X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=init.c;h=af7f2c8f14ec1cce773e3b3cc94c88f07382377c;hp=1bcd487941aee64e2e2107421968616e816e7436;hb=fdf3de27417d9a673466041851266ffb5c88cab8;hpb=549577a728fc3bacbfea8978c4f514cef5b073dc

diff --git a/init.c b/init.c
index 1bcd4879..af7f2c8f 100644
--- a/init.c
+++ b/init.c
@@ -18,329 +18,517 @@
 #include "fio.h"
 #include "parse.h"
 
-/*
- * The default options
- */
-#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_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)
-#define DEF_NR_FILES		(1)
-#define DEF_UNLINK		(0)
-#define DEF_WRITE_BW_LOG	(0)
-#define DEF_WRITE_LAT_LOG	(0)
+#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	= "description",
+		.type	= FIO_OPT_STR_STORE,
+		.off1	= td_var_offset(description),
+		.help	= "Text job description",
+	},
 	{
 		.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,
+		.off1	= td_var_offset(td_ddir),
+		.help	= "IO direction",
+		.def	= "read",
+		.posval = {
+			  { .ival = "read", .oval = TD_DDIR_READ },
+			  { .ival = "write", .oval = TD_DDIR_WRITE },
+			  { .ival = "randread", .oval = TD_DDIR_RANDREAD },
+			  { .ival = "randwrite", .oval = TD_DDIR_RANDWRITE },
+			  { .ival = "rw", .oval = TD_DDIR_RW },
+			  { .ival = "randrw", .oval = TD_DDIR_RANDRW },
+			  },
 	},
 	{
 		.name	= "ioengine",
 		.type	= FIO_OPT_STR,
 		.cb	= str_ioengine_cb,
+		.help	= "IO engine to use",
+		.def	= "sync",
+		.posval	= {
+			  { .ival = "sync", },
+#ifdef FIO_HAVE_LIBAIO
+			  { .ival = "libaio", },
+#endif
+#ifdef FIO_HAVE_POSIXAIO
+			  { .ival = "posixaio", },
+#endif
+			  { .ival = "mmap", },
+#ifdef FIO_HAVE_SPLICE
+			  { .ival = "splice", },
+#endif
+#ifdef FIO_HAVE_SGIO
+			  { .ival = "sg", },
+#endif
+			  { .ival = "null", }, { .ival = "net", },
+#ifdef FIO_HAVE_SYSLET
+			  { .ival = "syslet-rw", },
+#endif
+			  { .ival = "external", },
+			  },
+	},
+	{
+		.name	= "iodepth",
+		.type	= FIO_OPT_INT,
+		.off1	= td_var_offset(iodepth),
+		.help	= "Amount of IO buffers to keep in flight",
+		.def	= "1",
+	},
+	{
+		.name	= "iodepth_batch",
+		.type	= FIO_OPT_INT,
+		.off1	= td_var_offset(iodepth_batch),
+		.help	= "Number of IO to submit in one go",
+	},
+	{
+		.name	= "iodepth_low",
+		.type	= FIO_OPT_INT,
+		.off1	= td_var_offset(iodepth_low),
+		.help	= "Low water mark for queuing depth",
+	},
+	{
+		.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	= "file_service_type",
+		.type	= FIO_OPT_STR,
+		.off1	= td_var_offset(file_service_type),
+		.help	= "How to select which file to service next",
+		.def	= "roundrobin",
+		.posval	= {
+			  { .ival = "random", .oval = FIO_FSERVICE_RANDOM },
+			  { .ival = "roundrobin", .oval = FIO_FSERVICE_RR },
+			  },
+	},
+	{
+		.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,
+		.off1	= td_var_offset(mem_type),
+		.help	= "Backing type for IO buffers",
+		.def	= "malloc",
+		.posval	= {
+			  { .ival = "malloc", .oval = MEM_MALLOC },
+			  { .ival = "shm", .oval = MEM_SHM },
+#ifdef FIO_HAVE_HUGETLB
+			  { .ival = "shmhuge", .oval = MEM_SHMHUGE },
+#endif
+			  { .ival = "mmap", .oval = MEM_MMAP },
+#ifdef FIO_HAVE_HUGETLB
+			  { .ival = "mmaphuge", .oval = MEM_MMAPHUGE },
+#endif
+			  },
 	},
 	{
 		.name	= "verify",
 		.type	= FIO_OPT_STR,
-		.cb	= str_verify_cb,
+		.off1	= td_var_offset(verify),
+		.help	= "Verify sum function",
+		.def	= "0",
+		.posval = {
+			  { .ival = "0", .oval = VERIFY_NONE },
+			  { .ival = "crc32", .oval = VERIFY_CRC32 },
+			  { .ival = "md5", .oval = VERIFY_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	= "size",
-		.type	= FIO_OPT_STR_VAL,
-		.off1	= td_var_offset(total_file_size),
-	},
-	{
-		.name	= "bs",
-		.type	= FIO_OPT_STR_VAL,
-		.off1	= td_var_offset(bs),
-	},
-	{
-		.name	= "offset",
-		.type	= FIO_OPT_STR_VAL,
-		.off1	= td_var_offset(start_offset),
-	},
 	{
 		.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,
-	},
-	{
-		.name	= "bsrange",
-		.type	= FIO_OPT_RANGE,
-		.off1	= td_var_offset(min_bs),
-		.off2	= td_var_offset(max_bs),
-	},
-	{
-		.name	= "nrfiles",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(nr_files),
-	},
-	{
-		.name	= "iodepth",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(iodepth),
-	},
-	{
-		.name	= "fsync",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(fsync_blocks),
+		.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),
-		.max_val= 100,
+		.maxval	= 100,
+		.help	= "Percentage of mixed workload that is reads",
+		.def	= "50",
 	},
 	{
 		.name	= "rwmixwrite",
 		.type	= FIO_OPT_INT,
 		.off1	= td_var_offset(rwmixwrite),
-		.max_val= 100,
+		.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)
+		.off1	= td_var_offset(thinktime),
+		.help	= "Idle time between IO buffers (usec)",
+		.def	= "0",
 	},
 	{
-		.name	= "rate",
+		.name	= "thinktime_spin",
 		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(rate)
+		.off1	= td_var_offset(thinktime_spin),
+		.help	= "Start think time by spinning this amount (usec)",
+		.def	= "0",
 	},
 	{
-		.name	= "ratemin",
+		.name	= "thinktime_blocks",
 		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(ratemin)
+		.off1	= td_var_offset(thinktime_blocks),
+		.help	= "IO buffer period between 'thinktime'",
+		.def	= "1",
 	},
 	{
-		.name	= "ratecycle",
+		.name	= "rate",
 		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(ratecycle)
+		.off1	= td_var_offset(rate),
+		.help	= "Set bandwidth rate",
 	},
 	{
-		.name	= "startdelay",
+		.name	= "ratemin",
 		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(start_delay)
+		.off1	= td_var_offset(ratemin),
+		.help	= "The bottom limit accepted",
 	},
 	{
-		.name	= "timeout",
-		.type	= FIO_OPT_STR_VAL_TIME,
-		.off1	= td_var_offset(timeout)
+		.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_INT,
-		.off1	= td_var_offset(invalidate_cache)
+		.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_INT,
-		.off1	= td_var_offset(sync_io)
+		.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)
+		.off1	= td_var_offset(bw_avg_time),
+		.help	= "Time window over which to calculate bandwidth (msec)",
+		.def	= "500",
 	},
 	{
 		.name	= "create_serialize",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(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_INT,
-		.off1	= td_var_offset(create_fsync)
-	},
-	{
-		.name	= "loops",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(loops)
-	},
-	{
-		.name	= "numjobs",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(numjobs)
+		.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)
+		.off1	= td_var_offset(cpuload),
+		.help	= "Use this percentage of CPU",
 	},
 	{
 		.name	= "cpuchunks",
 		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(cpucycle)
-	},
-	{
-		.name	= "direct",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(odirect)
-	},
-	{
-		.name	= "overwrite",
-		.type	= FIO_OPT_INT,
-		.off1	= td_var_offset(overwrite)
+		.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_INT,
-		.off1	= td_var_offset(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_STR_SET,
+		.type	= FIO_OPT_BOOL,
 		.off1	= td_var_offset(unlink),
+		.help	= "Unlink created files after job has completed",
+		.def	= "0",
 	},
 	{
 		.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(thread),
+		.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,
@@ -386,14 +574,24 @@ static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
 		.has_arg	= no_argument,
 		.val		= 'v',
 	},
+	{
+		.name		= "help",
+		.has_arg	= no_argument,
+		.val		= 'h',
+	},
+	{
+		.name		= "cmdhelp",
+		.has_arg	= optional_argument,
+		.val		= 'c',
+	},
 	{
 		.name		= NULL,
 	},
 };
 
-static int def_timeout = DEF_TIMEOUT;
+static int def_timeout = 0;
 
-static char fio_version_string[] = "fio 1.7";
+static char fio_version_string[] = "fio 1.13";
 
 static char **ini_file;
 static int max_jobs = MAX_JOBS;
@@ -401,15 +599,24 @@ static int max_jobs = MAX_JOBS;
 struct thread_data def_thread;
 struct thread_data *threads = NULL;
 
-int rate_quit = 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 = DEF_WRITE_LAT_LOG;
-static int write_bw_log = DEF_WRITE_BW_LOG;
+static int write_lat_log = 0;
+int write_bw_log = 0;
+
+FILE *get_f_out()
+{
+	return f_out;
+}
+
+FILE *get_f_err()
+{
+	return f_err;
+}
 
 /*
  * Return a free job structure.
@@ -435,6 +642,9 @@ static void put_job(struct thread_data *td)
 	if (td == &def_thread)
 		return;
 
+	if (td->error)
+		fprintf(f_out, "fio: %s\n", td->verror);
+
 	memset(&threads[td->thread_number - 1], 0, sizeof(*td));
 	thread_number--;
 }
@@ -445,11 +655,6 @@ static void put_job(struct thread_data *td)
  */
 static void fixup_options(struct thread_data *td)
 {
-	if (!td->min_bs)
-		td->min_bs = td->bs;
-	if (!td->max_bs)
-		td->max_bs = td->bs;
-
 	if (!td->rwmixread && td->rwmixwrite)
 		td->rwmixread = 100 - td->rwmixwrite;
 
@@ -458,6 +663,100 @@ static void fixup_options(struct thread_data *td)
 		free(td->write_iolog_file);
 		td->write_iolog_file = NULL;
 	}
+
+	if (td->io_ops->flags & FIO_SYNCIO)
+		td->iodepth = 1;
+	else {
+		if (!td->iodepth)
+			td->iodepth = td->nr_files;
+	}
+
+	/*
+	 * only really works for sequential io for now, and with 1 file
+	 */
+	if (td->zone_size && td_random(td) && td->nr_files == 1)
+		td->zone_size = 0;
+
+	/*
+	 * Reads can do overwrites, we always need to pre-create the file
+	 */
+	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;
+
+	/*
+	 * thinktime_spin must be less than thinktime
+	 */
+	if (td->thinktime_spin > td->thinktime)
+		td->thinktime_spin = td->thinktime;
+
+	/*
+	 * The low water mark cannot be bigger than the iodepth
+	 */
+	if (td->iodepth_low > td->iodepth || !td->iodepth_low) {
+		/*
+		 * syslet work around - if the workload is sequential,
+		 * we want to let the queue drain all the way down to
+		 * avoid seeking between async threads
+		 */
+		if (!strcmp(td->io_ops->name, "syslet-rw") && !td_random(td))
+			td->iodepth_low = 1;
+		else
+			td->iodepth_low = td->iodepth;
+	}
+
+	/*
+	 * If batch number isn't set, default to the same as iodepth
+	 */
+	if (td->iodepth_batch > td->iodepth || !td->iodepth_batch)
+		td->iodepth_batch = td->iodepth;
+}
+
+/*
+ * This function leaks the buffer
+ */
+static char *to_kmg(unsigned int val)
+{
+	char *buf = malloc(32);
+	char post[] = { 0, 'K', 'M', 'G', 'P', 'E', 0 };
+	char *p = post;
+
+	do {
+		if (val & 1023)
+			break;
+
+		val >>= 10;
+		p++;
+	} while (*p);
+
+	snprintf(buf, 31, "%u%c", val, *p);
+	return buf;
 }
 
 /*
@@ -467,73 +766,32 @@ static void fixup_options(struct thread_data *td)
  */
 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 {
-	char *ddir_str[] = { "read", "write", "randread", "randwrite",
-			     "rw", NULL, "randrw" };
+	const char *ddir_str[] = { NULL, "read", "write", "rw", NULL,
+				   "randread", "randwrite", "randrw" };
 	struct stat sb;
-	int numjobs, ddir, i;
+	int numjobs, i;
 	struct fio_file *f;
 
-#ifndef FIO_HAVE_LIBAIO
-	if (td->io_engine == FIO_LIBAIO) {
-		log_err("Linux libaio not available\n");
-		return 1;
-	}
-#endif
-#ifndef FIO_HAVE_POSIXAIO
-	if (td->io_engine == FIO_POSIXAIO) {
-		log_err("posix aio not available\n");
-		return 1;
-	}
-#endif
-
-	fixup_options(td);
-
 	/*
 	 * the def_thread is just for options, it's not a real job
 	 */
 	if (td == &def_thread)
 		return 0;
 
-	/*
-	 * Set default io engine, if none set
-	 */
-	if (!td->io_ops) {
-		td->io_ops = load_ioengine(td, DEF_IO_ENGINE_NAME);
-		if (!td->io_ops) {
-			log_err("default engine %s not there?\n", DEF_IO_ENGINE_NAME);
-			return 1;
-		}
-	}
+	assert(td->io_ops);
 
-	if (td->io_ops->flags & FIO_SYNCIO)
-		td->iodepth = 1;
-	else {
-		if (!td->iodepth)
-			td->iodepth = td->nr_files;
-	}
-
-	/*
-	 * only really works for sequential io for now, and with 1 file
-	 */
-	if (td->zone_size && !td->sequential && td->nr_files == 1)
-		td->zone_size = 0;
-
-	/*
-	 * Reads can do overwrites, we always need to pre-create the file
-	 */
-	if (td_read(td) || td_rw(td))
-		td->overwrite = 1;
+	if (td->odirect)
+		td->io_ops->flags |= FIO_RAWIO;
 
 	td->filetype = FIO_TYPE_FILE;
-	if (!stat(jobname, &sb)) {
+	if (td->filename && !lstat(td->filename, &sb)) {
 		if (S_ISBLK(sb.st_mode))
 			td->filetype = FIO_TYPE_BD;
 		else if (S_ISCHR(sb.st_mode))
 			td->filetype = FIO_TYPE_CHAR;
 	}
 
-	if (td->odirect)
-		td->io_ops->flags |= FIO_RAWIO;
+	fixup_options(td);
 
 	if (td->filename)
 		td->nr_uniq_files = 1;
@@ -543,10 +801,19 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 	if (td->filetype == FIO_TYPE_FILE || td->filename) {
 		char tmp[PATH_MAX];
 		int len = 0;
-		int i;
 
-		if (td->directory && td->directory[0] != '\0')
-			sprintf(tmp, "%s/", td->directory);
+		if (td->directory && td->directory[0] != '\0') {
+			if (lstat(td->directory, &sb) < 0) {
+				log_err("fio: %s is not a directory\n", td->directory);
+				td_verror(td, errno, "lstat");
+				return 1;
+			}
+			if (!S_ISDIR(sb.st_mode)) {
+				log_err("fio: %s is not a directory\n", td->directory);
+				return 1;
+			}
+			len = sprintf(tmp, "%s/", td->directory);
+		}
 
 		td->files = malloc(sizeof(struct fio_file) * td->nr_files);
 
@@ -577,16 +844,9 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 		
 	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;
+	td->ts.clat_stat[0].min_val = td->ts.clat_stat[1].min_val = ULONG_MAX;
+	td->ts.slat_stat[0].min_val = td->ts.slat_stat[1].min_val = ULONG_MAX;
+	td->ts.bw_stat[0].min_val = td->ts.bw_stat[1].min_val = ULONG_MAX;
 
 	if (td->stonewall && td->thread_number > 1)
 		groupid++;
@@ -597,23 +857,34 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 		goto err;
 
 	if (td->write_lat_log) {
-		setup_log(&td->slat_log);
-		setup_log(&td->clat_log);
+		setup_log(&td->ts.slat_log);
+		setup_log(&td->ts.clat_log);
 	}
 	if (td->write_bw_log)
-		setup_log(&td->bw_log);
+		setup_log(&td->ts.bw_log);
 
 	if (!td->name)
 		td->name = strdup(jobname);
 
-	ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
-
 	if (!terse_output) {
 		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
-				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_ops->name, td->iodepth);
+			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, bs=%s-%s/%s-%s, ioengine=%s, iodepth=%u\n", td->name, td->groupid, ddir_str[td->td_ddir], c1, c2, c3, c4, td->io_ops->name, td->iodepth);
+
+				free(c1);
+				free(c2);
+				free(c3);
+				free(c4);
+			}
 		} else if (job_add_num == 1)
 			fprintf(f_out, "...\n");
 	}
@@ -648,7 +919,7 @@ err:
  */
 int init_random_state(struct thread_data *td)
 {
-	unsigned long seeds[4];
+	unsigned long seeds[5];
 	int fd, num_maps, blocks, i;
 	struct fio_file *f;
 
@@ -657,12 +928,12 @@ int init_random_state(struct thread_data *td)
 
 	fd = open("/dev/urandom", O_RDONLY);
 	if (fd == -1) {
-		td_verror(td, errno);
+		td_verror(td, errno, "open");
 		return 1;
 	}
 
 	if (read(fd, seeds, sizeof(seeds)) < (int) sizeof(seeds)) {
-		td_verror(td, EIO);
+		td_verror(td, EIO, "read");
 		close(fd);
 		return 1;
 	}
@@ -673,21 +944,26 @@ int init_random_state(struct thread_data *td)
 	os_random_seed(seeds[1], &td->verify_state);
 	os_random_seed(seeds[2], &td->rwmix_state);
 
-	if (td->sequential)
+	if (td->file_service_type == FIO_FSERVICE_RANDOM)
+		os_random_seed(seeds[3], &td->next_file_state);
+
+	if (!td_random(td))
 		return 0;
 
 	if (td->rand_repeatable)
-		seeds[3] = DEF_RANDSEED;
+		seeds[4] = FIO_RANDSEED * td->thread_number;
 
-	for_each_file(td, f, i) {
-		blocks = (f->file_size + td->min_bs - 1) / td->min_bs;
-		num_maps = blocks / 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));
+	if (!td->norandommap) {
+		for_each_file(td, f, i) {
+			blocks = (f->real_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);
+	os_random_seed(seeds[4], &td->random_state);
 	return 0;
 }
 
@@ -712,6 +988,8 @@ static int is_empty_or_comment(char *line)
 	for (i = 0; i < strlen(line); i++) {
 		if (line[i] == ';')
 			return 1;
+		if (line[i] == '#')
+			return 1;
 		if (!isspace(line[i]) && !iscntrl(line[i]))
 			return 0;
 	}
@@ -719,89 +997,60 @@ static int is_empty_or_comment(char *line)
 	return 1;
 }
 
-static int str_rw_cb(void *data, const 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)
 {
-	struct thread_data *td = data;
+	char *p = strstr(str, ":");
 
-	if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) {
-		td->ddir = DDIR_READ;
-		td->sequential = 1;
-		return 0;
-	} else if (!strncmp(mem, "randread", 8)) {
-		td->ddir = DDIR_READ;
-		td->sequential = 0;
-		return 0;
-	} else if (!strncmp(mem, "write", 5) || !strncmp(mem, "1", 1)) {
-		td->ddir = DDIR_WRITE;
-		td->sequential = 1;
-		return 0;
-	} else if (!strncmp(mem, "randwrite", 9)) {
-		td->ddir = DDIR_WRITE;
-		td->sequential = 0;
-		return 0;
-	} else if (!strncmp(mem, "rw", 2)) {
-		td->ddir = 0;
-		td->iomix = 1;
-		td->sequential = 1;
-		return 0;
-	} else if (!strncmp(mem, "randrw", 6)) {
-		td->ddir = 0;
-		td->iomix = 1;
-		td->sequential = 0;
-		return 0;
-	}
+	if (!p)
+		return NULL;
 
-	log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n");
-	return 1;
+	p++;
+	strip_blank_front(&p);
+	strip_blank_end(p);
+	return strdup(p);
 }
 
-static int str_verify_cb(void *data, const char *mem)
+static int str_mem_cb(void *data, const char *mem)
 {
 	struct thread_data *td = data;
 
-	if (!strncmp(mem, "0", 1)) {
-		td->verify = VERIFY_NONE;
-		return 0;
-	} else if (!strncmp(mem, "md5", 3) || !strncmp(mem, "1", 1)) {
-		td->verify = VERIFY_MD5;
-		return 0;
-	} else if (!strncmp(mem, "crc32", 5)) {
-		td->verify = VERIFY_CRC32;
-		return 0;
+	if (td->mem_type == MEM_MMAPHUGE || td->mem_type == MEM_MMAP) {
+		td->mmapfile = get_mmap_file(mem);
+		if (td->mem_type == MEM_MMAPHUGE && !td->mmapfile) {
+			log_err("fio: mmaphuge:/path/to/file\n");
+			return 1;
+		}
 	}
 
-	log_err("fio: verify types: md5, crc32\n");
-	return 1;
+	return 0;
 }
 
-static int str_mem_cb(void *data, const char *mem)
+/* External engines are specified by "external:name.o") */
+static const char *get_engine_name(const char *str)
 {
-	struct thread_data *td = data;
+	char *p = strstr(str, ":");
 
-	if (!strncmp(mem, "malloc", 6)) {
-		td->mem_type = MEM_MALLOC;
-		return 0;
-	} else if (!strncmp(mem, "shm", 3)) {
-		td->mem_type = MEM_SHM;
-		return 0;
-	} else if (!strncmp(mem, "mmap", 4)) {
-		td->mem_type = MEM_MMAP;
-		return 0;
-	}
+	if (!p)
+		return str;
 
-	log_err("fio: mem type: malloc, shm, mmap\n");
-	return 1;
+	p++;
+	strip_blank_front(&p);
+	strip_blank_end(p);
+	return p;
 }
 
 static int str_ioengine_cb(void *data, const char *str)
 {
 	struct thread_data *td = data;
+	const char *name = get_engine_name(str);
 
-	td->io_ops = load_ioengine(td, str);
+	td->io_ops = load_ioengine(td, name);
 	if (td->io_ops)
 		return 0;
 
-	log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice, cpu\n");
 	return 1;
 }
 
@@ -811,6 +1060,7 @@ static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
 	return 0;
 }
 
+#ifdef FIO_HAVE_IOPRIO
 static int str_prioclass_cb(void *data, unsigned int *val)
 {
 	struct thread_data *td = data;
@@ -826,6 +1076,7 @@ static int str_prio_cb(void *data, unsigned int *val)
 	td->ioprio |= *val;
 	return 0;
 }
+#endif
 
 static int str_exitall_cb(void)
 {
@@ -844,11 +1095,11 @@ static int str_cpumask_cb(void *data, unsigned int *val)
 /*
  * This is our [ini] type file parser.
  */
-int parse_jobs_ini(char *file, int stonewall_flag)
+static int parse_jobs_ini(char *file, int stonewall_flag)
 {
 	unsigned int global;
 	struct thread_data *td;
-	char *string, *name, *tmpbuf;
+	char *string, *name;
 	fpos_t off;
 	FILE *f;
 	char *p;
@@ -862,15 +1113,16 @@ int parse_jobs_ini(char *file, int stonewall_flag)
 
 	string = malloc(4096);
 	name = malloc(256);
-	tmpbuf = malloc(4096);
+	memset(name, 0, 256);
 
 	stonewall = stonewall_flag;
-	while ((p = fgets(string, 4096, f)) != NULL) {
-		if (ret)
+	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);
@@ -895,10 +1147,12 @@ int parse_jobs_ini(char *file, int stonewall_flag)
 		while ((p = fgets(string, 4096, f)) != NULL) {
 			if (is_empty_or_comment(p))
 				continue;
-			if (strstr(p, "["))
-				break;
 
 			strip_blank_front(&p);
+
+			if (p[0] == '[')
+				break;
+
 			strip_blank_end(p);
 
 			fgetpos(f, &off);
@@ -908,20 +1162,20 @@ int parse_jobs_ini(char *file, int stonewall_flag)
 			 * dump all the bad ones. Makes trial/error fixups
 			 * easier on the user.
 			 */
-			ret = parse_option(p, options, td);
+			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;
 }
@@ -936,37 +1190,14 @@ 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.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.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.nr_files = DEF_NR_FILES;
-	def_thread.unlink = DEF_UNLINK;
 	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
@@ -983,14 +1214,16 @@ static void usage(void)
 	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[])
 {
 	struct thread_data *td = NULL;
-	int c, ini_idx = 0, lidx, ret;
+	int c, ini_idx = 0, lidx, ret, dont_add_job = 0;
 
-	while ((c = getopt_long(argc, argv, "", long_options, &lidx)) != -1) {
+	while ((c = getopt_long_only(argc, argv, "", long_options, &lidx)) != -1) {
 		switch (c) {
 		case 't':
 			def_timeout = atoi(optarg);
@@ -1015,6 +1248,9 @@ static int parse_cmd_line(int argc, char *argv[])
 		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);
@@ -1038,19 +1274,24 @@ static int parse_cmd_line(int argc, char *argv[])
 					return 0;
 			}
 
-			parse_cmd_option(opt, val, options, td);
+			ret = parse_cmd_option(opt, val, options, td);
+			if (ret)
+				dont_add_job = 1;
 			break;
 		}
 		default:
-			printf("optarg <<%s>>\n", argv[optind]);
 			break;
 		}
 	}
 
 	if (td) {
-		ret = add_job(td, td->name ?: "fio", 0);
-		if (ret)
+		if (dont_add_job)
 			put_job(td);
+		else {
+			ret = add_job(td, td->name ?: "fio", 0);
+			if (ret)
+				put_job(td);
+		}
 	}
 
 	while (optind < argc) {
@@ -1147,6 +1388,8 @@ int parse_options(int argc, char *argv[])
 	f_out = stdout;
 	f_err = stderr;
 
+	options_init(options);
+
 	dupe_job_options();
 
 	if (setup_thread_area())
@@ -1168,7 +1411,6 @@ int parse_options(int argc, char *argv[])
 
 	if (!thread_number) {
 		log_err("No jobs defined(s)\n");
-		usage();
 		return 1;
 	}