Fix stack overflow due to large AF_INET6 requests

Resolves #16072 (CVE-2013-4458).

This patch fixes another stack overflow in getaddrinfo when it is
called with AF_INET6.  The AF_UNSPEC case was fixed as CVE-2013-1914,
but the AF_INET6 case went undetected back then.

(cherry picked from commit 7cbcdb3699)

Conflicts:
	NEWS

ChangeLog

@@ -1,3 +1,121 @@
+2013-10-25  Siddhesh Poyarekar  <siddhesh@redhat.com>
+
+	[BZ #16072]
+	* sysdeps/posix/getaddrinfo.c (gethosts): Allocate tmpbuf on
+	heap for large requests.
+
+2013-10-25  Aurelien Jarno  <aurelien@aurel32.net>
+
+	[BZ #9954]
+	* sysdeps/posix/getaddrinfo.c (rfc3484_sort): do not assign native
+	result if the result has no associated interface.
+	* sysdeps/posix/getaddrinfo.c (getaddrinfo): correctly detect
+	interface for all 127.X.Y.Z addresses.
+
+2013-09-23  Carlos O'Donell  <carlos@redhat.com>
+
+	[BZ #15754]
+	* sysdeps/generic/stackguard-macros.h: If PTRGUARD_LOCAL use
+	__pointer_chk_guard_local, otherwise __pointer_chk_guard.
+	* elf/Makefile: Define CFLAGS-tst-ptrguard1-static.c.
+
+2013-09-23  Carlos O'Donell  <carlos@redhat.com>
+
+	[BZ #15754]
+	* elf/Makefile (tests): Add tst-ptrguard1.
+	(tests-static): Add tst-ptrguard1-static.
+	(tst-ptrguard1-ARGS): Define.
+	(tst-ptrguard1-static-ARGS): Define.
+	* elf/tst-ptrguard1.c: New file.
+	* elf/tst-ptrguard1-static.c: New file.
+	* sysdeps/x86_64/stackguard-macros.h: Define POINTER_CHK_GUARD.
+	* sysdeps/i386/stackguard-macros.h: Likewise.
+	* sysdeps/powerpc/powerpc32/stackguard-macros.h: Likewise.
+	* sysdeps/powerpc/powerpc64/stackguard-macros.h: Likewise.
+	* sysdeps/s390/s390-32/stackguard-macros.h: Likewise.
+	* sysdeps/s390/s390-64/stackguard-macros.h: Likewise.
+	* sysdeps/sparc/sparc32/stackguard-macros.h: Likewise.
+	* sysdeps/sparc/sparc64/stackguard-macros.h: Likewise.
+
+2013-09-23  Hector Marco  <hecmargi@upv.es>
+	    Ismael Ripoll  <iripoll@disca.upv.es>
+	    Carlos O'Donell  <carlos@redhat.com>
+
+	[BZ #15754]
+	* sysdeps/generic/stackguard-macros.h: Define
+	__pointer_chk_guard_local and POINTER_CHK_GUARD.
+	* csu/libc-start.c [!SHARED && !THREAD_SET_POINTER_GUARD]:
+	Define __pointer_chk_guard_local.
+	(LIBC_START_MAIN) [!SHARED]: Call _dl_setup_pointer_guard.
+	Use THREAD_SET_POINTER_GUARD or set __pointer_chk_guard_local.
+
+2013-09-23  Siddhesh Poyarekar  <siddhesh@redhat.com>
+
+	[BZ #14547]
+	* string/tst-strcoll-overflow.c: New test case.
+	* string/Makefile (xtests): Add tst-strcoll-overflow.
+	* string/strcoll_l.c (STRCOLL): Skip allocating memory for
+	cache if string sizes may cause integer overflow.
+
+2013-09-23  Siddhesh Poyarekar  <siddhesh@redhat.com>
+
+	[BZ #14547]
+	* string/strcoll_l.c (coll_seq): New members rule, idx,
+	save_idx and back_us.
+	(get_next_seq_nocache): New function.
+	(do_compare_nocache): New function.
+	(STRCOLL): Use get_next_seq_nocache and do_compare_nocache
+	when malloc fails.
+
+2013-08-20  Siddhesh Poyarekar  <siddhesh@redhat.com>
+
+	* string/strcoll_l.c (coll_seq): New structure.
+	(get_next_seq_cached): New function.
+	(get_next_seq): New function.
+	(do_compare): New function.
+	(STRCOLL): Use GNU style definition.  Simplify implementation
+	by using get_next_seq, get_next_seq_cached and do_compare.
+
+2013-09-11  Will Newton  <will.newton@linaro.org>
+
+	[BZ #15857]
+	* malloc/malloc.c (__libc_memalign): Check the value of bytes
+	does not overflow.
+
+2013-09-11  Will Newton  <will.newton@linaro.org>
+
+	[BZ #15856]
+	* malloc/malloc.c (__libc_valloc): Check the value of bytes
+	does not overflow.
+
+2013-09-11  Will Newton  <will.newton@linaro.org>
+
+	[BZ #15855]
+	* malloc/malloc.c (__libc_pvalloc): Check the value of bytes
+	does not overflow.
+
+2013-08-16  Florian Weimer  <fweimer@redhat.com>
+
+	[BZ #14699]
+	CVE-2013-4237
+	* sysdeps/posix/dirstream.h (struct __dirstream): Add errcode
+	member.
+	* sysdeps/posix/opendir.c (__alloc_dir): Initialize errcode
+	member.
+	* sysdeps/posix/rewinddir.c (rewinddir): Reset errcode member.
+	* sysdeps/posix/readdir_r.c (__READDIR_R): Enforce NAME_MAX limit.
+	Return delayed error code.  Remove GETDENTS_64BIT_ALIGNED
+	conditional.
+	* sysdeps/unix/sysv/linux/wordsize-64/readdir_r.c: Do not define
+	GETDENTS_64BIT_ALIGNED.
+	* sysdeps/unix/sysv/linux/i386/readdir64_r.c: Likewise.
+	* manual/filesys.texi (Reading/Closing Directory): Document
+	ENAMETOOLONG return value of readdir_r.  Recommend readdir more
+	strongly.
+	* manual/conf.texi (Limits for Files): Add portability note to
+	NAME_MAX, PATH_MAX.
+	(Pathconf): Add portability note for _PC_NAME_MAX, _PC_PATH_MAX.
+
 2013-08-03  David S. Miller  <davem@davemloft.net>
 
 	* po/ko.po: Update Korean translation from translation project.

csu/libc-start.c

@@ -37,6 +37,12 @@ extern void __pthread_initialize_minimal (void);
    in thread local area.  */
 uintptr_t __stack_chk_guard attribute_relro;
 # endif
+# ifndef  THREAD_SET_POINTER_GUARD
+/* Only exported for architectures that don't store the pointer guard
+   value in thread local area.  */
+uintptr_t __pointer_chk_guard_local
+	attribute_relro attribute_hidden __attribute__ ((nocommon));
+# endif
 #endif
 
 #ifdef HAVE_PTR_NTHREADS
@@ -195,6 +201,16 @@ LIBC_START_MAIN (int (*main) (int, char **, char ** MAIN_AUXVEC_DECL),
 # else
   __stack_chk_guard = stack_chk_guard;
 # endif
+
+  /* Set up the pointer guard value.  */
+  uintptr_t pointer_chk_guard = _dl_setup_pointer_guard (_dl_random,
+							 stack_chk_guard);
+# ifdef THREAD_SET_POINTER_GUARD
+  THREAD_SET_POINTER_GUARD (pointer_chk_guard);
+# else
+  __pointer_chk_guard_local = pointer_chk_guard;
+# endif
+
 #endif
 
   /* Register the destructor of the dynamic linker if there is any.  */

elf/Makefile

@@ -121,7 +121,8 @@ endif
 tests = tst-tls1 tst-tls2 tst-tls9 tst-leaks1 \
 	tst-array1 tst-array2 tst-array3 tst-array4 tst-array5
 tests-static = tst-tls1-static tst-tls2-static tst-stackguard1-static \
-	       tst-leaks1-static tst-array1-static tst-array5-static
+	       tst-leaks1-static tst-array1-static tst-array5-static \
+	       tst-ptrguard1-static
 ifeq (yes,$(build-shared))
 tests-static += tst-tls9-static
 tst-tls9-static-ENV = \
@@ -145,7 +146,8 @@ tests += loadtest restest1 preloadtest loadfail multiload origtest resolvfail \
 	 tst-audit1 tst-audit2 tst-audit8 \
 	 tst-stackguard1 tst-addr1 tst-thrlock \
 	 tst-unique1 tst-unique2 tst-unique3 tst-unique4 \
-	 tst-initorder tst-initorder2 tst-relsort1 tst-null-argv
+	 tst-initorder tst-initorder2 tst-relsort1 tst-null-argv \
+	 tst-ptrguard1
 #	 reldep9
 test-srcs = tst-pathopt
 selinux-enabled := $(shell cat /selinux/enforce 2> /dev/null)
@@ -1016,6 +1018,12 @@ LDFLAGS-order2mod2.so = $(no-as-needed)
 tst-stackguard1-ARGS = --command "$(host-test-program-cmd) --child"
 tst-stackguard1-static-ARGS = --command "$(objpfx)tst-stackguard1-static --child"
 
+tst-ptrguard1-ARGS = --command "$(host-test-program-cmd) --child"
+# When built statically, the pointer guard interface uses
+# __pointer_chk_guard_local.
+CFLAGS-tst-ptrguard1-static.c = -DPTRGUARD_LOCAL
+tst-ptrguard1-static-ARGS = --command "$(objpfx)tst-ptrguard1-static --child"
+
 $(objpfx)tst-leaks1: $(libdl)
 $(objpfx)tst-leaks1-mem: $(objpfx)tst-leaks1.out
 	$(common-objpfx)malloc/mtrace $(objpfx)tst-leaks1.mtrace > $@

elf/tst-ptrguard1-static.c

@@ -0,0 +1 @@
+#include "tst-ptrguard1.c"

elf/tst-ptrguard1.c

@@ -0,0 +1,202 @@
+/* Copyright (C) 2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/wait.h>
+#include <stackguard-macros.h>
+#include <tls.h>
+#include <unistd.h>
+
+#ifndef POINTER_CHK_GUARD
+extern uintptr_t __pointer_chk_guard;
+# define POINTER_CHK_GUARD __pointer_chk_guard
+#endif
+
+static const char *command;
+static bool child;
+static uintptr_t ptr_chk_guard_copy;
+static bool ptr_chk_guard_copy_set;
+static int fds[2];
+
+static void __attribute__ ((constructor))
+con (void)
+{
+  ptr_chk_guard_copy = POINTER_CHK_GUARD;
+  ptr_chk_guard_copy_set = true;
+}
+
+static int
+uintptr_t_cmp (const void *a, const void *b)
+{
+  if (*(uintptr_t *) a < *(uintptr_t *) b)
+    return 1;
+  if (*(uintptr_t *) a > *(uintptr_t *) b)
+    return -1;
+  return 0;
+}
+
+static int
+do_test (void)
+{
+  if (!ptr_chk_guard_copy_set)
+    {
+      puts ("constructor has not been run");
+      return 1;
+    }
+
+  if (ptr_chk_guard_copy != POINTER_CHK_GUARD)
+    {
+      puts ("POINTER_CHK_GUARD changed between constructor and do_test");
+      return 1;
+    }
+
+  if (child)
+    {
+      write (2, &ptr_chk_guard_copy, sizeof (ptr_chk_guard_copy));
+      return 0;
+    }
+
+  if (command == NULL)
+    {
+      puts ("missing --command or --child argument");
+      return 1;
+    }
+
+#define N 16
+  uintptr_t child_ptr_chk_guards[N + 1];
+  child_ptr_chk_guards[N] = ptr_chk_guard_copy;
+  int i;
+  for (i = 0; i < N; ++i)
+    {
+      if (pipe (fds) < 0)
+	{
+	  printf ("couldn't create pipe: %m\n");
+	  return 1;
+	}
+
+      pid_t pid = fork ();
+      if (pid < 0)
+	{
+	  printf ("fork failed: %m\n");
+	  return 1;
+	}
+
+      if (!pid)
+	{
+	  if (ptr_chk_guard_copy != POINTER_CHK_GUARD)
+	    {
+	      puts ("POINTER_CHK_GUARD changed after fork");
+	      exit (1);
+	    }
+
+	  close (fds[0]);
+	  close (2);
+	  dup2 (fds[1], 2);
+	  close (fds[1]);
+
+	  system (command);
+	  exit (0);
+	}
+
+      close (fds[1]);
+
+      if (TEMP_FAILURE_RETRY (read (fds[0], &child_ptr_chk_guards[i],
+				    sizeof (uintptr_t))) != sizeof (uintptr_t))
+	{
+	  puts ("could not read ptr_chk_guard value from child");
+	  return 1;
+	}
+
+      close (fds[0]);
+
+      pid_t termpid;
+      int status;
+      termpid = TEMP_FAILURE_RETRY (waitpid (pid, &status, 0));
+      if (termpid == -1)
+	{
+	  printf ("waitpid failed: %m\n");
+	  return 1;
+	}
+      else if (termpid != pid)
+	{
+	  printf ("waitpid returned %ld != %ld\n",
+		  (long int) termpid, (long int) pid);
+	  return 1;
+	}
+      else if (!WIFEXITED (status) || WEXITSTATUS (status))
+	{
+	  puts ("child hasn't exited with exit status 0");
+	  return 1;
+	}
+    }
+
+  qsort (child_ptr_chk_guards, N + 1, sizeof (uintptr_t), uintptr_t_cmp);
+
+  /* The default pointer guard is the same as the default stack guard.
+     They are only set to default if dl_random is NULL.  */
+  uintptr_t default_guard = 0;
+  unsigned char *p = (unsigned char *) &default_guard;
+  p[sizeof (uintptr_t) - 1] = 255;
+  p[sizeof (uintptr_t) - 2] = '\n';
+  p[0] = 0;
+
+  /* Test if the pointer guard canaries are either randomized,
+     or equal to the default pointer guard value.
+     Even with randomized pointer guards it might happen
+     that the random number generator generates the same
+     values, but if that happens in more than half from
+     the 16 runs, something is very wrong.  */
+  int ndifferences = 0;
+  int ndefaults = 0;
+  for (i = 0; i < N; ++i)
+    {
+      if (child_ptr_chk_guards[i] != child_ptr_chk_guards[i+1])
+	ndifferences++;
+      else if (child_ptr_chk_guards[i] == default_guard)
+	ndefaults++;
+    }
+
+  printf ("differences %d defaults %d\n", ndifferences, ndefaults);
+
+  if (ndifferences < N / 2 && ndefaults < N / 2)
+    {
+      puts ("pointer guard values are not randomized enough");
+      puts ("nor equal to the default value");
+      return 1;
+    }
+
+  return 0;
+}
+
+#define OPT_COMMAND	10000
+#define OPT_CHILD	10001
+#define CMDLINE_OPTIONS	\
+  { "command", required_argument, NULL, OPT_COMMAND },  \
+  { "child", no_argument, NULL, OPT_CHILD },
+#define CMDLINE_PROCESS	\
+  case OPT_COMMAND:	\
+    command = optarg;	\
+    break;		\
+  case OPT_CHILD:	\
+    child = true;	\
+    break;
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"

malloc/malloc.c

@@ -3015,6 +3015,13 @@ __libc_memalign(size_t alignment, size_t bytes)
   /* Otherwise, ensure that it is at least a minimum chunk size */
   if (alignment <  MINSIZE) alignment = MINSIZE;
 
+  /* Check for overflow.  */
+  if (bytes > SIZE_MAX - alignment - MINSIZE)
+    {
+      __set_errno (ENOMEM);
+      return 0;
+    }
+
   arena_get(ar_ptr, bytes + alignment + MINSIZE);
   if(!ar_ptr)
     return 0;
@@ -3046,6 +3053,13 @@ __libc_valloc(size_t bytes)
 
   size_t pagesz = GLRO(dl_pagesize);
 
+  /* Check for overflow.  */
+  if (bytes > SIZE_MAX - pagesz - MINSIZE)
+    {
+      __set_errno (ENOMEM);
+      return 0;
+    }
+
   void *(*hook) (size_t, size_t, const void *) =
     force_reg (__memalign_hook);
   if (__builtin_expect (hook != NULL, 0))
@@ -3082,6 +3096,13 @@ __libc_pvalloc(size_t bytes)
   size_t page_mask = GLRO(dl_pagesize) - 1;
   size_t rounded_bytes = (bytes + page_mask) & ~(page_mask);
 
+  /* Check for overflow.  */
+  if (bytes > SIZE_MAX - 2*pagesz - MINSIZE)
+    {
+      __set_errno (ENOMEM);
+      return 0;
+    }
+
   void *(*hook) (size_t, size_t, const void *) =
     force_reg (__memalign_hook);
   if (__builtin_expect (hook != NULL, 0))

manual/conf.texi

@@ -1149,6 +1149,9 @@ typed ahead as input.  @xref{I/O Queues}.
 @deftypevr Macro int NAME_MAX
 The uniform system limit (if any) for the length of a file name component, not
 including the terminating null character.
+
+@strong{Portability Note:} On some systems, @theglibc{} defines
+@code{NAME_MAX}, but does not actually enforce this limit.
 @end deftypevr
 
 @comment limits.h
@@ -1157,6 +1160,9 @@ including the terminating null character.
 The uniform system limit (if any) for the length of an entire file name (that
 is, the argument given to system calls such as @code{open}), including the
 terminating null character.
+
+@strong{Portability Note:} @Theglibc{} does not enforce this limit
+even if @code{PATH_MAX} is defined.
 @end deftypevr
 
 @cindex limits, pipe buffer size
@@ -1476,6 +1482,9 @@ Inquire about the value of @code{POSIX_REC_MIN_XFER_SIZE}.
 Inquire about the value of @code{POSIX_REC_XFER_ALIGN}.
 @end table
 
+@strong{Portability Note:} On some systems, @theglibc{} does not
+enforce @code{_PC_NAME_MAX} or @code{_PC_PATH_MAX} limits.
+
 @node Utility Limits
 @section Utility Program Capacity Limits
 

manual/filesys.texi

@@ -444,9 +444,9 @@ symbols are declared in the header file @file{dirent.h}.
 @comment POSIX.1
 @deftypefun {struct dirent *} readdir (DIR *@var{dirstream})
 This function reads the next entry from the directory.  It normally
-returns a pointer to a structure containing information about the file.
-This structure is statically allocated and can be rewritten by a
-subsequent call.
+returns a pointer to a structure containing information about the
+file.  This structure is associated with the @var{dirstream} handle
+and can be rewritten by a subsequent call.
 
 @strong{Portability Note:} On some systems @code{readdir} may not
 return entries for @file{.} and @file{..}, even though these are always
@@ -461,19 +461,61 @@ conditions are defined for this function:
 The @var{dirstream} argument is not valid.
 @end table
 
-@code{readdir} is not thread safe.  Multiple threads using
-@code{readdir} on the same @var{dirstream} may overwrite the return
-value.  Use @code{readdir_r} when this is critical.
+To distinguish between an end-of-directory condition or an error, you
+must set @code{errno} to zero before calling @code{readdir}.  To avoid
+entering an infinite loop, you should stop reading from the directory
+after the first error.
+
+In POSIX.1-2008, @code{readdir} is not thread-safe.  In @theglibc{}
+implementation, it is safe to call @code{readdir} concurrently on
+different @var{dirstream}s, but multiple threads accessing the same
+@var{dirstream} result in undefined behavior.  @code{readdir_r} is a
+fully thread-safe alternative, but suffers from poor portability (see
+below).  It is recommended that you use @code{readdir}, with external
+locking if multiple threads access the same @var{dirstream}.
 @end deftypefun
 
 @comment dirent.h
 @comment GNU
 @deftypefun int readdir_r (DIR *@var{dirstream}, struct dirent *@var{entry}, struct dirent **@var{result})
-This function is the reentrant version of @code{readdir}.  Like
-@code{readdir} it returns the next entry from the directory.  But to
-prevent conflicts between simultaneously running threads the result is
-not stored in statically allocated memory.  Instead the argument
-@var{entry} points to a place to store the result.
+This function is a version of @code{readdir} which performs internal
+locking.  Like @code{readdir} it returns the next entry from the
+directory.  To prevent conflicts between simultaneously running
+threads the result is stored inside the @var{entry} object.
+
+@strong{Portability Note:} It is recommended to use @code{readdir}
+instead of @code{readdir_r} for the following reasons:
+
+@itemize @bullet
+@item
+On systems which do not define @code{NAME_MAX}, it may not be possible
+to use @code{readdir_r} safely because the caller does not specify the
+length of the buffer for the directory entry.
+
+@item
+On some systems, @code{readdir_r} cannot read directory entries with
+very long names.  If such a name is encountered, @theglibc{}
+implementation of @code{readdir_r} returns with an error code of
+@code{ENAMETOOLONG} after the final directory entry has been read.  On
+other systems, @code{readdir_r} may return successfully, but the
+@code{d_name} member may not be NUL-terminated or may be truncated.
+
+@item
+POSIX-1.2008 does not guarantee that @code{readdir} is thread-safe,
+even when access to the same @var{dirstream} is serialized.  But in
+current implementations (including @theglibc{}), it is safe to call
+@code{readdir} concurrently on different @var{dirstream}s, so there is
+no need to use @code{readdir_r} in most multi-threaded programs.  In
+the rare case that multiple threads need to read from the same
+@var{dirstream}, it is still better to use @code{readdir} and external
+synchronization.
+
+@item
+It is expected that future versions of POSIX will obsolete
+@code{readdir_r} and mandate the level of thread safety for
+@code{readdir} which is provided by @theglibc{} and other
+implementations today.
+@end itemize
 
 Normally @code{readdir_r} returns zero and sets @code{*@var{result}}
 to @var{entry}.  If there are no more entries in the directory or an
@@ -481,15 +523,6 @@ error is detected, @code{readdir_r} sets @code{*@var{result}} to a
 null pointer and returns a nonzero error code, also stored in
 @code{errno}, as described for @code{readdir}.
 
-@strong{Portability Note:} On some systems @code{readdir_r} may not
-return a NUL terminated string for the file name, even when there is no
-@code{d_reclen} field in @code{struct dirent} and the file
-name is the maximum allowed size.  Modern systems all have the
-@code{d_reclen} field, and on old systems multi-threading is not
-critical.  In any case there is no such problem with the @code{readdir}
-function, so that even on systems without the @code{d_reclen} member one
-could use multiple threads by using external locking.
-
 It is also important to look at the definition of the @code{struct
 dirent} type.  Simply passing a pointer to an object of this type for
 the second parameter of @code{readdir_r} might not be enough.  Some

ports/ChangeLog.ia64

@@ -1,3 +1,8 @@
+2013-09-22  Carlos O'Donell  <carlos@redhat.com>
+
+	[BZ #15754]
+	* sysdeps/ia64/stackguard-macros.h: Define POINTER_CHK_GUARD.
+
 2013-07-04  Andreas Jaeger  <aj@suse.de>
 
 	* sysdeps/unix/sysv/linux/ia64/sys/ptrace.h (PTRACE_LISTEN):

ports/ChangeLog.tile

@@ -1,3 +1,8 @@
+2013-09-22  Carlos O'Donell  <carlos@redhat.com>
+
+	[BZ #15754]
+	* sysdeps/tile/stackguard-macros.h: Define POINTER_CHK_GUARD.
+
 2013-07-22  Chris Metcalf  <cmetcalf@tilera.com>
 
 	[BZ #15759]

ports/sysdeps/ia64/stackguard-macros.h

@@ -2,3 +2,6 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("adds %0 = -8, r13;; ld8 %0 = [%0]" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("adds %0 = -16, r13;; ld8 %0 = [%0]" : "=r" (x)); x; })

ports/sysdeps/tile/stackguard-macros.h

@@ -4,11 +4,17 @@
 # if __WORDSIZE == 64
 #  define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("addi %0, tp, -16; ld %0, %0" : "=r" (x)); x; })
+#  define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("addi %0, tp, -24; ld %0, %0" : "=r" (x)); x; })
 # else
 #  define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("addi %0, tp, -8; ld4s %0, %0" : "=r" (x)); x; })
+#  define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("addi %0, tp, -12; ld4s %0, %0" : "=r" (x)); x; })
 # endif
 #else
 # define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("addi %0, tp, -8; lw %0, %0" : "=r" (x)); x; })
+# define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("addi %0, tp, -12; lw %0, %0" : "=r" (x)); x; })
 #endif

string/Makefile

@@ -57,6 +57,8 @@ tests		:= tester inl-tester noinl-tester testcopy test-ffs	\
 tests-ifunc := $(strop-tests:%=test-%-ifunc)
 tests += $(tests-ifunc)
 
+xtests = tst-strcoll-overflow
+
 include ../Rules
 
 tester-ENV = LANGUAGE=C

string/strcoll_l.c

@@ -41,11 +41,434 @@
 
 #include "../locale/localeinfo.h"
 
+/* Track status while looking for sequences in a string.  */
+typedef struct
+{
+  int len;			/* Length of the current sequence.  */
+  size_t val;			/* Position of the sequence relative to the
+				   previous non-ignored sequence.  */
+  size_t idxnow;		/* Current index in sequences.  */
+  size_t idxmax;		/* Maximum index in sequences.  */
+  size_t idxcnt;		/* Current count of indices.  */
+  size_t backw;			/* Current Backward sequence index.  */
+  size_t backw_stop;		/* Index where the backward sequences stop.  */
+  const USTRING_TYPE *us;	/* The string.  */
+  int32_t *idxarr;		/* Array to cache weight indices.  */
+  unsigned char *rulearr;	/* Array to cache rules.  */
+  unsigned char rule;		/* Saved rule for the first sequence.  */
+  int32_t idx;			/* Index to weight of the current sequence.  */
+  int32_t save_idx;		/* Save looked up index of a forward
+				   sequence after the last backward
+				   sequence.  */
+  const USTRING_TYPE *back_us;	/* Beginning of the backward sequence.  */
+} coll_seq;
+
+/* Get next sequence.  The weight indices are cached, so we don't need to
+   traverse the string.  */
+static void
+get_next_seq_cached (coll_seq *seq, int nrules, int pass,
+		     const unsigned char *rulesets,
+		     const USTRING_TYPE *weights)
+{
+  size_t val = seq->val = 0;
+  int len = seq->len;
+  size_t backw_stop = seq->backw_stop;
+  size_t backw = seq->backw;
+  size_t idxcnt = seq->idxcnt;
+  size_t idxmax = seq->idxmax;
+  size_t idxnow = seq->idxnow;
+  unsigned char *rulearr = seq->rulearr;
+  int32_t *idxarr = seq->idxarr;
+
+  while (len == 0)
+    {
+      ++val;
+      if (backw_stop != ~0ul)
+	{
+	  /* There is something pushed.  */
+	  if (backw == backw_stop)
+	    {
+	      /* The last pushed character was handled.  Continue
+		 with forward characters.  */
+	      if (idxcnt < idxmax)
+		{
+		  idxnow = idxcnt;
+		  backw_stop = ~0ul;
+		}
+	      else
+		{
+		  /* Nothing any more.  The backward sequence
+		     ended with the last sequence in the string.  */
+		  idxnow = ~0ul;
+		  break;
+		}
+	    }
+	  else
+	    idxnow = --backw;
+	}
+      else
+	{
+	  backw_stop = idxcnt;
+
+	  while (idxcnt < idxmax)
+	    {
+	      if ((rulesets[rulearr[idxcnt] * nrules + pass]
+		   & sort_backward) == 0)
+		/* No more backward characters to push.  */
+		break;
+	      ++idxcnt;
+	    }
+
+	  if (backw_stop == idxcnt)
+	    {
+	      /* No sequence at all or just one.  */
+	      if (idxcnt == idxmax)
+		/* Note that LEN is still zero.  */
+		break;
+
+	      backw_stop = ~0ul;
+	      idxnow = idxcnt++;
+	    }
+	  else
+	    /* We pushed backward sequences.  */
+	    idxnow = backw = idxcnt - 1;
+	}
+      len = weights[idxarr[idxnow]++];
+    }
+
+  /* Update the structure.  */
+  seq->val = val;
+  seq->len = len;
+  seq->backw_stop = backw_stop;
+  seq->backw = backw;
+  seq->idxcnt = idxcnt;
+  seq->idxnow = idxnow;
+}
+
+/* Get next sequence.  Traverse the string as required.  */
+static void
+get_next_seq (coll_seq *seq, int nrules, const unsigned char *rulesets,
+	      const USTRING_TYPE *weights, const int32_t *table,
+	      const USTRING_TYPE *extra, const int32_t *indirect)
+{
+#include WEIGHT_H
+  size_t val = seq->val = 0;
+  int len = seq->len;
+  size_t backw_stop = seq->backw_stop;
+  size_t backw = seq->backw;
+  size_t idxcnt = seq->idxcnt;
+  size_t idxmax = seq->idxmax;
+  size_t idxnow = seq->idxnow;
+  unsigned char *rulearr = seq->rulearr;
+  int32_t *idxarr = seq->idxarr;
+  const USTRING_TYPE *us = seq->us;
+
+  while (len == 0)
+    {
+      ++val;
+      if (backw_stop != ~0ul)
+	{
+	  /* There is something pushed.  */
+	  if (backw == backw_stop)
+	    {
+	      /* The last pushed character was handled.  Continue
+		 with forward characters.  */
+	      if (idxcnt < idxmax)
+		{
+		  idxnow = idxcnt;
+		  backw_stop = ~0ul;
+		}
+	      else
+		/* Nothing any more.  The backward sequence ended with
+		   the last sequence in the string.  Note that LEN
+		   is still zero.  */
+		break;
+	    }
+	  else
+	    idxnow = --backw;
+	}
+      else
+	{
+	  backw_stop = idxmax;
+
+	  while (*us != L('\0'))
+	    {
+	      int32_t tmp = findidx (&us, -1);
+	      rulearr[idxmax] = tmp >> 24;
+	      idxarr[idxmax] = tmp & 0xffffff;
+	      idxcnt = idxmax++;
+
+	      if ((rulesets[rulearr[idxcnt] * nrules]
+		   & sort_backward) == 0)
+		/* No more backward characters to push.  */
+		break;
+	      ++idxcnt;
+	    }
+
+	  if (backw_stop >= idxcnt)
+	    {
+	      /* No sequence at all or just one.  */
+	      if (idxcnt == idxmax || backw_stop > idxcnt)
+		/* Note that LEN is still zero.  */
+		break;
+
+	      backw_stop = ~0ul;
+	      idxnow = idxcnt;
+	    }
+	  else
+	    /* We pushed backward sequences.  */
+	    idxnow = backw = idxcnt - 1;
+	}
+      len = weights[idxarr[idxnow]++];
+    }
+
+  /* Update the structure.  */
+  seq->val = val;
+  seq->len = len;
+  seq->backw_stop = backw_stop;
+  seq->backw = backw;
+  seq->idxcnt = idxcnt;
+  seq->idxmax = idxmax;
+  seq->idxnow = idxnow;
+  seq->us = us;
+}
+
+/* Get next sequence.  Traverse the string as required.  This function does not
+   set or use any index or rule cache.  */
+static void
+get_next_seq_nocache (coll_seq *seq, int nrules, const unsigned char *rulesets,
+		      const USTRING_TYPE *weights, const int32_t *table,
+		      const USTRING_TYPE *extra, const int32_t *indirect,
+		      int pass)
+{
+#include WEIGHT_H
+  size_t val = seq->val = 0;
+  int len = seq->len;
+  size_t backw_stop = seq->backw_stop;
+  size_t backw = seq->backw;
+  size_t idxcnt = seq->idxcnt;
+  size_t idxmax = seq->idxmax;
+  int32_t idx = seq->idx;
+  const USTRING_TYPE *us = seq->us;
+
+  while (len == 0)
+    {
+      ++val;
+      if (backw_stop != ~0ul)
+	{
+	  /* There is something pushed.  */
+	  if (backw == backw_stop)
+	    {
+	      /* The last pushed character was handled.  Continue
+		 with forward characters.  */
+	      if (idxcnt < idxmax)
+		{
+		  idx = seq->save_idx;
+		  backw_stop = ~0ul;
+		}
+	      else
+		{
+		  /* Nothing anymore.  The backward sequence ended with
+		     the last sequence in the string.  Note that len is
+		     still zero.  */
+		  idx = 0;
+		  break;
+	        }
+	    }
+	  else
+	    {
+	      /* XXX Traverse BACKW sequences from the beginning of
+		 BACKW_STOP to get the next sequence.  Is ther a quicker way
+	         to do this?  */
+	      size_t i = backw_stop;
+	      us = seq->back_us;
+	      while (i < backw)
+		{
+		  int32_t tmp = findidx (&us, -1);
+		  idx = tmp & 0xffffff;
+		  i++;
+		}
+	      --backw;
+	      us = seq->us;
+	    }
+	}
+      else
+	{
+	  backw_stop = idxmax;
+	  int32_t prev_idx = idx;
+
+	  while (*us != L('\0'))
+	    {
+	      int32_t tmp = findidx (&us, -1);
+	      unsigned char rule = tmp >> 24;
+	      prev_idx = idx;
+	      idx = tmp & 0xffffff;
+	      idxcnt = idxmax++;
+
+	      /* Save the rule for the first sequence.  */
+	      if (__glibc_unlikely (idxcnt == 0))
+	        seq->rule = rule;
+
+	      if ((rulesets[rule * nrules + pass]
+		   & sort_backward) == 0)
+		/* No more backward characters to push.  */
+		break;
+	      ++idxcnt;
+	    }
+
+	  if (backw_stop >= idxcnt)
+	    {
+	      /* No sequence at all or just one.  */
+	      if (idxcnt == idxmax || backw_stop > idxcnt)
+		/* Note that len is still zero.  */
+		break;
+
+	      backw_stop = ~0ul;
+	    }
+	  else
+	    {
+	      /* We pushed backward sequences.  If the stream ended with the
+		 backward sequence, then we process the last sequence we
+		 found.  Otherwise we process the sequence before the last
+		 one since the last one was a forward sequence.  */
+	      seq->back_us = seq->us;
+	      seq->us = us;
+	      backw = idxcnt;
+	      if (idxmax > idxcnt)
+		{
+		  backw--;
+		  seq->save_idx = idx;
+		  idx = prev_idx;
+		}
+	      if (backw > backw_stop)
+		backw--;
+	    }
+	}
+
+      len = weights[idx++];
+      /* Skip over indices of previous levels.  */
+      for (int i = 0; i < pass; i++)
+	{
+	  idx += len;
+	  len = weights[idx];
+	  idx++;
+	}
+    }
+
+  /* Update the structure.  */
+  seq->val = val;
+  seq->len = len;
+  seq->backw_stop = backw_stop;
+  seq->backw = backw;
+  seq->idxcnt = idxcnt;
+  seq->idxmax = idxmax;
+  seq->us = us;
+  seq->idx = idx;
+}
+
+/* Compare two sequences.  This version does not use the index and rules
+   cache.  */
+static int
+do_compare_nocache (coll_seq *seq1, coll_seq *seq2, int position,
+		    const USTRING_TYPE *weights)
+{
+  int seq1len = seq1->len;
+  int seq2len = seq2->len;
+  size_t val1 = seq1->val;
+  size_t val2 = seq2->val;
+  int idx1 = seq1->idx;
+  int idx2 = seq2->idx;
+  int result = 0;
+
+  /* Test for position if necessary.  */
+  if (position && val1 != val2)
+    {
+      result = val1 > val2 ? 1 : -1;
+      goto out;
+    }
+
+  /* Compare the two sequences.  */
+  do
+    {
+      if (weights[idx1] != weights[idx2])
+	{
+	  /* The sequences differ.  */
+	  result = weights[idx1] - weights[idx2];
+	  goto out;
+	}
+
+      /* Increment the offsets.  */
+      ++idx1;
+      ++idx2;
+
+      --seq1len;
+      --seq2len;
+    }
+  while (seq1len > 0 && seq2len > 0);
+
+  if (position && seq1len != seq2len)
+    result = seq1len - seq2len;
+
+out:
+  seq1->len = seq1len;
+  seq2->len = seq2len;
+  seq1->idx = idx1;
+  seq2->idx = idx2;
+  return result;
+}
+
+/* Compare two sequences using the index cache.  */
+static int
+do_compare (coll_seq *seq1, coll_seq *seq2, int position,
+	    const USTRING_TYPE *weights)
+{
+  int seq1len = seq1->len;
+  int seq2len = seq2->len;
+  size_t val1 = seq1->val;
+  size_t val2 = seq2->val;
+  int32_t *idx1arr = seq1->idxarr;
+  int32_t *idx2arr = seq2->idxarr;
+  int idx1now = seq1->idxnow;
+  int idx2now = seq2->idxnow;
+  int result = 0;
+
+  /* Test for position if necessary.  */
+  if (position && val1 != val2)
+    {
+      result = val1 > val2 ? 1 : -1;
+      goto out;
+    }
+
+  /* Compare the two sequences.  */
+  do
+    {
+      if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
+	{
+	  /* The sequences differ.  */
+	  result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
+	  goto out;
+	}
+
+      /* Increment the offsets.  */
+      ++idx1arr[idx1now];
+      ++idx2arr[idx2now];
+
+      --seq1len;
+      --seq2len;
+    }
+  while (seq1len > 0 && seq2len > 0);
+
+  if (position && seq1len != seq2len)
+    result = seq1len - seq2len;
+
+out:
+  seq1->len = seq1len;
+  seq2->len = seq2len;
+  return result;
+}
+
 int
-STRCOLL (s1, s2, l)
-     const STRING_TYPE *s1;
-     const STRING_TYPE *s2;
-     __locale_t l;
+STRCOLL (const STRING_TYPE *s1, const STRING_TYPE *s2, __locale_t l)
 {
   struct __locale_data *current = l->__locales[LC_COLLATE];
   uint_fast32_t nrules = current->values[_NL_ITEM_INDEX (_NL_COLLATE_NRULES)].word;
@@ -56,34 +479,6 @@ STRCOLL (s1, s2, l)
   const USTRING_TYPE *weights;
   const USTRING_TYPE *extra;
   const int32_t *indirect;
-  uint_fast32_t pass;
-  int result = 0;
-  const USTRING_TYPE *us1;
-  const USTRING_TYPE *us2;
-  size_t s1len;
-  size_t s2len;
-  int32_t *idx1arr;
-  int32_t *idx2arr;
-  unsigned char *rule1arr;
-  unsigned char *rule2arr;
-  size_t idx1max;
-  size_t idx2max;
-  size_t idx1cnt;
-  size_t idx2cnt;
-  size_t idx1now;
-  size_t idx2now;
-  size_t backw1_stop;
-  size_t backw2_stop;
-  size_t backw1;
-  size_t backw2;
-  int val1;
-  int val2;
-  int position;
-  int seq1len;
-  int seq2len;
-  int use_malloc;
-
-#include WEIGHT_H
 
   if (nrules == 0)
     return STRCMP (s1, s2);
@@ -98,7 +493,6 @@ STRCOLL (s1, s2, l)
     current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_EXTRA,SUFFIX))].string;
   indirect = (const int32_t *)
     current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_INDIRECT,SUFFIX))].string;
-  use_malloc = 0;
 
   assert (((uintptr_t) table) % __alignof__ (table[0]) == 0);
   assert (((uintptr_t) weights) % __alignof__ (weights[0]) == 0);
@@ -106,18 +500,13 @@ STRCOLL (s1, s2, l)
   assert (((uintptr_t) indirect) % __alignof__ (indirect[0]) == 0);
 
   /* We need this a few times.  */
-  s1len = STRLEN (s1);
-  s2len = STRLEN (s2);
+  size_t s1len = STRLEN (s1);
+  size_t s2len = STRLEN (s2);
 
   /* Catch empty strings.  */
-  if (__builtin_expect (s1len == 0, 0) || __builtin_expect (s2len == 0, 0))
+  if (__glibc_unlikely (s1len == 0) || __glibc_unlikely (s2len == 0))
     return (s1len != 0) - (s2len != 0);
 
-  /* We need the elements of the strings as unsigned values since they
-     are used as indeces.  */
-  us1 = (const USTRING_TYPE *) s1;
-  us2 = (const USTRING_TYPE *) s2;
-
   /* Perform the first pass over the string and while doing this find
      and store the weights for each character.  Since we want this to
      be as fast as possible we are using `alloca' to store the temporary
@@ -127,411 +516,122 @@ STRCOLL (s1, s2, l)
 
      Please note that the localedef programs makes sure that `position'
      is not used at the first level.  */
-  if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
+
+  coll_seq seq1, seq2;
+  bool use_malloc = false;
+  int result = 0;
+
+  memset (&seq1, 0, sizeof (seq1));
+  seq2 = seq1;
+
+  size_t size_max = SIZE_MAX / (sizeof (int32_t) + 1);
+
+  if (MIN (s1len, s2len) > size_max
+      || MAX (s1len, s2len) > size_max - MIN (s1len, s2len))
     {
-      idx1arr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
-      idx2arr = &idx1arr[s1len];
-      rule1arr = (unsigned char *) &idx2arr[s2len];
-      rule2arr = &rule1arr[s1len];
+      /* If the strings are long enough to cause overflow in the size request,
+         then skip the allocation and proceed with the non-cached routines.  */
+    }
+  else if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
+    {
+      seq1.idxarr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
 
-      if (idx1arr == NULL)
-	/* No memory.  Well, go with the stack then.
-
-	   XXX Once this implementation is stable we will handle this
-	   differently.  Instead of precomputing the indeces we will
-	   do this in time.  This means, though, that this happens for
-	   every pass again.  */
-	goto try_stack;
-      use_malloc = 1;
+      /* If we failed to allocate memory, we leave everything as NULL so that
+	 we use the nocache version of traversal and comparison functions.  */
+      if (seq1.idxarr != NULL)
+	{
+	  seq2.idxarr = &seq1.idxarr[s1len];
+	  seq1.rulearr = (unsigned char *) &seq2.idxarr[s2len];
+	  seq2.rulearr = &seq1.rulearr[s1len];
+	  use_malloc = true;
+	}
     }
   else
     {
-    try_stack:
-      idx1arr = (int32_t *) alloca (s1len * sizeof (int32_t));
-      idx2arr = (int32_t *) alloca (s2len * sizeof (int32_t));
-      rule1arr = (unsigned char *) alloca (s1len);
-      rule2arr = (unsigned char *) alloca (s2len);
+      seq1.idxarr = (int32_t *) alloca (s1len * sizeof (int32_t));
+      seq2.idxarr = (int32_t *) alloca (s2len * sizeof (int32_t));
+      seq1.rulearr = (unsigned char *) alloca (s1len);
+      seq2.rulearr = (unsigned char *) alloca (s2len);
     }
 
-  idx1cnt = 0;
-  idx2cnt = 0;
-  idx1max = 0;
-  idx2max = 0;
-  idx1now = 0;
-  idx2now = 0;
-  backw1_stop = ~0ul;
-  backw2_stop = ~0ul;
-  backw1 = ~0ul;
-  backw2 = ~0ul;
-  seq1len = 0;
-  seq2len = 0;
-  position = rulesets[0] & sort_position;
-  while (1)
+  int rule = 0;
+
+  /* Cache values in the first pass and if needed, use them in subsequent
+     passes.  */
+  for (int pass = 0; pass < nrules; ++pass)
     {
-      val1 = 0;
-      val2 = 0;
+      seq1.idxcnt = 0;
+      seq1.idx = 0;
+      seq2.idx = 0;
+      seq1.backw_stop = ~0ul;
+      seq1.backw = ~0ul;
+      seq2.idxcnt = 0;
+      seq2.backw_stop = ~0ul;
+      seq2.backw = ~0ul;
 
-      /* Get the next non-IGNOREd element for string `s1'.  */
-      if (seq1len == 0)
-	do
-	  {
-	    ++val1;
+      /* We need the elements of the strings as unsigned values since they
+	 are used as indices.  */
+      seq1.us = (const USTRING_TYPE *) s1;
+      seq2.us = (const USTRING_TYPE *) s2;
 
-	    if (backw1_stop != ~0ul)
-	      {
-		/* The is something pushed.  */
-		if (backw1 == backw1_stop)
-		  {
-		    /* The last pushed character was handled.  Continue
-		       with forward characters.  */
-		    if (idx1cnt < idx1max)
-		      {
-			idx1now = idx1cnt;
-			backw1_stop = ~0ul;
-		      }
-		    else
-		      /* Nothing anymore.  The backward sequence ended with
-			 the last sequence in the string.  Note that seq1len
-			 is still zero.  */
-		      break;
-		  }
-		else
-		  idx1now = --backw1;
-	      }
-	    else
-	      {
-		backw1_stop = idx1max;
-
-		while (*us1 != L('\0'))
-		  {
-		    int32_t tmp = findidx (&us1, -1);
-		    rule1arr[idx1max] = tmp >> 24;
-		    idx1arr[idx1max] = tmp & 0xffffff;
-		    idx1cnt = idx1max++;
-
-		    if ((rulesets[rule1arr[idx1cnt] * nrules]
-			 & sort_backward) == 0)
-		      /* No more backward characters to push.  */
-		      break;
-		    ++idx1cnt;
-		  }
-
-		if (backw1_stop >= idx1cnt)
-		  {
-		    /* No sequence at all or just one.  */
-		    if (idx1cnt == idx1max || backw1_stop > idx1cnt)
-		      /* Note that seq1len is still zero.  */
-		      break;
-
-		    backw1_stop = ~0ul;
-		    idx1now = idx1cnt;
-		  }
-		else
-		  /* We pushed backward sequences.  */
-		  idx1now = backw1 = idx1cnt - 1;
-	      }
-	  }
-	while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
-      /* And the same for string `s2'.  */
-      if (seq2len == 0)
-	do
-	  {
-	    ++val2;
-
-	    if (backw2_stop != ~0ul)
-	      {
-		/* The is something pushed.  */
-		if (backw2 == backw2_stop)
-		  {
-		    /* The last pushed character was handled.  Continue
-		       with forward characters.  */
-		    if (idx2cnt < idx2max)
-		      {
-			idx2now = idx2cnt;
-			backw2_stop = ~0ul;
-		      }
-		    else
-		      /* Nothing anymore.  The backward sequence ended with
-			 the last sequence in the string.  Note that seq2len
-			 is still zero.  */
-		      break;
-		  }
-		else
-		  idx2now = --backw2;
-	      }
-	    else
-	      {
-		backw2_stop = idx2max;
-
-		while (*us2 != L('\0'))
-		  {
-		    int32_t tmp = findidx (&us2, -1);
-		    rule2arr[idx2max] = tmp >> 24;
-		    idx2arr[idx2max] = tmp & 0xffffff;
-		    idx2cnt = idx2max++;
-
-		    if ((rulesets[rule2arr[idx2cnt] * nrules]
-			 & sort_backward) == 0)
-		      /* No more backward characters to push.  */
-		      break;
-		    ++idx2cnt;
-		  }
-
-		if (backw2_stop >= idx2cnt)
-		  {
-		    /* No sequence at all or just one.  */
-		    if (idx2cnt == idx2max || backw2_stop > idx2cnt)
-		      /* Note that seq1len is still zero.  */
-		      break;
-
-		    backw2_stop = ~0ul;
-		    idx2now = idx2cnt;
-		  }
-		else
-		  /* We pushed backward sequences.  */
-		  idx2now = backw2 = idx2cnt - 1;
-	      }
-	  }
-	while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
-
-      /* See whether any or both strings are empty.  */
-      if (seq1len == 0 || seq2len == 0)
-	{
-	  if (seq1len == seq2len)
-	    /* Both ended.  So far so good, both strings are equal at the
-	       first level.  */
-	    break;
-
-	  /* This means one string is shorter than the other.  Find out
-	     which one and return an appropriate value.  */
-	  result = seq1len == 0 ? -1 : 1;
-	  goto free_and_return;
-	}
-
-      /* Test for position if necessary.  */
-      if (position && val1 != val2)
-	{
-	  result = val1 - val2;
-	  goto free_and_return;
-	}
-
-      /* Compare the two sequences.  */
-      do
-	{
-	  if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
-	    {
-	      /* The sequences differ.  */
-	      result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
-	      goto free_and_return;
-	    }
-
-	  /* Increment the offsets.  */
-	  ++idx1arr[idx1now];
-	  ++idx2arr[idx2now];
-
-	  --seq1len;
-	  --seq2len;
-	}
-      while (seq1len > 0 && seq2len > 0);
-
-      if (position && seq1len != seq2len)
-	{
-	  result = seq1len - seq2len;
-	  goto free_and_return;
-	}
-    }
-
-  /* Now the remaining passes over the weights.  We now use the
-     indeces we found before.  */
-  for (pass = 1; pass < nrules; ++pass)
-    {
       /* We assume that if a rule has defined `position' in one section
 	 this is true for all of them.  */
-      idx1cnt = 0;
-      idx2cnt = 0;
-      backw1_stop = ~0ul;
-      backw2_stop = ~0ul;
-      backw1 = ~0ul;
-      backw2 = ~0ul;
-      position = rulesets[rule1arr[0] * nrules + pass] & sort_position;
+      int position = rulesets[rule * nrules + pass] & sort_position;
 
       while (1)
 	{
-	  val1 = 0;
-	  val2 = 0;
-
-	  /* Get the next non-IGNOREd element for string `s1'.  */
-	  if (seq1len == 0)
-	    do
-	      {
-		++val1;
-
-		if (backw1_stop != ~0ul)
-		  {
-		    /* The is something pushed.  */
-		    if (backw1 == backw1_stop)
-		      {
-			/* The last pushed character was handled.  Continue
-			   with forward characters.  */
-			if (idx1cnt < idx1max)
-			  {
-			    idx1now = idx1cnt;
-			    backw1_stop = ~0ul;
-			  }
-			else
-			  {
-			    /* Nothing anymore.  The backward sequence
-			       ended with the last sequence in the string.  */
-			    idx1now = ~0ul;
-			    break;
-			  }
-		      }
-		    else
-		      idx1now = --backw1;
-		  }
-		else
-		  {
-		    backw1_stop = idx1cnt;
-
-		    while (idx1cnt < idx1max)
-		      {
-			if ((rulesets[rule1arr[idx1cnt] * nrules + pass]
-			     & sort_backward) == 0)
-			  /* No more backward characters to push.  */
-			  break;
-			++idx1cnt;
-		      }
-
-		    if (backw1_stop == idx1cnt)
-		      {
-			/* No sequence at all or just one.  */
-			if (idx1cnt == idx1max)
-			  /* Note that seq1len is still zero.  */
-			  break;
-
-			backw1_stop = ~0ul;
-			idx1now = idx1cnt++;
-		      }
-		    else
-		      /* We pushed backward sequences.  */
-		      idx1now = backw1 = idx1cnt - 1;
-		  }
-	      }
-	    while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
-	  /* And the same for string `s2'.  */
-	  if (seq2len == 0)
-	    do
-	      {
-		++val2;
-
-		if (backw2_stop != ~0ul)
-		  {
-		    /* The is something pushed.  */
-		    if (backw2 == backw2_stop)
-		      {
-			/* The last pushed character was handled.  Continue
-			   with forward characters.  */
-			if (idx2cnt < idx2max)
-			  {
-			    idx2now = idx2cnt;
-			    backw2_stop = ~0ul;
-			  }
-			else
-			  {
-			    /* Nothing anymore.  The backward sequence
-			       ended with the last sequence in the string.  */
-			    idx2now = ~0ul;
-			    break;
-			  }
-		      }
-		    else
-		      idx2now = --backw2;
-		  }
-		else
-		  {
-		    backw2_stop = idx2cnt;
-
-		    while (idx2cnt < idx2max)
-		      {
-			if ((rulesets[rule2arr[idx2cnt] * nrules + pass]
-			     & sort_backward) == 0)
-			  /* No more backward characters to push.  */
-			  break;
-			++idx2cnt;
-		      }
-
-		    if (backw2_stop == idx2cnt)
-		      {
-			/* No sequence at all or just one.  */
-			if (idx2cnt == idx2max)
-			  /* Note that seq2len is still zero.  */
-			  break;
-
-			backw2_stop = ~0ul;
-			idx2now = idx2cnt++;
-		      }
-		    else
-		      /* We pushed backward sequences.  */
-		      idx2now = backw2 = idx2cnt - 1;
-		  }
-	      }
-	    while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
+	  if (__glibc_unlikely (seq1.idxarr == NULL))
+	    {
+	      get_next_seq_nocache (&seq1, nrules, rulesets, weights, table,
+				    extra, indirect, pass);
+	      get_next_seq_nocache (&seq2, nrules, rulesets, weights, table,
+				    extra, indirect, pass);
+	    }
+	  else if (pass == 0)
+	    {
+	      get_next_seq (&seq1, nrules, rulesets, weights, table, extra,
+			    indirect);
+	      get_next_seq (&seq2, nrules, rulesets, weights, table, extra,
+			    indirect);
+	    }
+	  else
+	    {
+	      get_next_seq_cached (&seq1, nrules, pass, rulesets, weights);
+	      get_next_seq_cached (&seq2, nrules, pass, rulesets, weights);
+	    }
 
 	  /* See whether any or both strings are empty.  */
-	  if (seq1len == 0 || seq2len == 0)
+	  if (seq1.len == 0 || seq2.len == 0)
 	    {
-	      if (seq1len == seq2len)
+	      if (seq1.len == seq2.len)
 		/* Both ended.  So far so good, both strings are equal
 		   at this level.  */
 		break;
 
 	      /* This means one string is shorter than the other.  Find out
 		 which one and return an appropriate value.  */
-	      result = seq1len == 0 ? -1 : 1;
+	      result = seq1.len == 0 ? -1 : 1;
 	      goto free_and_return;
 	    }
 
-	  /* Test for position if necessary.  */
-	  if (position && val1 != val2)
-	    {
-	      result = val1 - val2;
-	      goto free_and_return;
-	    }
-
-	  /* Compare the two sequences.  */
-	  do
-	    {
-	      if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
-		{
-		  /* The sequences differ.  */
-		  result = (weights[idx1arr[idx1now]]
-			    - weights[idx2arr[idx2now]]);
-		  goto free_and_return;
-		}
-
-	      /* Increment the offsets.  */
-	      ++idx1arr[idx1now];
-	      ++idx2arr[idx2now];
-
-	      --seq1len;
-	      --seq2len;
-	    }
-	  while (seq1len > 0 && seq2len > 0);
-
-	  if (position && seq1len != seq2len)
-	    {
-	      result = seq1len - seq2len;
-	      goto free_and_return;
-	    }
+	  if (__glibc_unlikely (seq1.idxarr == NULL))
+	    result = do_compare_nocache (&seq1, &seq2, position, weights);
+	  else
+	    result = do_compare (&seq1, &seq2, position, weights);
+	  if (result != 0)
+	    goto free_and_return;
 	}
+
+      if (__glibc_likely (seq1.rulearr != NULL))
+	rule = seq1.rulearr[0];
+      else
+	rule = seq1.rule;
     }
 
   /* Free the memory if needed.  */
  free_and_return:
   if (use_malloc)
-    free (idx1arr);
+    free (seq1.idxarr);
 
   return result;
 }

string/tst-strcoll-overflow.c

@@ -0,0 +1,61 @@
+/* Copyright (C) 2013 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <locale.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+/* Verify that strcoll does not crash for large strings for which it cannot
+   cache weight lookup results.  The size is large enough to cause integer
+   overflows on 32-bit as well as buffer overflows on 64-bit.  The test should
+   work reasonably reliably when overcommit is disabled, but it obviously
+   depends on how much memory the system has.  There's a limitation to this
+   test in that it does not run to completion.  Actually collating such a
+   large string can take days and we can't have xcheck running that long.  For
+   that reason, we run the test for about 5 minutes and then assume that
+   everything is fine if there are no crashes.  */
+#define SIZE 0x40000000ul
+
+int
+do_test (void)
+{
+  if (setlocale (LC_COLLATE, "en_GB.UTF-8") == NULL)
+    {
+      puts ("setlocale failed, cannot test for overflow");
+      return 0;
+    }
+
+  char *p = malloc (SIZE);
+
+  if (p == NULL)
+    {
+      puts ("could not allocate memory");
+      return 1;
+    }
+
+  memset (p, 'x', SIZE - 1);
+  p[SIZE - 1] = 0;
+  printf ("%d\n", strcoll (p, p));
+  return 0;
+}
+
+#define TIMEOUT 300
+#define EXPECTED_SIGNAL SIGALRM
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"

sysdeps/generic/stackguard-macros.h

@@ -2,3 +2,11 @@
 
 extern uintptr_t __stack_chk_guard;
 #define STACK_CHK_GUARD __stack_chk_guard
+
+#ifdef PTRGUARD_LOCAL
+extern uintptr_t __pointer_chk_guard_local;
+# define POINTER_CHK_GUARD __pointer_chk_guard_local
+#else
+extern uintptr_t __pointer_chk_guard;
+# define POINTER_CHK_GUARD __pointer_chk_guard
+#endif

sysdeps/i386/stackguard-macros.h

@@ -2,3 +2,11 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("movl %%gs:0x14, %0" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({							\
+     uintptr_t x;					\
+     asm ("movl %%gs:%c1, %0" : "=r" (x)		\
+	  : "i" (offsetof (tcbhead_t, pointer_guard)));	\
+     x;							\
+   })

sysdeps/posix/dirstream.h

@@ -39,6 +39,8 @@ struct __dirstream
 
     off_t filepos;		/* Position of next entry to read.  */
 
+    int errcode;		/* Delayed error code.  */
+
     /* Directory block.  */
     char data[0] __attribute__ ((aligned (__alignof__ (void*))));
   };

sysdeps/posix/getaddrinfo.c

@@ -197,7 +197,22 @@ gaih_inet_serv (const char *servicename, const struct gaih_typeproto *tp,
 				&rc, &herrno, NULL, &localcanon));	      \
     if (rc != ERANGE || herrno != NETDB_INTERNAL)			      \
       break;								      \
-    tmpbuf = extend_alloca (tmpbuf, tmpbuflen, 2 * tmpbuflen);		      \
+    if (!malloc_tmpbuf && __libc_use_alloca (alloca_used + 2 * tmpbuflen))    \
+      tmpbuf = extend_alloca_account (tmpbuf, tmpbuflen, 2 * tmpbuflen,	      \
+				      alloca_used);			      \
+    else								      \
+      {									      \
+	char *newp = realloc (malloc_tmpbuf ? tmpbuf : NULL,		      \
+			      2 * tmpbuflen);				      \
+	if (newp == NULL)						      \
+	  {								      \
+	    result = -EAI_MEMORY;					      \
+	    goto free_and_return;					      \
+	  }								      \
+	tmpbuf = newp;							      \
+	malloc_tmpbuf = true;						      \
+	tmpbuflen = 2 * tmpbuflen;					      \
+      }									      \
   }									      \
   if (status == NSS_STATUS_SUCCESS && rc == 0)				      \
     h = &th;								      \
@@ -209,7 +224,8 @@ gaih_inet_serv (const char *servicename, const struct gaih_typeproto *tp,
 	{								      \
 	  __set_h_errno (herrno);					      \
 	  _res.options |= old_res_options & RES_USE_INET6;		      \
-	  return -EAI_SYSTEM;						      \
+	  result = -EAI_SYSTEM;						      \
+	  goto free_and_return;						      \
 	}								      \
       if (herrno == TRY_AGAIN)						      \
 	no_data = EAI_AGAIN;						      \
@@ -1735,13 +1751,13 @@ rfc3484_sort (const void *p1, const void *p2, void *arg)
 
 	      /* Fill in the results in all the records.  */
 	      for (int i = 0; i < src->nresults; ++i)
-		if (src->results[i].index == a1_index)
+		if (a1_index != -1 && src->results[i].index == a1_index)
 		  {
 		    assert (src->results[i].native == -1
 			    || src->results[i].native == a1_native);
 		    src->results[i].native = a1_native;
 		  }
-		else if (src->results[i].index == a2_index)
+		else if (a2_index != -1 && src->results[i].index == a2_index)
 		  {
 		    assert (src->results[i].native == -1
 			    || src->results[i].native == a2_native);
@@ -2601,7 +2617,14 @@ getaddrinfo (const char *name, const char *service,
 			  tmp.addr[0] = 0;
 			  tmp.addr[1] = 0;
 			  tmp.addr[2] = htonl (0xffff);
-			  tmp.addr[3] = sinp->sin_addr.s_addr;
+			  /* Special case for lo interface, the source address
+			     being possibly different than the interface
+			     address. */
+			  if ((ntohl(sinp->sin_addr.s_addr) & 0xff000000)
+			      == 0x7f000000)
+			    tmp.addr[3] = htonl(0x7f000001);
+			  else
+			    tmp.addr[3] = sinp->sin_addr.s_addr;
 			}
 		      else
 			{

sysdeps/posix/opendir.c

@@ -231,6 +231,7 @@ __alloc_dir (int fd, bool close_fd, int flags, const struct stat64 *statp)
   dirp->size = 0;
   dirp->offset = 0;
   dirp->filepos = 0;
+  dirp->errcode = 0;
 
   return dirp;
 }

sysdeps/posix/readdir_r.c

@@ -40,6 +40,7 @@ __READDIR_R (DIR *dirp, DIRENT_TYPE *entry, DIRENT_TYPE **result)
   DIRENT_TYPE *dp;
   size_t reclen;
   const int saved_errno = errno;
+  int ret;
 
   __libc_lock_lock (dirp->lock);
 
@@ -70,10 +71,10 @@ __READDIR_R (DIR *dirp, DIRENT_TYPE *entry, DIRENT_TYPE **result)
 		  bytes = 0;
 		  __set_errno (saved_errno);
 		}
+	      if (bytes < 0)
+		dirp->errcode = errno;
 
 	      dp = NULL;
-	      /* Reclen != 0 signals that an error occurred.  */
-	      reclen = bytes != 0;
 	      break;
 	    }
 	  dirp->size = (size_t) bytes;
@@ -106,29 +107,46 @@ __READDIR_R (DIR *dirp, DIRENT_TYPE *entry, DIRENT_TYPE **result)
       dirp->filepos += reclen;
 #endif
 
-      /* Skip deleted files.  */
+#ifdef NAME_MAX
+      if (reclen > offsetof (DIRENT_TYPE, d_name) + NAME_MAX + 1)
+	{
+	  /* The record is very long.  It could still fit into the
+	     caller-supplied buffer if we can skip padding at the
+	     end.  */
+	  size_t namelen = _D_EXACT_NAMLEN (dp);
+	  if (namelen <= NAME_MAX)
+	    reclen = offsetof (DIRENT_TYPE, d_name) + namelen + 1;
+	  else
+	    {
+	      /* The name is too long.  Ignore this file.  */
+	      dirp->errcode = ENAMETOOLONG;
+	      dp->d_ino = 0;
+	      continue;
+	    }
+	}
+#endif
+
+      /* Skip deleted and ignored files.  */
     }
   while (dp->d_ino == 0);
 
   if (dp != NULL)
     {
-#ifdef GETDENTS_64BIT_ALIGNED
-      /* The d_reclen value might include padding which is not part of
-	 the DIRENT_TYPE data structure.  */
-      reclen = MIN (reclen,
-		    offsetof (DIRENT_TYPE, d_name) + sizeof (dp->d_name));
-#endif
       *result = memcpy (entry, dp, reclen);
-#ifdef GETDENTS_64BIT_ALIGNED
+#ifdef _DIRENT_HAVE_D_RECLEN
       entry->d_reclen = reclen;
 #endif
+      ret = 0;
     }
   else
-    *result = NULL;
+    {
+      *result = NULL;
+      ret = dirp->errcode;
+    }
 
   __libc_lock_unlock (dirp->lock);
 
-  return dp != NULL ? 0 : reclen ? errno : 0;
+  return ret;
 }
 
 #ifdef __READDIR_R_ALIAS

sysdeps/posix/rewinddir.c

@@ -33,6 +33,7 @@ rewinddir (dirp)
   dirp->filepos = 0;
   dirp->offset = 0;
   dirp->size = 0;
+  dirp->errcode = 0;
 #ifndef NOT_IN_libc
   __libc_lock_unlock (dirp->lock);
 #endif

sysdeps/powerpc/powerpc32/stackguard-macros.h

@@ -2,3 +2,13 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("lwz %0,-28680(2)" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({												\
+     uintptr_t x;										\
+     asm ("lwz %0,%1(2)"									\
+	  : "=r" (x)										\
+	  : "i" (offsetof (tcbhead_t, pointer_guard) - TLS_TCB_OFFSET - sizeof (tcbhead_t))	\
+         );											\
+     x;												\
+   })

sysdeps/powerpc/powerpc64/stackguard-macros.h

@@ -2,3 +2,13 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("ld %0,-28688(13)" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({												\
+     uintptr_t x;										\
+     asm ("ld %0,%1(2)"										\
+	  : "=r" (x)										\
+	  : "i" (offsetof (tcbhead_t, pointer_guard) - TLS_TCB_OFFSET - sizeof (tcbhead_t))	\
+         );											\
+     x;												\
+   })

sysdeps/s390/s390-32/stackguard-macros.h

@@ -2,3 +2,14 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("ear %0,%%a0; l %0,0x14(%0)" : "=a" (x)); x; })
+
+/* On s390/s390x there is no unique pointer guard, instead we use the
+   same value as the stack guard.  */
+#define POINTER_CHK_GUARD \
+  ({							\
+     uintptr_t x;					\
+     asm ("ear %0,%%a0; l %0,%1(%0)"			\
+	  : "=a" (x)					\
+	  : "i" (offsetof (tcbhead_t, stack_guard)));	\
+     x;							\
+   })

sysdeps/s390/s390-64/stackguard-macros.h

@@ -2,3 +2,17 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("ear %0,%%a0; sllg %0,%0,32; ear %0,%%a1; lg %0,0x28(%0)" : "=a" (x)); x; })
+
+/* On s390/s390x there is no unique pointer guard, instead we use the
+   same value as the stack guard.  */
+#define POINTER_CHK_GUARD \
+  ({							\
+     uintptr_t x;					\
+     asm ("ear %0,%%a0;"				\
+	  "sllg %0,%0,32;"				\
+	  "ear %0,%%a1;"				\
+	  "lg %0,%1(%0)"				\
+	 : "=a" (x)					\
+	 : "i" (offsetof (tcbhead_t, stack_guard)));	\
+     x;							\
+   })

sysdeps/sparc/sparc32/stackguard-macros.h

@@ -2,3 +2,6 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("ld [%%g7+0x14], %0" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("ld [%%g7+0x18], %0" : "=r" (x)); x; })

sysdeps/sparc/sparc64/stackguard-macros.h

@@ -2,3 +2,6 @@
 
 #define STACK_CHK_GUARD \
   ({ uintptr_t x; asm ("ldx [%%g7+0x28], %0" : "=r" (x)); x; })
+
+#define POINTER_CHK_GUARD \
+  ({ uintptr_t x; asm ("ldx [%%g7+0x30], %0" : "=r" (x)); x; })

sysdeps/unix/sysv/linux/i386/readdir64_r.c

@@ -18,7 +18,6 @@
 #define __READDIR_R __readdir64_r
 #define __GETDENTS __getdents64
 #define DIRENT_TYPE struct dirent64
-#define GETDENTS_64BIT_ALIGNED 1
 
 #include <sysdeps/posix/readdir_r.c>
 

sysdeps/unix/sysv/linux/wordsize-64/readdir_r.c

@@ -1,5 +1,4 @@
 #define readdir64_r __no_readdir64_r_decl
-#define GETDENTS_64BIT_ALIGNED 1
 #include <sysdeps/posix/readdir_r.c>
 #undef readdir64_r
 weak_alias (__readdir_r, readdir64_r)

sysdeps/x86_64/multiarch/strstr.c

@@ -86,7 +86,7 @@
 /* Simple replacement of movdqu to address 4KB boundary cross issue.
    If EOS occurs within less than 16B before 4KB boundary, we don't
    cross to next page.  */
-static __m128i
+static inline __m128i
 __m128i_strloadu (const unsigned char * p, __m128i zero)
 {
   if (__builtin_expect ((int) ((size_t) p & 0xfff) > 0xff0, 0))

sysdeps/x86_64/stackguard-macros.h

@@ -4,3 +4,8 @@
   ({ uintptr_t x;						\
      asm ("mov %%fs:%c1, %0" : "=r" (x)				\
 	  : "i" (offsetof (tcbhead_t, stack_guard))); x; })
+
+#define POINTER_CHK_GUARD \
+  ({ uintptr_t x;						\
+     asm ("mov %%fs:%c1, %0" : "=r" (x)				\
+	  : "i" (offsetof (tcbhead_t, pointer_guard))); x; })