Merge branch 'master' into gfio
[fio.git] / filesetup.c
index d05050647ddbbcc78b1b22d59c91e8f7444d7b43..3054d9dab4803271de2bbee3369ac9ef66d1133d 100644 (file)
 #include "smalloc.h"
 #include "filehash.h"
 #include "os/os.h"
+#include "hash.h"
+#include "lib/axmap.h"
+
+#ifdef CONFIG_LINUX_FALLOCATE
+#include <linux/falloc.h>
+#endif
 
 static int root_warn;
 
@@ -66,18 +72,42 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
                return 1;
        }
 
-#ifdef FIO_HAVE_FALLOCATE
-       if (td->o.fallocate && !td->o.fill_device) {
-               dprint(FD_FILE, "fallocate file %s size %llu\n", f->file_name,
-                                                       f->real_file_size);
+#ifdef CONFIG_POSIX_FALLOCATE
+       if (!td->o.fill_device) {
+               switch (td->o.fallocate_mode) {
+               case FIO_FALLOCATE_NONE:
+                       break;
+               case FIO_FALLOCATE_POSIX:
+                       dprint(FD_FILE, "posix_fallocate file %s size %llu\n",
+                                f->file_name, f->real_file_size);
+
+                       r = posix_fallocate(f->fd, 0, f->real_file_size);
+                       if (r > 0) {
+                               log_err("fio: posix_fallocate fails: %s\n",
+                                               strerror(r));
+                       }
+                       break;
+#ifdef CONFIG_LINUX_FALLOCATE
+               case FIO_FALLOCATE_KEEP_SIZE:
+                       dprint(FD_FILE,
+                               "fallocate(FALLOC_FL_KEEP_SIZE) "
+                               "file %s size %llu\n",
+                               f->file_name, f->real_file_size);
+
+                       r = fallocate(f->fd, FALLOC_FL_KEEP_SIZE, 0,
+                                       f->real_file_size);
+                       if (r != 0)
+                               td_verror(td, errno, "fallocate");
 
-               r = posix_fallocate(f->fd, 0, f->real_file_size);
-               if (r > 0) {
-                       log_err("fio: posix_fallocate fails: %s\n",
-                                       strerror(r));
+                       break;
+#endif /* CONFIG_LINUX_FALLOCATE */
+               default:
+                       log_err("fio: unknown fallocate mode: %d\n",
+                               td->o.fallocate_mode);
+                       assert(0);
                }
        }
-#endif
+#endif /* CONFIG_POSIX_FALLOCATE */
 
        if (!new_layout)
                goto done;
@@ -208,11 +238,18 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
 static unsigned long long get_rand_file_size(struct thread_data *td)
 {
        unsigned long long ret, sized;
-       long r;
+       unsigned long r;
+
+       if (td->o.use_os_rand) {
+               r = os_random_long(&td->file_size_state);
+               sized = td->o.file_size_high - td->o.file_size_low;
+               ret = (unsigned long long) ((double) sized * (r / (OS_RAND_MAX + 1.0)));
+       } else {
+               r = __rand(&td->__file_size_state);
+               sized = td->o.file_size_high - td->o.file_size_low;
+               ret = (unsigned long long) ((double) sized * (r / (FRAND_MAX + 1.0)));
+       }
 
-       r = os_random_long(&td->file_size_state);
-       sized = td->o.file_size_high - td->o.file_size_low;
-       ret = (unsigned long long) ((double) sized * (r / (OS_RAND_MAX + 1.0)));
        ret += td->o.file_size_low;
        ret -= (ret % td->o.rw_min_bs);
        return ret;
@@ -245,7 +282,6 @@ static int bdev_size(struct thread_data *td, struct fio_file *f)
        r = blockdev_size(f, &bytes);
        if (r) {
                td_verror(td, r, "blockdev_size");
-               printf("fd is %d\n", f->fd);
                goto err;
        }
 
@@ -398,10 +434,16 @@ int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f)
                ret = errno;
 
        f->fd = -1;
+
+       if (f->shadow_fd != -1) {
+               close(f->shadow_fd);
+               f->shadow_fd = -1;
+       }
+
        return ret;
 }
 
-static int file_lookup_open(struct fio_file *f, int flags)
+int file_lookup_open(struct fio_file *f, int flags)
 {
        struct fio_file *__f;
        int from_hash;
@@ -426,6 +468,24 @@ static int file_lookup_open(struct fio_file *f, int flags)
        return from_hash;
 }
 
+static int file_close_shadow_fds(struct thread_data *td)
+{
+       struct fio_file *f;
+       int num_closed = 0;
+       unsigned int i;
+
+       for_each_file(td, f, i) {
+               if (f->shadow_fd == -1)
+                       continue;
+
+               close(f->shadow_fd);
+               f->shadow_fd = -1;
+               num_closed++;
+       }
+
+       return num_closed;
+}
+
 int generic_open_file(struct thread_data *td, struct fio_file *f)
 {
        int is_std = 0;
@@ -434,6 +494,11 @@ int generic_open_file(struct thread_data *td, struct fio_file *f)
 
        dprint(FD_FILE, "fd open %s\n", f->file_name);
 
+       if (td_trim(td) && f->filetype != FIO_TYPE_BD) {
+               log_err("fio: trim only applies to block device\n");
+               return 1;
+       }
+
        if (!strcmp(f->file_name, "-")) {
                if (td_rw(td)) {
                        log_err("fio: can't read/write to stdin/out\n");
@@ -448,14 +513,17 @@ int generic_open_file(struct thread_data *td, struct fio_file *f)
                        f_out = stderr;
        }
 
+       if (td_trim(td))
+               goto skip_flags;
        if (td->o.odirect)
                flags |= OS_O_DIRECT;
        if (td->o.sync_io)
                flags |= O_SYNC;
-       if (f->filetype != FIO_TYPE_FILE)
-               flags |= FIO_O_NOATIME;
        if (td->o.create_on_open)
                flags |= O_CREAT;
+skip_flags:
+       if (f->filetype != FIO_TYPE_FILE)
+               flags |= FIO_O_NOATIME;
 
 open_again:
        if (td_write(td)) {
@@ -469,7 +537,7 @@ open_again:
                        f->fd = dup(STDOUT_FILENO);
                else
                        from_hash = file_lookup_open(f, flags);
-       } else {
+       } else if (td_read(td)) {
                if (f->filetype == FIO_TYPE_CHAR && !read_only)
                        flags |= O_RDWR;
                else
@@ -479,6 +547,9 @@ open_again:
                        f->fd = dup(STDIN_FILENO);
                else
                        from_hash = file_lookup_open(f, flags);
+       } else { //td trim
+               flags |= O_RDWR;
+               from_hash = file_lookup_open(f, flags);
        }
 
        if (f->fd == -1) {
@@ -489,20 +560,40 @@ open_again:
                        flags &= ~FIO_O_NOATIME;
                        goto open_again;
                }
+               if (__e == EMFILE && file_close_shadow_fds(td))
+                       goto open_again;
 
-               snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name);
+               snprintf(buf, sizeof(buf), "open(%s)", f->file_name);
+
+               if (__e == EINVAL && (flags & OS_O_DIRECT)) {
+                       log_err("fio: looks like your file system does not " \
+                               "support direct=1/buffered=0\n");
+               }
 
                td_verror(td, __e, buf);
        }
 
        if (!from_hash && f->fd != -1) {
                if (add_file_hash(f)) {
-                       int ret;
+                       int fio_unused ret;
 
                        /*
-                        * OK to ignore, we haven't done anything with it
+                        * Stash away descriptor for later close. This is to
+                        * work-around a "feature" on Linux, where a close of
+                        * an fd that has been opened for write will trigger
+                        * udev to call blkid to check partitions, fs id, etc.
+                        * That polutes the device cache, which can slow down
+                        * unbuffered accesses.
                         */
-                       ret = generic_close_file(td, f);
+                       if (f->shadow_fd == -1)
+                               f->shadow_fd = f->fd;
+                       else {
+                               /*
+                                * OK to ignore, we haven't done anything
+                                * with it
+                                */
+                               ret = generic_close_file(td, f);
+                       }
                        goto open_again;
                }
        }
@@ -618,6 +709,12 @@ static unsigned long long get_fs_free_counts(struct thread_data *td)
        return ret;
 }
 
+uint64_t get_start_offset(struct thread_data *td)
+{
+       return td->o.start_offset +
+               (td->thread_number - 1) * td->o.offset_increment;
+}
+
 /*
  * Open the files and setup files sizes, creating files if necessary.
  */
@@ -631,7 +728,7 @@ int setup_files(struct thread_data *td)
        dprint(FD_FILE, "setup files\n");
 
        if (td->o.read_iolog_file)
-               return 0;
+               goto done;
 
        /*
         * if ioengine defines a setup() method, it's responsible for
@@ -679,7 +776,7 @@ int setup_files(struct thread_data *td)
        extend_size = total_size = 0;
        need_extend = 0;
        for_each_file(td, f, i) {
-               f->file_offset = td->o.start_offset;
+               f->file_offset = get_start_offset(td);
 
                if (!td->o.file_size_low) {
                        /*
@@ -710,8 +807,11 @@ int setup_files(struct thread_data *td)
 
                if (f->io_size == -1ULL)
                        total_size = -1ULL;
-               else
+               else {
+                        if (td->o.size_percent)
+                                f->io_size = (f->io_size * td->o.size_percent) / 100;
                        total_size += f->io_size;
+               }
 
                if (f->filetype == FIO_TYPE_FILE &&
                    (f->io_size + f->file_offset) > f->real_file_size &&
@@ -733,7 +833,7 @@ int setup_files(struct thread_data *td)
         */
        if (need_extend) {
                temp_stall_ts = 1;
-               if (!terse_output)
+               if (output_format == FIO_OUTPUT_NORMAL)
                        log_info("%s: Laying out IO file(s) (%u file(s) /"
                                 " %lluMB)\n", td->o.name, need_extend,
                                        extend_size >> 20);
@@ -778,6 +878,11 @@ int setup_files(struct thread_data *td)
         */
        if (!td->o.read_iolog_file)
                td->total_io_size = td->o.size * td->o.loops;
+
+done:
+       if (td->o.create_only)
+               td->done = 1;
+
        return 0;
 err_offset:
        log_err("%s: you need to specify valid offset=\n", td->o.name);
@@ -798,25 +903,77 @@ int pre_read_files(struct thread_data *td)
        return 1;
 }
 
+static int __init_rand_distribution(struct thread_data *td, struct fio_file *f)
+{
+       unsigned int range_size, seed;
+       unsigned long nranges;
+       uint64_t file_size;
+
+       range_size = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]);
+       file_size = min(f->real_file_size, f->io_size);
+
+       nranges = (file_size + range_size - 1) / range_size;
+
+       seed = jhash(f->file_name, strlen(f->file_name), 0) * td->thread_number;
+       if (!td->o.rand_repeatable)
+               seed = td->rand_seeds[4];
+
+       if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
+               zipf_init(&f->zipf, nranges, td->o.zipf_theta.u.f, seed);
+       else
+               pareto_init(&f->zipf, nranges, td->o.pareto_h.u.f, seed);
+
+       return 1;
+}
+
+static int init_rand_distribution(struct thread_data *td)
+{
+       struct fio_file *f;
+       unsigned int i;
+       int state;
+
+       if (td->o.random_distribution == FIO_RAND_DIST_RANDOM)
+               return 0;
+
+       state = td->runstate;
+       td_set_runstate(td, TD_SETTING_UP);
+       for_each_file(td, f, i)
+               __init_rand_distribution(td, f);
+       td_set_runstate(td, state);
+
+       return 1;
+}
+
 int init_random_map(struct thread_data *td)
 {
-       unsigned long long blocks, num_maps;
+       unsigned long long blocks;
        struct fio_file *f;
        unsigned int i;
 
-       if (td->o.norandommap || !td_random(td))
+       if (init_rand_distribution(td))
+               return 0;
+       if (!td_random(td))
                return 0;
 
        for_each_file(td, f, i) {
-               blocks = (f->real_file_size + td->o.rw_min_bs - 1) /
+               uint64_t file_size = min(f->real_file_size, f->io_size);
+
+               blocks = (file_size + td->o.rw_min_bs - 1) /
                                (unsigned long long) td->o.rw_min_bs;
-               num_maps = (blocks + BLOCKS_PER_MAP - 1) /
-                               (unsigned long long) BLOCKS_PER_MAP;
-               f->file_map = smalloc(num_maps * sizeof(int));
-               if (f->file_map) {
-                       f->num_maps = num_maps;
+               if (td->o.random_generator == FIO_RAND_GEN_LFSR) {
+                       unsigned long seed;
+
+                       seed = td->rand_seeds[FIO_RAND_BLOCK_OFF];
+                       
+                       if (!lfsr_init(&f->lfsr, blocks, seed))
+                               continue;
+               } else if (!td->o.norandommap) {
+                       f->io_axmap = axmap_new(blocks);
+                       if (f->io_axmap)
+                               continue;
+               } else if (td->o.norandommap)
                        continue;
-               }
+
                if (!td->o.softrandommap) {
                        log_err("fio: failed allocating random map. If running"
                                " a large number of jobs, try the 'norandommap'"
@@ -827,7 +984,6 @@ int init_random_map(struct thread_data *td)
 
                log_info("fio: file %s failed allocating random map. Running "
                         "job without.\n", f->file_name);
-               f->num_maps = 0;
        }
 
        return 0;
@@ -864,8 +1020,8 @@ void close_and_free_files(struct thread_data *td)
 
                sfree(f->file_name);
                f->file_name = NULL;
-               sfree(f->file_map);
-               f->file_map = NULL;
+               axmap_free(f->io_axmap);
+               f->io_axmap = NULL;
                sfree(f);
        }
 
@@ -885,10 +1041,13 @@ static void get_file_type(struct fio_file *f)
        else
                f->filetype = FIO_TYPE_FILE;
 
+       /* \\.\ is the device namespace in Windows, where every file is
+        * a block device */
+       if (strncmp(f->file_name, "\\\\.\\", 4) == 0)
+               f->filetype = FIO_TYPE_BD;
+
        if (!stat(f->file_name, &sb)) {
-               /* \\.\ is the device namespace in Windows, where every file is
-                * a block device */
-               if (S_ISBLK(sb.st_mode) || strncmp(f->file_name, "\\\\.\\", 4) == 0)
+               if (S_ISBLK(sb.st_mode))
                        f->filetype = FIO_TYPE_BD;
                else if (S_ISCHR(sb.st_mode))
                        f->filetype = FIO_TYPE_CHAR;
@@ -913,17 +1072,23 @@ int add_file(struct thread_data *td, const char *fname)
        }
 
        f->fd = -1;
-       fio_file_reset(f);
+       f->shadow_fd = -1;
+       fio_file_reset(td, f);
 
        if (td->files_size <= td->files_index) {
-               int new_size = td->o.nr_files + 1;
+               unsigned int new_size = td->o.nr_files + 1;
 
                dprint(FD_FILE, "resize file array to %d files\n", new_size);
 
                td->files = realloc(td->files, new_size * sizeof(f));
+               if (td->files == NULL) {
+                       log_err("fio: realloc OOM\n");
+                       assert(0);
+               }
                td->files_size = new_size;
        }
        td->files[cur_files] = f;
+       f->fileno = cur_files;
 
        /*
         * init function, io engine may not be loaded yet
@@ -950,7 +1115,7 @@ int add_file(struct thread_data *td, const char *fname)
                f->lock = fio_mutex_rw_init();
                break;
        case FILE_LOCK_EXCLUSIVE:
-               f->lock = fio_mutex_init(1);
+               f->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
                break;
        default:
                log_err("fio: unknown lock mode: %d\n", td->o.file_lock_mode);
@@ -1085,7 +1250,7 @@ static int recurse_dir(struct thread_data *td, const char *dirname)
        if (!D) {
                char buf[FIO_VERROR_SIZE];
 
-               snprintf(buf, FIO_VERROR_SIZE - 1, "opendir(%s)", dirname);
+               snprintf(buf, FIO_VERROR_SIZE, "opendir(%s)", dirname);
                td_verror(td, errno, buf);
                return 1;
        }
@@ -1097,7 +1262,7 @@ static int recurse_dir(struct thread_data *td, const char *dirname)
                if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
                        continue;
 
-               sprintf(full_path, "%s/%s", dirname, dir->d_name);
+               sprintf(full_path, "%s%s%s", dirname, FIO_OS_PATH_SEPARATOR, dir->d_name);
 
                if (lstat(full_path, &sb) == -1) {
                        if (errno != ENOENT) {
@@ -1154,7 +1319,7 @@ void dup_files(struct thread_data *td, struct thread_data *org)
                        assert(0);
                }
                __f->fd = -1;
-               fio_file_reset(__f);
+               fio_file_reset(td, __f);
 
                if (f->file_name) {
                        __f->file_name = smalloc_strdup(f->file_name);
@@ -1194,3 +1359,13 @@ void free_release_files(struct thread_data *td)
        td->files_index = 0;
        td->nr_normal_files = 0;
 }
+
+void fio_file_reset(struct thread_data *td, struct fio_file *f)
+{
+       f->last_pos = f->file_offset;
+       f->last_start = -1ULL;
+       if (f->io_axmap)
+               axmap_reset(f->io_axmap);
+       if (td->o.random_generator == FIO_RAND_GEN_LFSR)
+               lfsr_reset(&f->lfsr, td->rand_seeds[FIO_RAND_BLOCK_OFF]);
+}