[fio.git] / engines / mmap.c

/*
 * mmap engine
 *
 * IO engine that reads/writes from files by doing memcpy to/from
 * a memory mapped region of the file.
 *
 */
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/mman.h>

#include "../fio.h"
#include "../optgroup.h"
#include "../verify.h"

/*
 * Limits us to 1GiB of mapped files in total
 */
#define MMAP_TOTAL_SZ	(1 * 1024 * 1024 * 1024UL)

static unsigned long mmap_map_size;

struct fio_mmap_data {
	void *mmap_ptr;
	size_t mmap_sz;
	off_t mmap_off;
};

#ifdef CONFIG_HAVE_THP
struct mmap_options {
	void *pad;
	unsigned int thp;
};

static struct fio_option options[] = {
	{
		.name	= "thp",
		.lname	= "Transparent Huge Pages",
		.type	= FIO_OPT_INT,
		.off1	= offsetof(struct mmap_options, thp),
		.help	= "Memory Advise Huge Page",
		.category = FIO_OPT_C_ENGINE,
		.group	= FIO_OPT_G_MMAP,
	},
	{
		.name = NULL,
	},
};
#endif

static bool fio_madvise_file(struct thread_data *td, struct fio_file *f,
			     size_t length)

{
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
#ifdef CONFIG_HAVE_THP
	struct mmap_options *o = td->eo;

	/* Ignore errors on this optional advisory */
	if (o->thp)
		madvise(fmd->mmap_ptr, length, MADV_HUGEPAGE);
#endif

	if (!td->o.fadvise_hint)
		return true;

	if (!td_random(td)) {
		if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_SEQUENTIAL) < 0) {
			td_verror(td, errno, "madvise");
			return false;
		}
	} else {
		if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_RANDOM) < 0) {
			td_verror(td, errno, "madvise");
			return false;
		}
	}

	return true;
}

#ifdef CONFIG_HAVE_THP
static int fio_mmap_get_shared(struct thread_data *td)
{
	struct mmap_options *o = td->eo;

	if (o->thp)
		return MAP_PRIVATE;
	return MAP_SHARED;
}
#else
static int fio_mmap_get_shared(struct thread_data *td)
{
	return MAP_SHARED;
}
#endif

static int fio_mmap_file(struct thread_data *td, struct fio_file *f,
			 size_t length, off_t off)
{
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
	int flags = 0, shared = fio_mmap_get_shared(td);

	if (td_rw(td) && !td->o.verify_only)
		flags = PROT_READ | PROT_WRITE;
	else if (td_write(td) && !td->o.verify_only) {
		flags = PROT_WRITE;

		if (td->o.verify != VERIFY_NONE)
			flags |= PROT_READ;
	} else
		flags = PROT_READ;

	fmd->mmap_ptr = mmap(NULL, length, flags, shared, f->fd, off);
	if (fmd->mmap_ptr == MAP_FAILED) {
		fmd->mmap_ptr = NULL;
		td_verror(td, errno, "mmap");
		goto err;
	}

	if (!fio_madvise_file(td, f, length))
		goto err;

	if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_DONTNEED) < 0) {
		td_verror(td, errno, "madvise");
		goto err;
	}

#ifdef FIO_MADV_FREE
	if (f->filetype == FIO_TYPE_BLOCK)
		(void) posix_madvise(fmd->mmap_ptr, fmd->mmap_sz, FIO_MADV_FREE);
#endif

err:
	if (td->error && fmd->mmap_ptr)
		munmap(fmd->mmap_ptr, length);

	return td->error;
}

/*
 * Just mmap an appropriate portion, we cannot mmap the full extent
 */
static int fio_mmapio_prep_limited(struct thread_data *td, struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);

	if (io_u->buflen > mmap_map_size) {
		log_err("fio: bs too big for mmap engine\n");
		return EIO;
	}

	fmd->mmap_sz = mmap_map_size;
	if (fmd->mmap_sz  > f->io_size)
		fmd->mmap_sz = f->io_size;

	fmd->mmap_off = io_u->offset;

	return fio_mmap_file(td, f, fmd->mmap_sz, fmd->mmap_off);
}

/*
 * Attempt to mmap the entire file
 */
static int fio_mmapio_prep_full(struct thread_data *td, struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
	int ret;

	if (fio_file_partial_mmap(f))
		return EINVAL;
	if (io_u->offset != (size_t) io_u->offset ||
	    f->io_size != (size_t) f->io_size) {
		fio_file_set_partial_mmap(f);
		return EINVAL;
	}

	fmd->mmap_sz = f->io_size;
	fmd->mmap_off = 0;

	ret = fio_mmap_file(td, f, fmd->mmap_sz, fmd->mmap_off);
	if (ret)
		fio_file_set_partial_mmap(f);

	return ret;
}

static int fio_mmapio_prep(struct thread_data *td, struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
	int ret;

	/*
	 * It fits within existing mapping, use it
	 */
	if (io_u->offset >= fmd->mmap_off &&
	    io_u->offset + io_u->buflen <= fmd->mmap_off + fmd->mmap_sz)
		goto done;

	/*
	 * unmap any existing mapping
	 */
	if (fmd->mmap_ptr) {
		if (munmap(fmd->mmap_ptr, fmd->mmap_sz) < 0)
			return errno;
		fmd->mmap_ptr = NULL;
	}

	if (fio_mmapio_prep_full(td, io_u)) {
		td_clear_error(td);
		ret = fio_mmapio_prep_limited(td, io_u);
		if (ret)
			return ret;
	}

done:
	io_u->mmap_data = fmd->mmap_ptr + io_u->offset - fmd->mmap_off -
				f->file_offset;
	return 0;
}

static enum fio_q_status fio_mmapio_queue(struct thread_data *td,
					  struct io_u *io_u)
{
	struct fio_file *f = io_u->file;
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);

	fio_ro_check(td, io_u);

	if (io_u->ddir == DDIR_READ)
		memcpy(io_u->xfer_buf, io_u->mmap_data, io_u->xfer_buflen);
	else if (io_u->ddir == DDIR_WRITE)
		memcpy(io_u->mmap_data, io_u->xfer_buf, io_u->xfer_buflen);
	else if (ddir_sync(io_u->ddir)) {
		if (msync(fmd->mmap_ptr, fmd->mmap_sz, MS_SYNC)) {
			io_u->error = errno;
			td_verror(td, io_u->error, "msync");
		}
	} else if (io_u->ddir == DDIR_TRIM) {
		int ret = do_io_u_trim(td, io_u);

		if (!ret)
			td_verror(td, io_u->error, "trim");
	}


	/*
	 * not really direct, but should drop the pages from the cache
	 */
	if (td->o.odirect && ddir_rw(io_u->ddir)) {
		if (msync(io_u->mmap_data, io_u->xfer_buflen, MS_SYNC) < 0) {
			io_u->error = errno;
			td_verror(td, io_u->error, "msync");
		}
		if (posix_madvise(io_u->mmap_data, io_u->xfer_buflen, POSIX_MADV_DONTNEED) < 0) {
			io_u->error = errno;
			td_verror(td, io_u->error, "madvise");
		}
	}

	return FIO_Q_COMPLETED;
}

static int fio_mmapio_init(struct thread_data *td)
{
	struct thread_options *o = &td->o;

	if ((o->rw_min_bs & page_mask) &&
	    (o->odirect || o->fsync_blocks || o->fdatasync_blocks)) {
		log_err("fio: mmap options dictate a minimum block size of "
			"%llu bytes\n", (unsigned long long) page_size);
		return 1;
	}

	mmap_map_size = MMAP_TOTAL_SZ / o->nr_files;
	return 0;
}

static int fio_mmapio_open_file(struct thread_data *td, struct fio_file *f)
{
	struct fio_mmap_data *fmd;
	int ret;

	ret = generic_open_file(td, f);
	if (ret)
		return ret;

	fmd = calloc(1, sizeof(*fmd));
	if (!fmd) {
		int fio_unused __ret;
		__ret = generic_close_file(td, f);
		return 1;
	}

	FILE_SET_ENG_DATA(f, fmd);
	return 0;
}

static int fio_mmapio_close_file(struct thread_data *td, struct fio_file *f)
{
	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);

	FILE_SET_ENG_DATA(f, NULL);
	free(fmd);
	fio_file_clear_partial_mmap(f);

	return generic_close_file(td, f);
}

static struct ioengine_ops ioengine = {
	.name		= "mmap",
	.version	= FIO_IOOPS_VERSION,
	.init		= fio_mmapio_init,
	.prep		= fio_mmapio_prep,
	.queue		= fio_mmapio_queue,
	.open_file	= fio_mmapio_open_file,
	.close_file	= fio_mmapio_close_file,
	.get_file_size	= generic_get_file_size,
	.flags		= FIO_SYNCIO | FIO_NOEXTEND,
#ifdef CONFIG_HAVE_THP
	.options	= options,
	.option_struct_size = sizeof(struct mmap_options),
#endif
};

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

static void fio_exit fio_mmapio_unregister(void)
{
	unregister_ioengine(&ioengine);
}
Commit	Line	Data
2866c82d	1	/*
da751ca9 JA	2	* mmap engine
	3	*
	4	* IO engine that reads/writes from files by doing memcpy to/from
	5	* a memory mapped region of the file.
2866c82d JA	6	*
	7	*/
	8	#include <stdio.h>
	9	#include <stdlib.h>
2866c82d	10	#include <errno.h>
2866c82d	11	#include <sys/mman.h>
5f350952 JA	12
5f350952 JA	13	#include "../fio.h"
ff547caa	14	#include "../optgroup.h"
4f5af7b2	15	#include "../verify.h"
2866c82d	16
ac893112	17	/*
420b104a	18	* Limits us to 1GiB of mapped files in total
ac893112	19	*/
ff455a04	20	#define MMAP_TOTAL_SZ (1 * 1024 * 1024 * 1024UL)
2866c82d	21
ac893112	22	static unsigned long mmap_map_size;
2866c82d	23
03a32636 JA	24	struct fio_mmap_data {
	25	void *mmap_ptr;
	26	size_t mmap_sz;
	27	off_t mmap_off;
	28	};
	29
ff547caa KB	30	#ifdef CONFIG_HAVE_THP
	31	struct mmap_options {
	32	void *pad;
	33	unsigned int thp;
	34	};
	35
	36	static struct fio_option options[] = {
	37	{
	38	.name = "thp",
	39	.lname = "Transparent Huge Pages",
	40	.type = FIO_OPT_INT,
	41	.off1 = offsetof(struct mmap_options, thp),
	42	.help = "Memory Advise Huge Page",
	43	.category = FIO_OPT_C_ENGINE,
	44	.group = FIO_OPT_G_MMAP,
	45	},
	46	{
	47	.name = NULL,
	48	},
	49	};
	50	#endif
	51
c712c97a JA	52	static bool fio_madvise_file(struct thread_data td, struct fio_file f,
	53	size_t length)
	54
	55	{
	56	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
ff547caa KB	57	#ifdef CONFIG_HAVE_THP
	58	struct mmap_options *o = td->eo;
	59
	60	/* Ignore errors on this optional advisory */
	61	if (o->thp)
	62	madvise(fmd->mmap_ptr, length, MADV_HUGEPAGE);
	63	#endif
c712c97a JA	64
	65	if (!td->o.fadvise_hint)
	66	return true;
	67
	68	if (!td_random(td)) {
	69	if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_SEQUENTIAL) < 0) {
	70	td_verror(td, errno, "madvise");
	71	return false;
	72	}
	73	} else {
	74	if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_RANDOM) < 0) {
	75	td_verror(td, errno, "madvise");
	76	return false;
	77	}
	78	}
c712c97a JA	79
	80	return true;
	81	}
	82
ff547caa KB	83	#ifdef CONFIG_HAVE_THP
	84	static int fio_mmap_get_shared(struct thread_data *td)
	85	{
	86	struct mmap_options *o = td->eo;
	87
	88	if (o->thp)
	89	return MAP_PRIVATE;
	90	return MAP_SHARED;
	91	}
	92	#else
	93	static int fio_mmap_get_shared(struct thread_data *td)
	94	{
	95	return MAP_SHARED;
	96	}
	97	#endif
	98
ac893112 JA	99	static int fio_mmap_file(struct thread_data td, struct fio_file f,
ac893112 JA	100	size_t length, off_t off)
b5af8293	101	{
e19ccb55	102	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
ff547caa	103	int flags = 0, shared = fio_mmap_get_shared(td);
c97d8369	104
81a1ee61	105	if (td_rw(td) && !td->o.verify_only)
b5af8293	106	flags = PROT_READ \| PROT_WRITE;
81a1ee61	107	else if (td_write(td) && !td->o.verify_only) {
b5af8293 JA	108	flags = PROT_WRITE;
b5af8293 JA	109
2dc1bbeb	110	if (td->o.verify != VERIFY_NONE)
b5af8293 JA	111	flags \|= PROT_READ;
	112	} else
	113	flags = PROT_READ;
	114
ff547caa	115	fmd->mmap_ptr = mmap(NULL, length, flags, shared, f->fd, off);
03a32636 JA	116	if (fmd->mmap_ptr == MAP_FAILED) {
03a32636 JA	117	fmd->mmap_ptr = NULL;
ed47cbf7	118	td_verror(td, errno, "mmap");
b5af8293 JA	119	goto err;
	120	}
	121
c712c97a	122	if (!fio_madvise_file(td, f, length))
ecbfcd94	123	goto err;
ecbfcd94	124
c69f6bf3 JA	125	if (posix_madvise(fmd->mmap_ptr, length, POSIX_MADV_DONTNEED) < 0) {
	126	td_verror(td, errno, "madvise");
	127	goto err;
	128	}
	129
	130	#ifdef FIO_MADV_FREE
	131	if (f->filetype == FIO_TYPE_BLOCK)
	132	(void) posix_madvise(fmd->mmap_ptr, fmd->mmap_sz, FIO_MADV_FREE);
	133	#endif
	134
b5af8293	135	err:
03a32636 JA	136	if (td->error && fmd->mmap_ptr)
03a32636 JA	137	munmap(fmd->mmap_ptr, length);
93bcfd20	138
ed47cbf7	139	return td->error;
b5af8293 JA	140	}
b5af8293 JA	141
ed47cbf7 JA	142	/*
	143	* Just mmap an appropriate portion, we cannot mmap the full extent
	144	*/
	145	static int fio_mmapio_prep_limited(struct thread_data td, struct io_u io_u)
b5af8293	146	{
ac893112	147	struct fio_file *f = io_u->file;
e19ccb55	148	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
6977bcd0	149
ac893112 JA	150	if (io_u->buflen > mmap_map_size) {
ac893112 JA	151	log_err("fio: bs too big for mmap engine\n");
ed47cbf7	152	return EIO;
ac893112 JA	153	}
ac893112 JA	154
03a32636 JA	155	fmd->mmap_sz = mmap_map_size;
	156	if (fmd->mmap_sz > f->io_size)
	157	fmd->mmap_sz = f->io_size;
6977bcd0	158
03a32636	159	fmd->mmap_off = io_u->offset;
ac893112	160
03a32636	161	return fio_mmap_file(td, f, fmd->mmap_sz, fmd->mmap_off);
ed47cbf7 JA	162	}
	163
	164	/*
	165	* Attempt to mmap the entire file
	166	*/
	167	static int fio_mmapio_prep_full(struct thread_data td, struct io_u io_u)
	168	{
	169	struct fio_file *f = io_u->file;
e19ccb55	170	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
ed47cbf7 JA	171	int ret;
	172
	173	if (fio_file_partial_mmap(f))
	174	return EINVAL;
01626837 JA	175	if (io_u->offset != (size_t) io_u->offset \|\|
	176	f->io_size != (size_t) f->io_size) {
	177	fio_file_set_partial_mmap(f);
	178	return EINVAL;
	179	}
ed47cbf7	180
03a32636 JA	181	fmd->mmap_sz = f->io_size;
03a32636 JA	182	fmd->mmap_off = 0;
ed47cbf7	183
03a32636	184	ret = fio_mmap_file(td, f, fmd->mmap_sz, fmd->mmap_off);
ed47cbf7 JA	185	if (ret)
	186	fio_file_set_partial_mmap(f);
	187
6977bcd0	188	return ret;
b5af8293 JA	189	}
b5af8293 JA	190
ed47cbf7 JA	191	static int fio_mmapio_prep(struct thread_data td, struct io_u io_u)
	192	{
	193	struct fio_file *f = io_u->file;
e19ccb55	194	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
ed47cbf7 JA	195	int ret;
ed47cbf7 JA	196
8f933cae JA	197	/*
	198	* It fits within existing mapping, use it
	199	*/
03a32636	200	if (io_u->offset >= fmd->mmap_off &&
f23ff35e	201	io_u->offset + io_u->buflen <= fmd->mmap_off + fmd->mmap_sz)
ed47cbf7 JA	202	goto done;
ed47cbf7 JA	203
8f933cae JA	204	/*
	205	* unmap any existing mapping
	206	*/
03a32636 JA	207	if (fmd->mmap_ptr) {
03a32636 JA	208	if (munmap(fmd->mmap_ptr, fmd->mmap_sz) < 0)
8f933cae	209	return errno;
03a32636	210	fmd->mmap_ptr = NULL;
8f933cae JA	211	}
8f933cae JA	212
ed47cbf7 JA	213	if (fio_mmapio_prep_full(td, io_u)) {
	214	td_clear_error(td);
	215	ret = fio_mmapio_prep_limited(td, io_u);
	216	if (ret)
	217	return ret;
	218	}
	219
	220	done:
03a32636	221	io_u->mmap_data = fmd->mmap_ptr + io_u->offset - fmd->mmap_off -
ed47cbf7 JA	222	f->file_offset;
	223	return 0;
	224	}
	225
2e4ef4fb JA	226	static enum fio_q_status fio_mmapio_queue(struct thread_data *td,
2e4ef4fb JA	227	struct io_u *io_u)
ac893112 JA	228	{
ac893112 JA	229	struct fio_file *f = io_u->file;
e19ccb55	230	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
ac893112 JA	231
	232	fio_ro_check(td, io_u);
	233
	234	if (io_u->ddir == DDIR_READ)
	235	memcpy(io_u->xfer_buf, io_u->mmap_data, io_u->xfer_buflen);
	236	else if (io_u->ddir == DDIR_WRITE)
	237	memcpy(io_u->mmap_data, io_u->xfer_buf, io_u->xfer_buflen);
5f9099ea	238	else if (ddir_sync(io_u->ddir)) {
03a32636	239	if (msync(fmd->mmap_ptr, fmd->mmap_sz, MS_SYNC)) {
ac893112 JA	240	io_u->error = errno;
	241	td_verror(td, io_u->error, "msync");
	242	}
ff58fced JA	243	} else if (io_u->ddir == DDIR_TRIM) {
	244	int ret = do_io_u_trim(td, io_u);
	245
	246	if (!ret)
	247	td_verror(td, io_u->error, "trim");
ac893112 JA	248	}
ac893112 JA	249
ff58fced	250
ac893112 JA	251	/*
	252	* not really direct, but should drop the pages from the cache
	253	*/
ff58fced	254	if (td->o.odirect && ddir_rw(io_u->ddir)) {
ac893112 JA	255	if (msync(io_u->mmap_data, io_u->xfer_buflen, MS_SYNC) < 0) {
	256	io_u->error = errno;
	257	td_verror(td, io_u->error, "msync");
	258	}
03e20d68	259	if (posix_madvise(io_u->mmap_data, io_u->xfer_buflen, POSIX_MADV_DONTNEED) < 0) {
ac893112 JA	260	io_u->error = errno;
	261	td_verror(td, io_u->error, "madvise");
	262	}
	263	}
	264
	265	return FIO_Q_COMPLETED;
	266	}
	267
	268	static int fio_mmapio_init(struct thread_data *td)
	269	{
913ea0db	270	struct thread_options *o = &td->o;
ac893112	271
487197d9	272	if ((o->rw_min_bs & page_mask) &&
913ea0db JA	273	(o->odirect \|\| o->fsync_blocks \|\| o->fdatasync_blocks)) {
913ea0db JA	274	log_err("fio: mmap options dictate a minimum block size of "
4e0a8fa2	275	"%llu bytes\n", (unsigned long long) page_size);
913ea0db JA	276	return 1;
	277	}
	278
487197d9	279	mmap_map_size = MMAP_TOTAL_SZ / o->nr_files;
ac893112 JA	280	return 0;
	281	}
	282
03a32636 JA	283	static int fio_mmapio_open_file(struct thread_data td, struct fio_file f)
	284	{
	285	struct fio_mmap_data *fmd;
	286	int ret;
	287
	288	ret = generic_open_file(td, f);
	289	if (ret)
	290	return ret;
	291
	292	fmd = calloc(1, sizeof(*fmd));
	293	if (!fmd) {
8a68c41c JA	294	int fio_unused __ret;
8a68c41c JA	295	__ret = generic_close_file(td, f);
03a32636 JA	296	return 1;
	297	}
	298
e19ccb55	299	FILE_SET_ENG_DATA(f, fmd);
03a32636 JA	300	return 0;
	301	}
	302
	303	static int fio_mmapio_close_file(struct thread_data td, struct fio_file f)
	304	{
e19ccb55	305	struct fio_mmap_data *fmd = FILE_ENG_DATA(f);
03a32636	306
e19ccb55	307	FILE_SET_ENG_DATA(f, NULL);
03a32636	308	free(fmd);
965f0a41	309	fio_file_clear_partial_mmap(f);
03a32636 JA	310
	311	return generic_close_file(td, f);
	312	}
	313
5f350952	314	static struct ioengine_ops ioengine = {
2866c82d JA	315	.name = "mmap",
2866c82d JA	316	.version = FIO_IOOPS_VERSION,
ac893112 JA	317	.init = fio_mmapio_init,
ac893112 JA	318	.prep = fio_mmapio_prep,
2866c82d	319	.queue = fio_mmapio_queue,
03a32636 JA	320	.open_file = fio_mmapio_open_file,
03a32636 JA	321	.close_file = fio_mmapio_close_file,
df9c26b1	322	.get_file_size = generic_get_file_size,
0263882a	323	.flags = FIO_SYNCIO \| FIO_NOEXTEND,
ff547caa KB	324	#ifdef CONFIG_HAVE_THP
	325	.options = options,
	326	.option_struct_size = sizeof(struct mmap_options),
	327	#endif
2866c82d	328	};
5f350952 JA	329
	330	static void fio_init fio_mmapio_register(void)
	331	{
	332	register_ioengine(&ioengine);
	333	}
	334
	335	static void fio_exit fio_mmapio_unregister(void)
	336	{
	337	unregister_ioengine(&ioengine);
	338	}