X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=init.c;h=925da6aea27519abe5bcd78c9e45324fe5c07e00;hp=6bf338532d6b528714274f7b243932b5459276cb;hb=313cb206ffc6e50c089314d322ebf1c523f37531;hpb=3b70d7e51e0b672a8b337c57c8faf865c0b7f415 diff --git a/init.c b/init.c index 6bf33853..925da6ae 100644 --- a/init.c +++ b/init.c @@ -1,3 +1,6 @@ +/* + * This file contains job initialization and setup functions. + */ #include #include #include @@ -5,13 +8,19 @@ #include #include #include +#include +#include #include #include #include #include #include "fio.h" +#include "parse.h" +/* + * The default options + */ #define DEF_BS (4096) #define DEF_TIMEOUT (0) #define DEF_RATE_CYCLE (1000) @@ -20,8 +29,7 @@ #define DEF_IO_ENGINE_NAME "sync" #define DEF_SEQUENTIAL (1) #define DEF_RAND_REPEAT (1) -#define DEF_OVERWRITE (1) -#define DEF_CREATE (1) +#define DEF_OVERWRITE (0) #define DEF_INVALIDATE (1) #define DEF_SYNCIO (0) #define DEF_RANDSEED (0xb1899bedUL) @@ -39,24 +47,394 @@ #define DEF_RWMIX_CYCLE (500) #define DEF_RWMIX_READ (50) #define DEF_NICE (0) +#define DEF_NR_FILES (1) +#define DEF_UNLINK (1) +#define DEF_WRITE_BW_LOG (0) +#define DEF_WRITE_LAT_LOG (0) +#define DEF_NO_RAND_MAP (0) +#define DEF_HUGEPAGE_SIZE FIO_HUGE_PAGE + +#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 *); -static char fio_version_string[] = "fio 1.4"; +/* + * Map of job/command line options + */ +static struct fio_option options[] = { + { + .name = "name", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(name), + }, + { + .name = "directory", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(directory), + }, + { + .name = "filename", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(filename), + }, + { + .name = "rw", + .type = FIO_OPT_STR, + .cb = str_rw_cb, + }, + { + .name = "ioengine", + .type = FIO_OPT_STR, + .cb = str_ioengine_cb, + }, + { + .name = "mem", + .type = FIO_OPT_STR, + .cb = str_mem_cb, + }, + { + .name = "verify", + .type = FIO_OPT_STR, + .cb = str_verify_cb, + }, + { + .name = "write_iolog", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(write_iolog_file), + }, + { + .name = "read_iolog", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(read_iolog_file), + }, + { + .name = "exec_prerun", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(exec_prerun), + }, + { + .name = "exec_postrun", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(exec_postrun), + }, +#ifdef FIO_HAVE_IOSCHED_SWITCH + { + .name = "ioscheduler", + .type = FIO_OPT_STR_STORE, + .off1 = td_var_offset(ioscheduler), + }, +#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), + }, + { + .name = "zoneskip", + .type = FIO_OPT_STR_VAL, + .off1 = td_var_offset(zone_skip), + }, + { + .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), + }, + { + .name = "rwmixcycle", + .type = FIO_OPT_INT, + .off1 = td_var_offset(rwmixcycle), + }, + { + .name = "rwmixread", + .type = FIO_OPT_INT, + .off1 = td_var_offset(rwmixread), + .max_val= 100, + }, + { + .name = "rwmixwrite", + .type = FIO_OPT_INT, + .off1 = td_var_offset(rwmixwrite), + .max_val= 100, + }, + { + .name = "nice", + .type = FIO_OPT_INT, + .off1 = td_var_offset(nice), + }, +#ifdef FIO_HAVE_IOPRIO + { + .name = "prio", + .type = FIO_OPT_INT, + .cb = str_prio_cb, + }, + { + .name = "prioclass", + .type = FIO_OPT_INT, + .cb = str_prioclass_cb, + }, +#endif + { + .name = "thinktime", + .type = FIO_OPT_INT, + .off1 = td_var_offset(thinktime) + }, + { + .name = "rate", + .type = FIO_OPT_INT, + .off1 = td_var_offset(rate) + }, + { + .name = "ratemin", + .type = FIO_OPT_INT, + .off1 = td_var_offset(ratemin) + }, + { + .name = "ratecycle", + .type = FIO_OPT_INT, + .off1 = td_var_offset(ratecycle) + }, + { + .name = "startdelay", + .type = FIO_OPT_INT, + .off1 = td_var_offset(start_delay) + }, + { + .name = "timeout", + .type = FIO_OPT_STR_VAL_TIME, + .off1 = td_var_offset(timeout) + }, + { + .name = "invalidate", + .type = FIO_OPT_INT, + .off1 = td_var_offset(invalidate_cache) + }, + { + .name = "sync", + .type = FIO_OPT_INT, + .off1 = td_var_offset(sync_io) + }, + { + .name = "bwavgtime", + .type = FIO_OPT_INT, + .off1 = td_var_offset(bw_avg_time) + }, + { + .name = "create_serialize", + .type = FIO_OPT_INT, + .off1 = td_var_offset(create_serialize) + }, + { + .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) + }, + { + .name = "cpuload", + .type = FIO_OPT_INT, + .off1 = td_var_offset(cpuload) + }, + { + .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) + }, +#ifdef FIO_HAVE_CPU_AFFINITY + { + .name = "cpumask", + .type = FIO_OPT_INT, + .cb = str_cpumask_cb, + }, +#endif + { + .name = "end_fsync", + .type = FIO_OPT_INT, + .off1 = td_var_offset(end_fsync) + }, + { + .name = "unlink", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(unlink), + }, + { + .name = "exitall", + .type = FIO_OPT_STR_SET, + .cb = str_exitall_cb, + }, + { + .name = "stonewall", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(stonewall), + }, + { + .name = "thread", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(thread), + }, + { + .name = "write_bw_log", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(write_bw_log), + }, + { + .name = "write_lat_log", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(write_lat_log), + }, + { + .name = "norandommap", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(norandommap), + }, + { + .name = "bs_unaligned", + .type = FIO_OPT_STR_SET, + .off1 = td_var_offset(bs_unaligned), + }, + { + .name = "hugepage-size", + .type = FIO_OPT_STR_VAL, + .off1 = td_var_offset(hugepage_size), + }, + { + .name = NULL, + }, +}; + +#define FIO_JOB_OPTS (sizeof(options) / sizeof(struct fio_option)) +#define FIO_CMD_OPTS (16) +#define FIO_GETOPT_JOB (0x89988998) -static int repeatable = DEF_RAND_REPEAT; -static char *ini_file; +/* + * 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] = { + { + .name = "output", + .has_arg = required_argument, + .val = 'o', + }, + { + .name = "timeout", + .has_arg = required_argument, + .val = 't', + }, + { + .name = "latency-log", + .has_arg = required_argument, + .val = 'l', + }, + { + .name = "bandwidth-log", + .has_arg = required_argument, + .val = 'b', + }, + { + .name = "minimal", + .has_arg = optional_argument, + .val = 'm', + }, + { + .name = "version", + .has_arg = no_argument, + .val = 'v', + }, + { + .name = NULL, + }, +}; + +static int def_timeout = DEF_TIMEOUT; + +static char fio_version_string[] = "fio 1.9"; + +static char **ini_file; static int max_jobs = MAX_JOBS; struct thread_data def_thread; struct thread_data *threads = NULL; -int rate_quit = 0; -int write_lat_log = 0; -int write_bw_log = 0; int exitall_on_terminate = 0; +int terse_output = 0; unsigned long long mlock_size = 0; FILE *f_out = NULL; FILE *f_err = NULL; +static int write_lat_log = DEF_WRITE_LAT_LOG; +static int write_bw_log = DEF_WRITE_BW_LOG; + +/* + * Return a free job structure. + */ static struct thread_data *get_new_job(int global, struct thread_data *parent) { struct thread_data *td; @@ -68,39 +446,115 @@ static struct thread_data *get_new_job(int global, struct thread_data *parent) td = &threads[thread_number++]; *td = *parent; - td->name[0] = '\0'; - td->fd = -1; td->thread_number = thread_number; - return td; } static void put_job(struct thread_data *td) { + if (td == &def_thread) + return; + memset(&threads[td->thread_number - 1], 0, sizeof(*td)); thread_number--; } -static int add_job(struct thread_data *td, const char *jobname, int job_add_num) +/* + * 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) { - char *ddir_str[] = { "read", "write", "randread", "randwrite", - "rw", NULL, "randrw" }; - struct stat sb; - int numjobs, ddir; + if (!td->rwmixread && td->rwmixwrite) + td->rwmixread = 100 - td->rwmixwrite; -#ifndef FIO_HAVE_LIBAIO - if (td->io_engine == FIO_LIBAIO) { - log_err("Linux libaio not available\n"); - return 1; + if (td->write_iolog_file && td->read_iolog_file) { + log_err("fio: read iolog overrides write_iolog\n"); + free(td->write_iolog_file); + td->write_iolog_file = NULL; } -#endif -#ifndef FIO_HAVE_POSIXAIO - if (td->io_engine == FIO_POSIXAIO) { - log_err("posix aio not available\n"); - return 1; + + if (td->io_ops->flags & FIO_SYNCIO) + td->iodepth = 1; + else { + if (!td->iodepth) + td->iodepth = td->nr_files; } -#endif + + /* + * 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->min_bs[DDIR_READ]) + td->min_bs[DDIR_READ]= td->bs[DDIR_READ]; + if (!td->max_bs[DDIR_READ]) + td->max_bs[DDIR_READ] = td->bs[DDIR_READ]; + if (!td->min_bs[DDIR_WRITE]) + td->min_bs[DDIR_WRITE]= td->bs[DDIR_WRITE]; + if (!td->max_bs[DDIR_WRITE]) + td->max_bs[DDIR_WRITE] = td->bs[DDIR_WRITE]; + + td->rw_min_bs = min(td->min_bs[DDIR_READ], td->min_bs[DDIR_WRITE]); + + if (td_read(td) && !td_rw(td)) + td->verify = 0; + + if (td->norandommap && td->verify != VERIFY_NONE) { + log_err("fio: norandommap given, verify disabled\n"); + td->verify = VERIFY_NONE; + } + if (td->bs_unaligned && (td->odirect || td->io_ops->flags & FIO_RAWIO)) + log_err("fio: bs_unaligned may not work with raw io\n"); + + /* + * O_DIRECT and char doesn't mix, clear that flag if necessary. + */ + if (td->filetype == FIO_TYPE_CHAR && td->odirect) + td->odirect = 0; +} + +/* + * This function leaks the buffer + */ +static char *to_kmg(unsigned int val) +{ + char *buf = malloc(32); + char post[] = { 0, 'K', 'M', 'G', 'P', 0 }; + char *p = post; + + do { + if (val & 1023) + break; + + val >>= 10; + p++; + } while (*p); + + snprintf(buf, 31, "%u%c", val, *p); + return buf; +} + +/* + * Adds a job to the list of things todo. Sanitizes the various options + * to make sure we don't have conflicts, and initializes various + * members of td. + */ +static int add_job(struct thread_data *td, const char *jobname, int job_add_num) +{ + const char *ddir_str[] = { "read", "write", "randread", "randwrite", + "rw", NULL, "randrw" }; + struct stat sb; + int numjobs, ddir, i; + struct fio_file *f; /* * the def_thread is just for options, it's not a real job @@ -108,18 +562,19 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) if (td == &def_thread) return 0; - if (td->io_engine & FIO_SYNCIO) - td->iodepth = 1; - else { - if (!td->iodepth) - td->iodepth = 1; - } - /* - * only really works for sequential io for now + * Set default io engine, if none set */ - if (td->zone_size && !td->sequential) - td->zone_size = 0; + 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; td->filetype = FIO_TYPE_FILE; if (!stat(jobname, &sb)) { @@ -129,30 +584,53 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) td->filetype = FIO_TYPE_CHAR; } - if (td->filetype == FIO_TYPE_FILE) { + fixup_options(td); + + if (td->filename) + td->nr_uniq_files = 1; + else + td->nr_uniq_files = td->nr_files; + + if (td->filetype == FIO_TYPE_FILE || td->filename) { char tmp[PATH_MAX]; + int len = 0; if (td->directory && td->directory[0] != '\0') - sprintf(tmp, "%s/%s.%d", td->directory, jobname, td->jobnum); - else - sprintf(tmp, "%s.%d", jobname, td->jobnum); - td->file_name = strdup(tmp); - } else - td->file_name = strdup(jobname); + sprintf(tmp, "%s/", td->directory); + + td->files = malloc(sizeof(struct fio_file) * td->nr_files); + for_each_file(td, f, i) { + memset(f, 0, sizeof(*f)); + f->fd = -1; + + if (td->filename) + sprintf(tmp + len, "%s", td->filename); + else + sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i); + f->file_name = strdup(tmp); + } + } else { + td->nr_files = 1; + td->files = malloc(sizeof(struct fio_file)); + f = &td->files[0]; + + memset(f, 0, sizeof(*f)); + f->fd = -1; + f->file_name = strdup(jobname); + } + + for_each_file(td, f, i) { + f->file_size = td->total_file_size / td->nr_files; + f->file_offset = td->start_offset; + } + fio_sem_init(&td->mutex, 0); td->clat_stat[0].min_val = td->clat_stat[1].min_val = ULONG_MAX; td->slat_stat[0].min_val = td->slat_stat[1].min_val = ULONG_MAX; td->bw_stat[0].min_val = td->bw_stat[1].min_val = ULONG_MAX; - if (td->min_bs == -1U) - td->min_bs = td->bs; - if (td->max_bs == -1U) - td->max_bs = td->bs; - if (td_read(td) && !td_rw(td)) - td->verify = 0; - if (td->stonewall && td->thread_number > 1) groupid++; @@ -161,22 +639,40 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) if (setup_rate(td)) goto err; - if (write_lat_log) { + if (td->write_lat_log) { setup_log(&td->slat_log); setup_log(&td->clat_log); } - if (write_bw_log) + if (td->write_bw_log) setup_log(&td->bw_log); - if (td->name[0] == '\0') - snprintf(td->name, sizeof(td->name)-1, "client%d", td->thread_number); + if (!td->name) + td->name = strdup(jobname); ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2); - if (!job_add_num) - fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_engine_name, td->iodepth); - else if (job_add_num == 1) - fprintf(f_out, "...\n"); + 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]); + + fprintf(f_out, "%s: (g=%d): rw=%s, odir=%u, bs=%s-%s/%s-%s, rate=%u, ioengine=%s, iodepth=%u\n", td->name, td->groupid, ddir_str[ddir], td->odirect, c1, c2, c3, c4, td->rate, td->io_ops->name, td->iodepth); + + free(c1); + free(c2); + free(c3); + free(c4); + } + } else if (job_add_num == 1) + fprintf(f_out, "...\n"); + } /* * recurse add identical jobs, clear numjobs and stonewall options @@ -191,7 +687,6 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num) td_new->numjobs = 1; td_new->stonewall = 0; - td_new->jobnum = numjobs; job_add_num = numjobs - 1; if (add_job(td_new, jobname, job_add_num)) @@ -203,10 +698,18 @@ err: 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) { unsigned long seeds[4]; - int fd, num_maps, blocks; + int fd, num_maps, blocks, i; + struct fio_file *f; + + if (td->io_ops->flags & FIO_CPUIO) + return 0; fd = open("/dev/urandom", O_RDONLY); if (fd == -1) { @@ -229,14 +732,18 @@ int init_random_state(struct thread_data *td) if (td->sequential) return 0; - if (repeatable) + if (td->rand_repeatable) seeds[3] = DEF_RANDSEED; - blocks = (td->io_size + td->min_bs - 1) / td->min_bs; - num_maps = blocks / BLOCKS_PER_MAP; - td->file_map = malloc(num_maps * sizeof(long)); - td->num_maps = num_maps; - memset(td->file_map, 0, num_maps * sizeof(long)); + if (!td->norandommap) { + for_each_file(td, f, i) { + blocks = (f->file_size + td->rw_min_bs - 1) / td->rw_min_bs; + num_maps = (blocks + BLOCKS_PER_MAP-1)/ BLOCKS_PER_MAP; + f->file_map = malloc(num_maps * sizeof(long)); + f->num_maps = num_maps; + memset(f->file_map, 0, num_maps * sizeof(long)); + } + } os_random_seed(seeds[3], &td->random_state); return 0; @@ -256,201 +763,6 @@ static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu) #endif } -static unsigned long get_mult(char c) -{ - switch (c) { - case 'k': - case 'K': - return 1024; - case 'm': - case 'M': - return 1024 * 1024; - case 'g': - case 'G': - return 1024 * 1024 * 1024; - default: - return 1; - } -} - -/* - * convert string after '=' into decimal value, noting any size suffix - */ -static int str_cnv(char *p, unsigned long long *val) -{ - char *str; - int len; - - str = strchr(p, '='); - if (!str) - return 1; - - str++; - len = strlen(str); - - *val = strtoul(str, NULL, 10); - if (*val == ULONG_MAX && errno == ERANGE) - return 1; - - *val *= get_mult(str[len - 1]); - return 0; -} - -static int check_strcnv(char *p, char *name, unsigned long long *val) -{ - if (strncmp(p, name, strlen(name) - 1)) - return 1; - - return str_cnv(p, val); -} - -static void strip_blank_front(char **p) -{ - char *s = *p; - - while (isspace(*s)) - s++; -} - -static void strip_blank_end(char *p) -{ - char *s = p + strlen(p) - 1; - - while (isspace(*s) || iscntrl(*s)) - s--; - - *(s + 1) = '\0'; -} - -typedef int (str_cb_fn)(struct thread_data *, char *); - -static int check_str(char *p, char *name, str_cb_fn *cb, struct thread_data *td) -{ - char *s; - - if (strncmp(p, name, strlen(name))) - return 1; - - s = strstr(p, name); - if (!s) - return 1; - - s = strchr(s, '='); - if (!s) - return 1; - - s++; - strip_blank_front(&s); - return cb(td, s); -} - -static int check_strstore(char *p, char *name, char *dest) -{ - char *s; - - if (strncmp(p, name, strlen(name))) - return 1; - - s = strstr(p, name); - if (!s) - return 1; - - s = strchr(p, '='); - if (!s) - return 1; - - s++; - strip_blank_front(&s); - - strcpy(dest, s); - return 0; -} - -static int __check_range(char *str, unsigned long *val) -{ - char suffix; - - if (sscanf(str, "%lu%c", val, &suffix) == 2) { - *val *= get_mult(suffix); - return 0; - } - - if (sscanf(str, "%lu", val) == 1) - return 0; - - return 1; -} - -static int check_range(char *p, char *name, unsigned long *s, unsigned long *e) -{ - char option[128]; - char *str, *p1, *p2; - - if (strncmp(p, name, strlen(name))) - return 1; - - strcpy(option, p); - p = option; - - str = strstr(p, name); - if (!str) - return 1; - - p += strlen(name); - - str = strchr(p, '='); - if (!str) - return 1; - - /* - * 'p' now holds whatever is after the '=' sign - */ - p1 = str + 1; - - /* - * terminate p1 at the '-' sign - */ - p = strchr(p1, '-'); - if (!p) - return 1; - - p2 = p + 1; - *p = '\0'; - - if (!__check_range(p1, s) && !__check_range(p2, e)) - return 0; - - return 1; -} - -static int check_int(char *p, char *name, unsigned int *val) -{ - char *str; - - if (strncmp(p, name, strlen(name))) - return 1; - - str = strstr(p, name); - if (!str) - return 1; - - str = strchr(p, '='); - if (!str) - return 1; - - str++; - - if (sscanf(str, "%u", val) == 1) - return 0; - - return 1; -} - -static int check_strset(char *p, char *name) -{ - return strncmp(p, name, strlen(name)); -} - static int is_empty_or_comment(char *line) { unsigned int i; @@ -465,8 +777,10 @@ static int is_empty_or_comment(char *line) return 1; } -static int str_rw_cb(struct thread_data *td, char *mem) +static int str_rw_cb(void *data, const char *mem) { + struct thread_data *td = data; + if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) { td->ddir = DDIR_READ; td->sequential = 1; @@ -484,12 +798,12 @@ static int str_rw_cb(struct thread_data *td, char *mem) td->sequential = 0; return 0; } else if (!strncmp(mem, "rw", 2)) { - td->ddir = 0; + td->ddir = DDIR_READ; td->iomix = 1; td->sequential = 1; return 0; } else if (!strncmp(mem, "randrw", 6)) { - td->ddir = 0; + td->ddir = DDIR_READ; td->iomix = 1; td->sequential = 0; return 0; @@ -499,8 +813,10 @@ static int str_rw_cb(struct thread_data *td, char *mem) return 1; } -static int str_verify_cb(struct thread_data *td, char *mem) +static int str_verify_cb(void *data, const char *mem) { + struct thread_data *td = data; + if (!strncmp(mem, "0", 1)) { td->verify = VERIFY_NONE; return 0; @@ -516,70 +832,134 @@ static int str_verify_cb(struct thread_data *td, char *mem) return 1; } -static int str_mem_cb(struct thread_data *td, char *mem) +/* + * Check if mmap/mmaphuge has a :/foo/bar/file at the end. If so, return that. + */ +static char *get_mmap_file(const char *str) +{ + char *p = strstr(str, ":"); + + if (!p) + return NULL; + + p++; + strip_blank_front(&p); + strip_blank_end(p); + return strdup(p); +} + +static int str_mem_cb(void *data, const char *mem) { + struct thread_data *td = data; + if (!strncmp(mem, "malloc", 6)) { td->mem_type = MEM_MALLOC; return 0; - } else if (!strncmp(mem, "shm", 3)) { - td->mem_type = MEM_SHM; + } else if (!strncmp(mem, "mmaphuge", 8)) { +#ifdef FIO_HAVE_HUGETLB + /* + * mmaphuge must be appended with the actual file + */ + td->mmapfile = get_mmap_file(mem); + if (!td->mmapfile) { + log_err("fio: mmaphuge:/path/to/file\n"); + return 1; + } + + td->mem_type = MEM_MMAPHUGE; return 0; +#else + log_err("fio: mmaphuge not available\n"); + return 1; +#endif } else if (!strncmp(mem, "mmap", 4)) { + /* + * Check if the user wants file backed memory. It's ok + * if there's no file given, we'll just use anon mamp then. + */ + td->mmapfile = get_mmap_file(mem); td->mem_type = MEM_MMAP; return 0; + } else if (!strncmp(mem, "shmhuge", 7)) { +#ifdef FIO_HAVE_HUGETLB + td->mem_type = MEM_SHMHUGE; + return 0; +#else + log_err("fio: shmhuge not available\n"); + return 1; +#endif + } else if (!strncmp(mem, "shm", 3)) { + td->mem_type = MEM_SHM; + return 0; } - log_err("fio: mem type: malloc, shm, mmap\n"); + log_err("fio: mem type: malloc, shm, shmhuge, mmap, mmaphuge\n"); return 1; } -static int str_ioengine_cb(struct thread_data *td, char *str) +static int str_ioengine_cb(void *data, const char *str) { - if (!strncmp(str, "linuxaio", 8) || !strncmp(str, "aio", 3) || - !strncmp(str, "libaio", 6)) { - strcpy(td->io_engine_name, "libaio"); - td->io_engine = FIO_LIBAIO; - return 0; - } else if (!strncmp(str, "posixaio", 8)) { - strcpy(td->io_engine_name, "posixaio"); - td->io_engine = FIO_POSIXAIO; - return 0; - } else if (!strncmp(str, "sync", 4)) { - strcpy(td->io_engine_name, "sync"); - td->io_engine = FIO_SYNCIO; - return 0; - } else if (!strncmp(str, "mmap", 4)) { - strcpy(td->io_engine_name, "mmap"); - td->io_engine = FIO_MMAPIO; - return 0; - } else if (!strncmp(str, "sgio", 4)) { - strcpy(td->io_engine_name, "sgio"); - td->io_engine = FIO_SGIO; - return 0; - } else if (!strncmp(str, "splice", 6)) { - strcpy(td->io_engine_name, "splice"); - td->io_engine = FIO_SPLICEIO; + struct thread_data *td = data; + + td->io_ops = load_ioengine(td, str); + if (td->io_ops) return 0; - } - log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice\n"); + log_err("fio: ioengine= libaio, posixaio, sync, mmap, sgio, splice, cpu, null\n"); + log_err("fio: or specify path to dynamic ioengine module\n"); return 1; } +static int str_lockmem_cb(void fio_unused *data, unsigned long *val) +{ + mlock_size = *val; + return 0; +} + +#ifdef FIO_HAVE_IOPRIO +static int str_prioclass_cb(void *data, unsigned int *val) +{ + struct thread_data *td = data; + + td->ioprio |= *val << IOPRIO_CLASS_SHIFT; + return 0; +} + +static int str_prio_cb(void *data, unsigned int *val) +{ + struct thread_data *td = data; + + td->ioprio |= *val; + return 0; +} +#endif + +static int str_exitall_cb(void) +{ + exitall_on_terminate = 1; + return 0; +} + +static int str_cpumask_cb(void *data, unsigned int *val) +{ + struct thread_data *td = data; + + fill_cpu_mask(td->cpumask, *val); + return 0; +} + /* * This is our [ini] type file parser. */ -int parse_jobs_ini(char *file) +static int parse_jobs_ini(char *file, int stonewall_flag) { - unsigned int prioclass, prio, cpu, global, il; - unsigned long long ull; - unsigned long ul1, ul2; + unsigned int global; struct thread_data *td; - char *string, *name, *tmpbuf; + char *string, *name; fpos_t off; FILE *f; char *p; - int ret = 0; + int ret = 0, stonewall; f = fopen(file, "r"); if (!f) { @@ -589,14 +969,16 @@ int parse_jobs_ini(char *file) string = malloc(4096); name = malloc(256); - tmpbuf = malloc(4096); + memset(name, 0, 256); - while ((p = fgets(string, 4096, f)) != NULL) { - if (ret) + stonewall = stonewall_flag; + do { + p = fgets(string, 4095, f); + if (!p) break; if (is_empty_or_comment(p)) continue; - if (sscanf(p, "[%s]", name) != 1) + if (sscanf(p, "[%255s]", name) != 1) continue; global = !strncmp(name, "global", 6); @@ -609,292 +991,47 @@ int parse_jobs_ini(char *file) break; } + /* + * Seperate multiple job files by a stonewall + */ + if (!global && stonewall) { + td->stonewall = stonewall; + stonewall = 0; + } + fgetpos(f, &off); while ((p = fgets(string, 4096, f)) != NULL) { if (is_empty_or_comment(p)) continue; - if (strstr(p, "[")) - break; + strip_blank_front(&p); - strip_blank_end(p); - if (!check_int(p, "prio", &prio)) { -#ifndef FIO_HAVE_IOPRIO - log_err("io priorities not available\n"); - ret = 1; - break; -#endif - td->ioprio |= prio; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "prioclass", &prioclass)) { -#ifndef FIO_HAVE_IOPRIO - log_err("io priorities not available\n"); - ret = 1; - break; -#else - td->ioprio |= prioclass << IOPRIO_CLASS_SHIFT; - fgetpos(f, &off); - continue; -#endif - } - if (!check_int(p, "direct", &il)) { - td->odirect = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "rate", &td->rate)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "ratemin", &td->ratemin)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "ratecycle", &td->ratecycle)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "thinktime", &td->thinktime)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "cpumask", &cpu)) { -#ifndef FIO_HAVE_CPU_AFFINITY - log_err("cpu affinity not available\n"); - ret = 1; - break; -#endif - fill_cpu_mask(td->cpumask, cpu); - fgetpos(f, &off); - continue; - } - if (!check_int(p, "fsync", &td->fsync_blocks)) { - fgetpos(f, &off); - td->end_fsync = 1; - continue; - } - if (!check_int(p, "startdelay", &td->start_delay)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "timeout", &td->timeout)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "invalidate", &il)) { - td->invalidate_cache = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "iodepth", &td->iodepth)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "sync", &il)) { - td->sync_io = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "bwavgtime", &td->bw_avg_time)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "create_serialize", &il)) { - td->create_serialize = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "create_fsync", &il)) { - td->create_fsync = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "end_fsync", &il)) { - td->end_fsync = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "loops", &td->loops)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "numjobs", &td->numjobs)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "overwrite", &il)) { - td->overwrite = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "rwmixcycle", &td->rwmixcycle)) { - fgetpos(f, &off); - continue; - } - if (!check_int(p, "rwmixread", &il)) { - if (il > 100) - il = 100; - td->rwmixread = il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "rwmixwrite", &il)) { - if (il > 100) - il = 100; - td->rwmixread = 100 - il; - fgetpos(f, &off); - continue; - } - if (!check_int(p, "nice", &td->nice)) { - fgetpos(f, &off); - continue; - } - if (!check_range(p, "bsrange", &ul1, &ul2)) { - if (ul1 > ul2) { - td->max_bs = ul1; - td->min_bs = ul2; - } else { - td->max_bs = ul2; - td->min_bs = ul1; - } - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "bs", &ull)) { - td->bs = ull; - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "size", &td->file_size)) { - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "offset", &td->file_offset)) { - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "zonesize", &td->zone_size)) { - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "zoneskip", &td->zone_skip)) { - fgetpos(f, &off); - continue; - } - if (!check_strcnv(p, "lockmem", &mlock_size)) { - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "directory", tmpbuf)) { - td->directory = strdup(tmpbuf); - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "name", tmpbuf)) { - snprintf(td->name, sizeof(td->name)-1, "%s%d", tmpbuf, td->thread_number); - fgetpos(f, &off); - continue; - } - if (!check_str(p, "mem", str_mem_cb, td)) { - fgetpos(f, &off); - continue; - } - if (!check_str(p, "verify", str_verify_cb, td)) { - fgetpos(f, &off); - continue; - } - if (!check_str(p, "rw", str_rw_cb, td)) { - fgetpos(f, &off); - continue; - } - if (!check_str(p, "ioengine", str_ioengine_cb, td)) { - fgetpos(f, &off); - continue; - } - if (!check_strset(p, "create")) { - td->create_file = 1; - fgetpos(f, &off); - continue; - } - if (!check_strset(p, "exitall")) { - exitall_on_terminate = 1; - fgetpos(f, &off); - continue; - } - if (!check_strset(p, "stonewall")) { - td->stonewall = 1; - fgetpos(f, &off); - continue; - } - if (!check_strset(p, "thread")) { - td->use_thread = 1; - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "iolog", tmpbuf)) { - if (td->write_iolog) { - log_err("fio: read iolog overrides given write_iolog\n"); - free(td->iolog_file); - td->write_iolog = 0; - } - td->iolog_file = strdup(tmpbuf); - td->read_iolog = 1; - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "write_iolog", tmpbuf)) { - if (!td->read_iolog) { - td->iolog_file = strdup(tmpbuf); - td->write_iolog = 1; - } else - log_err("fio: read iolog overrides given write_iolog\n"); - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "exec_prerun", tmpbuf)) { - td->exec_prerun = strdup(tmpbuf); - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "exec_postrun", tmpbuf)) { - td->exec_postrun = strdup(tmpbuf); - fgetpos(f, &off); - continue; - } - if (!check_strstore(p, "ioscheduler", tmpbuf)) { -#ifndef FIO_HAVE_IOSCHED_SWITCH - log_err("io scheduler switching not available\n"); - ret = 1; + if (p[0] == '[') break; -#else - td->ioscheduler = strdup(tmpbuf); - fgetpos(f, &off); - continue; -#endif - } + + strip_blank_end(p); + + fgetpos(f, &off); /* * Don't break here, continue parsing options so we * dump all the bad ones. Makes trial/error fixups * easier on the user. */ - printf("Client%d: bad option %s\n",td->thread_number,p); - ret = 1; + ret |= parse_option(p, options, td); } if (!ret) { fsetpos(f, &off); ret = add_job(td, name, 0); + } else { + log_err("fio: job %s dropped\n", name); + put_job(td); } - if (ret) - break; - } + } while (!ret); free(string); free(name); - free(tmpbuf); fclose(f); return ret; } @@ -913,16 +1050,14 @@ static int fill_def_thread(void) */ def_thread.ddir = DDIR_READ; def_thread.iomix = 0; - def_thread.bs = DEF_BS; - def_thread.min_bs = -1; - def_thread.max_bs = -1; - def_thread.io_engine = DEF_IO_ENGINE; - strcpy(def_thread.io_engine_name, DEF_IO_ENGINE_NAME); + def_thread.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; - def_thread.timeout = DEF_TIMEOUT; - def_thread.create_file = DEF_CREATE; + def_thread.timeout = def_timeout; def_thread.overwrite = DEF_OVERWRITE; def_thread.invalidate_cache = DEF_INVALIDATE; def_thread.sync_io = DEF_SYNCIO; @@ -938,6 +1073,13 @@ static int fill_def_thread(void) 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; + def_thread.hugepage_size = DEF_HUGEPAGE_SIZE; #ifdef FIO_HAVE_DISK_UTIL def_thread.do_disk_util = 1; #endif @@ -945,81 +1087,98 @@ static int fill_def_thread(void) return 0; } -static void usage(char *name) +static void usage(void) { printf("%s\n", fio_version_string); - printf("\t-s IO is sequential\n"); - printf("\t-b Block size in KiB for each IO\n"); - printf("\t-t Runtime in seconds\n"); - printf("\t-R Exit all threads on failure to meet rate goal\n"); - printf("\t-o Use O_DIRECT\n"); - printf("\t-l Generate per-job latency logs\n"); - printf("\t-w Generate per-job bandwidth logs\n"); - printf("\t-f Job file (Required)\n"); - printf("\t-v Print version info and exit\n"); + printf("\t--output\tWrite output to file\n"); + printf("\t--timeout\tRuntime in seconds\n"); + printf("\t--latency-log\tGenerate per-job latency logs\n"); + printf("\t--bandwidth-log\tGenerate per-job bandwidth logs\n"); + printf("\t--minimal\tMinimal (terse) output\n"); + printf("\t--version\tPrint version info and exit\n"); } -static void parse_cmd_line(int argc, char *argv[]) +static int parse_cmd_line(int argc, char *argv[]) { - int c; + struct thread_data *td = NULL; + int c, ini_idx = 0, lidx, ret; - while ((c = getopt(argc, argv, "s:b:t:r:R:o:f:lwvhO:")) != EOF) { + while ((c = getopt_long(argc, argv, "", long_options, &lidx)) != -1) { switch (c) { - case 's': - def_thread.sequential = !!atoi(optarg); - break; - case 'b': - def_thread.bs = atoi(optarg); - def_thread.bs <<= 10; - if (!def_thread.bs) { - printf("bad block size\n"); - def_thread.bs = DEF_BS; - } - break; - case 't': - def_thread.timeout = atoi(optarg); - break; - case 'r': - repeatable = !!atoi(optarg); - break; - case 'R': - rate_quit = !!atoi(optarg); - break; - case 'o': - def_thread.odirect = !!atoi(optarg); - break; - case 'f': - ini_file = strdup(optarg); - break; - case 'l': - write_lat_log = 1; - break; - case 'w': - write_bw_log = 1; - break; - case 'O': - f_out = fopen(optarg, "w+"); - if (!f_out) { - perror("fopen output"); - exit(1); + case 't': + def_timeout = atoi(optarg); + break; + case 'l': + write_lat_log = 1; + break; + case 'w': + write_bw_log = 1; + break; + case 'o': + f_out = fopen(optarg, "w+"); + if (!f_out) { + perror("fopen output"); + exit(1); + } + f_err = f_out; + break; + case 'm': + terse_output = 1; + break; + case 'h': + usage(); + exit(0); + case 'v': + printf("%s\n", fio_version_string); + exit(0); + case FIO_GETOPT_JOB: { + const char *opt = long_options[lidx].name; + char *val = optarg; + + if (!strncmp(opt, "name", 4) && td) { + ret = add_job(td, td->name ?: "fio", 0); + if (ret) { + put_job(td); + return 0; } - f_err = f_out; - break; - case 'h': - usage(argv[0]); - exit(0); - case 'v': - printf("%s\n", fio_version_string); - exit(0); + td = NULL; + } + if (!td) { + int global = !strncmp(val, "global", 6); + + td = get_new_job(global, &def_thread); + if (!td) + return 0; + } + + ret = parse_cmd_option(opt, val, options, td); + if (ret) { + log_err("fio: job dropped\n"); + put_job(td); + td = NULL; + } + break; + } + default: + printf("optarg <<%s>>\n", argv[optind]); + break; } } - if (!ini_file && argc > 1 && argv[argc - 1][0] != '-') - ini_file = strdup(argv[argc - 1]); - if (!f_out) { - f_out = stdout; - f_err = stderr; + if (td) { + ret = add_job(td, td->name ?: "fio", 0); + if (ret) + put_job(td); } + + while (optind < argc) { + ini_idx++; + ini_file = realloc(ini_file, ini_idx * sizeof(char *)); + ini_file[ini_idx - 1] = strdup(argv[optind]); + optind++; + } + + return ini_idx; } static void free_shm(void) @@ -1033,6 +1192,11 @@ static void free_shm(void) } } +/* + * 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) { /* @@ -1040,9 +1204,9 @@ static int setup_thread_area(void) * we get a failure that looks like too large a shm segment */ do { - int s = max_jobs * sizeof(struct thread_data); + size_t size = max_jobs * sizeof(struct thread_data); - shm_id = shmget(0, s, IPC_CREAT | 0600); + shm_id = shmget(0, size, IPC_CREAT | 0600); if (shm_id != -1) break; if (errno != EINVAL) { @@ -1066,23 +1230,64 @@ static int setup_thread_area(void) return 0; } +/* + * Copy the fio options into the long options map, so we mirror + * job and cmd line options. + */ +static void dupe_job_options(void) +{ + struct fio_option *o; + unsigned int i; + + i = 0; + while (long_options[i].name) + i++; + + o = &options[0]; + while (o->name) { + long_options[i].name = o->name; + long_options[i].val = FIO_GETOPT_JOB; + if (o->type == FIO_OPT_STR_SET) + long_options[i].has_arg = no_argument; + else + long_options[i].has_arg = required_argument; + + i++; + o++; + assert(i < FIO_JOB_OPTS + FIO_CMD_OPTS); + } +} + int parse_options(int argc, char *argv[]) { + int job_files, i; + + f_out = stdout; + f_err = stderr; + + dupe_job_options(); + if (setup_thread_area()) return 1; if (fill_def_thread()) return 1; - parse_cmd_line(argc, argv); + job_files = parse_cmd_line(argc, argv); - if (!ini_file) { - log_err("Need job file\n"); - usage(argv[0]); - return 1; + for (i = 0; i < job_files; i++) { + if (fill_def_thread()) + return 1; + if (parse_jobs_ini(ini_file[i], i)) + return 1; + free(ini_file[i]); } - if (parse_jobs_ini(ini_file)) + free(ini_file); + + if (!thread_number) { + log_err("No jobs defined(s)\n"); return 1; + } return 0; }