2 * Copyright (C) 2018 Western Digital Corporation or its affiliates.
4 * This file is released under the GPL.
19 #include "oslib/asprintf.h"
26 * zbd_get_zoned_model - Get a device zoned model
27 * @td: FIO thread data
28 * @f: FIO file for which to get model information
30 int zbd_get_zoned_model(struct thread_data *td, struct fio_file *f,
31 enum zbd_zoned_model *model)
35 if (f->filetype == FIO_TYPE_PIPE) {
36 log_err("zonemode=zbd does not support pipes\n");
40 /* If regular file, always emulate zones inside the file. */
41 if (f->filetype == FIO_TYPE_FILE) {
46 if (td->io_ops && td->io_ops->get_zoned_model)
47 ret = td->io_ops->get_zoned_model(td, f, model);
49 ret = blkzoned_get_zoned_model(td, f, model);
51 td_verror(td, errno, "get zoned model failed");
52 log_err("%s: get zoned model failed (%d).\n",
60 * zbd_report_zones - Get zone information
61 * @td: FIO thread data.
62 * @f: FIO file for which to get zone information
63 * @offset: offset from which to report zones
64 * @zones: Array of struct zbd_zone
65 * @nr_zones: Size of @zones array
67 * Get zone information into @zones starting from the zone at offset @offset
68 * for the device specified by @f.
70 * Returns the number of zones reported upon success and a negative error code
71 * upon failure. If the zone report is empty, always assume an error (device
72 * problem) and return -EIO.
74 int zbd_report_zones(struct thread_data *td, struct fio_file *f,
75 uint64_t offset, struct zbd_zone *zones,
76 unsigned int nr_zones)
80 if (td->io_ops && td->io_ops->report_zones)
81 ret = td->io_ops->report_zones(td, f, offset, zones, nr_zones);
83 ret = blkzoned_report_zones(td, f, offset, zones, nr_zones);
85 td_verror(td, errno, "report zones failed");
86 log_err("%s: report zones from sector %llu failed (%d).\n",
87 f->file_name, (unsigned long long)offset >> 9, errno);
88 } else if (ret == 0) {
89 td_verror(td, errno, "Empty zone report");
90 log_err("%s: report zones from sector %llu is empty.\n",
91 f->file_name, (unsigned long long)offset >> 9);
99 * zbd_reset_wp - reset the write pointer of a range of zones
100 * @td: FIO thread data.
101 * @f: FIO file for which to reset zones
102 * @offset: Starting offset of the first zone to reset
103 * @length: Length of the range of zones to reset
105 * Reset the write pointer of all zones in the range @offset...@offset+@length.
106 * Returns 0 upon success and a negative error code upon failure.
108 int zbd_reset_wp(struct thread_data *td, struct fio_file *f,
109 uint64_t offset, uint64_t length)
113 if (td->io_ops && td->io_ops->reset_wp)
114 ret = td->io_ops->reset_wp(td, f, offset, length);
116 ret = blkzoned_reset_wp(td, f, offset, length);
118 td_verror(td, errno, "resetting wp failed");
119 log_err("%s: resetting wp for %llu sectors at sector %llu failed (%d).\n",
120 f->file_name, (unsigned long long)length >> 9,
121 (unsigned long long)offset >> 9, errno);
128 * zbd_get_max_open_zones - Get the maximum number of open zones
129 * @td: FIO thread data
130 * @f: FIO file for which to get max open zones
131 * @max_open_zones: Upon success, result will be stored here.
133 * A @max_open_zones value set to zero means no limit.
135 * Returns 0 upon success and a negative error code upon failure.
137 int zbd_get_max_open_zones(struct thread_data *td, struct fio_file *f,
138 unsigned int *max_open_zones)
142 if (td->io_ops && td->io_ops->get_max_open_zones)
143 ret = td->io_ops->get_max_open_zones(td, f, max_open_zones);
145 ret = blkzoned_get_max_open_zones(td, f, max_open_zones);
147 td_verror(td, errno, "get max open zones failed");
148 log_err("%s: get max open zones failed (%d).\n",
149 f->file_name, errno);
156 * zbd_zone_idx - convert an offset into a zone number
158 * @offset: offset in bytes. If this offset is in the first zone_size bytes
159 * past the disk size then the index of the sentinel is returned.
161 static uint32_t zbd_zone_idx(const struct fio_file *f, uint64_t offset)
165 if (f->zbd_info->zone_size_log2 > 0)
166 zone_idx = offset >> f->zbd_info->zone_size_log2;
168 zone_idx = offset / f->zbd_info->zone_size;
170 return min(zone_idx, f->zbd_info->nr_zones);
174 * zbd_zone_end - Return zone end location
175 * @z: zone info pointer.
177 static inline uint64_t zbd_zone_end(const struct fio_zone_info *z)
183 * zbd_zone_capacity_end - Return zone capacity limit end location
184 * @z: zone info pointer.
186 static inline uint64_t zbd_zone_capacity_end(const struct fio_zone_info *z)
188 return z->start + z->capacity;
192 * zbd_zone_full - verify whether a minimum number of bytes remain in a zone
194 * @z: zone info pointer.
195 * @required: minimum number of bytes that must remain in a zone.
197 * The caller must hold z->mutex.
199 static bool zbd_zone_full(const struct fio_file *f, struct fio_zone_info *z,
202 assert((required & 511) == 0);
205 z->wp + required > zbd_zone_capacity_end(z);
208 static void zone_lock(struct thread_data *td, const struct fio_file *f,
209 struct fio_zone_info *z)
211 struct zoned_block_device_info *zbd = f->zbd_info;
212 uint32_t nz = z - zbd->zone_info;
214 /* A thread should never lock zones outside its working area. */
215 assert(f->min_zone <= nz && nz < f->max_zone);
220 * Lock the io_u target zone. The zone will be unlocked if io_u offset
221 * is changed or when io_u completes and zbd_put_io() executed.
222 * To avoid multiple jobs doing asynchronous I/Os from deadlocking each
223 * other waiting for zone locks when building an io_u batch, first
224 * only trylock the zone. If the zone is already locked by another job,
225 * process the currently queued I/Os so that I/O progress is made and
228 if (pthread_mutex_trylock(&z->mutex) != 0) {
229 if (!td_ioengine_flagged(td, FIO_SYNCIO))
231 pthread_mutex_lock(&z->mutex);
235 static inline void zone_unlock(struct fio_zone_info *z)
240 ret = pthread_mutex_unlock(&z->mutex);
244 static bool is_valid_offset(const struct fio_file *f, uint64_t offset)
246 return (uint64_t)(offset - f->file_offset) < f->io_size;
249 static inline struct fio_zone_info *get_zone(const struct fio_file *f,
250 unsigned int zone_nr)
252 return &f->zbd_info->zone_info[zone_nr];
255 /* Verify whether direct I/O is used for all host-managed zoned drives. */
256 static bool zbd_using_direct_io(void)
258 struct thread_data *td;
263 if (td->o.odirect || !(td->o.td_ddir & TD_DDIR_WRITE))
265 for_each_file(td, f, j) {
267 f->zbd_info->model == ZBD_HOST_MANAGED)
275 /* Whether or not the I/O range for f includes one or more sequential zones */
276 static bool zbd_is_seq_job(struct fio_file *f)
278 uint32_t zone_idx, zone_idx_b, zone_idx_e;
283 zone_idx_b = zbd_zone_idx(f, f->file_offset);
284 zone_idx_e = zbd_zone_idx(f, f->file_offset + f->io_size - 1);
285 for (zone_idx = zone_idx_b; zone_idx <= zone_idx_e; zone_idx++)
286 if (get_zone(f, zone_idx)->has_wp)
293 * Verify whether offset and size parameters are aligned with zone boundaries.
295 static bool zbd_verify_sizes(void)
297 const struct fio_zone_info *z;
298 struct thread_data *td;
300 uint64_t new_offset, new_end;
305 for_each_file(td, f, j) {
308 if (f->file_offset >= f->real_file_size)
310 if (!zbd_is_seq_job(f))
313 if (!td->o.zone_size) {
314 td->o.zone_size = f->zbd_info->zone_size;
315 if (!td->o.zone_size) {
316 log_err("%s: invalid 0 zone size\n",
320 } else if (td->o.zone_size != f->zbd_info->zone_size) {
321 log_err("%s: job parameter zonesize %llu does not match disk zone size %llu.\n",
322 f->file_name, (unsigned long long) td->o.zone_size,
323 (unsigned long long) f->zbd_info->zone_size);
327 if (td->o.zone_skip % td->o.zone_size) {
328 log_err("%s: zoneskip %llu is not a multiple of the device zone size %llu.\n",
329 f->file_name, (unsigned long long) td->o.zone_skip,
330 (unsigned long long) td->o.zone_size);
334 zone_idx = zbd_zone_idx(f, f->file_offset);
335 z = get_zone(f, zone_idx);
336 if ((f->file_offset != z->start) &&
337 (td->o.td_ddir != TD_DDIR_READ)) {
338 new_offset = zbd_zone_end(z);
339 if (new_offset >= f->file_offset + f->io_size) {
340 log_info("%s: io_size must be at least one zone\n",
344 log_info("%s: rounded up offset from %llu to %llu\n",
345 f->file_name, (unsigned long long) f->file_offset,
346 (unsigned long long) new_offset);
347 f->io_size -= (new_offset - f->file_offset);
348 f->file_offset = new_offset;
350 zone_idx = zbd_zone_idx(f, f->file_offset + f->io_size);
351 z = get_zone(f, zone_idx);
353 if ((td->o.td_ddir != TD_DDIR_READ) &&
354 (f->file_offset + f->io_size != new_end)) {
355 if (new_end <= f->file_offset) {
356 log_info("%s: io_size must be at least one zone\n",
360 log_info("%s: rounded down io_size from %llu to %llu\n",
361 f->file_name, (unsigned long long) f->io_size,
362 (unsigned long long) new_end - f->file_offset);
363 f->io_size = new_end - f->file_offset;
371 static bool zbd_verify_bs(void)
373 struct thread_data *td;
378 for_each_file(td, f, j) {
383 zone_size = f->zbd_info->zone_size;
384 for (k = 0; k < FIO_ARRAY_SIZE(td->o.bs); k++) {
385 if (td->o.verify != VERIFY_NONE &&
386 zone_size % td->o.bs[k] != 0) {
387 log_info("%s: block size %llu is not a divisor of the zone size %llu\n",
388 f->file_name, td->o.bs[k],
389 (unsigned long long)zone_size);
398 static int ilog2(uint64_t i)
410 * Initialize f->zbd_info for devices that are not zoned block devices. This
411 * allows to execute a ZBD workload against a non-ZBD device.
413 static int init_zone_info(struct thread_data *td, struct fio_file *f)
416 struct fio_zone_info *p;
417 uint64_t zone_size = td->o.zone_size;
418 uint64_t zone_capacity = td->o.zone_capacity;
419 struct zoned_block_device_info *zbd_info = NULL;
422 if (zone_size == 0) {
423 log_err("%s: Specifying the zone size is mandatory for regular file/block device with --zonemode=zbd\n\n",
428 if (zone_size < 512) {
429 log_err("%s: zone size must be at least 512 bytes for --zonemode=zbd\n\n",
434 if (zone_capacity == 0)
435 zone_capacity = zone_size;
437 if (zone_capacity > zone_size) {
438 log_err("%s: job parameter zonecapacity %llu is larger than zone size %llu\n",
439 f->file_name, (unsigned long long) td->o.zone_capacity,
440 (unsigned long long) td->o.zone_size);
444 if (f->real_file_size < zone_size) {
445 log_err("%s: file/device size %"PRIu64" is smaller than zone size %"PRIu64"\n",
446 f->file_name, f->real_file_size, zone_size);
450 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
451 zbd_info = scalloc(1, sizeof(*zbd_info) +
452 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
456 mutex_init_pshared(&zbd_info->mutex);
457 zbd_info->refcount = 1;
458 p = &zbd_info->zone_info[0];
459 for (i = 0; i < nr_zones; i++, p++) {
460 mutex_init_pshared_with_type(&p->mutex,
461 PTHREAD_MUTEX_RECURSIVE);
462 p->start = i * zone_size;
464 p->type = ZBD_ZONE_TYPE_SWR;
465 p->cond = ZBD_ZONE_COND_EMPTY;
466 p->capacity = zone_capacity;
470 p->start = nr_zones * zone_size;
472 f->zbd_info = zbd_info;
473 f->zbd_info->zone_size = zone_size;
474 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
475 ilog2(zone_size) : 0;
476 f->zbd_info->nr_zones = nr_zones;
481 * Maximum number of zones to report in one operation.
483 #define ZBD_REPORT_MAX_ZONES 8192U
486 * Parse the device zone report and store it in f->zbd_info. Must be called
487 * only for devices that are zoned, namely those with a model != ZBD_NONE.
489 static int parse_zone_info(struct thread_data *td, struct fio_file *f)
492 struct zbd_zone *zones, *z;
493 struct fio_zone_info *p;
494 uint64_t zone_size, offset;
495 struct zoned_block_device_info *zbd_info = NULL;
496 int i, j, ret = -ENOMEM;
498 zones = calloc(ZBD_REPORT_MAX_ZONES, sizeof(struct zbd_zone));
502 nrz = zbd_report_zones(td, f, 0, zones, ZBD_REPORT_MAX_ZONES);
505 log_info("fio: report zones (offset 0) failed for %s (%d).\n",
510 zone_size = zones[0].len;
511 nr_zones = (f->real_file_size + zone_size - 1) / zone_size;
513 if (td->o.zone_size == 0) {
514 td->o.zone_size = zone_size;
515 } else if (td->o.zone_size != zone_size) {
516 log_err("fio: %s job parameter zonesize %llu does not match disk zone size %llu.\n",
517 f->file_name, (unsigned long long) td->o.zone_size,
518 (unsigned long long) zone_size);
523 dprint(FD_ZBD, "Device %s has %d zones of size %llu KB\n", f->file_name,
524 nr_zones, (unsigned long long) zone_size / 1024);
526 zbd_info = scalloc(1, sizeof(*zbd_info) +
527 (nr_zones + 1) * sizeof(zbd_info->zone_info[0]));
530 mutex_init_pshared(&zbd_info->mutex);
531 zbd_info->refcount = 1;
532 p = &zbd_info->zone_info[0];
533 for (offset = 0, j = 0; j < nr_zones;) {
535 for (i = 0; i < nrz; i++, j++, z++, p++) {
536 mutex_init_pshared_with_type(&p->mutex,
537 PTHREAD_MUTEX_RECURSIVE);
539 p->capacity = z->capacity;
541 case ZBD_ZONE_COND_NOT_WP:
542 case ZBD_ZONE_COND_FULL:
543 p->wp = p->start + p->capacity;
546 assert(z->start <= z->wp);
547 assert(z->wp <= z->start + zone_size);
553 case ZBD_ZONE_TYPE_SWR:
562 if (j > 0 && p->start != p[-1].start + zone_size) {
563 log_info("%s: invalid zone data\n",
570 offset = z->start + z->len;
573 nrz = zbd_report_zones(td, f, offset, zones,
574 min((uint32_t)(nr_zones - j),
575 ZBD_REPORT_MAX_ZONES));
578 log_info("fio: report zones (offset %llu) failed for %s (%d).\n",
579 (unsigned long long)offset,
586 zbd_info->zone_info[nr_zones].start = offset;
588 f->zbd_info = zbd_info;
589 f->zbd_info->zone_size = zone_size;
590 f->zbd_info->zone_size_log2 = is_power_of_2(zone_size) ?
591 ilog2(zone_size) : 0;
592 f->zbd_info->nr_zones = nr_zones;
602 static int zbd_set_max_open_zones(struct thread_data *td, struct fio_file *f)
604 struct zoned_block_device_info *zbd = f->zbd_info;
605 unsigned int max_open_zones;
608 if (zbd->model != ZBD_HOST_MANAGED || td->o.ignore_zone_limits) {
609 /* Only host-managed devices have a max open limit */
610 zbd->max_open_zones = td->o.max_open_zones;
614 /* If host-managed, get the max open limit */
615 ret = zbd_get_max_open_zones(td, f, &max_open_zones);
619 if (!max_open_zones) {
620 /* No device limit */
621 zbd->max_open_zones = td->o.max_open_zones;
622 } else if (!td->o.max_open_zones) {
623 /* No user limit. Set limit to device limit */
624 zbd->max_open_zones = max_open_zones;
625 } else if (td->o.max_open_zones <= max_open_zones) {
626 /* Both user limit and dev limit. User limit not too large */
627 zbd->max_open_zones = td->o.max_open_zones;
629 /* Both user limit and dev limit. User limit too large */
630 td_verror(td, EINVAL,
631 "Specified --max_open_zones is too large");
632 log_err("Specified --max_open_zones (%d) is larger than max (%u)\n",
633 td->o.max_open_zones, max_open_zones);
638 /* Ensure that the limit is not larger than FIO's internal limit */
639 zbd->max_open_zones = min_not_zero(zbd->max_open_zones,
640 (uint32_t) ZBD_MAX_OPEN_ZONES);
641 dprint(FD_ZBD, "%s: using max open zones limit: %"PRIu32"\n",
642 f->file_name, zbd->max_open_zones);
648 * Allocate zone information and store it into f->zbd_info if zonemode=zbd.
650 * Returns 0 upon success and a negative error code upon failure.
652 static int zbd_create_zone_info(struct thread_data *td, struct fio_file *f)
654 enum zbd_zoned_model zbd_model;
657 assert(td->o.zone_mode == ZONE_MODE_ZBD);
659 ret = zbd_get_zoned_model(td, f, &zbd_model);
667 case ZBD_HOST_MANAGED:
668 ret = parse_zone_info(td, f);
673 ret = init_zone_info(td, f);
678 td_verror(td, EINVAL, "Unsupported zoned model");
679 log_err("Unsupported zoned model\n");
683 f->zbd_info->model = zbd_model;
685 ret = zbd_set_max_open_zones(td, f);
687 zbd_free_zone_info(f);
694 void zbd_free_zone_info(struct fio_file *f)
700 pthread_mutex_lock(&f->zbd_info->mutex);
701 refcount = --f->zbd_info->refcount;
702 pthread_mutex_unlock(&f->zbd_info->mutex);
704 assert((int32_t)refcount >= 0);
711 * Initialize f->zbd_info.
713 * Returns 0 upon success and a negative error code upon failure.
715 * Note: this function can only work correctly if it is called before the first
718 static int zbd_init_zone_info(struct thread_data *td, struct fio_file *file)
720 struct thread_data *td2;
724 for_each_td(td2, i) {
725 for_each_file(td2, f2, j) {
726 if (td2 == td && f2 == file)
729 strcmp(f2->file_name, file->file_name) != 0)
731 file->zbd_info = f2->zbd_info;
732 file->zbd_info->refcount++;
737 ret = zbd_create_zone_info(td, file);
739 td_verror(td, -ret, "zbd_create_zone_info() failed");
743 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
745 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
746 struct fio_zone_info *z);
748 int zbd_init_files(struct thread_data *td)
753 for_each_file(td, f, i) {
754 if (zbd_init_zone_info(td, f))
760 void zbd_recalc_options_with_zone_granularity(struct thread_data *td)
765 for_each_file(td, f, i) {
766 struct zoned_block_device_info *zbd = f->zbd_info;
767 // zonemode=strided doesn't get per-file zone size.
768 uint64_t zone_size = zbd ? zbd->zone_size : td->o.zone_size;
773 if (td->o.size_nz > 0) {
774 td->o.size = td->o.size_nz * zone_size;
776 if (td->o.io_size_nz > 0) {
777 td->o.io_size = td->o.io_size_nz * zone_size;
779 if (td->o.start_offset_nz > 0) {
780 td->o.start_offset = td->o.start_offset_nz * zone_size;
782 if (td->o.offset_increment_nz > 0) {
783 td->o.offset_increment = td->o.offset_increment_nz * zone_size;
785 if (td->o.zone_skip_nz > 0) {
786 td->o.zone_skip = td->o.zone_skip_nz * zone_size;
791 int zbd_setup_files(struct thread_data *td)
796 if (!zbd_using_direct_io()) {
797 log_err("Using direct I/O is mandatory for writing to ZBD drives\n\n");
801 if (!zbd_verify_sizes())
804 if (!zbd_verify_bs())
807 for_each_file(td, f, i) {
808 struct zoned_block_device_info *zbd = f->zbd_info;
809 struct fio_zone_info *z;
815 f->min_zone = zbd_zone_idx(f, f->file_offset);
816 f->max_zone = zbd_zone_idx(f, f->file_offset + f->io_size);
819 * When all zones in the I/O range are conventional, io_size
820 * can be smaller than zone size, making min_zone the same
821 * as max_zone. This is why the assert below needs to be made
824 if (zbd_is_seq_job(f))
825 assert(f->min_zone < f->max_zone);
827 if (td->o.max_open_zones > 0 &&
828 zbd->max_open_zones != td->o.max_open_zones) {
829 log_err("Different 'max_open_zones' values\n");
832 if (zbd->max_open_zones > ZBD_MAX_OPEN_ZONES) {
833 log_err("'max_open_zones' value is limited by %u\n", ZBD_MAX_OPEN_ZONES);
837 for (zi = f->min_zone; zi < f->max_zone; zi++) {
838 z = &zbd->zone_info[zi];
839 if (z->cond != ZBD_ZONE_COND_IMP_OPEN &&
840 z->cond != ZBD_ZONE_COND_EXP_OPEN)
842 if (zbd_open_zone(td, f, zi))
845 * If the number of open zones exceeds specified limits,
846 * reset all extra open zones.
848 if (zbd_reset_zone(td, f, z) < 0) {
849 log_err("Failed to reest zone %d\n", zi);
858 static inline unsigned int zbd_zone_nr(const struct fio_file *f,
859 struct fio_zone_info *zone)
861 return zone - f->zbd_info->zone_info;
865 * zbd_reset_zone - reset the write pointer of a single zone
866 * @td: FIO thread data.
867 * @f: FIO file associated with the disk for which to reset a write pointer.
870 * Returns 0 upon success and a negative error code upon failure.
872 * The caller must hold z->mutex.
874 static int zbd_reset_zone(struct thread_data *td, struct fio_file *f,
875 struct fio_zone_info *z)
877 uint64_t offset = z->start;
878 uint64_t length = (z+1)->start - offset;
879 uint64_t data_in_zone = z->wp - z->start;
885 assert(is_valid_offset(f, offset + length - 1));
887 dprint(FD_ZBD, "%s: resetting wp of zone %u.\n", f->file_name,
889 switch (f->zbd_info->model) {
891 case ZBD_HOST_MANAGED:
892 ret = zbd_reset_wp(td, f, offset, length);
900 pthread_mutex_lock(&f->zbd_info->mutex);
901 f->zbd_info->sectors_with_data -= data_in_zone;
902 f->zbd_info->wp_sectors_with_data -= data_in_zone;
903 pthread_mutex_unlock(&f->zbd_info->mutex);
907 td->ts.nr_zone_resets++;
912 /* The caller must hold f->zbd_info->mutex */
913 static void zbd_close_zone(struct thread_data *td, const struct fio_file *f,
914 unsigned int zone_idx)
916 uint32_t open_zone_idx = 0;
918 for (; open_zone_idx < f->zbd_info->num_open_zones; open_zone_idx++) {
919 if (f->zbd_info->open_zones[open_zone_idx] == zone_idx)
922 if (open_zone_idx == f->zbd_info->num_open_zones)
925 dprint(FD_ZBD, "%s: closing zone %d\n", f->file_name, zone_idx);
926 memmove(f->zbd_info->open_zones + open_zone_idx,
927 f->zbd_info->open_zones + open_zone_idx + 1,
928 (ZBD_MAX_OPEN_ZONES - (open_zone_idx + 1)) *
929 sizeof(f->zbd_info->open_zones[0]));
930 f->zbd_info->num_open_zones--;
931 td->num_open_zones--;
932 get_zone(f, zone_idx)->open = 0;
936 * Reset a range of zones. Returns 0 upon success and 1 upon failure.
937 * @td: fio thread data.
938 * @f: fio file for which to reset zones
939 * @zb: first zone to reset.
940 * @ze: first zone not to reset.
942 static int zbd_reset_zones(struct thread_data *td, struct fio_file *f,
943 struct fio_zone_info *const zb,
944 struct fio_zone_info *const ze)
946 struct fio_zone_info *z;
947 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
952 dprint(FD_ZBD, "%s: examining zones %u .. %u\n", f->file_name,
953 zbd_zone_nr(f, zb), zbd_zone_nr(f, ze));
954 for (z = zb; z < ze; z++) {
955 uint32_t nz = zbd_zone_nr(f, z);
960 pthread_mutex_lock(&f->zbd_info->mutex);
961 zbd_close_zone(td, f, nz);
962 pthread_mutex_unlock(&f->zbd_info->mutex);
963 if (z->wp != z->start) {
964 dprint(FD_ZBD, "%s: resetting zone %u\n",
965 f->file_name, zbd_zone_nr(f, z));
966 if (zbd_reset_zone(td, f, z) < 0)
976 * Reset zbd_info.write_cnt, the counter that counts down towards the next
979 static void _zbd_reset_write_cnt(const struct thread_data *td,
980 const struct fio_file *f)
982 assert(0 <= td->o.zrf.u.f && td->o.zrf.u.f <= 1);
984 f->zbd_info->write_cnt = td->o.zrf.u.f ?
985 min(1.0 / td->o.zrf.u.f, 0.0 + UINT_MAX) : UINT_MAX;
988 static void zbd_reset_write_cnt(const struct thread_data *td,
989 const struct fio_file *f)
991 pthread_mutex_lock(&f->zbd_info->mutex);
992 _zbd_reset_write_cnt(td, f);
993 pthread_mutex_unlock(&f->zbd_info->mutex);
996 static bool zbd_dec_and_reset_write_cnt(const struct thread_data *td,
997 const struct fio_file *f)
999 uint32_t write_cnt = 0;
1001 pthread_mutex_lock(&f->zbd_info->mutex);
1002 assert(f->zbd_info->write_cnt);
1003 if (f->zbd_info->write_cnt)
1004 write_cnt = --f->zbd_info->write_cnt;
1006 _zbd_reset_write_cnt(td, f);
1007 pthread_mutex_unlock(&f->zbd_info->mutex);
1009 return write_cnt == 0;
1017 /* Calculate the number of sectors with data (swd) and perform action 'a' */
1018 static uint64_t zbd_process_swd(struct thread_data *td,
1019 const struct fio_file *f, enum swd_action a)
1021 struct fio_zone_info *zb, *ze, *z;
1023 uint64_t wp_swd = 0;
1025 zb = get_zone(f, f->min_zone);
1026 ze = get_zone(f, f->max_zone);
1027 for (z = zb; z < ze; z++) {
1029 zone_lock(td, f, z);
1030 wp_swd += z->wp - z->start;
1032 swd += z->wp - z->start;
1034 pthread_mutex_lock(&f->zbd_info->mutex);
1037 assert(f->zbd_info->sectors_with_data == swd);
1038 assert(f->zbd_info->wp_sectors_with_data == wp_swd);
1041 f->zbd_info->sectors_with_data = swd;
1042 f->zbd_info->wp_sectors_with_data = wp_swd;
1045 pthread_mutex_unlock(&f->zbd_info->mutex);
1046 for (z = zb; z < ze; z++)
1054 * The swd check is useful for debugging but takes too much time to leave
1055 * it enabled all the time. Hence it is disabled by default.
1057 static const bool enable_check_swd = false;
1059 /* Check whether the values of zbd_info.*sectors_with_data are correct. */
1060 static void zbd_check_swd(struct thread_data *td, const struct fio_file *f)
1062 if (!enable_check_swd)
1065 zbd_process_swd(td, f, CHECK_SWD);
1068 void zbd_file_reset(struct thread_data *td, struct fio_file *f)
1070 struct fio_zone_info *zb, *ze;
1073 if (!f->zbd_info || !td_write(td))
1076 zb = get_zone(f, f->min_zone);
1077 ze = get_zone(f, f->max_zone);
1078 swd = zbd_process_swd(td, f, SET_SWD);
1079 dprint(FD_ZBD, "%s(%s): swd = %" PRIu64 "\n", __func__, f->file_name,
1082 * If data verification is enabled reset the affected zones before
1083 * writing any data to avoid that a zone reset has to be issued while
1084 * writing data, which causes data loss.
1086 if (td->o.verify != VERIFY_NONE && td->runstate != TD_VERIFYING)
1087 zbd_reset_zones(td, f, zb, ze);
1088 zbd_reset_write_cnt(td, f);
1091 /* The caller must hold f->zbd_info->mutex. */
1092 static bool is_zone_open(const struct thread_data *td, const struct fio_file *f,
1093 unsigned int zone_idx)
1095 struct zoned_block_device_info *zbdi = f->zbd_info;
1098 assert(td->o.job_max_open_zones == 0 || td->num_open_zones <= td->o.job_max_open_zones);
1099 assert(td->o.job_max_open_zones <= zbdi->max_open_zones);
1100 assert(zbdi->num_open_zones <= zbdi->max_open_zones);
1102 for (i = 0; i < zbdi->num_open_zones; i++)
1103 if (zbdi->open_zones[i] == zone_idx)
1110 * Open a ZBD zone if it was not yet open. Returns true if either the zone was
1111 * already open or if opening a new zone is allowed. Returns false if the zone
1112 * was not yet open and opening a new zone would cause the zone limit to be
1115 static bool zbd_open_zone(struct thread_data *td, const struct fio_file *f,
1118 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
1119 struct fio_zone_info *z = get_zone(f, zone_idx);
1122 if (z->cond == ZBD_ZONE_COND_OFFLINE)
1126 * Skip full zones with data verification enabled because resetting a
1127 * zone causes data loss and hence causes verification to fail.
1129 if (td->o.verify != VERIFY_NONE && zbd_zone_full(f, z, min_bs))
1132 pthread_mutex_lock(&f->zbd_info->mutex);
1133 if (is_zone_open(td, f, zone_idx)) {
1135 * If the zone is already open and going to be full by writes
1136 * in-flight, handle it as a full zone instead of an open zone.
1138 if (z->wp >= zbd_zone_capacity_end(z))
1143 /* Zero means no limit */
1144 if (td->o.job_max_open_zones > 0 &&
1145 td->num_open_zones >= td->o.job_max_open_zones)
1147 if (f->zbd_info->num_open_zones >= f->zbd_info->max_open_zones)
1149 dprint(FD_ZBD, "%s: opening zone %d\n", f->file_name, zone_idx);
1150 f->zbd_info->open_zones[f->zbd_info->num_open_zones++] = zone_idx;
1151 td->num_open_zones++;
1156 pthread_mutex_unlock(&f->zbd_info->mutex);
1160 /* Anything goes as long as it is not a constant. */
1161 static uint32_t pick_random_zone_idx(const struct fio_file *f,
1162 const struct io_u *io_u)
1164 return io_u->offset * f->zbd_info->num_open_zones / f->real_file_size;
1168 * Modify the offset of an I/O unit that does not refer to an open zone such
1169 * that it refers to an open zone. Close an open zone and open a new zone if
1170 * necessary. The open zone is searched across sequential zones.
1171 * This algorithm can only work correctly if all write pointers are
1172 * a multiple of the fio block size. The caller must neither hold z->mutex
1173 * nor f->zbd_info->mutex. Returns with z->mutex held upon success.
1175 static struct fio_zone_info *zbd_convert_to_open_zone(struct thread_data *td,
1178 const uint32_t min_bs = td->o.min_bs[io_u->ddir];
1179 struct fio_file *f = io_u->file;
1180 struct fio_zone_info *z;
1181 unsigned int open_zone_idx = -1;
1182 uint32_t zone_idx, new_zone_idx;
1184 bool wait_zone_close;
1186 assert(is_valid_offset(f, io_u->offset));
1188 if (td->o.max_open_zones || td->o.job_max_open_zones) {
1190 * This statement accesses f->zbd_info->open_zones[] on purpose
1193 zone_idx = f->zbd_info->open_zones[pick_random_zone_idx(f, io_u)];
1195 zone_idx = zbd_zone_idx(f, io_u->offset);
1197 if (zone_idx < f->min_zone)
1198 zone_idx = f->min_zone;
1199 else if (zone_idx >= f->max_zone)
1200 zone_idx = f->max_zone - 1;
1201 dprint(FD_ZBD, "%s(%s): starting from zone %d (offset %lld, buflen %lld)\n",
1202 __func__, f->file_name, zone_idx, io_u->offset, io_u->buflen);
1205 * Since z->mutex is the outer lock and f->zbd_info->mutex the inner
1206 * lock it can happen that the state of the zone with index zone_idx
1207 * has changed after 'z' has been assigned and before f->zbd_info->mutex
1208 * has been obtained. Hence the loop.
1213 z = get_zone(f, zone_idx);
1215 zone_lock(td, f, z);
1216 pthread_mutex_lock(&f->zbd_info->mutex);
1218 if (z->cond != ZBD_ZONE_COND_OFFLINE &&
1219 td->o.max_open_zones == 0 && td->o.job_max_open_zones == 0)
1221 if (f->zbd_info->num_open_zones == 0) {
1222 dprint(FD_ZBD, "%s(%s): no zones are open\n",
1223 __func__, f->file_name);
1224 goto open_other_zone;
1229 * List of opened zones is per-device, shared across all threads.
1230 * Start with quasi-random candidate zone.
1231 * Ignore zones which don't belong to thread's offset/size area.
1233 open_zone_idx = pick_random_zone_idx(f, io_u);
1234 assert(!open_zone_idx ||
1235 open_zone_idx < f->zbd_info->num_open_zones);
1236 tmp_idx = open_zone_idx;
1237 for (i = 0; i < f->zbd_info->num_open_zones; i++) {
1240 if (tmp_idx >= f->zbd_info->num_open_zones)
1242 tmpz = f->zbd_info->open_zones[tmp_idx];
1243 if (f->min_zone <= tmpz && tmpz < f->max_zone) {
1244 open_zone_idx = tmp_idx;
1245 goto found_candidate_zone;
1251 dprint(FD_ZBD, "%s(%s): no candidate zone\n",
1252 __func__, f->file_name);
1253 pthread_mutex_unlock(&f->zbd_info->mutex);
1258 found_candidate_zone:
1259 new_zone_idx = f->zbd_info->open_zones[open_zone_idx];
1260 if (new_zone_idx == zone_idx)
1262 zone_idx = new_zone_idx;
1263 pthread_mutex_unlock(&f->zbd_info->mutex);
1268 /* Both z->mutex and f->zbd_info->mutex are held. */
1271 if (z->wp + min_bs <= zbd_zone_capacity_end(z)) {
1272 pthread_mutex_unlock(&f->zbd_info->mutex);
1277 /* Check if number of open zones reaches one of limits. */
1279 f->zbd_info->num_open_zones == f->max_zone - f->min_zone ||
1280 (td->o.max_open_zones &&
1281 f->zbd_info->num_open_zones == td->o.max_open_zones) ||
1282 (td->o.job_max_open_zones &&
1283 td->num_open_zones == td->o.job_max_open_zones);
1285 pthread_mutex_unlock(&f->zbd_info->mutex);
1287 /* Only z->mutex is held. */
1290 * When number of open zones reaches to one of limits, wait for
1291 * zone close before opening a new zone.
1293 if (wait_zone_close) {
1294 dprint(FD_ZBD, "%s(%s): quiesce to allow open zones to close\n",
1295 __func__, f->file_name);
1299 /* Zone 'z' is full, so try to open a new zone. */
1300 for (i = f->io_size / f->zbd_info->zone_size; i > 0; i--) {
1305 if (!is_valid_offset(f, z->start)) {
1307 zone_idx = f->min_zone;
1308 z = get_zone(f, zone_idx);
1310 assert(is_valid_offset(f, z->start));
1313 zone_lock(td, f, z);
1316 if (zbd_open_zone(td, f, zone_idx))
1320 /* Only z->mutex is held. */
1322 /* Check whether the write fits in any of the already opened zones. */
1323 pthread_mutex_lock(&f->zbd_info->mutex);
1324 for (i = 0; i < f->zbd_info->num_open_zones; i++) {
1325 zone_idx = f->zbd_info->open_zones[i];
1326 if (zone_idx < f->min_zone || zone_idx >= f->max_zone)
1328 pthread_mutex_unlock(&f->zbd_info->mutex);
1331 z = get_zone(f, zone_idx);
1333 zone_lock(td, f, z);
1334 if (z->wp + min_bs <= zbd_zone_capacity_end(z))
1336 pthread_mutex_lock(&f->zbd_info->mutex);
1338 pthread_mutex_unlock(&f->zbd_info->mutex);
1340 dprint(FD_ZBD, "%s(%s): did not open another zone\n", __func__,
1345 dprint(FD_ZBD, "%s(%s): returning zone %d\n", __func__, f->file_name,
1347 io_u->offset = z->start;
1349 assert(z->cond != ZBD_ZONE_COND_OFFLINE);
1353 /* The caller must hold z->mutex. */
1354 static struct fio_zone_info *zbd_replay_write_order(struct thread_data *td,
1356 struct fio_zone_info *z)
1358 const struct fio_file *f = io_u->file;
1359 const uint32_t min_bs = td->o.min_bs[DDIR_WRITE];
1361 if (!zbd_open_zone(td, f, zbd_zone_nr(f, z))) {
1363 z = zbd_convert_to_open_zone(td, io_u);
1367 if (z->verify_block * min_bs >= z->capacity) {
1368 log_err("%s: %d * %d >= %llu\n", f->file_name, z->verify_block,
1369 min_bs, (unsigned long long)z->capacity);
1371 * If the assertion below fails during a test run, adding
1372 * "--experimental_verify=1" to the command line may help.
1376 io_u->offset = z->start + z->verify_block * min_bs;
1377 if (io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1378 log_err("%s: %llu + %llu >= %llu\n", f->file_name, io_u->offset,
1379 io_u->buflen, (unsigned long long) zbd_zone_capacity_end(z));
1382 z->verify_block += io_u->buflen / min_bs;
1388 * Find another zone for which @io_u fits in the readable data in the zone.
1389 * Search in zones @zb + 1 .. @zl. For random workload, also search in zones
1392 * Either returns NULL or returns a zone pointer. When the zone has write
1393 * pointer, hold the mutex for the zone.
1395 static struct fio_zone_info *
1396 zbd_find_zone(struct thread_data *td, struct io_u *io_u,
1397 struct fio_zone_info *zb, struct fio_zone_info *zl)
1399 const uint32_t min_bs = td->o.min_bs[io_u->ddir];
1400 struct fio_file *f = io_u->file;
1401 struct fio_zone_info *z1, *z2;
1402 const struct fio_zone_info *const zf = get_zone(f, f->min_zone);
1405 * Skip to the next non-empty zone in case of sequential I/O and to
1406 * the nearest non-empty zone in case of random I/O.
1408 for (z1 = zb + 1, z2 = zb - 1; z1 < zl || z2 >= zf; z1++, z2--) {
1409 if (z1 < zl && z1->cond != ZBD_ZONE_COND_OFFLINE) {
1411 zone_lock(td, f, z1);
1412 if (z1->start + min_bs <= z1->wp)
1416 } else if (!td_random(td)) {
1419 if (td_random(td) && z2 >= zf &&
1420 z2->cond != ZBD_ZONE_COND_OFFLINE) {
1422 zone_lock(td, f, z2);
1423 if (z2->start + min_bs <= z2->wp)
1429 dprint(FD_ZBD, "%s: adjusting random read offset failed\n",
1435 * zbd_end_zone_io - update zone status at command completion
1437 * @z: zone info pointer
1439 * If the write command made the zone full, close it.
1441 * The caller must hold z->mutex.
1443 static void zbd_end_zone_io(struct thread_data *td, const struct io_u *io_u,
1444 struct fio_zone_info *z)
1446 const struct fio_file *f = io_u->file;
1448 if (io_u->ddir == DDIR_WRITE &&
1449 io_u->offset + io_u->buflen >= zbd_zone_capacity_end(z)) {
1450 pthread_mutex_lock(&f->zbd_info->mutex);
1451 zbd_close_zone(td, f, zbd_zone_nr(f, z));
1452 pthread_mutex_unlock(&f->zbd_info->mutex);
1457 * zbd_queue_io - update the write pointer of a sequential zone
1459 * @success: Whether or not the I/O unit has been queued successfully
1460 * @q: queueing status (busy, completed or queued).
1462 * For write and trim operations, update the write pointer of the I/O unit
1465 static void zbd_queue_io(struct thread_data *td, struct io_u *io_u, int q,
1468 const struct fio_file *f = io_u->file;
1469 struct zoned_block_device_info *zbd_info = f->zbd_info;
1470 struct fio_zone_info *z;
1477 zone_idx = zbd_zone_idx(f, io_u->offset);
1478 assert(zone_idx < zbd_info->nr_zones);
1479 z = get_zone(f, zone_idx);
1487 "%s: queued I/O (%lld, %llu) for zone %u\n",
1488 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1490 switch (io_u->ddir) {
1492 zone_end = min((uint64_t)(io_u->offset + io_u->buflen),
1493 zbd_zone_capacity_end(z));
1494 pthread_mutex_lock(&zbd_info->mutex);
1496 * z->wp > zone_end means that one or more I/O errors
1499 if (z->wp <= zone_end) {
1500 zbd_info->sectors_with_data += zone_end - z->wp;
1501 zbd_info->wp_sectors_with_data += zone_end - z->wp;
1503 pthread_mutex_unlock(&zbd_info->mutex);
1507 assert(z->wp == z->start);
1513 if (q == FIO_Q_COMPLETED && !io_u->error)
1514 zbd_end_zone_io(td, io_u, z);
1517 if (!success || q != FIO_Q_QUEUED) {
1518 /* BUSY or COMPLETED: unlock the zone */
1520 io_u->zbd_put_io = NULL;
1525 * zbd_put_io - Unlock an I/O unit target zone lock
1528 static void zbd_put_io(struct thread_data *td, const struct io_u *io_u)
1530 const struct fio_file *f = io_u->file;
1531 struct zoned_block_device_info *zbd_info = f->zbd_info;
1532 struct fio_zone_info *z;
1538 zone_idx = zbd_zone_idx(f, io_u->offset);
1539 assert(zone_idx < zbd_info->nr_zones);
1540 z = get_zone(f, zone_idx);
1545 "%s: terminate I/O (%lld, %llu) for zone %u\n",
1546 f->file_name, io_u->offset, io_u->buflen, zone_idx);
1548 zbd_end_zone_io(td, io_u, z);
1551 zbd_check_swd(td, f);
1555 * Windows and MacOS do not define this.
1558 #define EREMOTEIO 121 /* POSIX value */
1561 bool zbd_unaligned_write(int error_code)
1563 switch (error_code) {
1572 * setup_zbd_zone_mode - handle zoneskip as necessary for ZBD drives
1573 * @td: FIO thread data.
1574 * @io_u: FIO I/O unit.
1576 * For sequential workloads, change the file offset to skip zoneskip bytes when
1577 * no more IO can be performed in the current zone.
1578 * - For read workloads, zoneskip is applied when the io has reached the end of
1579 * the zone or the zone write position (when td->o.read_beyond_wp is false).
1580 * - For write workloads, zoneskip is applied when the zone is full.
1581 * This applies only to read and write operations.
1583 void setup_zbd_zone_mode(struct thread_data *td, struct io_u *io_u)
1585 struct fio_file *f = io_u->file;
1586 enum fio_ddir ddir = io_u->ddir;
1587 struct fio_zone_info *z;
1590 assert(td->o.zone_mode == ZONE_MODE_ZBD);
1591 assert(td->o.zone_size);
1593 zone_idx = zbd_zone_idx(f, f->last_pos[ddir]);
1594 z = get_zone(f, zone_idx);
1597 * When the zone capacity is smaller than the zone size and the I/O is
1598 * sequential write, skip to zone end if the latest position is at the
1599 * zone capacity limit.
1601 if (z->capacity < f->zbd_info->zone_size && !td_random(td) &&
1602 ddir == DDIR_WRITE &&
1603 f->last_pos[ddir] >= zbd_zone_capacity_end(z)) {
1605 "%s: Jump from zone capacity limit to zone end:"
1606 " (%llu -> %llu) for zone %u (%llu)\n",
1607 f->file_name, (unsigned long long) f->last_pos[ddir],
1608 (unsigned long long) zbd_zone_end(z), zone_idx,
1609 (unsigned long long) z->capacity);
1610 td->io_skip_bytes += zbd_zone_end(z) - f->last_pos[ddir];
1611 f->last_pos[ddir] = zbd_zone_end(z);
1615 * zone_skip is valid only for sequential workloads.
1617 if (td_random(td) || !td->o.zone_skip)
1621 * It is time to switch to a new zone if:
1622 * - zone_bytes == zone_size bytes have already been accessed
1623 * - The last position reached the end of the current zone.
1624 * - For reads with td->o.read_beyond_wp == false, the last position
1625 * reached the zone write pointer.
1627 if (td->zone_bytes >= td->o.zone_size ||
1628 f->last_pos[ddir] >= zbd_zone_end(z) ||
1629 (ddir == DDIR_READ &&
1630 (!td->o.read_beyond_wp) && f->last_pos[ddir] >= z->wp)) {
1635 f->file_offset += td->o.zone_size + td->o.zone_skip;
1638 * Wrap from the beginning, if we exceed the file size
1640 if (f->file_offset >= f->real_file_size)
1641 f->file_offset = get_start_offset(td, f);
1643 f->last_pos[ddir] = f->file_offset;
1644 td->io_skip_bytes += td->o.zone_skip;
1649 * zbd_adjust_ddir - Adjust an I/O direction for zonemode=zbd.
1651 * @td: FIO thread data.
1652 * @io_u: FIO I/O unit.
1653 * @ddir: I/O direction before adjustment.
1655 * Return adjusted I/O direction.
1657 enum fio_ddir zbd_adjust_ddir(struct thread_data *td, struct io_u *io_u,
1661 * In case read direction is chosen for the first random I/O, fio with
1662 * zonemode=zbd stops because no data can be read from zoned block
1663 * devices with all empty zones. Overwrite the first I/O direction as
1664 * write to make sure data to read exists.
1666 if (ddir != DDIR_READ || !td_rw(td))
1669 if (io_u->file->zbd_info->sectors_with_data ||
1670 td->o.read_beyond_wp)
1677 * zbd_adjust_block - adjust the offset and length as necessary for ZBD drives
1678 * @td: FIO thread data.
1679 * @io_u: FIO I/O unit.
1681 * Locking strategy: returns with z->mutex locked if and only if z refers
1682 * to a sequential zone and if io_u_accept is returned. z is the zone that
1683 * corresponds to io_u->offset at the end of this function.
1685 enum io_u_action zbd_adjust_block(struct thread_data *td, struct io_u *io_u)
1687 struct fio_file *f = io_u->file;
1688 uint32_t zone_idx_b;
1689 struct fio_zone_info *zb, *zl, *orig_zb;
1690 uint32_t orig_len = io_u->buflen;
1691 uint32_t min_bs = td->o.min_bs[io_u->ddir];
1699 assert(is_valid_offset(f, io_u->offset));
1700 assert(io_u->buflen);
1701 zone_idx_b = zbd_zone_idx(f, io_u->offset);
1702 zb = get_zone(f, zone_idx_b);
1706 /* Accept non-write I/Os for conventional zones. */
1707 if (io_u->ddir != DDIR_WRITE)
1710 * Make sure that writes to conventional zones
1711 * don't cross over to any sequential zones.
1713 if (!(zb + 1)->has_wp ||
1714 io_u->offset + io_u->buflen <= (zb + 1)->start)
1717 if (io_u->offset + min_bs > (zb + 1)->start) {
1719 "%s: off=%llu + min_bs=%u > next zone %llu\n",
1720 f->file_name, io_u->offset,
1721 min_bs, (unsigned long long) (zb + 1)->start);
1722 io_u->offset = zb->start + (zb + 1)->start - io_u->offset;
1723 new_len = min(io_u->buflen, (zb + 1)->start - io_u->offset);
1725 new_len = (zb + 1)->start - io_u->offset;
1727 io_u->buflen = new_len / min_bs * min_bs;
1732 * Accept the I/O offset for reads if reading beyond the write pointer
1735 if (zb->cond != ZBD_ZONE_COND_OFFLINE &&
1736 io_u->ddir == DDIR_READ && td->o.read_beyond_wp)
1739 zbd_check_swd(td, f);
1741 zone_lock(td, f, zb);
1743 switch (io_u->ddir) {
1745 if (td->runstate == TD_VERIFYING && td_write(td)) {
1746 zb = zbd_replay_write_order(td, io_u, zb);
1750 * Check that there is enough written data in the zone to do an
1751 * I/O of at least min_bs B. If there isn't, find a new zone for
1754 range = zb->cond != ZBD_ZONE_COND_OFFLINE ?
1755 zb->wp - zb->start : 0;
1756 if (range < min_bs ||
1757 ((!td_random(td)) && (io_u->offset + min_bs > zb->wp))) {
1759 zl = get_zone(f, f->max_zone);
1760 zb = zbd_find_zone(td, io_u, zb, zl);
1763 "%s: zbd_find_zone(%lld, %llu) failed\n",
1764 f->file_name, io_u->offset,
1769 * zbd_find_zone() returned a zone with a range of at
1772 range = zb->wp - zb->start;
1773 assert(range >= min_bs);
1776 io_u->offset = zb->start;
1779 * Make sure the I/O is within the zone valid data range while
1780 * maximizing the I/O size and preserving randomness.
1782 if (range <= io_u->buflen)
1783 io_u->offset = zb->start;
1784 else if (td_random(td))
1785 io_u->offset = zb->start +
1786 ((io_u->offset - orig_zb->start) %
1787 (range - io_u->buflen)) / min_bs * min_bs;
1789 * When zbd_find_zone() returns a conventional zone,
1790 * we can simply accept the new i/o offset here.
1795 * Make sure the I/O does not cross over the zone wp position.
1797 new_len = min((unsigned long long)io_u->buflen,
1798 (unsigned long long)(zb->wp - io_u->offset));
1799 new_len = new_len / min_bs * min_bs;
1800 if (new_len < io_u->buflen) {
1801 io_u->buflen = new_len;
1802 dprint(FD_IO, "Changed length from %u into %llu\n",
1803 orig_len, io_u->buflen);
1805 assert(zb->start <= io_u->offset);
1806 assert(io_u->offset + io_u->buflen <= zb->wp);
1809 if (io_u->buflen > f->zbd_info->zone_size) {
1810 td_verror(td, EINVAL, "I/O buflen exceeds zone size");
1812 "%s: I/O buflen %llu exceeds zone size %llu\n",
1813 f->file_name, io_u->buflen,
1814 (unsigned long long) f->zbd_info->zone_size);
1817 if (!zbd_open_zone(td, f, zone_idx_b)) {
1819 zb = zbd_convert_to_open_zone(td, io_u);
1821 dprint(FD_IO, "%s: can't convert to open zone",
1825 zone_idx_b = zbd_zone_nr(f, zb);
1827 /* Check whether the zone reset threshold has been exceeded */
1828 if (td->o.zrf.u.f) {
1829 if (f->zbd_info->wp_sectors_with_data >=
1830 f->io_size * td->o.zrt.u.f &&
1831 zbd_dec_and_reset_write_cnt(td, f)) {
1835 /* Reset the zone pointer if necessary */
1836 if (zb->reset_zone || zbd_zone_full(f, zb, min_bs)) {
1837 assert(td->o.verify == VERIFY_NONE);
1839 * Since previous write requests may have been submitted
1840 * asynchronously and since we will submit the zone
1841 * reset synchronously, wait until previously submitted
1842 * write requests have completed before issuing a
1847 if (zbd_reset_zone(td, f, zb) < 0)
1850 if (zb->capacity < min_bs) {
1851 td_verror(td, EINVAL, "ZCAP is less min_bs");
1852 log_err("zone capacity %llu smaller than minimum block size %d\n",
1853 (unsigned long long)zb->capacity,
1858 /* Make writes occur at the write pointer */
1859 assert(!zbd_zone_full(f, zb, min_bs));
1860 io_u->offset = zb->wp;
1861 if (!is_valid_offset(f, io_u->offset)) {
1862 td_verror(td, EINVAL, "invalid WP value");
1863 dprint(FD_ZBD, "%s: dropped request with offset %llu\n",
1864 f->file_name, io_u->offset);
1868 * Make sure that the buflen is a multiple of the minimal
1869 * block size. Give up if shrinking would make the request too
1872 new_len = min((unsigned long long)io_u->buflen,
1873 zbd_zone_capacity_end(zb) - io_u->offset);
1874 new_len = new_len / min_bs * min_bs;
1875 if (new_len == io_u->buflen)
1877 if (new_len >= min_bs) {
1878 io_u->buflen = new_len;
1879 dprint(FD_IO, "Changed length from %u into %llu\n",
1880 orig_len, io_u->buflen);
1883 td_verror(td, EIO, "zone remainder too small");
1884 log_err("zone remainder %lld smaller than min block size %d\n",
1885 (zbd_zone_capacity_end(zb) - io_u->offset), min_bs);
1891 case DDIR_SYNC_FILE_RANGE:
1902 assert(zb->cond != ZBD_ZONE_COND_OFFLINE);
1903 assert(!io_u->zbd_queue_io);
1904 assert(!io_u->zbd_put_io);
1905 io_u->zbd_queue_io = zbd_queue_io;
1906 io_u->zbd_put_io = zbd_put_io;
1908 * Since we return with the zone lock still held,
1909 * add an annotation to let Coverity know that it
1912 /* coverity[missing_unlock] */
1916 if (zb && zb->has_wp)
1921 /* Return a string with ZBD statistics */
1922 char *zbd_write_status(const struct thread_stat *ts)
1926 if (asprintf(&res, "; %llu zone resets", (unsigned long long) ts->nr_zone_resets) < 0)