[fio.git] / engines / pmemblk.c

/*
 * pmemblk: IO engine that uses PMDK libpmemblk to read and write data
 *
 * Copyright (C) 2016 Hewlett Packard Enterprise Development LP
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License,
 * version 2 as published by the Free Software Foundation..
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the Free
 * Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
 * Boston, MA 02110-1301, USA.
 */

/*
 * pmemblk engine
 *
 * IO engine that uses libpmemblk to read and write data
 *
 * To use:
 *   ioengine=pmemblk
 *
 * Other relevant settings:
 *   thread=1   REQUIRED
 *   iodepth=1
 *   direct=1
 *   unlink=1
 *   filename=/mnt/pmem0/fiotestfile,BSIZE,FSIZEMiB
 *
 *   thread must be set to 1 for pmemblk as multiple processes cannot
 *     open the same block pool file.
 *
 *   iodepth should be set to 1 as pmemblk is always synchronous.
 *   Use numjobs to scale up.
 *
 *   direct=1 is implied as pmemblk is always direct. A warning message
 *   is printed if this is not specified.
 *
 *   unlink=1 removes the block pool file after testing, and is optional.
 *
 *   The pmem device must have a DAX-capable filesystem and be mounted
 *   with DAX enabled.  filename must point to a file on that filesystem.
 *
 *   Example:
 *     mkfs.xfs /dev/pmem0
 *     mkdir /mnt/pmem0
 *     mount -o dax /dev/pmem0 /mnt/pmem0
 *
 *   When specifying the filename, if the block pool file does not already
 *   exist, then the pmemblk engine creates the pool file if you specify
 *   the block and file sizes.  BSIZE is the block size in bytes.
 *   FSIZEMB is the pool file size in MiB.
 *
 *   See examples/pmemblk.fio for more.
 *
 */

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/uio.h>
#include <errno.h>
#include <assert.h>
#include <string.h>
#include <libpmem.h>
#include <libpmemblk.h>

#include "../fio.h"

/*
 * libpmemblk
 */
typedef struct fio_pmemblk_file *fio_pmemblk_file_t;

struct fio_pmemblk_file {
	fio_pmemblk_file_t pmb_next;
	char *pmb_filename;
	uint64_t pmb_refcnt;
	PMEMblkpool *pmb_pool;
	size_t pmb_bsize;
	size_t pmb_nblocks;
};

static fio_pmemblk_file_t Cache;

static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER;

#define PMB_CREATE   (0x0001)	/* should create file */

fio_pmemblk_file_t fio_pmemblk_cache_lookup(const char *filename)
{
	fio_pmemblk_file_t i;

	for (i = Cache; i != NULL; i = i->pmb_next)
		if (!strcmp(filename, i->pmb_filename))
			return i;

	return NULL;
}

static void fio_pmemblk_cache_insert(fio_pmemblk_file_t pmb)
{
	pmb->pmb_next = Cache;
	Cache = pmb;
}

static void fio_pmemblk_cache_remove(fio_pmemblk_file_t pmb)
{
	fio_pmemblk_file_t i;

	if (pmb == Cache) {
		Cache = Cache->pmb_next;
		pmb->pmb_next = NULL;
		return;
	}

	for (i = Cache; i != NULL; i = i->pmb_next)
		if (pmb == i->pmb_next) {
			i->pmb_next = i->pmb_next->pmb_next;
			pmb->pmb_next = NULL;
			return;
		}
}

/*
 * to control block size and gross file size at the libpmemblk
 * level, we allow the block size and file size to be appended
 * to the file name:
 *
 *   path[,bsize,fsizemib]
 *
 * note that we do not use the fio option "filesize" to dictate
 * the file size because we can only give libpmemblk the gross
 * file size, which is different from the net or usable file
 * size (which is probably what fio wants).
 *
 * the final path without the parameters is returned in ppath.
 * the block size and file size are returned in pbsize and fsize.
 *
 * note that the user specifies the file size in MiB, but
 * we return bytes from here.
 */
static void pmb_parse_path(const char *pathspec, char **ppath, uint64_t *pbsize,
			   uint64_t *pfsize)
{
	char *path;
	char *s;
	uint64_t bsize;
	uint64_t fsizemib;

	path = strdup(pathspec);
	if (!path) {
		*ppath = NULL;
		return;
	}

	/* extract sizes, if given */
	s = strrchr(path, ',');
	if (s && (fsizemib = strtoull(s + 1, NULL, 10))) {
		*s = 0;
		s = strrchr(path, ',');
		if (s && (bsize = strtoull(s + 1, NULL, 10))) {
			*s = 0;
			*ppath = path;
			*pbsize = bsize;
			*pfsize = fsizemib << 20;
			return;
		}
	}

	/* size specs not found */
	strcpy(path, pathspec);
	*ppath = path;
	*pbsize = 0;
	*pfsize = 0;
}

static fio_pmemblk_file_t pmb_open(const char *pathspec, int flags)
{
	fio_pmemblk_file_t pmb;
	char *path = NULL;
	uint64_t bsize = 0;
	uint64_t fsize = 0;

	pmb_parse_path(pathspec, &path, &bsize, &fsize);
	if (!path)
		return NULL;

	pthread_mutex_lock(&CacheLock);

	pmb = fio_pmemblk_cache_lookup(path);
	if (!pmb) {
		pmb = malloc(sizeof(*pmb));
		if (!pmb)
			goto error;

		/* try opening existing first, create it if needed */
		pmb->pmb_pool = pmemblk_open(path, bsize);
		if (!pmb->pmb_pool && (errno == ENOENT) &&
		    (flags & PMB_CREATE) && (0 < fsize) && (0 < bsize)) {
			pmb->pmb_pool =
			    pmemblk_create(path, bsize, fsize, 0644);
		}
		if (!pmb->pmb_pool) {
			log_err("pmemblk: unable to open pmemblk pool file %s (%s)\n",
			     path, strerror(errno));
			goto error;
		}

		pmb->pmb_filename = path;
		pmb->pmb_next = NULL;
		pmb->pmb_refcnt = 0;
		pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool);
		pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool);

		fio_pmemblk_cache_insert(pmb);
	}

	pmb->pmb_refcnt += 1;

	pthread_mutex_unlock(&CacheLock);

	return pmb;

error:
	if (pmb) {
		if (pmb->pmb_pool)
			pmemblk_close(pmb->pmb_pool);
		pmb->pmb_pool = NULL;
		pmb->pmb_filename = NULL;
		free(pmb);
	}
	if (path)
		free(path);

	pthread_mutex_unlock(&CacheLock);
	return NULL;
}

static void pmb_close(fio_pmemblk_file_t pmb, const bool keep)
{
	pthread_mutex_lock(&CacheLock);

	pmb->pmb_refcnt--;

	if (!keep && !pmb->pmb_refcnt) {
		pmemblk_close(pmb->pmb_pool);
		pmb->pmb_pool = NULL;
		free(pmb->pmb_filename);
		pmb->pmb_filename = NULL;
		fio_pmemblk_cache_remove(pmb);
		free(pmb);
	}

	pthread_mutex_unlock(&CacheLock);
}

static int pmb_get_flags(struct thread_data *td, uint64_t *pflags)
{
	static int thread_warned = 0;
	static int odirect_warned = 0;

	uint64_t flags = 0;

	if (!td->o.use_thread) {
		if (!thread_warned) {
			thread_warned = 1;
			log_err("pmemblk: must set thread=1 for pmemblk engine\n");
		}
		return 1;
	}

	if (!td->o.odirect && !odirect_warned) {
		odirect_warned = 1;
		log_info("pmemblk: direct == 0, but pmemblk is always direct\n");
	}

	if (td->o.allow_create)
		flags |= PMB_CREATE;

	(*pflags) = flags;
	return 0;
}

static int fio_pmemblk_open_file(struct thread_data *td, struct fio_file *f)
{
	uint64_t flags = 0;
	fio_pmemblk_file_t pmb;

	if (pmb_get_flags(td, &flags))
		return 1;

	pmb = pmb_open(f->file_name, flags);
	if (!pmb)
		return 1;

	FILE_SET_ENG_DATA(f, pmb);
	return 0;
}

static int fio_pmemblk_close_file(struct thread_data fio_unused *td,
				  struct fio_file *f)
{
	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);

	if (pmb)
		pmb_close(pmb, false);

	FILE_SET_ENG_DATA(f, NULL);
	return 0;
}

static int fio_pmemblk_get_file_size(struct thread_data *td, struct fio_file *f)
{
	uint64_t flags = 0;
	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);

	if (fio_file_size_known(f))
		return 0;

	if (!pmb) {
		if (pmb_get_flags(td, &flags))
			return 1;
		pmb = pmb_open(f->file_name, flags);
		if (!pmb)
			return 1;
	}

	f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks;

	fio_file_set_size_known(f);

	if (!FILE_ENG_DATA(f))
		pmb_close(pmb, true);

	return 0;
}

static enum fio_q_status fio_pmemblk_queue(struct thread_data *td,
					   struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);

	unsigned long long off;
	unsigned long len;
	void *buf;

	fio_ro_check(td, io_u);

	switch (io_u->ddir) {
	case DDIR_READ:
	case DDIR_WRITE:
		off = io_u->offset;
		len = io_u->xfer_buflen;

		io_u->error = EINVAL;
		if (off % pmb->pmb_bsize)
			break;
		if (len % pmb->pmb_bsize)
			break;
		if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks)
			break;

		io_u->error = 0;
		buf = io_u->xfer_buf;
		off /= pmb->pmb_bsize;
		len /= pmb->pmb_bsize;
		while (0 < len) {
			if (io_u->ddir == DDIR_READ &&
			   0 != pmemblk_read(pmb->pmb_pool, buf, off)) {
				io_u->error = errno;
				break;
			} else if (0 != pmemblk_write(pmb->pmb_pool, buf, off)) {
				io_u->error = errno;
				break;
			}
			buf += pmb->pmb_bsize;
			off++;
			len--;
		}
		off *= pmb->pmb_bsize;
		len *= pmb->pmb_bsize;
		io_u->resid = io_u->xfer_buflen - (off - io_u->offset);
		break;
	case DDIR_SYNC:
	case DDIR_DATASYNC:
	case DDIR_SYNC_FILE_RANGE:
		/* we're always sync'd */
		io_u->error = 0;
		break;
	default:
		io_u->error = EINVAL;
		break;
	}

	return FIO_Q_COMPLETED;
}

static int fio_pmemblk_unlink_file(struct thread_data *td, struct fio_file *f)
{
	char *path = NULL;
	uint64_t bsize = 0;
	uint64_t fsize = 0;

	/*
	 * we need our own unlink in case the user has specified
	 * the block and file sizes in the path name.  we parse
	 * the file_name to determine the file name we actually used.
	 */

	pmb_parse_path(f->file_name, &path, &bsize, &fsize);
	if (!path)
		return ENOENT;

	unlink(path);
	free(path);
	return 0;
}

static struct ioengine_ops ioengine = {
	.name = "pmemblk",
	.version = FIO_IOOPS_VERSION,
	.queue = fio_pmemblk_queue,
	.open_file = fio_pmemblk_open_file,
	.close_file = fio_pmemblk_close_file,
	.get_file_size = fio_pmemblk_get_file_size,
	.unlink_file = fio_pmemblk_unlink_file,
	.flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_NOEXTEND | FIO_NODISKUTIL,
};

static void fio_init fio_pmemblk_register(void)
{
	register_ioengine(&ioengine);
}

static void fio_exit fio_pmemblk_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
43f3cec2	1	/*
363a5f65	2	* pmemblk: IO engine that uses PMDK libpmemblk to read and write data
43f3cec2 BB	3	*
	4	* Copyright (C) 2016 Hewlett Packard Enterprise Development LP
	5	*
	6	* This program is free software; you can redistribute it and/or
	7	* modify it under the terms of the GNU General Public License,
	8	* version 2 as published by the Free Software Foundation..
	9	*
	10	* This program is distributed in the hope that it will be useful,
	11	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	12	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	13	* GNU General Public License for more details.
	14	*
	15	* You should have received a copy of the GNU General Public
	16	* License along with this program; if not, write to the Free
fa07eaa6 SW	17	* Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
fa07eaa6 SW	18	* Boston, MA 02110-1301, USA.
43f3cec2 BB	19	*/
43f3cec2 BB	20
43f3cec2 BB	21	/*
	22	* pmemblk engine
	23	*
	24	* IO engine that uses libpmemblk to read and write data
	25	*
f1d85c6e BB	26	* To use:
f1d85c6e BB	27	* ioengine=pmemblk
43f3cec2 BB	28	*
43f3cec2 BB	29	* Other relevant settings:
0cbb3f53	30	* thread=1 REQUIRED
43f3cec2 BB	31	* iodepth=1
43f3cec2 BB	32	* direct=1
43f3cec2	33	* unlink=1
0cbb3f53	34	* filename=/mnt/pmem0/fiotestfile,BSIZE,FSIZEMiB
43f3cec2 BB	35	*
	36	* thread must be set to 1 for pmemblk as multiple processes cannot
	37	* open the same block pool file.
	38	*
	39	* iodepth should be set to 1 as pmemblk is always synchronous.
	40	* Use numjobs to scale up.
	41	*
0cbb3f53 RE	42	* direct=1 is implied as pmemblk is always direct. A warning message
	43	* is printed if this is not specified.
	44	*
	45	* unlink=1 removes the block pool file after testing, and is optional.
	46	*
	47	* The pmem device must have a DAX-capable filesystem and be mounted
	48	* with DAX enabled. filename must point to a file on that filesystem.
43f3cec2	49	*
0cbb3f53 RE	50	* Example:
	51	* mkfs.xfs /dev/pmem0
	52	* mkdir /mnt/pmem0
	53	* mount -o dax /dev/pmem0 /mnt/pmem0
43f3cec2 BB	54	*
43f3cec2 BB	55	* When specifying the filename, if the block pool file does not already
0cbb3f53	56	* exist, then the pmemblk engine creates the pool file if you specify
43f3cec2	57	* the block and file sizes. BSIZE is the block size in bytes.
0cbb3f53	58	* FSIZEMB is the pool file size in MiB.
43f3cec2	59	*
f1d85c6e BB	60	* See examples/pmemblk.fio for more.
f1d85c6e BB	61	*
43f3cec2 BB	62	*/
	63
	64	#include <stdio.h>
	65	#include <stdlib.h>
	66	#include <unistd.h>
	67	#include <sys/uio.h>
	68	#include <errno.h>
	69	#include <assert.h>
43f3cec2	70	#include <string.h>
cf8775b8 RE	71	#include <libpmem.h>
cf8775b8 RE	72	#include <libpmemblk.h>
43f3cec2 BB	73
	74	#include "../fio.h"
	75
43f3cec2 BB	76	/*
	77	* libpmemblk
	78	*/
c086b314	79	typedef struct fio_pmemblk_file *fio_pmemblk_file_t;
630431cc	80
43f3cec2	81	struct fio_pmemblk_file {
c086b314 JA	82	fio_pmemblk_file_t pmb_next;
	83	char *pmb_filename;
	84	uint64_t pmb_refcnt;
	85	PMEMblkpool *pmb_pool;
	86	size_t pmb_bsize;
	87	size_t pmb_nblocks;
43f3cec2	88	};
43f3cec2	89
630431cc	90	static fio_pmemblk_file_t Cache;
43f3cec2	91
c086b314	92	static pthread_mutex_t CacheLock = PTHREAD_MUTEX_INITIALIZER;
43f3cec2	93
c086b314	94	#define PMB_CREATE (0x0001) /* should create file */
43f3cec2	95
c086b314	96	fio_pmemblk_file_t fio_pmemblk_cache_lookup(const char *filename)
43f3cec2	97	{
c086b314	98	fio_pmemblk_file_t i;
43f3cec2 BB	99
43f3cec2 BB	100	for (i = Cache; i != NULL; i = i->pmb_next)
630431cc	101	if (!strcmp(filename, i->pmb_filename))
43f3cec2 BB	102	return i;
	103
	104	return NULL;
630431cc	105	}
43f3cec2	106
c086b314	107	static void fio_pmemblk_cache_insert(fio_pmemblk_file_t pmb)
43f3cec2 BB	108	{
	109	pmb->pmb_next = Cache;
	110	Cache = pmb;
630431cc	111	}
43f3cec2	112
c086b314	113	static void fio_pmemblk_cache_remove(fio_pmemblk_file_t pmb)
43f3cec2	114	{
c086b314	115	fio_pmemblk_file_t i;
43f3cec2 BB	116
	117	if (pmb == Cache) {
	118	Cache = Cache->pmb_next;
	119	pmb->pmb_next = NULL;
	120	return;
	121	}
	122
	123	for (i = Cache; i != NULL; i = i->pmb_next)
	124	if (pmb == i->pmb_next) {
	125	i->pmb_next = i->pmb_next->pmb_next;
	126	pmb->pmb_next = NULL;
	127	return;
	128	}
630431cc	129	}
43f3cec2 BB	130
	131	/*
	132	* to control block size and gross file size at the libpmemblk
	133	* level, we allow the block size and file size to be appended
	134	* to the file name:
	135	*
0cbb3f53	136	* path[,bsize,fsizemib]
43f3cec2 BB	137	*
	138	* note that we do not use the fio option "filesize" to dictate
	139	* the file size because we can only give libpmemblk the gross
	140	* file size, which is different from the net or usable file
	141	* size (which is probably what fio wants).
	142	*
	143	* the final path without the parameters is returned in ppath.
	144	* the block size and file size are returned in pbsize and fsize.
	145	*
0cbb3f53	146	* note that the user specifies the file size in MiB, but
43f3cec2 BB	147	* we return bytes from here.
43f3cec2 BB	148	*/
630431cc JA	149	static void pmb_parse_path(const char pathspec, char ppath, uint64_t pbsize,
630431cc JA	150	uint64_t *pfsize)
43f3cec2	151	{
c086b314 JA	152	char *path;
	153	char *s;
	154	uint64_t bsize;
0cbb3f53	155	uint64_t fsizemib;
43f3cec2 BB	156
43f3cec2 BB	157	path = strdup(pathspec);
630431cc	158	if (!path) {
43f3cec2 BB	159	*ppath = NULL;
	160	return;
	161	}
	162
	163	/* extract sizes, if given */
	164	s = strrchr(path, ',');
0cbb3f53	165	if (s && (fsizemib = strtoull(s + 1, NULL, 10))) {
43f3cec2 BB	166	*s = 0;
43f3cec2 BB	167	s = strrchr(path, ',');
c086b314	168	if (s && (bsize = strtoull(s + 1, NULL, 10))) {
43f3cec2	169	*s = 0;
c086b314	170	*ppath = path;
43f3cec2	171	*pbsize = bsize;
0cbb3f53	172	*pfsize = fsizemib << 20;
43f3cec2 BB	173	return;
	174	}
	175	}
	176
	177	/* size specs not found */
	178	strcpy(path, pathspec);
c086b314	179	*ppath = path;
43f3cec2 BB	180	*pbsize = 0;
43f3cec2 BB	181	*pfsize = 0;
630431cc	182	}
43f3cec2	183
630431cc	184	static fio_pmemblk_file_t pmb_open(const char *pathspec, int flags)
43f3cec2	185	{
c086b314 JA	186	fio_pmemblk_file_t pmb;
	187	char *path = NULL;
	188	uint64_t bsize = 0;
	189	uint64_t fsize = 0;
43f3cec2 BB	190
43f3cec2 BB	191	pmb_parse_path(pathspec, &path, &bsize, &fsize);
630431cc	192	if (!path)
43f3cec2 BB	193	return NULL;
43f3cec2 BB	194
de88999f	195	pthread_mutex_lock(&CacheLock);
43f3cec2 BB	196
43f3cec2 BB	197	pmb = fio_pmemblk_cache_lookup(path);
630431cc	198	if (!pmb) {
43f3cec2	199	pmb = malloc(sizeof(*pmb));
630431cc	200	if (!pmb)
43f3cec2 BB	201	goto error;
	202
	203	/* try opening existing first, create it if needed */
	204	pmb->pmb_pool = pmemblk_open(path, bsize);
630431cc	205	if (!pmb->pmb_pool && (errno == ENOENT) &&
c086b314 JA	206	(flags & PMB_CREATE) && (0 < fsize) && (0 < bsize)) {
	207	pmb->pmb_pool =
	208	pmemblk_create(path, bsize, fsize, 0644);
43f3cec2	209	}
630431cc	210	if (!pmb->pmb_pool) {
a0998600 RE	211	log_err("pmemblk: unable to open pmemblk pool file %s (%s)\n",
a0998600 RE	212	path, strerror(errno));
43f3cec2 BB	213	goto error;
	214	}
	215
	216	pmb->pmb_filename = path;
c086b314 JA	217	pmb->pmb_next = NULL;
	218	pmb->pmb_refcnt = 0;
	219	pmb->pmb_bsize = pmemblk_bsize(pmb->pmb_pool);
	220	pmb->pmb_nblocks = pmemblk_nblock(pmb->pmb_pool);
43f3cec2 BB	221
	222	fio_pmemblk_cache_insert(pmb);
	223	}
	224
	225	pmb->pmb_refcnt += 1;
	226
de88999f	227	pthread_mutex_unlock(&CacheLock);
c086b314	228
43f3cec2 BB	229	return pmb;
	230
	231	error:
630431cc JA	232	if (pmb) {
630431cc JA	233	if (pmb->pmb_pool)
43f3cec2	234	pmemblk_close(pmb->pmb_pool);
c086b314	235	pmb->pmb_pool = NULL;
43f3cec2 BB	236	pmb->pmb_filename = NULL;
	237	free(pmb);
	238	}
630431cc	239	if (path)
43f3cec2	240	free(path);
de88999f JA	241
de88999f JA	242	pthread_mutex_unlock(&CacheLock);
43f3cec2	243	return NULL;
630431cc	244	}
43f3cec2	245
630431cc	246	static void pmb_close(fio_pmemblk_file_t pmb, const bool keep)
43f3cec2	247	{
de88999f	248	pthread_mutex_lock(&CacheLock);
43f3cec2 BB	249
	250	pmb->pmb_refcnt--;
	251
630431cc	252	if (!keep && !pmb->pmb_refcnt) {
43f3cec2 BB	253	pmemblk_close(pmb->pmb_pool);
	254	pmb->pmb_pool = NULL;
	255	free(pmb->pmb_filename);
	256	pmb->pmb_filename = NULL;
	257	fio_pmemblk_cache_remove(pmb);
	258	free(pmb);
	259	}
	260
de88999f	261	pthread_mutex_unlock(&CacheLock);
630431cc	262	}
43f3cec2	263
630431cc	264	static int pmb_get_flags(struct thread_data td, uint64_t pflags)
43f3cec2	265	{
c086b314	266	static int thread_warned = 0;
43f3cec2 BB	267	static int odirect_warned = 0;
43f3cec2 BB	268
c086b314	269	uint64_t flags = 0;
43f3cec2 BB	270
	271	if (!td->o.use_thread) {
	272	if (!thread_warned) {
	273	thread_warned = 1;
a0998600	274	log_err("pmemblk: must set thread=1 for pmemblk engine\n");
43f3cec2 BB	275	}
	276	return 1;
	277	}
	278
	279	if (!td->o.odirect && !odirect_warned) {
	280	odirect_warned = 1;
a0998600	281	log_info("pmemblk: direct == 0, but pmemblk is always direct\n");
43f3cec2 BB	282	}
	283
	284	if (td->o.allow_create)
	285	flags \|= PMB_CREATE;
	286
	287	(*pflags) = flags;
	288	return 0;
630431cc	289	}
43f3cec2	290
c086b314	291	static int fio_pmemblk_open_file(struct thread_data td, struct fio_file f)
43f3cec2	292	{
c086b314 JA	293	uint64_t flags = 0;
c086b314 JA	294	fio_pmemblk_file_t pmb;
43f3cec2	295
630431cc	296	if (pmb_get_flags(td, &flags))
43f3cec2 BB	297	return 1;
	298
	299	pmb = pmb_open(f->file_name, flags);
630431cc	300	if (!pmb)
43f3cec2 BB	301	return 1;
43f3cec2 BB	302
710bf9c5	303	FILE_SET_ENG_DATA(f, pmb);
43f3cec2	304	return 0;
630431cc	305	}
43f3cec2	306
630431cc JA	307	static int fio_pmemblk_close_file(struct thread_data fio_unused *td,
630431cc JA	308	struct fio_file *f)
43f3cec2	309	{
710bf9c5	310	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);
43f3cec2 BB	311
43f3cec2 BB	312	if (pmb)
630431cc	313	pmb_close(pmb, false);
43f3cec2	314
710bf9c5	315	FILE_SET_ENG_DATA(f, NULL);
43f3cec2	316	return 0;
630431cc	317	}
43f3cec2	318
c086b314	319	static int fio_pmemblk_get_file_size(struct thread_data td, struct fio_file f)
43f3cec2	320	{
c086b314	321	uint64_t flags = 0;
710bf9c5	322	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);
43f3cec2 BB	323
	324	if (fio_file_size_known(f))
	325	return 0;
	326
630431cc JA	327	if (!pmb) {
630431cc JA	328	if (pmb_get_flags(td, &flags))
43f3cec2 BB	329	return 1;
43f3cec2 BB	330	pmb = pmb_open(f->file_name, flags);
630431cc	331	if (!pmb)
43f3cec2 BB	332	return 1;
	333	}
	334
	335	f->real_file_size = pmb->pmb_bsize * pmb->pmb_nblocks;
	336
	337	fio_file_set_size_known(f);
	338
710bf9c5	339	if (!FILE_ENG_DATA(f))
630431cc	340	pmb_close(pmb, true);
43f3cec2 BB	341
43f3cec2 BB	342	return 0;
630431cc	343	}
43f3cec2	344
2e4ef4fb JA	345	static enum fio_q_status fio_pmemblk_queue(struct thread_data *td,
2e4ef4fb JA	346	struct io_u *io_u)
43f3cec2	347	{
c086b314	348	struct fio_file *f = io_u->file;
710bf9c5	349	fio_pmemblk_file_t pmb = FILE_ENG_DATA(f);
43f3cec2	350
c086b314 JA	351	unsigned long long off;
	352	unsigned long len;
	353	void *buf;
43f3cec2 BB	354
	355	fio_ro_check(td, io_u);
	356
	357	switch (io_u->ddir) {
	358	case DDIR_READ:
43f3cec2 BB	359	case DDIR_WRITE:
	360	off = io_u->offset;
	361	len = io_u->xfer_buflen;
	362
	363	io_u->error = EINVAL;
630431cc	364	if (off % pmb->pmb_bsize)
43f3cec2	365	break;
630431cc	366	if (len % pmb->pmb_bsize)
43f3cec2 BB	367	break;
	368	if ((off + len) / pmb->pmb_bsize > pmb->pmb_nblocks)
	369	break;
	370
	371	io_u->error = 0;
	372	buf = io_u->xfer_buf;
	373	off /= pmb->pmb_bsize;
	374	len /= pmb->pmb_bsize;
	375	while (0 < len) {
cf8775b8 RE	376	if (io_u->ddir == DDIR_READ &&
	377	0 != pmemblk_read(pmb->pmb_pool, buf, off)) {
	378	io_u->error = errno;
	379	break;
	380	} else if (0 != pmemblk_write(pmb->pmb_pool, buf, off)) {
43f3cec2 BB	381	io_u->error = errno;
	382	break;
	383	}
	384	buf += pmb->pmb_bsize;
	385	off++;
	386	len--;
	387	}
	388	off *= pmb->pmb_bsize;
	389	len *= pmb->pmb_bsize;
	390	io_u->resid = io_u->xfer_buflen - (off - io_u->offset);
	391	break;
	392	case DDIR_SYNC:
	393	case DDIR_DATASYNC:
	394	case DDIR_SYNC_FILE_RANGE:
	395	/* we're always sync'd */
	396	io_u->error = 0;
	397	break;
	398	default:
	399	io_u->error = EINVAL;
	400	break;
	401	}
	402
	403	return FIO_Q_COMPLETED;
630431cc	404	}
43f3cec2	405
c086b314	406	static int fio_pmemblk_unlink_file(struct thread_data td, struct fio_file f)
43f3cec2	407	{
c086b314 JA	408	char *path = NULL;
	409	uint64_t bsize = 0;
	410	uint64_t fsize = 0;
43f3cec2 BB	411
	412	/*
	413	* we need our own unlink in case the user has specified
	414	* the block and file sizes in the path name. we parse
	415	* the file_name to determine the file name we actually used.
	416	*/
	417
	418	pmb_parse_path(f->file_name, &path, &bsize, &fsize);
630431cc	419	if (!path)
2442c935	420	return ENOENT;
43f3cec2 BB	421
	422	unlink(path);
	423	free(path);
43f3cec2	424	return 0;
630431cc	425	}
43f3cec2	426
13690c10	427	static struct ioengine_ops ioengine = {
c086b314 JA	428	.name = "pmemblk",
	429	.version = FIO_IOOPS_VERSION,
	430	.queue = fio_pmemblk_queue,
	431	.open_file = fio_pmemblk_open_file,
	432	.close_file = fio_pmemblk_close_file,
	433	.get_file_size = fio_pmemblk_get_file_size,
	434	.unlink_file = fio_pmemblk_unlink_file,
	435	.flags = FIO_SYNCIO \| FIO_DISKLESSIO \| FIO_NOEXTEND \| FIO_NODISKUTIL,
43f3cec2 BB	436	};
43f3cec2 BB	437
630431cc	438	static void fio_init fio_pmemblk_register(void)
43f3cec2 BB	439	{
	440	register_ioengine(&ioengine);
	441	}
	442
630431cc	443	static void fio_exit fio_pmemblk_unregister(void)
43f3cec2 BB	444	{
	445	unregister_ioengine(&ioengine);
	446	}