X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=init.c;h=719e46d3f3a9fd2619fb6b80efdca4df1a0c4aa7;hp=54e2ecae28202622d6954308d192439bb55dca2c;hb=9d6d019b47963fd1e79c0b045bb3197ab595dab5;hpb=3c9b60c1fb144950ca51220ffa18f485672c3ba6 diff --git a/init.c b/init.c index 54e2ecae..719e46d3 100644 --- a/init.c +++ b/init.c @@ -18,47 +18,11 @@ #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 DEF_NO_RAND_MAP (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 *); @@ -66,297 +30,610 @@ static int str_prioclass_cb(void *, unsigned int *); #endif static int str_exitall_cb(void); static int str_cpumask_cb(void *, unsigned int *); +static int str_fst_cb(void *, const char *); +static int str_filename_cb(void *, const char *); +static int str_directory_cb(void *, const char *); +static int str_opendir_cb(void *, const char *); + +#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), + .cb = str_directory_cb, + .help = "Directory to store files in", }, { .name = "filename", .type = FIO_OPT_STR_STORE, .off1 = td_var_offset(filename), + .cb = str_filename_cb, + .help = "File(s) to use for the workload", + }, + { + .name = "opendir", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(opendir), + .cb = str_opendir_cb, + .help = "Recursively add files from this directory and down", }, { .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, + .help = "Sequential read", + }, + { .ival = "write", + .oval = TD_DDIR_WRITE, + .help = "Sequential write", + }, + { .ival = "randread", + .oval = TD_DDIR_RANDREAD, + .help = "Random read", + }, + { .ival = "randwrite", + .oval = TD_DDIR_RANDWRITE, + .help = "Random write", + }, + { .ival = "rw", + .oval = TD_DDIR_RW, + .help = "Sequential read and write mix", + }, + { .ival = "randrw", + .oval = TD_DDIR_RANDRW, + .help = "Random read and write mix" + }, + }, }, { .name = "ioengine", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(ioengine), + .help = "IO engine to use", + .def = "sync", + .posval = { + { .ival = "sync", + .help = "Use read/write", + }, +#ifdef FIO_HAVE_LIBAIO + { .ival = "libaio", + .help = "Linux native asynchronous IO", + }, +#endif +#ifdef FIO_HAVE_POSIXAIO + { .ival = "posixaio", + .help = "POSIX asynchronous IO", + }, +#endif + { .ival = "mmap", + .help = "Memory mapped IO", + }, +#ifdef FIO_HAVE_SPLICE + { .ival = "splice", + .help = "splice/vmsplice based IO", + }, +#endif +#ifdef FIO_HAVE_SGIO + { .ival = "sg", + .help = "SCSI generic v3 IO", + }, +#endif + { .ival = "null", + .help = "Testing engine (no data transfer)", + }, + { .ival = "net", + .help = "Network IO", + }, +#ifdef FIO_HAVE_SYSLET + { .ival = "syslet-rw", + .help = "syslet enabled async pread/pwrite IO", + }, +#endif + { .ival = "cpuio", + .help = "CPU cycler burner engine", + }, + { .ival = "external", + .help = "Load external engine (append name)", + }, + }, + }, + { + .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 = "openfiles", + .type = FIO_OPT_INT, + .off1 = td_var_offset(open_files), + .help = "Number of files to keep open at the same time", + }, + { + .name = "file_service_type", .type = FIO_OPT_STR, - .cb = str_ioengine_cb, + .cb = str_fst_cb, + .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, + .help = "Choose a file at random", + }, + { .ival = "roundrobin", + .oval = FIO_FSERVICE_RR, + .help = "Round robin select files", + }, + }, + }, + { + .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, + .help = "Use malloc(3) for IO buffers", + }, + { .ival = "shm", + .oval = MEM_SHM, + .help = "Use shared memory segments for IO buffers", + }, +#ifdef FIO_HAVE_HUGETLB + { .ival = "shmhuge", + .oval = MEM_SHMHUGE, + .help = "Like shm, but use huge pages", + }, +#endif + { .ival = "mmap", + .oval = MEM_MMAP, + .help = "Use mmap(2) (file or anon) for IO buffers", + }, +#ifdef FIO_HAVE_HUGETLB + { .ival = "mmaphuge", + .oval = MEM_MMAPHUGE, + .help = "Like mmap, but use huge pages", + }, +#endif + }, }, { .name = "verify", .type = FIO_OPT_STR, - .cb = str_verify_cb, + .off1 = td_var_offset(verify), + .help = "Verify data written", + .def = "0", + .posval = { + { .ival = "0", + .oval = VERIFY_NONE, + .help = "Don't do IO verification", + }, + { .ival = "crc32", + .oval = VERIFY_CRC32, + .help = "Use crc32 checksums for verification", + }, + { .ival = "md5", + .oval = VERIFY_MD5, + .help = "Use md5 checksums for verification", + }, + }, }, { .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_INT, - .off1 = td_var_offset(bs[DDIR_READ]), - .off2 = td_var_offset(bs[DDIR_WRITE]), - }, - { - .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[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]), - }, - { - .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 (usecs)", + .def = "50000", }, #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 = "fsync_on_close", + .type = FIO_OPT_BOOL, + .off1 = td_var_offset(fsync_on_close), + .help = "fsync files on close", + .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 = "norandommap", - .type = FIO_OPT_STR_SET, - .off1 = td_var_offset(norandommap), + .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 = "bs_unaligned", + .name = "group_reporting", .type = FIO_OPT_STR_SET, - .off1 = td_var_offset(bs_unaligned), + .off1 = td_var_offset(group_reporting), + .help = "Do reporting on a per-group basis", }, { .name = NULL, @@ -402,14 +679,24 @@ 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 1.8"; +static char fio_version_string[] = "fio 1.14"; static char **ini_file; static int max_jobs = MAX_JOBS; @@ -417,15 +704,26 @@ 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; + +static int prev_group_jobs; + +FILE *get_f_out() +{ + return f_out; +} + +FILE *get_f_err() +{ + return f_err; +} /* * Return a free job structure. @@ -451,6 +749,9 @@ 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--; } @@ -474,13 +775,13 @@ static void fixup_options(struct thread_data *td) td->iodepth = 1; else { if (!td->iodepth) - td->iodepth = td->nr_files; + td->iodepth = td->open_files; } /* * only really works for sequential io for now, and with 1 file */ - if (td->zone_size && !td->sequential && td->nr_files == 1) + if (td->zone_size && td_random(td) && td->open_files == 1) td->zone_size = 0; /* @@ -509,6 +810,39 @@ static void fixup_options(struct thread_data *td) } if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO)) log_err("fio: bs_unaligned may not work with raw io\n"); + + /* + * 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; + + if (td->nr_files > td->files_index) + td->nr_files = td->files_index; + + if (td->open_files > td->nr_files || !td->open_files) + td->open_files = td->nr_files; } /* @@ -517,21 +851,35 @@ static void fixup_options(struct thread_data *td) static char *to_kmg(unsigned int val) { char *buf = malloc(32); - char post[] = { 0, 'K', 'M', 'G', 'P', -1 }; + char post[] = { 0, 'K', 'M', 'G', 'P', 'E', 0 }; char *p = post; - while (*p != -1) { + do { if (val & 1023) break; val >>= 10; p++; - } + } while (*p); snprintf(buf, 31, "%u%c", val, *p); return buf; } +/* External engines are specified by "external:name.o") */ +static const char *get_engine_name(const char *str) +{ + char *p = strstr(str, ":"); + + if (!p) + return str; + + p++; + strip_blank_front(&p); + strip_blank_end(p); + return p; +} + /* * Adds a job to the list of things todo. Sanitizes the various options * to make sure we don't have conflicts, and initializes various @@ -539,11 +887,13 @@ static char *to_kmg(unsigned int val) */ 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; + const char *ddir_str[] = { NULL, "read", "write", "rw", NULL, + "randread", "randwrite", "randrw" }; + unsigned int i; struct fio_file *f; + const char *engine; + char fname[PATH_MAX]; + int numjobs; /* * the def_thread is just for options, it's not a real job @@ -551,99 +901,73 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) if (td == &def_thread) return 0; - /* - * Set default io engine, if none set - */ + engine = get_engine_name(td->ioengine); + td->io_ops = load_ioengine(td, engine); 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; - } - } - - if (td->odirect) - td->io_ops->flags |= FIO_RAWIO; - - fixup_options(td); - - td->filetype = FIO_TYPE_FILE; - if (!stat(jobname, &sb)) { - if (S_ISBLK(sb.st_mode)) - td->filetype = FIO_TYPE_BD; - else if (S_ISCHR(sb.st_mode)) - td->filetype = FIO_TYPE_CHAR; + log_err("fio: failed to load engine %s\n", engine); + return 1; } - if (td->filename) - td->nr_uniq_files = 1; + if (td->use_thread) + nr_thread++; else - td->nr_uniq_files = td->nr_files; - - if (td->filetype == FIO_TYPE_FILE || td->filename) { - char tmp[PATH_MAX]; - int len = 0; - int i; + nr_process++; - if (td->directory && td->directory[0] != '\0') - sprintf(tmp, "%s/", td->directory); + if (td->odirect) + td->io_ops->flags |= FIO_RAWIO; - td->files = malloc(sizeof(struct fio_file) * td->nr_files); + if (!td->filename && !td->files_index) { + td->filename = strdup(jobname); - 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); + if (td->nr_files == 1) + add_file(td, td->filename); + else { + for (i = 0; i < td->nr_files; i++) { + sprintf(fname, "%s.%d.%d", td->filename, td->thread_number, i); + add_file(td, fname); + } } - } 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); } + fixup_options(td); + for_each_file(td, f, i) { - f->file_size = td->total_file_size / td->nr_files; - f->file_offset = td->start_offset; + if (td->directory && f->filetype == FIO_TYPE_FILE) { + sprintf(fname, "%s/%s", td->directory, f->file_name); + f->file_name = strdup(fname); + } } - fio_sem_init(&td->mutex, 0); + td->mutex = fio_sem_init(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; + 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) + if ((td->stonewall || td->numjobs > 1) && prev_group_jobs) { + prev_group_jobs = 0; groupid++; + } td->groupid = groupid; + prev_group_jobs++; if (setup_rate(td)) 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) + if (!strcmp(td->io_ops->name, "cpuio")) fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle); else { char *c1, *c2, *c3, *c4; @@ -653,7 +977,7 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) 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=%d, bs=%s-%s/%s-%s, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, c1, c2, c3, c4, td->rate, td->io_ops->name, td->iodepth); + 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); @@ -682,6 +1006,12 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) if (add_job(td_new, jobname, job_add_num)) goto err; } + + if (td->numjobs > 1) { + groupid++; + prev_group_jobs = 0; + } + return 0; err: put_job(td); @@ -694,21 +1024,22 @@ err: */ int init_random_state(struct thread_data *td) { - unsigned long seeds[4]; - int fd, num_maps, blocks, i; + unsigned long seeds[5]; + int fd, num_maps, blocks; struct fio_file *f; + unsigned int i; - if (td->io_ops->flags & FIO_CPUIO) + if (td->io_ops->flags & FIO_DISKLESSIO) return 0; 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; } @@ -719,23 +1050,30 @@ 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; if (!td->norandommap) { for_each_file(td, f, i) { - blocks = (f->file_size + td->rw_min_bs - 1) / td->rw_min_bs; + 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)); + if (!f->file_map) { + log_err("fio: failed allocating random map. If running a large number of jobs, try the 'norandommap' option\n"); + return 1; + } 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; } @@ -760,6 +1098,8 @@ 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; } @@ -767,90 +1107,35 @@ static int is_empty_or_comment(char *line) return 1; } -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; - 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; - } - - log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n"); - return 1; -} - -static int str_verify_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_opt_postfix(const char *str) { - struct thread_data *td = data; + char *p = strstr(str, ":"); - 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 (!p) + return NULL; - log_err("fio: verify types: md5, crc32\n"); - return 1; + 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; - return 0; - } else if (!strncmp(mem, "mmap", 4)) { - td->mem_type = MEM_MMAP; - return 0; + if (td->mem_type == MEM_MMAPHUGE || td->mem_type == MEM_MMAP) { + td->mmapfile = get_opt_postfix(mem); + if (td->mem_type == MEM_MMAPHUGE && !td->mmapfile) { + log_err("fio: mmaphuge:/path/to/file\n"); + return 1; + } } - log_err("fio: mem type: malloc, shm, mmap\n"); - return 1; -} - -static int str_ioengine_cb(void *data, const char *str) -{ - 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, cpu\n"); - return 1; + return 0; } static int str_lockmem_cb(void fio_unused *data, unsigned long *val) @@ -891,10 +1176,74 @@ static int str_cpumask_cb(void *data, unsigned int *val) return 0; } +static int str_fst_cb(void *data, const char *str) +{ + struct thread_data *td = data; + char *nr = get_opt_postfix(str); + + td->file_service_nr = 1; + if (nr) + td->file_service_nr = atoi(nr); + + return 0; +} + +static int str_filename_cb(void *data, const char *input) +{ + struct thread_data *td = data; + char *fname, *str, *p; + + p = str = strdup(input); + + strip_blank_front(&str); + strip_blank_end(str); + + if (!td->files_index) + td->nr_files = 0; + + while ((fname = strsep(&str, ":")) != NULL) { + if (!strlen(fname)) + break; + add_file(td, fname); + td->nr_files++; + } + + free(p); + return 0; +} + +static int str_directory_cb(void *data, const char fio_unused *str) +{ + struct thread_data *td = data; + struct stat sb; + + 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; + } + + return 0; +} + +static int str_opendir_cb(void *data, const char fio_unused *str) +{ + struct thread_data *td = data; + + if (!td->files_index) + td->nr_files = 0; + + return add_dir_files(td, td->opendir); +} + /* * 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; @@ -989,39 +1338,14 @@ static int fill_def_thread(void) } /* - * fill globals + * fill default options */ - def_thread.ddir = DDIR_READ; - def_thread.iomix = 0; - def_thread.bs[DDIR_READ] = DEF_BS; - def_thread.bs[DDIR_WRITE] = DEF_BS; - def_thread.min_bs[DDIR_READ] = def_thread.min_bs[DDIR_WRITE] = 0; - def_thread.max_bs[DDIR_READ] = def_thread.max_bs[DDIR_WRITE] = 0; - 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; - def_thread.norandommap = DEF_NO_RAND_MAP; + #ifdef FIO_HAVE_DISK_UTIL def_thread.do_disk_util = 1; #endif @@ -1038,14 +1362,16 @@ 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); @@ -1070,6 +1396,9 @@ 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); @@ -1094,23 +1423,23 @@ static int parse_cmd_line(int argc, char *argv[]) } ret = parse_cmd_option(opt, val, options, td); - if (ret) { - log_err("fio: job dropped\n"); - put_job(td); - td = NULL; - } + 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) { @@ -1207,6 +1536,8 @@ int parse_options(int argc, char *argv[]) f_out = stdout; f_err = stderr; + options_init(options); + dupe_job_options(); if (setup_thread_area())