X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=ioengines.c;h=7ab356add79f32629a6fc9a2c8fa543d81d597af;hp=c1648a016ee29bcb46b7994b039cc1c652852e9d;hb=f3e11d0575df5b93cd09f8658964e1fb5d38a774;hpb=860a3044085c676632f9832a78133cd9847cbe62

diff --git a/ioengines.c b/ioengines.c
index c1648a01..7ab356ad 100644
--- a/ioengines.c
+++ b/ioengines.c
@@ -17,23 +17,18 @@
 #include <assert.h>
 
 #include "fio.h"
-#include "os.h"
+#include "diskutil.h"
 
-static LIST_HEAD(engine_list);
+static FLIST_HEAD(engine_list);
 
 static int check_engine_ops(struct ioengine_ops *ops)
 {
 	if (ops->version != FIO_IOOPS_VERSION) {
-		log_err("bad ioops version %d (want %d)\n", ops->version, FIO_IOOPS_VERSION);
+		log_err("bad ioops version %d (want %d)\n", ops->version,
+							FIO_IOOPS_VERSION);
 		return 1;
 	}
 
-	/*
-	 * cpu thread doesn't need to provide anything
-	 */
-	if (ops->flags & FIO_CPUIO)
-		return 0;
-
 	if (!ops->queue) {
 		log_err("%s: no queue handler\n", ops->name);
 		return 1;
@@ -44,7 +39,7 @@ static int check_engine_ops(struct ioengine_ops *ops)
 	 */
 	if (ops->flags & FIO_SYNCIO)
 		return 0;
-	
+
 	if (!ops->event) {
 		log_err("%s: no event handler\n", ops->name);
 		return 1;
@@ -57,29 +52,31 @@ static int check_engine_ops(struct ioengine_ops *ops)
 		log_err("%s: no queue handler\n", ops->name);
 		return 1;
 	}
-		
+
 	return 0;
 }
 
 void unregister_ioengine(struct ioengine_ops *ops)
 {
-	list_del(&ops->list);
-	INIT_LIST_HEAD(&ops->list);
+	dprint(FD_IO, "ioengine %s unregistered\n", ops->name);
+	flist_del(&ops->list);
+	INIT_FLIST_HEAD(&ops->list);
 }
 
 void register_ioengine(struct ioengine_ops *ops)
 {
-	INIT_LIST_HEAD(&ops->list);
-	list_add_tail(&ops->list, &engine_list);
+	dprint(FD_IO, "ioengine %s registered\n", ops->name);
+	INIT_FLIST_HEAD(&ops->list);
+	flist_add_tail(&ops->list, &engine_list);
 }
 
 static struct ioengine_ops *find_ioengine(const char *name)
 {
 	struct ioengine_ops *ops;
-	struct list_head *entry;
+	struct flist_head *entry;
 
-	list_for_each(entry, &engine_list) {
-		ops = list_entry(entry, struct ioengine_ops, list);
+	flist_for_each(entry, &engine_list) {
+		ops = flist_entry(entry, struct ioengine_ops, list);
 		if (!strcmp(name, ops->name))
 			return ops;
 	}
@@ -93,6 +90,8 @@ static struct ioengine_ops *dlopen_ioengine(struct thread_data *td,
 	struct ioengine_ops *ops;
 	void *dlhandle;
 
+	dprint(FD_IO, "dload engine %s\n", engine_lib);
+
 	dlerror();
 	dlhandle = dlopen(engine_lib, RTLD_LAZY);
 	if (!dlhandle) {
@@ -120,6 +119,8 @@ struct ioengine_ops *load_ioengine(struct thread_data *td, const char *name)
 	struct ioengine_ops *ops, *ret;
 	char engine[16];
 
+	dprint(FD_IO, "load ioengine %s\n", name);
+
 	strncpy(engine, name, sizeof(engine) - 1);
 
 	/*
@@ -152,72 +153,132 @@ struct ioengine_ops *load_ioengine(struct thread_data *td, const char *name)
 
 void close_ioengine(struct thread_data *td)
 {
-	if (td->io_ops->cleanup)
+	dprint(FD_IO, "close ioengine %s\n", td->io_ops->name);
+
+	if (td->io_ops->cleanup) {
 		td->io_ops->cleanup(td);
+		td->io_ops->data = NULL;
+	}
 
 	if (td->io_ops->dlhandle)
 		dlclose(td->io_ops->dlhandle);
 
-#if 0
-	/* we can't do this for threads, so just leak it, it's exiting */
 	free(td->io_ops);
-#endif
 	td->io_ops = NULL;
 }
 
 int td_io_prep(struct thread_data *td, struct io_u *io_u)
 {
-	if (td->io_ops->prep)
-		return td->io_ops->prep(td, io_u);
+	dprint_io_u(io_u, "prep");
+	fio_ro_check(td, io_u);
+
+	lock_file(td, io_u->file, io_u->ddir);
+
+	if (td->io_ops->prep) {
+		int ret = td->io_ops->prep(td, io_u);
+
+		dprint(FD_IO, "->prep(%p)=%d\n", io_u, ret);
+		if (ret)
+			unlock_file(td, io_u->file);
+		return ret;
+	}
 
 	return 0;
 }
 
-int td_io_getevents(struct thread_data *td, int min, int max,
+int td_io_getevents(struct thread_data *td, unsigned int min, unsigned int max,
 		    struct timespec *t)
 {
-	if (min > 0 && td->io_ops->commit) {
-		int r = td->io_ops->commit(td);
+	int r = 0;
 
+	if (min > 0 && td->io_ops->commit) {
+		r = td->io_ops->commit(td);
 		if (r < 0)
-			return r;
+			goto out;
 	}
-	if (td->io_ops->getevents)
-		return td->io_ops->getevents(td, min, max, t);
-
-	return 0;
+	if (max > td->cur_depth)
+		max = td->cur_depth;
+	if (min > max)
+		max = min;
+
+	r = 0;
+	if (max && td->io_ops->getevents)
+		r = td->io_ops->getevents(td, min, max, t);
+out:
+	if (r >= 0)
+		io_u_mark_complete(td, r);
+	else
+		td_verror(td, -r, "get_events");
+
+	dprint(FD_IO, "getevents: %d\n", r);
+	return r;
 }
 
 int td_io_queue(struct thread_data *td, struct io_u *io_u)
 {
 	int ret;
 
+	dprint_io_u(io_u, "queue");
+	fio_ro_check(td, io_u);
+
 	assert((io_u->flags & IO_U_F_FLIGHT) == 0);
 	io_u->flags |= IO_U_F_FLIGHT;
 
+	assert(fio_file_open(io_u->file));
+
 	io_u->error = 0;
 	io_u->resid = 0;
 
 	if (td->io_ops->flags & FIO_SYNCIO) {
-		fio_gettime(&io_u->issue_time, NULL);
+		if (fio_fill_issue_time(td))
+			fio_gettime(&io_u->issue_time, NULL);
 
 		/*
-		 * for a sync engine, set the timeout upfront
+		 * only used for iolog
 		 */
-		if (mtime_since(&td->timeout_end, &io_u->issue_time) < IO_U_TIMEOUT)
-			io_u_set_timeout(td);
+		if (td->o.read_iolog_file)
+			memcpy(&td->last_issue, &io_u->issue_time,
+					sizeof(struct timeval));
 	}
 
-	if (io_u->ddir != DDIR_SYNC)
+	if (!ddir_sync(io_u->ddir))
 		td->io_issues[io_u->ddir]++;
 
 	ret = td->io_ops->queue(td, io_u);
 
-	if (ret == FIO_Q_QUEUED) {
+	unlock_file(td, io_u->file);
+
+	/*
+	 * Add warning for O_DIRECT so that users have an easier time
+	 * spotting potentially bad alignment. If this triggers for the first
+	 * IO, then it's likely an alignment problem or because the host fs
+	 * does not support O_DIRECT
+	 */
+	if (io_u->error == EINVAL && td->io_issues[io_u->ddir] == 1 &&
+	    td->o.odirect) {
+		log_info("fio: first direct IO errored. File system may not "
+			 "support direct IO, or iomem_align= is bad.\n");
+	}
+
+	if (!td->io_ops->commit) {
+		io_u_mark_submit(td, 1);
+		io_u_mark_complete(td, 1);
+	}
+
+	if (ret == FIO_Q_COMPLETED) {
+		if (!ddir_sync(io_u->ddir)) {
+			io_u_mark_depth(td, 1);
+			td->ts.total_io_u[io_u->ddir]++;
+		}
+	} else if (ret == FIO_Q_QUEUED) {
 		int r;
 
-		td->io_u_queued++;
-		if (td->io_u_queued > td->iodepth_batch) {
+		if (!ddir_sync(io_u->ddir)) {
+			td->io_u_queued++;
+			td->ts.total_io_u[io_u->ddir]++;
+		}
+
+		if (td->io_u_queued >= td->o.iodepth_batch) {
 			r = td_io_commit(td);
 			if (r < 0)
 				return r;
@@ -225,14 +286,15 @@ int td_io_queue(struct thread_data *td, struct io_u *io_u)
 	}
 
 	if ((td->io_ops->flags & FIO_SYNCIO) == 0) {
-		fio_gettime(&io_u->issue_time, NULL);
+		if (fio_fill_issue_time(td))
+			fio_gettime(&io_u->issue_time, NULL);
 
 		/*
-		 * async engine, set the timeout here
+		 * only used for iolog
 		 */
-		if (ret == FIO_Q_QUEUED &&
-		    mtime_since(&td->timeout_end, &io_u->issue_time) < IO_U_TIMEOUT)
-			io_u_set_timeout(td);
+		if (td->o.read_iolog_file)
+			memcpy(&td->last_issue, &io_u->issue_time,
+					sizeof(struct timeval));
 	}
 
 	return ret;
@@ -240,41 +302,179 @@ int td_io_queue(struct thread_data *td, struct io_u *io_u)
 
 int td_io_init(struct thread_data *td)
 {
-	if (td->io_ops->init)
-		return td->io_ops->init(td);
+	int ret = 0;
 
-	return 0;
+	if (td->io_ops->init) {
+		ret = td->io_ops->init(td);
+		if (ret && td->o.iodepth > 1) {
+			log_err("fio: io engine init failed. Perhaps try"
+				" reducing io depth?\n");
+		}
+	}
+
+	return ret;
 }
 
 int td_io_commit(struct thread_data *td)
 {
-	if (!td->cur_depth)
+	int ret;
+
+	dprint(FD_IO, "calling ->commit(), depth %d\n", td->cur_depth);
+
+	if (!td->cur_depth || !td->io_u_queued)
 		return 0;
 
+	io_u_mark_depth(td, td->io_u_queued);
 	td->io_u_queued = 0;
-	if (td->io_ops->commit)
-		return td->io_ops->commit(td);
+
+	if (td->io_ops->commit) {
+		ret = td->io_ops->commit(td);
+		if (ret)
+			td_verror(td, -ret, "io commit");
+	}
 
 	return 0;
 }
 
 int td_io_open_file(struct thread_data *td, struct fio_file *f)
 {
-	if (!td->io_ops->open_file(td, f)) {
-		f->open = 1;
-		td->nr_open_files++;
-		return 0;
+	assert(!fio_file_open(f));
+	assert(f->fd == -1);
+
+	if (td->io_ops->open_file(td, f)) {
+		if (td->error == EINVAL && td->o.odirect)
+			log_err("fio: destination does not support O_DIRECT\n");
+		if (td->error == EMFILE) {
+			log_err("fio: try reducing/setting openfiles (failed"
+				" at %u of %u)\n", td->nr_open_files,
+							td->o.nr_files);
+		}
+
+		assert(f->fd == -1);
+		assert(!fio_file_open(f));
+		return 1;
 	}
 
+	fio_file_reset(f);
+	fio_file_set_open(f);
+	fio_file_clear_closing(f);
+	disk_util_inc(f->du);
+
+	td->nr_open_files++;
+	get_file(f);
+
+	if (f->filetype == FIO_TYPE_PIPE) {
+		if (td_random(td)) {
+			log_err("fio: can't seek on pipes (no random io)\n");
+			goto err;
+		}
+	}
+
+	if (td->io_ops->flags & FIO_DISKLESSIO)
+		goto done;
+
+	if (td->o.invalidate_cache && file_invalidate_cache(td, f))
+		goto err;
+
+	if (td->o.fadvise_hint &&
+	    (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_FILE)) {
+		int flags;
+
+		if (td_random(td))
+			flags = POSIX_FADV_RANDOM;
+		else
+			flags = POSIX_FADV_SEQUENTIAL;
+
+		if (fadvise(f->fd, f->file_offset, f->io_size, flags) < 0) {
+			td_verror(td, errno, "fadvise");
+			goto err;
+		}
+	}
+
+#ifdef FIO_OS_DIRECTIO
+	/*
+	 * Some OS's have a distinct call to mark the file non-buffered,
+	 * instead of using O_DIRECT (Solaris)
+	 */
+	if (td->o.odirect) {
+		int ret = fio_set_odirect(f->fd);
+
+		if (ret) {
+			td_verror(td, ret, "fio_set_odirect");
+			goto err;
+		}
+	}
+#endif
+
+done:
+	log_file(td, f, FIO_LOG_OPEN_FILE);
+	return 0;
+err:
+	disk_util_dec(f->du);
+	if (td->io_ops->close_file)
+		td->io_ops->close_file(td, f);
 	return 1;
 }
 
-void td_io_close_file(struct thread_data *td, struct fio_file *f)
+int td_io_close_file(struct thread_data *td, struct fio_file *f)
 {
-	if (f->open) {
-		if (td->io_ops->close_file)
-			td->io_ops->close_file(td, f);
-		td->nr_open_files--;
-		f->open = 0;
+	if (!fio_file_closing(f))
+		log_file(td, f, FIO_LOG_CLOSE_FILE);
+
+	/*
+	 * mark as closing, do real close when last io on it has completed
+	 */
+	fio_file_set_closing(f);
+
+	disk_util_dec(f->du);
+	unlock_file_all(td, f);
+
+	return put_file(td, f);
+}
+
+int td_io_get_file_size(struct thread_data *td, struct fio_file *f)
+{
+	if (!td->io_ops->get_file_size)
+		return 0;
+
+	return td->io_ops->get_file_size(td, f);
+}
+
+static int do_sync_file_range(struct thread_data *td, struct fio_file *f)
+{
+	off64_t offset, nbytes;
+
+	offset = f->first_write;
+	nbytes = f->last_write - f->first_write;
+
+	if (!nbytes)
+		return 0;
+
+	return sync_file_range(f->fd, offset, nbytes, td->o.sync_file_range);
+}
+
+int do_io_u_sync(struct thread_data *td, struct io_u *io_u)
+{
+	int ret;
+
+	if (io_u->ddir == DDIR_SYNC) {
+		ret = fsync(io_u->file->fd);
+	} else if (io_u->ddir == DDIR_DATASYNC) {
+#ifdef FIO_HAVE_FDATASYNC
+		ret = fdatasync(io_u->file->fd);
+#else
+		ret = io_u->xfer_buflen;
+		io_u->error = EINVAL;
+#endif
+	} else if (io_u->ddir == DDIR_SYNC_FILE_RANGE)
+		ret = do_sync_file_range(td, io_u->file);
+	else {
+		ret = io_u->xfer_buflen;
+		io_u->error = EINVAL;
 	}
+
+	if (ret < 0)
+		io_u->error = errno;
+
+	return ret;
 }