]> git.ozlabs.org Git - ccan/blobdiff - ccan/failtest/failtest.c
failtest: use a linked list for history, not an array.
[ccan] / ccan / failtest / failtest.c
index 77aea9a7de6e185226f07a63b7ecb47dadf3a93b..f59ed03acf4abc70f48c8779fabf6684ab724444 100644 (file)
@@ -1,3 +1,5 @@
+/* Licensed under LGPL - see LICENSE file for details */
+#include <ccan/failtest/failtest.h>
 #include <stdarg.h>
 #include <string.h>
 #include <stdio.h>
 #include <sys/wait.h>
 #include <sys/stat.h>
 #include <sys/time.h>
+#include <signal.h>
 #include <assert.h>
+#include <ccan/time/time.h>
 #include <ccan/read_write_all/read_write_all.h>
 #include <ccan/failtest/failtest_proto.h>
-#include <ccan/failtest/failtest.h>
 #include <ccan/build_assert/build_assert.h>
+#include <ccan/str/str.h>
 
-enum failtest_result (*failtest_hook)(struct failtest_call *, unsigned);
+enum failtest_result (*failtest_hook)(struct tlist_calls *);
 
 static int tracefd = -1;
 
@@ -41,10 +45,9 @@ struct lock_info {
        int type;
 };
 
-bool (*failtest_exit_check)(struct failtest_call *history, unsigned num);
+bool (*failtest_exit_check)(struct tlist_calls *history);
 
-static struct failtest_call *history = NULL;
-static unsigned int history_num = 0;
+static struct tlist_calls history = TLIST_INIT(history);
 static int control_fd = -1;
 static struct timeval start;
 static unsigned int probe_count = 0;
@@ -56,6 +59,8 @@ static pid_t lock_owner;
 static struct lock_info *locks = NULL;
 static unsigned int lock_num = 0;
 
+static pid_t orig_pid;
+
 static const char info_to_arg[] = "mceoxprwf";
 
 /* Dummy call used for failtest_undo wrappers. */
@@ -67,24 +72,28 @@ static struct failtest_call *add_history_(enum failtest_call_type type,
                                          const void *elem,
                                          size_t elem_size)
 {
+       struct failtest_call *call;
+
        /* NULL file is how we suppress failure. */
        if (!file)
                return &unrecorded_call;
 
-       history = realloc(history, (history_num + 1) * sizeof(*history));
-       history[history_num].type = type;
-       history[history_num].file = file;
-       history[history_num].line = line;
-       history[history_num].cleanup = NULL;
-       memcpy(&history[history_num].u, elem, elem_size);
-       return &history[history_num++];
+       call = malloc(sizeof *call);
+       call->type = type;
+       call->file = file;
+       call->line = line;
+       call->cleanup = NULL;
+       memcpy(&call->u, elem, elem_size);
+       tlist_add_tail(&history, call, list);
+       return call;
 }
 
 #define add_history(type, file, line, elem) \
        add_history_((type), (file), (line), (elem), sizeof(*(elem)))
 
+/* We do a fake call inside a sizeof(), to check types. */
 #define set_cleanup(call, clean, type)                 \
-       (call)->cleanup = (void *)((void)sizeof(clean((type *)NULL)), (clean))
+       (call)->cleanup = (void *)((void)sizeof(clean((type *)NULL),1), (clean))
 
 static bool read_write_info(int fd)
 {
@@ -108,15 +117,18 @@ static bool read_write_info(int fd)
 
 static char *failpath_string(void)
 {
-       unsigned int i;
-       char *ret = malloc(history_num + 1);
-
-       for (i = 0; i < history_num; i++) {
-               ret[i] = info_to_arg[history[i].type];
-               if (history[i].fail)
-                       ret[i] = toupper(ret[i]);
+       struct failtest_call *i;
+       char *ret = strdup("");
+       unsigned len = 0;
+
+       /* Inefficient, but who cares? */
+       tlist_for_each(&history, i, list) {
+               ret = realloc(ret, len + 2);
+               ret[len] = info_to_arg[i->type];
+               if (i->fail)
+                       ret[len] = toupper(ret[len]);
+               ret[++len] = '\0';
        }
-       ret[i] = '\0';
        return ret;
 }
 
@@ -142,11 +154,23 @@ static void child_fail(const char *out, size_t outlen, const char *fmt, ...)
        exit(1);
 }
 
+static void trace(const char *fmt, ...)
+{
+       va_list ap;
+
+       if (tracefd == -1)
+               return;
+
+       va_start(ap, fmt);
+       vdprintf(tracefd, fmt, ap);
+       va_end(ap);
+}
+
 static pid_t child;
 
-static void hand_down(int signal)
+static void hand_down(int signum)
 {
-       kill(child, signal);
+       kill(child, signum);
 }
 
 static void release_locks(void)
@@ -215,18 +239,6 @@ static void get_locks(void)
        lock_owner = getpid();
 }
 
-static void trace_str(const char *str)
-{
-       ssize_t ret;
-
-       while ((ret = write(tracefd, str, strlen(str))) <= 0) {
-               str += ret;
-               if (!*str)
-                       return;
-       }
-       err(1, "Writing trace.");
-}
-
 struct saved_file {
        struct saved_file *next;
        int fd;
@@ -246,7 +258,8 @@ static struct saved_file *save_file(struct saved_file *next, int fd)
        s->len = lseek(fd, 0, SEEK_END);
        lseek(fd, 0, SEEK_SET);
        s->contents = malloc(s->len);
-       read(fd, s->contents, s->len);
+       if (read(fd, s->contents, s->len) != s->len)
+               err(1, "Failed to save %zu bytes", (size_t)s->len);
        lseek(fd, s->off, SEEK_SET);
        return s;
 }
@@ -258,23 +271,23 @@ static struct saved_file *save_file(struct saved_file *next, int fd)
 static struct saved_file *save_files(void)
 {
        struct saved_file *files = NULL;
-       int i;
+       struct failtest_call *i;
 
        /* Figure out the set of live fds. */
-       for (i = history_num - 2; i >= 0; i--) {
-               if (history[i].type == FAILTEST_OPEN) {
-                       int fd = history[i].u.open.ret;
+       tlist_for_each_rev(&history, i, list) {
+               if (i->type == FAILTEST_OPEN) {
+                       int fd = i->u.open.ret;
                        /* Only do successful, writable fds. */
                        if (fd < 0)
                                continue;
 
                        /* If it was closed, cleanup == NULL. */
-                       if (!history[i].cleanup)
+                       if (!i->cleanup)
                                continue;
 
-                       if ((history[i].u.open.flags & O_RDWR) == O_RDWR) {
+                       if ((i->u.open.flags & O_RDWR) == O_RDWR) {
                                files = save_file(files, fd);
-                       } else if ((history[i].u.open.flags & O_WRONLY)
+                       } else if ((i->u.open.flags & O_WRONLY)
                                   == O_WRONLY) {
                                /* FIXME: Handle O_WRONLY.  Open with O_RDWR? */
                                abort();
@@ -291,8 +304,11 @@ static void restore_files(struct saved_file *s)
                struct saved_file *next = s->next;
 
                lseek(s->fd, 0, SEEK_SET);
-               write(s->fd, s->contents, s->len);
-               ftruncate(s->fd, s->len);
+               if (write(s->fd, s->contents, s->len) != s->len)
+                       err(1, "Failed to restore %zu bytes", (size_t)s->len);
+               if (ftruncate(s->fd, s->len) != 0)
+                       err(1, "Failed to trim file to length %zu",
+                           (size_t)s->len);
                free(s->contents);
                lseek(s->fd, s->off, SEEK_SET);
                free(s);
@@ -300,40 +316,58 @@ static void restore_files(struct saved_file *s)
        }
 }
 
+static void free_files(struct saved_file *s)
+{
+       while (s) {
+               struct saved_file *next = s->next;
+               free(s->contents);
+               free(s);
+               s = next;
+       }
+}
+
+static void free_call(struct failtest_call *call)
+{
+       /* We don't do this in cleanup: needed even for failed opens. */
+       if (call->type == FAILTEST_OPEN)
+               free((char *)call->u.open.pathname);
+       tlist_del_from(&history, call, list);
+       free(call);
+}
+
 /* Free up memory, so valgrind doesn't report leaks. */
 static void free_everything(void)
 {
-       unsigned int i;
+       struct failtest_call *i;
 
-       /* We don't do this in cleanup: needed even for failed opens. */
-       for (i = 0; i < history_num; i++) {
-               if (history[i].type == FAILTEST_OPEN)
-                       free((char *)history[i].u.open.pathname);
-       }
-       free(history);
+       while ((i = tlist_top(&history, struct failtest_call, list)) != NULL)
+               free_call(i);
 }
 
 static NORETURN void failtest_cleanup(bool forced_cleanup, int status)
 {
-       int i;
+       struct failtest_call *i;
 
        /* For children, we don't care if they "failed" the testing. */
        if (control_fd != -1)
                status = 0;
 
-       if (forced_cleanup)
-               history_num--;
+       if (forced_cleanup) {
+               /* We didn't actually do final operation: remove it. */
+               i = tlist_tail(&history, struct failtest_call, list);
+               free_call(i);
+       }
 
        /* Cleanup everything, in reverse order. */
-       for (i = history_num - 1; i >= 0; i--) {
-               if (!history[i].cleanup)
+       tlist_for_each_rev(&history, i, list) {
+               if (!i->cleanup)
                        continue;
                if (!forced_cleanup) {
-                       printf("Leak at %s:%u\n",
-                              history[i].file, history[i].line);
+                       printf("Leak at %s:%u: --failpath=%s\n",
+                              i->file, i->line, failpath_string());
                        status = 1;
                }
-               history[i].cleanup(&history[i].u);
+               i->cleanup(&i->u);
        }
 
        free_everything();
@@ -351,7 +385,7 @@ static bool should_fail(struct failtest_call *call)
        struct saved_file *files;
 
        /* Are we probing? */
-       if (probe_count && --probe_count == 0)
+       if (probe_count && --probe_count == 0 && control_fd != -1)
                failtest_cleanup(true, 0);
 
        if (call == &unrecorded_call)
@@ -361,53 +395,59 @@ static bool should_fail(struct failtest_call *call)
                /* + means continue after end, like normal. */
                if (*failpath == '+')
                        failpath = NULL;
-               else {
-                       if (tolower(*failpath) != info_to_arg[call->type])
+               else if (*failpath == '\0') {
+                       /* Continue, but don't inject errors. */
+                       return call->fail = false;
+               } else {
+                       if (tolower((unsigned char)*failpath)
+                           != info_to_arg[call->type])
                                errx(1, "Failpath expected '%c' got '%c'\n",
                                     info_to_arg[call->type], *failpath);
-                       call->fail = isupper(*(failpath++));
+                       call->fail = cisupper(*(failpath++));
                        return call->fail;
                }
        }
 
        /* Attach debugger if they asked for it. */
-       if (debugpath && history_num == strlen(debugpath)) {
-               unsigned int i;
+       if (debugpath) {
+               char *path = failpath_string();
 
-               for (i = 0; i < history_num; i++) {
-                       char c = info_to_arg[history[i].type];
-                       if (history[i].fail)
-                               c = toupper(c);
-                       if (c != debugpath[i])
-                               break;
-               }
-               if (i == history_num) {
+               if (streq(path, debugpath)) {
                        char str[80];
 
                        /* Don't timeout. */
                        signal(SIGUSR1, SIG_IGN);
                        sprintf(str, "xterm -e gdb /proc/%d/exe %d &",
                                getpid(), getpid());
-                       system(str);
-                       sleep(5);
+                       if (system(str) == 0)
+                               sleep(5);
+               } else if (!strstarts(path, debugpath)) {
+                       fprintf(stderr, "--debugpath not followed: %s\n", path);
+                       debugpath = NULL;
                }
+               free(path);
        }
 
        if (failtest_hook) {
-               switch (failtest_hook(history, history_num)) {
+               switch (failtest_hook(&history)) {
                case FAIL_OK:
                        break;
+               case FAIL_PROBE:
+                       /* Already down probe path?  Stop now. */
+                       if (!probe_count) {
+                               /* FIXME: We should run *parent* and
+                                * run probe until calls match up again. */
+                               probe_count = 3;
+                               break;
+                       } else {
+                               /* Child should give up now. */
+                               if (control_fd != -1)
+                                       failtest_cleanup(true, 0);
+                               /* Parent, don't fail again. */
+                       }
                case FAIL_DONT_FAIL:
                        call->fail = false;
                        return false;
-               case FAIL_PROBE:
-                       /* Already down probe path?  Stop now. */
-                       if (probe_count)
-                               failtest_cleanup(true, 0);
-                       /* FIXME: We should run *parent* and run probe until
-                        * calls match up again. */
-                       probe_count = 3;
-                       break;
                default:
                        abort();
                }
@@ -428,29 +468,24 @@ static bool should_fail(struct failtest_call *call)
 
        if (child == 0) {
                if (tracefd != -1) {
-                       struct timeval now;
-                       char str[50], *p;
-                       gettimeofday(&now, NULL);
-                       if (now.tv_usec < start.tv_usec) {
-                               now.tv_sec--;
-                               now.tv_usec += 1000000;
-                       }
-                       now.tv_usec -= start.tv_usec;
-                       now.tv_sec -= start.tv_sec;
-                       sprintf(str, "%u (%u.%02u): ", getpid(),
-                               (int)now.tv_sec, (int)now.tv_usec / 10000);
-                       trace_str(str);
-                       p = failpath_string();
-                       trace_str(p);
-                       free(p);
-                       trace_str("(");
-                       p = strchr(history[history_num-1].file, '/');
+                       struct timeval diff;
+                       const char *p;
+                       char *failpath;
+                       struct failtest_call *c;
+
+                       c = tlist_tail(&history, struct failtest_call, list);
+                       diff = time_sub(time_now(), start);
+                       failpath = failpath_string();
+                       trace("%u->%u (%u.%02u): %s (", getppid(), getpid(),
+                             (int)diff.tv_sec, (int)diff.tv_usec / 10000,
+                             failpath);
+                       free(failpath);
+                       p = strrchr(c->file, '/');
                        if (p)
-                               trace_str(p+1);
+                               trace("%s", p+1);
                        else
-                               trace_str(history[history_num-1].file);
-                       sprintf(str, ":%u)\n", history[history_num-1].line);
-                       trace_str(str);
+                               trace("%s", c->file);
+                       trace(":%u)\n", c->line);
                }
                close(control[0]);
                close(output[0]);
@@ -459,6 +494,8 @@ static bool should_fail(struct failtest_call *call)
                if (output[1] != STDOUT_FILENO && output[1] != STDERR_FILENO)
                        close(output[1]);
                control_fd = control[1];
+               /* Valgrind spots the leak if we don't free these. */
+               free_files(files);
                return true;
        }
 
@@ -483,8 +520,13 @@ static bool should_fail(struct failtest_call *call)
                else
                        ret = poll(pfd, 2, failtest_timeout_ms);
 
-               if (ret <= 0)
+               if (ret == 0)
                        hand_down(SIGUSR1);
+               if (ret < 0) {
+                       if (errno == EINTR)
+                               continue;
+                       err(1, "Poll returned %i", ret);
+               }
 
                if (pfd[0].revents & POLLIN) {
                        ssize_t len;
@@ -575,14 +617,14 @@ void *failtest_malloc(size_t size, const char *file, unsigned line)
 
        p = add_history(FAILTEST_MALLOC, file, line, &call);
        if (should_fail(p)) {
-               p->u.calloc.ret = NULL;
+               p->u.malloc.ret = NULL;
                p->error = ENOMEM;
        } else {
-               p->u.calloc.ret = malloc(size);
+               p->u.malloc.ret = malloc(size);
                set_cleanup(p, cleanup_malloc, struct malloc_call);
        }
        errno = p->error;
-       return p->u.calloc.ret;
+       return p->u.malloc.ret;
 }
 
 static void cleanup_realloc(struct realloc_call *call)
@@ -591,28 +633,28 @@ static void cleanup_realloc(struct realloc_call *call)
 }
 
 /* Walk back and find out if we got this ptr from a previous routine. */
-static void fixup_ptr_history(void *ptr, unsigned int last)
+static void fixup_ptr_history(void *ptr)
 {
-       int i;
+       struct failtest_call *i;
 
        /* Start at end of history, work back. */
-       for (i = last - 1; i >= 0; i--) {
-               switch (history[i].type) {
+       tlist_for_each_rev(&history, i, list) {
+               switch (i->type) {
                case FAILTEST_REALLOC:
-                       if (history[i].u.realloc.ret == ptr) {
-                               history[i].cleanup = NULL;
+                       if (i->u.realloc.ret == ptr) {
+                               i->cleanup = NULL;
                                return;
                        }
                        break;
                case FAILTEST_MALLOC:
-                       if (history[i].u.malloc.ret == ptr) {
-                               history[i].cleanup = NULL;
+                       if (i->u.malloc.ret == ptr) {
+                               i->cleanup = NULL;
                                return;
                        }
                        break;
                case FAILTEST_CALLOC:
-                       if (history[i].u.calloc.ret == ptr) {
-                               history[i].cleanup = NULL;
+                       if (i->u.calloc.ret == ptr) {
+                               i->cleanup = NULL;
                                return;
                        }
                        break;
@@ -634,7 +676,9 @@ void *failtest_realloc(void *ptr, size_t size, const char *file, unsigned line)
                p->u.realloc.ret = NULL;
                p->error = ENOMEM;
        } else {
-               fixup_ptr_history(ptr, history_num-1);
+               /* Don't catch this one in the history fixup... */
+               p->u.realloc.ret = NULL;
+               fixup_ptr_history(ptr);
                p->u.realloc.ret = realloc(ptr, size);
                set_cleanup(p, cleanup_realloc, struct realloc_call);
        }
@@ -644,7 +688,7 @@ void *failtest_realloc(void *ptr, size_t size, const char *file, unsigned line)
 
 void failtest_free(void *ptr)
 {
-       fixup_ptr_history(ptr, history_num);
+       fixup_ptr_history(ptr);
        free(ptr);
 }
 
@@ -664,7 +708,7 @@ int failtest_open(const char *pathname,
        va_start(ap, line);
        call.flags = va_arg(ap, int);
        if (call.flags & O_CREAT) {
-               call.mode = va_arg(ap, mode_t);
+               call.mode = va_arg(ap, int);
                va_end(ap);
        }
        p = add_history(FAILTEST_OPEN, file, line, &call);
@@ -673,7 +717,7 @@ int failtest_open(const char *pathname,
                free((char *)call.pathname);
        p->u.open.ret = open(pathname, call.flags, call.mode);
 
-       if (!failpath && p->u.open.ret == -1) {
+       if (p->u.open.ret == -1) {
                p->fail = false;
                p->error = errno;
        } else if (should_fail(p)) {
@@ -882,7 +926,7 @@ add_lock(struct lock_info *locks, int fd, off_t start, off_t end, int type)
 /* We trap this so we can record it: we don't fail it. */
 int failtest_close(int fd, const char *file, unsigned line)
 {
-       int i;
+       struct failtest_call *i;
        struct close_call call;
        struct failtest_call *p;
 
@@ -899,30 +943,30 @@ int failtest_close(int fd, const char *file, unsigned line)
                return close(fd);
 
        /* Trace history to find source of fd. */
-       for (i = history_num-1; i >= 0; i--) {
-               switch (history[i].type) {
+       tlist_for_each_rev(&history, i, list) {
+               switch (i->type) {
                case FAILTEST_PIPE:
                        /* From a pipe? */
-                       if (history[i].u.pipe.fds[0] == fd) {
-                               assert(!history[i].u.pipe.closed[0]);
-                               history[i].u.pipe.closed[0] = true;
-                               if (history[i].u.pipe.closed[1])
-                                       history[i].cleanup = NULL;
+                       if (i->u.pipe.fds[0] == fd) {
+                               assert(!i->u.pipe.closed[0]);
+                               i->u.pipe.closed[0] = true;
+                               if (i->u.pipe.closed[1])
+                                       i->cleanup = NULL;
                                goto out;
                        }
-                       if (history[i].u.pipe.fds[1] == fd) {
-                               assert(!history[i].u.pipe.closed[1]);
-                               history[i].u.pipe.closed[1] = true;
-                               if (history[i].u.pipe.closed[0])
-                                       history[i].cleanup = NULL;
+                       if (i->u.pipe.fds[1] == fd) {
+                               assert(!i->u.pipe.closed[1]);
+                               i->u.pipe.closed[1] = true;
+                               if (i->u.pipe.closed[0])
+                                       i->cleanup = NULL;
                                goto out;
                        }
                        break;
                case FAILTEST_OPEN:
-                       if (history[i].u.open.ret == fd) {
-                               assert((void *)history[i].cleanup
+                       if (i->u.open.ret == fd) {
+                               assert((void *)i->cleanup
                                       == (void *)cleanup_open);
-                               history[i].cleanup = NULL;
+                               i->cleanup = NULL;
                                goto out;
                        }
                        break;
@@ -983,7 +1027,6 @@ int failtest_fcntl(int fd, const char *file, unsigned line, int cmd, ...)
        }
 
        p = add_history(FAILTEST_FCNTL, file, line, &call);
-       get_locks();
 
        if (should_fail(p)) {
                p->u.fcntl.ret = -1;
@@ -992,6 +1035,7 @@ int failtest_fcntl(int fd, const char *file, unsigned line, int cmd, ...)
                else
                        p->error = EDEADLK;
        } else {
+               get_locks();
                p->u.fcntl.ret = fcntl(p->u.fcntl.fd, p->u.fcntl.cmd,
                                       &p->u.fcntl.arg.fl);
                if (p->u.fcntl.ret == -1)
@@ -1011,10 +1055,19 @@ int failtest_fcntl(int fd, const char *file, unsigned line, int cmd, ...)
        return p->u.fcntl.ret;
 }
 
+pid_t failtest_getpid(const char *file, unsigned line)
+{
+       /* You must call failtest_init first! */
+       assert(orig_pid);
+       return orig_pid;
+}
+       
 void failtest_init(int argc, char *argv[])
 {
        unsigned int i;
 
+       orig_pid = getpid();
+               
        for (i = 1; i < argc; i++) {
                if (!strncmp(argv[i], "--failpath=", strlen("--failpath="))) {
                        failpath = argv[i] + strlen("--failpath=");
@@ -1026,13 +1079,18 @@ void failtest_init(int argc, char *argv[])
                        debugpath = argv[i] + strlen("--debugpath=");
                }
        }
-       gettimeofday(&start, NULL);
+       start = time_now();
+}
+
+bool failtest_has_failed(void)
+{
+       return control_fd != -1;
 }
 
 void failtest_exit(int status)
 {
        if (failtest_exit_check) {
-               if (!failtest_exit_check(history, history_num))
+               if (!failtest_exit_check(&history))
                        child_fail(NULL, 0, "failtest_exit_check failed\n");
        }