X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=c6d814bf6d8765904f74b9ac01c0dfeaf48cb77a;hp=0a3988699987c282f6004b832e4096bc2de49f88;hb=f686097271b5f9bfcb05f61bce297c29b1e4ece1;hpb=094e66cbf2bc85617b3705eadc81940b202e5014

diff --git a/io_u.c b/io_u.c
index 0a398869..c6d814bf 100644
--- a/io_u.c
+++ b/io_u.c
@@ -62,6 +62,7 @@ static uint64_t last_block(struct thread_data *td, struct fio_file *f,
 
 	/*
 	 * Hmm, should we make sure that ->io_size <= ->real_file_size?
+	 * -> not for now since there is code assuming it could go either.
 	 */
 	max_size = f->io_size;
 	if (max_size > f->real_file_size)
@@ -177,7 +178,7 @@ bail:
 	/*
 	 * Generate a value, v, between 1 and 100, both inclusive
 	 */
-	v = rand_between(&td->zone_state, 1, 100);
+	v = rand32_between(&td->zone_state, 1, 100);
 
 	zsi = &td->zone_state_index[ddir][v - 1];
 	stotal = zsi->size_perc_prev;
@@ -279,7 +280,7 @@ static bool should_do_random(struct thread_data *td, enum fio_ddir ddir)
 	if (td->o.perc_rand[ddir] == 100)
 		return true;
 
-	v = rand_between(&td->seq_rand_state[ddir], 1, 100);
+	v = rand32_between(&td->seq_rand_state[ddir], 1, 100);
 
 	return v <= td->o.perc_rand[ddir];
 }
@@ -328,7 +329,8 @@ static int get_next_rand_block(struct thread_data *td, struct fio_file *f,
 	if (!get_next_rand_offset(td, f, ddir, b))
 		return 0;
 
-	if (td->o.time_based) {
+	if (td->o.time_based ||
+	    (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)) {
 		fio_file_reset(td, f);
 		if (!get_next_rand_offset(td, f, ddir, b))
 			return 0;
@@ -361,8 +363,12 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 	if (f->last_pos[ddir] < f->real_file_size) {
 		uint64_t pos;
 
-		if (f->last_pos[ddir] == f->file_offset && o->ddir_seq_add < 0)
-			f->last_pos[ddir] = f->real_file_size;
+		if (f->last_pos[ddir] == f->file_offset && o->ddir_seq_add < 0) {
+			if (f->real_file_size > f->io_size)
+				f->last_pos[ddir] = f->io_size;
+			else
+				f->last_pos[ddir] = f->real_file_size;
+		}
 
 		pos = f->last_pos[ddir] - f->file_offset;
 		if (pos && o->ddir_seq_add) {
@@ -371,10 +377,21 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 			/*
 			 * If we reach beyond the end of the file
 			 * with holed IO, wrap around to the
-			 * beginning again.
+			 * beginning again. If we're doing backwards IO,
+			 * wrap to the end.
 			 */
-			if (pos >= f->real_file_size)
-				pos = f->file_offset;
+			if (pos >= f->real_file_size) {
+				if (o->ddir_seq_add > 0)
+					pos = f->file_offset;
+				else {
+					if (f->real_file_size > f->io_size)
+						pos = f->io_size;
+					else
+						pos = f->real_file_size;
+
+					pos += o->ddir_seq_add;
+				}
+			}
 		}
 
 		*offset = pos;
@@ -403,7 +420,7 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 				*is_random = 1;
 			} else {
 				*is_random = 0;
-				io_u_set(io_u, IO_U_F_BUSY_OK);
+				io_u_set(td, io_u, IO_U_F_BUSY_OK);
 				ret = get_next_seq_offset(td, f, ddir, &offset);
 				if (ret)
 					ret = get_next_rand_block(td, f, ddir, &b);
@@ -413,7 +430,7 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 			ret = get_next_seq_offset(td, f, ddir, &offset);
 		}
 	} else {
-		io_u_set(io_u, IO_U_F_BUSY_OK);
+		io_u_set(td, io_u, IO_U_F_BUSY_OK);
 		*is_random = 0;
 
 		if (td->o.rw_seq == RW_SEQ_SEQ) {
@@ -515,8 +532,8 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 	int ddir = io_u->ddir;
 	unsigned int buflen = 0;
 	unsigned int minbs, maxbs;
-	uint64_t frand_max;
-	unsigned long r;
+	uint64_t frand_max, r;
+	bool power_2;
 
 	assert(ddir_rw(ddir));
 
@@ -545,7 +562,7 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 			if (buflen < minbs)
 				buflen = minbs;
 		} else {
-			long perc = 0;
+			long long perc = 0;
 			unsigned int i;
 
 			for (i = 0; i < td->o.bssplit_nr[ddir]; i++) {
@@ -553,19 +570,19 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 
 				buflen = bsp->bs;
 				perc += bsp->perc;
-				if ((r <= ((frand_max / 100L) * perc)) &&
+				if (!perc)
+					break;
+				if ((r / perc <= frand_max / 100ULL) &&
 				    io_u_fits(td, io_u, buflen))
 					break;
 			}
 		}
 
-		if (td->o.verify != VERIFY_NONE)
-			buflen = (buflen + td->o.verify_interval - 1) &
-				~(td->o.verify_interval - 1);
-
-		if (!td->o.bs_unaligned && is_power_of_2(minbs))
+		power_2 = is_power_of_2(minbs);
+		if (!td->o.bs_unaligned && power_2)
 			buflen &= ~(minbs - 1);
-
+		else if (!td->o.bs_unaligned && !power_2) 
+			buflen -= buflen % minbs; 
 	} while (!io_u_fits(td, io_u, buflen));
 
 	return buflen;
@@ -601,7 +618,7 @@ static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 {
 	unsigned int v;
 
-	v = rand_between(&td->rwmix_state, 1, 100);
+	v = rand32_between(&td->rwmix_state, 1, 100);
 
 	if (v <= td->o.rwmix[DDIR_READ])
 		return DDIR_READ;
@@ -629,20 +646,24 @@ int io_u_quiesce(struct thread_data *td)
 	}
 
 	while (td->io_u_in_flight) {
-		int fio_unused ret;
+		int ret;
 
 		ret = io_u_queued_complete(td, 1);
 		if (ret > 0)
 			completed += ret;
 	}
 
+	if (td->flags & TD_F_REGROW_LOGS)
+		regrow_logs(td);
+
 	return completed;
 }
 
 static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
 {
 	enum fio_ddir odir = ddir ^ 1;
-	long usec, now;
+	long usec;
+	uint64_t now;
 
 	assert(ddir_rw(ddir));
 	now = utime_since_now(&td->start);
@@ -696,28 +717,22 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 	enum fio_ddir ddir;
 
 	/*
-	 * see if it's time to fsync
-	 */
-	if (td->o.fsync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_SYNC;
-
-	/*
-	 * see if it's time to fdatasync
-	 */
-	if (td->o.fdatasync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_DATASYNC;
-
-	/*
-	 * see if it's time to sync_file_range
+	 * See if it's time to fsync/fdatasync/sync_file_range first,
+	 * and if not then move on to check regular I/Os.
 	 */
-	if (td->sync_file_range_nr &&
-	   !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_SYNC_FILE_RANGE;
+	if (should_fsync(td)) {
+		if (td->o.fsync_blocks && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks))
+			return DDIR_SYNC;
+
+		if (td->o.fdatasync_blocks && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks))
+			return DDIR_DATASYNC;
+
+		if (td->sync_file_range_nr && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr))
+			return DDIR_SYNC_FILE_RANGE;
+	}
 
 	if (td_rw(td)) {
 		/*
@@ -741,8 +756,10 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 		ddir = DDIR_READ;
 	else if (td_write(td))
 		ddir = DDIR_WRITE;
-	else
+	else if (td_trim(td))
 		ddir = DDIR_TRIM;
+	else
+		ddir = DDIR_INVAL;
 
 	td->rwmix_ddir = rate_ddir(td, ddir);
 	return td->rwmix_ddir;
@@ -762,11 +779,11 @@ static void set_rw_ddir(struct thread_data *td, struct io_u *io_u)
 
 	io_u->ddir = io_u->acct_ddir = ddir;
 
-	if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) &&
+	if (io_u->ddir == DDIR_WRITE && td_ioengine_flagged(td, FIO_BARRIER) &&
 	    td->o.barrier_blocks &&
 	   !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) &&
 	     td->io_issues[DDIR_WRITE])
-		io_u_set(io_u, IO_U_F_BARRIER);
+		io_u_set(td, io_u, IO_U_F_BARRIER);
 }
 
 void put_file_log(struct thread_data *td, struct fio_file *f)
@@ -788,7 +805,7 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 		put_file_log(td, io_u->file);
 
 	io_u->file = NULL;
-	io_u_set(io_u, IO_U_F_FREE);
+	io_u_set(td, io_u, IO_U_F_FREE);
 
 	if (io_u->flags & IO_U_F_IN_CUR_DEPTH) {
 		td->cur_depth--;
@@ -801,7 +818,7 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 
 void clear_io_u(struct thread_data *td, struct io_u *io_u)
 {
-	io_u_clear(io_u, IO_U_F_FLIGHT);
+	io_u_clear(td, io_u, IO_U_F_FLIGHT);
 	put_io_u(td, io_u);
 }
 
@@ -817,11 +834,11 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 
 	td_io_u_lock(td);
 
-	io_u_set(__io_u, IO_U_F_FREE);
+	io_u_set(td, __io_u, IO_U_F_FREE);
 	if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir))
 		td->io_issues[ddir]--;
 
-	io_u_clear(__io_u, IO_U_F_FLIGHT);
+	io_u_clear(td, __io_u, IO_U_F_FLIGHT);
 	if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) {
 		td->cur_depth--;
 		assert(!(td->flags & TD_F_CHILD));
@@ -837,7 +854,7 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 {
 	unsigned int is_random;
 
-	if (td->io_ops->flags & FIO_NOIO)
+	if (td_ioengine_flagged(td, FIO_NOIO))
 		goto out;
 
 	set_rw_ddir(td, io_u);
@@ -1065,6 +1082,34 @@ static void io_u_mark_latency(struct thread_data *td, unsigned long usec)
 		io_u_mark_lat_msec(td, usec / 1000);
 }
 
+static unsigned int __get_next_fileno_rand(struct thread_data *td)
+{
+	unsigned long fileno;
+
+	if (td->o.file_service_type == FIO_FSERVICE_RANDOM) {
+		uint64_t frand_max = rand_max(&td->next_file_state);
+		unsigned long r;
+
+		r = __rand(&td->next_file_state);
+		return (unsigned int) ((double) td->o.nr_files
+				* (r / (frand_max + 1.0)));
+	}
+
+	if (td->o.file_service_type == FIO_FSERVICE_ZIPF)
+		fileno = zipf_next(&td->next_file_zipf);
+	else if (td->o.file_service_type == FIO_FSERVICE_PARETO)
+		fileno = pareto_next(&td->next_file_zipf);
+	else if (td->o.file_service_type == FIO_FSERVICE_GAUSS)
+		fileno = gauss_next(&td->next_file_gauss);
+	else {
+		log_err("fio: bad file service type: %d\n", td->o.file_service_type);
+		assert(0);
+		return 0;
+	}
+
+	return fileno >> FIO_FSERVICE_SHIFT;
+}
+
 /*
  * Get next file to service by choosing one at random
  */
@@ -1072,17 +1117,13 @@ static struct fio_file *get_next_file_rand(struct thread_data *td,
 					   enum fio_file_flags goodf,
 					   enum fio_file_flags badf)
 {
-	uint64_t frand_max = rand_max(&td->next_file_state);
 	struct fio_file *f;
 	int fno;
 
 	do {
 		int opened = 0;
-		unsigned long r;
 
-		r = __rand(&td->next_file_state);
-		fno = (unsigned int) ((double) td->o.nr_files
-				* (r / (frand_max + 1.0)));
+		fno = __get_next_fileno_rand(td);
 
 		f = td->files[fno];
 		if (fio_file_done(f))
@@ -1235,10 +1276,14 @@ static long set_io_u_file(struct thread_data *td, struct io_u *io_u)
 		put_file_log(td, f);
 		td_io_close_file(td, f);
 		io_u->file = NULL;
-		fio_file_set_done(f);
-		td->nr_done_files++;
-		dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name,
+		if (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)
+			fio_file_reset(td, f);
+		else {
+			fio_file_set_done(f);
+			td->nr_done_files++;
+			dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name,
 					td->nr_done_files, td->o.nr_files);
+		}
 	} while (1);
 
 	return 0;
@@ -1423,7 +1468,7 @@ again:
 
 	if (io_u) {
 		assert(io_u->flags & IO_U_F_FREE);
-		io_u_clear(io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT |
+		io_u_clear(td, io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT |
 				 IO_U_F_TRIMMED | IO_U_F_BARRIER |
 				 IO_U_F_VER_LIST);
 
@@ -1431,7 +1476,7 @@ again:
 		io_u->acct_ddir = -1;
 		td->cur_depth++;
 		assert(!(td->flags & TD_F_CHILD));
-		io_u_set(io_u, IO_U_F_IN_CUR_DEPTH);
+		io_u_set(td, io_u, IO_U_F_IN_CUR_DEPTH);
 		io_u->ipo = NULL;
 	} else if (td_async_processing(td)) {
 		/*
@@ -1466,7 +1511,7 @@ static bool check_get_trim(struct thread_data *td, struct io_u *io_u)
 			get_trim = 1;
 		}
 
-		if (get_trim && !get_next_trim(td, io_u))
+		if (get_trim && get_next_trim(td, io_u))
 			return true;
 	}
 
@@ -1588,7 +1633,7 @@ struct io_u *get_io_u(struct thread_data *td)
 	assert(fio_file_open(f));
 
 	if (ddir_rw(io_u->ddir)) {
-		if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
+		if (!io_u->buflen && !td_ioengine_flagged(td, FIO_NOIO)) {
 			dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
 			goto err_put;
 		}
@@ -1686,6 +1731,9 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 	if (td->parent)
 		td = td->parent;
 
+	if (!td->o.stats)
+		return;
+
 	if (no_reduce)
 		lusec = utime_since(&io_u->issue_time, &icd->time);
 
@@ -1710,16 +1758,18 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 		}
 	}
 
-	if (!td->o.disable_clat) {
-		add_clat_sample(td, idx, lusec, bytes, io_u->offset);
-		io_u_mark_latency(td, lusec);
-	}
+	if (ddir_rw(idx)) {
+		if (!td->o.disable_clat) {
+			add_clat_sample(td, idx, lusec, bytes, io_u->offset);
+			io_u_mark_latency(td, lusec);
+		}
 
-	if (!td->o.disable_bw)
-		add_bw_sample(td, idx, bytes, &icd->time);
+		if (!td->o.disable_bw && per_unit_log(td->bw_log))
+			add_bw_sample(td, io_u, bytes, lusec);
 
-	if (no_reduce)
-		add_iops_sample(td, idx, bytes, &icd->time);
+		if (no_reduce && per_unit_log(td->iops_log))
+			add_iops_sample(td, io_u, bytes);
+	}
 
 	if (td->ts.nr_block_infos && io_u->ddir == DDIR_TRIM) {
 		uint32_t *info = io_u_block_info(td, io_u);
@@ -1735,6 +1785,28 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 	}
 }
 
+static void file_log_write_comp(const struct thread_data *td, struct fio_file *f,
+				uint64_t offset, unsigned int bytes)
+{
+	int idx;
+
+	if (!f)
+		return;
+
+	if (f->first_write == -1ULL || offset < f->first_write)
+		f->first_write = offset;
+	if (f->last_write == -1ULL || ((offset + bytes) > f->last_write))
+		f->last_write = offset + bytes;
+
+	if (!f->last_write_comp)
+		return;
+
+	idx = f->last_write_idx++;
+	f->last_write_comp[idx] = offset;
+	if (f->last_write_idx == td->o.iodepth)
+		f->last_write_idx = 0;
+}
+
 static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 			 struct io_completion_data *icd)
 {
@@ -1745,7 +1817,7 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 	dprint_io_u(io_u, "io complete");
 
 	assert(io_u->flags & IO_U_F_FLIGHT);
-	io_u_clear(io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
+	io_u_clear(td, io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
 
 	/*
 	 * Mark IO ok to verify
@@ -1785,23 +1857,8 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 		if (!(io_u->flags & IO_U_F_VER_LIST))
 			td->this_io_bytes[ddir] += bytes;
 
-		if (ddir == DDIR_WRITE) {
-			if (f) {
-				if (f->first_write == -1ULL ||
-				    io_u->offset < f->first_write)
-					f->first_write = io_u->offset;
-				if (f->last_write == -1ULL ||
-				    ((io_u->offset + bytes) > f->last_write))
-					f->last_write = io_u->offset + bytes;
-			}
-			if (td->last_write_comp) {
-				int idx = td->last_write_idx++;
-
-				td->last_write_comp[idx] = io_u->offset;
-				if (td->last_write_idx == td->o.iodepth)
-					td->last_write_idx = 0;
-			}
-		}
+		if (ddir == DDIR_WRITE)
+			file_log_write_comp(td, f, io_u->offset, bytes);
 
 		if (ramp_time_over(td) && (td->runstate == TD_RUNNING ||
 					   td->runstate == TD_VERIFYING))
@@ -1848,7 +1905,7 @@ static void init_icd(struct thread_data *td, struct io_completion_data *icd,
 	icd->nr = nr;
 
 	icd->error = 0;
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		icd->bytes_done[ddir] = 0;
 }
 
@@ -1887,7 +1944,7 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
 		return -1;
 	}
 
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return 0;
@@ -1903,7 +1960,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 	int ret, ddir;
 	struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
-	dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts);
+	dprint(FD_IO, "io_u_queued_complete: min=%d\n", min_evts);
 
 	if (!min_evts)
 		tvp = &ts;
@@ -1926,7 +1983,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 		return -1;
 	}
 
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return ret;
@@ -1937,7 +1994,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
  */
 void io_u_queued(struct thread_data *td, struct io_u *io_u)
 {
-	if (!td->o.disable_slat) {
+	if (!td->o.disable_slat && ramp_time_over(td) && td->o.stats) {
 		unsigned long slat_time;
 
 		slat_time = utime_since(&io_u->start_time, &io_u->issue_time);
@@ -1964,7 +2021,7 @@ static struct frand_state *get_buf_state(struct thread_data *td)
 		return &td->buf_state;
 	}
 
-	v = rand_between(&td->dedupe_state, 1, 100);
+	v = rand32_between(&td->dedupe_state, 1, 100);
 
 	if (v <= td->o.dedupe_percentage)
 		return &td->buf_state_prev;