10.07: annocheck: scan gaps for NOPS.

diff --git a/annobin-global.h b/annobin-global.h
index 54f0a6337eea3a4e1241e47c49c1fc52dea961a8..745ee6b4754354b028308e9144454bb706ecfafb 100644 (file)
--- a/annobin-global.h
+++ b/annobin-global.h
@@ -19,7 +19,7 @@ extern "C" {
 /* The version of the package.
    NB/ This number is expected to be in the form "NNnn" where
    "NN" is major version number and "nn" is the minor version number.  */
-#define ANNOBIN_VERSION 1006
+#define ANNOBIN_VERSION 1007
 
 /* The version of the annotation specification supported.  */
 #define SPEC_VERSION  3

diff --git a/annocheck/hardened.c b/annocheck/hardened.c
index 3f397d4c8ae4329fbd2177c1be10f0b22991f26e..24ae0a569d5b42b0294e6311578a4f1711662ebb 100644 (file)
--- a/annocheck/hardened.c
+++ b/annocheck/hardened.c
@@ -804,7 +804,7 @@ parse_dw_at_language (annocheck_data * data, Dwarf_Attribute * attr)
     case DW_LANG_C_plus_plus_11:
     case DW_LANG_C_plus_plus_14:
       if (! fixed_format_messages)
-       einfo (VERBOSE, "%s: info: Written in C++", get_filename (data));
+       einfo (VERBOSE2, "%s: info: Written in C++", get_filename (data));
       set_lang (data, LANG_CXX, SOURCE_DW_AT_LANGUAGE);
       break;
 
@@ -3528,6 +3528,40 @@ check_seg (annocheck_data *    data,
   return true;
 }
 
+static bool
+is_nop_byte (annocheck_data * data ATTRIBUTE_UNUSED,
+            unsigned char    byte,
+            uint             index,
+            ulong            addr_bias)
+{
+  switch (per_file.e_machine)
+    {
+    case EM_PPC64:
+      /* NOP = 60000000 */
+      return (((addr_bias + index) & 3) == 3) && byte == 0x60;
+
+    case EM_AARCH64:
+      /* NOP = d503201f */
+      switch ((addr_bias + index) & 3)
+       {
+       case 0: return byte == 0x1f;
+       case 1: return byte == 0x20;
+       case 2: return byte == 0x03;
+       case 3: return byte == 0xd5;
+       default: return false;
+       }
+
+    case EM_S390:
+      /* NOP = 47000000 */
+      return (((addr_bias + index) & 3) == 3) && byte == 0x47;
+      
+    default:
+      /* FIXME: Add support for other architectures.  */
+      /* FIXME: Add support for alternative endianness.  */
+      return false;
+    }
+}
+
 /* Returns true if GAP is one that can be ignored.  */
 
 static bool
@@ -3535,9 +3569,11 @@ ignore_gap (annocheck_data * data, note_range * gap)
 {
   Elf_Scn * addr1_scn = NULL;
   Elf_Scn * addr2_scn = NULL;
+  Elf_Scn * prev_scn = NULL;
   Elf_Scn * scn = NULL;
   ulong     scn_end = 0;
   ulong     scn_name = 0;
+  ulong     addr1_bias = 0;
 
   /* These tests should be redundant, but just in case...  */
   if (ignore_gaps)
@@ -3550,6 +3586,8 @@ ignore_gap (annocheck_data * data, note_range * gap)
       return true;
     }
 
+  einfo (VERBOSE2, "Consider gap %#lx..%#lx", gap->start, gap->end);
+
   /* Gaps narrower than the alignment of the .text section are assumed
      to be padding between functions, and so can be ignored.  In theory
      there could be executable code in such gaps, and so we should also
@@ -3562,6 +3600,8 @@ ignore_gap (annocheck_data * data, note_range * gap)
       return true;
     }
 
+  gap->start = align (gap->start, per_file.text_section_alignment);
+  
   /* FIXME: The linker can create fill regions in the map that are larger
      than the text section alignment.  Not sure why, but it does happen.
      (cf lconvert in the qt5-qttools package which has a gap of 0x28 bytes
@@ -3583,19 +3623,61 @@ ignore_gap (annocheck_data * data, note_range * gap)
       while ((scn = elf_nextscn (data->elf, scn)) != NULL)
        {
          Elf32_Shdr * shdr = elf32_getshdr (scn);
+         ulong sec_end = shdr->sh_addr + shdr->sh_size;
 
-         if (addr1_scn == NULL
-             && shdr->sh_addr <= gap->start && ((shdr->sh_addr + shdr->sh_size) >= gap->start))
-           addr1_scn = scn;
+         /* We are only interested in code sections.  */
+         if (shdr->sh_type != SHT_PROGBITS
+             || (shdr->sh_flags & (SHF_ALLOC | SHF_EXECINSTR)) != (SHF_ALLOC | SHF_EXECINSTR))
+           continue;
 
-         if (addr2_scn == NULL)
+         if ((shdr->sh_addr <= gap->start) && (gap->start < sec_end))
            {
-             scn_end = shdr->sh_addr + shdr->sh_size;
-             scn_name = shdr->sh_name;
+             /* Record any section as a first match.  */
+             if (addr1_scn == NULL)
+               {
+                 addr1_scn = scn;
+                 addr1_bias = gap->start - shdr->sh_addr;
+                 scn_name = shdr->sh_name;
+                 scn_end = sec_end;
+               }
+             else
+               {
+                 /* FIXME: Which section should we select ?  */
+                 einfo (VERBOSE2, "multiple code sections (%x+%x vs %x+%x) contain gap start",
+                        shdr->sh_addr, shdr->sh_size,
+                        elf32_getshdr (addr1_scn)->sh_addr,
+                        elf32_getshdr (addr1_scn)->sh_size
+                        );
+               }
+           }
 
-             if (shdr->sh_addr <= gap->end && scn_end >= gap->end)
+         if ((shdr->sh_addr < gap->end) && (gap->end < sec_end))
+           {
+             /* Record any section as a first match.  */
+             if (addr2_scn == NULL)
                addr2_scn = scn;
+             else
+               {
+                 /* FIXME: Which section should we select ?  */
+                 einfo (VERBOSE2, "multiple code sections (%x+%x vs %x+%x) contain gap end",
+                        shdr->sh_addr, shdr->sh_size,
+                        elf32_getshdr (addr1_scn)->sh_addr,
+                        elf32_getshdr (addr1_scn)->sh_size);
+               }
+           }
+         else if (shdr->sh_addr == gap->end)
+           {
+             /* This gap ends at the start of the current section.
+                So it probably matches the previous section.  */
+             if (addr2_scn == NULL
+                 && prev_scn != NULL
+                 && prev_scn == addr1_scn)
+               {
+                 addr2_scn = prev_scn;
+               }
            }
+
+         prev_scn = scn;
        }
     }
   else
@@ -3603,30 +3685,75 @@ ignore_gap (annocheck_data * data, note_range * gap)
       while ((scn = elf_nextscn (data->elf, scn)) != NULL)
        {
          Elf64_Shdr * shdr = elf64_getshdr (scn);
+         ulong sec_end = shdr->sh_addr + shdr->sh_size;
 
-         if (addr1_scn == NULL
-             && shdr->sh_addr <= gap->start && ((shdr->sh_addr + shdr->sh_size) >= gap->start))
-           addr1_scn = scn;
+         /* We are only interested in code sections.  */
+         if (shdr->sh_type != SHT_PROGBITS
+             || (shdr->sh_flags & (SHF_ALLOC | SHF_EXECINSTR)) != (SHF_ALLOC | SHF_EXECINSTR))
+           continue;
 
-         if (addr2_scn == NULL)
+         if ((shdr->sh_addr <= gap->start) && (gap->start < sec_end))
            {
-             scn_end = shdr->sh_addr + shdr->sh_size;
-             scn_name = shdr->sh_name;
+             /* Record any section as a first match.  */
+             if (addr1_scn == NULL)
+               {
+                 addr1_scn = scn;
+                 addr1_bias = gap->start - shdr->sh_addr;
+                 scn_name = shdr->sh_name;
+                 scn_end = sec_end;
+               }
+             else
+               {
+                 /* FIXME: Which section should we select ?  */
+                 einfo (VERBOSE2, "multiple code sections (%lx+%lx vs %lx+%lx) contain gap start",
+                        shdr->sh_addr, shdr->sh_size,
+                        elf64_getshdr (addr1_scn)->sh_addr,
+                        elf64_getshdr (addr1_scn)->sh_size
+                        );
+               }
+           }
 
-             if (shdr->sh_addr <= gap->end && scn_end >= gap->end)
+         if ((shdr->sh_addr < gap->end) && (gap->end < sec_end))
+           {
+             /* Record any section as a first match.  */
+             if (addr2_scn == NULL)
                addr2_scn = scn;
+             else
+               {
+                 /* FIXME: Which section should we select ?  */
+                 einfo (VERBOSE2, "multiple code sections (%lx+%lx vs %lx+%lx) contain gap end",
+                        shdr->sh_addr, shdr->sh_size,
+                        elf64_getshdr (addr1_scn)->sh_addr,
+                        elf64_getshdr (addr1_scn)->sh_size);
+               }
+           }
+         else if (shdr->sh_addr == gap->end)
+           {
+             /* This gap ends at the start of the current section.
+                So it probably matches the previous section.  */
+             if (addr2_scn == NULL
+                 && prev_scn != NULL
+                 && prev_scn == addr1_scn)
+               {
+                 addr2_scn = prev_scn;
+               }
            }
+
+         prev_scn = scn;
        }
     }
 
   /* If the gap is not inside one or more sections, then something funny has gone on...  */
-  if (addr2_scn == NULL)
-    return false;
+  if (addr1_scn == NULL || addr2_scn == NULL)
+    {
+      einfo (VERBOSE2, "gap is strange: it does not start and/or end in a section - ignoring");
+      return true;
+    }
 
   /* If the gap starts in one section, but ends in a different section then we ignore it.  */
   if (addr1_scn != addr2_scn)
     {
-      einfo (VERBOSE2, "gap ignored - crosses section boundary");
+      einfo (VERBOSE2, "gap ignored: crosses section boundary");
       return true;
     }
 
@@ -3645,22 +3772,49 @@ ignore_gap (annocheck_data * data, note_range * gap)
        {
          if (strstr (sym, "glink_PLTresolve") || strstr (sym, "@plt"))
            {
-             einfo (VERBOSE2, "Ignoring gap %lx..%lx at end of ppc64 .text section - it contains PLT stubs",
-                    gap->start, gap->end);
+             einfo (VERBOSE2, "Ignoring gap at end of ppc64 .text section - it contains PLT stubs");
              return true;
            }
          else
-           einfo (VERBOSE2, "Potential PLT stub gap contains the symbol '%s', so the gap is not ignored", sym);
+           {
+             einfo (VERBOSE2, "Potential PLT stub gap contains the symbol '%s', so the gap is not ignored", sym);
+             return false;
+           }
        }
       else
        {
          /* Without symbol information we cannot be sure, but it is a reasonable supposition.  */
-         einfo (VERBOSE2, "Ignoring gap %lx..%lx at end of ppc64 .text section - it will contain PLT stubs",
-                gap->start, gap->end);
+         einfo (VERBOSE2, "Ignoring gap at end of ppc64 .text section - it will contain PLT stubs");
          return true;
        }
     }
 
+  /* Scan the contents of the gap.  If it is all zeroes or NOP instructions, then it can be ignored.  */
+  Elf_Data * sec_data;
+  sec_data = elf_getdata (addr1_scn, NULL);
+  /* Paranoia checks.  */
+  if (sec_data == NULL
+      || sec_data->d_off != 0
+      || sec_data->d_type != ELF_T_BYTE
+      || gap->start < addr1_bias /* This should never happen.  */
+      || (gap->end - addr1_bias) >= sec_data->d_size) /* Nor should this.  */
+    {
+      einfo (VERBOSE2, "could not check gap for NOPs!");
+      return false;
+    }
+
+  unsigned char * sec_bytes = ((unsigned char *) sec_data->d_buf) + addr1_bias;
+  uint i;
+  for (i = gap->end - gap->start; i--;)
+    if (sec_bytes[i] != 0 && ! is_nop_byte (data, sec_bytes[i], i, addr1_bias))
+      break;
+
+  if (i == (uint) -1)
+    {
+      einfo (VERBOSE2, "gap ignored - it contains padding and/or NOP instructions");
+      return true;
+    }
+
   return false;
 }