Fix io_u->buf calculation overflow
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index fa30019bfbeddd2bbbb756d069384468d343d1d2..5b58ab82252258a072d841c09c69a23f034f1cd2 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -46,6 +46,7 @@
 #include "profile.h"
 #include "lib/rand.h"
 #include "memalign.h"
+#include "server.h"
 
 unsigned long page_mask;
 unsigned long page_size;
@@ -54,13 +55,20 @@ unsigned long page_size;
        (char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
 
 int groupid = 0;
-int thread_number = 0;
-int nr_process = 0;
-int nr_thread = 0;
+unsigned int thread_number = 0;
+unsigned int nr_process = 0;
+unsigned int nr_thread = 0;
 int shm_id = 0;
 int temp_stall_ts;
 unsigned long done_secs = 0;
 
+/*
+ * Just expose an empty list, if the OS does not support disk util stats
+ */
+#ifndef FIO_HAVE_DISK_UTIL
+FLIST_HEAD(disk_list);
+#endif
+
 static struct fio_mutex *startup_mutex;
 static struct fio_mutex *writeout_mutex;
 static volatile int fio_abort;
@@ -70,11 +78,56 @@ static pthread_t disk_util_thread;
 static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
 
+unsigned long arch_flags = 0;
+
 struct io_log *agg_io_log[2];
 
-#define TERMINATE_ALL          (-1)
 #define JOB_START_TIMEOUT      (5 * 1000)
 
+static const char *fio_os_strings[os_nr] = {
+       "Invalid",
+       "Linux",
+       "AIX",
+       "FreeBSD",
+       "HP-UX",
+       "OSX",
+       "NetBSD",
+       "Solaris",
+       "Windows"
+};
+
+static const char *fio_arch_strings[arch_nr] = {
+       "Invalid",
+       "x86-64",
+       "x86",
+       "ppc",
+       "ia64",
+       "s390",
+       "alpha",
+       "sparc",
+       "sparc64",
+       "arm",
+       "sh",
+       "hppa",
+       "generic"
+};
+
+const char *fio_get_os_string(int nr)
+{
+       if (nr < os_nr)
+               return fio_os_strings[nr];
+
+       return NULL;
+}
+
+const char *fio_get_arch_string(int nr)
+{
+       if (nr < arch_nr)
+               return fio_arch_strings[nr];
+
+       return NULL;
+}
+
 void td_set_runstate(struct thread_data *td, int runstate)
 {
        if (td->runstate == runstate)
@@ -85,7 +138,7 @@ void td_set_runstate(struct thread_data *td, int runstate)
        td->runstate = runstate;
 }
 
-static void terminate_threads(int group_id)
+void fio_terminate_threads(int group_id)
 {
        struct thread_data *td;
        int i;
@@ -102,7 +155,9 @@ static void terminate_threads(int group_id)
                        /*
                         * if the thread is running, just let it exit
                         */
-                       if (td->runstate < TD_RAMP)
+                       if (!td->pid)
+                               continue;
+                       else if (td->runstate < TD_RAMP)
                                kill(td->pid, SIGTERM);
                        else {
                                struct ioengine_ops *ops = td->io_ops;
@@ -117,10 +172,15 @@ static void terminate_threads(int group_id)
 static void sig_int(int sig)
 {
        if (threads) {
-               log_info("\nfio: terminating on signal %d\n", sig);
-               fflush(stdout);
-               exit_value = 128;
-               terminate_threads(TERMINATE_ALL);
+               if (is_backend)
+                       fio_server_got_signal(sig);
+               else {
+                       log_info("\nfio: terminating on signal %d\n", sig);
+                       fflush(stdout);
+                       exit_value = 128;
+               }
+
+               fio_terminate_threads(TERMINATE_ALL);
        }
 }
 
@@ -133,7 +193,9 @@ static void *disk_thread_main(void *data)
                if (!threads)
                        break;
                update_io_ticks();
-               print_thread_status();
+
+               if (!is_backend)
+                       print_thread_status();
        }
 
        return NULL;
@@ -174,6 +236,13 @@ static void set_sig_handlers(void)
        act.sa_handler = sig_int;
        act.sa_flags = SA_RESTART;
        sigaction(SIGTERM, &act, NULL);
+
+       if (is_backend) {
+               memset(&act, 0, sizeof(act));
+               act.sa_handler = sig_int;
+               act.sa_flags = SA_RESTART;
+               sigaction(SIGPIPE, &act, NULL);
+       }
 }
 
 /*
@@ -201,7 +270,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
        if (mtime_since(&td->start, now) < 2000)
                return 0;
 
-       iops += td->io_blocks[ddir];
+       iops += td->this_io_blocks[ddir];
        bytes += td->this_io_bytes[ddir];
        ratemin += td->o.ratemin[ddir];
        rate_iops += td->o.rate_iops[ddir];
@@ -548,13 +617,14 @@ sync_done:
 
                /*
                 * if we can queue more, do so. but check if there are
-                * completed io_u's first.
+                * completed io_u's first. Note that we can get BUSY even
+                * without IO queued, if the system is resource starved.
                 */
-               full = queue_full(td) || ret == FIO_Q_BUSY;
+               full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
                if (full || !td->o.iodepth_batch_complete) {
                        min_events = min(td->o.iodepth_batch_complete,
                                         td->cur_depth);
-                       if (full && !min_events)
+                       if (full && !min_events && td->o.iodepth_batch_complete != 0)
                                min_events = 1;
 
                        do {
@@ -708,13 +778,15 @@ sync_done:
                        break;
 
                /*
-                * See if we need to complete some commands
+                * See if we need to complete some commands. Note that we
+                * can get BUSY even without IO queued, if the system is
+                * resource starved.
                 */
-               full = queue_full(td) || ret == FIO_Q_BUSY;
+               full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
                if (full || !td->o.iodepth_batch_complete) {
                        min_evts = min(td->o.iodepth_batch_complete,
                                        td->cur_depth);
-                       if (full && !min_evts)
+                       if (full && !min_evts && td->o.iodepth_batch_complete != 0)
                                min_evts = 1;
 
                        if (__should_check_rate(td, 0) ||
@@ -737,7 +809,7 @@ sync_done:
                if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
                        if (check_min_rate(td, &comp_time, bytes_done)) {
                                if (exitall_on_terminate)
-                                       terminate_threads(td->groupid);
+                                       fio_terminate_threads(td->groupid);
                                td_verror(td, EIO, "check_min_rate");
                                break;
                        }
@@ -761,7 +833,7 @@ sync_done:
        }
 
        if (td->trim_entries)
-               printf("trim entries %ld\n", td->trim_entries);
+               log_err("fio: %d trim entries leaked?\n", td->trim_entries);
 
        if (td->o.fill_device && td->error == ENOSPC) {
                td->error = 0;
@@ -864,12 +936,12 @@ static int init_io_u(struct thread_data *td)
                dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i);
 
                if (!(td->io_ops->flags & FIO_NOIO)) {
-                       io_u->buf = p + max_bs * i;
+                       io_u->buf = p;
                        dprint(FD_MEM, "io_u %p, mem %p\n", io_u, io_u->buf);
 
-                       if (td_write(td) && !td->o.refill_buffers)
+                       if (td_write(td))
                                io_u_fill_buffer(td, io_u, max_bs);
-                       else if (td_write(td) && td->o.verify_pattern_bytes) {
+                       if (td_write(td) && td->o.verify_pattern_bytes) {
                                /*
                                 * Fill the buffer with the pattern if we are
                                 * going to be doing writes.
@@ -881,6 +953,7 @@ static int init_io_u(struct thread_data *td)
                io_u->index = i;
                io_u->flags = IO_U_F_FREE;
                flist_add(&io_u->list, &td->io_u_freelist);
+               p += max_bs;
        }
 
        return 0;
@@ -965,8 +1038,10 @@ static int keep_running(struct thread_data *td)
 
 static void reset_io_counters(struct thread_data *td)
 {
-       td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0;
+       td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0;
        td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
+       td->stat_io_blocks[0] = td->stat_io_blocks[1] = 0;
+       td->this_io_blocks[0] = td->this_io_blocks[1] = 0;
        td->zone_bytes = 0;
        td->rate_bytes[0] = td->rate_bytes[1] = 0;
        td->rate_blocks[0] = td->rate_blocks[1] = 0;
@@ -978,11 +1053,6 @@ static void reset_io_counters(struct thread_data *td)
         */
        if (td->o.time_based || td->o.loops)
                td->nr_done_files = 0;
-
-       /*
-        * Set the same seed to get repeatable runs
-        */
-       td_fill_rand_seeds(td);
 }
 
 void reset_all_stats(struct thread_data *td)
@@ -1016,6 +1086,11 @@ static void clear_io_state(struct thread_data *td)
        close_files(td);
        for_each_file(td, f, i)
                fio_file_clear_done(f);
+
+       /*
+        * Set the same seed to get repeatable runs
+        */
+       td_fill_rand_seeds(td);
 }
 
 static int exec_string(const char *string)
@@ -1045,10 +1120,11 @@ static void *thread_main(void *data)
        pthread_condattr_t attr;
        int clear_state;
 
-       if (!td->o.use_thread)
+       if (!td->o.use_thread) {
                setsid();
-
-       td->pid = getpid();
+               td->pid = getpid();
+       } else
+               td->pid = gettid();
 
        dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
 
@@ -1092,6 +1168,22 @@ static void *thread_main(void *data)
                goto err;
        }
 
+       /*
+        * If we have a gettimeofday() thread, make sure we exclude that
+        * thread from this job
+        */
+       if (td->o.gtod_cpu)
+               fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu);
+
+       /*
+        * Set affinity first, in case it has an impact on the memory
+        * allocations.
+        */
+       if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) {
+               td_verror(td, errno, "cpu_set_affinity");
+               goto err;
+       }
+
        /*
         * May alter parameters that init_io_u() will use, so we need to
         * do this first.
@@ -1105,23 +1197,6 @@ static void *thread_main(void *data)
        if (td->o.verify_async && verify_async_init(td))
                goto err;
 
-       if (td->o.cpumask_set && fio_setaffinity(td->pid, td->o.cpumask) == -1) {
-               td_verror(td, errno, "cpu_set_affinity");
-               goto err;
-       }
-
-       /*
-        * If we have a gettimeofday() thread, make sure we exclude that
-        * thread from this job
-        */
-       if (td->o.gtod_cpu) {
-               fio_cpu_clear(&td->o.cpumask, td->o.gtod_cpu);
-               if (fio_setaffinity(td->pid, td->o.cpumask) == -1) {
-                       td_verror(td, errno, "cpu_set_affinity");
-                       goto err;
-               }
-       }
-
        if (td->ioprio_set) {
                if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
                        td_verror(td, errno, "ioprio_set");
@@ -1160,19 +1235,17 @@ static void *thread_main(void *data)
        }
 
        fio_gettime(&td->epoch, NULL);
-       getrusage(RUSAGE_SELF, &td->ts.ru_start);
+       getrusage(RUSAGE_SELF, &td->ru_start);
 
        clear_state = 0;
        while (keep_running(td)) {
                fio_gettime(&td->start, NULL);
-               memcpy(&td->ts.stat_sample_time[0], &td->start,
-                               sizeof(td->start));
-               memcpy(&td->ts.stat_sample_time[1], &td->start,
-                               sizeof(td->start));
+               memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
+               memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
                memcpy(&td->tv_cache, &td->start, sizeof(td->start));
 
                if (td->o.ratemin[0] || td->o.ratemin[1])
-                       memcpy(&td->lastrate, &td->ts.stat_sample_time,
+                       memcpy(&td->lastrate, &td->bw_sample_time,
                                                        sizeof(td->lastrate));
 
                if (clear_state)
@@ -1221,40 +1294,48 @@ static void *thread_main(void *data)
        td->ts.io_bytes[1] = td->io_bytes[1];
 
        fio_mutex_down(writeout_mutex);
-       if (td->ts.bw_log) {
+       if (td->bw_log) {
                if (td->o.bw_log_file) {
-                       finish_log_named(td, td->ts.bw_log,
+                       finish_log_named(td, td->bw_log,
                                                td->o.bw_log_file, "bw");
                } else
-                       finish_log(td, td->ts.bw_log, "bw");
+                       finish_log(td, td->bw_log, "bw");
        }
-       if (td->ts.lat_log) {
+       if (td->lat_log) {
                if (td->o.lat_log_file) {
-                       finish_log_named(td, td->ts.lat_log,
+                       finish_log_named(td, td->lat_log,
                                                td->o.lat_log_file, "lat");
                } else
-                       finish_log(td, td->ts.lat_log, "lat");
+                       finish_log(td, td->lat_log, "lat");
        }
-       if (td->ts.slat_log) {
+       if (td->slat_log) {
                if (td->o.lat_log_file) {
-                       finish_log_named(td, td->ts.slat_log,
+                       finish_log_named(td, td->slat_log,
                                                td->o.lat_log_file, "slat");
                } else
-                       finish_log(td, td->ts.slat_log, "slat");
+                       finish_log(td, td->slat_log, "slat");
        }
-       if (td->ts.clat_log) {
+       if (td->clat_log) {
                if (td->o.lat_log_file) {
-                       finish_log_named(td, td->ts.clat_log,
+                       finish_log_named(td, td->clat_log,
                                                td->o.lat_log_file, "clat");
                } else
-                       finish_log(td, td->ts.clat_log, "clat");
+                       finish_log(td, td->clat_log, "clat");
+       }
+       if (td->iops_log) {
+               if (td->o.iops_log_file) {
+                       finish_log_named(td, td->iops_log,
+                                               td->o.iops_log_file, "iops");
+               } else
+                       finish_log(td, td->iops_log, "iops");
        }
+
        fio_mutex_up(writeout_mutex);
        if (td->o.exec_postrun)
                exec_string(td->o.exec_postrun);
 
        if (exitall_on_terminate)
-               terminate_threads(td->groupid);
+               fio_terminate_threads(td->groupid);
 
 err:
        if (td->error)
@@ -1281,7 +1362,6 @@ err:
        if (td->o.write_iolog_file)
                write_iolog_close(td);
 
-       options_mem_free(td);
        td_set_runstate(td, TD_EXITED);
        return (void *) (unsigned long) td->error;
 }
@@ -1295,6 +1375,7 @@ static int fork_main(int shmid, int offset)
        struct thread_data *td;
        void *data, *ret;
 
+#ifndef __hpux
        data = shmat(shmid, NULL, 0);
        if (data == (void *) -1) {
                int __err = errno;
@@ -1302,6 +1383,12 @@ static int fork_main(int shmid, int offset)
                perror("shmat");
                return __err;
        }
+#else
+       /*
+        * HP-UX inherits shm mappings?
+        */
+       data = threads;
+#endif
 
        td = data + offset * sizeof(struct thread_data);
        ret = thread_main(td);
@@ -1312,10 +1399,12 @@ static int fork_main(int shmid, int offset)
 /*
  * Run over the job map and reap the threads that have exited, if any.
  */
-static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
+static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
+                        unsigned int *m_rate)
 {
        struct thread_data *td;
-       int i, cputhreads, realthreads, pending, status, ret;
+       unsigned int cputhreads, realthreads, pending;
+       int i, status, ret;
 
        /*
         * reap exited threads (TD_EXITED -> TD_REAPED)
@@ -1401,7 +1490,7 @@ reaped:
        }
 
        if (*nr_running == cputhreads && !pending && realthreads)
-               terminate_threads(TERMINATE_ALL);
+               fio_terminate_threads(TERMINATE_ALL);
 }
 
 static void *gtod_thread_main(void *data)
@@ -1455,7 +1544,7 @@ static void run_threads(void)
 {
        struct thread_data *td;
        unsigned long spent;
-       int i, todo, nr_running, m_rate, t_rate, nr_started;
+       unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
 
        if (fio_pin_memory())
                return;
@@ -1463,6 +1552,8 @@ static void run_threads(void)
        if (fio_gtod_offload && fio_start_gtod_thread())
                return;
 
+       set_sig_handlers();
+
        if (!terse_output) {
                log_info("Starting ");
                if (nr_thread)
@@ -1470,7 +1561,7 @@ static void run_threads(void)
                                                nr_thread > 1 ? "s" : "");
                if (nr_process) {
                        if (nr_thread)
-                               printf(" and ");
+                               log_info(" and ");
                        log_info("%d process%s", nr_process,
                                                nr_process > 1 ? "es" : "");
                }
@@ -1478,8 +1569,6 @@ static void run_threads(void)
                fflush(stdout);
        }
 
-       set_sig_handlers();
-
        todo = thread_number;
        nr_running = 0;
        nr_started = 0;
@@ -1488,10 +1577,8 @@ static void run_threads(void)
        for_each_td(td, i) {
                print_status_init(td->thread_number - 1);
 
-               if (!td->o.create_serialize) {
-                       init_disk_util(td);
+               if (!td->o.create_serialize)
                        continue;
-               }
 
                /*
                 * do file setup here so it happens sequentially,
@@ -1519,8 +1606,6 @@ static void run_threads(void)
                                        td_io_close_file(td, f);
                        }
                }
-
-               init_disk_util(td);
        }
 
        set_genesis_time();
@@ -1559,6 +1644,8 @@ static void run_threads(void)
                                break;
                        }
 
+                       init_disk_util(td);
+
                        /*
                         * Set state to created. Thread will transition
                         * to TD_INITIALIZED when it's done setting up.
@@ -1597,7 +1684,7 @@ static void run_threads(void)
                        dprint(FD_MUTEX, "wait on startup_mutex\n");
                        if (fio_mutex_down_timeout(startup_mutex, 10)) {
                                log_err("fio: job startup hung? exiting.\n");
-                               terminate_threads(TERMINATE_ALL);
+                               fio_terminate_threads(TERMINATE_ALL);
                                fio_abort = 1;
                                nr_started--;
                                break;
@@ -1665,50 +1752,40 @@ static void run_threads(void)
 
                reap_threads(&nr_running, &t_rate, &m_rate);
 
-               if (todo)
-                       usleep(100000);
+               if (todo) {
+                       if (is_backend)
+                               fio_server_idle_loop();
+                       else
+                               usleep(100000);
+               }
        }
 
        while (nr_running) {
                reap_threads(&nr_running, &t_rate, &m_rate);
-               usleep(10000);
+
+               if (is_backend)
+                       fio_server_idle_loop();
+               else
+                       usleep(10000);
        }
 
        update_io_ticks();
        fio_unpin_memory();
 }
 
-int main(int argc, char *argv[])
+int exec_run(void)
 {
-       long ps;
-
-       sinit();
-       init_rand(&__fio_rand_state);
-
-       /*
-        * We need locale for number printing, if it isn't set then just
-        * go with the US format.
-        */
-       if (!getenv("LC_NUMERIC"))
-               setlocale(LC_NUMERIC, "en_US");
+       struct thread_data *td;
+       int i;
 
-       ps = sysconf(_SC_PAGESIZE);
-       if (ps < 0) {
-               log_err("Failed to get page size\n");
-               return 1;
+       if (nr_clients)
+               return fio_handle_clients();
+       if (exec_profile) {
+               if (load_profile(exec_profile))
+                       return 1;
+               free(exec_profile);
+               exec_profile = NULL;
        }
-
-       page_size = ps;
-       page_mask = ps - 1;
-
-       fio_keywords_init();
-
-       if (parse_options(argc, argv))
-               return 1;
-
-       if (exec_profile && load_profile(exec_profile))
-               return 1;
-
        if (!thread_number)
                return 0;
 
@@ -1741,6 +1818,9 @@ int main(int argc, char *argv[])
                }
        }
 
+       for_each_td(td, i)
+               fio_options_free(td);
+
        cgroup_kill(cgroup_list);
        sfree(cgroup_list);
        sfree(cgroup_mnt);
@@ -1749,3 +1829,80 @@ int main(int argc, char *argv[])
        fio_mutex_remove(writeout_mutex);
        return exit_value;
 }
+
+void reset_fio_state(void)
+{
+       groupid = 0;
+       thread_number = 0;
+       nr_process = 0;
+       nr_thread = 0;
+       done_secs = 0;
+}
+
+static int endian_check(void)
+{
+       union {
+               uint8_t c[8];
+               uint64_t v;
+       } u;
+       int le = 0, be = 0;
+
+       u.v = 0x12;
+       if (u.c[7] == 0x12)
+               be = 1;
+       else if (u.c[0] == 0x12)
+               le = 1;
+
+#if defined(FIO_LITTLE_ENDIAN)
+       if (be)
+               return 1;
+#elif defined(FIO_BIG_ENDIAN)
+       if (le)
+               return 1;
+#else
+       return 1;
+#endif
+
+       if (!le && !be)
+               return 1;
+
+       return 0;
+}
+
+int main(int argc, char *argv[], char *envp[])
+{
+       long ps;
+
+       if (endian_check()) {
+               log_err("fio: endianness settings appear wrong.\n");
+               log_err("fio: please report this to fio@vger.kernel.org\n");
+               return 1;
+       }
+
+       arch_init(envp);
+
+       sinit();
+
+       /*
+        * We need locale for number printing, if it isn't set then just
+        * go with the US format.
+        */
+       if (!getenv("LC_NUMERIC"))
+               setlocale(LC_NUMERIC, "en_US");
+
+       ps = sysconf(_SC_PAGESIZE);
+       if (ps < 0) {
+               log_err("Failed to get page size\n");
+               return 1;
+       }
+
+       page_size = ps;
+       page_mask = ps - 1;
+
+       fio_keywords_init();
+
+       if (parse_options(argc, argv))
+               return 1;
+
+       return exec_run();
+}