qt-faststart - stco offset bug fix

when the last offsets in the stco atom are close to 4GB, the addition of
the moov atom size can overflow, causing corruption near the end of the
mp4 file.
this patch upgrades all stco atoms to co64 when such an edge case is
detected. in order to accomplish this, the implementation was changed to
walk the atom tree, instead of searching for the strings 'stco'/'co64'.
this was required since when an stco atom is changed to co64, its size
changes, and the sizes of all containing atoms (moov, trak, etc.) have
to be updated as well.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

tools/qt-faststart.c

@@ -28,6 +28,7 @@
 #include <stdlib.h>
 #include <inttypes.h>
 #include <string.h>
+#include <limits.h>
 
 #ifdef __MINGW32__
 #undef fseeko
@@ -43,8 +44,6 @@
 
 #define MIN(a,b) ((a) > (b) ? (b) : (a))
 
-#define BE_16(x) ((((uint8_t*)(x))[0] <<  8) | ((uint8_t*)(x))[1])
-
 #define BE_32(x) (((uint32_t)(((uint8_t*)(x))[0]) << 24) |  \
                              (((uint8_t*)(x))[1]  << 16) |  \
                              (((uint8_t*)(x))[2]  <<  8) |  \
@@ -59,6 +58,18 @@
                   ((uint64_t)(((uint8_t*)(x))[6]) <<  8) |  \
                   ((uint64_t)( (uint8_t*)(x))[7]))
 
+#define AV_WB32(p, val)    {                    \
+    ((uint8_t*)(p))[0] = ((val) >> 24) & 0xff;  \
+    ((uint8_t*)(p))[1] = ((val) >> 16) & 0xff;  \
+    ((uint8_t*)(p))[2] = ((val) >> 8) & 0xff;   \
+    ((uint8_t*)(p))[3] = (val) & 0xff;          \
+    }
+
+#define AV_WB64(p, val)    {                    \
+    AV_WB32(p, (val) >> 32)                     \
+    AV_WB32(p + 4, val)                         \
+    }
+
 #define BE_FOURCC(ch0, ch1, ch2, ch3)           \
     ( (uint32_t)(unsigned char)(ch3)        |   \
      ((uint32_t)(unsigned char)(ch2) <<  8) |   \
@@ -79,6 +90,10 @@
 #define UUID_ATOM QT_ATOM('u', 'u', 'i', 'd')
 
 #define CMOV_ATOM QT_ATOM('c', 'm', 'o', 'v')
+#define TRAK_ATOM QT_ATOM('t', 'r', 'a', 'k')
+#define MDIA_ATOM QT_ATOM('m', 'd', 'i', 'a')
+#define MINF_ATOM QT_ATOM('m', 'i', 'n', 'f')
+#define STBL_ATOM QT_ATOM('s', 't', 'b', 'l')
 #define STCO_ATOM QT_ATOM('s', 't', 'c', 'o')
 #define CO64_ATOM QT_ATOM('c', 'o', '6', '4')
 
@@ -86,6 +101,332 @@
 #define COPY_BUFFER_SIZE   33554432
 #define MAX_FTYP_ATOM_SIZE 1048576
 
+typedef struct {
+    uint32_t type;
+    uint32_t header_size;
+    uint64_t size;
+    unsigned char *data;
+} atom_t;
+
+typedef struct {
+    uint64_t moov_atom_size;
+    uint64_t stco_offset_count;
+    uint64_t stco_data_size;
+    int stco_overflow;
+    uint32_t depth;
+} update_chunk_offsets_context_t;
+
+typedef struct {
+    unsigned char *dest;
+    uint64_t original_moov_size;
+    uint64_t new_moov_size;
+} upgrade_stco_context_t;
+
+typedef int (*parse_atoms_callback_t)(void *context, atom_t *atom);
+
+static int parse_atoms(
+    unsigned char *buf,
+    uint64_t size,
+    parse_atoms_callback_t callback,
+    void *context)
+{
+    unsigned char *pos = buf;
+    unsigned char *end = pos + size;
+    atom_t atom;
+    int ret;
+
+    while (end - pos >= ATOM_PREAMBLE_SIZE) {
+        atom.size = BE_32(pos);
+        atom.type = BE_32(pos + 4);
+        pos += ATOM_PREAMBLE_SIZE;
+        atom.header_size = ATOM_PREAMBLE_SIZE;
+
+        switch (atom.size) {
+        case 1:
+            if (end - pos < 8) {
+                printf("not enough room for 64 bit atom size\n");
+                return -1;
+            }
+
+            atom.size = BE_64(pos);
+            pos += 8;
+            atom.header_size = ATOM_PREAMBLE_SIZE + 8;
+            break;
+
+        case 0:
+            atom.size = ATOM_PREAMBLE_SIZE + end - pos;
+            break;
+        }
+
+        if (atom.size < atom.header_size) {
+            printf("atom size %"PRIu64" too small\n", atom.size);
+            return -1;
+        }
+
+        atom.size -= atom.header_size;
+
+        if (atom.size > end - pos) {
+            printf("atom size %"PRIu64" too big\n", atom.size);
+            return -1;
+        }
+
+        atom.data = pos;
+        ret = callback(context, &atom);
+        if (ret < 0) {
+            return ret;
+        }
+
+        pos += atom.size;
+    }
+
+    return 0;
+}
+
+static int update_stco_offsets(update_chunk_offsets_context_t *context, atom_t *atom)
+{
+    uint32_t current_offset;
+    uint32_t offset_count;
+    unsigned char *pos;
+    unsigned char *end;
+
+    printf(" patching stco atom...\n");
+    if (atom->size < 8) {
+        printf("stco atom size %"PRIu64" too small\n", atom->size);
+        return -1;
+    }
+
+    offset_count = BE_32(atom->data + 4);
+    if (offset_count > (atom->size - 8) / 4) {
+        printf("stco offset count %"PRIu32" too big\n", offset_count);
+        return -1;
+    }
+
+    context->stco_offset_count += offset_count;
+    context->stco_data_size += atom->size - 8;
+
+    for (pos = atom->data + 8, end = pos + offset_count * 4;
+        pos < end;
+        pos += 4) {
+        current_offset = BE_32(pos);
+        if (current_offset > UINT_MAX - context->moov_atom_size) {
+            context->stco_overflow = 1;
+        }
+        current_offset += context->moov_atom_size;
+        AV_WB32(pos, current_offset);
+    }
+
+    return 0;
+}
+
+static int update_co64_offsets(update_chunk_offsets_context_t *context, atom_t *atom)
+{
+    uint64_t current_offset;
+    uint32_t offset_count;
+    unsigned char *pos;
+    unsigned char *end;
+
+    printf(" patching co64 atom...\n");
+    if (atom->size < 8) {
+        printf("co64 atom size %"PRIu64" too small\n", atom->size);
+        return -1;
+    }
+
+    offset_count = BE_32(atom->data + 4);
+    if (offset_count > (atom->size - 8) / 8) {
+        printf("co64 offset count %"PRIu32" too big\n", offset_count);
+        return -1;
+    }
+
+    for (pos = atom->data + 8, end = pos + offset_count * 8;
+        pos < end;
+        pos += 8) {
+        current_offset = BE_64(pos);
+        current_offset += context->moov_atom_size;
+        AV_WB64(pos, current_offset);
+    }
+
+    return 0;
+}
+
+static int update_chunk_offsets_callback(void *ctx, atom_t *atom)
+{
+    update_chunk_offsets_context_t *context = ctx;
+    int ret;
+
+    switch (atom->type) {
+    case STCO_ATOM:
+        return update_stco_offsets(context, atom);
+
+    case CO64_ATOM:
+        return update_co64_offsets(context, atom);
+
+    case MOOV_ATOM:
+    case TRAK_ATOM:
+    case MDIA_ATOM:
+    case MINF_ATOM:
+    case STBL_ATOM:
+        context->depth++;
+        if (context->depth > 10) {
+            printf("atoms too deeply nested\n");
+            return -1;
+        }
+
+        ret = parse_atoms(
+            atom->data,
+            atom->size,
+            update_chunk_offsets_callback,
+            context);
+        context->depth--;
+        return ret;
+    }
+
+    return 0;
+}
+
+static void set_atom_size(unsigned char *header, uint32_t header_size, uint64_t size)
+{
+    switch (header_size) {
+    case 8:
+        AV_WB32(header, size);
+        break;
+
+    case 16:
+        AV_WB64(header + 8, size);
+        break;
+    }
+}
+
+static void upgrade_stco_atom(upgrade_stco_context_t *context, atom_t *atom)
+{
+    unsigned char *pos;
+    unsigned char *end;
+    uint64_t new_offset;
+    uint32_t offset_count;
+    uint32_t original_offset;
+
+    /* Note: not performing validations since they were performed on the first pass */
+
+    offset_count = BE_32(atom->data + 4);
+
+    /* write the header */
+    memcpy(context->dest, atom->data - atom->header_size, atom->header_size + 8);
+    AV_WB32(context->dest + 4, CO64_ATOM);
+    set_atom_size(context->dest, atom->header_size, atom->header_size + 8 + offset_count * 8);
+    context->dest += atom->header_size + 8;
+
+    /* write the data */
+    for (pos = atom->data + 8, end = pos + offset_count * 4;
+        pos < end;
+        pos += 4) {
+        original_offset = BE_32(pos) - context->original_moov_size;
+        new_offset = (uint64_t)original_offset + context->new_moov_size;
+        AV_WB64(context->dest, new_offset);
+        context->dest += 8;
+    }
+}
+
+static int upgrade_stco_callback(void *ctx, atom_t *atom)
+{
+    upgrade_stco_context_t *context = ctx;
+    unsigned char *start_pos;
+    uint64_t copy_size;
+
+    switch (atom->type) {
+    case STCO_ATOM:
+        upgrade_stco_atom(context, atom);
+        break;
+
+    case MOOV_ATOM:
+    case TRAK_ATOM:
+    case MDIA_ATOM:
+    case MINF_ATOM:
+    case STBL_ATOM:
+        /* write the atom header */
+        memcpy(context->dest, atom->data - atom->header_size, atom->header_size);
+        start_pos = context->dest;
+        context->dest += atom->header_size;
+
+        /* parse internal atoms*/
+        if (parse_atoms(
+            atom->data,
+            atom->size,
+            upgrade_stco_callback,
+            context) < 0) {
+            return -1;
+        }
+
+        /* update the atom size */
+        set_atom_size(start_pos, atom->header_size, context->dest - start_pos);
+        break;
+
+    default:
+        copy_size = atom->header_size + atom->size;
+        memcpy(context->dest, atom->data - atom->header_size, copy_size);
+        context->dest += copy_size;
+        break;
+    }
+
+    return 0;
+}
+
+static int update_moov_atom(
+    unsigned char **moov_atom,
+    uint64_t *moov_atom_size)
+{
+    update_chunk_offsets_context_t update_context = { 0 };
+    upgrade_stco_context_t upgrade_context;
+    unsigned char *new_moov_atom;
+
+    update_context.moov_atom_size = *moov_atom_size;
+
+    if (parse_atoms(
+        *moov_atom,
+        *moov_atom_size,
+        update_chunk_offsets_callback,
+        &update_context) < 0) {
+        return -1;
+    }
+
+    if (!update_context.stco_overflow) {
+        return 0;
+    }
+
+    printf(" upgrading stco atoms to co64...\n");
+    upgrade_context.new_moov_size = *moov_atom_size +
+        update_context.stco_offset_count * 8 -
+        update_context.stco_data_size;
+
+    new_moov_atom = malloc(upgrade_context.new_moov_size);
+    if (new_moov_atom == NULL) {
+        printf("could not allocate %"PRIu64" bytes for updated moov atom\n",
+            upgrade_context.new_moov_size);
+        return -1;
+    }
+
+    upgrade_context.original_moov_size = *moov_atom_size;
+    upgrade_context.dest = new_moov_atom;
+
+    if (parse_atoms(
+        *moov_atom,
+        *moov_atom_size,
+        upgrade_stco_callback,
+        &upgrade_context) < 0) {
+        free(new_moov_atom);
+        return -1;
+    }
+
+    free(*moov_atom);
+    *moov_atom = new_moov_atom;
+    *moov_atom_size = upgrade_context.new_moov_size;
+
+    if (upgrade_context.dest != *moov_atom + *moov_atom_size) {
+        printf("unexpected - wrong number of moov bytes written\n");
+        return -1;
+    }
+
+    return 0;
+}
+
 int main(int argc, char *argv[])
 {
     FILE *infile  = NULL;
@@ -99,9 +440,6 @@ int main(int argc, char *argv[])
     unsigned char *ftyp_atom = NULL;
     uint64_t moov_atom_size;
     uint64_t ftyp_atom_size = 0;
-    uint64_t i, j;
-    uint32_t offset_count;
-    uint64_t current_offset;
     int64_t start_offset = 0;
     unsigned char *copy_buffer = NULL;
     int bytes_to_copy;
@@ -244,57 +582,9 @@ int main(int argc, char *argv[])
     fclose(infile);
     infile = NULL;
 
-    /* crawl through the moov chunk in search of stco or co64 atoms */
+    if (update_moov_atom(&moov_atom, &moov_atom_size) < 0) {
-    for (i = 4; i < moov_atom_size - 4; i++) {
-        atom_type = BE_32(&moov_atom[i]);
-        if (atom_type == STCO_ATOM) {
-            printf(" patching stco atom...\n");
-            atom_size = BE_32(&moov_atom[i - 4]);
-            if (atom_size < 16 || atom_size > moov_atom_size - i + 4) {
-                printf(" bad atom size\n");
-                goto error_out;
-            }
-            offset_count = BE_32(&moov_atom[i + 8]);
-            if (offset_count > (atom_size - 16) / 4) {
-                printf(" bad atom size/element count\n");
-                goto error_out;
-            }
-            for (j = 0; j < offset_count; j++) {
-                current_offset  = BE_32(&moov_atom[i + 12 + j * 4]);
-                current_offset += moov_atom_size;
-                moov_atom[i + 12 + j * 4 + 0] = (current_offset >> 24) & 0xFF;
-                moov_atom[i + 12 + j * 4 + 1] = (current_offset >> 16) & 0xFF;
-                moov_atom[i + 12 + j * 4 + 2] = (current_offset >>  8) & 0xFF;
-                moov_atom[i + 12 + j * 4 + 3] = (current_offset >>  0) & 0xFF;
-            }
-            i += atom_size - 4;
-        } else if (atom_type == CO64_ATOM) {
-            printf(" patching co64 atom...\n");
-            atom_size = BE_32(&moov_atom[i - 4]);
-            if (atom_size < 16 || atom_size > moov_atom_size - i + 4) {
-                printf(" bad atom size\n");
-                goto error_out;
-            }
-            offset_count = BE_32(&moov_atom[i + 8]);
-            if (offset_count > (atom_size - 16) / 8) {
-                printf(" bad atom size/element count\n");
         goto error_out;
     }
-            for (j = 0; j < offset_count; j++) {
-                current_offset  = BE_64(&moov_atom[i + 12 + j * 8]);
-                current_offset += moov_atom_size;
-                moov_atom[i + 12 + j * 8 + 0] = (current_offset >> 56) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 1] = (current_offset >> 48) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 2] = (current_offset >> 40) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 3] = (current_offset >> 32) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 4] = (current_offset >> 24) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 5] = (current_offset >> 16) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 6] = (current_offset >>  8) & 0xFF;
-                moov_atom[i + 12 + j * 8 + 7] = (current_offset >>  0) & 0xFF;
-            }
-            i += atom_size - 4;
-        }
-    }
 
     /* re-open the input file and open the output file */
     infile = fopen(argv[1], "rb");