X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=init.c;h=b044e28903a24f16bb549f647fba4a2b144b0cd1;hp=f4125b13c5f05c1064d07d5e12bf72d5295a034e;hb=127f686514ef8b1d9ab6ed702971b205454974e6;hpb=a4f4fdd7c9e46a50bc33ecef44d9f06036580ad4 diff --git a/init.c b/init.c index f4125b13..b044e289 100644 --- a/init.c +++ b/init.c @@ -9,7 +9,6 @@ #include #include #include -#include #include #include #include @@ -18,464 +17,34 @@ #include "fio.h" #include "parse.h" +static char fio_version_string[] = "fio 1.14a"; + #define FIO_RANDSEED (0xb1899bedUL) -#define td_var_offset(var) ((size_t) &((struct thread_data *)0)->var) +static char **ini_file; +static int max_jobs = MAX_JOBS; -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 *); +struct thread_data def_thread; +struct thread_data *threads = NULL; -#define __stringify_1(x) #x -#define __stringify(x) __stringify_1(x) +int exitall_on_terminate = 0; +int terse_output = 0; +unsigned long long mlock_size = 0; +FILE *f_out = NULL; +FILE *f_err = NULL; -/* - * 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, - .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", "net", "syslet-rw" }, - }, - { - .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 (usec)", - .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 = "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(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, - }, -}; +int write_bw_log = 0; -#define FIO_JOB_OPTS (sizeof(options) / sizeof(struct fio_option)) -#define FIO_CMD_OPTS (16) -#define FIO_GETOPT_JOB (0x89988998) +static int def_timeout = 0; +static int write_lat_log = 0; + +static int prev_group_jobs; /* * Command line options. These will contain the above, plus a few * extra that only pertain to fio itself and not jobs. */ -static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = { +static struct option long_options[FIO_NR_OPTIONS] = { { .name = "output", .has_arg = required_argument, @@ -513,7 +82,7 @@ static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = { }, { .name = "cmdhelp", - .has_arg = required_argument, + .has_arg = optional_argument, .val = 'c', }, { @@ -521,24 +90,15 @@ static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = { }, }; -static int def_timeout = 0; - -static char fio_version_string[] = "fio 1.11"; - -static char **ini_file; -static int max_jobs = MAX_JOBS; - -struct thread_data def_thread; -struct thread_data *threads = NULL; - -int exitall_on_terminate = 0; -int terse_output = 0; -unsigned long long mlock_size = 0; -FILE *f_out = NULL; -FILE *f_err = NULL; +FILE *get_f_out() +{ + return f_out; +} -static int write_lat_log = 0; -int write_bw_log = 0; +FILE *get_f_err() +{ + return f_err; +} /* * Return a free job structure. @@ -565,17 +125,49 @@ static void put_job(struct thread_data *td) return; if (td->error) - fprintf(f_out, "fio: %s\n", td->verror); + log_info("fio: %s\n", td->verror); memset(&threads[td->thread_number - 1], 0, sizeof(*td)); thread_number--; } +static int setup_rate(struct thread_data *td) +{ + unsigned long nr_reads_per_msec; + unsigned long long rate; + unsigned int bs; + + if (!td->rate && !td->rate_iops) + return 0; + + if (td_rw(td)) + bs = td->rw_min_bs; + else if (td_read(td)) + bs = td->min_bs[DDIR_READ]; + else + bs = td->min_bs[DDIR_WRITE]; + + if (td->rate) { + rate = td->rate; + nr_reads_per_msec = (rate * 1024 * 1000LL) / bs; + } else + nr_reads_per_msec = td->rate_iops * 1000UL; + + if (!nr_reads_per_msec) { + log_err("rate lower than supported\n"); + return -1; + } + + td->rate_usec_cycle = 1000000000ULL / nr_reads_per_msec; + td->rate_pending_usleep = 0; + return 0; +} + /* * Lazy way of fixing up options that depend on each other. We could also * define option callback handlers, but this is easier. */ -static void fixup_options(struct thread_data *td) +static int fixup_options(struct thread_data *td) { if (!td->rwmixread && td->rwmixwrite) td->rwmixread = 100 - td->rwmixwrite; @@ -590,13 +182,13 @@ 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; /* @@ -616,6 +208,9 @@ static void fixup_options(struct thread_data *td) td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]); + if (!td->file_size_high) + td->file_size_high = td->file_size_low; + if (td_read(td) && !td_rw(td)) td->verify = 0; @@ -627,10 +222,48 @@ static void fixup_options(struct thread_data *td) log_err("fio: bs_unaligned may not work with raw io\n"); /* - * O_DIRECT and char doesn't mix, clear that flag if necessary. + * 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->filetype == FIO_TYPE_CHAR && td->odirect) - td->odirect = 0; + 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; + + if ((td->rate && td->rate_iops) || (td->ratemin && td->rate_iops_min)) { + log_err("fio: rate and rate_iops are mutually exclusive\n"); + return 1; + } + if ((td->rate < td->ratemin) || (td->rate_iops < td->rate_iops_min)) { + log_err("fio: minimum rate exceeds rate\n"); + return 1; + } + + return 0; } /* @@ -639,7 +272,7 @@ 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', 0 }; + char post[] = { 0, 'K', 'M', 'G', 'P', 'E', 0 }; char *p = post; do { @@ -654,183 +287,52 @@ static char *to_kmg(unsigned int val) 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) +/* External engines are specified by "external:name.o") */ +static const char *get_engine_name(const char *str) { - 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)) - td->filetype = FIO_TYPE_BD; - else if (S_ISCHR(sb.st_mode)) - td->filetype = FIO_TYPE_CHAR; - } - - 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') { - if (lstat(td->directory, &sb) < 0) { - log_err("fio: %s is not a directory\n", td->directory); - td_verror(td, errno); - 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); - - 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->stonewall && td->thread_number > 1) - groupid++; - - td->groupid = groupid; - - if (setup_rate(td)) - goto err; - - if (td->write_lat_log) { - setup_log(&td->slat_log); - setup_log(&td->clat_log); - } - if (td->write_bw_log) - setup_log(&td->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 { - 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]); + char *p = strstr(str, ":"); - fprintf(f_out, "%s: (g=%d): rw=%s, bs=%s-%s/%s-%s, ioengine=%s, iodepth=%u\n", td->name, td->groupid, ddir_str[ddir], c1, c2, c3, c4, td->io_ops->name, td->iodepth); + if (!p) + return str; - free(c1); - free(c2); - free(c3); - free(c4); - } - } else if (job_add_num == 1) - fprintf(f_out, "...\n"); - } + p++; + strip_blank_front(&p); + strip_blank_end(p); + return p; +} - /* - * recurse add identical jobs, clear numjobs and stonewall options - * as they don't apply to sub-jobs - */ - numjobs = td->numjobs; - while (--numjobs) { - struct thread_data *td_new = get_new_job(0, td); +static int exists_and_not_file(const char *filename) +{ + struct stat sb; - if (!td_new) - goto err; + if (lstat(filename, &sb) == -1) + return 0; - td_new->numjobs = 1; - td_new->stonewall = 0; - job_add_num = numjobs - 1; + if (S_ISREG(sb.st_mode)) + return 0; - if (add_job(td_new, jobname, job_add_num)) - goto err; - } - return 0; -err: - put_job(td); - return -1; + return 1; } /* * Initialize the various random states we need (random io, block size ranges, * read/write mix, etc). */ -int init_random_state(struct thread_data *td) +static int init_random_state(struct thread_data *td) { - unsigned long seeds[4]; - int fd, num_maps, blocks, i; + unsigned long seeds[6]; + int fd, num_maps, blocks; struct fio_file *f; - - if (td->io_ops->flags & FIO_CPUIO) - return 0; + unsigned int i; 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; } @@ -841,223 +343,200 @@ 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); + + os_random_seed(seeds[5], &td->file_size_state); + + if (!td_random(td)) return 0; if (td->rand_repeatable) - seeds[3] = FIO_RANDSEED; + seeds[4] = FIO_RANDSEED * td->thread_number; 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)); + 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; } -static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu) -{ -#ifdef FIO_HAVE_CPU_AFFINITY - unsigned int i; - CPU_ZERO(&cpumask); +/* + * 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[] = { NULL, "read", "write", "rw", NULL, + "randread", "randwrite", "randrw" }; + unsigned int i; + struct fio_file *f; + const char *engine; + char fname[PATH_MAX]; + int numjobs, file_alloced; + + /* + * the def_thread is just for options, it's not a real job + */ + if (td == &def_thread) + return 0; + + engine = get_engine_name(td->ioengine); + td->io_ops = load_ioengine(td, engine); + if (!td->io_ops) { + log_err("fio: failed to load engine %s\n", engine); + goto err; + } + + if (td->use_thread) + nr_thread++; + else + nr_process++; + + if (td->odirect) + td->io_ops->flags |= FIO_RAWIO; + + file_alloced = 0; + if (!td->filename && !td->files_index) { + file_alloced = 1; - for (i = 0; i < sizeof(int) * 8; i++) { - if ((1 << i) & cpu) - CPU_SET(i, &cpumask); + if (td->nr_files == 1 && exists_and_not_file(jobname)) + add_file(td, jobname); + else { + for (i = 0; i < td->nr_files; i++) { + sprintf(fname, "%s.%d.%d", jobname, td->thread_number, i); + add_file(td, fname); + } + } } -#endif -} -static int is_empty_or_comment(char *line) -{ - unsigned int i; + if (fixup_options(td)) + goto err; - for (i = 0; i < strlen(line); i++) { - if (line[i] == ';') - return 1; - if (!isspace(line[i]) && !iscntrl(line[i])) - return 0; + for_each_file(td, f, i) { + if (td->directory && f->filetype == FIO_TYPE_FILE) { + sprintf(fname, "%s/%s", td->directory, f->file_name); + f->file_name = strdup(fname); + } } + + td->mutex = fio_sem_init(0); - return 1; -} - -static int str_rw_cb(void *data, const char *mem) -{ - struct thread_data *td = data; + 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 (!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 = DDIR_READ; - td->iomix = 1; - td->sequential = 1; - return 0; - } else if (!strncmp(mem, "randrw", 6)) { - td->ddir = DDIR_READ; - td->iomix = 1; - td->sequential = 0; - return 0; + if ((td->stonewall || td->numjobs > 1) && prev_group_jobs) { + prev_group_jobs = 0; + groupid++; } - log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n"); - return 1; -} - -static int str_verify_cb(void *data, const char *mem) -{ - struct thread_data *td = data; + td->groupid = groupid; + prev_group_jobs++; - 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 (init_random_state(td)) + goto err; - log_err("fio: verify types: md5, crc32\n"); - return 1; -} + if (setup_rate(td)) + goto err; -/* - * 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 (td->write_lat_log) { + setup_log(&td->ts.slat_log); + setup_log(&td->ts.clat_log); + } + if (td->write_bw_log) + setup_log(&td->ts.bw_log); - if (!p) - return NULL; + if (!td->name) + td->name = strdup(jobname); - p++; - strip_blank_front(&p); - strip_blank_end(p); - return strdup(p); -} + if (!terse_output) { + if (!job_add_num) { + if (!strcmp(td->io_ops->name, "cpuio")) + log_info("%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle); + else { + char *c1, *c2, *c3, *c4; -static int str_mem_cb(void *data, const char *mem) -{ - struct thread_data *td = data; + 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]); - if (!strncmp(mem, "malloc", 6)) { - td->mem_type = MEM_MALLOC; - return 0; - } 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; - } + log_info("%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); - 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; + free(c1); + free(c2); + free(c3); + free(c4); + } + } else if (job_add_num == 1) + log_info("...\n"); } - log_err("fio: mem type: malloc, shm, shmhuge, mmap, mmaphuge\n"); - return 1; -} - -static int str_ioengine_cb(void *data, const char *str) -{ - struct thread_data *td = data; + /* + * recurse add identical jobs, clear numjobs and stonewall options + * as they don't apply to sub-jobs + */ + numjobs = td->numjobs; + while (--numjobs) { + struct thread_data *td_new = get_new_job(0, td); - td->io_ops = load_ioengine(td, str); - if (td->io_ops) - return 0; + if (!td_new) + goto err; - 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; -} + td_new->numjobs = 1; + td_new->stonewall = 0; -static int str_lockmem_cb(void fio_unused *data, unsigned long *val) -{ - mlock_size = *val; - return 0; -} + if (file_alloced) { + td_new->filename = NULL; + td_new->files_index = 0; + td_new->files = NULL; + } -#ifdef FIO_HAVE_IOPRIO -static int str_prioclass_cb(void *data, unsigned int *val) -{ - struct thread_data *td = data; + job_add_num = numjobs - 1; - td->ioprio |= *val << IOPRIO_CLASS_SHIFT; - return 0; -} + if (add_job(td_new, jobname, job_add_num)) + goto err; + } -static int str_prio_cb(void *data, unsigned int *val) -{ - struct thread_data *td = data; + if (td->numjobs > 1) { + groupid++; + prev_group_jobs = 0; + } - td->ioprio |= *val; return 0; +err: + put_job(td); + return -1; } -#endif -static int str_exitall_cb(void) +static int is_empty_or_comment(char *line) { - exitall_on_terminate = 1; - return 0; -} + unsigned int i; -static int str_cpumask_cb(void *data, unsigned int *val) -{ - struct thread_data *td = data; + 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; + } - fill_cpu_mask(td->cpumask, *val); - return 0; + return 1; } /* @@ -1130,7 +609,7 @@ static 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 |= fio_option_parse(td, p); } if (!ret) { @@ -1160,7 +639,7 @@ static int fill_def_thread(void) /* * fill default options */ - fill_default_options(&def_thread, options); + fio_fill_default_options(&def_thread); def_thread.timeout = def_timeout; def_thread.write_bw_log = write_bw_log; @@ -1173,6 +652,55 @@ static int fill_def_thread(void) return 0; } +static void free_shm(void) +{ + struct shmid_ds sbuf; + + if (threads) { + shmdt((void *) threads); + threads = NULL; + shmctl(shm_id, IPC_RMID, &sbuf); + } +} + +/* + * The thread area is shared between the main process and the job + * threads/processes. So setup a shared memory segment that will hold + * all the job info. + */ +static int setup_thread_area(void) +{ + /* + * 1024 is too much on some machines, scale max_jobs if + * we get a failure that looks like too large a shm segment + */ + do { + size_t size = max_jobs * sizeof(struct thread_data); + + shm_id = shmget(0, size, IPC_CREAT | 0600); + if (shm_id != -1) + break; + if (errno != EINVAL) { + perror("shmget"); + break; + } + + max_jobs >>= 1; + } while (max_jobs); + + if (shm_id == -1) + return 1; + + threads = shmat(shm_id, NULL, 0); + if (threads == (void *) -1) { + perror("shmat"); + return 1; + } + + atexit(free_shm); + return 0; +} + static void usage(void) { printf("%s\n", fio_version_string); @@ -1189,9 +717,9 @@ static void usage(void) 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); @@ -1217,8 +745,7 @@ static int parse_cmd_line(int argc, char *argv[]) usage(); exit(0); case 'c': - ret = show_cmd_help(options, optarg); - exit(ret); + exit(fio_show_option_help(optarg)); case 'v': printf("%s\n", fio_version_string); exit(0); @@ -1242,12 +769,9 @@ static int parse_cmd_line(int argc, char *argv[]) return 0; } - ret = parse_cmd_option(opt, val, options, td); - if (ret) { - log_err("fio: job dropped\n"); - put_job(td); - td = NULL; - } + ret = fio_cmd_option_parse(td, opt, val); + if (ret) + dont_add_job = 1; break; } default: @@ -1256,9 +780,13 @@ static int parse_cmd_line(int argc, char *argv[]) } 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) { @@ -1271,82 +799,6 @@ static int parse_cmd_line(int argc, char *argv[]) return ini_idx; } -static void free_shm(void) -{ - struct shmid_ds sbuf; - - if (threads) { - shmdt((void *) threads); - threads = NULL; - shmctl(shm_id, IPC_RMID, &sbuf); - } -} - -/* - * The thread area is shared between the main process and the job - * threads/processes. So setup a shared memory segment that will hold - * all the job info. - */ -static int setup_thread_area(void) -{ - /* - * 1024 is too much on some machines, scale max_jobs if - * we get a failure that looks like too large a shm segment - */ - do { - size_t size = max_jobs * sizeof(struct thread_data); - - shm_id = shmget(0, size, IPC_CREAT | 0600); - if (shm_id != -1) - break; - if (errno != EINVAL) { - perror("shmget"); - break; - } - - max_jobs >>= 1; - } while (max_jobs); - - if (shm_id == -1) - return 1; - - threads = shmat(shm_id, NULL, 0); - if (threads == (void *) -1) { - perror("shmat"); - return 1; - } - - atexit(free_shm); - 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[]) { @@ -1355,9 +807,7 @@ int parse_options(int argc, char *argv[]) f_out = stdout; f_err = stderr; - options_init(options); - - dupe_job_options(); + fio_options_dup_and_init(long_options); if (setup_thread_area()) return 1;