(__attribute__((unused)) const stream_t *self, uint64
   return size;
 }
 
+static uint64_t stream_truncate(__attribute__((unused)) stream_t *self, __attribute__((unused)) uint64_t size) {
+  return size;
+}
+
 static void stream_close(__attribute__((unused)) stream_t *self) {
 }
 
-static stream_t stream = {stream_write, stream_read, stream_close};
+static stream_t stream = {stream_write, stream_read, stream_truncate, stream_close};
 
 static void on_key(uint8_t scancode) {
   const uint8_t pressed = !(scancode & 0x80);
diff --git a/src/kernel/pipe.c b/src/kernel/pipe.c
index 2dd7a7e..84c5198 100644
--- a/src/kernel/pipe.c
+++ b/src/kernel/pipe.c
@@ -23,17 +23,17 @@ typedef struct {
   pipe_t *pipe;
 } pipe_write_stream_t;
 
-static uint64_t null_read(__attribute__((unused)) const stream_t *self, __attribute__((unused)) uint64_t max,
+static uint64_t null_read(__attribute__((unused)) stream_t *self, __attribute__((unused)) uint64_t max,
                           __attribute__((unused)) char *to) {
   return 0;
 }
 
-static uint64_t null_write(__attribute__((unused)) const stream_t *self, __attribute__((unused)) const char *from,
+static uint64_t null_write(__attribute__((unused)) stream_t *self, __attribute__((unused)) const char *from,
                            __attribute__((unused)) uint64_t bytes) {
   return 0;
 }
 
-static uint64_t pipe_read(const stream_t *self, uint64_t max, char *to) {
+static uint64_t pipe_read(stream_t *self, uint64_t max, char *to) {
   pipe_t *pipe = ((pipe_read_stream_t *)self)->pipe;
 
   while (pipe->begin == pipe->end) {
@@ -60,7 +60,7 @@ static uint64_t pipe_read(const stream_t *self, uint64_t max, char *to) {
   return to_read;
 }
 
-static uint64_t pipe_write(const stream_t *self, const char *from, uint64_t bytes) {
+static uint64_t pipe_write(stream_t *self, const char *from, uint64_t bytes) {
   pipe_t *pipe = ((pipe_read_stream_t *)self)->pipe;
 
   uint64_t written = 0;
diff --git a/src/kernel/stream.h b/src/kernel/stream.h
index a4d73f4..8e6c8e5 100644
--- a/src/kernel/stream.h
+++ b/src/kernel/stream.h
@@ -27,12 +27,14 @@
 
 typedef struct stream stream_t;
 
-typedef uint64_t (*stream_write_t)(const stream_t *self, const char *from, uint64_t bytes);
-typedef uint64_t (*stream_read_t)(const stream_t *self, uint64_t max, char *to);
+typedef uint64_t (*stream_write_t)(stream_t *self, const char *from, uint64_t bytes);
+typedef uint64_t (*stream_read_t)(stream_t *self, uint64_t max, char *to);
+typedef uint64_t (*stream_truncate_t)(stream_t *self, uint64_t size);
 typedef void (*stream_close_t)(stream_t *self);
 
 typedef struct stream {
   stream_write_t write;
   stream_read_t read;
+  stream_truncate_t truncate;
   stream_close_t close;
 } stream_t;
diff --git a/src/kernel/syscall.c b/src/kernel/syscall.c
index c1d35e8..386ffae 100644
--- a/src/kernel/syscall.c
+++ b/src/kernel/syscall.c
@@ -182,6 +182,13 @@ static uint64_t close(uint64_t fd) {
   return 0;
 }
 
+static uint64_t truncate(uint64_t fd, uint64_t size) {
+  if (fd >= MAX_FDS || !current_process->fds[fd]) {
+    return (uint64_t)-1;
+  }
+  return current_process->fds[fd]->truncate(current_process->fds[fd], size);
+}
+
 static void exit(exit_code_t code) {
   current_process->state = PROCESS_ZOMBIE;
   current_process->code = code;
@@ -214,6 +221,8 @@ uint64_t syscall_dispatch(uint64_t func, uint64_t arg1, uint64_t arg2, uint64_t
     return open((const char *)arg1);
   case SYSCALL_CLOSE:
     return close(arg1);
+  case SYSCALL_TRUNCATE:
+    return truncate(arg1, arg2);
   case SYSCALL_EXIT:
     exit(arg1);
     return 0;
diff --git a/src/kernel/syscall.h b/src/kernel/syscall.h
index c7553ec..b5d3497 100644
--- a/src/kernel/syscall.h
+++ b/src/kernel/syscall.h
@@ -10,6 +10,7 @@
 #define SYSCALL_WAIT 5
 #define SYSCALL_OPEN 6
 #define SYSCALL_CLOSE 7
+#define SYSCALL_TRUNCATE 8
 #define SYSCALL_EXIT 60
 
 void syscall_init();
diff --git a/src/kernel/vga.c b/src/kernel/vga.c
index cb04e5e..96db46c 100644
--- a/src/kernel/vga.c
+++ b/src/kernel/vga.c
@@ -142,7 +142,7 @@ static void on_char_received(char c) {
   }
 }
 
-static uint64_t stream_write(__attribute__((unused)) const stream_t *self, const char *from, uint64_t bytes) {
+static uint64_t stream_write(__attribute__((unused)) stream_t *self, const char *from, uint64_t bytes) {
   uint64_t left = bytes;
   while (left--) {
     on_char_received(*from++);
@@ -150,15 +150,19 @@ static uint64_t stream_write(__attribute__((unused)) const stream_t *self, const
   return bytes;
 }
 
-static uint64_t stream_read(__attribute__((unused)) const stream_t *self, __attribute__((unused)) uint64_t max,
+static uint64_t stream_read(__attribute__((unused)) stream_t *self, __attribute__((unused)) uint64_t max,
                             __attribute__((unused)) char *to) {
   return 0;
 }
 
+static uint64_t stream_truncate(__attribute__((unused)) stream_t *self, __attribute__((unused)) uint64_t size) {
+  return size;
+}
+
 static void stream_close(__attribute__((unused)) stream_t *self) {
 }
 
-static stream_t stream = {stream_write, stream_read, stream_close};
+static stream_t stream = {stream_write, stream_read, stream_truncate, stream_close};
 
 stream_t *vga_init() {
   clear();
diff --git a/src/user/app/cp/cp.c b/src/user/app/cp/cp.c
new file mode 100644
index 0000000..23b3167
--- /dev/null
+++ b/src/user/app/cp/cp.c
@@ -0,0 +1,55 @@
+#include "src/lib/util.h"
+#include "src/user/syscall.h"
+
+#define BLOCK_SIZE 65536
+
+#define PRINT_S(s) write(1, s, sizeof(s) - 1);
+#define PRINT_D(s) write(1, s, string_length(s));
+
+exit_code_t main(uint64_t argc, const char **argv) {
+  if (argc != 3) {
+    PRINT_S("cp: requires source and target");
+    return EXIT_CODE_GENERAL_FAILURE;
+  }
+
+  uint64_t source = open(argv[1]);
+  if (source == (uint64_t)-1) {
+    PRINT_S("cp: source does not exist\n");
+    return EXIT_CODE_GENERAL_FAILURE;
+  }
+
+  uint64_t target = open(argv[2]);
+  if (target == (uint64_t)-1) {
+    PRINT_S("cp: target does not exist\n");
+    close(source);
+    return EXIT_CODE_GENERAL_FAILURE;
+  }
+
+  if (truncate(target, 0) == (uint64_t)-1) {
+    close(source);
+    close(target);
+    return EXIT_CODE_GENERAL_FAILURE;
+  }
+
+  static char buffer[BLOCK_SIZE];
+  uint64_t bytes;
+  while ((bytes = read(source, BLOCK_SIZE, buffer))) {
+    if (bytes == (uint64_t)-1) {
+      PRINT_S("cp: could not read file\n");
+      close(source);
+      close(target);
+      return EXIT_CODE_GENERAL_FAILURE;
+    }
+    if (write(target, buffer, bytes) == (uint64_t)-1) {
+      PRINT_S("cp: could not write file\n");
+      close(source);
+      close(target);
+      return EXIT_CODE_GENERAL_FAILURE;
+    }
+  }
+
+  close(source);
+  close(target);
+
+  return EXIT_CODE_OK;
+}
diff --git a/src/user/syscall.asm b/src/user/syscall.asm
index f4a195f..564bd37 100644
--- a/src/user/syscall.asm
+++ b/src/user/syscall.asm
@@ -48,6 +48,12 @@ close:
     syscall
     ret
 
+global truncate
+truncate:
+    mov rax, 8
+    syscall
+    ret
+
 global exit
 exit:
     mov rax, 60
diff --git a/src/user/syscall.h b/src/user/syscall.h
index 11af13a..ef60f3b 100644
--- a/src/user/syscall.h
+++ b/src/user/syscall.h
@@ -19,4 +19,6 @@ uint64_t open(const char *path);
 
 uint64_t close(uint64_t fd);
 
+uint64_t truncate(uint64_t fd, uint64_t size);
+
 void exit(exit_code_t code);