summaryrefslogtreecommitdiff
path: root/ioengines.c
blob: 9e3fcc9f681a46fa336ef247a6aa6497dade09af (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
/*
 * The io parts of the fio tool, includes workers for sync and mmap'ed
 * io, as well as both posix and linux libaio support.
 *
 * sync io is implemented on top of aio.
 *
 * This is not really specific to fio, if the get_io_u/put_io_u and
 * structures was pulled into this as well it would be a perfectly
 * generic io engine that could be used for other projects.
 *
 */
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <dlfcn.h>
#include <fcntl.h>
#include <assert.h>

#include "fio.h"
#include "diskutil.h"
#include "zbd.h"

static FLIST_HEAD(engine_list);

static bool 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);
		return true;
	}

	if (!ops->queue) {
		log_err("%s: no queue handler\n", ops->name);
		return true;
	}

	/*
	 * sync engines only need a ->queue()
	 */
	if (ops->flags & FIO_SYNCIO)
		return false;

	if (!ops->event || !ops->getevents) {
		log_err("%s: no event/getevents handler\n", ops->name);
		return true;
	}

	return false;
}

void unregister_ioengine(struct ioengine_ops *ops)
{
	dprint(FD_IO, "ioengine %s unregistered\n", ops->name);
	flist_del_init(&ops->list);
}

void register_ioengine(struct ioengine_ops *ops)
{
	dprint(FD_IO, "ioengine %s registered\n", ops->name);
	flist_add_tail(&ops->list, &engine_list);
}

static struct ioengine_ops *find_ioengine(const char *name)
{
	struct ioengine_ops *ops;
	struct flist_head *entry;

	flist_for_each(entry, &engine_list) {
		ops = flist_entry(entry, struct ioengine_ops, list);
		if (!strcmp(name, ops->name))
			return ops;
	}

	return NULL;
}

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);

	dlerror();
	dlhandle = dlopen(engine_lib, RTLD_LAZY);
	if (!dlhandle) {
		td_vmsg(td, -1, dlerror(), "dlopen");
		return NULL;
	}

	/*
	 * Unlike the included modules, external engines should have a
	 * non-static ioengine structure that we can reference.
	 */
	ops = dlsym(dlhandle, engine_lib);
	if (!ops)
		ops = dlsym(dlhandle, "ioengine");

	/*
	 * For some external engines (like C++ ones) it is not that trivial
	 * to provide a non-static ionengine structure that we can reference.
	 * Instead we call a method which allocates the required ioengine
	 * structure.
	 */
	if (!ops) {
		get_ioengine_t get_ioengine = dlsym(dlhandle, "get_ioengine");

		if (get_ioengine)
			get_ioengine(&ops);
	}

	if (!ops) {
		td_vmsg(td, -1, dlerror(), "dlsym");
		dlclose(dlhandle);
		return NULL;
	}

	td->io_ops_dlhandle = dlhandle;
	return ops;
}

static struct ioengine_ops *__load_ioengine(const char *name)
{
	char engine[64];

	snprintf(engine, sizeof(engine), "%s", name);

	/*
	 * linux libaio has alias names, so convert to what we want
	 */
	if (!strncmp(engine, "linuxaio", 8)) {
		dprint(FD_IO, "converting ioengine name: %s -> libaio\n", name);
		strcpy(engine, "libaio");
	}

	dprint(FD_IO, "load ioengine %s\n", engine);
	return find_ioengine(engine);
}

struct ioengine_ops *load_ioengine(struct thread_data *td)
{
	struct ioengine_ops *ops = NULL;
	const char *name;

	/*
	 * 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.
	 */
	name = td->o.ioengine_so_path ?: td->o.ioengine;

	/*
	 * 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);
	if (!ops)
		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;
	}

	/*
	 * Check that the required methods are there.
	 */
	if (check_engine_ops(ops))
		return NULL;

	return ops;
}

/*
 * For cleaning up an ioengine which never made it to init().
 */
void free_ioengine(struct thread_data *td)
{
	dprint(FD_IO, "free ioengine %s\n", td->io_ops->name);

	if (td->eo && td->io_ops->options) {
		options_free(td->io_ops->options, td->eo);
		free(td->eo);
		td->eo = NULL;
	}

	if (td->io_ops_dlhandle) {
		dlclose(td->io_ops_dlhandle);
		td->io_ops_dlhandle = NULL;
	}

	td->io_ops = NULL;
}

void close_ioengine(struct thread_data *td)
{
	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;
	}

	free_ioengine(td);
}

int td_io_prep(struct thread_data *td, struct io_u *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: io_u %p: ret=%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, unsigned int min, unsigned int max,
		    const struct timespec *t)
{
	int r = 0;

	/*
	 * For ioengine=rdma one side operation RDMA_WRITE or RDMA_READ,
	 * server side gets a message from the client
	 * side that the task is finished, and
	 * td->done is set to 1 after td_io_commit(). In this case,
	 * there is no need to reap complete event in server side.
	 */
	if (td->done)
		return 0;

	if (min > 0 && td->io_ops->commit) {
		r = td->io_ops->commit(td);
		if (r < 0)
			goto out;
	}
	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) {
		/*
		 * Reflect that our submitted requests were retrieved with
		 * whatever OS async calls are in the underlying engine.
		 */
		td->io_u_in_flight -= r;
		io_u_mark_complete(td, r);
	} else
		td_verror(td, r, "get_events");

	dprint(FD_IO, "getevents: %d\n", r);
	return r;
}

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 long buflen = io_u->xfer_buflen;
	enum fio_q_status ret;

	dprint_io_u(io_u, "queue");
	fio_ro_check(td, 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))
		pthread_mutex_unlock(&overlap_check);

	assert(fio_file_open(io_u->file));

	/*
	 * If using a write iolog, store this entry.
	 */
	log_io_u(td, io_u);

	io_u->error = 0;
	io_u->resid = 0;

	if (td_ioengine_flagged(td, FIO_SYNCIO) ||
		(td_ioengine_flagged(td, FIO_ASYNCIO_SYNC_TRIM) && 
		io_u->ddir == DDIR_TRIM)) {
		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(io_u->issue_time));
	}

	if (ddir_rw(ddir)) {
		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(io_u, ret);

	unlock_file(td, io_u->file);

	if (ret == FIO_Q_BUSY && ddir_rw(ddir)) {
		td->io_issues[ddir]--;
		td->io_issue_bytes[ddir] -= buflen;
		td->rate_io_issue_bytes[ddir] -= buflen;
		io_u_clear(td, io_u, IO_U_F_FLIGHT);
	}

	/*
	 * If an error was seen and the io engine didn't propagate it
	 * back to 'td', do so.
	 */
	if (io_u->error && !td->error)
		td_verror(td, io_u->error, "td_io_queue");

	/*
	 * 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] == 1 &&
	    td->o.odirect) {

		log_info("fio: first direct IO errored. File system may not "
			 "support direct IO, or iomem_align= is bad, or "
			 "invalid block size. Try setting direct=0.\n");
	}

	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);
		zbd_put_io_u(io_u);
	}

	if (ret == FIO_Q_COMPLETED) {
		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) {
		td->io_u_queued++;

		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)
			td_io_commit(td);
	}

	if (!td_ioengine_flagged(td, FIO_SYNCIO) &&
		(!td_ioengine_flagged(td, FIO_ASYNCIO_SYNC_TRIM) ||
		 io_u->ddir != DDIR_TRIM)) {
		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(io_u->issue_time));
	}

	return ret;
}

int td_io_init(struct thread_data *td)
{
	int ret = 0;

	if (td->io_ops->init) {
		ret = td->io_ops->init(td);
		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;
	}

	return ret;
}

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;

	io_u_mark_depth(td, td->io_u_queued);

	if (td->io_ops->commit) {
		ret = td->io_ops->commit(td);
		if (ret)
			td_verror(td, -ret, "io commit");
	}

	/*
	 * Reflect that events were submitted as async IO requests.
	 */
	td->io_u_in_flight += td->io_u_queued;
	td->io_u_queued = 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)
			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(td, 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_ioengine_flagged(td, FIO_DISKLESSIO))
		goto done;

	if (td->o.invalidate_cache && file_invalidate_cache(td, f))
		goto err;

	if (td->o.fadvise_hint != F_ADV_NONE &&
	    (f->filetype == FIO_TYPE_BLOCK || f->filetype == FIO_TYPE_FILE)) {
		int flags;

		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 if (td->o.fadvise_hint == F_ADV_SEQUENTIAL)
			flags = POSIX_FADV_SEQUENTIAL;
		else {
			log_err("fio: unknown fadvise type %d\n",
							td->o.fadvise_hint);
			flags = POSIX_FADV_NORMAL;
		}

		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");
		}
	}
#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;

		/*
		 * 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;
err:
	disk_util_dec(f->du);
	if (td->io_ops->close_file)
		td->io_ops->close_file(td, f);
	return 1;
}

int td_io_close_file(struct thread_data *td, struct fio_file *f)
{
	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);

	return put_file(td, f);
}

int td_io_unlink_file(struct thread_data *td, struct fio_file *f)
{
	if (td->io_ops->unlink_file)
		return td->io_ops->unlink_file(td, f);
	else {
		int ret;

		ret = unlink(f->file_name);
		if (ret < 0)
			return errno;

		return 0;
	}
}

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);
}

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;

	if (!engine || !*engine) {
		log_info("Available IO engines:\n");
		flist_for_each(entry, &engine_list) {
			io_ops = flist_entry(entry, struct ioengine_ops, list);
			log_info("\t%s\n", io_ops->name);
		}
		return 0;
	}
	sep = strchr(engine, ',');
	if (sep) {
		*sep = 0;
		sep++;
	}

	memset(&td, 0, sizeof(struct thread_data));
	td.o.ioengine = (char *)engine;
	io_ops = load_ioengine(&td);

	if (!io_ops) {
		log_info("IO engine %s not found\n", engine);
		return 1;
	}

	if (io_ops->options)
		ret = show_cmd_help(io_ops->options, sep);
	else
		log_info("IO engine %s has no options\n", io_ops->name);

	free_ioengine(&td);
	return ret;
}