Use chain-loading for Linux nacl_helper

chrome/nacl/nacl_helper_bootstrap_linux.x

Index: chrome/nacl/nacl_helper_bootstrap_linux.x
diff --git a/chrome/nacl/nacl_helper_bootstrap_linux.x b/chrome/nacl/nacl_helper_bootstrap_linux.x
new file mode 100644
index 0000000000000000000000000000000000000000..45a239602b20a4f4a0ff7b38c2beff154685c2f9
--- /dev/null
+++ b/chrome/nacl/nacl_helper_bootstrap_linux.x
@@ -0,0 +1,81 @@
+/* Copyright (c) 2011 The Chromium Authors. All rights reserved.
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ *
+ * This is a custom linker script used to build nacl_helper_bootstrap.
+ * It has a very special layout.  This script will only work with input
+ * that is kept extremely minimal.  If there are unexpected input sections
+ * not named here, the result will not be correct.
+ */
+
+/*
+ * Set the entry point to the symbol called _start, which we define in assembly.
+ */
+ENTRY(_start)
+
+/*
+ * This is the address where the program text starts.
+ * We set this as low as we think we can get away with.
+ * The common settings for sysctl vm.mmap_min_addr range from 4k to 64k.
+ */
+TEXT_START = 0x10000;
+
+/*
+ * This is the top of the range we are trying to reserve, which is 1G
+ * for x86-32 and ARM.  For an x86-64 zero-based sandbox, this really
+ * needs to be 36G.
+ */
+RESERVE_TOP = 1 << 30;
+
+/*
+ * We specify the program headers we want explicitly, to get the layout
+ * exactly right and to give the "reserve" segment p_flags of zero, so
+ * that it gets mapped as PROT_NONE.
+ */
+PHDRS {
+  text PT_LOAD FILEHDR PHDRS;
+  reserve PT_LOAD FLAGS(0);
+  stack PT_GNU_STACK FLAGS(6);	/* RW, no E */
+}
+
+/*
+ * Now we lay out the sections across those segments.
+ */
+SECTIONS {
+  /*
+   * Here is the program itself.
+   */
+  .text TEXT_START + SIZEOF_HEADERS : {
+    *(.note.gnu.build-id)
+    *(.text*)
+    *(.rodata*)
+    *(.eh_frame*)
+  } :text
+  etext = .;
+
+  /*
+   * Now we move up to the next p_align increment, and place the dummy
+   * segment there.  The linker emits this segment with the p_vaddr and
+   * p_memsz we want, which reserves the address space.  But the linker
+   * gives it a p_filesz of zero.  We have to edit the phdr after link
+   * time to give it a p_filesz matching its p_memsz.  That way, the
+   * kernel doesn't think we are preallocating a huge amount of memory.
+   * It just maps it from the file, i.e. way off the end of the file,

[Brad Chen, 2011/08/29 23:04:57]

Is this well-defined kernel behavior? I know you've checked the source, but it
would be nice to know it was specified someplace or that lots of people relied
on this behavior to be confident it won't change.

+   * which is perfect for reserving the address space.
+   */
+  . = ALIGN(CONSTANT(COMMONPAGESIZE));
+  RESERVE_START = .;
+  .reserve  : {
+    . = RESERVE_TOP - RESERVE_START;
+  } :reserve
+
+  /*
+   * These are empty input sections the linker generates.
+   * If we don't discard them, they pollute the flags in the output segment.
+   */
+  /DISCARD/ : {
+    *(.iplt)
+    *(.rel*)
+    *(.igot.plt)
+  }
+}