[PATCH] Change O_DIRECT vs buffered setup
[fio.git] / init.c
diff --git a/init.c b/init.c
index b310f6633236ac4720b4ef8f7375bcfddea5d276..5d6b0a904f8348af3ccd649054b24132e29de20d 100644 (file)
--- a/init.c
+++ b/init.c
@@ -1,6 +1,5 @@
 /*
- * This file contains the ini and command liner parser. It will create
- * and initialize the specified jobs.
+ * This file contains job initialization and setup functions.
  */
 #include <stdio.h>
 #include <stdlib.h>
 #include <ctype.h>
 #include <string.h>
 #include <errno.h>
+#include <getopt.h>
+#include <assert.h>
 #include <sys/ipc.h>
 #include <sys/shm.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 
 #include "fio.h"
+#include "parse.h"
+
+#define FIO_RANDSEED           (0xb1899bedUL)
+
+#define td_var_offset(var)     ((size_t) &((struct thread_data *)0)->var)
+
+static int str_rw_cb(void *, const char *);
+static int str_ioengine_cb(void *, const char *);
+static int str_mem_cb(void *, const char *);
+static int str_verify_cb(void *, const char *);
+static int str_lockmem_cb(void *, unsigned long *);
+#ifdef FIO_HAVE_IOPRIO
+static int str_prio_cb(void *, unsigned int *);
+static int str_prioclass_cb(void *, unsigned int *);
+#endif
+static int str_exitall_cb(void);
+static int str_cpumask_cb(void *, unsigned int *);
+
+#define __stringify_1(x)       #x
+#define __stringify(x)         __stringify_1(x)
+
+/*
+ * Map of job/command line options
+ */
+static struct fio_option options[] = {
+       {
+               .name   = "name",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(name),
+               .help   = "Name of this job",
+       },
+       {
+               .name   = "directory",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(directory),
+               .help   = "Directory to store files in",
+       },
+       {
+               .name   = "filename",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(filename),
+               .help   = "Force the use of a specific file",
+       },
+       {
+               .name   = "rw",
+               .type   = FIO_OPT_STR,
+               .cb     = str_rw_cb,
+               .help   = "IO direction",
+               .def    = "read",
+               .posval = { "read", "write", "randwrite", "randread", "rw",
+                               "randrw", },
+       },
+       {
+               .name   = "ioengine",
+               .type   = FIO_OPT_STR,
+               .cb     = str_ioengine_cb,
+               .help   = "IO engine to use",
+               .def    = "sync",
+               .posval = { "sync", "libaio", "posixaio", "mmap", "splice",
+                               "sg", "null", },
+       },
+       {
+               .name   = "iodepth",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(iodepth),
+               .help   = "Amount of IO buffers to keep in flight",
+               .def    = "1",
+       },
+       {
+               .name   = "size",
+               .type   = FIO_OPT_STR_VAL,
+               .off1   = td_var_offset(total_file_size),
+               .help   = "Size of device or file",
+       },
+       {
+               .name   = "bs",
+               .type   = FIO_OPT_STR_VAL_INT,
+               .off1   = td_var_offset(bs[DDIR_READ]),
+               .off2   = td_var_offset(bs[DDIR_WRITE]),
+               .help   = "Block size unit",
+               .def    = "4k",
+       },
+       {
+               .name   = "bsrange",
+               .type   = FIO_OPT_RANGE,
+               .off1   = td_var_offset(min_bs[DDIR_READ]),
+               .off2   = td_var_offset(max_bs[DDIR_READ]),
+               .off3   = td_var_offset(min_bs[DDIR_WRITE]),
+               .off4   = td_var_offset(max_bs[DDIR_WRITE]),
+               .help   = "Set block size range (in more detail than bs)",
+       },
+       {
+               .name   = "bs_unaligned",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(bs_unaligned),
+               .help   = "Don't sector align IO buffer sizes",
+       },
+       {
+               .name   = "offset",
+               .type   = FIO_OPT_STR_VAL,
+               .off1   = td_var_offset(start_offset),
+               .help   = "Start IO from this offset",
+               .def    = "0",
+       },
+       {
+               .name   = "randrepeat",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(rand_repeatable),
+               .help   = "Use repeatable random IO pattern",
+               .def    = "1",
+       },
+       {
+               .name   = "norandommap",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(norandommap),
+               .help   = "Accept potential duplicate random blocks",
+       },
+       {
+               .name   = "nrfiles",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(nr_files),
+               .help   = "Split job workload between this number of files",
+               .def    = "1",
+       },
+       {
+               .name   = "fsync",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(fsync_blocks),
+               .help   = "Issue fsync for writes every given number of blocks",
+               .def    = "0",
+       },
+       {
+               .name   = "direct",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(odirect),
+               .help   = "Use O_DIRECT IO (negates buffered)",
+               .def    = "0",
+       },
+       {
+               .name   = "buffered",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(odirect),
+               .neg    = 1,
+               .help   = "Use buffered IO (negates direct)",
+               .def    = "1",
+       },
+       {
+               .name   = "overwrite",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(overwrite),
+               .help   = "When writing, set whether to overwrite current data",
+               .def    = "0",
+       },
+       {
+               .name   = "loops",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(loops),
+               .help   = "Number of times to run the job",
+               .def    = "1",
+       },
+       {
+               .name   = "numjobs",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(numjobs),
+               .help   = "Duplicate this job this many times",
+               .def    = "1",
+       },
+       {
+               .name   = "startdelay",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(start_delay),
+               .help   = "Only start job when this period has passed",
+               .def    = "0",
+       },
+       {
+               .name   = "runtime",
+               .alias  = "timeout",
+               .type   = FIO_OPT_STR_VAL_TIME,
+               .off1   = td_var_offset(timeout),
+               .help   = "Stop workload when this amount of time has passed",
+               .def    = "0",
+       },
+       {
+               .name   = "mem",
+               .type   = FIO_OPT_STR,
+               .cb     = str_mem_cb,
+               .help   = "Backing type for IO buffers",
+               .def    = "malloc",
+               .posval =  { "malloc", "shm", "shmhuge", "mmap", "mmaphuge", },
+       },
+       {
+               .name   = "verify",
+               .type   = FIO_OPT_STR,
+               .cb     = str_verify_cb,
+               .help   = "Verify sum function",
+               .def    = "0",
+               .posval = { "crc32", "md5", },
+       },
+       {
+               .name   = "write_iolog",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(write_iolog_file),
+               .help   = "Store IO pattern to file",
+       },
+       {
+               .name   = "read_iolog",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(read_iolog_file),
+               .help   = "Playback IO pattern from file",
+       },
+       {
+               .name   = "exec_prerun",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(exec_prerun),
+               .help   = "Execute this file prior to running job",
+       },
+       {
+               .name   = "exec_postrun",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(exec_postrun),
+               .help   = "Execute this file after running job",
+       },
+#ifdef FIO_HAVE_IOSCHED_SWITCH
+       {
+               .name   = "ioscheduler",
+               .type   = FIO_OPT_STR_STORE,
+               .off1   = td_var_offset(ioscheduler),
+               .help   = "Use this IO scheduler on the backing device",
+       },
+#endif
+       {
+               .name   = "zonesize",
+               .type   = FIO_OPT_STR_VAL,
+               .off1   = td_var_offset(zone_size),
+               .help   = "Give size of an IO zone",
+               .def    = "0",
+       },
+       {
+               .name   = "zoneskip",
+               .type   = FIO_OPT_STR_VAL,
+               .off1   = td_var_offset(zone_skip),
+               .help   = "Space between IO zones",
+               .def    = "0",
+       },
+       {
+               .name   = "lockmem",
+               .type   = FIO_OPT_STR_VAL,
+               .cb     = str_lockmem_cb,
+               .help   = "Lock down this amount of memory",
+               .def    = "0",
+       },
+       {
+               .name   = "rwmixcycle",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(rwmixcycle),
+               .help   = "Cycle period for mixed read/write workloads (msec)",
+               .def    = "500",
+       },
+       {
+               .name   = "rwmixread",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(rwmixread),
+               .maxval = 100,
+               .help   = "Percentage of mixed workload that is reads",
+               .def    = "50",
+       },
+       {
+               .name   = "rwmixwrite",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(rwmixwrite),
+               .maxval = 100,
+               .help   = "Percentage of mixed workload that is writes",
+               .def    = "50",
+       },
+       {
+               .name   = "nice",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(nice),
+               .help   = "Set job CPU nice value",
+               .minval = -19,
+               .maxval = 20,
+               .def    = "0",
+       },
+#ifdef FIO_HAVE_IOPRIO
+       {
+               .name   = "prio",
+               .type   = FIO_OPT_INT,
+               .cb     = str_prio_cb,
+               .help   = "Set job IO priority value",
+               .minval = 0,
+               .maxval = 7,
+       },
+       {
+               .name   = "prioclass",
+               .type   = FIO_OPT_INT,
+               .cb     = str_prioclass_cb,
+               .help   = "Set job IO priority class",
+               .minval = 0,
+               .maxval = 3,
+       },
+#endif
+       {
+               .name   = "thinktime",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(thinktime),
+               .help   = "Idle time between IO buffers",
+               .def    = "0",
+       },
+       {
+               .name   = "thinktime_blocks",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(thinktime_blocks),
+               .help   = "IO buffer period between 'thinktime'",
+               .def    = "1",
+       },
+       {
+               .name   = "rate",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(rate),
+               .help   = "Set bandwidth rate",
+       },
+       {
+               .name   = "ratemin",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(ratemin),
+               .help   = "The bottom limit accepted",
+       },
+       {
+               .name   = "ratecycle",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(ratecycle),
+               .help   = "Window average for rate limits (msec)",
+               .def    = "1000",
+       },
+       {
+               .name   = "invalidate",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(invalidate_cache),
+               .help   = "Invalidate buffer/page cache prior to running job",
+               .def    = "1",
+       },
+       {
+               .name   = "sync",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(sync_io),
+               .help   = "Use O_SYNC for buffered writes",
+               .def    = "0",
+       },
+       {
+               .name   = "bwavgtime",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(bw_avg_time),
+               .help   = "Time window over which to calculate bandwidth (msec)",
+               .def    = "500",
+       },
+       {
+               .name   = "create_serialize",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(create_serialize),
+               .help   = "Serialize creating of job files",
+               .def    = "1",
+       },
+       {
+               .name   = "create_fsync",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(create_fsync),
+               .help   = "Fsync file after creation",
+               .def    = "1",
+       },
+       {
+               .name   = "cpuload",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(cpuload),
+               .help   = "Use this percentage of CPU",
+       },
+       {
+               .name   = "cpuchunks",
+               .type   = FIO_OPT_INT,
+               .off1   = td_var_offset(cpucycle),
+               .help   = "Length of the CPU burn cycles",
+       },
+#ifdef FIO_HAVE_CPU_AFFINITY
+       {
+               .name   = "cpumask",
+               .type   = FIO_OPT_INT,
+               .cb     = str_cpumask_cb,
+               .help   = "CPU affinity mask",
+       },
+#endif
+       {
+               .name   = "end_fsync",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(end_fsync),
+               .help   = "Include fsync at the end of job",
+               .def    = "0",
+       },
+       {
+               .name   = "unlink",
+               .type   = FIO_OPT_BOOL,
+               .off1   = td_var_offset(unlink),
+               .help   = "Unlink created files after job has completed",
+               .def    = "1",
+       },
+       {
+               .name   = "exitall",
+               .type   = FIO_OPT_STR_SET,
+               .cb     = str_exitall_cb,
+               .help   = "Terminate all jobs when one exits",
+       },
+       {
+               .name   = "stonewall",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(stonewall),
+               .help   = "Insert a hard barrier between this job and previous",
+       },
+       {
+               .name   = "thread",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(use_thread),
+               .help   = "Use threads instead of forks",
+       },
+       {
+               .name   = "write_bw_log",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(write_bw_log),
+               .help   = "Write log of bandwidth during run",
+       },
+       {
+               .name   = "write_lat_log",
+               .type   = FIO_OPT_STR_SET,
+               .off1   = td_var_offset(write_lat_log),
+               .help   = "Write log of latency during run",
+       },
+       {
+               .name   = "hugepage-size",
+               .type   = FIO_OPT_STR_VAL,
+               .off1   = td_var_offset(hugepage_size),
+               .help   = "When using hugepages, specify size of each page",
+               .def    = __stringify(FIO_HUGE_PAGE),
+       },
+       {
+               .name = NULL,
+       },
+};
+
+#define FIO_JOB_OPTS   (sizeof(options) / sizeof(struct fio_option))
+#define FIO_CMD_OPTS   (16)
+#define FIO_GETOPT_JOB (0x89988998)
 
 /*
- * The default options
+ * Command line options. These will contain the above, plus a few
+ * extra that only pertain to fio itself and not jobs.
  */
-#define DEF_BS                 (4096)
-#define DEF_TIMEOUT            (0)
-#define DEF_RATE_CYCLE         (1000)
-#define DEF_ODIRECT            (1)
-#define DEF_IO_ENGINE          (FIO_SYNCIO)
-#define DEF_IO_ENGINE_NAME     "sync"
-#define DEF_SEQUENTIAL         (1)
-#define DEF_RAND_REPEAT                (1)
-#define DEF_OVERWRITE          (1)
-#define DEF_CREATE             (1)
-#define DEF_INVALIDATE         (1)
-#define DEF_SYNCIO             (0)
-#define DEF_RANDSEED           (0xb1899bedUL)
-#define DEF_BWAVGTIME          (500)
-#define DEF_CREATE_SER         (1)
-#define DEF_CREATE_FSYNC       (1)
-#define DEF_LOOPS              (1)
-#define DEF_VERIFY             (0)
-#define DEF_STONEWALL          (0)
-#define DEF_NUMJOBS            (1)
-#define DEF_USE_THREAD         (0)
-#define DEF_FILE_SIZE          (1024 * 1024 * 1024UL)
-#define DEF_ZONE_SIZE          (0)
-#define DEF_ZONE_SKIP          (0)
-#define DEF_RWMIX_CYCLE                (500)
-#define DEF_RWMIX_READ         (50)
-#define DEF_NICE               (0)
-
-static int def_timeout = DEF_TIMEOUT;
-
-static char fio_version_string[] = "fio 1.5";
+static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
+       {
+               .name           = "output",
+               .has_arg        = required_argument,
+               .val            = 'o',
+       },
+       {
+               .name           = "timeout",
+               .has_arg        = required_argument,
+               .val            = 't',
+       },
+       {
+               .name           = "latency-log",
+               .has_arg        = required_argument,
+               .val            = 'l',
+       },
+       {
+               .name           = "bandwidth-log",
+               .has_arg        = required_argument,
+               .val            = 'b',
+       },
+       {
+               .name           = "minimal",
+               .has_arg        = optional_argument,
+               .val            = 'm',
+       },
+       {
+               .name           = "version",
+               .has_arg        = no_argument,
+               .val            = 'v',
+       },
+       {
+               .name           = "help",
+               .has_arg        = no_argument,
+               .val            = 'h',
+       },
+       {
+               .name           = "cmdhelp",
+               .has_arg        = required_argument,
+               .val            = 'c',
+       },
+       {
+               .name           = NULL,
+       },
+};
+
+static int def_timeout = 0;
+
+static char fio_version_string[] = "fio 1.11";
 
 static char **ini_file;
 static int max_jobs = MAX_JOBS;
@@ -57,15 +525,15 @@ static int max_jobs = MAX_JOBS;
 struct thread_data def_thread;
 struct thread_data *threads = NULL;
 
-int rate_quit = 0;
-int write_lat_log = 0;
-int write_bw_log = 0;
 int exitall_on_terminate = 0;
 int terse_output = 0;
 unsigned long long mlock_size = 0;
 FILE *f_out = NULL;
 FILE *f_err = NULL;
 
+static int write_lat_log = 0;
+static int write_bw_log = 0;
+
 /*
  * Return a free job structure.
  */
@@ -80,61 +548,46 @@ static struct thread_data *get_new_job(int global, struct thread_data *parent)
 
        td = &threads[thread_number++];
        *td = *parent;
-       td->name[0] = '\0';
 
-       td->fd = -1;
        td->thread_number = thread_number;
        return td;
 }
 
 static void put_job(struct thread_data *td)
 {
+       if (td == &def_thread)
+               return;
+
        memset(&threads[td->thread_number - 1], 0, sizeof(*td));
        thread_number--;
 }
 
 /*
- * Adds a job to the list of things todo. Sanitizes the various options
- * to make sure we don't have conflicts, and initializes various
- * members of td.
+ * Lazy way of fixing up options that depend on each other. We could also
+ * define option callback handlers, but this is easier.
  */
-static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
+static void fixup_options(struct thread_data *td)
 {
-       char *ddir_str[] = { "read", "write", "randread", "randwrite",
-                            "rw", NULL, "randrw" };
-       struct stat sb;
-       int numjobs, ddir;
+       if (!td->rwmixread && td->rwmixwrite)
+               td->rwmixread = 100 - td->rwmixwrite;
 
-#ifndef FIO_HAVE_LIBAIO
-       if (td->io_engine == FIO_LIBAIO) {
-               log_err("Linux libaio not available\n");
-               return 1;
+       if (td->write_iolog_file && td->read_iolog_file) {
+               log_err("fio: read iolog overrides write_iolog\n");
+               free(td->write_iolog_file);
+               td->write_iolog_file = NULL;
        }
-#endif
-#ifndef FIO_HAVE_POSIXAIO
-       if (td->io_engine == FIO_POSIXAIO) {
-               log_err("posix aio not available\n");
-               return 1;
-       }
-#endif
-
-       /*
-        * the def_thread is just for options, it's not a real job
-        */
-       if (td == &def_thread)
-               return 0;
 
-       if (td->io_engine & FIO_SYNCIO)
+       if (td->io_ops->flags & FIO_SYNCIO)
                td->iodepth = 1;
        else {
                if (!td->iodepth)
-                       td->iodepth = 1;
+                       td->iodepth = td->nr_files;
        }
 
        /*
-        * only really works for sequential io for now
+        * only really works for sequential io for now, and with 1 file
         */
-       if (td->zone_size && !td->sequential)
+       if (td->zone_size && !td->sequential && td->nr_files == 1)
                td->zone_size = 0;
 
        /*
@@ -143,6 +596,79 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
        if (td_read(td) || td_rw(td))
                td->overwrite = 1;
 
+       if (!td->min_bs[DDIR_READ])
+               td->min_bs[DDIR_READ]= td->bs[DDIR_READ];
+       if (!td->max_bs[DDIR_READ])
+               td->max_bs[DDIR_READ] = td->bs[DDIR_READ];
+       if (!td->min_bs[DDIR_WRITE])
+               td->min_bs[DDIR_WRITE]= td->bs[DDIR_WRITE];
+       if (!td->max_bs[DDIR_WRITE])
+               td->max_bs[DDIR_WRITE] = td->bs[DDIR_WRITE];
+
+       td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]);
+
+       if (td_read(td) && !td_rw(td))
+               td->verify = 0;
+
+       if (td->norandommap && td->verify != VERIFY_NONE) {
+               log_err("fio: norandommap given, verify disabled\n");
+               td->verify = VERIFY_NONE;
+       }
+       if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO))
+               log_err("fio: bs_unaligned may not work with raw io\n");
+
+       /*
+        * O_DIRECT and char doesn't mix, clear that flag if necessary.
+        */
+       if (td->filetype == FIO_TYPE_CHAR && td->odirect)
+               td->odirect = 0;
+}
+
+/*
+ * This function leaks the buffer
+ */
+static char *to_kmg(unsigned int val)
+{
+       char *buf = malloc(32);
+       char post[] = { 0, 'K', 'M', 'G', 'P', 0 };
+       char *p = post;
+
+       do {
+               if (val & 1023)
+                       break;
+
+               val >>= 10;
+               p++;
+       } while (*p);
+
+       snprintf(buf, 31, "%u%c", val, *p);
+       return buf;
+}
+
+/*
+ * Adds a job to the list of things todo. Sanitizes the various options
+ * to make sure we don't have conflicts, and initializes various
+ * members of td.
+ */
+static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
+{
+       const char *ddir_str[] = { "read", "write", "randread", "randwrite",
+                                  "rw", NULL, "randrw" };
+       struct stat sb;
+       int numjobs, ddir, i;
+       struct fio_file *f;
+
+       /*
+        * the def_thread is just for options, it's not a real job
+        */
+       if (td == &def_thread)
+               return 0;
+
+       assert(td->io_ops);
+
+       if (td->odirect)
+               td->io_ops->flags |= FIO_RAWIO;
+
        td->filetype = FIO_TYPE_FILE;
        if (!stat(jobname, &sb)) {
                if (S_ISBLK(sb.st_mode))
@@ -151,30 +677,53 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
                        td->filetype = FIO_TYPE_CHAR;
        }
 
-       if (td->filetype == FIO_TYPE_FILE) {
+       fixup_options(td);
+
+       if (td->filename)
+               td->nr_uniq_files = 1;
+       else
+               td->nr_uniq_files = td->nr_files;
+
+       if (td->filetype == FIO_TYPE_FILE || td->filename) {
                char tmp[PATH_MAX];
+               int len = 0;
 
                if (td->directory && td->directory[0] != '\0')
-                       sprintf(tmp, "%s/%s.%d", td->directory, jobname, td->thread_number);
-               else
-                       sprintf(tmp, "%s.%d", jobname, td->thread_number);
-               td->file_name = strdup(tmp);
-       } else
-               td->file_name = strdup(jobname);
+                       len = sprintf(tmp, "%s/", td->directory);
+
+               td->files = malloc(sizeof(struct fio_file) * td->nr_files);
+
+               for_each_file(td, f, i) {
+                       memset(f, 0, sizeof(*f));
+                       f->fd = -1;
+
+                       if (td->filename)
+                               sprintf(tmp + len, "%s", td->filename);
+                       else
+                               sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i);
+                       f->file_name = strdup(tmp);
+               }
+       } else {
+               td->nr_files = 1;
+               td->files = malloc(sizeof(struct fio_file));
+               f = &td->files[0];
+
+               memset(f, 0, sizeof(*f));
+               f->fd = -1;
+               f->file_name = strdup(jobname);
+       }
 
+       for_each_file(td, f, i) {
+               f->file_size = td->total_file_size / td->nr_files;
+               f->file_offset = td->start_offset;
+       }
+               
        fio_sem_init(&td->mutex, 0);
 
        td->clat_stat[0].min_val = td->clat_stat[1].min_val = ULONG_MAX;
        td->slat_stat[0].min_val = td->slat_stat[1].min_val = ULONG_MAX;
        td->bw_stat[0].min_val = td->bw_stat[1].min_val = ULONG_MAX;
 
-       if (td->min_bs == -1U)
-               td->min_bs = td->bs;
-       if (td->max_bs == -1U)
-               td->max_bs = td->bs;
-       if (td_read(td) && !td_rw(td))
-               td->verify = 0;
-
        if (td->stonewall && td->thread_number > 1)
                groupid++;
 
@@ -183,22 +732,38 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
        if (setup_rate(td))
                goto err;
 
-       if (write_lat_log) {
+       if (td->write_lat_log) {
                setup_log(&td->slat_log);
                setup_log(&td->clat_log);
        }
-       if (write_bw_log)
+       if (td->write_bw_log)
                setup_log(&td->bw_log);
 
-       if (td->name[0] == '\0')
-               snprintf(td->name, sizeof(td->name)-1, "client%d", td->thread_number);
+       if (!td->name)
+               td->name = strdup(jobname);
 
        ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
 
        if (!terse_output) {
-               if (!job_add_num)
-                       fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_engine_name, td->iodepth);
-               else if (job_add_num == 1)
+               if (!job_add_num) {
+                       if (td->io_ops->flags & FIO_CPUIO)
+                               fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
+                       else {
+                               char *c1, *c2, *c3, *c4;
+
+                               c1 = to_kmg(td->min_bs[DDIR_READ]);
+                               c2 = to_kmg(td->max_bs[DDIR_READ]);
+                               c3 = to_kmg(td->min_bs[DDIR_WRITE]);
+                               c4 = to_kmg(td->max_bs[DDIR_WRITE]);
+
+                               fprintf(f_out, "%s: (g=%d): rw=%s, odir=%u, bs=%s-%s/%s-%s, rate=%u, ioengine=%s, iodepth=%u\n", td->name, td->groupid, ddir_str[ddir], td->odirect, c1, c2, c3, c4, td->rate, td->io_ops->name, td->iodepth);
+
+                               free(c1);
+                               free(c2);
+                               free(c3);
+                               free(c4);
+                       }
+               } else if (job_add_num == 1)
                        fprintf(f_out, "...\n");
        }
 
@@ -233,7 +798,11 @@ err:
 int init_random_state(struct thread_data *td)
 {
        unsigned long seeds[4];
-       int fd, num_maps, blocks;
+       int fd, num_maps, blocks, i;
+       struct fio_file *f;
+
+       if (td->io_ops->flags & FIO_CPUIO)
+               return 0;
 
        fd = open("/dev/urandom", O_RDONLY);
        if (fd == -1) {
@@ -257,13 +826,17 @@ int init_random_state(struct thread_data *td)
                return 0;
 
        if (td->rand_repeatable)
-               seeds[3] = DEF_RANDSEED;
-
-       blocks = (td->io_size + td->min_bs - 1) / td->min_bs;
-       num_maps = blocks / BLOCKS_PER_MAP;
-       td->file_map = malloc(num_maps * sizeof(long));
-       td->num_maps = num_maps;
-       memset(td->file_map, 0, num_maps * sizeof(long));
+               seeds[3] = FIO_RANDSEED;
+
+       if (!td->norandommap) {
+               for_each_file(td, f, i) {
+                       blocks = (f->file_size + td->rw_min_bs - 1) / td->rw_min_bs;
+                       num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP;
+                       f->file_map = malloc(num_maps * sizeof(long));
+                       f->num_maps = num_maps;
+                       memset(f->file_map, 0, num_maps * sizeof(long));
+               }
+       }
 
        os_random_seed(seeds[3], &td->random_state);
        return 0;
@@ -283,230 +856,6 @@ static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu)
 #endif
 }
 
-static unsigned long get_mult_time(char c)
-{
-       switch (c) {
-               case 'm':
-               case 'M':
-                       return 60;
-               case 'h':
-               case 'H':
-                       return 60 * 60;
-               case 'd':
-               case 'D':
-                       return 24 * 60 * 60;
-               default:
-                       return 1;
-       }
-}
-
-static unsigned long get_mult_bytes(char c)
-{
-       switch (c) {
-               case 'k':
-               case 'K':
-                       return 1024;
-               case 'm':
-               case 'M':
-                       return 1024 * 1024;
-               case 'g':
-               case 'G':
-                       return 1024 * 1024 * 1024;
-               default:
-                       return 1;
-       }
-}
-
-/*
- * convert string after '=' into decimal value, noting any size suffix
- */
-static int str_to_decimal(char *p, unsigned long long *val, int kilo)
-{
-       char *str;
-       int len;
-
-       str = strchr(p, '=');
-       if (!str)
-               return 1;
-
-       str++;
-       len = strlen(str);
-
-       *val = strtoul(str, NULL, 10);
-       if (*val == ULONG_MAX && errno == ERANGE)
-               return 1;
-
-       if (kilo)
-               *val *= get_mult_bytes(str[len - 1]);
-       else
-               *val *= get_mult_time(str[len - 1]);
-       return 0;
-}
-
-static int check_str_bytes(char *p, char *name, unsigned long long *val)
-{
-       if (strncmp(p, name, strlen(name) - 1))
-               return 1;
-
-       return str_to_decimal(p, val, 1);
-}
-
-static int check_str_time(char *p, char *name, unsigned long long *val)
-{
-       if (strncmp(p, name, strlen(name) - 1))
-               return 1;
-
-       return str_to_decimal(p, val, 0);
-}
-
-static void strip_blank_front(char **p)
-{
-       char *s = *p;
-
-       while (isspace(*s))
-               s++;
-}
-
-static void strip_blank_end(char *p)
-{
-       char *s = p + strlen(p) - 1;
-
-       while (isspace(*s) || iscntrl(*s))
-               s--;
-
-       *(s + 1) = '\0';
-}
-
-typedef int (str_cb_fn)(struct thread_data *, char *);
-
-static int check_str(char *p, char *name, str_cb_fn *cb, struct thread_data *td)
-{
-       char *s;
-
-       if (strncmp(p, name, strlen(name)))
-               return 1;
-
-       s = strstr(p, name);
-       if (!s)
-               return 1;
-
-       s = strchr(s, '=');
-       if (!s)
-               return 1;
-
-       s++;
-       strip_blank_front(&s);
-       return cb(td, s);
-}
-
-static int check_strstore(char *p, char *name, char *dest)
-{
-       char *s;
-
-       if (strncmp(p, name, strlen(name)))
-               return 1;
-
-       s = strstr(p, name);
-       if (!s)
-               return 1;
-
-       s = strchr(p, '=');
-       if (!s)
-               return 1;
-
-       s++;
-       strip_blank_front(&s);
-
-       strcpy(dest, s);
-       return 0;
-}
-
-static int __check_range_bytes(char *str, unsigned long *val)
-{
-       char suffix;
-
-       if (sscanf(str, "%lu%c", val, &suffix) == 2) {
-               *val *= get_mult_bytes(suffix);
-               return 0;
-       }
-
-       if (sscanf(str, "%lu", val) == 1)
-               return 0;
-
-       return 1;
-}
-
-static int check_range_bytes(char *p, char *name, unsigned long *s,
-                            unsigned long *e)
-{
-       char option[128];
-       char *str, *p1, *p2;
-
-       if (strncmp(p, name, strlen(name)))
-               return 1;
-
-       strcpy(option, p);
-       p = option;
-
-       str = strstr(p, name);
-       if (!str)
-               return 1;
-
-       p += strlen(name);
-
-       str = strchr(p, '=');
-       if (!str)
-               return 1;
-
-       /*
-        * 'p' now holds whatever is after the '=' sign
-        */
-       p1 = str + 1;
-
-       /*
-        * terminate p1 at the '-' sign
-        */
-       p = strchr(p1, '-');
-       if (!p)
-               return 1;
-
-       p2 = p + 1;
-       *p = '\0';
-
-       if (!__check_range_bytes(p1, s) && !__check_range_bytes(p2, e))
-               return 0;
-
-       return 1;
-}
-
-static int check_int(char *p, char *name, unsigned int *val)
-{
-       char *str;
-
-       if (strncmp(p, name, strlen(name)))
-               return 1;
-
-       str = strstr(p, name);
-       if (!str)
-               return 1;
-
-       str = strchr(p, '=');
-       if (!str)
-               return 1;
-
-       str++;
-
-       if (sscanf(str, "%u", val) == 1)
-               return 0;
-
-       return 1;
-}
-
-static int check_strset(char *p, char *name)
-{
-       return strncmp(p, name, strlen(name));
-}
-
 static int is_empty_or_comment(char *line)
 {
        unsigned int i;
@@ -521,8 +870,10 @@ static int is_empty_or_comment(char *line)
        return 1;
 }
 
-static int str_rw_cb(struct thread_data *td, char *mem)
+static int str_rw_cb(void *data, const char *mem)
 {
+       struct thread_data *td = data;
+
        if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) {
                td->ddir = DDIR_READ;
                td->sequential = 1;
@@ -540,12 +891,12 @@ static int str_rw_cb(struct thread_data *td, char *mem)
                td->sequential = 0;
                return 0;
        } else if (!strncmp(mem, "rw", 2)) {
-               td->ddir = 0;
+               td->ddir = DDIR_READ;
                td->iomix = 1;
                td->sequential = 1;
                return 0;
        } else if (!strncmp(mem, "randrw", 6)) {
-               td->ddir = 0;
+               td->ddir = DDIR_READ;
                td->iomix = 1;
                td->sequential = 0;
                return 0;
@@ -555,8 +906,10 @@ static int str_rw_cb(struct thread_data *td, char *mem)
        return 1;
 }
 
-static int str_verify_cb(struct thread_data *td, char *mem)
+static int str_verify_cb(void *data, const char *mem)
 {
+       struct thread_data *td = data;
+
        if (!strncmp(mem, "0", 1)) {
                td->verify = VERIFY_NONE;
                return 0;
@@ -572,70 +925,134 @@ static int str_verify_cb(struct thread_data *td, char *mem)
        return 1;
 }
 
-static int str_mem_cb(struct thread_data *td, char *mem)
+/*
+ * Check if mmap/mmaphuge has a :/foo/bar/file at the end. If so, return that.
+ */
+static char *get_mmap_file(const char *str)
+{
+       char *p = strstr(str, ":");
+
+       if (!p)
+               return NULL;
+
+       p++;
+       strip_blank_front(&p);
+       strip_blank_end(p);
+       return strdup(p);
+}
+
+static int str_mem_cb(void *data, const char *mem)
 {
+       struct thread_data *td = data;
+
        if (!strncmp(mem, "malloc", 6)) {
                td->mem_type = MEM_MALLOC;
                return 0;
-       } else if (!strncmp(mem, "shm", 3)) {
-               td->mem_type = MEM_SHM;
+       } else if (!strncmp(mem, "mmaphuge", 8)) {
+#ifdef FIO_HAVE_HUGETLB
+               /*
+                * mmaphuge must be appended with the actual file
+                */
+               td->mmapfile = get_mmap_file(mem);
+               if (!td->mmapfile) {
+                       log_err("fio: mmaphuge:/path/to/file\n");
+                       return 1;
+               }
+
+               td->mem_type = MEM_MMAPHUGE;
                return 0;
+#else
+               log_err("fio: mmaphuge not available\n");
+               return 1;
+#endif
        } else if (!strncmp(mem, "mmap", 4)) {
+               /*
+                * Check if the user wants file backed memory. It's ok
+                * if there's no file given, we'll just use anon mamp then.
+                */
+               td->mmapfile = get_mmap_file(mem);
                td->mem_type = MEM_MMAP;
                return 0;
+       } else if (!strncmp(mem, "shmhuge", 7)) {
+#ifdef FIO_HAVE_HUGETLB
+               td->mem_type = MEM_SHMHUGE;
+               return 0;
+#else
+               log_err("fio: shmhuge not available\n");
+               return 1;
+#endif
+       } else if (!strncmp(mem, "shm", 3)) {
+               td->mem_type = MEM_SHM;
+               return 0;
        }
 
-       log_err("fio: mem type: malloc, shm, mmap\n");
+       log_err("fio: mem type: malloc, shm, shmhuge, mmap, mmaphuge\n");
        return 1;
 }
 
-static int str_ioengine_cb(struct thread_data *td, char *str)
+static int str_ioengine_cb(void *data, const char *str)
 {
-       if (!strncmp(str, "linuxaio", 8) || !strncmp(str, "aio", 3) ||
-           !strncmp(str, "libaio", 6)) {
-               strcpy(td->io_engine_name, "libaio");
-               td->io_engine = FIO_LIBAIO;
-               return 0;
-       } else if (!strncmp(str, "posixaio", 8)) {
-               strcpy(td->io_engine_name, "posixaio");
-               td->io_engine = FIO_POSIXAIO;
-               return 0;
-       } else if (!strncmp(str, "sync", 4)) {
-               strcpy(td->io_engine_name, "sync");
-               td->io_engine = FIO_SYNCIO;
-               return 0;
-       } else if (!strncmp(str, "mmap", 4)) {
-               strcpy(td->io_engine_name, "mmap");
-               td->io_engine = FIO_MMAPIO;
-               return 0;
-       } else if (!strncmp(str, "sgio", 4)) {
-               strcpy(td->io_engine_name, "sgio");
-               td->io_engine = FIO_SGIO;
-               return 0;
-       } else if (!strncmp(str, "splice", 6)) {
-               strcpy(td->io_engine_name, "splice");
-               td->io_engine = FIO_SPLICEIO;
+       struct thread_data *td = data;
+
+       td->io_ops = load_ioengine(td, str);
+       if (td->io_ops)
                return 0;
-       }
 
-       log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice\n");
+       log_err("fio: ioengine= libaio, posixaio, sync, mmap, sgio, splice, cpu, null\n");
+       log_err("fio: or specify path to dynamic ioengine module\n");
        return 1;
 }
 
+static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
+{
+       mlock_size = *val;
+       return 0;
+}
+
+#ifdef FIO_HAVE_IOPRIO
+static int str_prioclass_cb(void *data, unsigned int *val)
+{
+       struct thread_data *td = data;
+
+       td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
+       return 0;
+}
+
+static int str_prio_cb(void *data, unsigned int *val)
+{
+       struct thread_data *td = data;
+
+       td->ioprio |= *val;
+       return 0;
+}
+#endif
+
+static int str_exitall_cb(void)
+{
+       exitall_on_terminate = 1;
+       return 0;
+}
+
+static int str_cpumask_cb(void *data, unsigned int *val)
+{
+       struct thread_data *td = data;
+
+       fill_cpu_mask(td->cpumask, *val);
+       return 0;
+}
+
 /*
  * This is our [ini] type file parser.
  */
-int parse_jobs_ini(char *file)
+static int parse_jobs_ini(char *file, int stonewall_flag)
 {
-       unsigned int prioclass, prio, cpu, global, il;
-       unsigned long long ull;
-       unsigned long ul1, ul2;
+       unsigned int global;
        struct thread_data *td;
-       char *string, *name, *tmpbuf;
+       char *string, *name;
        fpos_t off;
        FILE *f;
        char *p;
-       int ret = 0, stonewall = 1;
+       int ret = 0, stonewall;
 
        f = fopen(file, "r");
        if (!f) {
@@ -645,14 +1062,16 @@ int parse_jobs_ini(char *file)
 
        string = malloc(4096);
        name = malloc(256);
-       tmpbuf = malloc(4096);
+       memset(name, 0, 256);
 
-       while ((p = fgets(string, 4096, f)) != NULL) {
-               if (ret)
+       stonewall = stonewall_flag;
+       do {
+               p = fgets(string, 4095, f);
+               if (!p)
                        break;
                if (is_empty_or_comment(p))
                        continue;
-               if (sscanf(p, "[%s]", name) != 1)
+               if (sscanf(p, "[%255s]", name) != 1)
                        continue;
 
                global = !strncmp(name, "global", 6);
@@ -677,294 +1096,35 @@ int parse_jobs_ini(char *file)
                while ((p = fgets(string, 4096, f)) != NULL) {
                        if (is_empty_or_comment(p))
                                continue;
-                       if (strstr(p, "["))
-                               break;
+
                        strip_blank_front(&p);
-                       strip_blank_end(p);
 
-                       if (!check_int(p, "prio", &prio)) {
-#ifndef FIO_HAVE_IOPRIO
-                               log_err("io priorities not available\n");
-                               ret = 1;
-                               break;
-#endif
-                               td->ioprio |= prio;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "prioclass", &prioclass)) {
-#ifndef FIO_HAVE_IOPRIO
-                               log_err("io priorities not available\n");
-                               ret = 1;
-                               break;
-#else
-                               td->ioprio |= prioclass << IOPRIO_CLASS_SHIFT;
-                               fgetpos(f, &off);
-                               continue;
-#endif
-                       }
-                       if (!check_int(p, "direct", &il)) {
-                               td->odirect = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "rand_repeatable", &il)) {
-                               td->rand_repeatable = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "rate", &td->rate)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "ratemin", &td->ratemin)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "ratecycle", &td->ratecycle)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "thinktime", &td->thinktime)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "cpumask", &cpu)) {
-#ifndef FIO_HAVE_CPU_AFFINITY
-                               log_err("cpu affinity not available\n");
-                               ret = 1;
-                               break;
-#endif
-                               fill_cpu_mask(td->cpumask, cpu);
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "fsync", &td->fsync_blocks)) {
-                               fgetpos(f, &off);
-                               td->end_fsync = 1;
-                               continue;
-                       }
-                       if (!check_int(p, "startdelay", &td->start_delay)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_time(p, "timeout", &ull)) {
-                               td->timeout = ul1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "invalidate", &il)) {
-                               td->invalidate_cache = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "iodepth", &td->iodepth)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "sync", &il)) {
-                               td->sync_io = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "bwavgtime", &td->bw_avg_time)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "create_serialize", &il)) {
-                               td->create_serialize = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "create_fsync", &il)) {
-                               td->create_fsync = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "end_fsync", &il)) {
-                               td->end_fsync = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "loops", &td->loops)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "numjobs", &td->numjobs)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "overwrite", &il)) {
-                               td->overwrite = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "rwmixcycle", &td->rwmixcycle)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "rwmixread", &il)) {
-                               if (il > 100)
-                                       il = 100;
-                               td->rwmixread = il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "rwmixwrite", &il)) {
-                               if (il > 100)
-                                       il = 100;
-                               td->rwmixread = 100 - il;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_int(p, "nice", &td->nice)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_range_bytes(p, "bsrange", &ul1, &ul2)) {
-                               if (ul1 > ul2) {
-                                       td->max_bs = ul1;
-                                       td->min_bs = ul2;
-                               } else {
-                                       td->max_bs = ul2;
-                                       td->min_bs = ul1;
-                               }
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "bs", &ull)) {
-                               td->bs = ull;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "size", &td->file_size)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "offset", &td->file_offset)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "zonesize", &td->zone_size)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "zoneskip", &td->zone_skip)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str_bytes(p, "lockmem", &mlock_size)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "directory", tmpbuf)) {
-                               td->directory = strdup(tmpbuf);
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "name", tmpbuf)) {
-                               snprintf(td->name, sizeof(td->name)-1, "%s%d", tmpbuf, td->thread_number);
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str(p, "mem", str_mem_cb, td)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str(p, "verify", str_verify_cb, td)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str(p, "rw", str_rw_cb, td)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_str(p, "ioengine", str_ioengine_cb, td)) {
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strset(p, "create")) {
-                               td->create_file = 1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strset(p, "exitall")) {
-                               exitall_on_terminate = 1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strset(p, "stonewall")) {
-                               td->stonewall = 1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strset(p, "thread")) {
-                               td->use_thread = 1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "iolog", tmpbuf)) {
-                               if (td->write_iolog) {
-                                       log_err("fio: read iolog overrides given write_iolog\n");
-                                       free(td->iolog_file);
-                                       td->write_iolog = 0;
-                               }
-                               td->iolog_file = strdup(tmpbuf);
-                               td->read_iolog = 1;
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "write_iolog", tmpbuf)) {
-                               if (!td->read_iolog) {
-                                       td->iolog_file = strdup(tmpbuf);
-                                       td->write_iolog = 1;
-                               } else
-                                       log_err("fio: read iolog overrides given write_iolog\n");
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "exec_prerun", tmpbuf)) {
-                               td->exec_prerun = strdup(tmpbuf);
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "exec_postrun", tmpbuf)) {
-                               td->exec_postrun = strdup(tmpbuf);
-                               fgetpos(f, &off);
-                               continue;
-                       }
-                       if (!check_strstore(p, "ioscheduler", tmpbuf)) {
-#ifndef FIO_HAVE_IOSCHED_SWITCH
-                               log_err("io scheduler switching not available\n");
-                               ret = 1;
+                       if (p[0] == '[')
                                break;
-#else
-                               td->ioscheduler = strdup(tmpbuf);
-                               fgetpos(f, &off);
-                               continue;
-#endif
-                       }
+
+                       strip_blank_end(p);
+
+                       fgetpos(f, &off);
 
                        /*
                         * Don't break here, continue parsing options so we
                         * dump all the bad ones. Makes trial/error fixups
                         * easier on the user.
                         */
-                       printf("Client%d: bad option %s\n",td->thread_number,p);
-                       ret = 1;
+                       ret |= parse_option(p, options, td);
                }
 
                if (!ret) {
                        fsetpos(f, &off);
                        ret = add_job(td, name, 0);
+               } else {
+                       log_err("fio: job %s dropped\n", name);
+                       put_job(td);
                }
-               if (ret)
-                       break;
-       }
+       } while (!ret);
 
        free(string);
        free(name);
-       free(tmpbuf);
        fclose(f);
        return ret;
 }
@@ -979,36 +1139,14 @@ static int fill_def_thread(void)
        }
 
        /*
-        * fill globals
+        * fill default options
         */
-       def_thread.ddir = DDIR_READ;
-       def_thread.iomix = 0;
-       def_thread.bs = DEF_BS;
-       def_thread.min_bs = -1;
-       def_thread.max_bs = -1;
-       def_thread.io_engine = DEF_IO_ENGINE;
-       strcpy(def_thread.io_engine_name, DEF_IO_ENGINE_NAME);
-       def_thread.odirect = DEF_ODIRECT;
-       def_thread.ratecycle = DEF_RATE_CYCLE;
-       def_thread.sequential = DEF_SEQUENTIAL;
+       fill_default_options(&def_thread, options);
+
        def_thread.timeout = def_timeout;
-       def_thread.create_file = DEF_CREATE;
-       def_thread.overwrite = DEF_OVERWRITE;
-       def_thread.invalidate_cache = DEF_INVALIDATE;
-       def_thread.sync_io = DEF_SYNCIO;
-       def_thread.mem_type = MEM_MALLOC;
-       def_thread.bw_avg_time = DEF_BWAVGTIME;
-       def_thread.create_serialize = DEF_CREATE_SER;
-       def_thread.create_fsync = DEF_CREATE_FSYNC;
-       def_thread.loops = DEF_LOOPS;
-       def_thread.verify = DEF_VERIFY;
-       def_thread.stonewall = DEF_STONEWALL;
-       def_thread.numjobs = DEF_NUMJOBS;
-       def_thread.use_thread = DEF_USE_THREAD;
-       def_thread.rwmixcycle = DEF_RWMIX_CYCLE;
-       def_thread.rwmixread = DEF_RWMIX_READ;
-       def_thread.nice = DEF_NICE;
-       def_thread.rand_repeatable = DEF_RAND_REPEAT;
+       def_thread.write_bw_log = write_bw_log;
+       def_thread.write_lat_log = write_lat_log;
+
 #ifdef FIO_HAVE_DISK_UTIL
        def_thread.do_disk_util = 1;
 #endif
@@ -1016,74 +1154,99 @@ static int fill_def_thread(void)
        return 0;
 }
 
-static void usage(char *name)
+static void usage(void)
 {
        printf("%s\n", fio_version_string);
-       printf("\t-o Write output to file\n");
-       printf("\t-t Runtime in seconds\n");
-       printf("\t-l Generate per-job latency logs\n");
-       printf("\t-w Generate per-job bandwidth logs\n");
-       printf("\t-m Minimal (terse) output\n");
-       printf("\t-f Job file (Required)\n");
-       printf("\t-v Print version info and exit\n");
+       printf("\t--output\tWrite output to file\n");
+       printf("\t--timeout\tRuntime in seconds\n");
+       printf("\t--latency-log\tGenerate per-job latency logs\n");
+       printf("\t--bandwidth-log\tGenerate per-job bandwidth logs\n");
+       printf("\t--minimal\tMinimal (terse) output\n");
+       printf("\t--version\tPrint version info and exit\n");
+       printf("\t--help\t\tPrint this page\n");
+       printf("\t--cmdhelp=cmd\tPrint command help, \"all\" for all of them\n");
 }
 
 static int parse_cmd_line(int argc, char *argv[])
 {
-       int c, idx = 1, ini_idx = 0;
+       struct thread_data *td = NULL;
+       int c, ini_idx = 0, lidx, ret;
 
-       while ((c = getopt(argc, argv, "t:o:f:lwvhm")) != EOF) {
+       while ((c = getopt_long(argc, argv, "", long_options, &lidx)) != -1) {
                switch (c) {
-                       case 't':
-                               def_timeout = atoi(optarg);
-                               idx++;
-                               break;
-                       case 'f':
-                               ini_idx++;
-                               ini_file = realloc(ini_file, ini_idx * sizeof(char *));
-                               ini_file[ini_idx - 1] = strdup(optarg);
-                               idx++;
-                               break;
-                       case 'l':
-                               write_lat_log = 1;
-                               idx++;
-                               break;
-                       case 'w':
-                               write_bw_log = 1;
-                               idx++;
-                               break;
-                       case 'o':
-                               f_out = fopen(optarg, "w+");
-                               if (!f_out) {
-                                       perror("fopen output");
-                                       exit(1);
+               case 't':
+                       def_timeout = atoi(optarg);
+                       break;
+               case 'l':
+                       write_lat_log = 1;
+                       break;
+               case 'w':
+                       write_bw_log = 1;
+                       break;
+               case 'o':
+                       f_out = fopen(optarg, "w+");
+                       if (!f_out) {
+                               perror("fopen output");
+                               exit(1);
+                       }
+                       f_err = f_out;
+                       break;
+               case 'm':
+                       terse_output = 1;
+                       break;
+               case 'h':
+                       usage();
+                       exit(0);
+               case 'c':
+                       ret = show_cmd_help(options, optarg);
+                       exit(ret);
+               case 'v':
+                       printf("%s\n", fio_version_string);
+                       exit(0);
+               case FIO_GETOPT_JOB: {
+                       const char *opt = long_options[lidx].name;
+                       char *val = optarg;
+
+                       if (!strncmp(opt, "name", 4) && td) {
+                               ret = add_job(td, td->name ?: "fio", 0);
+                               if (ret) {
+                                       put_job(td);
+                                       return 0;
                                }
-                               f_err = f_out;
-                               idx++;
-                               break;
-                       case 'm':
-                               terse_output = 1;
-                               idx++;
-                               break;
-                       case 'h':
-                               usage(argv[0]);
-                               exit(0);
-                       case 'v':
-                               printf("%s\n", fio_version_string);
-                               exit(0);
+                               td = NULL;
+                       }
+                       if (!td) {
+                               int global = !strncmp(val, "global", 6);
+
+                               td = get_new_job(global, &def_thread);
+                               if (!td)
+                                       return 0;
+                       }
+
+                       ret = parse_cmd_option(opt, val, options, td);
+                       if (ret) {
+                               log_err("fio: job dropped\n");
+                               put_job(td);
+                               td = NULL;
+                       }
+                       break;
+               }
+               default:
+                       break;
                }
        }
 
-       while (idx < argc) {
+       if (td) {
+               ret = add_job(td, td->name ?: "fio", 0);
+               if (ret)
+                       put_job(td);
+       }
+
+       while (optind < argc) {
                ini_idx++;
                ini_file = realloc(ini_file, ini_idx * sizeof(char *));
-               ini_file[ini_idx - 1] = strdup(argv[idx]);
-               idx++;
-       }
-               
-       if (!f_out) {
-               f_out = stdout;
-               f_err = stderr;
+               ini_file[ini_idx - 1] = strdup(argv[optind]);
+               optind++;
        }
 
        return ini_idx;
@@ -1138,30 +1301,66 @@ static int setup_thread_area(void)
        return 0;
 }
 
+/*
+ * Copy the fio options into the long options map, so we mirror
+ * job and cmd line options.
+ */
+static void dupe_job_options(void)
+{
+       struct fio_option *o;
+       unsigned int i;
+
+       i = 0;
+       while (long_options[i].name)
+               i++;
+
+       o = &options[0];
+       while (o->name) {
+               long_options[i].name = o->name;
+               long_options[i].val = FIO_GETOPT_JOB;
+               if (o->type == FIO_OPT_STR_SET)
+                       long_options[i].has_arg = no_argument;
+               else
+                       long_options[i].has_arg = required_argument;
+
+               i++;
+               o++;
+               assert(i < FIO_JOB_OPTS + FIO_CMD_OPTS);
+       }
+}
+
 int parse_options(int argc, char *argv[])
 {
        int job_files, i;
 
+       f_out = stdout;
+       f_err = stderr;
+
+       options_init(options);
+
+       dupe_job_options();
+
        if (setup_thread_area())
                return 1;
        if (fill_def_thread())
                return 1;
 
        job_files = parse_cmd_line(argc, argv);
-       if (!job_files) {
-               log_err("Need job file(s)\n");
-               usage(argv[0]);
-               return 1;
-       }
 
        for (i = 0; i < job_files; i++) {
                if (fill_def_thread())
                        return 1;
-               if (parse_jobs_ini(ini_file[i]))
+               if (parse_jobs_ini(ini_file[i], i))
                        return 1;
                free(ini_file[i]);
        }
 
        free(ini_file);
+
+       if (!thread_number) {
+               log_err("No jobs defined(s)\n");
+               return 1;
+       }
+
        return 0;
 }