PR4886: check build-id if able.

This provides sanity check of debuginfo file based on build-id. Many cases are considered, whether build-id exists in debuginfo file or not, whether module is loaded or not, whether build-id exists in runtime kernel/module. It will do work when LD >= 2.18 and kernel >=2.6.23, otherwise no check.

7 files changed, 147 insertions, 1 deletions

diff --git a/ChangeLog b/ChangeLog
index cc9dbd3d..8b1645e4 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2008-10-06  Wenji Huang  <wenji.huang@oracle.com>
+
+	PR 4886
+	* translate.cxx (dump_unwindsyms): Extract build-id from debuginfo.
+	(emit_module_init): Check build-id if able.
+
 2008-10-03  Jim Keniston  <jkenisto@us.ibm.com>
 
 	PR 6850
diff --git a/runtime/ChangeLog b/runtime/ChangeLog
index 44147429..0a12c0d3 100644
--- a/runtime/ChangeLog
+++ b/runtime/ChangeLog
@@ -1,3 +1,13 @@
+2008-10-06  Wenji Huang  <wenji.huang@oracle.com>
+
+        PR 4886
+        * sym.h (_stp_module): Add several fields to store build-id.
+        * sym.c (_stp_module_check): New function to check build-id.
+        * staprun/staprun.c (send_relocation_kernel): Send build-id of runtime
+	kernel if there.
+        * transport/symbols.c (_stp_do_relocation): Update note section.
+	(_stp_section_is_interesting): No filter to .note.gnu.build-id.
+
 2008-10-03  Mark Wielaard  <mjw@redhat.com>
 
 	* procfs.c (_stp_rmdir_proc_module): Remove debug output from warning.
diff --git a/runtime/staprun/staprun.c b/runtime/staprun/staprun.c
index f8a08876..d23945e6 100644
--- a/runtime/staprun/staprun.c
+++ b/runtime/staprun/staprun.c
@@ -370,6 +370,11 @@ int send_relocation_kernel ()
         }
       fclose (kallsyms);
       if (!done_with_kallsyms) srkrc = -1;
+      /* detect note section, send flag if there 
+       * NB: address=2 represents existed note, the real one in _stp_module
+       */ 
+      if (srkrc != -1 && !access("/sys/kernel/notes", R_OK))
+	 send_a_relocation ("kernel", ".note.gnu.build-id", 2);
     }
 
   return srkrc;
diff --git a/runtime/sym.c b/runtime/sym.c
index 1a9e26b2..97243929 100644
--- a/runtime/sym.c
+++ b/runtime/sym.c
@@ -160,6 +160,72 @@ static const char *_stp_kallsyms_lookup(unsigned long addr, unsigned long *symbo
 	return NULL;
 }
 
+/* Validate module/kernel based on build-id if there 
+*  The completed case is the following combination:
+*	   Debuginfo 		 Module			         Kernel	
+* 			   X				X
+* 	has build-id/not	unloaded		      has build-id/not	
+*				loaded && (has build-id/not)  
+*
+*  NB: build-id exists only if ld>=2.18 and kernel>= 2.6.23
+*/
+static int _stp_module_check(void)
+{
+	struct _stp_module *m = NULL;
+	unsigned long notes_addr, base_addr;
+	unsigned i,j;
+
+	for (i = 0; i < _stp_num_modules; i++)
+	{
+		m = _stp_modules[i];			
+
+		/* unloaded module */
+		if (m->notes_sect == 0) {
+              	     _stp_warn("skip checking %s\n", m->name);
+                    continue;
+                }
+		if (m->build_id_len > 0) { /* build-id in debuginfo file */
+		    dbug_sym(1, "validate %s based on build-id\n", m->name);
+
+		    /* loaded module/kernel, but without build-id */
+		    if (m->notes_sect == 1) {
+		   	_stp_error("missing build-id in %s\n", m->name);
+			return 1;
+		    } 
+		    /* notes end address */
+		    if (!strcmp(m->name, "kernel")) {
+		  	  notes_addr = m->build_id_offset;
+			  base_addr = _stp_module_relocate("kernel",
+							   "_stext", 0);
+                    } else {
+			  notes_addr = m->notes_sect + m->build_id_offset;
+			  base_addr = m->notes_sect;
+		    }
+		    /* notes start address */
+		    notes_addr -= m->build_id_len;
+		    if (notes_addr > base_addr) {
+		      for (j = 0; j < m->build_id_len; j++)	 
+		        if (*((unsigned char *) notes_addr+j) != 
+							*(m->build_id_bits+j)) 
+			{
+			  _stp_error("inconsistent bit (0x%x [%s] vs 0x%x [debuginfo]) of build-id\n", *((unsigned char *) notes_addr+j), m->name, *(m->build_id_bits+j));
+			  return 1;
+			} 
+		    } else { /* bug, shouldn't come here */
+			     _stp_error("unknown failure in checking %s\n", 
+								m->name);
+			     return 1;
+		    	   } /* end comparing */
+		} else { 
+			  /* build-id in module/kernel, absent in debuginfo */
+			  if (m->notes_sect > 1) {
+			    _stp_error("unexpected build-id in %s\n", m->name);
+			    return 1;
+			  }
+		 } /* end checking */
+	} /* end loop */
+	return 0;
+}
 
 /** Print an address symbolically.
  * @param address The address to lookup.
diff --git a/runtime/sym.h b/runtime/sym.h
index 5888d2c7..d23c1632 100644
--- a/runtime/sym.h
+++ b/runtime/sym.h
@@ -45,6 +45,16 @@ struct _stp_module {
 	uint32_t unwind_data_len;
 	uint32_t unwind_hdr_len;
 	uint32_t unwind_is_ehframe; /* unwind data comes from .eh_frame */
+	/* build-id information */
+	unsigned char *build_id_bits;
+	unsigned long  build_id_offset;
+	unsigned long  notes_sect; /* kernel: 1 - no build-id 
+				    *  	      2 - has build-id 
+				    * module: 0 - unloaded
+				    *	      1 - loaded and no build-id
+				    *	      Other - note section address  
+				   */
+	int build_id_len;
 };
 
 
diff --git a/runtime/transport/symbols.c b/runtime/transport/symbols.c
index 4bdd0904..faba0986 100644
--- a/runtime/transport/symbols.c
+++ b/runtime/transport/symbols.c
@@ -51,8 +51,16 @@ static void _stp_do_relocation(const char __user *buf, size_t count)
       if (strcmp (_stp_modules[mi]->name, msg.module))
         continue;
 
+      /* update note section to represent loaded */
+      if (_stp_modules[mi]->notes_sect == 0)
+	 _stp_modules[mi]->notes_sect = 1;	
       for (si=0; si<_stp_modules[mi]->num_sections; si++)
         {
+  	   if (!strcmp (".note.gnu.build-id", msg.reloc)) {
+		_stp_modules[mi]->notes_sect = msg.address;	
+		continue;
+	   }
+
           if (strcmp (_stp_modules[mi]->sections[si].name, msg.reloc))
             continue;
 
@@ -134,7 +142,8 @@ static int _stp_section_is_interesting(const char *name)
 {
 	int ret = 1;
 	if (!strncmp("__", name, 2)
-	    || !strncmp(".note", name, 5)
+	    || (!strncmp(".note", name, 5) 
+		&& strncmp(".note.gnu.build-id", name, 18))
 	    || !strncmp(".gnu", name, 4)
 	    || !strncmp(".mod", name, 4))
 		ret = 0;
diff --git a/translate.cxx b/translate.cxx
index e92c8483..0ee51792 100644
--- a/translate.cxx
+++ b/translate.cxx
@@ -1113,6 +1113,7 @@ c_unparser::emit_module_init ()
   o->newline(-1) << "}";
 
   // XXX: perform buildid-based checking if able
+  o->newline() << "if (_stp_module_check()) rc = -EINVAL;";
 
   o->newline(-1) << "}";
   o->newline() << "if (rc) goto out;";
@@ -4413,6 +4414,23 @@ dump_unwindsyms (Dwfl_Module *m,
   int syments = dwfl_module_getsymtab(m);
   assert(syments);
 
+  //extract build-id from debuginfo file
+  int build_id_len = 0;
+  unsigned char *build_id_bits;
+  GElf_Addr build_id_vaddr;
+
+  if ((build_id_len=dwfl_module_build_id(m,
+                                        (const unsigned char **)&build_id_bits,
+                                         &build_id_vaddr)) > 0)
+  {
+        if (c->session.verbose > 1) {
+           clog << "Found build-id in " << name
+                << ", length " << build_id_len;
+           clog << ", end at 0x" << hex << build_id_vaddr
+                << dec << endl;
+        }
+  }
+
   // Look up the relocation basis for symbols
   int n = dwfl_module_relocations (m);
 
@@ -4581,6 +4599,28 @@ dump_unwindsyms (Dwfl_Module *m,
   c->output << ".num_sections = sizeof(_stp_module_" << stpmod_idx << "_sections)/"
             << "sizeof(struct _stp_section), " << endl;
 
+  if (build_id_len > 0) {
+        c->output << ".build_id_bits = \"" ;
+        for (int j=0; j<build_id_len;j++)
+                c->output << "\\x" << hex
+                          << (unsigned short) *(build_id_bits+j) << dec;
+
+       c->output << "\", " << endl;
+       c->output << ".build_id_len = " << build_id_len << ", " << endl;
+
+       if (modname == "kernel")
+                c->output << ".build_id_offset = 0x" << hex << build_id_vaddr
+			  << dec << ", " << endl;
+       else
+                c->output << ".build_id_offset = 0x" << hex
+			  << build_id_vaddr - base
+ 			  << dec << ", " << endl;
+  } else
+        c->output << ".build_id_len = 0, " << endl;
+
+  //initialize the note section representing unloaded
+  c->output << ".notes_sect = 0," << endl; 
+
   c->output << "};" << endl << endl;
 
   c->undone_unwindsym_modules.erase (modname);