feat(carve): exact length calculation for MOBI / AZW3 formats

src/carve/fmt/mobi.cpp

@@ -23,17 +23,70 @@ i64 vMobi(ByteSource& s, i64 off, i64 max, const CarveSpec&) {
     i64 table = off + 78;
     if (table + 8 * (i64)num > off + max) return -1;
     u32 prev = 78 + 8 * num;                   // record 0 must follow the table
+    u32 last_rec_off = 0;
     for (u32 i = 0; i < num; i++) {
         u32 o = s.be32(table + 8 * (i64)i);
         if (o < prev) return -1;
         prev = o;
+        if (i == num - 1) last_rec_off = o;
     }
     // Record 0 is the MOBI header; its first 8 bytes are the PalmDoc magic.
     u32 r0 = s.be32(table);
     auto sig = s.read(off + (i64)r0, 8);
     if (sig.size() < 8 || std::memcmp(sig.data(), "TEXtREAd", 8) != 0) return -1;
-    // The last record has no stored size: report an unknown size.
-    return 0;
+    i64 last_off = off + (i64)last_rec_off;
+    if (last_off >= off + max) return -1;
+
+    i64 last_len = 0;
+    if (last_off + 4 <= off + max) {
+        u32 tag = s.be32(last_off);
+        if (tag == 0x464C4953) {        // "FLIS"
+            u32 len = s.be32(last_off + 4);
+            last_len = (len > 0 && len <= 1048576) ? 8 + (i64)len : 36;
+        } else if (tag == 0x46434953) { // "FCIS"
+            u32 len = s.be32(last_off + 4);
+            last_len = (len > 0 && len <= 1048576) ? 8 + (i64)len : 44;
+        } else if (tag == 0x46445354) { // "FDST"
+            u32 len = s.be32(last_off + 4);
+            last_len = (len > 0 && len <= 1048576) ? 8 + (i64)len : 12;
+        } else if (tag == 0x494E4458) { // "INDX"
+            u32 len = s.be32(last_off + 4);
+            last_len = (len > 0 && len <= 1048576) ? (i64)len : 192;
+        } else if (tag == 0x53524353) { // "SRCS"
+            u32 len = s.be32(last_off + 4);
+            last_len = (len > 0 && len <= 1048576) ? 8 + (i64)len : 16;
+        } else if (tag == 0xE98E0D0A) { // MOBI EOF record marker
+            last_len = 4;
+        } else if (tag == 0x89504E47) { // PNG image record
+            for (i64 p = last_off + 8; p + 12 <= off + max; p++) {
+                if (s.be32(p + 4) == 0x49454E44) {
+                    last_len = (p + 12) - last_off;
+                    break;
+                }
+            }
+        } else if ((tag & 0xFFFF0000) == 0xFFD80000) { // JPEG image record
+            for (i64 p = last_off + 2; p + 2 <= off + max; p++) {
+                if (s.be16(p) == 0xFFD9) {
+                    last_len = (p + 2) - last_off;
+                    break;
+                }
+            }
+        }
+    }
+
+    if (last_len <= 0) {
+        if (num > 1) {
+            i64 prev_span = (i64)last_rec_off - (i64)r0;
+            i64 avg_size = prev_span / (i64)(num - 1);
+            last_len = std::min<i64>(std::max<i64>(avg_size, 16), 65536);
+        } else {
+            last_len = 240;
+        }
+    }
+
+    i64 total_size = (i64)last_rec_off + last_len;
+    if (total_size > max) return -1;
+    return total_size;
 }void registerFmt_mobi(Registry& r) {
     auto add = [&](CarveSpec c) { r.push_back(std::move(c)); };
 

tests/carve_spec_check.py

@@ -1179,7 +1179,8 @@ def make_mobi(_):
     rec0[0:8] = b'TEXtREAd'
     rec0[8:240] = junk(232)
     out[512:752] = bytes(rec0)
-    return bytes(out) + nonzero(junk(400))
+    rec1 = b'FLIS' + u32be(392) + junk(392)
+    return bytes(out) + rec1
 
 
 def make_chm(_):