diff --git a/disk/limine/limine.conf b/disk/limine/limine.conf
index b82e1bb..af0ec66 100644
--- a/disk/limine/limine.conf
+++ b/disk/limine/limine.conf
@@ -2,7 +2,6 @@ timeout: 0
 quiet:   yes
 
 /Calcite
-protocol:       limine
-path:           boot():/calcite/kernel.elf
-module_path:    boot():/calcite/initfs.tar
-module_cmdline: initfs
+protocol: limine
+path:     boot():/calcite/kernel.elf
+cmdline:  root-drive=pata2,root-fs=iso9660
diff --git a/make-build.sh b/make-build.sh
index 17a1e28..8c44eb2 100644
--- a/make-build.sh
+++ b/make-build.sh
@@ -69,7 +69,8 @@ for dir in src/user-libs/*; do
   build_all $dir user_cc user_lib_ld lib$lib_name.o
 done
 
-apps=""
+cp_apps=""
+app_deps=""
 
 for dir in src/user-apps/*; do
 
@@ -81,21 +82,18 @@ for dir in src/user-apps/*; do
 
   app_name=$(echo $dir | sed -e 's/^src\/user-apps\///')
   build_all $dir user_cc user_app_ld $app_name.elf $lib_paths
-  build_dir=$(echo $dir | sed -e 's/^src/build/')
-  apps="$apps $build_dir/$app_name.elf"
+
+  build_elf=$(echo $dir | sed -e 's/^src/build/')/$app_name.elf
+  disk_dir=build/disk/calcite/apps/$app_name
+  disk_elf=$disk_dir/$app_name.elf
+
+  cp_apps=$(echo "$cp_apps" \
+    "mkdir -p $disk_dir &&" \
+    "cp $build_elf $disk_elf &&")
+  app_deps="$app_deps $build_elf"
 
 done
 
-echo "rule initfs" >> build.ninja
-echo "  command =" \
-  "rm -rf build/initfs &&" \
-  "mkdir -p build/initfs/apps &&" \
-  "cp build/user-apps/hello/hello.elf build/initfs/apps/ &&" \
-  "cd build/initfs &&" \
-  "tar cf ../initfs.tar *" >> build.ninja
-
-echo "build build/initfs.tar: initfs $apps" >> build.ninja
-
 echo "rule disk" >> build.ninja
 echo "  command =" \
   "rm -rf build/disk &&" \
@@ -107,10 +105,10 @@ echo "  command =" \
   "cp dependencies/limine/BOOTX64.EFI build/disk/EFI/BOOT/ &&" \
   "mkdir build/disk/calcite &&" \
   "cp build/kernel/kernel.elf build/disk/calcite/ &&" \
-  "cp build/initfs.tar build/disk/calcite/ &&" \
+  $cp_apps \
   "xorriso -as mkisofs -R -r -J -b limine/limine-bios-cd.bin -no-emul-boot -boot-load-size 4 -boot-info-table -hfsplus -apm-block-size 2048 --efi-boot limine/limine-uefi-cd.bin -efi-boot-part --efi-boot-image --protective-msdos-label build/disk -o build/disk.iso &&" \
   "dependencies/limine/limine bios-install build/disk.iso" >> build.ninja
 
-echo "build build/disk.iso: disk | build/kernel/kernel.elf build/initfs.tar" >> build.ninja
+echo "build build/disk.iso: disk | build/kernel/kernel.elf $app_deps" >> build.ninja
 
 echo "default build/disk.iso" >> build.ninja
diff --git a/src/kernel/drives.c b/src/kernel/drives.c
index 377a856..422cf8b 100644
--- a/src/kernel/drives.c
+++ b/src/kernel/drives.c
@@ -46,3 +46,10 @@ struct drive_info *add_drive() {
   return to_return;
 
 }
+
+const struct drive_info *look_up_drive(const char *name) {
+  for (int i = 0; i < drive_list_count; ++i)
+    if (strequ(drive_list[i].name, name))
+      return &drive_list[i];
+  return 0;
+}
diff --git a/src/kernel/drives.h b/src/kernel/drives.h
index 8eee84e..ee3992a 100644
--- a/src/kernel/drives.h
+++ b/src/kernel/drives.h
@@ -20,7 +20,8 @@
 #include <stdint.h>
 
 enum drive_access_result {
-  DAR_SUCCESS
+  DAR_SUCCESS,
+  DAR_HARDWARE_ERROR
 };
 
 struct drive_info {
@@ -32,7 +33,7 @@ struct drive_info {
 
   //start and count are both in blocks
   enum drive_access_result (*read_blocks)(
-    const void *driver_info, void *buffer, uint64_t start, uint64_t count);
+    const struct drive_info *info, void *buffer, uint64_t start, uint64_t count);
 
 };
 
@@ -40,4 +41,6 @@ struct drive_info {
 //the returned pointer might be invalidated by future calls.
 struct drive_info *add_drive();
 
+//returned pointer might be invalidated by future calls to add_drive.
+//returns 0 if there is no drive with that name.
 const struct drive_info *look_up_drive(const char *name);
diff --git a/src/kernel/entry.c b/src/kernel/entry.c
index 14c1c65..49506a8 100644
--- a/src/kernel/entry.c
+++ b/src/kernel/entry.c
@@ -22,12 +22,13 @@
 #include "syscalls.h"
 #include "process.h"
 #include "utility.h"
-#include "initfs.h"
+#include "drives.h"
 #include "paging.h"
 #include "panic.h"
 #include "heap.h"
 #include "pata.h"
 #include "ps2.h"
+#include "fs.h"
 
 #include <limine.h>
 
@@ -57,8 +58,8 @@ static volatile struct limine_memmap_request memmap_request = {
   .response = 0
 };
 
-static volatile struct limine_module_request module_request = {
-  .id = LIMINE_MODULE_REQUEST,
+static volatile struct limine_executable_cmdline_request cmdline_request = {
+  .id = LIMINE_EXECUTABLE_CMDLINE_REQUEST,
   .revision = 0,
   .response = 0
 };
@@ -84,8 +85,8 @@ static void map_kernel_region(
       physical_start + i, (uint8_t *)virtual_start + i, writable, executable);
 }
 
-static uint8_t *initfs_start;
-static uint64_t initfs_length;
+#define MAX_CMDLINE_LENGTH 1000
+static char cmdline_copy[MAX_CMDLINE_LENGTH + 1];
 
 [[noreturn]] static void with_kernel_page_tables();
 
@@ -95,9 +96,7 @@ static uint64_t initfs_length;
 
   if (fb_request.response == 0 || hhdm_request.response == 0 ||
       ka_request.response == 0 || memmap_request.response == 0 ||
-      module_request.response == 0 ||
-      fb_request.response->framebuffer_count == 0 ||
-      module_request.response->module_count == 0)
+      fb_request.response->framebuffer_count == 0)
     die();
 
   struct limine_framebuffer *fb = fb_request.response->framebuffers[0];
@@ -107,19 +106,6 @@ static uint64_t initfs_length;
       fb->blue_mask_shift != 0 || fb->blue_mask_size != 8)
     die();
 
-  //find the initfs, and die if we don't have one.
-
-  struct limine_file *initfs = 0;
-  struct limine_module_response *response = module_request.response;
-  for (uint64_t i = 0; i < response->module_count; ++i)
-    if (strequ(response->modules[i]->cmdline, "initfs")) {
-      initfs = response->modules[i];
-      break;
-    }
-
-  if (!initfs)
-    die();
-
   //set up page tables. we will mark the regions with bootloader structures as
   //usable, so we need to be careful not to allocate any physical pages until
   //after we are done using bootloader structures. we map the kernel into our
@@ -162,17 +148,17 @@ static uint64_t initfs_length;
   fb_height = fb->height;
   fb_pitch = fb->pitch;
 
-  //remap initfs module and store information.
+  //make a copy of the kernel cmdline
 
-  uint64_t initfs_physical_start =
-    (uint64_t)initfs->address - hhdm_request.response->offset;
+  const char *original_cmdline = cmdline_request.response->cmdline;
+  int cmdline_length = 0;
+  while (original_cmdline[cmdline_length] != 0)
+    ++cmdline_length;
 
-  initfs_length = initfs->size;
-  uint64_t initfs_length_rounded = ((initfs_length - 1) / 4096 + 1) * 4096;
+  if (cmdline_length > MAX_CMDLINE_LENGTH)
+    die();
 
-  initfs_start = find_free_kernel_region(initfs_length_rounded);
-  map_kernel_region(
-    initfs_physical_start, initfs_start, initfs_length_rounded, 0, 0);
+  memcpy(cmdline_copy, original_cmdline, cmdline_length + 1);
 
   //switch to kernel page tables!
 
@@ -180,11 +166,78 @@ static uint64_t initfs_length;
 
 }
 
+struct cmdline_pair {
+  const char *key;
+  const char *value;
+};
+
+static const struct cmdline_pair *cmdline_pairs;
+static int cmdline_pair_count;
+
+//returns the corresponding value if the key exists, otherwise returns 0.
+static const char *cmdline_look_up(const char *key) {
+  for (int i = 0; i < cmdline_pair_count; ++i)
+    if (strequ(cmdline_pairs[i].key, key))
+      return cmdline_pairs[i].value;
+  return 0;
+}
+
 [[noreturn]] static void with_kernel_page_tables() {
 
-  //set initfs
+  //store cmdline as key-value pairs
 
-  set_initfs(initfs_start, initfs_length);
+  if (cmdline_copy[0] == 0)
+    cmdline_pair_count = 0;
+
+  else {
+
+    int comma_count = 0;
+    for (int i = 0; cmdline_copy[i] != 0; ++i)
+      if (cmdline_copy[i] == ',')
+        ++comma_count;
+
+    cmdline_pair_count = comma_count + 1;
+    struct cmdline_pair *pairs = heap_alloc(cmdline_pair_count * sizeof(struct cmdline_pair));
+
+    int key_start = 0;
+    for (int i = 0; i < cmdline_pair_count; ++i) {
+
+      int key_end = key_start;
+      while (1) {
+        if (cmdline_copy[key_end] == '=')
+          break;
+        if (cmdline_copy[key_end] == ',' || cmdline_copy[key_end] == 0)
+          die();
+        ++key_end;
+      }
+
+      int value_start = key_end + 1;
+      int value_end = value_start;
+      while (1) {
+        if (cmdline_copy[value_end] == ',' || cmdline_copy[value_end] == 0)
+          break;
+        if (cmdline_copy[value_end] == '=')
+          die();
+        ++value_end;
+      }
+
+      char *key = heap_alloc(key_end - key_start + 1);
+      memcpy(key, &cmdline_copy[key_start], key_end - key_start);
+      key[key_end - key_start] = 0;
+      pairs[i].key = key;
+
+      char *value = heap_alloc(value_end - value_start + 1);
+      memcpy(value, &cmdline_copy[value_start], value_end - value_start);
+      value[value_end - value_start] = 0;
+      pairs[i].value = value;
+
+      key_start = value_end + 1;
+
+    }
+
+    cmdline_pairs = pairs;
+
+  }
 
   //set up interrupts
 
@@ -203,18 +256,40 @@ static uint64_t initfs_length;
 
   probe_pata_drives();
 
+  //load root file system
+
+  const char *root_drive_name = cmdline_look_up("root-drive");
+  if (!root_drive_name)
+    panic("root drive mising from cmdline")
+  const struct drive_info *root_drive = look_up_drive(root_drive_name);
+  if (!root_drive)
+    panic("could not find root drive")
+
+  const char *root_fs_type = cmdline_look_up("root-fs");
+  if (!root_fs_type)
+    panic("root fs missing from cmdline")
+  struct fs_info root_fs;
+  if (create_fs_info(root_drive, &root_fs, root_fs_type) != FAR_SUCCESS)
+    panic("could not create root file system info")
+
   //load hello and start it
 
   init_scheduler();
 
-  const uint8_t *hello_start;
-  uint64_t hello_length;
-  look_up_initfs_file("apps/hello.elf", &hello_start, &hello_length);
+  void *hello_node;
+  if ((*root_fs.look_up_file)(
+        &root_fs, &hello_node,
+        "calcite/apps/hello/hello.elf") != FAR_SUCCESS)
+    panic("could not look up hello.elf")
 
   struct process *hello = heap_alloc(sizeof(struct process));
   uint64_t hello_entry;
   create_process(hello);
-  load_elf(hello, &hello_entry, hello_start, hello_length);
+  if (load_elf(hello, &hello_entry, &root_fs, hello_node) != 1)
+    panic("could not load hello.elf")
+
+  if ((*root_fs.free_node)(&root_fs, hello_node) != FAR_SUCCESS)
+    panic("could not free hello.elf node")
 
   struct thread *hello_thread = heap_alloc(sizeof(struct thread));
   create_thread(hello, hello_thread);
diff --git a/src/kernel/fs.c b/src/kernel/fs.c
new file mode 100644
index 0000000..583908c
--- /dev/null
+++ b/src/kernel/fs.c
@@ -0,0 +1,30 @@
+/* Calcite, src/kernel/fs.c
+ * Copyright 2025 Benji Dial
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "iso9660.h"
+#include "utility.h"
+#include "fs.h"
+
+enum fs_access_result create_fs_info(
+  const struct drive_info *drive, struct fs_info *fs_out, const char *fs_type) {
+
+  if (strequ(fs_type, "iso9660"))
+    return create_iso9660_info(drive, fs_out);
+
+  return FAR_UNKNOWN_FS_TYPE;
+
+}
diff --git a/src/kernel/fs.h b/src/kernel/fs.h
new file mode 100644
index 0000000..7dc1886
--- /dev/null
+++ b/src/kernel/fs.h
@@ -0,0 +1,58 @@
+/* Calcite, src/kernel/fs.h
+ * Copyright 2025 Benji Dial
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include "drives.h"
+
+#include <stdint.h>
+
+enum fs_access_result {
+  FAR_SUCCESS,
+  FAR_HARDWARE_ERROR,
+  FAR_FORMAT_ERROR,
+  FAR_UNKNOWN_FS_TYPE,
+  FAR_NOT_FOUND
+};
+
+struct fs_stat {
+  uint64_t bytes;
+};
+
+struct fs_info {
+
+  const struct drive_info *drive;
+  const void *driver_info;
+
+  enum fs_access_result (*free_node)(
+    const struct fs_info *info, void *node);
+
+  enum fs_access_result (*look_up_file)(
+    const struct fs_info *info, void **node_out, const char *path);
+
+  enum fs_access_result (*stat_file)(
+    const struct fs_info *info, void *node, struct fs_stat *stat_out);
+
+  //start is in bytes
+  enum fs_access_result (*read_file)(
+    const struct fs_info *info, void *node,
+    void *buffer, uint64_t start, uint64_t bytes);
+
+};
+
+enum fs_access_result create_fs_info(
+  const struct drive_info *drive, struct fs_info *fs_out, const char *fs_type);
diff --git a/src/kernel/initfs.c b/src/kernel/initfs.c
deleted file mode 100644
index 9cf3fed..0000000
--- a/src/kernel/initfs.c
+++ /dev/null
@@ -1,74 +0,0 @@
-/* Calcite, src/kernel/initfs.c
- * Copyright 2025 Benji Dial
- *
- * This program is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
- * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
- * for more details.
- *
- * You should have received a copy of the GNU General Public License along
- * with this program. If not, see <https://www.gnu.org/licenses/>.
- */
-
-#include "initfs.h"
-
-static const uint8_t *initfs_start;
-static uint64_t initfs_length;
-
-void set_initfs(const uint8_t *start, uint64_t length) {
-  initfs_start = start;
-  initfs_length = length;
-}
-
-static uint64_t decode_octal(const char *from) {
-  uint64_t value = 0;
-  while (*from != '\0') {
-    value = value * 8 + *from - '0';
-    ++from;
-  }
-  return value;
-}
-
-void look_up_initfs_file(
-  const char *path, const uint8_t **start_out, uint64_t *length_out) {
-
-  int path_len = 0;
-  while (path[path_len] != '\0')
-    ++path_len;
-
-  uint64_t offset = 0;
-
-  while (1) {
-
-    if (offset + 512 > initfs_length) {
-      *start_out = 0;
-      return;
-    }
-
-    uint64_t length =
-      decode_octal((const char *)(initfs_start + offset + 0x7c));
-
-    int found_it = 1;
-
-    for (int i = 0; i < path_len; ++i)
-      if (initfs_start[offset + i] != path[i]) {
-        found_it = 0;
-        break;
-      }
-
-    if (found_it) {
-      *start_out = initfs_start + offset + 512;
-      *length_out = length;
-      return;
-    }
-
-    offset += 512 + ((length - 1) / 512 + 1) * 512;
-
-  }
-
-}
diff --git a/src/kernel/iso9660.c b/src/kernel/iso9660.c
new file mode 100644
index 0000000..6e7b83f
--- /dev/null
+++ b/src/kernel/iso9660.c
@@ -0,0 +1,375 @@
+/* Calcite, src/kernel/iso9660.c
+ * Copyright 2025 Benji Dial
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include "iso9660.h"
+#include "utility.h"
+#include "drives.h"
+#include "panic.h"
+#include "heap.h"
+#include "fs.h"
+
+//relevant sources:
+//  https://www.ecma-international.org/wp-content/uploads/ECMA-119_2nd_edition_december_1987.pdf
+
+//case-insensitive equality
+static bool ciequ(char c1, char c2) {
+  if (c1 >= 'a' && c1 <= 'z')
+    c1 &= 0xdf;
+  if (c2 >= 'a' && c2 <= 'z')
+    c2 &= 0xdf;
+  return c1 == c2;
+}
+
+struct iso9660_driver_info {
+  const void *path_table;
+  //in bytes
+  int path_table_length;
+};
+
+static const struct iso9660_driver_info *get_driver_info(const struct fs_info *info) {
+  return (const struct iso9660_driver_info *)info->driver_info;
+}
+
+struct iso9660_node {
+  uint32_t first_block;
+  //in bytes
+  uint32_t length;
+};
+
+static enum fs_access_result free_node_iso9660(const struct fs_info *info, void *node) {
+  (void)info;
+  heap_dealloc(node, sizeof(struct iso9660_node));
+  return FAR_SUCCESS;
+}
+
+static uint8_t directory_buffer[2048];
+
+static enum fs_access_result look_up_in_directory(
+  const struct fs_info *info, void **node_out,
+  const char *path, int path_length,
+  uint32_t first_directory_block) {
+
+  if (first_directory_block >= info->drive->block_count)
+    return FAR_FORMAT_ERROR;
+
+  switch ((*info->drive->read_blocks)(info->drive, directory_buffer, first_directory_block, 1)) {
+  case DAR_SUCCESS:
+    break;
+  case DAR_HARDWARE_ERROR:
+    return FAR_HARDWARE_ERROR;
+  default:
+    assert(0);
+  }
+
+  int length = *(uint32_t *)&directory_buffer[10];
+  if (length % 2048 != 0)
+    return FAR_FORMAT_ERROR;
+
+  uint32_t loaded_block = first_directory_block;
+  int offset_in_block = 0;
+  int total_offset = 0;
+
+  while (1) {
+
+    if (total_offset == length)
+      return FAR_NOT_FOUND;
+
+    if (offset_in_block == 2048 || directory_buffer[offset_in_block] == 0) {
+      ++loaded_block;
+      if (loaded_block >= info->drive->block_count)
+        return FAR_FORMAT_ERROR;
+      switch ((*info->drive->read_blocks)(info->drive, directory_buffer, loaded_block, 1)) {
+      case DAR_SUCCESS:
+        break;
+      case DAR_HARDWARE_ERROR:
+        return FAR_HARDWARE_ERROR;
+      default:
+        assert(0);
+      }
+      total_offset = total_offset - offset_in_block + 2048;
+      offset_in_block = 0;
+      if (total_offset == length)
+        return FAR_NOT_FOUND;
+    }
+
+    int dr_length = directory_buffer[offset_in_block];
+    if (offset_in_block + dr_length > 2048)
+      return FAR_FORMAT_ERROR;
+
+    int id_length = directory_buffer[offset_in_block + 32];
+    if (33 + id_length > dr_length)
+      return FAR_FORMAT_ERROR;
+
+    for (int i = 0; i < id_length; ++i)
+      if (directory_buffer[offset_in_block + 33 + i] == ';') {
+        id_length = i;
+        break;
+      }
+
+    if (id_length != path_length)
+      goto next_entry;
+
+    for (int i = 0; i < id_length; ++i)
+      if (!ciequ(directory_buffer[offset_in_block + 33 + i], path[i]))
+        goto next_entry;
+
+    struct iso9660_node *node = heap_alloc(sizeof(struct iso9660_node));
+    node->first_block = *(uint32_t *)&directory_buffer[offset_in_block + 2];
+    node->length = *(uint32_t *)&directory_buffer[offset_in_block + 10];
+
+    *node_out = node;
+    return FAR_SUCCESS;
+
+  next_entry:
+    offset_in_block += dr_length;
+    total_offset += dr_length;
+
+  }
+
+}
+
+static enum fs_access_result look_up_recursive(
+  const struct fs_info *info, void **node_out,
+  const char *path, int parent_path_table_offset,
+  int parent_directory_number) {
+
+  //FIXME: this function is pretty unoptimized
+
+recurse:
+
+  int slash_location = 0;
+  int path_length;
+  while (1) {
+    if (path[slash_location] == '/')
+      break;
+    if (path[slash_location] == 0) {
+      path_length = slash_location;
+      slash_location = -1;
+      break;
+    }
+    ++slash_location;
+  }
+
+  const struct iso9660_driver_info *driver_info = get_driver_info(info);
+
+  int on_offset = parent_path_table_offset;
+  int on_number = parent_directory_number;
+
+  if (slash_location == -1) {
+    if (on_offset + 6 > driver_info->path_table_length)
+      return FAR_NOT_FOUND;
+    uint32_t first_directory_block = *(uint32_t *)(driver_info->path_table + on_offset + 2);
+    return look_up_in_directory(info, node_out, path, path_length, first_directory_block);
+  }
+
+  while (1) {
+
+    if (on_offset >= driver_info->path_table_length)
+      return FAR_NOT_FOUND;
+
+    int old_id_length = *(uint8_t *)(driver_info->path_table + on_offset);
+    on_offset += 8 + old_id_length + (old_id_length & 0x01);
+    ++on_number;
+
+    if (on_offset + 8 > driver_info->path_table_length)
+      return FAR_NOT_FOUND;
+
+    int this_entry_parent = *(uint16_t *)(driver_info->path_table + on_offset + 6);
+    if (this_entry_parent != parent_directory_number)
+      continue;
+
+    int this_entry_id_length = *(uint8_t *)(driver_info->path_table + on_offset);
+    if (this_entry_id_length != slash_location)
+      continue;
+
+    if (on_offset + 8 + this_entry_id_length > driver_info->path_table_length)
+      return FAR_NOT_FOUND;
+
+    char *this_entry_id = (char *)(driver_info->path_table + on_offset + 8);
+    for (int i = 0; i < slash_location; ++i)
+      if (!ciequ(this_entry_id[i], path[i]))
+        goto next_entry;
+
+    path = path + slash_location + 1;
+    parent_path_table_offset = on_offset;
+    parent_directory_number = on_number;
+    goto recurse;
+
+  next_entry:
+    ;
+
+  }
+
+}
+
+static enum fs_access_result look_up_file_iso9660(
+  const struct fs_info *info, void **node_out, const char *path) {
+
+  return look_up_recursive(info, node_out, path, 0, 1);
+
+}
+
+static enum fs_access_result stat_file_iso9660(
+  const struct fs_info *info, void *node, struct fs_stat *stat_out) {
+
+  (void)info;
+  struct iso9660_node *iso9660_node = (struct iso9660_node *)node;
+  stat_out->bytes = iso9660_node->length;
+  return FAR_SUCCESS;
+
+}
+
+static uint8_t read_buffer[2048];
+
+static enum fs_access_result read_file_iso9660(
+  const struct fs_info *info, void *node, void *buffer, uint64_t start, uint64_t bytes) {
+
+  struct iso9660_node *iso9660_node = (struct iso9660_node *)node;
+  if (iso9660_node->first_block + (start + bytes - 1) / 2048 + 1 > info->drive->block_count)
+    return FAR_FORMAT_ERROR;
+
+  if (start % 2048 != 0) {
+
+    switch ((*info->drive->read_blocks)(
+              info->drive, read_buffer, iso9660_node->first_block + start / 2048, 1)) {
+    case DAR_SUCCESS:
+      break;
+    case DAR_HARDWARE_ERROR:
+      return FAR_HARDWARE_ERROR;
+    default:
+      assert(0);
+    }
+
+    if (start + bytes <= ((start - 1) / 2048 + 1) * 2048) {
+      memcpy(buffer, read_buffer + (start % 2048), bytes);
+      return FAR_SUCCESS;
+    }
+
+    int to_copy = 2048 - (start % 2048);
+    memcpy(buffer, read_buffer + (start % 2048), to_copy);
+    buffer += to_copy;
+    start += to_copy;
+    bytes -= to_copy;
+
+  }
+
+  switch ((*info->drive->read_blocks)(
+            info->drive, buffer,
+            iso9660_node->first_block + start / 2048,
+            bytes / 2048)) {
+  case DAR_SUCCESS:
+    break;
+  case DAR_HARDWARE_ERROR:
+    return FAR_HARDWARE_ERROR;
+  default:
+    assert(0);
+  }
+
+  buffer += (bytes / 2048) * 2048;
+  start += (bytes / 2048) * 2048;
+  bytes %= 2048;
+
+  if (bytes > 0) {
+
+    switch ((*info->drive->read_blocks)(
+              info->drive, read_buffer, iso9660_node->first_block + start / 2048, 1)) {
+    case DAR_SUCCESS:
+      break;
+    case DAR_HARDWARE_ERROR:
+      return FAR_HARDWARE_ERROR;
+    default:
+      assert(0);
+    }
+
+    memcpy(buffer, read_buffer + (start % 2048), bytes);
+
+  }
+
+  return FAR_SUCCESS;
+
+}
+
+static uint8_t vd_buffer[2048];
+
+enum fs_access_result create_iso9660_info(const struct drive_info *drive, struct fs_info *fs_out) {
+
+  if (drive->block_size != 2048)
+    panic("TODO")
+
+  int descriptor_number = 0;
+  while (1) {
+
+    if ((uint64_t)(16 + descriptor_number) >= drive->block_count)
+      return FAR_FORMAT_ERROR;
+
+    switch ((*drive->read_blocks)(drive, vd_buffer, 16 + descriptor_number, 1)) {
+    case DAR_SUCCESS:
+      break;
+    case DAR_HARDWARE_ERROR:
+      return FAR_HARDWARE_ERROR;
+    default:
+      assert(0)
+    }
+
+    if (vd_buffer[0] == 0xff)
+      return FAR_FORMAT_ERROR;
+
+    if (vd_buffer[0] == 0x01)
+      break;
+
+    ++descriptor_number;
+
+  }
+
+  uint32_t path_table_length = *(uint32_t *)&vd_buffer[132];
+  uint32_t path_table_start = *(uint32_t *)&vd_buffer[140];
+
+  uint64_t path_table_length_rounded_up = ((path_table_length - 1) / 2048 + 1) * 2048;
+  void *path_table = heap_alloc(path_table_length_rounded_up);
+
+  if (path_table_start >= drive->block_count)
+    return FAR_FORMAT_ERROR;
+
+  switch ((*drive->read_blocks)(
+            drive, path_table, path_table_start,
+            path_table_length_rounded_up / 2048)) {
+  case DAR_SUCCESS:
+    break;
+  case DAR_HARDWARE_ERROR:
+    heap_dealloc(path_table, path_table_length_rounded_up);
+    return FAR_HARDWARE_ERROR;
+  default:
+    assert(0);
+  }
+
+  heap_dealloc(
+    path_table + path_table_length,
+    path_table_length_rounded_up - path_table_length);
+
+  struct iso9660_driver_info *driver_info = heap_alloc(sizeof(struct iso9660_driver_info));
+  driver_info->path_table = path_table;
+  driver_info->path_table_length = path_table_length;
+
+  fs_out->drive = drive;
+  fs_out->driver_info = driver_info;
+  fs_out->free_node = &free_node_iso9660;
+  fs_out->look_up_file = &look_up_file_iso9660;
+  fs_out->stat_file = &stat_file_iso9660;
+  fs_out->read_file = &read_file_iso9660;
+  return FAR_SUCCESS;
+
+}
diff --git a/src/kernel/initfs.h b/src/kernel/iso9660.h
similarity index 71%
rename from src/kernel/initfs.h
rename to src/kernel/iso9660.h
index e03f7fa..be3f5c7 100644
--- a/src/kernel/initfs.h
+++ b/src/kernel/iso9660.h
@@ -1,4 +1,4 @@
-/* Calcite, src/kernel/initfs.h
+/* Calcite, src/kernel/iso9660.h
  * Copyright 2025 Benji Dial
  *
  * This program is free software: you can redistribute it and/or modify
@@ -17,10 +17,6 @@
 
 #pragma once
 
-#include <stdint.h>
+#include "fs.h"
 
-void set_initfs(const uint8_t *start, uint64_t length);
-
-//if the file does not exist, *start_out is set to a null pointer.
-void look_up_initfs_file(
-  const char *path, const uint8_t **start_out, uint64_t *length_out);
+enum fs_access_result create_iso9660_info(const struct drive_info *drive, struct fs_info *fs_out);
diff --git a/src/kernel/pata.c b/src/kernel/pata.c
index dbde548..1264a69 100644
--- a/src/kernel/pata.c
+++ b/src/kernel/pata.c
@@ -21,6 +21,11 @@
 #include "heap.h"
 #include "pata.h"
 
+//some relevant sources:
+//  ANSI: AT Attachment 8 - ATA/ATAPI Command Set
+//  Seagate: SCSI Commands Reference Manual
+//  OSDev Wiki: ATAPI, PCI IDE Controller
+
 //in any function that takes a "controller", 0 means the primary controller
 //and 1 means the secondary controller. in any function that takes a "drive",
 //bit 1 is the controller, and bit 0 is bit 4 of the device register.
@@ -116,6 +121,51 @@ static enum pata_result patapi_read_capacity(
 
 }
 
+static enum pata_result patapi_read(
+  uint8_t drive, uint16_t block_size, uint32_t start_block, uint32_t block_count, void *buffer) {
+
+  uint8_t controller = drive >> 1;
+
+  pata_set_device(controller, (drive & 0x01) << 4);
+  pata_set_features(controller, 0);
+  pata_set_lba(controller, (uint32_t)block_size << 8);
+  pata_set_command(controller, 0xa0);
+
+  enum pata_result result = wait_pio(controller);
+  if (result != PR_SUCCESS)
+    return result;
+
+  uint8_t cmd[12] = {
+    0xa8, 0,
+    start_block >> 24,
+    (start_block >> 16) & 0xff,
+    (start_block >> 8) & 0xff,
+    start_block & 0xff,
+    block_count >> 24,
+    (block_count >> 16) & 0xff,
+    (block_count >> 8) & 0xff,
+    block_count & 0xff,
+    0, 0 };
+  pata_write_data(controller, (uint16_t *)cmd, 8);
+
+  for (uint32_t i = 0; i < block_count; ++i) {
+
+    result = wait_pio(controller);
+    if (result != PR_SUCCESS)
+      return result;
+
+    uint32_t actual_response_size = pata_read_lba(controller) >> 8;
+    if (actual_response_size != block_size)
+      panic("TODO")
+
+    pata_read_data(controller, (uint16_t *)(buffer + i * block_size), block_size / 2);
+
+  }
+
+  return PR_SUCCESS;
+
+}
+
 static uint16_t ipd_buffer[256];
 
 struct pata_driver_info {
@@ -123,10 +173,19 @@ struct pata_driver_info {
 };
 
 static enum drive_access_result read_blocks_patapi(
-  const struct pata_driver_info *driver_info,
+  const struct drive_info *drive_info,
   void *buffer, uint64_t start, uint64_t count) {
 
-  panic("TODO");
+  if (count >= (1ULL << 32) || start >= (1ULL << 32))
+    panic("TODO")
+
+  uint8_t drive = ((const struct pata_driver_info *)drive_info->driver_info)->drive;
+
+  return
+    patapi_read(
+      drive, drive_info->block_size,
+      start, count, buffer) == PR_SUCCESS
+    ? DAR_SUCCESS : DAR_HARDWARE_ERROR;
 
 }
 
@@ -155,7 +214,7 @@ static void probe_pata_drive(uint8_t drive) {
   driver_info->drive = drive;
   di->driver_info = driver_info;
 
-  di->read_blocks = (void *)&read_blocks_patapi;
+  di->read_blocks = &read_blocks_patapi;
 
 }
 
diff --git a/src/kernel/process.c b/src/kernel/process.c
index f840299..925c112 100644
--- a/src/kernel/process.c
+++ b/src/kernel/process.c
@@ -16,6 +16,7 @@
  */
 
 #include "framebuffer.h"
+#include "fs.h"
 #include "scheduler.h"
 #include "process.h"
 #include "utility.h"
@@ -191,26 +192,38 @@ no_mem:
 
 }
 
-void load_elf(
+int load_elf(
   struct process *process, uint64_t *entry_out,
-  const void *elf_start, uint64_t elf_length) {
+  const struct fs_info *fs_info, void *fs_node) {
 
-  if (elf_length < 58)
-    panic("malformed elf")
+  struct fs_stat stat;
+  if ((*fs_info->stat_file)(fs_info, fs_node, &stat) != FAR_SUCCESS)
+    return 0;
 
-  *entry_out = *(uint64_t *)(elf_start + 24);
+  if (stat.bytes < 58)
+    return 0;
 
-  uint64_t phead_start = *(uint64_t *)(elf_start + 32);
-  uint16_t phead_entry_size = *(uint16_t *)(elf_start + 54);
-  uint16_t phead_entry_count = *(uint16_t *)(elf_start + 56);
+  uint8_t head_part[34];
+  if ((*fs_info->read_file)(fs_info, fs_node, head_part, 24, 34) != FAR_SUCCESS)
+    return 0;
 
-  if (elf_length < phead_start + phead_entry_count * phead_entry_size)
+  *entry_out = *(uint64_t *)head_part;
+
+  uint64_t phead_start = *(uint64_t *)(head_part + 8);
+  uint16_t phead_entry_size = *(uint16_t *)(head_part + 30);
+  uint16_t phead_entry_count = *(uint16_t *)(head_part + 32);
+
+  if (stat.bytes < phead_start + phead_entry_count * phead_entry_size)
     panic("malformed elf")
 
   for (uint16_t i = 0; i < phead_entry_count; ++i) {
 
-    uint64_t *entry =
-      (uint64_t *)(elf_start + phead_start + i * phead_entry_size);
+    uint64_t entry[6];
+    if ((*fs_info->read_file)(
+          fs_info, fs_node, entry,
+          phead_start + i * phead_entry_size,
+          6 * sizeof(uint64_t)) != FAR_SUCCESS)
+      return 0;
 
     if ((entry[0] & 0xffffffff) != 1)
       continue;
@@ -223,13 +236,13 @@ void load_elf(
     uint64_t file_length = entry[4];
     uint64_t virtual_length = ((entry[5] - 1) / 4096 + 1) * 4096;
 
-    if (elf_length < file_start + file_length)
-      panic("malformed elf")
+    if (stat.bytes < file_start + file_length)
+      return 0;
     if (file_length > virtual_length)
-      panic("malformed elf")
+      return 0;
 
     if ((uint64_t)virtual_start % 4096 != 0 || virtual_length % 4096 != 0)
-      panic("unaligned elf")
+      return 0;
 
     for (uint64_t i = 0; i < virtual_length; i += 4096) {
 
@@ -240,12 +253,20 @@ void load_elf(
       void *kvma = find_free_kernel_region(4096);
       map_in_kernel_page_table(pma, kvma, 1, 0);
 
-      if (i + 4096 <= file_length)
-        memcpy(kvma, elf_start + file_start + i, 4096);
+      if (i + 4096 <= file_length) {
+        if ((*fs_info->read_file)(
+              fs_info, fs_node, kvma,
+              file_start + i, 4096) != FAR_SUCCESS)
+          return 0;
+      }
       else if (i >= file_length)
         memzero(kvma, 4096);
       else {
-        memcpy(kvma, elf_start + file_start + i, file_length & 0xfff);
+        if ((*fs_info->read_file)(
+              fs_info, fs_node, kvma,
+              file_start + i,
+              file_length & 0xfff) != FAR_SUCCESS)
+          return 0;
         memzero(kvma + (file_length & 0xfff), 4096 - (file_length & 0xfff));
       }
 
@@ -255,6 +276,8 @@ void load_elf(
 
   }
 
+  return 1;
+
 }
 
 void destroy_process(struct process *process) {
diff --git a/src/kernel/process.h b/src/kernel/process.h
index a623429..e10bf0a 100644
--- a/src/kernel/process.h
+++ b/src/kernel/process.h
@@ -17,7 +17,7 @@
 
 #pragma once
 
-#include <stdint.h>
+#include "fs.h"
 
 struct process {
 
@@ -64,12 +64,10 @@ void unmap_page_for_process(
 void *find_free_process_region(
   struct process *process, uint64_t page_count);
 
-//loaded sections are copied to new pages. once storage drivers are written,
-//this should probably just take a file handle so it can load and parse the
-//header and then load sections directly into user pages.
-void load_elf(
+//returns 0 on failure, 1 on success.
+int load_elf(
   struct process *process, uint64_t *entry_out,
-  const void *elf_start, uint64_t elf_length);
+  const struct fs_info *fs_info, void *fs_node);
 
 void destroy_process(struct process *process);