X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=ioengines.c;h=e2316ee4e391d0d5899c960552fec4d19aa33ca2;hp=1b58168608046390146765b7f546c9b9d964736e;hb=c16dc793a3c45780f67ce65244b6e91323dee014;hpb=c7334fa3f3be87854354044615b0c0e473c50713

diff --git a/ioengines.c b/ioengines.c
index 1b581686..e2316ee4 100644
--- a/ioengines.c
+++ b/ioengines.c
@@ -9,20 +9,29 @@
  * generic io engine that could be used for other projects.
  *
  */
-#include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <string.h>
 #include <dlfcn.h>
 #include <fcntl.h>
 #include <assert.h>
+#include <sys/types.h>
+#include <dirent.h>
 
 #include "fio.h"
 #include "diskutil.h"
+#include "zbd.h"
 
 static FLIST_HEAD(engine_list);
 
-static bool check_engine_ops(struct ioengine_ops *ops)
+static inline bool async_ioengine_sync_trim(struct thread_data *td,
+					    struct io_u	*io_u)
+{
+	return td_ioengine_flagged(td, FIO_ASYNCIO_SYNC_TRIM) &&
+		io_u->ddir == DDIR_TRIM;
+}
+
+static bool check_engine_ops(struct thread_data *td, struct ioengine_ops *ops)
 {
 	if (ops->version != FIO_IOOPS_VERSION) {
 		log_err("bad ioops version %d (want %d)\n", ops->version,
@@ -41,6 +50,16 @@ static bool check_engine_ops(struct ioengine_ops *ops)
 	if (ops->flags & FIO_SYNCIO)
 		return false;
 
+	/*
+	 * async engines aren't reliable with offload
+	 */
+	if ((td->o.io_submit_mode == IO_MODE_OFFLOAD) &&
+	    (ops->flags & FIO_NO_OFFLOAD)) {
+		log_err("%s: can't be used with offloaded submit. Use a sync "
+			"engine\n", ops->name);
+		return true;
+	}
+
 	if (!ops->event || !ops->getevents) {
 		log_err("%s: no event/getevents handler\n", ops->name);
 		return true;
@@ -52,14 +71,12 @@ static bool check_engine_ops(struct ioengine_ops *ops)
 void unregister_ioengine(struct ioengine_ops *ops)
 {
 	dprint(FD_IO, "ioengine %s unregistered\n", ops->name);
-	flist_del(&ops->list);
-	INIT_FLIST_HEAD(&ops->list);
+	flist_del_init(&ops->list);
 }
 
 void register_ioengine(struct ioengine_ops *ops)
 {
 	dprint(FD_IO, "ioengine %s registered\n", ops->name);
-	INIT_FLIST_HEAD(&ops->list);
 	flist_add_tail(&ops->list, &engine_list);
 }
 
@@ -77,19 +94,46 @@ static struct ioengine_ops *find_ioengine(const char *name)
 	return NULL;
 }
 
+#ifdef CONFIG_DYNAMIC_ENGINES
+static void *dlopen_external(struct thread_data *td, const char *engine)
+{
+	char engine_path[PATH_MAX];
+	void *dlhandle;
+
+	sprintf(engine_path, "%s/fio-%s.so", FIO_EXT_ENG_DIR, engine);
+
+	dprint(FD_IO, "dlopen external %s\n", engine_path);
+	dlhandle = dlopen(engine_path, RTLD_LAZY);
+	if (!dlhandle)
+		log_info("Engine %s not found; Either name is invalid, was not built, or fio-engine-%s package is missing.\n",
+			 engine, engine);
+
+	return dlhandle;
+}
+#else
+#define dlopen_external(td, engine) (NULL)
+#endif
+
 static struct ioengine_ops *dlopen_ioengine(struct thread_data *td,
 					    const char *engine_lib)
 {
 	struct ioengine_ops *ops;
 	void *dlhandle;
 
-	dprint(FD_IO, "dload engine %s\n", engine_lib);
+	if (!strncmp(engine_lib, "linuxaio", 8) ||
+	    !strncmp(engine_lib, "aio", 3))
+		engine_lib = "libaio";
+
+	dprint(FD_IO, "dlopen engine %s\n", engine_lib);
 
 	dlerror();
 	dlhandle = dlopen(engine_lib, RTLD_LAZY);
 	if (!dlhandle) {
-		td_vmsg(td, -1, dlerror(), "dlopen");
-		return NULL;
+		dlhandle = dlopen_external(td, engine_lib);
+		if (!dlhandle) {
+			td_vmsg(td, -1, dlerror(), "dlopen");
+			return NULL;
+		}
 	}
 
 	/*
@@ -119,30 +163,54 @@ static struct ioengine_ops *dlopen_ioengine(struct thread_data *td,
 		return NULL;
 	}
 
-	td->io_ops_dlhandle = dlhandle;
+	ops->dlhandle = dlhandle;
 	return ops;
 }
 
-struct ioengine_ops *load_ioengine(struct thread_data *td, const char *name)
+static struct ioengine_ops *__load_ioengine(const char *engine)
 {
-	struct ioengine_ops *ops;
-	char engine[64];
+	/*
+	 * linux libaio has alias names, so convert to what we want
+	 */
+	if (!strncmp(engine, "linuxaio", 8) || !strncmp(engine, "aio", 3)) {
+		dprint(FD_IO, "converting ioengine name: %s -> libaio\n",
+		       engine);
+		engine = "libaio";
+	}
 
-	dprint(FD_IO, "load ioengine %s\n", name);
+	dprint(FD_IO, "load ioengine %s\n", engine);
+	return find_ioengine(engine);
+}
 
-	engine[sizeof(engine) - 1] = '\0';
-	strncpy(engine, name, sizeof(engine) - 1);
+struct ioengine_ops *load_ioengine(struct thread_data *td)
+{
+	struct ioengine_ops *ops = NULL;
+	const char *name;
 
 	/*
-	 * linux libaio has alias names, so convert to what we want
+	 * Use ->ioengine_so_path if an external ioengine path is specified.
+	 * In this case, ->ioengine is "external" which also means the prefix
+	 * for external ioengines "external:" is properly used.
 	 */
-	if (!strncmp(engine, "linuxaio", 8) || !strncmp(engine, "aio", 3))
-		strcpy(engine, "libaio");
+	name = td->o.ioengine_so_path ?: td->o.ioengine;
 
-	ops = find_ioengine(engine);
-	if (!ops)
+	/*
+	 * Try to load ->ioengine first, and if failed try to dlopen(3) either
+	 * ->ioengine or ->ioengine_so_path.  This is redundant for an external
+	 * ioengine with prefix, and also leaves the possibility of unexpected
+	 * behavior (e.g. if the "external" ioengine exists), but we do this
+	 * so as not to break job files not using the prefix.
+	 */
+	ops = __load_ioengine(td->o.ioengine);
+
+	/* We do re-dlopen existing handles, for reference counting */
+	if (!ops || ops->dlhandle)
 		ops = dlopen_ioengine(td, name);
 
+	/*
+	 * If ops is NULL, we failed to load ->ioengine, and also failed to
+	 * dlopen(3) either ->ioengine or ->ioengine_so_path as a path.
+	 */
 	if (!ops) {
 		log_err("fio: engine %s not loadable\n", name);
 		return NULL;
@@ -151,7 +219,7 @@ struct ioengine_ops *load_ioengine(struct thread_data *td, const char *name)
 	/*
 	 * Check that the required methods are there.
 	 */
-	if (check_engine_ops(ops))
+	if (check_engine_ops(td, ops))
 		return NULL;
 
 	return ops;
@@ -162,6 +230,8 @@ struct ioengine_ops *load_ioengine(struct thread_data *td, const char *name)
  */
 void free_ioengine(struct thread_data *td)
 {
+	assert(td != NULL && td->io_ops != NULL);
+
 	dprint(FD_IO, "free ioengine %s\n", td->io_ops->name);
 
 	if (td->eo && td->io_ops->options) {
@@ -170,8 +240,10 @@ void free_ioengine(struct thread_data *td)
 		td->eo = NULL;
 	}
 
-	if (td->io_ops_dlhandle)
-		dlclose(td->io_ops_dlhandle);
+	if (td->io_ops->dlhandle) {
+		dprint(FD_IO, "dlclose ioengine %s\n", td->io_ops->name);
+		dlclose(td->io_ops->dlhandle);
+	}
 
 	td->io_ops = NULL;
 }
@@ -198,7 +270,8 @@ int td_io_prep(struct thread_data *td, struct io_u *io_u)
 	if (td->io_ops->prep) {
 		int ret = td->io_ops->prep(td, io_u);
 
-		dprint(FD_IO, "->prep(%p)=%d\n", io_u, ret);
+		dprint(FD_IO, "prep: io_u %p: ret=%d\n", io_u, ret);
+
 		if (ret)
 			unlock_file(td, io_u->file);
 		return ret;
@@ -250,11 +323,11 @@ out:
 	return r;
 }
 
-int td_io_queue(struct thread_data *td, struct io_u *io_u)
+enum fio_q_status td_io_queue(struct thread_data *td, struct io_u *io_u)
 {
 	const enum fio_ddir ddir = acct_ddir(io_u);
-	unsigned long buflen = io_u->xfer_buflen;
-	int ret;
+	unsigned long long buflen = io_u->xfer_buflen;
+	enum fio_q_status ret;
 
 	dprint_io_u(io_u, "queue");
 	fio_ro_check(td, io_u);
@@ -262,6 +335,17 @@ int td_io_queue(struct thread_data *td, struct io_u *io_u)
 	assert((io_u->flags & IO_U_F_FLIGHT) == 0);
 	io_u_set(td, io_u, IO_U_F_FLIGHT);
 
+	/*
+	 * If overlap checking was enabled in offload mode we
+	 * can release this lock that was acquired when we
+	 * started the overlap check because the IO_U_F_FLIGHT
+	 * flag is now set
+	 */
+	if (td_offload_overlap(td)) {
+		int res = pthread_mutex_unlock(&overlap_check);
+		assert(res == 0);
+	}
+
 	assert(fio_file_open(io_u->file));
 
 	/*
@@ -272,25 +356,31 @@ int td_io_queue(struct thread_data *td, struct io_u *io_u)
 	io_u->error = 0;
 	io_u->resid = 0;
 
-	if (td_ioengine_flagged(td, FIO_SYNCIO)) {
-		if (fio_fill_issue_time(td))
+	if (td_ioengine_flagged(td, FIO_SYNCIO) ||
+		async_ioengine_sync_trim(td, io_u)) {
+		if (fio_fill_issue_time(td)) {
 			fio_gettime(&io_u->issue_time, NULL);
 
-		/*
-		 * only used for iolog
-		 */
-		if (td->o.read_iolog_file)
-			memcpy(&td->last_issue, &io_u->issue_time,
-					sizeof(struct timeval));
+			/*
+			 * only used for iolog
+			 */
+			if (td->o.read_iolog_file)
+				memcpy(&td->last_issue, &io_u->issue_time,
+						sizeof(io_u->issue_time));
+		}
 	}
 
+
 	if (ddir_rw(ddir)) {
-		td->io_issues[ddir]++;
-		td->io_issue_bytes[ddir] += buflen;
+		if (!(io_u->flags & IO_U_F_VER_LIST)) {
+			td->io_issues[ddir]++;
+			td->io_issue_bytes[ddir] += buflen;
+		}
 		td->rate_io_issue_bytes[ddir] += buflen;
 	}
 
 	ret = td->io_ops->queue(td, io_u);
+	zbd_queue_io_u(td, io_u, ret);
 
 	unlock_file(td, io_u->file);
 
@@ -318,45 +408,52 @@ int td_io_queue(struct thread_data *td, struct io_u *io_u)
 	    td->o.odirect) {
 
 		log_info("fio: first direct IO errored. File system may not "
-			 "support direct IO, or iomem_align= is bad. Try "
-			 "setting direct=0.\n");
+			 "support direct IO, or iomem_align= is bad, or "
+			 "invalid block size. Try setting direct=0.\n");
 	}
 
-	if (!td->io_ops->commit || io_u->ddir == DDIR_TRIM) {
+	if (zbd_unaligned_write(io_u->error) &&
+	    td->io_issues[io_u->ddir & 1] == 1 &&
+	    td->o.zone_mode != ZONE_MODE_ZBD) {
+		log_info("fio: first I/O failed. If %s is a zoned block device, consider --zonemode=zbd\n",
+			 io_u->file->file_name);
+	}
+
+	if (!td->io_ops->commit) {
 		io_u_mark_submit(td, 1);
 		io_u_mark_complete(td, 1);
 	}
 
 	if (ret == FIO_Q_COMPLETED) {
-		if (ddir_rw(io_u->ddir)) {
+		if (ddir_rw(io_u->ddir) ||
+		    (ddir_sync(io_u->ddir) && td->runstate != TD_FSYNCING)) {
 			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 (ddir_rw(io_u->ddir))
+		if (ddir_rw(io_u->ddir) ||
+		    (ddir_sync(io_u->ddir) && td->runstate != TD_FSYNCING))
 			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;
-		}
+		if (td->io_u_queued >= td->o.iodepth_batch)
+			td_io_commit(td);
 	}
 
-	if (!td_ioengine_flagged(td, FIO_SYNCIO)) {
-		if (fio_fill_issue_time(td))
+	if (!td_ioengine_flagged(td, FIO_SYNCIO) &&
+		!async_ioengine_sync_trim(td, io_u)) {
+		if (fio_fill_issue_time(td) &&
+			!td_ioengine_flagged(td, FIO_ASYNCIO_SETS_ISSUE_TIME)) {
 			fio_gettime(&io_u->issue_time, NULL);
 
-		/*
-		 * only used for iolog
-		 */
-		if (td->o.read_iolog_file)
-			memcpy(&td->last_issue, &io_u->issue_time,
-					sizeof(struct timeval));
+			/*
+			 * only used for iolog
+			 */
+			if (td->o.read_iolog_file)
+				memcpy(&td->last_issue, &io_u->issue_time,
+						sizeof(io_u->issue_time));
+		}
 	}
 
 	return ret;
@@ -368,28 +465,28 @@ int td_io_init(struct thread_data *td)
 
 	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");
-		}
+		if (ret)
+			log_err("fio: io engine %s init failed.%s\n",
+				td->io_ops->name,
+				td->o.iodepth > 1 ?
+				" Perhaps try reducing io depth?" : "");
+		else
+			td->io_ops_init = 1;
 		if (!td->error)
 			td->error = ret;
 	}
 
-	if (!ret && td_ioengine_flagged(td, FIO_NOIO))
-		td->flags |= TD_F_NOIO;
-
 	return ret;
 }
 
-int td_io_commit(struct thread_data *td)
+void td_io_commit(struct thread_data *td)
 {
 	int ret;
 
 	dprint(FD_IO, "calling ->commit(), depth %d\n", td->cur_depth);
 
 	if (!td->cur_depth || !td->io_u_queued)
-		return 0;
+		return;
 
 	io_u_mark_depth(td, td->io_u_queued);
 
@@ -404,14 +501,21 @@ int td_io_commit(struct thread_data *td)
 	 */
 	td->io_u_in_flight += td->io_u_queued;
 	td->io_u_queued = 0;
-
-	return 0;
 }
 
 int td_io_open_file(struct thread_data *td, struct fio_file *f)
 {
+	if (fio_file_closing(f)) {
+		/*
+		 * Open translates to undo closing.
+		 */
+		fio_file_clear_closing(f);
+		get_file(f);
+		return 0;
+	}
 	assert(!fio_file_open(f));
 	assert(f->fd == -1);
+	assert(td->io_ops->open_file);
 
 	if (td->io_ops->open_file(td, f)) {
 		if (td->error == EINVAL && td->o.odirect)
@@ -448,53 +552,56 @@ int td_io_open_file(struct thread_data *td, struct fio_file *f)
 	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)) {
+	if (td->o.fadvise_hint != F_ADV_NONE &&
+	    (f->filetype == FIO_TYPE_BLOCK || f->filetype == FIO_TYPE_FILE)) {
 		int flags;
 
-		if (td_random(td))
+		if (td->o.fadvise_hint == F_ADV_TYPE) {
+			if (td_random(td))
+				flags = POSIX_FADV_RANDOM;
+			else
+				flags = POSIX_FADV_SEQUENTIAL;
+		} else if (td->o.fadvise_hint == F_ADV_RANDOM)
 			flags = POSIX_FADV_RANDOM;
-		else
+		else if (td->o.fadvise_hint == F_ADV_SEQUENTIAL)
 			flags = POSIX_FADV_SEQUENTIAL;
-
-		if (posix_fadvise(f->fd, f->file_offset, f->io_size, flags) < 0) {
-			td_verror(td, errno, "fadvise");
-			goto err;
+		else {
+			log_err("fio: unknown fadvise type %d\n",
+							td->o.fadvise_hint);
+			flags = POSIX_FADV_NORMAL;
 		}
-	}
-#ifdef FIO_HAVE_STREAMID
-	if (td->o.fadvise_stream &&
-	    (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_FILE)) {
-		off_t stream = td->o.fadvise_stream;
 
-		if (posix_fadvise(f->fd, stream, f->io_size, POSIX_FADV_STREAMID) < 0) {
-			td_verror(td, errno, "fadvise streamid");
-			goto err;
+		if (posix_fadvise(f->fd, f->file_offset, f->io_size, flags) < 0) {
+			if (!fio_did_warn(FIO_WARN_FADVISE))
+				log_err("fio: fadvise hint failed\n");
 		}
 	}
-#endif
+#ifdef FIO_HAVE_WRITE_HINT
+	if (fio_option_is_set(&td->o, write_hint) &&
+	    (f->filetype == FIO_TYPE_BLOCK || f->filetype == FIO_TYPE_FILE)) {
+		uint64_t hint = td->o.write_hint;
+		int cmd;
 
-#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");
-			if (ret == ENOTTY) { /* ENOTTY suggests RAW device or ZFS */
-				log_err("fio: doing directIO to RAW devices or ZFS not supported\n");
-			} else {
-				log_err("fio: the file system does not seem to support direct IO\n");
-			}
+		/*
+		 * For direct IO, we just need/want to set the hint on
+		 * the file descriptor. For buffered IO, we need to set
+		 * it on the inode.
+		 */
+		if (td->o.odirect)
+			cmd = F_SET_FILE_RW_HINT;
+		else
+			cmd = F_SET_RW_HINT;
 
+		if (fcntl(f->fd, cmd, &hint) < 0) {
+			td_verror(td, errno, "fcntl write hint");
 			goto err;
 		}
 	}
 #endif
 
+	if (td->o.odirect && !OS_O_DIRECT && fio_set_directio(td, f))
+		goto err;
+
 done:
 	log_file(td, f, FIO_LOG_OPEN_FILE);
 	return 0;
@@ -515,11 +622,6 @@ int td_io_close_file(struct thread_data *td, struct fio_file *f)
 	 */
 	fio_file_set_closing(f);
 
-	disk_util_dec(f->du);
-
-	if (td->o.file_lock_mode != FILE_LOCK_NONE)
-		unlock_file_all(td, f);
-
 	return put_file(td, f);
 }
 
@@ -546,77 +648,50 @@ int td_io_get_file_size(struct thread_data *td, struct fio_file *f)
 	return td->io_ops->get_file_size(td, f);
 }
 
-static int do_sync_file_range(const struct thread_data *td,
-			      struct fio_file *f)
+#ifdef CONFIG_DYNAMIC_ENGINES
+/* Load all dynamic engines in FIO_EXT_ENG_DIR for enghelp command */
+static void
+fio_load_dynamic_engines(struct thread_data *td)
 {
-	off64_t offset, nbytes;
+	DIR *dirhandle = NULL;
+	struct dirent *dirent = NULL;
+	char engine_path[PATH_MAX];
 
-	offset = f->first_write;
-	nbytes = f->last_write - f->first_write;
+	dirhandle = opendir(FIO_EXT_ENG_DIR);
+	if (!dirhandle)
+		return;
 
-	if (!nbytes)
-		return 0;
+	while ((dirent = readdir(dirhandle)) != NULL) {
+		if (!strcmp(dirent->d_name, ".") ||
+		    !strcmp(dirent->d_name, ".."))
+			continue;
 
-	return sync_file_range(f->fd, offset, nbytes, td->o.sync_file_range);
-}
-
-int do_io_u_sync(const 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 CONFIG_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;
+		sprintf(engine_path, "%s/%s", FIO_EXT_ENG_DIR, dirent->d_name);
+		dlopen_ioengine(td, engine_path);
 	}
 
-	if (ret < 0)
-		io_u->error = errno;
-
-	return ret;
+	closedir(dirhandle);
 }
-
-int do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
-{
-#ifndef FIO_HAVE_TRIM
-	io_u->error = EINVAL;
-	return 0;
 #else
-	struct fio_file *f = io_u->file;
-	int ret;
-
-	ret = os_trim(f->fd, io_u->offset, io_u->xfer_buflen);
-	if (!ret)
-		return io_u->xfer_buflen;
-
-	io_u->error = ret;
-	return 0;
+#define fio_load_dynamic_engines(td) do { } while (0)
 #endif
-}
 
 int fio_show_ioengine_help(const char *engine)
 {
 	struct flist_head *entry;
 	struct thread_data td;
+	struct ioengine_ops *io_ops;
 	char *sep;
 	int ret = 1;
 
+	memset(&td, 0, sizeof(struct thread_data));
+
 	if (!engine || !*engine) {
 		log_info("Available IO engines:\n");
+		fio_load_dynamic_engines(&td);
 		flist_for_each(entry, &engine_list) {
-			td.io_ops = flist_entry(entry, struct ioengine_ops,
-						list);
-			log_info("\t%s\n", td.io_ops->name);
+			io_ops = flist_entry(entry, struct ioengine_ops, list);
+			log_info("\t%s\n", io_ops->name);
 		}
 		return 0;
 	}
@@ -626,9 +701,9 @@ int fio_show_ioengine_help(const char *engine)
 		sep++;
 	}
 
-	memset(&td, 0, sizeof(td));
+	td.o.ioengine = (char *)engine;
+	td.io_ops = load_ioengine(&td);
 
-	td.io_ops = load_ioengine(&td, engine);
 	if (!td.io_ops) {
 		log_info("IO engine %s not found\n", engine);
 		return 1;
@@ -640,6 +715,5 @@ int fio_show_ioengine_help(const char *engine)
 		log_info("IO engine %s has no options\n", td.io_ops->name);
 
 	free_ioengine(&td);
-
 	return ret;
 }