2887 lines
84 KiB
C
2887 lines
84 KiB
C
/* Mudflap: narrow-pointer bounds-checking by tree rewriting.
|
|
Copyright (C) 2002-2013 Free Software Foundation, Inc.
|
|
Contributed by Frank Ch. Eigler <fche@redhat.com>
|
|
and Graydon Hoare <graydon@redhat.com>
|
|
Splay Tree code originally by Mark Mitchell <mark@markmitchell.com>,
|
|
adapted from libiberty.
|
|
|
|
This file is part of GCC.
|
|
|
|
GCC is free software; you can redistribute it and/or modify it under
|
|
the terms of the GNU General Public License as published by the Free
|
|
Software Foundation; either version 3, or (at your option) any later
|
|
version.
|
|
|
|
GCC 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 General Public License
|
|
for more details.
|
|
|
|
Under Section 7 of GPL version 3, you are granted additional
|
|
permissions described in the GCC Runtime Library Exception, version
|
|
3.1, as published by the Free Software Foundation.
|
|
|
|
You should have received a copy of the GNU General Public License and
|
|
a copy of the GCC Runtime Library Exception along with this program;
|
|
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
<http://www.gnu.org/licenses/>. */
|
|
|
|
#include "config.h"
|
|
|
|
/* These attempt to coax various unix flavours to declare all our
|
|
needed tidbits in the system headers. */
|
|
#if !defined(__FreeBSD__) && !defined(__APPLE__)
|
|
#define _POSIX_SOURCE
|
|
#endif /* Some BSDs break <sys/socket.h> if this is defined. */
|
|
#define _GNU_SOURCE
|
|
#define _XOPEN_SOURCE
|
|
#define _BSD_TYPES
|
|
#define __EXTENSIONS__
|
|
#define _ALL_SOURCE
|
|
#define _LARGE_FILE_API
|
|
#define _XOPEN_SOURCE_EXTENDED 1
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <sys/types.h>
|
|
#include <sys/time.h>
|
|
#include <time.h>
|
|
#include <unistd.h>
|
|
#ifdef HAVE_EXECINFO_H
|
|
#include <execinfo.h>
|
|
#endif
|
|
#ifdef HAVE_SIGNAL_H
|
|
#include <signal.h>
|
|
#endif
|
|
#include <assert.h>
|
|
|
|
#include <string.h>
|
|
#include <limits.h>
|
|
#include <sys/types.h>
|
|
#include <signal.h>
|
|
#include <errno.h>
|
|
#include <ctype.h>
|
|
|
|
#include "mf-runtime.h"
|
|
#include "mf-impl.h"
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* Splay-tree implementation. */
|
|
|
|
typedef uintptr_t mfsplay_tree_key;
|
|
typedef void *mfsplay_tree_value;
|
|
|
|
/* Forward declaration for a node in the tree. */
|
|
typedef struct mfsplay_tree_node_s *mfsplay_tree_node;
|
|
|
|
/* The type of a function used to iterate over the tree. */
|
|
typedef int (*mfsplay_tree_foreach_fn) (mfsplay_tree_node, void *);
|
|
|
|
/* The nodes in the splay tree. */
|
|
struct mfsplay_tree_node_s
|
|
{
|
|
/* Data. */
|
|
mfsplay_tree_key key;
|
|
mfsplay_tree_value value;
|
|
/* Children. */
|
|
mfsplay_tree_node left;
|
|
mfsplay_tree_node right;
|
|
/* XXX: The addition of a parent pointer may eliminate some recursion. */
|
|
};
|
|
|
|
/* The splay tree itself. */
|
|
struct mfsplay_tree_s
|
|
{
|
|
/* The root of the tree. */
|
|
mfsplay_tree_node root;
|
|
|
|
/* The last key value for which the tree has been splayed, but not
|
|
since modified. */
|
|
mfsplay_tree_key last_splayed_key;
|
|
int last_splayed_key_p;
|
|
|
|
/* Statistics. */
|
|
unsigned num_keys;
|
|
|
|
/* Traversal recursion control flags. */
|
|
unsigned max_depth;
|
|
unsigned depth;
|
|
unsigned rebalance_p;
|
|
};
|
|
typedef struct mfsplay_tree_s *mfsplay_tree;
|
|
|
|
static mfsplay_tree mfsplay_tree_new (void);
|
|
static mfsplay_tree_node mfsplay_tree_insert (mfsplay_tree, mfsplay_tree_key, mfsplay_tree_value);
|
|
static void mfsplay_tree_remove (mfsplay_tree, mfsplay_tree_key);
|
|
static mfsplay_tree_node mfsplay_tree_lookup (mfsplay_tree, mfsplay_tree_key);
|
|
static mfsplay_tree_node mfsplay_tree_predecessor (mfsplay_tree, mfsplay_tree_key);
|
|
static mfsplay_tree_node mfsplay_tree_successor (mfsplay_tree, mfsplay_tree_key);
|
|
static int mfsplay_tree_foreach (mfsplay_tree, mfsplay_tree_foreach_fn, void *);
|
|
static void mfsplay_tree_rebalance (mfsplay_tree sp);
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* Utility macros */
|
|
|
|
#define CTOR __attribute__ ((constructor))
|
|
#define DTOR __attribute__ ((destructor))
|
|
|
|
|
|
/* Codes to describe the context in which a violation occurs. */
|
|
#define __MF_VIOL_UNKNOWN 0
|
|
#define __MF_VIOL_READ 1
|
|
#define __MF_VIOL_WRITE 2
|
|
#define __MF_VIOL_REGISTER 3
|
|
#define __MF_VIOL_UNREGISTER 4
|
|
#define __MF_VIOL_WATCH 5
|
|
|
|
/* Protect against recursive calls. */
|
|
|
|
static void
|
|
begin_recursion_protect1 (const char *pf)
|
|
{
|
|
if (__mf_get_state () == reentrant)
|
|
{
|
|
write (2, "mf: erroneous reentrancy detected in `", 38);
|
|
write (2, pf, strlen(pf));
|
|
write (2, "'\n", 2); \
|
|
abort ();
|
|
}
|
|
__mf_set_state (reentrant);
|
|
}
|
|
|
|
#define BEGIN_RECURSION_PROTECT() \
|
|
begin_recursion_protect1 (__PRETTY_FUNCTION__)
|
|
|
|
#define END_RECURSION_PROTECT() \
|
|
__mf_set_state (active)
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* Required globals. */
|
|
|
|
#define LOOKUP_CACHE_MASK_DFL 1023
|
|
#define LOOKUP_CACHE_SIZE_MAX 65536 /* Allows max CACHE_MASK 0xFFFF */
|
|
#define LOOKUP_CACHE_SHIFT_DFL 2
|
|
|
|
struct __mf_cache __mf_lookup_cache [LOOKUP_CACHE_SIZE_MAX];
|
|
uintptr_t __mf_lc_mask = LOOKUP_CACHE_MASK_DFL;
|
|
unsigned char __mf_lc_shift = LOOKUP_CACHE_SHIFT_DFL;
|
|
#define LOOKUP_CACHE_SIZE (__mf_lc_mask + 1)
|
|
|
|
struct __mf_options __mf_opts;
|
|
int __mf_starting_p = 1;
|
|
|
|
#ifdef LIBMUDFLAPTH
|
|
#if defined(HAVE_TLS) && !defined(USE_EMUTLS)
|
|
__thread enum __mf_state_enum __mf_state_1 = reentrant;
|
|
#endif
|
|
#else
|
|
enum __mf_state_enum __mf_state_1 = reentrant;
|
|
#endif
|
|
|
|
#ifdef LIBMUDFLAPTH
|
|
pthread_mutex_t __mf_biglock =
|
|
#ifdef PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP
|
|
PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP;
|
|
#else
|
|
PTHREAD_MUTEX_INITIALIZER;
|
|
#endif
|
|
#endif
|
|
|
|
/* Use HAVE_PTHREAD_H here instead of LIBMUDFLAPTH, so that even
|
|
the libmudflap.la (no threading support) can diagnose whether
|
|
the application is linked with -lpthread. See __mf_usage() below. */
|
|
#if HAVE_PTHREAD_H
|
|
#ifdef _POSIX_THREADS
|
|
#pragma weak pthread_join
|
|
#else
|
|
#define pthread_join NULL
|
|
#endif
|
|
#endif
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* stats-related globals. */
|
|
|
|
static unsigned long __mf_count_check;
|
|
static unsigned long __mf_lookup_cache_reusecount [LOOKUP_CACHE_SIZE_MAX];
|
|
static unsigned long __mf_count_register;
|
|
static unsigned long __mf_total_register_size [__MF_TYPE_MAX+1];
|
|
static unsigned long __mf_count_unregister;
|
|
static unsigned long __mf_total_unregister_size;
|
|
static unsigned long __mf_count_violation [__MF_VIOL_WATCH+1];
|
|
static unsigned long __mf_sigusr1_received;
|
|
static unsigned long __mf_sigusr1_handled;
|
|
/* not static */ unsigned long __mf_reentrancy;
|
|
#ifdef LIBMUDFLAPTH
|
|
/* not static */ unsigned long __mf_lock_contention;
|
|
#endif
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* mode-check-related globals. */
|
|
|
|
typedef struct __mf_object
|
|
{
|
|
uintptr_t low, high; /* __mf_register parameters */
|
|
const char *name;
|
|
char type; /* __MF_TYPE_something */
|
|
char watching_p; /* Trigger a VIOL_WATCH on access? */
|
|
unsigned read_count; /* Number of times __mf_check/read was called on this object. */
|
|
unsigned write_count; /* Likewise for __mf_check/write. */
|
|
unsigned liveness; /* A measure of recent checking activity. */
|
|
unsigned description_epoch; /* Last epoch __mf_describe_object printed this. */
|
|
|
|
uintptr_t alloc_pc;
|
|
struct timeval alloc_time;
|
|
char **alloc_backtrace;
|
|
size_t alloc_backtrace_size;
|
|
#ifdef LIBMUDFLAPTH
|
|
pthread_t alloc_thread;
|
|
#endif
|
|
|
|
int deallocated_p;
|
|
uintptr_t dealloc_pc;
|
|
struct timeval dealloc_time;
|
|
char **dealloc_backtrace;
|
|
size_t dealloc_backtrace_size;
|
|
#ifdef LIBMUDFLAPTH
|
|
pthread_t dealloc_thread;
|
|
#endif
|
|
} __mf_object_t;
|
|
|
|
/* Live objects: splay trees, separated by type, ordered on .low (base address). */
|
|
/* Actually stored as static vars within lookup function below. */
|
|
|
|
/* Dead objects: circular arrays; _MIN_CEM .. _MAX_CEM only */
|
|
static unsigned __mf_object_dead_head[__MF_TYPE_MAX_CEM+1]; /* next empty spot */
|
|
static __mf_object_t *__mf_object_cemetary[__MF_TYPE_MAX_CEM+1][__MF_PERSIST_MAX];
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* Forward function declarations */
|
|
|
|
void __mf_init () CTOR;
|
|
static void __mf_sigusr1_respond ();
|
|
static unsigned __mf_find_objects (uintptr_t ptr_low, uintptr_t ptr_high,
|
|
__mf_object_t **objs, unsigned max_objs);
|
|
static unsigned __mf_find_objects2 (uintptr_t ptr_low, uintptr_t ptr_high,
|
|
__mf_object_t **objs, unsigned max_objs, int type);
|
|
static unsigned __mf_find_dead_objects (uintptr_t ptr_low, uintptr_t ptr_high,
|
|
__mf_object_t **objs, unsigned max_objs);
|
|
static void __mf_adapt_cache ();
|
|
static void __mf_describe_object (__mf_object_t *obj);
|
|
static unsigned __mf_watch_or_not (void *ptr, size_t sz, char flag);
|
|
static mfsplay_tree __mf_object_tree (int type);
|
|
static void __mf_link_object (__mf_object_t *node);
|
|
static void __mf_unlink_object (__mf_object_t *node);
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* Configuration engine */
|
|
|
|
static void
|
|
__mf_set_default_options ()
|
|
{
|
|
memset (& __mf_opts, 0, sizeof (__mf_opts));
|
|
|
|
__mf_opts.adapt_cache = 1000003;
|
|
__mf_opts.abbreviate = 1;
|
|
__mf_opts.verbose_violations = 1;
|
|
__mf_opts.free_queue_length = 4;
|
|
__mf_opts.persistent_count = 100;
|
|
__mf_opts.crumple_zone = 32;
|
|
__mf_opts.backtrace = 4;
|
|
__mf_opts.timestamps = 1;
|
|
__mf_opts.mudflap_mode = mode_check;
|
|
__mf_opts.violation_mode = viol_nop;
|
|
#ifdef HAVE___LIBC_FREERES
|
|
__mf_opts.call_libc_freeres = 1;
|
|
#endif
|
|
__mf_opts.heur_std_data = 1;
|
|
#ifdef LIBMUDFLAPTH
|
|
__mf_opts.thread_stack = 0;
|
|
#endif
|
|
|
|
/* PR41443: Beware that the above flags will be applied to
|
|
setuid/setgid binaries, and cannot be overriden with
|
|
$MUDFLAP_OPTIONS. So the defaults must be non-exploitable.
|
|
|
|
Should we consider making the default violation_mode something
|
|
harsher than viol_nop? OTOH, glibc's MALLOC_CHECK_ is disabled
|
|
by default for these same programs. */
|
|
}
|
|
|
|
static struct mudoption
|
|
{
|
|
char *name;
|
|
char *description;
|
|
enum
|
|
{
|
|
set_option,
|
|
read_integer_option,
|
|
} type;
|
|
unsigned value;
|
|
unsigned *target;
|
|
}
|
|
options [] =
|
|
{
|
|
{"mode-nop",
|
|
"mudflaps do nothing",
|
|
set_option, (unsigned)mode_nop, (unsigned *)&__mf_opts.mudflap_mode},
|
|
{"mode-populate",
|
|
"mudflaps populate object tree",
|
|
set_option, (unsigned)mode_populate, (unsigned *)&__mf_opts.mudflap_mode},
|
|
{"mode-check",
|
|
"mudflaps check for memory violations",
|
|
set_option, (unsigned)mode_check, (unsigned *)&__mf_opts.mudflap_mode},
|
|
{"mode-violate",
|
|
"mudflaps always cause violations (diagnostic)",
|
|
set_option, (unsigned)mode_violate, (unsigned *)&__mf_opts.mudflap_mode},
|
|
|
|
{"viol-nop",
|
|
"violations do not change program execution",
|
|
set_option, (unsigned)viol_nop, (unsigned *)&__mf_opts.violation_mode},
|
|
{"viol-abort",
|
|
"violations cause a call to abort()",
|
|
set_option, (unsigned)viol_abort, (unsigned *)&__mf_opts.violation_mode},
|
|
{"viol-segv",
|
|
"violations are promoted to SIGSEGV signals",
|
|
set_option, (unsigned)viol_segv, (unsigned *)&__mf_opts.violation_mode},
|
|
{"viol-gdb",
|
|
"violations fork a gdb process attached to current program",
|
|
set_option, (unsigned)viol_gdb, (unsigned *)&__mf_opts.violation_mode},
|
|
{"trace-calls",
|
|
"trace calls to mudflap runtime library",
|
|
set_option, 1, &__mf_opts.trace_mf_calls},
|
|
{"verbose-trace",
|
|
"trace internal events within mudflap runtime library",
|
|
set_option, 1, &__mf_opts.verbose_trace},
|
|
{"collect-stats",
|
|
"collect statistics on mudflap's operation",
|
|
set_option, 1, &__mf_opts.collect_stats},
|
|
#ifdef SIGUSR1
|
|
{"sigusr1-report",
|
|
"print report upon SIGUSR1",
|
|
set_option, 1, &__mf_opts.sigusr1_report},
|
|
#endif
|
|
{"internal-checking",
|
|
"perform more expensive internal checking",
|
|
set_option, 1, &__mf_opts.internal_checking},
|
|
{"print-leaks",
|
|
"print any memory leaks at program shutdown",
|
|
set_option, 1, &__mf_opts.print_leaks},
|
|
#ifdef HAVE___LIBC_FREERES
|
|
{"libc-freeres",
|
|
"call glibc __libc_freeres at shutdown for better leak data",
|
|
set_option, 1, &__mf_opts.call_libc_freeres},
|
|
#endif
|
|
{"check-initialization",
|
|
"detect uninitialized object reads",
|
|
set_option, 1, &__mf_opts.check_initialization},
|
|
{"verbose-violations",
|
|
"print verbose messages when memory violations occur",
|
|
set_option, 1, &__mf_opts.verbose_violations},
|
|
{"abbreviate",
|
|
"abbreviate repetitive listings",
|
|
set_option, 1, &__mf_opts.abbreviate},
|
|
{"timestamps",
|
|
"track object lifetime timestamps",
|
|
set_option, 1, &__mf_opts.timestamps},
|
|
{"ignore-reads",
|
|
"ignore read accesses - assume okay",
|
|
set_option, 1, &__mf_opts.ignore_reads},
|
|
{"wipe-stack",
|
|
"wipe stack objects at unwind",
|
|
set_option, 1, &__mf_opts.wipe_stack},
|
|
{"wipe-heap",
|
|
"wipe heap objects at free",
|
|
set_option, 1, &__mf_opts.wipe_heap},
|
|
{"heur-proc-map",
|
|
"support /proc/self/map heuristics",
|
|
set_option, 1, &__mf_opts.heur_proc_map},
|
|
{"heur-stack-bound",
|
|
"enable a simple upper stack bound heuristic",
|
|
set_option, 1, &__mf_opts.heur_stack_bound},
|
|
{"heur-start-end",
|
|
"support _start.._end heuristics",
|
|
set_option, 1, &__mf_opts.heur_start_end},
|
|
{"heur-stdlib",
|
|
"register standard library data (argv, errno, stdin, ...)",
|
|
set_option, 1, &__mf_opts.heur_std_data},
|
|
{"free-queue-length",
|
|
"queue N deferred free() calls before performing them",
|
|
read_integer_option, 0, &__mf_opts.free_queue_length},
|
|
{"persistent-count",
|
|
"keep a history of N unregistered regions",
|
|
read_integer_option, 0, &__mf_opts.persistent_count},
|
|
{"crumple-zone",
|
|
"surround allocations with crumple zones of N bytes",
|
|
read_integer_option, 0, &__mf_opts.crumple_zone},
|
|
/* XXX: not type-safe.
|
|
{"lc-mask",
|
|
"set lookup cache size mask to N (2**M - 1)",
|
|
read_integer_option, 0, (int *)(&__mf_lc_mask)},
|
|
{"lc-shift",
|
|
"set lookup cache pointer shift",
|
|
read_integer_option, 0, (int *)(&__mf_lc_shift)},
|
|
*/
|
|
{"lc-adapt",
|
|
"adapt mask/shift parameters after N cache misses",
|
|
read_integer_option, 1, &__mf_opts.adapt_cache},
|
|
{"backtrace",
|
|
"keep an N-level stack trace of each call context",
|
|
read_integer_option, 0, &__mf_opts.backtrace},
|
|
#ifdef LIBMUDFLAPTH
|
|
{"thread-stack",
|
|
"override thread stacks allocation: N kB",
|
|
read_integer_option, 0, &__mf_opts.thread_stack},
|
|
#endif
|
|
{0, 0, set_option, 0, NULL}
|
|
};
|
|
|
|
static void
|
|
__mf_usage ()
|
|
{
|
|
struct mudoption *opt;
|
|
|
|
fprintf (stderr,
|
|
"This is a %s%sGCC \"mudflap\" memory-checked binary.\n"
|
|
"Mudflap is Copyright (C) 2002-2013 Free Software Foundation, Inc.\n"
|
|
"\n"
|
|
"Unless setuid, a program's mudflap options be set by an environment variable:\n"
|
|
"\n"
|
|
"$ export MUDFLAP_OPTIONS='<options>'\n"
|
|
"$ <mudflapped_program>\n"
|
|
"\n"
|
|
"where <options> is a space-separated list of \n"
|
|
"any of the following options. Use `-no-OPTION' to disable options.\n"
|
|
"\n",
|
|
#if HAVE_PTHREAD_H
|
|
(pthread_join ? "multi-threaded " : "single-threaded "),
|
|
#else
|
|
"",
|
|
#endif
|
|
#if LIBMUDFLAPTH
|
|
"thread-aware "
|
|
#else
|
|
"thread-unaware "
|
|
#endif
|
|
);
|
|
/* XXX: The multi-threaded thread-unaware combination is bad. */
|
|
|
|
for (opt = options; opt->name; opt++)
|
|
{
|
|
int default_p = (opt->value == * opt->target);
|
|
|
|
switch (opt->type)
|
|
{
|
|
char buf[128];
|
|
case set_option:
|
|
fprintf (stderr, "-%-23.23s %s", opt->name, opt->description);
|
|
if (default_p)
|
|
fprintf (stderr, " [active]\n");
|
|
else
|
|
fprintf (stderr, "\n");
|
|
break;
|
|
case read_integer_option:
|
|
strncpy (buf, opt->name, 128);
|
|
strncpy (buf + strlen (opt->name), "=N", 2);
|
|
fprintf (stderr, "-%-23.23s %s", buf, opt->description);
|
|
fprintf (stderr, " [%d]\n", * opt->target);
|
|
break;
|
|
default: abort();
|
|
}
|
|
}
|
|
|
|
fprintf (stderr, "\n");
|
|
}
|
|
|
|
|
|
int
|
|
__mf_set_options (const char *optstr)
|
|
{
|
|
int rc;
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
rc = __mfu_set_options (optstr);
|
|
/* XXX: It's not really that easy. A change to a bunch of parameters
|
|
can require updating auxiliary state or risk crashing:
|
|
free_queue_length, crumple_zone ... */
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
return rc;
|
|
}
|
|
|
|
|
|
int
|
|
__mfu_set_options (const char *optstr)
|
|
{
|
|
struct mudoption *opts = 0;
|
|
char *nxt = 0;
|
|
long tmp = 0;
|
|
int rc = 0;
|
|
const char *saved_optstr = optstr;
|
|
|
|
/* XXX: bounds-check for optstr! */
|
|
|
|
while (*optstr)
|
|
{
|
|
switch (*optstr) {
|
|
case ' ':
|
|
case '\t':
|
|
case '\n':
|
|
optstr++;
|
|
break;
|
|
|
|
case '-':
|
|
if (*optstr+1)
|
|
{
|
|
int negate = 0;
|
|
optstr++;
|
|
|
|
if (*optstr == '?' ||
|
|
strncmp (optstr, "help", 4) == 0)
|
|
{
|
|
/* Caller will print help and exit. */
|
|
return -1;
|
|
}
|
|
|
|
if (strncmp (optstr, "no-", 3) == 0)
|
|
{
|
|
negate = 1;
|
|
optstr = & optstr[3];
|
|
}
|
|
|
|
for (opts = options; opts->name; opts++)
|
|
{
|
|
if (strncmp (optstr, opts->name, strlen (opts->name)) == 0)
|
|
{
|
|
optstr += strlen (opts->name);
|
|
assert (opts->target);
|
|
switch (opts->type)
|
|
{
|
|
case set_option:
|
|
if (negate)
|
|
*(opts->target) = 0;
|
|
else
|
|
*(opts->target) = opts->value;
|
|
break;
|
|
case read_integer_option:
|
|
if (! negate && (*optstr == '=' && *(optstr+1)))
|
|
{
|
|
optstr++;
|
|
tmp = strtol (optstr, &nxt, 10);
|
|
if ((optstr != nxt) && (tmp != LONG_MAX))
|
|
{
|
|
optstr = nxt;
|
|
*(opts->target) = (int)tmp;
|
|
}
|
|
}
|
|
else if (negate)
|
|
* opts->target = 0;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
|
|
default:
|
|
fprintf (stderr,
|
|
"warning: unrecognized string '%s' in mudflap options\n",
|
|
optstr);
|
|
optstr += strlen (optstr);
|
|
rc = -1;
|
|
break;
|
|
}
|
|
}
|
|
|
|
/* Special post-processing: bound __mf_lc_mask and free_queue_length for security. */
|
|
__mf_lc_mask &= (LOOKUP_CACHE_SIZE_MAX - 1);
|
|
__mf_opts.free_queue_length &= (__MF_FREEQ_MAX - 1);
|
|
|
|
/* Clear the lookup cache, in case the parameters got changed. */
|
|
/* XXX: race */
|
|
memset (__mf_lookup_cache, 0, sizeof(__mf_lookup_cache));
|
|
/* void slot 0 */
|
|
__mf_lookup_cache[0].low = MAXPTR;
|
|
|
|
TRACE ("set options from `%s'\n", saved_optstr);
|
|
|
|
/* Call this unconditionally, in case -sigusr1-report was toggled. */
|
|
__mf_sigusr1_respond ();
|
|
|
|
return rc;
|
|
}
|
|
|
|
|
|
#ifdef PIC
|
|
|
|
void
|
|
__mf_resolve_single_dynamic (struct __mf_dynamic_entry *e)
|
|
{
|
|
char *err;
|
|
|
|
assert (e);
|
|
if (e->pointer) return;
|
|
|
|
#if HAVE_DLVSYM
|
|
if (e->version != NULL && e->version[0] != '\0') /* non-null/empty */
|
|
e->pointer = dlvsym (RTLD_NEXT, e->name, e->version);
|
|
else
|
|
#endif
|
|
e->pointer = dlsym (RTLD_NEXT, e->name);
|
|
|
|
err = dlerror ();
|
|
|
|
if (err)
|
|
{
|
|
fprintf (stderr, "mf: error in dlsym(\"%s\"): %s\n",
|
|
e->name, err);
|
|
abort ();
|
|
}
|
|
if (! e->pointer)
|
|
{
|
|
fprintf (stderr, "mf: dlsym(\"%s\") = NULL\n", e->name);
|
|
abort ();
|
|
}
|
|
}
|
|
|
|
|
|
static void
|
|
__mf_resolve_dynamics ()
|
|
{
|
|
int i;
|
|
for (i = 0; i < dyn_INITRESOLVE; i++)
|
|
__mf_resolve_single_dynamic (& __mf_dynamic[i]);
|
|
}
|
|
|
|
|
|
/* NB: order must match enums in mf-impl.h */
|
|
struct __mf_dynamic_entry __mf_dynamic [] =
|
|
{
|
|
{NULL, "calloc", NULL},
|
|
{NULL, "free", NULL},
|
|
{NULL, "malloc", NULL},
|
|
{NULL, "mmap", NULL},
|
|
#ifdef HAVE_MMAP64
|
|
{NULL, "mmap64", NULL},
|
|
#endif
|
|
{NULL, "munmap", NULL},
|
|
{NULL, "realloc", NULL},
|
|
{NULL, "DUMMY", NULL}, /* dyn_INITRESOLVE */
|
|
#ifdef LIBMUDFLAPTH
|
|
{NULL, "pthread_create", PTHREAD_CREATE_VERSION},
|
|
{NULL, "pthread_join", NULL},
|
|
{NULL, "pthread_exit", NULL}
|
|
#endif
|
|
};
|
|
|
|
#endif /* PIC */
|
|
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
|
|
/* Lookup & manage automatic initialization of the five or so splay trees. */
|
|
static mfsplay_tree
|
|
__mf_object_tree (int type)
|
|
{
|
|
static mfsplay_tree trees [__MF_TYPE_MAX+1];
|
|
assert (type >= 0 && type <= __MF_TYPE_MAX);
|
|
if (UNLIKELY (trees[type] == NULL))
|
|
trees[type] = mfsplay_tree_new ();
|
|
return trees[type];
|
|
}
|
|
|
|
|
|
/* not static */void
|
|
__mf_init ()
|
|
{
|
|
char *ov = 0;
|
|
|
|
/* Return if initialization has already been done. */
|
|
if (LIKELY (__mf_starting_p == 0))
|
|
return;
|
|
|
|
#if defined(__FreeBSD__) && defined(LIBMUDFLAPTH)
|
|
pthread_self();
|
|
LOCKTH ();
|
|
UNLOCKTH ();
|
|
#endif /* Prime mutex which calls calloc upon first lock to avoid deadlock. */
|
|
|
|
/* This initial bootstrap phase requires that __mf_starting_p = 1. */
|
|
#ifdef PIC
|
|
__mf_resolve_dynamics ();
|
|
#endif
|
|
__mf_starting_p = 0;
|
|
|
|
__mf_set_state (active);
|
|
|
|
__mf_set_default_options ();
|
|
|
|
if (getuid () == geteuid () && getgid () == getegid ()) /* PR41433, not setuid */
|
|
ov = getenv ("MUDFLAP_OPTIONS");
|
|
if (ov)
|
|
{
|
|
int rc = __mfu_set_options (ov);
|
|
if (rc < 0)
|
|
{
|
|
__mf_usage ();
|
|
exit (1);
|
|
}
|
|
}
|
|
|
|
/* Initialize to a non-zero description epoch. */
|
|
__mf_describe_object (NULL);
|
|
|
|
#define REG_RESERVED(obj) \
|
|
__mf_register (& obj, sizeof(obj), __MF_TYPE_NOACCESS, # obj)
|
|
|
|
REG_RESERVED (__mf_lookup_cache);
|
|
REG_RESERVED (__mf_lc_mask);
|
|
REG_RESERVED (__mf_lc_shift);
|
|
/* XXX: others of our statics? */
|
|
|
|
/* Prevent access to *NULL. */
|
|
__mf_register (MINPTR, 1, __MF_TYPE_NOACCESS, "NULL");
|
|
__mf_lookup_cache[0].low = (uintptr_t) -1;
|
|
}
|
|
|
|
|
|
|
|
int
|
|
__wrap_main (int argc, char* argv[])
|
|
{
|
|
extern char **environ;
|
|
extern int main ();
|
|
extern int __real_main ();
|
|
static int been_here = 0;
|
|
|
|
if (__mf_opts.heur_std_data && ! been_here)
|
|
{
|
|
unsigned i;
|
|
|
|
been_here = 1;
|
|
__mf_register (argv, sizeof(char *)*(argc+1), __MF_TYPE_STATIC, "argv[]");
|
|
for (i=0; i<argc; i++)
|
|
{
|
|
unsigned j = strlen (argv[i]);
|
|
__mf_register (argv[i], j+1, __MF_TYPE_STATIC, "argv element");
|
|
}
|
|
|
|
for (i=0; ; i++)
|
|
{
|
|
char *e = environ[i];
|
|
unsigned j;
|
|
if (e == NULL) break;
|
|
j = strlen (environ[i]);
|
|
__mf_register (environ[i], j+1, __MF_TYPE_STATIC, "environ element");
|
|
}
|
|
__mf_register (environ, sizeof(char *)*(i+1), __MF_TYPE_STATIC, "environ[]");
|
|
|
|
__mf_register (& errno, sizeof (errno), __MF_TYPE_STATIC, "errno area");
|
|
|
|
#if !(defined(__sun__) && defined(__svr4__))
|
|
/* Conflicts with the automatic registration of __iob[]. */
|
|
__mf_register (stdin, sizeof (*stdin), __MF_TYPE_STATIC, "stdin");
|
|
__mf_register (stdout, sizeof (*stdout), __MF_TYPE_STATIC, "stdout");
|
|
__mf_register (stderr, sizeof (*stderr), __MF_TYPE_STATIC, "stderr");
|
|
#endif
|
|
|
|
/* Make some effort to register ctype.h static arrays. */
|
|
#if defined(__sun__) && defined(__svr4__)
|
|
/* __ctype[] is declared without size, but MB_CUR_MAX is the last
|
|
member. There seems to be no proper way to determine the size. */
|
|
__mf_register (__ctype, &MB_CUR_MAX - &__ctype[0] + 1, __MF_TYPE_STATIC, "__ctype");
|
|
/* __ctype_mask points at _C_masks[1]. The size can only determined
|
|
using nm on libc.so.1. */
|
|
__mf_register (__ctype_mask - 1, 1028, __MF_TYPE_STATIC, "_C_masks");
|
|
#endif
|
|
/* On modern Linux GLIBC, these are thread-specific and changeable, and are dealt
|
|
with in mf-hooks2.c. */
|
|
}
|
|
|
|
#ifdef PIC
|
|
return main (argc, argv, environ);
|
|
#else
|
|
return __real_main (argc, argv, environ);
|
|
#endif
|
|
}
|
|
|
|
|
|
|
|
extern void __mf_fini () DTOR;
|
|
void __mf_fini ()
|
|
{
|
|
TRACE ("__mf_fini\n");
|
|
__mfu_report ();
|
|
|
|
#ifndef PIC
|
|
/* Since we didn't populate the tree for allocations in constructors
|
|
before __mf_init, we cannot check destructors after __mf_fini. */
|
|
__mf_opts.mudflap_mode = mode_nop;
|
|
#endif
|
|
}
|
|
|
|
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* __mf_check */
|
|
|
|
void __mf_check (void *ptr, size_t sz, int type, const char *location)
|
|
{
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
__mfu_check (ptr, sz, type, location);
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
}
|
|
|
|
|
|
void __mfu_check (void *ptr, size_t sz, int type, const char *location)
|
|
{
|
|
unsigned entry_idx = __MF_CACHE_INDEX (ptr);
|
|
struct __mf_cache *entry = & __mf_lookup_cache [entry_idx];
|
|
int judgement = 0; /* 0=undecided; <0=violation; >0=okay */
|
|
uintptr_t ptr_low = (uintptr_t) ptr;
|
|
uintptr_t ptr_high = CLAMPSZ (ptr, sz);
|
|
struct __mf_cache old_entry = *entry;
|
|
|
|
if (UNLIKELY (__mf_opts.sigusr1_report))
|
|
__mf_sigusr1_respond ();
|
|
if (UNLIKELY (__mf_opts.ignore_reads && type == 0))
|
|
return;
|
|
|
|
TRACE ("check ptr=%p b=%u size=%lu %s location=`%s'\n",
|
|
ptr, entry_idx, (unsigned long)sz,
|
|
(type == 0 ? "read" : "write"), location);
|
|
|
|
switch (__mf_opts.mudflap_mode)
|
|
{
|
|
case mode_nop:
|
|
/* It is tempting to poison the cache here similarly to
|
|
mode_populate. However that eliminates a valuable
|
|
distinction between these two modes. mode_nop is useful to
|
|
let a user count & trace every single check / registration
|
|
call. mode_populate is useful to let a program run fast
|
|
while unchecked.
|
|
*/
|
|
judgement = 1;
|
|
break;
|
|
|
|
case mode_populate:
|
|
entry->low = ptr_low;
|
|
entry->high = ptr_high;
|
|
judgement = 1;
|
|
break;
|
|
|
|
case mode_check:
|
|
{
|
|
unsigned heuristics = 0;
|
|
|
|
/* Advance aging/adaptation counters. */
|
|
static unsigned adapt_count;
|
|
adapt_count ++;
|
|
if (UNLIKELY (__mf_opts.adapt_cache > 0 &&
|
|
adapt_count > __mf_opts.adapt_cache))
|
|
{
|
|
adapt_count = 0;
|
|
__mf_adapt_cache ();
|
|
}
|
|
|
|
/* Looping only occurs if heuristics were triggered. */
|
|
while (judgement == 0)
|
|
{
|
|
DECLARE (void, free, void *p);
|
|
__mf_object_t* ovr_obj[1];
|
|
unsigned obj_count;
|
|
__mf_object_t** all_ovr_obj = NULL;
|
|
__mf_object_t** dealloc_me = NULL;
|
|
unsigned i;
|
|
|
|
/* Find all overlapping objects. Be optimistic that there is just one. */
|
|
obj_count = __mf_find_objects (ptr_low, ptr_high, ovr_obj, 1);
|
|
if (UNLIKELY (obj_count > 1))
|
|
{
|
|
/* Allocate a real buffer and do the search again. */
|
|
DECLARE (void *, malloc, size_t c);
|
|
unsigned n;
|
|
all_ovr_obj = CALL_REAL (malloc, (sizeof (__mf_object_t *) *
|
|
obj_count));
|
|
if (all_ovr_obj == NULL) abort ();
|
|
n = __mf_find_objects (ptr_low, ptr_high, all_ovr_obj, obj_count);
|
|
assert (n == obj_count);
|
|
dealloc_me = all_ovr_obj;
|
|
}
|
|
else
|
|
{
|
|
all_ovr_obj = ovr_obj;
|
|
dealloc_me = NULL;
|
|
}
|
|
|
|
/* Update object statistics. */
|
|
for (i = 0; i < obj_count; i++)
|
|
{
|
|
__mf_object_t *obj = all_ovr_obj[i];
|
|
assert (obj != NULL);
|
|
if (type == __MF_CHECK_READ)
|
|
obj->read_count ++;
|
|
else
|
|
obj->write_count ++;
|
|
obj->liveness ++;
|
|
}
|
|
|
|
/* Iterate over the various objects. There are a number of special cases. */
|
|
for (i = 0; i < obj_count; i++)
|
|
{
|
|
__mf_object_t *obj = all_ovr_obj[i];
|
|
|
|
/* Any __MF_TYPE_NOACCESS hit is bad. */
|
|
if (UNLIKELY (obj->type == __MF_TYPE_NOACCESS))
|
|
judgement = -1;
|
|
|
|
/* Any object with a watch flag is bad. */
|
|
if (UNLIKELY (obj->watching_p))
|
|
judgement = -2; /* trigger VIOL_WATCH */
|
|
|
|
/* A read from an uninitialized object is bad. */
|
|
if (UNLIKELY (__mf_opts.check_initialization
|
|
/* reading */
|
|
&& type == __MF_CHECK_READ
|
|
/* not written */
|
|
&& obj->write_count == 0
|
|
/* uninitialized (heap) */
|
|
&& obj->type == __MF_TYPE_HEAP))
|
|
judgement = -1;
|
|
}
|
|
|
|
/* We now know that the access spans no invalid objects. */
|
|
if (LIKELY (judgement >= 0))
|
|
for (i = 0; i < obj_count; i++)
|
|
{
|
|
__mf_object_t *obj = all_ovr_obj[i];
|
|
|
|
/* Is this access entirely contained within this object? */
|
|
if (LIKELY (ptr_low >= obj->low && ptr_high <= obj->high))
|
|
{
|
|
/* Valid access. */
|
|
entry->low = obj->low;
|
|
entry->high = obj->high;
|
|
judgement = 1;
|
|
}
|
|
}
|
|
|
|
/* This access runs off the end of one valid object. That
|
|
could be okay, if other valid objects fill in all the
|
|
holes. We allow this only for HEAP and GUESS type
|
|
objects. Accesses to STATIC and STACK variables
|
|
should not be allowed to span. */
|
|
if (UNLIKELY ((judgement == 0) && (obj_count > 1)))
|
|
{
|
|
unsigned uncovered = 0;
|
|
for (i = 0; i < obj_count; i++)
|
|
{
|
|
__mf_object_t *obj = all_ovr_obj[i];
|
|
int j, uncovered_low_p, uncovered_high_p;
|
|
uintptr_t ptr_lower, ptr_higher;
|
|
|
|
uncovered_low_p = ptr_low < obj->low;
|
|
ptr_lower = CLAMPSUB (obj->low, 1);
|
|
uncovered_high_p = ptr_high > obj->high;
|
|
ptr_higher = CLAMPADD (obj->high, 1);
|
|
|
|
for (j = 0; j < obj_count; j++)
|
|
{
|
|
__mf_object_t *obj2 = all_ovr_obj[j];
|
|
|
|
if (i == j) continue;
|
|
|
|
/* Filter out objects that cannot be spanned across. */
|
|
if (obj2->type == __MF_TYPE_STACK
|
|
|| obj2->type == __MF_TYPE_STATIC)
|
|
continue;
|
|
|
|
/* Consider a side "covered" if obj2 includes
|
|
the next byte on that side. */
|
|
if (uncovered_low_p
|
|
&& (ptr_lower >= obj2->low && ptr_lower <= obj2->high))
|
|
uncovered_low_p = 0;
|
|
if (uncovered_high_p
|
|
&& (ptr_high >= obj2->low && ptr_higher <= obj2->high))
|
|
uncovered_high_p = 0;
|
|
}
|
|
|
|
if (uncovered_low_p || uncovered_high_p)
|
|
uncovered ++;
|
|
}
|
|
|
|
/* Success if no overlapping objects are uncovered. */
|
|
if (uncovered == 0)
|
|
judgement = 1;
|
|
}
|
|
|
|
|
|
if (dealloc_me != NULL)
|
|
CALL_REAL (free, dealloc_me);
|
|
|
|
/* If the judgment is still unknown at this stage, loop
|
|
around at most one more time. */
|
|
if (judgement == 0)
|
|
{
|
|
if (heuristics++ < 2) /* XXX parametrize this number? */
|
|
judgement = __mf_heuristic_check (ptr_low, ptr_high);
|
|
else
|
|
judgement = -1;
|
|
}
|
|
}
|
|
|
|
}
|
|
break;
|
|
|
|
case mode_violate:
|
|
judgement = -1;
|
|
break;
|
|
}
|
|
|
|
if (__mf_opts.collect_stats)
|
|
{
|
|
__mf_count_check ++;
|
|
|
|
if (LIKELY (old_entry.low != entry->low || old_entry.high != entry->high))
|
|
/* && (old_entry.low != 0) && (old_entry.high != 0)) */
|
|
__mf_lookup_cache_reusecount [entry_idx] ++;
|
|
}
|
|
|
|
if (UNLIKELY (judgement < 0))
|
|
__mf_violation (ptr, sz,
|
|
(uintptr_t) __builtin_return_address (0), location,
|
|
((judgement == -1) ?
|
|
(type == __MF_CHECK_READ ? __MF_VIOL_READ : __MF_VIOL_WRITE) :
|
|
__MF_VIOL_WATCH));
|
|
}
|
|
|
|
|
|
static __mf_object_t *
|
|
__mf_insert_new_object (uintptr_t low, uintptr_t high, int type,
|
|
const char *name, uintptr_t pc)
|
|
{
|
|
DECLARE (void *, calloc, size_t c, size_t n);
|
|
|
|
__mf_object_t *new_obj;
|
|
new_obj = CALL_REAL (calloc, 1, sizeof(__mf_object_t));
|
|
new_obj->low = low;
|
|
new_obj->high = high;
|
|
new_obj->type = type;
|
|
new_obj->name = name;
|
|
new_obj->alloc_pc = pc;
|
|
#if HAVE_GETTIMEOFDAY
|
|
if (__mf_opts.timestamps)
|
|
gettimeofday (& new_obj->alloc_time, NULL);
|
|
#endif
|
|
#if LIBMUDFLAPTH
|
|
new_obj->alloc_thread = pthread_self ();
|
|
#endif
|
|
|
|
if (__mf_opts.backtrace > 0 && (type == __MF_TYPE_HEAP || type == __MF_TYPE_HEAP_I))
|
|
new_obj->alloc_backtrace_size =
|
|
__mf_backtrace (& new_obj->alloc_backtrace,
|
|
(void *) pc, 2);
|
|
|
|
__mf_link_object (new_obj);
|
|
return new_obj;
|
|
}
|
|
|
|
|
|
static void
|
|
__mf_uncache_object (__mf_object_t *old_obj)
|
|
{
|
|
/* Remove any low/high pointers for this object from the lookup cache. */
|
|
|
|
/* Can it possibly exist in the cache? */
|
|
if (LIKELY (old_obj->read_count + old_obj->write_count))
|
|
{
|
|
uintptr_t low = old_obj->low;
|
|
uintptr_t high = old_obj->high;
|
|
struct __mf_cache *entry;
|
|
unsigned i;
|
|
if ((high - low) >= (__mf_lc_mask << __mf_lc_shift))
|
|
{
|
|
/* For large objects (>= cache size - 1) check the whole cache. */
|
|
entry = & __mf_lookup_cache [0];
|
|
for (i = 0; i <= __mf_lc_mask; i++, entry++)
|
|
{
|
|
/* NB: the "||" in the following test permits this code to
|
|
tolerate the situation introduced by __mf_check over
|
|
contiguous objects, where a cache entry spans several
|
|
objects. */
|
|
if (entry->low == low || entry->high == high)
|
|
{
|
|
entry->low = MAXPTR;
|
|
entry->high = MINPTR;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
/* Object is now smaller then cache size. */
|
|
unsigned entry_low_idx = __MF_CACHE_INDEX (low);
|
|
unsigned entry_high_idx = __MF_CACHE_INDEX (high);
|
|
if (entry_low_idx <= entry_high_idx)
|
|
{
|
|
entry = & __mf_lookup_cache [entry_low_idx];
|
|
for (i = entry_low_idx; i <= entry_high_idx; i++, entry++)
|
|
{
|
|
/* NB: the "||" in the following test permits this code to
|
|
tolerate the situation introduced by __mf_check over
|
|
contiguous objects, where a cache entry spans several
|
|
objects. */
|
|
if (entry->low == low || entry->high == high)
|
|
{
|
|
entry->low = MAXPTR;
|
|
entry->high = MINPTR;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
/* Object wrapped around the end of the cache. First search
|
|
from low to end of cache and then from 0 to high. */
|
|
entry = & __mf_lookup_cache [entry_low_idx];
|
|
for (i = entry_low_idx; i <= __mf_lc_mask; i++, entry++)
|
|
{
|
|
/* NB: the "||" in the following test permits this code to
|
|
tolerate the situation introduced by __mf_check over
|
|
contiguous objects, where a cache entry spans several
|
|
objects. */
|
|
if (entry->low == low || entry->high == high)
|
|
{
|
|
entry->low = MAXPTR;
|
|
entry->high = MINPTR;
|
|
}
|
|
}
|
|
entry = & __mf_lookup_cache [0];
|
|
for (i = 0; i <= entry_high_idx; i++, entry++)
|
|
{
|
|
/* NB: the "||" in the following test permits this code to
|
|
tolerate the situation introduced by __mf_check over
|
|
contiguous objects, where a cache entry spans several
|
|
objects. */
|
|
if (entry->low == low || entry->high == high)
|
|
{
|
|
entry->low = MAXPTR;
|
|
entry->high = MINPTR;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void
|
|
__mf_register (void *ptr, size_t sz, int type, const char *name)
|
|
{
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
__mfu_register (ptr, sz, type, name);
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
}
|
|
|
|
|
|
void
|
|
__mfu_register (void *ptr, size_t sz, int type, const char *name)
|
|
{
|
|
TRACE ("register ptr=%p size=%lu type=%x name='%s'\n",
|
|
ptr, (unsigned long) sz, type, name ? name : "");
|
|
|
|
if (__mf_opts.collect_stats)
|
|
{
|
|
__mf_count_register ++;
|
|
__mf_total_register_size [(type < 0) ? 0 :
|
|
(type > __MF_TYPE_MAX) ? 0 :
|
|
type] += sz;
|
|
}
|
|
|
|
if (UNLIKELY (__mf_opts.sigusr1_report))
|
|
__mf_sigusr1_respond ();
|
|
|
|
switch (__mf_opts.mudflap_mode)
|
|
{
|
|
case mode_nop:
|
|
break;
|
|
|
|
case mode_violate:
|
|
__mf_violation (ptr, sz, (uintptr_t) __builtin_return_address (0), NULL,
|
|
__MF_VIOL_REGISTER);
|
|
break;
|
|
|
|
case mode_populate:
|
|
/* Clear the cache. */
|
|
/* XXX: why the entire cache? */
|
|
/* XXX: race */
|
|
memset (__mf_lookup_cache, 0, sizeof(__mf_lookup_cache));
|
|
/* void slot 0 */
|
|
__mf_lookup_cache[0].low = MAXPTR;
|
|
break;
|
|
|
|
case mode_check:
|
|
{
|
|
__mf_object_t *ovr_objs [1];
|
|
unsigned num_overlapping_objs;
|
|
uintptr_t low = (uintptr_t) ptr;
|
|
uintptr_t high = CLAMPSZ (ptr, sz);
|
|
uintptr_t pc = (uintptr_t) __builtin_return_address (0);
|
|
|
|
/* Treat unknown size indication as 1. */
|
|
if (UNLIKELY (sz == 0)) sz = 1;
|
|
|
|
/* Look for objects only of the same type. This will e.g. permit a registration
|
|
of a STATIC overlapping with a GUESS, and a HEAP with a NOACCESS. At
|
|
__mf_check time however harmful overlaps will be detected. */
|
|
num_overlapping_objs = __mf_find_objects2 (low, high, ovr_objs, 1, type);
|
|
|
|
/* Handle overlaps. */
|
|
if (UNLIKELY (num_overlapping_objs > 0))
|
|
{
|
|
__mf_object_t *ovr_obj = ovr_objs[0];
|
|
|
|
/* Accept certain specific duplication pairs. */
|
|
if (((type == __MF_TYPE_STATIC) || (type == __MF_TYPE_GUESS))
|
|
&& ovr_obj->low == low
|
|
&& ovr_obj->high == high
|
|
&& ovr_obj->type == type)
|
|
{
|
|
/* Duplicate registration for static objects may come
|
|
from distinct compilation units. */
|
|
VERBOSE_TRACE ("harmless duplicate reg %p-%p `%s'\n",
|
|
(void *) low, (void *) high,
|
|
(ovr_obj->name ? ovr_obj->name : ""));
|
|
break;
|
|
}
|
|
|
|
/* Alas, a genuine violation. */
|
|
else
|
|
{
|
|
/* Two or more *real* mappings here. */
|
|
__mf_violation ((void *) ptr, sz,
|
|
(uintptr_t) __builtin_return_address (0), NULL,
|
|
__MF_VIOL_REGISTER);
|
|
}
|
|
}
|
|
else /* No overlapping objects: AOK. */
|
|
__mf_insert_new_object (low, high, type, name, pc);
|
|
|
|
/* We could conceivably call __mf_check() here to prime the cache,
|
|
but then the read_count/write_count field is not reliable. */
|
|
break;
|
|
}
|
|
} /* end switch (__mf_opts.mudflap_mode) */
|
|
}
|
|
|
|
|
|
void
|
|
__mf_unregister (void *ptr, size_t sz, int type)
|
|
{
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
__mfu_unregister (ptr, sz, type);
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
}
|
|
|
|
|
|
void
|
|
__mfu_unregister (void *ptr, size_t sz, int type)
|
|
{
|
|
DECLARE (void, free, void *ptr);
|
|
|
|
if (UNLIKELY (__mf_opts.sigusr1_report))
|
|
__mf_sigusr1_respond ();
|
|
|
|
TRACE ("unregister ptr=%p size=%lu type=%x\n", ptr, (unsigned long) sz, type);
|
|
|
|
switch (__mf_opts.mudflap_mode)
|
|
{
|
|
case mode_nop:
|
|
break;
|
|
|
|
case mode_violate:
|
|
__mf_violation (ptr, sz,
|
|
(uintptr_t) __builtin_return_address (0), NULL,
|
|
__MF_VIOL_UNREGISTER);
|
|
break;
|
|
|
|
case mode_populate:
|
|
/* Clear the cache. */
|
|
/* XXX: race */
|
|
memset (__mf_lookup_cache, 0, sizeof(__mf_lookup_cache));
|
|
/* void slot 0 */
|
|
__mf_lookup_cache[0].low = MAXPTR;
|
|
break;
|
|
|
|
case mode_check:
|
|
{
|
|
__mf_object_t *old_obj = NULL;
|
|
__mf_object_t *del_obj = NULL; /* Object to actually delete. */
|
|
__mf_object_t *objs[1] = {NULL};
|
|
unsigned num_overlapping_objs;
|
|
|
|
num_overlapping_objs = __mf_find_objects2 ((uintptr_t) ptr,
|
|
CLAMPSZ (ptr, sz), objs, 1, type);
|
|
|
|
/* Special case for HEAP_I - see free & realloc hook. They don't
|
|
know whether the input region was HEAP or HEAP_I before
|
|
unmapping it. Here we give HEAP a try in case HEAP_I
|
|
failed. */
|
|
if ((type == __MF_TYPE_HEAP_I) && (num_overlapping_objs == 0))
|
|
{
|
|
num_overlapping_objs = __mf_find_objects2 ((uintptr_t) ptr,
|
|
CLAMPSZ (ptr, sz), objs, 1, __MF_TYPE_HEAP);
|
|
}
|
|
|
|
old_obj = objs[0];
|
|
if (UNLIKELY ((num_overlapping_objs != 1) /* more than one overlap */
|
|
|| ((sz == 0) ? 0 : (sz != (old_obj->high - old_obj->low + 1))) /* size mismatch */
|
|
|| ((uintptr_t) ptr != old_obj->low))) /* base mismatch */
|
|
{
|
|
__mf_violation (ptr, sz,
|
|
(uintptr_t) __builtin_return_address (0), NULL,
|
|
__MF_VIOL_UNREGISTER);
|
|
break;
|
|
}
|
|
|
|
__mf_unlink_object (old_obj);
|
|
__mf_uncache_object (old_obj);
|
|
|
|
/* Wipe buffer contents if desired. */
|
|
if ((__mf_opts.wipe_stack && old_obj->type == __MF_TYPE_STACK)
|
|
|| (__mf_opts.wipe_heap && (old_obj->type == __MF_TYPE_HEAP
|
|
|| old_obj->type == __MF_TYPE_HEAP_I)))
|
|
{
|
|
memset ((void *) old_obj->low,
|
|
0,
|
|
(size_t) (old_obj->high - old_obj->low + 1));
|
|
}
|
|
|
|
/* Manage the object cemetary. */
|
|
if (__mf_opts.persistent_count > 0
|
|
&& (unsigned) old_obj->type <= __MF_TYPE_MAX_CEM)
|
|
{
|
|
old_obj->deallocated_p = 1;
|
|
old_obj->dealloc_pc = (uintptr_t) __builtin_return_address (0);
|
|
#if HAVE_GETTIMEOFDAY
|
|
if (__mf_opts.timestamps)
|
|
gettimeofday (& old_obj->dealloc_time, NULL);
|
|
#endif
|
|
#ifdef LIBMUDFLAPTH
|
|
old_obj->dealloc_thread = pthread_self ();
|
|
#endif
|
|
|
|
if (__mf_opts.backtrace > 0 && old_obj->type == __MF_TYPE_HEAP)
|
|
old_obj->dealloc_backtrace_size =
|
|
__mf_backtrace (& old_obj->dealloc_backtrace,
|
|
NULL, 2);
|
|
|
|
/* Encourage this object to be displayed again in current epoch. */
|
|
old_obj->description_epoch --;
|
|
|
|
/* Put this object into the cemetary. This may require this plot to
|
|
be recycled, and the previous resident to be designated del_obj. */
|
|
{
|
|
unsigned row = old_obj->type;
|
|
unsigned plot = __mf_object_dead_head [row];
|
|
|
|
del_obj = __mf_object_cemetary [row][plot];
|
|
__mf_object_cemetary [row][plot] = old_obj;
|
|
plot ++;
|
|
if (plot == __mf_opts.persistent_count) plot = 0;
|
|
__mf_object_dead_head [row] = plot;
|
|
}
|
|
}
|
|
else
|
|
del_obj = old_obj;
|
|
|
|
if (__mf_opts.print_leaks)
|
|
{
|
|
if ((old_obj->read_count + old_obj->write_count) == 0 &&
|
|
(old_obj->type == __MF_TYPE_HEAP
|
|
|| old_obj->type == __MF_TYPE_HEAP_I))
|
|
{
|
|
/* The problem with a warning message here is that we may not
|
|
be privy to accesses to such objects that occur within
|
|
uninstrumented libraries. */
|
|
#if 0
|
|
fprintf (stderr,
|
|
"*******\n"
|
|
"mudflap warning: unaccessed registered object:\n");
|
|
__mf_describe_object (old_obj);
|
|
#endif
|
|
}
|
|
}
|
|
|
|
if (del_obj != NULL) /* May or may not equal old_obj. */
|
|
{
|
|
if (__mf_opts.backtrace > 0)
|
|
{
|
|
CALL_REAL(free, del_obj->alloc_backtrace);
|
|
if (__mf_opts.persistent_count > 0)
|
|
{
|
|
CALL_REAL(free, del_obj->dealloc_backtrace);
|
|
}
|
|
}
|
|
CALL_REAL(free, del_obj);
|
|
}
|
|
|
|
break;
|
|
}
|
|
} /* end switch (__mf_opts.mudflap_mode) */
|
|
|
|
|
|
if (__mf_opts.collect_stats)
|
|
{
|
|
__mf_count_unregister ++;
|
|
__mf_total_unregister_size += sz;
|
|
}
|
|
}
|
|
|
|
|
|
|
|
struct tree_stats
|
|
{
|
|
unsigned obj_count;
|
|
unsigned long total_size;
|
|
unsigned live_obj_count;
|
|
double total_weight;
|
|
double weighted_size;
|
|
unsigned long weighted_address_bits [sizeof (uintptr_t) * 8][2];
|
|
};
|
|
|
|
|
|
|
|
static int
|
|
__mf_adapt_cache_fn (mfsplay_tree_node n, void *param)
|
|
{
|
|
__mf_object_t *obj = (__mf_object_t *) n->value;
|
|
struct tree_stats *s = (struct tree_stats *) param;
|
|
|
|
assert (obj != NULL && s != NULL);
|
|
|
|
/* Exclude never-accessed objects. */
|
|
if (obj->read_count + obj->write_count)
|
|
{
|
|
s->obj_count ++;
|
|
s->total_size += (obj->high - obj->low + 1);
|
|
|
|
if (obj->liveness)
|
|
{
|
|
unsigned i;
|
|
uintptr_t addr;
|
|
|
|
/* VERBOSE_TRACE ("analyze low=%p live=%u name=`%s'\n",
|
|
(void *) obj->low, obj->liveness, obj->name); */
|
|
|
|
s->live_obj_count ++;
|
|
s->total_weight += (double) obj->liveness;
|
|
s->weighted_size +=
|
|
(double) (obj->high - obj->low + 1) *
|
|
(double) obj->liveness;
|
|
|
|
addr = obj->low;
|
|
for (i=0; i<sizeof(uintptr_t) * 8; i++)
|
|
{
|
|
unsigned bit = addr & 1;
|
|
s->weighted_address_bits[i][bit] += obj->liveness;
|
|
addr = addr >> 1;
|
|
}
|
|
|
|
/* Age the liveness value. */
|
|
obj->liveness >>= 1;
|
|
}
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
|
|
static void
|
|
__mf_adapt_cache ()
|
|
{
|
|
struct tree_stats s;
|
|
uintptr_t new_mask = 0;
|
|
unsigned char new_shift;
|
|
float cache_utilization;
|
|
float max_value;
|
|
static float smoothed_new_shift = -1.0;
|
|
unsigned i;
|
|
|
|
memset (&s, 0, sizeof (s));
|
|
|
|
mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_HEAP), __mf_adapt_cache_fn, (void *) & s);
|
|
mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_HEAP_I), __mf_adapt_cache_fn, (void *) & s);
|
|
mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_STACK), __mf_adapt_cache_fn, (void *) & s);
|
|
mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_STATIC), __mf_adapt_cache_fn, (void *) & s);
|
|
mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_GUESS), __mf_adapt_cache_fn, (void *) & s);
|
|
|
|
/* Maybe we're dealing with funny aging/adaptation parameters, or an
|
|
empty tree. Just leave the cache alone in such cases, rather
|
|
than risk dying by division-by-zero. */
|
|
if (! (s.obj_count > 0) && (s.live_obj_count > 0) && (s.total_weight > 0.0))
|
|
return;
|
|
|
|
/* Guess a good value for the shift parameter by finding an address bit that is a
|
|
good discriminant of lively objects. */
|
|
max_value = 0.0;
|
|
for (i=0; i<sizeof (uintptr_t)*8; i++)
|
|
{
|
|
float value = (float) s.weighted_address_bits[i][0] * (float) s.weighted_address_bits[i][1];
|
|
if (max_value < value) max_value = value;
|
|
}
|
|
for (i=0; i<sizeof (uintptr_t)*8; i++)
|
|
{
|
|
float shoulder_factor = 0.7; /* Include slightly less popular bits too. */
|
|
float value = (float) s.weighted_address_bits[i][0] * (float) s.weighted_address_bits[i][1];
|
|
if (value >= max_value * shoulder_factor)
|
|
break;
|
|
}
|
|
if (smoothed_new_shift < 0) smoothed_new_shift = __mf_lc_shift;
|
|
/* Converge toward this slowly to reduce flapping. */
|
|
smoothed_new_shift = 0.9*smoothed_new_shift + 0.1*i;
|
|
new_shift = (unsigned) (smoothed_new_shift + 0.5);
|
|
assert (new_shift < sizeof (uintptr_t)*8);
|
|
|
|
/* Count number of used buckets. */
|
|
cache_utilization = 0.0;
|
|
for (i = 0; i < (1 + __mf_lc_mask); i++)
|
|
if (__mf_lookup_cache[i].low != 0 || __mf_lookup_cache[i].high != 0)
|
|
cache_utilization += 1.0;
|
|
cache_utilization /= (1 + __mf_lc_mask);
|
|
|
|
new_mask |= 0xffff; /* XXX: force a large cache. */
|
|
new_mask &= (LOOKUP_CACHE_SIZE_MAX - 1);
|
|
|
|
VERBOSE_TRACE ("adapt cache obj=%u/%u sizes=%lu/%.0f/%.0f => "
|
|
"util=%u%% m=%p s=%u\n",
|
|
s.obj_count, s.live_obj_count, s.total_size, s.total_weight, s.weighted_size,
|
|
(unsigned)(cache_utilization*100.0), (void *) new_mask, new_shift);
|
|
|
|
/* We should reinitialize cache if its parameters have changed. */
|
|
if (new_mask != __mf_lc_mask ||
|
|
new_shift != __mf_lc_shift)
|
|
{
|
|
__mf_lc_mask = new_mask;
|
|
__mf_lc_shift = new_shift;
|
|
/* XXX: race */
|
|
memset (__mf_lookup_cache, 0, sizeof(__mf_lookup_cache));
|
|
/* void slot 0 */
|
|
__mf_lookup_cache[0].low = MAXPTR;
|
|
}
|
|
}
|
|
|
|
|
|
|
|
/* __mf_find_object[s] */
|
|
|
|
/* Find overlapping live objecs between [low,high]. Return up to
|
|
max_objs of their pointers in objs[]. Return total count of
|
|
overlaps (may exceed max_objs). */
|
|
|
|
unsigned
|
|
__mf_find_objects2 (uintptr_t ptr_low, uintptr_t ptr_high,
|
|
__mf_object_t **objs, unsigned max_objs, int type)
|
|
{
|
|
unsigned count = 0;
|
|
mfsplay_tree t = __mf_object_tree (type);
|
|
mfsplay_tree_key k = (mfsplay_tree_key) ptr_low;
|
|
int direction;
|
|
|
|
mfsplay_tree_node n = mfsplay_tree_lookup (t, k);
|
|
/* An exact match for base address implies a hit. */
|
|
if (n != NULL)
|
|
{
|
|
if (count < max_objs)
|
|
objs[count] = (__mf_object_t *) n->value;
|
|
count ++;
|
|
}
|
|
|
|
/* Iterate left then right near this key value to find all overlapping objects. */
|
|
for (direction = 0; direction < 2; direction ++)
|
|
{
|
|
/* Reset search origin. */
|
|
k = (mfsplay_tree_key) ptr_low;
|
|
|
|
while (1)
|
|
{
|
|
__mf_object_t *obj;
|
|
|
|
n = (direction == 0 ? mfsplay_tree_successor (t, k) : mfsplay_tree_predecessor (t, k));
|
|
if (n == NULL) break;
|
|
obj = (__mf_object_t *) n->value;
|
|
|
|
if (! (obj->low <= ptr_high && obj->high >= ptr_low)) /* No overlap? */
|
|
break;
|
|
|
|
if (count < max_objs)
|
|
objs[count] = (__mf_object_t *) n->value;
|
|
count ++;
|
|
|
|
k = (mfsplay_tree_key) obj->low;
|
|
}
|
|
}
|
|
|
|
return count;
|
|
}
|
|
|
|
|
|
unsigned
|
|
__mf_find_objects (uintptr_t ptr_low, uintptr_t ptr_high,
|
|
__mf_object_t **objs, unsigned max_objs)
|
|
{
|
|
int type;
|
|
unsigned count = 0;
|
|
|
|
/* Search each splay tree for overlaps. */
|
|
for (type = __MF_TYPE_NOACCESS; type <= __MF_TYPE_GUESS; type++)
|
|
{
|
|
unsigned c = __mf_find_objects2 (ptr_low, ptr_high, objs, max_objs, type);
|
|
if (c > max_objs)
|
|
{
|
|
max_objs = 0;
|
|
objs = NULL;
|
|
}
|
|
else /* NB: C may equal 0 */
|
|
{
|
|
max_objs -= c;
|
|
objs += c;
|
|
}
|
|
count += c;
|
|
}
|
|
|
|
return count;
|
|
}
|
|
|
|
|
|
|
|
/* __mf_link_object */
|
|
|
|
static void
|
|
__mf_link_object (__mf_object_t *node)
|
|
{
|
|
mfsplay_tree t = __mf_object_tree (node->type);
|
|
mfsplay_tree_insert (t, (mfsplay_tree_key) node->low, (mfsplay_tree_value) node);
|
|
}
|
|
|
|
/* __mf_unlink_object */
|
|
|
|
static void
|
|
__mf_unlink_object (__mf_object_t *node)
|
|
{
|
|
mfsplay_tree t = __mf_object_tree (node->type);
|
|
mfsplay_tree_remove (t, (mfsplay_tree_key) node->low);
|
|
}
|
|
|
|
/* __mf_find_dead_objects */
|
|
|
|
/* Find overlapping dead objecs between [low,high]. Return up to
|
|
max_objs of their pointers in objs[]. Return total count of
|
|
overlaps (may exceed max_objs). */
|
|
|
|
static unsigned
|
|
__mf_find_dead_objects (uintptr_t low, uintptr_t high,
|
|
__mf_object_t **objs, unsigned max_objs)
|
|
{
|
|
if (__mf_opts.persistent_count > 0)
|
|
{
|
|
unsigned count = 0;
|
|
unsigned recollection = 0;
|
|
unsigned row = 0;
|
|
|
|
assert (low <= high);
|
|
assert (max_objs == 0 || objs != NULL);
|
|
|
|
/* Widen the search from the most recent plots in each row, looking
|
|
backward in time. */
|
|
recollection = 0;
|
|
while (recollection < __mf_opts.persistent_count)
|
|
{
|
|
count = 0;
|
|
|
|
for (row = 0; row <= __MF_TYPE_MAX_CEM; row ++)
|
|
{
|
|
unsigned plot;
|
|
unsigned i;
|
|
|
|
plot = __mf_object_dead_head [row];
|
|
for (i = 0; i <= recollection; i ++)
|
|
{
|
|
__mf_object_t *obj;
|
|
|
|
/* Look backward through row: it's a circular buffer. */
|
|
if (plot > 0) plot --;
|
|
else plot = __mf_opts.persistent_count - 1;
|
|
|
|
obj = __mf_object_cemetary [row][plot];
|
|
if (obj && obj->low <= high && obj->high >= low)
|
|
{
|
|
/* Found an overlapping dead object! */
|
|
if (count < max_objs)
|
|
objs [count] = obj;
|
|
count ++;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (count)
|
|
break;
|
|
|
|
/* Look farther back in time. */
|
|
recollection = (recollection * 2) + 1;
|
|
}
|
|
|
|
return count;
|
|
} else {
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
/* __mf_describe_object */
|
|
|
|
static void
|
|
__mf_describe_object (__mf_object_t *obj)
|
|
{
|
|
static unsigned epoch = 0;
|
|
if (obj == NULL)
|
|
{
|
|
epoch ++;
|
|
return;
|
|
}
|
|
|
|
if (__mf_opts.abbreviate && obj->description_epoch == epoch)
|
|
{
|
|
fprintf (stderr,
|
|
"mudflap %sobject %p: name=`%s'\n",
|
|
(obj->deallocated_p ? "dead " : ""),
|
|
(void *) obj, (obj->name ? obj->name : ""));
|
|
return;
|
|
}
|
|
else
|
|
obj->description_epoch = epoch;
|
|
|
|
fprintf (stderr,
|
|
"mudflap %sobject %p: name=`%s'\n"
|
|
"bounds=[%p,%p] size=%lu area=%s check=%ur/%uw liveness=%u%s\n"
|
|
"alloc time=%lu.%06lu pc=%p"
|
|
#ifdef LIBMUDFLAPTH
|
|
" thread=%u"
|
|
#endif
|
|
"\n",
|
|
(obj->deallocated_p ? "dead " : ""),
|
|
(void *) obj, (obj->name ? obj->name : ""),
|
|
(void *) obj->low, (void *) obj->high,
|
|
(unsigned long) (obj->high - obj->low + 1),
|
|
(obj->type == __MF_TYPE_NOACCESS ? "no-access" :
|
|
obj->type == __MF_TYPE_HEAP ? "heap" :
|
|
obj->type == __MF_TYPE_HEAP_I ? "heap-init" :
|
|
obj->type == __MF_TYPE_STACK ? "stack" :
|
|
obj->type == __MF_TYPE_STATIC ? "static" :
|
|
obj->type == __MF_TYPE_GUESS ? "guess" :
|
|
"unknown"),
|
|
obj->read_count, obj->write_count, obj->liveness,
|
|
obj->watching_p ? " watching" : "",
|
|
obj->alloc_time.tv_sec, obj->alloc_time.tv_usec,
|
|
(void *) obj->alloc_pc
|
|
#ifdef LIBMUDFLAPTH
|
|
, (unsigned) obj->alloc_thread
|
|
#endif
|
|
);
|
|
|
|
if (__mf_opts.backtrace > 0)
|
|
{
|
|
unsigned i;
|
|
for (i=0; i<obj->alloc_backtrace_size; i++)
|
|
fprintf (stderr, " %s\n", obj->alloc_backtrace[i]);
|
|
}
|
|
|
|
if (__mf_opts.persistent_count > 0)
|
|
{
|
|
if (obj->deallocated_p)
|
|
{
|
|
fprintf (stderr, "dealloc time=%lu.%06lu pc=%p"
|
|
#ifdef LIBMUDFLAPTH
|
|
" thread=%u"
|
|
#endif
|
|
"\n",
|
|
obj->dealloc_time.tv_sec, obj->dealloc_time.tv_usec,
|
|
(void *) obj->dealloc_pc
|
|
#ifdef LIBMUDFLAPTH
|
|
, (unsigned) obj->dealloc_thread
|
|
#endif
|
|
);
|
|
|
|
|
|
if (__mf_opts.backtrace > 0)
|
|
{
|
|
unsigned i;
|
|
for (i=0; i<obj->dealloc_backtrace_size; i++)
|
|
fprintf (stderr, " %s\n", obj->dealloc_backtrace[i]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
static int
|
|
__mf_report_leaks_fn (mfsplay_tree_node n, void *param)
|
|
{
|
|
__mf_object_t *node = (__mf_object_t *) n->value;
|
|
unsigned *count = (unsigned *) param;
|
|
|
|
if (count != NULL)
|
|
(*count) ++;
|
|
|
|
fprintf (stderr, "Leaked object %u:\n", (*count));
|
|
__mf_describe_object (node);
|
|
|
|
return 0;
|
|
}
|
|
|
|
|
|
static unsigned
|
|
__mf_report_leaks ()
|
|
{
|
|
unsigned count = 0;
|
|
|
|
(void) mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_HEAP),
|
|
__mf_report_leaks_fn, & count);
|
|
(void) mfsplay_tree_foreach (__mf_object_tree (__MF_TYPE_HEAP_I),
|
|
__mf_report_leaks_fn, & count);
|
|
|
|
return count;
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* __mf_report */
|
|
|
|
void
|
|
__mf_report ()
|
|
{
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
__mfu_report ();
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
}
|
|
|
|
void
|
|
__mfu_report ()
|
|
{
|
|
if (__mf_opts.collect_stats)
|
|
{
|
|
fprintf (stderr,
|
|
"*******\n"
|
|
"mudflap stats:\n"
|
|
"calls to __mf_check: %lu\n"
|
|
" __mf_register: %lu [%luB, %luB, %luB, %luB, %luB]\n"
|
|
" __mf_unregister: %lu [%luB]\n"
|
|
" __mf_violation: [%lu, %lu, %lu, %lu, %lu]\n",
|
|
__mf_count_check,
|
|
__mf_count_register,
|
|
__mf_total_register_size[0], __mf_total_register_size[1],
|
|
__mf_total_register_size[2], __mf_total_register_size[3],
|
|
__mf_total_register_size[4], /* XXX */
|
|
__mf_count_unregister, __mf_total_unregister_size,
|
|
__mf_count_violation[0], __mf_count_violation[1],
|
|
__mf_count_violation[2], __mf_count_violation[3],
|
|
__mf_count_violation[4]);
|
|
|
|
fprintf (stderr,
|
|
"calls with reentrancy: %lu\n", __mf_reentrancy);
|
|
#ifdef LIBMUDFLAPTH
|
|
fprintf (stderr,
|
|
" lock contention: %lu\n", __mf_lock_contention);
|
|
#endif
|
|
|
|
/* Lookup cache stats. */
|
|
{
|
|
unsigned i;
|
|
unsigned max_reuse = 0;
|
|
unsigned num_used = 0;
|
|
unsigned num_unused = 0;
|
|
|
|
for (i = 0; i < LOOKUP_CACHE_SIZE; i++)
|
|
{
|
|
if (__mf_lookup_cache_reusecount[i])
|
|
num_used ++;
|
|
else
|
|
num_unused ++;
|
|
if (max_reuse < __mf_lookup_cache_reusecount[i])
|
|
max_reuse = __mf_lookup_cache_reusecount[i];
|
|
}
|
|
fprintf (stderr, "lookup cache slots used: %u unused: %u peak-reuse: %u\n",
|
|
num_used, num_unused, max_reuse);
|
|
}
|
|
|
|
{
|
|
unsigned live_count;
|
|
live_count = __mf_find_objects (MINPTR, MAXPTR, NULL, 0);
|
|
fprintf (stderr, "number of live objects: %u\n", live_count);
|
|
}
|
|
|
|
if (__mf_opts.persistent_count > 0)
|
|
{
|
|
unsigned dead_count = 0;
|
|
unsigned row, plot;
|
|
for (row = 0; row <= __MF_TYPE_MAX_CEM; row ++)
|
|
for (plot = 0 ; plot < __mf_opts.persistent_count; plot ++)
|
|
if (__mf_object_cemetary [row][plot] != 0)
|
|
dead_count ++;
|
|
fprintf (stderr, " zombie objects: %u\n", dead_count);
|
|
}
|
|
}
|
|
if (__mf_opts.print_leaks && (__mf_opts.mudflap_mode == mode_check))
|
|
{
|
|
unsigned l;
|
|
extern void * __mf_wrap_alloca_indirect (size_t c);
|
|
|
|
/* Free up any remaining alloca()'d blocks. */
|
|
__mf_wrap_alloca_indirect (0);
|
|
#ifdef HAVE___LIBC_FREERES
|
|
if (__mf_opts.call_libc_freeres)
|
|
{
|
|
extern void __libc_freeres (void);
|
|
__libc_freeres ();
|
|
}
|
|
#endif
|
|
|
|
__mf_describe_object (NULL); /* Reset description epoch. */
|
|
l = __mf_report_leaks ();
|
|
fprintf (stderr, "number of leaked objects: %u\n", l);
|
|
}
|
|
}
|
|
|
|
/* __mf_backtrace */
|
|
|
|
size_t
|
|
__mf_backtrace (char ***symbols, void *guess_pc, unsigned guess_omit_levels)
|
|
{
|
|
void ** pc_array;
|
|
unsigned pc_array_size = __mf_opts.backtrace + guess_omit_levels;
|
|
unsigned remaining_size;
|
|
unsigned omitted_size = 0;
|
|
unsigned i;
|
|
DECLARE (void, free, void *ptr);
|
|
DECLARE (void *, calloc, size_t c, size_t n);
|
|
DECLARE (void *, malloc, size_t n);
|
|
|
|
pc_array = CALL_REAL (calloc, pc_array_size, sizeof (void *) );
|
|
#ifdef HAVE_BACKTRACE
|
|
pc_array_size = backtrace (pc_array, pc_array_size);
|
|
#else
|
|
#define FETCH(n) do { if (pc_array_size >= n) { \
|
|
pc_array[n] = __builtin_return_address(n); \
|
|
if (pc_array[n] == 0) pc_array_size = n; } } while (0)
|
|
|
|
/* Unroll some calls __builtin_return_address because this function
|
|
only takes a literal integer parameter. */
|
|
FETCH (0);
|
|
#if 0
|
|
/* XXX: __builtin_return_address sometimes crashes (!) on >0 arguments,
|
|
rather than simply returning 0. :-( */
|
|
FETCH (1);
|
|
FETCH (2);
|
|
FETCH (3);
|
|
FETCH (4);
|
|
FETCH (5);
|
|
FETCH (6);
|
|
FETCH (7);
|
|
FETCH (8);
|
|
if (pc_array_size > 8) pc_array_size = 9;
|
|
#else
|
|
if (pc_array_size > 0) pc_array_size = 1;
|
|
#endif
|
|
|
|
#undef FETCH
|
|
#endif
|
|
|
|
/* We want to trim the first few levels of the stack traceback,
|
|
since they contain libmudflap wrappers and junk. If pc_array[]
|
|
ends up containing a non-NULL guess_pc, then trim everything
|
|
before that. Otherwise, omit the first guess_omit_levels
|
|
entries. */
|
|
|
|
if (guess_pc != NULL)
|
|
for (i=0; i<pc_array_size; i++)
|
|
if (pc_array [i] == guess_pc)
|
|
omitted_size = i;
|
|
|
|
if (omitted_size == 0) /* No match? */
|
|
if (pc_array_size > guess_omit_levels)
|
|
omitted_size = guess_omit_levels;
|
|
|
|
remaining_size = pc_array_size - omitted_size;
|
|
|
|
#ifdef HAVE_BACKTRACE_SYMBOLS
|
|
*symbols = backtrace_symbols (pc_array + omitted_size, remaining_size);
|
|
#else
|
|
{
|
|
/* Let's construct a buffer by hand. It will have <remaining_size>
|
|
char*'s at the front, pointing at individual strings immediately
|
|
afterwards. */
|
|
void *buffer;
|
|
char *chars;
|
|
char **pointers;
|
|
enum { perline = 30 };
|
|
buffer = CALL_REAL (malloc, remaining_size * (perline + sizeof(char *)));
|
|
pointers = (char **) buffer;
|
|
chars = (char *)buffer + (remaining_size * sizeof (char *));
|
|
for (i = 0; i < remaining_size; i++)
|
|
{
|
|
pointers[i] = chars;
|
|
sprintf (chars, "[0x%p]", pc_array [omitted_size + i]);
|
|
chars = chars + perline;
|
|
}
|
|
*symbols = pointers;
|
|
}
|
|
#endif
|
|
CALL_REAL (free, pc_array);
|
|
|
|
return remaining_size;
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
/* __mf_violation */
|
|
|
|
void
|
|
__mf_violation (void *ptr, size_t sz, uintptr_t pc,
|
|
const char *location, int type)
|
|
{
|
|
char buf [128];
|
|
static unsigned violation_number;
|
|
DECLARE(void, free, void *ptr);
|
|
|
|
TRACE ("violation pc=%p location=%s type=%d ptr=%p size=%lu\n",
|
|
(void *) pc,
|
|
(location != NULL ? location : ""), type, ptr, (unsigned long) sz);
|
|
|
|
if (__mf_opts.collect_stats)
|
|
__mf_count_violation [(type < 0) ? 0 :
|
|
(type > __MF_VIOL_WATCH) ? 0 :
|
|
type] ++;
|
|
|
|
/* Print out a basic warning message. */
|
|
if (__mf_opts.verbose_violations)
|
|
{
|
|
unsigned dead_p;
|
|
unsigned num_helpful = 0;
|
|
struct timeval now = { 0, 0 };
|
|
#if HAVE_GETTIMEOFDAY
|
|
gettimeofday (& now, NULL);
|
|
#endif
|
|
|
|
violation_number ++;
|
|
fprintf (stderr,
|
|
"*******\n"
|
|
"mudflap violation %u (%s): time=%lu.%06lu "
|
|
"ptr=%p size=%lu\npc=%p%s%s%s\n",
|
|
violation_number,
|
|
((type == __MF_VIOL_READ) ? "check/read" :
|
|
(type == __MF_VIOL_WRITE) ? "check/write" :
|
|
(type == __MF_VIOL_REGISTER) ? "register" :
|
|
(type == __MF_VIOL_UNREGISTER) ? "unregister" :
|
|
(type == __MF_VIOL_WATCH) ? "watch" : "unknown"),
|
|
now.tv_sec, now.tv_usec,
|
|
(void *) ptr, (unsigned long)sz, (void *) pc,
|
|
(location != NULL ? " location=`" : ""),
|
|
(location != NULL ? location : ""),
|
|
(location != NULL ? "'" : ""));
|
|
|
|
if (__mf_opts.backtrace > 0)
|
|
{
|
|
char ** symbols;
|
|
unsigned i, num;
|
|
|
|
num = __mf_backtrace (& symbols, (void *) pc, 2);
|
|
/* Note: backtrace_symbols calls malloc(). But since we're in
|
|
__mf_violation and presumably __mf_check, it'll detect
|
|
recursion, and not put the new string into the database. */
|
|
|
|
for (i=0; i<num; i++)
|
|
fprintf (stderr, " %s\n", symbols[i]);
|
|
|
|
/* Calling free() here would trigger a violation. */
|
|
CALL_REAL(free, symbols);
|
|
}
|
|
|
|
|
|
/* Look for nearby objects. For this, we start with s_low/s_high
|
|
pointing to the given area, looking for overlapping objects.
|
|
If none show up, widen the search area and keep looking. */
|
|
|
|
if (sz == 0) sz = 1;
|
|
|
|
for (dead_p = 0; dead_p <= 1; dead_p ++) /* for dead_p in 0 1 */
|
|
{
|
|
enum {max_objs = 3}; /* magic */
|
|
__mf_object_t *objs[max_objs];
|
|
unsigned num_objs = 0;
|
|
uintptr_t s_low, s_high;
|
|
unsigned tries = 0;
|
|
unsigned i;
|
|
|
|
s_low = (uintptr_t) ptr;
|
|
s_high = CLAMPSZ (ptr, sz);
|
|
|
|
while (tries < 16) /* magic */
|
|
{
|
|
if (dead_p)
|
|
num_objs = __mf_find_dead_objects (s_low, s_high, objs, max_objs);
|
|
else
|
|
num_objs = __mf_find_objects (s_low, s_high, objs, max_objs);
|
|
|
|
if (num_objs) /* good enough */
|
|
break;
|
|
|
|
tries ++;
|
|
|
|
/* XXX: tune this search strategy. It's too dependent on
|
|
sz, which can vary from 1 to very big (when array index
|
|
checking) numbers. */
|
|
s_low = CLAMPSUB (s_low, (sz * tries * tries));
|
|
s_high = CLAMPADD (s_high, (sz * tries * tries));
|
|
}
|
|
|
|
for (i = 0; i < min (num_objs, max_objs); i++)
|
|
{
|
|
__mf_object_t *obj = objs[i];
|
|
uintptr_t low = (uintptr_t) ptr;
|
|
uintptr_t high = CLAMPSZ (ptr, sz);
|
|
unsigned before1 = (low < obj->low) ? obj->low - low : 0;
|
|
unsigned after1 = (low > obj->high) ? low - obj->high : 0;
|
|
unsigned into1 = (high >= obj->low && low <= obj->high) ? low - obj->low : 0;
|
|
unsigned before2 = (high < obj->low) ? obj->low - high : 0;
|
|
unsigned after2 = (high > obj->high) ? high - obj->high : 0;
|
|
unsigned into2 = (high >= obj->low && low <= obj->high) ? high - obj->low : 0;
|
|
|
|
fprintf (stderr, "Nearby object %u: checked region begins %uB %s and ends %uB %s\n",
|
|
num_helpful + i + 1,
|
|
(before1 ? before1 : after1 ? after1 : into1),
|
|
(before1 ? "before" : after1 ? "after" : "into"),
|
|
(before2 ? before2 : after2 ? after2 : into2),
|
|
(before2 ? "before" : after2 ? "after" : "into"));
|
|
__mf_describe_object (obj);
|
|
}
|
|
num_helpful += num_objs;
|
|
}
|
|
|
|
fprintf (stderr, "number of nearby objects: %u\n", num_helpful);
|
|
}
|
|
|
|
/* How to finally handle this violation? */
|
|
switch (__mf_opts.violation_mode)
|
|
{
|
|
case viol_nop:
|
|
break;
|
|
case viol_segv:
|
|
kill (getpid(), SIGSEGV);
|
|
break;
|
|
case viol_abort:
|
|
abort ();
|
|
break;
|
|
case viol_gdb:
|
|
|
|
snprintf (buf, 128, "gdb --pid=%u", (unsigned) getpid ());
|
|
system (buf);
|
|
/* XXX: should probably fork() && sleep(GDB_WAIT_PARAMETER)
|
|
instead, and let the forked child execlp() gdb. That way, this
|
|
subject process can be resumed under the supervision of gdb.
|
|
This can't happen now, since system() only returns when gdb
|
|
dies. In that case, we need to beware of starting a second
|
|
concurrent gdb child upon the next violation. (But if the first
|
|
gdb dies, then starting a new one is appropriate.) */
|
|
break;
|
|
}
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------ */
|
|
|
|
|
|
unsigned __mf_watch (void *ptr, size_t sz)
|
|
{
|
|
unsigned rc;
|
|
LOCKTH ();
|
|
BEGIN_RECURSION_PROTECT ();
|
|
rc = __mf_watch_or_not (ptr, sz, 1);
|
|
END_RECURSION_PROTECT ();
|
|
UNLOCKTH ();
|
|
return rc;
|
|
}
|
|
|
|
unsigned __mf_unwatch (void *ptr, size_t sz)
|
|
{
|
|
unsigned rc;
|
|
LOCKTH ();
|
|
rc = __mf_watch_or_not (ptr, sz, 0);
|
|
UNLOCKTH ();
|
|
return rc;
|
|
}
|
|
|
|
|
|
static unsigned
|
|
__mf_watch_or_not (void *ptr, size_t sz, char flag)
|
|
{
|
|
uintptr_t ptr_high = CLAMPSZ (ptr, sz);
|
|
uintptr_t ptr_low = (uintptr_t) ptr;
|
|
unsigned count = 0;
|
|
|
|
TRACE ("%s ptr=%p size=%lu\n",
|
|
(flag ? "watch" : "unwatch"), ptr, (unsigned long) sz);
|
|
|
|
switch (__mf_opts.mudflap_mode)
|
|
{
|
|
case mode_nop:
|
|
case mode_populate:
|
|
case mode_violate:
|
|
count = 0;
|
|
break;
|
|
|
|
case mode_check:
|
|
{
|
|
__mf_object_t **all_ovr_objs;
|
|
unsigned obj_count;
|
|
unsigned n;
|
|
DECLARE (void *, malloc, size_t c);
|
|
DECLARE (void, free, void *p);
|
|
|
|
obj_count = __mf_find_objects (ptr_low, ptr_high, NULL, 0);
|
|
VERBOSE_TRACE (" %u:", obj_count);
|
|
|
|
all_ovr_objs = CALL_REAL (malloc, (sizeof (__mf_object_t *) * obj_count));
|
|
if (all_ovr_objs == NULL) abort ();
|
|
n = __mf_find_objects (ptr_low, ptr_high, all_ovr_objs, obj_count);
|
|
assert (n == obj_count);
|
|
|
|
for (n = 0; n < obj_count; n ++)
|
|
{
|
|
__mf_object_t *obj = all_ovr_objs[n];
|
|
|
|
VERBOSE_TRACE (" [%p]", (void *) obj);
|
|
if (obj->watching_p != flag)
|
|
{
|
|
obj->watching_p = flag;
|
|
count ++;
|
|
|
|
/* Remove object from cache, to ensure next access
|
|
goes through __mf_check(). */
|
|
if (flag)
|
|
__mf_uncache_object (obj);
|
|
}
|
|
}
|
|
CALL_REAL (free, all_ovr_objs);
|
|
}
|
|
break;
|
|
}
|
|
|
|
return count;
|
|
}
|
|
|
|
|
|
void
|
|
__mf_sigusr1_handler (int num)
|
|
{
|
|
__mf_sigusr1_received ++;
|
|
}
|
|
|
|
/* Install or remove SIGUSR1 handler as necessary.
|
|
Also, respond to a received pending SIGUSR1. */
|
|
void
|
|
__mf_sigusr1_respond ()
|
|
{
|
|
static int handler_installed;
|
|
|
|
#ifdef SIGUSR1
|
|
/* Manage handler */
|
|
if (__mf_opts.sigusr1_report && ! handler_installed)
|
|
{
|
|
signal (SIGUSR1, __mf_sigusr1_handler);
|
|
handler_installed = 1;
|
|
}
|
|
else if(! __mf_opts.sigusr1_report && handler_installed)
|
|
{
|
|
signal (SIGUSR1, SIG_DFL);
|
|
handler_installed = 0;
|
|
}
|
|
#endif
|
|
|
|
/* Manage enqueued signals */
|
|
if (__mf_sigusr1_received > __mf_sigusr1_handled)
|
|
{
|
|
__mf_sigusr1_handled ++;
|
|
assert (__mf_get_state () == reentrant);
|
|
__mfu_report ();
|
|
handler_installed = 0; /* We may need to re-enable signal; this might be a SysV library. */
|
|
}
|
|
}
|
|
|
|
|
|
/* XXX: provide an alternative __assert_fail function that cannot
|
|
fail due to libmudflap infinite recursion. */
|
|
#ifndef NDEBUG
|
|
|
|
static void
|
|
write_itoa (int fd, unsigned n)
|
|
{
|
|
enum x { bufsize = sizeof(n)*4 };
|
|
char buf [bufsize];
|
|
unsigned i;
|
|
|
|
for (i=0; i<bufsize-1; i++)
|
|
{
|
|
unsigned digit = n % 10;
|
|
buf[bufsize-2-i] = digit + '0';
|
|
n /= 10;
|
|
if (n == 0)
|
|
{
|
|
char *m = & buf [bufsize-2-i];
|
|
buf[bufsize-1] = '\0';
|
|
write (fd, m, strlen(m));
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void
|
|
__assert_fail (const char *msg, const char *file, unsigned line, const char *func)
|
|
{
|
|
#define write2(string) write (2, (string), strlen ((string)));
|
|
write2("mf");
|
|
#ifdef LIBMUDFLAPTH
|
|
write2("(");
|
|
write_itoa (2, (unsigned) pthread_self ());
|
|
write2(")");
|
|
#endif
|
|
write2(": assertion failure: `");
|
|
write (2, msg, strlen (msg));
|
|
write2("' in ");
|
|
write (2, func, strlen (func));
|
|
write2(" at ");
|
|
write (2, file, strlen (file));
|
|
write2(":");
|
|
write_itoa (2, line);
|
|
write2("\n");
|
|
#undef write2
|
|
abort ();
|
|
}
|
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/* Adapted splay tree code, originally from libiberty. It has been
|
|
specialized for libmudflap as requested by RMS. */
|
|
|
|
static void
|
|
mfsplay_tree_free (void *p)
|
|
{
|
|
DECLARE (void, free, void *p);
|
|
CALL_REAL (free, p);
|
|
}
|
|
|
|
static void *
|
|
mfsplay_tree_xmalloc (size_t s)
|
|
{
|
|
DECLARE (void *, malloc, size_t s);
|
|
return CALL_REAL (malloc, s);
|
|
}
|
|
|
|
|
|
static void mfsplay_tree_splay (mfsplay_tree, mfsplay_tree_key);
|
|
static mfsplay_tree_node mfsplay_tree_splay_helper (mfsplay_tree,
|
|
mfsplay_tree_key,
|
|
mfsplay_tree_node *,
|
|
mfsplay_tree_node *,
|
|
mfsplay_tree_node *);
|
|
|
|
|
|
/* Help splay SP around KEY. PARENT and GRANDPARENT are the parent
|
|
and grandparent, respectively, of NODE. */
|
|
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_splay_helper (mfsplay_tree sp,
|
|
mfsplay_tree_key key,
|
|
mfsplay_tree_node * node,
|
|
mfsplay_tree_node * parent,
|
|
mfsplay_tree_node * grandparent)
|
|
{
|
|
mfsplay_tree_node *next;
|
|
mfsplay_tree_node n;
|
|
int comparison;
|
|
|
|
n = *node;
|
|
|
|
if (!n)
|
|
return *parent;
|
|
|
|
comparison = ((key > n->key) ? 1 : ((key < n->key) ? -1 : 0));
|
|
|
|
if (comparison == 0)
|
|
/* We've found the target. */
|
|
next = 0;
|
|
else if (comparison < 0)
|
|
/* The target is to the left. */
|
|
next = &n->left;
|
|
else
|
|
/* The target is to the right. */
|
|
next = &n->right;
|
|
|
|
if (next)
|
|
{
|
|
/* Check whether our recursion depth is too high. Abort this search,
|
|
and signal that a rebalance is required to continue. */
|
|
if (sp->depth > sp->max_depth)
|
|
{
|
|
sp->rebalance_p = 1;
|
|
return n;
|
|
}
|
|
|
|
/* Continue down the tree. */
|
|
sp->depth ++;
|
|
n = mfsplay_tree_splay_helper (sp, key, next, node, parent);
|
|
sp->depth --;
|
|
|
|
/* The recursive call will change the place to which NODE
|
|
points. */
|
|
if (*node != n || sp->rebalance_p)
|
|
return n;
|
|
}
|
|
|
|
if (!parent)
|
|
/* NODE is the root. We are done. */
|
|
return n;
|
|
|
|
/* First, handle the case where there is no grandparent (i.e.,
|
|
*PARENT is the root of the tree.) */
|
|
if (!grandparent)
|
|
{
|
|
if (n == (*parent)->left)
|
|
{
|
|
*node = n->right;
|
|
n->right = *parent;
|
|
}
|
|
else
|
|
{
|
|
*node = n->left;
|
|
n->left = *parent;
|
|
}
|
|
*parent = n;
|
|
return n;
|
|
}
|
|
|
|
/* Next handle the cases where both N and *PARENT are left children,
|
|
or where both are right children. */
|
|
if (n == (*parent)->left && *parent == (*grandparent)->left)
|
|
{
|
|
mfsplay_tree_node p = *parent;
|
|
|
|
(*grandparent)->left = p->right;
|
|
p->right = *grandparent;
|
|
p->left = n->right;
|
|
n->right = p;
|
|
*grandparent = n;
|
|
return n;
|
|
}
|
|
else if (n == (*parent)->right && *parent == (*grandparent)->right)
|
|
{
|
|
mfsplay_tree_node p = *parent;
|
|
|
|
(*grandparent)->right = p->left;
|
|
p->left = *grandparent;
|
|
p->right = n->left;
|
|
n->left = p;
|
|
*grandparent = n;
|
|
return n;
|
|
}
|
|
|
|
/* Finally, deal with the case where N is a left child, but *PARENT
|
|
is a right child, or vice versa. */
|
|
if (n == (*parent)->left)
|
|
{
|
|
(*parent)->left = n->right;
|
|
n->right = *parent;
|
|
(*grandparent)->right = n->left;
|
|
n->left = *grandparent;
|
|
*grandparent = n;
|
|
return n;
|
|
}
|
|
else
|
|
{
|
|
(*parent)->right = n->left;
|
|
n->left = *parent;
|
|
(*grandparent)->left = n->right;
|
|
n->right = *grandparent;
|
|
*grandparent = n;
|
|
return n;
|
|
}
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
mfsplay_tree_rebalance_helper1 (mfsplay_tree_node n, void *array_ptr)
|
|
{
|
|
mfsplay_tree_node **p = array_ptr;
|
|
*(*p) = n;
|
|
(*p)++;
|
|
return 0;
|
|
}
|
|
|
|
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_rebalance_helper2 (mfsplay_tree_node * array, unsigned low,
|
|
unsigned high)
|
|
{
|
|
unsigned middle = low + (high - low) / 2;
|
|
mfsplay_tree_node n = array[middle];
|
|
|
|
/* Note that since we're producing a balanced binary tree, it is not a problem
|
|
that this function is recursive. */
|
|
if (low + 1 <= middle)
|
|
n->left = mfsplay_tree_rebalance_helper2 (array, low, middle - 1);
|
|
else
|
|
n->left = NULL;
|
|
|
|
if (middle + 1 <= high)
|
|
n->right = mfsplay_tree_rebalance_helper2 (array, middle + 1, high);
|
|
else
|
|
n->right = NULL;
|
|
|
|
return n;
|
|
}
|
|
|
|
|
|
/* Rebalance the entire tree. Do this by copying all the node
|
|
pointers into an array, then cleverly re-linking them. */
|
|
static void
|
|
mfsplay_tree_rebalance (mfsplay_tree sp)
|
|
{
|
|
mfsplay_tree_node *all_nodes, *all_nodes_1;
|
|
|
|
if (sp->num_keys <= 2)
|
|
return;
|
|
|
|
all_nodes = mfsplay_tree_xmalloc (sizeof (mfsplay_tree_node) * sp->num_keys);
|
|
|
|
/* Traverse all nodes to copy their addresses into this array. */
|
|
all_nodes_1 = all_nodes;
|
|
mfsplay_tree_foreach (sp, mfsplay_tree_rebalance_helper1,
|
|
(void *) &all_nodes_1);
|
|
|
|
/* Relink all the nodes. */
|
|
sp->root = mfsplay_tree_rebalance_helper2 (all_nodes, 0, sp->num_keys - 1);
|
|
|
|
mfsplay_tree_free (all_nodes);
|
|
}
|
|
|
|
|
|
/* Splay SP around KEY. */
|
|
static void
|
|
mfsplay_tree_splay (mfsplay_tree sp, mfsplay_tree_key key)
|
|
{
|
|
if (sp->root == 0)
|
|
return;
|
|
|
|
/* If we just splayed the tree with the same key, do nothing. */
|
|
if (sp->last_splayed_key_p &&
|
|
(sp->last_splayed_key == key))
|
|
return;
|
|
|
|
/* Compute a maximum recursion depth for a splay tree with NUM nodes.
|
|
The idea is to limit excessive stack usage if we're facing
|
|
degenerate access patterns. Unfortunately such patterns can occur
|
|
e.g. during static initialization, where many static objects might
|
|
be registered in increasing address sequence, or during a case where
|
|
large tree-like heap data structures are allocated quickly.
|
|
|
|
On x86, this corresponds to roughly 200K of stack usage.
|
|
XXX: For libmudflapth, this could be a function of __mf_opts.thread_stack. */
|
|
sp->max_depth = 2500;
|
|
sp->rebalance_p = sp->depth = 0;
|
|
|
|
mfsplay_tree_splay_helper (sp, key, &sp->root, NULL, NULL);
|
|
if (sp->rebalance_p)
|
|
{
|
|
mfsplay_tree_rebalance (sp);
|
|
|
|
sp->rebalance_p = sp->depth = 0;
|
|
mfsplay_tree_splay_helper (sp, key, &sp->root, NULL, NULL);
|
|
|
|
if (sp->rebalance_p)
|
|
abort ();
|
|
}
|
|
|
|
|
|
/* Cache this splay key. */
|
|
sp->last_splayed_key = key;
|
|
sp->last_splayed_key_p = 1;
|
|
}
|
|
|
|
|
|
|
|
/* Allocate a new splay tree. */
|
|
static mfsplay_tree
|
|
mfsplay_tree_new ()
|
|
{
|
|
mfsplay_tree sp = mfsplay_tree_xmalloc (sizeof (struct mfsplay_tree_s));
|
|
sp->root = NULL;
|
|
sp->last_splayed_key_p = 0;
|
|
sp->num_keys = 0;
|
|
|
|
return sp;
|
|
}
|
|
|
|
|
|
|
|
/* Insert a new node (associating KEY with DATA) into SP. If a
|
|
previous node with the indicated KEY exists, its data is replaced
|
|
with the new value. Returns the new node. */
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_insert (mfsplay_tree sp, mfsplay_tree_key key, mfsplay_tree_value value)
|
|
{
|
|
int comparison = 0;
|
|
|
|
mfsplay_tree_splay (sp, key);
|
|
|
|
if (sp->root)
|
|
comparison = ((sp->root->key > key) ? 1 :
|
|
((sp->root->key < key) ? -1 : 0));
|
|
|
|
if (sp->root && comparison == 0)
|
|
{
|
|
/* If the root of the tree already has the indicated KEY, just
|
|
replace the value with VALUE. */
|
|
sp->root->value = value;
|
|
}
|
|
else
|
|
{
|
|
/* Create a new node, and insert it at the root. */
|
|
mfsplay_tree_node node;
|
|
|
|
node = mfsplay_tree_xmalloc (sizeof (struct mfsplay_tree_node_s));
|
|
node->key = key;
|
|
node->value = value;
|
|
sp->num_keys++;
|
|
if (!sp->root)
|
|
node->left = node->right = 0;
|
|
else if (comparison < 0)
|
|
{
|
|
node->left = sp->root;
|
|
node->right = node->left->right;
|
|
node->left->right = 0;
|
|
}
|
|
else
|
|
{
|
|
node->right = sp->root;
|
|
node->left = node->right->left;
|
|
node->right->left = 0;
|
|
}
|
|
|
|
sp->root = node;
|
|
sp->last_splayed_key_p = 0;
|
|
}
|
|
|
|
return sp->root;
|
|
}
|
|
|
|
/* Remove KEY from SP. It is not an error if it did not exist. */
|
|
|
|
static void
|
|
mfsplay_tree_remove (mfsplay_tree sp, mfsplay_tree_key key)
|
|
{
|
|
mfsplay_tree_splay (sp, key);
|
|
sp->last_splayed_key_p = 0;
|
|
if (sp->root && (sp->root->key == key))
|
|
{
|
|
mfsplay_tree_node left, right;
|
|
left = sp->root->left;
|
|
right = sp->root->right;
|
|
/* Delete the root node itself. */
|
|
mfsplay_tree_free (sp->root);
|
|
sp->num_keys--;
|
|
/* One of the children is now the root. Doesn't matter much
|
|
which, so long as we preserve the properties of the tree. */
|
|
if (left)
|
|
{
|
|
sp->root = left;
|
|
/* If there was a right child as well, hang it off the
|
|
right-most leaf of the left child. */
|
|
if (right)
|
|
{
|
|
while (left->right)
|
|
left = left->right;
|
|
left->right = right;
|
|
}
|
|
}
|
|
else
|
|
sp->root = right;
|
|
}
|
|
}
|
|
|
|
/* Lookup KEY in SP, returning VALUE if present, and NULL
|
|
otherwise. */
|
|
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_lookup (mfsplay_tree sp, mfsplay_tree_key key)
|
|
{
|
|
mfsplay_tree_splay (sp, key);
|
|
if (sp->root && (sp->root->key == key))
|
|
return sp->root;
|
|
else
|
|
return 0;
|
|
}
|
|
|
|
|
|
/* Return the immediate predecessor KEY, or NULL if there is no
|
|
predecessor. KEY need not be present in the tree. */
|
|
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_predecessor (mfsplay_tree sp, mfsplay_tree_key key)
|
|
{
|
|
int comparison;
|
|
mfsplay_tree_node node;
|
|
/* If the tree is empty, there is certainly no predecessor. */
|
|
if (!sp->root)
|
|
return NULL;
|
|
/* Splay the tree around KEY. That will leave either the KEY
|
|
itself, its predecessor, or its successor at the root. */
|
|
mfsplay_tree_splay (sp, key);
|
|
comparison = ((sp->root->key > key) ? 1 :
|
|
((sp->root->key < key) ? -1 : 0));
|
|
|
|
/* If the predecessor is at the root, just return it. */
|
|
if (comparison < 0)
|
|
return sp->root;
|
|
/* Otherwise, find the rightmost element of the left subtree. */
|
|
node = sp->root->left;
|
|
if (node)
|
|
while (node->right)
|
|
node = node->right;
|
|
return node;
|
|
}
|
|
|
|
/* Return the immediate successor KEY, or NULL if there is no
|
|
successor. KEY need not be present in the tree. */
|
|
|
|
static mfsplay_tree_node
|
|
mfsplay_tree_successor (mfsplay_tree sp, mfsplay_tree_key key)
|
|
{
|
|
int comparison;
|
|
mfsplay_tree_node node;
|
|
/* If the tree is empty, there is certainly no successor. */
|
|
if (!sp->root)
|
|
return NULL;
|
|
/* Splay the tree around KEY. That will leave either the KEY
|
|
itself, its predecessor, or its successor at the root. */
|
|
mfsplay_tree_splay (sp, key);
|
|
comparison = ((sp->root->key > key) ? 1 :
|
|
((sp->root->key < key) ? -1 : 0));
|
|
/* If the successor is at the root, just return it. */
|
|
if (comparison > 0)
|
|
return sp->root;
|
|
/* Otherwise, find the leftmost element of the right subtree. */
|
|
node = sp->root->right;
|
|
if (node)
|
|
while (node->left)
|
|
node = node->left;
|
|
return node;
|
|
}
|
|
|
|
/* Call FN, passing it the DATA, for every node in SP, following an
|
|
in-order traversal. If FN every returns a non-zero value, the
|
|
iteration ceases immediately, and the value is returned.
|
|
Otherwise, this function returns 0.
|
|
|
|
This function simulates recursion using dynamically allocated
|
|
arrays, since it may be called from mfsplay_tree_rebalance(), which
|
|
in turn means that the tree is already uncomfortably deep for stack
|
|
space limits. */
|
|
static int
|
|
mfsplay_tree_foreach (mfsplay_tree st, mfsplay_tree_foreach_fn fn, void *data)
|
|
{
|
|
mfsplay_tree_node *stack1;
|
|
char *stack2;
|
|
unsigned sp;
|
|
int val = 0;
|
|
enum s { s_left, s_here, s_right, s_up };
|
|
|
|
if (st->root == NULL) /* => num_keys == 0 */
|
|
return 0;
|
|
|
|
stack1 = mfsplay_tree_xmalloc (sizeof (mfsplay_tree_node) * st->num_keys);
|
|
stack2 = mfsplay_tree_xmalloc (sizeof (char) * st->num_keys);
|
|
|
|
sp = 0;
|
|
stack1 [sp] = st->root;
|
|
stack2 [sp] = s_left;
|
|
|
|
while (1)
|
|
{
|
|
mfsplay_tree_node n;
|
|
enum s s;
|
|
|
|
n = stack1 [sp];
|
|
s = stack2 [sp];
|
|
|
|
/* Handle each of the four possible states separately. */
|
|
|
|
/* 1: We're here to traverse the left subtree (if any). */
|
|
if (s == s_left)
|
|
{
|
|
stack2 [sp] = s_here;
|
|
if (n->left != NULL)
|
|
{
|
|
sp ++;
|
|
stack1 [sp] = n->left;
|
|
stack2 [sp] = s_left;
|
|
}
|
|
}
|
|
|
|
/* 2: We're here to traverse this node. */
|
|
else if (s == s_here)
|
|
{
|
|
stack2 [sp] = s_right;
|
|
val = (*fn) (n, data);
|
|
if (val) break;
|
|
}
|
|
|
|
/* 3: We're here to traverse the right subtree (if any). */
|
|
else if (s == s_right)
|
|
{
|
|
stack2 [sp] = s_up;
|
|
if (n->right != NULL)
|
|
{
|
|
sp ++;
|
|
stack1 [sp] = n->right;
|
|
stack2 [sp] = s_left;
|
|
}
|
|
}
|
|
|
|
/* 4: We're here after both subtrees (if any) have been traversed. */
|
|
else if (s == s_up)
|
|
{
|
|
/* Pop the stack. */
|
|
if (sp == 0) break; /* Popping off the root note: we're finished! */
|
|
sp --;
|
|
}
|
|
|
|
else
|
|
abort ();
|
|
}
|
|
|
|
mfsplay_tree_free (stack1);
|
|
mfsplay_tree_free (stack2);
|
|
return val;
|
|
}
|