diff --git a/src/Makefile b/src/Makefile
index 83e2e81..cd3f89d 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -18,7 +18,7 @@ LDFLAGS = -mregparmcall -Wl,--gc-sections -Os -nostdlib -flto
 fat12boot.bin:
 	$(OBJCOPY) -O binary $< $@
 
-fat12boot.elf: fat12boot.o fat12.o bootsect.S
+fat12boot.elf: fat12boot.o fat12.o bootsect.S stdlib.o
 fat12boot.elf: LDFLAGS += -T bootsect.ld
 
 fat12boot.bin: fat12boot.elf
diff --git a/src/fat12.c b/src/fat12.c
index 4484184..bd12c17 100644
--- a/src/fat12.c
+++ b/src/fat12.c
@@ -1,12 +1,18 @@
 #include <stdint.h>
 #include <stdlib.h>
+#include <string.h>
+
+#include "polos.h"
 
 #define kSectorsPerCluster 2
 #define kSectorsPerTrack 9
 #define kHeads 2
+#define kTracks 40
 #define kFatSizeSectors 2
 #define kRootDirSizeSectors 7
+#define kBytesPerCluster 1024
 #define kDataRegionStartSector (1 + kFatSizeSectors * 2 + kRootDirSizeSectors)
+#define kNumClusters ((kSectorsPerTrack * kHeads * kTracks - kDataRegionStartSector) / kSectorsPerCluster + 2)
 
 typedef struct {
     char name[8];
@@ -22,23 +28,12 @@ typedef struct {
 static uint8_t* gFat;
 static direntry* gRootdir;
 
-static int readsector(int c, int h, int s, uint8_t* addr) {
-    register uint8_t* dest asm("bx") = addr;
-    register uint8_t nsects asm("al") = 1;
-    register uint8_t func asm("ah") = 0x02;
-    register uint8_t sect asm("cl") = s;
-    register uint8_t cyl asm("ch") = c;
-    register uint8_t head asm("dh") = h;
-    register uint8_t drive asm("dl") = 0;
-    register uint16_t seg asm("es") = 0;
-
-    register uint8_t ret asm("ah");
-
+static uint8_t rwsector(uint8_t w, int c, int h, int s, void* addr) {
+    uint16_t ret;
+    uint16_t cmd = w ? 0x0301 : 0x0201;
+    // retry for a total of up to 3 attempts if drive not ready
     for (int i = 0; i < 3; i++) {
-        asm volatile("int $0x13"
-                     : "=r"(ret)
-                     : "r"(dest), "r"(nsects), "r"(func), "r"(sect), "r"(cyl),
-                       "r"(head), "r"(drive), "r"(seg));
+        ret = __readwritesector__(cmd, addr, c, h, s) >> 8;
         if (ret != 0x80) {
             break;
         }
@@ -47,6 +42,14 @@ static int readsector(int c, int h, int s, uint8_t* addr) {
     return ret;
 }
 
+static uint8_t readsector(int c, int h, int s, void* addr) {
+    return rwsector(0, c, h, s, addr);
+}
+
+static uint8_t writesector(int c, int h, int s, void* addr) {
+    return rwsector(1, c, h, s, addr);
+}
+
 static int readcluster(int cluster) {
     int offs = cluster * 3 / 2;
     if (cluster % 2) {
@@ -57,6 +60,28 @@ static int readcluster(int cluster) {
     }
 }
 
+static int writecluster(int cluster, int value) {
+    int offs = cluster * 3 / 2;
+    if (cluster % 2) {
+        gFat[offs] = (gFat[offs] & 0x0f) + ((value & 0x0f) << 4);
+        gFat[offs+1] = value >> 4;
+        // high nibble is lsb + whole byte
+    } else {
+        gFat[offs] = value & 0xff;
+        gFat[offs+1] = (gFat[offs+1] & 0xf0) + (value >> 8);
+    }
+    return value;
+}
+
+static int firstfreecluster() {
+    for (int i = 2; i < kNumClusters; i++) {
+        if (readcluster(i) == 0) {
+            return i;
+        }
+    }
+    return -1;
+}
+
 static void cluster2chs(int cluster, int* c, int* h, int* s) {
     int logicalsector =
             kDataRegionStartSector + (cluster - 2) * kSectorsPerCluster;
@@ -65,19 +90,9 @@ static void cluster2chs(int cluster, int* c, int* h, int* s) {
     *c = logicalsector / (kHeads * kSectorsPerTrack);
 }
 
-static int strncmp(const char* s1, const char* s2, size_t len) {
-    for (int i = 0; i < len && s1[i]; i++) {
-        if (s1[i] != s2[i]) {
-            return 1;
-        }
-    }
-    // XXX: really we should return 0 only if *s1 == *s2 here too
-    return 0;
-}
-
 static int loadfile(direntry* entry, void* addr) {
     int cluster = entry->cluster;
-    for (int i = 0; i < entry->size; i += 1024) {
+    for (int i = 0; i < entry->size; i += kBytesPerCluster) {
         int c, h, s;
         cluster2chs(cluster, &c, &h, &s);
         if (readsector(c, h, s, addr + i)) {
@@ -88,10 +103,10 @@ static int loadfile(direntry* entry, void* addr) {
             s = 1;
             h++;
         }
-        if (h > 1) {
-            h = 0;
-            c++;
-        }
+        // if (h > 1) {  // this will never happen
+        //     h = 0;
+        //     c++;
+        // }
         if (readsector(c, h, s, addr + i + 512)) {
             return -5;
         };
@@ -100,6 +115,94 @@ static int loadfile(direntry* entry, void* addr) {
     return 0;
 }
 
+static direntry* findfile(const char* name) {
+    direntry* file = 0;
+
+    for (int i = 0; i < 16; i++) {
+        direntry* entry = &gRootdir[i];
+        if (!strncmp(entry->name, name, 11)) {
+            file = entry;
+            break;
+        }
+    }
+    return file;
+}
+
+static direntry* newfile(const char* name) {
+    direntry* file = 0;
+
+    for (int i = 0; i < 16; i++) {
+        direntry* entry = &gRootdir[i];
+        if (entry->name[0] == 0) {
+            file = entry;
+            break;
+        }
+    }
+    if (!file) {
+        return file;
+    }
+
+    memcpy(file->name, name, 11);
+    file->size = 0;
+    file->cluster = 0;
+
+    return file;
+}
+
+static int allocatecluster(int previous) {
+    int cluster = firstfreecluster();
+    writecluster(cluster, 0xfff);
+    if (previous != 0) {
+        writecluster(previous, cluster);
+    }
+    return cluster;
+}
+
+static void freeclusterchain(int cluster) {
+    if (cluster < 2 || cluster >= 0xf00) {
+        return;
+    }
+    freeclusterchain(readcluster(cluster));
+    writecluster(cluster, 0);
+}
+
+static int savefile(direntry* file, void* addr, int size) {
+    freeclusterchain(file->cluster);
+    file->cluster = 0;
+    int cluster = 0;
+    for (int i = 0; i < size; i += kBytesPerCluster) {
+        cluster = allocatecluster(cluster);
+        if (file->cluster == 0) {
+            file->cluster = cluster;
+        }
+        int c, h, s;
+        cluster2chs(cluster, &c, &h, &s);
+        if (writesector(c, h, s, addr + i)) {
+            return -7;
+        }
+        s++;
+        if (s > 9) {
+            s = 1;
+            h++;
+        }
+        // if (h > 1) {  // no cluster will cross a cylinder boundary
+        //     h = 0;
+        //     c++;
+        // }
+        if (writesector(c, h, s, addr + i + 512)) {
+            return -7;
+        };
+    }
+    file->size = size;
+    if (writesector(0, 0, 2, gFat)) {
+        return -8;
+    }
+    if (writesector(0, 0, 6, gRootdir)) {
+        return -9;
+    }
+    return 0;
+}
+
 int fat12_init(void* fat_addr, void* rootdir_addr) {
     if (readsector(0, 0, 2, fat_addr)) {
         return -2;
@@ -115,21 +218,22 @@ int fat12_init(void* fat_addr, void* rootdir_addr) {
 }
 
 int fat12_readfile(const char* name, void* addr) {
-    direntry* file = 0;
-
-    for (int i = 0; i < 16; i++) {
-        direntry* entry = &gRootdir[i];
-        if (entry->name[0] == 0) {
-            break;
-        }
-        if (!strncmp(entry->name, name, 11)) {
-            file = entry;
-            break;
-        }
-    }
+    direntry* file = findfile(name);
     if (!file) {
         return -4;
     }
 
     return loadfile(file, addr);
 }
+
+int fat12_writefile(const char* name, void* addr, int size) {
+    direntry* file = findfile(name);
+    if (!file) {
+        file = newfile(name);
+    }
+    if (!file) {
+        // no moar space
+        return -6;
+    }
+    return savefile(file, addr, size);
+}
diff --git a/src/polos.h b/src/polos.h
index 83d9da0..5613ec4 100644
--- a/src/polos.h
+++ b/src/polos.h
@@ -8,6 +8,9 @@
 
 int runcomms(char until_idle);
 
+int putchar(int);
+int puts(const char*);
+
 inline static uint8_t __get_far_u8__(uint16_t addr, uint16_t seg) {
     register uint16_t ad asm("di") = addr;
     register uint8_t ret asm("al");
@@ -18,3 +21,39 @@ inline static uint8_t __get_far_u8__(uint16_t addr, uint16_t seg) {
         : "es");
     return ret;
 }
+
+inline static void __set_far_u8__(uint16_t addr, uint16_t seg, uint8_t val) {
+    register uint16_t ad asm("di") = addr;
+    register uint8_t v asm("al") = val;
+    asm("mov %0, %%es              \t\n"
+    "stosb      \n\t" ::"r"(seg), "r"(ad), "r"(v) : "es", "memory");
+}
+
+inline static uint16_t __readwritesector__(uint16_t cmd, uint8_t* addr, uint8_t c, uint8_t h, uint8_t s) {
+    register uint8_t* dest asm("bx") = addr;
+    register uint8_t sect asm("cl") = s;
+    register uint8_t cyl asm("ch") = c;
+    register uint8_t head asm("dh") = h;
+    register uint8_t drive asm("dl") = 0;
+    register uint16_t ret asm("ax");
+
+    asm volatile("xor %%ax, %%ax \n\t"
+            "mov %%ax, %%es \n\t"
+            "mov %1, %%ax \n\t"
+            "int $0x13"
+            : "=r"(ret)
+            : "ir"(cmd), "r"(dest), "r"(sect), "r"(cyl),
+            "r"(head), "r"(drive)
+            );
+
+    return ret;
+}
+
+inline static uint16_t __readsector__(uint8_t* addr, uint8_t c, uint8_t h, uint8_t s) {
+    return __readwritesector__(0x0201, addr, c, h, s);
+}
+
+inline static uint16_t __writesector__(uint8_t* addr, uint8_t c, uint8_t h, uint8_t s) {
+    return __readwritesector__(0x0301, addr, c, h, s);
+}
+
diff --git a/src/stdlib.c b/src/stdlib.c
index 71e80a9..43f1b4b 100644
--- a/src/stdlib.c
+++ b/src/stdlib.c
@@ -46,7 +46,17 @@ void* memset(void* ptr, int val, size_t len) {
     return ptr;
 }
 
-int _int80h_2(uint8_t fun, uint16_t arg1, uint16_t arg2) {
+int strncmp(const char* s1, const char* s2, size_t len) {
+    for (int i = 0; i < len && s1[i]; i++) {
+        if (s1[i] != s2[i]) {
+            return 1;
+        }
+    }
+    // XXX: really we should return 0 only if *s1 == *s2 here too
+    return 0;
+}
+
+static int int80h_2(uint8_t fun, uint16_t arg1, uint16_t arg2) {
     register uint16_t a1 asm("dx") = arg1;
     register uint16_t a2 asm("cx") = arg2;
     register uint8_t ah asm("ah") = fun;
@@ -79,12 +89,12 @@ int read(int fd, void* buf, size_t size) {
     if (fd != kLpt1) {
         return -1;
     }
-    return _int80h_2(0x05, (uint16_t)buf, size);
+    return int80h_2(0x05, (uint16_t)buf, size);
 }
 
 int write(int fd, void* buf, size_t size) {
     if (fd != kLpt1) {
         return -1;
     }
-    return _int80h_2(0x06, (uint16_t)buf, size);
+    return int80h_2(0x06, (uint16_t)buf, size);
 }