[PATCH] Support for mmap of hugetlb files as memory backing
[fio.git] / init.c
diff --git a/init.c b/init.c
index 732cceb993755a7ac2c5b6f3672c13c5950664b8..edaec144f3aaaa5895339df4f2a703bdff7d9c48 100644 (file)
--- a/init.c
+++ b/init.c
@@ -29,7 +29,7 @@
 #define DEF_IO_ENGINE_NAME     "sync"
 #define DEF_SEQUENTIAL         (1)
 #define DEF_RAND_REPEAT                (1)
-#define DEF_OVERWRITE          (1)
+#define DEF_OVERWRITE          (0)
 #define DEF_INVALIDATE         (1)
 #define DEF_SYNCIO             (0)
 #define DEF_RANDSEED           (0xb1899bedUL)
 #define DEF_RWMIX_READ         (50)
 #define DEF_NICE               (0)
 #define DEF_NR_FILES           (1)
-#define DEF_UNLINK             (0)
+#define DEF_UNLINK             (1)
 #define DEF_WRITE_BW_LOG       (0)
 #define DEF_WRITE_LAT_LOG      (0)
+#define DEF_NO_RAND_MAP                (0)
 
 #define td_var_offset(var)     ((size_t) &((struct thread_data *)0)->var)
 
@@ -59,8 +60,10 @@ 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 *);
 
@@ -137,8 +140,9 @@ static struct fio_option options[] = {
        },
        {
                .name   = "bs",
-               .type   = FIO_OPT_STR_VAL,
-               .off1   = td_var_offset(bs),
+               .type   = FIO_OPT_STR_VAL_INT,
+               .off1   = td_var_offset(bs[DDIR_READ]),
+               .off2   = td_var_offset(bs[DDIR_WRITE]),
        },
        {
                .name   = "offset",
@@ -163,8 +167,10 @@ static struct fio_option options[] = {
        {
                .name   = "bsrange",
                .type   = FIO_OPT_RANGE,
-               .off1   = td_var_offset(min_bs),
-               .off2   = td_var_offset(max_bs),
+               .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",
@@ -342,6 +348,16 @@ static struct fio_option options[] = {
                .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 = NULL,
        },
@@ -393,7 +409,7 @@ static struct option long_options[FIO_JOB_OPTS + FIO_CMD_OPTS] = {
 
 static int def_timeout = DEF_TIMEOUT;
 
-static char fio_version_string[] = "fio 1.7";
+static char fio_version_string[] = "fio 1.9";
 
 static char **ini_file;
 static int max_jobs = MAX_JOBS;
@@ -401,7 +417,6 @@ 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;
@@ -473,12 +488,53 @@ static void fixup_options(struct thread_data *td)
        if (td_read(td) || td_rw(td))
                td->overwrite = 1;
 
-       if (!td->min_bs)
-               td->min_bs = td->bs;
-       if (!td->max_bs)
-               td->max_bs = td->bs;
+       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;
 }
 
 /*
@@ -488,25 +544,12 @@ static void fixup_options(struct thread_data *td)
  */
 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
 {
-       char *ddir_str[] = { "read", "write", "randread", "randwrite",
-                            "rw", NULL, "randrw" };
+       const char *ddir_str[] = { "read", "write", "randread", "randwrite",
+                                  "rw", NULL, "randrw" };
        struct stat sb;
        int numjobs, ddir, i;
        struct fio_file *f;
 
-#ifndef FIO_HAVE_LIBAIO
-       if (td->io_engine == FIO_LIBAIO) {
-               log_err("Linux libaio not available\n");
-               return 1;
-       }
-#endif
-#ifndef FIO_HAVE_POSIXAIO
-       if (td->io_engine == FIO_POSIXAIO) {
-               log_err("posix aio not available\n");
-               return 1;
-       }
-#endif
-
        /*
         * the def_thread is just for options, it's not a real job
         */
@@ -524,7 +567,8 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
                }
        }
 
-       fixup_options(td);
+       if (td->odirect)
+               td->io_ops->flags |= FIO_RAWIO;
 
        td->filetype = FIO_TYPE_FILE;
        if (!stat(jobname, &sb)) {
@@ -534,8 +578,7 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
                        td->filetype = FIO_TYPE_CHAR;
        }
 
-       if (td->odirect)
-               td->io_ops->flags |= FIO_RAWIO;
+       fixup_options(td);
 
        if (td->filename)
                td->nr_uniq_files = 1;
@@ -545,7 +588,6 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
        if (td->filetype == FIO_TYPE_FILE || td->filename) {
                char tmp[PATH_MAX];
                int len = 0;
-               int i;
 
                if (td->directory && td->directory[0] != '\0')
                        sprintf(tmp, "%s/", td->directory);
@@ -607,8 +649,21 @@ static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
                if (!job_add_num) {
                        if (td->io_ops->flags & FIO_CPUIO)
                                fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
-                       else
-                               fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_ops->name, td->iodepth);
+                       else {
+                               char *c1, *c2, *c3, *c4;
+
+                               c1 = to_kmg(td->min_bs[DDIR_READ]);
+                               c2 = to_kmg(td->max_bs[DDIR_READ]);
+                               c3 = to_kmg(td->min_bs[DDIR_WRITE]);
+                               c4 = to_kmg(td->max_bs[DDIR_WRITE]);
+
+                               fprintf(f_out, "%s: (g=%d): rw=%s, 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");
        }
@@ -674,12 +729,14 @@ int init_random_state(struct thread_data *td)
        if (td->rand_repeatable)
                seeds[3] = DEF_RANDSEED;
 
-       for_each_file(td, f, i) {
-               blocks = (f->file_size + td->min_bs - 1) / td->min_bs;
-               num_maps = blocks / BLOCKS_PER_MAP;
-               f->file_map = malloc(num_maps * sizeof(long));
-               f->num_maps = num_maps;
-               memset(f->file_map, 0, num_maps * sizeof(long));
+       if (!td->norandommap) {
+               for_each_file(td, f, i) {
+                       blocks = (f->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);
@@ -735,12 +792,12 @@ static int str_rw_cb(void *data, const 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;
@@ -776,15 +833,46 @@ static int str_mem_cb(void *data, const char *mem)
        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
+               char *hugefile;
+
+               /*
+                * mmaphuge must be appended with the actual file
+                */
+               hugefile = strstr(mem, ":");
+               if (!hugefile) {
+                       log_err("fio: mmaphuge:/path/to/file\n");
+                       return 1;
+               }
+
+               hugefile++;
+               strip_blank_front(&hugefile);
+               strip_blank_end(hugefile);
+               td->hugefile = strdup(hugefile);
+               td->mem_type = MEM_MMAPHUGE;
                return 0;
+#else
+               log_err("fio: mmaphuge not available\n");
+               return 1;
+#endif
        } else if (!strncmp(mem, "mmap", 4)) {
                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;
 }
 
@@ -796,7 +884,8 @@ static int str_ioengine_cb(void *data, const char *str)
        if (td->io_ops)
                return 0;
 
-       log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice, cpu\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;
 }
 
@@ -806,6 +895,7 @@ static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
        return 0;
 }
 
+#ifdef FIO_HAVE_IOPRIO
 static int str_prioclass_cb(void *data, unsigned int *val)
 {
        struct thread_data *td = data;
@@ -821,6 +911,7 @@ static int str_prio_cb(void *data, unsigned int *val)
        td->ioprio |= *val;
        return 0;
 }
+#endif
 
 static int str_exitall_cb(void)
 {
@@ -839,7 +930,7 @@ static int str_cpumask_cb(void *data, unsigned int *val)
 /*
  * This is our [ini] type file parser.
  */
-int parse_jobs_ini(char *file, int stonewall_flag)
+static int parse_jobs_ini(char *file, int stonewall_flag)
 {
        unsigned int global;
        struct thread_data *td;
@@ -860,8 +951,9 @@ int parse_jobs_ini(char *file, int stonewall_flag)
        memset(name, 0, 256);
 
        stonewall = stonewall_flag;
-       while ((p = fgets(string, 4096, f)) != NULL) {
-               if (ret)
+       do {
+               p = fgets(string, 4095, f);
+               if (!p)
                        break;
                if (is_empty_or_comment(p))
                        continue;
@@ -890,10 +982,12 @@ int parse_jobs_ini(char *file, int stonewall_flag)
                while ((p = fgets(string, 4096, f)) != NULL) {
                        if (is_empty_or_comment(p))
                                continue;
-                       if (strstr(p, "["))
-                               break;
 
                        strip_blank_front(&p);
+
+                       if (p[0] == '[')
+                               break;
+
                        strip_blank_end(p);
 
                        fgetpos(f, &off);
@@ -903,16 +997,17 @@ int parse_jobs_ini(char *file, int stonewall_flag)
                         * dump all the bad ones. Makes trial/error fixups
                         * easier on the user.
                         */
-                       ret = parse_option(p, options, td);
+                       ret |= parse_option(p, options, td);
                }
 
                if (!ret) {
                        fsetpos(f, &off);
                        ret = add_job(td, name, 0);
+               } else {
+                       log_err("fio: job %s dropped\n", name);
+                       put_job(td);
                }
-               if (ret)
-                       break;
-       }
+       } while (!ret);
 
        free(string);
        free(name);
@@ -934,9 +1029,10 @@ static int fill_def_thread(void)
         */
        def_thread.ddir = DDIR_READ;
        def_thread.iomix = 0;
-       def_thread.bs = DEF_BS;
-       def_thread.min_bs = 0;
-       def_thread.max_bs = 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;
@@ -961,6 +1057,7 @@ static int fill_def_thread(void)
        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
@@ -1032,7 +1129,12 @@ static int parse_cmd_line(int argc, char *argv[])
                                        return 0;
                        }
 
-                       parse_cmd_option(opt, val, options, td);
+                       ret = parse_cmd_option(opt, val, options, td);
+                       if (ret) {
+                               log_err("fio: job dropped\n");
+                               put_job(td);
+                               td = NULL;
+                       }
                        break;
                }
                default:
@@ -1162,7 +1264,6 @@ int parse_options(int argc, char *argv[])
 
        if (!thread_number) {
                log_err("No jobs defined(s)\n");
-               usage();
                return 1;
        }