[fio.git] / filesetup.c

#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <assert.h>
#include <dirent.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <sys/types.h>

#include "fio.h"
#include "smalloc.h"
#include "filehash.h"

static int root_warn;

static int extend_file(struct thread_data *td, struct fio_file *f)
{
	int r, new_layout = 0, unlink_file = 0, flags;
	unsigned long long left;
	unsigned int bs;
	char *b;

	if (read_only) {
		log_err("fio: refusing extend of file due to read-only\n");
		return 0;
	}

	/*
	 * check if we need to lay the file out complete again. fio
	 * does that for operations involving reads, or for writes
	 * where overwrite is set
	 */
	if (td_read(td) || (td_write(td) && td->o.overwrite))
		new_layout = 1;
	if (td_write(td) && !td->o.overwrite)
		unlink_file = 1;

	if (unlink_file || new_layout) {
		if ((unlink(f->file_name) < 0) && (errno != ENOENT)) {
			td_verror(td, errno, "unlink");
			return 1;
		}
	}

	flags = O_WRONLY | O_CREAT;
	if (new_layout)
		flags |= O_TRUNC;

	dprint(FD_FILE, "open file %s, flags %x\n", f->file_name, flags);
	f->fd = open(f->file_name, flags, 0644);
	if (f->fd < 0) {
		td_verror(td, errno, "open");
		return 1;
	}

	if (!new_layout)
		goto done;

	dprint(FD_FILE, "truncate file %s, size %llu\n", f->file_name,
							f->real_file_size);
	if (ftruncate(f->fd, f->real_file_size) == -1) {
		td_verror(td, errno, "ftruncate");
		goto err;
	}

	dprint(FD_FILE, "fallocate file %s, size %llu\n", f->file_name,
							f->real_file_size);
	if (posix_fallocate(f->fd, 0, f->real_file_size) < 0) {
		td_verror(td, errno, "posix_fallocate");
		goto err;
	}

	b = malloc(td->o.max_bs[DDIR_WRITE]);
	memset(b, 0, td->o.max_bs[DDIR_WRITE]);

	left = f->real_file_size;
	while (left && !td->terminate) {
		bs = td->o.max_bs[DDIR_WRITE];
		if (bs > left)
			bs = left;

		r = write(f->fd, b, bs);

		if (r == (int) bs) {
			left -= bs;
			continue;
		} else {
			if (r < 0)
				td_verror(td, errno, "write");
			else
				td_verror(td, EIO, "write");

			break;
		}
	}

	if (td->terminate)
		unlink(f->file_name);
	else if (td->o.create_fsync)
		fsync(f->fd);

	free(b);
done:
	close(f->fd);
	f->fd = -1;
	return 0;
err:
	close(f->fd);
	f->fd = -1;
	return 1;
}

static unsigned long long get_rand_file_size(struct thread_data *td)
{
	unsigned long long ret;
	long r;

	r = os_random_long(&td->file_size_state);
	ret = td->o.file_size_low + (unsigned long long) ((double) (td->o.file_size_high - td->o.file_size_low) * (r / (RAND_MAX + 1.0)));
	ret -= (ret % td->o.rw_min_bs);
	return ret;
}

static int file_size(struct thread_data *td, struct fio_file *f)
{
	struct stat st;

	if (fstat(f->fd, &st) == -1) {
		td_verror(td, errno, "fstat");
		return 1;
	}

	f->real_file_size = st.st_size;
	return 0;
}

static int bdev_size(struct thread_data *td, struct fio_file *f)
{
	unsigned long long bytes;
	int r;

	r = blockdev_size(f->fd, &bytes);
	if (r) {
		td_verror(td, r, "blockdev_size");
		return 1;
	}

	f->real_file_size = bytes;
	return 0;
}

static int get_file_size(struct thread_data *td, struct fio_file *f)
{
	int ret = 0;

	if (f->flags & FIO_SIZE_KNOWN)
		return 0;

	if (f->filetype == FIO_TYPE_FILE)
		ret = file_size(td, f);
	else if (f->filetype == FIO_TYPE_BD)
		ret = bdev_size(td, f);
	else
		f->real_file_size = -1;

	if (ret)
		return ret;

	if (f->file_offset > f->real_file_size) {
		log_err("%s: offset extends end (%Lu > %Lu)\n", td->o.name, f->file_offset, f->real_file_size);
		return 1;
	}

	f->flags |= FIO_SIZE_KNOWN;
	return 0;
}

int file_invalidate_cache(struct thread_data *td, struct fio_file *f)
{
	int ret = 0;

	dprint(FD_IO, "invalidate cache (%d)\n", td->o.odirect);

	if (td->o.odirect)
		return 0;

	/*
	 * FIXME: add blockdev flushing too
	 */
	if (f->mmap)
		ret = madvise(f->mmap, f->io_size, MADV_DONTNEED);
	else if (f->filetype == FIO_TYPE_FILE)
		ret = fadvise(f->fd, f->file_offset, f->io_size, POSIX_FADV_DONTNEED);
	else if (f->filetype == FIO_TYPE_BD) {
		ret = blockdev_invalidate_cache(f->fd);
		if (ret < 0 && errno == EACCES && geteuid()) {
			if (!root_warn) {
				log_err("fio: only root may flush block devices. Cache flush bypassed!\n");
				root_warn = 1;
			}
			ret = 0;
		}
	} else if (f->filetype == FIO_TYPE_CHAR || f->filetype == FIO_TYPE_PIPE)
		ret = 0;

	if (ret < 0) {
		td_verror(td, errno, "invalidate_cache");
		return 1;
	}

	return ret;
}

int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f)
{
	int ret = 0;

	dprint(FD_FILE, "fd close %s\n", f->file_name);

	remove_file_hash(f);

	if (close(f->fd) < 0)
		ret = errno;

	f->fd = -1;
	return ret;
}

int generic_open_file(struct thread_data *td, struct fio_file *f)
{
	struct fio_file *__f;
	int is_std = 0;
	int flags = 0;
	int from_hash = 0;

	dprint(FD_FILE, "fd open %s\n", f->file_name);

	if (!strcmp(f->file_name, "-")) {
		if (td_rw(td)) {
			log_err("fio: can't read/write to stdin/out\n");
			return 1;
		}
		is_std = 1;

		/*
		 * move output logging to stderr, if we are writing to stdout
		 */
		if (td_write(td))
			f_out = stderr;
	}

	if (td->o.odirect)
		flags |= OS_O_DIRECT;
	if (td->o.sync_io)
		flags |= O_SYNC;
	if (f->filetype != FIO_TYPE_FILE)
		flags |= O_NOATIME;

open_again:
	if (td_write(td)) {
		assert(!read_only);

		flags |= O_RDWR;

		if (f->filetype == FIO_TYPE_FILE)
			flags |= O_CREAT;

		if (is_std)
			f->fd = dup(STDOUT_FILENO);
		else {
			__f = lookup_file_hash(f->file_name);
			if (__f) {
				f->sem = __f->sem;
				f->fd = dup(__f->fd);
				f->references++;
				from_hash = 1;
			} else
				f->fd = open(f->file_name, flags, 0600);
		}
	} else {
		if (f->filetype == FIO_TYPE_CHAR && !read_only)
			flags |= O_RDWR;
		else
			flags |= O_RDONLY;

		if (is_std)
			f->fd = dup(STDIN_FILENO);
		else {
			__f = lookup_file_hash(f->file_name);
			if (__f) {
				f->sem = __f->sem;
				f->fd = dup(__f->fd);
				f->references++;
				from_hash = 1;
			} else
				f->fd = open(f->file_name, flags);
		}
	}

	if (f->fd == -1) {
		char buf[FIO_VERROR_SIZE];
		int __e = errno;

		if (errno == EPERM && (flags & O_NOATIME)) {
			flags &= ~O_NOATIME;
			goto open_again;
		}

		snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name);

		td_verror(td, __e, buf);
	}

	if (get_file_size(td, f))
		goto err;

	if (!from_hash && f->fd != -1) {
		if (add_file_hash(f)) {
			int ret;

			/*
			 * OK to ignore, we haven't done anything with it
			 */
			ret = generic_close_file(td, f);
			goto open_again;
		}
	}

	return 0;
err:
	close(f->fd);
	return 1;
}

int open_files(struct thread_data *td)
{
	struct fio_file *f;
	unsigned int i;
	int err = 0;

	dprint(FD_FILE, "open files\n");

	for_each_file(td, f, i) {
		err = td_io_open_file(td, f);
		if (err) {
			if (td->error == EMFILE) {
				log_err("fio: limited open files to: %d\n", td->nr_open_files);
				td->o.open_files = td->nr_open_files;
				err = 0;
				clear_error(td);
			}
			break;
		}

		if (td->o.open_files == td->nr_open_files)
			break;
	}

	if (!err)
		return 0;

	for_each_file(td, f, i)
		td_io_close_file(td, f);

	return err;
}

/*
 * open/close all files, so that ->real_file_size gets set
 */
static int get_file_sizes(struct thread_data *td)
{
	struct fio_file *f;
	unsigned int i;
	int err = 0;

	for_each_file(td, f, i) {
		if (td->io_ops->open_file(td, f)) {
			if (td->error != ENOENT) {
				log_err("%s\n", td->verror);
				err = 1;
			}
			clear_error(td);
		} else {
			if (td->io_ops->close_file)
				td->io_ops->close_file(td, f);
		}

		if (f->real_file_size == -1ULL && td->o.size)
			f->real_file_size = td->o.size / td->o.nr_files;
	}

	return err;
}

/*
 * Open the files and setup files sizes, creating files if necessary.
 */
int setup_files(struct thread_data *td)
{
	unsigned long long total_size, extend_size;
	struct fio_file *f;
	unsigned int i;
	int err = 0, need_extend;

	dprint(FD_FILE, "setup files\n");

	/*
	 * if ioengine defines a setup() method, it's responsible for
	 * opening the files and setting f->real_file_size to indicate
	 * the valid range for that file.
	 */
	if (td->io_ops->setup)
		err = td->io_ops->setup(td);
	else
		err = get_file_sizes(td);

	if (err)
		return err;

	/*
	 * check sizes. if the files/devices do not exist and the size
	 * isn't passed to fio, abort.
	 */
	total_size = 0;
	for_each_file(td, f, i) {
		if (f->real_file_size == -1ULL)
			total_size = -1ULL;
		else
			total_size += f->real_file_size;
	}

	/*
	 * device/file sizes are zero and no size given, punt
	 */
	if ((!total_size || total_size == -1ULL) && !td->o.size &&
	    !(td->io_ops->flags & FIO_NOIO) && !td->o.fill_device) {
		log_err("%s: you need to specify size=\n", td->o.name);
		td_verror(td, EINVAL, "total_file_size");
		return 1;
	}

	/*
	 * now file sizes are known, so we can set ->io_size. if size= is
	 * not given, ->io_size is just equal to ->real_file_size. if size
	 * is given, ->io_size is size / nr_files.
	 */
	extend_size = total_size = 0;
	need_extend = 0;
	for_each_file(td, f, i) {
		f->file_offset = td->o.start_offset;

		if (!td->o.file_size_low) {
			/*
			 * no file size range given, file size is equal to
			 * total size divided by number of files. if that is
			 * zero, set it to the real file size.
			 */
			f->io_size = td->o.size / td->o.nr_files;
			if (!f->io_size)
				f->io_size = f->real_file_size - f->file_offset;
		} else if (f->real_file_size < td->o.file_size_low ||
			   f->real_file_size > td->o.file_size_high) {
			if (f->file_offset > td->o.file_size_low) 
				goto err_offset;
			/*
			 * file size given. if it's fixed, use that. if it's a
			 * range, generate a random size in-between.
			 */
			if (td->o.file_size_low == td->o.file_size_high)
				f->io_size = td->o.file_size_low - f->file_offset;
			else
				f->io_size = get_rand_file_size(td) - f->file_offset;
		} else
			f->io_size = f->real_file_size - f->file_offset;

		if (f->io_size == -1ULL)
			total_size = -1ULL;
		else
			total_size += f->io_size;

		if (f->filetype == FIO_TYPE_FILE &&
		    (f->io_size + f->file_offset) > f->real_file_size &&
		    !(td->io_ops->flags & FIO_DISKLESSIO)) {
			need_extend++;
			extend_size += (f->io_size + f->file_offset);
			f->flags |= FIO_FILE_EXTEND;
		}	
	}

	if (!td->o.size || td->o.size > total_size)
		td->o.size = total_size;

	/*
	 * See if we need to extend some files
	 */
	if (need_extend) {
		temp_stall_ts = 1;
		log_info("%s: Laying out IO file(s) (%u file(s) / %LuMiB)\n",
			td->o.name, need_extend, extend_size >> 20);

		for_each_file(td, f, i) {
			if (!(f->flags & FIO_FILE_EXTEND))
				continue;

			assert(f->filetype == FIO_TYPE_FILE);
			f->flags &= ~FIO_FILE_EXTEND;
			f->real_file_size = (f->io_size + f->file_offset);
			err = extend_file(td, f);
			if (err)
				break;
		}
		temp_stall_ts = 0;
	}

	if (err)
		return err;

	if (!td->o.zone_size)
		td->o.zone_size = td->o.size;

	/*
	 * iolog already set the total io size, if we read back
	 * stored entries.
	 */
	if (!td->o.read_iolog_file)
		td->total_io_size = td->o.size * td->o.loops;
	return 0;
err_offset:
	log_err("%s: you need to specify valid offset=\n", td->o.name);
	return 1;
}

int init_random_map(struct thread_data *td)
{
	unsigned long long blocks, num_maps;
	struct fio_file *f;
	unsigned int i;

	if (td->o.norandommap || !td_random(td))
		return 0;

	for_each_file(td, f, i) {
		blocks = (f->real_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(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;
	}

	return 0;
}

void close_files(struct thread_data *td)
{
	struct fio_file *f;
	unsigned int i;

	for_each_file(td, f, i)
		td_io_close_file(td, f);
}

void close_and_free_files(struct thread_data *td)
{
	struct fio_file *f;
	unsigned int i;

	dprint(FD_FILE, "close files\n");

	for_each_file(td, f, i) {
		if (td->o.unlink && f->filetype == FIO_TYPE_FILE)
			unlink(f->file_name);

		td_io_close_file(td, f);

		sfree(f->file_name);
		f->file_name = NULL;

		if (f->file_map) {
			sfree(f->file_map);
			f->file_map = NULL;
		}
		sfree(f);
	}

	td->o.filename = NULL;
	free(td->files);
	td->files = NULL;
	td->o.nr_files = 0;
}

static void get_file_type(struct fio_file *f)
{
	struct stat sb;

	if (!strcmp(f->file_name, "-"))
		f->filetype = FIO_TYPE_PIPE;
	else
		f->filetype = FIO_TYPE_FILE;

	if (!lstat(f->file_name, &sb)) {
		if (S_ISBLK(sb.st_mode))
			f->filetype = FIO_TYPE_BD;
		else if (S_ISCHR(sb.st_mode))
			f->filetype = FIO_TYPE_CHAR;
		else if (S_ISFIFO(sb.st_mode))
			f->filetype = FIO_TYPE_PIPE;
	}
}

int add_file(struct thread_data *td, const char *fname)
{
	int cur_files = td->files_index;
	char file_name[PATH_MAX];
	struct fio_file *f;
	int len = 0;

	dprint(FD_FILE, "add file %s\n", fname);

	f = smalloc(sizeof(*f));
	f->fd = -1;

	td->files = realloc(td->files, (cur_files + 1) * sizeof(f));

	td->files[cur_files] = f;

	/*
	 * init function, io engine may not be loaded yet
	 */
	if (td->io_ops && (td->io_ops->flags & FIO_DISKLESSIO))
		f->real_file_size = -1ULL;

	if (td->o.directory)
		len = sprintf(file_name, "%s/", td->o.directory);

	sprintf(file_name + len, "%s", fname);
	f->file_name = smalloc_strdup(file_name);

	get_file_type(f);

	if (td->o.lockfile)
		f->sem = fio_sem_init(1);

	td->files_index++;
	if (f->filetype == FIO_TYPE_FILE)
		td->nr_normal_files++;

	return cur_files;
}

void get_file(struct fio_file *f)
{
	dprint(FD_FILE, "get file %s, ref=%d\n", f->file_name, f->references);
	assert(f->flags & FIO_FILE_OPEN);
	f->references++;
}

int put_file(struct thread_data *td, struct fio_file *f)
{
	int ret = 0;

	dprint(FD_FILE, "put file %s, ref=%d\n", f->file_name, f->references);

	if (!(f->flags & FIO_FILE_OPEN))
		return 0;

	assert(f->references);
	if (--f->references)
		return 0;

	if (should_fsync(td) && td->o.fsync_on_close)
		fsync(f->fd);

	if (td->io_ops->close_file)
		ret = td->io_ops->close_file(td, f);

	td->nr_open_files--;
	f->flags &= ~FIO_FILE_OPEN;
	return ret;
}

void lock_file(struct thread_data *td, struct fio_file *f)
{
	if (f && f->sem) {
		if (f->sem_owner == td && f->sem_batch--)
			return;

		fio_sem_down(f->sem);
		f->sem_owner = td;
		f->sem_batch = td->o.lockfile_batch;
	}
}

void unlock_file(struct fio_file *f)
{
	if (f && f->sem) {
		int sem_val;

		if (f->sem_batch)
			return;

		sem_getvalue(&f->sem->sem, &sem_val);
		if (!sem_val)
			f->sem_owner = NULL;
		fio_sem_up(f->sem);
	}
}

static int recurse_dir(struct thread_data *td, const char *dirname)
{
	struct dirent *dir;
	int ret = 0;
	DIR *D;

	D = opendir(dirname);
	if (!D) {
		char buf[FIO_VERROR_SIZE];

		snprintf(buf, FIO_VERROR_SIZE - 1, "opendir(%s)", dirname);
		td_verror(td, errno, buf);
		return 1;
	}

	while ((dir = readdir(D)) != NULL) {
		char full_path[PATH_MAX];
		struct stat sb;

		if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
			continue;

		sprintf(full_path, "%s/%s", dirname, dir->d_name);

		if (lstat(full_path, &sb) == -1) {
			if (errno != ENOENT) {
				td_verror(td, errno, "stat");
				return 1;
			}
		}

		if (S_ISREG(sb.st_mode)) {
			add_file(td, full_path);
			td->o.nr_files++;
			continue;
		}
		if (!S_ISDIR(sb.st_mode))
			continue;

		if ((ret = recurse_dir(td, full_path)) != 0)
			break;
	}

	closedir(D);
	return ret;
}

int add_dir_files(struct thread_data *td, const char *path)
{
	int ret = recurse_dir(td, path);

	if (!ret)
		log_info("fio: opendir added %d files\n", td->o.nr_files);

	return ret;
}

void dup_files(struct thread_data *td, struct thread_data *org)
{
	struct fio_file *f;
	unsigned int i;
	size_t bytes;

	if (!org->files)
		return;

	bytes = org->files_index * sizeof(f);
	td->files = malloc(bytes);
	memcpy(td->files, org->files, bytes);

	for_each_file(td, f, i) {
		struct fio_file *__f;

		__f = smalloc(sizeof(*__f));

		if (f->file_name)
			__f->file_name = smalloc_strdup(f->file_name);

		td->files[i] = __f;
	}
}

/*
 * Returns the index that matches the filename, or -1 if not there
 */
int get_fileno(struct thread_data *td, const char *fname)
{
	struct fio_file *f;
	unsigned int i;

	for_each_file(td, f, i)
		if (!strcmp(f->file_name, fname))
			return i;

	return -1;
}

/*
 * For log usage, where we add/open/close files automatically
 */
void free_release_files(struct thread_data *td)
{
	close_files(td);
	td->files_index = 0;
	td->nr_normal_files = 0;
}
Commit	Line	Data
	1	#include <unistd.h>
	2	#include <fcntl.h>
	3	#include <string.h>
	4	#include <assert.h>
	5	#include <dirent.h>
	6	#include <sys/stat.h>
	7	#include <sys/mman.h>
	8	#include <sys/types.h>
	9
	10	#include "fio.h"
	11	#include "smalloc.h"
	12	#include "filehash.h"
	13
	14	static int root_warn;
	15
	16	static int extend_file(struct thread_data td, struct fio_file f)
	17	{
	18	int r, new_layout = 0, unlink_file = 0, flags;
	19	unsigned long long left;
	20	unsigned int bs;
	21	char *b;
	22
	23	if (read_only) {
	24	log_err("fio: refusing extend of file due to read-only\n");
	25	return 0;
	26	}
	27
	28	/*
	29	* check if we need to lay the file out complete again. fio
	30	* does that for operations involving reads, or for writes
	31	* where overwrite is set
	32	*/
	33	if (td_read(td) \|\| (td_write(td) && td->o.overwrite))
	34	new_layout = 1;
	35	if (td_write(td) && !td->o.overwrite)
	36	unlink_file = 1;
	37
	38	if (unlink_file \|\| new_layout) {
	39	if ((unlink(f->file_name) < 0) && (errno != ENOENT)) {
	40	td_verror(td, errno, "unlink");
	41	return 1;
	42	}
	43	}
	44
	45	flags = O_WRONLY \| O_CREAT;
	46	if (new_layout)
	47	flags \|= O_TRUNC;
	48
	49	dprint(FD_FILE, "open file %s, flags %x\n", f->file_name, flags);
	50	f->fd = open(f->file_name, flags, 0644);
	51	if (f->fd < 0) {
	52	td_verror(td, errno, "open");
	53	return 1;
	54	}
	55
	56	if (!new_layout)
	57	goto done;
	58
	59	dprint(FD_FILE, "truncate file %s, size %llu\n", f->file_name,
	60	f->real_file_size);
	61	if (ftruncate(f->fd, f->real_file_size) == -1) {
	62	td_verror(td, errno, "ftruncate");
	63	goto err;
	64	}
	65
	66	dprint(FD_FILE, "fallocate file %s, size %llu\n", f->file_name,
	67	f->real_file_size);
	68	if (posix_fallocate(f->fd, 0, f->real_file_size) < 0) {
	69	td_verror(td, errno, "posix_fallocate");
	70	goto err;
	71	}
	72
	73	b = malloc(td->o.max_bs[DDIR_WRITE]);
	74	memset(b, 0, td->o.max_bs[DDIR_WRITE]);
	75
	76	left = f->real_file_size;
	77	while (left && !td->terminate) {
	78	bs = td->o.max_bs[DDIR_WRITE];
	79	if (bs > left)
	80	bs = left;
	81
	82	r = write(f->fd, b, bs);
	83
	84	if (r == (int) bs) {
	85	left -= bs;
	86	continue;
	87	} else {
	88	if (r < 0)
	89	td_verror(td, errno, "write");
	90	else
	91	td_verror(td, EIO, "write");
	92
	93	break;
	94	}
	95	}
	96
	97	if (td->terminate)
	98	unlink(f->file_name);
	99	else if (td->o.create_fsync)
	100	fsync(f->fd);
	101
	102	free(b);
	103	done:
	104	close(f->fd);
	105	f->fd = -1;
	106	return 0;
	107	err:
	108	close(f->fd);
	109	f->fd = -1;
	110	return 1;
	111	}
	112
	113	static unsigned long long get_rand_file_size(struct thread_data *td)
	114	{
	115	unsigned long long ret;
	116	long r;
	117
	118	r = os_random_long(&td->file_size_state);
	119	ret = td->o.file_size_low + (unsigned long long) ((double) (td->o.file_size_high - td->o.file_size_low) * (r / (RAND_MAX + 1.0)));
	120	ret -= (ret % td->o.rw_min_bs);
	121	return ret;
	122	}
	123
	124	static int file_size(struct thread_data td, struct fio_file f)
	125	{
	126	struct stat st;
	127
	128	if (fstat(f->fd, &st) == -1) {
	129	td_verror(td, errno, "fstat");
	130	return 1;
	131	}
	132
	133	f->real_file_size = st.st_size;
	134	return 0;
	135	}
	136
	137	static int bdev_size(struct thread_data td, struct fio_file f)
	138	{
	139	unsigned long long bytes;
	140	int r;
	141
	142	r = blockdev_size(f->fd, &bytes);
	143	if (r) {
	144	td_verror(td, r, "blockdev_size");
	145	return 1;
	146	}
	147
	148	f->real_file_size = bytes;
	149	return 0;
	150	}
	151
	152	static int get_file_size(struct thread_data td, struct fio_file f)
	153	{
	154	int ret = 0;
	155
	156	if (f->flags & FIO_SIZE_KNOWN)
	157	return 0;
	158
	159	if (f->filetype == FIO_TYPE_FILE)
	160	ret = file_size(td, f);
	161	else if (f->filetype == FIO_TYPE_BD)
	162	ret = bdev_size(td, f);
	163	else
	164	f->real_file_size = -1;
	165
	166	if (ret)
	167	return ret;
	168
	169	if (f->file_offset > f->real_file_size) {
	170	log_err("%s: offset extends end (%Lu > %Lu)\n", td->o.name, f->file_offset, f->real_file_size);
	171	return 1;
	172	}
	173
	174	f->flags \|= FIO_SIZE_KNOWN;
	175	return 0;
	176	}
	177
	178	int file_invalidate_cache(struct thread_data td, struct fio_file f)
	179	{
	180	int ret = 0;
	181
	182	dprint(FD_IO, "invalidate cache (%d)\n", td->o.odirect);
	183
	184	if (td->o.odirect)
	185	return 0;
	186
	187	/*
	188	* FIXME: add blockdev flushing too
	189	*/
	190	if (f->mmap)
	191	ret = madvise(f->mmap, f->io_size, MADV_DONTNEED);
	192	else if (f->filetype == FIO_TYPE_FILE)
	193	ret = fadvise(f->fd, f->file_offset, f->io_size, POSIX_FADV_DONTNEED);
	194	else if (f->filetype == FIO_TYPE_BD) {
	195	ret = blockdev_invalidate_cache(f->fd);
	196	if (ret < 0 && errno == EACCES && geteuid()) {
	197	if (!root_warn) {
	198	log_err("fio: only root may flush block devices. Cache flush bypassed!\n");
	199	root_warn = 1;
	200	}
	201	ret = 0;
	202	}
	203	} else if (f->filetype == FIO_TYPE_CHAR \|\| f->filetype == FIO_TYPE_PIPE)
	204	ret = 0;
	205
	206	if (ret < 0) {
	207	td_verror(td, errno, "invalidate_cache");
	208	return 1;
	209	}
	210
	211	return ret;
	212	}
	213
	214	int generic_close_file(struct thread_data fio_unused td, struct fio_file f)
	215	{
	216	int ret = 0;
	217
	218	dprint(FD_FILE, "fd close %s\n", f->file_name);
	219
	220	remove_file_hash(f);
	221
	222	if (close(f->fd) < 0)
	223	ret = errno;
	224
	225	f->fd = -1;
	226	return ret;
	227	}
	228
	229	int generic_open_file(struct thread_data td, struct fio_file f)
	230	{
	231	struct fio_file *__f;
	232	int is_std = 0;
	233	int flags = 0;
	234	int from_hash = 0;
	235
	236	dprint(FD_FILE, "fd open %s\n", f->file_name);
	237
	238	if (!strcmp(f->file_name, "-")) {
	239	if (td_rw(td)) {
	240	log_err("fio: can't read/write to stdin/out\n");
	241	return 1;
	242	}
	243	is_std = 1;
	244
	245	/*
	246	* move output logging to stderr, if we are writing to stdout
	247	*/
	248	if (td_write(td))
	249	f_out = stderr;
	250	}
	251
	252	if (td->o.odirect)
	253	flags \|= OS_O_DIRECT;
	254	if (td->o.sync_io)
	255	flags \|= O_SYNC;
	256	if (f->filetype != FIO_TYPE_FILE)
	257	flags \|= O_NOATIME;
	258
	259	open_again:
	260	if (td_write(td)) {
	261	assert(!read_only);
	262
	263	flags \|= O_RDWR;
	264
	265	if (f->filetype == FIO_TYPE_FILE)
	266	flags \|= O_CREAT;
	267
	268	if (is_std)
	269	f->fd = dup(STDOUT_FILENO);
	270	else {
	271	__f = lookup_file_hash(f->file_name);
	272	if (__f) {
	273	f->sem = __f->sem;
	274	f->fd = dup(__f->fd);
	275	f->references++;
	276	from_hash = 1;
	277	} else
	278	f->fd = open(f->file_name, flags, 0600);
	279	}
	280	} else {
	281	if (f->filetype == FIO_TYPE_CHAR && !read_only)
	282	flags \|= O_RDWR;
	283	else
	284	flags \|= O_RDONLY;
	285
	286	if (is_std)
	287	f->fd = dup(STDIN_FILENO);
	288	else {
	289	__f = lookup_file_hash(f->file_name);
	290	if (__f) {
	291	f->sem = __f->sem;
	292	f->fd = dup(__f->fd);
	293	f->references++;
	294	from_hash = 1;
	295	} else
	296	f->fd = open(f->file_name, flags);
	297	}
	298	}
	299
	300	if (f->fd == -1) {
	301	char buf[FIO_VERROR_SIZE];
	302	int __e = errno;
	303
	304	if (errno == EPERM && (flags & O_NOATIME)) {
	305	flags &= ~O_NOATIME;
	306	goto open_again;
	307	}
	308
	309	snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name);
	310
	311	td_verror(td, __e, buf);
	312	}
	313
	314	if (get_file_size(td, f))
	315	goto err;
	316
	317	if (!from_hash && f->fd != -1) {
	318	if (add_file_hash(f)) {
	319	int ret;
	320
	321	/*
	322	* OK to ignore, we haven't done anything with it
	323	*/
	324	ret = generic_close_file(td, f);
	325	goto open_again;
	326	}
	327	}
	328
	329	return 0;
	330	err:
	331	close(f->fd);
	332	return 1;
	333	}
	334
	335	int open_files(struct thread_data *td)
	336	{
	337	struct fio_file *f;
	338	unsigned int i;
	339	int err = 0;
	340
	341	dprint(FD_FILE, "open files\n");
	342
	343	for_each_file(td, f, i) {
	344	err = td_io_open_file(td, f);
	345	if (err) {
	346	if (td->error == EMFILE) {
	347	log_err("fio: limited open files to: %d\n", td->nr_open_files);
	348	td->o.open_files = td->nr_open_files;
	349	err = 0;
	350	clear_error(td);
	351	}
	352	break;
	353	}
	354
	355	if (td->o.open_files == td->nr_open_files)
	356	break;
	357	}
	358
	359	if (!err)
	360	return 0;
	361
	362	for_each_file(td, f, i)
	363	td_io_close_file(td, f);
	364
	365	return err;
	366	}
	367
	368	/*
	369	* open/close all files, so that ->real_file_size gets set
	370	*/
	371	static int get_file_sizes(struct thread_data *td)
	372	{
	373	struct fio_file *f;
	374	unsigned int i;
	375	int err = 0;
	376
	377	for_each_file(td, f, i) {
	378	if (td->io_ops->open_file(td, f)) {
	379	if (td->error != ENOENT) {
	380	log_err("%s\n", td->verror);
	381	err = 1;
	382	}
	383	clear_error(td);
	384	} else {
	385	if (td->io_ops->close_file)
	386	td->io_ops->close_file(td, f);
	387	}
	388
	389	if (f->real_file_size == -1ULL && td->o.size)
	390	f->real_file_size = td->o.size / td->o.nr_files;
	391	}
	392
	393	return err;
	394	}
	395
	396	/*
	397	* Open the files and setup files sizes, creating files if necessary.
	398	*/
	399	int setup_files(struct thread_data *td)
	400	{
	401	unsigned long long total_size, extend_size;
	402	struct fio_file *f;
	403	unsigned int i;
	404	int err = 0, need_extend;
	405
	406	dprint(FD_FILE, "setup files\n");
	407
	408	/*
	409	* if ioengine defines a setup() method, it's responsible for
	410	* opening the files and setting f->real_file_size to indicate
	411	* the valid range for that file.
	412	*/
	413	if (td->io_ops->setup)
	414	err = td->io_ops->setup(td);
	415	else
	416	err = get_file_sizes(td);
	417
	418	if (err)
	419	return err;
	420
	421	/*
	422	* check sizes. if the files/devices do not exist and the size
	423	* isn't passed to fio, abort.
	424	*/
	425	total_size = 0;
	426	for_each_file(td, f, i) {
	427	if (f->real_file_size == -1ULL)
	428	total_size = -1ULL;
	429	else
	430	total_size += f->real_file_size;
	431	}
	432
	433	/*
	434	* device/file sizes are zero and no size given, punt
	435	*/
	436	if ((!total_size \|\| total_size == -1ULL) && !td->o.size &&
	437	!(td->io_ops->flags & FIO_NOIO) && !td->o.fill_device) {
	438	log_err("%s: you need to specify size=\n", td->o.name);
	439	td_verror(td, EINVAL, "total_file_size");
	440	return 1;
	441	}
	442
	443	/*
	444	* now file sizes are known, so we can set ->io_size. if size= is
	445	* not given, ->io_size is just equal to ->real_file_size. if size
	446	* is given, ->io_size is size / nr_files.
	447	*/
	448	extend_size = total_size = 0;
	449	need_extend = 0;
	450	for_each_file(td, f, i) {
	451	f->file_offset = td->o.start_offset;
	452
	453	if (!td->o.file_size_low) {
	454	/*
	455	* no file size range given, file size is equal to
	456	* total size divided by number of files. if that is
	457	* zero, set it to the real file size.
	458	*/
	459	f->io_size = td->o.size / td->o.nr_files;
	460	if (!f->io_size)
	461	f->io_size = f->real_file_size - f->file_offset;
	462	} else if (f->real_file_size < td->o.file_size_low \|\|
	463	f->real_file_size > td->o.file_size_high) {
	464	if (f->file_offset > td->o.file_size_low)
	465	goto err_offset;
	466	/*
	467	* file size given. if it's fixed, use that. if it's a
	468	* range, generate a random size in-between.
	469	*/
	470	if (td->o.file_size_low == td->o.file_size_high)
	471	f->io_size = td->o.file_size_low - f->file_offset;
	472	else
	473	f->io_size = get_rand_file_size(td) - f->file_offset;
	474	} else
	475	f->io_size = f->real_file_size - f->file_offset;
	476
	477	if (f->io_size == -1ULL)
	478	total_size = -1ULL;
	479	else
	480	total_size += f->io_size;
	481
	482	if (f->filetype == FIO_TYPE_FILE &&
	483	(f->io_size + f->file_offset) > f->real_file_size &&
	484	!(td->io_ops->flags & FIO_DISKLESSIO)) {
	485	need_extend++;
	486	extend_size += (f->io_size + f->file_offset);
	487	f->flags \|= FIO_FILE_EXTEND;
	488	}
	489	}
	490
	491	if (!td->o.size \|\| td->o.size > total_size)
	492	td->o.size = total_size;
	493
	494	/*
	495	* See if we need to extend some files
	496	*/
	497	if (need_extend) {
	498	temp_stall_ts = 1;
	499	log_info("%s: Laying out IO file(s) (%u file(s) / %LuMiB)\n",
	500	td->o.name, need_extend, extend_size >> 20);
	501
	502	for_each_file(td, f, i) {
	503	if (!(f->flags & FIO_FILE_EXTEND))
	504	continue;
	505
	506	assert(f->filetype == FIO_TYPE_FILE);
	507	f->flags &= ~FIO_FILE_EXTEND;
	508	f->real_file_size = (f->io_size + f->file_offset);
	509	err = extend_file(td, f);
	510	if (err)
	511	break;
	512	}
	513	temp_stall_ts = 0;
	514	}
	515
	516	if (err)
	517	return err;
	518
	519	if (!td->o.zone_size)
	520	td->o.zone_size = td->o.size;
	521
	522	/*
	523	* iolog already set the total io size, if we read back
	524	* stored entries.
	525	*/
	526	if (!td->o.read_iolog_file)
	527	td->total_io_size = td->o.size * td->o.loops;
	528	return 0;
	529	err_offset:
	530	log_err("%s: you need to specify valid offset=\n", td->o.name);
	531	return 1;
	532	}
	533
	534	int init_random_map(struct thread_data *td)
	535	{
	536	unsigned long long blocks, num_maps;
	537	struct fio_file *f;
	538	unsigned int i;
	539
	540	if (td->o.norandommap \|\| !td_random(td))
	541	return 0;
	542
	543	for_each_file(td, f, i) {
	544	blocks = (f->real_file_size + td->o.rw_min_bs - 1) / (unsigned long long) td->o.rw_min_bs;
	545	num_maps = (blocks + BLOCKS_PER_MAP-1)/ (unsigned long long) BLOCKS_PER_MAP;
	546	f->file_map = smalloc(num_maps * sizeof(long));
	547	if (!f->file_map) {
	548	log_err("fio: failed allocating random map. If running a large number of jobs, try the 'norandommap' option\n");
	549	return 1;
	550	}
	551	f->num_maps = num_maps;
	552	}
	553
	554	return 0;
	555	}
	556
	557	void close_files(struct thread_data *td)
	558	{
	559	struct fio_file *f;
	560	unsigned int i;
	561
	562	for_each_file(td, f, i)
	563	td_io_close_file(td, f);
	564	}
	565
	566	void close_and_free_files(struct thread_data *td)
	567	{
	568	struct fio_file *f;
	569	unsigned int i;
	570
	571	dprint(FD_FILE, "close files\n");
	572
	573	for_each_file(td, f, i) {
	574	if (td->o.unlink && f->filetype == FIO_TYPE_FILE)
	575	unlink(f->file_name);
	576
	577	td_io_close_file(td, f);
	578
	579	sfree(f->file_name);
	580	f->file_name = NULL;
	581
	582	if (f->file_map) {
	583	sfree(f->file_map);
	584	f->file_map = NULL;
	585	}
	586	sfree(f);
	587	}
	588
	589	td->o.filename = NULL;
	590	free(td->files);
	591	td->files = NULL;
	592	td->o.nr_files = 0;
	593	}
	594
	595	static void get_file_type(struct fio_file *f)
	596	{
	597	struct stat sb;
	598
	599	if (!strcmp(f->file_name, "-"))
	600	f->filetype = FIO_TYPE_PIPE;
	601	else
	602	f->filetype = FIO_TYPE_FILE;
	603
	604	if (!lstat(f->file_name, &sb)) {
	605	if (S_ISBLK(sb.st_mode))
	606	f->filetype = FIO_TYPE_BD;
	607	else if (S_ISCHR(sb.st_mode))
	608	f->filetype = FIO_TYPE_CHAR;
	609	else if (S_ISFIFO(sb.st_mode))
	610	f->filetype = FIO_TYPE_PIPE;
	611	}
	612	}
	613
	614	int add_file(struct thread_data td, const char fname)
	615	{
	616	int cur_files = td->files_index;
	617	char file_name[PATH_MAX];
	618	struct fio_file *f;
	619	int len = 0;
	620
	621	dprint(FD_FILE, "add file %s\n", fname);
	622
	623	f = smalloc(sizeof(*f));
	624	f->fd = -1;
	625
	626	td->files = realloc(td->files, (cur_files + 1) * sizeof(f));
	627
	628	td->files[cur_files] = f;
	629
	630	/*
	631	* init function, io engine may not be loaded yet
	632	*/
	633	if (td->io_ops && (td->io_ops->flags & FIO_DISKLESSIO))
	634	f->real_file_size = -1ULL;
	635
	636	if (td->o.directory)
	637	len = sprintf(file_name, "%s/", td->o.directory);
	638
	639	sprintf(file_name + len, "%s", fname);
	640	f->file_name = smalloc_strdup(file_name);
	641
	642	get_file_type(f);
	643
	644	if (td->o.lockfile)
	645	f->sem = fio_sem_init(1);
	646
	647	td->files_index++;
	648	if (f->filetype == FIO_TYPE_FILE)
	649	td->nr_normal_files++;
	650
	651	return cur_files;
	652	}
	653
	654	void get_file(struct fio_file *f)
	655	{
	656	dprint(FD_FILE, "get file %s, ref=%d\n", f->file_name, f->references);
	657	assert(f->flags & FIO_FILE_OPEN);
	658	f->references++;
	659	}
	660
	661	int put_file(struct thread_data td, struct fio_file f)
	662	{
	663	int ret = 0;
	664
	665	dprint(FD_FILE, "put file %s, ref=%d\n", f->file_name, f->references);
	666
	667	if (!(f->flags & FIO_FILE_OPEN))
	668	return 0;
	669
	670	assert(f->references);
	671	if (--f->references)
	672	return 0;
	673
	674	if (should_fsync(td) && td->o.fsync_on_close)
	675	fsync(f->fd);
	676
	677	if (td->io_ops->close_file)
	678	ret = td->io_ops->close_file(td, f);
	679
	680	td->nr_open_files--;
	681	f->flags &= ~FIO_FILE_OPEN;
	682	return ret;
	683	}
	684
	685	void lock_file(struct thread_data td, struct fio_file f)
	686	{
	687	if (f && f->sem) {
	688	if (f->sem_owner == td && f->sem_batch--)
	689	return;
	690
	691	fio_sem_down(f->sem);
	692	f->sem_owner = td;
	693	f->sem_batch = td->o.lockfile_batch;
	694	}
	695	}
	696
	697	void unlock_file(struct fio_file *f)
	698	{
	699	if (f && f->sem) {
	700	int sem_val;
	701
	702	if (f->sem_batch)
	703	return;
	704
	705	sem_getvalue(&f->sem->sem, &sem_val);
	706	if (!sem_val)
	707	f->sem_owner = NULL;
	708	fio_sem_up(f->sem);
	709	}
	710	}
	711
	712	static int recurse_dir(struct thread_data td, const char dirname)
	713	{
	714	struct dirent *dir;
	715	int ret = 0;
	716	DIR *D;
	717
	718	D = opendir(dirname);
	719	if (!D) {
	720	char buf[FIO_VERROR_SIZE];
	721
	722	snprintf(buf, FIO_VERROR_SIZE - 1, "opendir(%s)", dirname);
	723	td_verror(td, errno, buf);
	724	return 1;
	725	}
	726
	727	while ((dir = readdir(D)) != NULL) {
	728	char full_path[PATH_MAX];
	729	struct stat sb;
	730
	731	if (!strcmp(dir->d_name, ".") \|\| !strcmp(dir->d_name, ".."))
	732	continue;
	733
	734	sprintf(full_path, "%s/%s", dirname, dir->d_name);
	735
	736	if (lstat(full_path, &sb) == -1) {
	737	if (errno != ENOENT) {
	738	td_verror(td, errno, "stat");
	739	return 1;
	740	}
	741	}
	742
	743	if (S_ISREG(sb.st_mode)) {
	744	add_file(td, full_path);
	745	td->o.nr_files++;
	746	continue;
	747	}
	748	if (!S_ISDIR(sb.st_mode))
	749	continue;
	750
	751	if ((ret = recurse_dir(td, full_path)) != 0)
	752	break;
	753	}
	754
	755	closedir(D);
	756	return ret;
	757	}
	758
	759	int add_dir_files(struct thread_data td, const char path)
	760	{
	761	int ret = recurse_dir(td, path);
	762
	763	if (!ret)
	764	log_info("fio: opendir added %d files\n", td->o.nr_files);
	765
	766	return ret;
	767	}
	768
	769	void dup_files(struct thread_data td, struct thread_data org)
	770	{
	771	struct fio_file *f;
	772	unsigned int i;
	773	size_t bytes;
	774
	775	if (!org->files)
	776	return;
	777
	778	bytes = org->files_index * sizeof(f);
	779	td->files = malloc(bytes);
	780	memcpy(td->files, org->files, bytes);
	781
	782	for_each_file(td, f, i) {
	783	struct fio_file *__f;
	784
	785	__f = smalloc(sizeof(*__f));
	786
	787	if (f->file_name)
	788	__f->file_name = smalloc_strdup(f->file_name);
	789
	790	td->files[i] = __f;
	791	}
	792	}
	793
	794	/*
	795	* Returns the index that matches the filename, or -1 if not there
	796	*/
	797	int get_fileno(struct thread_data td, const char fname)
	798	{
	799	struct fio_file *f;
	800	unsigned int i;
	801
	802	for_each_file(td, f, i)
	803	if (!strcmp(f->file_name, fname))
	804	return i;
	805
	806	return -1;
	807	}
	808
	809	/*
	810	* For log usage, where we add/open/close files automatically
	811	*/
	812	void free_release_files(struct thread_data *td)
	813	{
	814	close_files(td);
	815	td->files_index = 0;
	816	td->nr_normal_files = 0;
	817	}