2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License as published by
4 * the Free Software Foundation; either version 2 of the License, or
5 * (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
19 * @brief Various functions to aid in debugging
21 * @copyright 2013 The FreeRADIUS server project
22 * @copyright 2013 Arran Cudbard-Bell <a.cudbardb@freeradius.org>
25 #include <freeradius-devel/libradius.h>
28 #if defined(HAVE_MALLOPT) && defined(HAVE_MALLOC_H)
33 * runtime backtrace functions are not POSIX but are included in
34 * glibc, OSX >= 10.5 and various BSDs
37 # include <execinfo.h>
40 #ifdef HAVE_SYS_PRCTL_H
41 # include <sys/prctl.h>
44 #ifdef HAVE_SYS_RESOURCE_H
45 # include <sys/resource.h>
49 # define PTHREAD_MUTEX_LOCK pthread_mutex_lock
50 # define PTHREAD_MUTEX_UNLOCK pthread_mutex_unlock
52 # define PTHREAD_MUTEX_LOCK(_x)
53 # define PTHREAD_MUTEX_UNLOCK(_x)
57 # define MAX_BT_FRAMES 128
58 # define MAX_BT_CBUFF 65536 //!< Should be a power of 2
60 # ifdef HAVE_PTHREAD_H
61 static pthread_mutex_t fr_debug_init = PTHREAD_MUTEX_INITIALIZER;
64 typedef struct fr_bt_info {
65 void *obj; //!< Memory address of the block of allocated memory.
66 void *frames[MAX_BT_FRAMES]; //!< Backtrace frame data
67 int count; //!< Number of frames stored
71 void *obj; //!< Pointer to the parent object, this is our needle
72 //!< when we iterate over the contents of the circular buffer.
73 fr_cbuff_t *cbuff; //!< Where we temporarily store the backtraces
77 static char panic_action[512]; //!< The command to execute when panicking.
78 static fr_fault_cb_t panic_cb = NULL; //!< Callback to execute whilst panicking, before the
80 static fr_fault_log_t fr_fault_log = NULL; //!< Function to use to process logging output.
81 static int fr_fault_log_fd = STDERR_FILENO; //!< Where to write debug output.
83 static int fr_debugger_present = -1; //!< Whether were attached to by a debugger.
85 #ifdef HAVE_SYS_RESOURCE_H
86 static struct rlimit core_limits;
89 #define FR_FAULT_LOG(fmt, ...) fr_fault_log(fmt "\n", ## __VA_ARGS__)
91 /** Stub callback to see if the SIGTRAP handler is overriden
93 * @param signum signal raised.
95 static void _sigtrap_handler(UNUSED int signum)
97 fr_debugger_present = 0;
98 signal(SIGTRAP, SIG_DFL);
101 /** Break in debugger (if were running under a debugger)
103 * If the server is running under a debugger this will raise a
104 * SIGTRAP which will pause the running process.
106 * If the server is not running under debugger then this will do nothing.
108 void fr_debug_break(void)
110 if (fr_debugger_present == -1) {
111 fr_debugger_present = 0;
112 signal(SIGTRAP, _sigtrap_handler);
114 } else if (fr_debugger_present == 1) {
120 /** Generate a backtrace for an object during destruction
122 * If this is the first entry being inserted
124 static int _fr_do_bt(fr_bt_marker_t *marker)
128 if (!fr_assert(marker->obj) || !fr_assert(marker->cbuff)) {
132 bt = talloc_zero(marker->cbuff, fr_bt_info_t);
136 bt->count = backtrace(bt->frames, MAX_BT_FRAMES);
137 fr_cbuff_rp_insert(marker->cbuff, bt);
142 /** Print backtrace entry for a given object
144 * @param cbuff to search in.
145 * @param obj pointer to original object
147 void backtrace_print(fr_cbuff_t *cbuff, void *obj)
154 while ((p = fr_cbuff_rp_next(cbuff, NULL))) {
155 if ((p == obj) || !obj) {
157 frames = backtrace_symbols(p->frames, p->count);
159 fprintf(stderr, "Stacktrace for: %p\n", p);
160 for (i = 0; i < p->count; i++) {
161 fprintf(stderr, "%s\n", frames[i]);
164 /* We were only asked to look for one */
172 fprintf(stderr, "No backtrace available for %p", obj);
176 /** Inserts a backtrace marker into the provided context
178 * Allows for maximum laziness and will initialise a circular buffer if one has not already been created.
180 * Code augmentation should look something like:
182 // Create a static cbuffer pointer, the first call to backtrace_attach will initialise it
183 static fr_cbuff *my_obj_bt;
185 my_obj_t *alloc_my_obj(TALLOC_CTX *ctx) {
188 this = talloc(ctx, my_obj_t);
190 // Attach backtrace marker to object
191 backtrace_attach(&my_obj_bt, this);
197 * Then, later when a double free occurs:
199 (gdb) call backtrace_print(&my_obj_bt, <pointer to double freed memory>)
202 * which should print a limited backtrace to stderr. Note, this backtrace will not include any argument
203 * values, but should at least show the code path taken.
205 * @param cbuff this should be a pointer to a static *fr_cbuff.
206 * @param obj we want to generate a backtrace for.
208 fr_bt_marker_t *fr_backtrace_attach(fr_cbuff_t **cbuff, TALLOC_CTX *obj)
210 fr_bt_marker_t *marker;
212 if (*cbuff == NULL) {
213 PTHREAD_MUTEX_LOCK(&fr_debug_init);
214 /* Check again now we hold the mutex - eww*/
215 if (*cbuff == NULL) {
218 ctx = fr_autofree_ctx();
219 *cbuff = fr_cbuff_alloc(ctx, MAX_BT_CBUFF, true);
221 PTHREAD_MUTEX_UNLOCK(&fr_debug_init);
224 marker = talloc(obj, fr_bt_marker_t);
229 marker->obj = (void *) obj;
230 marker->cbuff = *cbuff;
232 talloc_set_destructor(marker, _fr_do_bt);
237 void backtrace_print(UNUSED fr_cbuff_t *cbuff, UNUSED void *obj)
239 fprintf(stderr, "Server built without fr_backtrace_* support, requires execinfo.h and possibly -lexecinfo\n");
241 fr_bt_marker_t *fr_backtrace_attach(UNUSED fr_cbuff_t **cbuff, UNUSED TALLOC_CTX *obj)
243 fprintf(stderr, "Server built without fr_backtrace_* support, requires execinfo.h and possibly -lexecinfo\n");
246 #endif /* ifdef HAVE_EXECINFO */
248 static int _panic_on_free(UNUSED char *foo)
251 return -1; /* this should make the free fail */
254 /** Insert memory into the context of another talloc memory chunk which
255 * causes a panic when freed.
257 * @param ctx TALLOC_CTX to monitor for frees.
259 void fr_panic_on_free(TALLOC_CTX *ctx)
263 ptr = talloc(ctx, char);
264 talloc_set_destructor(ptr, _panic_on_free);
267 /** Set the dumpable flag, also controls whether processes can PATTACH
269 * @param dumpable whether we should allow core dumping
271 #if defined(HAVE_SYS_PRCTL_H) && defined(PR_SET_DUMPABLE)
272 static int fr_set_dumpable_flag(bool dumpable)
274 if (prctl(PR_SET_DUMPABLE, dumpable ? 1 : 0) < 0) {
275 fr_strerror_printf("Cannot re-enable core dumps: prctl(PR_SET_DUMPABLE) failed: %s",
283 static int fr_set_dumpable_flag(UNUSED bool dumpable)
285 fr_strerror_printf("Changing value of PR_DUMPABLE not supported on this system");
290 /** Get the processes dumpable flag
293 #if defined(HAVE_SYS_PRCTL_H) && defined(PR_GET_DUMPABLE)
294 static int fr_get_dumpable_flag(void)
298 ret = prctl(PR_GET_DUMPABLE);
300 fr_strerror_printf("Cannot get dumpable flag: %s", fr_syserror(errno));
305 * Linux is crazy and prctl sometimes returns 2 for disabled
307 if (ret != 1) return 0;
311 static int fr_get_dumpable_flag(void)
313 fr_strerror_printf("Getting value of PR_DUMPABLE not supported on this system");
319 /** Get the current maximum for core files
321 * Do this before anything else so as to ensure it's properly initialized.
323 int fr_set_dumpable_init(void)
325 #ifdef HAVE_SYS_RESOURCE_H
326 if (getrlimit(RLIMIT_CORE, &core_limits) < 0) {
327 fr_strerror_printf("Failed to get current core limit: %s", fr_syserror(errno));
334 /** Enable or disable core dumps
336 * @param allow_core_dumps whether to enable or disable core dumps.
338 int fr_set_dumpable(bool allow_core_dumps)
341 * If configured, turn core dumps off.
343 if (!allow_core_dumps) {
344 #ifdef HAVE_SYS_RESOURCE_H
345 struct rlimit no_core;
347 no_core.rlim_cur = 0;
348 no_core.rlim_max = 0;
350 if (setrlimit(RLIMIT_CORE, &no_core) < 0) {
351 fr_strerror_printf("Failed disabling core dumps: %s", fr_syserror(errno));
359 if (fr_set_dumpable_flag(true) < 0) return -1;
362 * Reset the core dump limits to their original value.
364 #ifdef HAVE_SYS_RESOURCE_H
365 if (setrlimit(RLIMIT_CORE, &core_limits) < 0) {
366 fr_strerror_printf("Cannot update core dump limit: %s", fr_syserror(errno));
374 /** Check to see if panic_action file is world writeable
376 * @return 0 if file is OK, else -1.
378 static int fr_fault_check_permissions(void)
386 * Try and guess which part of the command is the binary, and check to see if
387 * it's world writeable, to try and save the admin from their own stupidity.
389 * @fixme we should do this properly and take into account single and double
392 if ((q = strchr(panic_action, ' '))) {
394 * need to use a static buffer, because mallocing memory in a signal handler
395 * is a bad idea and can result in deadlock.
397 len = snprintf(filename, sizeof(filename), "%.*s", (int)(q - panic_action), panic_action);
398 if (is_truncated(len, sizeof(filename))) {
399 fr_strerror_printf("Failed writing panic_action to temporary buffer (truncated)");
407 if (stat(p, &statbuf) == 0) {
409 if ((statbuf.st_mode & S_IWOTH) != 0) {
410 fr_strerror_printf("panic_action file \"%s\" is globally writable", p);
419 /** Prints a simple backtrace (if execinfo is available) and calls panic_action if set.
423 void fr_fault(int sig)
425 char cmd[sizeof(panic_action) + 20];
427 size_t left = sizeof(cmd), ret;
429 char const *p = panic_action;
435 * Makes the backtraces slightly cleaner
437 memset(cmd, 0, sizeof(cmd));
439 FR_FAULT_LOG("CAUGHT SIGNAL: %s", strsignal(sig));
442 * Check for administrator sanity.
444 if (fr_fault_check_permissions() < 0) {
445 FR_FAULT_LOG("Refusing to execute panic action: %s", fr_strerror());
450 * Run the callback if one was registered
452 if (panic_cb && (panic_cb(sig) < 0)) goto finish;
455 * Produce a simple backtrace - They've very basic but at least give us an
456 * idea of the area of the code we hit the issue in.
458 * See below in fr_fault_setup() and
459 * https://sourceware.org/bugzilla/show_bug.cgi?id=16159
460 * for why we only print backtraces in debug builds if we're using GLIBC.
462 #if defined(HAVE_EXECINFO) && (!defined(NDEBUG) || !defined(__GNUC__))
464 size_t frame_count, i;
465 void *stack[MAX_BT_FRAMES];
468 frame_count = backtrace(stack, MAX_BT_FRAMES);
470 FR_FAULT_LOG("Backtrace of last %zu frames:", frame_count);
473 * Only use backtrace_symbols() if we don't have a logging fd.
474 * If the server has experienced memory corruption, there's
475 * a high probability that calling backtrace_symbols() which
476 * mallocs more memory, will fail.
478 if (fr_fault_log_fd < 0) {
479 strings = backtrace_symbols(stack, frame_count);
480 for (i = 0; i < frame_count; i++) {
481 FR_FAULT_LOG("%s", strings[i]);
485 backtrace_symbols_fd(stack, frame_count, fr_fault_log_fd);
490 /* No panic action set... */
491 if (panic_action[0] == '\0') {
492 FR_FAULT_LOG("No panic action set");
496 /* Substitute %p for the current PID (useful for attaching a debugger) */
497 while ((q = strstr(p, "%p"))) {
498 out += ret = snprintf(out, left, "%.*s%d", (int) (q - p), p, (int) getpid());
501 FR_FAULT_LOG("Panic action too long");
507 if (strlen(p) >= left) goto oob;
508 strlcpy(out, p, left);
510 FR_FAULT_LOG("Calling: %s", cmd);
513 bool disable = false;
516 * Here we temporarily enable the dumpable flag so if GBD or LLDB
517 * is called in the panic_action, they can pattach tot he running
520 if (fr_get_dumpable_flag() == 0) {
521 if ((fr_set_dumpable_flag(true) < 0) || !fr_get_dumpable_flag()) {
522 FR_FAULT_LOG("Failed setting dumpable flag, pattach may not work: %s", fr_strerror());
526 FR_FAULT_LOG("Temporarily setting PR_DUMPABLE to 1");
532 * We only want to error out here, if dumpable was originally disabled
533 * and we managed to change the value to enabled, but failed
534 * setting it back to disabled.
537 FR_FAULT_LOG("Resetting PR_DUMPABLE to 0");
538 if (fr_set_dumpable_flag(false) < 0) {
539 FR_FAULT_LOG("Failed reseting dumpable flag to off: %s", fr_strerror());
540 FR_FAULT_LOG("Exiting due to insecure process state");
546 FR_FAULT_LOG("Panic action exited with %i", code);
550 if (sig == SIGUSR1) {
558 /** Work around debuggers which can't backtrace past the signal handler
560 * At least this provides us some information when we get talloc errors.
562 static void _fr_talloc_fault(char const *reason)
564 fr_fault_log("talloc abort: %s\n", reason);
569 /** Wrapper to pass talloc log output to our fr_fault_log function
572 static void _fr_talloc_log(char const *msg)
574 fr_fault_log("%s\n", msg);
577 /** Generate a talloc memory report for a context and print to stderr/stdout
579 * @param ctx to generate a report for, may be NULL in which case the root context is used.
581 int fr_log_talloc_report(TALLOC_CTX *ctx)
584 char const *null_ctx = NULL;
588 fd = dup(fr_fault_log_fd);
590 fr_strerror_printf("Couldn't write memory report, failed to dup log fd: %s", fr_syserror(errno));
593 log = fdopen(fd, "w");
596 fr_strerror_printf("Couldn't write memory report, fdopen failed: %s", fr_syserror(errno));
600 fprintf(log, "Current state of talloced memory:\n");
602 null_ctx = talloc_get_name(NULL);
606 talloc_report_full(NULL, log);
608 fprintf(log, "Context level %i", i++);
610 talloc_report_full(ctx, log);
611 } while ((ctx = talloc_parent(ctx)) && (talloc_get_name(ctx) != null_ctx)); /* Stop before we hit NULL ctx */
618 /** Signal handler to print out a talloc memory report
622 static void _fr_fault_mem_report(int sig)
624 fr_fault_log("CAUGHT SIGNAL: %s\n", strsignal(sig));
626 if (fr_log_talloc_report(NULL) < 0) fr_perror("memreport");
629 static int _fr_disable_null_tracking(UNUSED bool *p)
631 talloc_disable_null_tracking();
635 /** Registers signal handlers to execute panic_action on fatal signal
637 * May be called multiple time to change the panic_action/program.
639 * @param cmd to execute on fault. If present %p will be substituted
640 * for the parent PID before the command is executed, and %e
641 * will be substituted for the currently running program.
642 * @param program Name of program currently executing (argv[0]).
643 * @return 0 on success -1 on failure.
645 int fr_fault_setup(char const *cmd, char const *program)
647 static bool setup = false;
649 char *out = panic_action;
650 size_t left = sizeof(panic_action), ret;
656 /* Substitute %e for the current program */
657 while ((q = strstr(p, "%e"))) {
658 out += ret = snprintf(out, left, "%.*s%s", (int) (q - p), p, program ? program : "");
661 fr_strerror_printf("Panic action too long");
667 if (strlen(p) >= left) goto oob;
668 strlcpy(out, p, left);
670 *panic_action = '\0';
674 * Check for administrator sanity.
676 if (fr_fault_check_permissions() < 0) return -1;
678 /* Unsure what the side effects of changing the signal handler mid execution might be */
681 if (fr_set_signal(SIGSEGV, fr_fault) < 0) return -1;
684 if (fr_set_signal(SIGBUS, fr_fault) < 0) return -1;
687 if (fr_set_signal(SIGABRT, fr_fault) < 0) return -1;
689 * Use this instead of abort so we get a
690 * full backtrace with broken versions of LLDB
692 talloc_set_abort_fn(_fr_talloc_fault);
695 if (fr_set_signal(SIGFPE, fr_fault) < 0) return -1;
699 if (fr_set_signal(SIGUSR1, fr_fault) < 0) return -1;
703 if (fr_set_signal(SIGUSR2, _fr_fault_mem_report) < 0) return -1;
707 * Setup the default logger
709 if (!fr_fault_log) fr_fault_set_log_fn(NULL);
710 talloc_set_log_fn(_fr_talloc_log);
713 * Needed for memory reports
715 * Disable null tracking on exit, else valgrind complains
718 TALLOC_CTX *autofree;
721 talloc_enable_null_tracking();
723 autofree = talloc_autofree_context();
724 marker = talloc(autofree, bool);
725 talloc_set_destructor(marker, _fr_disable_null_tracking);
729 * If were using glibc malloc > 2.4 this scribbles over
730 * uninitialised and freed memory, to make memory issues easier
733 #if defined(HAVE_MALLOPT) && !defined(NDEBUG)
734 mallopt(M_PERTURB, 0x42);
735 mallopt(M_CHECK_ACTION, 3);
738 #if defined(HAVE_EXECINFO) && defined(__GNUC__) && !defined(NDEBUG)
740 * We need to pre-load lgcc_s, else we can get into a deadlock
741 * in fr_fault, as backtrace() attempts to dlopen it.
743 * Apparently there's a performance impact of loading lgcc_s,
744 * so only do it if this is a debug build.
746 * See: https://sourceware.org/bugzilla/show_bug.cgi?id=16159
751 backtrace(stack, 10);
760 /** Set a callback to be called before fr_fault()
762 * @param func to execute. If callback returns < 0
763 * fr_fault will exit before running panic_action code.
765 void fr_fault_set_cb(fr_fault_cb_t func)
770 /** Default logger, logs output to stderr
773 static void CC_HINT(format (printf, 1, 2)) _fr_fault_log(char const *msg, ...)
778 vfprintf(stderr, msg, ap);
783 /** Set a file descriptor to log panic_action output to.
785 * @param func to call to output log messages.
787 void fr_fault_set_log_fn(fr_fault_log_t func)
789 fr_fault_log = func ? func : _fr_fault_log;
792 /** Set a file descriptor to log memory reports to.
794 * @param fd to write output to.
796 void fr_fault_set_log_fd(int fd)
798 fr_fault_log_fd = fd;
802 #ifdef WITH_VERIFY_PTR
805 * Verify a VALUE_PAIR
807 inline void fr_verify_vp(VALUE_PAIR const *vp)
809 (void) talloc_get_type_abort(vp, VALUE_PAIR);
811 if (vp->data.ptr) switch (vp->da->type) {
817 if (!talloc_get_type(vp->data.ptr, uint8_t)) {
818 fr_perror("Type check failed for attribute \"%s\"", vp->da->name);
819 (void) talloc_get_type_abort(vp->data.ptr, uint8_t);
822 len = talloc_array_length(vp->vp_octets);
823 if (vp->length > len) {
824 fr_perror("VALUE_PAIR length %zu does not equal uint8_t buffer length %zu", vp->length, len);
835 if (!talloc_get_type(vp->data.ptr, char)) {
836 fr_perror("Type check failed for attribute \"%s\"", vp->da->name);
837 (void) talloc_get_type_abort(vp->data.ptr, char);
840 len = (talloc_array_length(vp->vp_strvalue) - 1);
841 if (vp->length > len) {
842 fr_perror("VALUE_PAIR %s length %zu is too small for char buffer length %zu",
843 vp->da->name, vp->length, len);
847 if (vp->vp_strvalue[vp->length] != '\0') {
848 fr_perror("VALUE_PAIR %s buffer not \\0 terminated", vp->da->name);
863 void fr_verify_list(TALLOC_CTX *expected, VALUE_PAIR *vps)
869 for (vp = fr_cursor_init(&cursor, &vps);
871 vp = fr_cursor_next(&cursor)) {
874 parent = talloc_parent(vp);
875 if (expected && (parent != expected)) {
876 fr_perror("Expected VALUE_PAIR (%s) to be parented by %p (%s), "
877 "but parented by %p (%s)",
879 expected, talloc_get_name(expected),
880 parent, parent ? talloc_get_name(parent) : "NULL");
882 fr_log_talloc_report(expected);
883 if (parent) fr_log_talloc_report(parent);