X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=filesetup.c;h=fa09219f62a2e2b93f0e001b2e7254da72e04e99;hb=243bfe190245a10e9d0981bf2a7c722edc4c43d4;hp=1372177efc67a6499c4a3c48483ff6b92981d981;hpb=58726a1006aa33303e41951faac1ceca2bf09ec5;p=fio.git

diff --git a/filesetup.c b/filesetup.c
index 1372177e..fa09219f 100644
--- a/filesetup.c
+++ b/filesetup.c
@@ -3,6 +3,7 @@
 #include <string.h>
 #include <assert.h>
 #include <dirent.h>
+#include <libgen.h>
 #include <sys/stat.h>
 #include <sys/mman.h>
 #include <sys/types.h>
@@ -10,9 +11,25 @@
 #include "fio.h"
 #include "smalloc.h"
 #include "filehash.h"
+#include "options.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;
 
+static FLIST_HEAD(filename_list);
+
+static inline void clear_error(struct thread_data *td)
+{
+	td->error = 0;
+	td->verror[0] = '\0';
+}
+
 /*
  * Leaves f->fd open on success, caller must close
  */
@@ -33,10 +50,11 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
 	 * does that for operations involving reads, or for writes
 	 * where overwrite is set
 	 */
-	if (td_read(td) || (td_write(td) && td->o.overwrite) ||
+	if (td_read(td) ||
+	   (td_write(td) && td->o.overwrite && !td->o.file_append) ||
 	    (td_write(td) && td->io_ops->flags & FIO_NOEXTEND))
 		new_layout = 1;
-	if (td_write(td) && !td->o.overwrite)
+	if (td_write(td) && !td->o.overwrite && !td->o.file_append)
 		unlink_file = 1;
 
 	if (unlink_file || new_layout) {
@@ -58,26 +76,63 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
 		return 1;
 	}
 
+#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,
+				 (unsigned long long) 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,
+				(unsigned long long) 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");
+
+			break;
+#endif /* CONFIG_LINUX_FALLOCATE */
+		default:
+			log_err("fio: unknown fallocate mode: %d\n",
+				td->o.fallocate_mode);
+			assert(0);
+		}
+	}
+#endif /* CONFIG_POSIX_FALLOCATE */
+
 	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;
+	/*
+	 * The size will be -1ULL when fill_device is used, so don't truncate
+	 * or fallocate this file, just write it
+	 */
+	if (!td->o.fill_device) {
+		dprint(FD_FILE, "truncate file %s, size %llu\n", f->file_name,
+					(unsigned long long) f->real_file_size);
+		if (ftruncate(f->fd, f->real_file_size) == -1) {
+			if (errno != EFBIG) {
+				td_verror(td, errno, "ftruncate");
+				goto err;
+			}
+		}
 	}
 
-#ifdef FIO_HAVE_FALLOCATE
-	dprint(FD_FILE, "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));
-#endif
-
 	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) {
@@ -85,15 +140,26 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
 		if (bs > left)
 			bs = left;
 
+		fill_io_buffer(td, b, bs, bs);
+
 		r = write(f->fd, b, bs);
 
-		if (r == (int) bs) {
-			left -= bs;
+		if (r > 0) {
+			left -= r;
 			continue;
 		} else {
-			if (r < 0)
+			if (r < 0) {
+				int __e = errno;
+
+				if (__e == ENOSPC) {
+					if (td->o.fill_device)
+						break;
+					log_info("fio: ENOSPC on laying out "
+						 "file, stopping\n");
+					break;
+				}
 				td_verror(td, errno, "write");
-			else
+			} else
 				td_verror(td, EIO, "write");
 
 			break;
@@ -109,6 +175,13 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
 			goto err;
 		}
 	}
+	if (td->o.fill_device && !td_write(td)) {
+		fio_file_clear_size_known(f);
+		if (td_io_get_file_size(td, f))
+			goto err;
+		if (f->io_size > f->real_file_size)
+			f->io_size = f->real_file_size;
+	}
 
 	free(b);
 done:
@@ -121,12 +194,15 @@ err:
 
 static int pre_read_file(struct thread_data *td, struct fio_file *f)
 {
-	int r, did_open = 0, old_runstate;
+	int ret = 0, r, did_open = 0, old_runstate;
 	unsigned long long left;
 	unsigned int bs;
 	char *b;
 
-	if (!(f->flags & FIO_FILE_OPEN)) {
+	if (td->io_ops->flags & FIO_PIPEIO)
+		return 0;
+
+	if (!fio_file_open(f)) {
 		if (td->io_ops->open_file(td, f)) {
 			log_err("fio: cannot pre-read, failed to open file\n");
 			return 1;
@@ -134,14 +210,19 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
 		did_open = 1;
 	}
 
-	old_runstate = td->runstate;
-	td_set_runstate(td, TD_PRE_READING);
+	old_runstate = td_bump_runstate(td, TD_PRE_READING);
 
 	bs = td->o.max_bs[DDIR_READ];
 	b = malloc(bs);
 	memset(b, 0, bs);
 
-	lseek(f->fd, f->file_offset, SEEK_SET);
+	if (lseek(f->fd, f->file_offset, SEEK_SET) < 0) {
+		td_verror(td, errno, "lseek");
+		log_err("fio: failed to lseek pre-read file\n");
+		ret = 1;
+		goto error;
+	}
+
 	left = f->io_size;
 
 	while (left && !td->terminate) {
@@ -159,22 +240,31 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
 		}
 	}
 
-	td_set_runstate(td, old_runstate);
+error:
+	td_restore_runstate(td, old_runstate);
 
 	if (did_open)
 		td->io_ops->close_file(td, f);
+
 	free(b);
-	return 0;
+	return ret;
 }
 
 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;
@@ -195,7 +285,7 @@ static int file_size(struct thread_data *td, struct fio_file *f)
 
 static int bdev_size(struct thread_data *td, struct fio_file *f)
 {
-	unsigned long long bytes;
+	unsigned long long bytes = 0;
 	int r;
 
 	if (td->io_ops->open_file(td, f)) {
@@ -204,7 +294,7 @@ static int bdev_size(struct thread_data *td, struct fio_file *f)
 		return 1;
 	}
 
-	r = blockdev_size(f->fd, &bytes);
+	r = blockdev_size(f, &bytes);
 	if (r) {
 		td_verror(td, r, "blockdev_size");
 		goto err;
@@ -216,23 +306,61 @@ static int bdev_size(struct thread_data *td, struct fio_file *f)
 	}
 
 	f->real_file_size = bytes;
+	td->io_ops->close_file(td, f);
+	return 0;
+err:
+	td->io_ops->close_file(td, f);
+	return 1;
+}
+
+static int char_size(struct thread_data *td, struct fio_file *f)
+{
+#ifdef FIO_HAVE_CHARDEV_SIZE
+	unsigned long long bytes = 0;
+	int r;
+
+	if (td->io_ops->open_file(td, f)) {
+		log_err("fio: failed opening blockdev %s for size check\n",
+			f->file_name);
+		return 1;
+	}
+
+	r = chardev_size(f, &bytes);
+	if (r) {
+		td_verror(td, r, "chardev_size");
+		goto err;
+	}
+
+	if (!bytes) {
+		log_err("%s: zero sized char device?\n", f->file_name);
+		goto err;
+	}
+
+	f->real_file_size = bytes;
+	td->io_ops->close_file(td, f);
 	return 0;
 err:
 	td->io_ops->close_file(td, f);
 	return 1;
+#else
+	f->real_file_size = -1ULL;
+	return 0;
+#endif
 }
 
 static int get_file_size(struct thread_data *td, struct fio_file *f)
 {
 	int ret = 0;
 
-	if (f->flags & FIO_SIZE_KNOWN)
+	if (fio_file_size_known(f))
 		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 if (f->filetype == FIO_TYPE_CHAR)
+		ret = char_size(td, f);
 	else
 		f->real_file_size = -1;
 
@@ -240,12 +368,13 @@ static int get_file_size(struct thread_data *td, struct fio_file *f)
 		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);
+		log_err("%s: offset extends end (%llu > %llu)\n", td->o.name,
+					(unsigned long long) f->file_offset,
+					(unsigned long long) f->real_file_size);
 		return 1;
 	}
 
-	f->flags |= FIO_SIZE_KNOWN;
+	fio_file_set_size_known(f);
 	return 0;
 }
 
@@ -255,26 +384,30 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
 {
 	int ret = 0;
 
-	if (len == -1ULL) {
+	if (len == -1ULL)
 		len = f->io_size;
-		if (len == -1ULL && td->o.fill_device)
-			return 0;
-	}
 	if (off == -1ULL)
 		off = f->file_offset;
 
+	if (len == -1ULL || off == -1ULL)
+		return 0;
+
 	dprint(FD_IO, "invalidate cache %s: %llu/%llu\n", f->file_name, off,
 								len);
 
 	/*
 	 * FIXME: add blockdev flushing too
 	 */
-	if (f->mmap)
-		ret = madvise(f->mmap, len, MADV_DONTNEED);
-	else if (f->filetype == FIO_TYPE_FILE) {
-		ret = fadvise(f->fd, off, len, POSIX_FADV_DONTNEED);
+	if (f->mmap_ptr) {
+		ret = posix_madvise(f->mmap_ptr, f->mmap_sz, POSIX_MADV_DONTNEED);
+#ifdef FIO_MADV_FREE
+		if (f->filetype == FIO_TYPE_BD)
+			(void) posix_madvise(f->mmap_ptr, f->mmap_sz, FIO_MADV_FREE);
+#endif
+	} else if (f->filetype == FIO_TYPE_FILE) {
+		ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED);
 	} else if (f->filetype == FIO_TYPE_BD) {
-		ret = blockdev_invalidate_cache(f->fd);
+		ret = blockdev_invalidate_cache(f);
 		if (ret < 0 && errno == EACCES && geteuid()) {
 			if (!root_warn) {
 				log_err("fio: only root may flush block "
@@ -300,10 +433,10 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
 
 int file_invalidate_cache(struct thread_data *td, struct fio_file *f)
 {
-	if (!(f->flags & FIO_FILE_OPEN))
+	if (!fio_file_open(f))
 		return 0;
 
-	return __file_invalidate_cache(td, f, -1, -1);
+	return __file_invalidate_cache(td, f, -1ULL, -1ULL);
 }
 
 int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f)
@@ -318,10 +451,17 @@ 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;
+	}
+
+	f->engine_data = 0;
 	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;
@@ -333,9 +473,6 @@ static int file_lookup_open(struct fio_file *f, int flags)
 		 * racy, need the __f->lock locked
 		 */
 		f->lock = __f->lock;
-		f->lock_owner = __f->lock_owner;
-		f->lock_batch = __f->lock_batch;
-		f->lock_ddir = __f->lock_ddir;
 		from_hash = 1;
 	} else {
 		dprint(FD_FILE, "file not found in hash %s\n", f->file_name);
@@ -346,6 +483,24 @@ 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;
@@ -354,6 +509,11 @@ 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");
@@ -368,14 +528,24 @@ 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.oatomic) {
+		if (!FIO_O_ATOMIC) {
+			td_verror(td, EINVAL, "OS does not support atomic IO");
+			return 1;
+		}
+		flags |= OS_O_DIRECT | FIO_O_ATOMIC;
+	}
 	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)) {
@@ -389,7 +559,7 @@ 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
@@ -399,30 +569,53 @@ 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) {
 		char buf[FIO_VERROR_SIZE];
 		int __e = errno;
 
-		if (errno == EPERM && (flags & FIO_O_NOATIME)) {
+		if (__e == EPERM && (flags & FIO_O_NOATIME)) {
 			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 pollutes 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;
 		}
 	}
@@ -463,20 +656,112 @@ static int get_file_sizes(struct thread_data *td)
 	return err;
 }
 
+struct fio_mount {
+	struct flist_head list;
+	const char *base;
+	char __base[256];
+	unsigned int key;
+};
+
+/*
+ * Get free number of bytes for each file on each unique mount.
+ */
+static unsigned long long get_fs_free_counts(struct thread_data *td)
+{
+	struct flist_head *n, *tmp;
+	unsigned long long ret = 0;
+	struct fio_mount *fm;
+	FLIST_HEAD(list);
+	struct fio_file *f;
+	unsigned int i;
+
+	for_each_file(td, f, i) {
+		struct stat sb;
+		char buf[256];
+
+		if (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_CHAR) {
+			if (f->real_file_size != -1ULL)
+				ret += f->real_file_size;
+			continue;
+		} else if (f->filetype != FIO_TYPE_FILE)
+			continue;
+
+		strcpy(buf, f->file_name);
+
+		if (stat(buf, &sb) < 0) {
+			if (errno != ENOENT)
+				break;
+			strcpy(buf, ".");
+			if (stat(buf, &sb) < 0)
+				break;
+		}
+
+		fm = NULL;
+		flist_for_each(n, &list) {
+			fm = flist_entry(n, struct fio_mount, list);
+			if (fm->key == sb.st_dev)
+				break;
+
+			fm = NULL;
+		}
+
+		if (fm)
+			continue;
+
+		fm = malloc(sizeof(*fm));
+		strcpy(fm->__base, buf);
+		fm->base = basename(fm->__base);
+		fm->key = sb.st_dev;
+		flist_add(&fm->list, &list);
+	}
+
+	flist_for_each_safe(n, tmp, &list) {
+		unsigned long long sz;
+
+		fm = flist_entry(n, struct fio_mount, list);
+		flist_del(&fm->list);
+
+		sz = get_fs_size(fm->base);
+		if (sz && sz != -1ULL)
+			ret += sz;
+
+		free(fm);
+	}
+
+	return ret;
+}
+
+uint64_t get_start_offset(struct thread_data *td, struct fio_file *f)
+{
+	struct thread_options *o = &td->o;
+
+	if (o->file_append && f->filetype == FIO_TYPE_FILE)
+		return f->real_file_size;
+
+	return td->o.start_offset +
+		(td->thread_number - 1) * td->o.offset_increment;
+}
+
 /*
  * 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 thread_options *o = &td->o;
 	struct fio_file *f;
-	unsigned int i;
+	unsigned int i, nr_fs_extra = 0;
 	int err = 0, need_extend;
+	int old_state;
+	const unsigned int bs = td_min_bs(td);
+	uint64_t fs = 0;
 
 	dprint(FD_FILE, "setup files\n");
 
-	if (td->o.read_iolog_file)
-		return 0;
+	old_state = td_bump_runstate(td, TD_SETTING_UP);
+
+	if (o->read_iolog_file)
+		goto done;
 
 	/*
 	 * if ioengine defines a setup() method, it's responsible for
@@ -489,7 +774,7 @@ int setup_files(struct thread_data *td)
 		err = get_file_sizes(td);
 
 	if (err)
-		return err;
+		goto err_out;
 
 	/*
 	 * check sizes. if the files/devices do not exist and the size
@@ -503,14 +788,32 @@ int setup_files(struct thread_data *td)
 			total_size += f->real_file_size;
 	}
 
+	if (o->fill_device)
+		td->fill_device_size = get_fs_free_counts(td);
+
 	/*
 	 * 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);
+	if ((!total_size || total_size == -1ULL) && !o->size &&
+	    !(td->io_ops->flags & FIO_NOIO) && !o->fill_device &&
+	    !(o->nr_files && (o->file_size_low || o->file_size_high))) {
+		log_err("%s: you need to specify size=\n", o->name);
 		td_verror(td, EINVAL, "total_file_size");
-		return 1;
+		goto err_out;
+	}
+
+	/*
+	 * Calculate per-file size and potential extra size for the
+	 * first files, if needed.
+	 */
+	if (!o->file_size_low) {
+		uint64_t all_fs;
+
+		fs = o->size / o->nr_files;
+		all_fs = fs * o->nr_files;
+
+		if (all_fs < o->size)
+			nr_fs_extra = (o->size - all_fs) / bs;
 	}
 
 	/*
@@ -521,29 +824,35 @@ 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, f);
 
-		if (!td->o.file_size_low) {
+		if (!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.
+			 * total size divided by number of files. If that is
+			 * zero, set it to the real file size. If the size
+			 * doesn't divide nicely with the min blocksize,
+			 * make the first files bigger.
 			 */
-			f->io_size = td->o.size / td->o.nr_files;
+			f->io_size = fs;
+			if (nr_fs_extra) {
+				nr_fs_extra--;
+				f->io_size += bs;
+			}
+
 			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)
+		} else if (f->real_file_size < o->file_size_low ||
+			   f->real_file_size > o->file_size_high) {
+			if (f->file_offset > 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 {
+			if (o->file_size_low == o->file_size_high)
+				f->io_size = o->file_size_low - f->file_offset;
+			else {
 				f->io_size = get_rand_file_size(td)
 						- f->file_offset;
 			}
@@ -552,49 +861,55 @@ int setup_files(struct thread_data *td)
 
 		if (f->io_size == -1ULL)
 			total_size = -1ULL;
-		else
+		else {
+                        if (o->size_percent)
+                                f->io_size = (f->io_size * 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 &&
 		    !(td->io_ops->flags & FIO_DISKLESSIO)) {
-			if (!td->o.create_on_open) {
+			if (!o->create_on_open) {
 				need_extend++;
 				extend_size += (f->io_size + f->file_offset);
 			} else
 				f->real_file_size = f->io_size + f->file_offset;
-			f->flags |= FIO_FILE_EXTEND;
+			fio_file_set_extend(f);
 		}
 	}
 
-	if (!td->o.size || td->o.size > total_size)
-		td->o.size = total_size;
+	if (!o->size || o->size > total_size)
+		o->size = total_size;
 
 	/*
 	 * See if we need to extend some files
 	 */
 	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) /"
-				 " %LuMiB)\n", td->o.name, need_extend,
+				 " %lluMB)\n", o->name, need_extend,
 					extend_size >> 20);
 
 		for_each_file(td, f, i) {
-			unsigned long long old_len, extend_len;
+			unsigned long long old_len = -1ULL, extend_len = -1ULL;
 
-			if (!(f->flags & FIO_FILE_EXTEND))
+			if (!fio_file_extend(f))
 				continue;
 
 			assert(f->filetype == FIO_TYPE_FILE);
-			f->flags &= ~FIO_FILE_EXTEND;
-			old_len = f->real_file_size;
-			extend_len = f->io_size + f->file_offset - old_len;
+			fio_file_clear_extend(f);
+			if (!o->fill_device) {
+				old_len = f->real_file_size;
+				extend_len = f->io_size + f->file_offset -
+						old_len;
+			}
 			f->real_file_size = (f->io_size + f->file_offset);
 			err = extend_file(td, f);
 			if (err)
 				break;
-			
+
 			err = __file_invalidate_cache(td, f, old_len,
 								extend_len);
 			close(f->fd);
@@ -606,20 +921,28 @@ int setup_files(struct thread_data *td)
 	}
 
 	if (err)
-		return err;
+		goto err_out;
 
-	if (!td->o.zone_size)
-		td->o.zone_size = td->o.size;
+	if (!o->zone_size)
+		o->zone_size = 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;
+	if (!o->read_iolog_file)
+		td->total_io_size = o->size * o->loops;
+
+done:
+	if (o->create_only)
+		td->done = 1;
+
+	td_restore_runstate(td, old_state);
 	return 0;
 err_offset:
-	log_err("%s: you need to specify valid offset=\n", td->o.name);
+	log_err("%s: you need to specify valid offset=\n", o->name);
+err_out:
+	td_restore_runstate(td, old_state);
 	return 1;
 }
 
@@ -637,25 +960,78 @@ 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_bump_runstate(td, TD_SETTING_UP);
+
+	for_each_file(td, f, i)
+		__init_rand_distribution(td, f);
+
+	td_restore_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) /
-				(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;
+		uint64_t file_size = min(f->real_file_size, f->io_size);
+
+		blocks = file_size / (unsigned long long) td->o.rw_min_bs;
+
+		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, 0))
+				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'"
@@ -666,7 +1042,6 @@ 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;
@@ -677,8 +1052,10 @@ 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);
+	for_each_file(td, f, i) {
+		if (fio_file_open(f))
+			td_io_close_file(td, f);
+	}
 }
 
 void close_and_free_files(struct thread_data *td)
@@ -689,28 +1066,30 @@ void close_and_free_files(struct thread_data *td)
 	dprint(FD_FILE, "close files\n");
 
 	for_each_file(td, f, i) {
+		if (fio_file_open(f))
+			td_io_close_file(td, f);
+
+		remove_file_hash(f);
+
 		if (td->o.unlink && f->filetype == FIO_TYPE_FILE) {
 			dprint(FD_FILE, "free unlink %s\n", f->file_name);
 			unlink(f->file_name);
 		}
 
-		td_io_close_file(td, f);
-		remove_file_hash(f);
-
 		sfree(f->file_name);
 		f->file_name = NULL;
-
-		if (f->file_map) {
-			sfree(f->file_map);
-			f->file_map = NULL;
-		}
+		axmap_free(f->io_axmap);
+		f->io_axmap = NULL;
 		sfree(f);
 	}
 
 	td->o.filename = NULL;
 	free(td->files);
+	free(td->file_locks);
 	td->files_index = 0;
 	td->files = NULL;
+	td->file_locks = NULL;
+	td->o.file_lock_mode = FILE_LOCK_NONE;
 	td->o.nr_files = 0;
 }
 
@@ -723,7 +1102,12 @@ static void get_file_type(struct fio_file *f)
 	else
 		f->filetype = FIO_TYPE_FILE;
 
-	if (!lstat(f->file_name, &sb)) {
+	/* \\.\ 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)) {
 		if (S_ISBLK(sb.st_mode))
 			f->filetype = FIO_TYPE_BD;
 		else if (S_ISCHR(sb.st_mode))
@@ -733,32 +1117,133 @@ static void get_file_type(struct fio_file *f)
 	}
 }
 
-int add_file(struct thread_data *td, const char *fname)
+static int __is_already_allocated(const char *fname)
 {
-	int cur_files = td->files_index;
-	char file_name[PATH_MAX];
-	struct fio_file *f;
-	int len = 0;
+	struct flist_head *entry;
+	char *filename;
 
-	dprint(FD_FILE, "add file %s\n", fname);
+	if (flist_empty(&filename_list))
+		return 0;
+
+	flist_for_each(entry, &filename_list) {
+		filename = flist_entry(entry, struct file_name, list)->filename;
+
+		if (strcmp(filename, fname) == 0)
+			return 1;
+	}
+
+	return 0;
+}
+
+static int is_already_allocated(const char *fname)
+{
+	int ret;
+
+	fio_file_hash_lock();
+	ret = __is_already_allocated(fname);
+	fio_file_hash_unlock();
+	return ret;
+}
+
+static void set_already_allocated(const char *fname)
+{
+	struct file_name *fn;
+
+	fn = malloc(sizeof(struct file_name));
+	fn->filename = strdup(fname);
+
+	fio_file_hash_lock();
+	if (!__is_already_allocated(fname)) {
+		flist_add_tail(&fn->list, &filename_list);
+		fn = NULL;
+	}
+	fio_file_hash_unlock();
+
+	if (fn) {
+		free(fn->filename);
+		free(fn);
+	}
+}
+
+
+static void free_already_allocated(void)
+{
+	struct flist_head *entry, *tmp;
+	struct file_name *fn;
+
+	if (flist_empty(&filename_list))
+		return;
+
+	fio_file_hash_lock();
+	flist_for_each_safe(entry, tmp, &filename_list) {
+		fn = flist_entry(entry, struct file_name, list);
+		free(fn->filename);
+		flist_del(&fn->list);
+		free(fn);
+	}
+
+	fio_file_hash_unlock();
+}
+
+static struct fio_file *alloc_new_file(struct thread_data *td)
+{
+	struct fio_file *f;
 
 	f = smalloc(sizeof(*f));
 	if (!f) {
 		log_err("fio: smalloc OOM\n");
 		assert(0);
+		return NULL;
 	}
-		
+
 	f->fd = -1;
+	f->shadow_fd = -1;
+	fio_file_reset(td, f);
+	return f;
+}
+
+int add_file(struct thread_data *td, const char *fname, int numjob, int inc)
+{
+	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);
+
+	if (td->o.directory)
+		len = set_name_idx(file_name, td->o.directory, numjob);
+
+	sprintf(file_name + len, "%s", fname);
+
+	/* clean cloned siblings using existing files */
+	if (numjob && is_already_allocated(file_name))
+		return 0;
+
+	f = alloc_new_file(td);
 
 	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);
+		}
+		if (td->o.file_lock_mode != FILE_LOCK_NONE) {
+			td->file_locks = realloc(td->file_locks, new_size);
+			if (!td->file_locks) {
+				log_err("fio: realloc OOM\n");
+				assert(0);
+			}
+			td->file_locks[cur_files] = FILE_LOCK_NONE;
+		}
 		td->files_size = new_size;
 	}
 	td->files[cur_files] = f;
+	f->fileno = cur_files;
 
 	/*
 	 * init function, io engine may not be loaded yet
@@ -766,26 +1251,22 @@ int add_file(struct thread_data *td, const char *fname)
 	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);
 	if (!f->file_name) {
 		log_err("fio: smalloc OOM\n");
 		assert(0);
 	}
-	
+
 	get_file_type(f);
 
 	switch (td->o.file_lock_mode) {
 	case FILE_LOCK_NONE:
 		break;
 	case FILE_LOCK_READWRITE:
-		f->lock = fio_mutex_rw_init();
+		f->rwlock = fio_rwlock_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);
@@ -796,16 +1277,41 @@ int add_file(struct thread_data *td, const char *fname)
 	if (f->filetype == FIO_TYPE_FILE)
 		td->nr_normal_files++;
 
+	set_already_allocated(file_name);
+
+	/*
+	 * For adding files after the fact - if openfiles= isn't
+	 * given as an option, ensure we allow at least one file open
+	 */
+	if (!td->o.open_files)
+		td->o.open_files = 1;
+
+	if (inc)
+		td->o.nr_files++;
+
 	dprint(FD_FILE, "file %p \"%s\" added at %d\n", f, f->file_name,
 							cur_files);
 
 	return cur_files;
 }
 
+int add_file_exclusive(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 add_file(td, fname, 0, 1);
+}
+
 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);
+	assert(fio_file_open(f));
 	f->references++;
 }
 
@@ -815,8 +1321,10 @@ int put_file(struct thread_data *td, struct fio_file *f)
 
 	dprint(FD_FILE, "put file %s, ref=%d\n", f->file_name, f->references);
 
-	if (!(f->flags & FIO_FILE_OPEN))
+	if (!fio_file_open(f)) {
+		assert(f->fd == -1);
 		return 0;
+	}
 
 	assert(f->references);
 	if (--f->references)
@@ -832,7 +1340,8 @@ int put_file(struct thread_data *td, struct fio_file *f)
 		ret = f_ret;
 
 	td->nr_open_files--;
-	f->flags &= ~FIO_FILE_OPEN;
+	fio_file_clear_open(f);
+	assert(f->fd == -1);
 	return ret;
 }
 
@@ -841,57 +1350,36 @@ void lock_file(struct thread_data *td, struct fio_file *f, enum fio_ddir ddir)
 	if (!f->lock || td->o.file_lock_mode == FILE_LOCK_NONE)
 		return;
 
-	if (f->lock_owner == td && f->lock_batch--)
-		return;
-
 	if (td->o.file_lock_mode == FILE_LOCK_READWRITE) {
 		if (ddir == DDIR_READ)
-			fio_mutex_down_read(f->lock);
+			fio_rwlock_read(f->rwlock);
 		else
-			fio_mutex_down_write(f->lock);
+			fio_rwlock_write(f->rwlock);
 	} else if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE)
 		fio_mutex_down(f->lock);
 
-	f->lock_owner = td;
-	f->lock_batch = td->o.lockfile_batch;
-	f->lock_ddir = ddir;
+	td->file_locks[f->fileno] = td->o.file_lock_mode;
 }
 
 void unlock_file(struct thread_data *td, struct fio_file *f)
 {
 	if (!f->lock || td->o.file_lock_mode == FILE_LOCK_NONE)
 		return;
-	if (f->lock_batch)
-		return;
-
-	if (td->o.file_lock_mode == FILE_LOCK_READWRITE) {
-		const int is_read = f->lock_ddir == DDIR_READ;
-		int val = fio_mutex_getval(f->lock);
-
-		if ((is_read && val == 1) || (!is_read && val == -1))
-			f->lock_owner = NULL;
-
-		if (is_read)
-			fio_mutex_up_read(f->lock);
-		else
-			fio_mutex_up_write(f->lock);
-	} else if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE) {
-		int val = fio_mutex_getval(f->lock);
-
-		if (val == 0)
-			f->lock_owner = NULL;
 
+	if (td->o.file_lock_mode == FILE_LOCK_READWRITE)
+		fio_rwlock_unlock(f->rwlock);
+	else if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE)
 		fio_mutex_up(f->lock);
-	}
+
+	td->file_locks[f->fileno] = FILE_LOCK_NONE;
 }
 
 void unlock_file_all(struct thread_data *td, struct fio_file *f)
 {
-	if (f->lock_owner != td)
+	if (td->o.file_lock_mode == FILE_LOCK_NONE || !td->file_locks)
 		return;
-
-	f->lock_batch = 0;
-	unlock_file(td, f);
+	if (td->file_locks[f->fileno] != FILE_LOCK_NONE)
+		unlock_file(td, f);
 }
 
 static int recurse_dir(struct thread_data *td, const char *dirname)
@@ -904,7 +1392,7 @@ 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;
 	}
@@ -916,7 +1404,7 @@ 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) {
@@ -926,8 +1414,7 @@ static int recurse_dir(struct thread_data *td, const char *dirname)
 		}
 
 		if (S_ISREG(sb.st_mode)) {
-			add_file(td, full_path);
-			td->o.nr_files++;
+			add_file(td, full_path, 0, 1);
 			continue;
 		}
 		if (!S_ISDIR(sb.st_mode))
@@ -964,25 +1451,29 @@ void dup_files(struct thread_data *td, struct thread_data *org)
 
 	td->files = malloc(org->files_index * sizeof(f));
 
+	if (td->o.file_lock_mode != FILE_LOCK_NONE)
+		td->file_locks = malloc(org->files_index);
+
 	for_each_file(org, f, i) {
 		struct fio_file *__f;
 
-		__f = smalloc(sizeof(*__f));
-		if (!__f) {
-			log_err("fio: smalloc OOM\n");
-			assert(0);
-		}
-	
+		__f = alloc_new_file(td);
+
 		if (f->file_name) {
 			__f->file_name = smalloc_strdup(f->file_name);
 			if (!__f->file_name) {
 				log_err("fio: smalloc OOM\n");
 				assert(0);
 			}
-	
+
 			__f->filetype = f->filetype;
 		}
 
+		if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE)
+			__f->lock = f->lock;
+		else if (td->o.file_lock_mode == FILE_LOCK_READWRITE)
+			__f->rwlock = f->rwlock;
+
 		td->files[i] = __f;
 	}
 }
@@ -1011,3 +1502,31 @@ 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]);
+}
+
+int fio_files_done(struct thread_data *td)
+{
+	struct fio_file *f;
+	unsigned int i;
+
+	for_each_file(td, f, i)
+		if (!fio_file_done(f))
+			return 0;
+
+	return 1;
+}
+
+/* free memory used in initialization phase only */
+void filesetup_mem_free(void)
+{
+	free_already_allocated();
+}